CESifo Working Paper no. 6251 econstor Make Your Publications Visible. A Service of zbw Leibniz-Informationszentrum Wirtschaft Leibniz Information Centre for Economics Roth, Christopher; Wohlfart, Johannes Working Paper Experienced Inequality and Preferences for Redistribution CESifo Working Paper, No. 6251 Provided in Cooperation with: Ifo Institute – Leibniz Institute for Economic Research at the University of Munich Suggested Citation: Roth, Christopher; Wohlfart, Johannes (2016) : Experienced Inequality and Preferences for Redistribution, CESifo Working Paper, No. 6251, Center for Economic Studies and ifo Institute (CESifo), Munich This Version is available at: http://hdl.handle.net/10419/149338 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. www.econstor.eu Experienced Inequality and Preferences for Redistribution Christopher Roth Johannes Wohlfart CESIFO WORKING PAPER NO. 6251 CATEGORY 2: PUBLIC CHOICE DECEMBER 2016 An electronic version of the paper may be downloaded • from the SSRN website: www.SSRN.com • from the RePEc website: www.RePEc.org • from the CESifo website: Twww.CESifo-group.org/wp T ISSN 2364-1428 http://www.ssrn.com/ http://www.repec.org/ http://www.cesifo-group.de/ CESifo Working Paper No. 6251 Experienced Inequality and Preferences for Redistribution Abstract We examine in how far people’s experiences of income inequality affect their preferences for redistribution. We use several large nationally representative datasets to provide evidence that people who have experienced more inequality while growing up are less in favor of redistribution, after controlling for income, demographics, unemployment experiences and current macro-economic conditions. They are also less likely to consider the prevailing distribution of incomes to be unfair, suggesting that inequality experiences affect the reference point about what is a fair division of overall income. Finally, we conduct an experiment to show that individuals randomly exposed to environments promoting inequality in the experience stage of the experiment redistribute less in a subsequent behavioral measure. JEL-Codes: P160, E600, Z130. Keywords: inequality, redistribution, macroeconomic experiences, experiment. Christopher Roth Department of Economics University of Oxford & CSAE Manor Road Building, Manor Road, United Kingdom - Oxford OX1 3UQ christopher.roth@economics.ox.ac Johannes Wohlfart Department of Economics Goethe University Frankfurt Theodor-W.-Adorno-Platz 3, PF H32 Germany - 60323 Frankfurt am Main wohlfart@econ.uni-frankfurt.de December 7, 2016 We would like to thank Alberto Alesina, Heike Auerswald, Peter Bent, Enzo Cerletti, Anujit Chakraborty, Ester Faia, Eliana La Ferrara, Nicola Fuchs-Schündeln, Ingar Haaland, Michalis Haliassos, Emma Harrington, Michael Kosfeld, Ulrike Malmendier, Salvatore Nunnari, Matthew Rabin, Sonja Settele, Guido Tabellini, Bertil Tungodden, Ferdinand von Siemens and seminar participants at Goethe University Frankfurt and ZEW Mannheim as well as participants at the 10th Workshop on Political Economy at ifo Dresden for helpful comments. We also thank Ulrike Malmendier and Stefan Nagel for sharing code. 1 Introduction Understanding the origins of individuals' preferences for redistribution is a key question in po- litical economy. People's demand for redistribution in�uences the levels of government spending and taxation and thereby a�ects the degree of economic inequality. For example, people's taste for redistribution can explain di�erences in the generosity of welfare systems between European countries and the US (Alesina et al., 2001). Experiences of adverse macroeconomic conditions, such as recessions, have been shown to be an important determinant of redistributive preferences (Giuliano and Spilimbergo, 2014). At the same time, people's aversion to inequality has been singled out as a key factor in shaping redistributive choices (Fehr and Schmidt, 1999). However, no evidence exists on how experiences of inequality a�ect people's aversion to inequality and their demand for redistribution. In this paper we examine how growing up in times of high income inequality a�ects views on inequality and preferences for redistribution. On the one hand, experiencing inequality could make people more used to an unequal distribution of incomes and therefore lower their taste for redistribution. On the other hand, people who have lived through times of high inequality could be particularly aware of potential adverse e�ects of inequality and could be more in favor of redistribution. Our evidence comes from several nationally representative datasets: the US General Social Survey, the German General Social Survey as well as the European Social Survey.1 To mea- sure our respondents' experiences of income inequality, we focus on the average level of income inequality that prevailed in their country while they were between 18 and 25 years old. This period of life is sometimes referred to as the �impressionable years� and has been identi�ed as particularly important for the formation of political attitudes and beliefs (Giuliano and Spilim- bergo, 2014; Krosnick and Alwin, 1989; Mannheim, 1970). Speci�cally, we calculate, for each birth-cohort in our datasets, the share of total income held by the top �ve percent of earners2 during their impressionable years.3 We show that our results are robust to using alternative 1While our main evidence comes from the United States and Germany, we also leverage data from a variety of other OECD countries, such as Canada, Denmark, Finland, France, Italy, the Netherlands, Norway, Portugal, Sweden, Switzerland and the United Kingdom. We also replicate our main �ndings using the International Social Survey Program (ISSP) on Social Inequality. 2Top income shares are very commonly used measures of income inequality. The inequality data are taken from �The World Wealth and Income Database� (Alvaredo et al., 2011). 3Our results are robust to using alternative measures of income inequality, namely the share of total income held by the top ten percent of earners, the share of total income held by the top one percent of earners, as well as the Gini coe�cient of equivalized household income. 1 measures of income inequality experiences following the methodology in Malmendier and Nagel (2011). In all of our main speci�cations we control for age �xed e�ects and year �xed e�ects, i.e. we identify our key coe�cient of interest making use of between cohort di�erences in inequality experiences within age groups and years. By including age �xed e�ects, we rule out that our �ndings result from changes in preferences over people's life-time, for example by people becoming more conservative as they get older. The inclusion of year �xed e�ects ensures that our results are not driven by common shocks that a�ect everyone in a given year. In addition, we control for cohort-group �xed e�ects (cohort group brackets of 25 years) which mitigates the concern that our �ndings are driven by di�erences in political attitudes across cohorts associated with longer-term changes in zeitgeist.4 We also control for income and a number of socioeconomic characteristics as well as the national unemployment rate people experienced in their impressionable years which could be correlated with inequality experiences. Across datasets, we provide evidence that individuals who witnessed high levels of income inequality in their impressionable years are less in favor of redistributive policies and are less likely to identify with and to vote for left-wing parties. They are also more likely to believe that inequality increases motivation and that inequality arises due to di�erences in e�ort rather than luck.5 We also �nd that people who have grown up in times of high inequality are less likely to consider the prevailing distribution of incomes to be unfair, suggesting that inequality experiences alter someone's reference point about what is a fair division of resources.6 Combined, our �ndings suggest that being used to an unequal distribution of incomes lowers people's distaste for inequality and reduces their demand for redistribution. To provide causal evidence that experiencing inequality can a�ect people's reference points about the approporiate amount of redistribution and thereby alter their re-distributive pref- erences, we conduct an online experiment on Amazon Mechanical Turk. In the experiment respondents make a hypothetical distributive choice for two other individuals. Respondents are told that these two individuals have previously completed di�erent numbers of tasks for us on MTurk.7 In the �rst stage of the experiment we randomly assign our respondents either to an 4Since we control for both age and year �xed e�ects, we cannot also include dummies for every individual cohort (Campbell, 2001). In addition, inequality experiences vary at the cohort-level, which prohibits separate identi�cation of unrestricted cohort e�ects. 5This evidence on changes in beliefs is in line with the seminal theoretical work by Piketty (1995) who argues that economic circumstances could alter a person's belief about the drivers of success. 6We provide evidence that inequality experiences exert the largest in�uence on political attitudes and beliefs during the impressionable years as compared to other periods in our respondents' lives. 7This is related to the behavioral measure employed by Almås et al. (2016) and others. 2 inequality condition or to an equality condition. Individuals in the inequality condition choose between two options that result in highly unequal outcomes for the other workers, while people in the equality condition choose between two options that result in more equal outcomes for the other workers. In the second stage our respondents are again asked to distribute money between two other workers, but this time they all face the same choice set of potential payo�s to the two workers. In line with the observational evidence, we �nd that individuals who have experienced in- equality in the �rst stage of the experiment are less likely to redistribute in stage two compared to people in the equality condition. Since participants are in the role of a spectator who observes inequality between two other workers, our design rules out channels that work through the par- ticipants' own outcomes. This experimental evidence highlights that exposure to an institutional environment that promotes inequality can in�uence people's reference point about what is a fair division of resources and thereby a�ect people's preferences for redistribution. We also use the observational data to examine alternative channels through which experienc- ing inequality could a�ect beliefs and redistributive preferences. First, we test whether people form their redistributive preferences based on the e�ect inequality had on them personally. It could be the case that only people who personally bene�ted from high inequality while growing up adjust their redistributive preferences. The e�ects are not signi�cantly di�erent for individuals with better starting conditions or more success in life, providing evidence against this mecha- nism. Second, we show that the e�ects are unlikely to operate through changes in perceived social status. To provide evidence against the possibility that our e�ects are driven by cohort-speci�c changes in zeitgeist accompanied with changes in general political preferences, we conduct a se- ries of placebo tests. We provide evidence that inequality experiences do not a�ect how conser- vative individuals are in matters unrelated to redistribution and inequality, such as nationalism, attitudes towards democracy, attitudes towards guns or attitudes towards immigrants. This is consistent with our interpretation that inequality experiences are driving the changes in redis- tributive preferences, rather than picking up more general di�erences in political attitudes across cohorts. Moreover, we demonstrate the robustness of our results to controlling for other experiences during people's impressionable years, such as experiencing a crisis (Giuliano and Spilimbergo, 2014), the experienced growth rate of real per capita GDP, the experienced political ideology of 3 the leading party as well as the experienced size of the government. Our results barely change after controlling for these other experiences, indicating that omitted variable bias from other experiences during impressionable years is unlikely. We contribute to a growing literature on the origins and determinants of redistributive pref- erences (Alesina et al., 2013; Durante et al., 2014; Alesina and Giuliano, 2010; Alesina and La Ferrara, 2005) and beliefs about inequality (Piketty, 1995).8 Researchers have established that redistributive preferences are in�uenced by culture (Luttmer and Singhal, 2012; Alesina and Giuliano, 2010), political regimes (Alesina and Fuchs-Schuendeln, 2007; Pan and Xu, 2015), relative income (Karadja et al., 2016; Cruces et al., 2013) and historical experiences (Chen et al., 2016; Roland and Yang, 2016).9 Our paper is most closely related to Giuliano and Spilimbergo (2014) who show that individ- uals who have experienced a recession in their formative years believe that success in life depends more on luck than e�ort, support more government redistribution, and tend to vote for left-wing parties. Our paper shows that people's experiences of unequal distributions of incomes matter on top of the e�ects of experiencing a crisis. We also contribute to the literature on the relationship between inequality and the demand for redistribution. Inequality and preferences for redistribution are negatively correlated in the aggregate across countries, but this pattern vanishes when looking at within-country movements of inequality (Kenworthy and McCall, 2008). Changes in aggregate inequality in a country could a�ect the average demand for redistribution through various channels, such as changes in incomes of di�erent groups, changes in beliefs about social mobility or fairness concerns.10 Kerr (2014) �nds that short-run increases in inequality within countries or within U.S. regions are associated with greater acceptance for wage di�erentials but also with higher support for redistributive policies at the individual level, conditional on individual characteristics. We identify the e�ect of growing up in times of high inequality conditional on e�ects of contemporaneous inequality that are common across cohorts by including year �xed e�ects. Our �ndings highlight a channel through which long-run trends in inequality could be re�ected in the average demand for redistribution in a country. When there is a long-term increase in inequality, younger generations could be more used to this inequality and exhibit lower distaste for 8More generally, our paper is related to the literature on endogenous preferences (Nunn and Wantchekon, 2011; Kosse et al., 2016; Becker et al., 2016; Schildberg-Hörisch et al., 2014). 9For excellent reviews, see Alesina and Giuliano (2010) and Nunn (2012). 10Fairness concerns are commonly modeled as inequality aversion, i.e. the idea that people have a distaste for unequal distributions of income. 4 it relative to older generations. These e�ects could either amplify a potential negative relationship between long-run changes in inequality and preferences for redistribution, or they could mitigate a positive relationship between the two. Our paper is also related to Kuziemko et al. (2015) who �nd that preferences for redistribu- tion do not respond to information about inequality. We extend their paper in two ways: �rst, we provide �eld evidence that people's experiences of inequality a�ect their preferences for redis- tribution; second, we show that people's exposure to an institutional environment that gives rise to inequality can change people's view on what distribution of resources is fair. Our experimental �ndings are also related to recent work by Charité et al. (2015) who show that reference points matter for people's redistributive choices when subjects are given the opportunity to redistribute unequal, unearned initial endowments between two anonymous recipients. At a more general level, our paper also complements the growing literature on the e�ects of lifetime experiences on belief formation and economic behavior (Hertwig et al., 2004; Nisbett and Ross, 1980; Weber et al., 1993). For instance, Malmendier and Nagel (2011) provide evidence that having experienced negative macroeconomic shocks makes people less likely to invest in stocks. Moreover, Malmendier and Nagel (2015) show that people's experienced in�ation rates predict their contemporaneous in�ation expectations. Fuchs-Schuendeln and Schuendeln (2015) provide evidence that people's experience of living in a democracy makes them more likely to support democratic regimes. Our paper contributes to this literature by highlighting that experiences of income inequality alter people's views about fairness and distributive justice and that they shape people's political preferences and beliefs. More generally, our paper highlights that life-time experiences could a�ect preferences and beliefs later in life by changing people's reference-points. At a method- ological level, our paper is the �rst one that provides experimental evidence that experiences could a�ect preferences and beliefs through a reference point channel. The paper proceeds as follows: Section 2 describes the data. In section 3, we present the main results of our analysis. Section 4 highlights potential mechanisms and we conduct a series of robustness checks in section 5. Section 6 presents the experimental design and the experimental results. Finally, the paper concludes. 5 2 Data 2.1 General Social Survey (US) We leverage rich data on political preferences and beliefs from the General Social Survey (GSS). This dataset consists of repeated cross-sections from 1972 to 2014 that are representative for the US and has been widely used in previous research in economics (Alesina and La Ferrara, 2000; Giuliano and Spilimbergo, 2014). Following Giuliano and Spilimbergo (2014) we focus on questions in which respondents are asked about their preferences for redistribution to the poor. In addition, we examine people's beliefs about the determinants of success in life, in line with the idea that individuals who believe that luck rather than hard work is a major determinant of success are more likely to be in favor of government redistribution (Piketty, 1995). Speci�cally we examine the following measures of redistributive preferences: • Help Poor: People's view on whether the government in Washington should do everything to improve the standard of living of all poor Americans or whether it is not the government's responsibility, and that each person should take care of himself. • Pro-welfare: People's opinion on whether the government is not spending enough money on assistance to the poor. • Success due to luck: People's view on whether success is mostly due to luck or owing to hard work. We also consider people's self-placement on a conservative-liberal scale, their party a�liation, and their self-reported past voting behavior. We examine whether people identify more as Democrat or Republican and whether they report having voted for Democrats or Republicans. We code all variables such that high values mean that they are more in favor of redistribution and more likely to vote for Democrats. We also use questions that allow us to shed light on the mechanisms behind our �ndings. We look at people's self-assessed social and economic position in society. In Appendix C, we provide more details on these variables. In Table A13 we display the summary statistics for our sample from the General Social Survey. 2.2 German General Social Survey The German General Social Survey (Allbus) collects data on political attitudes and behavior, as well as a large set of demographics in Germany. Every two years since 1980 a representative 6 cross section of the population is surveyed using both constant and variable questions. We use data from the waves from 1980 to 2014. The previous literature emphasizes that support for redistribution depends on people's beliefs about the sources of economic inequality (Benabou and Tirole, 2006; Alesina et al., 2001; Alesina and Angeletos, 2005; Fong, 2001). The German General Social Survey contains unique data on views about the sources and consequences of inequality: • Inequality is Unfair: People's opinion on whether the social inequalities prevailing in Germany are unfair. • Inequality does not increase motivation: People's beliefs about the e�ect of inequality on people's motivation. • Inequality re�ects luck: People's disagreement to the statement that di�erences in rank between people are acceptable as they essentially re�ect how people used their opportuni- ties. We code the variables such that high values stand for less favorable attitudes to inequality. In addition, we focus on outcomes that are similar to the outcomes we use in the General Social Survey. Speci�cally, we look at political behavior as measured through voting intentions, self- reported past voting beavior and people's self-assesment on a political scale. These variables are described in detail in Appendix C. In Table A14 we show summary statistics for our sample from the Allbus. 2.3 European Social Survey The European Social Survey (ESS) is a dataset containing rich information about political atti- tudes, beliefs and behavioral patterns of the various populations in Europe. It also contains data on a rich set of demographic variables. The ESS has been widely used to study redistributive preferences, see for example Luttmer and Singhal (2012). We make use of all available waves from the ESS (2002-2014).1112 Our key outcome variables of interest are a measure capturing whether people are in favor of redistribution as well as people's self-reported voting behavior and their self-placement on a 11Most of our sample from the ESS comes from three countries: France, Germany and the United Kingdom, each of which makes up for around 20 percent of the sample. Denmark, Finland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden and Switzerland all together constitute about 40 percent of the overall sample. 12Due to lacking inequality data we drop all respondents currently living in Eastern Germany and focus only on Western German Respondents. 7 political scale. As in the other datasets, we code all outcome variables such that higher values represent more left-wing views. These outcomes are described in more detail in Appendix C. In Table A15 we provide summary statistics for our sample from the ESS. 2.4 Normalizations, Controls and Missings The outcome variables we use in our analysis are mostly self-placements between left and right or between agreement and disagreement to a particular statement on 4-point, 5-point or 10-point scales. We normalize all outcome variables as well as all experience variables using the mean and the standard deviation of the respective variables in our �nal samples of interest. These normalizations enable us to compare e�ect sizes across outcomes and across datasets. We construct a consistent set of controls for key demographics, such as income, gender, marital status, education, religious a�liation and employment status for all of the datasets of interest. In Appendix D we describe the exact controls we include for each of the di�erent datasets.13 2.5 Inequality and Unemployment Data We use data on top income shares from the �The World Wealth and Income Database� (WID) (Alvaredo et al., 2011) which is the most extensive data source of internationally comparable measures of income inequality. The database contains very rich data on the share of overall national income earned by people at the top of the distribution. We focus on the share of total gross income earned by the top 1, the top 5 and the top 10 percent of earners respectively. We also make use of data on the Gini coe�cient of equivalized disposable household incomes taken from the �Chartbook of Economic Inequality� (Atkinson and Morelli, 2014). For most countries data on the Gini coe�cient are available only from a much later point in time than data on top income shares. In our main analysis we therefore focus on the experienced share of total income earned by the top �ve percent of earners.14 In Appendix E, we provide a detailed overview on the inequality data that are available for each country and the respective cohorts we are able to use in our analysis. 13To deal with missing values and to keep the sample as large as possible, for each of the above categories of controls we code missings as zero and include a dummy variable indicating missing values in that category. 14Note that our data on top income shares refer to total earnings before taxes and transfers, while our data on the Gini coe�cient are based on disposable household income after taxes. The reason for this discrepancy is di�erent data availability between the two measures. We do not consider this aspect material for our analysis, as we focus on one measure within each estimation and because movements in pre- and post-tax inequality are highly correlated. 8 15 20 25 30 35 T op 5 % in co m e sh ar e 1880 1900 1920 1940 1960 1980 2000 2020 Year DE FR UK US Figure 1: Top 5 percent share in total income over time and countries. Figure 1 illustrates the evolution of the share of total income going to the top �ve percent in the largest countries that are part of our sample. We observe quite substantial variation of this measure over the last 100 years, both across countries and over time. 9 20 25 30 35 40 T op 5 % in co m e sh ar e 1900 1920 1940 1960 1980 2000 2020 Year New England Middle Atlantic East North Central West North Central South Atlantic 15 20 25 30 35 40 T op 5 % in co m e sh ar e 1900 1920 1940 1960 1980 2000 2020 Year East South Central West South Central Mountain Pacific Figure 2: Top 5 percent share in total income over time and US census divisions. In addition, we use data on top income shares in US census divisions. Figure 2 shows the evolution of the income share of the top �ve percent in the di�erent census divisions over time. While the trends are similar across regions, there is still substantial variation across regions at any point in time. In our analysis we focus on those countries for which we could obtain historical inequality data from the World Wealth and Income Database. We use linear interpolation to impute data for years in which inequality data are missing. We impute inequality data if the gap between any two data points for which inequality data are available, is at most six years.15 We also use historical data on national unemployment rates from Global Financial Data (GFD) and use the same rule to impute missing values. 15This allows us to use much larger samples of individuals in our analyses. We have made sure that our results are robust to using di�erent maximum gaps for the imputation of the inequality data. 10 2.6 Construction of Experience Variable In most of our estimations we focus on the level of income inequality that our respondents ex- perienced while they were between 18 and 25 years old, an age range sometimes referred to as the �impressionable years�. This age range corresponds to the time when most individuals begin to participate in political life. Previous literature has identi�ed this time period as particularly important for the formation of political attitudes and beliefs. For instance, Krosnick and Alwin (1989) provide evidence that individuals' susceptibility to attitude change is high during the im- pressionable years and drops considerably thereafter. Giuliano and Spilimbergo (2014) �nd that experiencing a recession while aged between 18 and 25 signi�cantly a�ects political preferences later in life, while similar experiences in other age ranges do not seem to matter. Following this literature, we calculate, for each birth-cohort in our datasets, the average share of total income held by the top �ve percent of earners while this birth cohort was in their impressionable years. In our main speci�cations we focus on the national-level inequality that our respondents experienced during their impressionable years in their country of residence, IEit. In an alternative speci�cation we use region-speci�c inequality experiences, IEirt. The GSS provides data on the census division the respondents lived in at age 16, and we compute someone's experienced inequality during his or her impressionable years using historical data on top income shares in this census division. This method relies on the assumption that our respondents did not move when they were aged between 16 and 25.16 Our datasets do not contain direct questions of the level of inequality that our respondents perceived while they were young. Our measures of experienced inequality are therefore based on the actual level of inequality that prevailed during our respondents' formative years. In Appendix B we use data from the International Social Survey Program on Social Inequality (ISSP) to show that people's perceived levels of inequality closely co-move with actual inequality in their country of residence. We show that people believe that they live in a more unequal society when inequality is higher. Similarly, people report higher estimates of pay gaps between CEOs, cabinet ministers and doctors on the one hand, and unskilled workers on the other hand, when inequality is high. These results are robust to including country and time �xed e�ects as well as demographic controls. We report these �ndings in tables A21 and A22. While these �ndings indicate that our measure of inequality experiences is valid, the extent to 16We provide evidence that our results are robust to excluding movers (de�ned as people living in a di�erent census division when they are interviewed than the census division they lived in at age 16). 11 20 25 30 35 40 E xp er ie nc ed to p 5 % in co m e sh ar e (a ge 1 8- 25 ) 1860 1880 1900 1920 1940 1960 1980 2000 Cohort DE FR UK US Figure 3: Experienced top 5 percent income share (age 18-25) against cohort across countries. which individuals have �experienced inequality� depends on individual-level characteristics like people's media consumption, their place of residence or their work place during their formative years. This means that our measure of �inequality experience� is measured with noise. However, this measurement error does not constitute a threat to the internal validity of our �ndings and, if anything, will bias our estimates towards zero. Figure 3 plots the average income share of the top �ve percent experienced over impressionable years against cohort for the largest countries in our sample. We observe that in the US and in the UK cohorts born from around 1960 onward experienced higher levels of inequality during their impressionable years relative to earlier cohorts. The pattern is reversed for France. In the case of Germany, experienced inequality is the lowest for people born around 1960 and higher for those born before that or after. Figure 4 shows experienced inequality for cohorts growing up in the di�erent US census divisions. The large di�erences across census divisions provide an additional source of variation that we exploit in our estimations. 12 25 30 35 40 45 E xp er ie nc ed to p 5 % in co m e sh ar e (a ge 1 8- 25 ) 1880 1900 1920 1940 1960 1980 2000 Cohort New England Middle Atlantic East North Central West North Central South Atlantic 20 25 30 35 40 E xp er ie nc ed to p 5 % in co m e sh ar e (a ge 1 8- 25 ) 1880 1900 1920 1940 1960 1980 2000 Cohort East South Central West South Central Mountain Pacific Figure 4: Experienced top 5 percent income share (age 18-25) against cohort across US census divisions. 13 Similarly, we calculate the average experienced national unemployment rate during our re- spondents' impressionable years, UEit, to account for other macroeconomic shocks that could be correlated with inequality experiences. As our experience variables are reliant on having lived through the impressionable years (age 18 to 25) we restrict our attention to people of age 26 and older in most of our estimations. 3 Empirical Strategy and Results 3.1 Empirical Speci�cation: GSS and Allbus We estimate the e�ect of inequality experiences, IEit, on people's redistributive preferences, yirt. In our preferred speci�cation we also control for other macroeconomic experiences that might a�ect redistributive preferences (Giuliano and Spilimbergo, 2014). In particular, we control for peoples' national-level unemployment experiences, UEit. Moreover, we include a vector of household controls, Xit. 17 In addition, we also account for age �xed e�ects, δit, regional �xed e�ects18, ρr, cohort group �xed e�ects, πi 19, and year �xed e�ects, βt. Speci�cally, we estimate the following equation: yirt = α1IEit + α2UEit + Π T Xit + δit + ρr + βt + πi + εirt (1) We also use region-speci�c inequality experiences, IEirt, for the GSS. In these estimations we control for �xed e�ects for the census division our respondent lived in at age 16, ρ16i, interacted with age �xed e�ects, δit, cohort group �xed e�ects, πi, as well as year �xed e�ects, βt. This in turn allows us to non-parametrically control for age-speci�c trends at the regional level, dif- ferences across cohort groups at the regional level, as well as shocks that are correlated within groups of people growing up in the same census division. The speci�cation is given as follows: yirt = α1IEirt + α2UEit + Π T Xit + ρ16it × δit + ρ16it × βt + ρ16it × πi + ρr + εirt (2) 17This is a vector controlling for household income, household size, the respondent's marital status, religion, educational level and employment status. 18In the US this corresponds to census division and in Germany to the federal state. 19We include dummy variables for the cohorts born between 1876 and 1900, between 1901 and 1925, between 1926 and 1950, between 1951 and 1975, and 1976 or later, respectively. 14 3.2 Empirical Speci�cation: ESS The empirical speci�cation for the European Social Survey is very similar to the speci�cation that uses region-speci�c variation in inequality experiences in the US. We estimate the e�ect of country-speci�c inequality experiences during impressionable years, IEict, on people's redis- tributive preferences, yict. We control for national-level unemployment experiences during im- pressionable years, UEict, and a vector of household controls, Xit. In addition, we account for country-�xed e�ects, ρc, interacted with both time �xed e�ects, βt, and cohort group �xed e�ects, πi, as well as country-speci�c age trends, ageit × ρc.20 yict = α1IEict + α2UEict + Π T Xit + ρc × ageit + ρc × βt + ρc × πi + εict (3) For all of the previous three empirical speci�cations, we report standard errors that are two-way clustered by the respondents' age and cohort as we might expect large intra-cluster correlations in these non-nested clusters (Cameron and Miller, 2013). Importantly, our results are robust to clustering standard errors just by cohort or age.21 Since we test for a large set of outcome variables, we account for multiple hypothesis testing. For our main tables, we adjust the p-values using the �sharpened q-value approach� (Benjamini et al., 2006; Anderson, 2008). For each family of outcomes, we control for a false discovery rate of 5 percent, i.e. the expected proportion of rejections that are type I errors (Anderson, 2008).22 3.3 Results In Table 1, we present the results from the General Social Survey. In Panel A we report the results on national-level inequality experiences during impressionable years, while Panel B shows the results using regional inequality experiences. As can be seen in Columns 1 and 2 in Panel A, we �nd strong evidence that individuals with higher inequality experiences are less likely to be in favor of helping the poor and less in favor of welfare. In Column 3, we show that people who 20Since each country is part of the ESS only in a few waves (sometimes only one) and since the time dimension of the ESS is short (2000-2015), we do not have enough variation of inequality experiences within country-age groups to include an interaction of age �xed e�ects and country �xed e�ects. Our independent variable varies at the country-cohort level, so in the extreme case of observing observations from a particular country only in one year, all the variation in the independent variable would be absorbed by the interaction of age �xed e�ects and country �xed e�ects. 21For all of these datasets we make use of population weights. This makes sure that we can make statements about a sample that is representative of the general population. 22These adjusted p-values are displayed in the tables as FDR-adjusted p-values. 15 Table 1: Main Results: General Social Survey (US) (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A Inequality Experiences -0.0370** -0.0234* -0.0147 -0.0383*** -0.0476*** -0.0414*** (0.0147) (0.0126) (0.0112) (0.0123) (0.0126) (0.0129) FDR-adjusted p-values [.009]*** [.027]** [.067]* [.004]*** [.001]*** [.004]*** Observations 23,199 26,135 29,083 40,136 46,327 32,907 R-squared 0.108 0.128 0.024 0.078 0.146 0.200 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Panel B Inequality Experiences -0.0415** -0.0268* -0.0142 -0.0598*** -0.0522*** -0.0377*** (Regional) (0.0179) (0.0138) (0.0111) (0.0134) (0.0123) (0.0141) FDR-adjusted p-values [.022]** [.045]** [.075]* [.002]*** [.001]** [.002]** Observations 22,987 25,831 28,670 39,632 45,703 32,597 R-squared 0.139 0.159 0.054 0.099 0.170 0.226 Census div 16 FE x Age FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Year FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. The p-values adjusted for a false discovery rate of �ve percent are presented in brackets. Inequality experiences in Panel A are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years. Inequality experiences in Panel B are based on the regional experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations in Panel A control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects, as well as region �xed e�ects. In Panel B, we control for age �xed e�ects, year �xed e�ects and cohort group �xed e�ects interacted with census division at age 16 �xed e�ects and we also control for current census division �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. experienced higher inequality become marginally signi�cantly more likely to attribute success in life more to e�ort than to luck.23 In Columns 4 to 6, we provide consistent evidence that people with higher levels of inequality experience are less likely to be liberal and less likely to vote for democrats. Across speci�cations, we �nd the e�ects to be quite similar between national and regional experiences in terms of both signi�cance and magnitude. In Table 2, we show the results from the German General Social Survey (ALLBUS). In Columns 1 to 3 we demonstrate that people with experiences of higher inequality hold di�erent views on inequality. Speci�cally, these people are less likely to consider the prevailing level of 23One interpretation of this �nding is that higher inequality experiences increase people's perceived income risk. Consequently, they demand more insurance which can come in the form of redistribution by the government. 16 Table 2: Main Results: German General Social Survey (Allbus) (1) (2) (3) (4) (5) (6) Inequality: Inequality does not Inequality Left-wing Intention to Vote: Voted: Left Unfair increase motivation re�ects luck Left Inequality Experiences -0.0543* -0.0428 -0.0684* -0.0957*** -0.0836*** -0.0961** (0.0307) (0.0296) (0.0349) (0.0196) (0.0298) (0.0457) FDR-adjusted p-values [.065]* [.08]* [.053]* [.001]*** [.013]** [.049]*** Observations 10,401 10,357 10,309 18,979 14,691 9,533 R-squared 0.071 0.044 0.068 0.080 0.109 0.111 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. The p-values adjusted for a false discovery rate of �ve percent are presented in brackets. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects . All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. inequality as unfair (Column 1), suggesting that inequality experiences a�ect perceptions of what is a fair division of resources. Moreover, they are more likely to consider inequality important for motivation (Column 2) and to attribute di�erences in income to e�ort rather than luck (Column 3). In Columns 4 to 6 we examine the e�ects of inequality experiences on people's self-assessment on a political scale, their voting intentions as well as their voting behavior in the last federal election. In line with the previous �ndings, we �nd that experiences of higher inequality decrease people's support for left-wing parties. Table 3 displays the results from the European Social Survey. As can be seen in Column 1, we provide evidence that people with high inequality experiences are less likely to agree to the statement that �the government should take measures to reduce di�erences in income levels�. In addition, we �nd that people with more inequality experiences place themselves less on the left on a political scale and are less likely to have voted for a left-wing party in the last election. As can be seen in Tables 1 to 3 all of our key results are robust to taking into account multiple-hypothesis testing.24 To illustrate the magnitude of the e�ects, we compare the e�ect sizes found in the di�erent 24In Tables A17 - A19 in appendix A, we also show our main results including all relevant controls. We �nd evidence that the controls predict preferences for redistribution in line with the previous literature (Alesina and Giuliano, 2013, 2010). For instance, individuals with higher incomes and more education are more against redistribution, while females are more in favor of redistribution. 17 Table 3: Main Results: European Social Survey (1) (2) (3) Pro-redistribution Left-wing Voted: Left Experienced Inequality -0.0390* -0.117*** -0.121*** (0.0234) (0.0200) (0.0389) FDR-adjusted p-values [.096]* [.001]*** [.002]*** Observations 85,529 81,167 25,462 R-squared 0.143 0.079 0.153 Country FE x Age trends Yes Yes Yes Country FE x Year FE Yes Yes Yes Country FE x Cohort group FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. The p-values adjusted for a false discovery rate of �ve percent are presented in brackets. Inequality experiences are based on the experienced share of total national income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age trends, year �xed e�ects and cohort group �xed e�ects, each interacted with country �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 18 samples to the e�ects of other important determinants of preferences for redistribution. In this exercise we focus on our respondents' self-placement on a political scale, as this variable is available across the datasets used. According to our estimates using national-level inequality experiences and the General Social Survey (US), a one standard deviation increase in inequality experiences leads to a decrease of 3.8 percent of a standard deviation in people's tendency to consider themselves as left-wing. Moving from the inequality experiences of the cohort born in 1950 (very low inequality experiences) to the cohort born in 1980 (high inequality experiences) implies a 10.3 percent of a standard deviation decrease in the dependent variable. For comparison, the e�ect of being female is an increase by 12.5 percent of a standard deviation, while the e�ect of holding a highschool degree is a decrease by around 9.5 percent of a standard deviation. We obtain larger e�ect sizes in the sample from the German General Social Survey. Here, a one standard deviation increase in inequality experiences leads to a decrease of people's tendency to consider themselves left-wing by around 9.6 percent of a standard deviation. Moving from the low inequality experience of people born in 1950 to high inequality experiences of the cohort of 1980 implies a decrease in the dependent variable by 21.2 percent of a standard deviation. For comparison, being female increases the self-assessment as left-wing by around 7.7 percent of a standard deviation. Moving from the lowest to the highest quintile in the income distribution leads to a decrease in the dependent variable by 19.2 percent of a standard deviation. According to our estimations on the cross-country sample from the ESS, a one standard deviation increase in inequality experiences leads to a decrease in people's self-classi�cation as left-wing by around 11.7 percent of a standard deviation. For the cohort born in 1980, moving from the country where this cohort has the lowest inequality experience (Denmark) to the country where this cohort has the highest inequality experience (UK) implies a decrease in the tendency of people to consider themselves left-wing by almost 50 percent of a standard deviation. We also replicate our key results using data on voting behavior and support for redistribu- tion from the International Social Survey Program Module on Social Inequality. Importantly, our estimates are fairly similar in terms of magnitude and signi�cance, which provides us with additional con�dence in our results. We present the �ndings from this additional dataset in Appendix B. Our �ndings are not contradictory to Kerr (2014) who �nds that short-run increases in inequality within countries or U.S. regions are associated with greater demand for redistribution. He identi�es e�ects of short-term changes in inequality that operate uniformly across cohorts, 19 which are absorbed by year �xed e�ects in our analysis. Thus, we identify the e�ect of growing up in times of high inequality on top of these e�ects. While all cohorts may exhibit a distaste for inequality, our �ndings suggest that the strength of this concern depends on someone's experience while growing up. The results of this paper are also in line with Alesina and Fuchs-Schuendeln (2007) who provide evidence that people who grew up and lived in East Germany under the communist regime are more in favor of redistribution than are people from West Germany. Our �nding of a negative e�ect of inequality experiences on demand for redistribution provides an additional explanation for higher demand for redistribution in formerly communist countries, where income inequality was often low. 4 Mechanisms 4.1 Reference Points and Fairness It could be that experiences of inequality alter people's reference points and thereby a�ect peo- ple's perception of appropriate levels of inequality. For example, individuals with a higher ref- erence point about inequality due to experiences would perceive current levels of inequality as less severe and therefore exhibit a lower demand for redistribution. Indeed, recent evidence by Charité et al. (2015) shows that reference points play an important role in determining people's taste for redistribution. Above, we presented evidence that people are less likely to perceive the prevailing distribution of incomes as unfair if they have higher inequality experiences. Since everyone in a given year faces the same aggregate level of inequality, this suggests that people interpret the fairness of the prevailing distribution in light of their inequality experiences. We provide experimental evidence for this mechanism in section 6. 4.2 Extrapolation from own circumstances The negative e�ect of experiencing inequality on preferences for redistribution could be driven by individuals who bene�ted personally from high levels of inequality while they were young. If this was the case we would expect the e�ect to be stronger for those who had better starting conditions in life and for those who were more successful in life. To shed light on this mechanism, we examine 20 heterogeneous e�ects by a variety of proxies for starting conditions in life and economic status. For each of our main outcomes, we estimate the following speci�cation: yirt = α1IEit + α2IEit × interactit + α3interactit + α4UEit + ΠT Xit + δit + ρr + βt + πi + εirt (4) where interactit is our interaction variable of interest. We then calculate the estimated average e�ect sizes (AES) for the coe�cients α1 and α2 across the six speci�cation we estimate in the GSS or the Allbus, respectively (Kling et al., 2005; Giuliano and Spilimbergo, 2014).25 Using the AES instead of individual coe�cients increases our e�ective statistical power. This is particularly important for the heterogeneity analysis for which we have lower statistical power. In Table 4 we show that there is no signi�cant heterogeneity by relative family income at age 16 and by father's education in our sample from the GSS.26 27 This suggests that the e�ect is not driven by those who had better starting conditions in life. Moreover, the e�ect is only marginally signi�cantly larger for those with high current relative income, and not signi�cantly di�erent for those with high education. In Table 5 we show that also in the Allbus sample the e�ects are fairly uniform across groups. Taken together, these homogeneous results suggest that extrapolation from own circumstances is an unlikely explanation for the e�ect of inequality experiences on redistributive preferences.28 4.3 Relative income Experiences of inequality could also change people's beliefs about their economic status. Specif- ically, people who grew up in times of high income inequality, and who are therefore used to more inequality, could be less likely to perceive their current relative income as low. People's beliefs about their relative position in the income distribution have been shown to change peo- 25The AES is de�ned as the average of all coe�cient estimates across a family of estimations, where each coe�cient is divided by the standard deviation of the respective outcome. All our outcomes are normalized, so the AES is the simple average of the estimated coe�cients. We calculate p-values for the AES based on simultaneous estimation of the six regressions. 26These variables are coded as one if the respondent considered the income of his family at age 16 to be at least average and if the respondent's father had at least high school education, respectively. 27We also do not �nd heterogeneity according to education of the mother. 28We also examined heterogeneity according to age, but do not report the results for brevity. We found that the e�ect is fairly uniform across age groups, suggesting that the e�ects persist over the lives of the respondents. In addition, we checked whether the e�ects vary by the degree someone's perceived relative income increased or decreased between his or her youth and the survey year. We found no evidence for heterogeneous e�ects along this dimension. 21 Table 4: Heterogeneous E�ects: General Social Survey (GSS) (1) (2) (3) (4) AES AES AES AES Inequality Experiences -.0213*** -.0435*** -.0278*** -0.0291** [0.001] [0.000] [0.000] [0.010] Inequality Experiences × High relative income at 16 -.0104 [0.128] Inequality Experiences × High father's education .008 [0.336] Inequality Experiences × High relative income -.0114 [0.120] Inequality Experiences × High education -.006 [0.401] Observations 25,078 24,818 30,271 31,919 Age FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Region FE Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes HH controls Yes Yes Yes Yes P-values from simultaneous estimation clustered by cohort are displayed in parentheses. The number of observations refers to the average number used for the estimation of a given AES. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects . All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 22 Table 5: Heterogeneous E�ects: German General Social Survey (Allbus) (1) (2) (3) AES AES AES Inequality Experiences -.0744*** -.0691*** -.0681*** [0.000] [0.005] [0.000] Inequality Experiences × High father's education .0059 [0.769] Inequality Experiences × High relative income -.0177 [0.353] Inequality Experiences × High education -.0204 [0.375] Observations 14,052 10,308 14,122 Age FE Yes Yes Yes Year FE Yes Yes Yes Cohort group FE Yes Yes Yes Region FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes P-values from simultaneous estimation clustered by cohort are displayed in parentheses. The number of observations refers to the average number used for the estimation of a given AES. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 23 Table 6: Other outcomes: GSS and Allbus (1) (2) (3) GSS (national inequality experiences) Allbus Low relative Low social Low social income position position Inequality Experiences 0.00340 0.00467 -0.0368 (0.0121) (0.0118) (0.0234) Observations 43,234 44,402 15,025 R-squared 0.257 0.205 0.204 Age FE Yes Yes Yes Year FE Yes Yes Yes Cohort group FE Yes Yes Yes Region FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. Inequality ex- periences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national un- employment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as regions �xed e�ects. All speci�cations con- trol for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. ple's demand for redistribution (Cruces et al., 2013; Karadja et al., 2016). We therefore test whether experiences of high inequality lower people's perceived relative income. Similarly, we test whether inequality experiences a�ect people's self-perceived social class. Table 6 shows the results of these estimations for the GSS and Allbus, respectively. We �nd no evidence for a signi�cant e�ect of inequality experiences on perceived relative income and social class. 5 Robustness 5.1 �Impressionable Years� versus Other Years In our main speci�cation we examined the e�ect of inequality experiences during impressionable years, i.e. when people are aged between 18 and 25. We focused on these years as the previous literature on the formation of beliefs suggests that this period in life is particularly important for shaping beliefs and preferences (Mannheim, 1970; Krosnick and Alwin, 1989; Giuliano and 24 Spilimbergo, 2014). In what follows we examine the in�uence of experiences in other periods of life on people's preferences for redistribution. In particular, we use the Allbus and the GSS to examine the e�ect of inequality experiences during di�erent eight year intervals (2�9, 10�17, 26�33, 34�41, 42�49, and 50�57) in our respondents' lives. We closely follow the empirical strategy in Giuliano and Spilimbergo (2014). In Panels A to G in Tables A1 to A3, we show how inequality experiences in di�erent life periods a�ect people's preferences for redistribution. While we still �nd signi�cantly negative e�ects of experiences in life periods surrounding the impressionable years (10-17 and 26-33, respectively), the e�ects are weaker or vanish completely for other age ranges.29 All in all, this evidence corroborates our view that inequality experiences during the impressionable years are vital in shaping people's beliefs, values and political preferences.30 5.2 Life-time Experiences We �nd very similar e�ects of inequality experiences when we use the methodology developed by Malmendier and Nagel (2011). While in our main estimations we look at the e�ect of inequality experiences during someone's formative years, this alternative measure is based on a weighted average of top income shares experienced over a respondent's lifetime until the time of the interview. Thus, in contrast to our previous measures, we now allow more recent experiences to still have some e�ect. In line with the above �ndings that earlier experiences matter more than later experiences, we use a weighting factor that gives more weight to early experiences and in which experiences only matter beginning from age 18.31 We re-estimate our main speci�cations using the same set of controls but employing these alternative measures of experienced inequality and experienced unemployment. In Panels H of Tables A1 to A3 we show that we obtain very similar results in terms of e�ect size and statistical signi�cance when we use this alternative measure of inequality experiences. 29The di�erences in experiences across cohorts stem from medium- to longer-term changes in inequality. This is a key di�erence to Giuliano and Spilimbergo (2014) who examine the e�ect of experiencing macroeconomic shocks while young, which occur at a higher frequency. Finding some e�ect of inequality experienced during periods surrounding the impressionable years, which is correlated with inequality experienced from age 18 to 25, is therefore what one would expect. 30Given the nature of the dataset, it is di�cult to compare the importance of experiences during impressionable years versus experiences during other periods of life. Since in each estimation we only focus on individuals who have lived through the relevant life period, we cannot hold constant the sample size and sample composition in the di�erent speci�cations. 31For details regarding the construction of this alternative measure see Appendix F. 25 5.3 Placebo Outcomes It could be the case that our estimates merely pick up cohort di�erences in political preferences and in particular in how left-wing people in di�erent cohorts generally are. The inclusion of 25- year cohort-group �xed e�ects in our main speci�cations ensures that our results are not driven by longer-term general shifts in preferences across cohorts. To further address this concern, we show that other political attitudes that di�er between the left and the right of the political spectrum, but that are not directly related to inequality and redistribution, are not a�ected by our measures of inequality experiences. In Tables A4 - A6 we provide evidence that inequality experiences do not signi�cantly a�ect nationalism, attitudes towards guns, attitudes towards immigrants32, attitudes towards democ- racy, attitudes towards the uni�cation of the EU and people's belief in god.33 5.4 Other Experiences during Impressionable Years We also examine in how far our results are sensitive to controlling for other macroeconomic experiences during impressionable years. Speci�cally, we control for whether our respondents experienced an economic crisis during their impressionable years. As in Giuliano and Spilim- bergo (2014) we de�ne a crisis experience as at least once having experienced a drop of real per capita GDP by 3.8 percent or more during the impressionable years. Next, we examine whether our estimates are sensitive to the inclusion of the average growth rate of real GDP per capita during the impressionable years. In order to control for the experienced size of the government, we include the average ratio of total tax revenue relative to GDP experienced during the impres- sionable years. Finally, we also include a proxy for experienced political ideology, namely the fraction of someone's impressionable years in which a Republican president (US) or conservative chancellor (Germany) was in o�ce. When we control for these other experiences in our estimations on the Allbus and the GSS our main results barely change (see Tables A7 and A8).34 This indicates that our results are not driven by other experiences people made while growing up which are correlated with inequality experiences. 32In the Allbus we focus only on attitudes towards immigrants that are not related to economic concerns. The respective variables in the GSS and the ESS mainly refer to whether the number of immigrants should be increased or decreased. 33In Appendix C, we provide detailed information on the placebo variables used in our analysis. 34We demonstrate robustness of our main speci�cation by including these other experiences one at a time. Since all experience variables vary at the cohort level, and since macroeconomic variables tend to be highly correlated, including all these other experiences at once would lead to problems of multicollinearity. 26 5.5 Other Robustness Checks In Tables A9-A12 we examine how sensitive our results are to a variety of robustness checks. Our results are robust to using di�erent de�nitions of income inequality based on (i) the share of income earned by the top ten percent, (ii) the share of income earned by the top one percent as well as (iii) the Gini coe�cient of equivalized disposable household incomes which is available for a much smaller sample of respondents.35 In contrast to top income shares which are based on before-tax incomes, the Gini coe�cient measures after-tax income inequality. We obtain very similar results when we use these alternative inequality measures. If anything, we �nd larger e�ect sizes when we use the Gini coe�cient instead of top income shares. In addition, we show that our results remain unchanged when we exclude all individuals with missing values in any of the controls. Our �ndings are also robust to not controlling for people's national unemployment experiences which alleviates concerns that inequality experiences operate through unobservable long-run e�ects of unemployment experiences. Moreover, the results are una�ected when we control for age trends rather than age �xed e�ects or when we exclude movers from our estimations on the GSS which rely on regional variation in income inequality experiences.36 As a �nal robustness check we exclude the 25-year cohort group �xed e�ects from our speci�cations and obtain very similar results. 6 Experimental Evidence In this section, we examine the causal e�ect of exposure to di�erent levels of inequality on people's redistributive preferences by conducting a series of tightly controlled online experiments on Amazon Mechanical Turk. We experimentally vary our respondents' exposure to inequality in stage one of the experiment. Then, we assess in how far exposure to inequality in stage 1 a�ects people's redistributive preferences in stage 2 of the experiment. We hypothesize that individuals exposed to inequality in the �rst stage of the experiment will �nd high levels of inequality more acceptable and therefore redistribute less in the second stage. The use of an experiment improves upon the observational evidence in three dimensions: �rst, it allows us to more cleanly identify the causal e�ect of experiencing inequality. It is not 35While a lot of historical data on the Gini coe�cient exist in the US, much less reliable data on the Gini coe�cient are available for most European countries in our sample. This implies that the samples we can use in our analysis for the ESS are much smaller than for the measures of top income inequality. 36We de�ne movers as people who live in a di�erent census division when they are interviewed than when they were aged 16. 27 possible to randomly assign individuals to di�erent life-time experiences and it is very di�cult to exploit exogenous variation in inequality that is not correlated with other economic variables. Our experiment o�ers us tight control over the agent's decision environment and allows us to keep constant payo�s and economic circumstances of agents, while only manipulating the degree of inequality our agents face in stage one of the experiment. Second, it allows us to measure redistributive preferences with behavioral measures rather than self-reports. Third, it allows us to test for mechanisms. In particular, by using a spectator design we shut down any mechanism involving self-interest or extrapolation from own circumstances, i.e. channels operating through our agents' own outcomes. The experiment captures the idea of an experienced distribution of income in a stylized manner. However, we believe that the experiment highlights a mechanism that could similarly operate for inequality experienced during someone's formative years. Speci�cally, the experiment shows that people's acceptance of inequality and their demand for redistribution may be highly dependent on the degree of inequality people are used to. 6.1 Experimental Design 6.1.1 Stage 1: Treatment We randomly assign individuals into choice environments in which they make hypothetical choices about the payo�s of two other workers on mTurk, which we refer to as A and B. Individuals in the inequality condition choose between two options that are highly unequal between A and B, while people in the equality condition choose between two options that result in more equal outcomes for worker A and B. In other words, we signi�cantly manipulate our agents' choice sets which in turn a�ects the degree of inequality they experience in stage 1 of the experiment. One group of individuals is randomly assigned to choose unequal outcomes, while another group of individuals is forced to choose more equal outcomes. This in turn implies that the agents operate either in environments in which institutions promote redistribution and thereby lower inequality, or in environments without redistribution and with high levels of inequality. Speci�cally, our participants are allocated to one of the following two conditions: • Equality Condition: �Person A and B are both mTurk workers and they previously worked for us on a task. Worker A completed 2 tasks correctly, while worker B completed 8 tasks correctly. The number of correctly completed tasks depends on both the worker's 28 e�ort and luck. We would like to pay them a total of $1 for their work. How much money would you like to pay them?� � Option A: 50 cents for worker A and 50 cents for worker B. � Option B: 48 cents for worker A and 52 cents for worker B. • Inequality Condition: �Person A and B are both mTurk workers and they previously worked for us on a task. Worker A completed 2 tasks correctly, while worker B completed 8 tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. We would like to pay them a total of $1 for their work. How much money would you like to pay them?� � Option A: 22 cents for worker A and 78 cents for worker B. � Option B: 20 cents for worker A and 80 cents for worker B. 6.1.2 Stage 2: Redistribution Game In the second stage our players again make hypothetical distributive choices between two workers who have completed a real-e�ort task.37 Player C has completed three tasks correctly, while player D has completed seven tasks correctly. Our respondents are also told that �the number of correctly completed tasks depends on both the worker's e�ort and luck.� Individuals decide how to split $1 between the two workers. They can give a higher share of the $1 to the worker who has completed more tasks correctly or the worker who completed less tasks correctly, or they can split the money equally between the two.38 We code our measure of redistributive preferences, yi, such that higher values correspond to larger shares of the money going to the lower-achieving worker. 6.1.3 Experiment 2 We also conduct an additional experiment (which we refer to as experiment 2) to assess the robustness of the �rst experiment. We make several changes: �rst, we use a di�erent choice set for our main measure of re-distributive preferences. In particular, we restrict the choice set of our respondent in stage 2 of the experiment by not allowing them to give a larger share of the $1 to the worker who completed less tasks correctly. Second, we use a di�erent choice set for 37We tell our respondents that �these workers are NOT the same people as from the previous task�. 38A more precise description of this behavioral measure can be found in Appendix G. 29 respondents in the inequality condition. Speci�cally, we let them choose between giving 20 cents to worker A and 80 cents to worker B, or nothing to worker A and 100 cents to worker B. 6.2 Sample We ran our experiments on Amazon Mechanical Turk (MTurk), an online crowdsourcing mar- ketplace commonly used to conduct online experiments. The pool of available workers is very large and more representative of the US population than student samples. Many experiments have been replicated using MTurk samples. MTurk participants produce high-quality data (Ma- son and Suri, 2012; Horton et al., 2011; Buhrmester et al., 2011), and are more attentive to instructions than college students (Hauser and Schwarz, 2016). In order to participate in the experiment, people had to be based in the United States, have an overall rating of more than 95% and have completed more than 500 tasks on MTurk. These commonly applied restrictions are important in order to get high-quality data, as demonstrated by Peer et al. (2014). The experiments were run in June and July 2016. 200 participants completed experiment 1 and 202 individuals completed experiment 2.39 Our sample from the experiments is fairly similar to the general population. The median income in our sample is 46,000 while it is 53,000 in the US. The average age in our sample is 36 which is substantially below the average age of the US population. The proportion of people having a bachelor's degree is higher in our sample than in the general population. An overview of the sample composition in our mTurk sample is given in Table A16. 6.3 Results To compare the behavior of people in the equality condition with that of people in the inequality condition, we regress our measure of redistributive preferences, yi, on a treatment indicator, Inequalityi, which takes the value one for people who are in the inequality condition, and the value zero for all the other participants. Speci�cally, the equation that we estimate is: yi = π0 + π1Inequalityi + εi 39Only eight people dropped out of experiment 1, while three individuals dropoed out of experiment 2. 30 Table 7: Results from the Online Experiment (1) (2) (3) (4) Experiment 1 Experiment 2 Redistribute Redistribute (z) Redistribute Redistribute (z) Panel A Inequality -0.727∗∗ -0.342∗∗ -0.928∗∗∗ -0.564∗∗∗ (0.290) (0.137) (0.202) (0.123) Observations 201 201 202 202 R-squared 0.031 0.031 0.096 0.096 Panel B: with controls Inequality -0.694∗∗ -0.327∗∗ -0.969∗∗∗ -0.589∗∗∗ (0.297) (0.140) (0.207) (0.126) Observations 200 200 202 202 R-squared 0.115 0.115 0.157 0.157 We apply robust standard errors. In Panel B we control for household income, education, employ- ment status, household size, religion, and gender.* p < 0.10, ** p < 0.05, *** p < 0.01. where εi is an individual-speci�c error term. 40 Second, we re-estimate the speci�cation above including a set of demographic controls.41 In Table 7 we provide strong evidence that individuals exposed to environments that promote higher levels of inequality are less in favor of redistribution as measured by our behavioral measure of redistribution.42 In Panel A, we present our main results by just regressing the outcome measure on an inequality treatment indicator, while Panel B includes a set of demographic controls. In Columns 1 and 2 of Table 7 we present the results from experiment 1. In Columns 3 and 4 we show the results from experiment 2. The data from both experiments clearly show that people in the inequality condition system- atically give a lower share of the $1 to the lower-achieving worker. The e�ect sizes we observe are large. People in the inequality condition redistribute between .35 and .56 of a standard deviation less relative to people in the equality condition. These results are barely a�ected by the inclusion 40For all the speci�cations, we use robust standard errors. 41Speci�cally, we control for household income, education, employment status, household size, religion and gender. 42It is important to note that our behavioral measure of redistribution is signi�cantly correlated with people's political a�liation. Speci�cally, Republicans redistribute less compared to Democrats which suggests that our behavioral measure does have a high degree of external validity. 31 of controls.43 Taken together, experiencing inequality seems to increase our respondents' inequality ac- ceptance. The design of the experiment rules out channels working through people's personal outcomes. Our results therefore suggest that experiences of how money is distributed a�ect what is considered a fair allocation through the formation of reference points. 6.3.1 Alternative Mechanisms Our treatment conditions could induce di�erential feelings of fairness as they mimic di�erent fairness principles. In particular, the allocation of resources in the equality condition represents the �equality principle� according to which people with di�erent performance will be paid equal amounts. The distribution of resources in the inequality condition, on the other hand, is closer to the �equity principle� which posits that people should be paid according to their performance in the task. As the performance di�erences across workers depend both on e�ort and luck, it is not obvious which of the two options in each of the two treatment conditions should be considered fair. In the equality condition about two thirds of participants prefer option B, while the rest prefers option A. In the inequality condition about 70 percent of participants prefer option B. If most people actually considered the equity principle as fair (Abeler et al., 2010), then this would imply that participants in the equality condition would be forced to implement an unfair choice, while people in the inequality condition would be forced to implement a fair choice. This in turn could in theory a�ect people's behavior in stage two of the experiment. However, it is not clear why people forced to implement an �unfair� (according to the �equity principle�) choice in stage 1 of the experiment, would systematically prefer to implement more equal payments between the workers in stage 2 of the experiment. It could also be the case that respondents interpret the choice set in stage 1 of the experiment as a signal about the relative roles of e�ort and luck in determining the number of tasks that the workers have completed. In particular, people in the inequality condition could believe that success is due to e�ort rather than luck and therefore redistribute less. To address this concern, we asked our respondents whether they think that the number of correctly completed tasks by the workers depended more on the workers' e�ort or more on the workers' luck. People's responses 43The e�ect sizes are quite large and therefore unlikely can be explained by experimenter demand e�ects (Zizzo, 2010). Evidence by de Quidt et al. (2016) suggests that demand e�ects when people change their beliefs about the experimenter's hypotheses are rather small. 32 to this question do not di�er between the participants in the �inequality condition� and those in the �equality condition�.44 This suggests that in our experiment exposure to environments promoting inequality a�ect redistributive preferences independently of e�ects working through beliefs about the determinants of success in life. 7 Conclusion We use several large nationally representative datasets to highlight that people who have grown up in times of higher inequality are less left-wing as measured by their self-reported redistributive preferences as well as their party a�liation. We also show that people with more inequality experience hold di�erent beliefs about inequality and are less likely to consider the prevailing level of inequality as too high. This suggests that experienced inequality could a�ect the way people evaluate current levels of inequality. We then test the conjecture that reference points may play an important role in shaping people's redistributive choices in a tightly controlled experimental set-up. We also provide evidence against alternative channels through which our e�ects could operate: �rst, we show that the e�ects are equally strong for those with worse starting conditions or less success in life, suggesting that people do not form their redistributive preferences based on the e�ect inequality had on them personally. Second, we demonstrate that our e�ects are unlikely to operate through changes in perceived relative income. The results of this paper suggest that preferences for redistribution are shaped by the level of inequality that prevailed during people's formative years. This implies that the increases in inequality over the last decades are re�ected in lower preferences for redistribution among younger generations relative to older generations. While fairness concerns may have led to increasing demand for redistribution across cohorts (Kerr, 2014), these concerns seem to be weaker for younger generations who are more used to high levels of inequality. Going forward, the longer high levels of inequality prevail, the higher will be the the average level of experienced inequality among voters. Our �ndings suggest that the forces pushing society back towards lower levels of inequality may become weaker the longer high levels of inequality prevail. 44This zero is percisely measured. 33 References Abeler, Johannes, Ste�en Altmann, Sebastian Kube, and Matthias Wibral, �Gift Exchange and Workers' Fairness Concerns: When Equality is Unfair,� Journal of the European Economic Association, 2010, 8 (6), 1299�1324. Alesina, Alberto and Eliana La Ferrara, �Participation in Heterogeneous Communities,� The Quarterly Journal of Economics, 2000, pp. 847�904. and , �Preferences for Redistribution in the Land of Opportunities,� Journal of Public Economics, 2005, 89 (5), 897�931. and George-Marios Angeletos, �Fairness and Redistribution: US versus Europe,� The American Economic Review, 2005. and Nicola Fuchs-Schuendeln, �Good bye Lenin (or not?): The E�ect of Communism on People's Preferences,� The American Economic Review, 2007. and Paola Giuliano, �Preferences for Redistribution,� Handbook of Social Economics, 2010. and , �Culture and Institutions,� National Bureau of Economic Research, 2013. , Edward Glaeser, and Bruce Sacerdote, �Why doesn't the US have a European-style Welfare System?,� Brookings Paper on Economics Activity, 2001. Alesina, Alberto F, Paola Giuliano, and Nathan Nunn, �On the Origins of Gender Roles: Women and the Plough,� The Quarterly Journal of Economics, 2013. Almås, Ingvild, Alexander Cappelen, and Bertil Tungodden, �Are Americans more Meritocratic and E�ciency seeking than Scandinavians?,� Working Paper, 2016. Alvaredo, Facundo, Anthony Barnes Atkinson, Thomas Piketty, and Emmanuel Saez, The World Top Incomes Database 2011. Anderson, Michael L, �Multiple Inference and Gender Di�erences in the E�ects of Early Inter- vention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects,� Journal of the American Statistical Association, 2008, 103 (484). Atkinson, Anthony B and Salvatore Morelli, �Chartbook of Economic Inequality,� ECINEQ WP, 2014, 324. 34 Becker, Anke, Thomas Dohmen, Benjamin Enke, and Armin Falk, �The Ancient Origins of the Cross-country Heterogeneity in Risk Preferences,� Working Paper, 2016. Benabou, Roland and Jean Tirole, �Belief in a Just World and Redistributive Politics,� The Quarterly Journal of Economics, 2006. Benjamini, Yoav, Abba M Krieger, and Daniel Yekutieli, �Adaptive Linear Step-up Procedures that Control the False Discovery Rate,� Biometrika, 2006, 93 (3), 491�507. Buhrmester, Michael, Tracy Kwang, and Samuel D Gosling, �Amazon's Mechanical Turk: A New Source of Inexpensive, yet High-quality Data?,� Perspectives on Psychological Science, 2011, 6 (1), 3�5. Cameron, A Colin and Douglas L Miller, �A Practitioner's Guide to Cluster-Robust Infer- ence,� Forthcoming in Journal of Human Resources, 2013. Campbell, John Y, �A Comment on James M. Poterba's `Demographic Structure and Asset Returns',� Review of Economics and Statistics, 2001, 83 (4), 585�588. Charité, Jimmy, Raymond Fisman, and Ilyana Kuziemko, �Reference Points and Redis- tributive Preferences: Experimental Evidence,� National Bureau of Economic Research, 2015. Chen, Yuyu, Hui Wang, and David Y Yang, �Salience of History and the Preference for Redistribution,� Available at SSRN 2717651, 2016. Cruces, Guillermo, Ricardo Perez-Truglia, and Martin Tetaz, �Biased Perceptions of Income Distribution and Preferences for Redistribution: Evidence from a Survey Experiment,� Journal of Public Economics, 2013, 98, 100�112. de Quidt, Jonathan, Johannes Haushofer, and Christopher Roth, �Measuring and Bounding Experimenter Demand E�ects,� Working Paper, 2016. Durante, Ruben, Louis Putterman, and Joël Weele, �Preferences for Redistribution and Perception of Fairness: An Experimental Study,� Journal of the European Economic Associa- tion, 2014, 12 (4), 1059�1086. Fehr, Ernst and Klaus M Schmidt, �A Theory of Fairness, Competition, and Cooperation,� Quarterly journal of Economics, 1999, pp. 817�868. 35 Fong, Christina, �Social Preferences, Self-interest, and the Demand for Redistribution,� Journal of Public Economics, 2001, 82 (2), 225�246. Fuchs-Schuendeln, Nicola and Matthias Schuendeln, �On the Endogeneity of Political Preferences: Evidence from Individual Experience with Democracy,� Science, 2015, 347 (6226), 1145�1148. Giuliano, Paola and Antonio Spilimbergo, �Growing up in a Recession,� The Review of Economic Studies, 2014, 81 (2), 787�817. Hauser, David J and Norbert Schwarz, �Attentive Turkers: MTurk Participants Perform Better on Online Attention Checks Than do Subject Pool Participants,� Behavior Research Methods, 2016, 48 (1), 400�7. Hertwig, Ralph, Greg Barron, Elke U Weber, and Ido Erev, �Decisions from Experience and the E�ect of Rare Events in Risky Choice,� Psychological Science, 2004, 15 (8), 534�539. Horton, John J, David G Rand, and Richard J Zeckhauser, �The Online Laboratory: Conducting Experiments in a Real Labor Market,� Experimental Economics, 2011, 14 (3), 399�425. Huber, John and Ronald Inglehart, �Expert Interpretations of Party Space and Party Locations in 42 Societies,� Party Politics, 1995, 1 (1), 73�111. Karadja, Mounir, Johanna Mollerstrom, and David Seim, �Richer (and Holier) Than Thou? The E�ect of Relative Income Improvements on Demand for Redistribution,� Review of Economics and Statistics, 2016. Kenworthy, Lane and Leslie McCall, �Inequality, Public Opinion and Redistribution,� Socio- Economic Review, 2008, 6 (1), 35�68. Kerr, William R, �Income inequality and social preferences for redistribution and compensation di�erentials,� Journal of Monetary Economics, 2014, 66, 62�78. Kiatpongsan, Sorapop and Michael I Norton, �How much (more) should CEOs make? A Universal Desire for More Equal Pay,� Perspectives on Psychological Science, 2014, 9 (6), 587�593. 36 Kling, Je�rey R, Jens Ludwig, and Lawrence F Katz, �Neighborhood E�ects on Crime for Female and Male Youth: Evidence from a Randomized Housing Voucher Experiment,� The Quarterly Journal of Economics, 2005, pp. 87�130. Kosse, Fabian, Thomas Deckers, Hannah Schildberg-Hörisch, and Armin Falk, �The Formation of Prosociality: Causal Evidence on the Role of Social Environment,� Technical Report, Institute for the Study of Labor (IZA) 2016. Krosnick, Jon A and Duane F Alwin, �Aging and Susceptibility to Attitude Change.,� Journal of Personality and Social Psychology, 1989, 57 (3), 416. Kuziemko, Ilyana, Michael I Norton, Emmanuel Saez, and Stefanie Stantcheva, �How Elastic are Preferences for Redistribution? Evidence from Randomized Survey Experiments,� The American Economic Review, 2015, 105 (4), 1478�1508. Luttmer, Erzo FP and Monica Singhal, �Culture, Context, and the Taste for Redistribu- tion,� American Economic Journal: Economic Policy, 2012. Malmendier, Ulrike and Leslie S Shen, �Scarred Consumption,� Technical Report, Working paper, UC-Berkeley 2015. and Stefan Nagel, �Depression Babies: Do Macroeconomic Experiences A�ect Risk- taking?,� The Quarterly Journal of Economics, 2011. and , �Learning from In�ation Experiences*,� The Quarterly Journal of Economics, 2015, p. qjv037. Mannheim, Karl, �The Problem of Generations,� Psychoanalytic Review, 1970, 57 (3), 378. Mason, Winter and Siddharth Suri, �Conducting Behavioral Research on Amazon's Me- chanical Turk,� Behavior Research Methods, 2012, 44 (1), 1�23. Nisbett, Richard E and Lee Ross, Human Inference: Strategies and Shortcomings of Social Judgment 1980. Norton, Michael I and Dan Ariely, �Building a better America�One Wealth Quintile at a Time,� Perspectives on Psychological Science, 2011, 6 (1), 9�12. Nunn, Nathan, �Culture and the Historical Process,� Economic History of Developing Regions, 2012, 27 (sup1), S108�S126. 37 and Leonard Wantchekon, �The Slave Trade and the Origins of Mistrust in Africa,� The American Economic Review, 2011. Pan, Jennifer and Yiqing Xu, �China's Ideological Spectrum,� MIT Political Science Depart- ment Research Paper, 2015. Peer, Eyal, Joachim Vosgerau, and Alessandro Acquisti, �Reputation as a Su�cient Condition for Data Quality on Amazon Mechanical Turk,� Behavior Research Methods, 2014, 46 (4), 1023�1031. Piketty, Thomas, �Social Mobility and Redistributive Politics,� The Quarterly Journal of Eco- nomics, 1995, pp. 551�584. Roland, Gerard and David Y Yang, �China's Lost Generation: Changes in Beliefs and their Intergenerational Transmission,� Working Paper, 2016. Schildberg-Hörisch, Hannah, Thomas Deckers, Armin Falk, and Fabian Kosse, �How Does Socio-Economic Status Shape a Child's Personality?,� Working Paper, 2014. Weber, Elke U, Ulf Böckenholt, Denis J Hilton, and Brian Wallace, �Determinants of Diagnostic Hypothesis Generation: E�ects of Information, Base Rates, and Experience,� Journal of Experimental Psychology: Learning, Memory, and Cognition, 1993, 19 (5), 1151. Zizzo, Daniel John, �Experimenter Demand E�ects in Economic Experiments,� Experimental Economics, 2010, 13 (1), 75�98. 38 A Additional Tables Table A1: GSS: National Inequality Experiences in Other Periods of Life (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A: 2 to 9 Inequality Experiences -0.0269 -0.0313 0.0358 -0.106*** -0.0309 0.00160 (0.0332) (0.0253) (0.0362) (0.0327) (0.0369) (0.0361) Observations 21,998 23,703 26,806 37,492 42,790 30,219 Panel B: 10 to 17 Inequality Experiences -0.0588** -0.0361* -0.00608 -0.0533** -0.0346 -0.0502** (0.0232) (0.0207) (0.0227) (0.0212) (0.0215) (0.0215) Observations 22,881 25,299 28,331 39,313 45,158 32,054 Panel C: 18 to 25 (main) Inequality Experiences -0.0370** -0.0234* -0.0147 -0.0383*** -0.0476*** -0.0414*** (0.0147) (0.0126) (0.0112) (0.0123) (0.0126) (0.0129) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel D: 26 to 33 Inequality Experiences -0.0403*** -0.0279** -0.00735 -0.0422*** -0.0664*** -0.0422*** (0.0144) (0.0119) (0.0111) (0.00920) (0.00972) (0.0131) Observations 18,663 20,942 23,487 32,320 37,467 27,738 Panel E: 34 to 41 Inequality Experiences -0.00458 -0.0206 -0.0183 0.0235 -0.0831*** -0.0626*** (0.0254) (0.0183) (0.0186) (0.0174) (0.0189) (0.0203) Observations 14,181 16,070 18,015 24,608 28,798 21,873 Panel F: 42 to 49 Inequality Experiences 0.0369 -0.0174 -0.0278 0.0473*** -0.0544*** -0.0230 (0.0331) (0.0229) (0.0186) (0.0182) (0.0178) (0.0178) Observations 10,325 11,821 13,253 17,976 21,124 16,363 Panel G: 50 to 57 Inequality Experiences 0.0285 -0.0441 -0.0333 -0.0102 -0.0906*** -0.0725*** (0.0665) (0.0317) (0.0278) (0.0285) (0.0332) (0.0258) Observations 7,118 8,035 9,147 12,296 14,505 11,394 Panel H: Life-time experiences λ = −1 Inequality Experiences -0.0463*** -0.0162 -0.00953 -0.0224 -0.0359*** -0.0256** (0.0140) (0.0163) (0.0133) (0.0149) (0.0116) (0.0130) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during di�erent life periods. Unemployment experiences are based on the experienced national unemployment rate during the di�erent life periods. We control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel H we display the results on the e�ect of life-time inequality experiences based on the methodology developed by Malmendier and Nagel (2011) - we use a weighting factor of λ = -1. * p < 0.10, ** p < 0.05, *** p < 0.01. 39 Table A2: GSS: Regional Inequality Experiences in Other Periods of Life (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A: 2 to 9 Inequality Experiences -0.0398 -0.0478* 0.0201 -0.0903*** -0.0534** -0.00692 (Regional) (0.0338) (0.0252) (0.0243) (0.0213) (0.0247) (0.0271) Observations 21,829 23,460 26,458 37,063 42,270 29,970 Panel B: 10 to 17 Inequality Experiences -0.0287 -0.00502 -0.00782 -0.0381** -0.00525 -0.0333* (Regional) (0.0188) (0.0210) (0.0215) (0.0168) (0.0146) (0.0174) Observations 22,704 25,037 27,963 38,865 44,603 31,784 Panel C: 18 to 25 (main) Inequality Experiences -0.0415** -0.0268* -0.0142 -0.0598*** -0.0522*** -0.0377*** (Regional) (0.0179) (0.0138) (0.0111) (0.0134) (0.0123) (0.0141) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel D: 26 to 33 Inequality Experiences -0.0384** -0.0406*** -0.00866 -0.0509*** -0.0696*** -0.0377*** (Regional) (0.0165) (0.0150) (0.0135) (0.0115) (0.0105) (0.0142) Observations 18,500 20,699 23,162 31,916 36,956 27,455 Panel E: 34 to 41 Inequality Experiences 0.00763 -0.0125 -0.0244 0.00213 -0.0581*** -0.0527*** (Regional) (0.0249) (0.0174) (0.0181) (0.0161) (0.0177) (0.0189) Observations 14,065 15,880 17,767 24,303 28,403 21,631 Panel F: 42 to 49 Inequality Experiences 0.0550 -0.0183 -0.0282 0.0485*** -0.0470** -0.00858 (Regional) (0.0425) (0.0271) (0.0242) (0.0186) (0.0232) (0.0237) Observations 10,236 11,674 13,067 17,745 20,826 16,165 Panel G: 50 to 57 Inequality Experiences 0.0361 0.0170 -0.0189 0.0135 -0.120*** -0.0913*** (Regional) (0.0487) (0.0382) (0.0353) (0.0299) (0.0400) (0.0321) Observations 7,037 7,921 9,011 12,121 14,279 11,241 Census div 16 FE x Age FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Year FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. Inequality experiences are based on the regional experienced share of total income earned by the top 5 percent during the di�erent periods of life. Unemployment experiences are based on the experienced national unemployment rate during the di�erent periods of life. We control for age �xed e�ects, year �xed e�ects and cohort group �xed e�ects, each interacted with census division at age 16 �xed e�ects and we also control for current census division �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 40 Table A3: Allbus: Inequality Experiences in Other Periods of Life (1) (2) (3) (4) (5) (6) Inequality: Inequality does not Inequality Left-wing Intention to Vote: Voted: Left Unfair increase motivation re�ects luck Left Panel A: 2 to 9 Inequality Experiences 0.0502 -0.0671 0.00994 0.0205 0.0672** 0.0455 (0.0898) (0.0677) (0.0571) (0.0541) (0.0329) (0.0528) Observations 7,506 7,427 7,408 14,072 10,940 6,855 Panel B: 10 to 17 Inequality Experiences -0.137*** -0.0555 -0.112*** -0.0358 -0.120*** -0.148*** (0.0483) (0.0454) (0.0331) (0.0287) (0.0232) (0.0408) Observations 8,942 8,885 8,852 16,248 12,524 8,023 Panel C: 18 to 25 (main) Inequality Experiences -0.0543* -0.0428 -0.0684* -0.0957*** -0.0836*** -0.0961** (0.0307) (0.0296) (0.0349) (0.0196) (0.0298) (0.0457) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel D: 26 to 33 Inequality Experiences -0.00389 0.00454 -0.00287 -0.0196* -0.00724 -0.00974 (0.0171) (0.0203) (0.0183) (0.0101) (0.0145) (0.0157) Observations 9,740 9,669 9,628 17,462 13,758 9,123 Panel E: 34 to 41 Inequality Experiences 0.0315** 0.0222 0.0505*** -0.0152 0.0140 0.0432** (0.0141) (0.0302) (0.0152) (0.0116) (0.0126) (0.0171) Observations 8,411 8,329 8,300 14,933 11,916 7,989 Panel F: 42 to 49 Inequality Experiences -0.00839 0.0190 0.0107 -0.0275 0.00227 -0.0296 (0.0297) (0.0298) (0.0163) (0.0233) (0.0230) (0.0202) Observations 6,826 6,726 6,710 12,355 9,993 6,685 Panel G: 50 to 57 Inequality Experiences 0.0417 0.102*** 0.0733*** 0.0140 0.0563*** 0.0533** (0.0276) (0.0358) (0.0252) (0.0162) (0.0147) (0.0236) Observations 5,111 5,013 5,006 9,387 7,743 5,130 Panel H: Life-time experiences λ = −1 Inequality Experiences -0.0430** -0.0532*** -0.0698*** -0.0610*** -0.0845*** -0.112*** (0.0192) (0.0188) (0.0232) (0.00827) (0.0181) (0.0219) Observations 7,556 7,532 7,499 14,163 10,963 7,249 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors two-way clustered by age and cohort are displayed in parentheses. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the di�erent periods of life. Unemployment experiences are based on the experienced national unemployment rate during the di�erent periods of life. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. In Panel H we display the results on the e�ect of life-time inequality experiences based on the methodology developed by Malmendier and Nagel (2011) - we use a weighting factor of λ = -1. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 41 Table A4: GSS: Placebos (1) (2) (3) Pro-immigration Pro-guns God exists Inequality Experiences 0.0329 -0.000870 0.00392 (0.0343) (0.0104) (0.0112) Observations 8,266 30,527 15,322 R-squared 0.098 0.084 0.301 Age FE Yes Yes Yes Year FE Yes Yes Yes Cohort group FE Yes Yes Yes Census div FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experi- ences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impression- able years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations con- trol for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 42 Table A5: Allbus: Placebos (1) (2) (3) Pro-immigration Nationalism Nature determines life Inequality Experiences -0.0330 -0.0298 -0.0712 (0.0272) (0.0344) (0.0623) Observations 11,057 5,666 4,178 R-squared 0.235 0.122 0.090 Age FE Yes Yes Yes Year FE Yes Yes Yes Cohort group FE Yes Yes Yes Region FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. Table A6: ESS: Placebos (1) (2) (3) Pro-immigration Pro-EU uni�cation Pro-democratic Inequality Experiences -0.00592 -0.0210 0.0419 (0.0213) (0.0217) (0.0273) Observations 81,136 55,907 48,045 R-squared 0.209 0.109 0.070 Country FE x Age trends Yes Yes Yes Country FE x Year FE Yes Yes Yes Country FE x Cohort group FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age trends, year �xed e�ects and cohort group �xed e�ects, each interacted with country �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 43 Table A7: GSS (national): Other Experiences during the Impressionable Years (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A: Unemployment (main) Inequality Experiences -0.0370** -0.0234* -0.0147 -0.0383*** -0.0476*** -0.0414*** (0.0147) (0.0126) (0.0112) (0.0123) (0.0126) (0.0129) Unemployment Experiences 0.00868 0.000302 0.00397 0.0177** -0.00213 0.00609 (0.0158) (0.00841) (0.00823) (0.00823) (0.00757) (0.00627) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel B: Crisis Inequality Experiences -0.0331** -0.0249** -0.0115 -0.0271** -0.0529*** -0.0398*** (0.0150) (0.0109) (0.00885) (0.0115) (0.0105) (0.0123) Experienced a Crisis -0.00768 0.0268 -0.0222 -0.0584** 0.0692*** 0.0247* (0.0286) (0.0284) (0.0335) (0.0242) (0.0222) (0.0138) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel C: GDP Growth Inequality Experiences -0.0363** -0.0225** -0.00855 -0.0306** -0.0455*** -0.0317*** (0.0153) (0.0110) (0.00959) (0.0127) (0.0103) (0.0109) Experienced GDP growth -0.00925 0.00219 0.0129 0.00119 0.00910* 0.0203*** (0.0124) (0.00620) (0.00873) (0.00761) (0.00539) (0.00707) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel D: Tax Revenue Inequality Experiences -0.0430** -0.0123 -0.0378** -0.0668*** -0.0425*** -0.0501** (0.0204) (0.0162) (0.0147) (0.0179) (0.0151) (0.0199) Experienced Tax Revenue rel to GDP -0.0435 0.0433 -0.175*** -0.159** -0.0247 -0.108* (0.0957) (0.0562) (0.0536) (0.0632) (0.0471) (0.0650) Observations 22,411 24,409 27,498 38,335 43,856 31,063 Panel E: Political ideology Inequality Experiences -0.0351** -0.0226** -0.0113 -0.0317*** -0.0465*** -0.0370*** (0.0139) (0.0108) (0.00918) (0.0121) (0.0101) (0.0123) Experienced Republican President 0.0163** -0.00559 -0.0122** 0.00726 -0.0163** -0.0107 (0.00763) (0.00553) (0.00598) (0.00768) (0.00767) (0.00723) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A we show the main results. In Panel B, we show the results controlling for crisis experiences. In Panel C, we show the results controlling for experienced growth of real GDP per capita. In Panel D, we control for the experienced size of the goverment as the experienced ratio of total tax revenue to GDP. In Panel E, we control for the fraction of the impresionable years in which a republican president was in o�ce. * p < 0.10, ** p < 0.05, *** p < 0.01. 44 Table A8: Allbus: Other Experiences during the Impressionable Years (1) (2) (3) (4) (5) (6) Inequality: Inequality does not Inequality Left-wing Intention to Vote: Voted: Left Unfair increase motivation re�ects luck Left Panel A: Unemployment (main) Inequality Experiences -0.0544* -0.0428 -0.0684** -0.0956*** -0.0837*** -0.0962** (0.0307) (0.0296) (0.0349) (0.0196) (0.0298) (0.0456) Unemployment Experiences 0.108* 0.0293 0.0994* 0.00953 0.0254 0.0949 (0.0614) (0.0465) (0.0539) (0.0337) (0.0499) (0.0734) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel B: Crisis Inequality Experiences -0.0892*** -0.0522** -0.101*** -0.0978*** -0.0902*** -0.124*** (0.0212) (0.0242) (0.0271) (0.0220) (0.0266) (0.0431) Experienced a Crisis -0.249*** -0.0380 0.0728 -0.0723 -0.0185 0.107 (0.0831) (0.129) (0.289) (0.0837) (0.0745) (0.129) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel C: GDP Growth Inequality Experiences -0.0891*** -0.0518** -0.105*** -0.102*** -0.0952*** -0.127*** (0.0215) (0.0248) (0.0276) (0.0219) (0.0277) (0.0446) Experienced GDP Growth 0.00369 0.00314 -0.0253 -0.0119 -0.0131 -0.0186 (0.0341) (0.0244) (0.0305) (0.0139) (0.00937) (0.0231) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel D: Tax Revenue Inequality Experiences -0.0902*** -0.0725*** -0.0968*** -0.118*** -0.111*** -0.130*** (0.0211) (0.0246) (0.0251) (0.0209) (0.0292) (0.0498) Experienced Tax Revenue rel to GDP 0.111 0.144 -0.00542 -0.0157 -0.0111 -0.0637 (0.0853) (0.0989) (0.0490) (0.0573) (0.0824) (0.112) Panel E: Political Ideology Inequality Experiences -0.0835*** -0.0620** -0.0939*** -0.121*** -0.0993*** -0.129** (0.0204) (0.0251) (0.0277) (0.0224) (0.0288) (0.0518) Experienced a conservative chancellor 0.0139 0.0126 -0.0104 0.00211 -0.0254* -0.0158 (0.0164) (0.0134) (0.0129) (0.00987) (0.0145) (0.0241) Observations 9,974 9,955 9,902 17,740 13,602 9,046 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects . All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A we show the main results. In Panel B, we show the results controlling for crisis experiences. In Panel C, we show the results controlling for experienced growth of real GDP per capita. In Panel D, we control for the experienced size of the goverment as the experienced ratio of total tax revenue to GDP. In Panel E, we control for the fraction of the impresionable years in which a conservative chancellor was in o�ce. * p < 0.10, ** p < 0.05, *** p < 0.01. 45 Table A9: GSS: Robustness (National Inequality Experiences) (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A: Main Inequality Experiences -0.0370** -0.0234* -0.0147 -0.0383*** -0.0476*** -0.0414*** (0.0147) (0.0126) (0.0112) (0.0123) (0.0126) (0.0129) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel B: Top 1 percent Inequality Experiences -0.0398*** -0.0215* -0.0125 -0.0451*** -0.0469*** -0.0411*** (0.0141) (0.0126) (0.0112) (0.0124) (0.0123) (0.0129) Observations 23,238 26,325 29,233 40,302 46,587 33,097 Panel C: Top 10 percent Inequality Experiences -0.0361** -0.0258** -0.0163 -0.0347*** -0.0515*** -0.0426*** (0.0158) (0.0129) (0.0113) (0.0127) (0.0128) (0.0131) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel D: No missings Inequality Experiences -0.0458*** -0.0373*** -0.0221* -0.0467*** -0.0510*** -0.0437*** (0.0155) (0.0126) (0.0128) (0.0139) (0.0124) (0.0141) Observations 20,935 23,817 26,311 36,458 40,736 29,298 Panel E: No unemployment experience controls Inequality Experiences -0.0337** -0.0233** -0.0129 -0.0310** -0.0485*** -0.0383*** (0.0143) (0.0105) (0.00891) (0.0122) (0.0107) (0.0121) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel F: Age trend Inequality Experiences -0.0441*** -0.0284** -0.0183* -0.0303*** -0.0547*** -0.0428*** (0.0142) (0.0131) (0.0102) (0.0114) (0.0117) (0.0113) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Panel G: Gini coe�cient Inequality Experiences -0.0494*** -0.0157 -0.00360 -0.0750*** -0.0481*** -0.0517*** (0.0122) (0.0127) (0.00907) (0.0113) (0.00914) (0.0130) Observations 19,626 20,246 23,272 33,004 37,253 25,906 Panel H: No cohort group FE Inequality Experiences -0.0254** -0.0298*** -0.0133 -0.0294*** -0.0211** -0.0240*** (0.0113) (0.00863) (0.00825) (0.00902) (0.0107) (0.00833) Observations 23,199 26,135 29,083 40,136 46,327 32,907 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years, unless otherwise stated. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A, we show the main results. In Panel B, we show the results using the top 1 percent income share as our measure of inequality. In Panel C, we show the results using the top 10 percent income share as our measure of inequality. In Panel D, we only use observations for which we do not have missings in any of the controls. In Panel E, we do not make use of unemployment experience controls. In Panel F, we use an age trend rather than age �xed e�ects. In Panel G, we show the results using the Gini coe�cient as our measure of income inequality. In Panel H, we do not make use of cohort group �xed e�ects. * p < 0.10, ** p < 0.05, *** p < 0.01. 46 Table A10: GSS: Robustness (Regional Inequality Experiences) (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Panel A: Main Inequality Experiences -0.0415** -0.0268* -0.0142 -0.0598*** -0.0522*** -0.0377*** (0.0179) (0.0138) (0.0111) (0.0134) (0.0123) (0.0141) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel B: Top 1 percent Inequality Experiences -0.0398*** -0.0215* -0.0125 -0.0451*** -0.0469*** -0.0411*** (0.0141) (0.0126) (0.0112) (0.0124) (0.0123) (0.0129) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel C: Top 10 percent Inequality Experiences -0.0419** -0.0241* -0.0168 -0.0564*** -0.0498*** -0.0351** (0.0179) (0.0143) (0.0119) (0.0137) (0.0125) (0.0144) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel D: No missings Inequality Experiences -0.0523*** -0.0455*** -0.0180 -0.0687*** -0.0591*** -0.0417*** (0.0185) (0.0146) (0.0130) (0.0147) (0.0123) (0.0154) Observations 20,748 23,545 25,957 36,017 40,234 29,033 Panel E: No unemployment experience controls Inequality Experiences -0.0370** -0.0276** -0.0129 -0.0526*** -0.0529*** -0.0358*** (0.0176) (0.0123) (0.00923) (0.0135) (0.0109) (0.0138) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel F: Age trend Inequality Experiences -0.0479*** -0.0272* -0.0174* -0.0494*** -0.0575*** -0.0355*** (0.0176) (0.0140) (0.00966) (0.0125) (0.0111) (0.0130) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Panel G: No movers Inequality Experiences -0.0409** -0.0239* -0.0135 -0.0499*** -0.0584*** -0.0376*** (0.0203) (0.0128) (0.0141) (0.0153) (0.0128) (0.0146) Observations 17,653 19,914 22,108 30,481 35,252 24,833 Panel H: No cohort group FE Inequality Experiences -0.0274** -0.0310*** -0.0122* -0.0355*** -0.0184* -0.0168** (0.0109) (0.0100) (0.00731) (0.00943) (0.00981) (0.00733) Observations 22,987 25,831 28,670 39,632 45,703 32,597 Census div 16 FE x Age FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Year FE Yes Yes Yes Yes Yes Yes Census div 16 FE x Cohort group FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years, unless otherwise stated. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects and cohort group �xed e�ects, each interacted with census division at 16 �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A, we show the main results. In Panel B, we show the results using the top 1 percent income share as our measure of inequality. In Panel C, we show the results using the top 10 percent income share as our measure of inequality. In Panel D, we only use observations for which we do not have missings in any of the controls. In Panel E, we do not make use of unemployment experience controls. In Panel F, we use an age trend rather than age �xed e�ects. In Panel G, we show the results excluding those who moved to a di�erent census division between age 16 and the time of the interview. In Panel H, we do not make use of cohort group �xed e�ects. * p < 0.10, ** p < 0.05, *** p < 0.01. 47 Table A11: Allbus: Robustness (1) (2) (3) (4) (5) (6) Inequality: Inequality does not Inequality Left-wing Intention to Vote: Voted: Left Unfair increase motivation re�ects luck Left Panel A: Main Inequality Experiences -0.0544* -0.0428 -0.0684** -0.0956*** -0.0837*** -0.0962** (0.0307) (0.0296) (0.0349) (0.0196) (0.0298) (0.0456) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel B: Top 1 percent Inequality Experiences -0.109*** -0.0650* -0.123*** -0.125*** -0.138*** -0.141*** (0.0333) (0.0352) (0.0286) (0.0229) (0.0287) (0.0430) Observations 13,635 13,537 13,478 25,555 20,236 12,930 Panel C: Top 10 percent Inequality Experiences -0.0678 -0.0697 -0.0579 -0.127*** -0.0742* -0.0824 (0.0466) (0.0436) (0.0525) (0.0312) (0.0394) (0.0729) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel D: No missings Inequality Experiences -0.0382 -0.0610* -0.0626 -0.102*** -0.0987*** -0.0833 (0.0365) (0.0339) (0.0411) (0.0254) (0.0352) (0.0562) Observations 7,719 7,686 7,669 13,609 10,997 6,831 Panel E: No unemployment experience controls Inequality Experiences -0.0896*** -0.0523** -0.101*** -0.0983*** -0.0904*** -0.124*** (0.0212) (0.0242) (0.0271) (0.0219) (0.0266) (0.0430) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel F: Age trend Inequality Experiences -0.0548** -0.0498** -0.0864*** -0.0982*** -0.105*** -0.0987** (0.0236) (0.0213) (0.0239) (0.0186) (0.0295) (0.0478) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Panel G: Gini coe�cient Inequality Experiences -0.0188 -0.0616*** -0.0458** -0.0530*** -0.0695*** -0.0539** (0.0216) (0.0172) (0.0226) (0.00863) (0.0123) (0.0221) Observations 9,783 9,766 9,716 17,318 13,242 8,827 Panel H: No cohort group FE Inequality Experiences -0.0347* -0.0398** -0.0512** -0.0492*** -0.0674*** -0.0979*** (0.0200) (0.0202) (0.0239) (0.0132) (0.0178) (0.0294) Observations 10,401 10,357 10,309 18,979 14,691 9,533 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Cohort group FE Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the national level experienced share of total income earned by the top 5 percent during the impressionable years, unless otherwise stated. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, year �xed e�ects, cohort group �xed e�ects as well as region �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A, we show the main results. In Panel B, we show the results using the top 1 percent income share as our measure of inequality. In Panel C, we show the results using the top 10 percent income share as our measure of inequality. In Panel D, we only use observations for which we do not have missings in any of the controls. In Panel E, we do not make use of unemployment experience controls. In Panel F, we use an age trend rather than age �xed e�ects. In Panel G, we show the results using the Gini coe�cient as our measure of income inequality. In Panel H, we do not make use of cohort group �xed e�ects. * p < 0.10, ** p < 0.05, *** p < 0.01. 48 Table A12: ESS: Robustness (1) (2) (3) Pro-redistribution Left-wing Voted: Left Panel A: Main Inequality Experiences -0.0390* -0.117*** -0.121*** (0.0234) (0.0200) (0.0389) Observations 85,529 81,167 25,462 Panel B: Top 1 percent Inequality Experiences -0.0509* -0.124*** -0.127*** (0.0279) (0.0193) (0.0371) Observations 92,831 87,731 28,591 Panel C: Top 10 percent Inequality Experiences -0.0412* -0.108*** -0.114*** (0.0223) (0.0203) (0.0404) Observations 84,485 79,676 25,438 Panel D: No missings Inequality Experiences -0.0486** -0.124*** -0.118*** (0.0244) (0.0251) (0.0453) Observations 68,937 66,498 21,899 Panel E: No unemployment experience controls Inequality Experiences -0.0435* -0.118*** -0.118*** (0.0241) (0.0193) (0.0367) Observations 85,529 81,167 25,462 Panel F: Gini coe�cient Inequality Experiences -0.0850** -0.155** -0.183** (0.0344) (0.0613) (0.0795) Observations 44,670 42,077 15,852 Panel G: No cohort group FE Inequality Experiences -0.0491*** -0.106*** -0.139*** (0.0144) (0.00980) (0.0245) Observations 85,529 81,167 25,462 Country FE x Age trend Yes Yes Yes Country FE x Year FE Yes Yes Yes Country FE x Cohort group FE Yes Yes Yes Unemployment Experiences Yes Yes Yes HH controls Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years unless otherwise stated. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age trends, year �xed e�ects as well as cohort group �xed e�ects, each interacted with country �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. In Panel A, we show the main results. In Panel B, we show the results using the top 1 percent income share as our measure of inequality. In Panel C, we show the results using the top 10 percent income share as our measure of inequality. In Panel D, we only use observations for which we do not have missings in any of the controls. In Panel E, we do not make use of unemployment experience controls. In Panel F, we show the results using the Gini coe�cient as our measure of income inequality. In Panel G, we do not make use of cohort group �xed e�ects. * p < 0.10, ** p < 0.05, *** p < 0.01. 49 Table A13: GSS: Summary Stats Variable Mean Std. Dev. Min. Max. N Vote democrat 0.466 0.499 0 1 32907 Share top 10 during impr years (national) 34.879 4.16 31.779 46.294 47207 Share top 5 during impr years (national) 23.587 3.785 20.679 33.926 47207 Share top 1 during impr years (national) 10.375 2.882 7.875 17.805 47207 Gini during impr year (national) 38.566 2.035 36.838 45.225 36080 Unemployment during impr years 6.545 2.913 3.875 17.988 47207 Share top 10 during impr years (regional) 35.411 3.898 28.246 50.609 46544 Share top 5 during impr years (regional) 24.458 3.514 18.935 39.275 46544 Share top 1 during impr years (regional) 11.429 2.705 7.565 23.098 46544 Full-time employed 0.517 0.5 0 1 47207 Part-time employed 0.096 0.294 0 1 47207 Temporarily not working 0.024 0.152 0 1 47207 Unemployed 0.03 0.17 0 1 47207 Retired 0.135 0.342 0 1 47207 In School 0.011 0.103 0 1 47207 Keeping house 0.168 0.374 0 1 47207 Other labor force 0.021 0.142 0 1 47207 Married 0.675 0.468 0 1 47207 Widowed 0.072 0.258 0 1 47207 Divorced 0.107 0.31 0 1 47207 Separated 0.028 0.164 0 1 47207 Single 0.118 0.323 0 1 47207 Age 48.249 14.975 26 88 47207 Less than high school 0.207 0.405 0 1 47207 High school 0.570 0.495 0 1 47207 College 0.221 0.415 0 1 47207 Female 0.547 0.498 0 1 47207 White 0.839 0.367 0 1 47207 Black 0.133 0.339 0 1 47207 Other race 0.028 0.165 0 1 47207 Born in US 0.84 0.367 0 1 47207 Household Size 2.853 1.28 1 5 47203 Urban 0.466 0.499 0 1 47207 Protestant 0.614 0.487 0 1 47207 Catholic 0.234 0.423 0 1 47207 Jewish 0.019 0.137 0 1 47207 No religion 0.097 0.297 0 1 47207 Other religion 0.032 0.175 0 1 47207 50 Table A14: Allbus: Summary Stats Variable Mean Std. Dev. Min. Max. N Share top 10 during impr years -0.049 0.925 -0.652 5.612 20712 Share top 5 during impr years -0.06 0.885 -0.883 8.556 20712 Share top 1 during impr years -0.352 0.88 -1.278 9.106 20712 Gini during impr years -0.017 0.966 -0.842 3.322 18626 Unemployment during impr years -0.129 0.927 -1.129 3.529 20712 Age 42.72 12.798 26 102 20712 Female 0.517 0.5 0 1 20712 Education: No schooling 0.012 0.107 0 1 20712 Education: �Hauptschule� 0.422 0.494 0 1 20712 Education: Middle school 0.284 0.451 0 1 20712 Education: A-levels 0.277 0.448 0 1 20712 Married 0.686 0.464 0 1 20712 Separated 0.016 0.126 0 1 20712 Widowed 0.04 0.195 0 1 20712 Divorced 0.071 0.256 0 1 20712 Single 0.186 0.389 0 1 20712 Full-time employed 0.544 0.498 0 1 20712 Part-time employed 0.118 0.322 0 1 20712 Out of labor force 0.314 0.464 0 1 20712 Unemployed 0.007 0.081 0 1 20712 Retired 0.014 0.117 0 1 20712 Student 0.002 0.049 0 1 20712 Other employment 0 0.02 0 1 20712 Protestant 0.412 0.492 0 1 20712 Catholic 0.412 0.492 0 1 20712 Other religion 0.015 0.123 0 1 20712 No Religion 0.157 0.363 0 1 20712 Household Size 2.851 1.228 1 5 20665 51 Table A15: ESS: Summary Stats Variable Mean Std. Dev. Min. Max. N Share top 10 during impr years 31.468 3.126 21.839 43.379 79640 Share top 5 during impr years 20.898 2.622 13.343 32.606 86303 Share top 1 during impr years 8.48 1.918 4.024 17.25 86303 Gini during impr years 28.411 3.717 20.569 37.15 43918 Unemployment during impr years 6.854 4.284 0.01 35.287 86303 Male 0.479 0.5 0 1 86303 Age 47.236 12.989 26 102 86303 Below high school 0.264 0.441 0 1 86303 High school 0.451 0.498 0 1 86303 College 0.278 0.448 0 1 86303 Married 0.601 0.49 0 1 86303 Separated 0.014 0.116 0 1 86303 Divorced 0.088 0.283 0 1 86303 Widowed 0.036 0.185 0 1 86303 Never married 0.207 0.405 0 1 86303 Employed 0.789 0.408 0 1 86303 Self-employed 0.13 0.336 0 1 86303 Not in paid work 0.055 0.229 0 1 86303 Religion: Catholic 0.265 0.441 0 1 86303 Religion: Protestant 0.197 0.398 0 1 86303 Religion: Eastern Orthodox 0.001 0.023 0 1 86303 Religion: other christian 0.013 0.114 0 1 86303 Religion: Jewish 0.001 0.034 0 1 86303 Religion: Islamic 0.006 0.078 0 1 86303 Religion: Other 0.005 0.074 0 1 86303 Religion: None 0.403 0.49 0 1 86303 Household Size 2.778 1.218 1 5 86270 Income bracket (waves 1-3) 7.29 2.169 1 12 31690 Income bracket (waves 4-7) 5.905 2.748 1 10 43220 Denmark 0.021 0.145 0 1 86303 Finland 0.019 0.137 0 1 86303 France 0.189 0.391 0 1 86303 Germany 0.238 0.426 0 1 86303 Great Britain 0.21 0.408 0 1 86303 Italy 0.044 0.205 0 1 86303 Netherlands 0.078 0.268 0 1 86303 Norway 0.02 0.14 0 1 86303 Portugal 0.019 0.138 0 1 86303 Spain 0.09 0.286 0 1 86303 Sweden 0.041 0.199 0 1 86303 Switzerland 0.03 0.171 0 1 86303 52 Table A16: Experimental Sample: Summary stats Variable Mean Std. Dev. N Income 46044.776 26498.068 402 Grade 12 or less 0.015 0.121 405 Graduated high school 0.16 0.368 405 Some college no degree 0.279 0.449 405 Associate degree 0.099 0.299 405 Bachelor degree 0.323 0.468 405 Postgrad 0.116 0.321 405 Full-time employed 0.625 0.485 405 Part-time employed 0.193 0.395 405 Unemployed 0.069 0.254 405 Unemployed: not looking for a job 0.047 0.212 405 Retired 0.017 0.13 405 Other employment 0.042 0.201 405 White 0.741 0.438 402 Black 0.085 0.279 402 Hispanic 0.067 0.251 402 Age 35.706 11.222 402 53 Table A17: GSS (national inequality experiences): Main Results showing Key Controls (1) (2) (3) (4) (5) (6) Help poor Pro welfare Success due to luck Liberal Party: Democrat Voted: Democrat Inequality Experiences -0.0370** -0.0234* -0.0147 -0.0383*** -0.0476*** -0.0414*** (0.0147) (0.0126) (0.0112) (0.0123) (0.0126) (0.0129) Unemployment Experiences 0.00868 0.000302 0.00397 0.0177** -0.00213 0.00609 (0.0158) (0.00841) (0.00823) (0.00823) (0.00757) (0.00627) Female 0.137*** 0.0130 -0.101*** 0.125*** 0.126*** 0.117*** (0.0171) (0.0165) (0.0156) (0.0124) (0.0141) (0.0146) Part-time employed 0.0176 0.0452 0.0540** 0.0204 -0.0384** 0.00530 (0.0259) (0.0278) (0.0233) (0.0132) (0.0159) (0.0184) Temporarily not working 0.0397 0.0739 -0.0147 0.109*** 0.0420 0.0478 (0.0416) (0.0486) (0.0442) (0.0384) (0.0308) (0.0359) Unemployed 0.157*** 0.226*** 0.110** 0.0481* 0.0806*** 0.0913*** (0.0377) (0.0458) (0.0480) (0.0267) (0.0305) (0.0296) Retired 0.0986*** 0.120*** -0.000732 0.0152 0.0677*** 0.0836*** (0.0373) (0.0192) (0.0282) (0.0231) (0.0182) (0.0255) In school 0.0151 0.300*** 0.142*** 0.0821 0.0827 0.0421 (0.0570) (0.0506) (0.0512) (0.0569) (0.0504) (0.0640) Keeping the house 0.0631*** 0.148*** -0.0248 -0.0797*** -0.0509*** -0.0259 (0.0215) (0.0179) (0.0201) (0.0203) (0.0157) (0.0216) Other labor force 0.250*** 0.326*** -0.0295 0.0619* 0.0385 0.0635* (0.0534) (0.0503) (0.0497) (0.0372) (0.0379) (0.0351) Married -0.121*** -0.151*** -0.0789*** -0.186*** -0.115*** -0.134*** (0.0207) (0.0287) (0.0227) (0.0212) (0.0149) (0.0216) Widowed -0.0928** -0.0928*** -0.0883*** -0.0997*** -0.0287 -0.0933*** (0.0414) (0.0339) (0.0310) (0.0280) (0.0182) (0.0301) Divorced -0.0329 -0.0285 -0.00705 -0.0210 -0.0331* -0.0613** (0.0216) (0.0269) (0.0306) (0.0245) (0.0194) (0.0247) Separated 0.00189 -0.0145 -0.0570 -0.00536 -0.0853*** -0.0858*** (0.0447) (0.0489) (0.0383) (0.0329) (0.0276) (0.0308) High-school -0.244*** -0.123*** 0.0624*** -0.0958*** -0.176*** -0.190*** (0.0232) (0.0162) (0.0196) (0.0147) (0.0198) (0.0206) College -0.319*** 0.0310 0.100*** -0.0137 -0.246*** -0.137*** (0.0260) (0.0223) (0.0261) (0.0176) (0.0269) (0.0237) Black 0.526*** 0.604*** 0.180*** 0.317*** 0.888*** 1.027*** (0.0199) (0.0282) (0.0202) (0.0195) (0.0179) (0.0143) Other race 0.160*** 0.234*** 0.0784* 0.157*** 0.339*** 0.441*** (0.0349) (0.0546) (0.0459) (0.0395) (0.0307) (0.0519) Income bracket 2 0.103 0.0769 -0.0732 -0.116 0.0377 0.0604 (0.0979) (0.0756) (0.0869) (0.0851) (0.0484) (0.0752) Income bracket 3 0.0830 0.0170 -0.00125 -0.0922 0.105** 0.171** (0.112) (0.0694) (0.0867) (0.0979) (0.0535) (0.0853) Income bracket 4 0.131 -0.107 -0.107 -0.0490 0.159*** 0.164** (0.105) (0.0820) (0.0796) (0.0955) (0.0498) (0.0804) Income bracket 5 0.119 -0.109 -0.0625 -0.111 0.174*** 0.141* (0.103) (0.0731) (0.0704) (0.0944) (0.0485) (0.0751) Income bracket 6 0.111 -0.171** -0.145** -0.0804 0.115** 0.119 (0.116) (0.0741) (0.0649) (0.0841) (0.0472) (0.0764) Income bracket 7 0.00347 -0.233*** -0.0755 -0.0977 0.185*** 0.108 (0.101) (0.0785) (0.0774) (0.0896) (0.0464) (0.0825) Income bracket 8 -0.0282 -0.243*** -0.0536 -0.0447 0.183*** 0.140* (0.102) (0.0780) (0.0604) (0.0833) (0.0482) (0.0829) Income bracket 9 -0.0445 -0.345*** -0.0811 -0.0533 0.181*** 0.103 (0.0955) (0.0742) (0.0608) (0.0799) (0.0440) (0.0735) Income bracket 10 -0.0354 -0.417*** -0.152** -0.0854 0.149*** 0.0780 (0.0851) (0.0679) (0.0638) (0.0780) (0.0443) (0.0768) Income bracket 11 -0.141* -0.409*** -0.137* -0.115 0.112** 0.0423 (0.0842) (0.0646) (0.0727) (0.0802) (0.0470) (0.0724) Income bracket 12 -0.234*** -0.498*** -0.138** -0.134* 0.0209 -0.0655 (0.0878) (0.0684) (0.0625) (0.0781) (0.0415) (0.0702) Protestant -0.160*** -0.144*** -0.158*** -0.535*** -0.348*** -0.535*** (0.0188) (0.0231) (0.0244) (0.0235) (0.0188) (0.0241) Catholic -0.0755*** -0.107*** -0.114*** -0.401*** -0.0534*** -0.317*** (0.0247) (0.0233) (0.0260) (0.0247) (0.0191) (0.0232) Jewish 0.0504 0.200*** 0.0814 0.0642 0.378*** 0.196*** (0.0523) (0.0505) (0.0555) (0.0461) (0.0443) (0.0486) Other religion -0.0429 -0.112*** -0.0184 -0.327*** -0.176*** -0.240*** (0.0355) (0.0396) (0.0455) (0.0486) (0.0337) (0.0430) Cohort: 1876 - 1900 -0.464 -0.191** 0.0247 0.0176 -0.132 0.0393 (0.303) (0.0757) (0.106) (0.135) (0.136) (0.0841) Cohort: 1901 - 1925 -0.0569 -0.0102 0.0636 0.0858 -0.0332 -0.00948 (0.0601) (0.0623) (0.0523) (0.0875) (0.0778) (0.0726) Cohort: 1926 -1950 -0.0832* 0.0153 0.0477 0.0487 -0.0792 -0.0349 (0.0473) (0.0550) (0.0417) (0.0719) (0.0574) (0.0563) Cohort: 1951 - 1975 -0.0662* 0.00629 0.00460 -0.0184 -0.151*** -0.106*** (0.0399) (0.0478) (0.0333) (0.0534) (0.0418) (0.0408) Observations 23,199 26,135 29,083 40,136 46,327 32,907 R-squared 0.108 0.128 0.024 0.078 0.146 0.200 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Census div FE Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, region �xed e�ects as well as year �xed e�ects . All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 54 Table A18: Allbus: Main Results showing Key Controls (1) (2) (3) (4) (5) (6) Inequality: Inequality does not Inequality Left-wing Intention to Vote: Voted: Left Unfair increase motivation re�ects luck Left Inequality Experiences -0.0543* -0.0428 -0.0684* -0.0957*** -0.0836*** -0.0961** (0.0307) (0.0296) (0.0349) (0.0196) (0.0298) (0.0457) Unemployment Experiences 0.104* 0.0279 0.0950* 0.00877 0.0243 0.0909 (0.0586) (0.0443) (0.0515) (0.0321) (0.0477) (0.0701) Female 0.0884*** 0.126*** 0.0969*** 0.0771*** 0.0547*** 0.0425 (0.0231) (0.0240) (0.0232) (0.0132) (0.0200) (0.0322) Education: �Hauptschule� -0.185** -0.0912 -0.0306 -0.0140 -0.0189 0.0780 (0.0944) (0.117) (0.101) (0.0594) (0.0796) (0.0689) Education: �Middle School� -0.239** 0.0269 0.122 0.0226 -0.0402 0.0414 (0.0974) (0.122) (0.104) (0.0573) (0.0755) (0.0561) Education: �A-level� -0.171* 0.196* 0.370*** 0.249*** 0.119 0.244*** (0.0906) (0.116) (0.0897) (0.0593) (0.0739) (0.0653) Married -0.0541 -0.0294 -0.0389 -0.137*** -0.131*** -0.183*** (0.0341) (0.0284) (0.0289) (0.0248) (0.0214) (0.0351) Separated 0.0162 -0.0822* -0.00468 -0.119* 0.00804 -0.121 (0.0633) (0.0495) (0.0697) (0.0640) (0.0635) (0.0808) Widowed -0.0283 0.0562 0.0585 -0.163*** -0.0767** -0.127** (0.0710) (0.0573) (0.0909) (0.0436) (0.0361) (0.0552) Divorced 0.103** 0.0298 0.0797** -0.0806** 0.0261 -0.0569 (0.0521) (0.0500) (0.0360) (0.0379) (0.0381) (0.0387) Part-time employed 0.0453 0.00123 0.0389 0.0757*** 0.0958*** 0.125*** (0.0318) (0.0454) (0.0362) (0.0176) (0.0289) (0.0317) Out of the labor force 0.0837*** 0.0231 0.0536* 0.0422* 0.0748*** 0.0804*** (0.0278) (0.0285) (0.0277) (0.0229) (0.0209) (0.0224) Unemployed 0.379*** 0.164 0.378*** 0.317*** 0.325*** 0.336*** (0.0956) (0.114) (0.100) (0.123) (0.0292) (0.0995) Retired 0.0112 0.0109 0.0520 0.0405 0.207*** 0.162 (0.110) (0.136) (0.167) (0.0523) (0.0741) (0.125) Student 0.445*** 0.675*** 0.186 0.473*** 0.203* 0.289** (0.0871) (0.105) (0.163) (0.147) (0.119) (0.117) Other employment 0.573 -0.643 -0.316 0.206 0.485 0.611*** (0.502) (0.426) (0.576) (0.224) (0.453) (0.127) Protestant -0.0492** -0.0692** -0.0562* -0.177*** -0.230*** -0.204*** (0.0250) (0.0319) (0.0317) (0.0196) (0.0159) (0.0176) Catholic -0.0793*** -0.0634* -0.0505 -0.323*** -0.490*** -0.506*** (0.0242) (0.0372) (0.0351) (0.0245) (0.0224) (0.0322) Other religion -0.137* 0.0117 -0.0476 -0.112* -0.0247 -0.113 (0.0823) (0.0926) (0.0837) (0.0585) (0.0762) (0.107) Income quintile: 2 -0.0793** 0.0145 -0.000689 -0.0736* -0.130*** -0.130*** (0.0328) (0.0501) (0.0371) (0.0386) (0.0346) (0.0484) Income quintile: 3 -0.108*** -0.0784* -0.0760** -0.0946*** -0.143*** -0.135*** (0.0364) (0.0461) (0.0319) (0.0342) (0.0330) (0.0398) Income quintile: 4 -0.176*** -0.0870* -0.0970** -0.0985*** -0.150*** -0.173*** (0.0378) (0.0466) (0.0453) (0.0339) (0.0345) (0.0389) Income quintile: 5 -0.296*** -0.153*** -0.186*** -0.192*** -0.298*** -0.295*** (0.0332) (0.0465) (0.0377) (0.0389) (0.0307) (0.0438) Cohort: 1876 - 1900 0.692** -0.165 0.441 0.720*** 0.945** 0.825** (0.345) (0.342) (0.404) (0.265) (0.449) (0.400) Cohort: 1901 - 1925 0.130 0.328 (0.266) (0.422) Cohort: 1926 - 1950 -0.117** -0.0695 -0.101 -0.106* -0.0842 -0.0754 (0.0546) (0.0819) (0.0635) (0.0637) (0.0691) (0.125) Cohort: 1951 - 1975 -0.0975* -0.0474 -0.0866 -0.123** -0.0149 -0.0318 (0.0583) (0.0739) (0.0541) (0.0574) (0.0676) (0.121) Observations 10,401 10,357 10,309 18,979 14,691 9,533 R-squared 0.071 0.044 0.068 0.080 0.109 0.111 Age FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced national unemployment rate during the impressionable years. All speci�cations control for age �xed e�ects, region �xed e�ects as well as year �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 55 Table A19: ESS: Main Results showing Key Controls (1) (2) (3) Pro-redistribution Voted: Left Left-wing Inequality Experiences -0.0390* -0.117*** -0.121*** (0.0234) (0.0200) (0.0389) Unemployment Experiences -0.0328** -0.00740 0.0164 (0.0163) (0.0140) (0.0285) Male -0.119*** -0.0870*** -0.101*** (0.00963) (0.0118) (0.0163) High school -0.0756*** -0.0298** -0.0682** (0.0138) (0.0121) (0.0291) College -0.204*** 0.140*** 0.141*** (0.0114) (0.0142) (0.0273) Married -0.0751*** -0.118*** -0.0747*** (0.0130) (0.0128) (0.0230) Separated -0.0626 -0.219*** -0.120* (0.0405) (0.0537) (0.0654) Divorced -0.0259 -0.0397* -0.0404 (0.0178) (0.0211) (0.0385) Widowed -0.0550** -0.0882*** -0.0813* (0.0225) (0.0318) (0.0485) Self-Employed -0.183*** -0.232*** -0.228*** (0.0123) (0.0194) (0.0170) Not in paid work -0.0387 -0.0548*** 0.00122 (0.0255) (0.0202) (0.0381) Religion: Catholic -0.127*** -0.353*** -0.421*** (0.0158) (0.0179) (0.0232) Religion: Protestant -0.113*** -0.269*** -0.247*** (0.0119) (0.0146) (0.0227) Religion: Eastern Orthodox -0.170 0.117 -0.142 (0.192) (0.128) (0.272) Religion: Other Christian 0.0235 -0.226*** -0.0571 (0.0485) (0.0384) (0.0822) Religion: Jewish -0.502*** -0.216* -0.318* (0.158) (0.123) (0.189) Religion: Islamic 0.170*** 0.158** 0.445*** (0.0638) (0.0789) (0.0904) Religion: Other 0.0714 0.109* 0.135 (0.0630) (0.0634) (0.0931) Income bracket: 1 -0.0118 -0.141 -0.0517 (0.246) (0.193) (0.246) Income bracket: 2 0.0142 -0.202 -0.128 (0.227) (0.195) (0.240) Income bracket: 3 -0.0558 -0.155 -0.0891 (0.239) (0.192) (0.223) Income bracket: 4 -0.0943 -0.275 -0.200 (0.240) (0.196) (0.233) Income bracket: 5 -0.115 -0.257 -0.170 (0.241) (0.185) (0.238) Income bracket: 6 -0.162 -0.272 -0.190 (0.233) (0.195) (0.227) Income bracket: 7 -0.209 -0.235 -0.185 (0.231) (0.193) (0.234) Income bracket: 8 -0.275 -0.281 -0.243 (0.240) (0.189) (0.244) Income bracket: 9 -0.398* -0.298 -0.206 (0.222) (0.187) (0.223) Income bracket: 10 -0.669*** -0.443** -0.415* (0.243) (0.195) (0.232) Observations 85,529 81,167 25,462 R-squared 0.143 0.079 0.153 Country FE x Age trends Yes Yes Yes Country FE x Year FE Yes Yes Yes Country FE x Cohort group FE Yes Yes Yes HH controls Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during the impressionable years. Unemployment experiences are based on the experienced unemployment rate during the impressionable years. All speci�cations control for age trends, year �xed e�ects as well as cohort group �xed e�ects, each interacted with country �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored. * p < 0.10, ** p < 0.05, *** p < 0.01. 56 B Additional Results from the ISSP B.1 Description of the ISSP We also make use of a unique dataset containing rich data on perceptions about inequality, the International Social Survey Program (ISSP) module on Social Inequality. The ISSP has been widely used to study perceptions of social inequality, see for example Kiatpongsan and Norton (2014) or Norton and Ariely (2011). There are in total four waves of the social inequality module: one in 1987, one in 1992, one in 1999 and the last available one in 2009. On the one hand, the ISSP allows us to examine whether perceived and actual income inequality co-move. On the other hand, we provide an additional robustness check by replicating our main results on the ISSP. In table A20 we report summary statistics for the sample from the ISSP that we use to replicate our main �ndings.45 Most of our sample from the ISSP comes from six countries: Australia, France, Germany46, Norway, the United Kingdom and the US, each of which makes up for around ten percent of the sample. Canada, Denmark, Finland, Italy, Japan, the Netherlands, New Zealand, Portugal, Spain, Sweden and Switzerland together constitute about 40 percent of the overall sample. B.2 Co-movement between actual and perceived inequality B.2.1 Outcome variables: Perceptions of inequality First, we create a variable capturing people's beliefs about how much inequality there is in their countries based on their response to the following question: �These �ve diagrams show di�erent types of society. Please read the descriptions and look at the diagrams and decide which you think best describes [ COUNTRY ]: • Type A: A small elite at the top, very few people in the middle and the great mass of people at the bottom. • Type B: A society like a pyramid with a small elite at the top, more people in the middle, and most at the bottom. 45We can use a slightly larger sample to examine the correlation between actual inequality and perceived inequality because we can also use respondents who are younger than 26. 46Due to lacking inequality data we drop all respondents currently living in Eastern Germany and focus only on Western German Respondents. 57 Table A20: Summary Stats: ISSP Variable Mean Std. Dev. Min. Max. N Share top 10 during impr years 30.994 3.936 21.839 44.81 38239 Share top 5 during impr years 20.36 3.329 13.343 38.555 38974 Share top 1 during impr years 8.147 2.346 4.024 18.709 40508 Gini during impr years 31.137 5.857 20.569 44.712 18246 Unemployment during impr years 5.293 3.962 0.01 35.287 40663 Age 49.962 15.334 26 98 44918 Female 0.524 0.499 0 1 44918 Below secondary 0.45 0.498 0 1 44918 Secondary 0.227 0.419 0 1 44918 Above secondary 0.309 0.462 0 1 44918 Married 0.667 0.471 0 1 44918 Widowed 0.076 0.265 0 1 44918 Divorced 0.076 0.265 0 1 44918 Separated 0.02 0.139 0 1 44918 Single 0.154 0.361 0 1 44918 Full-time employed 0.402 0.49 0 1 44918 Part-time employed 0.082 0.275 0 1 44918 Unemployed 0.031 0.174 0 1 44918 Student 0.012 0.109 0 1 44918 Retired 0.18 0.384 0 1 44918 Other employment 0.118 0.323 0 1 44918 Catholic 0.273 0.445 0 1 44918 Church of England 0.086 0.28 0 1 44918 Protestant 0.104 0.306 0 1 44918 No religion 0.21 0.407 0 1 44918 Other religion 0.261 0.439 0 1 44918 Household Size 2.774 1.294 1 5 43151 Australia 0.138 0.345 0 1 44918 Canada 0.035 0.183 0 1 44918 Denmark 0.025 0.157 0 1 44918 Finland 0.015 0.122 0 1 44918 France 0.094 0.292 0 1 44918 Germany 0.111 0.314 0 1 44918 Great Britain 0.072 0.259 0 1 44918 Italy 0.021 0.143 0 1 44918 Japan 0.051 0.22 0 1 44918 Netherlands 0.03 0.171 0 1 44918 Norway 0.082 0.274 0 1 44918 NZL 0.059 0.235 0 1 44918 Portugal 0.051 0.22 0 1 17269 Spain 0.046 0.209 0 1 44918 Sweden 0.063 0.242 0 1 44918 Switzerland 0.025 0.157 0 1 44918 US 0.114 0.318 0 1 44918 • Type C: A pyramid except that just a few people are at the bottom. • Type D: A society with most people in the middle. • Type E: Many people near the top, and only a few near the bottom. 58 What type of society is [ COUNTRY ] today � which diagram comes closest?� We code this variable such that high values mean that people think that the country they live in today is more unequal, ranking perceived society progressively as more equal moving from type A to type E. Second, we use unique data on people's beliefs about earnings in di�erent occupations to construct measures of beliefs about the pay gaps between CEOs and unskilled workers; Cabinet ministers and unskilled workers; and doctors and unskilled workers. For example, the respondents are asked: �How much do you think an unskilled worker in a factory earns before taxes?�; or they are asked: �How much do you think a chairman of a large national company earns before taxes?� We calculate pay gaps as the ratios between the estimates for the higher-earning professions and th estimate for unskilled workers. To account for outliers we winsorize the estimated pay gaps at the 99th percentile. B.2.2 Results: Perceptions of inequality In tables A21 and A22 we show the results from regressing beliefs about inequality on actual top income shares. In some speci�cations we add country and year �xed e�ects and a set of demographic controls. Across speci�cations, we �nd that actual inequality strongly predicts people's perceived level of inequality. Table A21: ISSP: Perceptions of inequality (1) (2) (3) Belief: high inequality Belief: high inequality Belief: high inequality Current Income Share of Top 5 % 0.0371*** 0.0490*** 0.0488*** (0.00142) (0.00520) (0.00524) Observations 33,052 33,052 33,052 R-squared 0.025 0.126 0.157 Year FE No Yes Yes Country FE No Yes Yes HH controls No No Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 59 Table A22: ISSP: Estimated pay gaps (1) (2) (3) Estimated CEO Estimated cabinet minister Estimated doctor worker paygap worker paygap worker paygap Current Income Share of Top 5 % 1.786*** 0.227*** 0.153*** (0.200) (0.0368) (0.0198) Observations 43,841 43,809 44,191 R-squared 0.182 0.129 0.128 Year FE Yes Yes Yes Country FE Yes Yes Yes HH controls Yes Yes Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 B.3 Replication of main results on the ISSP B.3.1 Outcomes: Experienced inequality Our main outcome variables of interest on preferences for redistribution focus on the role the government should play and are given as follows:47 • Too much inequality: �Di�erences in income in [ COUNTRY ] are too large.� We code this variable such that high values in this question correspond to more agreement to this statement. • Tax the rich more: Moreover, we use a question on people's desired tax levels of people with di�erent income levels: �Do you think people with high incomes should pay a larger share of their income in taxes than those with low incomes, the same share, or a smaller share?� High values mean that individuals want higher shares of taxes for richer people. • Do not reduce bene�ts to the poor: �The government should spend less on bene�ts for the poor�. We code this variable such that high values indicate disagreement with this statement. • Party a�liation: Left: We also use data on people's party a�liation and their voting intention. In particular, individuals are asked for which party they intend to vote in the 47These questions are answered on a 5 point scale where 1 means �strongly agree� and 5 means �strongly disagree�. 60 next election. The data provided by the ISSP then classi�es people's voting behavior on a scale from (1) far right to (5) far left.48 • Voted: Left: Moreover, individuals are asked about their voting behavior in the last election. As before we use the derived data from the ISSP that classi�es the voting intention on a scale ranging from (1) far right to (5) far left. B.3.2 Results: Experienced inequality We show the results from the replication of our main �ndings on the ISSP sample in A23. We �nd that high inequality experiences are associated with less agreement that there is too much inequality in the respondent's country (Column 1). We �nd no signi�cant e�ect on agreement to the statement that the rich should be taxed more than the poor, even though the sign of the coe�cient goes into the expected direction (Column 2). However, people who have experienced high inequality are signi�cantly more likely to be in favor of reducing the bene�ts to the poor (Column 3) and are signi�cantly less likely to be a�liated to a left-wing party or to vote for a left-wing party (Colums 4 and 5). Taken together, the results from the ISSP strongly replicate our earlier �ndings on the samples from the GSS, Allbus and ESS. This provides us with additional con�dence in the robustness of our results. 48We set this variable to missing for all individuals who either do not intend to vote, or intend to vote for another party not part of this left-right spectrum. 61 Table A23: ISSP: Replication of main �ndings (1) (2) (3) (4) (5) Too much Tax the Do not reduce Party a�liation Voted: inequality rich more bene�ts to the poor Left Left Inequality Experiences -0.0535** -0.0101 -0.0612** -0.121** -0.0529** (0.0242) (0.0215) (0.0280) (0.0560) (0.0218) Observations 34,439 33,445 19,100 7,761 26,048 R-squared 0.142 0.075 0.103 0.058 0.075 Country FE x Age trends Yes Yes Yes Yes Yes Country FE x Year FE Yes Yes Yes Yes Yes Country FE x Cohort group FE Yes Yes Yes Yes Yes Unemployment Experiences Yes Yes Yes Yes Yes HH controls Yes Yes Yes Yes Yes Standard errors are two-way clustered by age and cohort. Inequality experiences are based on the experienced share of total income earned by the top 5 percent during impressionable years. Unemployment experiences are based on the experienced national unemployment during impressionable years. All speci�cation control for age trends, year �xed e�ects and cohort group �xed e�ects, each interacted with country �xed e�ects. All speci�cations control for a large set of controls: household income, marital status, education, employment status, household size, religion, and gender. All outcome measures are z-scored.* p < 0.10, ** p < 0.05, *** p < 0.01. 62 C Description of Outcomes C.1 General Social Survey C.1.1 Main Outcomes • Help Poor: �Some people think that the government in Washington should do everything to improve the standard of living of all poor Americans (they are at point 5 on this card). Other people think it is not the government's responsibility, and that each person should take care of himself (they are at point 1). Where are you placing yourself in this scale?� • Pro-welfare: �We are faced with many problems in this country, none of which can be solved easily or inexpensively. I am going to name some of these problems, and for each one I would like you to tell me whether you think we are spending too much money on it, too little money or about the right amount.� We focus on people's answer to that question on the issue of �assistance to the poor.� We code this variable such that higher values indicate too little assistance to the poor. • Success due to luck: �Some people say that people get ahead by their own hard work; others say that lucky breaks or help from other people are more important. Which do you think is most important?� The answer can take a value from 1 to 3: hard work is most important (1), hard work and luck are equally important (2), luck is most important (3). • Liberal: �We hear a lot of talk these days about liberals and conservatives. I am going to show you a seven-point scale on which the political views that people might hold are arranged from extremely liberal to extremely conservative. Where would you place yourself in this scale?� We coded the question such that high values mean that the respondent is liberal. • Party: Democrat: �Generally speaking, do you usually think of yourself as a Republican, Democrat, Independent, or what?� We coded this variable such that higher values corre- spond to support of the democratic party and lower values to the support of Republicans. We set observations as missing if respondents identi�ed with another party. • Voted: Democrat: We also look at people's past voting behavior. Speci�cally, this variable takes value one if the respondent voted for the democratic candidate in the last presidential election and takes value zero if the respondent voted for the republican can- 63 didate. We set this measure to missing if the respondent did not vote in the presidential election or if the respondent voted for an independent candidate. C.1.2 Mechanisms • Low relative income: People's self-assessed position in the income distribution on a �ve-point scale reaching from �Far below average� to �Far above average�. We code this variable such that high values correspond to perceived low relative income. • Low social position: People's self-assessed position in society on a four-point scale, reaching from �Lower class� to �Upper class� We code this variable such that high values indicate a perceived low social position. C.1.3 Placebo Outcomes • Pro-immigration: People's view on whether the number of immigrants should be in- creased or decreased on a �ve-point scale from (1) decrease a lot to (5) increase a lot. • Pro-guns: This variable takes value one for people opposing a law which would require a person to obtain a police permit before he or she could buy a gun. • God exists: People's belief in god on a six-point scale from (1) people not believing in God to (6) people stating that they know God really exists and that they have no doubts about it. C.2 Allbus C.2.1 Main Outcomes • Inequality: Unfair: Disagreement on 4-point scale to the statement: �I think the so- cial inequalities in this country are fair.� We coded this variable such that higher values correspond to more distaste of inequality. • Inequality does not increase motivation: This variable captures people's beliefs about the e�ect of inequality on motivation. High values mean that people think that inequality does not increase motivation. 64 • Inequality re�ects luck: Disagreement on 4-point scale to the statement: �Di�erences in rank between people are acceptable as they essentially re�ect how people used their opportunities.� High values mean that people disagree with this statement. • Left-wing: People's self-assessment of their political views on a 10-point scale. We coded this variable such that high values indicate a more left-wing self-assessment. • Intention to vote: Left: We classi�ed each party based on the classi�cation of parties on the left-right spectrum from Huber and Inglehart (1995). Higher values correspond to intentions to vote for more left-wing parties. • Voted: Left: As above we create an index for each party that our respondent voted for using the classi�cation of parties on the left-right spectrum from Huber and Inglehart (1995). Higher values of this variable mean that people voted for more left-wing parties. C.2.2 Mechanisms • Low social position: �In our society there are people who are at the top and people who are at the bottom. Where would you place yourself on such a scale?� This is coded such that high values mean that people think that they are closer to the bottom of the distribution. C.2.3 Placebo Outcomes • Pro-immigration: We construct an index of attitudes towards immigrants by looking at the following questions on a scale from (1) strongly disagree to (7) strongly agree. � Immigrants should adapt to German customs. � Immigrants should not have any right to participate politically. � Immigrants should not be allowed to marry Germans. The index is coded such that more disagreement to these statements receives higher values. • Nationalism: People's nationalism is measured on a four point scale ranging from (1) very proud to be German to (4) not very proud to be German. We code this variable such that high values indicate high nationalism. 65 • Nature determines life: People's agreement to the statement �in the �nal analysis, our life is determined by the laws of nature.� on a scale from (1) strongly agree to (5) strongly disagree. We code this variable such that high values indicate agreement to this statement. C.3 ESS C.3.1 Main Outcomes • Pro-redistribution: �The government should take measures to reduce di�erences in in- come levels.� We code this variable such that high values correspond to agreement to this statement. • Left-wing: �In politics people sometimes talk of `left' and `right'. Where would you place yourself on this scale, where 0 means the left and 10 means the right?� We recode this variable such that high values refer to people placing themselves on the left. • Voted: Left: People's voting behavior in the last election. In particular, we coded up this voting behavior on a right-left scale, taking higher values for left-wing parties and lower values for right-wing parties. As in Giuliano and Spilimbergo (2014), we used the classi�cation of parties on the left-right spectrum from Huber and Inglehart (1995). If the party was not part of Huber's classi�cation or if a person did not vote, we coded the observation as missing. C.3.2 Placebo Outcomes • Pro-immigration: We construct an index of attitudes towards immigrants by looking at the following questions with scales from (1) to (4) and (0) to (10). � �Allow many immigrants of same race/ethnic group� (4) vs. �Allow no immigrants of the same race/ethnic group� (1). � �Allow many immigrants of di�erent race/ethnic group� (4) vs. �Allow no immigrants of di�erent race/ethnic group� (1). � �Allow many immigrants from poorer countries to Europe� (4) vs. �Allow no immi- grants from poorer countries to Europe� (1). � �Immigration is good for the economy� (10) vs. �Immigration is bad for the economy� (0). 66 � �Immigration is good for cultural life� (10) vs, �Immigration is bad for cultural life� (0). � �Immigration makes this country a better place to live� (10) vs. �Immigration makes this country a worse place to live� (0). We code the index such that high values indicate more positive attitudes towards immi- grants. • Pro-EU uni�cation: �European uni�cation should go further� (10) or �European uni�- cation has gone too far� (0). • Pro-democratic: People's agreement on a 5-point scale to the statement �Political parties that wish to overthrow democracy should be banned.� 67 D Data description: Control Variables D.1 General Social Survey We control for our respondents' employment status by including dummy variables on whether the respondent is employed part-time, temporarily not working, unemployed, retired, in school, keeping the house or in other employment (the base category is full-time employment). To account for the respondent's marital status, we include the following dummies: whether the respondent is married, widowed, divorced or separated (the omitted category is �never married�). We include the following set of indicator variables to capture our respondent's educational attainment: an indicator for whether our respondent completed at most high school as well as a dummy for whether our respondent completed college (�below highschool� is the omitted category). We also include a dummy for whether our respondent is black. Following Giuliano and Spilimbergo (2014) we include dummies for each of the 12 income brackets available in the GSS to control for absolute household income. In addition, we include a set of dummies for our respondents' household size. Finally, we also control for our respondent's religion by including dummies for whether they are protestant, catholic, Jewish or whether they have another religion. Finally, we include a dummy indicating the gender of the respondent. D.2 German General Social Survey (Allbus) We control for key demographics, such as income, gender, marital status, education, religious a�liation and employment status. In particular, we control for education by including dummy variables for the type of schooling our respondent completed.49 We control for marital status by including dummy variables for whether our respondent is married, widowed, divorced or separated (single is the omitted category). We account for people's employment status by dummies for whether our respondent is part- time employed, unemployed, out of the labor force, student, retired, or in other employment (the omitted category is full-time employment). We also control for people's position in the income distribution in a given year by including dummies for quintiles of self-reported monthly household income. We also control for our respondent's religion by including dummy variables for whether our 49In particular, we include dummies for �Hauptschule�, �Realschule� and �Abitur/FH�. Below �Hauptschule� is the omitted category. 68 respondent is catholic, protestant or member of another religion (the omitted category is �no religion�). Finally, we also include a dummy variable indicating the gender of the respondent. D.3 European Social Survey We control for education by including dummy variables for whether our respondent completed at most high-school or holds a college degree (no completion of high school is the omitted category). We control for marital status by including dummy variables for whether our respondent is mar- ried, widowed, divorced or separated (single is the omitted category). We account for people's employment status by including dummy variables for whether they are self-employed or not in paid work (the omitted category is that they are employed). We also control for people's income level. For waves one to three we make use of the only available income variable which measures absolute household income levels categorized into 12 brackets. For waves four to seven we use a variable on the country-speci�c income decile that our respondent's household belongs to. We also control for household size with a set of dummies indicating whether there is one person in the household, two, three, four or more than �ve. We also control for our respondent's religion by including dummies for whether our respondent is catholic, protestant, a�liated to another Christian religion, Islamic, Jewish or a�liated to another religion (the omitted category is �no religion�). Finally, we also include a dummy variable indicating the gender of our respondent. 69 E Inequality Data We now provide an overview of the inequality data we use in our analysis. We linearly interpolate missing inequality data up to gaps of six years. In our analysis we make use of those cohorts for which this method gives inequality data for their full �impressionable years� (age 18-25). Table A24 shows the years for which inequality data are available for the di�erent countries in our sample. Table A24: Availability of Inequality Data Country Share top 10 percent Share top 5 percent Share top 1 percent Gini coe�cient Australia 1941-2010 1939-2010 1921-2010 1981-2010 Canada 1941-2010 1920-2010 1920-2010 1976-2011 Denmark 1903-2010 1903-2010 1903-2010 - Finland 1920-2009 1920-2009 1920-2009 1966-2011 France 1919-2012 1915-2012 1915-2012 1956-2011 Germany 1891-1936; 1961-2008 1891-1938; 1961-2008 1891-1938; 1957-2008 1962-2010 Italy 1974-2009 1974-2009 1974-2009 1967-2010 Ireland 1975-2009 - 1975-2009 - Japan 1947-2010 1907-1924; 1947-2010 1886-2010 1962-2001 Netherlands 1914-2012 1914-2012 1914-2012 1977-2008 New Zealand 1924-2012 1921-2012 1921-2012 1982-2009 Norway 1948-2011 1948-2011 1948-2011 1986-2011 Portugal 1976-2005 1976-2005 1976-2005 1993-2011 Spain 1981-2012 1981-2012 1981-2012 1990-2011 Sweden 1903-1920; 1930-2013 1903-1920; 1930-2013 1903-1920; 1930-2013 1975-2011 Switzerland 1933-2010 1933-2010 1933-2010 - United Kingdom 1949-2012 1949-2012 1949-2012 1961-2011 United States (national) 1917-2014 1917-2014 1913-2014 1944-2012 United States (state-level) 1917-2015 1917-2015 1917-2015 - In this table we provide an overview of the available inequality data for the countries in our sample. These data are taken from �The World Wealth and Income Database� (Alvaredo et al., 2011) and from the �Chartbook of Economic Inequality� (Atkinson and Morelli, 2014). 70 F Construction of Life-time Experiences As in Malmendier and Nagel (2011) and Malmendier and Nagel (2015), we construct a weighted average of past national-level income shares of the top �ve percent50 for each individual i in country c and in year t, using a speci�cation of weights that introduces merely one additional parameter to measure past experiences (Malmendier and Shen, 2015): IEict(λ) = ageit−1∑ k=1 wit(k,λ)Ic,t−k (5) where wit(k,λ) = (ageit − k)λ∑ageit−1 k=1 (ageit − k)λ (6) where Ic,t−k is the share of total income held by the top �ve percent of earners in year t-k. Given that the empirical literature on the role of experiences in the formation of political attitudes posits a big importance of early experiences and in particular experiences during the impressionable years (Giuliano and Spilimbergo, 2014; Krosnick and Alwin, 1989), we assume that experiences before age 18 do not matter. In other words, we construct the experience measures as the weighted average of experiences from age 18 onwards. The weights wit(k,λ) are a function of k, i.e. how distant the inequality was experienced relative to the individual's age at time t, and of the weighting parameter λ. The value of λ determines the relative importance of distant experiences compared to more recent experiences. In our estimations we use a weight of λ = −1 which gives rise to a weight that increases linearly when one moves further into the past from the survey year.51 This weighting scheme gives more importance to people's early experiences, while still allowing for some impact of more recent experiences in life.52 50We used the exact same methodology to look at alternative measures of inequality. The results looked very similar and are omitted for brevity. 51We obtain very similar results when we use weights of λ = −0.5 or λ = −2 instead. 52If λ > 0, the weights are decreasing in lag k, i.e. income inequality experienced closer to current age at time t receives higher weight. 71 G Experimental Instructions G.1 Experiment 1 G.1.1 Introduction This study is conducted by researchers from Goethe University Frankfurt and the University of Oxford. Participants will be asked to answer a few questions about their preferences, as well as a set of demographic questions. Participation in the study typically takes 2 minutes and is strictly anonymous. In order to be paid, it is necessary to �nish the survey. If you complete the survey, you will receive a �xed payment of 30 cents. Each person is only allowed to participate in the experiment once. If you encounter a technical problem, please do not restart the experiment, but contact us at dphiloxfordecon@gmail.com. If participants have further questions about this study or their rights, or if they wish to lodge a complaint or concern, they may contact us at dphiloxforde- con@gmail.com • I have read the information provided on the previous page. • I have had the opportunity to ask questions about the study. • I understand that I may withdraw from the study at any time. • I understand how to raise a concern or make a complaint. • I understand that I can only participate in this experiment once. • I understand that close attention to the survey is required for my responses to count. If you are 18 years of age or older, agree with the statements above, and freely consent to participate in the study, please click on the �I Agree� button to begin the experiment. • I Agree • I Disagree 72 G.1.2 Treatment Conditions Equality Condition: We will now ask you to complete a hypothetical task. Imagine that there are two worker who worked for us on several tasks. Worker A completed two tasks correctly, while worker B completed eight tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker A and worker B? • Give 50 cents to player A and 50 cents to player B. • Give 48 cents to player A and 52 cents to player B. Inequality Condition: We will now ask you to complete a hypothetical task. Imagine that there are two worker who worked for us on several tasks. Worker A completed two tasks correctly, while worker B completed eight tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker A and worker B? • 22 cents for worker A and 78 cents for worker B. • 20 cents for worker A and 80 cents for worker B. G.1.3 Outcome Measure: Behavioral Measure of Redistribution We will now ask you to complete a hypothetical task. Imagine that there are two workers who worked for us on several tasks. Note, these workers are NOT the same people as from the pre- vious task. 73 Worker C completed three tasks correctly, while worker D completed seven tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker C and worker D? • Give 80 cents to worker C and 20 cents to worker D. • Give 75 cents to worker C and 25 cents to worker D. • Give 70 cents to worker C and 30 cents to worker D. • Give 65 cents to worker C and 35 cents to worker D. • Give 60 cents to worker C and 40 cents to worker D. • Give 55 cents to worker C and 45 cents to worker D. • Give 50 cents to worker C and 50 cents to worker D. • Give 45 cents to worker C and 55 cents to worker D. • Give 40 cents to worker C and 60 cents to worker D. • Give 35 cents to worker C and 65 cents to worker D. • Give 30 cents to worker C and 70 cents to worker D. • Give 25 cents to worker C and 75 cents to worker D. • Give 20 cents to worker C and 80 cents to worker D. G.1.4 Demographics The main part of the survey is now over. We will now just ask you some general questions about yourself. Which of these describes you more accurately? [Male, Female] What year were you born? In which state do you currently reside? How many people are there in your household including yourself? 74 What was your annual household income (before taxes) in 2015? [Less than $10,000, Between $10,000 and $19,999, Between $20,000 and $29,999, Between $30,000 and $39,999, Between $40,000 and $49,999, Between $50,000 and $59,999, Between $60,000 and $69,999, Between $70,000 and $79,999, Between $80,000 and $99,999, More than $100,000] What is the highest level of education you have completed? [12th grade or less; Graduated high school or equivalent; Some college, no degree; Associate degree; Bachelor's degree; Post-graduate degree] What is your religion? [Christianity, Judaism, Islam, Hinduism, None, Other] What is your ethnicity? [White, Black, Hispanic, Asian, Other] What category would best describe your political orientation? [Republican, Democrat, Other] Which of these describes your current situation most accurately? [Employed full-time, Employed part-time, Unemployed and looking for a job, Unemployed but not looking for a job, Retired, Other] 75 G.2 Experimental Instructions: Experiment 2 G.2.1 Introduction This study is conducted by researchers from Goethe University Frankfurt and the University of Oxford. Participants will be asked to answer a few questions about their preferences, as well as a set of demographic questions. Participation in the study typically takes 2 minutes and is strictly anonymous. In order to be paid, it is necessary to �nish the survey. If you complete the survey, you will receive a �xed payment of 30 cents. Each person is only allowed to participate in the experiment once. If you encounter a technical problem, please do not restart the experiment, but contact us at dphiloxfordecon@gmail.com. If participants have further questions about this study or their rights, or if they wish to lodge a complaint or concern, they may contact us at dphiloxforde- con@gmail.com • I have read the information provided on the previous page. • I have had the opportunity to ask questions about the study. • I understand that I may withdraw from the study at any time. • I understand how to raise a concern or make a complaint. • I understand that I can only participate in this experiment once. • I understand that close attention to the survey is required for my responses to count. If you are 18 years of age or older, agree with the statements above, and freely consent to participate in the study, please click on the �I Agree� button to begin the experiment. • I Agree • I Disagree G.2.2 Treatment Conditions Equality Condition: 76 We will now ask you to complete a hypothetical task. Imagine that there are two worker who worked for us on several tasks. Worker A completed two tasks correctly, while worker B completed eight tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker A and worker B? • Give 50 cents to player A and 50 cents to player B. • Give 48 cents to player A and 52 cents to player B. Inequality Condition: We will now ask you to complete a hypothetical task. Imagine that there are two worker who worked for us on several tasks. Worker A completed two tasks correctly, while worker B completed eight tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker A and worker B? • Nothing for worker A and 100 cents for worker B. • 20 cents for worker A and 80 cents for worker B. G.2.3 Outcome Measure: Behavioral Measure of Redistribution We will now ask you to complete a hypothetical task. Imagine that there are two workers who worked for us on several tasks. Note, these workers are NOT the same people as from the pre- vious task. Worker C completed three tasks correctly, while worker D completed seven tasks correctly. The number of correctly completed tasks depends on both the worker's e�ort and luck. How would you like to split one dollar between worker C and worker D? 77 • Give 50 cents to worker C and 50 cents to worker D. • Give 45 cents to worker C and 55 cents to worker D. • Give 40 cents to worker C and 60 cents to worker D. • Give 35 cents to worker C and 65 cents to worker D. • Give 30 cents to worker C and 70 cents to worker D. • Give 25 cents to worker C and 75 cents to worker D. • Give 20 cents to worker C and 80 cents to worker D. 78 CESifo Working Paper No. 6251 Category 2: Public Choice December 2016 Abstract Wohlfahrt Experienced inequality and preferences for redistribution.pdf Introduction Data General Social Survey (US) German General Social Survey European Social Survey Normalizations, Controls and Missings Inequality and Unemployment Data Construction of Experience Variable Empirical Strategy and Results Empirical Specification: GSS and Allbus Empirical Specification: ESS Results Mechanisms Reference Points and Fairness Extrapolation from own circumstances Relative income Robustness ``Impressionable Years'' versus Other Years Life-time Experiences Placebo Outcomes Other Experiences during Impressionable Years Other Robustness Checks Experimental Evidence Experimental Design Stage 1: Treatment Stage 2: Redistribution Game Experiment 2 Sample Results Alternative Mechanisms Conclusion Additional Tables Additional Results from the ISSP Description of the ISSP Co-movement between actual and perceived inequality Outcome variables: Perceptions of inequality Results: Perceptions of inequality Replication of main results on the ISSP Outcomes: Experienced inequality Results: Experienced inequality Description of Outcomes General Social Survey Main Outcomes Mechanisms Placebo Outcomes Allbus Main Outcomes Mechanisms Placebo Outcomes ESS Main Outcomes Placebo Outcomes Data description: Control Variables General Social Survey German General Social Survey (Allbus) European Social Survey Inequality Data Construction of Life-time Experiences Experimental Instructions Experiment 1 Introduction Treatment Conditions Outcome Measure: Behavioral Measure of Redistribution Demographics Experimental Instructions: Experiment 2 Introduction Treatment Conditions Outcome Measure: Behavioral Measure of Redistribution