Democracy and Resilient Pro-Social Behavioral Change: An Experimental Study

Recent experimental research has shown that a policy put in place endogenously (i.e., through their votes) can be more effective at inducing cooperation in dilemma situations. This paper shows that a similar result holds when a mild sanction is implemented in a voluntary contribution dilemma. Moreover, the effect of endogenous selection of the policy influences the behavior of those concerned even in their interactions in an exogenous setting. The strongest positive democracy effects are seen with those who are favorable of introducing the policy; and especially those whose cooperative dispositions are near the median of the sample and male subjects.


Introduction
Suppose that you are a student who smokes. Imagine that students including you collectively Endogenous collective decision processes, in which rules are collectively selected by voting, are used in most societies and organizations. Empirical research and experiments suggest that the process by which a policy is implemented may substantially change its outcome by a¤ecting civic behaviors on the relevant voted issue. Empirical studies include Frey (1997) for tax laws, Bardhan (2000) for irrigation rules, and Ichniowski, Shaw and Prennushi (1997) and Ichniowski and Shaw (2003) for human resource management practices. Experimental evidence comes from Bortolami and Mittone (2009) for the probability of being audited and the type of sanction and reward, Dal  for a policy that changes the material payo¤ of a game from a prisoner's dilemma to a coordination game, Tyran and Feld (2006) for a non-deterrent law, and Sutter et al. (2010) for an informal sanctioning opportunity. 2 Out of the experimental evidence, Dal Bó et al. and Tyran and Feld furthermore …nd that the e¤ects of endogenous decision processes may exist only when a policy that can resolve or mitigate a problem is endogenously selected. The e¤ects of endogenous processes are currently known to have three e¤ects among others: the e¤ects of signals sent via the voting process, the e¤ects of a selection bias through endogenous processes (i.e., selection e¤ects), and the democracy premium, which is an e¤ect directly in ‡uencing people's behavior.
However, two important issues regarding the democracy premium remain to be studied. First, little is known about the characteristics of the subjects a¤ected by the endogenously imposed policies. Conceptually, the endogenous e¤ect can di¤er by voter type. For instance, supporters of a policy may exhibit a positive endogenous e¤ect due to positive reciprocity when a policy is selected re ‡ecting their votes. Opponents of the policy, by contrast, may exhibit a negative endogenous e¤ect due to negative reciprocity as the policy was chosen despite their dissenting votes. There is competing evidence concerning its e¤ects on assent and dissent. Dal Bó et al.
(2010) …nd that when the policy is endogenously chosen, only those who support a policy that encourages cooperation in a social dilemma behave more cooperatively in one-period play. Tyran and Feld (2006), by contrast, …nd that the act of voting itself changes the behavior of subjects, irrespective of whether they support a policy, in the direction the policy intends to change, if it is endogenously imposed. Furthermore, to our knowledge, the relationship between an individual's cooperative dispositions and the democracy premium is an underexplored issue. Studying the democracy premium by cooperative disposition would add valuable insights to the rich body of experimental literature that has documented that people have heterogeneous social preferences and their behavior di¤ers by their cooperative type (e.g., Fischbacher et al. 2001, Kamei 2011, Kocher et al. 2008, Kurzban and Houser 2005. The degree and direction of the democracy e¤ect that each cooperative type exhibit may di¤er by the type. Lastly, the gender di¤erence in the impact of endogenously imposed policies has not yet been deeply studied. Experimental research in recent decades has found fruitful evidence on the signi…cant gender di¤erences in behavior (Croson and Gneezy 2009). Speci…cally, recent studies show that men are more a¤ected by social norms in charitable giving decisions (Croson et al. 2010), or in cooperation decisions of a voluntary contribution game (Zetland and Della Biusta 2011). The democracy premium therefore may di¤er by gender as male subjects may be more a¤ected than female subjects if an endogenous imposition of a policy changes their perception of the environment and creates cooperative norms in a group.
Second, the possibility of indirect e¤ects of democracy has not received much attention in experimental research despite that there is a large literature on spill-over or transfer of institutional e¤ects on people's behavior from perspectives of other behavioral theories. For instance, it is shown that, in the behavioral game theory, subjects'decisions can be in ‡uenced by their other decisions made in di¤erent environments (e.g., Bednar et al. 2012, Cason et al. 2012. Other studies have demonstrated that the e¤ect of the use of monetary incentives in a task can be transferred to tasks without it (Frey 1993, Frey and Benz 2001, Osterloh and Frey 2000, Frey and Jegen 2001.
The indirect institutional e¤ect in the context of democratic decision-making would include an increase in the level of compliance with exogenously implemented policies among subjects who have experienced successful collective decision-making of policies. In the example of the No Smoking Policy, you might behave more pro-socially, complying with the No Smoking Policy, even in public libraries or in another region where the rule was autocratically implemented.
Moreover, researchers in political science have proposed that civic participation and endogenous decision-making may have a substantial indirect impact on those involved, although these studies su¤er from econometric problems such as a self-selection bias. One indirect e¤ect is of the citizen's initiative -which is one form of direct democracy in the United States -on an increase in voter turnout (Schlozman and Yohai 2008, Smith 2001, Tolbert et al. 2001, Tolbert et al. 2003, Tolbert et al. 2009). Another possible e¤ect is that participating in a public problem-solving activity or in collective decision-making changes a citizen's belief about how he or she should behave as a member of society. It also increases the likelihood that the citizen will engage in future public activities (Haney, Borgida and Farr 2002). The e¤ects of democratic participation here in the literature include the e¤ects of the endogenous decision process of policies that are then successfully implemented.
In order to study the prevalence of indirect e¤ects of democracy as well as to provide a detailed analysis on the democracy premium by subject characteristics, we examine whether the level of people's compliance with exogenously imposed policies rises when they endogenously impose a policy in another setting. We design our experiment by using the framework of a simple linear voluntary contribution game (public goods game), in which full free-riding is privately optimal but full contribution is socially optimal, and a non-deterrent or mild sanction policy. The mild sanction policy just mitigates the incentive to defect in social dilemmas; thus, the mutual full free-riding remains the unique equilibrium in the game even after the policy has been implemented (Tyran and Feld 2006). 3 There are many real examples of mild sanctions, ranging from global policies such as environmental policy to various ordinances such as prohibition of littering.
The key feature of our experimental design is that each subject belongs to two groups with distinct partners and simultaneously plays two modi…ed voluntary contribution games. For some subjects, a mild sanction policy is implemented endogenously in one group, and by the computer in the other group. For other subjects, the policy is implemented exogenously by the computer in both of their assigned two groups. The former are treatment subjects, and the latter are control subjects.
In this paper, we call a group in which the implementation decision is endogenously (exogenously) made the "endogenous (exogenous) group," irrespective of whether the policy is selected. When we measure the e¤ects of the endogenous process, we control for selection e¤ects. Controlling for selection e¤ects is important because endogenous processes are channels through which subjects can choose institutional conditions; subjects' voting decisions and subsequent actions in a game may be correlated with unobservable factors. 4 In order to control for selection e¤ects, in the experiment, every individual makes a voting decision and then the computer randomly assigns them endogenous or exogenous institutional conditions as in Dal . We use the term "direct e¤ect of democracy" for the e¤ect of endogenous processes on a social dilemma that is not captured by the e¤ect of an exogenously imposed policy and by the signaling or selection e¤ects. We use the term "spill-over e¤ect of democracy"for the e¤ect of a person's exposure to the successful endogenous imposition of a policy in one setting on her level of compliance in another setting. We refer to these two democratic premium together as the "e¤ects of democracy," or the "democracy e¤ects." Our experiment result indicates that when the policy is endogenously imposed, treatment subjects who support the policy contribute to the endogenous group signi…cantly more than control subjects who have the same preference. Strikingly, the former subjects in addition contribute to their exogenous group in which the policy is also put in place signi…cantly more than the latter subjects do. The magnitude of the spill-over e¤ect of democracy is similar to the direct e¤ect of democracy. The substantial behavioral change that extends beyond the endogenous group is striking, considering that the policy implemented only mitigates the incentive to defect, and that the social dilemma remains.
Furthermore, we …nd that out of the supporters of the policy only those whose cooperative dispositions are in the middle 50 percent exhibit the statistically signi…cant positive e¤ects of democracy. By contrast, out of the opponents of the policy, we …nd that highly cooperative subjects contribute signi…cantly less when the policy was endogenously chosen. Moreover, our data indicates that the positive e¤ects of democracy for those who support the policy are driven by the male, not female, subjects.
In addition to the contribution of our paper to the literature of organization economics, (experimental) political economy and behavioral game theory mentioned above, our …ndings also have broad implications for the methods used to measure the treatment e¤ect of a policy. That is, since the impact of an endogenously imposed policy on people's behavior may be substantial and may 4 See Dal Bó et al. (2010) and Tyran and Feld (2006). not be restricted to one endogenous setting, the treatment e¤ect of the policy might be underestimated if the possible e¤ect on external settings is ignored when a policy is implemented through endogenous institutions.
The rest of the paper is organized as follows: Section 2 provides details of the experimental design and our identi…cation strategy. Section 3 reports and discusses results from the experiment. Section 4 concludes. An Appendix providing supporting material, including instructions, is available on-line.

Experimental Design
We use a voluntary contribution game framework in the experiment. We employ a modi…ed version of the voting procedure proposed by Dal  to control for selection e¤ects. In this section, we …rst provide the details of our voting procedure (the main piece of our design) and an identi…cation strategy to measure the two kinds of democracy premiums at the same time (Section 2.1). We then describe the whole design including the schematic diagram of our experiment (Section 2.2).

The Voting Procedure and the Triad Device
This section …rst brie ‡y sketches the identi…cation strategy employed in Dal Bó et al. (also see Tyran and Feld for their method). They use a prisoner's dilemma game as an environment. Their interest is to compare cooperation rates between endogenously versus exogenously modi…ed games so as to measure the direct e¤ect of democracy. The issue is that if subjects' voting decisions in ‡uence their institutional choices, then the simply averaged behavior under a selected scheme would not be comparable between the two implementation processes due to the selection e¤ects.
The experimental device to deal with the selection problems, used in Dal Bó et al., is to let all subjects make voting decisions. In this procedure, after everyone votes on a policy, the computer randomly assigns either an endogenous or exogenous implementation condition to each group. Only when the endogenous condition is given to a group, their votes collectively determine the policy implementation. When the exogenous condition is assigned, by contrast, the computer randomly decides whether or not to use their policy in that group. Their identi…cation strategy is to use individual-level data conditioning on their voting decisions, when comparing behavior between the endogenous and exogenous processes. The selection e¤ects no longer exist if we compare behaviors 6 of those whose voting preferences are the same across the implementation processes.
In our experiment, we identify not only the direct e¤ect, but also the spill-over e¤ect of democracy. For this purpose, we modify the voting procedure proposed by Dal Bó et al. so that the following …ve issues are handled. First, we need to generate the following two kinds of subjects (treatment subjects and control subjects) after the voting stage. The …rst subset of subjects (treatment subjects) contains those who are exposed to the successful endogenous implementation of a policy, belong to the endogenous group, and at the same time belong to another group in which the policy is exogenously imposed. The second subset of subjects (control subjects) is composed of those who do not experience any endogenous process and participate only in groups in which the policy is exogenously imposed. A comparison of behavior between these two sets of subjects makes it possible to gauge the e¤ects of democracy. Second, to make the two sets of subjects fully comparable, we should design the experimental set-up so that the former subset of subjects should belong to the same number of groups as the latter subset of subjects. Third, we need to let all subjects vote before an implementation condition is assigned in order to control for selection effects. So as to elicit subjects'voting preferences in an incentive-compatible manner, they must not know beforehand whether an endogenous or exogenous condition is given to each of their groups in the voting stage. Fourth, we need enough observations to undertake statistical and econometric analyses. Speci…cally, it is essential to have the adequate number of independent observations, despite the fact that each subject belongs to more than one group in the experiment. Fifth, results should not be subject to how each group is formed (i.e., network structure). In order to meet these desirable requirements, we design the experiment so that each subject simultaneously belongs to two groups, and our voting procedure uses an inventive device, linking triplets of subjects into "triads" of the type shown in Fig.1 (a); during the experiment, three groups are linked with one another. The group size is two, which is the minimum. 5 The triad structure is unknown to subjects in order to meet the …fth requirement above; they only know that they randomly belong to two groups, each with a di¤erent partner (see Appendix A.1 for instructions).
Our voting procedure operates as follows. First, each individual votes on whether to use a policy separately in each of their two assigned groups. The policy that is used in the experiment is a mild sanction policy as we discuss later, which helps mitigate a social dilemma with non-deterrent sanctions. The number of votes cast by an individual is two, unlike Dal Bó et al. and other studies, as they belong to two distinct groups. We refer to our subjects as follows: "Yes-Yes voters" are those who supported the policy in both of their assigned two groups, "No-No voters"are those who voted against it in both of their assigned two groups, and "Yes-No voters" are those who voted in favor of the policy in one group and against it in the other. After every subject makes their voting decisions, the triad device comes into play: in each triad the computer randomly assigns the endogenous condition to one group and the exogenous condition to the remaining two groups.
The probability with which the endogenous (exogenous) condition is assigned to a group is exactly 1/3 (2/3). As a result, two-thirds of the subjects are exposed to the endogenous process in one group and one-third are never exposed to it (see Fig.1 (b)). Under the endogenous condition, the computer counts the votes and the majority determines whether to implement the policy in that group. When one individual votes in favor of the policy and the other votes against it, the computer randomly breaks the tie. Under the exogenous condition, by contrast, the computer does not consider the votes and randomly decides whether to use the policy. Subjects are not told the exact process either as the existence of the triad structure; they are only told that their votes are randomly counted for each of their two assigned groups. This voting procedure resolves the …rst to third issues described in the previous paragraph. In the paper, we use the following notation for the implementation outcomes: "Endo"indicates that majority rule determined whether to use the policy, and "Exo"means that the computer randomly determined whether to use the policy. "Imp" means that the policy was implemented, and "Not" means that the policy was not implemented.
There are seven pairs of possible implementation outcomes for each subject since they belong to two groups. These are: "Endo Imp, Exo Imp," "Endo Imp, Exo Not," "Endo Not, Exo Imp," "Endo Not, Exo Not," "Exo Imp, Exo Imp," "Exo Imp, Exo Not," and "Exo Not, Exo Not." To better solve the aforementioned fourth issue concerning the number of observations, in addition to the use of the minimum group size, we only focus on the implementation outcomes in which a policy is imposed in both groups ("Endo Imp, Exo Imp," "Exo Imp, Exo Imp"); and adjust one experimental element. This is because our interest is to compare the e¤ects of an endogenously imposed policy and those of an exogenously imposed policy as discussed in Section 1. Accordingly, we manipulate the probability with which a policy is imposed as follows. Under the exogenous condition, or when the vote share is 50% under the endogenous condition, the computer will implement the policy with a probability of 90% and will not implement the policy with a probability of 10%. 6 With this experimental set-up, the outcome "Exo Imp" is much more frequent than the outcome "Exo Not," and the outcome "Endo Imp" is much more frequent than the outcome "Endo Not" when the vote share is 50%. As a result, the unequal probabilities set in the experiment make it possible to produce as many as subjects in "Endo Imp, Exo Imp" (treatment condition) and "Exo Imp, Exo Imp"(control condition). Fig.2 describes treatment and control subjects disassembled from the triad device; which we use in analyses.
As discussed, we identify the e¤ects of democracy by using these two sets of subjects. The two democracy premiums are measured conditioning on their voting decisions to deal with selection e¤ects as in Strategy 1. Subjects are not told the exact probabilities; they are only told that the computer decides randomly in these circumstances. 7 Strategy 1 The E¤ ects of Democracy for voter type s 2 fY Y; N N; Y N g : (1) The Direct E¤ ect of Democracy = C Endo "Endo;Exo;"s C Exo "Exo;Exo;"s (2) The Spill-Over E¤ ect of Democracy = C Exo "Endo;Exo;"s C Exo "Exo;Exo;"s Here, the two contribution decisions in "Endo Imp, Exo Imp"by voter type s 2 fY Y; N N; Y N g are written as C Endo "Endo;Exo;"s and C Exo "Endo;Exo;"s . The former denotes the contribution decision in the endogenous group, and the latter denotes the contribution decision in the exogenous group. Y Y indicates a Yes-Yes voter, N N indicates a No-No voter and Y N indicates a Yes-No voter. As it turns out later, the majority of subjects are either Yes-Yes or No-No voters; therefore, Yes-No voters are not used in most of our analyses except subsection a in Section 3.2. The two contribution decisions in "Exo Imp, Exo Imp" by voter type s 2 fY Y; N N; Y N g are written as C Exo "Exo;Exo;"s : C Endo "Endo;Exo;"s , C Exo "Endo;Exo;"s and C Exo "Exo;Exo;"s denote the average of C Endo "Endo;Exo;"s , C Exo "Endo;Exo;"s and C Exo "Exo;Exo;"s , respectively.

The Experiment
The experiment has two periods. In each period, subjects play two voluntary contribution games, each of which has a group size of two. At the onset of the experiment, each subject is randomly assigned to two groups of two individuals. The other individuals in the two groups are di¤erent from one another. To prevent repeated encounters, regardless of his or her decision in Period 1, each individual is assigned to two new groups with two di¤erent individuals after that period is over. In other words, we employ the perfect stranger matching protocol. In Period 2, as in Period 1, the other individuals in the two groups are di¤erent from each other.
In Period 1, each subject is given an endowment of 20 points (experimental currency unit where 6 points = US$1) in each of their two assigned groups. Subjects then decide how to allocate their 20 points between a private account and a group account in each group. For their choices, only integers between 0 and 20 are allowed. The marginal per capita return (MPCR) from the public account equals 0.6 in each group. Thus, subject i 's total earnings in Period 1 are the sum of their earnings in each group as follows: where C ik is the contribution of subject i in the k th group (k = 1 or 2).
On the one hand, since the MPCR equals 0.6, contributing nothing to the public account in each group is a strictly dominant strategy for an individual. On the other hand, since 0.6 times 2 is 1.2, which is greater than 1, if two individuals in a group contribute 20 points to the public account, the total material payo¤ achieves its maximum. Thus, the unique Nash Equilibrium (NE) in each group is full free-riding (contributing 0 points to the public account), whereas the social or Pareto optimum is full cooperation (contributing 20 points to the public account). Period 1 plays a role in familiarizing subjects with the voluntary contribution game before their voting decisions.
Also, we use the information of the very …rst contribution decisions in analysis in order to classify subjects as how much they contribute to the public account in period 1 elucidates possible human cooperative types (e.g., Gunnthorsdottir et al. 2007). 8 Subjects are aware of their counterparts' allocations to the public account and of their own earnings at the end of Period 1.
In Period 2, a mild sanction policy may be imposed in each group. The MPCR equals 0.6 in this period as well. When the policy is put in place, each point that a subject allocates to his or her private account results in a …ne of 0.3 points. In other words, the earnings in each group are as follows: when subject i contributes C i to the public account, he or she obtains: where I fimplementg = 1 if the policy is implemented; = 0 otherwise.
As in Period 1, since each individual belongs to two groups, their total earnings in Period 2 are the sum of their earnings in each group. Since 1 :3 > :6, regardless of whether the policy is in use, contributing nothing to the public account is still the strictly dominant strategy. As a result, this is the unique NE for a subject who is maximizing material payo¤s, while contributing the full endowment to the public account is the social optimum. Whether the sanction policy is in place in a group depends on their voting decisions and on the computer's random choices. The voting stage is included at the onset of Period 2, which is just before the above contribution stage. The voting procedure described in Section 2.1 is used.

Information feedback after the voting stage:
Under the endogenous condition, a subject obtains partial information about the distribution of votes (i.e., the vote of his or her counterpart) in a group from the vote outcome. For instance, suppose that a subject votes not to use the policy, but the policy is endogenously imposed in that group. In that case, the subject is immediately aware that his or her counterpart voted in favor of using the policy. If we assume that a subject may make contribution decisions and may develop beliefs about partners' future cooperation based on the partial information, the di¤erence in the level of information between "Endo Imp, Exo Imp" and "Exo Imp, Exo Imp" becomes a problem when identifying the e¤ects of democracy, since our purpose is not to study the e¤ects of signals sent via their voting decisions, but to study the democracy premiums alone. Therefore, we need to cancel out the di¤erence in the level of information so as to eliminate a competing hypothesis.
Dal Bó et al. studied the direct e¤ect of democracy by also making that information arti…cially available in the exogenous condition as well. Speci…cally, each of their subjects were informed of whether "at most 50%," or "at least 50%" of them voted in favor of the policy in their group, regardless of whether they were in endogenous or exogenous condition. They conclude that the direct e¤ect of democracy is present even though the opportunity of the signaling is balanced. In our paper, we also include the counterfactual information feedback in both the endogenous and exogenous conditions to balance the level of information. Accordingly, the only di¤erence between an endogenous group and an exogenous group in our experiment is their experience concerning the endogenous implementation processes of the policy. 9 In short, every individual is informed of the following after the voting decision: (1) whether "at most"one or "at least"one person in each group voted to implement the policy (we refer to the former as the "information At-Most,"and the latter as the "information At-Least" in the paper); (2) whether votes or the computer decided on the implementation of the policy; (3) whether the policy will be implemented in each of their assigned two groups.
Some subjects are simultaneously assigned to an endogenous group and an exogenous group; these two groups are di¤erent from each other even if the same policy is implemented in each group, because the implementation process is di¤erent in each group. In addition to the announcement A schematic diagram of the experiment is included in Fig.3. As described above, a rational, self-interested subject will contribute nothing in Period 2, regardless of the outcomes of the voting and implementation conditions. In other words, the sanction policy, if implemented, will cause him or her to pay a maximum …ne of 6 points (= :3 20). Since the group size is two, each subject's vote is pivotal in a group. Thus, under common knowledge that the probability with which their votes will be considered is positive, as well as that the probability with which the computer will implement the policy when the endogenous condition is assigned to a group but the vote share is 50% is also positive, standard theory predicts that each subject will vote against the use of the policy in both assigned groups in this stage (Subgame Perfect Nash Equilibrium Besides the spill-over or transfer of institutional e¤ects discussed in Section 1, income inequalityaverse models such as prominent Fehr and Schmidt (1999) also predicts positive spill-over e¤ects for the Yes-Yes voters, if the direct e¤ect of democracy is present with them. If an inequality-averse Yes-Yes voter in the treatment condition expects that more mutual cooperation is realized in her endogenous group, then she might raise her compliance and increase her level of cooperation also in her exogenous group; as the income inequality between her and the subject in their exogenous group would increase otherwise. After

Results
A total of three hundred undergraduate students at Brown University participated in the experiment in April and May, 2010. In total, 15 sessions were conducted. 11 No subject participated in more than one session. The sessions lasted around an hour on average. The average earnings were $21.38 (including a $5 participation fee) with a standard deviation of $5.04. The instructions for all experimental treatments were neutrally framed, and a few comprehension questions were 1 0 The CRT consists of three questions. See Frederick (2005) for details. A subject earns seven points for each correct response to the three questions. In the beauty contest game, each subject chooses an integer between 0 and 100. The subject with the closest number to two thirds of the average of all numbers in a session earns 120 points.
1 1 The experiment was conducted at a computer laboratory at Brown University. Participants were recruited from the general student population there, using the Brown University Social Science Experiment Lab (BUSSEL) on-line recruitment system. 46 out of the 300 subjects listed economics as one of their concentrations, which was similar to the proportion of economics majors in the university's undergraduate population. Other participants came from …elds ranging from the humanities to the sciences. The number of participants per session was either 21, 18 or 15. All subjects were …rst-time participants in voluntary contribution game experiments. included with them (see Appendix A for the instructions). The experiment was programmed in z-tree (Fischbacher 2007). Communication between subjects was not permitted during a session.
The average contribution in Period 1 was 5.96, with a standard deviation of 6.06. Contributing 30% of the endowment in a one-shot experiment is not unusual. This is near the lower end of the range that is predicted from previous experiments in the literature, given the lower e¢ ciency of the group optimum (see Ledyard (1995) and Zelmer (2003)).

Voting Decisions and Collective Outcomes
The majority of the subjects hold a …rm voting preference, as shown in Table 1. Out of the 300 subjects, around 42% (48%) of subjects voted in favor of (against) the introduction of sanctions in both of their assigned two groups, and only 10% of them voted di¤erently in their assigned two groups. Individual voting decisions often reveal their heterogeneous tastes towards imposing a policy. Appendix Table B.1 reports estimates of probit regressions, where the dependent variable is each subject's voting decision (which equals 1 if the subject voted in favor of the introduction of sanctions; 0 otherwise), and the independent variables are their various demographic variables.
We …nd that the period 1 average contributions variable obtains a signi…cantly positive coe¢ cient at the 1% level, implying that cooperative disposition is an important source of selections in our experiment. 12 At the same time, we …nd that female subjects were signi…cantly more likely to vote against the introduction of the sanction policy, suggesting gender is another source of selections. The female subjects'dislike of using sanctions is consistent with Putterman et al. (2011), which experimentally shows that female subjects are less likely to choose an e¢ cient sanctioning parameter, given the opportunity to construct a centralized sanction scheme capable of resolving a free-riding problem.
The triadic device randomly assigned 200 subjects (two-thirds of the subjects) into one endogenous condition and one exogenous condition, and 100 subjects (one-third of the subjects) into two exogenous conditions. Out of the 200 subjects, 60% of the subjects faced the treatment condition ("Endo Imp, Exo Imp") in period 2. 67 out of the 120 subjects were Yes-Yes voters, and 37 were No-No voters. Of the 100 subjects who belonged to two exogenous groups, 76% of them faced the control condition ("Exo Imp, Exo Imp") in period 2. 31 out of the 76 subjects were Yes-Yes voters and 37 were No-No voters. The number of Yes-Yes voters in the treatment condition and the control condition is proportionally balanced, as the number in the treatment condition is double of that in the control condition, as shown in Table 2.

The Direct and Spill-Over E¤ects of Democracy
In the experiment, all subjects received either the information At-Least or the information At-Most, regardless of whether they belonged to an endogenous or exogenous group. This feature in principle balances the level of information between the two implementation processes. Appendix Table B.3 reports the number of subjects that received the information At-Least by implementation outcome and by voter type (Yes-Yes, No-No). In the treatment condition ("Endo Imp, Exo Imp"), out of the 67 Yes-Yes voters, 52 (48)  This section …rst identi…es the average and median treatment e¤ects in aggregate for each voter type. We …nd that both the direct and spill-over e¤ects are substantial only for the Yes-Yes voters (subsection a). We then explore the signi…cance of the democracy e¤ects by cooperative disposition. Our results show that the degree to which subjects exhibit the e¤ects of democracy di¤er not only by their voter type, but also by their cooperative type (subsection b). We further …nd that the strongest democracy e¤ects are seen in the male Yes-Yes voters (subsection c).

a. The E¤ ects of Democracy in Aggregate
The Yes-Yes voters in the treatment condition ("Endo Imp, Exo Imp") on average contributed 16.55 points to their endogenous groups and 15.73 points to their exogenous groups, as shown in Appendix Table B.2. By contrast, the Yes-Yes voters in the control condition ("Exo Imp, Exo Imp") contributed much smaller amounts, 12.05 points on average, to their exogenous groups. The direct e¤ect of democracy is therefore 4.5 points on average (= 16:55 12:05), which is 22.5% of the endowment, and the spill-over e¤ect of democracy is 3.7 points on average (= 15:73 12:05), which is 18.4% of the endowment. A regression analysis, shown in column (1) and (5) of Table   3, …nds that both the direct e¤ect and the spill-over e¤ect of democracy are signi…cant at the 1% and 5% level, respectively. 13 The sizes of the direct e¤ect and of the spill-over e¤ect decrease to 3.35 points and 3.05 points, which are 25.5% and 17.7% reduction, respectively, if control variables are added, as in columns (2) and (6), but these are signi…cant at the 10% level. 14 The size of the direct e¤ect of democracy is greater than that of the spill-over e¤ect of democracy. They are signi…cantly di¤erent at the 10% level if we use one-tailed Wilcoxon signed ranks test (p-value = 0.080), although this is suggestive evidence, due to a lack of statistical power, of the treatment subjects discriminating their own behaviors between the two groups.
The e¤ects of democracy can alternatively be measured by the use of the median. The median contribution of the Yes-Yes voters in the treatment condition is 20 points, whether in the endogenous group or in the exogenous group. By contrast, it is 15 points for the Yes-Yes voters in the control condition. The direct and spill-over e¤ects of democracy are both 5 points (= 20 15), which is 25% of the endowment, as in columns (3) and (7). These e¤ects are signi…cant at the 1% or 5% level, regardless of whether control variables are included or not, shown in columns (3), (4), (7) and (8). Our result that the Yes-Yes voters contribute more in the endogenous group in which the social dilemma remains (the unique equilibrium is the same, full free-riding, regardless of whether sanctions are imposed or not) is similar to the result in Dal Bó et al. (2010) that shows that 1 3 The average contribution of 16.55 points in the treatment condition and 12.05 points in the control condition are signi…cantly di¤erent according to an individual-level Mann-Whitney test (p-value = .0051, two-tailed), so are 15.73 points and 12.05 points (p-value = .018, two-tailed), although some individual characteristics were not balanced between the treatment condition and the control condition (see footnote 14). The regression models were estimated separately for each of the direct and spill-over e¤ects in the paper. This is because including contribution decisions both in their endogenous and exogenous groups along with two dummy variables in a single equation to measure the direct and spill-over e¤ect requires clustering the standard errors by individual ID. However, the …nite sample bias of clustering is known to be greater when the size of clusters is small. 1 4 Although we employed a random assignment procedure as in other experiments, there were some chances that individual characteristics were not balanced between the treatment condition and the control condition. In fact, we …nd that the period 1 average contribution of the Yes-Yes voters is signi…cantly di¤erent between the treatment and the control condition. Furthermore, we …nd that the proportions of female Yes-Yes voters are signi…cantly di¤erent between the treatment and control conditions. It is, therefore, important to conduct a robustness check. We do so by estimating the e¤ects of democracy while controlling for various control variables as in columns (2) and (6). when a prisoner's dilemma game was endogenously changed to a coordination game where the mutual cooperation is also an equilibrium, the supporters of the modi…cation cooperate signi…cantly more. Our result on the direct e¤ect is also consistent with Tyran and Feld (2006), which show the possibility of the signi…cant endogenous e¤ects in the case of a non-deterrent sanction law in a social dilemma. The results of ours and the two previous studies resonate with the idea that Yes-Yes voters exhibit positive reciprocity to the outcome that a policy was endogenously imposed re ‡ecting their approving votes. The result that Yes-Yes voters exposed to the successful endogenous imposition of the policy also contribute more in their exogenous groups in which the policy was also put in use implies that the impact may spill over to their behavior in a second domain. 15 Result 1: Yes-Yes voters exhibit both the direct and spill-over e¤ ects of democracy in aggregate.

The impacts of democracy on No-No voters appear to be di¤erent from those on Yes-Yes voters.
The average or median contributions of the No-No voters are statistically similar in aggregate between the treatment condition and the control condition. On the one hand, the No-No voters in "Endo Imp, Exo Imp" on average contributed 7.35 points in their endogenous groups and 7.65 points in their exogenous group. On the other hand, the No-No voters in "Exo Imp, Exo Imp"on average contributed 5.84 points in their exogenous groups (see Appendix Table B.2). The direct e¤ect of democracy, 1.51 points on average (= 7:35 5:84), and the spill-over e¤ect of democracy, 1.81 points on average (= 7:65 5:84), are both insigni…cant, as shown in Appendix Table B.5. 16 Likewise, the median contribution of the No-No voters in the treatment condition is 5 points (3 points) in their endogenous groups (their exogenous groups), which is higher than that of No-No voters in the control treatment (0 points). However, the di¤erence in the median between the two conditions is not signi…cant (see Appendix Table B.5).
The result that No-No voters in aggregate do not exhibit the e¤ects of democracy, either direct or spill-over e¤ect, in one-period play is similar to the result in Dal Bó et al. (2010), but not consistent with Tyran and Feld, which …nd that the endogenous e¤ects may also be substantially positive for the  The di¤erence in behavior between ours and Tyran and Feld can be due to the di¤erence in the subject pool (e.g., the United States versus Switzerland).
If we hypothesize that voting for (against) the mild sanction policy is one way of sending signals to (not to) cooperate, the Yes-No voters can be categorized as the strategic subjects who want to cooperate in one group but want to free ride in the other group. This behavior is veri…ed through the Yes-No voters'decisions under the control condition ("Exo Imp, Exo Imp"). The Yes-No voters in the control contributed 15.6 points on average to the group in which they voted for using the policy, and 9.63 points on average to the other group in which they voted against it as in Appendix Table B.6. We …nd the di¤erence in the average contribution between the two groups is signi…cant according to a two-sided Wilcoxon signed ranks test (p-value = .0187).
By contrast, the Yes-No voters in the treatment condition ("Endo Imp, Exo Imp") contributed insigni…cantly more in a group in which they voted for using the policy than in the other group where they voted against it. This is perhaps due to some e¤ects of the endogenous process.
However, a comparison of the Yes-No voters' contribution decisions between the treatment and control condition …nds that the Yes-No voters did not exhibit the e¤ects of democracy (Appendix Table B.6).

b. Cooperative Dispositions and the E¤ ects of Democracy
We next study how each human cooperative type exhibits the e¤ects of democracy. For this purpose, we statistically and objectively classify all subjects into three subsets, using the upper and lower quartiles (25th and 75th percentiles) of their period 1 average contributions; and then, for each subset, we compare their degree of cooperation in period 2 between the treatment and control conditions. 18 The resulting three subsets of the subjects are intuitive. The bottom 25% of the subjects can be interpreted as "free-rider"types as they contributed nothing to both of their assigned two groups.
This subset of subjects are hereafter called "free-rider"in this paper. 19 The top 25% of the subjects can be interpreted as "cooperator" types, which is hereafter used to refer to this subset, as they contributed more than or equal to 50% of their endowment without sanctions. 20 As for the rest of the subjects, the middle 50% (the interquartile range), we simply call them the "intermediate"

types.
The democracy premiums of the Yes-Yes voters appear to di¤er by their cooperative disposition, as shown in Table 4. The "free-rider" and "intermediate" types who faced the treatment condition ("Endo Imp, Exo Imp") contributed substantially more, than those faced with the control condition ("Exo Imp, Exo Imp"). The "cooperator"types, by contrast, contributed almost similar amounts, around 18 points, to their public account, regardless of the implementation processes when sanctions were imposed in both of their assigned two groups. A regression analysis, found in Appendix Table B.7, con…rms that the "free-rider" and "intermediate" types exhibit relatively large, positive average and median e¤ects of democracy. We further …nd that not only the direct e¤ect, but also the spill-over e¤ect of democracy is statistically signi…cant for the "intermediate" types. 21 As for the "free-rider"types, only the direct e¤ect of democracy (not the spill-over e¤ect) is often signi…cant at the 10% level and sometimes at the 5% level, depending on the speci…cations.
The democracy e¤ects are not signi…cant for the "cooperator" types.
Result 3: The democracy premiums of Yes-Yes voters di¤ ers by cooperative disposition. The "in-1 8 See Appendix Fig. B.3. for the empirical distribution of their period 1 average contributions. The lower quartile is 0, whereas the upper quartile is 9.5 1 9 Fischbacher et al. (2001) classify those who contribute nothing to a public good, no matter how much other subjects contribute, as free riders. Note that although the bottom 25% of our subjects contributed nothing in period 1, some subjects might not be considered as free riders according to their classi…cation method. This is because some subjects might have fully defected due to a very pessimistic belief that their counterparts fully free ride.
2 0 Kurzban and Houser (2005) classify subjects into three categories: free riders, reciprocators and cooperators. The cooperators in Kurzban and Houser are those who contribute more than half of their endowment to a public good no matter how many points their counterpart contribute. Note that some of the top 25% of our subjects might not satisfy the condition set by Kurzban and Houser. For example, a conditional cooperator who reciprocates exactly the same amount as their counterpart's contribution would contribute more than half of his endowment if he believes that his partner also does so; but, he would contribute nothing if he believes that his partner also does so. In this paper he is classi…ed as a cooperator if he has an optimistic belief, whereas he is not in Kurzban and Houser. 2 1 The average contribution of the intermediate types in the endogenous (exogenous) group of "Endo Imp, Exo Imp" is signi…cantly di¤erent from that in the exogenous group of "Exo Imp, Exo Imp," according to an individual-level Mann-Whitney test (p-value = .0291 (.0318), two-tailed).
termediate" types, faced with the treatment condition ("Endo Imp, Exo Imp"), exhibit both the direct and spill-over e¤ ects of democracy statistically signi…cantly.
How does the behavior of the No-No voters di¤er by their cooperative disposition? Out of the No-No voters, the "free-rider"types appear to be more cooperative under the treatment condition than under the control condition as seen in Table 4. However, neither the direct e¤ect nor the spill-over e¤ect of democracy is signi…cant. In addition, unlike the case for the Yes-Yes voters, the "intermediate" types contributed almost similar amounts, when sanctions were imposed for both of their assigned two groups, regardless of the implementation processes. However, the most cooperative category, the "cooperator" types, contributed signi…cantly less in the treatment condition than in the control condition. This resonates with the idea that No-No voters exhibit negative reciprocity to the outcome that a policy was endogenously imposed despite their opposition.

Result 4: The democracy premiums of No-No voters di¤ ers by their cooperative disposition. The
"cooperator"types of the No-No voters, faced with the treatment condition ("Endo Imp, Exo Imp"), contributed signi…cantly less than those in the control condition ("Exo Imp, Exo Imp") did. The other two categories of the No-No voters did not exhibit the e¤ ects of democracy.
Results 3 and 4 suggest that consideration of heterogeneous cooperative dispositions can be important and helpful in studying the impacts of democracy, as also shown in other social dilemma experiments (Fischbacher and Gächter 2010).

c. Gender Di¤ erence in the Degree of the Democracy E¤ ects
The male Yes-Yes voters in the treatment condition ("Endo Imp, Exo Imp") contributed 15.87 points on average to their endogenous groups, and 14.91 points on average to their exogenous groups. By contrast, the male Yes-Yes voters in the control condition ("Exo Imp, Exo Imp") contributed much smaller amounts, 9.04 points on average, to their group accounts. The direct e¤ect of democracy for the male Yes-Yes voters is therefore 6.8 points on average (= 15:87 9:04), which is 34.2% of the endowment, and the spill-over e¤ect of democracy is 5.9 points on average (= 14:91 9:04), which is 29.4% of the endowment. These direct and spill-over e¤ects are signi…cant at the 1% and 5% level, respectively, as shown in columns (1) and (9) of Table 5. 22 As shown in columns (2) and (10), the e¤ects are shrunk by around 20% if control variables are included, but these are signi…cant at the 5% or 10% level. Results are similar even if we use median to gauge the e¤ects of democracy.
By contrast, the female Yes-Yes voters, regardless of the implementation processes, contributed somewhat larger amounts once the policy was introduced. Faced with the treatment condition ("Endo Imp, Exo Imp"), they contributed 18.15 points on average to their endogenous groups, and 17.65 points on average to their exogenous groups. By contrast, faced with the control condition ("Exo Imp, Exo Imp"), they contributed somewhat lower amounts, 14.53 points on average. We …nd that the democracy premiums by the female Yes-Yes voters (18:15 14:53 = 3:62 or 17:65 14:53 = 3:12) are much less than those by the male Yes-Yes voters. The premiums are not signi…cant except for one speci…cation, as shown in columns (5) to (8) and (13) to (16). 23 This suggests that the e¤ects of democracy are driven by the male Yes-Yes voters. This supports the idea that men are more a¤ected by emerged cooperation norms as discussed in Section 1.
Appendix Table B.10 reports results for No-No voters by gender. The male No-No voters exhibit almost no e¤ects of democracy in aggregate as shown in Panels (Ia) and (IIa), regardless of its speci…cation. This seems to suggest that the negative reciprocity exhibited by the male No-No voters cancels out the e¤ect of emerged cooperation norms. The female No-No voters, by contrast, exhibit around 5 points on average, which is around 25% of the endowment, of the direct and spill-over e¤ects of democracy. These e¤ects are sometimes signi…cant at the 10% level, depending on the speci…cations (see Panels (Ib) and (IIb) of Table B.10).
Result 6: The male Yes-Yes voters exhibit both the direct and spill-over e¤ ects of democracy, but the female Yes-Yes voters do not. The male No-No voters do not exhibit either the direct or spill-over e¤ ect of democracy, but the female No-No voters do at the 10% level.

Discussion and Conclusions
This paper provides evidence that the successful endogenous imposition of a policy may generate not only a direct e¤ect in the endogenous group, but also a spill-over e¤ect on behaviors beyond the di¤erent according to an individual-level Mann-Whitney test (p-value = .0059, two-tailed), so are that of 14.91 points and 9.04 points (p-value = .023, two-tailed).
2 3 The average contribution of 18.15 points in the treatment condition and 14.53 points in the control are not signi…cantly di¤erent according to an individual-level Mann-Whitney test (p-value = .1839, two-tailed), nor are that of 14.91 points and 9.04 points (p-value = .1780, two-tailed). group. Those who experienced an endogenous change in the payo¤ structure in one group contribute to the group signi…cantly more than those who experienced the same change only exogenously. The study further shows that when those who have experienced the endogenous change are placed in other groups in which the change was imposed exogenously, they still contribute signi…cantly more than subjects who have never experienced the endogenous change, even though the other members in the two groups are di¤erent from one another. Moreover, we …nd that in aggregate these positive e¤ects were only limited for those assenting; the e¤ects were not seen in those dissenting to the policy. Our detailed analysis indicates that the degree to which they exhibit the e¤ects of democracy di¤er by their cooperative disposition; and that the democracy e¤ects in our experiment are driven by the male Yes-Yes voters.
Our results have broad implications in economic policy, political economy and organization economics. Dilemmas exist at many levels in a society: from concerns about the environment (such as global warming) and tax compliance at the national level, to concerns about air pollution and illegal parking at the community level. Policies at various levels have been implemented to resolve these dilemmas, either by the government or through some forms of endogenous processes. The result of this paper implies that the denser the network of endogenous processes, the more likely citizens might become pro-social on various issues and might ful…ll their obligations as citizens.
Resilient pro-social behavioral changes beyond one issue might facilitate social interactions with other citizens with similarly high public spirits. In a …rm with more participatory oriented management, workers that support policies may have a higher perceived self-determination and intrinsic motivation when the participatory management is e¤ectively used to impose the policies. This may result in better performances of employees and it may also make it easier to coordinate employees within and across sections, making the …rm more productive. In evaluating the e¤ect of a policy, it is worth noting that when the policy is implemented through an endogenous process, the extra e¤ects of an endogenously implemented policy (i.e., the democracy e¤ects) are not limited to the relevant social dilemma, but may also emerge in another social dilemma through resilient behavioral changes of those assenting to the policy, triggered by the endogenous process. Furthermore, it is noteworthy that even when introducing some rules does not change the substance of the problem in a material sense, as in the mild sanction policy, introducing the rules endogenously may lead to an opportunity to address not only the problem targeted by such rule, but also other problems that are indirectly, or even not at all, related to the …rst problem by raising their compliance with policies.
The mechanism behind our results on the spill-over e¤ect of democracy on the Yes-Yes voters can be understood by using various models or concepts, including (a) the income inequality-averse model as discussed in Section 2.2, (b) the behavioral spill-over and (c) motivation transfer e¤ects.
As for (b), for example, Bednar et al. (2012) describe that in an environment where subjects simultaneously play two distinct games, if they play a game that encourages sel…sh behavior (cooperation) they are more likely to behave in a sel…sh manner (cooperate) in the other game as well; a phenomena which they call the behavioral spillover. 24 Our result on the democracy e¤ect that extends from the endogenous group to the exogenous group is consistent with their …nding, considering that cooperation was more encouraged and strategic uncertainty was conceivably lower in the endogenous group than in the exogenous group. Consequently, the Yes-Yes voters might have applied a high level of their compliance in the endogenous group to their decisions in the other exogenous group.
The motivational transfer e¤ect in (c) is a damaging e¤ect triggered by monetary incentives that spreads to areas indirectly related to the incentivized area (Frey and Benz 2001). The spillover e¤ect of democracy observed in our experiment is similar to their …ndings, in that subjects' decisions in one domain are in ‡uenced by an institution experienced in the other domain; although the e¤ect in Frey and Benz (2001) transferred subjects'motivation negatively and the democracy e¤ects found in our experiment impacted subjects positively.
We will now turn to directions for future research. We believe that although our results on the spill-over e¤ect are relatively clear, we should treat our paper as the …rst step in studying the spill-over e¤ect of endogenously imposed policies; and that we cannot generalize our …ndings only with the present paper. There are many future research possibilities worthwhile being addressed in order to establish behavioral regularity of our results and explore the driving forces behind them.
First, in our experiment, each subject simultaneously played two voluntary contribution games.
However, if they play the games sequentially, or if they play distinct two games, then the magnitude of the spill-over e¤ect of democracy may di¤er. In addition, it would be worth investigating whether the spill-over e¤ect is a long-lasting e¤ect extending to many environments. Moreover, it could also be worth exploring the signi…cance of the indirect e¤ect of an endogenously imposed policy when there are no policies in the exogenous environment, as people's successful imposition of a policy a¤ect their voluntary cooperation behavior even without policies.
Second, although we believe that our voting procedure is carefully designed, we are aware that there are some limitations in the current design. For example, in our experiment, each subject belongs to only two groups, and the size of the groups is two, the minimum. The scale of the democracy e¤ects may di¤er by the number of groups and group size.
Third, it should also be noted that this study shares similarities with other experimental work in terms of external validity of laboratory experiments. Replication studies as well as …eld experiments on the present topics would be important future research.
As a …nal remark, we note that the result on spill-over e¤ect also has a signi…cant methodological implication in identifying the e¤ects of democracy. We have two potential methods to measure the e¤ects of democracy, whether experiments or empirical studies. The …rst method is to let each subject play games both in the treatment condition and the control condition, and then to compare their behaviors between the two conditions. The second method is, by contrast, to divide subjects into two groups: a treatment group and a control group, let them play the game only in one assigned condition and compare behaviors between the two groups. Which method should we use? Our results indicate that the …rst method may not be accurate, due to the spill-over e¤ect of democracy, and that employing the second method may be a more conservative way to measure the democracy e¤ects.

b) Assignment of implementation process
Note: In this triad, subject 1 and subject 2 are candidates of treatment subjects, whereas subject 3 is a candidate of control subjects.

Fig. 2. Treatment Subject and Control Subject (disassembled from a triad device)
Notes: is the contribution of a voter of type s in their endogenous group under the treatment condition ("Endo Imp, Exo Imp").
is the contribution of a voter of type s in their exogenous group under the same treatment condition.
is the average contribution of a voter of type s in their exogenous groups under the control condition ("Exo Imp, Exo Imp")  -Instructions for Phase 2 -Group composition changes "Endo" condition: Majority rule is used. "Exo" condition: The computer randomly decides.

(I) Subjects' views
Each subject is randomly assigned to a triad one group of a triad two groups of a triad "Endo" condition "Exo" condition -Two approving votes: sanctions imposed -Two dissenting votes: sanctions not imposed -The split of the vote: sanctions imposed with a 90% prob.

(II) Computer's operation
Either "Endo" or "Exo" condition is randomly assigned in each group Voting decision in each of their assigned two groups (group 1, group 2)  Notes: "Yes" and "no" in the column heading indicate that the individual voted in favor of and against the introduction of the sanction policy in a group, respectively. Numbers whose row condition is "X, Y" and whose column condition is 'a, b' are the number of subjects that voted a in a group having an implementation outcome X and that voted b in the other group having an outcome Y. Contribution decisions in the exogenous group by the Yes-Yes voters in treatment condition ("Endo Imp, Exo Imp"), and contribution decisions by voters in the treatment condition ("Endo Imp, Exo Imp"), and contribution decisions the Yes-Yes voters in the control condition ("Exo Imp, Exo Imp") (Reference group).
by the Yes-Yes voters in the control condition ("Exo Imp, Exo Imp") (Reference group).    (1), (2), (5) and (6) and median (quantile) regressions in columns (3), (4), (7) and (8). The reference group in each regression is the average contribution of the Yes-Yes voters in "Exo Imp, Exo Imp" condition; in this condition, since each subject makes two contribution decisions in the same environment, the average of the two contributions is used as each individual's data. #1 Four IQ measures were elicited from subjects in the experiment (beauty contest game scores, cognitive reflection test scores, SAT (math) scores and SAT (verbal) scores). The first principal components were created using the 67 subjects in "Endo Imp, Exo Imp," and the 31 subjects in "Exo Imp, Exo Imp" in order to deal with a multicolinearity problem. Appendix Table B.4 reports supplementary regression analyses when using one of the four IQ measures, instead of the first principal components, as an IQ measure, finding that the direct and spill-over effects are significant at the 5% or 10% level. #2 Eight subjects were excluded since they did not answer their SAT scores. *, **, and *** indicate significance at the .10 level, at the 0.05 level and at the .01 level, respectively.   .04*** 7.59*** 13.00*** 2.03 14.5*** 12.9*** 17.0*** 16.4*** (2.14)

Average Effects
(2. Notes: OLS regressions in columns (1), (2), (5) and (6), and median (quantile) regressions in columns (3), (4), (7) and (8). The reference group is the average contribution of the Yes-Yes voters in "Exo Imp, Exo Imp" condition; in this condition, since each subject makes two contribution decisions in the same environment, the average of the two contributions is used as each individual's data. #1 IQ measures include beauty contest game scores, cognitive reflection test scores, SAT (math) scores and SAT (verbal) scores. The first principal component was created using the 67 subjects in "Endo Imp, Exo Imp," and the 31 subjects in "Exo Imp, Exo Imp" in order to deal with a multicolinearity problem. Appendix Table B.9 reports supplementary regression results when using one of the four IQ measures, instead of the first principal component, finding that both the direct and spill-over effects are significant often at the 5%, and sometimes at the 1% or 10% level for the male Yes-Yes voters. #2 Three subjects were excluded since they did not answer their SAT scores. #3 Five subjects were excluded since they did not answer their either SAT (Math) or SAT (verbal) scores. *, **, and *** indicate significance at the .10 level, at the 0.05 level and at the .01 level, respectively.
(II) The Spill-Over Effect of Democracy Dependent variable: Contribution decisions in the exogenous group by Yes-Yes voters in the treatment condition ("Endo Imp, Exo Imp"), and contribution decisions by the Yes-Yes voters in the control condition ("Exo Imp, Exo Imp") (Reference group).  .04*** 9.04*** 13.00*** 9.63** 14.5*** 14.6*** 17.0*** 18. Notes: OLS regressions in columns (9), (10), (13) and (14), and median (quantile) regressions in columns (11), (12), (15) and (16). The reference group is the average contribution of the Yes-Yes voters in "Exo Imp, Exo Imp" condition; in this condition, since each subject makes two contribution decisions in the same environment, the average of the two contributions is used as each individual's data. #1 IQ measures include beauty contest game scores, cognitive reflection test scores, SAT (math) scores and SAT (verbal) scores. The first principal component was created using the 67 subjects in "Endo Imp, Exo Imp," and the 31 subjects in "Exo Imp, Exo Imp" in order to deal with a multicolinearity problem. Appendix Table B.9 reports supplementary regression results when using one of the four IQ measures, instead of the first principal component, finding that both the direct and spill-over effects are significant often at the 5%, and sometimes at the 1% or 10% level for the male Yes-Yes voters. #2 Three subjects were excluded since they did not answer their SAT scores. #3 Five subjects were excluded since they did not answer their either SAT (Math) or SAT (verbal) scores. *, **, and *** indicate significance at the .10 level, at the 0.05 level and at the .01 level, respectively.