Following Alesina and Fuchs-Schündeln (2007) , several studies have looked at political and social preferences in the context of German reunification. Svallfors (2010) uses the International Social Survey Program data for 1990, 1996, and 2006. He finds that attitudes towards state intervention of East Germans converge quickly to those of West Germans, and that there is no East-West difference in attitudes among the younger cohorts. Brosig-Koch et al. (2011) , on the other hand, see no convergence in attitudes towards solidarity, cooperation, and fairness between East and West Germans, relying on laboratory experiments among students. Finally, Avdeenko (2018) shows that East Germans living in the border regions are less likely to vote for left parties, which the author interprets as retrospective electoral punishment.
In this paper, we revisit the long-term effects of having lived under Communism versus Capitalism on preferences. Adding the 2017 release of the German Socio-Economic Panel, we update the results obtained by Alesina and Fuchs-Schündeln (2007) . We show that East Germans continue to be significantly more pro-state compared to West Germans even 27 years post reunification. Moreover, we find that the speed of convergence slowed down significantly. The convergence of preferences between 2002 and 2017 is mainly driven by the change in the composition of the population, with younger cohorts replacing older cohorts: On the individual level, there was on average no further significant convergence of preferences between 2002 and 2017. Finally, we present evidence on the intergenerational transmission of preferences: we find that young individuals born after reunification to parents who used to live in the East are significantly more likely to exhibit pro-state preferences than young individuals born to parents who used to live in the West, independent of their own current residence. Linking children to their parents, we document a strong correlation between the preferences of parents and children. This intergenerational transmission of preferences implies that the effects of political regimes on preferences can be extremely persistent.
Experiences shape preferences. This has meanwhile been shown for a wide range of preferences, from risk and inflation tolerance ( Malmendier and Nagel 2011 ; Malmendier 2021 ) to support of democracy ( Fuchs-Schündeln and Schündeln 2015 ). Alesina and Fuchs-Schündeln (2007) analyze the effect of having lived under Communism versus Capitalism on preferences for a strong welfare state. Based on the 1997 and 2002 releases of the German Socio-Economic Panel, Alesina and Fuchs-Schündeln (2007) find East Germans to be significantly more in favor of the government taking a strong role in caring for vulnerable groups than West Germans. At the same time, they document convergence of preferences between 1997 and 2002 and conclude that it would take 20–40 years for the views of East and West Germans to converge if convergence were to continue at the same (linear) speed. By documenting convergence of preferences on the individual level, and stronger East-West differences for older individuals who lived under the different systems for a longer time, they can establish a causal effect of living under different systems on preferences.
In addition to the main explanatory variable East – which refers to the residence before reunification, independent of the current residence – we include two year dummies, as well as interaction terms between East and the year dummies (East*2002 and East*2017) to evaluate convergence of preferences between East and West Germans. The vector of control variables X includes age, its square and cube, gender, number of children and number of adults in the household, dummy variables for education (with fewer than nine years of schooling as the omitted category), dummy variables for marital status (with single as the omitted category), dummy variables for labor force status (with employed as the omitted category), dummy variables for the type of work (with blue-collar as the omitted category), and the log of household income. Explanatory variables are summarized in Appendix Tables A3 for the unbalanced sample and A4 for the balanced sample.
Figure 1 shows the share of the respondents from West and East who state that the government should be responsible for providing financial security for the five different vulnerable groups. The majority of the respondents agree that the state should provide financial support to the unemployed – in 2017, 74% of East Germans and 69% of West Germans had pro-state preferences with respect to this question. At the same time, only 48% of the respondents from the East and 35% of the respondents from the West think that it is the government’s duty to support the family, and the averages are in a similar range for the other outcome variables. Over time, the West German sample becomes more pro-state: the fraction of individuals who consider that the state is responsible for financial security increases between 1997 and 2017 across all five variables. The East German sample, on the other hand, becomes on average less pro-state between 1997 and 2002. However, this pattern reverses between 2002 and 2017, when the opinion moves more pro-state in both samples, with the exception of the opinion on the state taking care of unemployed individuals in the East sample.
Our basic sample is constructed as follows. For the main analyses, we use data from both individual and household questionnaires for 1997, 2002, and 2017. We only include individuals born before 1989. We use the initial West and East German samples (A and C), as well as all general refreshment samples (samples E, F, H, J, and K). Appendix Tables A1 and A2 provide summary statistics of the preferences separately for the East and West samples and years, once on the full sample and once on the balanced sample that includes only individuals who answer in all three survey waves. On average, East Germans in the unbalanced sample spent roughly 30 years under Communism. For the analysis of the youngest cohorts in Section 4 , we focus on individuals born between 1990 and 1999, and link them to their parents based on the biography questionnaire.
As in Alesina and Fuchs-Schündeln (2007) , we use the German Socio-Economic Panel (SOEP) as our primary data source. SOEP is a large representative panel of German households. The survey started in 1984, first sampling only West German households, but adding an East German sample already in the spring of 1990. The respondents of the survey answer a wide array of questions, providing information about their socio-economic background, family ties, labor market experiences, attitudes, and political views. The question concerning preferences about the role of the government in providing financial security was asked only in the 1997, 2002, and 2017 waves. This question reads: “In addition to the state, private individuals such as free-market companies, organizations, associations or individual citizens are responsible for a large number of social tasks in our society today. What is your opinion on who should be responsible for the following areas: financial protection of families/financial security for unemployment/financial protection in case of illness/financial security in old age/financial security for people in need of care”. The answers are given on a scale of 1–5, which correspond to “only the state”, “mostly the state”, “both the state and private forces”, “especially private forces”, and “only private forces”. We use the answers to this question to generate five independent dummy variables. Specifically, the dummy variables take the value of 1 when the respondent states that financial security is the responsibility of “only the state” or “mostly the state”, and the value of 0 otherwise. Thus, our dependent variables indicate whether an individual has pro-state preferences. Our main explanatory variable is a dummy variable East, which is equal to 1 if the respondent lived in East Germany before reunification, independent of the current residence.
3 Main Results
Table 1 replicates the basic regression specification from Alesina and Fuchs-Schündeln (2007), but extends the sample to include data from 2017, in addition to the years 1997 and 2002. The dependent variable is a dummy equal to one if the individual supports an active role of the state. The right-hand side variables of interest are the East dummy and the interaction terms between East and the years 2002 and 2017.
Table 1: Baseline regressions. Dependent variable: responsibility for financial security… …when …when …of the …when …when requiring unemployed sick family old care East 0.438*** 0.465*** 0.420*** 0.436*** 0.377*** (0.030) (0.028) (0.028) (0.028) (0.028) Year 2002 0.043** 0.161*** −0.004 −0.023 0.113*** (0.020) (0.020) (0.021) (0.020) (0.020) East*2002 −0.137*** −0.170*** −0.097*** −0.157*** −0.161*** (0.035) (0.033) (0.033) (0.032) (0.033) Year 2017 0.202*** 0.275*** 0.130*** 0.153*** 0.267*** (0.024) (0.024) (0.024) (0.024) (0.023) East*2017 −0.333*** −0.207*** −0.119*** −0.182*** −0.162*** (0.040) (0.038) (0.038) (0.038) (0.038) Age −0.020* 0.023** −0.009 0.009 0.009 (0.011) (0.011) (0.011) (0.011) (0.011) Age squared (*103) 0.495** −0.373* 0.216 −0.149 −0.214 (0.214) (0.207) (0.211) (0.207) (0.205) Age cubed (*105) −0.337*** 0.190 −0.158 0.078 0.120 (0.129) (0.125) (0.128) (0.125) (0.124) College −0.193*** −0.297*** −0.221*** −0.316*** −0.126** (0.064) (0.059) (0.059) (0.059) (0.058) Vocational training −0.157** −0.221*** −0.249*** −0.226*** −0.105* (0.061) (0.056) (0.057) (0.057) (0.056) Secondary schooling −0.160** −0.147** −0.151** −0.132** −0.051 (0.063) (0.058) (0.059) (0.058) (0.058) Intermediate schooling −0.181*** −0.191*** −0.268*** −0.178*** −0.118* (0.069) (0.065) (0.066) (0.065) (0.064) Male −0.039** −0.046*** −0.015 0.002 0.027* (0.015) (0.015) (0.015) (0.015) (0.015) Number of children 0.046*** 0.038*** 0.072*** 0.029*** 0.017* (0.011) (0.010) (0.010) (0.010) (0.010) Number of adults 0.023*** 0.028*** 0.021*** 0.020*** 0.009** (0.004) (0.004) (0.004) (0.004) (0.004) Married 0.026 0.050** 0.025 0.043* 0.045* (0.026) (0.025) (0.025) (0.025) (0.025) Divorced 0.007 −0.032 0.023 −0.000 0.003 (0.035) (0.034) (0.034) (0.034) (0.033) Married but separated −0.022 −0.010 0.001 0.020 0.050 (0.053) (0.051) (0.052) (0.052) (0.051) Widowed 0.011 0.013 0.010 0.019 0.033 (0.039) (0.038) (0.038) (0.037) (0.037) Log HH income −0.149*** −0.209*** −0.169*** −0.239*** −0.168*** (0.017) (0.016) (0.016) (0.016) (0.016) Civil servant −0.189*** −0.222*** 0.049 −0.116*** −0.168*** (0.040) (0.041) (0.041) (0.041) (0.039) Self-employed −0.324*** −0.328*** −0.266*** −0.394*** −0.241*** (0.035) (0.036) (0.037) (0.036) (0.035) White-collar worker −0.046* −0.048** −0.025 −0.101*** −0.076*** (0.023) (0.022) (0.023) (0.022) (0.022) Unemployed 0.122*** 0.002 0.117*** 0.000 −0.015 (0.039) (0.036) (0.036) (0.036) (0.036) Retired −0.125*** −0.113*** 0.005 −0.069* −0.031 (0.038) (0.036) (0.037) (0.036) (0.036) Maternity 0.002 −0.072 0.016 −0.150** −0.069 (0.061) (0.058) (0.058) (0.058) (0.058) Nonworking −0.038 −0.035 0.060** −0.030 −0.019 (0.030) (0.029) (0.029) (0.029) (0.029) Training −0.093 −0.066 −0.203*** −0.071 −0.065 (0.059) (0.057) (0.058) (0.057) (0.056) Other nonworking −0.014 −0.075** −0.003 −0.050 −0.095** (0.040) (0.038) (0.038) (0.038) (0.038) Observations 36,564 36,599 36,532 36,636 36,637 Individuals 24,819 24,840 24,801 24,847 24,848 Log likelihood −22,180 −24,421 −23,638 −24,330 −24,961
As Table 1 shows, individuals who lived in East Germany before reunification have consistently and significantly stronger pro-state preferences than individuals from West Germany in 1997, as expressed by the positive and significant coefficients on the East dummy. Having lived under different systems for up to 45 years shaped the preferences of East and West Germans. Moreover, the preferences of East and West Germans converge significantly between 1997 and 2002 – the coefficients of the interaction terms East*2002 are all significantly negative. This is in line with a common effect on preferences from living under the same system since 1990. The results so far are also quantitatively very similar to the results in Alesina and Fuchs-Schündeln (2007). Extending the time frame to include the 2017 survey has thus no effect on the results from the previous years.
What is interesting, though, is how preferences evolved between 2002 and 2017. First focusing on West Germans, there was already some movement toward stronger state preferences between 1997 and 2002 – the Year 2002 dummy is positive and significant in 3 out of 5 cases – but this movement became even stronger between 2002 and 2017: the Year 2017 dummy is positive and significant for all five variables. It is also significantly different from the Year 2002 dummy (Appendix Table A5). Thus, West German preferences moved more in favor of a strong government in the two decades between 1997 and 2017. Turning to the coefficients on the interaction terms with the East dummy, the preference differences between East and West Germans continued to narrow: the coefficients of East*2017 are not only significantly negative but also larger in absolute magnitude than the East*2002 coefficients. Yet, only in the case of taking care of unemployed individuals is the East*2017 coefficient significantly different from the East*2002 coefficient. Thus, there is significant convergence between 2002 and 2017 only in one of the five preference variables. In the other four cases, the convergence is not only not significant in this period, but according to the point estimates also substantially slower than in the period 1997 to 2002. This is true even for the only variable in which significant convergence still occurs between 2002 and 2017: taking into account that this is a 15-year period, compared to the 5-year period between 1997 and 2002, the speed of convergence slowed down significantly. The coefficients on the control variables are very close to the estimates in Alesina and Fuchs-Schündeln (2007) both in sign and magnitude. Individuals with more education and higher monthly household income, as well as the self-employed and white-collar workers, are consistently less pro-state.
The marginal effects corresponding to Table 1 are shown in Appendix Table A7. They allow us to talk about the speed of convergence more concretely. In 1997, individuals from East Germany were between 14.8 and 17.8 percentage points more likely to have pro-government preferences than individuals from West Germany. Compared to 1997, the East-West difference in the probability of being pro-state declined by between 3.6 and 6.5 percentage points in 2002, and by between 4.4 and 11.5 percentage points in 2017. Clearly, the convergence of preferences between East and West Germans is not complete – even 27 years after reunification the preferences regarding the active role of state differ between East and West Germans. If the speed of convergence continues as it did between 2002 and 2017, then preferences for taking care of the unemployed should converge by 2025. However, for the other three variables which still show convergence between 2002 and 2017, final convergence of preferences would take significantly longer. In the case of taking care of the sick, for instance, it will take a further 98 years after 2017 for preferences to converge; both because the difference is still quite large in 2017 (East Germans are 9.8 percentage points more likely to answer that the state should take care of sick individuals), and because convergence is slow (the East-West difference declined by only 1.5 percentage points over the 15 year period 2002 to 2017). For the other two variables, the much lower speed of convergence between 2002 and 2017 implies that full convergence of preferences will take between 144 (taking care of old individuals) and 208 years (taking care of individuals requiring care). Thus, not only is the convergence between 2002 and 2017 not statistically significant, neither is it economically significant.
Individual preferences for a strong government could be affected not only by the individual economic situation, but also by general economic circumstances, either because individuals are altruistic and care for people around them, or because they could potentially be affected by these general economic circumstances themselves in the future. Thus, we include as an additional control variable the state-level unemployment rate in the survey year. As Table 2 shows, indeed individuals who reside in states with higher unemployment rates tend to exhibit stronger preferences for an active role of the state: the coefficients on the state-level unemployment rate are positive and statistically significant across all five dependent variables. The inclusion of this control variable decreases the magnitude of the East coefficients, which suggests that part of the East effect can be explained by higher unemployment rates in the Eastern states. Nevertheless, the East dummy is still a significant predictor of preferences. The convergence results are also quantitatively affected by the inclusion of this control variable: the inclusion of the state-level unemployment rate leads to stronger convergence between 1997 and 2002, and weaker convergence or even divergence between 2002 and 2017.
Table 2: Regressions controlling for state-level unemployment rate. Dependent variable: responsibility for financial security… …when …when …of the …when …when requiring unemployed sick family old care East 0.292*** 0.359*** 0.311*** 0.330*** 0.258*** (0.039) (0.036) (0.036) (0.036) (0.036) Year 2002 0.079*** 0.187*** 0.023 0.003 0.142*** (0.021) (0.021) (0.021) (0.021) (0.021) East*2002 −0.161*** −0.187*** −0.115*** −0.175*** −0.181*** (0.036) (0.033) (0.033) (0.033) (0.033) Year 2017 0.291*** 0.339*** 0.195*** 0.217*** 0.339*** (0.028) (0.027) (0.028) (0.027) (0.027) East*2017 −0.217*** −0.122*** −0.032 −0.098** −0.067 (0.045) (0.042) (0.042) (0.042) (0.042) Regional unemployment rate 0.019*** 0.014*** 0.014*** 0.014*** 0.016*** (0.003) (0.003) (0.003) (0.003) (0.003) All control variables Yes Yes Yes Yes Yes Observations 36,563 36,598 36,531 36,635 36,636 Individuals 24,818 24,839 24,800 24,846 24,847 Log likelihood −22,160 −24,409 −23,625 −24,318 −24,946
In Table 3, we split the Eastern sample into East German states by current residence, as in Table 8 of Alesina and Fuchs-Schündeln (2007). We drop all individuals who are originally from the East and now reside in a West state and vice versa. The table shows that the patterns we found before are quite homogeneous and are thus observed across individual states as well. First, in 1997 individuals from all Eastern states are significantly more pro-state than their West German counterparts. Secondly, all the significant coefficients of the interaction terms measuring convergence are negative (except for one coefficient for Brandenburg in 2002 and one for Mecklenburg-Vorpommern in 2017). For 2002, 15 of the 25 convergence coefficients are negative and significant, and the same is true for 2017. Convergence is stronger in Sachsen-Anhalt, Thüringen, and Sachsen than in the other three states. Generally, this analysis confirms that the baseline results are not driven by a single state.
Table 3: Splitting East into states, unbalanced sample. Dependent variable: responsibility for financial security… …when …when …of the …when …when requiring unemployed sick family old care Mecklenburg-Vorpommern 0.300*** 0.337*** 0.229*** 0.305*** 0.401*** (0.076) (0.070) (0.070) (0.070) (0.070) Brandenburg 0.407*** 0.380*** 0.416*** 0.462*** 0.404*** (0.063) (0.058) (0.057) (0.058) (0.058) Sachsen-Anhalt 0.502*** 0.585*** 0.464*** 0.581*** 0.462*** (0.062) (0.055) (0.055) (0.055) (0.055) Thüringen 0.611*** 0.451*** 0.441*** 0.444*** 0.372*** (0.064) (0.054) (0.054) (0.055) (0.055) Sachsen 0.478*** 0.536*** 0.477*** 0.456*** 0.376*** (0.049) (0.044) (0.044) (0.044) (0.044) Year 2002 0.046** 0.154*** 0.000 −0.022 0.115*** (0.020) (0.020) (0.021) (0.020) (0.020) Mecklenburg-Vorpommern*2002 0.003 0.008 −0.002 −0.068 −0.172* (0.100) (0.092) (0.089) (0.088) (0.090) Brandenburg*2002 0.145* 0.093 0.078 −0.011 −0.064 (0.080) (0.070) (0.068) (0.071) (0.071) Sachsen-Anhalt*2002 −0.156** −0.276*** −0.146** −0.319*** −0.289*** (0.074) (0.065) (0.066) (0.065) (0.066) Thüringen*2002 −0.544*** −0.181*** −0.217*** −0.141** −0.063 (0.073) (0.066) (0.065) (0.066) (0.064) Sachsen*2002 −0.136** −0.225*** −0.155*** −0.200*** −0.162*** (0.058) (0.053) (0.052) (0.052) (0.053) Year 2017 0.198*** 0.259*** 0.127*** 0.156*** 0.269*** (0.024) (0.024) (0.024) (0.024) (0.024) Mecklenburg-Vorpommern*2017 −0.192* −0.084 0.183* −0.090 −0.073 (0.109) (0.103) (0.101) (0.101) (0.105) Brandenburg*2017 −0.104 −0.013 0.028 −0.121 −0.069 (0.090) (0.080) (0.081) (0.082) (0.082) Sachsen-Anhalt*2017 −0.326*** −0.283*** −0.200** −0.223*** −0.130 (0.091) (0.080) (0.079) (0.079) (0.081) Thüringen*2017 −0.829*** −0.367*** −0.369*** −0.478*** −0.346*** (0.084) (0.078) (0.078) (0.079) (0.079) Sachsen*2017 −0.298*** −0.165*** −0.113* −0.109* −0.161*** (0.069) (0.063) (0.062) (0.062) (0.062) All control variables Yes Yes Yes Yes Yes Observations 33,845 33,887 33,825 33,916 33,915 Individuals 23,048 23,070 23,036 23,076 23,077 Log likelihood −20,545 −22,547 −21,777 −22,452 −23,057
3.1 Heterogeneity by Cohort Table 4 replicates the results presented in Table 3 of Alesina and Fuchs-Schündeln (2007), in which we allow for separate cohort patterns in East and West. As in Alesina and Fuchs-Schündeln (2007), we find that older cohorts from the East are significantly more pro-state compared to younger cohorts, while the opposite is true for West Germans. Thus, the East-West differences in preferences are increasing in the time the individuals lived under separate regimes. This is convincing evidence of the effect of the regime on preferences. Table 4: Regressions with cohorts interacted with East. Dependent variable: responsibility for financial security… …when …when …of the …when …when requiring unemployed sick family old care East 0.295*** 0.284*** 0.231*** 0.187*** 0.214*** (0.053) (0.049) (0.050) (0.050) (0.050) Year 2002 0.036* 0.133*** −0.035 −0.062*** 0.086*** (0.021) (0.021) (0.022) (0.021) (0.021) East*2002 −0.133*** −0.159*** −0.086*** −0.146*** −0.152*** (0.036) (0.033) (0.033) (0.033) (0.033) Year 2017 0.183*** 0.184*** 0.022 0.027 0.170*** (0.035) (0.034) (0.035) (0.034) (0.033) East*2017 −0.304*** −0.169*** −0.076** −0.130*** −0.125*** (0.041) (0.038) (0.038) (0.038) (0.038) Born 1961–1975 −0.042 −0.185*** −0.215*** −0.274*** −0.155*** (0.040) (0.038) (0.039) (0.039) (0.038) Born 1946–1960 −0.123** −0.336*** −0.336*** −0.435*** −0.285*** (0.055) (0.053) (0.054) (0.053) (0.052) Born before 1946 −0.137* −0.358*** −0.420*** −0.513*** −0.367*** (0.074) (0.072) (0.073) (0.072) (0.071) Born 1961–1975*East −0.047 0.050 0.047 0.053 0.030 (0.054) (0.050) (0.051) (0.051) (0.050) Born 1946–1960*East 0.164*** 0.160*** 0.148*** 0.206*** 0.128** (0.054) (0.050) (0.051) (0.051) (0.050) Born before 1946*East 0.339*** 0.347*** 0.377*** 0.505*** 0.341*** (0.054) (0.050) (0.051) (0.051) (0.050) All control variables Yes Yes Yes Yes Yes Observations 36,564 36,599 36,532 36,636 36,637 Individuals 24,819 24,840 24,801 24,847 24,848 Log likelihood −22,126 −24,353 −23,565 −24,196 −24,902
3.2 Results for Balanced Sample Finally, Table 5 again builds on an exercise by Alesina and Fuchs-Schündeln (2007). Here, we restrict the sample to a balanced sample. This allows us to understand whether the shift in state-related preferences is driven by the change in the cohort composition of the sample, or whether it occurs on the individual level. Given that younger cohorts have more similar preferences than older cohorts, some convergence will naturally occur over time as the relative role of younger cohorts increases. This effect is controlled for if we only include individuals whom we observe in all sample years. Since here we include only individuals who are observed over two decades, the sample size drops from around 36,500 observations to only roughly 7000 observations. Table 5: Balanced sample: regressions with individuals who answer in 1997, 2002 and 2017. Dependent variable: responsibility for financial security… …when …when …of the …when …when requiring unemployed sick family old care East 0.412*** 0.401*** 0.402*** 0.280*** 0.300*** (0.060) (0.057) (0.057) (0.057) (0.057) Year 2002 0.070 0.194*** 0.005 −0.010 0.192*** (0.046) (0.045) (0.047) (0.046) (0.045) East*2002 −0.143** −0.139** −0.024 −0.027 −0.157** (0.073) (0.069) (0.070) (0.067) (0.068) Year 2017 0.144** 0.319*** 0.096* 0.159*** 0.340*** (0.057) (0.054) (0.056) (0.056) (0.055) East*2017 −0.157** −0.165** −0.031 0.037 −0.162** (0.079) (0.073) (0.071) (0.072) (0.074) All control variables Yes Yes Yes Yes Yes Observations 6981 6981 6981 6981 6981 Individuals 2327 2327 2327 2327 2327 Log likelihood −4122 −4651 −4534 −4627 −4744 The results based on the balanced sample remain qualitatively unchanged: Individuals from East Germany have significantly stronger preferences for an active role of the state in providing financial security, and the East-West difference in preferences becomes smaller over time. Thus, the convergence of preferences between East and West is not explained by a shift in the cohort composition alone. Similar to the unbalanced sample, the pace of convergence is faster between 1997 and 2002 and is heterogeneous across the dimensions of preferences. In the balanced sample, we still observe convergence of preferences between 2002 and 2017 in all dimensions but taking care of the old, but this convergence is never significant: none of the coefficients on the interactions of East*2002 and East*2017 are significantly different. Looking at the marginal effects across the two samples (Appendix Tables A7 and A8), we notice that the estimated convergence between 2002 and 2017 is in fact quite a bit faster in the unbalanced sample. For instance, the probability of being pro-state when it comes to the role of the state in case of unemployment decreased by 6.8 percentage points in 2017 compared to 2002 in the unbalanced sample, but only by 0.5 percentage points in the balanced sample. This means that the overall convergence between 2002 and 2017 is driven mainly by the change in the composition of the sample. The coefficient of Year 2017 is significant and positive in the balanced sample as well – West Germans became more pro-state compared to both 1997 and 2002. Relative to the West Germans, there is no significant development of East German preferences away from the state. One might be concerned that selective attrition drives the differences in speed of convergence between the balanced and unbalanced samples. In Appendix Table A9, we test whether the probability of staying in the sample between two consecutive sample periods (1997 and 2002 or 2002 and 2017, respectively) is correlated with the preferences in the first of the two respective sample periods. We find that there is no systematic pattern, with five coefficients being positive and five negative, and only three of the ten coefficients – one positive and two negative – being statistically significant. Thus, the differences in convergence results between the two samples do not stem from selective attrition.