Away from Political Parties into Lifestyle Politics: Young People in Advanced Democracies

In an era of dramatic decline in youth electoral participation (Blais and Rubenson 2013; Klingemann 2014), it is particularly surprising that better-educated and better-informed younger generations in long-standing democracies are disengaging from traditional democratic practices such as voting and aligning with a political party. We know that young people engage differently in politics, preferring ad-hoc, issue-based, elite-challenging forms of participation (Norris 2003; Sloam 2016). Do young people participate differently because societal transformations have created differential generational characteristics? If so, are these lasting characteristics, or are a sudden change in political behaviour that is particular to a cohort and fades away in subsequent cohorts?

Most explanations follow some version of the generational replacement phenomenon, which capitalize on earlier works (Dalton, 2008; Inglehart, 1977; Klingemann, 2014; Norris, 2011): younger cohorts with distinct characteristics replace their older counterparts in the electorate. The most prominent and well-tested theory, the social modernisation account, emphasizes a gradual increase in generational gaps due to subsequent value change. Younger cohorts are more economically secure – meaning they are more educated, politically sophisticated, socially independent, critical citizens who have post-materialist values compared to survivalist values. Citizenship, to them, is a right and not a duty (Dalton, 2007). Hence, abstention is less stigmatizing. Abstention occurs because these cohorts, by virtue of their enhanced cognitive resources, are more critical of the workings of their government (Dalton, 2007; Ferrin & Kriesi, 2016; Norris, 1999). Their increasing share in the electorate will progressively reduce turnout and partisan alignment.

A key idea that parallels this theory is the cognitive mobilisation (CM) thesis, which explains the diminishing role of political parties as mobilising agents. According to Inglehart (1970:47), cognitive mobilisation is the process where formal education “increases the individual’s capacity to receive and interpret messages”. Cognitive mobilisation in advanced industrial democracies has led to the increase in the number of apartisans (Dalton, 1984, 2007) – sophisticated individuals with no party ties; who have enough cognitive resources (from formal education) to grapple with the complexities of politics. Despite their limited electoral experience, cognitive mobilisation is expected to be higher in the younger generations. The cognitive mobilisation thesis indicates a generational component (Dalton, 1984, p.286): younger citizens have higher education levels compared to their elders and therefore can better engage with political information (more easily available through mass media; Baker et al., 1981).

But how does the cognitive mobilisation thesis stand today? Donovan (2018) questions the temporal and dynamic manifestations of cognitive mobilization. In 2022, for example, should we still expect these social forces such as rising levels of education and changes in mass media to have the effects they were said to be having in the 1960s? If social forces have operated since then, would we not expect expanded access to education to have neutralised differences in cognitive mobilization across age cohorts at some point? Related to this, have some applications of the cognitive mobilisation thesis mistaken a lifecycle effect (in any era, young people vote less and protest more) for generational change? How does the thesis stand today given the new, fragmented media environment with self-selection of news (Lodge and Taylor 2000; Nyhan and Reifer 2014) and news avoidance (Prior 2005)?

All said, voters in advanced democracies are different today – the world has changed. A larger proportion of citizens have high educational attainment, and access to mass media is fundamentally different. Parties play a different role. This article reviews a cognitively sophisticated electorate as an explanation for decline in electoral turnout and rise in partisan dealignment. Using survey data in the Comparative Study of Electoral Systems (CSES) and multi-level models to control for competing lifecycle and period effects, it shows that societal modernisation is a long evolutionary process, creating electorates which increasingly mobilise cognitively across generations. Overall, this article provides a clear and robust understanding of how exactly electorates are changing in advanced democracies. It also underscores the need for political parties to look for new ways to reconnect with cognitive apartisans who do not necessarily need them.

What is a generation?

A generation or birth cohort refers to a time span representing social change. Troll’s (1970) idea Zeitgeist is similar to Mannheim’s (1927; 1959) generational unit where members develop distinctive world views during late adolescence and early adulthood (typically between the ages 18-27). There are two premises embedded in this argument (García-Albacete 2014). First, young people are more susceptible to their political context and influenced by societal transformations during their formative years. So, if you socialised in an era where women in the workforce is a norm rather than an exception, then you will demand more women’s rights, and be sensitive to violations of such rights.

Second, values, orientations and attitudes formed during this time will persist over the course of one’s life. Persistence refers to the tendency to structure inputs, as well as reject dissonant items, based on previous cognitive design (Ryder, 1965: 856). If youth participation in traditional forms is declining due to unique cohort characteristics, then we can expect a long-term change in the participation of citizens in advanced democracies.

To understand the concept of generation, it is important to understand what it is not. A generational effect is different to a lifecycle effect that manifests during the developmental stage – for instance, the phenomenon that political involvement is not the same during youth and adulthood. Interest and involvement with politics increases with the accumulation of resources as one approaches middle-age and then decreases again with certain events like retirement.

For a generation effect, the socio-political environment is different from events that give rise to period effects which impact the population’s political engagement in general, regardless of age. But these events can leave stronger impacts on those in their formative years than others. Therefore, these manifest as long-term consequences for newer cohorts and not their older counterparts. Similarly, older events may have impacted older cohorts in a way that does not impact newer cohorts which did not socialise under those old events.

This brings me to my next crucial point: these time effects present the age-period-cohort (APC) identification problem where they are exact linear functions (Period – Age = Cohort) (Yang & Land 2006, 2008). In short, when we find evidence for one of these three, we cannot be confident of which of the three is driving the effect. This means a robust analysis would attempt to estimate the unique effect of one while controlling for the other two.  Although existing studies provide theoretically robust explanations for youth political behaviour, there is a lack of methodologically rigorous enquiries on which of the three time effects – age, period or cohort (APC) – drives youth engagement.

Cognitive mobilisation: a continuous process?

Social transformations such as the increase in education levels, development of new technologies and the rise of economic wellbeing have given rise to higher levels of cognitive engagement among younger cohorts. This has also paralleled decline in traditional political participation. Dalton (1984, 2007, 2012) and Norris’ (1999a) answer to the paradox implies that citizens in advanced democracies possess skills and resources needed to politically engage and do not need traditional institutions like parties to provide strong cues. This process of cognitive mobilisation has increased apartisans – sophisticated citizens who pay more attention to the performance of the government and react to it. This idea of ‘critical citizens’ is applicable more to younger cohorts who have reached higher levels of education and are more familiar with new technological tools. Therefore, the modernisation process results in an increase in ‘elite-challenging’, at the expense of ‘elite-directed’, political participation. (Inglehart, 1990; Inglehart and Welzel, 2005).

How much do key assumptions about CM carry forward over time in the established democracies? Dalton (1984; 2008) has long maintained that cognitive mobilisation is a constant (continuous) process where forces of change continue to disrupt especially younger age cohorts. But did he then refer to social transformations as a long term evolutionary societal change continually concentrating postmaterialist values in the electorate? Are these linear developments different from period effects brought about by major historical events, which are country-specific impacts on generations and thus are separate from societal transformations that could have affected several countries?

Donovan (2018) reasons there are three possible scenarios in which CM manifests. Under the first scenario, there was a finite period of societal transformations after WWII, but this has ran its course, producing contemporary, better-educated and more-interested cohorts. Donovan refers to this as the ceiling effect, meaning there were limits to how many people experienced the major transformative political effects of education and innovations in media. There may have been a phase shift which was limited to the mid-20th century, where local prints and radio mediums provided wide and novel access to global perspectives. For example, a person’s access to information about the Vietnam War may have been fundamentally different from a similar person’s access to information about WWII, but not all too different (or of lower quality) compared to information about the Iraq War.

The second scenario is that an ongoing, continuous process of political transformation where increasing proportion of citizens in the established democracies continue to become more educated and more interested in politics. Indeed, in his recent work, Dalton refers to “a changing public” (Dalton 2013:29). It is a dynamic process where the “the need for [partisan] cues declines as the political skills of the voters increase and information costs decrease,” and where “the dramatic spread of education” is leading to an expansion of political sophistication (Dalton 1984:265). This process may have started at the post-WWII time when very few people had access to quality secondary or tertiary education. This changed over time where increasingly larger proportions of a country’s population are experiencing more education. This means that it could take several generations, or centuries, for most or all citizens to reach higher levels of cognitive sophistication that results from increased access to education.

The changes in mass media can also be seen as a continuing process. In the 1950s, print newspapers and radio were replaced by broadcast televisions as a lower-cost medium to access political information. By the 1990s cable televisions and global satellite technologies outperformed broadcast television in scope, quantity, immediacy and cost-effectiveness of information. What further transformed and broadened access to political information is the growth of mobile devices, Internet and social media in the 21st century. Undoubtedly, social networking sites like Facebook, Twitter, Instagram and TikTok have changed the modern political campaigning scene. New media was instrumental in destabilising authoritarian regimes during the Arab Spring. As a continuing transformation, advents in technologies are increasing the proportion of people who are interested and involved with politics.

Finally, under the third scenario, CM is a process unique to a particular generation, where some point in the past young people may have become less partisan but more interested in politics, but those generational differences are no more evident. Donovan (2018) argues that whereas the first two scenarios reflect broad social forces affecting many countries simultaneously in the same manner, unique cohort effect scenarios could be more idiosyncratic. He uses the US example: social and political changes during the post-WWII through to the 1960s period shaped a generation with high distrust of government. This was precipitated by the Vietnam War and the Watergate scandal, experienced by the first wave of first-generation university students coming of age at that time. Or there may have been influence of forces such as a short-term sorting in the two-party system precipitated by the civil rights era. Consequently, the proportion of highly interested (younger) might increase – but as this generation ages and is replaced, the changes that might be attributed to cognitive mobilisation could decay over time.

The key question really is whether greater formal education, and greater use of mass media after the 1950s left advanced democracies with new batches of high cognition apartisans and cognitive partisans – where countries quickly reached a fairly static new equilibrium mix of traditional partisans, cognitive partisans, and apartisans. Or are electorates, over time, increasingly defined as being cognitively mobilized such that traditional partisans are gradually being replaced by apartisans? Donovan’s (2018) crude, albeit commendable, analysis doesn’t find evidence for CM be a continuous process in Australia and America, looking into partisanship and political interest, but it does not control for confounding lifecycle and period effects and does not look into a variety of advanced democracies. There is still, therefore, a need to test proposed hypothesis: Each subsequent cohort engage less with traditional, elite-directed practices such a casting a ballot compared to previous cohorts. If there is a gradual decline across generations, then CM is a continuous process rather than a one-off generational feature.

Analysing survey data

Here, I use post-election survey data from the Comparative Study of Electoral Systems (CSES) Integrated Module Database (IMD) between 1996-2016. For the purposes of this study on advanced industrialised democracies, I subset the dataset to include respondents from thirty-five Organisation for Economic Co-operation and Development (OECD) countries, who have experienced similar socio-historic transformations and political arrangements. This is important to assume that individuals from the same generations underwent similar experiences in their formative years across the different countries.

I use two indicators – turnout and party identification – as measures for traditional engagement with politics. The two survey questions are: Did you cast a ballot? and Do you feel close to a political party? They both are dichotomous variables with yes/no responses. The turnout variable reports “yes” when the respondent casts a vote in any of the following elections: main election, presidential elections in round 1 or 2 in the survey year, elections in the lower or the upper house. The party identification variable reports “yes” to any of the following questions: Are you close to any political party? or Do you feel closer to one party?

The main independent variables of interest are the three features of time progress. In line with previous research on generational trends, these are the age[1], period and cohort effects (Dassonneville, 2013; Grasso, 2014; Smets & Neundorf, 2014). I operationalise the generation variable by transforming the continuous year of birth/age variable into a five-category cohort variable, comprising: Post-WWII generation (birth year: 1926-1945, era: 1946-1965); 60s-70s generation (birth year:1946-1957, era: 1966-1977); 80s generation (birth year: 1958-1968, era: 1978-1988); 90s generation (birth year:1969-1979, era: 1989-1999); and, 00s generation (birth year:1980-1998, era: 2000-2008/16).

Individual level engagement is affected by other socio-demographic factors. Thereby, I also include individual-level socio-economic controls (Grasso, 2013; Grasso and Guigni 2022), namely gender (Henn and Foard, 2014), household income and education (Solt, 2008; Verba et al., 1995). Gender is a binary variable with two values (male or female). The household income variable reports the income quintiles based on the gross annual income, before tax and deductions, from all sources of all members in the family. Education attainment is a categorical variable based on the highest educational attainment and not enrolment (none, primary, high secondary, post-secondary (non-university), university (and beyond)). Education is well-regarded as a key factor which boosts political participation (Stoker 2006; Tenn 2007; Sloam 2012; Flanagan et al. 2012). In order to investigate whether education has an independent effect on the dependent variables, or it moderates the relationship between generations and the dependent variables, I also include an interaction term in my models. 

A challenge of this study is to disentangle the highly collinear age, period and cohort effects using hierarchical models. The age variable represents the biological process of aging. A cohort (or generation) refers to a group of individuals who were born in the same time and had formative ages in the same political, economic and social context (Mannheim, 1928). A period variable, like the year of the election/survey, which effects all ages in the same way and varies independent of individuals. In repeated cross-sectional surveys[2], individuals are clustered in cells cross-classified by two types of social context – cohorts and periods (Yang & Land, 2008, p.86). Fixed models fall short in accounting for the hierarchical structure of the data (Yang, 2008, p.212). In contrast, multilevel mixed models acknowledge the hierarchy whereby individuals sharing the same context are nested in cohorts and periods (Bell & Jones, 2014, 2015). In this study, the hierarchical age-period-cohort (HAPC) models[3] with random intercepts account for error-correlation (Dassonneville, 2013; Smets & Neundorf, 2014, p.43). This attempts to break the linearity of the APC model.

Figure 2 presents the findings from two cross-classified multilevel models which distinguish generational effects from period effects, while also taking into account age differences in attitudes and behaviours towards democracy. In these models, age is a fixed effect (same regression intercept for all individuals) whereas generations (cohorts) and election years (period) are specified as random effects (where regression intercepts vary among groups). Each ordinal response variable has been re-coded as binary variable for the sake of parsimony; hence, all models are regressed as logit models. Only the fixed effects of the two models are presented in coefficient plots in Figure 2: here, the estimated coefficient of each variable (with 95% confidence intervals) show how the effect of each predictor differs from zero. Those in the zero-line have no significant association with the outcome variables.

In all cases, age-squared term sits on the zero line, meaning that there is no association between one’s age and the two outcome variables. This suggests no lifecycle effect: people with a lower age are no different from those who are older. Therefore, some other time effect is at play here. It is evident that generational differences are significantly different from zero: each younger cohort, significantly have lower odds of engaging with the two traditional practices – voting and party affiliation- compared to the reference category, Post WWII generation. This is strong evidence in support of the CM hypothesis that each subsequent cohort – even when controlling for lifecycle and period effects – is gradually disengaging from traditional, elite-directed, conventions forms of participation.

Looking at socio-demographic variables, females have lower odds of voting (although not significant) and aligning with a political party. Respondents from the upper household income quintiles have higher odds of engaging compared to those with a low household income. This is as expected (Lachat, 2007).

Education has a positive but independent effect on each generation. An edu*cohort interaction variable yields insignificant coefficients for all the outcome variables. It is noteworthy that, among all independent factors, university education has the largest positive association with both democratic behaviours. There’s something beyond education, because when we disaggregate education from generational effects, there are still fixed generational effects. This means that there are generational characteristics or values that are accumulating across generations, that explain much of the variation in traditional engagement.

What does this mean?

This article finds that generations differ due to slow evolutionary changes (Ryder, 1965, p. 851). The underlying mechanism is characterised by rise in education and the development of new technologies, which are not disruptive events like a war or pandemic but accumulate permanent resources across generations. That is, these linear developments are different from period effects brought about by major historical events and are societal transformations that have affected several countries.

Although education has an independent positive effect on all outcome variables, the current study demonstrates that decline in participation occurs even amongst the better educated. In line with modernisation theory, the CM offers an explanation for this surprising trend. Dalton (2007) insists that cognitive resources can shape both engagement and disengagement. Focusing on partisanship, he distinguishes two groups with high cognitive resources based on their affinity to political parties. They are cognitive partisans and apartisans (Dalton, 1984, 2007). Cognitive partisans have strong party ties together with psychological involvement in politics in places where party cues lack. Despite their limited electoral experience, cognitive mobilisation is higher in younger generations (Dalton, 1984, p.268): younger citizens have higher education levels compared to their elders and therefore can better engage with the political information (Baker et al., 1981). As such, one might expect cognitive partisans to engage in traditional activities such as voting as well.

But what explains disengagement better than cognitive partisans is the second group – the apartisans. Apartisans have high cognitive resources like higher education but lack party ties. Although these individuals do not need party cues to make political decisions, it does not necessarily mean that they will not engage in electoral processes. It is true, nonetheless, that cognitive resources such as higher education allows one to distinguish between effective and inactive participation. The vote, for example, in its aggregate form is powerful but not blunt; that is, it provides very little information and does not guide the behaviour of the elected. For the individual, the vote is weak because they cannot disaggregate the effectiveness of their vote in terms of the extent to which they moved the decision makers to align with their preferences (Verba, 1967, p.73).

In contrast to voting, participation in activities that do not aim to achieve a policy goal, but rather bring selective group or individual benefits, may not be as powerful in the aggregate sense. Yet, it is powerful for the group or individual in terms of conveying a specific message. It appears that better-educated younger cohorts are reluctant to engage in processes that have unclear policy implications. And the analysis above shows that this is because of the increase in cognitive apartisans over time in a continuous CM process.

In an era of declining control of political parties, the ever-evolving communication media has provided broadened opportunities for mobilisation (Kriesi 2008, pp. 156–7). High hopes placed on new technologies as a quick, low-cost and suitable channel for mobilizing citizens (Norris, 2002, pp. 207–12), in an era of weakened party loyalties and the weakening of parties as mobilization agents. Social media, for example, provides a powerful tool for organizing protest rallies and petitions and lowering the costs – both time and money wise – of mobilizing people. It has hosted and provided exposure to various social movement organisations, which tackle a variety of issues relevant to younger people. This media environment is characterised by less distinct boundaries between political and non-political activities, lowering the thresholds of engagement (Ekström and Shehata, 2018). Therefore, political parties need to revaluate their connections with voters as newer generations replace older counterparts in the electorate. Using the new media to this end may be a good strategy, but there is a need for more research in understanding the dynamic impacts of the new media on young people.

Figure 1. Younger generations are better educated than older generations. Source: CSES IMD (1996-2016).

Figure 2. Coefficient plots with 95% confidence intervals for fixed effects from cross-classified hierarchical models.


[1] For the purposes of the analysis, those below the age of 18 and over 90 are removed such that all respondents have had an opportunity to participate. Younger citizens have very little chance to participate while older people have mobility issues (Grasso, 2014, p.69). To avoid issues from multicollinearity, I replace the age variable with a mean-centred age-squared term (age-47.55 squared). When fitting a regression model, multicollinearity- when predictors are highly correlated- can be problem. This can make the estimates very sensitive, which may erroneously change in response to minor changes in the model or the data. None of the other predictors are highly correlated except for age and cohort (r= 0.91), obviously because both, one’s age and the generation one is in, depends on their birth year. Mean-centring the age-squared variable reduces Pearson’s correlation coefficient, r, from 0.91 to 0.11. This transformation is useful also because an expected curvilinear relationship of age, especially with turnout (Smets & Neundorf, 2014, p.45).

[2] In repeated cross-sectional surveys, individual respondents from the same sampling frame (i.e. countries) are surveyed repeatedly over time (after each national election).

[3] The HAPC cross-classified random effects model (CCREM) for a dichotomous dependent variable can be specified as a logistic regression model (see chapter 3 for details).

Further Reading on E-International Relations

Editorial Credit(s)

Milena Marjanovic

Please Consider Donating

Before you download your free e-book, please consider donating to support open access publishing.

E-IR is an independent non-profit publisher run by an all volunteer team. Your donations allow us to invest in new open access titles and pay our bandwidth bills to ensure we keep our existing titles free to view. Any amount, in any currency, is appreciated. Many thanks!

Donations are voluntary and not required to download the e-book - your link to download is below.


Get our weekly email