Grasso, M.T. (2014) “Age, Period and Cohort Analysis in a Comparative Context: Political Generations and Political Participation Repertoires in Western Europe.\" Electoral Studies 33: 63-76

Electoral Studies 33 (2014) 63–76

Contents lists available at SciVerse ScienceDirect

Electoral Studies journal homepage: www.elsevier.com/locate/electstud

Age, period and cohort analysis in a comparative context: Political generations and political participation repertoires in Western Europe Maria T. Grasso* Department of Politics, University of Sheffield, Elmfield, Northumberland Road, Sheffield S10 2TU, United Kingdom

a r t i c l e i n f o

a b s t r a c t

Article history: Received 14 June 2013 Accepted 24 June 2013

This paper presents a method for studying age-period-cohort effects in a comparative context where repeated cross-sectional data are available covering a suitably long period of time. The method consists in the application of multi-level models with country as the higher level of analysis and random coefficients to model variables which vary at the country-level. Additionally, the application of generalized additive models (GAMs) and generalized additive mixed models (GAMMs) provides robust empirical tests of cohort categorizations applied in this and previous studies to estimate otherwise collinear effects. To illustrate the method, I derive and test the theory that generations will be differentiated in their patterns of participation based on the ascendancy of certain repertoires in the era of their political socialization. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Participation Political generations Western Europe Multi-level models Generalized additive models (GAMs) Generalized additive mixed models (GAMMs)

1. Introduction In political science research it is often crucial to analyse the relative importance of age, period and cohort effects to understand the origins and trajectories of social change. Social ageing, historical context and generational membership are all three related to the passing of time but often have divergent effects from each other. These three time effects have different implications for what we can expect from the future, given inter-generational replacement. The fundamental importance of disentangling these effects for explaining the occurrence and emergence of various social phenomena means that we must devise strategies to deal with the age, period and cohort “identification problem” in different research contexts (see Introduction to this Special Symposium by Neundorf and Niemi, 2014). The ‘identification problem’ stems from the fact that three effects cannot be estimated simultaneously. This is since age period and cohort are in a linear relationship with each other. As soon as we know two of the values (someone’s age and the year in which * Tel.: þ44 (0) 1142 221 702. E-mail address: m.grasso@sheffield.ac.uk. 0261-3794/$ – see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.electstud.2013.06.003

they were surveyed, for example), the third value in the relationship (i.e., in this case, their year of birth) is automatically known. To deal with this methodological hurdle and allow for the estimation of all three effects simultaneously, we must devise strategies, or different methodological approaches, that allow us to ‘break’ this linearity. The way in which we choose to ‘break’ the linearity of the age-period-cohort relationship will be largely influenced by the theoretical expectations of the research and the availability of ‘side-information’ to support simplifying assumptions or constraints on one or more of the three effects. Indeed, it is the substantive meaning attached to each of the three effects in terms of the research at hand which normally holds the key for determining which simplifying assumptions are the most legitimate and useful in a given research context (Glenn 1976, Tilley 2002, Tilley and Evans 2014). With this in mind, this paper presents a method for studying age, period and cohort effects in a comparative context where repeated cross-sectional data are available covering a suitably long period of time so that members of the same cohorts are observed at different historical moments and in different phases of their life-time. More specifically, the method presented in this paper applies multi-level

64

M.T. Grasso / Electoral Studies 33 (2014) 63–76

models (Snijders and Bosker, 1999) with country as the higher level of analysis and random coefficients to model those variables which vary at the country-level. This modelling strategy has the distinct advantage of accurately reflecting the fact that observations are nested within countries and that not all variables have the same effects cross-nationally. This kind of approach is useful in a comparative context so as to allow for the correct modelling of those effects which vary between countries while at the same time recognising that there is some random variability at the country-level. This is a significant improvement on a cumbersome fixed effects approach with interactions which would estimate an inconvenient number of parameters and also importantly ignore the random variability at the country-level. To ‘break’ the linearity in the age-period-cohort relationship and simultaneously estimate all three effects in the analysis, the method applied here relies on the transformation of the continuous year of birth variable into a fivecategory cohort variable. This means that people who are born within a given period are set to have equal cohort effects. As Rosow (1978: 69) pointed out, “the general bounding criteria for cohorts [cannot] be clearly established independent of specific analytic questions to delineate them”. However, any categorization of cohort, no matter how theoretically sound, always runs the risk of losing information or applying the wrong ‘cuts’. Spitzer (1973: 1358) points out that there is always going to be a boundary problem of where to delineate social generations in the “seamless continuum of daily births” and that there is always unavoidable ambiguity in terms of where to apply the ‘cuts’. This problem becomes even more important if the cohort analysis is done in a comparative context. Cohorts of the same birth year might differ as they experienced different formative events in their respective home-countries. This paper takes this criticism of a priori theoretical categorizations seriously, and unlike previous studies, provides a robust and novel empirical test of the categorization of cohorts, developed from theory. This is accomplished through the application of generalized additive models (GAMs) and generalized additive mixed models (GAMMs). Both types of analysis allow us to plot the non-parametric smoothed curve for the effect of year of birth (for example, see Tilley, 2002 for an application of GAMs to study political generations in the UK). The utility of the application of the GAMs to plot the countryby-country smoothed cohort effects is that it allows us visually check whether cohort effects are similar across countries. Diagnostic country-by-country logistic regressions were also estimated to allow for the most accurate set-up of the multi-level models. The advantage of the application of GAMMs, on the other hand, is that it allows us, just like in the multi-level models, to include random effects for those variables which vary at the country-level. Thus, by plotting the non-parametric smoothed curve for the effect of year of birth for the whole sample without ignoring the nested structure of the data the GAMMs crucially provide a means to visualise the shape of the cohort effects and overcome the need for categorizations in this context. This gives us greater confidence in our results. While categorizing cohorts is still necessary to estimate the

multi-level models, GAMMs allow us to visualise the shape of the cohort effects and thus provide a robust and novel empirical test to show that the theoretically-motivated cohort cut-offs applied for the multi-level models did not lead to biased results. 2. Political generations and political participation in Western Europe I illustrate this method for age, period and cohort analysis by examining generational differences in various political activities in Western Europe. I hypothesize that certain generations are more likely to engage in specific political acts than other generations, based on the relative importance of different repertoires of participation in the historical context of a generation’s coming of age. In particular, this theorising suggests that older generations, coming of age in a period when mass parties and elections shaped social cleavages and were fundamental to the existence of democratic government, will have higher levels of party membership. In contrast, the generation coming of age in 1960s and 1970s, during the ascendancy of ‘unconventional’ modes of participation, are more likely than both older generations, but also than younger generations coming of age in subsequent, less politicised political contexts, to protest, petition and participate in social movement organisations (SMOs) (see also Grasso, 2011; Grasso, 2013a, 2013b for more on this). The five-category distinction between cohorts applied in the multi-level models assumes that the historical periods in which individuals have spent the majority of their ‘formative years’ (here understood to be 15–25 years of age, but see Bartels and Jackman (2014) in this Special Symposium for evidence that this crucial ‘impressionable’ period might come earlier) are sufficiently different as to warrant the distinction of five generations each with their particular values and proclivities. It stands to reason that coming of age in periods as different as pre-WWII, post-WWII, 1960s– 1970s, 1980s and 1990s will present divergent experiences. I analyse data from the European Values Study 19812008 (EVS, 2011) on observations of individuals born between 1909 and 1981 in ten advanced Western European countries: Belgium, Denmark, France, West Germany, Great Britain, Ireland, Italy, the Netherlands, Spain and Sweden. Despite intricate national trajectories, the broad historical patterns identified as salient for determining generational differences in the modes of political action in this study are common to all ten countries analysed here. In all of them, political parties and the social cleavages they represented, particularly around class, but also around religion and language/region, were the fundamental structuring boundaries of democratic competition at least until the 1960s. All ten nations, even though some to a greater and some to a lesser extent, shared a period of economic affluence and heightened radicalism around educational institutions and youth in the late-1960s and 1970s. Finally, the de-politicisation of public life in the wake of the Cold War, the convergence of mainstream parties on the centre of the ideological spectrum, and the withering ideological struggle between grand-narratives of Left and Right in the age of what Francis Fukuyama famously dubbed ‘the End of Politics’

M.T. Grasso / Electoral Studies 33 (2014) 63–76

is another development which regardless of national specificities is common to all ten countries included in this study. Mannheim (1928: 232) famously argued that “youth experiencing the same concrete historical problems may be said to be part of the same actual generation”. These experiences are understood to crystallise and differentiate generations in the population even as they mature through the life-course so that generational differences are constant through the life-cycle. It is thus with respect to theoretical expectations about the effect of the historic period of socialization on cohorts’ political attitudes and behaviours as distinct from those of previous (and later) cohorts that demographic cohorts can be given content, and come to be identified as distinct political generations. The method applied in this study allows us to investigate age, period and cohort effects to provide results which support the political generations narrative and show that the period in which a generation experienced its formative years leaves a lasting imprint on the participatory proclivities of generations relative to those coming of age in other eras. The analysis investigates the following substantive research questions to illustrate the method for age-periodcohort analysis in a comparative context applied in this study: Are generations coming of age in given periods when certain modes of participation are ascendant more likely to engage in these repertoires than generations coming of age in other periods? In other words, is the 1960s–70s Generation the most likely to engage in protest activism, i.e. demonstrate, petition, and participate through SMOs? Are the two older generations more likely to be involved with parties, i.e. engage in institutional activism? In order to investigate these research questions, we need to draw on historical evidence and previous scholarship on generational theory and categorization of generations in Western Europe. 3. Method 3.1. Categorizing political generations Jennings (1987: 368) explained how individuals coming of age during periods of pronounced stress and drama, epochal events or rapid socio-economic change are often said to be uniquely identified in a political sense – hence such labels as the “depression generation” the “silent generation” etc. Becker (1990: 2; 1992: 222) defined a generation as “characterised by a specific historical setting and by common characteristics”. For Becker, these generations were the pre-war generation, the silent generation, the protest generation, the lost generation, and the pragmatic generation. This categorization, which is also employed in Van Deth and Elff (2000) for Western Europe, and those from other studies, are presented in Table 1 and compared to the very similar one developed for this study. However, once a categorization of cohorts is devised, it is also crucial to establish how individuals born in different years are assigned to a generation. The key advantage of the method for doing so applied in this study is that individuals are assigned to a political generation based on the period in which they spent the majority of their formative years (15– 25 years of age). For example, someone born in 1948 would

65

have turned 15 in 1963, spending three formative years in the post-WWII era; however, she would have spent most of her formative years in the 1960s–70s era (1966–1977). As such, it would be wrong to categorize her as a member of the post-WWII generation; she is more accurately classified as part of the 1960s–70s Generation.1 As noted in the introduction, we can isolate the confounding effects of ageing and time period in order to examine ‘pure’ cohort effects by using repeated crosssectional data over a prolonged period of time since this allows us to include individuals from the same cohorts interviewed at different historical moments and in different phases of their life-course. Table 2 illustrates this point for the respondents included in this study and shows the advantage of having data for repeated cross-sections spanning the 1980s through to the late 2000s. This is a crucial precondition to be able to estimate ‘pure’ cohort effects, net of age and/or period effects in the models. By including terms for cohort, age and year of survey in the models we can thus separate age, period and cohort effects and obtain coefficients for ‘pure’ cohort effects. This in turn allows us to say something meaningful about generational differences in participation. Without the inclusion of all three effects in the models to be estimated with data from repeated cross-sections so that individuals are observed at different times in their life-course and during different historical periods, it would be unclear whether the effects for year-of-birth-groupings are actually capturing the effects of socialization in a given historical period as opposed to the effects of biological ageing and/or differential contextual influences in a given year of survey (Mattei, et al.1990; Jennings and Niemi,1991; Evans and De Graaf 1996; Van Den Broek, 1995; Andersen and Fetner, 2008). More specifically, in order to be able to estimate cohort effects net of age and period effects, the multi-level models presented in the analysis section of this paper include: a five-category variable for cohort (Pre-WWII Generation, Post-WWII Generation [reference category], 1960s–70s Generation, 1980s Generation, 1990s Generation); a fourcategory factor for year of survey (1981 [reference category], 1990, 1999, 2008); age both as a linear term and as a quadratic term in order to capture the linear and curvilinear components of age effects on participation. For the purpose of illustrating this method for ageperiod-cohort analysis in a comparative context, in the remainder of this paper, I will test a theory for generational differences in participation derived from Mannheim’s (1928) seminal work on political generations in general. The theory, outlined in previous sections, is that political generations are differentiated in their patterns of participation based on the ascendancy of certain repertoires of political action in the historic context of their political socialization.

1 Although other studies in the literature appear to employ similar categorization of cohorts, the way in which they classify individuals into these categories is not the same as the one applied in this study. This is partly due to the fact that the research questions under investigation differ between studies, which means that different methods for categorizing individuals are more or less appropriate. For example, Clarke et al. (2004) are mainly interested in turnout and vote choice and therefore they base their classification on when an individual turned 18.

66

M.T. Grasso / Electoral Studies 33 (2014) 63–76

Table 1 Cohort classification based on socio-historic period of formative years compared to others in the literature. The categorization employed in this study:

Pre-WWII Generation

Post-WWII Generation

60s–70s Generation

80s Generation

90s Generation

Context

Era/period Year of birth

1929–1945 1909–1925

1946–1965 1926–1945

1966–1977 1946–1957

1978–1988 1958–1968

1989–2001 1969–1981

Western Europe

The categorization employed in Van Deth and Elff (2000)

Pre-war Generation

Silent Generation

Protest Generation

Lost Generation

Pragmatic Generation

Context

Formative years

Depression, war

Reconstruction

Crisis, individualism 1956–1970

Pragmatism

Western Europe

Year of birth

1910–1930

1931–1940

Affluence, radicalism 1941–1955

The categorization employed in Heath and Park (1997):

Pre-war Generation

Post-war Generation

60s Generation

80s Generation

Context

Formative years Year of birth

1920s/30s Before 1926

1940s/50s 1927–1945

1960s/70s 1946–1960

1970s/80s After 1960

Great Britain

The categorization employed in Clarke et al. (2004):

Post-WWII Generation

Macmillan Generation

Wilson/Callaghan Generation

Thatcher Generation

Blair Generation

Context

Formative years Year of birth

Before 1950 Before 1932

1951–1964 1933–1946

1964–1979 1946–1961

1979–1992 1961–1974

Post-1992 Post-1975

Great Britain

I will call this the political generations theory, since it is inspired by the approach to political generations in general exemplified by Mannheim’s (1928) work. For the purpose of the discussion of results in the analysis, I will contrast the expectations of this theory to those implied instead by one version of the societal modernization account. In contrast to the political generations theory, the societal modernization account does not locate generational differences in participation in the specific politico-historical context of socialization of different cohorts. Rather, changing patterns of participation in all advanced postindustrial democracies are understood to be underpinned by a similar underlying process of societal modernization which leads ‘younger cohorts’ with more ‘post materialist’ value-orientations to shun ‘hierarchical, bureaucratic’ forms of institutional (or ‘elite-directed’) participation – such as voting, party membership and campaign activities – in favour of more fluid, ad hoc types of extra-institutional (or ‘elite-challenging’) participation – such as protest, engagement in social movement organisations (SMOs), petitioning, boycotting, etc. (Inglehart, 1977, 1990; Inglehart and Catterberg, 2002). In turn, the implied inter-generational shift from ‘elitedirected’ to ‘elite-challenging’ activities is seen to be the cause of an aggregate-level rise in extra-institutional participation and decreasing levels of institutional participation in the population as older cohorts are replaced by younger ones through natural processes of generational replacement. So, according to the societal modernization account of generational differences in participation, the 1980s and 1990s Generations should protest more, not less, than the 1960s–70s Generation, even though the latter experienced political socialization in a more radical and politicised historical period.

Post-1970

It is worth stating here that even in those versions of the societal modernization account which allow for some fluctuation in this pattern between successive cohorts, this is not attributed to the political context of socialization, but rather to levels of economic security obtaining during prepubescent years. This is since one of the main premises of the societal modernization account is that affluence during childhood leads to the development of values which foster extra-institutional participation and suppress institutional participation, as discussed above. However, given that the 1960–70s and 1990s Generations exhibited similar levels of affluence during childhood in contrast to the 1980s Generation (the late 1970s–1980s were marked by economic crisis and high unemployment across Europe), provided that our evidence will show that the 1990s Generations does not revert to participation patterns akin to those of the 1960s Generation, as would be assumed even by this ‘less strict’ version of the societal modernization account, then this suggests that variations in economic affluence at childhood cannot account for generational differences that do not conform to the ‘strict’ societal modernization account. Now, spelling out the expectations of the political generations theory and the ‘strict’ societal modernization account, the Pre-WWII Generation in particular, but also the Post-WWII Generation, experienced their political socialization (respectively, 1929–1945 and 1946–1965) during the classic period of cleavage congealment (Lipset and Rokkan, 1967) when mass parties were very popular. Based on the political generations account, we would expect these generations to be the most likely to engage in institutional political activities (e.g., party membership). However, since key changes in the weakening popularity of mass parties were in effect by the 1960s, by the same account, we would not

M.T. Grasso / Electoral Studies 33 (2014) 63–76

67

Table 2 Cohort representation in each survey wave, EVS 1981–2008. Pre-WWII Generation

Post-WWII Generation

60s–70s Generation

80s Generation

Era/period

Until 1945

1946–1965

1966–1977

1978–1988

90s Generation 1989–2001

Year of birth

1909–1925

1926–1945

1946–1957

1969–1981

1981 N %

At age 56–72 2140 21

At age 36–55 3070 30.3

At age 24–35 2723 27

1958–1968 At age 18*–23 2198 21.7

1990 N %

At age 65–81 1837 14.1

At age 45–64 3982 30.7

At age 33–44 2984 23

At age 22–32 3166 24.3

At age 18*–21 1039 7.9

13,008 100

1999 N %

At age 74–90 753 6.8

At age 54–73 2932 26.5

At age 42–53 2280 20.5

At age 31–41 2596 23.5

At age 18–30 2510 22.7

11,071 100

2008 N %

At age 83–90* 316 3.1

At age 63–82 2687 26.1

At age 51–62 2394 23.3

At age 40–50 2348 22.9

At age 27–39 2525 24.6

10,270 100

Totals %

5046 11.4

12,671 28.5

10,381 23.3

10,308 23.2

6074 13.6

44,480 100

Total 10,131 100

* Notes: All respondents younger than 18 and those older than 90 were dropped from the sample so that all respondents would have had the opportunity for participation (very young people would not have had many opportunities to participate before 18 and very old people tend to have mobility issues).

expect significant differences in this type of participation between the 1960s–70s Generation and, respectively, the 1980s and 1990s Generations or between the 1980s and the 1990s Generations (H1). On the other hand, for the societal modernization account to be supported, each younger cohort should engage in institutional activities less than the previous cohort (H2). For institutional participation, the Pre-WWII Generation should be more likely than the Post-WWII Generation to participate through political parties; the Post-WWII Generation should be more likely than the 1960s–70s Generation to join a party, and so forth. Compared to the previous two generations, the formative years of the 1960s–70s Generation (1966–1977) were marked by high levels of political militancy and ideological polarisation across Western Europe. Left-wing movements and parties became strong and young people for the first time emerged as a social and political force in society. Perhaps best symbolised by the Mai 1968 uprisings across Western Europe (but also around the world), the late-1960s and 1970s resound in popular imaginary as period of social upheaval and radicalism (Della Porta and Diani, 2006). This was also a period when many structural transformations emerging out of WWII and post-war developments were coming to the fore. The Cold War was in full sway and student organisations emerged to challenge the perceived hypocrisy of both ideological blocs. The 1960s–70s Generation coming of age in this period, also experienced rising societal affluence, the boom of mass production, and the expansion of higher education. For the first time, young people from different social backgrounds came together, sharing ideas and experiences, but also building solidarity and a shared identity by fighting together to redress common grievances. In particular, this generation did not have obvious conventional routes - the voting age was still 21 (Katz, 1997: 218–229) – to challenge the traditional structures of authority which confronted it and therefore resorted to ‘unconventional’ political repertoires such as protest and occupations, nowadays seen as

legitimate if not mainstream methods of democratic engagement (Van Aelst and Walgrave, 2001). It is in a sense perfectly fitting that challenging the ‘conventional’ realm of politics and societal hierarchy, also demanded that the methods employed be new. However, this period of unbridled optimism and of the expansion of youth engagement in society and widening opportunities for young people soon dissipated and the late 1970s brought with them economic crisis and mass youth unemployment. The socialization context of the 1980s Generation was thus quite different from that of the 1960s–70s Generation. As discussed above, some scholars called them ‘the lost generation’ since the unfavourable societal conditions at the time meant that youth were unable to enjoy many of the advantages of the previous generation. Moreover, the political zeitgeist through the 1980s was turning increasingly conservative in many countries, and emphasis was placed on individual responsibility as opposed to political and grouplevel solutions to social problems such as inequality. Finally, by the time that the 1990s Generation was coming of age the political landscape was even further transformed with the end of the Cold War and the symbolic victory of market capitalism in the face of the demise of really-existing socialism. As a result of these events, the Left (whether it had remained faithful or had distanced itself from Moscow) lost many supporters across Western Europe while once MarxistLeninist or Communist parties subscribed to reformist political programs and accepted market capitalism with some welfare provisions, also increasingly recoiling from grandideological narratives (Furedi, 2004), preferring instead to focus on more managerial and technocratic approaches to government, characteristic of what Francis Fukuyama’s understood as the post-Cold War ‘End of History’. Given the very different historical contexts of socialization of the cohorts in this study, the expectation here is that if the political generations theory is correct, then the 1960s-70s Generation should be more likely than both previous (the PreWWII and Post-WWII) and subsequent (the 1980s and 1990s)

68

M.T. Grasso / Electoral Studies 33 (2014) 63–76

generations to engage in extra-institutional forms of participation (i.e., with SMOs, protest and petitioning) (H3). On the other hand, for the societal modernization account to be supported, each younger cohort should engage in extra-institutional activities more than each previous cohort (H4). So, for example, for extra-institutional participation, the 1990s Generation should be more likely than the 1980s Generation to engage in demonstrating, etc.), the 1980s Generation should be more likely than the 1960s–70s Generation to do so, and so forth.

up to include a random effect for time at the country-level to model this variability in the relationship between time and participation across countries. As specified above, allowing us to include these random effects at the country level is a distinct advantage of multi-level models. The logistic (since all outcomes are dichotomous) multi-level model, predicting each of the four types of participation, being fit to the data is: p

log 1ijpij ¼ g00 þ g1 ageij þ g2 age2 ij

3.2. Multi-level models

þ g3 1990ij þ g4 1999ij þ g5 2008ij

As discussed in the Introduction, this paper presents a method for studying age, period and cohort effects in a comparative context where repeated cross-sectional data are available covering a suitably long period of time so that members of the same cohorts are observed at different historical moments and in different phases of their life-time. More specifically, the method presented in this paper applies multi-level models with country as the higher level of analysis and random coefficients to model those variables which vary at the country level. This modelling strategy has the distinct advantage of accurately reflecting the fact that observations are nested within countries and also that not all variables have the same effects cross-nationally. This kind of approach is very useful for applications in a comparative context so as to correctly model the effects which vary between countries while at the same time recognising that there is some random variability at the country-level. This is a significant improvement on a cumbersome fixed effects approach with interactions which would estimate an inconvenient number of parameters and also importantly ignore the random variability at the country level. However, in order to decide how to precisely set up our multi-level models to accurately model the patterns obtaining across the different countries, it is important to run diagnostic country-by-country analyses first. These are necessary in order to decide how to best set up the models. The first and most basic type of such analysis was to run country-by-country logistic regression models for each dependent variable and observe the nature of the ageperiod-cohort effects cross-nationally.2 The results of these diagnostic models where results were statistically significant showed that the patterns of generational differences obtaining for each mode of participation were very similar across countries. This finding fits nicely with the theoretical premises of the study and also means, in terms of the set-up of the multi-level models, that random effects need not be applied for cohort at the country-level. Age effects were also very consistent between countries. However, results showed that the effect of time on participation varied between countries. In some countries time exhibited a linear effect on participation whereas in others it exhibited a curvilinear effect. As such, the multi-level models are set

þg6 Pre WWIIij þ g7 1960s 70sij þ g8 1980sij þ g9 1990sij

2 The results for these eighty regression models (ten countries, four dependent variables and two versions of the model, one with age and age-squared and one with categorical age groups to mimic the set up of the GAMs), are not presented here but are available upon request.

þgk Xij þd0j þ d1j 1990ij þ d2j 1999ij þ d3j 2008ij þ εij (1) In this equation,

g00 þ g1 ageij þ g2 age2 ij þ g3 1990ij þ g4 1999ij þ g5 2008ij þg6 Pre WWIIij þ g7 1960s 70sij þ g8 1980sij þ g9 1990sij þgk Xij represents the fixed effects, and

þd0j þ d1j 1990ij þ d2j 1999ij þ d3j 2008ij þ εij represents the random effects. The j subscripts indicate that parameters vary between countries, whereas the ij subscripts indicate that each variable varies between respondents and countries. g00 is the constant; all explanatory variables, with the exception of the survey year dummies, are included solely as fixed effects in the model. εij is the individual-level error, d0j represents the random variability associated with group-level characteristics, d1i 1990ij þ d2i 1999ij þ d3i 2008ij represent the random effects for the year of survey dummies. Fitting a random slope for survey year thus takes into account the variability in the relationship between time and participation between countries which was uncovered in the exploratory country-by-country analyses. gk Xij represents the controls. It is important to note that d0j ; dzj ; and εij are all separately independently normally distributed with mean zero and some variance parameter that is estimated. It is also worth noting at this stage that while generally a higher level N of at least 30 is ideal for multi-level models, in this context we cannot add further higher level units since the theory being tested is only applicable to Western European advanced industrial democracies and there was no repeated cross-sectional data available spanning 1981–2008 for other countries which conformed the requirements of the theory. The alternative, as outlined above, would have been to use fixed effects models with interactions by country to substitute for the random effects but these models would have erroneously ignored the random variability at the country level and also estimated a great deal more parameters making them cumbersome.

M.T. Grasso / Electoral Studies 33 (2014) 63–76

3.3. Generalized additive models (GAMs) and generalized additive mixed models (GAMMs) As just presented, both the diagnostic logistic regression models and the final multi-level models rely on splitting year of birth into a five category factor, where each level represents a different cohort grouping, to estimate the otherwise collinear effects of age-period-cohort. However, while this simplifying assumption is absolutely necessary to ‘break’ the linearity in the age-period-cohort relationship and estimate all three effects, it comes at the price of loss of information. Theory and a number of other studies in the literature suggest that these cohort categorizations are robust, however, this method means that people that are born within a given period are set to have equal cohort effects. On the other hand, Spitzer (1973: 1358) points out that there is always going to be a boundary problem of where to delineate social generations in the “seamless continuum of daily births” and that there is always unavoidable ambiguity in terms of where to apply the ‘cuts’. This problem becomes even more important if the cohort analysis is done in a comparative context. Cohorts of the same birth year might differ as they experienced different formative events in their respective home-countries. This paper takes this criticism of a priori theoretical categorization seriously, and unlike previous studies, the method presented here also provides a robust and novel empirical test of the categorization of cohorts, developed from theory, applied in this study. This is accomplished through the application of both generalized additive models (GAMs) and generalized additive mixed models (GAMMs). In order to check (in a similar fashion to what was accomplished in the diagnostic country-by-country logistic regression models) whether cohort effects are similar across countries, logistic generalized additive models are fit for each country:

log 1p ¼ a p

þ b1 age group 1 þ b2 age group 2 þ b3 1990 þ b4 1999 þ b5 2008 þs ðyear of birthÞ

(2)

þb k Xk þε In order to provide an empirical robustness check of the cohort categorization applied in the multi-level models with the random effects for time, I fit the following logistic generalized additive mixed model for each dependent variable: p

log 1ijpij ¼ g00 þ g1 age group 1ij þ g2 age group 2ij þ g3 1990ij þ g4 1999ij þ g5 2008ij þsðyear of birthÞ

(3)

þgk Xij

69

models year of birth (i.e., the identified cohort effect in this model) – replacing the categorical variable from the multilevel models – and allows us to plot the smoothed nonlinear cohort effect since year of birth is estimated as smoothly changing. There are different smoothing functions which could be applied; smoothing splines are used here.3 The rest of the model from Eq. 3 is exactly the same as the logistic multi-level model from Eq. 1 except that in this model age needs to be included as a three-level categorical variable to estimate it since year of birth is included as a continuous smoothed term (and not as the five-category variable from the logistic multi-level model in Eq. 1). (See Table 2 and Section 3.1 above for more information). The combined method, employing both logistic regression/logistic multi-level models and generalized additive models/generalized additive mixed models with random effects for time at the country level applied in this study thus allows for the disentangling and thus the analysis of age-period-cohort effects with repeated cross-sectional, cross-national, survey data covering the period 1981– 2008 (EVS, 2011) while also taking into account the variability in time effects between countries. Importantly, it also provides a means for checking the empirical accuracy of applying multi-level models which assume cohort effects will be similar across countries, by plotting the smoothed cohort effects by country with the GAMs and then also overcoming the need for categorizations of generations applied in the multi-level models, by plotting the smoothed cohort effects for all observations nested within their relative country with the GAMMs. 4. Data and variables 4.1. Data and case selection To apply this method for comparative age-periodcohort analysis and investigate whether generations coming of age in certain historic periods are more likely than other generations to engage in activities ascendant in that period, this paper employs 1981–2008 European Values Study data (EVS, 2011). This survey conveniently contains observations on both institutional (party membership) and extra-institutional (SMO participation, protest, and signing a petition) types of political participation in repeated cross-sections over a prolonged period of time. All respondents younger than 18 and those older than 90 were dropped from the sample so that all respondents would have had the opportunity for participation (very young people would not have had many opportunities to participate before 18 and very old people tend to have mobility issues). For the analyses, all cases that had missing values on any of our dependent or independent variables were also deleted so that samples are the same across all models. The data examined are for political participation in ten Western European democracies: Belgium, Denmark, France, West Germany, Great Britain, Ireland, Italy, the Netherlands,

þd0j þ d1j 1990ij þ d2j 1999ij þ d3j 2008ij þ εij The important component to note in Eqs. 2 and 3 above is the s (year of birth) term. This is the smooth function which

3 The software package selected the smoothing parameter by generalized cross-validation.

70

M.T. Grasso / Electoral Studies 33 (2014) 63–76

Spain and Sweden. These countries fulfil the requirements for testing theories of participation in advanced industrial democracies, and they are all included in all four waves of the EVS i.e. 1981, 1990, 1999, 2008 providing enough time points to estimate effects adequately. The countries all experienced similar socio-historic transformations4 and have similar political arrangements. The former in particular is crucial for justifying the assumption that the individuals in the same generation can be said to have experienced their formative years in relatively similar historical contexts across the different countries included in this study. Other countries which could have in theory fit the requirements for testing the theories were not included because of the unavailability of over time EVS data.

4.2. Dependent variables Since this paper looks to empirically investigate whether, as it is often simply assumed, the patterns of generational differences are similar for indicators such as petitioning and protest traditionally grouped in the same theoretical category (alternately called extra-institutional, elite-challenging, unconventional, etc. participation), here each indicator is investigated as a stand-alone dependent variable in the models and additive scales are not constructed for the extra-institutional repertoire. The distinction between the institutional and extrainstitutional dependent variables is based on Barnes and Kaase’s (1979) seminal work. Inglehart (1977, 1990) and Inglehart and Catterberg (2002) apply an almost identical distinction between ‘elite-directed’ and ‘elite-directing’ or ‘elite-challenging’ participation. The four dependent variables (party membership, demonstrating, signing a petition and participation in social movement organisations) were recoded as dichotomies (1 yes, 0 no) based on actual participation (response categories ‘have done’ for demonstrating and petitioning and ‘mentioned’ for party membership and SMO participation). This is a significant advantage over protest potential scales (based on the three possible survey responses to the demonstration and petition survey items from the EVS: 1 would not do, 2 might do, 3 have done) used in many other studies of extra-institutional participation as dependent variables. This is a major improvement, since using actual participation as a dependent variable avoids the logical conflation between participation with professed approval of participation in protest activities. Moreover, more educated, liberal and tolerant individuals (such as those in the youngest cohorts) are more likely to say that they would be prepared to engage in political activities without necessarily participating in practice and therefore avoiding protest potential scales as dependent variables allows us to better capture generational differences in actual participation as opposed to generational differences in positive attitudes to participation. The dependent variable for SMO participation is constructed from six EVS variables (membership or unpaid work

4 Spain, with its more recent history of democracy can perhaps be seen as the exception here, but diagnostic results discussed above showed that cohort and other effects except for time are remarkably similar to those for the other Western European countries examined here.

in environmental, ecology and conservation groups, animal rights organisations, and third world development organizations and human rights organisations). Four-wave data was not available for other types of SMO participation but these were the two most popular types of SMO activism in the data. 4.3. Control variables The models also include a number of controls which are standard in the literature on political participation and which capture important changes in social structure and valueorientations in advanced industrial societies in the period under investigation: gender (male), education, socioeconomic status, and political values. Education level is measured by the survey item that asks ‘at what age did you complete your education?’ since there are no other education variables for all four survey waves in the EVS. This is included in the models as a linear term. A dummy variable for manual occupation is also included. While a more fine-tuned class schema would have been ideal, this was the only comparable measure of class across the survey waves. A control for political values is also included since they changed in the 20th century and have an effect on engagement in participation in extra-institutional activities (Inglehart,1990). A four-category variable for left-libertarian (1), left-authoritarian (2), rightlibertarian (3), and right-authoritarian (4) values was constructed from two value scales based on various attitudinal questions on left-right economic values (redistribution, public vs private ownership, etc.) and libertarianauthoritarian social values (tolerance, gender equality, etc.).5 5. Results 5.1. Multi-level models Table 3 presents the results for the multi-level models of institutional (party membership) and extra-institutional (SMO participation, demonstrating and signing a petition) participation. For institutional participation, given the context of their socialization, we would have expected the Pre- and Post-WWII Generations to be the most engaged with parties but there was no expectation of significant differences between the 1960s–70s Generation and, respectively, the 1980s and 1990s Generations and between the 1980s and 1990s Generation (H1). On the other hand, for the societal modernization account to be supported, each younger cohort should engage in institutional activities less than each previous cohort (H2). Table 3 shows that the 1960s–70s Generation, 1980s, and also 1990s Generation are all significantly less likely than the Post-WWII cohort to engage with parties. So far, this evidence is consistent with both accounts. However, Wald-tests of differences between cohort coefficients6 show that there

5 Further details of the survey items used to construct these values variables are available from the author. 6 The models in Table 3 compare the coefficients for each cohort with that of the Post-WWII Generation. To check whether the 1990s Generation was significantly different from the 1980s Generation, the 1960s–70s Generation, and so on, one needs testing for coefficient differences via the Wald test.

M.T. Grasso / Electoral Studies 33 (2014) 63–76

71

Table 3 Multi-level parameter estimates for institutional and extra-institutional political participation, EVS 1981–2008. N ¼ 43,368, countries:10

Party membership

SMO participation

Attending a demonstration

Signing a petition

Fixed Effects

Estimate (s. e.)

Estimate (s. e.)

Estimate (s. e.)

Estimate (s.e.)

Constant Cohort Pre-WWII Generation Post-WWII Generation (reference category) 60s–70s Generation 80s Generation 90s Generation Age Age2 Survey year 1981 (reference category) 1990 1999 2008 Gender (male) Education (year completed) Manual occupation Political values Left-libertarian (reference category) Left-authoritarian Right-libertarian Right-authoritarian

5.15 (0.33)***

5.27 (0.25)***

2.74 (0.17)***

2.88 (0.16)***

0.18 (0.09)*

0.09 (0.07)

0.03 (0.06)

0.18 (0.09)* 0.52 (0.14)*** 0.76 (0.20)*** 0.04 (0.01)*** 0.00 (0.00)***

0.03 (0.07) 0.08 (0.11) 0.35 (0.15)* 0.03 (0.01)*** 0.00 (0.00)***

0.14 (0.05)* 0.16 (0.08)* 0.44 (0.12)*** 0.01 (0.01) 0.00 (0.00)

0.19 (0.05)*** 0.10 (0.07) 0.10 (0.09) 0.05 (0.01)*** 0.00 (0.00)***

0.11 (0.07) 0.27 (0.11)* 0.32 (0.14)* 0.66 (0.05)*** 0.08 (0.01)*** 0.46 (0.05)***

0.65 (0.06)*** 0.71 (0.09)*** 0.70 (0.11)*** 0.10 (0.03)** 0.12 (0.01)*** 0.37 (0.04)***

0.52 (0.05)*** 0.91 (0.06)*** 0.89 (0.09)*** 0.52 (0.02)*** 0.10 (0.003)*** 0.21 (0.03)***

0.42 (0.04)*** 0.83 (0.05)*** 0.73 (0.07)*** 0.13 (0.02)*** 0.10 (0.003)*** 0.20 (0.03)***

0.33 (0.07)*** 0.21 (0.06)*** 0.15 (0.07)*

0.47 (0.05)*** 0.18 (0.04)*** 0.46 (0.05)***

0.63 (0.03)*** 0.79 (0.03)*** 0.88 (0.04)***

0.57 (0.03)*** 0.24 (0.03)*** 0.51 (0.03)***

Variance (std. d.)

Variance (std. d.)

Variance (std. d.)

Variance (std. d.)

0.0006 (0.025)

0.0006 (0.025)

0.0006 (0.025)

0.0006 (0.025)

0.0021 (0.046) 0.0025 (0.050) 0.0027 (0.052)

0.0021 (0.046) 0.0025 (0.050) 0.0027 (0.052)

0.0021 (0.046) 0.0025 (0.050) 0.0027 (0.052)

0.0021 (0.046) 0.0025 (0.050) 0.0027 (0.052)

Random Effects

0.01 (0.11)

Between-country variance Survey year 1981 (reference category) 1989 1999 2008 AIC BIC LogLikelihood Deviance

17454 17680 8701 17402

27770 27996 13859 27718

44428 44654 22188 44376

54145 54371 27047 54093

Significance levels: *p  0.05; **p  0.01; ***p  0.001.

are no significant differences between the Pre-WWII Generation and the 1960s–70s Generation (p ¼ 0.258). There are, however, significant differences between the Pre-WWII Generation and, respectively, both the 1980s Generation (p ¼ 0.014) and the 1990s Generation (p ¼ 0.006). Moreover, there are significant differences in party membership between the 1960s–70s Generation and the 1980s Generation (p ¼ 0.0001), between the 1960s–70s Generation and the 1990s Generation (p ¼ 0.0001) and between the 1980s and 1990s Generations (p ¼ 0.038). While there are no differences between the Pre-WWII and 1960s–70s Generations; the 1980s and 1990s Generation are both significantly less likely to support parties than the 1960s–70s Generation. The evidence presented in Table 3 and discussed above for institutional participation can thus be seen to partially support the claims of the societal modernization account. However, this piece of evidence alone does not necessarily mean that the mechanism suggested by the societal modernization is correct since there are other competing explanations in the literature for why the generations coming of age since the sweeping transformations of the late 1960s should be less likely than previous generations to engage with parties and electoral politics (Blais et al.,

2004; Fieldhouse et al., 2007; Franklin, 2004; Plutzer, 2002; Putnam, 2000; Heath and Taylor, 1999; Mair, 2006; Mair and Van Biezen, 2001). For example, the convergence between major parties and coalitions on the centre of the political spectrum, the increasing professionalisation and managerialism of party elites in contrast to the bottomup grass-roots approach of the mass parties of the first half of the 20th century and the fact that parties do not provide adequate representation for those issues which concern newer generations since they see them as a lost constituency, are just some options for alternative explanations for the observed empirical patterns. Future research should explore whether this effect is also present for other types of institutional participation since the mechanism hypothesized by the societal modernization account relates to all institutional activities. Moreover, it might be that social structural transformations affecting socialization into party allegiances have further deteriorated from the 1960s, to the 1980s, to the 1990s. Turning to extra-institutional participation, the expectation here was that if the political generations theory is correct, then the 1960s–70s Generation should be more likely than both previous and subsequent generations to engage in extra-institutional forms of participation (i.e.,

72

M.T. Grasso / Electoral Studies 33 (2014) 63–76

SMO participation, protest and signing a petition) (H3). On the other hand, for the societal modernization account to be supported by evidence, each younger cohort should engage in extra-institutional activities more than each previous cohort (H4). The evidence for participation in social movement organisations presented in Table 3 shows that the 1990s Generation is significantly less likely than the, older, PostWWII Generation to engage in this activity. Additionally, there are no significant differences between the PostWWII Generation and both the 1960s–70s and 1980s Generation. Wald-tests of differences between cohort coefficients show that while there are no significant differences between the 1960s–70s and 1980s Generation (p ¼ 0.076), the 1960s–70s Generation is significantly more likely than the 1990s Generation to engage in this extra-institutional activity (p ¼ 0.0002); additionally, the 1980s Generation is also significantly more likely than the 1990s Generation to participate in SMOs (p ¼ 0.0001). Therefore, these results show that younger cohorts are not more likely than older cohorts to engage in all types of extra-institutional activities as suggested by the societal modernization account. Rather, this evidence supports the political generations theory: the 1960s–70s Generation coming of age in the radicalising period exemplified by the ’68 student revolts across Western Europe and around the globe, are more likely even than more recent generations (in this case, the generation coming of age in the 1990s) to engage with social movements. It is worth noting here that these results are also in contrast with the ‘less strict’ version of the societal modernization account discussed above. If fluctuations between generations were down to economic conditions at the time of socialization we would expect to see the 1980s Generation as significantly less participatory than both the 1960s and 1990s Generation. However, this is not the case, suggesting that the dramatically different political context of the socialization of the 1990s Generation with the end of the Cold War, the decline of the Left, and the other developments discussed above may hold more promise for explaining why they are less likely to engage with SMOs than both the 1960–70s and 1980s Generations. As for the other two extra-institutional modes of participation, the results for attending a demonstration presented in Table 3 show that while the 1960s–70s Generation is significantly more likely than the Post-WII Generation to engage in this activity, both the 1980s and 1990s Generations are significantly less protest prone than the much older Post-WWII generation. It is thus also evident that the 1960s–70s Generation are significantly more likely than both the 1980s and 1990s Generations to demonstrate. Again, this evidence runs counter to the expectations of the societal modernization account. The fact that the 1960s–70s Generation are more likely than younger cohorts to engage in demonstrations rather supports the political generations theory presented in this study, following Mannheim’s (1928) seminal work on political generations in general. Again, this evidence also runs counter to the ‘less strict’ version of the societal modernization account since despite more favourable economic conditions during socialization, the 1990s Generation is

just as little protest prone as the 1980s Generation, relative to the 1960s-70s Generation. The models presented in Table 3 also show that the 1960s–70s Generation, following the expectations of the political generations theory, is significantly more likely than the Post-WWII Generation to sign a petition. However, running against the societal modernization account, there are no significant differences between the Post-WWII Generation and the 1980s and 1990s Generations for petitioning. Waldtests of differences between cohort coefficients reveal that, as with demonstrating, the 1960s–70s Generation are also significantly more likely than both the 1980s (p ¼ 0.024) and the 1990s Generation (p ¼ 0.0000) to sign a petition. Once again, this evidence runs counter to the societal modernization account and provides further evidence for the political generations theory of participation: the 1960s– 70s Generation coming of age in a more radical and ideologically-polarised period are more likely to engage in extra-institutional, protest activities - even more moderate ones, such as petitioning - than younger cohorts. Once again, these results also run counter the ‘less strict’ version of the societal modernization account since economic conditions during socialization do not account for the fluctuations between generations. 5.2. Generalized additive models (GAMs) and generalized additive mixed models (GAMMs) As discussed in the methods section, the method presented here also allows for robust and novel empirical tests of results presented in the multi-level models by estimating the effect of year of birth as smoothly changing through the application of both generalized additive models (GAMs) and generalized additive mixed models (GAMMs). I focus on the latter results here since the country-by-country GAMs were mainly run as diagnostics to check that cohort effects were similar across countries, which had already been shown through the country-bycountry logistic regression models developed to determine the correct modelling strategy for the multi-level models and suggesting the need for the inclusion of random effects for time but not cohort. As an illustrative example, Fig. 1 presents the country-by country GAMs plotting the smoothed cohort effects for demonstrating.7 As can be seen, and also taking into consideration the confidence intervals, the shape of the cohort effects is similar enough across the countries, including Spain, which was theoretically more troubling given its more recent democratization. In order to provide empirical tests of the categorization of cohort applied in the multi-level models with the random effects for time and check that the results were not biased by this five-factor schema necessary to estimate the otherwise collinear effects of age-period-cohort, generalized additive mixed models (GAMMs) also including random effects for time at the country-level were applied. Figs. 2–5 plot the smoothed cohort effects from these

7 Country-by-country GAMs for the other three dependent variables are available from the author upon request.

M.T. Grasso / Electoral Studies 33 (2014) 63–76

73

Fig. 1. Attending a demonstration by country, EVS 1981–2008. Smoothed cohort effect (GAMs).

models for each of the four indicators of participation for all the observations nested in their respective country. Fig. 2 plots the smoothed cohort effect on party membership. It can be seen that the curve slopes down steeply around the mid-1940s, which is just about around the time when the first members of the 1960s–70s Generation would have been born, based on the cohort classification schema applied in this study (i.e., in 1947, see Table 1 for more details). This robustness check thus supports the results from the random effects multi-level models for party membership presented in Table 3, and provides evidence that regardless of whether one categorizes cohorts or applies GAMMs to overcome the need for categorizations and to plot the

smoothed cohort effect, the evidence points to the 1960s– 70s Generation, and more so, the 1980s and 1990s Generations as being less likely than the Post-WWII Generation cohort to be members of a party. Additionally, also supporting the multi-level model results for party membership discussed previously, Fig. 2 shows that the curve keeps sloping down over the years of birth of the 1980s Generation (1958–1968) and those of 1990s Generation (1969– 1981). Moving on to the indicators of extra-institutional participation, Fig. 3 plots the smoothed cohort effect for SMO participation; it shows that the curve starts sloping down during the 1940s which is just before when the 1960–70s Generation would have been born. Supporting

74

M.T. Grasso / Electoral Studies 33 (2014) 63–76

Fig. 2. Party membership EVS 1981–2008. Smoothed cohort effect (GAMM).

Fig. 4. Attending a demonstration EVS 1981–2008. Smoothed cohort effect (GAMM).

Fig. 3. SMO participation EVS 1981–2008. Smoothed cohort effect (GAMM). Fig. 5. Signing a petition EVS 1981–2008. Smoothed cohort effect (GAMM).

the results in Table 3 the curve keeps sloping down steeply, also over the years of birth of 1980s and 1990s Generation. Therefore, the results once again show that despite the fact that SMO membership is often theoretically categorized as belonging to the extra-institutional repertoire, the empirical generational patterns for this indicator are rather more similar to those for party membership – an activity which is clearly part of the institutional repertoire. Perhaps the issue with these results is the fact that, due to over-time data limitations, the dependent variable only captures membership work for environmental, ecology, conservation, animal rights, and development organisations and that patterns for generational differences would have been more similar to those for extra-institutional activities had data been available for other types of SMO participation across the four waves. Alternatively, it might be that older generations are generally more likely to be involved in organisations – regardless of whether they are “old” or “new”. Perhaps the increasing emphasis on individual responsibility as opposed to political solutions to social problems meant that younger generations were less likely to see organisational participation as an effective means for social change. Another option could be that higher levels of party membership amongst older generations means that they are more easily mobilised, through the networks of

Left, social-democratic and Green parties, also to join other organisations such as SMOs (Passy and Giugni, 2001). Finally, examining the GAMM results for protest activism, Fig. 4 plots the smoothed cohort effect for demonstrating; Fig. 5 that for signing a petition. They both show that the peaks map on quite neatly to the years of birth of the 1960s– 70s Generation and that younger cohorts are generally more likely than older cohorts to engage in protest activism, but not to the same extent as the 1960s–70s Generation. For both forms of protest action, the 1960–70s Generation stands out as the most participatory, confirming once again the expectations of the political generations theory. To conclude the analysis of results, the application of generalized additive models (GAMs) provided a useful means to visualise whether cohort effects were similar across countries for the different modes of participation justifying, along with the country-by-country diagnostic logistic regression models, the set-up of the multi-level models with random effects for time but not cohort. Similarly, the application of generalized additive mixed models (GAMMs) to plot the smoothed cohort effects provided a useful empirical check of the results from the multi-level models showing that they were not biased by the application of the five-level factor for cohort applied in this study:

M.T. Grasso / Electoral Studies 33 (2014) 63–76

examining the smoothed cohort effect curves plotted from the GAMMs supported the multi-level modelling results. 6. Discussion and conclusions Using data from the European Values Study 1981–2008 (EVS, 2011), this paper analysed generational differences in institutional and extra-institutional political participation in ten advanced Western European democracies. In doing so, this paper presented a method for age, period and cohort analysis with repeated cross-sectional, cross-national, survey data covering a suitably long period of time so individuals from the same cohort are observed in different periods and in different stages of the life-course and where generations can be categorized based on theory. The application of multi-level models with random effects allowed to model those variables which through diagnostic country-by-country logistic regressions were found to vary in their effect between countries – in this case, time. These multi-level models relied on the inclusion of cohort as a five-category factor, along with period as survey year dummies, and both linear and squared age terms, in order to estimate all three effects simultaneously and ‘break’ the linearity of the age-period-cohort relationship to analyse generational differences. While the categorization of cohorts applied in this study was rooted in theory and previous research, it nonetheless represented a loss of information and potential source of bias since the ‘cuts’ applied to distinguish between cohorts in any categorization will always be arbitrary to an extent. To deal with this problem common to all research applying cohort categorizations to break the linearity of age-period-cohort, the method applied here also provided an empirical test of cohort categorizations applied in this and previous studies by plotting smoothed cohort effects. This was accomplished both through the GAMs, as a diagnostic to support the country-by-country logistic regression results and therefore the set-up of the multi-level models and also, more importantly, through the GAMMs, to check the results of the overall multi-level models since the GAMMs also allowed to include the random effects for time at the country-level. In this way, both the GAMs and the GAMMs showed that the results for generational differences in participation were robust and that they were not simply the artefact of potentially arbitrary cut-off points for the distinct historic eras in which to have located the divergent formative experiences of our five generations. Substantively, the results reported in this study showed that the evidence for party membership provided some support for the societal modernization account. However, to provide conclusive evidence that generational differences in participation support this account future research needs to assess whether this effect is also present for other types of institutional participation since the mechanism hypothesized by the societal modernization account demands that each younger generation should be less likely than the previous one to participate in all types of institutional participation. If this is only true of party membership but not campaign activities or contacting political representatives, for example, then other explanations for generational

75

differences pertaining specifically to party membership may be more adequate to account for why we observe the empirical patterns presented in this study. On the other hand, the evidence for the three types of extra-institutional participation (SMO participation, demonstrating and signing a petition) was overwhelmingly supportive of the political generations theory of participation developed in this study and inspired by Mannheim’s (1928) classic work on political generations in general. The evidence presented here for SMO participation, participation in protests and also for petition signing shows that the 1960s–70s Generation is more participatory than the 1980s and also the 1990s Generations in three modes of action which according to the societal modernization account should be practiced increasingly more by each younger generation. The evidence presented here for extra-institutional participation thus undermines the expectations of both versions of the societal modernization account. The more favourable economic conditions, relative to those of the 1980s Generation, at the time of the socialization of the 1990s Generation did not lead them to participate more than this generation. The 1960s–70s Generation were more likely than both the 1980s and 1990s Generations to demonstrate and petition and more likely than the 1990s Generation to participate in social movement organisations (SMOs). Given this, it appears that the more radical and ideologically-polarised context of the political socialization of the 1960s–70s Generation meant that they are more likely to engage in ‘unconventional’ or ‘new’ activities even than younger generations. This striking finding supports Mannheim’s (1928: 226) crucial point from his seminal essay on The Problem of Generations that: “nothing is more false than the usual assumption uncritically shared by most students of generations, that the younger generation is progressive and the older generation eo ipso conservative”. In other words, the evidence presented here does not support the idea of an ongoing inter-generational shift from institutional, or ‘elite-directed’, to extra-institutional, or ‘elite-challenging’, participation as suggested by the societal modernization account. The results of this study have shown that the extent to which members of a generation will engage in politics – through whatever repertoires of action available to them – and demand political change of the world forged by previous historical battles, will depend in large part on the dynamics and political characteristics of the era in which they came of age and experienced political socialization. It could be considered that the 1980s and 1990s Generations have not matched the levels of political engagement of the 1960–70s Generation since they were able, in their youth, to enjoy many of the benefits emerging from the social and political battles which the 1960s–70s Generation fought and won. However, today, with a deteriorating economy, youth unemployment reaching 40 per cent in several Western European countries, the growing curtailment of civil liberties, and the expansion of surveillance in the name of security, where even employment means nonetheless precarious conditions and fewer safety nets or guarantees, to name only a few recent concerns, a new generation of activists

76

M.T. Grasso / Electoral Studies 33 (2014) 63–76

engaging politically to over-come these contemporary challenges may soon re-emerge.

Acknowledgements I am indebted to Anja Neundorf and Richard Niemi for their comments on earlier versions of this paper. I am also grateful to the anonymous reviewer, the participants of the “Beyond Political Socialization: New Approaches in Age, Period, Cohort Analysis” workshop, Nuffield College, University of Oxford, March 16–17, 2012 and to Stephen Fisher, Robert Andersen, David Armstrong, Geoffrey Evans, Mark Franklin, Marco Giugni, Anthony Heath, Chaeyoon Lim, Peter Mair, James Tilley, and Andries van den Broek for feedback on previous versions of this research; to the UK Economic and Social Research Council (ESRC) for funding my graduate studies at the University of Oxford and to the European Values Study (EVS) for providing the data. Any errors that remain are entirely my own.

References Andersen, R., Fetner, T., 2008. Cohort differences in tolerance of homosexuality: attitudinal change in Canada and the United States, 1981– 2000. Public Opinion Quarterly 1 (0), 1–20. Barnes, S., Kaase, M., 1979. Political Action: Mass Participation in Five Western Societies. Sage, Thousand Oaks, California. Bartels, L., Jackman, S., 2014. A Generational Model of Political Learning. Electoral Studies 33, 7–18. Becker, H.A., 1990. Dynamics of life histories and generations research. In: Becker, H.A. (Ed.), Life Histories and Generations. ISOR, Utrecht. Becker, H.A., 1992. A pattern of generations and its consequences. In: Becker, H.A. (Ed.), Dynamics of Cohorts and Generations Research. Thesis Publishers, Amsterdam. Blais, A., Gidengil, E., Nevitte, N., Nadeau, R., 2004. Where does turnout decline come from? European Journal of Political Research 43, 221–236. Clarke, H., Sanders, D., Stewart, M., Whiteley, P., 2004. Political Choice in Britain. Oxford University Press, Oxford. Della Porta, D., Diani, M., 2006. Social Movements: an Introduction, second ed. Blackwell, Oxford. Evans, G., De Graaf, N.D., 1996. Why are the young more postmaterialist? A cross-national analysis of individual and contextual influences on postmaterial values. Comparative Political Studies 28 (4), 608–635. EVS, 2011. European values study 1981–2008, longitudinal data file. In: Archive, G.D. (Ed.), GESIS Data Archive. Germany, Cologne. Fieldhouse, E., Tranmer, M., Russell, A., 2007. Something about young people or something about elections? Electoral participation of young people in Europe: evidence from a multilevel analysis of the European Social Survey. European Journal of Political Research 46, 797–822. Franklin, M.N., 2004. Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945. Cambridge University Press, Cambridge. Furedi, F., 2004. Reflections on some uncomfortable realities. In: Todd, M., Taylor, G. (Eds.), Democracy and Participation: Popular Protest and New Social Movements. Merlin Press: viii–xviii., London. Glenn, N.D., 1976. Cohort analysts’ futile quest: statistical attempts to separate age, period and cohort effects. American Sociological Review 41 (5), 900–904. Grasso, M.T., 2011. Political Participation in Western Europe. Nuffield College, University of Oxford. DPhil.

Grasso, M.T., 2013a. The differential impact of education on young people’s political activism: comparing Italy and the United Kingdom. Comparative Sociology 12, 1–30. Grasso, M.T., 2013b. What are the Distinguishing Generational, Life Course, and Historical Drivers of Changing Identities? Future Identities: Changing Identities in the UK – the Next 10 years. Foresight: the Future of Identity in the UK. The Government Office for Science, London. Heath, A., Park, A., 1997. Thatcher’s children? In: Jowell, R. (Ed.), British Social Attitudes: the end of Conservative values? (the 14th Report). Sage, London, pp. 1–22. Heath, A., Taylor, B., 1999. New sources of abstention? In: Evans, G., Norris, P. (Eds.), Critical Elections: British Parties and Elections in Long-term Perspective. Sage, London. Inglehart, R., 1977. The Silent Revolution: Changing Values and Political Styles among Western Publics. Princeton University Press, Princeton. Inglehart, R., 1990. Culture Shift in Advanced Industrial Society. Princeton University Press, Princeton. Inglehart, R., Catterberg, G., 2002. Trends in political action: the developmental trend and the post-honeymoon decline. International Journal of Comparative Sociology 43, 300–316. Jennings, M.K., 1987. Residues of a movement: the aging of the American protest generation. American Political Science Review 81 (2), 367–382. Jennings, M.K., Niemi, R.G., 1991. The Youth-parent Socialisation Panel Study. Princeton University Press, Princeton. Katz, R., 1997. Democracy and Elections. Oxford University Press, Oxford. Lipset, S., Rokkan, S., 1967. Party Systems and Voter Alignments: Crossnational Perspectives. Collier-MacMillan, London. Mair, P., 2006. Ruling the void? The hollowing of Western democracy. New Left Review 42, 25–51. Mair, P., Van Biezen, I., 2001. Party membership in twenty European democracies, 1980–2000. Party Politics 7 (1), 5–21. Mannheim, K., 1928. The Problem of Generations Essays on the Sociology of Knowlegde. Routledge, London. Mattei, F., Niemi, R.G., Powell, G.B., 1990. On the depth and persistence of generational change: evidence from Italy. Comparative Political Studies 23 (3), 334–354. Neundorf, A., Niemi, R.G., 2014. Beyond political socialization: New approaches to age, period, cohort analysis. Electoral Studies 33, 1–6. Passy, F., Giugni, M., 2001. “Social Networks and Individual Perceptions: Explaining Differential Participation in Social Movements.” Sociological Forum 16 (1). Plutzer, E., 2002. Becoming a habitual voter: inertia, resources, and growth in young adulthood. American Political Science Review 96, 41–56. Putnam, R., 2000. Bowling Alone: the Collapse and Revival of American Community. Simon & Schuster, New York. Rosow, I., 1978. What is a cohort and why? Human Development 21, 65–75. Snijders, T., Bosker, R., 1999. Multilevel Analysis: an Introduction to Basic and Advanced Multilevel Modelling. Sage, London. Spitzer, A.B., 1973. The historical problem of generations. American Historical Review 78, 1353–1385. Tilley, J., 2002. Political generations and partisanship in the UK, 1964– 1997. Journal of the Royal Statistical Society, Series A (Statistics in Society) 165 (1), 121–135. Tilley, J., Evans, G., 2014. Ageing and generational effects on vote choice: combining cross-sectional and panel data to estimate APC effects. Electoral Studies 33, 19–27. Van Aelst, P., Walgrave, S., 2001. Who is that (wo)man in the street? From the normalisation of protest to the normalisation of the protester. European Journal of Political Research 39, 461–486. Van Den Broek, A., 1995. Cultural change: the impact of cohort replacement and the absence of generations. In: Workshop on "Generational Change in Post-War Europe", Second Conference of the European Sociological Association, Budapest, August 30–September 2, 1995. Van Deth, J.W., Elff, M., 2000. Political Involvement and Apathy in Europe 1973–1998. Mannheimer Zentrum für Europäische Sozialforschung, pp. 1–47. Working Paper.