What Makes Bad Policy Win: The Effects of Competing for Political Power Kai Ou Abstract Does the competition for political power affect political behavior? Would it undermine the quality of democratic choice? This study examines these questions using results generated from a laboratory experiment. The experimental design establishes elections consisting of a bad policy and a socially desirable policy. By varying whether individuals economic or political status is earned or randomly assigned, I measure voting behavior across institutions and estimate of the probability that the socially desirable policy wins. I find that the competition for different status has differential effects on political behavior: Competing for economic status decreases individuals political participation, but competing for political power increases voter turnout. The competition for political power itself does not undermine the quality of democracy as compared to the effect of competition for economic status. However, when economic status directly determines the distribution of political power, policy outcomes are less likely to be social welfare maximizing. Keywords: Political Competition, Inequality, Voting, Distributive Policy, Institutions, Experiment The financial support from New York University is gratefully acknowledged. I thank Chris Dawes, Eric Dickson, Guillaume Fréchette, Sanford Gordon, Jens Großer, Zhaotian Luo, Matthew Pietryka, Xiangdong Qin, and Jean-Robert Tyran for their useful comments. I am especially indebted to Rebecca Morton for her constant guidance, support, and advice in the development of this project. An earlier version of this paper was presented at Shanghai Jiao Tong University, New York University, and Florida State University. I thank the participants for their comments and suggestions. All errors remain my own responsibility. Assistant Professor, Department of Political Science, Florida State University, and faculty of the xs/fs Experimental Social Science Research Group at Florida State University, 551 Bellamy Hall, Tallahassee, FL 32306-2230. kou@fsu.edu.
Public policies, such as farm subsidies in agriculture, the Trans-Pacific Trade deal, and the health care reform, affect the distribution of wealth. To resolve conflicts between interest groups and balance interests of different social communities, democratic elections are devised and widely used to decide public policies (Przeworski, Rivero and Xi, 2015). The equity of voting rights is supposed to protect a society from biased and extreme policies, a hallmark of the democratic ideal. However, in the history of democracies and in modern American politics, property owners and taxpayers have strong influence on the social-welfare spending and public goods provision (Scheve and Stasavage, 2017; Bartels, 2016; Ansell and Samuels, 2014), and interest groups including lawyer and medical associations decide the operation of social systems (e.g., Eggers and Hainmueller, 2014; Baybeck, Berry and Siegel, 2011; Bernhagen and Bräuninger, 2005). In other words, equal political rights do not guarantee political equality. Favorable policy outcomes are often more attainable by those who have more political influence. Naturally, political influence is a scarce resource that individuals and groups want to earn and compete for. Competition has been found to affect individuals economic behavior (e.g., Pan and Houser, 2017; Herrera, Reuben and Ting, 2017; Markussen, Reuben and Tyran, 2014; Alesina and Angeletos, 2005). Political competition may have a crucial influence on the formation of political preferences (Druckman and Lupia, 2016). However, the competition in politics is fundamentally different from the competition in economics, and thus the effect of competition for political power is not straightforward. 1 political influence affect voting behavior? Does the competition for Would it undermine the quality of democratic choice? Despite the extensive literature on the effect of institutions on political behavior (e.g., Herrera, Morelli and Nunnari, 2016; Michelitch, 2015; Woon, 2014; Bendor, Diermeier, Siegel and Ting, 2011; Dal Bó, Foster and Putterman, 2010) and on the equilibrium effect of voting behavior on collective choices (e.g., Diermeier and Li, 2017; Dal Bó, Dal Bó and Eyster, 2017; Agranov and Palfrey, 2015; Casella, Llorente-Saguer and Palfrey, 2012), there 1 In the arena of politics, an individual does not need to be rich to have more political power. An individual could become an influential politician or a member of a politically influential interest group in the absence of being wealthy in the first place. 1
is still no clear answer to these questions. Understanding these questions is important. When political inequality is just a social reality, who has more power, how one acquires more power, and what one does when obtaining this power has a crucial effect on policy outcomes. In this article, I investigate how competition for political status influences individuals incentives and their choices of distributive policies. Based on the individual level data, I study the competition between individuals and groups and explore the effect of political institutions on the efficiency and quality of policy choices. To investigate the effect of competing for political influence on voting behavior and policy outcome, the use of observational data is fraught with two challenges. First is the challenge of constructing the appropriate counterfactual by which to compare the results of the study. Not only are data on political inequality difficult to access, record, and measure, but there is a fundamental problem in that economic and political inequality are observationally equivalent in many environments, with both co-evolving in many societies. It is exceedingly difficult to identify the effect of competition for political influence and establish causal relationships between economic inequality and political inequality using historical or contemporary data. Furthermore, the links between economic inequality and political inequality vary across countries and cultures. Therefore, when using observational data to do a comparative study, it is difficult to measure the impact caused by the differences such as political institutions, human capital, and development levels. Second, and critically, it is important to develop a method to cleanly evaluate the institutional effect on the quality of democratic choice. Although political inequality is generally acknowledged in academic studies and in the minds of the public, there is not a consensus on the definition or implication of political inequality (Przeworski, 2008). It is important to study the behavioral and policy effect of competing for political power in a broad sense so that the results of this study could be applied to the understanding of the policy outcomes of particular forms of political inequality. I address the first problem by conducting the study in a laboratory setting, which allows me to precisely investigate the behavior of individuals within different institutions and control for various challenging confounders in the analysis. In the experiment, I vary the 2
mechanism of formulating institutions in terms of whether subjects can compete for more political influence, whether subjects can earn more income than others, or both. In one case I hold individuals economic status constant (flat income). Then by varying whether individuals political status is earned or randomly assigned, I can identify the effect of competition for political power. In another case, I set one s political status to be randomly assigned, and assess the effect of competition for economic status by varying whether one can earn better economic status or remain at a constant level. I then study the joint effect of competition for both economic and political status by comparing the voting decisions and policy choices between the institution in which one could compete for both economic and political status and institutions in which one could earn either economic or political power (but not both). I address the second challenge by establishing an intuitive environment of political inequality such that there is a privileged group that is the minority of the population but has more political influence in the policy making process than an unprivileged group that constitutes the majority of the population. Moreover, I consider elections consist of two distributive policies: a policy that is favorable to the minority and privileged group, and another policy that is favorable to the majority and unprivileged group. This design captures the conflict of interest in distributive public policies between social groups, as most public policies are seen as benefiting some at the expense of others (Pateman, 1970; Dahl, 1973). The experimental design also reflects the benefit of earning membership in a politically influential group. That is, policy outcomes are determined by the real distribution of political power rather than a simple comparison of group sizes or voter numbers. I control for the aggregate social welfare and the distribution of political power between institutions. Thus, by evaluating the efficiency and implications of collectively chosen policies, I can compare the quality of democratic institutions. The experimental results show that economic preferences and political preferences are not necessarily allied: Competition for economic status decreases the average political participation by approximately 15%, while competition for political status in general increases voter turnout by 52%. When both a higher income and membership into privileged group 3
can be earned, it increases political participation dramatically and is most significant when compared to situations in which individuals compete for only economic status. These results identify a non-zero institutional effect on voting at the individual level. To determine whether the individual level findings are economically significant, I then consider how the changes in voting behavior translate into the efficiency and quality of policy outcomes. Specifically, how often do the majority group preferred policies win in elections? I find that when income is constant, the social welfare maximizing (majority group preferred) policies are more likely to be chosen when individuals compete for political status than when political status is randomly assigned. In the absence of competition for political status, the majority preferred policies win more frequently when allowing competition for economic status. These results suggest that the effect of competition for economic or political status on policy outcomes is positive as compared to the counterfactual without any competition for social status. However, I also find that compared to situations in which individuals only compete for either economic status or political status, when better economic status guarantees better political status, the policy outcome is less likely to be social welfare maximizing. These findings imply that competition for political influence per se does not undermine the health of democracy. What impacts the efficiency and quality of policy outcomes is the relationship between economic status and political status. This paper has important implications for the understanding of differences in distributive policies that exist across democracies (e.g., Weingast, Shepsle and Johnsen, 1981; Wittman, 1995; Alesina, Glaeser and Sacerdote, 2001; Iversen and Soskice, 2006). The experimental results of this study suggest that in countries where economic inequality is highly correlated with political inequality, policy outcomes are more favorable to the minority but politically influential groups. This finding, through a different perspective, provides evidence of the detrimental effect of inequality that is consistent with other studies using observational data (e.g., Piketty, 1995; Benabou and Ok, 2001; Gilens, 2012; Giger, Rosset and Bernauer, 2012). This study contributes to the literature in that I isolate the effect of competition for political influence and I identify that the mechanism that undermines the quality of democratic choice 4
may be the linkage between economic and political inequality. I find robustness evidence supporting that the way in which an institution decides the distribution of welfare and political power has a fundamental effect on the health of democracy. This study also contributes to the growing body of studies that investigate the effect of the origins of unequal social status on the relationship between economic and policy preferences (e.g., Osborne and Slivinski, 1996; Acemoglu and Robinson, 2005; Benoit and Laver, 2006; Carey, 2007; Besley, Persson and Sturm, 2010; Berry, 2015). The results of this study show that effort expended to earn political status in a competitive process affects subsequent political behavior, but the effect of competition for political status is different from the conventional wisdom of the effect of earning economic status. This study identifies the relationship between competing for political status and the quality of democratic choice. Research Design: Conceptual Framework In this section, I present a simple conceptual framework to define a theoretical model developed for this study and demonstrate the micro rationality of how institutions affect individuals preferences over policies. The theoretical discussion serves as the reasoning underpinning the research design and it provides the theoretical foundation for the empirical analysis. Voting Over Distributive Policies To establish a normative benchmark for the research design, I consider a society in which there are N voters who collectively choose one of two distributive policies, {A, B}. The electorate is split into two groups of voters and there is asymmetry in political preferences and political power between the two groups. Group A has βn voters and Group B has (1 β)n voters, where 0 < β < 0.5. All Group i players receive the same payoffs if Policy j is elected, so that u j i > 0. Moreover, Group i voters receive higher payoffs if Policy i is elected; that is, u i i > u j i and i j. Group A is a privileged group in which each of its voters has more political power than each Group B voter. The political influence of a privileged 5
voter is given by k, where k > 1. k is a broadly defined political advantage. It could represent the unequal representation, more weight in voting, higher ability to take the voting cost, and others. Voting for a Policy incurs individual i a voting cost c, while abstaining is free. The size of the electorate and the number of each type of voter is common knowledge to all. For the sake of simplicity, an election winner is determined by a random dictator rule (that is similar to the voting rule used in Feddersen, Gailmard and Sandroni (2009); Dal Bó, Dal Bó and Eyster (2017)) in which a randomly selected voter s ballot is chosen to decide the winner. Specifically, if the randomly selected ballot voted for Policy A (B), then Policy A (B) is declared the winner; if the randomly selected ballot is an abstention, then an additional ballot is randomly drawn until a winner is determined. In the case that all voters abstain, then each policy wins with a probability of 50%. The random dictator rule introduces a random effect on the outcome, which helps to identify a unique symmetric equilibrium of the voting game. Moreover, it results in some uncertainty over the outcome of the election which mirrors the scenario in natural elections. That is, because of the strategic voting behavior and random shocks or variations, there is a probability that a black swan event occurs and the underdog could win the election. Finally, the randomness helps to relax the artificial condition that voters know the exact distribution of political power and voter types, which bolsters the case for external validity. The theoretical analysis follows the logic of a pivotal voter model in which a voter is motivated by her individual assessment of the likelihood that her vote will be pivotal in the ultimate decision. Given the random dictator rule, it is straightforward to show that any vote in favor of a policy increases the probability that such policy wins the election. Hence, for any distribution of voter choices, the only choice voters face is whether to vote for their own preferred policy or abstain. Define π i as the increase in the probability of Policy i winning when a voter of Group i chooses to vote rather than abstain, then π i (u i i u j i ) is the expected benefit of voter i voting for Policy i. Given that the cost of voting and payoffs of each possible outcome is common knowledge, the symmetric strategy profile of an individual voter is characterized by the mixed strategy, σ(p, q ), where p is the probability that a 6
Group A player votes for Policy A and q is the probability that a Group B player votes for B on equilibria. I assume that voters make their voting decisions using the standard calculus of voting. That is, an individual will vote when the expected benefit of voting is higher than the voting cost, and abstain otherwise. When π i (u i i u j i ) = c, a voter is indifferent between voting and abstaining, which yields the symmetric mixed-strategy Bayesian Nash equilibrium to the voting game. A detailed derivation of the pivotal probabilities can be found in Appendix A.1. When the parameters of the payoff and voting cost are determined, the calculation of the symmetric mixed-strategy equilibrium behavior as well as the probability of a policy winning the election is determined. Remark 1 By the symmetry assumptions, the probability of Policy B winning the election can be simplified to Prob. (B wins) = q (1 β)n p kβn + q (1 β)n Remark 1 defines the measurement of the efficiency and quality of policy outcomes investigated in this study, and it establishes the normative benchmark of the theoretical analysis. This remark follows by noticing that the probability of Policy B winning depends on the proportion of the privileged Group relative to the electorate population and on the magnitude of unequal political power k that the privileged group members have. There exist situations in which the privileged group is the absolute minority in the population but has sufficient political power and advantages in an election that enables them to generate a favorable outcome for their group. Then, regardless of the finally implemented policy, the expected benefit of being a member of the privileged group is always greater than the expected benefit of being a member of the unprivileged group. Institutional Effects on Democratic Choices To be specific, an institution in this study refers to a set of rules that determine who can do what under what situations; an institution decides the cost and benefit of an individual s choices, and in this study it determines the origins of one s economic and political status and 7 (1)
the relationship between them. From authoritarian regimes to democratic societies, there exist politically privileged groups in different forms. The formation of a privileged group is typically not an exogenous event but rather a consequence of a societal, geographic, or historical selection process, which makes it extremely difficult to establish causality and conduct a comparative study to cleanly investigate institutional effects on political behavior and policy choices. In this article, I study a privileged group that is only politically privileged. That is, I consider a setting in which citizens have no initial (economic or political) endowment; they should exert effort to earn their economic and/or political status. I focus on an institutional effect that is caused by the relationship between economic status and political status and the determination of status. Even with the same policy selection process, the origins of the power and how the membership of the privileged group is determined may have a crucial influence on policy preferences and election results. I construct a counterfactual benchmark such that one s economic status cannot be earned and whether one can become a member of the privileged group is randomly determined. By establishing treatments in which individuals are allowed to compete for either economic or political status, I can test whether and how individuals competing for economic or political power may lead them to form political preferences over distributive policies. At the micro level, a theoretical assumption underlines the research design is that human beings place more value on the earned status. 2 Assume an individual has two mental accounts (Tversky and Kahneman, 1981; Thaler, 1985), one is for economic status and the other is for political status. Between the two accounts, when an individual earned economic but not political status, she would place more value on economic account and relatively less value on political account; vice versa. The value placed on each account affects one s valuation of economic or political activities: The competition for economic status may make individuals become self-regarding and increase in-group rivalry, which in turn decreases coordination in a collective action and thus make people place fewer value on political actions. However, the competition for political status may make people place more value on political account, which 2 I justify this theoretical assumption using the experimental data and report the results after I analyze the main findings. 8
in turn increases people s interest in politics and their willingness to participate in political actions. As a result, these analyses would predict a negative effect on political participation caused by competition for economic status and a positive effect caused by competition for political status. The Experiment In this section, I explain how I establish different institutional environments in the experiment to study effects of competition for status. Setup In the laboratory experiment I used the environment discussed in the conceptual framework with a particular choice of parameters. There are 4 Group A voters and 6 Group B voters. Each Group A member has two ballots while each Group B member has one ballot, so that Group A is the privileged Group. If a Group A member votes for Policy A (B), it means that both of her ballots are for Policy A (B). The cost of voting is 7 experimental currency units (ECU). I consider two elections: Election E1 and E2. Table 1 presents the payoffs of the elections. Table 1: Payoffs of Elections Election E1 Election E2 A wins B wins A wins B wins Group A 250 200 250 230 Group B 50 200 50 230 The experiment reported herein presents at least two methodological challenges. The first stems from the desire to design an experiment that is relatively easy to explain to subjects, especially the key message that they have greater political power and advantages as members of the privileged group. Having more ballots than the voters of the unprivileged group in an election is the most straightforward way to relay to participants the advantages of being privileged. The experimental design used in this study is intuitive, and it can be used to 9
examine the trade-off underlying the proposed theory and to empirically test the behavioral hypotheses. As noted earlier, political advantages studied in this paper are broadly defined. Political inequality established in this experiment can be thought of as a metaphor for a wide variety of setting, such as more representation, more weight in voting, higher ability to take the voting cost, etc. Second is my interest in controlling for the benefits of greater political power and establishing a clean measure of the effects of competing for political influence on policy outcome. For this reason, the same distribution of political power is applied to every treatment. In each institution investigated in this paper, the level of political inequality is the same. To measure the health of democracy with political inequality, it is important to have a clear distinction between policies. Undoubtedly, there is no perfect policy that satisfies everyone. To this end, I considered a good policy from the perspectives of equality, minimum payoff, and aggregate social welfare. In the design of payoffs of elections, I used the criteria defined in Feddersen, Gailmard and Sandroni (2009) and made Policy B to be a socially desirable policy. The implication is that in terms of income distribution, when Policy B wins it (a) minimizes the inequality in earnings between Group A and Group B members, (b) increases the minimum payoff of the losing group from 50 ECU to at least 200, and (c) maximizes aggregate payoffs of the entire society when Group B members make up the majority of the society. Then, when I analyze the experimental results, I focus on the frequency that Policy B wins and I use it to estimate the efficiency and quality of policy outcomes. The difference between Election E1 and Election E2 is that when Policy B wins E2, the outcome represents an even wealthier society, which means Policy B in E2 is even better than Policy B in E1. This design provides an opportunity for investigating the efficiency loss when a better policy is available. Procedures At the beginning of a session, participants were told that the experiment consisted of two stages that were relevant to the determination of a subject s economic and political status in the experiment. First, subjects engaged in a Task Stage in which subjects were 10
asked to exert effort to solve problems for twenty minutes. Specifically, subjects were asked to do a counting-zero task (Abeler, Falk, Goette and Huffman, 2011): subjects counted the number of zeros in tables containing 150 randomly ordered zeros and ones, with performance recorded according to a point system in which correct counts received one point. Depending on what treatment one participated in, more points would result in better status, with subject performance ranked at the end of the task stage. When the Task Stage was concluded, subjects received feedback on their performance. Subjects were ranked based on their earned points and then grouped based on the treatment being tested. Subjects were then engaged in an election process (hereafter, the Voting Stage). The setting of the Voting Stage is exactly as aforementioned. Of the 10 participants in each session, 4 were assigned to Group A and 6 to Group B. After each election-period, when all ten voters had made their choice, subjects received feedback that specified the distribution of turnout in that election, and a voter s own payoff was displayed on the screen, broken down into three parts: the winner of the election, whether a voting cost had occurred to the voter, and the payoff of that voter in that election. Each session consisted of 60 elections. For the two elections, E1 and E2, they were conducted as two blocks. The first 30 elections were for one election and the second 30 elections were for the other election. The sequence of the elections was switched across sessions to avoid order effects. Subjects understood that one election of each block would be randomly selected as the election for which they were paid. A strategic assessment of an election environment is sophisticated. By making subjects engage in voting repeatedly, I wanted to facilitate learning so that I could test whether subjects updated their beliefs and strategies and whether the results converged to strategic voting equilibria. The winner of each election was determined by the random dictator rule as aforementioned. While one of the ten participants was drawing the winning ballot, the others were required to wait until the winner was determined. There was no connection between the ballots shown on the computer screen and seat number. Participants were then paid based on the combination of whether or not they voted and the election outcome. Based on a predetermined exchange rate and the realization of 11
the randomly selected election-period, the earned experimental currency was converted to dollars and paid to subjects. Voting procedures were identical across all treatments, but the mechanisms for determining task earnings and political roles varied depending on the treatment in question. The experiment was programmed using z-tree (Fischbacher, 2007) and conducted at the Center for Experimental Social Science at New York University. 100 undergraduate student subjects participated in the study. Subjects were recruited from a subject pool consisting of undergraduate students from various departments. Participants, who were limited to one session each, were identified by their experiment specific ID numbers. Instructions were read by the same researcher at all sessions. I had no prior connections to any subject who participated in this study. Sessions averaged approximately 90 minutes each and participants were paid based on the number of experimental currency they earned during the experiment (exchange rate of 50 ECU = $1). The average total payment, including an $8 show-up fee, was $23. Treatments and Hypotheses The main experiment involved four treatments with two sessions conducted for each treatment. Following the conceptual framework, I considered whether economic advantages could be extended to political advantages, and I used a between-subject 2 2 factorial design for four treatments. Table 2 reports a summary of the treatment variables with indications of the respective source of variation. Essentially, I considered how one s economic status and political status was determined. Four different institutions on the assignment of economic status and political status were studied. In Treatment EarnN, subjects were paid a constant earning and their political status was not determined by their task performance. This treatment served as the counterfactual benchmark to identify the effects of competition for economic and political status. When the task performance was not correlated with the determination of economic status (as in EarnN and EarnP), subjects were paid 100 ECU regardless of how many questions were answered correctly. When economic status was determined by the relative performance 12
Table 2: Description of Treatment Variables Treatment What can be earned? economic success political power EarnN No No EarnE Yes No EarnP No Yes EarnEP Yes Yes of subjects (as in EarnE and EarnEP), participants who ranked first or second received 200 ECU, those ranked third and fourth received 150 ECU, those ranked fifth, sixth and seventh received 75 ECU, and the rest received 25 ECU. It is essential to note that, in every treatment, the sum of task earnings was 1000 ECU. This design provided the incentives to compete for better economic status when available, and enabled me to isolate the effect of competition only for political influence. Importantly, it allowed for aggregate payoffs to be equivalent across treatments, thereby supporting comparisons across treatments. On political status assignments, in EarnN and EarnE subjects were randomly assigned to either Group A or B. In EarnP and EarnEP subjects were assigned to either Group A or B based on their economic status such that the four with higher incomes were assigned to Group A and the rest were assigned to Group B. The discrete and performance-based wealth/power assignment system could be thought of an abstract of a wide variety of institutions. This design allowed me to control for the distribution of political inequality across treatments. Based on the analysis reported in the conceptual framework, for a political preference to be formed and affected in the competition process, it must be that the competition for political status increases individuals interest in politics and their willingness to participate in politics. It leads to the following hypothesis. Hypothesis 1 When individuals compete for membership of the privileged group, they are more likely to engage in voting and vote for their group favored policy. To interpret the effect of changes of individuals political behavior on policy outcomes, I revisit the conceptual framework and use the particular values of the parameters used in the 13
experiment to derive the hypotheses that I test with the experimental results. I solve for the symmetric mixed strategy equilibria of the voting subgames and calculate the probability that Policy B wins. Normatively, Policy B should win in about 57% of the time in E1 and 81% of the time in E2. A detailed report of calculations can be found in Appendix A.2. The theoretical analysis based on the experimental setting establishes a normative benchmark that proves the democratic ideal: even if the privileged minority is politically more powerful, good policy (Policy B in the context of this study) is more likely to be the outcome of democratic choice. This leads to the following hypothesis. Hypothesis 2 Policy B is more likely to win in elections, and the probability that Policy B wins is greater in E2 than in E1. Following Remark 1, it is apparent that to what extent the quality of policy outcomes is affected depends on how institutions motivate different groups engaging in politics. Hypothesis 3 When the competition for status motivates more B s voting, it increases the quality of policy outcomes. Otherwise, it decreases the quality of democratic choice. Results In this section, following the sequence of the theoretical analysis, I first analyze voting observed in different treatments, then I measure the quality of democratic choice based on the voting data. In most cases Group A (B) voters voted for Policy A (B); subjects occasionally voted for the other group s favored policy, but the frequency of voting for the other policy is not statistically higher than zero. In this section, I focus on the results of voting for one s own group preferred policy. 3 Then I analyze the effect of value on political preference and empirically measure how much value subjects placed on earned status. Finally I report the results of a robustness check experiment that demonstrate that the main findings are robust to heterogeneity. 3 I conducted robustness checks (including whether the results are robust to possible experience and learning effects) and report additional results in Appendix B. The main results reported in this section are robust to additional investigations and robustness checks. 14
Main Findings Among a number of interesting results, there are a number of important findings. I conducted the Mann Whitney Wilcoxon statistical test to explore average treatment effects. Compared to the counterfactual benchmark (EarnN) where everyone has the same economic status and political status was randomly determined, the competition for economic status (EarnE) on average decreased the average of political participation by approximately 15% (33% vs 28%, z = 2.535, p = 0.011), while the competition for political status (EarnP) on average increased voter turnout by 52% (33% vs 50%, z = 8.545, p < 0.001). The different effects on aggregate voting can be seen in Figure 1. Average E1 E2 Averages of Voting.2.3.4.5.6 0.33 0.28 0.50 0.42 Averages of Voting.2.3.4.5.6 0.30 0.24 0.51 0.42 Averages of Voting.2.3.4.5.6 0.35 0.32 0.49 0.43 EarnN EarnE EarnP EarnEP EarnN EarnE EarnP EarnEP EarnN EarnE EarnP EarnEP Figure 1: Average Treatment Effects on Voting Result 1 Whether an individual could compete for economic status (as in EarnE) or political power (as in EarnP) has a strong but differential effect on political participation. Competing for better economic status significantly decreases voter turnout; competing for more political power significantly increases voter turnout. Support When I break the analysis down by group and election, I find the same result. In E1, allowing individuals to compete for economic status significantly decreased A s voting from 44% to 24% (z = 5.602, p < 0.001), and in E2 it decreased it from 38% to 28% (z = 3.517, p < 0.001). As for Group B voters, the difference of voting between EarnE 15
and EarnN is statistically indistinguishable. 4 However, allowing individuals to earn more political power (as in EarnP) significantly increased political participation. Compared to EarnN, A s voting was increased from 24% to 51% in E1 (z = 2.175, p = 0.015) and from 28% to 46% in E2 (z = 2.314, p = 0.010), B s voting was increased from 21% to 50% in E1 (z = 6.161, p < 0.001) and from 33% to 50% in E2 (z = 4.252, p < 0.001). In EarnEP, the interactive effect of competition for both economic and political status seems to be dominated by the effect of competition for political power. For Group A voters, A s voting in EarnEP was more than it was in EarnE. This effect was significant both in E1 (47% vs 24%, z = 5.639, p < 0.001) and in E2 (48% vs 28%, z = 6.295, p < 0.001). However, compared to EarnP, A s voting in EarnEP was statistically undistinguishable in E1 (47% vs 51%, z = 1.169, p = 0.242) and in E2 (48% vs 46%, z = 1.200, p = 0.115). For Group B voters, they voted for Policy B 38% of the time in EarnEP in E1 and 39% in E2, which was significantly greater as compared to EarnE (in E1 z = 5.267, p < 0.001; in E2 z = 1.873, p = 0.031) and significantly smaller as compared to EarnP (in E1 z = 1.970, p = 0.024; in E2 z = 3.033, p = 0.001). These results suggest a non-zero treatment effect on voting as predicted, which are strong evidence supporting Hypothesis 1. How did the changes of voting translate into the efficiency and quality of policy outcomes? The dynamics of Group A and Group B voters interactions were shaped by the origins of their economic/political status. Questions about the quality and efficiency of an institution thus reduce to: (1) how did institutions, characterized by the relationship between earning more scarce economic or political resources, shape policy outcomes? and (2) in what type of institution was the socially desirable policy likely to win? As noted earlier, Policy B is arguably a socially desirable policy. Based on the experimental data, I estimate the efficiency and quality of policy outcomes as the percentage of times Policy B wins in elections and obtain the following result. Result 2 The frequency of Policy B winning is greater in E2 than in E1. However, Policy 4 For E1, 21% (EarnN) vs 24% (EarnE), z = 0.502, p = 0.308; for E2, 33% (EarnN) vs 34% (EarnE), z =-statistic is 0.142, p = 0.556. 16
B wins significantly less than the normative benchmark, and Policy A wins more frequently than Policy B. Support The theoretical analysis predicts that in E1 Policy B should win with a chance of 57%, but the estimate based on the experimental results is only 40%, which is significantly smaller (t = 10.979, p < 0.001). Policy B is predicted to win with a chance of 81% in E2 but the estimate is about 43%, which is also significantly smaller (t = 33.663, p < 0.001). Policy A won more frequently than Policy B both in E1 (t = 6.438, p < 0.001) and in E2 (t = 6.676, p < 0.001). When comparing the estimates of E1 and E2, the probability of Policy B winning was significantly greater in E2 than E1 (z = 2.888, p = 0.002). This result is in line with the theoretical prediction, but overall Result 2 reports a mixed finding that partially supports Hypothesis 2. Theoretically the socially desirable policy should have a greater chance to win, but the experimental results demonstrate an opposite outcome that undermines the democratic ideal. While Policy B is relatively more likely to win in E2, it is essential to note that the difference between the estimate and the normative benchmark is about 39% in E2 while it is 17% in E1. The difference between differences suggests a significantly greater efficiency loss in E2 (t = 11.312, p < 0.001). Result 2 implies that when a better policy is available (as noted earlier the Policy B in E2 is better than the Policy B in E1), the detrimental effect is more salient, which in turn leads to a quantitatively worse democracy. Result 3 Compared to EarnN, in institutions where individuals can either compete for economic or political status, the socially desirable policy is relatively more likely to be selected. An institution that allows competing for economic status but not political status generates a similar effect on policy outcome as compared to an institution that allows competing for political status but not economic status. However, an institution in which earned economic status guarantees more political power significantly reduces the probability that the socially desirable policy is chosen in elections. Support The importance of the main results, and the different effects responsible for the decline in the probability of choosing Policy B can be seen in Figure 2 in which I report the 17
differences between estimates of Policy B winning in elections by treatment. In E1, compared to EarnN in which Policy B won about 27% of the time, the estimate was significantly greater in EarnE (51%, t = 5.447, p < 0.001) and in EarnP (42%, t = 4.985, p < 0.001). In E2, compared to EarnN in which Policy B won about 37% of the time, the estimate was significantly greater in EarnE (49%, t = 3.429, p < 0.001) and in EarnP (45%, t = 2.934, p = 0.002). The interactive effect of competing for both economic and political status resulted in a negative effect. In E1, it significantly decreased the quality of policy outcomes from 51% in EarnE (and 42% in EarnP) to 40% in EarnEP (t = 2.293, p = 0.012), although the difference between EarnP and EarnEP was statistically undistinguishable (t = 0.582, p = 0.281). In E2, it significantly decreased the quality of the policy outcome from 49% in EarnE (and 45% in EarnP) to 38% in EarnEP (t = 3.568, p < 0.001; t = 3.187, p = 0.001). What creates the observed institutional effects? There are two channels that could increase the chance of Policy B winning. First, when Group B s voting is constant, the decreased Group A s voting increases B s winning. Second, when Group A s voting is constant or when Group B s voting increased sufficiently more than the growth of A s voting, Policy B wins more frequently. Both of the two effects are observed in the experiment. From EarnN to EarnE, B s voting was statistically undistinguishable, but A s voting was significantly decreased because of competition for economic status. From EarnN to EarnP, both A s and B s voted more in elections, but the competition for political status motivated sufficiently more B s voting than A s voting. The difference between EarnEP and EarnP is more nuanced. A s voting in both of the two treatments was identical, but B s voting was less in EarnEP than in EarnP. Group B voters were economically disadvantaged in EarnEP compared to EarnP. This may suggest, ceteris paribus, that when economic inequality guaranteed political inequality, lower income group/class was less likely to participate in politics. Taken together, these differential effects explain why Policy B won more in EarnE and EarnP, meanwhile they indicate clear evidence supporting Hypothesis 3. Was Policy B more likely to win in EarnE than EarnP? If the competition for political 18
vs EarnN (Avg.) vs EarnN (E1) vs EarnN (E2) Effects on Policy B Winning -.1 0.1.2.3 Effects on Policy B Winning -.1 0.1.2.3 Effects on Policy B Winning -.1 0.1.2.3 Effects on Policy B Winning -.1 0.1.2.3 vs EarnEP (Avg.) vs EarnEP (E1) vs EarnEP (E2) Effects on Policy B Winning -.1 0.1.2.3 Effects on Policy B Winning -.1 0.1.2.3 EarnE EarnP 95% CI Zero Difference Note 1: The results are based on two-sided t statistical tests. Note 2: When the zero difference benchmark lies in the 95% confidence interval, effects are not significant. Figure 2: Estimates of Policy B Winning by Treatment and Election status is controlled, would it result in better policy outcomes? The experimental design reported in this study could not make causal claims to answer these questions. However, if we concentrate on the estimate in the context of this study, we can find some suggestive information that may be relevant to answering these inquires. In E1, Policy B won marginally more (t = 1.875, p = 0.063) in EarnE (51%) than EarnP (42%), but in E2, the difference between EarnE (49%) and EarnP (45%) was statistically undistinguishable (t = 1.302, p = 0.195). This suggests that quantitatively the competition for political status yields an effect on policy outcomes that has few to no difference as compared to the effect of competition for economic status. It follows that what determines the efficiency and quality of policy outcomes is the relationship between economic status and political status. It is important to understand the exact comparisons of efficiency and welfare between the two institutions 19
(EarnE and EarnP) in future studies. This study takes an important step toward this direction. Measuring Value on Status The heart of this study lies in an assumption that human beings place more value on earned status, as noted in the conceptual framework. In this section, I analyze the effect of value on political preference and voting behavior. I consider a simple logit model and incorporate a value component into a voter s utility function. 5 To illustrate, define π vote as the expected monetary payoff to a voter from voting, and π abstain as the expected monetary payoff to a voter from abstaining, respectively. Let ν denote the value that an individual puts on the payoff of voting, where ν is a function of whether political status is earned. Then, a voter s utility function is represented by: u i (π vote, π abstain ) = { π vote + ν π abstain if voted if abstained (2) I use the experimental data to estimate the parameters of Equation (2). The maximumlikelihood estimation on the binary-response results uses a logit specification: Pr(Vote) = exp{γ(π vote + ν)} exp{γ(π vote + ν)} + exp{γ(π abstain )} where the parameter γ reflects the sensitivity of the choices to utility differences. When γ = 0, this model is reduced to a random choice model with equal probability. When γ is arbitrarily large, the probability of choosing the action with higher utility approaches one (McFadden, 1981). Table 3 reports the results of the parameter estimation by election and treatment. To investigate whether ν is significantly different from zero, I estimate the case when ν = 0 and compare the estimations of voting between ν = 0 with ν estimated. The results are reported in Appendix B.6. The estimations that include ν not surprisingly provide a better fit for the data than the case where ν = 0, since the χ 2 statistic of a likelihood ratio 5 More details of the model are reported in Appendix B.6. 20
Table 3: Structure Estimations of γ and ν E1 E2 EarnN EarnE EarnP EarnN EarnE EarnP A s Voting Obs. 0.44 0.24 0.51 0.38 0.28 0.46 A s Voting Est. 0.39 0.26 0.50 0.37 0.32 0.46 B s Voting Obs. 0.21 0.24 0.50 0.33 0.34 0.50 B s Voting Est. 0.23 0.23 0.51 0.34 0.31 0.50 γ 1.90 0.40 0.10 1.20 0.30 0.40 ν -0.30-2.75 0.05-0.45-2.45-0.15 log-likelihood -351.61-331.04-415.87-387.39-374.36-415.21 Note: The payoff values were rescaled by 100 times to facilitate the convergence in estimations. test for EarnN and EarnE rejects the null hypothesis that ν equals zero at the 1 percent level. While the null hypothesis for EarnP cannot be rejected, it is because the estimations of ν are sufficiently close to zero when both γ and ν are endogenous and estimated simultaneously. To explore the effects of competing status on ν, I constrained γ at the estimated value when all treatments are combined and considered whether I could reject the null hypothesis that ν is unaffected by changes of institutions. For E1, the statistic equals 68.12 for the comparison between EarnN and EarnP, which rejects the null hypothesis at the 1 percent level. Compared to EarnN, the statistic equals 3.7 for EarnP, which also rejects the null hypothesis but at a marginal level (10%). For E2, the statistic equals 27.16 for the comparison between EarnN and EarnP and 43.05 for the comparison between EarnN and EarnE; both of which also can be rejected at the 1 percent level. The analysis provides evidence of the selective valuation assumption. These results suggest that subjects place an intrinsic value on voting and they place relatively higher value on voting when they earned political status, which supports the theoretical foundation of this study. Heterogeneity For different participants, they may have different inborn traits or their desires for political power could be heterogenous. To address this concern, I conducted a robustness-check experiment with 20 subjects over two sessions. The robustness-check experiment was conducted in a comparable way to the implementation of the principal experiment. Subjects 21
were asked to engage in the same counting-zero task for the Task Stage and the same voting game for the Voting Stage. In the new experiment, I focused on Election E1 and treatments EarnEP and EarnE. In each session, the same group of subjects engaged in both of the two treatments so that I could control for unobservable confounding factors. More details of the robustness check experiment is reported in Appendix B.7. In general I find a qualitatively similar result that is consistent with the finding reported earlier. Specifically, Group A voters voted for Policy A 55% of the time in EarnEP but 29% in EarnE, which is a significant difference (Z = 4.562, p < 0.001) and consistent with the main findings. Similarly, Group B voters voted for Policy B about 42% of the time in EarnEP but 35% in EarnE (Z = 1.795, p = 0.036). The estimate of the probability that Policy B wins in EarnEP is about 38% but about 63% in EarnE (Z = 3.655, p < 0.001), which suggests that Policy B was significantly less likely to win in EarnEP than in EarnE. Given that the main results continue to hold with the addition of heterogeneity controls, it is safe to conclude that the main findings are robust to heterogeneity. One important methodological implication of these results is that in experiments where subjects compete for political status, the competition process may introduce unintentional distortion in the outcomes. That is, those who worked the hardest to be a member of the privileged group may have different preferences due to selection, which may lead to a misinterpretation of the results. The results reported in this robustness check experiment may largely reduce this concern. External Validity The main findings arise from a counterfactual established in the laboratory and administered to a sample of student subjects. It is important, therefore, to discuss the extent to which one can generalize from the results of this paper. The external validity of the main findings can be approached at three levels of generality: First, I identify the existence of a (non-zero) causal effect of competing for political status on voter turnout and the qualify of democratic choice. This finding should be seen as providing empirical support for a broad class of theoretical models that emphasize the institutional 22