What do Germans think and know about income inequality? A survey experiment

What do Germans think and know about income inequality? A survey experiment Abstract Germans are unable to assess their own position in the income distribution of their country and do not know much about income inequality and stratification. They are well aware of their ignorance. Germans would prefer society to be more egalitarian than they perceive it. Providing accurate information about the income distribution does not change this preference for more redistribution—except among those who learn that they are net contributors in the German tax-transfer system. 1. Introduction What do Germans know and think about income inequality in their country? In a nutshell: they do not know much. In particular, they do not know their own position in the income distribution. They know that they do not know much—but across all income groups they think that inequality should be reduced. They do not change their minds when they learn more about inequality—only those who learn that they are net contributors to the tax transfer system become less supportive of more redistribution. These are the main observations from a survey experiment on the perceptions and preferences of Germans with respect to income inequality and redistribution that we conducted in early 2015 and that we report in this article. While there is some international evidence (surveyed in Section 2) that perceived inequality does not coincide with measured, ‘objective’ inequality, a detailed analysis for Germany has, to our knowledge, not been available so far. We conducted a survey in a representative sample of 1100 German households that included two randomized information treatments (see Section 3). Participants were asked for the income of their household, for their perceived own rank in the German income distribution, for their opinions on the current level of inequality and about their perceptions and preferences of social stratification. Our first observation is that survey respondents systematically fail to locate their own position in the income scale even roughly. Relatively poor respondents tend to overestimate their own rank while relatively rich respondents tend to underestimate their relative income. Cum grano salis, this suggests that the income distribution is perceived to be more equalized than it actually is. When respondents were asked which of several stylized shapes best describes the German society today, they were right only slightly more often than by chance. An unexpected second observation is that respondents across all income groups asked for more redistribution. Not only was this preference omnipresent—its strength is fairly constant across income deciles. Moreover, asked for their most-preferred pattern of stratification, respondents selected the most egalitarian ones out of the choices we gave them. In two information treatments we checked how far redistributive preference is driven by biased perceptions. The treatment group was informed about their true position in the income distribution. This information did not alter preferences for redistribution, though. One potential interpretation is that, given that already the pretreatment preferences for more redistribution were strong, a treatment that, if anything, taught respondents that income inequality was higher than they had previously thought, cannot have much effect. In a second step, members of the (first) treatment group were informed whether they were net contributors to, or net beneficiaries from, the tax-transfer system in Germany. Respondents who learned that they were losing from redistribution asked for less redistribution afterwards. We would like to emphasize the agnostic nature of our survey. In particular, we did not conduct it with a view that respondents hold—or should hold—a stable, consistent or well-argued view on redistribution. We just wanted to find out what respondents really know and think. Still, it is tempting to distill some coherence out of the responses we obtained. As far as possible, we tried; more far-reaching interpretations would be speculative and not backed by the survey data themselves. The rest of this article is organized as follows: Section 2 embeds our survey experiment and its findings into the extant literature. Section 3 describes the survey and our sample. Section 4 documents the biases in the self-assessment of income positions. Section 5 turns to the strong preferences for redistribution, both before and after informational treatment. Perceptions of and preferences for social stratification are discussed in Section 6. Section 7 shows that pocketbook attenuate preferences for redistribution. Some conclusions are offered in Section 8. Additional material is collected in an Appendix. 2. Related literature Our survey experiment on the correlations between (mis-)perceptions of inequality and views on redistribution is related to a number of contributions in the literature. 2.1 Similar surveys To the best of our knowledge, studies of similar type so far only exist for Argentina, Sweden and Norway. Cruces et al. (2013) collected data on household incomes and on the self-assessments of income ranks in the Argentine income distribution. Their study finds that the relatively poor tend to overestimate their relative positions while the relatively rich tend to underestimate theirs. When biased subjects were confronted with accurate information, (only) the preferences of the relatively poor changed in the direction of calling for more redistribution. Karadja et al. (2014) ran a similar survey experiment for Sweden. Roughly three-quarters of their respondents missed their relative position by more than 10 percentage points, and 92% of this group underestimated their position. An information treatment was largely ineffective; only conservative respondents who learned that they were richer than they thought demanded less redistribution. In a study on the effects of income transparency on well-being in Norway, Perez-Truglia (2016) finds biases in respondents’ perceptions of their own relative income ranks. An information treatment moderated these biases and made respondents change their preferences for redistribution (the gradient between redistribution preferences and actual income rank should increased). We transfer the setting of Cruces et al. (2013) and Karadja et al. (2014) to the German case. Our study differs, however, by including assessments of social stratification (for a motivation, see below) and pocketbook concerns. Our first finding—that the poor think they are richer and vice versa—is in line with previous observations. Our second observation—that better knowledge does not change minds—adds a piece of negative evidence to the mixed collection of results on information treatments. 2.2 Misperceptions of inequality Our survey respondents substantially misperceive the income distribution in Germany: they systematically fail to locate their own position on the income scale and they get the assessment of the (stylized) social stratification in Germany right only slightly more often than by mere chance. Such misperceptions on income inequality are not uncommon, irrespective of how (perceived) inequality is measured.1 Using perceived wage differences between various occupations, Osberg and Smeeding (2006) find a massive underestimation of wage inequality in USA. Kenworthy and McCall (2008) calculate perceived relative wage levels for different countries and show that perceived and actual time trends of inequality are inconsistent. Norton and Ariely (2011) exhibit a dramatic underestimation of wealth inequality in the US population. Engelhardt and Wagener (2014) construct hypothetical perceived income distributions for 26 OECD countries by aggregating the self-positioning among International Social Survey Programme (ISSP) respondents; they find that the inequality in these perceived distributions is considerably below actual inequality. Not all studies find that populations underestimate inequality in their societies. Using the ISSP question which type of society, visualized by rhomb- or pyramid-shaped graphs, best describes the society respondents were living in, Niehues (2014) and Gimpelson and Treisman (2015) show that knowledge of social stratification is low, but involves an overestimation of inequality. This suggests that studies based on individual incomes, wages or wealth and the attending self-positioning biases observe an underestimation of inequality while studies using perceived social stratification detect an overestimation of inequality. Both approaches, called the ‘comparative’ and the ‘normative’ view in Clark and D’Ambrosio (2015), differ conceptually: the first presupposes that the perceived structure of the society (or at least of its income distribution) is derived from one’s own position, relative to some reference group. By contrast, the second approach operates with the structure of society as a whole and does not require that individuals position themselves in the perceived or desired society. We combine both approaches in our survey—and indeed confirm for Germany that biases go into different directions. 2.3 Information treatments Methodologically, our survey design follows a strand of literature that uses information as an experimental treatment in a field setting. Some studies support the knowledge gap theory proposed by Tichenor et al. (1970), arguing that differences in decision quality are, to some degree, based on different levels of knowledge. In a rational choice approach this would imply that, when new information arrives, people update their beliefs (e.g. in a Bayesian fashion), which might affect their revealed preferences. For example, Duflo and Saez (2003) show this with regard to retirement plans and Jensen (2010) with regard to educational decisions. Other studies provide evidence for the knowledge–behavior gap theory, due to Hornik (1989), positing that additional information will only affect decisions and actions if it successfully changes the underlying beliefs, habits, emotions etc. on which decisions are based. The results of our rather ineffective first information treatment indicate—in line with knowledge-behavior gap theory—that information does not suffice to change minds; what matters is whether beliefs or constraints are addressed by the information treatment. Interestingly, however, the more effective second information treatment with its direct appeal to individual monetary (dis-)advantages shows a potentially promising way to make knowledge updates change behavior. 3. Survey and sample 3.1 The survey The online survey was conducted in February 2015 and interviewed a random sample of 1100 households in Germany. Data collection was performed by Norstat company.2 All participants were asked for their incomes, for a set of individual and household characteristics and general political attitudes as well as for their views and knowledge on income inequality in Germany. Two informational treatments (detailed below) followed. In terms of income, we asked respondents for the average monthly income of their household in 2014.3 To enable respondents to make meaningful comparisons of households of different size, we broadly explained to them the concept of equivalent incomes, and then informed them about their monthly net household income corrected by the modified OECD equivalence weight. We then asked them: What do you think, how many households in Germany have an equal or lower standard of living than yours? Response categories were given in deciles. We then compared respondents’ perceived decile to their actual income decile. These objective deciles were calculated from the boundaries of deciles of the German monthly net household income distribution, corrected by the modified OECD equivalence weights, based on the then most recent German Socio-Economic Panel (GSOEP 2012, v29). Social stratification was expressed by five different stylized types of society (for details see Section 6). We introduced them to our respondents and asked them to state which type best describes German society today and of which type Germany ought to be. The experiment proceeded as follows: after a first set of questions on attitudes toward income inequality and social stratification, incomes and self-assessment in the income distribution for everybody, we randomly split the group of participants into two halves who would continue with different questionnaires. Each questionnaire again posed questions on respondents’ preferences for redistribution and social stratification, but they differed in the amount of information we provided to participants: before we asked them to (re-)state their preferences, individuals in the treatment group were informed about the income distribution, their actual relative position and their self-positioning bias; the control group did not get any such information. Our design is inspired by the information treatment in Cruces et al. (2013), but we provided the treatment group with detailed information about the actual income distribution, their relative position in it and which income belongs to the relative position they estimated to be associated with. All information was given graphically and in written. To test the role of self-interest and pocketbook concerns we implemented, within the first treatment group, a second information treatment.4 Here, we informed respondents whether they are (likely to be) a net payer or net beneficiary from the German tax-transfer system. This information does not target at respondents’ beliefs but at their budget constraints. Our design allows us to use difference-in-differences approaches when assessing outcomes. We included two information treatments to check the robustness of the stated preferences for (more) redistribution. In Sections 4 and 6 we will report on the first information treatment. The second treatment will be dealt with in Section 7. 3.2 The sample Our survey was quoted according to age and gender, which consequently lead to a representative age structure in the sample.5 Treatment and control groups are also balanced along key variables like education, income, political ideology and so on. Around 92% of the respondents were born in Germany. Our sample is slightly more educated and their mean income is lower than in the general population (probably because the sample did not include earners of very high incomes).6 A comparison of selective main characteristics between the general German population and our survey sample can be seen in Table A1 in the Appendix. Table 1. Self-positioning bias by income decile (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 Notes: Bias is defined as perceived income decile—objective income decile. See Table A2 for detailed definitions. Table 1. Self-positioning bias by income decile (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 Notes: Bias is defined as perceived income decile—objective income decile. See Table A2 for detailed definitions. We dropped the first (net income below €400) and the hundredth percentile (net income above €5000) from our sample for data cleaning reasons. The remaining sample consists of 859 observations. The left-hand panel of Figure 1 shows the income distribution of our sample by income deciles, taken from the GSOEP, v29. A fully representative sample should exhibit a 10% density in every decile. In our sample, low income deciles are somewhat overrepresented, while high income deciles are underrepresented. Otherwise, the inaccuracies in the distribution of incomes are negligible. Figure 1. View largeDownload slide Distribution of objective and perceived income decile. Figure 1. View largeDownload slide Distribution of objective and perceived income decile. To capture potential correlates of attitudes towards redistribution, we constructed a number of variables for our sample (see Table A2 in the Appendix for a list). To measure whether the availability of social comparisons shapes perceptions and positions on income inequality, we defined dummy variable (reference group) with value of one when a respondent stated that his/her reference group encompasses all social classes.7 This holds for 13% of respondents; 26% state to be mainly in contact with the lower class, 62% with the middle class and 3% with the upper class. Table 2. Determinants of perceived own income decile (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 Notes: Robust standard errors (SE) in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: perceived own income decile. See Table A2 for detailed definitions. Table 2. Determinants of perceived own income decile (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 Notes: Robust standard errors (SE) in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: perceived own income decile. See Table A2 for detailed definitions. Bartels (2005, 2008) argues that perceptions of inequality are systematically shaped by political ideology, with conservatives being less aware of (changes in) inequality, even when controlling for their general political knowledge. We let respondents self-locate their ideological position on a scale from 1 (‘left’) to 10 (‘right’), from which we constructed variable ideology. The demand for redistribution can also be associated to individuals’ views on the fairness of the income distribution. Following Corneo and Grüner (2002), we asked respondents (as in the ISSP) ‘How important is hard work for getting ahead in life?’, with five categories from ‘essential’ to ‘not important at all’. We include this as a regressor (hard work), too. Media consumption may be relevant, too. We asked respondents how often (daily, weekly, monthly, rarely or never) they used different media (newspaper, TV, Internet). In all, 75% of respondents watch news in TV or read news in the internet daily, and 37% read a daily newspaper. We constructed a variable informed to summarize all media usage, with greater numbers indicating higher levels of usage. 4. Biases in self-assessments 4.1 Measurement and descriptives The right-hand part of Figure 1 shows the distribution of our respondents across (actual) income deciles, based on their self-assessments. This distribution is considerably less dispersed than the objective one. Moreover, lower income groups tend to overestimate their relative income position while higher income groups tend to underestimate the relative income. If, as in Cruces et al. (2013), these biases are the result of respondents’ flawed inferences from own social experiences with differences in incomes to the entire income distribution, our observations indicate a widespread underestimation. We defined as variable bias the difference between perceived and actual decile. A negative [positive] bias indicates an underestimation [overestimation] of one’s income decile. Table 1 provides a detailed picture of respondents’ self-positioning biases, sorted by actual income deciles. Column (1) shows that the average perceived own decile ranges from 3.106 (in the first decile) to 6.240 (in the 10th decile). In the middle of the income distribution the mean bias [column (2)] is relatively small, but it increases towards both ends. Perceptions of relatively poor respondents are positively based [columns (3) and (4)], while the relatively rich tend to underestimate their relative income position which leads to negative biases [columns (5) and (6)]. The distribution of biases is also shown in Figure A1 in the Appendix. In Table 2, we report correlates of perceived deciles other than objective relative income. As can be seen in column (1), the objective income rank is a statistically highly significant correlate. Due to the systematic bias, the regression coefficient is lower than one. This observation remains stable after including individual characteristics as age, gender, education level and political ideology. Perceived income positions are not correlated with age or gender. Regression coefficients are fairly small and not statistically significant. A higher education level—measured in highest degree—decreases the perceived relative income rank: a higher level of education level is positively correlated with a negative bias (underestimation of one’s own income rank) and negatively correlated with a positive bias (overestimation of relative income).8 The coefficient of political ideology is positive but close to zero, indicating that a more conservative ideology goes along with a slightly higher perception of one’s relative income. There is no significant correlation between the self-positioning bias and the self-assessment of respondents’ reference group, measured by variable reference group. Regressing informedness (i.e. variable informed) on bias groups, we observe, however, a significant negative correlation with positive bias and a significant positive correlation with negative bias. However, there is also a high correlation between informed and income; if we control for income, the significance of the correlations vanishes. We do not have evidence, thus, that respondents who are well informed about current affairs have a more precise picture of their income rank. As respondents systematically fail to locate themselves in the income distribution, we want to know how sure they were in their answers. A mere 14% of respondents reported that they were sure or very sure about their self-positioning, 48% were somewhat sure and 38% not sure at all. Overall, people seem to know that they do not know very much. Interestingly, the reported levels of confidence do not vary across perceived income deciles. Therefore, we can refute the objection that respondents choose middle categories for their self-positioning if they have no clue. To sum up: respondents know little about their relative income and they are aware of this fact. 5. Preferences for redistribution 5.1 Initial preferences Even before the self-assessments we had asked respondents about their general opinion on redistribution in Germany. Answers were coded in seven categories, ranging from 1 (‘There is too much effort to equalize incomes’) over 4 (‘It is fine as it is’) to 7 (‘Income inequality is far too high and should be reduced’). We take respondents’ answers as their revealed preference for more/less redistribution. An overwhelming majority of 83% of the respondents asks for more redistribution (categories 5–7), 11% are satisfied with the status quo (category 4) and merely 6% think there is too much income equalization in Germany (categories 1–3). This is a strong and surprising observation, and we had a more detailed look at respondents’ preferences for redistribution. We first study the mean preferences for redistribution by income deciles.9 As can be seen in column (1) of Table 3, mean preferences for redistribution range from 4.125 in the top perceived decile to 6.100 in the lowest perceived decile. This indicates that there is a quite uniform (average) preference for more redistribution and greater income equality across all deciles of perceived income (recall that 4 signifies a preference for the status quo). Moreover, between the second and ninth decile, the average preferences for redistribution do not show much variation. Preferences are, thus, remarkably homogeneous across income groups. Moreover, we do not observe any differences in preferences for redistribution within and across the political spectrum. Especially, the correlation between preferences for redistribution and stated political ideology is very low ( ρ=−0.16). The generally very high popularity of more redistribution from left to right is also reflected in the programs of all major German parties (excluding, possibly, the small liberal party) that all advocate ‘more’ social justice. Table 3. (Mean) Preferences for redistribution by perceived income decile (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. Table 3. (Mean) Preferences for redistribution by perceived income decile (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. These observations are in harmony with the pattern one gets from the ISSP. Summary statistics for answers from Germany can be seen in Table A4, column (1) and Figure A2. Again, mean preferences are quite similar across perceived income deciles: virtually everybody asks for more redistribution.10 Table 4. (Perceived) distribution and preferences for redistribution: experimental results (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) Notes: Robust standard errors in parentheses. *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: preferences for redistribution. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate [overestimate] their relative income rank. Initial preferences for less redistribution indicate an initial preference ≤4. Above average bias indicates a bias above mean bias (positive bias >2.5; negative bias < −2.8). Table 4. (Perceived) distribution and preferences for redistribution: experimental results (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) Notes: Robust standard errors in parentheses. *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: preferences for redistribution. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate [overestimate] their relative income rank. Initial preferences for less redistribution indicate an initial preference ≤4. Above average bias indicates a bias above mean bias (positive bias >2.5; negative bias < −2.8). 5.2 Informed preferences Preferences generally depend on the perceived relative position. If an information treatment just confirms individual perceptions, nothing should happen. But if the perception is initially wrong and then corrected by an information treatment, preferences might be updated. First insights emerge from columns (2) and (3) in Table 3.11 The columns report mean preferences for redistribution in the perceived income deciles for, respectively, treatment and control group. Differences are small, and the values do not visibly differ from the replies to the initial question, reported in column (1). Changes in preferences for redistribution by initial self-positioning bias can be seen in Figure A4. Changes are shown for treatment and control group, separately. The graphs show that the magnitudes in changes are small across all degrees of misperceptions (the large values at the extreme ends are not very informative as only very few respondents erred that dramatically in their initial assessments). To identify whether or not there is a treatment effect, we used both the simple first-difference estimator as well as a difference in differences estimator. As our sample size is too small, we cannot meaningfully estimate potential treatment effects for each pair of perceived and factual income decile. We, therefore, choose plausible larger subgroups and partitioned respondents into those who held no bias, a positive bias or a negative bias in their income assessment. For the ‘no-bias-respondents’ the information treatment just confirms their beliefs and we, thus, do not expect any treatment effect. But the information treatment (truthfully) may change the beliefs of respondents with a bias. Results are reported in Table 4. Columns (1) to (3) show the average preferences for redistribution of those who, respectively, underestimated, correctly assessed, and overestimated their relative income positions. Panel A uses the full sample—and shows that the treatment did not generate any statistically significant effects, neither for simple differences nor for differences in differences. Furthermore, differences in differences coefficients are not only statistically insignificant, but also close to zero. A plausible explanation for the lack of effect is, of course, that our information treatment—which should make most respondents conclude that inequality is more pronounced than they initially thought—further cemented their strong preference for redistribution (recall that more than 83% of the respondents stated a preference for more redistribution already at the outset). To check whether the information treatment at least impacted on those who had initially not been for more redistribution, we restrict our sample to these respondents. This shrinks the number of observations dramatically. However, as reported in Panel B of Table 4, we do not observe any statistically significant difference between control and treatment group. Nevertheless, we cannot completely reject moderate differences with the statistical power we have. In Panel C, we restrict attention to respondents who initially held a large (i.e. above-average) bias, as these individuals might have had much reason to change their views, becoming aware of their great knowledge gap. Here, the first difference estimator in column (3) reports a statistically significant difference of stated preferences between control and treatment group, which vanishes, however, when diff-in-diff is implemented. That the first difference is non-zero in a statistically significant way is, thus, likely to be an artefact of having systematic differences among those members in treatment and control group who held a positive bias. Respondents with a leftist leaning (Panel D) do not show any treatment effect either. Neither do those who attach great importance to hard work, i.e. who (initially) considered the process of income generation as fair (Panel E). With exception of column 1 in Panel D, all diff-in-diff coefficients are not just statistically insignificant but close to zero, too. Hence, overall preferences for redistribution proved to be immune against our informative update. Still, 26% of treatment group members change their preferences upon treatment (only 14% in the control group). Thus, we indeed observe a higher variation of preferences in the treatment group, but the difference is not statistically significant.12 By design, we cannot say why the treatment was ineffective: according to knowledge–behavior gap theory (see Section 2), it might not have changed individuals’ perceptions or, if it did, did not translate into changes in preferences for redistribution. Within the short time span of the survey, respondents may fail to update their beliefs on inequality or might not see enough reason to give up cherished views on the social meaning of inequality. Neither do we observe any impact of the respondents’ degree of confidence in their initial self-assessment on their disposition to change (or not to change) their views towards redistribution after the information treatment.13 6. Perceived and preferred types of society The dimensions of inequality discussed so far refer to a comparative view. We now examine perceptions and preferences for a different concept of inequality—social stratification—based on a normative view of inequality. In a stylized way, the degree of social stratification in a society can be depicted by simple graphs. As in the ISSP questionnaire, we presented five types of society to our respondents (see Figure 2) and asked them to state (a) which type best describes German society today and (b) of which type they think Germany ought to be. Figure 2. View largeDownload slide Types of society. Source: ISSP 2009 Social Inequality IV questionnaire. Figure 2. View largeDownload slide Types of society. Source: ISSP 2009 Social Inequality IV questionnaire. 6.1 Perceptions and preferences Interpreting the society types in Figure 2 as representations of the income distribution, the type which best describes today’s Germany is Type C. When asked about the actual type of the Germany society at the beginning of the survey, 29% of respondents opted for Type C [see column (1) in Table 5]. About 57% chose one of the more unequal Types A or B, and 15% thought it was one of the more equal Types D or E. For reference, a random choice among these five types of society would lead to the right answer in 20% of the time, i.e. the majority of respondents misjudge social stratification. Interestingly, they tend to overestimate inequality.14 Table 5. What type of society is Germany today? (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 Table 5. What type of society is Germany today? (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 If we turn to the ‘What do you think Germany ought to be like’ question the picture reverses: more than 80% of respondents think that the rather equal Types D and E are desirable [see column (1) in Table 6]. About 10% vouch for Type C and a mere 7% prefer the rather unequal Types A and B. Table 6. What do you think Germany ought to be like? (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 Table 6. What do you think Germany ought to be like? (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 6.2 Information treatment Our treatment does not inform respondents about the actual type of the German society but only provides additional information about the income distribution. Still, this could have helped treated participants to improve their assessments. We therefore asked both the ‘is’ and the ‘ought’ question on stratification again, after the treatment. As can be seen from Tables 5 and 6 there are indeed small differences between treatment group [column (2)] and control group [column (3)]. For a more detailed analysis of these differences we again estimated first differences and difference in differences. Tables A6 and A7 in the Appendix show the results for, respectively, the ‘is’ question and the ‘ought’ question.15 They convey similar messages as Table 4 in the previous section: there are no significant differences between treatment and control group, and the full sample specification coefficients are also close to zero. This holds irrespective of whether individuals overestimated, correctly estimated or underestimated their income position, prior to the treatment. Hence, we were either unable to alter our respondents’ beliefs or changes in beliefs did not translate into changes in preferences. Table 7. Net payer/beneficiary and preferences for redistribution (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: change in preferences which is defined as after-treatment preference minus before-treatment preferences. See Table A2 for detailed definitions. Table 7. Net payer/beneficiary and preferences for redistribution (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: change in preferences which is defined as after-treatment preference minus before-treatment preferences. See Table A2 for detailed definitions. 7. Net contributor or beneficiary? In a second treatment, we triggered pocketbook concerns. In the spirit of rational choice-approaches to income redistribution as in Meltzer and Richard (1981), the idea was to check whether learning that one belongs to the net payers or net beneficiaries in the German tax-transfer system affects one’s views on redistribution. For example, high-income earners who tend to underestimate their relative position might change their preference for more redistribution once they get informed that they would financially suffer from further inequality reduction. The treatment informs individuals about their ‘payer status’ where we (generously) described individuals up to the 65th percentile as ‘net receivers’, individuals between the 65th and 75th percentile as ‘rather neutral’, and individuals above the 75th percentile of the income distribution as ‘net payers’; these brackets were calculated from GSOEP data by subtracting (equivalized) net incomes from (equivalized) market incomes. While learning their rank in the income distribution in the first treatment—as well as in Cruces et al. (2013) and Karadja et al. (2014)—might provide respondents with a rough idea on whether they benefit or suffer from the tax-transfer system, our second treatment captures this more directly. Since treatment effects are likely to be heterogeneous again (e.g. they might vary with payer status), we generate dummy variable pay which takes value 1 if the respondent belongs to the seventh decile or higher and zero otherwise.16 About 34% of the (treated) respondents are the net payers. Among these, 94% underestimated their relative position in the income distribution, 4% held no bias and 2% overestimated their relative income. The second information treatment was only (randomly) applied to those in the previous treatment group, i.e. all individuals knew about their relative income position. Therefore, our empirical analysis focuses on the first difference between stated preferences before and after the second information treatment. Regressing this difference in preferences on pay provides a statistically highly significant coefficient of negative sign, as can be seen in column (1) of Table 7: learning to be a net payer decreases one’s preference for redistribution. Interestingly, this change of mind occurs irrespective of respondents’ political leanings and fairness perceptions: the sign and magnitude of preference changes when being informed about one’s net payer status do not vary when we control for political attitudes and the hard work-variable [see columns (2) and (3) in Table 7]. In summary, learning that they are richer or poorer than they had previously thought has no effect on individuals’ demand for redistribution – but learning that they are likely to lose from redistribution does decrease their demand. A potential explanation (beyond the scope of our survey) is that, unless primed so, people do not think about redistribution in terms of financial costs and benefits but in general and principled terms of good or fair societies. At this abstract level, the own income position might not matter much when assessing the status quo and expressing a normative view, leading to a weak correlation between income rank and preferences. Once the veil of ignorance is lifted—which happens in the second information treatment—pocketbook concerns set in. The initial information treatment had no effect: those who learn (or are assured) that they are relatively rich obviously did not consider or grasp the personal financial implications of this (new) information. This might be understandable, given that at this stage of the survey the financing of the welfare state had not played any role at all. Once the (personalized) price tag on redistribution becomes salient, rich respondents partly reconsider their policy preferences. The inertia most (rich) respondents show after the first treatment and their preference shift after the second are in line with a ‘cheap talk’-interpretation of preferences towards redistribution: people hold cherished views on how large inequality is and ought to be, and they do not change these ‘expressive’ views until they are given reason to think through its consequences for themselves. 8. Discussion and conclusion All in all, our results show that Germans are poorly informed about their own relative income, that their perception of social stratification is just slightly better, and that they are aware of their ignorance. Still, they have outspoken preferences for more redistribution. A surprising observation from our survey is the stable and strong preference for more redistribution across all incomes.17 This is in contrast to what one would expect, in particular towards the top of the income distribution and in a country that, in international comparison, a large welfare state with a considerable degree of progressivity. Part of the puzzle might be resolved when noticing that the average survey participant from income deciles 8 to 10 locates herself between deciles 5 and 6 and, thus, holds a massively negative self-perception bias. Obviously, these comparatively rich people believe that there is a considerable share of the population with (even) higher incomes—and that there are, thus, substantial resources available that could be redistributed downwards. Hence, additional redistribution looks easily feasible. Moreover, underestimating their relative position, the upper income groups might not think of themselves as massive net contributors to redistributive schemes (some might even hope to benefit financially from redistribution), which makes a greater degree egalitarianism appear costless to themselves.18 Given their preference for more redistribution we expected that it would be mainly those relatively rich respondents who underestimate their actual income position who reacted to the information treatment by expressing lower enthusiasm for redistribution. This did not happen, however. A tentative explanation—apart from the possibility that the information treatment did not help respondents to better understand inequality—would be that high-income earners indeed harbor sincere egalitarian or pro-poor preferences. Inferring from the information treatment how big the gap between rich and poor in Germany actually is or how low incomes in the poorer strata really are, might strengthen their desire for more redistribution, even after taking potential pocketbook concerns into account. Our results trigger the question why we do not observe more redistribution, given that all Germans seem to cherish it. Tentative answers are: first, the preference for more egalitarianism (which also shows up in the platforms of all major political parties in Germany) is mostly cheap talk—if programs are actually proposed, pocketbook concerns override well-intentioned preference statements. Our experiment supports this explanation: already informational clues towards net payer status weaken the preference for redistribution noticeably. Second, the ‘political system’ (government, parliament, lobby groups etc.) holds different preferences on redistribution than the citizenry. As there is no direct voting (on redistribution) in Germany, political processes might produce results that deviate from what a popular vote would dictate. Third, financial feasibility and government budget constraints limit the scope for more redistribution, even if it is wished for by voters. (Mis-)Perceptions of reality in the citizenry matter in democracies: normative views on the desirability (or lack thereof) of policy changes—more or less redistribution, say—are shaped, among others, by perceptions of the status quo. Distorted perceptions might lead to biased political choices. The links between citizens’ views and preferences and actual redistribution policies certainly deserve greater attention. Supplementary material Supplementary material is available at Socio-Economic Review Journal online. Footnotes 1 Popular misperceptions also prevail with other issues (inflation, corruption, risks etc.). See Stevenson and Duch (2013) for a discussion. A potential common root is that individuals make inferences about objective reality from the limited sample of their own experiences and observations. For example, their reference group—relatives, friends, neighbors and colleagues—is typically not a cross-section of society but less heterogeneous. This biased and limited availability of social comparison leads to biased inferences (e.g. Runciman, 1966; Evans and Kelley, 2004). 2 Norstat is a market research company (http://opinion-people.com/de). Participants in Norstat panels can collect points that can be exchanged for money. 3 In 2015—the survey year—a minimum wage was implemented in Germany. Therefore, we restrict our analysis to the previous period to avoid (unknown) biases resulting from this reform. 4 The first and the second treatment are separate events, and (potential) changes in preferences were separately surveyed. 5 Potential respondents did not know that they were going to be asked for their views on inequality or redistribution. Selection into the survey, thus, did not a priori favor people with strong views or interest in these issues. Respondents spent on average 5 min on the survey, with a median time of 4 min, 30 sec. Given the moderate length of the survey, these answering times appear appropriate. Unfortunately, we have no information about attrition rates. For people who aborted the survey, no data were saved. 6 We also checked whether or not results are robust against re-weighting our observations according to income deciles by six different age groups. Mean and median income increase noticeably, nevertheless, our results are robust against using sample weights. Main results of regressions using sample weights can be seen in Table OA.1 in the Online Appendix. 7 We asked this question at the end of the survey, after having uncovered the actual type of society. 8 Precise correlations between controls and bias groups are available on request. 9 We report results for perceived income decile because perceptions might matter more for political preferences than the (unknown) actual position. Results for mean preferences in actual income deciles are reported in Table A3. They are qualitatively the same as for perceived deciles. 10 Columns (2) and (3) in Table A4 and Figure A3 show that this phenomenon is not confined to Germany: the mean preferences for redistribution by perceived income decile for Sweden and Argentina—the two countries for which comparable studies exist, i.e. Cruces et al. (2013) and Karadja et al. (2014)—also lie in a relatively small range. 11 See columns (2) and (3) in Table A3 for mean preferences by income decile. 12 From a technical point of view, the missing treatment effect is no surprise. Mean preferences for redistribution are uniformly distributed over income deciles. Thus, if we inform respondents about their self-positioning bias, the decile changes do not imply different preferences on average (see Table A3). 13 Whether a respondent was initially wrong (and might, thus, have felt uncomfortable with the initial view or its correction) or right (and might, thus, feel encouraged to reinforce the initial view) does not play any role. Neither does the specific way in which biases are measured. See Table OA.2 in the Online Appendix. 14 Verify from column (1) in Table A5 that responses in our sample are in line with those in the ISSP 2009. The same holds for the ‘ought’-question; see column (2) in Table A5. 15 As in Section 5, results are presented by bias group because of potentially heterogeneous treatment effects. Furthermore, we ran estimates for each ‘before treatment’-type and for the full sample. Diff-in-Diff is only applied in the latter case, because both methods—first difference and diff-in-diff—obviously coincide in the former ones. 16 We used the 70th percentile rather than the more precise 75th percentile as the threshold because we framed the entire survey in deciles (e.g. respondents were informed that they belonged to the seventh or eighth decile). Our results also hold if we set the threshold at the eighth decile. 17 By contrast, Cruces et al. (2013, Figure 4) find a u-shaped pattern in the preference for more redistribution over the Argentinian income distribution. 18 Observe from Table 3 that some respondents who believe to belong to the top deciles of the income distribution indeed do ask for more redistribution (38 respondents locating themselves in the eighth decile, 4 in the ninth and 8 in the top). Still there is no bias resulting from a potentially above-average leftist ideology in this sub-sample. Acknowledgement We thank Maximilian Stockhausen for helpful suggestions. References Bartels L. M. ( 2005 ) ‘ Homer Gets a Tax Cut: Inequality and Public Policy in the American Mind’, Perspectives on Politics , 3 , 15 – 31 . Google Scholar CrossRef Search ADS Bartels L. M. ( 2008 ) Unequal Democracy: The Political Economy of the New Gilded Age , Princeton, NJ , Princeton University Press . Clark A. E. , D’Ambrosio C. ( 2015 ) ‘Attitudes to Income Inequality: Experimental and Survey Evidence’. In Atkinson A. B. , Bourguignon F. (eds) Handbook of Income Distribution , Vol. 2 , Amsterdam , North-Holland , pp. 1147 – 1208 . Corneo G. , Grüner H.-P. ( 2002 ) ‘ Individual Preferences for Political Redistribution’, Journal of Public Economics , 83 , 83 – 107 . Google Scholar CrossRef Search ADS Cruces G. , Perez-Truglia R. , Tetaz M. ( 2013 ) ‘ Biased Perception of Income Distribution and Preferences for Redistribution: Evidence from a Survey Experiment’, Journal of Public Economics , 98 , 100 – 112 . Google Scholar CrossRef Search ADS Duflo E. , Saez E. ( 2003 ) ‘ The Role of Information and Social Interactions in Retirement Plan Decisions: Evidence from a Randomized Experiment’, The Quarterly Journal of Economics , 118 , 815 – 842 . Google Scholar CrossRef Search ADS Engelhardt C. , Wagener A. ( 2014 ) Biased Perceptions of Income Inequality and Redistribution, CESifo Working Paper No. 4838, Munich, CESifo. Evans M. , Kelley J. ( 2004 ) ‘ Subjective Social Location: Data from 21 Nations’, International Journal of Public Opinion Research , 16 , 3 – 38 . Google Scholar CrossRef Search ADS Gimpelson V. , Treisman D. ( 2015 ) Misperceiving Inequality, NBER Working Paper No. 21174, Cambridge, MA, National Bureau of Economic Research. Hornik R. ( 1989 ) ‘The Knowledge-Behavior Gap in Public Information Campaigns: A Development Communication View’. In Salmon C. T. (ed.) Information Campaigns: Balancing Social Values and Social Change , London , Sage , pp. 113 – 138 . Jensen R. ( 2010 ) ‘ The (Perceived) Returns to Education and the Demand for Schooling’, Quarterly Journal of Economics , 125 , 515 – 548 . Google Scholar CrossRef Search ADS Karadja M. , Moellerstroem J. , Seim D. ( 2014 ) Richer (And Holier) Than Thou? The Effect of Relative Income Improvements on Demand For Redistribution, IFN Working Paper No. 1042, Stockholm, Research Institute of Industrial Economics (IFN). Kenworthy L. , McCall L. ( 2008 ) ‘ Inequality, Public Opinion and Redistribution’ , Socio-Economic Review , 6 , 35 – 68 . Google Scholar CrossRef Search ADS Meltzer A. H. , Richard S. F. ( 1981 ) ‘ A Rational Theory of the Size of Government’, The Journal of Political Economy , 89 , 914 – 927 . Google Scholar CrossRef Search ADS Niehues J. ( 2014 ) Subjective Perceptions Of Inequality And Redistributive Preferences: An International Comparison. Discussion Paper, Cologne, Cologne Institute for Economic Research (IW). Norton M. I. , Ariely D. ( 2011 ) ‘ Building a Better America—One Wealth Quintile at a Time’, Perspectives on Psychological Science , 6 , 9 – 12 . Google Scholar CrossRef Search ADS PubMed Osberg L. , Smeeding T. ( 2006 ) ‘ “Fair” Inequality? Attitudes toward Pay Differentials: The United States in Comparative Perspective’, American Sociological Review , 71 , 450 – 473 . Google Scholar CrossRef Search ADS Perez-Truglia R. ( 2016 ) ‘The Effects of Income Transparency on Well-Being: Evidence from a Natural Experiment’, accessed at SSRN http://ssrn.com/abstract=2657808 or http://dx.doi.org/10.2139/ssrn.2657808. Runciman W. G. ( 1966 ) Relative Deprivation and Social Justice: A Study of Attitudes to Social Inequality in Twentieth-Century England , London , Routledge Kegan Paul . Stevenson R. T. , Duch R. ( 2013 ) ‘ The Meaning and use of Subjective Perceptions in Studies of Economic Voting’, Electoral Studies , 32 , 305 – 320 . Google Scholar CrossRef Search ADS Tichenor P. J. , Donohue G. A. , Olien C. N. ( 1970 ) ‘ Mass Media Flow and Differential Growth in Knowledge’, Public Opinion Quarterly , 34 , 159 – 170 . Google Scholar CrossRef Search ADS Appendix Table A1. Comparison of survey respondents and population (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 Source: Own survey and micro census 2011. View Large Table A1. Comparison of survey respondents and population (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 Source: Own survey and micro census 2011. View Large Table A2. Variable definitions and descriptive statistics Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Notes: n = 859 for all variables. View Large Table A2. Variable definitions and descriptive statistics Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Notes: n = 859 for all variables. View Large Table A3. (Mean) Preferences for redistribution by income decile (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. View Large Table A3. (Mean) Preferences for redistribution by income decile (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. View Large Table A4. (Mean) Preferences for redistribution: ISSP 2009 (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) Notes: Preferences for redistribution are coded from 1 to 5. Survey question: ‘Differences in income in R’s country are too large.’ (5) strongly agree (4) agree (3) neither agree nor disagree (2) disagree (1) strongly disagree. See Engelhardt and Wagener (2014) for the perceived income decile in the ISSP. Source: ISSP 2009 Social Inequality IV. View Large Table A4. (Mean) Preferences for redistribution: ISSP 2009 (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) Notes: Preferences for redistribution are coded from 1 to 5. Survey question: ‘Differences in income in R’s country are too large.’ (5) strongly agree (4) agree (3) neither agree nor disagree (2) disagree (1) strongly disagree. See Engelhardt and Wagener (2014) for the perceived income decile in the ISSP. Source: ISSP 2009 Social Inequality IV. View Large Table A5. Answers ISSP 2009 (for Germany) (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 Source: ISSP 2009 Social Inequality IV. View Large Table A5. Answers ISSP 2009 (for Germany) (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 Source: ISSP 2009 Social Inequality IV. View Large Table A6. Perceived types today: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type today. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A6. Perceived types today: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type today. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A7. Preferred society types: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type preferred. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A7. Preferred society types: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type preferred. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Figure A1. View largeDownload slide Distribution of variable bias in our sample. See Table A1 for definition. Figure A1. View largeDownload slide Distribution of variable bias in our sample. See Table A1 for definition. Figure A2. View largeDownload slide Germans mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey Preferences are normalized to (0,1). Figure A2. View largeDownload slide Germans mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey Preferences are normalized to (0,1). Figure A3. View largeDownload slide Mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey. Preferences are normalized to (0,1). Figure A3. View largeDownload slide Mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey. Preferences are normalized to (0,1). Figure A4. View largeDownload slide Mean change in preferences for redistribution by initial self-positioning bias. Figure A4. View largeDownload slide Mean change in preferences for redistribution by initial self-positioning bias. © The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Socio-Economic Review Oxford University Press

What do Germans think and know about income inequality? A survey experiment

Loading next page...
 
/lp/ou_press/what-do-germans-think-and-know-about-income-inequality-a-survey-2thT7Vh3hU
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1475-1461
eISSN
1475-147X
D.O.I.
10.1093/ser/mwx036
Publisher site
See Article on Publisher Site

Abstract

Abstract Germans are unable to assess their own position in the income distribution of their country and do not know much about income inequality and stratification. They are well aware of their ignorance. Germans would prefer society to be more egalitarian than they perceive it. Providing accurate information about the income distribution does not change this preference for more redistribution—except among those who learn that they are net contributors in the German tax-transfer system. 1. Introduction What do Germans know and think about income inequality in their country? In a nutshell: they do not know much. In particular, they do not know their own position in the income distribution. They know that they do not know much—but across all income groups they think that inequality should be reduced. They do not change their minds when they learn more about inequality—only those who learn that they are net contributors to the tax transfer system become less supportive of more redistribution. These are the main observations from a survey experiment on the perceptions and preferences of Germans with respect to income inequality and redistribution that we conducted in early 2015 and that we report in this article. While there is some international evidence (surveyed in Section 2) that perceived inequality does not coincide with measured, ‘objective’ inequality, a detailed analysis for Germany has, to our knowledge, not been available so far. We conducted a survey in a representative sample of 1100 German households that included two randomized information treatments (see Section 3). Participants were asked for the income of their household, for their perceived own rank in the German income distribution, for their opinions on the current level of inequality and about their perceptions and preferences of social stratification. Our first observation is that survey respondents systematically fail to locate their own position in the income scale even roughly. Relatively poor respondents tend to overestimate their own rank while relatively rich respondents tend to underestimate their relative income. Cum grano salis, this suggests that the income distribution is perceived to be more equalized than it actually is. When respondents were asked which of several stylized shapes best describes the German society today, they were right only slightly more often than by chance. An unexpected second observation is that respondents across all income groups asked for more redistribution. Not only was this preference omnipresent—its strength is fairly constant across income deciles. Moreover, asked for their most-preferred pattern of stratification, respondents selected the most egalitarian ones out of the choices we gave them. In two information treatments we checked how far redistributive preference is driven by biased perceptions. The treatment group was informed about their true position in the income distribution. This information did not alter preferences for redistribution, though. One potential interpretation is that, given that already the pretreatment preferences for more redistribution were strong, a treatment that, if anything, taught respondents that income inequality was higher than they had previously thought, cannot have much effect. In a second step, members of the (first) treatment group were informed whether they were net contributors to, or net beneficiaries from, the tax-transfer system in Germany. Respondents who learned that they were losing from redistribution asked for less redistribution afterwards. We would like to emphasize the agnostic nature of our survey. In particular, we did not conduct it with a view that respondents hold—or should hold—a stable, consistent or well-argued view on redistribution. We just wanted to find out what respondents really know and think. Still, it is tempting to distill some coherence out of the responses we obtained. As far as possible, we tried; more far-reaching interpretations would be speculative and not backed by the survey data themselves. The rest of this article is organized as follows: Section 2 embeds our survey experiment and its findings into the extant literature. Section 3 describes the survey and our sample. Section 4 documents the biases in the self-assessment of income positions. Section 5 turns to the strong preferences for redistribution, both before and after informational treatment. Perceptions of and preferences for social stratification are discussed in Section 6. Section 7 shows that pocketbook attenuate preferences for redistribution. Some conclusions are offered in Section 8. Additional material is collected in an Appendix. 2. Related literature Our survey experiment on the correlations between (mis-)perceptions of inequality and views on redistribution is related to a number of contributions in the literature. 2.1 Similar surveys To the best of our knowledge, studies of similar type so far only exist for Argentina, Sweden and Norway. Cruces et al. (2013) collected data on household incomes and on the self-assessments of income ranks in the Argentine income distribution. Their study finds that the relatively poor tend to overestimate their relative positions while the relatively rich tend to underestimate theirs. When biased subjects were confronted with accurate information, (only) the preferences of the relatively poor changed in the direction of calling for more redistribution. Karadja et al. (2014) ran a similar survey experiment for Sweden. Roughly three-quarters of their respondents missed their relative position by more than 10 percentage points, and 92% of this group underestimated their position. An information treatment was largely ineffective; only conservative respondents who learned that they were richer than they thought demanded less redistribution. In a study on the effects of income transparency on well-being in Norway, Perez-Truglia (2016) finds biases in respondents’ perceptions of their own relative income ranks. An information treatment moderated these biases and made respondents change their preferences for redistribution (the gradient between redistribution preferences and actual income rank should increased). We transfer the setting of Cruces et al. (2013) and Karadja et al. (2014) to the German case. Our study differs, however, by including assessments of social stratification (for a motivation, see below) and pocketbook concerns. Our first finding—that the poor think they are richer and vice versa—is in line with previous observations. Our second observation—that better knowledge does not change minds—adds a piece of negative evidence to the mixed collection of results on information treatments. 2.2 Misperceptions of inequality Our survey respondents substantially misperceive the income distribution in Germany: they systematically fail to locate their own position on the income scale and they get the assessment of the (stylized) social stratification in Germany right only slightly more often than by mere chance. Such misperceptions on income inequality are not uncommon, irrespective of how (perceived) inequality is measured.1 Using perceived wage differences between various occupations, Osberg and Smeeding (2006) find a massive underestimation of wage inequality in USA. Kenworthy and McCall (2008) calculate perceived relative wage levels for different countries and show that perceived and actual time trends of inequality are inconsistent. Norton and Ariely (2011) exhibit a dramatic underestimation of wealth inequality in the US population. Engelhardt and Wagener (2014) construct hypothetical perceived income distributions for 26 OECD countries by aggregating the self-positioning among International Social Survey Programme (ISSP) respondents; they find that the inequality in these perceived distributions is considerably below actual inequality. Not all studies find that populations underestimate inequality in their societies. Using the ISSP question which type of society, visualized by rhomb- or pyramid-shaped graphs, best describes the society respondents were living in, Niehues (2014) and Gimpelson and Treisman (2015) show that knowledge of social stratification is low, but involves an overestimation of inequality. This suggests that studies based on individual incomes, wages or wealth and the attending self-positioning biases observe an underestimation of inequality while studies using perceived social stratification detect an overestimation of inequality. Both approaches, called the ‘comparative’ and the ‘normative’ view in Clark and D’Ambrosio (2015), differ conceptually: the first presupposes that the perceived structure of the society (or at least of its income distribution) is derived from one’s own position, relative to some reference group. By contrast, the second approach operates with the structure of society as a whole and does not require that individuals position themselves in the perceived or desired society. We combine both approaches in our survey—and indeed confirm for Germany that biases go into different directions. 2.3 Information treatments Methodologically, our survey design follows a strand of literature that uses information as an experimental treatment in a field setting. Some studies support the knowledge gap theory proposed by Tichenor et al. (1970), arguing that differences in decision quality are, to some degree, based on different levels of knowledge. In a rational choice approach this would imply that, when new information arrives, people update their beliefs (e.g. in a Bayesian fashion), which might affect their revealed preferences. For example, Duflo and Saez (2003) show this with regard to retirement plans and Jensen (2010) with regard to educational decisions. Other studies provide evidence for the knowledge–behavior gap theory, due to Hornik (1989), positing that additional information will only affect decisions and actions if it successfully changes the underlying beliefs, habits, emotions etc. on which decisions are based. The results of our rather ineffective first information treatment indicate—in line with knowledge-behavior gap theory—that information does not suffice to change minds; what matters is whether beliefs or constraints are addressed by the information treatment. Interestingly, however, the more effective second information treatment with its direct appeal to individual monetary (dis-)advantages shows a potentially promising way to make knowledge updates change behavior. 3. Survey and sample 3.1 The survey The online survey was conducted in February 2015 and interviewed a random sample of 1100 households in Germany. Data collection was performed by Norstat company.2 All participants were asked for their incomes, for a set of individual and household characteristics and general political attitudes as well as for their views and knowledge on income inequality in Germany. Two informational treatments (detailed below) followed. In terms of income, we asked respondents for the average monthly income of their household in 2014.3 To enable respondents to make meaningful comparisons of households of different size, we broadly explained to them the concept of equivalent incomes, and then informed them about their monthly net household income corrected by the modified OECD equivalence weight. We then asked them: What do you think, how many households in Germany have an equal or lower standard of living than yours? Response categories were given in deciles. We then compared respondents’ perceived decile to their actual income decile. These objective deciles were calculated from the boundaries of deciles of the German monthly net household income distribution, corrected by the modified OECD equivalence weights, based on the then most recent German Socio-Economic Panel (GSOEP 2012, v29). Social stratification was expressed by five different stylized types of society (for details see Section 6). We introduced them to our respondents and asked them to state which type best describes German society today and of which type Germany ought to be. The experiment proceeded as follows: after a first set of questions on attitudes toward income inequality and social stratification, incomes and self-assessment in the income distribution for everybody, we randomly split the group of participants into two halves who would continue with different questionnaires. Each questionnaire again posed questions on respondents’ preferences for redistribution and social stratification, but they differed in the amount of information we provided to participants: before we asked them to (re-)state their preferences, individuals in the treatment group were informed about the income distribution, their actual relative position and their self-positioning bias; the control group did not get any such information. Our design is inspired by the information treatment in Cruces et al. (2013), but we provided the treatment group with detailed information about the actual income distribution, their relative position in it and which income belongs to the relative position they estimated to be associated with. All information was given graphically and in written. To test the role of self-interest and pocketbook concerns we implemented, within the first treatment group, a second information treatment.4 Here, we informed respondents whether they are (likely to be) a net payer or net beneficiary from the German tax-transfer system. This information does not target at respondents’ beliefs but at their budget constraints. Our design allows us to use difference-in-differences approaches when assessing outcomes. We included two information treatments to check the robustness of the stated preferences for (more) redistribution. In Sections 4 and 6 we will report on the first information treatment. The second treatment will be dealt with in Section 7. 3.2 The sample Our survey was quoted according to age and gender, which consequently lead to a representative age structure in the sample.5 Treatment and control groups are also balanced along key variables like education, income, political ideology and so on. Around 92% of the respondents were born in Germany. Our sample is slightly more educated and their mean income is lower than in the general population (probably because the sample did not include earners of very high incomes).6 A comparison of selective main characteristics between the general German population and our survey sample can be seen in Table A1 in the Appendix. Table 1. Self-positioning bias by income decile (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 Notes: Bias is defined as perceived income decile—objective income decile. See Table A2 for detailed definitions. Table 1. Self-positioning bias by income decile (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 (1) (2) (3) (4) (5) (6) Objective decile Average perceived own decile Mean bias Proportion with positive bias Average positive bias Proportion with negative bias Average negative bias 1 3.106 2.106 0.695 3.031 0.000 0.000 2 3.330 1.330 0.582 2.547 0.154 −1.000 3 3.725 0.725 0.418 2.684 0.319 −1.241 4 4.055 0.055 0.397 1.828 0.438 −1.531 5 4.174 −0.826 0.174 1.600 0.640 −1.727 6 4.369 −1.631 0.131 1.455 0.810 −2.250 7 4.695 −2.305 0.061 1.200 0.805 −2.955 8 4.930 −3.070 0.012 1.000 0.965 −3.193 9 4.947 −4.053 0.000 0.000 1.000 −4.053 10 6.240 −3.760 0.000 0.000 1.000 −3.760 Notes: Bias is defined as perceived income decile—objective income decile. See Table A2 for detailed definitions. We dropped the first (net income below €400) and the hundredth percentile (net income above €5000) from our sample for data cleaning reasons. The remaining sample consists of 859 observations. The left-hand panel of Figure 1 shows the income distribution of our sample by income deciles, taken from the GSOEP, v29. A fully representative sample should exhibit a 10% density in every decile. In our sample, low income deciles are somewhat overrepresented, while high income deciles are underrepresented. Otherwise, the inaccuracies in the distribution of incomes are negligible. Figure 1. View largeDownload slide Distribution of objective and perceived income decile. Figure 1. View largeDownload slide Distribution of objective and perceived income decile. To capture potential correlates of attitudes towards redistribution, we constructed a number of variables for our sample (see Table A2 in the Appendix for a list). To measure whether the availability of social comparisons shapes perceptions and positions on income inequality, we defined dummy variable (reference group) with value of one when a respondent stated that his/her reference group encompasses all social classes.7 This holds for 13% of respondents; 26% state to be mainly in contact with the lower class, 62% with the middle class and 3% with the upper class. Table 2. Determinants of perceived own income decile (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 Notes: Robust standard errors (SE) in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: perceived own income decile. See Table A2 for detailed definitions. Table 2. Determinants of perceived own income decile (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 (1) (2) Mean SE Mean SE Objective decile 0.275*** (0.022) 0.275*** (0.023) Age 0.0042 (0.005) Women −0.0238 (0.125) Education −0.107* (0.062) Ideology 0.0694** (0.034) constant 2.818*** (0.124) 2.702*** (0.411) R2 0.158 0.168 n 859 859 Notes: Robust standard errors (SE) in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: perceived own income decile. See Table A2 for detailed definitions. Bartels (2005, 2008) argues that perceptions of inequality are systematically shaped by political ideology, with conservatives being less aware of (changes in) inequality, even when controlling for their general political knowledge. We let respondents self-locate their ideological position on a scale from 1 (‘left’) to 10 (‘right’), from which we constructed variable ideology. The demand for redistribution can also be associated to individuals’ views on the fairness of the income distribution. Following Corneo and Grüner (2002), we asked respondents (as in the ISSP) ‘How important is hard work for getting ahead in life?’, with five categories from ‘essential’ to ‘not important at all’. We include this as a regressor (hard work), too. Media consumption may be relevant, too. We asked respondents how often (daily, weekly, monthly, rarely or never) they used different media (newspaper, TV, Internet). In all, 75% of respondents watch news in TV or read news in the internet daily, and 37% read a daily newspaper. We constructed a variable informed to summarize all media usage, with greater numbers indicating higher levels of usage. 4. Biases in self-assessments 4.1 Measurement and descriptives The right-hand part of Figure 1 shows the distribution of our respondents across (actual) income deciles, based on their self-assessments. This distribution is considerably less dispersed than the objective one. Moreover, lower income groups tend to overestimate their relative income position while higher income groups tend to underestimate the relative income. If, as in Cruces et al. (2013), these biases are the result of respondents’ flawed inferences from own social experiences with differences in incomes to the entire income distribution, our observations indicate a widespread underestimation. We defined as variable bias the difference between perceived and actual decile. A negative [positive] bias indicates an underestimation [overestimation] of one’s income decile. Table 1 provides a detailed picture of respondents’ self-positioning biases, sorted by actual income deciles. Column (1) shows that the average perceived own decile ranges from 3.106 (in the first decile) to 6.240 (in the 10th decile). In the middle of the income distribution the mean bias [column (2)] is relatively small, but it increases towards both ends. Perceptions of relatively poor respondents are positively based [columns (3) and (4)], while the relatively rich tend to underestimate their relative income position which leads to negative biases [columns (5) and (6)]. The distribution of biases is also shown in Figure A1 in the Appendix. In Table 2, we report correlates of perceived deciles other than objective relative income. As can be seen in column (1), the objective income rank is a statistically highly significant correlate. Due to the systematic bias, the regression coefficient is lower than one. This observation remains stable after including individual characteristics as age, gender, education level and political ideology. Perceived income positions are not correlated with age or gender. Regression coefficients are fairly small and not statistically significant. A higher education level—measured in highest degree—decreases the perceived relative income rank: a higher level of education level is positively correlated with a negative bias (underestimation of one’s own income rank) and negatively correlated with a positive bias (overestimation of relative income).8 The coefficient of political ideology is positive but close to zero, indicating that a more conservative ideology goes along with a slightly higher perception of one’s relative income. There is no significant correlation between the self-positioning bias and the self-assessment of respondents’ reference group, measured by variable reference group. Regressing informedness (i.e. variable informed) on bias groups, we observe, however, a significant negative correlation with positive bias and a significant positive correlation with negative bias. However, there is also a high correlation between informed and income; if we control for income, the significance of the correlations vanishes. We do not have evidence, thus, that respondents who are well informed about current affairs have a more precise picture of their income rank. As respondents systematically fail to locate themselves in the income distribution, we want to know how sure they were in their answers. A mere 14% of respondents reported that they were sure or very sure about their self-positioning, 48% were somewhat sure and 38% not sure at all. Overall, people seem to know that they do not know very much. Interestingly, the reported levels of confidence do not vary across perceived income deciles. Therefore, we can refute the objection that respondents choose middle categories for their self-positioning if they have no clue. To sum up: respondents know little about their relative income and they are aware of this fact. 5. Preferences for redistribution 5.1 Initial preferences Even before the self-assessments we had asked respondents about their general opinion on redistribution in Germany. Answers were coded in seven categories, ranging from 1 (‘There is too much effort to equalize incomes’) over 4 (‘It is fine as it is’) to 7 (‘Income inequality is far too high and should be reduced’). We take respondents’ answers as their revealed preference for more/less redistribution. An overwhelming majority of 83% of the respondents asks for more redistribution (categories 5–7), 11% are satisfied with the status quo (category 4) and merely 6% think there is too much income equalization in Germany (categories 1–3). This is a strong and surprising observation, and we had a more detailed look at respondents’ preferences for redistribution. We first study the mean preferences for redistribution by income deciles.9 As can be seen in column (1) of Table 3, mean preferences for redistribution range from 4.125 in the top perceived decile to 6.100 in the lowest perceived decile. This indicates that there is a quite uniform (average) preference for more redistribution and greater income equality across all deciles of perceived income (recall that 4 signifies a preference for the status quo). Moreover, between the second and ninth decile, the average preferences for redistribution do not show much variation. Preferences are, thus, remarkably homogeneous across income groups. Moreover, we do not observe any differences in preferences for redistribution within and across the political spectrum. Especially, the correlation between preferences for redistribution and stated political ideology is very low ( ρ=−0.16). The generally very high popularity of more redistribution from left to right is also reflected in the programs of all major German parties (excluding, possibly, the small liberal party) that all advocate ‘more’ social justice. Table 3. (Mean) Preferences for redistribution by perceived income decile (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. Table 3. (Mean) Preferences for redistribution by perceived income decile (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) (1) (2) (3) (4) Initial preferences 1st treatment (treated) 1st treatment (control) 2nd treatment (of the treated) Perceived decile Mean SE Mean SE Mean SE Mean SE 1 6.100 (1.241) 6.243 (0.955) 5.906 (1.304) 6.162 (0.958) 2 5.763 (1.319) 5.933 (1.087) 5.731 (1.430) 5.683 (1.308) 3 5.730 (1.186) 5.702 (1.144) 6.048 (0.877) 5.606 (1.280) 4 5.679 (1.212) 5.707 (1.250) 5.839 (1.059) 5.480 (1.379) 5 5.430 (1.426) 5.347 (1.465) 5.500 (1.422) 5.236 (1.477) 6 5.543 (1.456) 5.452 (1.418) 5.653 (1.451) 5.310 (1.490) 7 5.592 (1.308) 5.714 (1.132) 5.765 (1.208) 5.476 (1.194) 8 5.500 (1.767) 5.700 (1.342) 6.118 (1.054) 5.600 (1.603) 9 6.000 (1.155) 6.500 (0.707) 7.000 (0.000) 6.000 (1.414) 10 4.125 (1.642) 4.250 (0.957) 4.250 (1.258) 3.750 (2.217) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. These observations are in harmony with the pattern one gets from the ISSP. Summary statistics for answers from Germany can be seen in Table A4, column (1) and Figure A2. Again, mean preferences are quite similar across perceived income deciles: virtually everybody asks for more redistribution.10 Table 4. (Perceived) distribution and preferences for redistribution: experimental results (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) Notes: Robust standard errors in parentheses. *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: preferences for redistribution. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate [overestimate] their relative income rank. Initial preferences for less redistribution indicate an initial preference ≤4. Above average bias indicates a bias above mean bias (positive bias >2.5; negative bias < −2.8). Table 4. (Perceived) distribution and preferences for redistribution: experimental results (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) (1) (2) (3) Negative bias No bias Positive bias Panel A: full sample  Treatment group [observations (obs.)] 5.694 (248) 5.696 (69) 5.679 (131)  Control group (obs.) 5.681 (220) 5.897 (68) 5.923 (118)  Difference [standard error (s.e.)] 0.012 (0.110) −0.201 (0.222) −0.244 (0.169)  Diff-in-Diff (s.e.) 0.079 (0.162) −0.060 (0.332) 0.005 (0.242) Panel B: initial preference for less Redistribution  Treatment group (obs.) 4.098 (41) 3.875 (16) 4.000 (24)  Control group (obs.) 3.897 (39) 4.000 (8) 3.950 (20)  Difference (s.e.) 0.200 (0.257) −0.125 (0.552) 0.050 (0.479)  Diff-in-Diff (s.e.) 0.249 (0.329) −0.750 (0.697) −0.108 (0.570) Panel C: above average bias  Treatment group (obs.) 5.631 (122) 5.696 (69) 5.435 (46)  Control group (obs.) 5.570 (114) 5.897 (68) 5.978 (45)  Difference (s.e.) 0.061 (0.159) −0.201 (0.222) −0.543* (0.292)  Diff-in-Diff (s.e.) 0.032 (0.235) −0.060 (0.332) −0.173 (0.427) Panel D: leftist attitude  Treatment group (obs.) 6.000 (78) 6.259 (27) 5.700 (40)  Control group (obs.) 5.989 (90) 5.885 (26) 6.119 (42)  Difference (s.e.) 0.011 (0.157) 0.375 (0.282) −0.419* (0.280)  Diff-in-Diff (s.e.) 0.830 (0.237) −0.001 (0.414) −0.020 (0.427) Panel E: hard work is important  Treatment group (obs.) 5.677 (164) 5.636 (44) 5.677 (96)  Control group (obs.) 5.711 (152) 5.744 (39) 6.108 (83)  Difference (s.e.) −0.034 (0.133) −0.107 (0.309) −0.431** (0.189)  Diff-in-Diff (s.e.) 0.045 (0.197) 0.013 (0.458) 0.007 (0.272) Notes: Robust standard errors in parentheses. *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: preferences for redistribution. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate [overestimate] their relative income rank. Initial preferences for less redistribution indicate an initial preference ≤4. Above average bias indicates a bias above mean bias (positive bias >2.5; negative bias < −2.8). 5.2 Informed preferences Preferences generally depend on the perceived relative position. If an information treatment just confirms individual perceptions, nothing should happen. But if the perception is initially wrong and then corrected by an information treatment, preferences might be updated. First insights emerge from columns (2) and (3) in Table 3.11 The columns report mean preferences for redistribution in the perceived income deciles for, respectively, treatment and control group. Differences are small, and the values do not visibly differ from the replies to the initial question, reported in column (1). Changes in preferences for redistribution by initial self-positioning bias can be seen in Figure A4. Changes are shown for treatment and control group, separately. The graphs show that the magnitudes in changes are small across all degrees of misperceptions (the large values at the extreme ends are not very informative as only very few respondents erred that dramatically in their initial assessments). To identify whether or not there is a treatment effect, we used both the simple first-difference estimator as well as a difference in differences estimator. As our sample size is too small, we cannot meaningfully estimate potential treatment effects for each pair of perceived and factual income decile. We, therefore, choose plausible larger subgroups and partitioned respondents into those who held no bias, a positive bias or a negative bias in their income assessment. For the ‘no-bias-respondents’ the information treatment just confirms their beliefs and we, thus, do not expect any treatment effect. But the information treatment (truthfully) may change the beliefs of respondents with a bias. Results are reported in Table 4. Columns (1) to (3) show the average preferences for redistribution of those who, respectively, underestimated, correctly assessed, and overestimated their relative income positions. Panel A uses the full sample—and shows that the treatment did not generate any statistically significant effects, neither for simple differences nor for differences in differences. Furthermore, differences in differences coefficients are not only statistically insignificant, but also close to zero. A plausible explanation for the lack of effect is, of course, that our information treatment—which should make most respondents conclude that inequality is more pronounced than they initially thought—further cemented their strong preference for redistribution (recall that more than 83% of the respondents stated a preference for more redistribution already at the outset). To check whether the information treatment at least impacted on those who had initially not been for more redistribution, we restrict our sample to these respondents. This shrinks the number of observations dramatically. However, as reported in Panel B of Table 4, we do not observe any statistically significant difference between control and treatment group. Nevertheless, we cannot completely reject moderate differences with the statistical power we have. In Panel C, we restrict attention to respondents who initially held a large (i.e. above-average) bias, as these individuals might have had much reason to change their views, becoming aware of their great knowledge gap. Here, the first difference estimator in column (3) reports a statistically significant difference of stated preferences between control and treatment group, which vanishes, however, when diff-in-diff is implemented. That the first difference is non-zero in a statistically significant way is, thus, likely to be an artefact of having systematic differences among those members in treatment and control group who held a positive bias. Respondents with a leftist leaning (Panel D) do not show any treatment effect either. Neither do those who attach great importance to hard work, i.e. who (initially) considered the process of income generation as fair (Panel E). With exception of column 1 in Panel D, all diff-in-diff coefficients are not just statistically insignificant but close to zero, too. Hence, overall preferences for redistribution proved to be immune against our informative update. Still, 26% of treatment group members change their preferences upon treatment (only 14% in the control group). Thus, we indeed observe a higher variation of preferences in the treatment group, but the difference is not statistically significant.12 By design, we cannot say why the treatment was ineffective: according to knowledge–behavior gap theory (see Section 2), it might not have changed individuals’ perceptions or, if it did, did not translate into changes in preferences for redistribution. Within the short time span of the survey, respondents may fail to update their beliefs on inequality or might not see enough reason to give up cherished views on the social meaning of inequality. Neither do we observe any impact of the respondents’ degree of confidence in their initial self-assessment on their disposition to change (or not to change) their views towards redistribution after the information treatment.13 6. Perceived and preferred types of society The dimensions of inequality discussed so far refer to a comparative view. We now examine perceptions and preferences for a different concept of inequality—social stratification—based on a normative view of inequality. In a stylized way, the degree of social stratification in a society can be depicted by simple graphs. As in the ISSP questionnaire, we presented five types of society to our respondents (see Figure 2) and asked them to state (a) which type best describes German society today and (b) of which type they think Germany ought to be. Figure 2. View largeDownload slide Types of society. Source: ISSP 2009 Social Inequality IV questionnaire. Figure 2. View largeDownload slide Types of society. Source: ISSP 2009 Social Inequality IV questionnaire. 6.1 Perceptions and preferences Interpreting the society types in Figure 2 as representations of the income distribution, the type which best describes today’s Germany is Type C. When asked about the actual type of the Germany society at the beginning of the survey, 29% of respondents opted for Type C [see column (1) in Table 5]. About 57% chose one of the more unequal Types A or B, and 15% thought it was one of the more equal Types D or E. For reference, a random choice among these five types of society would lead to the right answer in 20% of the time, i.e. the majority of respondents misjudge social stratification. Interestingly, they tend to overestimate inequality.14 Table 5. What type of society is Germany today? (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 Table 5. What type of society is Germany today? (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 (1) (2) (3) Initial type today (%) First treatment (treated) (%) First treatment (control) (%) Type A 21.65 26.34 21.90 Type B 34.92 29.91 32.36 Type C 28.52 27.23 30.90 Type D 10.71 11.83 9.49 Type E 4.19 4.69 5.35 If we turn to the ‘What do you think Germany ought to be like’ question the picture reverses: more than 80% of respondents think that the rather equal Types D and E are desirable [see column (1) in Table 6]. About 10% vouch for Type C and a mere 7% prefer the rather unequal Types A and B. Table 6. What do you think Germany ought to be like? (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 Table 6. What do you think Germany ought to be like? (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 (1) (2) (3) Initial type preferred (%) First treatment (treated) (%) First treatment (control) (%) Type A 1.86 2.01 1.48 Type B 5.01 4.91 6.42 Type C 10.48 13.39 12.10 Type D 63.33 62.28 60.49 Type E 19.32 17.41 19.51 6.2 Information treatment Our treatment does not inform respondents about the actual type of the German society but only provides additional information about the income distribution. Still, this could have helped treated participants to improve their assessments. We therefore asked both the ‘is’ and the ‘ought’ question on stratification again, after the treatment. As can be seen from Tables 5 and 6 there are indeed small differences between treatment group [column (2)] and control group [column (3)]. For a more detailed analysis of these differences we again estimated first differences and difference in differences. Tables A6 and A7 in the Appendix show the results for, respectively, the ‘is’ question and the ‘ought’ question.15 They convey similar messages as Table 4 in the previous section: there are no significant differences between treatment and control group, and the full sample specification coefficients are also close to zero. This holds irrespective of whether individuals overestimated, correctly estimated or underestimated their income position, prior to the treatment. Hence, we were either unable to alter our respondents’ beliefs or changes in beliefs did not translate into changes in preferences. Table 7. Net payer/beneficiary and preferences for redistribution (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: change in preferences which is defined as after-treatment preference minus before-treatment preferences. See Table A2 for detailed definitions. Table 7. Net payer/beneficiary and preferences for redistribution (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 (1) (2) (3) Coef. SE Coef. SE Coef. SE Pay −0.251*** (0.001) −0.251*** (0.073) −0.257*** (0.073) Hard work −0.087 (0.074) Ideology 0.017 (0.018) Constant −0.078* (0.042) −0.019 (0.066) −0.160 (0.101) R2 0.026 0.029 0.028 n 448 448 448 Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: change in preferences which is defined as after-treatment preference minus before-treatment preferences. See Table A2 for detailed definitions. 7. Net contributor or beneficiary? In a second treatment, we triggered pocketbook concerns. In the spirit of rational choice-approaches to income redistribution as in Meltzer and Richard (1981), the idea was to check whether learning that one belongs to the net payers or net beneficiaries in the German tax-transfer system affects one’s views on redistribution. For example, high-income earners who tend to underestimate their relative position might change their preference for more redistribution once they get informed that they would financially suffer from further inequality reduction. The treatment informs individuals about their ‘payer status’ where we (generously) described individuals up to the 65th percentile as ‘net receivers’, individuals between the 65th and 75th percentile as ‘rather neutral’, and individuals above the 75th percentile of the income distribution as ‘net payers’; these brackets were calculated from GSOEP data by subtracting (equivalized) net incomes from (equivalized) market incomes. While learning their rank in the income distribution in the first treatment—as well as in Cruces et al. (2013) and Karadja et al. (2014)—might provide respondents with a rough idea on whether they benefit or suffer from the tax-transfer system, our second treatment captures this more directly. Since treatment effects are likely to be heterogeneous again (e.g. they might vary with payer status), we generate dummy variable pay which takes value 1 if the respondent belongs to the seventh decile or higher and zero otherwise.16 About 34% of the (treated) respondents are the net payers. Among these, 94% underestimated their relative position in the income distribution, 4% held no bias and 2% overestimated their relative income. The second information treatment was only (randomly) applied to those in the previous treatment group, i.e. all individuals knew about their relative income position. Therefore, our empirical analysis focuses on the first difference between stated preferences before and after the second information treatment. Regressing this difference in preferences on pay provides a statistically highly significant coefficient of negative sign, as can be seen in column (1) of Table 7: learning to be a net payer decreases one’s preference for redistribution. Interestingly, this change of mind occurs irrespective of respondents’ political leanings and fairness perceptions: the sign and magnitude of preference changes when being informed about one’s net payer status do not vary when we control for political attitudes and the hard work-variable [see columns (2) and (3) in Table 7]. In summary, learning that they are richer or poorer than they had previously thought has no effect on individuals’ demand for redistribution – but learning that they are likely to lose from redistribution does decrease their demand. A potential explanation (beyond the scope of our survey) is that, unless primed so, people do not think about redistribution in terms of financial costs and benefits but in general and principled terms of good or fair societies. At this abstract level, the own income position might not matter much when assessing the status quo and expressing a normative view, leading to a weak correlation between income rank and preferences. Once the veil of ignorance is lifted—which happens in the second information treatment—pocketbook concerns set in. The initial information treatment had no effect: those who learn (or are assured) that they are relatively rich obviously did not consider or grasp the personal financial implications of this (new) information. This might be understandable, given that at this stage of the survey the financing of the welfare state had not played any role at all. Once the (personalized) price tag on redistribution becomes salient, rich respondents partly reconsider their policy preferences. The inertia most (rich) respondents show after the first treatment and their preference shift after the second are in line with a ‘cheap talk’-interpretation of preferences towards redistribution: people hold cherished views on how large inequality is and ought to be, and they do not change these ‘expressive’ views until they are given reason to think through its consequences for themselves. 8. Discussion and conclusion All in all, our results show that Germans are poorly informed about their own relative income, that their perception of social stratification is just slightly better, and that they are aware of their ignorance. Still, they have outspoken preferences for more redistribution. A surprising observation from our survey is the stable and strong preference for more redistribution across all incomes.17 This is in contrast to what one would expect, in particular towards the top of the income distribution and in a country that, in international comparison, a large welfare state with a considerable degree of progressivity. Part of the puzzle might be resolved when noticing that the average survey participant from income deciles 8 to 10 locates herself between deciles 5 and 6 and, thus, holds a massively negative self-perception bias. Obviously, these comparatively rich people believe that there is a considerable share of the population with (even) higher incomes—and that there are, thus, substantial resources available that could be redistributed downwards. Hence, additional redistribution looks easily feasible. Moreover, underestimating their relative position, the upper income groups might not think of themselves as massive net contributors to redistributive schemes (some might even hope to benefit financially from redistribution), which makes a greater degree egalitarianism appear costless to themselves.18 Given their preference for more redistribution we expected that it would be mainly those relatively rich respondents who underestimate their actual income position who reacted to the information treatment by expressing lower enthusiasm for redistribution. This did not happen, however. A tentative explanation—apart from the possibility that the information treatment did not help respondents to better understand inequality—would be that high-income earners indeed harbor sincere egalitarian or pro-poor preferences. Inferring from the information treatment how big the gap between rich and poor in Germany actually is or how low incomes in the poorer strata really are, might strengthen their desire for more redistribution, even after taking potential pocketbook concerns into account. Our results trigger the question why we do not observe more redistribution, given that all Germans seem to cherish it. Tentative answers are: first, the preference for more egalitarianism (which also shows up in the platforms of all major political parties in Germany) is mostly cheap talk—if programs are actually proposed, pocketbook concerns override well-intentioned preference statements. Our experiment supports this explanation: already informational clues towards net payer status weaken the preference for redistribution noticeably. Second, the ‘political system’ (government, parliament, lobby groups etc.) holds different preferences on redistribution than the citizenry. As there is no direct voting (on redistribution) in Germany, political processes might produce results that deviate from what a popular vote would dictate. Third, financial feasibility and government budget constraints limit the scope for more redistribution, even if it is wished for by voters. (Mis-)Perceptions of reality in the citizenry matter in democracies: normative views on the desirability (or lack thereof) of policy changes—more or less redistribution, say—are shaped, among others, by perceptions of the status quo. Distorted perceptions might lead to biased political choices. The links between citizens’ views and preferences and actual redistribution policies certainly deserve greater attention. Supplementary material Supplementary material is available at Socio-Economic Review Journal online. Footnotes 1 Popular misperceptions also prevail with other issues (inflation, corruption, risks etc.). See Stevenson and Duch (2013) for a discussion. A potential common root is that individuals make inferences about objective reality from the limited sample of their own experiences and observations. For example, their reference group—relatives, friends, neighbors and colleagues—is typically not a cross-section of society but less heterogeneous. This biased and limited availability of social comparison leads to biased inferences (e.g. Runciman, 1966; Evans and Kelley, 2004). 2 Norstat is a market research company (http://opinion-people.com/de). Participants in Norstat panels can collect points that can be exchanged for money. 3 In 2015—the survey year—a minimum wage was implemented in Germany. Therefore, we restrict our analysis to the previous period to avoid (unknown) biases resulting from this reform. 4 The first and the second treatment are separate events, and (potential) changes in preferences were separately surveyed. 5 Potential respondents did not know that they were going to be asked for their views on inequality or redistribution. Selection into the survey, thus, did not a priori favor people with strong views or interest in these issues. Respondents spent on average 5 min on the survey, with a median time of 4 min, 30 sec. Given the moderate length of the survey, these answering times appear appropriate. Unfortunately, we have no information about attrition rates. For people who aborted the survey, no data were saved. 6 We also checked whether or not results are robust against re-weighting our observations according to income deciles by six different age groups. Mean and median income increase noticeably, nevertheless, our results are robust against using sample weights. Main results of regressions using sample weights can be seen in Table OA.1 in the Online Appendix. 7 We asked this question at the end of the survey, after having uncovered the actual type of society. 8 Precise correlations between controls and bias groups are available on request. 9 We report results for perceived income decile because perceptions might matter more for political preferences than the (unknown) actual position. Results for mean preferences in actual income deciles are reported in Table A3. They are qualitatively the same as for perceived deciles. 10 Columns (2) and (3) in Table A4 and Figure A3 show that this phenomenon is not confined to Germany: the mean preferences for redistribution by perceived income decile for Sweden and Argentina—the two countries for which comparable studies exist, i.e. Cruces et al. (2013) and Karadja et al. (2014)—also lie in a relatively small range. 11 See columns (2) and (3) in Table A3 for mean preferences by income decile. 12 From a technical point of view, the missing treatment effect is no surprise. Mean preferences for redistribution are uniformly distributed over income deciles. Thus, if we inform respondents about their self-positioning bias, the decile changes do not imply different preferences on average (see Table A3). 13 Whether a respondent was initially wrong (and might, thus, have felt uncomfortable with the initial view or its correction) or right (and might, thus, feel encouraged to reinforce the initial view) does not play any role. Neither does the specific way in which biases are measured. See Table OA.2 in the Online Appendix. 14 Verify from column (1) in Table A5 that responses in our sample are in line with those in the ISSP 2009. The same holds for the ‘ought’-question; see column (2) in Table A5. 15 As in Section 5, results are presented by bias group because of potentially heterogeneous treatment effects. Furthermore, we ran estimates for each ‘before treatment’-type and for the full sample. Diff-in-Diff is only applied in the latter case, because both methods—first difference and diff-in-diff—obviously coincide in the former ones. 16 We used the 70th percentile rather than the more precise 75th percentile as the threshold because we framed the entire survey in deciles (e.g. respondents were informed that they belonged to the seventh or eighth decile). Our results also hold if we set the threshold at the eighth decile. 17 By contrast, Cruces et al. (2013, Figure 4) find a u-shaped pattern in the preference for more redistribution over the Argentinian income distribution. 18 Observe from Table 3 that some respondents who believe to belong to the top deciles of the income distribution indeed do ask for more redistribution (38 respondents locating themselves in the eighth decile, 4 in the ninth and 8 in the top). Still there is no bias resulting from a potentially above-average leftist ideology in this sub-sample. Acknowledgement We thank Maximilian Stockhausen for helpful suggestions. References Bartels L. M. ( 2005 ) ‘ Homer Gets a Tax Cut: Inequality and Public Policy in the American Mind’, Perspectives on Politics , 3 , 15 – 31 . Google Scholar CrossRef Search ADS Bartels L. M. ( 2008 ) Unequal Democracy: The Political Economy of the New Gilded Age , Princeton, NJ , Princeton University Press . Clark A. E. , D’Ambrosio C. ( 2015 ) ‘Attitudes to Income Inequality: Experimental and Survey Evidence’. In Atkinson A. B. , Bourguignon F. (eds) Handbook of Income Distribution , Vol. 2 , Amsterdam , North-Holland , pp. 1147 – 1208 . Corneo G. , Grüner H.-P. ( 2002 ) ‘ Individual Preferences for Political Redistribution’, Journal of Public Economics , 83 , 83 – 107 . Google Scholar CrossRef Search ADS Cruces G. , Perez-Truglia R. , Tetaz M. ( 2013 ) ‘ Biased Perception of Income Distribution and Preferences for Redistribution: Evidence from a Survey Experiment’, Journal of Public Economics , 98 , 100 – 112 . Google Scholar CrossRef Search ADS Duflo E. , Saez E. ( 2003 ) ‘ The Role of Information and Social Interactions in Retirement Plan Decisions: Evidence from a Randomized Experiment’, The Quarterly Journal of Economics , 118 , 815 – 842 . Google Scholar CrossRef Search ADS Engelhardt C. , Wagener A. ( 2014 ) Biased Perceptions of Income Inequality and Redistribution, CESifo Working Paper No. 4838, Munich, CESifo. Evans M. , Kelley J. ( 2004 ) ‘ Subjective Social Location: Data from 21 Nations’, International Journal of Public Opinion Research , 16 , 3 – 38 . Google Scholar CrossRef Search ADS Gimpelson V. , Treisman D. ( 2015 ) Misperceiving Inequality, NBER Working Paper No. 21174, Cambridge, MA, National Bureau of Economic Research. Hornik R. ( 1989 ) ‘The Knowledge-Behavior Gap in Public Information Campaigns: A Development Communication View’. In Salmon C. T. (ed.) Information Campaigns: Balancing Social Values and Social Change , London , Sage , pp. 113 – 138 . Jensen R. ( 2010 ) ‘ The (Perceived) Returns to Education and the Demand for Schooling’, Quarterly Journal of Economics , 125 , 515 – 548 . Google Scholar CrossRef Search ADS Karadja M. , Moellerstroem J. , Seim D. ( 2014 ) Richer (And Holier) Than Thou? The Effect of Relative Income Improvements on Demand For Redistribution, IFN Working Paper No. 1042, Stockholm, Research Institute of Industrial Economics (IFN). Kenworthy L. , McCall L. ( 2008 ) ‘ Inequality, Public Opinion and Redistribution’ , Socio-Economic Review , 6 , 35 – 68 . Google Scholar CrossRef Search ADS Meltzer A. H. , Richard S. F. ( 1981 ) ‘ A Rational Theory of the Size of Government’, The Journal of Political Economy , 89 , 914 – 927 . Google Scholar CrossRef Search ADS Niehues J. ( 2014 ) Subjective Perceptions Of Inequality And Redistributive Preferences: An International Comparison. Discussion Paper, Cologne, Cologne Institute for Economic Research (IW). Norton M. I. , Ariely D. ( 2011 ) ‘ Building a Better America—One Wealth Quintile at a Time’, Perspectives on Psychological Science , 6 , 9 – 12 . Google Scholar CrossRef Search ADS PubMed Osberg L. , Smeeding T. ( 2006 ) ‘ “Fair” Inequality? Attitudes toward Pay Differentials: The United States in Comparative Perspective’, American Sociological Review , 71 , 450 – 473 . Google Scholar CrossRef Search ADS Perez-Truglia R. ( 2016 ) ‘The Effects of Income Transparency on Well-Being: Evidence from a Natural Experiment’, accessed at SSRN http://ssrn.com/abstract=2657808 or http://dx.doi.org/10.2139/ssrn.2657808. Runciman W. G. ( 1966 ) Relative Deprivation and Social Justice: A Study of Attitudes to Social Inequality in Twentieth-Century England , London , Routledge Kegan Paul . Stevenson R. T. , Duch R. ( 2013 ) ‘ The Meaning and use of Subjective Perceptions in Studies of Economic Voting’, Electoral Studies , 32 , 305 – 320 . Google Scholar CrossRef Search ADS Tichenor P. J. , Donohue G. A. , Olien C. N. ( 1970 ) ‘ Mass Media Flow and Differential Growth in Knowledge’, Public Opinion Quarterly , 34 , 159 – 170 . Google Scholar CrossRef Search ADS Appendix Table A1. Comparison of survey respondents and population (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 Source: Own survey and micro census 2011. View Large Table A1. Comparison of survey respondents and population (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 (1) (2) Survey Census (2011) Mean SE Mean Age (18–70), years 45.2 (14.541) 44.2 Women 0.505 (0.500) 0.512 Household net income (monthly) 2405 (2319) 2988 Primary education 0.002 (0.048) 0.047 Lower secondary education 0.112 (0.315) 0.356 Secondary education 0.359 (0.480) 0.269 Higher secondary education 0.527 (0.500) 0.283 Retired 0.212 (0.409) 0.237 Source: Own survey and micro census 2011. View Large Table A2. Variable definitions and descriptive statistics Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Notes: n = 859 for all variables. View Large Table A2. Variable definitions and descriptive statistics Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Variable Definition Mean SD Equiv. household net income (Monthly) Net income divided by equivalence weight based on the modified OECD scale 1461.25 745.88 Objective income decile Respondent’s relative income rank corresponding to the deciles of the GSOEP v29 (equivalence weighted) 4.960 2.886 Perceived own income decile Respondent’s stated perceived own decile 4.184 1.998 Survey question: What is the share of households in Germany that have a lower standard of living than yours? Answer categories were given in deciles Bias Perceived own income decile minus objective income decile −0.774 2.781 Preference for redistribution Respondent’s stated attitude towards actual income inequality (scale ranges from 1 to 7) (1) there is too much effort to equalize incomes, (4) satisfied with status quo, (7) there should be much more effort to equalize incomes. 5.651 1.341 Type today Survey question: which type (see Figure 2) best describes German society today?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 2.409 1.069 Preferred type Survey question: which type (see Figure 2) the German society ought to be like?: (1) Type A, (2) Type B, (3) Type C, (4) Type D, (5) Type E 3.932 0.814 Pay Indicator variable = 1 if respondents objective income decile is 7 or higher 0.341 0.474 Informed Sum of respondents stated (news) media consumption 12.744 2.287 We asked for news in TV, Internet and newspaper, response categories: (1) never, (2) seldom, (3) monthly, (4) weekly, (5) daily Confidence How confident respondent feels with her answer on the own income decile: (1) not sure, (2) somewhat sure, (3) sure, (4) very sure 1.790 0.757 Confident Indicator variable = 1 if confidence is (2), (3) or (4) 0.618 0.486 Reference group Indicator variable = 1 if the respondent states that she hast friends from all social classes Ideology Respondent’s stated political leaning on a range from (1) left to (10) right 5.013 1.875 Left Indicator variable = 1 if ideology is equal or lower (4) 0.484 0.500 Hard work Indicator variable = 1 if respondent states that hard work is important to get ahead in life 0.675 0.469 Survey question: How important is hard work to get ahead in life?: (1) essential, (2) very important, (3) important, (4) somewhat important, (5) not important. Age Age in years 45.213 14.541 Women Indicator variable = 1 if respondent is female 0.505 0.500 Education Respondent’s highest degree: (1) primary education, (2) lower secondary education, (3) secondary education, (4) higher secondary education (Fachhochschulreife), (5) higher secondary education (Abitur) 3.794 1.081 Notes: n = 859 for all variables. View Large Table A3. (Mean) Preferences for redistribution by income decile (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. View Large Table A3. (Mean) Preferences for redistribution by income decile (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) (1) (2) (3) (4) Initial First treatment First treatment Second treatment Preferences (treated) (control) (of the treated) Decile Mean SE Mean SE Mean SE Mean SE 1 5.801 (1.410) 5.920 (1.217) 5.831 (1.485) 5.813 (1.302) 2 6.011 (1.260) 5.867 (1.198) 6.022 (1.252) 5.756 (1.417) 3 5.670 (1.367) 5.551 (1.355) 5.976 (1.129) 5.449 (1.542) 4 5.904 (1.238) 5.750 (1.368) 6.024 (0.987) 5.656 (1.405) 5 5.663 (1.184) 5.556 (1.120) 6.025 (0.974) 5.600 (1.031) 6 5.500 (1.303) 5.780 (1.250) 5.529 (1.107) 5.700 (1.233) 7 5.573 (1.248) 5.762 (1.246) 5.718 (1.169) 5.476 (1.348) 8 5.453 (1.621) 5.523 (1.577) 5.595 (1.547) 5.136 (1.564) 9 5.520 (1.107) 5.571 (0.966) 5.606 (1.059) 5.143 (1.458) 10 5.060 (1.420) 5.292 (1.083) 5.230 (1.306) 5.167 (1.167) Notes: Preferences for redistribution are coded from 1 to 7. For more details see Table A2. View Large Table A4. (Mean) Preferences for redistribution: ISSP 2009 (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) Notes: Preferences for redistribution are coded from 1 to 5. Survey question: ‘Differences in income in R’s country are too large.’ (5) strongly agree (4) agree (3) neither agree nor disagree (2) disagree (1) strongly disagree. See Engelhardt and Wagener (2014) for the perceived income decile in the ISSP. Source: ISSP 2009 Social Inequality IV. View Large Table A4. (Mean) Preferences for redistribution: ISSP 2009 (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) (1) (2) (3) Germany Argentina Sweden Perceived decile Mean SE Mean SE Mean SE 1 4.750 (0.866) 4.310 (0.541) 4.636 (0.505) 2 4.667 (0.620) 4.167 (0.794) 4.385 (0.768) 3 4.713 (0.580) 4.339 (0.712) 4.236 (0.860) 4 4.600 (0.670) 4.274 (0.676) 4.395 (0.786) 5 4.460 (0.720) 4.236 (0.738) 4.157 (0.837) 6 4.351 (0.810) 4.168 (0.833) 3.987 (0.868) 7 4.184 (0.876) 4.286 (0.749) 3.600 (1.068) 8 4.071 (0.956) 4.436 (0.640) 3.518 (1.210) 9 4.071 (1.141) 4.222 (0.833) 3.455 (1.368) 10 4.500 (1.000) 5.000 (0.000) 3.571 (1.505) Notes: Preferences for redistribution are coded from 1 to 5. Survey question: ‘Differences in income in R’s country are too large.’ (5) strongly agree (4) agree (3) neither agree nor disagree (2) disagree (1) strongly disagree. See Engelhardt and Wagener (2014) for the perceived income decile in the ISSP. Source: ISSP 2009 Social Inequality IV. View Large Table A5. Answers ISSP 2009 (for Germany) (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 Source: ISSP 2009 Social Inequality IV. View Large Table A5. Answers ISSP 2009 (for Germany) (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 (1) (2) Type today Type preferred Type A 18.80% 1.49% Type B 35.38% 10.36% Type C 23.03% 18.21% Type D 18.57% 57.06% Type E 4.22% 12.87% N 1255 1274 Source: ISSP 2009 Social Inequality IV. View Large Table A6. Perceived types today: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type today. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A6. Perceived types today: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group (observations (obs.)] 1.264 (53) 1.111 (18) 1.394 (33)  Control group (obs.) 1.200 (40) 1.467 (15) 1.222 (27)  Difference (s.e.) 0.064 (0.143) −0.356 (0.277) 0.172 (0.227) Type B  Treatment group (obs.) 2.107 (75) 2.154 (26) 2.220 (50)  Control group (obs.) 2.025 (79) 1.957 (23) 2.149 (47) Type C  Treatment group (obs.) 2.868 (76) 2.867 (15) 2.939 (33)  Control group (obs.) 3.000 (76) 2.833 (24) 2.857 (21)  Difference (s.e.) −0.132 (0.870) 0.033 (0.180) 0.082 (0.166) Type D  Treatment group (obs.) 3.541 (37) 3.625 (8) 3.889 (9)  Control group (obs.) 3.905 (21) 3.750 (4) 3.769 (13)  Difference (s.e.) −0.364* (0.190) −0.125 (0.314) 0.120 (0.352) Type E  Treatment group (obs.) 4.714 (7) 3.500 (2) 3.000 (6)  Control group (obs.) 4.500 (8) 5.000 (2) 4.182 (11)  Difference (s.e.) 0.214 (0.564) −1.500 (1.500) −1.182 (0.846) Full sample  Treatment group (obs.) 2.448 (248) 2.246 (69) 2.344 (131)  Control group (obs.) 2.473 (224) 2.353 (68) 2.429 (119)  Difference (s.e.) −0.026 (0.100) −0.107 (0.179) −0.085 (0.154)  Diff-in-Diff (s.e.) −0.046 (0.138) −0.044 (0.251) 0.086 (0.211) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type today. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A7. Preferred society types: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type preferred. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Table A7. Preferred society types: experimental results (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) (1) (2) (3) Negative bias No bias Positive bias Type A  Treatment group [observations(obs.)] 1.000 (2) – (–) 1.625 (8)  Control group (obs.) 1 (2) 1.000 (1) 2.333 (3)  Difference [standard error (s.e.)] – (–) – (–) – (–) Type B  Treatment group (obs.) 2.500 (10) 3.000 (2) 2.462 (13)  Control group (obs.) 2.000 (11) 2.000 (2) 2.000 (5)  Difference (s.e.) 0.500* (0.256) – (–) 0.462 (0.400) Type C  Treatment group (obs.) 3.148 (27) 3.375 (8) 3.000 (13)  Control group (obs.) 2.947 (19) 3.143 (7) 3.143 (14)  Difference (s.e.) 0.201 (0.145) 0.232 (0.312) −0.143 (0.101) Type D  Treatment group (obs.) 3.927 (164) 3.957 (46) 3.946 (74)  Control group (obs.) 3.918 (147) 4.024 (41) 3.956 (68)  Difference (s.e.) 0.008 (0.043) −0.068 (0.090) −0.010 (0.058) Type E  Treatment group (obs.) 4.867 (45) 4.769 (13) 4.826 (23)  Control group (obs.) 4.707 (41) 4.750 (16) 4.750 (28)  Difference (s.e.) 0.159 (0.128) 0.019 (0.242) 0.076 (0.182) Full sample  Treatment group (obs.) 3.931 (248) 4.014 (69) 3.718 (131)  Control group (obs.) 3.860 (220) 4.000 (67) 3.924 (118)  Difference (s.e.) 0.072 (0.072) 0.014 (0.130) −0.206* (0.118)  Diff-in-Diff (s.e.) 0.069 (0.099) 0.029 (0.178) 0.057 (0.171) Notes: Robust standard errors in parentheses: *** P < 0.01, **P < 0.05, *P < 0.1. Dependent variable: type preferred. Columns show average preferences for redistribution by bias group. Respondents with negative (positive) bias underestimate (overestimate) their relative income rank. View Large Figure A1. View largeDownload slide Distribution of variable bias in our sample. See Table A1 for definition. Figure A1. View largeDownload slide Distribution of variable bias in our sample. See Table A1 for definition. Figure A2. View largeDownload slide Germans mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey Preferences are normalized to (0,1). Figure A2. View largeDownload slide Germans mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey Preferences are normalized to (0,1). Figure A3. View largeDownload slide Mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey. Preferences are normalized to (0,1). Figure A3. View largeDownload slide Mean preferences for redistribution by perceived income decile. Comparison of ISSP and authors’ survey. Preferences are normalized to (0,1). Figure A4. View largeDownload slide Mean change in preferences for redistribution by initial self-positioning bias. Figure A4. View largeDownload slide Mean change in preferences for redistribution by initial self-positioning bias. © The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

Socio-Economic ReviewOxford University Press

Published: Oct 20, 2017

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off