Inequality and redistribution behavior in a give-or-take game

Abstract

Political polarization and extremism are widely thought to be driven by the surge in economic inequality in many countries around the world. Understanding why inequality persists depends on knowing the causal effect of inequality on individual behavior. We study how inequality affects redistribution behavior in a randomized "give-or-take" experiment that created equality, advantageous inequality, or disadvantageous inequality between two individuals before offering one of them the opportunity to either take from or give to the other. We estimate the causal effect of inequality in representative samples of German and American citizens (n = 4,966) and establish two main findings. First, individuals imperfectly equalize payoffs: On average, respondents transfer 12% of the available endowments to realize more equal wealth distributions. This means that respondents tolerate a considerable degree of inequality even in a setting in which there are no costs to redistribution. Second, redistribution behavior in response to disadvantageous and advantageous inequality is largely asymmetric: Individuals who take from those who are richer do not also tend to give to those who are poorer, and individuals who give to those who are poorer do not tend to take from those who are richer. These behavioral redistribution types correlate in meaningful ways with support for heavy taxes on the rich and the provision of welfare benefits for the poor. Consequently, it seems difficult to construct a majority coalition willing to back the type of government interventions needed to counter rising inequality.

Keywords: democracy; experiment; inequality; policy preferences; redistribution.

Fig. 1.

Advantageous (a-)inequality (own richer), equality, and disadvantageous (d-)inequality (own poorer) cause different types of redistribution behavior as measured by the $/€ taken/given in the (A) pooled data, the (B) United States, and (C) Germany. Error bars indicate 95% confidence intervals calculated from robust standard errors. All differences are statistically significant (P < 0.001). n(total) = 4,966; n(United States) = 2,749; n(Germany) = 2,217.

Fig. 2.

The distributions of individual aversion to a-inequality (white bars) and d-inequality (gray bars) in the give-or-take game differ significantly from each other in the pooled data (A). The distributions of (B) disadvantageous and (C) advantageous inequality aversion also differs significantly between Germany (white bars) and the United States (gray bars). The results are based on a Kolmogorov–Smirnov test of the null hypothesis of no difference between the distributions. The inequality aversion parameters are estimated in a linear regression of the amount taken/given on the difference between individuals’ gift card values in the give-or-take game using respondents’ conditional redistribution schedules (SI Appendix, Fig. S4). See Estimating Individual-Level Aversion to Inequality and SI Appendix, Materials and Methods for details on the underlying estimation procedure. n(Pooled) = 4,796; n(United States) =2,645 (d-inequality), 2,735 (a-inequality); n(Germany) = 2,170 (d-inequality), 2,208 (a-inequality).

Fig. 3.

Redistribution type predicts variation in policy views. Shown are marginal effects of (A) d-equalizer and (B) a-equalizer redistribution types on policy views compared with nonequalizers in the pooled data, United States, and Germany. We use a linear regression to model policy views as a function of redistribution type (using binary indicator variables) and a full set of sociodemographic and political covariates as well as country-fixed effects (SI Appendix, Tables S13–S17 report the underlying estimates in detail). Policy views are measured on a 5-point scale (strongly disagree–strongly agree). Dots with vertical lines indicate point estimates with robust 95% confidence intervals. n(total) = 4,921, n(United States) = 2,733, and n(Germany) = 2,188.