برداشت جانبدارانه از توزیع درآمد و ترجیحات برای توزیع مجدد: شواهدی از یک آزمایش بررسی

Individual perceptions of income distribution play a vital role in political economy and public finance models, yet there is little evidence regarding their origins or accuracy. This study examines how individuals form these perceptions and explores their potential impact on preferences for redistribution. A tailored household survey provides original evidence on systematic biases in individuals' evaluations of their own relative position in the income distribution. The study discusses one of the mechanisms that may generate such biases, based on the extrapolation of information from endogenous reference groups, and presents some suggestive evidence that this mechanism has significant explanatory power. The impact of these biased perceptions on attitudes toward redistributive policies is studied by means of an experimental design that was incorporated into the survey, which provided consistent information on the own-ranking within the income distribution to a randomly selected group of respondents. The evidence suggests that those who had overestimated their relative position and thought that they were relatively richer than they were tend to demand higher levels of redistribution when informed of their true ranking.

The shape of the income distribution plays a key role in the determination of policies with redistributive components (such as those dealing with social security, health care, government transfers and taxation) in political economy models. However, the main policy determinant is not its actual shape, but rather how it is perceived by agents in the economy. Additionally, individuals' perceptions of the income distribution can affect how they will react to redistributive policies (for instance, through tax evasion), which is a key input for public finance models. This study fills a gap in the literature by exploring the origins and consequences of systematic biases in individuals' perceptions of aggregate income distributions. The findings presented in this paper contribute to the recent literature on the incorporation of subjective perceptions and inference problems into the determination of political economy outcomes (for a seminal contribution, see Piketty, 1995). For instance, when forming their views on public policies, agents may need to infer the importance of effort and of predetermined factors in the income generation process (Piketty, 1995 and Bénabou and Tirole, 2006); in so doing, they may evaluate the prospects for economic mobility (Bénabou and Ok, 2001 and Alesina and La Ferrara, 2005), or they may arrive at conclusions as to the causes of poverty and the fairness of socioeconomic outcomes in general (Alesina and Glaeser, 2004). To form their judgments, views and attitudes, agents need to make complicated inferences about distributional outcomes (e.g., inequality, mobility) based on limited information and within given time constraints, but there is as yet little evidence on the origins or the accuracy of the inferences they make in this regard. This paper also makes a contribution to a growing body of work that attempts to document agents' expectations and subjective probabilities (Manski, 2004 and Hurd, 2009) and to explain how they are formed (Zafar, 2011). In an application to distributional issues, Norton and Ariely (2011) elicit information on Americans' perceptions of the wealth distribution in their society and find significant discrepancies between actual and perceived levels of inequality. This paper documents systematic differences between objective and subjective income distributions and sheds light on the origins of these discrepancies. Most importantly, an experimental design makes it possible to test whether correcting these biases has an impact on stated preferences for redistribution. The assessment of an income distribution by an economic agent can be regarded, fundamentally, as a statistical inference problem. Individuals observe the income levels of no more than a sub-sample of the population and must then infer the entire distribution from that information. If agents do not fully account for the selection process involved in the formation of the sample that they observe, their inferences will be systematically biased. This failure may be due to limitations in the available information set which arise from the fact that information may be costly or difficult to obtain. Alternatively, agents may have the necessary information, but they may sometimes fail to use it correctly, as argued in the cognitive bias literature. Irrespective of whether agents have limited information or bounded rationality, this rationalization of distributional perceptions provides a series of corollaries that can be tested with data on objective and perceived distributions. The same data can be used to study biases and preferences for redistribution, which is the main focus of this study. The empirical results presented in this paper are based on the Survey on Distributional Perceptions and Redistribution, a study of 1100 representative households in Greater Buenos Aires in Argentina. The survey was designed and implemented in 2009 for the specific purpose of testing the posited mechanisms for the formation of distributional perceptions. Data were collected on each respondent's household income and on his or her assessment of its ranking (to the closest decile) in the overall income distribution. The first finding is that systematic biases are present in perceptions of own income rank: a significant portion of poorer individuals place themselves in higher positions than they actually occupy, while a significant proportion of richer individuals underestimate their rank. Moreover, as predicted, the bias is significantly correlated with the respondents' relative positions within the reference group (as proxied by area of residence). Also in keeping with the posited mechanisms, respondents with friends from heterogeneous social backgrounds are less prone to these biases. Finally, the study explores how these misperceptions about the income distribution may affect attitudes toward redistribution. For instance, self-interest might induce poor individuals to demand less redistribution if they think they are relatively richer than they actually are. This study presents the results from a unique randomized experiment that was implemented within the survey: for a randomly assigned treatment group, the interviewer highlighted any discrepancy between the subjective assessment of the respondent's ranking and that respondent's actual position, effectively correcting any bias that was present. This survey field experiment contributes to the literature on information provision as a treatment (Duflo and Saez, 2003, Chetty and Saez, 2009 and Card et al., 2010). An original feature is that perceptions are not only contrasted with reality (as in Olken, 2009, among others). In addition, in this survey experiment, biased subjects were provided with feedback and were actually confronted with accurate information. The results from the experiment indicate that confronting agents' biased perceptions with accurate information had a significant effect on their stated preferences for redistribution. Those who underestimated their income ranking did not change their attitudes toward redistribution when provided with accurate information about their income ranking. However, those who overestimated their relative position (i.e., who thought that they were relatively richer than they are) and who were provided with accurate information demanded more redistribution than those in the control group. To the degree that the information treatment managed to correct biased distributional perceptions, these results can be interpreted as evidence of the effect of biases in distributional perceptions on political attitudes. This finding constitutes an alternative to theories that posit prospects of upward mobility (Bénabou and Ok, 2001) or other factors as accounting for the relatively low levels of demand for redistribution in modern democracies. This paper is organized as follows. The next section discusses the formation of subjective income distributions and individuals' perceptions of their income rank and then goes on to explore these factors' implications for attitudes toward redistribution. The third section describes the household survey and outlines the randomized experiment that was designed to answer these questions. The fourth section presents the empirical results on biased perceptions of income distribution, and the fifth section describes the identification strategy and the results from the experiment on biases and preferences for redistribution. The last section concludes.

The motivation for undertaking this study was the apparent lack of evidence regarding the accuracy and origins of perceptions of the income distribution, which play a crucial (though implicit) role in political economy and public finance models. The study focuses on a simple mechanism, whereby agents extrapolate from their reference group without accounting for the selection process underlying the formation of the group owing to either informational or cognitive limitations. A tailored household survey provided evidence of the presence of sizeable systematic biases in perceptions about relative income in Argentina. Furthermore, the analysis uncovered suggestive evidence that the incorrect extrapolation of information from reference groups can be a powerful explanation for the observed biases. The systematic biases documented in this paper and their consequences can arise in any society. This study involved the implementation of an original survey experiment in the field in which a randomly assigned group of respondents was provided with accurate information about their ranking in the income distribution as a form of feedback concerning their responses. Confronting agents who had biased perceptions with this information had a significant effect on their stated preferences for redistribution: those who overestimated their relative position (who thought that they were relatively richer than they were) demanded more redistribution. To the degree that the information treatment managed to correct biased distributional perceptions, these results can be interpreted as evidence of the causal effect of misperceptions on political attitudes. This mechanism provides an alternative explanation for the low levels of redistribution observed in modern democracies.26 Having accurate information about the income distribution might induce agents to better calibrate their demands for redistribution. The results in this paper support Romer's (2003) discussion of the possible welfare-improving effects of subsidizing information and Besley's (2007) remarks about the potential of information for improving policies, although the impact of the biases in the efficiency of redistribution should also be considered (Acemoglu and Robinson, 2001). The role of misconceptions in political economy has been studied before (Romer, 2003 and Slemrod, 2006). While Besley (2007) highlights the benefits to be derived from incorporating notions of dispersed and limited information for modern political economy, building-in more specific factors, such as biased perceptions of the distribution, can further enrich political economy models and empirical applications. It can also provide explanations for other puzzles in the literature, such as those pointed out by Bartels' (2008) results on the reduced responsiveness of representatives to low-income voters. More generally, concepts such as inequality, self-interest and the median voter can be adapted in their application to political economy outcomes when misperceptions and misconceptions play a role. The findings in this paper indicate how perceptions may affect stated preferences for redistribution. Further research could focus on the impact of biases and information on actual behavior, such as voting patterns. Moreover, it would be interesting to explore how misperceptions affect individuals' reactions to redistributive policies (e.g., as expressed through charitable contributions and tax evasion) and to learn whether the provision of information on broader issues that go beyond such matters as rules and regulations (e.g., Chetty and Saez, 2009) may have implications for public finance models. Finally, the results of the analyses conducted in connection with this study could be attributable either to limited information or to limited cognitive ability — further research will be needed in order to pinpoint the source of the observed biases in distributional perceptions.