Beyond individual characteristics: Generational differences in redistribution preferences

Transcription

1 Beyond individual characteristics: Generational differences in redistribution preferences Anja Neundorf (University of Oxford) Stuart Soroka (McGill University) Abstract The body of work focused on attitudes towards the welfare state has produced a number of important insights into how citizens differ in their preferences on income redistribution. Existing research focuses mainly on individual-level characteristics such as income, education, labor force status or economic risk exposure. This paper seeks to add to this literature by taking into account another set of characteristics, namely, formative economic and political experiences and circumstances from the period during which citizens come of age. It is postulated here that these factors shape the redistribution preferences of whole generations. We conduct a hierarchical age-periodcohort analysis of economic preferences using repeated cross-sectional data from the British Social Attitudes Surveys ( ). Doing so allows us to examine generations coming of age between the end of World War I and today. The analysis thus permits a rather detailed look at the potentially profound and lasting impact of these aspects of political socialization on both political and policy preferences. We are also, more specifically, able to examine whether the documented shift towards more rightist redistribution preferences in the British electorate is (at least partly) explained by generational replacement. Keywords: Britain. Redistribution preferences; cohort analysis; political socialization; Great Paper to be presented at the Annual EPOP Conference, Oxford, 7-10 September We are grateful to David Soskice, Matt Golder, Charlotte Cavaille, Lucy Barnes and Michael Becher for taking the time to comment on an earlier version of this paper. anja.neundorf@nuffield.ox.ac.uk stuart.soroka@mcgill.ca. 1

2 Introduction The post-war era has seen a striking rise in the size of the welfare state. Indeed, social welfare policy, and redistributive policy more broadly, forms a considerable portion of what modern states do. Redistributive policies are accordingly a central focus of party programs, on both the left and the right. They are a regular, central concern for citizens. And redistributive policies have been particularly salient in the wake of the Great Recession, the Occupy movement, and tensions in the Eurozone. This seems like a particularly pertinent time to ask, then, What are the sources of public attitudes about redistribution? There are already considerable bodies of work on the issue, admittedly. But as we shall see below, much of this work focuses on current individual-level interests typically economic interests (or some combination, for instance, of expected future income and anticipated tax returns (Meltzer and Richard 1981)), though sometimes on political interests and institutional environments as well. In almost all cases, however, research focuses on current factors. We wish to shift the focus here to what we believe is another, fundamental source of citizens attitudes towards redistribution. Like a limited body of past work (discussed below), we wish to analyze the effects of past economic and political experiences. Put more precisely, we wish to consider the possibility that citizens economic and political experiences during their formative years have a marked and lasting impact on their core policy preferences (such as preferences for redistributive policy). Existing work already makes clear that a range of political values and attitudes are crystallized early in adulthood (Alwin and Krosnick 1991; Jennings 1989; Sears and Valentino 1997); relatedly, some work suggests that regime type essentially, the regime in which one grows up can affect attitudes toward state intervention (Alesina and Fuchs-Schündeln 2008; Svallfors 2010). But relatively few have considered directly the possibility that experiences during ones formative years can affect redistributive preferences; and the few who do focus exclusively on macroeconomic factors (Giuliano and Spilimbergo 2009; Malmendier and Nagel 2011). We wish to both improve and expand this line of analysis here. In short, we wish to consider the role of both economic and political experiences in early adulthood in determining lasting preferences for more or less redistribution. In this paper we suggest that individuals who share the same birth year, and hence grow up during the same historical period, share some underlying believes about redistributive policy. These durable preferences, evident throughout peoples lives, are in part a product of some kind of formative experience. We test this possibility below, by building upon existing studies (Alesina and Fuchs-Schündeln 2008; Giuliano and Spilimbergo 2009; Malmendier and Nagel 2011), but also by developing these further through (1) combining both economic and political factors during one s formative years, and (2) contrasting the effects of the current political and economic situation (period effects) with these much more distant socialization experiences (cohort effects). We report analyses of the British Social Attitude Survey, which covers the time period 1983 to These repeated cross-sectional surveys include respondents that came of age between 1916 and today, covering an exceptionally interesting and variable time in modern British history indeed, covering the rise of the redistributive welfare state very nearly in its entirety. Results suggest that cohorts collective redistribution preferences are strongly influenced by the level of unemployment, general economic growth, and 2

3 whether the Labour Party was in power during respondents formative years. They further suggest that for the measures of redistributive preferences used here current economic and political situation matters relatively little. Overall, these results present strong evidence that the political and economic conditions during our youth play an important role in shaping our economic preferences that persist throughout later life. What we know about redistribution preferences The existing literature on preferences for redistribution focuses on a combination of individual-level current, past and contextual factors. That said, the bulk of the existing evident relates to the former. That literature focuses in particular on one s own economic circumstances, both in form of income and economic risk, i.e., the likelihood of losing one s job, or finding another one (Kuhn 2010; Rehm 2009; Svallfors 1997, 2003). In short, those who are wealthy have less incentive to support redistribution; as do those least at risk of losing work. (Indeed, there is also some evidence that the gap between the interests of the rich and poor may be increasing over time, see, e.g., Reich 1991.) There has also been a focus on the role of more general attitudes relating to support for redistribution, in particular ideas about luck and hard work (Alesina and Glaeser 2004; Alesina and Angeletos 2005). Related to both is a growing literature on the impact that attitudes about national identity and ethnic diversity, in conjunction with economic circumstances, affect support for redistribution (Johnston et al. 2010). Many of these individual-levels factors have been investigated at the past and/or contextual levels as well. There is work examining the impact of past experiences with poverty and misfortune on attitudes about redistribution (Piketty 1995); work on the potential effect of growing income inequality (Weakliem et al. 2005); and research focusing on the impact of the ethic-diversity context (Soroka et al. 2006). Research has in addition considered factors such as socio-cultural experiences and political indoctrination (Alesina and Glaeser 2004; Alesina and Giuliano 2009), as well as the effects of existing welfare state institutions on public attitudes, where the emphasis is on the tendency for existing institutions to have an impact on citizens beliefs about what the state should or shouldn t be doing (Rothstein 1998). A particularly interesting body of work from the perspective of the current paper is the research that considers the impact of parental socialization (Todd 1985; Esping-Andersen 1999; Benabou and Tirole 2005). This is just a sample of the vast body of work on redistributive attitudes, of course. (For valuable review, see Alesina and Giuliano 2009). And these various bodies of work are overlapping redistribution attitudes are not driven by individual or contextual factors, current or past conditions, but all of them. The contextual literature certainly makes this clear, and our own work below follows suit. Like much of the individual-level literature, we are acutely interested in the impact of economic factors; though we care not just about current economic circumstances, but past ones. Like the contextual literature, we see preferences for redistribution as cumulative, affected by experiences over one s lifetime. In this case, however, we are particularly interested in the impact of a very particular period in one s life. This perhaps most similar to work on parental socialization, work that emphasizes the importance of one s formative years. But here we are not focused on parent s attitudes, but rather the political-economic context in which one comes of age. 3

4 In short, there is a considerable literature probing attitudes about redistribution a literature on which we hope to build below. Formative experiences, economic and political There are certainly diverging views as to whether political attitudes and preferences are stable over one s life span, but the importance of the impressionable or formative years between childhood and adulthood is generally accepted (see e.g., Braungart and Braungart 1986; Sears and Funk 1999; Highton and Wolfinger 2001; Plutzer 2002). Young citizens, it is believed, are not yet set in their political ways and are subsequently more easily influenced by external factors (Jennings 1989; Alwin and Krosnick 1991; Sears and Valentino 1997). Hyman (1959) was among the first to draw attention to the necessity of studying processes of early political socialization. He defined political socialization as an individual s learning of social patterns corresponding to his societal position as mediated through various agencies of society (Hyman 1959: 25). Such agencies can be diverse: family, peers, school, mass media, and as is the focus of our study even the economic and political context. The idea is that the political and economic circumstances under which citizens grew up have a lasting effect on the economic and ideological preferences of the respective generation. If, for example, individuals grow up under an economic recession, they might always prefer more redistribution. Or they could develop the opposite sentiment perhaps as they faced economic hardship, they became more individualistic. The consequence is a lasting predisposition towards more, or less, redistribution. We are aware of only four studies that have investigated the socialization effects of macro contexts on redistributive or economic attitudes. Where economic factors are concerned, the two critical studies are by Giuliano and Spilimbergo (2009) and Malmendier and Nagel (2011). The latter focuses on economic risk taking, accounting for individual s formative experiences with macroeconomic shocks and general trends in stock market returns. They conclude that the memory of these events vanishes over time, though it is also the case that effect of extreme events such as the Great Depression (...) can last for a long time (Malmendier and Nagel 2011: 378). On the other hand, Giuliano and Spilimbergo (2009) find that general experience with recession (not necessarily as dramatic as during the Great Depression) affects individuals economic preferences for their entire lives. Where political factors are concerned, both Alesina and Fuchs-Schündeln (2008) and Svallfors (2010) analyze redistribution preferences in East and West Germany, with an eye on preferences about the role of the state given individuals experiences with Communism (in the East). Both studies conclude that Communism had a lasting impact on individuals preferences namely, it produced a lasting higher acceptance of state interventions in the economy. The specification of the political content in these studies is rather light, however. No study has investigated whether party discourse and popularity, for instance, or direct government activities (such as spending) during one s formative years has a lasting impact on preferences. Indeed, no study has directly considered the possibility that, in a state in which the partisanship of government varies, that partisanship during one s formative years may have a lasting effect. 4

5 We seek to do this here, with a focus on preferences for redistributive policy, and with an eye towards including similar economic and political variables both in the past, and in the present. In short, we are testing the impressionable years hypothesis, which states that redistribution preferences crystallize during a period of mental plasticity in adolescents and remain the base of later-life attitudes. On the other hand (and as shown in the study by Malmendier and Nagel (2011)), these formative experiences might be transcended by more recent events. We are therefore also testing the current events hypothesis, which states that redistribution preferences are mainly influenced by the current economic and political context. That said, as this analysis is new (in terms of both data and models) our hypotheses are rather limited. We expect that the context (whether past, current or both) matters, both economically and politically; but we do not yet have a good sense for how it matters. We regard the models below as a first, mainly investigative, test. That test occurs in the UK, a country that we have selected for both pragmatic and theoretical reasons. Pragmatically speaking, the UK is one of the few countries in the world where there has been an available, nationally-representative opinion survey capturing preferences for redistribution over an extended period of time. But the UK also works well for the kind of analysis we wish to pursue. It is a country which has seen a major extension of the welfare state in the post-war era, and in which debates about the size and scope of that welfare state have been prominent, and ongoing. It is a country that has had a relatively stable two-party system for the last century, and while this is not a requirement for there to be lasting effects of government partisanship during one s formative years, it does make capturing those potential effects somewhat easier. It is also a country in which there have been some relatively distinct periods of government partisanship the Thatcher years stand out as one obvious example, but there are of course others as well. In the case of the UK, in short, the argument that there are cohorts that come of age in quite different political and economic environments seems relatively easy to make. Whether that has any lasting effect of preferences for redistribution is the subject of the sections that follow. Empirical Test Data & variables To assess whether the economic and political context has a lasting imprint on a generation s redistribution preferences, we make use of the British Social Attitude Survey (BSA) for which data is available annually from 1983 to 2009 (except for 1988 and 1992). The cumulative data set contains nationally representative cross-sectional surveys across 25 points in time. 1 The main dependent variable which we label no wealth equality is a measure of an underlying value of fair distribution of a nation s wealth. 2 The question wording reads: How much do you agree that: Ordinary working people do not get their fair share of 1 The data is available for download on 2 The variable was first included in the survey in 1986 and was then asked annually except in 1994 and We hence end up with 20 time points covering 24 years. 5

6 the nation s wealth. Respondents can choose between four categories, from strongly disagree to strongly agree. Looking over the entire period, roughly 63 percent of the British public agrees with this statement. This is a very general question, of course, and it may be relatively immune to current conditions. There is however some space for the impact of current affairs the question asks about the current distribution of income, after all, and so current levels of redistribution may matter. We accordingly include both formative and current versions of our variables capturing the economic and political context. We also tested these models using a different dependent variable, which we label more tax/spending: About the government choosing between these three options. Which do you think it should choose? (1) Reduce taxes and spend less on health, education and social benefits OR (2) keep taxes and spending on these services at the same level as now OR (3) increase taxes and spend more on health, education and social benefits. Given that very few respondents selection the first option, we group responses (1) and (2) into one category to which we contrast those in favor of more spending as well as higher taxes. This is the question which, in our view, is most clearly directed at current conditions. Our interest here, then, is in the possibility that this variable shows a somewhat stronger link to current economic and political condition than the no wealth equality variable. As we are interested in generational differences, the most important variable in our analysis is one measuring cohorts. Following standard strategies in the literature, we group respondents into five-year birth cohorts (Fienberg and Mason 1979; Mason et al. 1973). The key then is to match the information of the economic and political context during the crucial formative years to each five-year cohort. However, as Delli Carpini (1989: 20) notes there is some disagreement on the specific years involved. He notes that age ranges considered crucial can be as diverse as 17 to 25 and 20 to 30. One recent study by Bartels and Jackman (2012) provides an empirical test for what the critical years are. This model predicts that late adolescents (up to 20 years olds) are most affected by current political affairs; this corresponds nicely to ideas about attitude crystallization noted above. We accordingly use 15 to 20 years old as the crucial period. Every respondent who turned 15 during each five-year interval is accordingly grouped into one cohort. To give an illustration: consider for example our first cohort that spans over the years 1916 to This cohort encompasses respondents born between 1901 and The oldest in this group (born 1901) had all their critical years within , while the youngest (born 1905) turned 15 in Using this logic, we group respondents into a total of 20 cohorts from 1916 and To capture the economic and political context, we attach to each respondent a series of economic and political indicators, averaged over the five-year interval corresponding to their socialization experience. There are of course many possible options, but for this first analysis we rely on a relatively simple set of three economic and one political variable. For the economy, we use the unemployment rate, the rate of inflation, and per capita GDP growth. The annual unemployment rate is drawn from the British Office of National Statistic (ONS) data back to 1965, and before that we rely on data reported in Mitchell (1988); the inflation rate is drawn entirely from the long-term measure of prices available through the ONS; and per capita GDP growth is based on data from the OECD back to 1970 and Mitchell (1988) before. 3 Our political variable is very simple: 3 Whenever two separate series are merged, we use backward interpolation interpolating the newest 6

7 Table 1: Cohorts and their socialization context Cohort (socializ.) N Obs Unempl. Inflation GDP growth Labour gov , , , , , , , , , , , , , , , it is the number of years for each five-year period during which there was a Labour Government. A value of 5 corresponds to Labour control over the entire cohort, then; a value of 0 corresponds to a period of Conservative domination. Table 1 shows basic descriptive data for each of our contextual measures, for each cohort. The design of the political variable is relatively clear here: the cohort that came of age (15-20 years old) between 2001 and 2005 was 100% exposed to Labour government, for instance, while the previous cohort had only a 80% exposure (as the Labour Party was still in the opposition in 1996). The table makes clear that there is real variance in the economic context across cohorts as well. Note that for each cohort-level variable there is a current equivalent (which are not reported). That is, we include the five-year average of the unemployment rate for respondents coming of age in a given cohort; we also include the current unemployment rate in the year in which the respondent was surveyed. In this way, we capture the impact of context both in the formative years, and presently. We also include two other contextual variables (for which there are not current equivalents): a time trend, to capture the possibility that preferences for government intervention drift upwards (or downwards) over time; and a variable equal to 1 for cohorts coming of age during one of the two world wars. These were likely very powerful experiences, both politically and economically, and so we test for a resulting shift in redistributive preferences, upwards or downwards, we really aren t sure. Indeed, for both past and current contextual variables, our expectations where the direction of impact is concerned are at this time a little vague. The literature is divided about whether current economic hardship pushes individuals towards or away from series backward using percentage changes in the old series to ensure that there is not a punctuation that is a consequence of merging slightly different data series. 7

8 support for redistribution (Erikson et al. 2002; Soroka and Wlezien 2010). We can imagine either possibility where past economic experiences are concerned as well it is unclear whether economic hardship will push young adults towards support for more or less redistribution. It is similarly unclear whether the party in Government tends to push young adults in their direction, or away from them. Again, we can imagine either possibility. Parties in government have more control over the political discourse, get more media attention and have a larger influence on the political agenda. They are, in short, more prominent and have a higher impact on the generation that comes of age during their time in power (Tilley 2002). This would be in line with previous work by on the impact of communism, discussed above (Alesina and Fuchs-Schündeln 2008; Svallfors 2010). We might thus expect that generations coming of age during a Labour government to be more supportive of redistribution. 4 At the same time, there is a growing body of work suggesting that, in the short term at least, citizens react thermostatically to policy (Wlezien 1995; Soroka and Wlezien 2010). It may well be that growing up in a period of Labour or Conservative hegemony moves one to the other side of the ideological spectrum. Existing research on preferences redistribution, discussed in the preceding section, suggests a number of individual-level predictors that we include here as well. We do not interpret these in much detail below they are for the current purposes control variables, drawn directly from previous work and intended to account for individual variation so that we can better isolate generational differences. They are as follows: age (in years), gender (1=female), education (1=higher education degree), marital status (1 = married), belonging to religious denomination, geographical region, house ownership, income (quintiles), employment status (1 = working full-time or part-time; 2 = unemployed; reference is not working, i.e. retired) and union membership. 5 The particular importance of including age is discussed in the following section; though that note it, along with gender, education, income, house ownership, marital status, employment, and region are linked to socio-economic status, and are thus standard predictors of preferences on redistributive policy. Religion and union membership are expected to be related to the ideological predispositions of respondents. Modeling In order to single out factors that induce cohort differences in redistribution preferences it is essential to account for age as well as period effects. The most problematic features of age-related research have to do with the difficulty of separating out the effects of age, (current) period, and cohort (Glenn 2005). The three factors cannot be identified in simple cross-sectional survey data, of course, since each factor is (almost) completely determined by the other two. But if we fail to account for all three effects, we cannot know whether an observed economic attitude (Y i jt ) of an individual i (i = 1,...,I) is because she belongs to a specific cohort j (C j ; j = 1,...,J) or because of her age (A it ) or the current time 4 Of course, we acknowledge that the party changed their ideological appeal in the mid-1990s to be less pro-redistribution. However, the Labour Party was still more left-wing in economic issues than the rival Conservative Party (Budge et al. 2001). 5 Appendix I reports the descriptive statistics of all variables used here. Appendix II further reports the estimates of these individual-level variables obtained from the hierarchical age-period-cohort (HAPC) regression models presented in Tables 2 and 4. 8

9 t (P t ; t = 1,...,T ; which is typically measured by the survey year). These three factors are related in the following way C j = P t A it (1) Recent advances in social statistical analysis of age, period, cohort (APC) models by Yang and her colleagues (Yang 2008; Yang and Land 2006, 2008) propose the use of hierarchical modelling to overcome this identification problem. More specifically, Yang suggests the use of mixed (fixed and random effects) models allowing for random-intercepts to account for the cross-classified grouping of cohorts and periods (survey years). Using these random effects for cohort and period breaks the linearity of the APC model. Furthermore, in our models the grouping of cohorts into five-year intervals forces these grouped birth-years to have identical effects. This is an additional (more traditional) way to deal with the identification problem in APC-models (Fienberg and Mason 1979; Mason et al. 1973). The rationale behind the cross-classified random intercepts of cohort and period effects is quite simple. Cohorts are usually defined as a group of individuals who were born at the same time and hence grew up under the same political, economic, and social circumstances. As citizens who came of age at roughly the same time share common circumstances, we can assume the errors in a model explaining their economic and political preferences are dependent. It is thus necessary to account for this error-correlation by applying random intercept models. Moreover, these cohorts are clustered within the same survey year. In repeated cross-sectional surveys such as the BSA used in this research, individuals are nested within cells created by the cross-classification of two types of social context: birth cohorts and survey years (Yang and Land 2006: 86). Hence the models presented below estimate fixed effects of age and other individual-level covariates as well as cross-classified random effects of period and cohort. 6 Once we have taken into account the nested character of the data it is possible to evaluate the influence of context-specific variables such as the unemployment or party in power during a respondent s formative years on her redistribution preferences. Such a hierarchical age-period-cohort (HAPC) regression model for the dependent variable no equal wealth distribution the ith respondent for i = 1,...,n jt within the jth cohort for j = 1,...,J socialization cohorts and the tth time period (or survey year) for T = 1,...,20 can be specified as follows: 7 NoEqualWealth i jt = α 0 jt + β 1 Age it + M m=2 β m X mi + ε i jt, ε i jt N(0,σ 2 ) (2) 6 Yang and Land (2008: 323) note that it is technically possible to group the data from an age-by-period matrix with cohort year as the within-cell predictor and period and age (or age group) being the Level 2 predictors. Substantively, however, we believe it is most sensible to use the former specification. As Yang (2006) notes, the age variable in APC analyses is associated with the biological process of aging internal to individuals. By contrast, period and cohort effects reflect the influences of forces that are external to individuals and operate in different ways. 7 See Snijders and Bosker (1999: ) for a general introductory discussion of these cross-classified random models. 9

10 A first model that is reported below, only includes the age of respondents [Pure APC Model]. The second model that is estimated (as presented in equation 2) controls for m individual characteristics (m = 2,..,M) X such as income, education and others described above [Demographic Model]. 8 Most importantly, model 2 includes a random intercept α 0 jt, which specifies whether the overall mean pro-redistribution varies significantly from cohort to cohort and from period to period. These hierarchical models allow us, then, to capture the contextual effects of period and cohort; they can reveal the extent to which individuals attitudes whether the nation s wealth is equally distributed is shaped by the economic and political environment during their impressionable years. Consider: α 0 jt = γ 0 + L l=1 γ l Z l j + u 0 j0 + ν 00t (3) where γ 0 is the mean effect of all time periods across all cohorts. u 0 j0 denotes a cohort specific error term (u 0 j0 N(0,τ u )) and ν 00t a time specific error (ν 00t N(0,τ ν )). This model includes (l = 1,..,L) covariates Z that account for the political and economical context specific factors during the formative years of respondent i. These are assumed to explain intercohort variation in levels of redistribution preferences. This is the third model that is estimated [Socialization Context Model]. In a final step we add k (k = 1,..,K) covariates P that also take into account the current economic and political situation. The final forth model to be estimated is hence [Current Context Model]: α 0 jt = γ 0 + L l=1 γ l Z l j + K k=1 δ k P kt + u 0 j0 + ν 00t (4) Results Results of the four models of no wealth equality are shown in Table 2. The first column shows estimates from the Pure APC Model; the second column adds demographics; the third and fourth add past (cohort) and current contexts. Coefficients actually change relatively little from one model to the next, but there are some important differences in the estimated variance components. Note, for instance, that the estimated variance across cohorts more than quadruples from the first to the second model. This is a product of more accurate estimates of cohort effects once we have accounted for demographic differences (particularly income and education). To illustrate these cohort effects, Figure 1 displays the predicted random effect coefficients for cohorts. Clearly, cohort differences exist in regard to redistribution preferences. Cohorts that came of age in the 1950s to early 1970s are most likely to agree with the statement that the nation s wealth is unequally distributed. On the other hand, those socialized during the economic crises in the late 1970s and early 1980s seem to collectively disagree more with this statement that the older cohorts. 8 As this model does not include a random slope-coefficient it is not necessary to center the individual-level explanatory variables (Snijders and Bosker 1999: 80-81). 10

11 Table 2: Linear HAPC models on no wealth equality : cross-classified random effects MODEL 1 MODEL 2 MODEL 3 MODEL 4 Pure Demo- Socializ. Current APC graphics Context Context Intercept (.022) (.038) (.148) (.301) Individual Fixed Effects Age (.000) (.001) (.002) (.002) Included Gender, married, union member, own house, degree, religion, income, region, labour force status. [See Appendix II.1 for estimates]. Socialization Contextual Fixed Effects at c Unemployment (at c) (.213) (.213) Inflation (at c) (.002) (.002) GDP growth (at c) (.546) (.546) Labour in gov. (at c) (.022) (.022) War (at c) (.027) (.027) Time trend (.008) (.012) Current Contextual Fixed Effects at t Unemployment (at t) (1.511) Inflation (at t) (.009) GDP growth (at t) (.655) Labour in gov. (at t) (.081) Variance Components Cohort ( ) (.007) (.011) (.007) (.007) Period ( ) (.012) (.011) (.009) (.009) Residual (.003) (.003) (.003) (.003) N of obs. 50,785 LogLik -67,312-65,463-65,452-65,450 Significance levels: p<.05, p<.01 p<.001. Data: British Social Attitude Survey ( ). Note: Entries are OLS-coefficients estimated by a mixed generalized linear model. Dependent variable: Ordinary working people do not get their fair share of the nation s wealth. -100: disagree strongly; +100 agree strongly. 11

12 .15 APC + Demographics Predicted random effect Cohort.15 APC + Demographics + Socialization Context Predicted random effect Cohort Figure 1: Predicted random cohort effects on no wealth equality (including 95% C.I.) Once we take into account the socialization context by adding the unemployment, inflation, GDP growth rate, and the government participation of the Labour Party, the observed cohort differences vanish. This is very clear in the estimated effects shown in Figure 1, and the related variance estimates in model three (in Table 2). The implication of this decreased variance is that much of what we saw in the preceding figure was a product of economic and political context; put differently, context plays an important role in the creation of systematic differences across cohorts. In fact, these variables account for a little over half the variance across cohorts. And one principle driver in this case is the unemployment rate. A one-percentage-point increase in the average unemployment during respondents formative years is associated with an average decrease in their preferences for redistribution of about -.8. to be clear: economic hardship, ceteris paribus, makes individuals less supportive of redistribution. That said, unemployment is not the only contextual effect that matters. GDP growth also seems to have a lasting effect on preferences for redistribution, and the effect here partly reinforces what we have seen for unemployment. GDP growth is positively correlated with support for redistribution so just as hardship (in the form of unemployment) pushes support down, success (in the form of GDP growth) pushes support up. It is not 12

13 Table 3: Model Comparison (Dep. Var: no wealth equality ) Model LOGLIK RATIO TEST (Prob > chi 2 ) 1. Pure APC 2. Demographics To pure APC To demog. To socializ. Context Models 3. Socializ. context Socializ. + current context Note: Entries are model fit statistics of cross-classified random effects models on no wealth equality. surprising, though perhaps somewhat disappointing, to see that support for redistribution is greatest amongst those who needed it least. These findings are in line with what has been found in other recent work focusing on the current effect of economic factors on policy preferences (Soroka and Wlezien 2010). That said, the final model in Table 2 reveals little impact of current economic conditions on the dependent variable. The limited impact of current conditions is clear in the rather small reduction in the estimated variance for periods (survey years); it is also readily apparent given that no current contextual variable is statistically significant. We can hence reject the current events hypothesis, at least for the variables explored here. There is one current variable that just narrowly misses statistical significance, however; and indeed it s past/cohort equivalent does as well: Labour Government. In both cases, we see hints of a thermostatic reaction to partisanship. The finding regarding current effects is no surprise there is a growing body of evidence of this thermostatic reactiveness on the part of the public, in the UK and elsewhere. That Labour governments during our formative years may push us towards the Conservatives, however, is a relatively new and unique finding. It stands in contrast with what past work has found regarding the experience with state-socialism (Alesina and Fuchs-Schündeln 2008; Svallfors 2010) or governments lasting imprint in partisanship (Tilley 2002). But it makes some sense in a democratic context. Growing up in an environment that is dominated by one party (or the other) tends to shift young citizens views towards the other side of the political spectrum not just temporarily, but permanently. To sum, the results of the HAPC models predicting cohort differences in attitudes about the equality of wealth distribution show that most importantly cohort differences exist. Confirming the impressionable years hypothesis, these differences can be partly explained by the economic situation during a cohort s late adolescence. Current political and economic factors do not seem to matter once this formative experience is accounted for. These results are confirmed when conducting basic log-likelihood ratiotests reported in Table 3. As the first column reveals it is very important to take into account individual-level characteristics. All models perform better than the pure APC model, which only includes the age (individual fixed-effect), cohort, and period variables (as cross-classified grouping variables). However, comparing the models that only account for demographics to the two models including formative experiences confirms that socialization matters. The model fit is significantly improved for the context mod- 13

14 els accounting for the political and economic situation during a cohorts formative years. Table 3 also reveals that the current context does not matter for explaining differences in economic attitudes. Including these variables into the model does not improve the overall fit. Robustness of findings One reason for the non-existence of period effect could be the choice of our dependent variable, which might not be sensitive to the current environment, but rather taps a stable underlying attitude. In a next step, we therefore repeated our analyses using a different dependent variable, which we believe may be more sensitive to capture the current political and economic situation. Table 4 reports the results of the HAPC models predicting attitudes for more taxing and spending. Note that the dependent variable is dichotomized, which yields a logistic regression model. The coefficient size is hence not comparable to those presented in Table 2. The models presented in Table 4 overwhelmingly confirm the results presented and discussed above. 9 Current period effects do not matter even using another dependent variable. However, again high unemployment and low GDP growth during a cohort s formative years lead to anti-redistribution preferences in later life. As confirmed in Table 4 and illustrated in Figure 2, which displays the predicted random effect coefficients of more tax/spending, cohort differences exist. Most astonishing are of course the strong effects of the socialization experience on attitudes whether the government should tax and spend more. These significantly reduce the variance of cohort differences. Conclusion and discussion We see this work as a good first step in an ongoing analysis of how redistributive preferences are affected by economic and political conditions as citizens come of age. We have the benefit of being able to build on a vast body of work on preferences for redistribution, both in the UK and elsewhere. We are also able to draw on a growing body of work on political socialization and related APC models. That said, there clearly is much more to do. There are a few relatively simple adjustments, to start. We would like to consider other specifications of our economic and political variables; and we would like to add levels of government spending to the models as well. We would like to consider dependent variables that capture even more directly the impact of current as well as past conditions. We need to experiment with different cohorts defining cohorts in narrower (2-year?) intervals, for instance. We also need to consider the possibility of heterogeneity of effects. It may be, for instance, that the shift in mean differences among cohorts mainly comes 9 We also tested identical models on several other measures of redistributive preferences: (1) The government should redistribute income from the better-off to those who are less well off. ; (2) The government should spend more money on welfare benefits for the poor, even if it leads to higher taxes ; and (3) If welfare benefits were not so generous people would learn to stand on their own two feet. All questions follow the same strongly disagree to strongly agree continuum as our no wealth equality measure. Results are also largely the same, and so we do not report results here. 14

15 Table 4: Logistic HAPC models on tax/spending pref.: cross-classified random effects MODEL 1 MODEL 2 MODEL 3 MODEL 4 Pure Demo- Socializ. Current APC graphics Context Context Intercept (.146) (.166) (.538) (1.018) Individual Fixed Effects Age (.002) (.002) (.005) (.006) Included Gender, married, union member, own house, degree, religion, income, region, labour force status. [See Appendix II.2 for estimates]. Socialization Contextual Fixed Effects at c Unemployment (at c) (1.200) (1.200) Inflation (at c) (.010) (.010) GDP growth (at c) (2.800) (2.800) Labour in gov. (at c) (.121) (.120) War (at c) (.152) (.152) Time trend (.029) (.033) Current Contextual Fixed Effects at t Unemployment (at t) (7.033) Inflation (at t) (.046) GDP growth (at t) (3.671) Labour in gov. (at t) (.381) Variance Components Cohort ( ) (.044) (.050) (.032) (.031) Period ( ) (.051) (.051) (.052) (.049) N of obs. 54,656 LogLik -36,450-36,151-36,142-36,140 Significance levels: p<.05, p<.01 p<.001. Data: British Social Attitude Survey ( ). Note: Entries are logit coefficients estimated by a mixed generalized linear model. Dependent variable: 1 = government should increase taxes and spending; 0 = government should tax/spend less or the same. 15

16 .6 APC + Demographics Predicted random effect Cohort.6 APC + Demographics + Socialization Context Predicted random effect Cohort Figure 2: Predicted random cohort effects on tax/spend more (including 95% C.I.) 16

17 from shifts among the middle class; while the poor are always pro-redistribution and the wealthy are always against it. We would expect that the middle class will be strongly against redistribution if the economy is doing badly, after all, since they are the ones most likely to pay more. Intra-cohort variance may be different across income groups, in short, and we need to explore this further. There are also some potentially conflating factors that require further investigation. Inter-cohort differences in redistribution preferences might partly be a product in cohort differences in their overall success in life, which is influenced by their entrance to the labour market (c.f. Brunner and Kuhn 2009; Bell and Blanchflower 2011; Ehrmann and Tzamourani 2012). The argument in this case is that there is some unobserved heterogeneity in initial economic conditions and cohort characteristics, which produce differences in economic preferences. Maybe in the end it is all just about individual-level characteristics, but these are skewed as a product of different entrance conditions. Alternatively, cohorts entering the labour market during economic crises may develop higher risk-aversions as well as lower expected life-income. Note however that this should actually lead to preferences of higher redistribution Giuliano and Spilimbergo (2009), and that is not what we see here. That said, our analysis of five different dependent variables, each of which produces roughly similar estimates, suggests that further work will not change dramatically our general results. We expect that, pending additional robustness checks of course, we will end up in about the same place we are now. Which is to say that we expect to find that (a) we can account for a good amount of the variance in cohort effects with a relatively simple group of variables capturing the economic and political context at the time at which respondents came of age, (b) economic hardship during one s formative years seems to depress support for redistribution, and (c) the partisanship of government tends to lead those coming of age towards the opposite side of the political spectrum. In sum, there is more work to do, but we believe our results thus far are promising. Past work has shown that current conditions are important to individuals attitudes towards redistribution. Past experiences can matter as well. And here, we have some preliminary evidence that the economic and political conditions during one s formative years may be particularly influential. 17

18 References Alesina, Alberto and George-Marios Angeletos Fairness and Redistribution: US vs. Europe. American Economic Review 95: Alesina, Alberto and Nicola Fuchs-Schündeln Good-Bye Lenin (or Not?): The Effect of Communism on People s Preferences. American Economic Review 97: Alesina, Alberto and Edward L. Glaeser Fighting poverty in the US and Europe: A world of difference. Oxford University Press. Alesina, Alberto F. and Paola Giuliano Preferences for redistribution. Working Paper 14825, National Bureau of Economic Research. Alwin, Duane F. and Jon A. Krosnick Aging, Cohorts and the Stability of Sociopolitical Orientations over the Life Span. American Journal of Sociology 97(1): Bartels, Larry M. and Simon Jackman A generational model of political learning., presented at the workshop New approaches in age, period, cohort analysis ; Oxford. Bell, David N.F. and David G. Blanchflower Young people and the Great Recession. Oxford Review of Economic Policy 27(2): Benabou, Roland and Jean Tirole Belief in a just world and redistributive politics. Quarterly Journal of Economics 121(2): Braungart, Richard G. and Margaret M. Braungart Life-course and Generational Politics. Annual Review of Sociology 12: Brunner, Beatrice and Andreas Kuhn To Shape the Future: How Labor Market Entry Conditions Affect Individuals Long-Run Wage Profiles., Institute for Empirical Research in Economics, University of Zurich Working Paper No Available at SSRN: or Budge, I., H-D. Klingemann, A. Volkens, J. Bara, and E. Tanenbaum, eds Mapping policy preferences: Estimates for parties, electors, and governments Oxford: Oxford University Press. Delli Carpini, Michael X Age and history: Generations and sociopolitical change. In Political Learning in Adulthood, Roberta S. Sigel, ed., pages 11 55, Chicago: University of Chicago Press. Ehrmann, Michael and Panagiota Tzamourani Memories of high inflation. European Journal of Political Economy 28: Erikson, Robert S., Michael B. MacKuen, and James A. Stimson The Macro Polity. Cambridge: Cambridge University Press. Esping-Andersen, Gosta Social foundations of postindustrial economies. Oxford: Oxford University Press. Fienberg, Stephen E. and William M. Mason Identification and estimation of age-periodcohort models in the analysis of discrete archival data. Sociological methodology 10:1 67. Giuliano, Paolo and Antonio Spilimbergo Growing up in a Recession: Beliefs and the Macroeconomy. Working Paper 15321, National Bureau of Economic Research. Glenn, Norval D Cohort Analysis. Thousand Oaks, CA: Sage, 2 ed. Highton, Benjamin and Raymond E. Wolfinger The First Seven Years of the Political Life Cycle. American Journal of Political Science 45(1): Hyman, Herbert H Political socialization: A study in the psychology of political behavior. Glencoe: Free Press. Jennings, M. Kent The Crystallization of Orientations. In Continuities in political action, Samuel H. Barnes, Jan W. van Deth, and M. Kent Jennings, eds., Berlin and New York: de Gruyter. Johnston, Richard, Keith Banting, Will Kymlicka, and Stuart Soroka National identity and 18