Essays on Small and Micro Enterprises by Mongoljin Batsaikhan B.A., University of Tokyo, 2006 M.A., Brown University, 2007 Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in the Department of Economics at Brown University Providence, Rhode Island May 2012 © Copyright 2012 by Mongoljin Batsaikhan This dissertation by Mongoljin Batsaikhan is accepted in its present form by the Department of Economics as satisfying the dissertation requirement for the degree of Doctor of Philosophy. Date____________ _______________________________________ Andrew D. Foster, Advisor Recommended to the Graduate Council Date____________ _______________________________________ Louis Putterman, Reader Date____________ _______________________________________ Nathaniel Baum-Snow, Reader Approved by the Graduate Council Date____________ _______________________________________ Peter M. Weber, Dean of the Graduate School iii Vita The author was born in Ulaanbaatar, Mongolia on June 22, 1981. He attended the University of Tokyo and graduated with a degree of Economics in March 2006. During this time he joined several field surveys in Mongolia, which led him to pursue his career in Development Economics and to begin a doctoral program in Economics at Brown University in August 2006. This dissertation represents the culmination of his efforts to complete the requirements for the Degree of Doctor of Philosophy in the Department of Economics at Brown University. iv Acknowledgements This dissertation could not be completed without the guidance from my advisors. I wish to thank Andrew Foster for his support and helping me to acquire critical research skills and deeper insights of Development Economics. I also wish to thank Louis Putterman for introducing me to Experimental Economics and guiding me through this exciting field. I thank Nathaniel Baum-Snow for investing a tremendous amount of time in helping me to apply Urban Economics to my dissertation. This dissertation have greatly benefitted from comments from Pedro Dal Bó, among other professors and graduate students in Department of Economics at Brown University. I would like to thank my co-authors in Chapter III for investing their time on conducting the experiment in other locations and contributing to the paper. I am grateful to the Sales Network Department of Newtel LLC for providing me with the data and helping me to conduct my field surveys and laboratory experiments. I also thank local research assistants, office managers at Mobicom centers in Ulaanbaatar, Mongolia, and laboratory managers and instructors at the Science and Technology University of Mongolia for making this project into fruitful data collection. This dissertation is generously funded by Hazeltine Fellowship, the William R. Rhodes Center for International Economics and Finance at the Watson Institute for International Studies, Newtel LLC, the National Science Foundation, Brown Graduate School and Department of Economics at Brown University. I would like to thank Louis Putterman, Ross Levine and Brian Knight for funding opportunities, and Mary Ellen Brown, Angelica Vargas, and Carrie Honeman for administrating my complicated budgets. Finally, I wish to thank my family and friends for constant support throughout my education. I am particularly grateful to Munkhjin Batsaikhan, Ariunaa Jambaldorj, and Batsaikhan Byambaa for helping my data collection and being the most reliable people in the field. The papers also have greatly benefited from Alissa MacMillan, Niki Clements and other writing center associates at Brown University. v Contents List of Tables and Figures……………………………………………….……………………………..vii I. GROWTH FOR MICRO ENTERPRISES IN DEVELOPING COUNTRIES: EVIDENCE FROM PHONE CARD WHOLESALERS IN MONGOLIA ................................ 1 I.1 Introduction..................................................................................................................................... 1 I.2 Background and Data....................................................................................................................... 6 I.3 Conceptual Framework .................................................................................................................... 9 I.4 Empirical Strategy .......................................................................................................................... 13 I.5 Interpretation ................................................................................................................................ 17 I.6 Conclusion ..................................................................................................................................... 21 II. TRUST, TRUSTWORTHINESS, AND SUCCESS IN BUSINESS ........................... 22 II.1 Introduction .............................................................................................................................. 22 II.2 Experiments and Data ............................................................................................................... 26 II.3 Conceptual Framework.............................................................................................................. 30 II.4 Empirical Results ....................................................................................................................... 32 II.5 Conclusion ................................................................................................................................. 38 III. SECURITY OF PROPERTY AS A PUBLIC GOOD: EXPERIMENTAL BEHAVIOR AND SOCIAL NORMS IN FIVE COUNTRIES .................................... 40 III.1 Introduction .............................................................................................................................. 40 III.2 Experimental design and predictions ......................................................................................... 45 III.3 Results ....................................................................................................................................... 57 III.4 Conclusion ................................................................................................................................. 75 BIBLIOGRAPHY ....................................................................................................................... 80 APPENDIX………………………………………………………………………………..…….….…….…..133 vi List of Tables and Figures Tables: ............................................................................................................................................................................................................. 89 Table I.1. The Effect of Mobile Banking on Wholesaler’ Business Expansion............................................................................................. 89 Table I.1. The Effect of Mobile Banking on Wholesaler’ Business Expansion............................................................................................. 90 Table II.1.b: Correlation between variables ................................................................................................................................................ 92 Table II.1.c: p-values of Mann-Whitney tests of difference in trust (t1) and trustworthiness, (t2p) ......................................................... 92 Table II.2: Correlation between Trust and Sales ......................................................................................................................................... 93 Table II.3: Correlation between sales and absolute deviation from the optimal trust, own answers used............................................... 94 Table II.4: Correlation between sales and absolute deviation from the optimal trust, average t2p used................................................. 95 Table II.5: Sales and dev2, the absolute difference between t2p and its subjective probability (same alpha and beta across subjects) . 96 Table II.6: First risk measure and its correlation with other variables ....................................................................................................... 97 Table II.7: Second risk measure and its correlation with other variables................................................................................................... 98 Table III. 1: Wealth production schedule .................................................................................................................................................... 99 Table III. 2: Treatments and group (subject) numbers by site .................................................................................................................. 100 Table III. 3: Predicted and actual average behaviors and outcomes by treatment .................................................................................. 101 Table III. 4: p-values of Mann-Whitney tests of difference in allocations across treatments .................................................................. 102 Table III. 4b: p-values of Mann-Whitney tests of difference in allocations across treatments among entrepreneur subjects ............... 103 Table III. 4c: p-values of Mann-Whitney tests of difference in allocations across treatments between student and entrepreneur subjects ..................................................................................................................................................................................................... 103 Table III. 5: Kruskal-Wallis tests of difference in allocations across countries ......................................................................................... 104 Table III. 6: Comparison of institutional choice of student subjects in five countries and entrepreneur subjects .................................. 105 Table III. 7: Comparison of the number of promises to steal between entrepreneur and students subjects in S-CHAT treatment ....... 106 vii Figures: .......................................................................................................................................................................................................... 107 Figure I.1. Spatial Distribution of Wholesalers in Ulaanbaatar ................................................................................................................. 107 Fugure I.2. Spatial Distribution of Mobicom Retailers in Ulaanbaatar ..................................................................................................... 108 Figure I.3. County Populations in Ulaanbaatar ......................................................................................................................................... 109 Figure I.4. Comparison of Actual versus Predicted Number of Retailers by Each County ........................................................................ 110 Figure I. 5. Average Distance from Wholesalers to New Retailers in Entry Mouth .................................................................................. 111 Figure II.1: Number of subjects for each trust level in student and entrepreneur subjects .................................................................... 112 Figure II.2: Percentage of subjects for each trust level in student and entrepreneur subjects ............................................................... 112 Figure II.3: Average amount and percentage of t2 given t1 by subject groups........................................................................................ 113 Figure II.4: Expected earnings for any given t1, calculated from own answers ....................................................................................... 113 Figure II.5: Correlation between trust and sales ...................................................................................................................................... 114 Figure II.6: Correlation between sales and absolute deviation from the optimal trust ........................................................................... 114 Figure II.7: Correlation between the average of t2p and its subjective probability ................................................................................. 115 Figure II.8: Correlation between sales and the deviation from the optimal trust, subjective probability calculated from alpha and beta that varies across individuals .................................................................................................................................................................... 115 Figure III. 1.a: Session timelines for each treatment ................................................................................................................................ 116 Figure III. 1.b: Timelines of stage games for each treatment ................................................................................................................... 117 Figure III. 2.a: Average allocation to production by period and treatment .............................................................................................. 118 Figure III. 2.b: Average allocation to collective protection by period and treatment .............................................................................. 118 Figure III. 2.c: Average allocation to private protection by period and treatment .................................................................................. 119 Figure III. 2.d: Average allocation to theft by period and treatment ....................................................................................................... 119 Figure III. 2.e: Average earnings by period and treatment ....................................................................................................................... 120 Figure III. 3: Average share of endowment allocated to each activity by country and treatment ........................................................... 120 viii Figure III. 4.a: Incidence of property crimes and allocations to private protection, all treatments......................................................... 122 Figure III. 4.b: Incidence of property crimes and allocations to collective protection, all treatments .................................................... 123 Figure III. 4.c: Incidence of property crimes and allocations to collective protection, period one only .................................................. 123 Figure III. 5.a: Perception of safety and allocations to private protection, all treatments ...................................................................... 124 Figure III. 5.b: Perception of safety and allocations to collective protection, all treatments .................................................................. 124 Figure III. 5.c: Perception of safety and allocations to collective protection, period one only ................................................................ 125 Figure III. 6: Governance index and share of individual votes for the mandatory scheme ...................................................................... 125 Figure III. 7: Trust Index and production in CHAT treatment. .................................................................................................................. 126 Figure III. 8a: Average Allocation to Private protection by age group and period ................................................................................... 126 Figure III. 8b: Average Allocation to Theft by age group and period ........................................................................................................ 127 Figure III. 8c: Average Earnings by age group and period ........................................................................................................................ 127 Figure III. 9a: Average Earnings by treatment and period among Entrepreneurs.................................................................................... 128 Figure III. 9b: Average Theft Allocation by treatment and period among Entrepreneurs ....................................................................... 128 Figure III. 10a: Correlation between Sales and Theft Allocation in the first period ................................................................................. 129 Figure III. 10b: Correlation between Sales and Production Allocation in the first period ........................................................................ 129 Figure III. 11a: Distribution of Theft Allocation by period ........................................................................................................................ 130 Figure III. 11b: Distribution of Production Allocation by period ............................................................................................................... 130 Figure III. 11c: Distribution of Collective Protection Allocations by period .............................................................................................. 131 Figure III. 11c: Distribution of Private Protection Allocation by period .................................................................................................... 131 Figure III. 12: Coefficient of retaliation to theft over 24 rounds .............................................................................................................. 132 ix I. Growth for Micro Enterprises in Developing Countries: Evidence from Phone Card Wholesalers in Mongolia I.1 Introduction Since private companies contributed a significant amount to OECD countries’ development, it has been recognized that sustainable development must begin with a foundation in private sector contributions. Most studies in development economics pay attention to household economics. And it is a known fact that developed countries developed without international aid, large fractions of which nowadays are directed to households, including investment in education and health. However, the lack of available data due to imperfect tax and registration systems in developing countries makes research difficult. For example, there are numerous studies on the missing middle class, while there are a few on the missing middle-size firms. We have not quite studied why small firms in developing countries stay small for a long time and why only a few succeed in expanding their business. Development economics has been concerned with establishing the factors that transform an under-developed economy into a developed one. While Lewis (1954) claimed that there are unlimited supplies of labor with a very low or negative productivity in the developing world, his claim also applies to the large number of small and micro enterprises (SMEs) in developing countries that fail to grow. The transformation of these 1 unproductive or subsistence SMEs into productive entrepreneurs is the main concern of this study. At macroeconomic level, Schumpter (1942) emphasized the role of innovation and entrepreneurs in economic growth, and Aghion and Howitt (1992), opening the field to an enormous amount of literature, documented the importance of technological improvement. Despite the macroeconomic and theoretical implications of this topic, little work has been done to identify the causes of transformation at a micro level. This paper contributes to the literature by taking three different approaches. First, it contributes to the urban and development economics by introducing new evidence that technology can build an alternative infrastructure and that this alternative infrastructure reduces the cost for small firms and increases their size. Second, I develop a simple model to estimate the cost function and to show the importance of cost for small firms in developing countries. Although traditional industrial organization papers documented the importance of cost for entry in developed countries, little evidence has been documented in developing countries. Third, I show some microeconomic evidence about the channel of Schumpterian growth through an analysis of positive externalities from innovations by large companies to small enterprises in developing countries. Figure I.6 shows that new technology introduction by a large company has a significant impact on the entry behavior of small firms. Even though innovations are not generated by small firms, they benefit from the innovations of large firms, thus expanding their business. Compared to the smooth size-distribution of firms in developed countries, medium and large sized firms are lacking in developing countries (Beck et al 2004 and Herrera and Leora 2005); it has been a puzzle in economics literature why the size distribution is significantly different and what factors hinder the growth of small firms in 2 developing countries. Schoar (2009) argues that there are systematically different types of entrepreneurs in developing countries and she separates subsistence entrepreneurs from transformational ones. Subsistence entrepreneurs run small businesses that provide alternative employment opportunities; such businesses do not expand or create employment opportunities for others. In contrast, transformational entrepreneurs achieve rapid business growth and create job opportunities for others. De Mel et al. (2008) and Ardagna and Lusandi (2008) also show that there are significant differences between these two types of entrepreneurs, and that subsistence SMEs do not grow into transformational ones in most developing countries (Mondragon-Velez et al 2008). Several hypotheses have been proposed as to why transformation is restricted in developing countries. One claim is that cultural and regulatory constraints, such as entry barriers, corruptions, and appreciation for entrepreneurs, restrict SME growth (Adragna 2008, Klapper et al 2010 and Schrabl 2008). Bruhn et al. (2010) hypothesize that managerial capital, an important factor for firm growth, is lacking in developing countries. In addition, large fractions of the literature suggest that access to capital, or lack thereof, contributes to restricted transformation. The results are mixed and the impacts are not always positive for SMEs, or at least for subsistence SMEs. Karlan and Morduch (2010) summarize the results and conclude that microfinance alone cannot solve the problem of restricted economical transformation in developing countries. Although studies in microfinance have been the leader in this case and gradually data collections are being extended to non-financial variables, data collection on firms is not easy and there is little known about firms in developing countries. The lack of available data has been the main reason why small firms have not been studied in 3 development economics. Official statistical measures are in general not as good as those in developed countries. Furthermore, small business owners usually do not keep records of their business activities and in most cases avoid taxes, creating unreliable data for official records. Recently, led by the World Bank, surveys on firms started in developing countries, which allows for more thorough analysis of their structure. To overcome this difficulty, I collected data on small firms and merged to an existing data set. In developing countries, typical businesses are small shops and they are all over the cities. I concentrate my study on prepaid cell phone recharging cards, which are often sold at these shops. These cards are identical and the price is fixed by mobile carriers. Therefore, this particular product gives economic agents only a few choice variables, which enables me to look at the business structure and pattern more clearly than in other cases. Another advantage of looking at this particular industry is data availability. By construction, the mobile phone industry collects data naturally for every transaction. I take advantage of this and merge my survey data to a data set collected by server computers. This gives me more accurate data on businesses in developing countries. Moreover, mobile phones are becoming more and more common developing countries. Over the past years, the number of subscriptions increased much faster in developing countries than in developed countries; even the lowest income group in developing countries tends to have their own cell phones. Cell phones have become one of the most important tools for business in developing countries, where landlines are not necessarily developed. The cell phone industry has created additional infrastructure benefit of which firms and their customer can enjoy. Therefore, understanding the 4 business structure of this particular industry itself can be a contribution to the current development economics. Developing countries lack in infrastructure and it is usually hard to build infrastructure due both to its high cost and corruption in developing countries. However, studies suggest that infrastructure has a positive impact on economic activities including urban growth, trade, real income, employments, productivity, and real GDP (Baum-Snow 2007, Dinkelman 2010, Donaldson 2010, Duflo et al 2009, Duranton and Turner 2011, 2012, and Fernald 1999, ). Moreover, recent literature show that the infrastructure that affects economic outcomes is not limited to just traditional physical infrastructures. Especially, improvement in mobile technology positively affects the market (Jensen 2007, Allen 2011). In this study, I show that mobile technology builds an alternative infrastructure, which is developed by private sector firms, and the new infrastructure reduces costs for small firms. To study the impact of a technology, following the industrial organization literature, I estimate the cost for small firms from a structural model. This exercise shows that the cost is one of the important factors for small firms in developing countries and adds an additional hypothesis for why small firms do not grow in developing countries. The importance of cost has been documented empirically in industrial organization literature (for details, see Berry and Reiss 2007). However, due to data limitation, similar studies have not been possible in developing countries. Entry models require structural estimations and those estimations require detailed data on individual firms. However, many small firms in developing countries are not officially registered and even if they are registered they do not reveal business data to economists due to tax issues. To overcome 5 this difficulty, I have collected my own data on a large set of firms in Ulaanbaatar, Mongolia and combined the data with computer collected sales data provided by a mobile company in the country. The data set allows me to analyze the entry behavior at firm level and estimate the cost function. The estimation of cost function and evidence from a natural experiment demonstrate that a cost-reducing technology can help SMEs expand their business and provides microeconomic evidence of Schumpeterian growth in a developing country. Using an estimated model, I show that a cost-reduction technology helps SMEs grow and improves social welfare. The structure of the paper is as follows. In Section I.2, I introduce the telecommunications industry in Mongolia, and the data used in this study. In Sections I.3 and I.4, I build a simple model to clarify the business structure, estimate the cost function and explain the empirical strategy used. I interpret the results in Section I.5 and conclude in Section I.6. I.2 Background and Data Cell phone card wholesaling was started in 2006 by Mobicom, the largest mobile phone provider in Mongolia. Mobicom did not have local offices in most areas, whereas their customers were all over the country and most of the customers used prepaid service, which requires buying cell phone cards. The wholesaling system increased the geographical coverage of its card dealers. Initially, Mobicom let the dealers become 6 wholesalers and buy cards at a discounted price. In exchange, wholesalers made contracts with other retailers to sell the cards. The number of retailers who sold the card significantly increased because dealers who chose to become wholesalers facilitated the geographical expansion of contract retailers. Wholesalers buy prepaid cards from Mobicom at a 5% discount and distribute them to other retailers. Retailers buy cards from the wholesalers at a 3% discount and sell them directly to customers. The cards have a specific monetary value. For example, the typical amounts on the card are 1000MNT (MNT stands for Mongolian Tugrik and 1250MNT is approximately 1USD), 3000MNT and 5000MNT; customers charge their cell phone with these prepaid amounts. Therefore, there is no negotiation on the price of this product and retailers sell the cards by the price written on the card. Customers should be indifferent to the store they choose because the quality and price of the product is fixed, unless the stores make other managerial efforts through other products to attract customers. These cards are small enough not to occupy a big space in stores and are most often sold instantly. In early 2008, Mobicom introduced a service called Top-up, which allows customers and retailers to transfer credit over cell phones. This is an electronic alternative to physical cards and makes up 90% of the total sales of prepaid cards today. When customers come to retailers to buy cards, they no longer have to buy the physical card. Instead, they pay the amount of the card to retailers, who in return instantly send the credit from their cell phones to the customer’s cell phone. The role of retailers in this case is to collect the money from customers in different locations. However, retailers also have to buy credits from wholesalers, who, in turn, have to buy credit from Mobicom. 7 The advantage of this system is that Mobicom automatically collects data on all transactions made through the top-up service. Mobicom also has information, including the wholesalers’ addresses, contract dates, and the cell phone numbers of all of the retailers under each wholesaler. The first data set was delivered by a legal agreement between Mobicom and Brown University. The data include the business addresses of wholesalers in Ulaanbaatar, the capital city of Mongolia, monthly transactions by every retailer since 2009; and the matching information of retailers and wholesalers. Because the address information in Ulaanbaatar is not digitalized, I geocoded the location of the wholesalers into a digital map. There are 1203 wholesalers in Ulaanbaatar. I also conducted a field survey to collect the data on the locations of retailers. I hired survey enumerators to walk around the city and identify all Mobicom signs, which indicate the stores that sell the prepaid cards from Mobicom. In this case, enumerators used GPS devices to acquire the GIS data of the exact location. This covered approximately 1200 retailers. For those without an identified location, we pulled out all retailers with positive sales in May 2010 and asked by phone for the location. Given this information, enumerators went to the retailers to acquire the GIS data. This survey collected the locations of 2734 active retailers in the city. The spatial distribution of wholesalers and retailers is shown in Figures I.1 and I.2. Mobicom agreed to share the sales data between wholesalers and retailer, and retailers and customers for each month since 2009. Whenever digital cards are transferred from one mobile phone to another, including the one between wholesalers and retailers, 8 and between retailers and customers, the data are collected by Mobicom’s server computers. The data are shown in monthly aggregate level by retailers. The data also show the network between wholesalers and retailers. Therefore, summing up the data at a wholesaler will give the sales data for that wholesaler. I also obtained the data for demographic characteristics of counties in Ulaanbaatar. The data include the population, density, and area of each county, except a distant county, Nalaikh. Excluding Nalaikh, there are 120 counties in the city and each of them will be considered as a market in this paper. Population distribution is shown in Figure I.3. I.3 Conceptual Framework In this section, I develop a simple theoretical model to understand the economic structure of the industry and to make clear the empirical estimation. Since I do not observe the cost for wholesalers and retailers, I estimate their cost function using the model that I develop in this section. In most of the literature, the cost is either estimated from a model or surveyed by multiple methods, which usually suffer various reporting bias, especially in developing countries. The cost data has been collected by surveys in development literature. However, asking small business owners in developing countries what the exact cost of their business is always a challenge. Most of them do not know how to define cost. In case of shop owners, even after explaining what the business cost is for them, a large fraction of them do not keep records of how much of the products that they have is consumed by their family. On contrary, studies in industrial organization estimates the cost from a structural model with rich data sets. Since my data set allows a simple 9 structural estimation and the structure of the business in this case is fairly simple, I follow the industrial organization literature in this study. To simplify the complexity of optimization, I assume that firms decide whether or not to enter each market independently of their decisions about entry into other markets. Firms, which are wholesalers, are indexed by i, whereas markets, indexed by m, are counties in Ulaanbaatar, Mongolia. The profit function is defined as the difference between their market share as revenue and costs. Since the cell phone cards are very small and thin, the only cost is the transportation cost from the firm to the market, which is , and the fixed cost of having an additional firm, , once having entered the market. Fixed cost in this case is more like a marginal cost of adding more retailers in each market once entering the market. But this cost varies by market depending on how dense the market is. I also introduce a market specific idiosyncratic profit into the profit function. This can be an unobserved match between the firm and the market, or entrepreneurial and managerial efforts. where is the profit for firm i when entering the market m is the number of retailers with which firm i makes contracts in market m is the number of retailers with contracts with other firms 10 is the demand of the market is the fixed cost of adding one more retailer in the market once having entered the market is the idiosyncratic profit for firm i entering market m Therefore, is the market share by firm i in market m, and is the transportation cost of firm i in market m. A firm maximizes its profit in each market with respect to the number of retailers in that market and decides on how many retailers to make contracts with. The first order condition with respect to : By re-arranging the equation, the number of retailers for firm i in market m is By plugging this into the original profit equation: Firms with non-negative profits will enter the market. Therefore, I will find the firm that has zero profit. I specify this firm with an asterisk. For this boundary firm, the profit is zero. In other words: 11 Solving this quadratic equation for as a function of transportation cost and idiosyncratic error of the boundary firm: I define t as the difference between firm specific transportation cost and the idionsyncratic profit. For the boundary firm noted with an asterisk, t would be: I. Firms with t lower than t* will enter the market m because their cost is lower than the revenue. In other words, the number of retailers in the market should be equal to the cumulative density of t up to t*. II. where g(t) is the density function of t. It is the difference between the transportation cost and idiosyncratic terms. There are two unknowns in Equation (1) and (2): and . The equilibrium number of retailers in market m will be the solution to this system of equations. 12 I.4 Empirical Strategy The solution to the system of equations presented above predicts the number of retailers in any given market. I use the method of simulated moments (McFadden 1989) to estimate the parameters of the model. However, in this case, I simulate the idiosyncratic error term from a distribution. Since the model has a constant, I assume that the mean of this distribution is zero. Since these terms represent revenue, I first assume that a log- normal distribution and it will be the preferred distribution. However, I also show that normal distribution can be used but the fit is not as good as log-normal distribution. The survey data used in this study provides me with all possible distances among current retailers and all wholesalers. However, since the survey data do not contain information for all potential retailers, I assume that all possible distances contained in the data are the proxy for distribution. In addition, the transportation cost function should be specified; most urban literature assumes that this function is either linear or quadratic. In this case, I assume the former. The final parameter that must be specified is , the fixed cost of having an additional retailer once having entered the market. The idea behind this parameter is that there must be a small cost of traveling within the market. Since there must be more retailers in denser markets, I specify the fixed cost as a fraction of the density of the market. The fraction coefficient (f) does not vary across markets. In other words, 13 It is clear from the model that I do not estimate demand in each market. The underlying assumption is that firms’ locations are exogenous and they perfectly observe the demand of the market before entering. In other words, D is observable for firms. The justification of this assumption is that wholesaling system emerged in 2006 when the retailers at the time had an option to become a wholesaler after they had their fixed business location. Therefore, the current wholesalers were retailers in the beginning and their location was fixed before they become a wholesaler. In the estimation, I use two variables for demand. The first variable is the actual sales of cell phone cards sold in that market in May 2010. The second is the total population within that market. Clearly, the actual sales data provides a better prediction than the population, and the interpretations of the coefficients are straight-forward. However, I will also show that population provides a reasonable proxy for demand. Given the assumptions above it is possible to estimate the parameters of the cost function: First, I generate random variables from a log-normal distribution with mean zero for each firm and market combination. This represents the idiosyncratic profit. To obtain the cumulative density of the values, I calculate the term below for each combination and sort them in ascending order. 14 Next, I calculate the following equation for each order number j from the density above: I then calculate the difference between the two and choose the order number that gives the closest value to zero. Due to random variable generation, I do not expect the difference to be zero, which would represent equilibrium. However, I select the closest value to zero and identify this as the equilibrium number of retailers in that market. This is the simplest procedure to find the fixed point that gives the solution to the system of equations. The logic of this procedure is shown in Figure I.1. Figure I.1. Finding the fixed point: t* Sales, pr Profit, Revenue, Positive Cost profit Number of Entries Nm By repeating the procedure from an initial value of each parameter, I identify parameters that give the minimum of the summation of the difference between the actual number of entry and the estimated number of entry in each market. In other words, 15 I also specify the span of values for each iteration. After considering the memory that has to be used due to the large sample size, the value of was changed by 0.1, f was by 100, by 50, and by 0.333. Following this procedure above, the estimated values are: Figure I.6 shows the result of the predicted and actual number of entries in each market, sorted by predicted number of entries, when the above values are used. The x-axis specifies each of the 120 counties in the city in this estimation. The result indicates a very strong fit of the model. I also use normal distribution for idiosyncratic profit. The result is similar and is shown in Appendix, Figure A I.1. To show the estimation process more clearly, I also include graphs that show how the predicted number of entries in each market changes when parameters are altered. See Appendix, Figure A I.2, A I.3, A I.4, and A I.5. To emphasize the importance of cost structure in developing countries, I show several counter-factual experiments using the parameters estimated. First, I compute the number of retailers in each market if Mobicom were the only distributor. This is the 16 situation in which there is no wholesaling system. This is an important exercise since a significant amount of business in developing countries use intermediate wholesalers. This might be due to the high transportation cost in developing countries. Without wholesalers, Mobicom would have very few retailers in each market due to its huge travel. Only 14 counties are estimated to have a retailer that serves Mobicom prepaid cards. This result shows that the wholesaling system increased the market coverage for Mobicom, making the product available in distant markets. Customers potentially benefitted from this change since their nearest retailer have become close to their location. To differentiate the effect of cost from that of density, which is usually considered to be the main determinant of spatial distribution of firms, I use a simple probit model to show what determines the entry behavior of wholesalers. Table I. 2 shows the results from probit regressions and it is clear that the distance from wholesaler to market or county is strongly significant after controlling for density, whereas density is no longer significant when the distance is controlled for. I.5 Interpretation The fit of the model indicates that, in this case, cost plays an important role for SME business. Since the cost is the major factor in SME business structures, I examine how adoption of a novel cost-reducing technology affects SME business structure and growth. Generally, wholesalers collect sales from their retailers once or twice a week. Upon collection, wholesalers deposit the funds into their Mobicom account, and the money is converted into digital cards. These digital cards are then placed into retailers’ mobile 17 phones and sold directly to customers once retailers have confirmed the amount in their own mobile banking account. Since these transactions rely on manual pick-ups and deposits, wholesalers must travel between their retailers and Mobicom, which generates substantial travel costs. In late 2008, Mobicom began offering mobile banking services to its wholesalers and retailers. Both parties can use this service to digitally transfer money between them and into their own Mobicom accounts. For example, when retailers need to buy new cards and have sufficient funds in their bank accounts, they can use an application on their cell phone to transfer money from their bank account to the wholesalers. Upon confirmation of the transfer, wholesalers can deposit the sales into Mobicom’s bank account. Once Mobicom confirms the deposit, the equivalent number of digital cards are transferred into wholesalers’ cell phones, and then to retailers’ cell phones. Mobile banking reduces the frequent and obligatory travel to retailers and Mobicom offices to pick up or deposit sales; thus, this service significantly reduces travel costs for wholesalers. Typically, Mobicom gradually introduces new products into different market. Mobile banking was first introduced to the Center Point office, one of four Mobicom offices, in October 2008. Wholesalers who are registered under this office could take advantage of the service, whereas all other wholesalers were not permitted to use it. Using this key difference between Center Point wholesalers and other Mobicom wholesalers, I show that the new mobile banking technology only affected the entry behavior of wholesalers registered at the Center Point office. Figure I.6 shows the average distance between a wholesaler and their new retailers by retailer entry month for 18 the four Mobicom offices. There is a sharp increase in retailer distance for Center Point wholesalers who travelled to obtain new retailer contracts in October 2008, when mobile banking became available to them. For an average month, the wholesaler-retailer distance for Center Point wholesalers is 6870 meters. However, in October 2008, wholesalers traveled more than 28000 meters to obtain new contracts. This indicates that mobile banking reduced travel costs and permitted these wholesalers to obtain contracts at greater distances. There are only three wholesalers who did not use mobile banking at the Center Point office. All three were registered immediately before our data were generated, in either April or May 2010. All 27 retailers that registered under Center Point in October 2008 used mobile banking. To better estimate the effects of mobile banking, I regress the distance from one wholesaler to each of their retailers on the interaction of the following dummy variables: whether the wholesaler uses mobile banking, whether the wholesaler was registered at Center Point office, and whether the retailers were registered in October 2008. Additionally, I account for the month and Mobicom office fixed effects, and whether the wholesaler use mobile banking technology. The results of this analysis are shown in Table I. 1; the most conservative estimate shows that wholesalers traveled 21664 additional meters to obtain contracts with retailers after the technology had been introduced. This effect is most likely due to a reduction in travel cost because of the use of mobile banking. Using the coefficients of the travel cost function estimated earlier, I calculate the benefit of mobile banking for each wholesaler-retailer combination in each month. On average, the monthly savings for a wholesaler per retailer is 19 500+13.333*21664=289346MNT when mobile banking is used. This is slightly more than $240 or 38% of average retailer monthly sales if January 2009 sales data are used. The sales data are available starting January 2009. The average retailer sales in January 2009 was 749378MNT; thus, in January 2009, the retailer benefit from technology improvement was about 38% of sales. The point estimate for this amount is 396135.5 MNT when the sales of January 2009 is regressed on the interaction of dummy variables of whether the wholesaler uses the mobile banking, whether the wholesaler was registered at Center Point office, and whether the retailer was registered in October 2008, and accounted for monthly and Mobicom office fixed effects. Although this amount is not statistically significantly different from zero, we cannot reject that the amount is different from 289346MNT, which is the estimated amount using the lowest point estimate for distance and the travel cost function estimated earlier from the model. The total benefit of mobile banking technology would be the summation of all possible benefits from future months for retailers and the benefit for customers. Although it is difficult to estimate the welfare improvement for customers, the social benefit from the technology is at least the amount that the wholesalers saved due to its implementation. Nevertheless, these data show that a cost-reducing technology helps SMEs expand their business, and demonstrates that the magnitude of this expansion is approximately 16% of retailer monthly sales, which is also 16% at the wholesaler level since the number is multiplied by the average. 20 I.6 Conclusion In this paper, a straightforward model for SMEs that enter different markets is established and the parameters of the cost function are estimated with the data collected from a field survey and by server computers of Mobicom. The result indicates that cost structure is important for determining an SME’s business model and where they locate retailer contracts. A simple test of this hypothesis involves analysis of the entry pattern after the introduction of a new cost-reducing technology. I estimated the effect of mobile banking technology, introduced in October 2008 on retailer and wholesaler sales, and showed that the benefit from the technology is approximately 16% of retailers’ and wholesalers’ sales each month. I estimated this effect using two different approaches: directly from the model and from reduced-form estimation. In both cases, the estimated magnitudes are similar: 396135.5MNT and 319991MNT. This benefit is due to cost reductions afforded by decreased travel and to the geographical expansion of the business. In most developing countries, SMEs do not grow and there is a lack of middle- sized firms. A substantial amount of literature is dedicated to investigating why SMEs do not flourish. This study provides microeconomic evidence of Schumpeterian growth, demonstrates that cost plays a major role in SME business structure and expansion, and shows that implementing new cost-reducing technologies can positively contribute to the growth of SMEs. 21 II. Trust, Trustworthiness, and Success in Business II.1 Introduction When there is no institutional support to force contracts among economic agents, social capital helps to improve market outcomes. In a world where social dilemmas exist and cooperation can improve economic outcomes for agents without any binding contracts, one has to commit to economic transactions with others on the basis of trust. Trust and trustworthiness among agents in this case can work as social capital to solve the social dilemma. This paper studies how trust can be associated with business outcomes and how successful entrepreneurs invest in social capital such as trust in developing countries. I use data collected from trust game, a commonly used method to measure trust and trustworthiness in experimental economics, and link this data to a unique data set collected from the field that measures real business outcomes. This paper is the first (to my knowledge) to link sales data to trust and trustworthiness measure collected in the laboratory, and to show that subjects with higher trust in the laboratory tend to be more successful in business in real life. In this setup, trust game also gives detailed information on how subjects invested in trust and allows me to analyze how successful entrepreneurs invest differently in the laboratory. Trust is a highly profitable action in this case and subjects who invest more in trust tend to be more successful in real business. Moreover, successful entrepreneurs tend to invest the amount that is close to the optimal level of trust by successfully guessing the underlying parameters of the population. 22 Defining trust is difficult and varies even across fields in economics. In this study, I use the trust measure which is commonly used in experimental economics (originally discussed in Berg, Dickhaut, and McCabe 1995) to measure individual trust levels (for a detailed discussion of the definition of trust, see Eckel and Wilson 2009). Trust game has been implemented in a large fraction of experimental literature to analyze the behavior of subjects facing a social dilemma problem in a laboratory setup. Trust game involves two players: A and B. Each of them is endowed with 6000 Mongolian Tugriks (MNT), approximately $5US. Player A makes the first move and decides how much of her endowment to send to Player B. Player A can send a zero amount as well. Any positive amount sent by Player A will be tripled by the experimenter and will be passed to Player B. After receiving the tripled amount, Player B has the option of sending back some or all of the received amount. Player B also has an option to not return any. The subgame perfect equilibrium for self-interested rational agents is when Player A sends zero and Player B has no option to return. However, in most laboratory experiments, subjects deviate from the game theory prediction and Player A sends a positive amount and Player B returns a positive amount as well. In experimental economics, the amount that Player A sends to B is considered as “trust” whereas the amount that Player B returns to A is considered as “trustworthiness”. If trustworthiness is expected in a society and exists as social capital to overcome social dilemmas, Player A has an incentive to trust Player B and invest a positive amount into trust. This situation can be observed in many occasions in developing countries where contracts are hard to enforce or are even non-existent. Business transactions in developing countries, in 23 particular, often require one individual to trust the other. In that sense, studying the relationship between trust and business outcomes in developing countries is important. Trust game has been implemented in many countries with various subjects in a laboratory setup. A meta-analysis of trust games shows that Player A sent 50.8 percent of their endowment to Player B and Player B returned 36.5 percent of that amount on average (Johnson and Mislin 2011). Subjects largely deviated from the game theory prediction. This paper shows that this result holds in Mongolian micro-entrepreneur subjects. In contrast, although trust is barely profitable in the meta-analysis, it is largely profitable in this case and successful entrepreneurs invest in trust more than less successful ones; the investment in trust is close to the optimal amount of trust to maximize the profit. Whether trust is profitable or not depends on trustworthiness. If Player B returns more than 33.3 percent of the tripled amount that Player A sent to B, trust is profitable. If trust is profitable, or at least if subjects predict that trust is profitable in their society (i.e. the subjects expect higher trustworthiness from other subjects), trusting others more can be the optimal strategy for Player A. In most cases of trust game in a laboratory setup, trust has been reported to be barely profitable (Johnson and Mislin 2011); returning only 36.5% of the tripled amount. Barr (2003) shows that trust has even negative return in Zimbabwe. However, I will show that trust is quite profitable among Mongolian microentrepreneurs; returning 51.5% of the tripled amount. Even if trust is profitable, the optimal amount of trust that maximizes earnings depends on the underlying distribution of trustworthiness and subjects’ perception if it. Player A maximizes its profit based on his or her belief of how much Player B will return. 24 Player A has to guess the distribution of the proportion that Player B returns out of each tripled amount that Player A sends and chooses the amount that maximizes its overall profit. During a two-person zero-sum game, Palacios-Huerta (2003) and Palacios-Huerta and Volij (2008) show that professional soccer players accurately capture the underlying distribution of actions that their opponents take and make the decisions that are close to the game theory predictions. De Mel, McKenzie and Woodruff (2008) show that microentrepreneurs in Sri Lanka with higher IQ measures are more successful in business. Thanks to the rich data set collected from trust game, this study allows me to observe the underlying distribution of trustworthiness and to relate that to actual trust level chosen by subjects. I further link this result to sales data from real life and show that successful entrepreneurs access the underlying distribution better and determine their optimal level of trust. This study successfully connects real life data to experimental outcomes. The external validity of experimental results cannot be established just in the lab and there has been a need to study behaviors of subjects outside of the laboratory. Even though recent studies have enriched the experimental literature by linking the data from the laboratory to real life, the number of studies is still very limited and the results are contradicting to each other and theoretical predictions. Karlan (2005) shows that trustworthiness predicts whether a borrower pays their loan back but trust predicts lower savings; concluding that trust is not something that promotes economic activity. However, it is hard to tell whether people without savings are successful in business or if the result is site specific. It might be that people with low savings are successful in business because they invest more. Bloom, Sadun, and Van Reenen (2012) theorize that higher trust (measured in a survey) 25 increases firm size, and their empirical evidence suggests that trust promotes firm decentralization and increases firm size. This paper will show that trust, but not trustworthiness, is in fact associated with higher sales (thus, larger firm size); this result is the first to be documented. Not only do I show the link between trust and business success measured in sales, but I also show that trusting more (or investing in trust more) generates higher earnings. This means that successful entrepreneurs are more trusting presumably because trusting others more is more profitable. I also show that successful entrepreneurs invest an amount that is close to the optimal level of trust that maximizes the earning at individual level as well. The paper is organized as follows. In Section 2, I explain the experiments and data in detail. Section 3 develops a simple model and Section 4 reports the results. I conclude in Section 5. II.2 Experiments and Data II.2.a Subjects I conducted three experiments with 120 subjects recruited from wholesalers and retailers in Ulaanbaatar, Mongolia. Wholesalers have contracts with local mobile companies to distribute prepaid phone cards to retailers in the city. Although most wholesalers are specialized in prepaid card wholesaling, retailers tend to sell other products, typically groceries, cleaning supplies, and alcoholic beverages. Sales data are drawn from one of the mobile carriers, Mobicom, the largest mobile carrier in Mongolia. The sales data 26 measure the revenue of Mobicom prepaid cards that each subject posted in May 2010. Mobicom prepaid card is now in digital form and is no longer a scratch card anymore. When a customer pays a certain amount to a retailer, the retailer deposits an equal amount from his or her business cell phone account to the customer’s cell phone account. This transaction is recorded at the Mobicom’s server computers. I have obtained the monthly total sales for each retailer from Mobicom. I observe which wholesaler is connected to which retailer and this information gives me the sales of each wholesaler as well. There were 2734 active retailers in May, 2010 and 1203 wholesalers registered with Mobicom (for details of the original population data, see Batsaikhan 2012). I circulated the advertisement of the experiment through local Mobicom offices to recruit retailers and wholesalers. I divided applicants into six groups and invited them to the laboratory. In each section of all six sections, 22 to 27 people were invited and there were two subjects who could not participate in the experiment due to over-enrolment and earned only the show-up payment of 5000MNT. I also recruited 40 students from local universities and conducted the same experiments with them. II.2.b Experiments Three experiments were conducted in each section; only the first and last experiments are used here to analyze trust. The first one is trust game described above. The second one is a property rights game and the data from this experiment is used for a different paper (for details, see Campos-Ortiz et al 2012). The last one is a risk measure game, similar to the risk game proposed in Holt and Laury (2002). 27 In trust game, each subject is asked to fill their choices as if they played the role of both Player A and B. Translation of the questions is attached in Appendix (AF II.1). In the beginning of the experiment, subjects are explained the rules of the experiment and how they are paid. After subjects make all possible choices, the experimenter randomly assigned subjects into Player A and B; matched them by their random order to calculate the trust game payment. This was also explained to subjects during the instruction to ensure subjects did not and would not know who they were paired with. The data include the measure of trust, the amount that each subject sends as Player A (noted as t1 variable), and all possible returns for any given t1 (noted as t2 variable). The variable t1 starts from 0 and increases by 500 until 6000MNT. For each t1, I calculated the percentage of return out of tripled amount of t1. The average of all possible percentages is the measure of trustworthiness, noted as t2p. In other words, Individual risk attitude might affect the decision of how much one would send as Player A since this decision faces the uncertainty of whether Player B returns any amount. Eckel and Wilson (2004) show that trust is not correlated with various risk measures. Bohnet et al (2008) indicate the existence of betrayal aversion as another type of risk. To affirm that trust measured here is not merely a simple risk attitude, I use different types of risk measures to analyze whether risk plays a role in trust game and sales. In the risk measure game, each subject answers five different questions, each of which offers a choice between a certain payment and uncertain payment. Translation of risk measure game is attached in Appendix (AF II.2). The first question asks if the 28 subject prefers 1500MNT with certainty or a 50% chance of getting 0MNT and 50% chance of getting 2700MNT. The last amount in the question increases as it progresses: 3000MNT, 3500MNT, 4000MNT, 4500MNT. After the subject makes the decision, the subject rolls a die and it determines the payment to the subject based on which question out of the five is chosen. If the die shows six, the experimenter asks the subject to roll the die until the subject gets a number less than 6, because the questionnaire has only five questions. The first measure, frisk, is an order number (from 1 to 5) for after which question the subject starts choosing risky assets consistently. 1 is the riskiest and 5 is the most risk averse. However, there are a small fraction of subjects who did not make a consistent choice and switched more than once within five questions. To include the subjects who responded inconsistently in the risk measure, I also use the risk measure discussed in Durante and Putterman (2009). This alternative measure, riskall, tries to capture the overall risk attitude of subjects. (See Appendix II.1 for details of constructing this risk measure) After the experiment, the subject takes an exit survey. The survey notes information on age, gender, the amount of bank loan, interest rate on the loan, whether the subject is a wholesaler, the length of business, occupation, and whether the subject owns property. The summary statistics are shown in Table II. 1. 29 II.3 Conceptual Framework To analyze how subjects make choices in the experiment, I have developed a simple model. The model will also help to estimate parameters and to exercise counterfactual experiments. Each subject evaluates the expected earnings from trust game. However, subjects face uncertainty about how much the second mover will return and have to rely on their beliefs concerning others’ behavior. There are three cases when subjects decide on the amount of t1 that maximizes the expected earnings. Cases t1=0 t1>0, t2=0 t1>0, t2= Probability NA 1-p p Earnings 6000 6000- is each subject’s belief concerning the fraction of the tripled amount that Player B will return. This variable varies across subjects depending on their personal beliefs concerning others. The subject will choose the that maximizes the expected earnings: If the subject chooses to send positive , the subject maximizes the expected earning as well: I assume that every subject in the experiment shares the same and it is a linear function of . Then the expected earnings when positive T1 is sent is: 30 If , subject chooses one of the corner solutions. In that case, t1=0 or t1=6000MNT. When , the expected utility function is concave and has an interior solution. The first order condition with respect to : And the optimal amount of trust that subject chooses is: The optimal expected earnings when positive T1 is sent: For any , this amount is larger than 6000. This means that subject will choose this particular for any . There are two cases of how and are determined. First, these parameters can be specific to individual subject. Each subject has different parameters and these can be estimated using 12 different responses for each . Second, every subject can share the same parameters. In this case, these parameters can be estimated using all responses made by all subjects: 1440 observations. In the data, I observe t1 for each individual. Once and are estimated, one can back out the p parameter from: 31 II.4 Empirical Results II.4.a Trust game On average, entrepreneurs sent 3092MNT or 51.53% of their endowment as Player A. This amount is consistent with other results from trust game, which indicate that subjects send 50.8% of their endowment on average. Student subjects, however, sent only 34.17% of their endowment. The trust level among student subjects is lower than the trust among entrepreneurs. This could be because the endowment might be perceived slightly less valuable for entrepreneurs than student subjects. However, Johnson and Mislin (2011) show that student subjects actually send more than non-students subjects when fixed effect model is used, and Anderson et al (2011) show that stakes do not make a difference in Player A’s decision in a ultimatum game (in which Player A makes a similar decision to trust game) unless stakes are significantly large. Moreover, results in this case are also consistent with the behavior among students in the property rights game that we conducted after trust game. Students tend to steal more from each other in the property rights game; the results are shown in Campos- Ortiz et al (2012). The trust comparison between student and entrepreneur subjects is shown in Figure 1 and Figure 2. Students’ is concentrated around the lower and the fraction is shown in Figure 2, to account for the difference in number of subjects (40 students and 120 entrepreneurs). 32 For any given amount of trust, entrepreneurs returned a greater amount than students as Player B. On average, entrepreneurs returned 51.5% and students returned 42.25%. Both numbers are higher than the results from other geographical locations documented in Johnson and Mislin (2011). In this case, trust is highly profitable. Figure 3 presents the proportion and the actual amount of t2 for every t1. In both cases, the percentage of t2 is higher than 33.33%, which is the boundary above which trust is profitable. Although the percentage of t2 falls as t1 increases, the earnings increase as t1 increases. I calculate expected earnings for each subject if the subject plays with herself, i.e. if the subject’s answer as Player B is what the subject expects from others. For each t1, define m as t1=500m. Then the expected earnings for subject i is: where is how much subject i returns when t1=500m. For every m, I calculate the average expected earning across subjects in order to calculate the mean earning when 500m is sent by Player A. The result is shown in Figure 4 and it indicates that expected earning reaches the maximum when the subject sends the whole endowment. This means that subjects should send higher t1 to maximize, and the outcome of trust from the subjects in this case is lower than the socially optimal trust. The result holds the same when average t2 is used instead of their own t2 for each t1. 33 II.4.b Trust and its link to non-experimental variables Trust is profitable and it pays off more when a larger amount is invested in this sample. If this is the case, subjects who invest in trust more should be more successful in real life since the social capital pays off. Figure 5 is a simple scatter plot of trust and sales, and it shows a strong correlation between the sales data and trust measure from the experiment. Table II. 2 exhibits the robustness of this correlation, even after controlling for trustworthiness, demographic variables, and other business related indicators such as bank loans, whether the subject is a wholesaler or not, and the risk measure. This result is the first to show that the trust measure from experiments is associated with real business data. Previous literature tended to find the correlation between trustworthiness and other real life data, but not with trust. For example, Karlan (2005) shows that trustworthiness predicts whether one pays her loan back, but trust from the laboratory experiment is likely to depict merely a risky action. In contrast, this study shows that trust is in fact more than a simple risky action and it predicts how successful one is in real business. However, consistent with Karlan (2005), trustworthiness in this study predicts the amount of bank loans (Table II. 1.b). Furthermore, I find that male subjects tend to send more as Player A while age or age-squared does not predict t1. Similar results about gender were found in Vietnam and Thailand as well (Carpenter et al 2004). Trustworthiness is correlated with age and business length in this case but risk measures predict neither trust nor trustworthiness. I will discuss different risk measures in Section 4.d in detail. 34 II.4.c Optimal trust The results in 4.a. Trust game indicate that subjects maximize expected earnings when they exhibit the maximum trust. However, the optimal trust in Section 4.a. is calculated from the average of all 120 subjects. Each individual subject might have a different optimum given her belief about how much Player B will return. Subjects might have different risk attitudes or different perceptions of expected return. The risk measure argument will be discussed in the next sub-section. Here, I analyze how the perception of others’ behavior affects the decision in three ways. First, I infer each subject’s perception from their own choice for t2. Second, I use the average of t2 in the sample and show how each subject deviates from that value. Third, I estimate their subjective parameters from the model developed earlier. First, I calculate individual earnings for each t1 using their own t2 for every t1 and found the t1 that maximizes expected earnings. In other words, I assume here that each subject’s own t2 is their guess for how much t2 others would return. Mathematically, for every i, find t1=500m that: This is the optimal trust for each subject and it might be different from their actual choice of t1depending on how accurately each subject guesses the underlying parameters. I calculate the absolute difference between the optimal and actual trust and show its association with real business data. Figure 6 shows that entrepreneurs with higher sales tend to deviate less from the optimal choice. The result is robust after controlling for 35 other variables, and it indicates that successful entrepreneurs optimize their expected earnings better than less successful ones (Table II. 3). Second, I calculate the optimal trust using the average of t2 for each t1. For each t1, I calculate the average percentage of t2p of all 120 subjects. I assume that this is the subjective perception for t2p for each subject and find the optimal trust that maximizes their expected earnings. Table II. 4 shows these results and confirms the results reported in the previous paragraph. Third, p in the model represents subject’s perception of t2p. As discussed in the end of the conceptual framework, there are two ways to estimate this parameter. One way is to assume that every subject shares the same and , and to estimate them from the data. The point estimates and standard errors (in parenthesis) from regressing t2p on t1 with all 120 subjects are: (0.0094) and (2.5695* ) In the equilibrium, I back out the p parameter from each subject’s choice of t1: I calculate each p according to this equation using the and estimated above and the subject’s own t1. The correlation between p and t2p is shown Figure 7. If subjects have the perfect information on others’ behavior, this correlation should be perfect as well. I calculate the difference between this subjective probability, p, and the realization of this, the average of t2p for each subject. I note the absolute difference between these two probabilities as dev2. Table II.5 exhibits the correlation between this variable and 36 sales, both with and without various controls. Contrary to the previous measures, this new measure does not indicate a statistically significant difference. The second way is to estimate and for each subject. However, since this has to be done with 12 observations for each subject, the point estimates give values that are larger than 1 or less than 0 for a probability parameter, which should be between 0 and 1. It also gives positive values for beta. To account for this problem, I dropped the values that are not feasible and created a new variable, psub_corr. The estimated and corrected values for every subject (of subjective probability of t2p and and with standard errors) are reported in Appendix Table II.1. In this case, the correlation between sales and the absolute difference (between subjective probably and actual t2p) is slightly negative but statistically insignificant with only 56 observations that have both data (Figure 8). Lastly, I calculate the magnitude of trust and sales. Table II.2 presents the correlation between trust and sales, and trust measure in this case varies from 0 to 6000. I take the coefficient from specification (6) of Table II.2, the most conservative estimate among the six specifications, and calculate the impact of trust. Increasing trust measure from 0 to 6000 increases sales by at least 9,828,000MNT. This is approximately 1.12 standard deviations of sales. II.4.d Risk Measure The first risk measure, frisk, does not explain the trust or trustworthiness measure, both with or without additional controls. It is not associated with sales data, either. Table II.6 shows the results. The second risk measure, riskall, also does not explain trust, trustworthiness, and sales (Table II. 7). Both risk measures do not affect the correlation 37 between sales and trust. These results indicate that trust here cannot be induced to the risky decision to gamble. II.5 Conclusion Trust functions as an essential social capital for economic activities. It is especially important in developing countries where legal contracts are not binding. Literature in experimental economics has set up social dilemma problems in the laboratory and shows that subjects use their social norms outside of the laboratory to overcome these dilemmas. However, the measures obtained in the laboratory do not necessarily correlate with variables outside of the laboratory. In particular, no studies so far showed that the trust measure from experimental economics is positively related to economic variables outside of the laboratory. This study implements the traditional trust game with small business owners in Mongolia to show that trust from experimental economics is in fact associated with sales in real business. The unique data in this study included the additional risk measures and real business data from the field. The results indicate that trusting others is a highly profitable action in this case and that successful business owners tend to trust others more. Moreover, successful entrepreneurs also invest close to the optimal amount of trust in different measures because they successfully access the underlying parameters of the population. 38 Trust in this case cannot be explained simply by risk attitudes of individuals and is associated with gender of the subject. Trustworthiness, in contrast, does not explain sales in real business but it does predict the amount of bank loans that subjects borrowed. Trustworthiness is also associated with age and the length of business but not with gender. The results from experimental economics are often questioned its external validity of the results. This study links the trust in the laboratory to real business data and shows that successful entrepreneurs show higher level of trust. Trust is highly profitable and serves as social capital to overcome social dilemmas in the laboratory. The results from this study indicate that successful entrepreneurs utilize trust to overcome social dilemmas and successfully access the information of underlying population, making higher profits by investing the optimal amount of trust. 39 III. Security of Property as a Public Good: Experimental Behavior and Social Norms in Five Countries 1 III.1 Introduction Private property might strike us as the antithesis of a public good. Yet efficient protection of individuals’ rights to property is to a large extent a problem of collective action. Where property rights are not protected by some combination of fear of penalties and voluntary norm-compliance, individuals are forced to devote time and other resources to defending whatever wealth they are able to obtain, and their incentives to invest and to produce may be greatly attenuated. Societies that fail to achieve well- enforced property rights can therefore be expected to be poorer than those that do. While social norms of desisting from theft contribute to a public good of secure property, private investment in defense of property (e.g., locks, alarm systems, barbed wire, and so forth) is also observed in every society. On top of that, well-functioning 1 Most of this chapter consists of the working draft of a joint paper by Francisco Campos-Ortiz, Louis Putterman, T.K. Ahn, Loukas Balafoutas, Mongoljin Batsaikhan, and Matthias Sutter. However, Batsaikhan alone is responsible for the material discussing the experiments with entrepreneur subjects and the extra treatment with student subjects in Mongolia, noted as S-CHAT later in the paper. The other authors are not responsible for any errors in that material, which appears on Section 2.e., pp13 of Section 2.f., Section 3.e., and all tables and figures discussed in these sections.Putterman, T.K. Ahn, Loukas Balafoutas, Mongoljin Batsaikhan, and Matthias Sutter. However, Batsaikhan alone is responsible for the material discussing the experiments with entrepreneur subjects and the extra treatment with student subjects in Mongolia, noted as S- CHAT later in the paper. The other authors are not responsible for any errors in that material, which appears on Section 2.e., pp13 of Section 2.f., Section 3.e., and all tables and figures discussed in these sections. 40 modern societies assign much of the task of protection to collective institutions—police forces, courts, prison systems—capable of protecting the property of large numbers of individuals and thus achieving economies of scale. Of course, the mix of norm compliance, private protection, and collective protection of property varies a lot across societies (Tabellini, 2008), making it an intriguing question how to protect private property most effectively. In this paper, we present an experiment on the protection of private property in five economically, institutionally, and culturally distinct countries: Austria, Mexico, Mongolia, South Korea and the United States. These countries cover five out of eight different regions in the world value survey cultural map (see Inglehart and Welzel, 2005, p. 63), allowing us to study how cultural and socio-political differences relate to experimental behavior of student subjects. In total, we have 795 participants across the five countries. We also compare the results from students to the results from business owners in Mongolia. 120 small business owners from Mongolia participated in the experiment and we analyze the correlation between the behavior in the laboratory and the real business data. Within each country, we study experimentally a world in which individuals, organized in micro-societies of five subjects, experience the temptation to devote resources to theft because allocations to production have diminishing returns but those to theft do not. We conduct four treatments in each country that share the same production, private protection and theft technologies. In three of the treatments we add a more cost-effective collective protection technology that in its most basic form suffers from free-riding incentives. 41 Behaviors within each of our subject pools respond to treatment differences in qualitatively similar ways: without collective protection, the frequency of theft is above the social optimum, but less than half of what standard theory would predict. When collective protection depends only on voluntary contributions, we observe statistically significant, but economically modest improvement. Only when collective action is taken by a binding majority vote on a tax or when group members can chat with each other―thereby establishing social norms of desisting from theft―we observe substantial efficiency gains through increased production (that substitutes resource allocation to theft or protection from theft). Despite these similarities in the reaction to treatment differences, the key contribution of our paper is to show significant cross-country differences that are related to socio-political factors within the countries of our experiment. First, subjects from countries with a higher incidence of property crimes or lower perceptions of safety tend to devote more resources to protection from plundering. Second, subjects from countries with high-quality political institutions are more prone to support the funding of collective protection with a mandatory tax. Third, the observance of explicit non-theft agreements is likelier to be sustained in countries with higher levels of survey-assessed trust. And fourth, theft is less observed among business owners with high sales in real life and it has affected the other business owners in the same cohort of during the experiment. Therefore, business owners achieved higher earnings than student subject in Mongolia. These findings suggest an important role of socio-political factors in determining the success of both formal and informal institutions at securing property rights and thus promoting productive activities. Our results can also be seen as consistent with the 42 hypothesis that differences in social capital help to explain differences in the quality of institutions and in economic performance (Knack and Keefer, 1997; Tabellini, 2008). A number of economists, including Grossman (1991), Hirshleifer (1991), Skaperdas (1992), and Grossman and Kim (1995), have engaged in the theoretical study of the security of property by analyzing equilibrium allocations of resources between productive, protective, and appropriative activities in the absence of either external enforcement or norms. The basic general equilibrium framework of such papers has been extended to investigate the conditions under which the introduction of government favors the allocation of resources to productive activities (Grossman, 2002), or the stability of the “anarchic” equilibrium (Hirshleifer, 1995). 2 Scholars have also explored the implications of appropriative activities for economic growth (Tornell and Lane, 1999), income distribution (Skaperdas and Syropoulos, 1997), and decisions to redistribute resources so as to avoid conflict (Grossman, 1994). Experimental studies motivated by the theoretical literature on appropriative conflict began in the late 1990s. Durham, Hirshleifer and Smith (1998) test, and largely confirm, the predictions of Hirshleifer’s (1991) “Paradox of Power” hypothesis, which suggests that weaker or poorer parties might improve their position relative to stronger or richer opponents by engaging in conflict. Carter and Anderton (2001) subject Grossman and Kim’s (1995) predator-prey framework to experimental testing. Duffy and Kim (2005) assess the stability of an equilibrium in which agents devote resources to production, predation and defense against predation, as well as the effect of the 2Hirshleifer (1995, p. 26) defines anarchy as "a system in which participants can seize and defend resources without regulation from above, is not chaos but a spontaneous order." 43 introduction of a government. Powell and Wilson (2008) study experimentally the evolution of institutions in stateless societies by analyzing the level of efficiency in a Hobbesian state of nature, then offering subjects the opportunity to pledge support to a non-binding agreement not to engage in theft. Finally, Kimbrough, Smith and Wilson (2010) permit theft and communication in a continuous-time experiment studying endogenous specialization and trade. Our experiment differs from those mentioned in several respects. Our subjects are neither assigned to nor required to choose between specialized producer or predator roles. Our focus on collective action, institutions, and the role of norms leads us to introduce a novel collective protection technology with greater social but lower private returns than private protection. We can compare the effectiveness of collective protection technologies when they are either introduced through formal voting or by voluntary contributions. We study how communication within groups affects social norms and the propensity to contribute to collective protection. Most importantly, however, ours is the first appropriation experiment to include subject pools in a diverse set of countries, which offers the possibility of assessing in a controlled way the operation of the same set of exogenously imposed institutions in different societies. The rest of the paper proceeds as follows. In Section III.2, we spell out our experimental design and discuss the predictions of standard economic theory. Section III.3 discusses our results. Section III.4 concludes. 44 III.2 Experimental design and predictions In each country, we study four treatments that share a common core structure. Additionally, we added a simplified version of one of the four treatments in Mongolia. In each treatment, subjects in sessions of fifteen or twenty participants are randomly assigned to fixed-partner groups of five to play a repeated game for 24 periods. Each period, each of the five is endowed with ten “effort tokens” that he or she must allocate among three activities: 3 (1) a productive activity that produces “wealth tokens” with diminishing returns; (2) theft directed at other group members’ accumulation of wealth tokens; and (3) private protection of own accumulations from theft. In the following, let mi , Ti = Σj≠i tij, and pi indicate the numbers of tokens individual i devotes to production (making wealth tokens), theft (where tij indicates the theft tokens i directs at a specific individual j), and private protection. To make theft a live consideration from the outset, each subject is endowed with an initial accumulation of 100 wealth tokens. Table III. 1 shows the production function from effort to wealth tokens. Marginal returns decrease from 15 wealth tokens to one wealth token. In contrast to production, each effort token devoted to theft transfers a constant 10 wealth tokens from targeted individual j’s accumulation at the beginning of the period to the targeting individual i with probability of success 1 – Pj, where 0 ≤ Pj ≤ 1 is j’s total level of protection stated as a probability that a given theft attempt against j will be thwarted. Each of the pj effort tokens j devotes to the private protection of her wealth accumulation raises Pj by 0.1. 3 A fourth activity, collective protection, is available in three of the treatments and will be explained when these treatments are introduced. 45 Each theft attempt by some individual i against individual j is governed by an independent random draw with the indicated probability, such that if, for example, i devotes 3 effort tokens to stealing from j this period, 30 tokens are transferred from j to i with probability 1 – Pj and 0 tokens are transferred with probability Pj.4 At the end of each period, subjects learn the number of wealth tokens they and each other group member accumulated by production and theft and the number lost by theft, and cumulative information on these categories is subsequently available in a “stats” screen that can be opened at any time.5 III.2.a. NCP treatment – No Collective Protection In our first treatment, which we call No Collective Protection or NCP, subjects determined their allocations among production, theft and private protection simultaneously. We made collective protection unavailable to give us a benchmark against which to measure its effects when present. For subjects in this treatment, the experiment as a whole consisted of six four-period phases separated by one minute breaks, as shown by Panel A of Figure III. 1.a, with the structure of the individual period shown by Panel A of Figure III. 1.b. 4 The only exception to the rule regarding number of wealth tokens transferred occurs when a targeted subject’s accumulation balance reaches 0. Because we prevent the balance from becoming negative, those engaging in theft can split between them no more than the total accumulation a targeted subject has at the beginning of a period. We stipulate that this splitting is proportionate to the number of tokens each had allocated to theft from the targeted individual. Given that statistics on others’ accumulations are always available and that those accumulations grow fairly large with time, the limitation was rarely binding. In more than 18,000 period*subject observations, the rule took effect only six times. 5 Group members have fixed letter identifiers throughout their sessions, although which actual individuals constitute a group is never made known to the subjects. Summary information on theft does not reveal who stole from whom, although that can be deduced if there is only one successful theft in a period. 46 Considering the per-period constraint mi + Ti + pi = 10, we ask what vector of non-negative integers (mi, Ti, pi) maximizes i’s expected utility subject to the behaviors of i’s four counterparts. Initially, assume i to be risk-neutral, strictly self-interested, and thus expected to maximize profits. If there were no allocations to protection, each effort token devoted to theft would yield for i 10 wealth tokens, so only the first three tokens devoted to production could compete with theft in terms of expected returns (see Table III. 1). Consider, then, the possible equilibrium vector (3, 7, 0) in which each member of a group devotes three tokens to production and seven tokens to theft. Would it ever be advantageous to deviate from it and to devote tokens to private protection? Assuming that others devote seven effort tokens to theft and that she herself is thus on average targeted by seven theft tokens, a subject expects to reduce her losses to theft by an average of 0.1x70 = 7 wealth tokens for each token devoted to protection, versus the ten she can gain from theft. So a risk-neutral agent would engage in no protective effort, and (3, 7, 0) would be the unique equilibrium of the stage game. Risk-aversion will not weigh in favor of private protection, since allocating tokens to it increases rather than reduces risk even as it reduces expected earnings.6 Unless individuals can credibly feign preferences other than payoff-maximization, the (3, 7, 0) profile is also the subgame-perfect outcome of the finitely repeated game. 6 Whereas expected gains are 39 + 70 – 70 = 39 if the (3, 7, 0) strategy were followed by all, an individual subject who deviated by putting a token into private protection, hence to (3, 6, 1), would earn 39 + 60 – l, where l is a random variable—loss from theft—that can take values of 0, 10,…, 280 depending on how many, if any, of up to four theft attempts against i, each involving 0 to 7 theft tokens, are thwarted by i’s 10% private protection level. Since E(l) = 63 assuming a 10% protection level, expected earnings under the strategy are 36 tokens. Deviation from any profile of theft and private protection that others are assumed to adopt in favor of more private protection for oneself likewise reduces expected earnings but raises their variance, whereas deviation from such a profile towards less private protection for oneself increases expected earnings and reduces their variance. As an aside, the possibility that others randomly assign all seven theft tokens against any one individual can be discounted if subjects are assumed to be even slightly risk-averse, since the fact that the success of each theft attempt by i against j is governed by an independent random draw makes diversification of the attempts across all other group members the most attractive strategy, all else being equal. 47 It is clear that in the NCP treatment, our subjects face a social dilemma. If all resist the temptation to engage in theft and put ten tokens each period into production, each earns 70 tokens per period, versus the 39 tokens that are the equilibrium prediction for selfish, rational, non-risk-loving agents. Not stealing can accordingly be thought of as a public good, and the (3, 7, 0) equilibrium represents a failure of public goods provision. With this in mind we introduced, in the remaining three treatments, a “second best” mechanism of collective action to establish collective protection. III.2.b VCP treatment – Voluntary Collective Protection In the remaining three treatments, the second stage of each period consists of simultaneous decisions on allocating effort tokens to production, theft, and private protection, just as in the allocation stage of NCP. However, that stage is now preceded by a prior stage with an opportunity for collective protection (see Panel B of Figure III. 1.b). In treatment VCP, contributions to collective protection are voluntary by putting effort tokens into a collective protection fund. Each unit any group member devotes to the fund up to a maximum of 12 tokens raises P (the probability of protecting one’s wealth against theft) of all members by 0.06, with a thirteenth or higher token having no additional effect. 7 We impose a ceiling on the level of collective protection of 72% (a 28% probability of a theft succeeding) because we deem it realistic that property cannot be made 100% secure by public policing alone. 7Note that since decisions are made simultaneously and without communication, over-allocation is possible. Group members learn the total contributions provided, but not the contribution of any individual member. 48 Private and collective protection combine to determine j’s total protection Pj = min[0.1pj + min(0.06Σck, 0.72), 1], where c indicates contributions to collective protection and k indexes any group member including i and j. Notice that tokens allocated to private protection raise the protection level of only the allocator’s accumulation by 10 percentage points, whereas tokens allocated to collective protection raise all group members’ protection levels by 6 percentage points, making free-riding on collective protection a dominant strategy. Denoting the number of tokens that individual i allocates to collective protection by ci,, we can denote i’s strategy by (mi, Ti, pi, ci), where mi + Ti + pi + ci = 10. Since we have already demonstrated that a risk-averse or risk-neutral subject wishing to maximize her earnings will allocate no tokens to private protection, it is clear that standard theory assuming self-interested agents also predicts that there will be no tokens allocated to collective protection, since a token so allocated yields even less in anticipated gains (i.e., avoided losses) to i than does a token allocated to private protection. So the equilibrium prediction is that all will adopt profiles of (3, 7, 0, 0). Of course, this constitutes an inefficient social dilemma outcome. Assuming that the social optimum of 100% production and zero theft is out of reach, by putting only three tokens each into collective protection in the first stage subjects can render it individually rational to assign the remaining seven tokens of each to production, leading to outputs of 64 wealth tokens per period instead of the 39 wealth token output that is otherwise predicted.8 While rational subjects concerned only with their own payoffs and 8 Clearly, it would be still more efficient were two subjects to allocate three tokens and three to allocate two tokens each to collective protection, leaving two more tokens for production; but we discount this possibility as 49 assuming the same about others will put nothing into collective protection, non-trivial contributions to public goods in laboratory experiments and in real world environments are widely documented, so positive contributions might be anticipated on behavioral grounds. III.2.c VOTE treatment – Voting on collective protection Our third treatment differs from VCP in that groups are given the opportunity to solve the free-riding problem surrounding collective protection by voting to make contributions mandatory—a scheme analogous to using taxes to fund police, prisons, etc. Following a first phase of four periods in which no collective protection is available, group members vote before each of the remaining five phases (of four periods each) on whether to make contributions to collective protection mandatory or keep them voluntary. If a majority prefers mandatory contributions, then in the first stage of each of the following four periods, group members indicate their preferred level of contribution knowing that the median proposal will bind all; otherwise, periods take the same form as in VCP. Panel B of Figure III. 1.a shows the timeline of this treatment, while Panels B and C of Figure III. 1.b illustrate the timelines of the stage game for each of the two possible scenarios. Under standard theoretical assumptions, the choice between the voluntary and mandatory schemes should be a simple one. As Section 2.b showed, equilibrium play under the voluntary scheme entails no contributions either to collective or to private protection, with each adhering to the (3, 7, 0, 0) allocation and earning an average of 39 largely infeasible in the absence of a coordination device. We discuss an alternative scenario in which all allocate two tokens to collective protection later. 50 wealth tokens each period. If the mandatory scheme is adopted, by contrast, subjects can vote to mandate contributions of either two or three tokens to collective protection and thus make it individually rational for each to put six to eight tokens into production and have expected earnings of approximately 64 wealth tokens. 9 A subject perceiving a positive probability of being pivotal should accordingly vote for the mandatory scheme, and without the means to coordinate voting, it is reasonable to expect all to vote this way.10 This yields 64 wealth tokens as expected earnings according to standard theory, or 91% of the potential earnings. This is much better than the expected 39 wealth tokens (or 56% of the maximum) in NCP and VCP. 9 If the group mandates a contribution of 3, there will be a joint “overprovision” by a total of 2 tokens, but with only 28% chance of a theft succeeding and hence an expected gain of only 2.8 wealth tokens from each token allocated to theft, it is unambiguous that all 7 remaining tokens should be devoted to production, yielding expected earnings of 64 wealth tokens per period. If the group chooses a mandatory allocation of 2, the 10 tokens in collective protection will yield a 60% protection level and thus an expected gain of 4 wealth tokens from a token allocated to theft, causing a risk-neutral subject to put no more than 7 tokens into production and to be indifferent between putting a seventh token into production versus putting only six into production and one into theft. Assuming risk-neutrality, subjects would choose randomly between 6 versus 7 tokens in production, so expected earnings would be 62 tokens (= 0.5x60 + 0.5x64), making the three token requirement the better choice. Assuming subjects were slightly risk-averse, however, all would choose to put a seventh token into production given a 60% protection rate, so earnings would be 64 wealth tokens, the same as when each contributes three tokens (mandatorily) to collective protection. With even stronger risk-aversion, subjects facing 60% collective protection might even prefer putting an eighth token into production, with certain return of 3 wealth tokens, to putting one effort token into theft, with expected return of 4 wealth tokens but a 40% chance of obtaining nothing. Thus, assumptions of strong risk-aversion could lead to a preference for mandating a 2 rather than 3 token allocation to collective protection, since the number of wealth tokens each produces could be 67 under this assumption. Although the earnings outcome associated with a 2 token requirement thus ranges from 62 to 67 wealth tokens, depending on degree of risk aversion, we treat 64 as the benchmark predicted earnings under the mandatory allocations scheme, since it is achieved exactly in two plausible scenarios and is close to the unweighted average outcome of the four scenarios considered. 10 Being unable to know for certain how others are voting, a subject cannot rule out that she will be pivotal, and this should eliminate her indifference. A trembling hand perfection argument can similarly be enlisted in favor of the prediction of uniform voting for the mandatory scheme. 51 III.2.d CHAT treatment We lastly implement a CHAT treatment that allows group members to communicate with one another via typed messages.11 Decision interactions continued to be in six four-period phases under the same protocol as in VCP, but each phase was preceded by several minutes for exchanging messages within each group. The duration of the chat periods was four minutes before phase one, then fell in half minute increments until the last two phases, before each of which subjects had two minutes for chat. Panel C in Figure III. 1.a shows details of the treatment’s timing. If groups reached cooperative agreements on desisting from theft and almost all subjects adhered to them, the within-group identity of any thief could be inferred by the others from the available “stats” information. In that case, group retaliation by theft might serve as a credible deterrent. However, given the finite nature of the game and the fact that no exogenous mechanism to enforce agreements was available, standard theory predicts that no self-interested rational subject will adhere to a promise not to steal, since each has an incentive to steal in the last period and thus, expecting that all will steal in that period, each can expect all rational selfish group members to steal in every period regardless of chat messages. Hence, communication does not move the theoretical equilibrium allocations from the (3, 7, 0, 0) profile predicted in VCP. Nonetheless, past observations of cooperation in dilemma experiments in which pre-play communication was available (Sally, 1995; Brosig, Ockenfels and Weimann, 2003; Zelmer, 2003) lead to 11Subjects were instructed to: (a) not reveal their real identity or ask the identity of others (to preserve anonymity); (b) not threaten punishments or promise rewards outside of the experiment domain; and (c) not use obscene language. They were truthfully informed that all messages would be screened by the experimenters and that violations would lead to forfeit of all earnings. This proved sufficient to assure that the instructions were observed. 52 behavioral expectations of at least some departure towards a more cooperative outcome in CHAT. III.2.e. S-CHAT treatment This is a simplified version of chat treatment, where subjects can send only a letter that indicates whether the subject is willing to engage in theft instead of having a full chat. Subjects have three options: (a) I will steal no matter what; (b) I will steal only when others steal; (c) I will NOT steal no matter what. This treatment was implemented only in Mongolia, in addition to the original chat treatment with students, in order to simplify the chat for business owner subjects. Since some business owners are not very familiar with modern online chats, we introduced this treatment. To compare student subjects to business owners, we implemented S-CHAT treatment with student subjects as well. However, the usual CHAT treatment was not implemented with business owners. III.2.f Subject pools Because responses to the dilemma of property rights seemed likely to vary with normative orientations, beliefs, and institutions, we decided to conduct our experiment using subjects in a number of different countries having different historical and contemporary characteristics. 12 The five countries in which the experiments were conducted—Austria, South Korea, Mexico, Mongolia and the U.S.—represent a broad 12Noteworthy experiments suggesting cross-national differences between subject pools include Roth et al. (1992), Henrich et al. (2001), Herrmann, Thöni and Gächter (2008), Bohnet et al. (2008) and Bohnet, Herrmann and Zeckhauser (2010). 53 range of characteristics. Austria and the U.S. are economically developed, politically democratic societies, with Austria having considerably greater ethnic homogeneity and a long-standing social democratic institutional caste compared to the more individualistic free market qualities of the U.S. South Korea provides a more recently industrialized and democratized Asian setting with a less extensive welfare state, Confucian paternalistic traditions and a heavy dose of Western, Christian and modern technological influences. Mexico is an upper middle income developing country with a population of mixed Amerindian and Spanish origin which has experienced intermittent economic growth, partly facilitated by proximity to the United States, with a reputation for political instability, corruption, and, like South Korea, relatively recent effective democratization. Mongolia, which shares a high level of ethnic homogeneity with Austria and South Korea, is the least economically developed country in the sample, is the only one to have gone through three generations under Communist rule before beginning a transition to free market capitalism in the 1990s, and is also the only one whose economy and society were based on semi-nomadic pastoralism rather than settled agriculture before modern times. Our sample accordingly represents three continents, five cultures (Inglehart and Welzel, 2005), a wide spectrum of economic development levels, a variety of levels of ethnic homogeneity, a range of experiences with democracy, and, as we will see, a range of perceived levels of quality of government, of social trust, and of experienced security of property. At each site, sessions of all four treatments were conducted in a university computer lab using college-age students as subjects, each participating in no more than one session and thus only one treatment. In total, we had 795 student participants, with 54 six to eight groups of five members each per treatment and country (see Table III. 2). All student participants were similar in age, education and socio-economic position in their respective countries. Specific sites were the University of Innsbruck (Austria), the Instituto Tecnológico Autónomo de México or ITAM (Mexico City), the Mongolian University of Science and Technology or MUST (Ulaanbaatar), Korea University (Seoul) and Brown University (Providence, Rhode Island, U.S.) 13 One can of course argue that a more diverse subject pool should contain subjects more typical of their societies than college students, and going further, that the latter might share much in common despite having different nationalities. We agree with the spirit of this concern, but nonetheless find that these subjects, used in our study for logistical reasons, display differences by nationality that correspond to documented national differences. We recruited business owners of local wholesalers and retailers in Ulaanbaatar, Mongolia. Wholesalers have contracts with local mobile companies to distribute prepaid phone cards to retailers in the city. Although most wholesalers specialize in prepaid card wholesaling, retailers tend to sell other products, typically groceries, cleaning supplies, and alcoholic beverages. Sales data are drawn from one of the mobile carriers, Mobicom, the largest mobile carrier in Mongolia. The sales data measure the revenue of Mobicom prepaid cards that each subject posted in May 2010. Mobicom prepaid cards are now in digital form and are no longer a scratch card. When a 13At these four universities, subjects were drawn from the general undergraduate programs. However, the case of MUST is slightly different. This institution was selected as our site in Mongolia because it offered one of the few facilities in Ulaanbaatar with an adequate computer lab, but Mongolian student subjects were recruited from a total of nine institutions, of which three, MUST, Mongolian National University, and Institute of Finance and Economics, account for the lion’s share (the others being Mongolian University of Pedagogy, Institute of Commerce and Business, University of Medical Sciences, University of Humanities, University of Agriculture, and Soyol Erdem University). We recruited from multiple universities in Ulaanbaatar because MUST lacks social science and humanities students, making its students less diverse than those in the other countries’ subject pools. 55 customer pays a certain amount to a retailer, the retailer deposits an equal amount from his or her business cell phone account to the customer’s cell phone account. This transaction is recorded by Mobicom’s server computers. I have obtained the total monthly sales for each retailer from Mobicom. I observe which wholesaler is connected to which retailer, and this information gives me the sales of each wholesaler as well. There were 2734 active retailers in May 2010 and 1203 wholesalers registered with Mobicom (for details of the original population data, see Batsaikhan 2012). I circulated the advertisement of the experiment through local Mobicom offices to recruit retailers and wholesalers. I divided applicants into six sessions and invited them to the laboratory. In each section, 22 to 27 people were invited and there were two subjects who could not attend the experiment due to over-enrollment and earned only the show-up payment of 5000MNT. III.2.g Procedures Experiments were conducted between January and June of 2011 on computers programmed in Multistage (software initially developed at U.C.L.A. and Caltech). At the beginning of each session, instructions were read aloud in the relevant language while subjects read along on paper.14Subjects then became accustomed to entering decisions (dictated by the experimenter) and viewing outcomes in the relevant screen formats in sets of controlled practice periods. In NCP, VCP, and CHAT, all instructions and 14 Instructions were translated from English to German, Korean, Mongolian and Spanish by native-speakers of each language belonging to our team and underwent “back-translation” to English by a different bilingual individual who had not read the English version to check for consistency. Instructions and practice scripts for all treatments in English are included in Appendix D. 56 practice took place before phase one. In VOTE, the initial instructions and practice before phase one resembled those of NCP (except that subjects were told that additional instructions would follow that phase) and this was followed by further instructions describing collective protection, how to vote on it and determine its level. Subjects were invited to ask questions of clarification before payoff-determining play commenced. III.3 Results III.3.a Comparing play by treatment To simplify exposition, we first pool the data from our five sites with students subjects in the main treatments (excluding the S-CHAT treatment) and focus on differences among treatments, then turn to comparisons across sites in section 3.c and comparison of student and business owner subjects in section 3.e. Figure III. 2.a–d display plots of average allocations to each of the four possible activities—a. production, b. collective protection, c. private protection, and d. theft—by period, and Figure III. 2.e shows the resulting average earnings per subject and period. This compares our theoretical benchmarks to the actual average choices and outcomes by treatment. Our first general observation is that in the NCP treatment, average token allocations to production (4.3) and theft (2.9) lie between the equilibrium for rational, selfish players (3, 7, 0) and the social optimum (10, 0, 0). There are also substantial allocations to private protection—averaging 2.9 tokens—which is at odds with the zero allocation predicted even for risk-averse agents. By allocating 1.3 more tokens per period to production than predicted for rational, selfish players, subjects earned an average of 57 46.6 tokens per period rather than the predicted 39, thus capturing about a quarter of the potential gain from cooperation15 but leaving the remaining three quarters “on the Table III..” While we later report evidence that some subjects may have been refraining from theft due to moral restraint and the desire to achieve a more cooperative outcome, the allocations to private protection are high enough to explain most of the desistance from theft. Therefore, explaining why subjects engaged in so much private protection is an important task, to which we turn in section 3.b. In the VCP treatment, the average voluntary contribution to collective protection begins at 1.5 effort tokens per subject in period one, but declines rapidly, yielding an overall average of 0.43 tokens per period. Taking into account the average allocations of 2.74 tokens to private protection, the average subject’s total protection level is about 40% in VCP (versus 29% in NCP). This level renders the expected return to theft for a hypothetical subject with perfect foresight 6 wealth tokens, one less than the certain return on a 5th token assigned to production. Presumably in part because of this higher protection, average allocations to production were 0.54 tokens higher than in NCP (4.83 vs. 4.29) and those to theft 0.84 tokens lower (2.01 vs. 2.85)—both differences being significant at the 1% level according to a Mann-Whitney test based on group averages as independent observations (see Table III. 4). Average earnings were thus 50.35 per period, 3.7 tokens higher than in the NCP treatment, a difference that is also significant at the 1% level. While modest, the introduction of a collective protection technology raises the percentage of potential cooperative surplus obtained by subjects by 12 percentage points, to 36.6% (cf. Table III. 3). 15The potential gains from cooperation are 31, which is the difference between 70 (if all tokens are invested into production and no theft occurs) and 39 (the earnings in equilibrium). 58 Recall that in theory, the VOTE treatment offers subjects their best opportunity to attain higher efficiency on the basis of individual rationality and self-interest. By voting to mandate the contribution of two or three tokens per subject to collective protection, sufficient protection can be assured so that allocating the remaining seven tokens to production becomes rational and thus about 80% of potential efficiency gains are attained. Figure III. 2.a and 2.e show that subjects did boost production and earnings in VOTE relative to the first two treatments, with Figure III. 2.b showing increased collective protection (Table III.4Error! Reference source not found. shows that these differences are statistically significant with p < 0.01 according to Mann-Whitney tests). In this respect, our VOTE treatment successfully illustrates the emergence of a tax-financed public policing institution. The impact is less than predicted, however, since the average efficiency gain in the five phases when the mandatory collective protection scheme was available is slightly under 50%, rather than the predicted 80%. This is largely explained by the facts that majorities voted to use the more efficient mandatory scheme in only 64% of the available opportunities and that the mandated collective protection level when the scheme was selected was not always ideal. Groups set that level at three tokens in 10.3% of votes and at two tokens in 59.5%, so an efficient scheme with mandatory contributions of either two or three tokens was in place in only about 45% (≈ (.103+.595)*.64) of Phase 2–6 periods. Mandatory contributions of zero tokens, one token, and four tokens were chosen, in 5%, 25% and 0.2% of votes, respectively. Even in those periods in which groups selected the mandatory contributions of two or three tokens, allocations to production averaged only 6.05 rather than the privately optimal seven effort tokens, so earnings per period averaged only 58.81 wealth tokens; this is significantly more than the 59 50.35 of the VCP treatment but still below the feasible 64 tokens of the second-best benchmark outcome. Also, we again see a surprising attraction to private protection. Subjects assigned an average of 2.35 (1.13) tokens to private protection when playing under the voluntary (mandatory) contribution scheme. While in the VOTE treatment only 1 group out of 38 achieved the predicted 80% efficiency gain or better, performance in the CHAT treatment far exceeds the theoretical expectation of full free riding. Of the 40 groups participating, 23 met or exceeded the cooperative benchmark’s 80.6% share of potential efficiency gain (i.e., earned an average of 64 or above), including 21 groups with zero allocation to theft in the first 20 periods, 20 groups with average earnings of 67 tokens or better (95.7% of potential maximum earnings) and 3 groups achieving the full social optimum (100% allocations to production). Figure III. 2.a and 2.e make clear that the differences in production allocation and efficiency achieved in the other three treatments are dwarfed by those of the CHAT treatment. Average allocation to production in CHAT is 8.12 tokens, well above the 5.32 tokens in the next most efficient treatment, VOTE. Not surprisingly, the difference between allocations under the CHAT treatment and all others is statistically significant at the 1% level (see Table III. 4). While a few groups discussed the possibility of using collective protection to deter theft during their chat periods, the average allocation to collective protection was in fact lowest in CHAT among the three treatments in which it is available, at 0.20 tokens per period. Allocations to theft and to private protection both averaged between 0.8 and 0.9 tokens per period. Analysis of chat contents supports the conclusion that the closer approach to efficiency in the treatment is mainly attributable to 60 the creation of and adherence to agreements to desist from theft, agreements sometimes explicitly backed by the threat that if any group member broke it and stole from others, the remaining members would punish that individual by stealing from him or her. The data show that that threat was exercised effectively on numerous occasions. III.3.b Why is there lower-than-predicted theft and higher-than-predicted private protection? While the ability to enter agreements supported by relatively high levels of trust in keeping the agreements helps to explain the much lower-than-predicted levels of theft observed in the CHAT treatment, what might explain lower-than-predicted theft in the NCP and VCP treatments, and during periods in the VOTE treatment in which groups failed to mandate high levels of collective protection? An aspect of the answer mentioned earlier is the higher-than-predicted level of investment in private protection, which was in many periods sufficient to make the observed (low) levels of theft rational. But these allocations to protection remain unexplained by either payoff-maximization or risk- aversion. We briefly explore four alternative explanations: loss aversion, moral reservation, asymmetric protective motives, and errors. In Section 2, we saw that a self-interested, rational and non-risk-loving subject would expect group members to allocate seven tokens to theft and three to production each period. Relative to that choice, withdrawing a token from theft reduces the decision- maker’s earnings by ten tokens and devoting that token instead to private protection increases expected earnings by maintaining possession of seven tokens that would 61 otherwise be forfeited. A subject might prefer this alternative if she values seven tokens that are in her existing accumulation more than she values ten tokens she could steal from another’s. Possible reasons for such a preference are loss aversion and the devaluation of stolen tokens by moral taint. Loss aversion can be modeled formally by assuming that subject i seeks to maximize the sum of utility from final wealth and gain-loss utility taking the form used by Koszegi and Rabin (2006): x if x  0  ( x)    x if x  0 where η is the weight placed on gain-loss utility, x are the gains ( x  0) or losses ( x  0) , and λ>1 is the coefficient of loss aversion. Then, treating wealth tokens accumulated through theft as gains and wealth tokens lost to theft as losses, it can be shown that the utility for subject i is higher under profile (mi, Ti -1, pi +1) than under profile (mi, Ti, pi) as long as λ > 10(1 - 0.1pi)/Ti. For example, if all members of a group are devoting four tokens to theft and two to private protection, it follows from the formula that subject i is better off allocating a third token to protection as long as λ >2. A λ-value of two may be plausible, since Tversky and Kahneman (1992) provide a median estimate of 2.25 for λ. Reluctance to steal on moral grounds is another factor that might plausibly account for lower-than-predicted allocations to theft. It is noteworthy that in the first period of the NCP treatment, where subjects determined their allocations to theft and private protection in the complete absence of signals of others’ plans, roughly one quarter of subjects did not allocate any token to theft. Since allocations to private protection large 62 enough to make even a one-token allocation to theft unprofitable were only a remote possibility, this much forbearance from theft probably indicates that substantial numbers of subjects were reluctant to steal before being given the “moral green light” to do so that would come from others’ stealing. Although substantial allocations to private protection may explain lower-than- expected theft levels later on in NCP and other experiment sessions, some causation could run in the opposite direction. In particular, if an individual is reluctant to steal from others but believes that this is less true of others, it would be in her interest to invest some tokens in private protection. In Appendix B we conduct an exercise to check whether private protection could have been motivated in this way. We make the extreme assumption that each time a subject spent less than seven tokens on theft, she does so due to moral scruples and then decide her allocation between production and private protection so as to maximize her earnings subject to those scruples. We calculate the associated optimal allocation to private protection and include it in a regression to predict the amount actually allocated to it. This conditional payoff-maximizing allocation to private protection turns out to be a statistically significant predictor of the actual allocation, but explains no more than a fifth of the protection allocations observed. Our third explanatory factor, asymmetric protective motives, refers to anticipation of retaliation after own engagement in theft. Whereas so far we assumed theft tokens to be directed randomly (and thus distributed equally among other group members in expectation), one could argue that theft attempts against, say subject i, could have been prompted by i's successful stealing in the preceding period. Assuming that i anticipates this, she may find that the returns to investing a token in private protection are greater 63 than the expected returns to theft (especially when some protection by other group members is in place), thereby making higher-than-predicted private protection more likely. In Appendix B we present evidence of a significant positive correlation between stealing in the previous period and allocations to private protection in the current one (see columns 3, 6, 7, and 10 in Table III. B1). Of course, some allocations to private protection despite higher returns from theft may simply be explained by errors and bounded rationality. The interactions in which subjects engaged were unfamiliar, and few are likely to be good at calculating optimal conditions on the fly. Both an impulse to allocate a little to each available activity, given lack of clarity about the best course of action, and a desire to protect existing accumulations, not always subjected to a carefully calculated comparison with other strategies, probably explain some allocations to private protection. III.3.c Comparing play by country The four panels of Figure III. 3 show for each treatment the decisions taken in the five different countries separately, and also pooled over all countries. Before pointing out differences, it is useful to note again the considerable number of qualitative similarities across countries. At all five sites, production and earnings are lowest in NCP followed by VCP and VOTE, and highest in CHAT. Allocation to collective protection is highest in VOTE, intermediate in VCP, and lowest in CHAT in every country except South Korea, where contributions in VCP and CHAT are approximately the same. The trends of 64 declining allocations to collective protection and increasing allocations to theft over time are also observed for each subject pool. Generally speaking, moreover, subjects from all five countries behaved similarly under our simplest environments, NCP and VCP, as indicated by the results from Kruskal-Wallis tests based on group averages as independent observations and country as the grouping variable. Panels A and B in Table III. 3 show that in these two treatments, allocations to all activities except production are statistically indistinguishable across countries.16 Summing up, these results suggest that under the more primitive conditions, individuals from widely distinct countries with different cultures and histories tend to exhibit similar behaviors. There are, on the other hand, statistically significant and economically interesting differences among the subject pools related to the impact of more sophisticated institutions than the voluntarily-funded collective protection. In the VOTE treatment, we find considerable variation in institutional preferences among subject pools, with the proportion of individuals voting in favor of the mandatory scheme ranging from 29.5% in Mongolia to 69.7% in Austria, with the U.S. (58%), Mexico (61.1%) and South Korea (63%) in between. The frequency of majority selection of the scheme follows a similar but not identical order, ranging from 22.5% in Mongolia, to 62.5% in the U.S., 75% in South Korea, 80% in Austria and 82.9% in Mexico. Panel C(i) in shows that such 16Mann-Whitney tests for every pair of countries reveal that the difference in allocations to production in NCP are statistically significant for Austria and South Korea (p = 0.018), Austria and Mongolia (p = 0.015), South Korea and Mexico (p = 0.082), South Korea and the U.S. (p = 0.036), Mexico and Mongolia (p = 0.063, and Mongolia and the U.S. (p = 0.010). Parallel tests show the difference in allocations to production in VCP is driven by the outlier status of Austria in the treatment. Significant differences are observed between Austria and South Korea (p = 0.049), Austria and Mongolia (p = 0.007), and Austria and the U.S. (p = 0.021). For any other pair of countries, the difference is statistically insignificant. 65 differences in the preferences for and choice between the two schemes are statistically significant according to Kruskal-Wallis tests. Conditional on the choice of the mandatory scheme, however, Panel C(ii) shows that behaviors are statistically indistinguishable across countries. When groups choose the independent contributions scheme, on the other hand, we observe statistically significant differences in the support for the mandatory regime and allocations to protective activities; the former stemming from the reluctance of Mongolian subjects to vote for the mandatory scheme, whereas the latter being determined by Mexican and Mongolian subjects’ heftier allocations to collective protection but smaller to private protection relative to the other three subject pools.17 In the CHAT treatment, we find statistically significant differences for all allocation decisions (see Panel D in Table III. 5). These differences mainly stem from the failure of Mongolian subjects to adhere to non-theft agreements. Indeed, results from pair-wise Mann-Whitney tests show that allocations to theft and production differ significantly between Mongolia and each other country with p < 0.01, whereas they are statistically indistinguishable between any other pair of countries. 17 Results from par-wise Mann-Whitney tests indicate that under the voluntary contributions scheme, Mongolian subjects were significantly (at the 10% level) less prone to vote for the mandatory institution than subjects from any of the other countries; for any other pair of countries the difference is statistically insignificant. Allocations to collective protection are statistically different with p < 0.05 between South Korea and Mongolia, and Mexico and the U.S.; and with p = 0.065 between South Korea and Mexico. Allocations to private protection are statistically different with p < 0.05 between South Korea and Mexico, South Korea and Mongolia, Mexico and the U.S., and Mongolia and the U.S. 66 III.3.d How experimental choices relate to socio-political factors in the five countries Are subjects’ behaviors in the experiment in any way indicative of the underlying socio- political environment in their country? Evidently so. Subjects from countries in which property is less secure tend to allocate more resources to protection. We considered two measures of each country’s protection of property rights. The first focuses on the incidence of property crimes, and is constructed by using data obtained from the United Nations’ International Crime Victim Survey (ICVS).18 We employed data from the years 2000 (South Korea and Mongolia), 2004 (Mexico and the U.S.) and 2005 (Austria) and computed the share of respondents who experienced burglary or robbery as the fraction of respondents from each country who answered “yes” to the questions (i) “Over the past 5 years, did anyone actually get into your house or flat without permission and steal or try to steal something?,” or (ii) “Over the past 5 years, has anyone taken something from you, by using force, or threatening you? Or did anyone try to do so?” Thus, a higher score reflects a higher incidence of property crimes. According to this measure, Austria and the U.S. are the countries with the lowest incidence of property crimes, followed by South Korea and Mexico, while Mongolia exhibits the highest incidence (see in Appendix C for detailed numbers). Figure III. 4.a displays the positive association between our measure of incidence of property crimes and average allocations to private protection across all four treatments and periods. Likewise, Figure III. 4.b exhibits a positive correlation between incidence of property crimes and average allocations to collective protection, although the fit is noticeably weaker. Delving further into the evidence, we find that the free-rider problem, 18 http://www.unodc.org/unodc/en/data-and-analysis/Crime-Victims-Survey.html 67 which the experimental literature on public goods widely identifies as a more dominant force in later periods, could have eroded the positive association between incidence of property crimes and average contributions to collective protection as the experiment progressed. In fact, Figure III. 4.c shows a much clearer positive association between incidence of property crimes and contributions to collective protection in the first period of the experiment. Our second measure of the degree of protection of property rights in each country uses data from the ICVS to construct a composite index aimed at capturing how safe people in each country feel. We built this index via factor analysis of the responses to the survey questions (i) “How safe do you feel walking alone in your area after dark? (1=very safe 2=fairly safe, 3=a bit unsafe, 4=very unsafe),” and (ii) “What would you say are the chances that over the next twelve months someone will try to break into your home? (1=very likely, 2=likely; 3=not very likely)”.19 A higher value of the index reflects a perception that people and their possessions are at higher risk. Our results indicate that, among our sites, Mongolians feel the least safe, followed by Mexicans and South Koreans; Austrians and Americans exhibit the highest perceptions of safety (see Appendix C for detailed Figure III.). Figure III. 5.a and 5.b show that subjects from countries where people feel less safe tended, on average across treatments and periods, to allocate more resources to both private and collective protection. Once again, moreover, we observe an even clearer association between perceptions of safety and average allocations to collective protection 19See Johnson and Wichern (2002) for a detailed description of factor analysis methods. We implemented this technique using the factor/predict commands in Stata. 68 if we restrict our analysis to the first period, before the free-rider problem becomes a more dominant driver of contribution choices (see Figure III. 5.c). The VOTE treatment invites cross-country comparison because it is the only one in which our subjects decide on the use of an institution and the level of a tax by voting. We wondered whether differences in the quality of the political institutions among the countries represented could help to explain some of the cross-country variation in the support for provision of collective protection by mandating tax-like contributions. To explore this issue, we constructed a composite “Governance Index” applying factor analysis methods to three variables included in the World Bank’s Worldwide Governance Indicators (WGI) dataset: (a) government effectiveness, (b) rule of law, and (c) control of corruption. A higher value of our Governance Index reflects political institutions of higher quality. Of the countries included in this study, Austria exhibits the highest Governance Index, followed by the U.S., South Korea and Mexico, with Mongolia having the lowest score (see in Appendix C, where we also provide definitions for the components of the index). Figure III. 6 shows a positive association between our Governance Index and the share of individual votes for the mandatory scheme. Although the positive correlation is mainly driven by the two countries on the extremes of the governance spectrum, Mongolia and Austria, the overall pattern suggests that subjects from countries with political institutions of higher quality are more prone to support the government-like institution meant to foster efficiency. 69 Finally, we consider the CHAT treatment. The effectiveness of communication in groups as a means to increase efficiency in interaction depends crucially on mutual trust in each others’ statements and promises. Therefore, we relate allocations in production to the level of trust in the five countries. We employ a “Trust Index” that captures the difference between the share of national respondents to the most recent World Values Survey or similar regional survey who chose “Most people can be trusted” and the share of respondents who chose “You can’t be too careful” in response to the question “Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?”.20 It is readily seen in Figure III. 7 that subjects from countries with larger trust index values were more prone to engage in productive activities rather than protection and theft, thereby achieving larger efficiency gains than subjects from countries with lower trust scores. III.3.e Experiments with non-student subjects in Mongolia A different type of information about external validity comes from conducting the same experiment with a different subject pool. We had the opportunity to carry out sessions of the VCP, VOTE and S-CHAT treatments with 120 subjects. In VCP and VOTE treatment, we additionally divided them into two different age groups: 32 years old or younger and 38 years old or older. This is done to separate the effect of communism. Mongolia was a socialist country until 1990, when the country accepted democracy 20The Trust Index captures the difference between the shares of responses rather than just the fraction of respondents who chose “Most people can be trusted” in order to adjust for the "No Answer" option that is offered in some of the regional surveys, or for slight differences in wording (e.g., by framing the question as a statement respondents would agree or disagree with). 70 following most East European countries. Subjects who are 32 years old or younger at the time of experiment were 12 years old or younger when the system changed. This means that these subjects were in an elementary school during the change, and their mind-set was not greatly affected by the old system. On the other hand, subjects who are 38 years old or older were at least 18 years old at the time of the institutional change. Their beliefs must have formed during the old system. First, we compared the subjects’ behavior during 24 rounds of VCP and VOTE treatments by three different age groups: 32 or under, 38 or over, and students (Figure III. 8). There is no significant difference between the two older groups. However, there is a distinct difference in pattern between Mongolian student and entrepreneur subjects. Over 24 rounds, the theft allocation by students increased dramatically whereas it decreased among entrepreneur subjects. The tokens allocated to private protection dropped in earlier rounds among entrepreneur subjects while student subjects experienced a gradual decrease. Overall, entrepreneurs’ earnings increased steadily and achieved higher total than students; whose earnings stayed relatively stable over 24 rounds. Student subjects earned much less because a large number of tokens were devoted to theft. In particular, the Mongolian entrepreneurs achieved 3.5% (8.7%) higher earnings than the Mongolian students in the VCP (VOTE) treatments, even exceeding the average earnings of student subjects in the five countries as a whole by 0.4% (1.2%), although still earning less than the best-performing student subject pools. The finding that social cooperation is, if anything, somewhat greater in subject populations of older adults is a common one (see Sutter and Kocher, 2007). 71 The difference between students and entrepreneurs is also very distinct in terms of the choice of preferred institutions in VOTE treatment. Student subjects in Mongolia preferred the independent contributions scheme most of the time and they differed from students in the four other countries. However, entrepreneur subjects in Mongolia preferred the voting scheme more, and this is consistent with the students in four countries except Mongolia. However, there is no clear difference between the young and old entrepreneurs. Second, we compared different treatments among entrepreneurs. While treatments improve the outcomes among student subjects in five countries, there is no difference among entrepreneur subjects. This can be confirmed in Figure III. 9. VOTE, VCP and S- CHAT have very similar patterns over 24 rounds and the level of theft and profit in all rounds are the same as well. The only difference is during the first four rounds in VOTE, where collective protection is offered. These four rounds operate essentially under the same conditions as in NCP. To further analyze why treatments do not have any effect on outcome variables, we looked at the first round. Using the sales data of each entrepreneur subject, we looked at the correlation between sales in real life and the number of tokens devoted to production and theft in the first round. Figure III. 10 shows that there is a strong correlation between the sales in real life and how many tokens the entrepreneur subject allocated to production. However, the subjects with higher sales allocated less of their tokens to theft. This means that pro-social entrepreneurs (who allocated less to theft) tend to be more successful in real business. This result is consistent with Batsaikhan (2012): entrepreneurs with a high trust measure tend to be more successful in business. 72 Next, we analyzed how behaviors in the experiment evolved over 24 rounds among entrepreneur subjects. Figure III. 11 shows that entrepreneurs increased the number of tokens allocated to production over 24 rounds and decreased the tokens for private and collective protections. Figure III. 12 further shows the distribution of these allocations in each round. Theft allocations are highly concentrated around 2 and 3 tokens in the first round. However, this changes over 24 rounds, and most subjects chose not to allocate any tokens to theft in the last round. This is especially interesting because this is the last round and subjects in the other subject pools tend to allocate more to theft. Theoretically, it is also rational to allocate more tokens to theft in the last period because there is no more interaction among subjects anymore and theft is more rewarding. Student subjects actually allocated more to theft toward the end of the experiment (Figure III. 2). Collective protection also shows an interesting pattern. In the first round, entrepreneurs allocated an extremely large amount of tokens to collective protection. However, this amount decreased quickly in two ways: the total number of collective protection tokens and the number of subjects who allocated a large amount of collective protection tokens. Every subject spent more than 6 tokens on collective protection in the first period but only a few allocated more than 6 in the last 5 rounds. Private protection has a similar trend with theft; it gradually decreased over 24 rounds. As it can be predicted, the number of production tokens and the number of subjects who allocated a large amount of tokens to production both increased as the experiment progresses; with the majority of the subjects allocating more than 5 tokens to production. 73 Overall, the number of theft tokens remained low compared to student subjects. However, production tokens increased while protection tokens decreased. Allocating tokens to protection is an inefficient way of using tokens compared to the social optimum, where no one allocates tokens to theft. This seems to be the case for entrepreneurs. Even though availability of a voting institution and a simple chat did not have an impact on the outcome of the experiment, successful entrepreneurs seem to change the dynamics of the experiment and to lead to a socially optimal equilibrium. When there is no institutional support to force contracts among economic agents, social capital helps to improve market outcomes. In this case, entrepreneurs might be using their social capital outside of the laboratory to increase the efficiency inside of it. To further access why entrepreneur subjects engaged in theft less than student subjects, we compared the answers in S-CHAT between student and entrepreneur subjects. Subjects in this treatment have an option to send a message indicating whether or not, or in what conditions, they would steal from others. There is no clear difference in the first two options between students and entrepreneurs (Table III. 7). “Never steal” means that the subject sent a message: “I will NOT steal no matter what. “ “Steal back” means that the subject sent a message: “I will steal only when others steal. “ “Always steal” means that the subject sent a message: “I will steal no matter what,” and this is where there is a slight difference in the last chat round: thirteen students chose this option in the last chat round compared to 3 entrepreneurs in the same round. Since theft was very low among the entrepreneur subjects, entrepreneurs could have punished the subjects with high theft allocation. To confirm this retaliatory behavior, we have regressed the loss from theft in period i on the theft allocation in period i-1. The 74 regression coefficient in each round is shown in Figure III. 12. There seems to be a clear retaliation behavior in the first four rounds. The coefficient becomes insignificant after that. Although this retaliatory behavior might have changed the dynamics of the experiment for entrepreneur subjects, it is inconclusive and requires further analysis. III.4 Conclusion We used laboratory decision-making experiments to study how groups of individuals may attempt to establish secure rights to property that permit a socially efficient allocation of resources to production. In addition to a purely anarchic setting (NCP) in which voluntary abstinence from theft and a private protection technology are the only ways to make property secure, we studied three treatments that incorporate a technology of collective property protection simulating real world counterparts (police, courts, prisons). This collective protection technology adds a second social dilemma element reinforcing the idea that property rights are a public good. We conducted these four treatments with undergraduate subjects in five economically, institutionally, and culturally distinct countries: Austria, Mexico, Mongolia, South Korea and the U.S. Our results in the treatments without voting or communication, i.e., in NCP and VCP, echo those of more standard voluntary cooperation experiments. Attempts to cooperate are rarely entirely absent, especially in the initial periods of play, as indicated in our data by the fact that 30 – 40% of subjects completely refrained from theft in first period play in the NCP treatment in the Austrian and U.S. subject pools. But cooperation tended to unravel with repetition much as in the canonical voluntary contribution 75 mechanism (Ledyard, 1995), so overall efficiency was closer to the non-cooperative equilibrium prediction than to the social optimum. About a quarter of potential gains from cooperation were achieved in NCP, and slightly over a third in VCP, our basic treatment with voluntary collective protection option. In our VOTE treatment, a majority of subjects voted rationally to fund collective protection by a mandatory levy, illustrating how governments help to address the dilemma of property in modern societies. With a substantial minority of votes favoring the non-mandatory institution and with frequent choice of lower-than-efficient tax levels, however, the institutional solution fell short of its theoretical potential, performing especially badly in subject pools exhibiting more distrust of government, such as in Mongolia. Communication has often proven to be a powerful tool for building cooperation in the laboratory, and our specific design and subject pools provide an unusually dramatic example of this. In contrast to the potential efficiency gains of 25%, 37%, and 48% in NCP, VCP, and VOTE, in treatment CHAT 77% of potential efficiency gains were achieved, with fully half of all groups exceeding an impressive 95% of potential gains and three out of twenty groups reaching 100% efficiency. Ours appears to be the most effective laboratory demonstration to date of the power of verbal agreements to translate into respect for property. Although subjects in each of our five countries were university students, we found considerable variation across countries correlating with differences in country characteristics that are suggested by large-scale surveys. We consider this the most 76 important contribution of our paper. Our findings support the view that social norms are important to the solution of collective action dilemmas and that the strength of the norms in question varies in a manner which also affects the effectiveness of institutions. One plausible interpretation of our findings is that many individuals are willing to refrain from theft conditional on others not stealing, thereby making expectations of the proportion of others who would steal critical to outcomes (e.g., only 10% of Mongolian subjects refrained from first period theft in the same treatment that saw three to four times more Austrian and U.S. subjects do so). Assuming that expectations of the frequency of theft within subject pools are correlated with the incidence of theft in their societies helps to explain observed cross-country variation in allocations to protection. These differences may reinforce the disparities in social trust which help to explain the dramatic heterogeneity in the effects of pre-play communication in our five countries (the average Austrian group went 20 periods without a single theft whereas the average Mongolian group saw theft in the very first period of the CHAT treatment). Moreover, differences in quality of government go some way towards explaining the variation in subjects' inclination to employ a government-like mechanism to fund collective protection from theft: almost 70% of Austrian subjects but less than 30% of Mongolian ones voted to make contributions to collective protection mandatory in the VOTE treatment. A general message of the cross-country differences in our study is that high trust, good governance, and low incidence of crime in the world outside the lab are associated with better resolution of the dilemma of property inside the lab. While more ultimate causes of differences in societal traits are beyond our scope, it is interesting to recall that Mongolia is not just the only country in our sample that spent decades under Communism 77 but also the only one whose society was for centuries based on migratory herding without towns, cities, or strong centralized institutions. In his recent book on the history of violence, the psychologist Steven Pinker cites studies suggesting a connection between herding and violence. “Herders all over the world cultivate a hair trigger for violent retaliation.” Pinker relies especially on the account of Cohen and Nisbett (1997), who use the idea to explain higher rates of violence in the U.S. south’s mountainous frontier due to its settlement by “Scots-Irish, many of them sheepherders, who hailed from the mountainous periphery of the British Isles beyond the reach of the central government” (Pinker, 2011). Many of our results invite interpretations applicable to a broad class of collective action dilemmas. Nevertheless, we would like to conclude by noting that our experiment delivers several findings particularly relevant to the problem of property rights. The choices of our experimental subjects support the argument that normative constraints may play a part in making property secure, but that they require supportive initial beliefs and channels of reinforcement. The operation of institutions to support collective action is likewise shown to be possible, but not automatic. The underpinnings of effective norms and good institutional choices are to a significant degree historically and culturally contingent. Cross-country evidence from outside of the lab may also be called on in support of the idea that secure property rights are requirements of more productive economies. Without attempting to identify underlying causal chains, it is worth noting that the per capita incomes of the five countries from which our subjects were drawn are negatively correlated with the reported incidence of crime, positively correlated with 78 social trust and quality of government institutions, and positively correlated with the efficiencies achieved in the lab by our subjects. 79 Bibliography Aghion, Philip and Peter Howitt (1992). “A Model of Growth through Creative Destruction”, Econometrica, Econometric Society, vol. 60(2): 323-51 Ardagna, Silvia and Annamaria Lusardi, (2008), “Explaining International Differences in Entrepreneurship: The Role of Individual Characteristics and Regulatory Constraints, in: International Differences in Entrepreneurship”, edited by: Josh Lerner and Antoinette Schoar, NBER Andersen, Steffen, Ertac, Seda, Gneezy, Uri, Hoffman, Moshe, and List, John, A., (2011). "Stakes Matter in Ultimatum Games," American Economic Review, 101:3427-39 Barr, Abigail Margaret, (2003). “Trust and Expected Trustworthiness: Experimental Evidence from Zimbabwean Villages”, Economic Journal, Vol. 113, pp. 614-630 Batsaikhan, Mongoljin, (2012). “Growth for Micro-enterprises: Evidence from Phone Card Wholesalers in Mongolia”, Working Paper Baum-Snow, Nathaniel, (2007) “Did highways cause suburbanization?”, Quarterly Journal of Economics 122(2):775–805 Beck, Thorsten, Asli Demirguc-Kunt, and Ross Levine, (2004). “Small and medium enterprises, growth, and poverty: cross-country evidence”, Policy Research Working Paper Series 3178, The World Bank 80 Berg, J., Dickhaut, J., & McCabe, K. (1995). “Trust, reciprocity, and social-history”, Games and Economic Behavior, 10: 122–142. Berry, Steven and Reiss, Peter, (2007). "Empirical Models of Entry and Market Structure," Handbook of Industrial Organization, Elsevier Bertrand, Marianne, David Thesmar and Antoinette Schoar, (2005). “Banking Deregulation and Industry Structure: Evidence from the 1985 French Banking Act,” The Journal of Finance. Bloom, Nick, Sadun, Rafaella, and Van Reene, John, (2012). “Organization of Firms Across Countries”, the Quarterly Journal of Economics, Forthcoming. Bohnet, Iris, Fiona Greig, Benedikt Herrmann and Richard Zeckhauser, (2008), “Betrayal Aversion: Evidence from Brazil, China, Oman, Switzerland, Turkey, and the United States,” American Economic Review 98 (1): 294 – 310. Bohnet, Iris, Benedikt Herrmann and Richard Zeckhauser, (2010), “Trust and the Reference Points for Trustworthiness in Gulf and Western Countries,” Quarterly Journal of Economics 125 (2): 811 – 828. Brosig, Jeannette, Axel Ockenfels and Joachim Weimann, (2003), “The Effect of Communication Media on Cooperation,” German Economic Review 4: 217-242. Bruhn,Miriam, Dean Karlan and Antoinette Schoar, (2010). “What Capital is Missing in Developing Countries” , American Economic Review Papers & Proceedings. Campos-Ortiz, Francisco, Putterman, Louis, Ahn, T.K., Balafoutas, Loukas, Batsaikhan, Mongoljin, and Sutter, Matthias, (2012). “Security of Property as a Public Good: 81 Experimental Behavior and Social Norms in Five Countries”, Dissertation Chapter, Brown University. Carpenter, Jeffrey P., Daniere, Amrita G., and Takahashi, Lois M., (2004). "Cooperation, trust, and social capital in Southeast Asian urban slums," Journal of Economic Behavior & Organization, 55: 533-551 Carter, John R. and Charles H. Anderton, (2001), “An Experimental Test of a Predator- Prey Model of Appropriation,” Journal of Economic Behavior and Organization, 45: 83-97. Cohen, Dov and Richard E. Nisbett, (1997), “Field experiments examining the culture of honor: the role of institutions in perpetuating norms about violence,” Personality and Social Psychology Bulletin, 11(23): 1188-1199. De Mel, Suresh, McKenzie, David, and Woodruff, Christopher, (2008). "Returns to Capital in Microenterprises: Evidence from a Field Experiment," The Quarterly Journal of Economics, 123: 1329-1372 Donaldson, Dave (2010). “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure”, NBER Working Papers, 16487. Duffy, John and Minseong Kim, (2005), “Anarchy in the Laboratory (and the Role of the State),” Journal of Economic Behavior and Organization, 56: 297-329. Durante, Ruben, and Putterman, Louis, (2009). "Preferences For Redistribution and Perception of Fairness: An Experimental Study," Working Paper. 82 Duranton, Gilles. and Turner, Matthew A, (2012a). “The fundamental law of road congestion: Evidence from US cities” American Economic Review 101 (6), 2616-2652. Duranton, Gilles. and Turner, Matthew A, (2012b). “Urban growth and transportation”, Review of Economic Studies, forthcoming. Durham, Yvonne, Jack Hirshleifer and Vernon L. Smith, (1998), “Do the Rich Get Richer and the Poor Poorer? Experimental Tests of a Model of Power,” American Economic Review, 88: 970-83. Eckel, Catherine C., and Wilson, Rick K., (2004). "Is trust a risky decision?", Journal of Economic Behavior & Organization, Elsevier, 55: 447-465 Eckel,Catherine C., and Wilson, Rick K., (2009). “Trust and Social Exchange”, prepared for the Handbook of Experimental Political Science. Fernald, John G, (1999). “Roads to prosperity? Assessing the link between public capital and productivity”, American Economic Review 89(3):619–638 Grossman, Herschel, (1991), “A General Equilibrium Model of Insurrections,” American Economic Review, 81: 912-21. Grossman, Herschel, (1994), “Production, Appropriation, and Land Reform,” American Economic Review, 84: 705-12. Grossman, Herschel, (2002), “Make Us a King: Anarchy, Predation, and the State,” European Journal of Political Economy, 18: 31-46. 83 Grossman, Herschel and Minseong Kim (1995), “Swords of Plowshares? A Theory of the Security of Claims to Property,” Journal of Political Economy, 103: 1275-88. Henrich, Joseph, Robert Boyd, Samuel Bowles, Colin Camerer, Herbert Gintis, Richard McElreath and Ernst Fehr, (2001), “In Search of Homo Economicus: Experiments in 15 Small-Scale Societies,” American Economic Review 91: 73-79. Herrmann, Benedikt, Christian Thöni and Simon Gächter, (2008), “Antisocial punishment across societies,” Science 319: 1362-7. Herrera, Ana Maria and Eduardo Lora (2005). “Why So Small? Explaining the Size of Firms in Latin America”, Unpublished working paper Hirshleifer, Jack, (1991), “The Paradox of Power,” Economics and Politics, 3: 177-200. Hirshleifer, Jack, (1995), “Anarchy and its Breakdown,” Journal of Political Economy, 103: 26-52. Holt, C. A., and Laury, S.K., (2002). “Risk Aversion and Incentive Effects”, American Economic Review, 92: 1644–1655. Inglehart, Ronald and Christian Welzel, 2005, Modernization, Cultural Change and Democracy. New York: Cambridge University Press. Jensen, Robert, (2007). “The Digital Provide: Information (technology), market performance and welfare in the South Indian fisheries sector”, Quarterly Journal of Economics Johnson, Noel D. and Mislin, Alexandra, (2011), “Trust Games: A Meta-Analysis.” Journal of Economic Psychology, Forthcoming. 84 Johnson, Richard and Dean Wichern, (2002), Applied Multivariate Statistical Analysis, 5th Edition, Upper Saddle River, NJ: Prentice Hall. Karlan, Dean S., (2005). "Using Experimental Economics to Measure Social Capital and Predict Financial Decisions," American Economic Review, 95: 1688-1699 Karlan, Dean and Jonathan Morduch, (2010) “Access to Finance”, Handbook of Development Economics, Chapter 2, Volume 5 Kimbrough, Erik, Vernon Smith and Bart Wilson, (2010), “Exchange, Theft, and the Social Formation of Property,” Journal of Economic Behavior and Organization 74: 206-229. Knack, Stephen and P. Keefer, (1997), “Does Social Capital Have an Economic Payoff? A Cross-Country Investigation,” Quarterly Journal of Economics, 112:1251-1288. Koszegi, Botond and Matthew Rabin, (2006), “A Model of Reference-Dependent Preferences,” Quarterly Journal of Economics 121 (4): 1133-1166. Ledyard, John O. (1995), “Public Goods: A Survey of Experimental Research,” in John Kagel and Alvin Roth, eds., The Handbook of Experimental Economics. Princeton: Princeton University Press. Leora F. Klapper, and Inessa Love, (2010). "The impact of the business environment on young firm financing," Policy Research Working Paper Series 5322, The World Bank. Lewis, W. A. (1954), “Economic Development with Unlimited Supplies of Labour”, The Manchster School, volume 22, no. 2, pp. 139-191. 85 McFadden, Daniel, (1989), “A Method of Simulated Moments for Estimation of Discrete Response Models Without Numerical Integration”, Econometrica, Vol. 57, No. 5pp. 995-1026 Mondragon-Velez and Ximena Pena-Parga, (2008). “Business Ownership and Self- Employment in Developing Economies: The Colombian Case”, in International Differences in Entrepreneurship, edited by: Josh Lerner and Antoinette Schoar, NBER. Mullainathan, Sendhil and Philipp Schnabl (2008), “Does less market entry regulation generate more entrepreneurs? Evidence from a regulatory reform in Peru”, in: International Differences in Entrepreneurship, edited by: Josh Lerner and Antoinette Schoar, NBER Palacios-Huerta, Ignacio, (2003). "Professionals Play Minimax," Review of Economic Studies, 70: 395-415. Palacios-Huerta, Ignacio & Volij, Oscar, (2008). "Experientia Docet: Professionals Play Minimax in Laboratory Experiments," Econometrica, 76: 71-115 Pinker, Steven, (2011), The Better Angels of Our Nature: Why Violence Has Declined. New York: Viking. Powell, Benjamin and Bart Wilson, (2008), “An Experimental Investigation of the Hobessian Jungles,” Journal of Economic Behavior and Organization, 66: 669- 686. 86 Roth, Al, Vesna Prasnikar, Masahiro Okuno-Fujiwara and Shmuel Zamir, (1992), “Bargaining and Market Behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo: An Experimental Study,” American Economic Review 81: 1068-1095. Sally, David, (1995), “Conversation and Cooperation in Social Dilemmas: A Meta- analysis of Experiments from 1958 to 1992,” Rationality and Society 7(1): 58-92. Schoar, Antoinette, (2009). “The Divide between Subsistence and Transformational Entrepreneurship” NBER Working Paper. Schumpeter, J. A. (1942). “Capitalism, Socialism, and Democracy”, New York: Harper Brothers. Skaperdas, Stergios, (1992), “Cooperation, Conflict, and the Power in the Absence of Property Rights,” American Economic Review, 82: 720-39. Skaperdas, Stergios and Constantinos Syropoulos, (1997), “The Distribution of Income in the Presence of Appropriative Activities,” Economica, 64: 101-117. Smith, Adam, David Skarbeck and Bart Wilson, (2012), “Anarchy, Groups and Conflict: An Experiment on the Emergence of Protective Associations,” Social Choice and Welfare, 38(2): 325-353. Suresh de Mel, David McKenzie and Christopher Woodruff, (2005), "Returns to Capital in Microenterprises: Evidence from a Field Experiment," Quarterly Journal of Economics, 123(4), pp. 1329-1372. Suresh de Mel, David McKenzie and Christopher Woodruff, (2008) “Who are the Microenterprise Owners? Evidence from Sri Lanka on Tokman v. de Soto”, in: 87 International Differences in Entrepreneurship, edited by: Josh Lerner and Antoinette Schoar, NBER. Sutter, Matthias and Martin Kocher, (2007), “Trust and Trustworthiness across Different Age Groups,” Games and Economic Behavior, 29: 364-382. Tabellini, Guido, (2008), “Institutions and Culture,” Journal of the European Economic Association, 6(2-3): 255 -294. Tornell, Aaron and Philip Lane, (1999), “The Voracity Effect,” American Economic Review 89: 22-46. Tversky, Amos and Daniel Kahneman, (1992), “Advances in Prospect Theory: Cumulative Representation of Uncertainty,” Journal of Risk and Uncertainty 5: 297-323. Zelmer, Jennifer, (2003), “Linear Public Goods Experiments: A Meta-Analysis,” Experimental Economics 6: 299-310. 88 Tables: Table I.1. The Effect of Mobile Banking on Wholesaler’ Business Expansion OLS Estimates Dependent Variable: Distance from wholesaler to new retailer (1) (2) (3) (4) (5) (6) (7) (8) Mobile Banking * Center Point * Oct 2008 22,947 22,099 22,038 22,038 22,038 22,500 22,500 21,664 (3754) (3760) (3762) (3763) (3764) (10325) (8671) (10489) Controlling for: Mobile Banking * Center Point N N Y Y Y Y Y Y Mobile Banking N Y N Y Y Y Y Y Mobicom Office Fixed Effects N N N N Y Y Y Y Month Fixed Effects N N N N N Y Y Y Robust Standard Errors Clustered N N N N N N Y Y County Fixed Effects N N N N N N N Y Observations 2,148 2,148 2,148 2,148 2,148 2,135 2,135 2,135 R-squared 0.017 0.021 0.021 0.021 0.021 0.027 0.027 0.026 Notes: The registration data for wholesalers and the survey data for retailers are used. However, the retailers with missing data of the Mobicom office, wholesaler identification numbers are dropped. Only active retailers in May 2010 are included the whole sample. The dependent variable is the distance from wholesaler to retailer. The explanatory variable of interest is the interaction of dummies whether the wholesaler used mobile banking, whether it belongs to Center Point Office, and whether the contract with retailers is initiated in Octobor 2008. Mobile banking indicates whether the wholesaler use mobile banking technology. Four Mobicom offices in Ulaanbaatar are included. Standard errors in parentheses 89 Table I.1. The Effect of Mobile Banking on Wholesaler’ Business Expansion OLS Estimates Dependent Variable: Distance from wholesaler to new retailer (1) (2) (3) (4) (5) (6) (7) (8) Mobile Banking * Center Point * Oct 2008 22,947 22,099 22,038 22,038 22,038 22,500 22,500 21,664 (3754) (3760) (3762) (3763) (3764) (10325) (8671) (10489) Controlling for: Mobile Banking * Center Point N N Y Y Y Y Y Y Mobile Banking N Y N Y Y Y Y Y Mobicom Office Fixed Effects N N N N Y Y Y Y Month Fixed Effects N N N N N Y Y Y Robust Standard Errors Clustered N N N N N N Y Y County Fixed Effects N N N N N N N Y Observations 2,148 2,148 2,148 2,148 2,148 2,135 2,135 2,135 R-squared 0.017 0.021 0.021 0.021 0.021 0.027 0.027 0.026 Notes: The registration data for wholesalers and the survey data for retailers are used. However, the retailers with missing data of the Mobicom office, wholesaler identification numbers are dropped. Only active retailers in May 2010 are included the whole sample. The dependent variable is the distance from wholesaler to retailer. The explanatory variable of interest is the interaction of dummies whether the wholesaler used mobile banking, whether it belongs to Center Point Office, and whether the contract with retailers is initiated in Octobor 2008. Mobile banking indicates whether the wholesaler use mobile banking technology. Four Mobicom offices in Ulaanbaatar are included. Standard errors in parentheses 90 Table II.1.a: Summary Statistics Variables Obs Mean Std. Dev. Min Max sales: Monthly Sales in May 2010 116 11891.49 8752.159 150 52516 t1: Trust Measure 119 3092.437 1733.232 0 6000 t2p: Average Percentage of T2 119 51.50276 21.21507 0 100 t2ave: Average Amount of T2 119 4913.229 2226.802 0 9750 risk: Risk Measure 0-6 120 1.95 2.073644 0 6 ws: wholesaler Dummy 120 0.4416667 0.4986677 0 1 male: Gender Dummy 120 0.3166667 0.4671266 0 1 age 118 37.61864 10.62083 19 67 bus_duration: Length of Business 116 7.258621 4.827675 1 24 loan: Amount of Loan from Banks 97 5635.052 11948.8 0 55000 bank_rate: Monthly Interest Rate 97 0.778866 1.084979 0 3.5 91 Table II.1.b: Correlation between variables risk sales t1 t2p risk age male Loan length ws dummy sales 1 t1: trust 0.3434* 1 t2p: trustworthiness 0.0607 0.3862* 1 risk 0.0322 0.0458 0.1617* 1 age 0.1151 0.0364 0.1795* 0.0422 1 male 0.2536* 0.2533* 0.1122 0.1032 -0.0413 1 Amount of Loan from Banks 0.1573 0.156 0.2482* 0.1093 0.1768* 0.17* 1 Length of Business in years 0.1209 0.1541 0.1679* -0.0461 0.4317* 0.11 0.43* 1 Wholesaler dummy 0.3148* 0.1136 0.1896* 0.0134 0.2275* 0.08 0.46* 0.29* 1 Risk Dummy 0.1956* 0.1521* 0.2066* 0.6060* 0.1152 0.24* 0.14 0.09 0.24* 1 Note: * indicates 5% significance. Risk dummy is a dummy for whether the subject consistently choose the question in the risk measure game Table II.1.c: p-values of Mann-Whitney tests of difference in trust (t1) and trustworthiness, (t2p) t1 t2p p-value <.01 0.0112 92 Table II.2: Correlation between Trust and Sales VARIABLES Monthly Sales in 1000MNT, May 2010 data used OLS estimates (1) (2) (3) (4) (5) (6) (7) t1: Trust measure in MNT 1.712*** 1.883*** 1.911*** 1.672*** 1.638*** 1.639*** (0.44) (0.48) (0.48) (0.48) (0.47) (0.47) t2p: Trustworthiness as a Percentage returned 25.02 -36.11 -48.31 -51.96 -66.55* -67.14* (38.57) (39.48) (40.26) (39.67) (38.56) (39.11) age 103.2 115.2 70.88 70.67 (76.02) (75.06) (73.83) (74.20) male 3,618** 3,133* 3,118* (1,706) (1,653) (1,667) Wholesaler dummy 4,687*** 4,688*** (1,543) (1,550) risk measure: higher number indicates risk averse 38.91 (360) Constant 6,593*** 10,613*** 7,908*** 4,671 4,006 4,577 4,538 (1,559) (2,132) (2,122) (3,301) (3,265) (3,154) (3,189) Observations 116 116 116 114 114 114 114 R-squared 0.118 0.004 0.124 0.139 0.173 0.238 0.239 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 93 Table II.3: Correlation between sales and absolute deviation from the optimal trust, own answers used VARIABLES Absolute Deviation from Optimal Trust sales -0.0473*** -0.0501*** -0.0450** -0.0464** -0.0463** -0.0468** -0.0119 (0.01720) (0.01820) (0.01870) (0.01820) (0.01830) (0.01840) (0.01630) ws 152.8 161.6 320 320.9 309.7 186.1 (320.00000) (319.70000) (316.90000) (317.60000) (320.10000) (269.10000) male -384.7 -285.5 -269.9 -293.9 -1.245 (334.00) (327.40) (328.80) (330.00) (280.50) t2p -19.03*** -18.29** -18.92*** -1.93E+00 (7.12500) (7.21500) (7.16500) (6.55400) risk -51.54 (72.08000) riskall -8.73E+00 (30.15000) t1 -0.559*** (0.08360) Constant 2,985*** 2,951*** 3,009*** 3,897*** 3,952*** 3,946*** 4,310*** (254.20) (265.20) (269.60) (423.60) (431.50) (458.30) (364.00) Observations 116 116 116 116 116 116 116 R-squared 0.062 0.064 0.075 0.131 0.135 0.131 0.382 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 94 Table II.4: Correlation between sales and absolute deviation from the optimal trust, average t2p used VARIABLES Absolute Difference from Optimal Trust, average t2p used sales -0.0603*** -0.0622*** -0.0546*** -0.0562*** -0.0563*** -0.0554*** -0.0165 (0.01760) (0.01860) (0.01890) (0.01830) (0.01830) (0.01840) (0.01530) ws 102.9 116 295 294 318 141.1 (326.00000) (323.40000) (318.00000) (318.50000) (320.80000) (252.30000) male -572.5* -460.4 -477.6 -441.3 -133.6 (337.90) (328.50) (329.80) (330.70) (263.00) t2p -21.51*** -22.32*** -21.75*** -1.85E+00 (7.15100) (7.23700) (7.17900) (6.14600) risk 57.03 (72.31000) riskall 1.96E+01 (30.21000) t1 -0.643*** (0.07830) Constant 3,235*** 3,211*** 3,299*** 4,302*** 4,241*** 4,191*** 4,777*** (258.80) (270.10) (272.80) (425.10) (432.80) (459.20) (341.30) Observations 116 116 116 116 116 116 116 R-squared 0.094 0.095 0.117 0.184 0.188 0.187 0.494 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 95 Table II.5: Sales and dev2, the absolute difference between t2p and its subjective probability (same alpha and beta across subjects) VARIABLES dev2 sales 0.000147 -0.00000392 -0.0000727 -1.94E-08 -2.38E-08 7.14E-07 (0.00023) (0.00024) (0.00024) (0.00000) (0.00000) (0.00000) ws 8.400** 8.280** -0.0206 -0.0207 -0.0235 (4.12500) (4.11700) (0.03050) (0.03060) (0.03060) male 5.229 0.0297 0.0285 0.0346 (4.30) (0.03) (0.03) (0.03) t2p 0.997*** 0.997*** 0.998*** (0.00069) (0.00069) (0.00075) risk 0.00403 0.00371 (0.00694) (0.00692) t1 -1.19E-05 (0.00001) Constant 48.76*** 46.86*** 46.06*** -0.464*** -0.468*** -0.459*** (3.33300) (3.41800) (3.47300) (0.04070) (0.04150) (0.04200) Observations 116.00 116.00 116.00 116.00 116.00 116.00 R-squared 0.004 0.039 0.051 1 1 1 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 96 Table II.6: First risk measure and its correlation with other variables VARIABLES frisk t1 0.0000545 -0.0000233 -0.0000353 -5.00E-05 -5.04E-05 (0.00011) (0.00012) (0.00013) (0.00013) (0.00013) t2p 0.0157* 0.0165* 0.0162 0.0158 0.015 (0.00887) (0.00966) (0.00997) (0.01000) (0.01030) sales 0.00000769 0.00000431 0.00000221 (0.00) (0.00) (0.00) male 3.28E-01 3.71E-01 (0.43700) (0.45100) age 0.00492 (0.01960) Constant 1.756*** 1.115** 1.149** 1.113* 1.112* 9.96E-01 (0.38900) (0.49400) (0.52500) (0.56600) (0.56700) (0.85100) Observations 119.00000 119.00000 119.00000 116.00000 116.00000 114.00000 R-squared 0.00200 0.02600 0.02600 0.02500 0.03000 0.03100 Standard errors in parentheses, * p<0.05 97 Table II.7: Second risk measure and its correlation with other variables VARIABLES riskall t1 (0.00026) (0.00032) (0.00022) (0.00018) (0.00021) -0.000264 -0.000287 -0.000308 -0.000312 -0.000317 t2p 0.00210 0.01210 0.01170 0.01270 0.01430 -0.0217 -0.0235 -0.024 -0.0241 -0.0249 sales (0.00) (0.00) (0.00) -0.000057 -5.81E-05 -5.95E-05 male (0.88500) (0.74300) -1.052 -1.082 age (0.01560) -4.71E-02 Constant 5.719*** 4.808*** 5.273*** 5.767*** 5.769*** 6.328*** -0.936 -1.207 -1.277 -1.363 -1.365 -2.043 Observations 119.00000 119.00000 119.00000 116.00000 116.00000 114.00000 R-squared 0.008 0 0.01 0.021 0.027 0.027 Standard errors in parentheses, * p<0.05 98 Table III. 1: Wealth production schedule # Effort Tokens # Wealth Tokens Produced 1 15 2 28 3 39 4 48 5 55 6 60 7 64 8 67 9 69 10 70 99 Table III. 2: Treatments and group (subject) numbers by site Number of Groups (Subjects) Treatment Description Austria Mexico Mongolia South Korea U.S. Total NCP Identical period structure with (No 7 7 6 8 8 36 simultaneous allocation of endowments Collective (35) (35) (30) (40) (40) (180) among three activities only. Protection) VCP Identical period structure with stage 1 (Voluntary 7 8 6 8 8 37 allocations to collective protection, stage 2 Collective (35) (40) (30) (40) (40) (185) allocations to remaining three activities. Protection) Phase 1 like NCP, then vote on independent versus mandatory voted 7 7 8 8 8 38 VOTE allocations to collective protection at (35) (35) (40) (40) (40) (190) beginning of each of phases 2–6. Like VCP but text message sharing before 8 8 8 8 8 40 CHAT each phase. (40) (40) (40) (40) (40) (200) 29 30 28 32 32 151 Total (145) (150) (140) (160) (160) (755) 100 Table III. 3: Predicted and actual average behaviors and outcomes by treatment Private Collective % of Max. Efficiency Production Theft Protection Protection Earnings Gain NCP 3 4.29 7 2.85 0 2.87 n.a. n.a. 39 46.64 0% 24.6% VCP 3 4.83 7 2.01 0 2.74 0 0.43 39 50.35 0% 36.6% VOTE 7 5.32 0 or 1 1.85 0 1.57 2 or 3 1.26 64 53.89 80.6% 48.0% Voluntary scheme 3 4.45 7 2.82 0 2.35 0 0.39 39 48.21 0% 29.7% Mandatory scheme 7 5.82 0 or 1 1.29 0 1.13 2 or 3 1.76 64 57.12 80.6% 58.5% CHAT 3 8.12 7 0.87 0 0.82 0 0.2 39 62.93 0% 77.2% Notes: Bold entries are predicted values assuming rational self-interested decision-makers with common knowledge of type. For the VOTE treatment, entries refer to phases 2–6 when choice between two methods of contributing to collective protection is available. Earnings are assumed equal to 64 regardless of whether 2 or 3 tokens are mandated to collective protection assuming that slight risk-aversion leads subjects to allocate a seventh token to production rather than theft despite an equal expected return. Percentage of maximum efficiency gain is the proportion of the 31 wealth token difference between earnings predicted in conditions without mandatory collective protection (39) and socially optimal earnings (70). 101 Table III. 4: p-values of Mann-Whitney tests of difference in allocations across treatments VCP VOTE CHAT Collective Protection VCP - <.01 <.01 VOTE - - <.01 Production NCP <.01 <.01 <.01 VCP - <.01 <.01 VOTE - - <.01 Private Protection NCP 0.33 <.01 <.01 VCP - <.01 <.01 VOTE - - <.01 Theft NCP <.01 <.01 <.01 VCP - 0.28 <.01 VOTE - - <.01 Earnings per Period NCP <.01 <.01 <.01 VCP - <.01 <.01 VOTE - - <.01 For the VOTE treatment, only results from phases 2–6, when choice between two methods of contributing to collective protection is available, are taken into account. 102 Table III. 4b: p-values of Mann-Whitney tests of difference in allocations across treatments among entrepreneur subjects VOTE S-CHAT Collective Protection VCP .1556 .3358 VOTE - 1.0 Production VCP .7987 .9310 VOTE - .9808 Private Protection VCP .5443 .5932 VOTE - .9157 Theft VCP .5035 .038 VOTE - <.01 Earnings per Period VCP .6270 .9923 VOTE - .7876 Table III. 5c: p-values of Mann-Whitney tests of difference in allocations across treatments between student and entrepreneur subjects VCP VOTE S-CHAT Collective Protection <.01 <.01 <.01 Production <.01 <.01 <.01 Private Protection .017 .19 <.01 Theft <.01 <.01 <.01 Earnings per Period .014 .083 .021 103 Table III. 5: Kruskal-Wallis tests of difference in allocations across countries χ2(4) adjusted for ties p-value Panel A: NCP Production 12.57 0.014 Private protection 2.43 0.657 Theft 4.63 0.328 Panel B: VCP Collective protection 5.32 0.256 Production 8.44 0.077 Private protection 6.29 0.178 Theft 7.20 0.126 Panel C: VOTE (phases 2-6) (i) Institutional preferences Support for mandatory scheme 13.69 0.008 Selection of mandatory scheme 10.62 0.031 (ii) Under mandatory scheme Support for mandatory scheme 4.24 0.375 Collective protection 5.70 0.223 Production 6.42 0.170 Private protection 7.52 0.111 Theft 5.20 0.268 (iii) Under independent contributions Support for mandatory scheme 14.66 0.006 Collective protection 9.49 0.050 Production 1.05 0.902 Private protection 9.56 0.049 Theft 3.41 0.491 Panel D: CHAT Collective protection 15.57 0.004 Production 18.30 0.001 Private protection 15.79 0.003 Theft 18.87 0.001 104 Table III. 6: Comparison of institutional choice of student subjects in five countries and entrepreneur subjects Rounds Rounds 5 9 13 17 21 5 9 13 17 21 US Mongolia Session 1 Session 1 Group 1 I V V V V Group 1 I I I I I Group 2 I I V V V Group 2 I I I I I Group 3 V V V I V Group 3 I V I I I Group 4 V V V V V Group 4 I I I I I Session 2 Session 2 Group 1 V V V V V Group 1 I V I I V Group 2 I I I I I Group 2 V V V V V Group 3 I I I I I Group 3 I V I I I Group 4 I V V V V Group 4 I I I I I 5 9 13 17 21 5 9 13 17 21 Mexico Austria Session 1 Session 1 Group 1 V V V V V Group 1 V V V V V Group 2 V V V V V Group 2 I I V V I Group 3 I V V V V Group 3 V V V V V Session 2 Group 4 V V V V V Group 1 V V V V V Session 2 Group 2 V V V V V Group 1 I V I V V Group 3 V V V V V Group 2 I V V V V Group 4 I I I I I Group 3 V I V V V 105 Entrepreneur Subjects 5 9 13 17 21 5 9 13 17 21 Under 32 Over 38 Group 1 V I V V V Group 1 V I I V I Group 2 I I V I I Group 2 V V V V V Group 3 V V V V V Group 3 V I V I I Group 4 I V I V V Group 4 V V V V V Notes: "V" stands for voting scheme; "I" stands for independent contributions. Table III. 7: Comparison of the number of promises to steal between entrepreneur and students subjects in S-CHAT treatment Entrepreneurs Choice Round 0 Round 4 Round 8 Round 12 Round 16 Round 20 Never Steal 17 14 23 21 13 17 Steal back 17 19 12 11 20 20 Always Steal 6 6 4 7 7 3 Students Choice Round 0 Round 4 Round 8 Round 12 Round 16 Round 20 Never Steal 13 22 13 18 13 12 Steal back 21 11 22 16 15 13 Always Steal 6 7 5 4 9 13 Note: Each cell shows the number of subjects made that particular choice in each round 106 Figures: Figure I.1. Spatial Distribution of Wholesalers in Ulaanbaatar Figure 1 shows the spatial distribution of wholesalers in Ulaanbaatar. The registration data of Mobicom, retrieved in May, 2010, is used. County borders are shown as the main layer and the paved roads are depicted in blue line. Measure is in kilometers. 107 Fugure I.2. Spatial Distribution of Mobicom Retailers in Ulaanbaatar Figure 2 shows the spatial distribution of mobi-card retailers in Ulaanbaatar. The registration data of Mobicom, retrieved in May, 2010, is used. County borders are shown as the main layer and the paved roads are depicted in blue line. Measure is in kilometers. 108 Figure I.3. County Populations in Ulaanbaatar Figure 3 shows the population of each counties in Ulaanbaatar by five categories. Source is 2010 National Statistics Division data. County borders are shown as the main layer. Measure is in kilometers. 109 Figure I.4. Comparison of Actual versus Predicted Number of Retailers by Each County actual predict 60 ed 50 40 30 20 10 0 109 101 105 113 117 1 5 9 25 33 13 17 21 29 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 Notes: Y-axis shows the actual and predicted number of entries in each county. X-axis shows each counties sorted by the predicted number of entries in descending order. Log-normal distribution is used for idiosyncratic profit and sales data is used as a proxy for demand 110 Figure I. 5. Average Distance from Wholesalers to New Retailers in Entry Mouth. 30000 tedy1 Average Distance between Wholesalers and Retailers, tedy2 25000 tedy3 20000 in meters 15000 10000 5000 0 200903 201002 200804 200809 200810 200811 200812 200901 200904 200905 200906 200907 200908 200909 200910 200911 200912 201001 201003 201004 201005 201006 Months Notes: Y-axis shows the average distance from wholesalers to their new retailers added in each month. X-axis shows the months of retailer registration. The registration data at Mobicom, retrieved in June 2010 is used. All distances are in meters. The graph separates the four Mobicom offices in Ulaanbaatar: tedy1, 2, 3 and Center Point. The dotted square highlights the impact of mobile banking technology, which only affected Center Point office in October 2008. 111 Figure II.1: Number of subjects for each trust level in student and entrepreneur subjects T1 for Entrepreneurs T1 for Students 12 10 8 6 4 2 0 0 0 0 0 0 0 0 0 0 0 0 0 Figure II.2: Percentage of subjects for each trust level in student and entrepreneur subjects t1 entrepreneurs t1 students 0.3 0.25 0.2 0.15 0.1 0.05 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 112 Figure II.3: Average amount and percentage of t2 given t1 by subject groups Entrepreneurs Student Entrepreneurs % Student % 10000 0.7 9000 0.6 8000 7000 0.5 6000 0.4 5000 0.3 4000 3000 0.2 2000 0.1 1000 0 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 Figure II.4: Expected earnings for any given t1, calculated from own answers Entrepreneurs Students 10000 9000 8000 Expected Earnings in MNT 7000 6000 5000 4000 3000 2000 1000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 113 Figure II.5: Correlation between trust and sales 50000 40000 30000 20000 10000 0 0 2000 4000 6000 trust measure, the amount sent as first mover in trust game Figure II.6: Correlation between sales and absolute deviation from the optimal trust 50000 40000 30000 20000 10000 0 0 2000 4000 6000 Absolute value of Deviation 114 Figure II.7: Correlation between the average of t2p and its subjective probability .85 .8 .75 .7 .65 .6 0 .2 .4 .6 .8 1 average fraction of t2 Figure II.8: Correlation between sales and the deviation from the optimal trust, subjective probability calculated from alpha and beta that varies across individuals 50000 40000 30000 20000 10000 0 0 .1 .2 .3 .4 .5 diff_abs_psubj 115 Figure III. 1.a: Session timelines for each treatment Panel A: NCP and VCP Instructions and Break of 1 Break of 1 Break of 1 Break of 1 Break of 1 Debriefing two practice* rounds minute minute minute minute minute questionnaire** 4 periods 4 periods 4 periods 4 periods 4 periods 4 periods of play of play of play of play of play of play 1 4 8 12 16 20 24 Period Panel B: VOTE Instructions of schemes and vote; two practice* rounds (one using each of the schemes), and 1 st Vote Instructions of NCP condition and Debriefing two practice* rounds questionnaire** 2nd Vote 3rd Vote 4th Vote 5 th Vote 4 periods of play under 4 periods under 4 periods under 4 periods under 4 periods under 4 periods under NCP chosen scheme chosen scheme chosen scheme chosen scheme chosen scheme 1 4 8 12 16 20 24 Period Panel C: CHAT Instructions, two practice* rounds (including 1 min of chatting), and 1st 2 nd Chat 3rd Chat 4th Chat 5 th Chat (2 6th Chat Debriefing Chat (4 min) (3.5 min) (3 min) (2.5 min) min) (2 min) questionnaire** 4 periods 4 periods 4 periods 4 periods 4 periods 4 periods of play of play of play of play of play of play 1 4 8 12 16 20 24 Period * Practice rounds were guided by experimenter directions for familiarization with the software interface and without indications of others' likely choices. ** In Austria, Korea, Mongolia and the U.S., sessions ended with a debriefing questionnaire. In Mexico, subjects completed the questionnaire several days before their participation in the lab; sessions ended with subjects writing down their comments about the experiment. 116 Figure III. 1.b: Timelines of stage games for each treatment Panel A: NCP Each team member selects Members see outcome mᵢ, pᵢ, and tij for all j≠i and summary statistics First Stage Second Stage Panel B: VCP, CHAT, and VOTE if the independent contributions scheme is chosen Each team Each team member selects member selects cᵢ mᵢ, pᵢ, and tij for all j≠i Members see outcome Members learn ∑cᵢ and summary statistics First Stage Second Stage Third Stage Panel C: VOTE if the mandatory contributions scheme is chosen Each team member Each team member selects proposes cᵢ mᵢ, pᵢ, and tij for all j≠i Members learn median proposed cᵢ and computer deducts it Members see outcome automatically from each member's and summary statistics effort token endowment First Stage Second Stage Third Stage 117 Figure III. 2.a: Average allocation to production by period and treatment 9 8 7 6 5 4 0 5 10 15 20 25 Period NCP VCP VOTE CHAT Figure III. 2.b: Average allocation to collective protection by period and treatment 1.5 1 .5 0 0 5 10 15 20 25 Period VCP VOTE CHAT 118 Figure III. 2.c: Average allocation to private protection by period and treatment 3 2.5 2 1.5 1 .5 0 5 10 15 20 25 Period NCP VCP VOTE CHAT Figure III. 2.d: Average allocation to theft by period and treatment 3 Avg. allocation to theft 2 1 0 0 5 10 15 20 25 Period NCP VCP VOTE CHAT 119 Figure III. 2.e: Average earnings by period and treatment 65 60 Avg. earnings per round 55 50 45 40 0 5 10 15 20 25 Period NCP VCP VOTE CHAT Figure III. 3: Average share of endowment allocated to each activity by country and treatment Panel A: NCP 2.51 2.67 3.25 3.09 2.75 2.85 4.73 4.32 3.80 3.89 4.62 4.29 2.76 3.01 2.94 3.02 2.63 2.87 Austria Mexico Mongolia South USA Overall Korea Private protection Production Theft 120 Panel B: VCP 1.66 1.59 2.01 2.47 2.17 2.22 4.82 5.47 4.65 4.63 4.83 4.57 3.08 2.33 2.74 2.49 2.89 2.77 0.38 0.51 0.64 0.28 0.39 0.43 Austria Mexico Mongolia South Korea USA Overall Collective protection Private protection Production Theft Panel C: VOTE 1.60 1.39 1.63 1.77 1.85 2.75 5.49 5.34 5.81 5.43 5.32 4.62 1.25 1.72 1.62 1.57 1.33 1.88 1.87 1.26 1.31 1.18 1.26 0.75 Austria Mexico Mongolia South Korea USA Overall Collective protection Private protection Production Theft 121 Panel D: CHAT 0.25 0.53 0.29 0.72 0.87 2.58 8.45 8.80 8.12 9.56 5.04 8.72 2.03 0.12 0.78 0.61 0.82 0.07 0.24 0.34 0.30 0.030.54 0.20 Austria Mexico Mongolia South Korea USA Overall Collective protection Private protection Production Theft Figure III. 4.a: Incidence of property crimes and allocations to private protection, all treatments 2.4 Mongolia 2.2 South Korea Mexico 2 USA 1.8 Austria 1.6 .1 .15 .2 .25 .3 .35 Share of population who experienced burglary or robbery 122 Figure III. 4.b: Incidence of property crimes and allocations to collective protection, all treatments .8 Mexico .7 .6 South Korea Mongolia Austria .5 USA .1 .15 .2 .25 .3 .35 Share of population who experienced burglary or robbery Figure III. 4.c: Incidence of property crimes and allocations to collective protection, period one only 1.8 Mongolia 1.6 1.4 Mexico 1.2 South Korea 1 Austria USA .8 .1 .15 .2 .25 .3 .35 Share of population who experienced burglary or robbery 123 Figure III. 5.a: Perception of safety and allocations to private protection, all treatments 2.4 2.2 Mongolia South Korea Mexico 2 USA 1.8 Austria 1.6 -.2 -.1 0 .1 .2 Composite index of perception of safety Figure III. 5.b: Perception of safety and allocations to collective protection, all treatments .8 Mexico .7 .6 South Korea Mongolia Austria .5 USA -.2 -.1 0 .1 .2 Composite index of perception of safety 124 Figure III. 5.c: Perception of safety and allocations to collective protection, period one only 1.8 Mongolia 1.6 1.4 Mexico 1.2 South Korea 1 Austria USA .8 -.2 -.1 0 .1 .2 Composite index of perception of safety Figure III. 6: Governance index and share of individual votes for the mandatory scheme .7 Austria South Korea Mexico .6 USA .5 .4 .3 Mongolia -.5 0 .5 1 1.5 2 Governance Index 125 Figure III. 7: Trust Index and production in CHAT treatment. 10 Austria 9 South Korea USA Mexico 8 7 6 Mongolia 5 20 40 60 80 Trust Index Figure III. 8a: Average Allocation to Private protection by age group and period 3 2.5 2 1.5 1 0 5 10 15 20 25 Round Entrepreneurs under 32 Entrepreneurs over 38 Students 126 Figure III. 8b: Average Allocation to Theft by age group and period 3.5 3 Avg. allocation to theft 2.5 2 1.5 1 0 5 10 15 20 25 Round Entrepreneurs under 32 Entrepreneurs over 38 Students Figure III. 8c: Average Earnings by age group and period 60 Avg. earnings per period 55 50 45 0 5 10 15 20 25 Round Entrepreneurs under 32 Entrepreneurs over 38 Students 127 Figure III. 9a: Average Earnings by treatment and period among Entrepreneurs 60 55 50 45 0 5 10 15 20 25 round Profit in each round, VCP Profit in each round, VOTE Profit in each round, S-CHAT Figure III. 9b: Average Theft Allocation by treatment and period among Entrepreneurs 3 2.5 2 1.5 1 .5 0 5 10 15 20 25 round Theft allocate in each round, VCP Theft allocate in each round, VOTE Theft allocate in each round, S-CHAT 128 Figure III. 10a: Correlation between Sales and Theft Allocation in the first period 50000 40000 30000 20000 10000 0 0 2 4 6 8 Number of tokens allocated to Theft in the first round Figure III. 10b: Correlation between Sales and Production Allocation in the first period 50000 40000 30000 20000 10000 0 0 2 4 6 8 10 production allocate in the first period of property rights game 129 Figure III. 11a: Distribution of Theft Allocation by period 0 1 2 3 4 1.5 1 .5 0 5 6 7 8 9 1.5 1 .5 0 10 11 12 13 14 1.5 Density 1 .5 0 15 16 17 18 19 1.5 1 .5 0 0 5 10 20 21 22 23 1.5 1 .5 0 0 5 10 0 5 10 0 5 10 0 5 10 theft_allocate_total Graphs by round Figure III. 11b: Distribution of Production Allocation by period 0 1 2 3 4 1 .5 0 5 6 7 8 9 1 .5 0 10 11 12 13 14 1 Density .5 0 15 16 17 18 19 1 .5 0 0 5 10 20 21 22 23 1 .5 0 0 5 10 0 5 10 0 5 10 0 5 10 production_allocate Graphs by round 130 Figure III. 11c: Distribution of Collective Protection Allocations by period 0 1 2 3 4 10 20 30 0 5 6 7 8 9 10 20 30 0 10 11 12 13 14 10 20 30 Density 0 15 16 17 18 19 10 20 30 0 0 .2 .4 .6 .8 20 21 22 23 10 20 30 0 0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6 .8 collective_protection Graphs by round Figure III. 11c: Distribution of Private Protection Allocation by period 0 1 2 3 4 1.5 1 .5 0 5 6 7 8 9 1.5 1 .5 0 10 11 12 13 14 1.5 Density 1 .5 0 15 16 17 18 19 1.5 1 .5 0 0 5 10 20 21 22 23 1.5 1 .5 0 0 5 10 0 5 10 0 5 10 0 5 10 private_protect_allocate Graphs by round 131 Figure III. 12: Coefficient of retaliation to theft over 24 rounds 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 -1 -2 -3 Note: This Figure III. reports the coefficient from regression the loss from a theft in period i on the theft allocation on period i-1 over 24 rounds with entrepreneur subjects. The 95% confidence interval is shown in dotted lines 132 Appendix Figure A I.1. Comparison of Actual versus Predicted Number of Retailers by Each County 45 40 Actual Number of 35 Entries 30 Number of Entries 25 20 15 10 5 0 81 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 85 89 93 97 101 105 109 113 117 Counties Notes: Y-axis shows the actual and predicted number of entries in each county. X-axis shows each counties sorted by the predicted number of entries in descending order. Normal distribution is used for idiosyncratic profit and sales data is used as a proxy for demand. 133 Figure A I.2. Predicted Number of Entries when Standard Deviation is Changed. Notes: Y-axis shows the actual and predicted number of entries in each county. The blue line shows the actual number of entries. X-axis shows each counties sorted by the predicted number of entries in descending order. Log-normal distribution is used for idiosyncratic profit and sales data is used as a proxy for demand. This note applies to A3, A4 and A5 as well Figure A I.3. Predicted Number of Entries when f is Changed. Notes: See Figure A I.2. 134 Figure A I.4. Predicted Number of Entries when alpha is changed. Notes: See Figure A I.2. Figure A I.5. Predicted Number of Entries when beta is changed. Notes: See Figure A I.2. 135 Appendix II.1: Construction of riskall. Risk measure questionnaire consisted of five choices between a certain amount and a lottery with increasing expected value. If the subject chooses the certain amount in a particular question, I mark that question as C. If the subject prefers the risky choice of two outcomes in a particular question, I mark that question as R. If a question is marked as R, there is no certainty involved. Therefore, that question gets 0 in this risk measure. If a question is marked as C, that question gets the value of the order number of the question. For example: CONSISTENT CHOICES INCONSISTENT CHOICES CCCCC = 1+2+3+4+5 = 15 (with riskall=5) RCCCC = 0+2+3+4+5 = 14 CCCCR = 1+2+3+4+0 = 10 (with riskall=4) CRCRC = 1+0+3+0+5 = 9 CCRRR = 1+2+0+0+0 = 3 (with riskall=2) CRCCR = 1+0+3+4+0 = 8 RRRRR= 0+0+0+0=0=0 (with riskall=0) RCRCR = 0+2+0+4+0 = 6 136 Appendix Table II. 1: Estimates of alpha, beta and subjective probability of t2p ID alpha s.e. beta s.e. t1 t2p psubj psubj_corr 1 0.50 0.06 0.000071 0.000016 2000 0.73 0.42 2 0.42 0.11 0.000078 0.000029 6000 0.17 0.65 3 0.59 0.10 0.000046 0.000028 1500 0.44 0.74 0.74 4 0.50 0.14 0.000065 0.000039 4000 0.71 0.33 5 0.16 0.21 0.000052 0.000058 5000 0.32 0.50 6 0.40 0.04 0.000034 0.000011 4000 0.29 2.67 7 0.63 0.10 0.000056 0.000027 1500 0.45 0.71 0.71 8 0.61 0.14 0.000112 0.000039 5000 0.24 0.65 9 0.36 0.04 0.000002 0.000010 2000 0.36 0.94 0.94 10 0.33 0.01 0.000001 0.000002 1000 0.33 1.00 11 0.34 0.20 0.000054 0.000055 1000 0.52 0.74 12 0.63 0.04 0.000027 0.000011 2000 0.54 0.64 0.64 13 0.68 0.03 0.000012 0.000009 3000 0.64 0.55 0.55 14 0.72 0.15 0.000120 0.000042 5000 0.33 0.69 15 0.27 0.06 0.000011 0.000017 2000 0.24 1.46 16 0.48 0.15 0.000003 0.000041 6000 0.48 0.65 17 0.47 0.09 0.000025 0.000024 5000 0.55 0.46 18 0.54 0.04 0.000036 0.000010 3000 0.43 1.02 19 0.26 0.02 0.000040 0.000007 3000 0.13 16.19 20 0.28 0.04 0.000030 0.000011 500 0.18 1.34 21 0.64 0.18 0.000034 0.000048 5500 0.53 1.27 22 0.35 0.04 0.000063 0.000011 1500 0.56 0.62 23 0.26 0.03 0.000002 0.000009 2000 0.27 1.22 24 0.42 0.06 0.000036 0.000017 1000 0.31 0.95 0.95 25 0.78 0.12 0.000069 0.000032 2000 0.56 0.66 0.66 26 0.67 0.20 0.000069 0.000055 3000 0.45 1.27 27 0.62 0.19 0.000006 0.000052 4000 0.60 0.58 0.58 28 0.56 0.11 0.000051 0.000030 4000 0.39 2.29 29 0.30 0.04 0.000049 0.000012 3000 0.46 0.56 30 0.35 0.06 0.000029 0.000016 5000 0.44 0.53 31 0.51 0.05 0.000036 0.000014 6000 0.63 0.35 32 1.02 0.10 0.000052 0.000027 6000 0.85 0.83 0.83 33 0.78 0.05 0.000026 0.000015 6000 0.69 0.71 0.71 34 0.56 0.20 0.000020 0.000054 1000 0.50 0.63 0.63 35 0.64 0.12 0.000066 0.000033 500 0.43 0.58 0.58 36 0.60 0.16 0.000040 0.000042 3000 0.47 0.92 0.92 37 0.50 0.05 0.000027 0.000014 6000 0.59 0.40 38 0.00 0.00 0.000000 0.000000 0 0.00 39 0.80 0.21 0.000007 0.000057 2000 0.78 0.43 0.43 137 40 0.67 0.00 0.000000 0.000000 2000 0.67 0.50 0.50 41 0.46 0.06 0.000017 0.000017 4000 0.52 0.56 42 0.67 0.00 0.000000 0.000000 6000 0.67 0.50 0.50 43 0.79 0.15 0.000054 0.000041 3000 0.61 0.72 0.72 44 0.53 0.05 0.000021 0.000014 1000 0.46 0.69 0.69 45 0.19 0.04 0.000040 0.000012 6000 0.07 1.19 46 0.00 0.00 0.000000 0.000000 2000 0.00 47 0.79 0.12 0.000087 0.000033 1000 0.51 0.54 0.54 48 0.40 0.04 0.000016 0.000011 1500 0.34 0.96 0.96 49 0.58 0.11 0.000020 0.000029 1000 0.51 0.62 0.62 50 0.53 0.10 0.000014 0.000028 2000 0.58 0.57 51 0.43 0.06 0.000021 0.000016 2000 0.36 0.97 0.97 52 0.64 0.22 0.000073 0.000060 1000 0.40 0.67 0.67 53 0.42 0.07 0.000076 0.000019 4000 0.17 1.76 54 0.81 0.06 0.000046 0.000016 500 0.67 0.43 0.43 55 0.23 0.14 0.000093 0.000037 4000 0.53 0.35 56 0.43 0.06 0.000012 0.000017 3000 0.47 0.67 57 0.41 0.07 0.000007 0.000020 3000 0.39 0.91 0.91 58 1.01 0.07 0.000034 0.000019 5000 0.90 0.50 0.50 59 0.40 0.08 0.000071 0.000022 1000 0.16 1.32 60 0.49 0.11 0.000053 0.000031 3000 0.66 0.42 61 0.71 0.08 0.000025 0.000021 2500 0.79 0.40 62 0.30 0.08 0.000007 0.000022 2000 0.32 1.02 63 0.79 0.20 0.000066 0.000055 2000 0.57 0.64 0.64 64 1.00 0.00 0.000000 0.000000 5000 1.00 0.33 0.33 65 0.21 0.04 0.000038 0.000010 500 0.09 1.94 66 0.00 0.00 0.000000 0.000000 0.00 67 0.94 0.17 0.000067 0.000047 500 0.72 0.38 0.38 68 1.03 0.11 0.000061 0.000029 5000 0.84 0.78 0.78 69 0.50 0.08 0.000017 0.000021 5000 0.45 1.01 70 0.54 0.03 0.000011 0.000009 2000 0.50 0.68 0.68 71 0.44 0.19 0.000021 0.000050 1000 0.37 0.83 0.83 72 0.67 0.12 0.000007 0.000034 1500 0.69 0.48 73 0.95 0.03 0.000007 0.000009 6000 0.93 0.38 0.38 74 1.00 0.00 0.000000 0.000000 6000 1.00 0.33 0.33 75 0.35 0.06 0.000012 0.000017 500 0.31 0.98 0.98 76 0.46 0.05 0.000024 0.000015 2000 0.53 0.60 77 0.53 0.13 0.000031 0.000035 3000 0.63 0.47 78 0.50 0.05 0.000020 0.000014 2000 0.44 0.78 0.78 79 0.82 0.09 0.000128 0.000024 3000 0.40 7.10 80 0.45 0.11 0.000059 0.000029 3000 0.26 3.28 81 0.66 0.03 0.000021 0.000007 4000 0.72 0.40 82 0.74 0.07 0.000069 0.000020 1000 0.52 0.55 0.55 138 83 0.34 0.04 0.000067 0.000010 1500 0.55 0.62 84 0.26 0.12 0.000027 0.000033 5000 0.35 0.63 85 0.67 0.00 0.000000 0.000000 6000 0.67 0.50 0.50 86 0.43 0.08 0.000004 0.000022 2000 0.42 0.79 0.79 87 0.43 0.03 0.000001 0.000008 2500 0.43 0.78 0.78 88 0.36 0.11 0.000019 0.000029 3500 0.42 0.67 89 0.64 0.06 0.000022 0.000016 5500 0.72 0.38 90 0.62 0.04 0.000015 0.000010 2000 0.57 0.59 0.59 91 0.67 0.00 0.000000 0.000000 6000 0.67 0.50 0.50 92 0.72 0.05 0.000026 0.000015 5000 0.64 0.72 0.72 93 0.48 0.12 0.000001 0.000033 2500 0.47 0.71 0.71 94 0.54 0.03 0.000011 0.000009 3000 0.50 0.71 0.71 95 0.70 0.03 0.000035 0.000008 4000 0.81 0.34 96 0.68 0.06 0.000041 0.000016 3000 0.55 0.77 0.77 97 1.00 0.00 0.000000 0.000000 5000 1.00 0.33 0.33 98 0.83 0.07 0.000032 0.000018 5500 0.93 0.28 99 0.51 0.05 0.000017 0.000014 1000 0.45 0.70 0.70 100 0.78 0.03 0.000024 0.000008 5000 0.86 0.33 101 0.79 0.06 0.000095 0.000016 3000 0.48 1.56 102 0.89 0.08 0.000051 0.000022 3000 0.72 0.57 0.57 103 0.74 0.08 0.000111 0.000022 3500 0.38 8.23 104 0.51 0.10 0.000061 0.000028 3000 0.31 2.30 105 0.33 0.00 0.000000 0.000000 2000 0.33 1.00 1.00 106 0.44 0.07 0.000012 0.000020 1500 0.40 0.83 0.83 107 0.74 0.06 0.000023 0.000016 3000 0.81 0.38 108 0.69 0.03 0.000049 0.000008 5000 0.85 0.28 109 0.52 0.11 0.000054 0.000029 1000 0.69 0.53 110 0.61 0.18 0.000049 0.000049 5000 0.76 0.31 111 0.58 0.04 0.000005 0.000011 1500 0.56 0.59 0.59 112 0.36 0.11 0.000013 0.000029 500 0.31 0.97 0.97 113 0.56 0.11 0.000051 0.000030 2500 0.39 1.11 114 0.69 0.03 0.000051 0.000007 5000 0.86 0.28 115 0.66 0.10 0.000088 0.000027 4000 0.38 8.94 116 0.64 0.02 0.000015 0.000005 3000 0.59 0.61 0.61 117 0.64 0.09 0.000037 0.000026 2500 0.52 0.74 0.74 118 0.59 0.08 0.000017 0.000023 4500 0.54 0.76 0.76 119 0.67 0.00 0.000000 0.000000 6000 0.67 0.50 0.50 120 0.55 0.09 0.000047 0.000023 3000 0.40 1.24 139 Appendix, Form II.1: Translation of trust game instruction Your first decisions in today’s experiment will determine part of your earnings. From this first portion, you can earn at least 3000MNT, but depending on your decision and the decision of another participant with whom you will be paired for this part, you might earn less or more than that (anything between 0 and 3000). You will indicate your choices for this portion by circling numbers on a form. In all, you must circle one number on each of eleven different lines. First, we will explain the general idea of this part of the experiment. You and the participant with whom you are paired will each begin this portion being credited with money in the amount of 6000MNT. One of you will end up being assigned to the role of first decision maker, or Participant A. The other will be second decision maker, or Participant B. A makes only one decision: whether to send to B some, all, or none of A’s 6000MNT. If A sends some amount of money to B, the experimenter triples that amount of money, causing B to receive three times what A sent. For example, if A decides to send 2000MNT, B receives 6000MNT. If A decides to send 4000MNT, B receives 12000MNT. B then decides how much, if any, of the money received B sends back to A. B can keep the entire amount, return the entire amount, or send back any amount in. Amounts that B sends back to A are not tripled, so if B sends back to A, say 2000MNT, A receives 2000MNT. Given the decisions made by A and B, A will earn from this portion of the experiment any part of A’s initial 6000MNT that A does not send to B plus any part of the tripled amount that B sends back to A. B will earn B’s initial 6000MNT plus the tripled amount, if anything, sent by A, minus any part of the tripled amount that B chooses to send back to A. Each individual will engage in this kind of interaction only one time. The money that you earn from this interaction will be added to other money you earn today and paid to you in cash at the end of the experiment. The results of this first portion of today’s experiment will not be announced to you until all parts of the experiment are over. (Even then, you will never learn the identity of the participant with whom you were paired.) Therefore, an individual in the B role does not know in advance how much (if anything) their counterpart in the A role has decided to send. Instead, each individual makes a decision for every possible amount that A might send. B decides how much, if anything, to send back to A if A sends B 500MNT (and B receives 1500MNT); how much, if anything, to send back to B if A sends B 1000MNT (and B receives 3000MNT); and so on. (If A sends 0, there is no real decision to be made by B.) Which decision of B’s actually ends up determining B’s earnings depends on what participant A he is randomly paired with and on what that participant A has decided to do. Whether you are going to be in role A or in role B is going to be determined randomly after you fill out the decision form. Therefore, every participant is asked to complete the form in its entirety, without knowing whether you are in role A or role B. You are to begin by selecting an amount (if anything) you want to send to your counterpart B if you end up being in role A. Then you will select an amount (if anything) you want to send back to your counterpart A for each possible amount that A might send if you end up being in role B. Before you begin, please look at the decision sheet. Then raise your hand if you have any questions, and I will come to you. Do not begin until you are sure that you understand your task. 140 Decision Sheet for 1st Portion of Experiment Instructions: For each numbered line (except 2.), circle clearly only one of the possible items. 1. If I am in role A, I wish to send to my counterpart in role B (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 2. (If I am in role B and my counterpart A sends me 0, I have no decision to make.) 3. If I am in role B and my counterpart A sends me 500 (I receive 1500), I choose to return to A (circle one number) 0 500 1000 1500 4. If I am in role B and my counterpart A sends me 1000 (I receive 3000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 5. If I am in role B and my counterpart A sends me 1500 (I receive 4500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 6. If I am in role B and my counterpart A sends me 2000 (I receive 6000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 7. If I am in role B and my counterpart A sends me 2500 (I receive 7500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8. If I am in role B and my counterpart A sends me 3000 (I receive 9000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9. If I am in role B and my counterpart A sends me 3500 (I receive 10500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 10. If I am in role B and my counterpart A sends me 4000 (I receive 12000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 141 11. If I am in role B and my counterpart A sends me 4500 (I receive 13500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 12. If I am in role B and my counterpart A sends me 5000 (I receive 15000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000 13. If I am in role B and my counterpart A sends me 5500 (I receive 16500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000 15500 16000 16500 14. If I am in role B and my counterpart A sends me 6000 (I receive 18000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000 15500 16000 16500 17000 17500 18000 142 Appendix Form II.2: Risk measure game instruction 1) Please indicate whether you would like to receive 1500MNT or would prefer to let the die randomly select for you either 0 or 2700MNT (each with a 50% probability). a. Guaranteed 1500MNT b. Let the die decide either 0 or 2700MNT 2) Please indicate whether you would like to receive 1500MNT or would prefer to let the die randomly select for you either 0 or 2700MNT (each with a 50% probability). a. Guaranteed 1500MNT b. Let the die decide either 0 or 3000MNT 3) Please indicate whether you would like to receive 1500MNT or would prefer to let the die randomly select for you either 0 or 2700MNT (each with a 50% probability). a. Guaranteed 1500MNT b. Let the die decide either 0 or 3500MNT 4) Please indicate whether you would like to receive 1500MNT or would prefer to let the die randomly select for you either 0 or 2700MNT (each with a 50% probability). a. Guaranteed 1500MNT b. Let the die decide either 0 or 4000MNT 5) Please indicate whether you would like to receive 1500MNT or would prefer to let the die randomly select for you either 0 or 2700MNT (each with a 50% probability). a. Guaranteed 1500MNT b. Let the die decide either 0 or 4500MNT 143 Appendix A – Representativeness of student subjects A.1 Student and general population responses to survey questions One way to investigate whether our student subject pools are representative of the general population in their countries is to compare the survey responses they provided at the end of their experiment sessions (or in the case of the Mexican subjects, one or more weeks before those sessions * ) with those of larger surveys, such as the World Values Survey. Two questions conducive to such comparison are one regarding self-positioning on a left-to-right political spectrum and one regarding the trustworthiness of others. We display the relevant data in Table III. A1. On trust, we have subject responses from three of our subject pools to a question about the likelihood of a lost wallet being returned, and for all five countries, data from the World Values Survey (WVS), East Asia Barometer (EAB), or the European Values Survey (EVS) on what has come to be called the “generalized trust question” (“Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?”). For the three countries for which both measures are available, there is a consistent ordering, with both the highest expectation that the wallet would be returned and the highest generalized trust in the U.S., the lowest response on both in Mexico, and a middle position for Austria. Table III. A1: Average survey responses by subject pool in country surveys Country Category and variable Description and source Austria*** Mexico* Mongolia** South Korea* U.S.* Trust a 2.83 2.51 n.a. n.a. 3.02 Wallet return Post- or pre-experiment survey Generalized trust b WVS/EAB/EVS 36.8 15.6 10.2 28.2 39.3 Post- or pre-experiment survey 1.52 3.23 1.98 1.76 2.11 Political Outlook c WVS/EAB/EVS 2.7 3.1 n.a. 2.9 2.85 Sources: *World Values Survey Wave 5 (2005-2008); **East Asia Barometer (2006); ***European Values Survey (1999) a 1=0-20%; 2=21-40%; 3=41-60%; 4=61-80%; 5=81-100% b % of respondents saying "most can be trusted" to the generalized trus question c 1=very liberal,..., 5=very conservative *In order to minimize the danger of influencing subjects’ behaviors during the experiment by asking questions about attitudes towards theft prior to their experiment session, the pre-experiment survey of ITAM students in Mexico City included more than three times the number of questions as the post-experiment surveys administered elsewhere, with questions on theft interspersed among questions on various other political and social topics. 144 We have data on self-reported political outlook from both our subject survey and the WVS/EVS for four countries, data for Mongolia being unavailable in the EAB. In this case, our own survey question wording is identical to that in the WVS and EVS. For these four countries there is consistency between the two sources insofar as Austria is the most liberal and Mexico the most conservative. The orderings for the two countries in the middle, South Korea and the U.S., differ by survey, although their WVS values are essentially the same. In the case of political outlook, the numbers suggest that with the exception of our Mexican site, the university students were more politically liberal than the general populations of their countries. 145 Appendix B – Regression analysis of ‘normative constrained optimizing’ and asymmetric protective motives for private protection Based on the assumption that each observed allocation of less than seven tokens to theft reflects the decision-maker’s operative moral constraint, we calculated the individual’s expected earnings-maximizing allocation to private protection by individual and period in the NCP and VCP treatments on the (strong) assumption that he or she correctly anticipated the average amount of theft in which others would engage in each period (presumably based on previous observations). † We then estimated equations in which this “optimal allocation to private protection” is included in a regression model of individual period-specific expenditure on private protection that also includes individual and period fixed effects. Table III. B1: Allocations to private protection Dependent variable: allocations to private protection NCP VCP NCP & VCP (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Optimal allocation 0.794*** 0.780*** 0.772*** 0.760*** 0.744*** 0.781*** 0.770*** (0.077) (0.074) (0.083) (0.081) (0.082) (0.057) (0.055) Loss through theft in t-1 -0.002* 0.001 0.001 -0.001 (0.001) (0.002) (0.002) (0.001) Loss through theft in t-2 0.001 0.003** 0.003** 0.001 (0.001) (0.001) (0.001) (0.001) Accumulated profit thru t-1 0.000 0.000 0.000 0.000 (0.001) (0.001) (0.001) (0.001) Collective Protection -1.335* (0.740) Gains from theft in t-1 0.010*** 0.010*** 0.009*** 0.009*** (0.002) (0.003) (0.003) (0.002) Accumulated gains from theft thru t-1 0.001* 0.000 0.000 0.001* (0.000) (0.001) (0.001) (0.000) Constant 2.883*** 2.244*** 1.687* 2.330*** 1.741*** 1.444 1.704 2.759*** 2.657*** 2.576*** (0.187) (0.207) (0.933) (0.207) (0.255) -(1.369) (1.349) (0.092) (0.089) (0.155) Individual FE Y Y Y Y Y Y Y Y Y Y Round FE Y Y Y Y Y Y Y Y Y Y Observations 3,960 3,960 3,960 4,070 4,070 4,070 4,070 8,030 8,030 8,030 R-squared 0.005 0.216 0.234 0.009 0.134 0.146 0.149 0.003 0.167 0.18 Notes: Heteroskedasticity-robust standard errors clustered at the group level are shown in parentheses. *** p<0.01, ** p<0.05, * p<0.1 † Calculations done for the VCP treatment deduct the individual’s allocation (if any) to collective protection to determine that number of tokens available for production or private protection. Estimates assuming all tokens not used for theft to be available for either production or private protection give essentially the same results, since allocations to collective protection are usually small. 146 Our baseline specification only includes the intercept, individual and period fixed effects. The results are shown in columns (1), (4) and (8). Next, we add our measure of “optimal allocation to private protection,” which yields the estimates shown in columns (2), (5) and (9). With wealth accumulations and the proportions of those accumulations gained from theft coming to vary over time, individuals might reasonably anticipate (arguably because of retaliation) being asymmetrically targeted for theft in some periods. In the remaining regressions, we add controls for a subject’s accumulated wealth tokens, accumulated gains from theft, and gain from theft in the most recent period. We also control for the possibility that recent losses to theft increase an individual’s propensity to invest in private protection. Finally, for the VCP treatment, we include a specification that controls for the level of collective protection in the group during the period in question. 147 Appendix C – Measures of crime incidence, perception of safety and governance The first two columns in Table III. C 1 show the share of respondents who answered “yes” to each of the two questions we considered in order to gauge the frequency of property crimes in the countries where we conducted the experiments. As mentioned in the text, the source of the data is the United Nations’ International Crime Victim Survey (ICVS). We used information from years 2000 in the case of Mongolia (N=944) and South Korea (N=2,043), 2004 for Mexico (N=1,992) and the U.S. (N=2,011), and 2005 for Austria (N=2,004). The third column presents our measure for the incidence of crime, defined as the share of respondents who replied positively to either question (this is the measure employed in Figure III. 4.a-c). Table III. C 1: Incidence of crime Share (%) of respondents who answered "yes" to the question: Over the past 5 years, has Over the past 5 years, did anyone anyone taken something actually get into your house or from you, by using force, or flat without permission and steal threatening you? Or did or try to steal something? anyone try to do so? Incidence of crime Austria 6.14 2.00 7.88 Mexico 10.89 9.54 18.72 Mongolia 27.21 10.85 34.04 South Korea 15.02 1.42 15.91 U.S.A. 5.99 3.33 8.35 The first two columns in Table III. C 2 present the shares of respondents who reported in the ICVS feeling unsafe, as judged by their perceptions of safety when walking in the dark or the chances that their homes get broken into over the course of the following year. The third column shows the composite measure 148 of safety perceptions obtained by applying factor analysis on the previous two metrics (this is the index used in Figure III. 5.a-c). As explained in the text, the Government Index was constructed from three variables— government effectiveness, rule of law, and control of corruption—included in a dataset of governance measures assembled by the World Bank. Government effectiveness is meant to capture “perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies.” Rule of law represents “perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence.” And control of corruption embodies “perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests.”‡ Table III. C 2: Perceptions of safety Share (%) of respondents Share (%) of respondents who who report feeling “a bit reckon it is “likely” or “very unsafe” or “very unsafe” likely” that over the next twelve when walking alone after months someone will try to break dark into their homes Perception of safety Austria 21.7 27.74 -0.08 Mexico 34.59 37.00 0.17 Mongolia 53.81 20.87 0.19 South Korea 21.93 31.03 0.02 U.S.A. 20.07 14.92 -0.15 Each of these governance measures, shown in the first three columns of Table III. C 3, is expressed in units ranging from -2.5 to 2.5, with higher values denoting better governance ‡ For more information on the measures, see http://info.worldbank.org/governance/wgi/index.asp 149 outcomes. We averaged each country’s score in each dimension across the 2000-2009 period and computed via factor analysis the Governance Index, which appears in the last column (this is the measure used in Figure III. 6). Table III. C 3: Governance index Government effectiveness Rule of law Control of corruption Governance Index Austria 1.78 1.83 1.94 1.90 Mexico 0.17 -0.42 -0.20 -0.14 Mongolia -0.46 -0.16 -0.42 -0.37 South Korea 1.02 0.90 0.47 0.78 U.S.A. 1.60 1.53 1.53 1.59 Appendix D – Instructions and practice scripts§ D.1 NCP Instructions General Information This is an experiment aimed at studying decision-making while interacting with other individuals. During the experiment, you will be earning money in the form of wealth tokens . At the end of the experiment you will be paid in cash in real dollars (1 wealth token = 1.4¢). The amount you will earn will depend on your and others’ decisions. Please read and listen carefully to make sure you understand the decision process. At the end of the instructions you will have a chance to ask questions. The experiment will conclude with a brief questionnaire. Your Group At the beginning of the experiment you will be randomly assigned to a group consisting of yourself and four others. Each member of the group will be randomly assigned a subject number (denoted Sub 0, Sub 1,…, Sub 4) which remains fixed throughout the experiment. You will interact exclusively with the people in your group of five throughout the entire experiment. All decisions are made anonymously, so no participant knows the identities of the other decision makers, nor will you ever be informed who was in your group. Payments are anonymous and will be made in cash at the end of the session. § Instructions in German, Korean, Mongolian and Spanish are available from the authors upon request. 150 Experiment Structure The experiment consists of 24 rounds, organized in 6 sets of 4 rounds. Between each set of four rounds, there will be a brief pause. In total, we expect the experiment to last no more than two hours, including these instructions and practice rounds. Communication and questions Communication is not allowed at any time during the experiment. If you have any questions, please raise your hand and we will come to assist you. Do not hesitate to call on us. Other kinds of tokens During the experiment, you will have the chance to use two types of tokens. Only wealth tokens will be converted into cash at the end of the experiment. Effort tokens will also play a part, but they only have value insofar as they help you to earn or to conserve wealth tokens. Instructions At the beginning of the first round, each member of the group (yourself included) will be endowed with 100 wealth tokens . Each round, each of you will receive 10 effort tokens . Effort tokens have no money value, but they can be used to help you earn or conserve wealth tokens. In every round, you have to allocate your effort tokens among three alternatives: # Effort # Wealth Tokens Tokens Produced 1 15 2 28 3 39 4 48 5 55 151 6 60  Produce new wealth tokens . The table on the right shows the relation 7 64 between effort tokens (the input) and 8 67 wealth tokens (the output). For instance, 9 69 if you allocate 5 effort tokens to 10 70 production, you produce 55 wealth tokens which are added to your accumulation.  Steal others’ wealth tokens . You can assign effort tokens to stealing from any other member(s) of the group, such that for every effort token you direct to stealing from, say, member 2, you will have the chance to steal 10 wealth tokens from him/her. The likelihood that your theft attempts succeed depends upon the other member’s total degree of security, as described below.  Protect your own wealth tokens from being stolen . For every effort token you assign to protection, the probability that you will keep your wealth tokens in the event that someone attempts to steal them rises by 10 percentage points. For example, if you put 4 effort tokens into protection, the likelihood that an attempt to steal wealth tokens from you will fail becomes 4 X 10 = 40% (in other words, the likelihood that the attempted theft succeeds is only 60%). The example below illustrates the stealing mechanism as well as the determination of chances. Example: Member 2 puts 3 effort tokens into protection and member 3 puts 1 effort token into protection. Hence, member 2’s degree of protection is 3 X 10 = 30%, while 3’s level of protection is 1 X 10 = 10%. If you direct 2 effort tokens to an attempt to steal from member 2 and 1 effort token to attempting to steal from member 3, 20 of 2’s wealth tokens and/or 10 of 3’s wealth tokens may be transferred to you. The likelihood of your theft from 2 succeeding is 100 – 30 = 70%, and the likelihood of your theft from 3 succeeding is 100 – 10 = 90%. The computer will ultimately determine according to the aforementioned probability of 70% whether 0 or 20 wealth tokens are transferred to you from member 2 (note: there’s a 70% chance of 20 tokens being transferred, a 30% chance of 0 tokens being transferred, and no chance of an intermediate amount being transferred). Likewise, the computer will independently decide using the probability 90% whether 0 or 10 wealth tokens are transferred to you from member 3. If a theft is successful, the wealth tokens are deducted from 2’s and/or 3’s accumulation and are added to yours. 152 Exception: If the total number of wealth tokens that other members would successfully steal from a particular group member, say member 2, exceeds 2’s existing accumulation, then the computer will adjust the size of the transfers, since member 2 cannot end up with a negative number of wealth tokens. For example, suppose members 0, 1, 3 and 4 each direct 3 effort tokens to attempting to steal from Member 2, and that none of the theft attempts is prevented by the action of Member 2’s degree of protection, so that a total of 3 X 10 X 4 = 120 wealth tokens would be taken from member 2. And suppose that member 2 has only 100 wealth tokens at this point. Then members 0, 1, 3 and 4 would each receive only 100 / 4 = 25 rather than 30 wealth tokens. The decision of each to direct 3 effort tokens against member 2 is nevertheless irrevocable; each would have spent 3 effort tokens, though gaining only 25 rather than 30 wealth tokens from it. At the end of each round, the total number of wealth tokens will be computed according to the following formula: Wealth tokens = Wealth tokens held at the beginning of the round + New wealth tokens produced + Wealth tokens you stole from other members – Wealth tokens other members stole from you At the end of each round you will learn the statistics of your performance in that round as well as the cumulative statistics through that round. That is, you will find out: (i) the number of wealth tokens produced; (ii) the total number of wealth tokens you sought to steal from others and the number of wealth tokens that you successfully stole; (iii) the number of wealth tokens other group members sought to steal from you and the number of wealth tokens they successfully stole. (Note that the information on (iii) is given as an aggregate; you won’t be told which particular group members attempted or succeeded to steal from you.) These amounts will be added/subtracted from the number of wealth tokens that you started the round with, according to the formula described above. You will also learn about the total accumulation of wealth tokens in the hands of each other group member through that round, and the number of wealth tokens that each group member has thus far obtained by production, the number each has thus far obtained by stealing from others, and the number each has thus far lost by means of theft. This information will be available to you anytime in the following round as you make your allocation decisions. Payoffs Your earnings from this experiment will be the $5 that is guaranteed to you simply for participating, plus 1.4¢ for every wealth token that you have accumulated by the end of the 24 rounds. Notice that your 153 earnings do not depend on how much you accumulate in comparison to others, only on how much you accumulate. To make sure that you understand how the different choices operate in the experiment, we’ll now provide some examples. Examples of protection possibilities Example 1: Member 1 allocates 5 effort tokens to protection, member 2 allocates 3 effort tokens to protection, and members 3, 4 and 5 each use 1 effort token for protection. Their corresponding levels of security are 50%, 30%, 10%, 10% and 10%, respectively. Example 2: Group members use various numbers of effort tokens for protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 60%, 40%, 30%, 10% and 0%, respectively. Examples of the use of effort tokens and payoffs Example 1: You begin with 100 wealth tokens and each round you use all 10 effort tokens to produce wealth tokens. No group member ever attempts to steal wealth tokens from you. You earn 70 wealth tokens each round, accumulating a total of 100 + (70 X 24) = 1780 wealth tokens. Your earnings in dollars would be $5 + (1.4¢ X 1780) = $29.92 Example 2: You begin with 100 wealth tokens and each round you allocate 2 effort tokens to protection and 8 effort tokens to stealing from others. You can steal up to 80 wealth tokens, and there is an 80% (= 100 – (2 X 10)) chance that any given attempt by others to steal wealth tokens from you will succeed. Your maximum accumulation could be 100 + (80 X 24) = 2020 wealth tokens, but you may earn less than this, possibly much less, if others successfully steal wealth tokens from you and/or if others use effort tokens to provide some security for their accumulations. Using the maximum estimate of 2020 wealth tokens, your accumulated earnings in dollars would be $5 + (1.4¢ X 2020) = $33.28 Example 3: You begin with 100 wealth tokens and each round you and others in your group use four effort tokens for production, two effort tokens for protection, and four effort tokens for stealing, assigning one token to stealing from each other group member. You produce 48 wealth tokens, and your protection level each period is 20%. The other members of your group each attempt to steal 10 wealth tokens from you and succeed in 80% of attempts, hence reducing your wealth token accumulation by 4 X 10 X 0.8 = 32 wealth tokens per period, on average. Your four attempts to steal 10 wealth tokens from other members succeed on 80% of attempts, thus adding 4 X 10 X 0.8 = 32 wealth tokens to your accumulation per period, on average. Your wealth token accumulation thus rises by an average total of 48 per period, earning you a total of 100 + (48 X 24) = 1252 wealth tokens, for earnings of $5 + (1.4¢ X 1252) = $22.53 154 Example 4: You begin with 100 wealth tokens and each round you put 3 effort tokens into protection and assign the remaining 7 effort tokens to production. Suppose other members also put 3 effort tokens into protection each period. The likelihood that a theft attempt will succeed is 100 – (10 X 3) = 70%. Your 7 effort tokens produce 64 wealth tokens for you each round. You can accumulate up to 100 + (64 X 24) = 1636 wealth tokens, but you may earn less if others steal tokens from you. Using the maximum estimate of 1636 wealth tokens, your total earnings would be $5 + (1.4¢ X 1636) = $27.90. Note that the behavior does not change across rounds in these examples, but this is just for the sake of making these illustrations easy to understand. In fact, your strategy can change over time. As can be seen, there are an almost infinite number of possible outcomes, depending on your decisions and the decisions of others in your group. After questions are answered and we go through two practice rounds that don’t affect your earnings, you will engage in the first four periods of the experiment. Any questions? Practice Scripts Before the real decision-making begins, we’re going to go through two practice rounds the purpose of which is to familiarize you with the way that you enter your choices on your computer screen, the order of choice, and the information you get back after each decision. The earnings shown on your screen for these practice rounds are only illustrative, reflecting decisions I’ll be asking you to enter. They have no effect on your real earnings in the experiment. Also, the participants with whom you’ll interact in the practice rounds are not the ones in your group for the real decision periods. Please follow our instructions as closely as possible and do not click any buttons until told to. The first screen you’ll see tells you that you and each other person in your group has 100 wealth tokens at the beginning of the experiment. Please click next. The next screen you’ll see in each round tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. The next screen is where you have to enter your allocation choices to production, protection and theft. You have to click on at least some of the boxes below the lines saying “production tokens,” “protection tokens” and “theft tokens”, until you finish allocating all your effort tokens. 155 Please use your effort tokens as follows (and do not click submit until we tell you to):  Assign 5 effort tokens to the production of new wealth tokens and 3 to the protection of your existing wealth tokens.  Randomly assign your remaining 2 effort tokens to stealing from other members of your group. Before leaving this screen, please notice the following: (a) you can click on the button labeled “Stats” at the top to learn the number of wealth tokens held by the others in your group; (b) when you enter a number of production tokens, you’ll see immediately below that box the number of Wealth tokens you’ll produce if you stick with that choice; (c) when you enter a number of protection tokens, you’ll see below that box the total level of security of your wealth token accumulation. Before clicking on the submit button, please also notice that it is possible to reconsider and to change your allocations at any time until you hit submit. Simply delete any entry you want to change and enter a new value. Also note that you don’t need to enter a value into each box. If you enter no value under production tokens, for instance, the computer will understand your production tokens choice to be 0. The same applies to protection and theft tokens. You can always choose 0 in any case. If the number of effort tokens you use for the round does not sum to 10 in total, you will receive an error message and will have to change entries until all of your effort tokens have been used. If you are ready, click on Submit now. Note that if all participants haven’t yet made their decisions and clicked submit, you’ll see a screen saying “Waiting for Others”. There is no possibility of seeing what others decide first and then making your own decision. When everyone has submitted their decisions you’ll see a “Round Performance Summary.” Please read it and raise your hand if you have any questions about this screen. When you’re done, please click on Next. The next screen gives you information about the wealth token accumulations of each member in your group. Notice that there are two boxes: (1) the upper box shows the accumulations of each group member broken up into gains and losses through production and theft through the period that just ended; (2) the lower box shows the accumulations of each group member broken up into gains and losses through production and theft in the period that just ended. Please click on Next Round. We’ll now run through a second and final practice round. Remember, the practice rounds do not count towards determining your earnings. Again, please follow our instructions for this one last practice round. 156 Notice that your first screen of the round tells you your updated accumulation of wealth tokens. Note that you begin every round with the same number of effort tokens, 10. Please click next. Again, the next screen is for indicating your allocation choices to production, protection and theft. Notice that you can also click on the Stats button here to view again the accumulations of each group member and the break-down according to production, gains from theft and losses due to theft. If you want the stats to disappear, you can click Hide, otherwise they will go away automatically when you click submit, but you can view them again at the next decision stage. At the second allocation screen, please allocate your effort tokens as follows:  6 to production,  1 to protection, and  For your theft choices, please open again the Stats window. In the real experiment, you may want to use the information in that window to help you decide whom you try to steal from. For this practice round, as an illustration, please assign 2 effort tokens to whichever member of your group (excluding yourself) has the smallest accumulation of wealth tokens after Practice 1 (check the stats window). If more than one member is tied for smallest accumulation, choose one randomly to assign your effort tokens to. Finally, assign 1 effort token to whichever member of your group has the 2nd smallest accumulation of wealth tokens. When everyone has submitted their decisions you’ll see the Round Performance Summary, as before. When you’re ready, please click on Next. Finally, as before, the next screen gives you information about the wealth token accumulations of each group member broken up into gains and losses through production and theft. Before we begin the rounds that count toward your real earnings, please note that the time it will take to complete all 24 rounds depends on all of your rates of progress. No individual can move forward until all participants make the corresponding decisions at each stage of the process, which includes, at the end of each period, that you finish viewing your round performance summary and click “next.” Please focus on the task and click the appropriate “submit” and “next” buttons as soon as you are ready to do so, so that the process can proceed for all in a timely fashion. To help make sure that we finish before _ _ _, we’ll remind you to continue if we see that progress stalls. (We can track on our monitor whether actions have been taken but not which actions they were, that information is stored only for later analysis.) O.k.? Please begin. 157 D.2 VCP Instructions General Information This is an experiment aimed at studying decision-making while interacting with other individuals. During the experiment, you will be earning money in the form of wealth tokens . At the end of the experiment you will be paid in cash in real dollars (1 wealth token = 1.4¢). The amount you will earn will depend on your and others’ decisions. Please read and listen carefully to make sure you understand the decision process. At the end of the instructions you will have a chance to ask questions. The experiment will conclude with a brief questionnaire. Your Group At the beginning of the experiment you will be randomly assigned to a group consisting of yourself and four others. Each member of the group will be randomly assigned a subject number (denoted Sub 0, Sub 1,…, Sub 4) which remains fixed throughout the experiment. You will interact exclusively with the people in your group of five throughout the entire experiment. All decisions are made anonymously, so no participant knows the identities of the other decision makers, nor will you ever be informed who was in your group. Payments are anonymous and will be made in cash at the end of the session. Experiment Structure The experiment consists of 24 rounds, organized in 6 sets of 4 rounds. Between each set of four rounds, there will be a brief pause. In total, we expect the experiment to last no more than two hours, including these instructions and practice rounds. Communication and questions Communication is not allowed at any time during the experiment. If you have any questions, please raise your hand and we will come to assist you. Do not hesitate to call on us. Other kinds of tokens During the experiment, you will have the chance to use two types of tokens. Only wealth tokens will be converted into cash at the end of the experiment. Effort tokens will also play a part, but they only have value insofar as they help you to earn or to conserve wealth tokens. 158 Instructions At the beginning of the first round, each member of the group (yourself included) will be endowed with 100 wealth tokens . Each round, each of you will receive 10 effort tokens . Effort tokens have no money value, but they can be used to help you earn or conserve wealth tokens. There are three activities you can perform using your effort tokens: - Produce more wealth tokens - Steal others’ wealth tokens - Protect your own wealth tokens from being stolen in one or both of two ways: collective protection , which adds to the security of all members’ wealth tokens, and private protection , which adds to the security of the wealth tokens of the person who pays for it only We next explain the order of decision-making and provide more details about the activities of production, theft, and protection. In every round, decisions are made in two stages:  Stage 1: you have to decide how many of your 10 effort tokens you contribute to collective protection , whereby the wealth tokens of every member are equally protected if another member seeks to steal them. For every effort token that any member of the group puts into collective protection, the probability that you (and each person in your group) will keep your wealth tokens if someone attempts to steal them will rise by 6 percentage points on a scale of 0- 100%. The highest level of security that can be attained through collective protection is reached if the sum of everyone’s contributions is 12 effort tokens (or more). In that case, there is a 12 X 6 = 72% chance that a member’s wealth tokens will remain with them in the event of an attempt to steal them. Greater contributions beyond 12 effort tokens do not increase the probability that you will keep your wealth tokens. At the end of Stage 1, you will learn the total number of effort tokens put into collective protection and the resulting degree of security at which theft is prevented, but you will not learn how many tokens each individual put into collective protection.  Stage 2: In this stage, you have to allocate your remaining effort tokens among three alternatives: private protection, production and theft.  Private protection is meant to provide further protection to your own wealth tokens only. For every effort token you assign to private protection, the probability that you will keep your wealth tokens in the event that someone attempts to steal them rises by 10 percentage points, which will be added to the level of security already achieved through collective protection. For example, if collective protection is 54% (i.e., 9 effort tokens were contributed in total) and you put 4 effort tokens into private protection, the likelihood that an attempt to steal wealth tokens from you will fail increases to 54 + (4 X 10) = 94% (in other words, the likelihood that the attempted theft succeeds is only 6%). Of course, you cannot increase the likelihood of failure beyond 100%, so at some point, 159 additional effort tokens lose their effect. In our example, if you put 5 effort tokens into private protection, the 5th effort token raises the likelihood of a theft attempt failing by 6% only, to 100%, and a 6th effort token would have no effect. Note that effort tokens put into collective protection benefit all members equally, and that everyone knows this degree of security in each round; however, private protection benefits only you, and only you know how many effort tokens you use for private protection, thereby your own total level of protection.  Production of new wealth tokens. # Effort # Wealth Tokens The table on the right shows the relation Tokens Produced between effort tokens (the input) and 1 15 wealth tokens (the output). For instance, 2 28 if you allocate 5 effort tokens to 3 39 production, you produce 55 wealth tokens 4 48 which are added to your accumulation. 5 55 6 60 7 64 8 67 9 69 10 70  Theft enables any member to steal wealth tokens from any other member(s) of the group with a likelihood of success that depends upon the other members’ total degree of security, as already described. You can assign effort tokens to stealing from any of the other members, such that for every effort token you direct to stealing from, say, member 2, you will have the chance to steal 10 wealth tokens from him/her. The example below illustrates the stealing mechanism as well as the determination of chances. Example: Group members put a total of 5 effort tokens into collective protection, so the degree of security provided through collective protection is 5 X 6% = 30%. Member 2 puts 3 effort tokens into private protection and member 3 puts no effort tokens into private protection. Hence, 2’s total degree of protection is 30 + (3 X 10) = 60%, while 3’s total level of protection is 30 + (0 X 10) = 30%. If you direct 2 effort tokens to an attempt to steal from member 2 and 1 effort token to attempting to steal from member 3, 20 of 2’s wealth tokens and/or 10 of 3’s wealth tokens may be transferred to you. The likelihood of your theft from 2 succeeding is 100 – 60 = 40%, and the likelihood of your theft from 3 succeeding is 100 – 30 = 70%. The computer will ultimately determine according to the aforementioned probability of 40% whether 0 or 20 wealth tokens are transferred to you from member 2 (note: there’s a 40% chance of 20 tokens being transferred, a 60% chance of 0 tokens being transferred, and no chance of an intermediate amount being 160 transferred). Likewise, the computer will independently decide using the probability 70% whether 0 or 10 wealth tokens are transferred to you from member 3. If a theft is successful, the wealth tokens are deducted from 2’s and/or 3’s accumulation and are added to yours. Exception: If the total number of wealth tokens that other members would successfully steal from a particular group member, say member 2, exceeds 2’s existing accumulation, then the computer will adjust the size of the transfers, since member 2 cannot end up with a negative number of wealth tokens. For example, suppose that members 0, 1, 3 and 4 each direct 3 effort tokens to attempting to steal from member 2, and that none of the theft attempts is prevented by the action of 2’s degree of protection, so that a total of 3 X 10 X 4 = 120 wealth tokens would be taken from member 2. And suppose that member 2 has only 100 wealth tokens at this point. Then members 0, 1, 3 and 4 would each receive only 100 / 4 = 25 rather than 30 wealth tokens. The decision of each to direct 3 effort tokens against member 2 is nevertheless irrevocable; each would have spent 3 effort tokens, though gaining only 25 rather than 30 wealth tokens from it. At the end of each round, the total number of wealth tokens will be computed according to the following formula: Wealth tokens = Wealth tokens held at the beginning of the round + New wealth tokens produced + Wealth tokens you stole from other members – Wealth tokens other members stole from you At the end of each round you will learn the statistics of your performance in that round as well as the cumulative statistics through that round. That is, you will find out: (i) the number of wealth tokens produced; (ii) the total number of wealth tokens you sought to steal from others and the number of wealth tokens that you successfully stole; (iii) the number of wealth tokens other group members sought to steal from you and the number of wealth tokens they successfully stole. (Note that the information on (iii) is given as an aggregate; you won’t be told which particular group members attempted or succeeded to steal from you.) These amounts will be added/subtracted from the number of wealth tokens that you started the round with, according to the formula described above. You will also learn about the total accumulation of wealth tokens in the hands of each other group member through that round, and the number of wealth tokens that each group member has thus far obtained by production, the number each has thus far obtained by stealing from others, and the number each has thus far lost by means of theft. This information will be available to you anytime in the following round as you make your allocation decisions. Payoffs 161 Your earnings from this experiment will be the $5 that is guaranteed to you simply for participating, plus 1.4¢ for every wealth token that you have accumulated by the end of the 24 rounds. Notice that your earnings do not depend on how much you accumulate in comparison to others, only on how much you accumulate. To make sure that you understand how collective and private protection work in the experiment, we’ll now provide some examples. Examples of protection possibilities Example 1: Three members each contribute two effort tokens and two members each contribute three effort tokens to collective protection, for a total of 12, giving a protection level of 72% to everyone. Nothing is spent on private protection, so every member equally enjoys a 72% level of protection. Example 2: No member contributes any effort tokens to collective protection. Each member uses three effort tokens for private protection. They each achieve a security level of 30%. Example 3: Two members each contribute two effort tokens to collective protection, while the other members contribute three, one, and zero effort tokens, respectively. In total, 8 effort tokens are contributed to collective protection, which provides a security level of 48% to everyone. Members individually use various numbers of effort tokens for private protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 100%, 88%, 78%, 58% and 48%, respectively. Examples of the use of effort tokens and payoffs Example 1: You begin with 100 wealth tokens and each round you use all 10 effort tokens to produce wealth tokens. No one ever contributes to collective protection. No group member ever attempts to steal wealth tokens from you. You earn 70 wealth tokens each round, accumulating a total of 100 + (70 X 24) = 1780 wealth tokens. Your earnings in dollars would be $5 + (1.4¢ X 1780) = $29.92 Example 2: You begin with 100 wealth tokens and each round you allocate 2 effort tokens to private protection and 8 effort tokens to stealing from others. No one ever contributes to collective protection. You can steal up to 80 wealth tokens, and there is an 80% (= 100 – (2 X 10)) chance that any given attempt by others to steal wealth tokens from you will succeed. Your maximum accumulation could be 100 + (80 X 24) = 2020 wealth tokens, but you may earn less than this, possibly much less, if others successfully steal wealth tokens from you and/or if others use effort tokens to provide some security for their accumulations. Using the maximum estimate of 2020 wealth tokens, your accumulated earnings in dollars would be $5 + (1.4¢ X 2020) = $33.28 162 Example 3: You begin with 100 wealth tokens and each round you and others in your group use four effort tokens for production, two effort tokens for private protection, and four effort tokens for stealing, assigning one token to stealing from each other group member. No one ever contributes to collective protection. You produce 48 wealth tokens, and your protection level each period is 20%. The other members of your group each attempt to steal 10 wealth tokens from you and succeed in 80% of attempts, hence reducing your wealth token accumulation by 4 X 10 X 0.8 = 32 wealth tokens per period, on average. Your four attempts to steal 10 wealth tokens from other members succeed on 80% of attempts, thus adding 4 X 10 X 0.8 = 32 wealth tokens to your accumulation per period, on average. Your wealth token accumulation thus rises by an average total of 48 per period, earning you a total of 100 + (48 X 24) = 1252 wealth tokens, for earnings of $5 + (1.4¢ X 1252) = $22.53 Example 4: You begin with 100 wealth tokens and each round you put 3 effort tokens into collective protection and assign the remaining 7 effort tokens to production. Suppose other members also put 3 effort tokens into collective protection each period. With 3 X 5 = 15 tokens in collective protection, the likelihood that a theft attempt will succeed is 100 – (12 X 6) = 100 – 72 = 28%. Your 7 effort tokens produce 64 wealth tokens for you each round. You can accumulate up to 100 + (64 X 24) = 1636 wealth tokens, but you may earn less if others attempt to steal tokens from you, although in this case they have a 28% chance of success each time (versus the 80% chance of success in the previous example). Using the maximum estimate of 1636 wealth tokens, your total earnings would be $5 + (1.4¢ X 1636) = $27.90 Note that the behavior does not change across rounds in these examples, but this is just for the sake of making these illustrations easy to understand. In fact, your strategy can change over time. As can be seen, there are an almost infinite number of possible outcomes, depending on your decisions and the decisions of others in your group. After questions are answered and we go through two practice rounds that don’t affect your earnings, you will engage in the first four periods of the experiment. Any questions? Practice Scripts Before the real decision-making begins, we’re going to go through two practice rounds the purpose of which is to familiarize you with the way that you enter your choices on your computer screen, the order of choice, and the information you get back after each decision. The earnings shown on your screen for these practice rounds are only illustrative, reflecting decisions I’ll be asking you to enter. They have no effect on your real earnings in the experiment. Also, the participants with whom you’ll interact in the practice rounds are not the ones in your group for the real decision periods. Please follow our instructions as closely as possible and do not click any buttons until told to. 163 The first screen you’ll see tells you that you and each other person in your group has 100 wealth tokens at the beginning of the experiment. Please click next. The next screen you’ll see in each round tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. The next screen is the first in which you have to enter a decision. You should enter the number of effort tokens you want to contribute to collective protection under the heading “Put in Collective Protection.” In this first practice round, please allocate the following number of effort tokens to the group account if the last digit of your seat number is even and < 6: 1 effort token odd and < 6: 2 effort tokens even and > 5: 3 effort tokens odd and > 5: 4 effort tokens Notice that in the lower part of the window you can see the number of effort tokens that remain in your account, which automatically updates as you enter your contribution into collective protection. Now click Submit. Note that if all participants haven’t yet made their decisions and clicked submit, you’ll see a screen saying “Waiting for Others”. There is no possibility of seeing what others decide first and then making your own decision. The next screen shows the total number of effort tokens all group members have contributed to collective protection and the consequent level of collective protection. Click continue. The next screen you’ll see is the decision screen for allocating the remainder of your effort tokens for this period. Notice that the number of effort tokens you have left appears in the top of the window, and this number is updated as you allocate these effort tokens to production, private protection and theft. You have to click on at least some of the boxes below the lines saying “production tokens,” “private protection tokens” and “theft tokens”, until you finish allocating your effort tokens. Please use your remaining effort tokens as follows (and do not click submit until we tell you to):  Assign 3 effort tokens to production and 3 to private protection.  If you still have any effort tokens left, use them to try stealing from other group members. 164  Assign no more than one effort token to stealing from an individual and if you have multiple effort tokens assign them randomly, for instance if you have 2, assign them to any two decision-makers in your group. Before leaving this screen, please notice the following: (a) you can click on the button labeled “Stats” at the top to learn the number of wealth tokens held by the others in your group; (b) when you enter a number of production tokens, you’ll see immediately below that box the number of Wealth tokens you’ll produce if you stick with that choice; (c) when you enter a number of private protection tokens, you’ll see below that box the total level of security of your wealth token accumulation. Before clicking on the submit button, please also notice that it is possible to reconsider and to change your allocations at any time until you hit submit. Simply delete any entry you want to change and enter a new value. Also note that you don’t need to enter a value into each box. If you enter no value under production tokens, for instance, the computer will understand your production tokens choice to be 0. The same applies to private protection and theft tokens. You can always choose 0 in any case. If the number of effort tokens you use for the round does not sum to 10 in total, you will receive an error message and will have to change entries until all of your effort tokens have been used. If you are ready, click on Submit now. When everyone has submitted their decisions you’ll see a “Round Performance Summary.” Please read it and raise your hand if you have any questions about this screen. When you’re done, please click on Next. The next screen gives you information about the wealth token accumulations of each member in your group. Notice that there are two boxes: (1) the upper box shows the accumulations of each group member broken up into gains and losses through production and theft through the period that just ended; (2) the lower box shows the accumulations of each group member broken up into gains and losses through production and theft in the period that just ended. Please click on Next Round. We’ll now run through a second and final practice round. Remember, the practice rounds do not count towards determining your earnings. Again, please follow our instructions for this one last practice round. Notice that your first screen of the round tells you your updated accumulation of wealth tokens. Note that you begin every round with the same number of effort tokens, 10. Please click next. Again, the next screen is for indicating how many effort tokens you want to put in the group account. Notice that you can also click on the Stats button here to view again the accumulations of each group member and the break-down according to production, gains from theft and losses due to theft. If you want the stats to disappear, you can click Hide, otherwise they will go away automatically when you click submit, but you can view them again at the next decision stage. 165 For this last practice round, please allocate the following number of effort tokens to collective protection if the last digit of your seat number is even and < 6: 4 effort token odd and < 6: 3 effort tokens even and > 5: 2 effort tokens odd and > 5: 1 effort tokens Now click Submit. At the second allocation screen, please allocate your effort tokens as follows:  3 to production,  1 to private protection, and  For your theft choices, please open again the Stats window. In the real experiment, you may want to use the information in that window to help you decide whom you try to steal from. For this practice round, as an illustration, please assign 2 effort tokens to whichever member of your group (excluding yourself) has the smallest accumulation of wealth tokens after Practice 1 (check the stats window). If more than one member is tied for smallest accumulation, choose one randomly to assign your effort tokens to. If you have more effort tokens left, assign 1 token to whichever member of your group has the 2nd smallest accumulation of wealth tokens. If you have more effort tokens left, assign them as you will among the other group members you haven’t tried to steal from yet. When everyone has submitted their decisions you’ll see the Round Performance Summary, as before. When you’re ready, please click on Next. Finally, as before, the next screen gives you information about the wealth token accumulations of each group member broken up into gains and losses through production and theft. Before we begin the rounds that count toward your real earnings, please note that the time it will take to complete all 24 rounds depends on all of your rates of progress. No individual can move forward until all participants make the corresponding decisions at each stage of the process, which includes, at the end of each period, that you finish viewing your round performance summary and click “next.” Please focus on the task and click the appropriate “submit” and “next” buttons as soon as you are ready to do so, so that the process can proceed for all in a timely fashion. To help make sure that we finish before _ _ _, we’ll remind you to continue if we see that progress stalls. (We can track on our monitor whether actions have been taken but not which actions they were, that information is stored only for later analysis.) 166 O.k.? Please begin. D.3 VOTE Instructions General Information This is an experiment aimed at studying decision-making while interacting with other individuals. During the experiment, you will be earning money in the form of wealth tokens . At the end of the experiment you will be paid in cash in real dollars (1 wealth token = 1.4¢). The amount you will earn will depend on your and others’ decisions. Please read and listen carefully to make sure you understand the decision process. At the end of the instructions you will have a chance to ask questions. The experiment will conclude with a brief questionnaire. Experiment Structure The experiment consists of 24 rounds, organized in 6 sets of 4 rounds. The initial instructions cover the first 4 rounds, which will be followed by further instructions and the remaining 5 sets of 4 rounds. (The first rounds are labeled 1 – 4, the remaining ones are labeled 1 – 20.) In total, we expect the experiment to last no more than two hours, including the instructions and practice rounds. Your Group At the beginning of the experiment you will be randomly assigned to a group consisting of yourself and four others. Each member of the group will be randomly assigned a subject number (denoted Sub 0, Sub 1,…, Sub 4) which remains fixed. You will interact exclusively with the people in your group of five throughout the entire experiment. All decisions are made anonymously, so no participant knows the identities of the other decision-makers, nor will you ever be informed who was in your group. Payments are anonymous and will be made in cash at the end of the session. Communication and questions Communication is not allowed at any time during the experiment. If you have any questions, please raise your hand and we will come to assist you. Do not hesitate to call on us. Other kinds of tokens 167 During the experiment, you will have the chance to use two types of tokens. Only wealth tokens will be converted into cash at the end of the experiment. Effort tokens will also play a part, but they only have value insofar as they help you to earn or to conserve wealth tokens. Instructions At the beginning of the first round, each member of the group (yourself included) will be endowed with 100 wealth tokens . Each round, each of you will receive 10 effort tokens . Effort tokens have no money value, but they can be used to help you earn or conserve wealth tokens. In every round, you have to allocate your effort tokens among three alternatives:  Produce new wealth tokens . # Effort # Wealth Tokens The table on the right shows the relation Tokens Produced between effort tokens (the input) and 1 15 wealth tokens (the output). For instance, 2 28 if you allocate 5 effort tokens to 3 39 production, you produce 55 wealth tokens 4 48 which are added to your accumulation. 5 55 6 60 7 64 8 67 9 69 10 70  Steal others’ wealth tokens . You can assign effort tokens to stealing from any other member(s) of the group, such that for every effort token you direct to stealing from, say, member 2, you will have the chance to steal 10 wealth tokens from him/her. The likelihood that your theft attempts succeed depends upon the other member’s total degree of security, as described below. 168  Protect your own wealth tokens from being stolen . For every effort token you assign to protection, the probability that you will keep your wealth tokens in the event that someone attempts to steal them rises by 10 percentage points. For example, if you put 4 effort tokens into protection, the likelihood that an attempt to steal wealth tokens from you will fail becomes 4 X 10 = 40% (in other words, the likelihood that the attempted theft succeeds is only 60%). The example below illustrates the stealing mechanism as well as the determination of chances. Example: Member 2 puts 3 effort tokens into protection and member 3 puts 1 effort token into protection. Hence, member 2’s degree of protection is 3 X 10 = 30%, while 3’s level of protection is 1 X 10 = 10%. If you direct 2 effort tokens to an attempt to steal from member 2 and 1 effort token to attempting to steal from member 3, 20 of 2’s wealth tokens and/or 10 of 3’s wealth tokens may be transferred to you. The likelihood of your theft from 2 succeeding is 100 – 30 = 70%, and the likelihood of your theft from 3 succeeding is 100 – 10 = 90%. The computer will ultimately determine according to the aforementioned probability of 70% whether 0 or 20 wealth tokens are transferred to you from member 2 (note: there’s a 70% chance of 20 tokens being transferred, a 30% chance of 0 tokens being transferred, and no chance of an intermediate amount being transferred). Likewise, the computer will independently decide using the probability 90% whether 0 or 10 wealth tokens are transferred to you from member 3. If a theft is successful, the wealth tokens are deducted from 2’s and/or 3’s accumulation and are added to yours. Exception: If the total number of wealth tokens that other members would successfully steal from a particular group member, say member 2, exceeds 2’s existing accumulation, then the computer will adjust the size of the transfers, since member 2 cannot end up with a negative number of wealth tokens. For example, suppose members 0, 1, 3 and 4 each direct 3 effort tokens to attempting to steal from Member 2, and that none of the theft attempts is prevented by the action of Member 2’s degree of protection, so that a total of 3 X 10 X 4 = 120 wealth tokens would be taken from member 2. And suppose that member 2 has only 100 wealth tokens at this point. Then members 0, 1, 3 and 4 would each receive only 100 / 4 = 25 rather than 30 wealth tokens. The decision of each to direct 3 effort tokens against member 2 is nevertheless irrevocable; each would have spent 3 effort tokens, though gaining only 25 rather than 30 wealth tokens from it. At the end of each round, the total number of wealth tokens will be computed according to the following formula: Wealth tokens = Wealth tokens held at the beginning of the round + New wealth tokens produced + Wealth tokens you stole from other members – Wealth tokens other members stole from you At the end of each round you will learn the statistics of your performance in that round as well as the cumulative statistics through that round. That is, you will find out: (i) the number of wealth tokens 169 produced; (ii) the total number of wealth tokens you sought to steal from others and the number of wealth tokens that you successfully stole; (iii) the number of wealth tokens other group members sought to steal from you and the number of wealth tokens they successfully stole. (Note that the information on (iii) is given as an aggregate; you won’t be told which particular group members attempted or succeeded to steal from you.) These amounts will be added/subtracted from the number of wealth tokens that you started the round with, according to the formula described above. You will also learn about the total accumulation of wealth tokens in the hands of each other group member through that round, and the number of wealth tokens that each group member has thus far obtained by production, the number each has thus far obtained by stealing from others, and the number each has thus far lost by means of theft. This information will be available to you anytime in the following round as you make your allocation decisions. Payoffs Your earnings from this experiment will be the $5 that is guaranteed to you simply for participating, plus 1.4¢ for every wealth token that you have accumulated by the end of the 24 rounds. Notice that your earnings do not depend on how much you accumulate in comparison to others, only on how much you accumulate. To make sure that you understand how the different choices operate in the experiment, we’ll now provide some examples. Examples of protection possibilities Example 1: Member 1 allocates 5 effort tokens to protection, member 2 allocates 3 effort tokens to protection, and members 3, 4 and 5 each use 1 effort token for protection. Their corresponding levels of security are 50%, 30%, 10%, 10% and 10%, respectively. Example 2: Group members use various numbers of effort tokens for protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 60%, 40%, 30%, 10% and 0%, respectively. Examples of the use of effort tokens and payoffs In the following examples, we illustrate possible behaviors and the earnings these behaviors would lead to if followed for the entire 24 rounds of the experiment. While the set of decisions to be made will change in some respects after the first 4 rounds, enough remains the same so that calculating payoffs on a 24 round basis is a useful way for you to grasp the earnings consequences of a given scenario. 170 Example 1: You begin with 100 wealth tokens and each round you use all 10 effort tokens to produce wealth tokens. No group member ever attempts to steal wealth tokens from you. You earn 70 wealth tokens each round, accumulating a total of 100 + (70 X 24) = 1780 wealth tokens. Your earnings in dollars would be $5 + (1.4¢ X 1780) = $29.92 Example 2: You begin with 100 wealth tokens and each round you allocate 2 effort tokens to protection and 8 effort tokens to stealing from others. You can steal up to 80 wealth tokens, and there is an 80% (= 100 – (2 X 10)) chance that any given attempt by others to steal wealth tokens from you will succeed. Your maximum accumulation could be 100 + (80 X 24) = 2020 wealth tokens, but you may earn less than this, possibly much less, if others successfully steal wealth tokens from you and/or if others use effort tokens to provide some security for their accumulations. Using the maximum estimate of 2020 wealth tokens, your accumulated earnings in dollars would be $5 + (1.4¢ X 2020) = $33.28 Example 3: You begin with 100 wealth tokens and each round you and others in your group use four effort tokens for production, two effort tokens for protection, and four effort tokens for stealing, assigning one token to stealing from each other group member. You produce 48 wealth tokens, and your protection level each period is 20%. The other members of your group each attempt to steal 10 wealth tokens from you and succeed in 80% of attempts, hence reducing your wealth token accumulation by 4 X 10 X 0.8 = 32 wealth tokens per period, on average. Your four attempts to steal 10 wealth tokens from other members succeed on 80% of attempts, thus adding 4 X 10 X 0.8 = 32 wealth tokens to your accumulation per period, on average. Your wealth token accumulation thus rises by an average total of 48 per period, earning you a total of 100 + (48 X 24) = 1252 wealth tokens, for earnings of $5 + (1.4¢ X 1252) = $22.53 Example 4: You begin with 100 wealth tokens and each round you put 3 effort tokens into protection and assign the remaining 7 effort tokens to production. Suppose other members also put 3 effort tokens into protection each period. The likelihood that a theft attempt will succeed is 100 – (10 X 3) = 70%. Your 7 effort tokens produce 64 wealth tokens for you each round. You can accumulate up to 100 + (64 X 24) = 1636 wealth tokens, but you may earn less if others steal tokens from you. Using the maximum estimate of 1636 wealth tokens, your total earnings would be $5 + (1.4¢ X 1636) = $27.90 Note that the behavior does not change across rounds in these examples, but this is just for the sake of making these illustrations easy to understand. In fact, your strategy can change over time. As can be seen, there are an almost infinite number of possible outcomes, depending on your decisions and the decisions of others in your group. After questions are answered and we go through two practice rounds that don’t affect your earnings, you will engage in the first four periods of the experiment. Any questions? 171 Instructions for the remaining 20 rounds The following are the additional instructions that you will need for the remaining 20 rounds of the experiment. You will continue to interact with the same group of participants that you interacted with during the first 4 rounds. As before, you will not be identified to one another other than by the subject number, either during or after the experiment, and there must be no communication. During the remaining rounds, you continue to receive 10 effort tokens at the beginning of each round, which you can use to produce more wealth tokens, to steal others’ wealth tokens, or to protect your wealth tokens from being stolen by others. The computer has recorded your wealth token accumulation so far and will count it towards your final earnings. However, each of you will begin the remaining rounds with a fresh allotment of 100 wealth tokens and you will be shown your earnings in the remainder of the experiment accumulating from this new starting point. In addition to the three uses of effort tokens that are already familiar to you, there is now an additional way in which effort tokens can be used, called collective protection . Collective protection differs from the private protection activity that has been available until now in three ways. First, an effort token allocated to collective protection increases the security of the wealth accumulations of all members of the group, not only that of the person who allocated it. Second, an effort token allocated to collective protection reduces the probability of a successful theft by 6%, not the 10% associated with private protection tokens (but note that this 6% affects all five group members whereas the 10% impact of a token allocated to private protection affects only one group member—you). Third, allocation of tokens to collective protection takes place in a separate, initial stage of the round, and the total level of collective protection is known to all group members when they make their allocation decisions regarding the remaining three activities (production, stealing, and private protection), whereas the level of an individual’s private protection is not known to others when they make those decisions. A further detail about collective protection is that there is a maximum achievable level of collective protection that is reached if group members put a total of 12 or more tokens into this activity. That is, 72% (6% X 12) is the largest amount by which the probability of successful theft can be reduced by collective protection. If group members allocate 13 or more effort tokens to collective protection, the protection level remains at 72% (an attempt at theft succeeds with 28% probability). However, allocations to private protection add to the total protection level of an individual. For example, suppose that 7 tokens are allocated to collective protection, giving a protection level of 6% X 7 = 42%. If you then allocate 4 tokens to private protection, your own protection level is 42% + (10% X 4) = 82%. Of course, your level of protection from theft cannot exceed 100%, so at some point additional tokens allocated to private protection have no effect. For example, if the collective protection level is 72% and you allocate 4 tokens to private protection, the third of your 4 tokens brings your protection level to 100% and the fourth produces no further change. 172 Although allocations to collective protection always take place in a distinct first stage of each round, there are two different ways in which the allocation decisions can be made. At the beginning of each set of 4 rounds, your group decides by majority vote which of these two schemes will be used in those rounds.  Scheme 1 (Decide Individually): at the start of every round, each group member decides independently how many of her/his 10 effort tokens to contribute to collective protection.  Scheme 2 (Decide by Vote): at the start of every round, each group member votes for a number of effort tokens s/he would like all five group members to be required to put into collective protection. The choices of the five of you will be ordered from lowest to highest, and the amount that lies in the middle will be selected. That amount will then be deducted automatically from the ten effort tokens with which each member begins the round. Once the choice between Schemes 1 and 2 has been made by majority vote in your group, it will be in force for the next four periods of the experiment, after which your group will vote again on the scheme that will be used to define the individual contributions to collective protection for the subsequent four rounds. Whereas votes on schemes take place before each set of four rounds, each round itself consists of two stages. In stage 1, each group member either makes her/his contribution to collective protection (if Scheme 1 is in place), or votes for a contribution amount and has the amount decided on deducted automatically (if Scheme 2 is in place). In stage 2, each group member has to allocate her/his remaining effort tokens among the three remaining alternatives of private protection, production and theft. The impact of tokens allocated to each of those three activities remains exactly as in the first four rounds. To make sure that you understand how collective and private protection work in the experiment, we’ll now provide some examples beginning with Scheme 1 (Decide Individually). Examples with individual choice of collective protection (Scheme 1) Example 1. Three members each contribute two effort tokens and two members each contribute three effort tokens to collective protection, for a total of 12, giving a protection level of 72% to everyone. Nothing is spent on private protection, so every member equally enjoys a 72% level of protection. Example 2. No member contributes any effort tokens to collective protection. Each member uses three effort tokens for private protection. They each achieve a security level of 30%. Example 3: Two members each contribute two effort tokens to collective protection, while the other members each contribute three, one, and zero effort tokens, respectively. In total, 8 effort tokens are contributed to collective protection, which provides a security level of 48% to everyone. Members individually use various numbers of effort tokens for private protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 100%, 88%, 78%, 58% and 48%, respectively. 173 Examples with voted choice of collective protection (Scheme 2) Example 1: Subject 0 votes to require that 4 effort tokens be contributed to collective protection; subject 1 votes for 2; subject 2 for 3; subject 3 for 0 and subject 4 for 1. From lowest to highest, we arrange the amounts for which subjects voted as: 0, 1, 2, 3, 4. The amount that lies in the middle is 2. Therefore, every group member will have 2 effort tokens deducted from her/his endowment of 10 effort tokens and assigned to collective protection. The level of collective protection from the 2 X 5 = 10 tokens is calculated as before, i.e., 10 X 6% = 60%. Nothing is spent on private protection, so every member equally enjoys a 60% level of protection. Example 2: Subject 0 votes for requiring 3 effort tokens to be contributed to collective protection; subject 1 votes for 1; subject 2 for 0; subject 3 for 1 and subject 4 for 2. From lowest to highest, we arrange the proposals as follows: 0, 1, 1, 2, 3. Although there are two votes for 1, one of these counts as the middle proposal. Therefore, every group member is obligated to contribute 1 effort token to collective protection, so there is an automatic deduction of 1 effort token from each individual’s endowment of 10 effort tokens. The 5 tokens put into collective protection yield a collective protection level of 5 X 6% = 30%. Members individually use various amounts of effort tokens for private protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 90%, 70%, 60%, 40% and 30%, respectively. Other examples on the determination of contributions to collective protection under Scheme 2: Votes are for 0, 0, 0, 2, 3 Decision: everyone contributes 0 Votes are for 0, 1, 2, 3, 3 Decision: everyone contributes 2 Votes are for 0, 1, 3, 4, 5 Decision: everyone contributes 3 Votes are for 0, 3, 4, 4, 7 Decision: everyone contributes 4 Examples illustrating how the full set of decisions can lead to different overall earnings in the experiment were given in the previous instructions. Are there any questions before we begin the practice rounds? Practice Scripts Practice for first 4 rounds Before the first real set of 4 rounds begins, we’re going to go through two practice rounds the purpose of which is to familiarize you with the way that you enter your choices on your computer screen, the order of choice, and the information you get back after each decision. The earnings shown on your screen for these practice rounds are only illustrative, reflecting decisions I’ll be asking you to enter. They have no effect on your real earnings in the experiment. Also, the participants with whom you’ll interact in the practice rounds are not the ones in your group for the real decision periods. Please follow our instructions as closely as possible and do not click any buttons until told to. 174 The first screen you’ll see tells you that you and each other person in your group has 100 wealth tokens at the beginning of the experiment. Please click next. The next screen you’ll see in each round tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. The next screen is where you have to enter your allocation choices to production, protection and theft. You have to click on at least some of the boxes below the lines saying “production tokens,” “protection tokens” and “theft tokens”, until you finish allocating all your effort tokens. Please use your effort tokens as follows (and do not click submit until we tell you to):  Assign 5 effort tokens to the production of new wealth tokens and 3 to the protection of your existing wealth tokens.  Randomly assign your remaining 2 effort tokens to stealing from other members of your group. Before leaving this screen, please notice the following: (a) you can click on the button labeled “Stats” at the top to learn the number of wealth tokens held by the others in your group; (b) when you enter a number of production tokens, you’ll see immediately below that box the number of Wealth tokens you’ll produce if you stick with that choice; (c) when you enter a number of protection tokens, you’ll see below that box the total level of security of your wealth token accumulation. Before clicking on the submit button, please also notice that it is possible to reconsider and to change your allocations at any time until you hit submit. Simply delete any entry you want to change and enter a new value. Also note that you don’t need to enter a value into each box. If you enter no value under production tokens, for instance, the computer will understand your production tokens choice to be 0. The same applies to protection and theft tokens. You can always choose 0 in any case. If the number of effort tokens you use for the round does not sum to 10 in total, you will receive an error message and will have to change entries until all of your effort tokens have been used. If you are ready, click on Submit now. Note that if all participants haven’t yet made their decisions and clicked submit, you’ll see a screen saying “Waiting for Others”. There is no possibility of seeing what others decide first and then making your own decision. When everyone has submitted their decisions you’ll see a “Round Performance Summary.” Please read it and raise your hand if you have any questions about this screen. When you’re done, please click on Next. 175 The next screen gives you information about the wealth token accumulations of each member in your group. Notice that there are two boxes: (1) the upper box shows the accumulations of each group member broken up into gains and losses through production and theft through the period that just ended; (2) the lower box shows the accumulations of each group member broken up into gains and losses through production and theft in the period that just ended. Please click on Next Round. We’ll now run through a second practice round, which will be the last before the first real rounds begin. Remember, the practice rounds do not count towards determining your earnings. Again, please follow our instructions for this second practice round. Notice that your first screen of the round tells you your updated accumulation of wealth tokens. Note that you begin every round with the same number of effort tokens, 10. Please click next. Again, the next screen is for indicating your allocation choices to production, protection and theft. Notice that you can also click on the Stats button here to view again the accumulations of each group member and the break-down according to production, gains from theft and losses due to theft. If you want the stats to disappear, you can click Hide, otherwise they will go away automatically when you click submit, but you can view them again at the next decision stage. At the second allocation screen, please allocate your effort tokens as follows:  6 to production,  1 to protection, and  For your theft choices, please open again the Stats window. In the real experiment, you may want to use the information in that window to help you decide whom you try to steal from. For this practice round, as an illustration, please assign 2 effort tokens to whichever member of your group (excluding yourself) has the smallest accumulation of wealth tokens after Practice 1 (check the stats window). If more than one member is tied for smallest accumulation, choose one randomly to assign your effort tokens to. Finally, assign 1 effort token to whichever member of your group has the 2nd smallest accumulation of wealth tokens. When everyone has submitted their decisions you’ll see the Round Performance Summary, as before. When you’re ready, please click on Next. Finally, as before, the next screen gives you information about the wealth token accumulations of each group member broken up into gains and losses through production and theft. 176 Before we begin the rounds that count toward your real earnings, please note that the time it will take to complete all 24 rounds depends on all of your rates of progress. No individual can move forward until all participants make the corresponding decisions at each stage of the process, which includes, at the end of each period, that you finish viewing your round performance summary and click “next.” Please focus on the task and click the appropriate “submit” and “next” buttons as soon as you are ready to do so, so that the process can proceed for all in a timely fashion. To help make sure that we finish before _ _ _, we’ll remind you to continue if we see that progress stalls. (We can track on our monitor whether actions have been taken but not which actions they were, that information is stored only for later analysis.) O.k.? Please begin. Practice Rounds for remaining 20 rounds (Practice under independent contribution) Before proceeding with the remaining 20 rounds, we’re going to go through two practice rounds the purpose of which is to familiarize you with the procedure for voting on which of the two schemes for determining the level of collective protection will be used by your group, as well as to familiarize you with the collective protection technology itself. As before, the earnings shown on your screen for these practice rounds are only illustrative and have no effect on your real earnings in the experiment. Please follow our instructions as closely as possible and do not click any buttons until told to. The first screen you’ll see tells you that you and each other person in your group has 100 wealth tokens at the beginning of the remaining rounds. Please click next. The next screen you’ll see asks you to indicate which of the two schemes under which the level of collective protection can be determined you prefer. For this practice round, please all vote for “Decide individually”, whereby each group member decides independently how many effort tokens to contribute to collective protection. Click next. The next screen shows the result of the vote. Since everyone in the group voted for determining contributions into collective protection independently, this is in fact the scheme that will be used for this practice round. The next screen you’ll see in each round tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. 177 The next screen is for indicating how many effort tokens you want to contribute to collective protection. Please allocate the following number of effort tokens to the group account if the last digit of your seat number is even and < 6: 1 effort token odd and < 6: 2 effort tokens even and > 5: 3 effort tokens odd and > 5: 4 effort tokens Notice that in the lower part of the window you can see the number of effort tokens that remain in your account, which automatically updates as you enter your contribution into collective protection. Click submit. The next screen shows the total number of effort tokens all group members have contributed to collective protection and the consequent level of collective protection. Click continue. The next screen you’ll see is the decision screen for allocating the remainder of your effort tokens for this period. Notice that the number of effort tokens you have left appears in the top of the window, and this number is updated as you allocate these effort tokens to production, private protection and theft. You have to click on at least some of the boxes below the lines saying “production tokens,” “private protection tokens” and “theft tokens”, until you finish allocating your effort tokens. For purposes of this practice round, please use your remaining effort tokens as follows (and do not click submit until we tell you to):  Assign 3 effort tokens to production and 3 to private protection.  If you still have any effort tokens left, use them to try stealing from other group members, distributing them randomly among the other group members. Before leaving this screen, please notice that when you enter a number of private protection tokens, you’ll see below that box the total (i.e., collective protection + private protection) level of security of your Wealth token accumulation. Click on Submit now. As before, the next two screens provide the “Round Performance” summary and the information about the wealth token accumulations of each member in your group. Please click on Next and Next Round, accordingly. (Practice under voting scheme) 178 We’ll now have a practice round using the scheme in which the contributions to collective protection are determined by vote. There is again a first screen telling you that you have an endowment of 100 wealth tokens. This screen will in fact only appear at the beginning of the first of the remaining rounds.. The next screen is again the one in which you vote for one or the other of the two schemes for determining contributions to collective protection. (Recall that you will actually vote on the scheme only at the beginning of each set of four rounds, not in every round. The remaining 20 rounds are renumbered beginning from 1, so the votes are before round 1, round 5, round 9, round 13, and round 17. Here you’re voting on the scheme during consecutive rounds in order to get you familiar with both schemes.) Please all vote for “Decide by vote (middle preference binding)” whereby each group member votes for a number of effort tokens he or she would like all five group members to be required to put into collective protection, and the amount that lies in the middle will be selected. Click next. The next screen shows the result of the vote. Since everyone in the group voted for determining contributions into collective protection by majority vote, this is in fact the scheme that will be used for this practice round. The next screen you’ll see tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. In the next screen, you’ll put a number into the box under the heading “Enter the number of tokens you wish to have all group members including yourself put into Collective Protection.” Please vote for having the following number of effort tokens put into collective protection: if the last digit of your seat number is even and < 6: 4 effort tokens odd and < 6: 3 effort tokens even and > 5: 2 effort tokens odd and > 5: 1 effort token When everyone has cast their vote, you’ll see what the group selected as the number of effort tokens that will be contributed by each group member in the round and the consequent level of collective protection. You will also see the number of effort tokens that remain in your account. Notice that the amount that the group decided on (i.e., the middle number among those selected by the five group members) is automatically deducted from your endowment of 10 effort tokens. Click next. 179 At the second allocation screen, please enter the following decisions (and do not click submit until we tell you to):  First, assign 3 effort tokens to production and 1 to private protection.  Then, allocate the remainder of your effort tokens to theft, assigning them randomly among the other members of your group. You may now submit the allocation. The next two screens show your performance summary and the accumulation of wealth tokens by each group member. Click Next as you finish looking at this information. This ends our two practice rounds. Any questions? Please begin. D.2 CHAT Instructions General Information This is an experiment aimed at studying decision-making while interacting with other individuals. During the experiment, you will be earning money in the form of wealth tokens . At the end of the experiment you will be paid in cash in real dollars (1 wealth token = 1.4¢). The amount you will earn will depend on your and others’ decisions. Please read and listen carefully to make sure you understand the decision process. At the end of the instructions you will have a chance to ask questions. The experiment will conclude with a brief questionnaire. Your Group At the beginning of the experiment you will be randomly assigned to a group consisting of yourself and four others. Each member of the group will be randomly assigned a subject number (denoted Sub 0, Sub 1,…, Sub 4) which remains fixed throughout the experiment. You will interact exclusively with the people in your group of five throughout the entire experiment. All decisions are made anonymously, so no participant knows the identities of the other decision makers, nor will you ever be informed who was in your group. Payments are anonymous and will be made in cash at the end of the session. Experiment Structure The experiment consists of 24 rounds, organized in 6 sets of 4 rounds. Before each set of four rounds, you and the others in your group will have an opportunity to exchange written messages in a chat room. In total, we expect the experiment to last no more than two hours, including these instructions and practice rounds. 180 Communication and questions Apart from the written messages you can send to your group’s chat room, communication is not allowed at any other time during the experiment. If you have any questions, please raise your hand and we will come to assist you. Do not hesitate to call on us. Other kinds of tokens During the experiment, you will have the chance to use two types of tokens. Only wealth tokens will be converted into cash at the end of the experiment. Effort tokens will also play a part, but they only have value insofar as they help you to earn or to conserve wealth tokens. Instructions At the beginning of the first round, each member of the group (yourself included) will be endowed with 100 wealth tokens . Each round, each of you will receive 10 effort tokens . Effort tokens have no money value, but they can be used to help you earn or conserve wealth tokens. There are three activities you can perform using your effort tokens: - Produce more wealth tokens - Steal others’ wealth tokens - Protect your own wealth tokens from being stolen in one or both of two ways: collective protection , which adds to the security of all members’ wealth tokens, and private protection , which adds to the security of the wealth tokens of the person who pays for it only. We next explain the order of decision-making and provide more details about the activities of production, theft, and protection. In every round, decisions are made in two stages:  Stage 1: you have to decide how many of your 10 effort tokens you contribute to collective protection , whereby the wealth tokens of every member are equally protected if another member seeks to steal them. For every effort token that any member of the group puts into collective protection, the probability that you (and each person in your group) will keep your wealth tokens if someone attempts to steal them will rise by 6 percentage points on a scale of 0- 100%. The highest level of security that can be attained through collective protection is reached if the sum of everyone’s contributions is 12 effort tokens (or more). In that case, there is a 12 X 6 = 72% chance that a member’s wealth tokens will remain with them in the event of an attempt to 181 steal them. Greater contributions beyond 12 effort tokens do not increase the probability that you will keep your wealth tokens. At the end of Stage 1, you will learn the total number of effort tokens put into collective protection and the resulting degree of security at which theft is prevented, but you will not learn how many tokens each individual put into collective protection.  Stage 2: In this stage, you have to allocate your remaining effort tokens among three alternatives: private protection, production and theft.  Private protection is meant to provide further protection to your own wealth tokens only. For every effort token you assign to private protection, the probability that you will keep your wealth tokens in the event that someone attempts to steal them rises by 10 percentage points, which will be added to the level of security already achieved through collective protection. For example, if collective protection is 54% (i.e., 9 effort tokens were contributed in total) and you put 4 effort tokens into private protection, the likelihood that an attempt to steal wealth tokens from you will fail increases to 54 + (4 X 10) = 94% (in other words, the likelihood that the attempted theft succeeds is only 6%). Of course, you cannot increase the likelihood of failure beyond 100%, so at some point, additional effort tokens lose their effect. In our example, if you put 5 effort tokens into private protection, the 5th effort token raises the likelihood of a theft attempt failing by 6% only, to 100%, and a 6th effort token would have no effect. Note that effort tokens put into collective protection benefit all members equally, and that everyone knows this degree of security in each round; however, private protection benefits only you, and only you know how many effort tokens you use for private protection, thereby your own total level of protection.  Production of new wealth tokens. # Effort # Wealth Tokens The table on the right shows the relation Tokens Produced between effort tokens (the input) and 1 15 wealth tokens (the output). For instance, 2 28 if you allocate 5 effort tokens to 3 39 production, you produce 55 wealth tokens 4 48 which are added to your accumulation. 5 55 6 60 7 64 8 67 9 69 10 70  Theft enables any member to steal wealth tokens from any other member(s) of the group with a likelihood of success that depends upon the other members’ total degree 182 of security, as already described. You can assign effort tokens to stealing from any of the other members, such that for every effort token you direct to stealing from, say, member 2, you will have the chance to steal 10 wealth tokens from him/her. The example below illustrates the stealing mechanism as well as the determination of chances. Example: Group members put a total of 5 effort tokens into collective protection, so the degree of security provided through collective protection is 5 X 6% = 30%. Member 2 puts 3 effort tokens into private protection and member 3 puts no effort tokens into private protection. Hence, 2’s total degree of protection is 30 + (3 X 10) = 60%, while 3’s total level of protection is 30 + (0 X 10) = 30%. If you direct 2 effort tokens to an attempt to steal from member 2 and 1 effort token to attempting to steal from member 3, 20 of 2’s wealth tokens and/or 10 of 3’s wealth tokens may be transferred to you. The likelihood of your theft from 2 succeeding is 100 – 60 = 40%, and the likelihood of your theft from 3 succeeding is 100 – 30 = 70%. The computer will ultimately determine according to the aforementioned probability of 40% whether 0 or 20 wealth tokens are transferred to you from member 2 (note: there’s a 40% chance of 20 tokens being transferred, a 60% chance of 0 tokens being transferred, and no chance of an intermediate amount being transferred). Likewise, the computer will independently decide using the probability 70% whether 0 or 10 wealth tokens are transferred to you from member 3. If a theft is successful, the wealth tokens are deducted from 2’s and/or 3’s accumulation and are added to yours. Exception: If the total number of wealth tokens that other members would successfully steal from a particular group member, say member 2, exceeds 2’s existing accumulation, then the computer will adjust the size of the transfers, since member 2 cannot end up with a negative number of wealth tokens. For example, suppose that members 0, 1, 3 and 4 each direct 3 effort tokens to attempting to steal from member 2, and that none of the theft attempts is prevented by the action of 2’s degree of protection, so that a total of 3 X 10 X 4 = 120 wealth tokens would be taken from member 2. And suppose that member 2 has only 100 wealth tokens at this point. Then members 0, 1, 3 and 4 would each receive only 100 / 4 = 25 rather than 30 wealth tokens. The decision of each to direct 3 effort tokens against member 2 is nevertheless irrevocable; each would have spent 3 effort tokens, though gaining only 25 rather than 30 wealth tokens from it. At the end of each round, the total number of wealth tokens will be computed according to the following formula: Wealth tokens = Wealth tokens held at the beginning of the round + New wealth tokens produced + Wealth tokens you stole from other members – Wealth tokens other members stole from you At the end of each round you will learn the statistics of your performance in that round as well as the cumulative statistics through that round. That is, you will find out: (i) the number of wealth tokens produced; (ii) the total number of wealth tokens you sought to steal from others and the number of wealth 183 tokens that you successfully stole; (iii) the number of wealth tokens other group members sought to steal from you and the number of wealth tokens they successfully stole. (Note that the information on (iii) is given as an aggregate; you won’t be told which particular group members attempted or succeeded to steal from you.) These amounts will be added/subtracted from the number of wealth tokens that you started the round with, according to the formula described above. You will also learn about the total accumulation of wealth tokens in the hands of each other group member through that round, and the number of wealth tokens that each group member has thus far obtained by production, the number each has thus far obtained by stealing from others, and the number each has thus far lost by means of theft. This information will be available to you anytime in the following round as you make your allocation decisions. Payoffs Your earnings from this experiment will be the $5 that is guaranteed to you simply for participating, plus 1.4¢ for every wealth token that you have accumulated by the end of the 24 rounds. Notice that your earnings do not depend on how much you accumulate in comparison to others, only on how much you accumulate. To make sure that you understand how collective and private protection work in the experiment, we’ll now provide some examples. Examples of protection possibilities Example 1: Three members each contribute two effort tokens and two members each contribute three effort tokens to collective protection, for a total of 12, giving a protection level of 72% to everyone. Nothing is spent on private protection, so every member equally enjoys a 72% level of protection. Example 2: No member contributes any effort tokens to collective protection. Each member uses three effort tokens for private protection. They each achieve a security level of 30%. Example 3: Two members each contribute two effort tokens to collective protection, while the other members contribute three, one, and zero effort tokens, respectively. In total, 8 effort tokens are contributed to collective protection, which provides a security level of 48% to everyone. Members individually use various numbers of effort tokens for private protection; for example 6 tokens, 4 tokens, 3 tokens, 1 token, and no tokens. They achieve the corresponding levels of security: 100%, 88%, 78%, 58% and 48%, respectively. Examples of the use of effort tokens and payoffs 184 Example 1: You begin with 100 wealth tokens and each round you use all 10 effort tokens to produce wealth tokens. No one ever contributes to collective protection. No group member ever attempts to steal wealth tokens from you. You earn 70 wealth tokens each round, accumulating a total of 100 + (70 X 24) = 1780 wealth tokens. Your earnings in dollars would be $5 + (1.4¢ X 1780) = $29.92 Example 2: You begin with 100 wealth tokens and each round you allocate 2 effort tokens to private protection and 8 effort tokens to stealing from others. No one ever contributes to collective protection. You can steal up to 80 wealth tokens, and there is an 80% (= 100 – (2 X 10)) chance that any given attempt by others to steal wealth tokens from you will succeed. Your maximum accumulation could be 100 + (80 X 24) = 2020 wealth tokens, but you may earn less than this, possibly much less, if others successfully steal wealth tokens from you and/or if others use effort tokens to provide some security for their accumulations. Using the maximum estimate of 2020 wealth tokens, your accumulated earnings in dollars would be $5 + (1.4¢ X 2020) = $33.28 Example 3: You begin with 100 wealth tokens and each round you and others in your group use four effort tokens for production, two effort tokens for private protection, and four effort tokens for stealing, assigning one token to stealing from each other group member. No one ever contributes to collective protection. You produce 48 wealth tokens, and your protection level each period is 20%. The other members of your group each attempt to steal 10 wealth tokens from you and succeed in 80% of attempts, hence reducing your wealth token accumulation by 4 X 10 X 0.8 = 32 wealth tokens per period, on average. Your four attempts to steal 10 wealth tokens from other members succeed on 80% of attempts, thus adding 4 X 10 X 0.8 = 32 wealth tokens to your accumulation per period, on average. Your wealth token accumulation thus rises by an average total of 48 per period, earning you a total of 100 + (48 X 24) = 1252 wealth tokens, for earnings of $5 + (1.4¢ X 1252) = $22.53 Example 4: You begin with 100 wealth tokens and each round you put 3 effort tokens into collective protection and assign the remaining 7 effort tokens to production. Suppose other members also put 3 effort tokens into collective protection each period. With 3 X 5 = 15 tokens in collective protection, the likelihood that a theft attempt will succeed is 100 – (12 X 6) = 100 – 72 = 28%. Your 7 effort tokens produce 64 wealth tokens for you each round. You can accumulate up to 100 + (64 X 24) = 1636 wealth tokens, but you may earn less if others attempt to steal tokens from you, although in this case they have a 28% chance of success each time (versus the 80% chance of success in the previous example). Using the maximum estimate of 1636 wealth tokens, your total earnings would be $5 + (1.4¢ X 1636) = $27.90 Note that the behavior does not change across rounds in these examples, but this is just for the sake of making these illustrations easy to understand. In fact, your strategy can change over time. As can be seen, there are an almost infinite number of possible outcomes, depending on your decisions and the decisions of others in your group. Chat rooms 185 After questions are answered and we go through two practice rounds that don’t affect your earnings, you will be able to exchange written messages with the others in your group in a chat room before engaging in the first four periods of the experiment. This initial chat room opportunity will last for four minutes, after which will come the first four periods of decision-making that count towards your accumulation of wealth tokens. When those periods end, you will again be able to exchange messages in the chat room, with another such chat period before each four periods of decisions. The time available for chat messages will gradually decline, to three and a half minutes before period 5, three minutes before period 9, two and a half minutes before period 13, and two minutes each before periods 17 and 21. Messages sent to your chat room are seen only by members of your own group and by the experimenters. The experimenters will monitor messages to make sure that three rules are adhered to. First, you are not permitted to send comments that divulge your personal identity, including references to where you are seated, your clothing, gender, and of course your name. Second, you are not to use offensive language. Third, while you are free to discuss courses of action during the experiment, you may not promise to reward or threaten to penalize other members of your group after the experiment ends. If the experimenters find a violation of these rules, you will forfeit your earnings from the experiment. Any questions? Practice Scripts Before the real experiment begins, we’re going to go through a short period of practice using the chat room and two practice decision-making rounds, the purpose of which is to familiarize you with the way that you enter your choices on your computer screen, the order of choice, and the information you get back after each decision. For this chat period and these decision rounds, you will be randomly grouped with a set of participants who are not (barring random overlap) the ones in your group during the periods that determine your money earnings. The earnings shown on your screen for these practice rounds are only illustrative, reflecting decisions I’ll be asking you to enter. They have no effect on your real earnings in the experiment. Please follow our instructions as closely as possible and do not click any buttons until told to. The first screen you’ll see tells you that you and each other person in your group has 100 wealth tokens at the beginning of the experiment. Do not click next yet!. We’re about to initialize the chat room interface for one minute. When the chat window pops up, please type a message of about twenty words in which you describe today’s weather, a book you read recently, or what you ate for breakfast today, remembering to adhere to the chat room rule that you must not write anything that would identify you (note that your subject id appears automatically with your messages). If there’s time, you can also respond to someone else’s message. The purpose is just to get comfortable with how the messages including your own will appear in the chat window. Once the chatting time expires, you’ll be automatically directed to the next screen. O.k., you can now click next. (Participants chat for one minute) 186 The next screen you’ll see in each round tells you the number of wealth tokens and the number of effort tokens with which you begin the round. Please click next. The next screen is the first in which you have to enter a decision. You should enter the number of effort tokens you want to contribute to collective protection under the heading “Put in Collective Protection.” In this first practice round, please allocate the following number of effort tokens to the group account if the last digit of your seat number is even and < 6: 1 effort token odd and < 6: 2 effort tokens even and > 5: 3 effort tokens odd and > 5: 4 effort tokens Notice that in the lower part of the window you can see the number of effort tokens that remain in your account, which automatically updates as you enter your contribution into collective protection. Now click Submit. Note that if all participants haven’t yet made their decisions and clicked submit, you’ll see a screen saying “Waiting for Others”. There is no possibility of seeing what others decide first and then making your own decision. The next screen shows the total number of effort tokens all group members have contributed to collective protection and the consequent level of collective protection. Click continue. The next screen you’ll see is the decision screen for allocating the remainder of your effort tokens for this period. Notice that the number of effort tokens you have left appears in the top of the window, and this number is updated as you allocate these effort tokens to production, private protection and theft. You have to click on at least some of the boxes below the lines saying “production tokens,” “private protection tokens” and “theft tokens”, until you finish allocating your effort tokens. Please use your remaining effort tokens as follows (and do not click submit until we tell you to):  Assign 3 effort tokens to production and 3 to private protection.  If you still have any effort tokens left, use them to try stealing from other group members.  Assign no more than one effort token to stealing from an individual and if you have multiple effort tokens assign them randomly, for instance if you have 2, assign them to any two decision-makers in your group. 187 Before leaving this screen, please notice the following: (a) you can click on the button labeled “Stats” at the top to learn the number of wealth tokens held by the others in your group; (b) when you enter a number of production tokens, you’ll see immediately below that box the number of Wealth tokens you’ll produce if you stick with that choice; (c) when you enter a number of private protection tokens, you’ll see below that box the total level of security of your wealth token accumulation. Before clicking on the submit button, please also notice that it is possible to reconsider and to change your allocations at any time until you hit submit. Simply delete any entry you want to change and enter a new value. Also note that you don’t need to enter a value into each box. If you enter no value under production tokens, for instance, the computer will understand your production tokens choice to be 0. The same applies to private protection and theft tokens. You can always choose 0 in any case. If the number of effort tokens you use for the round does not sum to 10 in total, you will receive an error message and will have to change entries until all of your effort tokens have been used. If you are ready, click on Submit now. When everyone has submitted their decisions you’ll see a “Round Performance Summary.” Please read it and raise your hand if you have any questions about this screen. When you’re done, please click on Next. The next screen gives you information about the wealth token accumulations of each member in your group. Notice that there are two boxes: (1) the upper box shows the accumulations of each group member broken up into gains and losses through production and theft through the period that just ended; (2) the lower box shows the accumulations of each group member broken up into gains and losses through production and theft in the period that just ended. Please click on Next Round. We’ll now run through a second practice round, which will be the last before the real rounds begin. Remember, the practice rounds do not count towards determining your earnings. Again, please follow our instructions for this second practice round. Notice that your first screen of the round tells you your updated accumulation of wealth tokens. Note that you begin every round with the same number of effort tokens, 10. Please click next. Again, the next screen is for indicating how many effort tokens you want to put in the group account. Notice that you can also click on the Stats button here to view again the accumulations of each group member and the break-down according to production, gains from theft and losses due to theft. If you want the stats to disappear, you can click Hide, otherwise they will go away automatically when you click submit, but you can view them again at the next decision stage. For this last practice round, please allocate the following number of effort tokens to collective protection 188 if the last digit of your seat number is even and < 6: 4 effort tokens odd and < 6: 3 effort tokens even and > 5: 2 effort tokens odd and > 5: 1 effort token Now click Submit. At the second allocation screen, please allocate your effort tokens as follows:  3 to production,  1 to private protection, and  For your theft choices, please open again the Stats window. In the real experiment, you may want to use the information in that window to help you decide whom you try to steal from. For this practice round, as an illustration, please assign 2 effort tokens to whichever member of your group (excluding yourself) has the smallest accumulation of wealth tokens after Practice 1 (check the stats window). If more than one member is tied for smallest accumulation, choose one randomly to assign your effort tokens to. If you have more effort tokens left, assign 1 token to whichever member of your group has the 2nd smallest accumulation of wealth tokens. If you have more effort tokens left, assign them as you will among the other group members you haven’t tried to steal from yet. When everyone has submitted their decisions you’ll see the Round Performance Summary, as before. When you’re ready, please click on Next. Finally, as before, the next screen gives you information about the wealth token accumulations of each group member broken up into gains and losses through production and theft. Before we begin the chat periods and the rounds that count toward your real earnings, please note that the time it will take to complete all 24 rounds depends on all of your rates of progress. No individual can move forward until all participants make the corresponding decisions at each stage of the process, which includes, at the end of each period, that you finish viewing your round performance summary and click “next.” Please focus on the task and click the appropriate “submit” and “next” buttons as soon as you are ready to do so, so that the process can proceed for all in a timely fashion. To help make sure that we finish before _ _ _, we’ll remind you to continue if we see that progress stalls. (We can track on our monitor whether actions have been taken but not which actions they were, that information is stored only for later analysis.) O.k.? 189 Please begin the first chat session with the participants with whom you will be grouped for the remainder of the experiment. This first chat session lasts four minutes, followed by four decision periods. Remember the three rules: (a) don’t identify yourself, (b) no threats or promises about what happens outside of the session, and (c) no obscenities. Thank you. 190