Transmission and Its Endurance: The Role of Parental Politicization

Based on an individual's perceptions about their parents’ political views, the early political socialization research suggested an almost photographic reproduction of political values within the family.Footnote ¹⁶ This pattern left little room for a more nuanced exploration of possible sources of variability in the rates of political reproduction. Studies based on data collected independently on children and parents challenged this established wisdom, however.Footnote ¹⁷ Reported coefficients of association between parents and children were only modest, calling for an investigation of the conditions that favour the transmission of parental political beliefs.Footnote ¹⁸ Drawing on the insights of social learning theory,Footnote ¹⁹ few studies have pointed to the importance of parental politicization. After analysing a long series of political measures, Jennings, Stoker and Bowers conclude that ‘[f]amilies marked by parent political engagement and frequent political interchanges are families fostering the transmission of political attitudes and identities from parent to child’.Footnote ²⁰ Looking at 2006 high school seniors, Wolak reaches the same conclusion.Footnote ²¹ By discussing politics at home, parents produce less ambiguous partisan signals and thus facilitate reproduction, either as the outcome of role modelling or as a result of overt transmission.Footnote ²²

As offspring grow up, however, the initial impact of parental socialization seems to diminish.Footnote ²³ Although the question of how this influence wanes as the offspring age has not been systematically addressed, some studies have reconsidered the role of political engagement. Beck and Jennings found that young adults who came from highly politicized families in the United States were more likely to switch into independence from 1965 to 1973 than those from less politicized families.Footnote ²⁴ The authors interpreted these results in light of the turbulent political period of the 1960s and early 1970s, arguing that the more politicized youth were more susceptible to the forces of the times.Footnote ²⁵ Likewise, Jennings, Stoker and Bowers found that in the 1960s, children from politicized homes were more likely to disagree with their parents on the issue of school integration than those from apolitical homes.Footnote ²⁶ The former were more likely to hold pro-integration views, presumably due to their greater exposure to the civil rights movement.

As with these examples, the argument developed here emphasizes the indirect role that parental politicization plays in determining the longevity of parents’ influence. The initial premise is based on Converse's theory of how party identification develops over the lifespan. According to Converse, political attitudes in early adulthood are a function of prior attitudes, shaped by parental socialization, and new experiences and communications.Footnote ²⁷ Achen formalized this intuition by modelling the formation of partisan preferences as the outcome of a Bayesian learning process that pulls young people towards the current trend within society.Footnote ²⁸ Any deviation from parents’ views will thus reflect the direction of new information flows during early adulthood.

Although the partisan direction of these cues will depend on the time period and the specific social locations of individuals, the magnitude of their influence on youth's partisan outlooks should vary according to the individual's latent predisposition to receive such cues, which in turn will depend on her level of political involvement. Although selective exposure will shape some information flows,Footnote ²⁹ others can be thought of as operating exogenously. People do not tend to choose colleges or jobs on partisan grounds. And while sorting is strong in the selection of spouses,Footnote ³⁰ it is much weaker when it comes to whom a young person interacts with in college or the work environment. Exposure to political discussion in such contexts may challenge previously held views.

Social learning theories provide ample support for this argument. Mutz and Mondak show that the social context with the highest potential to generate cross-cutting political influences is the workplace.Footnote ³¹ Klofstad analyses panel data of undergraduate students who were randomly assigned a roommate in their first year of studies.Footnote ³² He finds that political discussions with the roommate boosted civic engagement and political participation. These findings reaffirm the importance of ‘weak ties’: people often lack the necessary discretion to control their agenda of social contacts.Footnote ³³ Thus, as Huckfeldt and Sprague argue, ‘people who are strategically located to affect political preferences and choices are not necessarily close friends or family but rather the structurally provided associates with whom we share a common space’.Footnote ³⁴

Political discussions are typically fuelled by the occurrence of political events, which often carry significant partisan connotations. Previous research has shown that individuals’ political attitudes are particularly sensitive to new events during the years of early adulthood.Footnote ³⁵ In the words of Alwin and McCammon, this ‘unique intersection of biography and history’ gives room for what Mannheim referred to as the ‘historical dimension of the social process’.Footnote ³⁶

The occurrence of new political events, and the presence of new political networks and contexts, will not necessarily lead to a change in PID, however. Two further attributes are essential. First, the individual must be attentive to the political cues and messages that others are providing. The probability of exposure to such cues and the attention given to them are both positively associated with higher levels of political awareness and interest.Footnote ³⁷ Secondly, young adults must develop accurate beliefs about others’ opinions. This is true both when parents are the providers of political stimuli and when the young individual acquires new extrafamilial discussion partners.Footnote ³⁸ Perceptual accuracy is higher when the attitudinal domain is more salient to the recipient of these cues. Again, this is more likely to be the case when young adults are interested in politics.Footnote ³⁹

Finally, politically interested young adults are likely to have grown up in politicized families. An adolescent acquainted with political discussions within the family will probably continue to talk about politics in her new social contexts. A person whose early socialization promoted knowledge of and interest in politics will likely continue to be attentive and interested as she moves through her adult years. This contact will bring her closer to external influences that might induce her to revise her political worldview. These ideas lead, first, to the following hypothesis:

Hypothesis 1 is predicated on the idea that individuals whose parents are highly politically involved are more likely to be exposed to (and affected by) political cues transmitted during their adulthood. Drawing on the ideas developed above, at least two factors provide such cues: new social contexts and political events. We should thus expect individuals from highly politicized homes to be more likely to:

A change in partisanship should, in most cases, not result from a direct confrontation between peers’ PID and parental PID. Rather, it should follow the accumulation of information that questions the initial political views of adolescence. The literature on the development of party identification has emphasized the dynamic interplay between issues and partisanship.Footnote ⁴⁰ Partisan divergence should thus be accompanied by a shift in issue orientations, away from parental views:

The empirical analysis starts by examining whether parental politicization increases their offspring's deviance from parental partisan preferences during early adulthood (Hypothesis 1). I then assess the indirect role of parental politicization, looking at whether young adults from highly politicized families are more likely to qualify their partisanship as a result of (1) entering a new social context (Hypothesis 2A); or (2) experiencing a salient political event (Hypothesis 2B). The next test examines whether young individuals from more politically engaged families are also more likely to abandon their parents’ views on two salient issues of the period under investigation (Hypothesis 3).

This analysis is followed by three important extensions. First, I test an observational implication of Hypothesis 1. The Southern realignment coincided with the period of early adulthood of our respondents. My theory predicts that the shift away from the Democrats was primarily driven by the politicized youth of that period. Secondly, I explore the long-term trajectory of parent-child similarity during the offsprings’ lifespan. Thirdly, I test the generalizability of Hypothesis 1 in different political settings and periods. The concluding remarks discuss how the findings from this analysis complement important insights from recent studies that focus on the role of parental influence in the development of PID.

Data

For the core set of analyses I use data from the Jennings-Niemi-Stoker Parent-Youth Socialisation Study.Footnote ⁴¹ The first wave of this study took place in 1965 when a nation-wide random sample of high school seniors (second generation, born in 1947 or 1948), together with one or both of their parents (first generation), were interviewed.Footnote ⁴² Both cohorts were re-interviewed in 1973 and again in 1982. The last wave of this study came in 1997 when the initial youth cohort, then fifty, was re-interviewed for a fourth time.

The use of this particular dataset is justified for three reasons. First, it is one of the very few studies that is based on a random sample of young individuals.Footnote ⁴³ Secondly, since it is a long-term panel study, it allows an examination of change in the level of parent-child convergence as children move through adulthood. Thirdly, it interviews the parents of the 1965 generation, thus providing direct measures of the political views and level of political involvement of the parents when the children were growing up.

Model

The core hypothesis (Hypothesis 1) asserts that parental politicization causes greater partisan deviations between parents and children when the latter leave their parents’ home. The empirical configuration of this statement implies a difference-in-difference framework, as denoted by Equation 1:

X denotes the matrix of control variables and c denotes the vector of coefficients associated with X. The subscript s in e_is denotes a clustering of the errors at the school level. The key coefficient is b. Given that Hypothesis 1 does not have a directional component, I examine the absolute partisan difference between parents and offspring. The expectation is that b > 0, reflecting the positive impact of parental politicization on deviations from parental partisanship between 1965 and 1973.

I begin with the period 1965–1973 because it covers what is thought to be the most important period of adult political development – the so-called impressionable years.Footnote ⁴⁴ During this time, the younger generation aged from eighteen to twenty-six and experienced many important developments in their lives. Most of them had finished their studies, found a job and gotten married by 1973, as was the norm at that time.Footnote ⁴⁵ Using the three first waves of the same study, Plutzer found that the only significant predictors of the growth curve of electoral participation after the age of eighteen took place during this period.Footnote ⁴⁶ The greatest gains in attitudinal stability also occurred between 1965 and 1973.Footnote ⁴⁷ I also consider later adult years in the long-run analyses presented below.

Measurement

The estimation of Equation 1 needs appropriate measures for each of its terms. Three points need to be made about measurement. The first regards how parental partisanship was constructed to represent the views of both parents in two-parent homes. The second refers to the measurement of the differenced dependent variable. The third involves the measurement of family politicization.

The Transmission of Political Interest

For the theoretical argument to hold, parental political interest needs to be transmitted to the offspring and to persist when the latter leaves the parental home. To see whether this is the case, four measures of 1973 offspring politicization were constructed. The first constitutes an encompassing measure of media consumption that consists of the following four indicators: reading magazines about politics, watching political programmes on TV, reading the newspaper and listening to political shows on the radio. The resulting scale ranges from 0 to 13, with a mean value of 8.7. The second is a 1 (not interested) to 4 (very interested) measure of interest in public affairs. The third is a dummy denoting respondents who talk very often with their spouses about politics. The fourth is a political knowledge index, ranging from 1 (no correct answer) to 8 (six correct answers). To gain a point of reference about the strength of the association between parental politicization in 1965 and youth political interest in 1973, I compare the former with parental education, using two dummies about whether each of the parents has attended college. The parental politcization variable [ParPol] is also recoded so that it ranges from 0 to 1.

With respect to media consumption, the ordinary least square (OLS) coefficient attached to ParPol is 1.81 (0.487). The equivalent figures for the two parental education dummies are 0.438 (0.262) for the father and −0.252 (0.284) for the mother. Using the measure of interest in political affairs, the equivalent figures are 0.373 (0.108) for ParPol and 0.118 (0.077) and −0.056 (0.826) for father's and mother's education, respectively. Focusing on the frequency of political discussions with the spouse, we get the following OLS coefficients: ParPol 0.200 (0.073); father education: 0.015 (0.048); mother education: 0.020 (0.049). Finally, the political knowledge OLS regression provides a similar pattern: ParPol 1.35 (0.235); father education: 0.662 (0.131); mother education: 0.235 (0.152).

It is thus safe to conclude that parents leave their imprint on their offspring's levels of political attentiveness. The transmission of political interest does not evaporate as young individuals leave their parental homes. Rather, engaging in political discussions at home and participating in politics seems to contribute to young adults’ interest in political affairs.

Controls

Before moving to the estimation of Equation 1, it is necessary to describe the set of controls included in the equation. Two possible confounds need to be controlled for, namely the strength of partisan attachment and the degree of partisan homogeneity within the family. Both of these characteristics may boost political interest within the family and may also account for the longevity of parental partisan signals.Footnote ⁵⁶ Although strength of partisanship can be regarded as post-treatment, its inclusion in the model is also an important way to account for differential ceiling effects. Offspring with parental PID = 3 can score above or equal to −3 and below or equal to 3 in the dependent variable. The possible range of values for offspring with strong Republican or strong Democrat parents is −6 to 6. If highly politicized parents are also more likely to be strong partisans, this gap may produce an upward bias in our estimates. Further controls include race, gender and two dummies denoting whether the father or mother of single-interviewed families was interviewed and whether both or one of the parents were interviewed.

Results: HYPOTHESIS 1

The results appear in Table 1. The first column shows the model without controls, whereas the second column includes all controls described above. ParPol exerts a positive effect in both cases, with a bivariate coefficient of 0.092 and a multivariate coefficient of 0.067. The reduction in the magnitude of the coefficient is primarily due to the inclusion of the variable measuring parental strength of partisanship. Excluding this variable, the coefficient of ParPol becomes 0.085. Recall that the dependent variable varies from −6 to 6 (SD = 1.33), while ParPol varies from 0 to 8 (SD = 1.78). The 0.067 net increase in the partisan gap between parents and offspring for a unit increase in the scale of ParPol amounts to a shift of 0.538 when comparing those from most and least politicized families, which is 40 per cent of a standard deviation change in the dependent variable.Footnote ⁵⁷

Table 1

The Impact of Parental Politicization on Change in Partisan Similarity between Parents and Offspring

Note: entries are OLS coefficients, standard errors, clustered at the school level, in brackets. When factorizing parental strength of PID and parental partisan homogeneity, ParPol becomes 0.060 (0.027).

Following the arguments elaborated above, this pattern is attributed to the fact that political influences during early adulthood are given more weight among those coming from more politicized homes. Online Appendix C addresses two competing explanations for the role attributed here to parental politicization. It shows that neither parental partisan stability, as a marker of the consistency of the parental cue, nor socio-demographic concordance wipes out the effect of parental politicization. The next sections try to gauge the mechanism driving this effect, focusing on the role of new social contexts, emerging events and salient political issues.

Identifying the Mechanism I: New Social Contexts (HYPOTHESIS 2A)

Leaving home at the start of adulthood, young adults either go to college or find a job. Here I focus on the first option, because this is the only one that can generate expectations about the partisan direction of a new context in which many of the high school seniors found themselves after 1965. Different work environments generate diverse partisan cues, thus without specific information about each individual's working environment we cannot formulate expectations about their effects on PID. Universities, however, are traditionally liberal institutions in the United States.

This was probably never more the case than in the 1960s, especially in its second half, when colleges were prominent sites of mobilization for liberal causes.Footnote ⁵⁸ Although in that period the link between ideology and parties was not as strong as it appears to be in the beginning of the twenty-first century,Footnote ⁵⁹ a liberal context should work more in favour of the Democrats than the Republicans.

If children from more politicized families are more likely to question their parental cues due to being more attentive to competing partisan cues encountered within the university, parental politicization should boost the impact of higher education on PID. That is to say, those from politicized homes should move more decidedly in the pro-Democratic direction as a result of acquiring a college degree than those from apolitical homes. The dependent variable used to test this hypothesis is the difference between offspring PID in 1973 and parental PID in 1965. PID is always measured on a 0 to 6 scale, from strong Democrat to strong Republican. This variable is regressed on ParPol, college attendance (0 = no college; 1 = some college; 2 = completed college studies), their interaction and the set of controls. The marginal effect of college attendance, conditional on ParPol, is depicted in the first graph of Figure 1. College seems to cause a shift to more Democratic scores but in a non-homogenous way. Providing support for Hypothesis 2A, this effect appears to be only significant for offspring from highly politicized families.

Fig. 1

Deviating from parental partisanship as a result of college attendance. Note: the solid line of the first graph depicts the average effect of college attendance on OffspringPID₇₃-FamilyPID₆₅ across all levels of parental politicization. PID ranges from 0 = strong Democrat to 6 = strong Republican. Thus the dependent variable ranges from −6 {Parent strong Republican in 1965, child strong Democrat in 1973} to 6 {Parent strong Democrat in 1965, child strong Republican in 1973}. The solid curve of the second curve presents the difference in the probability of denoting a Democrat ID while stemming from a Republican background (both offspring and parents declaring themselves Republicans in 1965), as a result of completing college education versus not going to college. The marginal effects shown in the first graph are based on an OLS model, whereas the predicted values of the second graph come from a probit analysis. In both graphs, the dotted curves display the associated 95 per cent confidence intervals. The histograms show the density of parental politicization among respondents not attending college and the kernel density (solid black, Epanechnikov kernel) curve denotes the same density for those who completed college education. Despite some imbalance, there is support in the data for higher politicization values, reducing the danger of extreme out-of-sample extrapolations.

The second graph of Figure 1 addresses a more specific scenario: what is the change in the probability of a young Republican in 1965 (from a Republican family) switching to the Democrats as a result of finishing college versus not going to university, according to her level of parental politicization? Negative values in the vertical axis indicate that graduating from university decreases the chances of denoting Democrat partisanship. Positive values denote an increase in this probability. As Figure 1 makes clear, only young Republicans coming from politicized families switched to the Democrats as a result of their experience in university during the 1965–1973 period.

A potential point of caution could be that this finding may conceal some heterogeneity between most prestigious universities (where a liberal culture did prevail) and community colleges or parochial universities (where it probably did not). If offspring from more politicized families are more likely to go to large universities, the effect attributed to parental politicization might stem from this increased probability of receiving liberal signals in college. To evaluate this competing explanation, I further distinguish whether individuals attended a public or non-denominational private university versus a community or denominational college. ParPol matters more in determining the effect of college attendance for the former, as expected. Replicating the analysis of the second graph of Figure 1 separately for the two types of colleges, the interaction term between college attendance and ParPol is 0.082 (0.032) and 0.036 (0.062) for the first and second groups, respectively.

Identifying the Mechanism II: Political Events (HYPOTHESIS 2B)

Are people from politicized families more influenced by events taking place during their early adulthood? In the political context of the period covered in our data, one salient event that endured throughout the period between the two waves of the study was US involvement in the Vietnam War. The military intervention of the United States had already started by early 1965 but peaked in 1968 and went on until 1975. Between 1965 and 1973 it remained high on the public agenda and engendered polarizing sentiments, especially among the young.Footnote ⁶⁰ Although it caused a severe decline in support for both parties, it mainly affected the Republicans since they were in office from 1969–1973, when public opinion turned against the war.Footnote ⁶¹ This means that by 1973, young adults who approved of the US policy in Vietnam should have become more attached to the Republican Party, whereas those against the war should have developed a more pro-Democrat identification.

But the core question is whether any deviation from parental PID as the result of the Vietnam War is systematically related to the level of parental politicization. To address this question, I replicated the analysis of the previous section; the only difference was that a binary variable about whether US involvement in Vietnam was a good (0) or bad (1) idea was interacted with parental politicization. Again the first graph uses the difference between OffspringPID ₇₃ and ParentPID ₆₅ as the dependent variable and the second graph depicts a more specific scenario: for those who came from Republican homes, how does the probability of identifying as a Democrat in 1973 vary by level of family politicization and attitude toward the Vietnam War? As shown in Figure 2, children from politicized homes were more likely to abandon their parental partisanship as a result of their attitudes toward the Vietnam War. Objecting to the war boosts the probability of switching to the Democrats by 20 per cent among offspring from highly politicized Republican families, but exerts no significant effect among offspring from less politicized families.

Fig 2

Deviating from parental partisanship as a result of the Vietnam War. Note: the structure is identical to Figure 1, the only difference being that now ParPol is interacted with attitudes towards the war. The solid line of the first graph depicts the average effect of attitudes towards the war on {OffspringPID₇₃-FamilyPID₆₅} across all levels of parental politicization. The solid curve of the second graph presents the difference in the probability of denoting a Democrat ID while stemming from a Republican background (both offspring and parents declaring themselves Republicans in 1965), as a result of disapproving of the US intervention in Vietnam. The marginal effects shown in the first graph are based on an OLS model, whereas the predicted values of the second graph come from a probit analysis. In both graphs, the dotted curves display the associated 95 per cent confidence intervals. The histograms show the density of parental politicization among respondents favouring the war and the kernel density curve denotes the same density for those disapproving the war. Here, there is clearly support in the data for both groups.

Identifying the Mechanism III: Salient Issues (HYPOTHESIS 3)

The last hypothesis examines the extent to which the shift away from parental partisanship is also coupled by a change in issue attitudes. Although examining whether changes in issue opinions drive partisan change or vice versa is impossible with the data at hand, it is still important to see whether, apart from changing their PID, offspring from more politicized families were also more likely to change their issue opinions in a way that is consistent with the political mood of their generation.

School integration and the role of religion in public high schools were two issues that engendered public discussion during the 1960s and 1970s in the United States. Accompanied by an egalitarian media framing, the ‘tide of opinion’ in elite communications or elsewhere was moving in a liberal direction on these issues over this period.Footnote ⁶² In a systematic study of the dynamics of issue attitudes in the United States, Mayer shows that American public opinion was shifting likewise. He divides the years between 1960 and 1988 into four periods. The second period starts in 1966 and ends in 1973, coinciding almost perfectly with the first two waves of the Socialisation study. He classifies various issues into three categories, with regard to whether public opinion became more liberal, exhibited little or no change, or became more conservative. Both school prayers and racial integration in schools are included in the first category.Footnote ⁶³

According to Hypothesis 3, we should expect that young people from highly politicized homes were getting that message and moving along the same lines. Questions on school prayer and school integration were asked of both generations of Socialisation Study respondents in 1965 and 1973. The first asked whether public schools should start every day with a prayer. The second asked respondents if they think the government should take measures to racially integrate the schools. Respondents could only state whether they agreed or disagreed with these statements, without giving an indication of how strongly they felt about each issue. Each variable is coded −1 if respondents believed government should not induce school integration/schools should start with a prayer and 1 otherwise (with 0 denoting neutral stances). The 1973 offspring measures are regressed on the 1965 parental measures plus parental politicization and the list of controls that are also included in all models above.Footnote ⁶⁴

The entries in Table 2 are log odds from a generalized ordinal logit, which tests the proportional odds assumption.Footnote ⁶⁵ Only one coefficient is shown for parental politicization, because this assumption appears to hold in both occasions for this variable. As can be seen, the effect of ParPol is positive in both issues, although it is much lower in the first column. The last part of the table presents the change in the probability that the offspring will adopt a liberal stance in either issue when shifting ParPol from the 25th to the 75th percentile. In both cases, the 1965 parental values are set to −1. Parental politicization increases the chances of shifting from parental views towards more liberal attitudes by three and eight percentage points in the issues of school integration and prayers, respectively.

Table 2

Change in Issue Attitudes towards More Liberal Stances from 1965 to 1973

Note: entries in the first part of the table are log odds from generalized ordinal choice models. The 1965 parental and the 1973 offspring scales take the following values: −1 (do not integrate schools/start with prayer); 0 (depends, no opinion); 1 (integrate schools/do not start with prayer). Robust standard errors, clustered at the school level, are in brackets. The last row denotes the change in the probability of adopting a liberal position on each issue in 1973 if parents adopted a conservative stance in both issues between low and high levels of ParPol.

An Observable Implication: Parental Politicization and the Southern Realignment

The previous results seem to suggest that members of the class of 1965 who grew up in politicized homes were especially likely to shift to the left during their early adult years. Is the direction of this shift driven by the context-specific partisan influences of this period, or does it signal a more general tendency, whereby parent-child divergence only occurs when the political context prompts youth to develop more liberal stances? The former is consistent with the argument of this article, while the latter would undermine it.

To examine this question, we need to find a political process characterized by salient anti-Democrat cues. A good case for this purpose is the Southern Democratic dealignment. While the Vietnam War was leaving its stigma in American politics writ large, a parallel and equally important development was taking place in a specific area of the country: since the mid-1960s and until at least the end of the 1970s, the until-then solid Democratic South gradually but steadily moved away from the party. There has been solid evidence that young cohorts of that period constituted the frontline of this shift.Footnote ⁶⁶ Even if the realigning process had already initiated before 1965, our young cohort was in the heart of this change. If parental political involvement is important because it introduces the youth to new political stimuli – rather than simply pushing the offspring to more liberal attitudes – we should observe that the offspring of more politicized Southern families are more likely to abandon the Democrats than those coming from less politicized homes.

Empirical evidence addressing this question is provided in Figure 3. The first graph again uses the difference between OffspringPID ₇₃ and ParentPID ₆₅ as the dependent variable and shows the gap between Southerners and non-Southerners across all levels of ParPol. The second shows the probability of identifying as a Democrat in 1973 among Southerners and non-Southerners. In generating the predicted values used in the graph, the following scenario was assumed: all high school seniors came from Democratic homes and declared themselves as Democrats in 1965. This exact scenario is at the heart of the Southern realignment. As the figure shows, young Southerners growing up in politically engaged homes primarily drove the shift away from the Democrats.

Fig. 3

Abandoning the Democrats in the South. Note: the structure is again similar to Figures 1 and 2. The solid line of the first graph depicts the average difference in {OffspringPID₇₃-FamilyPID₆₅} among Southerners and non-Southerners, across all levels of parental politicization. The solid curve of the second graph presents the difference in the probability of denoting a Democrat ID while stemming from a Democratic background (both offspring and parents declaring themselves as Democrats in 1965), among Southerners and non-Southerners. The marginal effects shown in the first graph are based on an OLS model, whereas the predicted values of the second graph come from a probit analysis. The histogram and the density curve suggest that the distributions of ParPol in the two areas are quite similar.

In the Long Run: Tracing the Footprints of Parental Influence over the Offsprings’ Life Trajectory

Do the differences attributed to the level of parental politicization persist after 1973 or do they evaporate as the class of 1965 becomes older and as the political environment of the late 1960s and early 1970s is replaced by that of the 1980s and 1990s? To explore these long-term patterns I use all four waves of the youth socialization panel. The third wave came in 1982 and the fourth wave took place in 1997.Footnote ⁶⁷ I employ a growth curve model to estimate the long-term impact of parental politicization. Growth curve models are based on an interaction between the predictor of interest [ParPol] and a function of time. By coding time accordingly, the main effect of ParPol represents its effect on the starting level (1965), whereas the interaction term indicates its effect on the growth rate of this relationship. To avoid functional form assumptions regarding how parental-child similarity evolves over the offsprings’ lifespan, time is modelled using a set of dummies denoting the year of the panel waves. Thus instead of one parameter indicating the growth rate of parental-child convergence, three such parameters are estimated, each one corresponding to a given wave of the panel study.

Following the logic of Hypothesis 1, the dependent variable measures the absolute partisan distance between the parent in 1965 and the offspring in each of the four waves. The model used in the analysis is:

where i = 1,…,N indexes individuals, t = 0,…,T–1 indexes the number of observations for each individual and α_ki is the growth curve parameter for i with k = {0,1,2}.

Growth curve models are based on the idea that the baseline and growth curve parameters vary across individuals. By obtaining estimates for α_ki for each i, we have N different values for these parameters, which we can then treat as random variables. This leads to an intercept- and slopes-as-outcomes model, represented by the level-2 equations, given below:

Since both the main variable of interest [ParPol] and all controls are measured in 1965, all variables included in the model (X) are deemed to influence both the starting level (α₀) and the growth rate (α_1, α_2, α₃). The estimation substitutes (3), (4), (5) and (6) into (1), giving room to a three-level mixed-effects model.Footnote ⁶⁸ Level-3 units represent the school clusters. Level-2 units represent the respondents available in any of the four waves and level-1 units consist of the repeated observations (individuals stacked within the four waves of the study). In other words, repeated measures of observations are nested within individuals, who are nested within high schools.

In this model, α_0is denotes the level of parent-child partisan similarity when individuals are eighteen years old. Accordingly, γ_01s estimates the impact of parental politicization on the starting level of parent-child convergence, that is, before individuals leave their parental home. α_1i measures the growth rate of this relationship by the time youngsters are twenty-six years old. γ_11s estimates the impact of parental politicization on the change in the probability of parent-child partisan congruence from 1965 to 1973 and γ_21s and γ_31s estimate the same effect but compare 1965 to 1982 and 1997, respectively.

Table 3 presents the results and Figure 4 visualizes them. The small circles in Figure 4 display the average absolute distance in partisanship between children and parents for the less politicized group, that is, those located at the 25th percentile of the empirical distribution of the variable. The triangles represent the more politically engaged families: those located at the 75th percentile of the empirical distribution of the variable. The vertical spikes denote the 95 per cent confidence intervals accompanying each point estimate.

Table 3

The Impact of Parental Politicization on the Starting Point and the Growth Curve of Parent-Child Partisan Concordance, 1965–1997

Note: entries in the first row are log odds from a mixed-effects logit model, standard errors in brackets. Controls, and their interactions with time, are not shown. Level-3 N: 97; Level-2 N: 1,335; Level-1 N: 4,526.

Fig. 4

Parent-child partisan divergence in the long run. Note: the black circles denote the average distance (in a 0 to 6 scale) between parents and offspring among respondents coming from families with lower scores of political engagement. The triangles represent the equivalent estimates for offspring coming from politicized homes. The solid and dashed spikes indicate the 95 per cent confidence intervals. The predictions are based on the growth curve estimates shown in Table 3.

Offspring from more politicized families are more likely to change their views during early adulthood; doing so makes their increased level of initial similarity with their parents disappear. Parental politicization boosts the starting point but decreases the growth rate of parent-child partisan similarity. On the whole, the results of the growth curve analysis confirm the findings from the period between 1965 and 1973. What is new from this analysis, and crucial for the appreciation of the importance of the ‘impressionable years’ in shaping people's partisan profiles, is that deviations occurring during early adulthood largely persist over the individual's lifespan. I discuss the implications of this finding in the concluding section after I examine the generalizability of Hypothesis 1 in different political generations and political contexts.

A Generalizable Dynamic?

The youth sample from the Jennings-Niemi Socialisation Study entered into early adulthood during a very turbulent period in American politics. Between the mid-1960s and the early 1970s, American youth experienced various dramatic events that provided a common bond that helped this cohort develop into an actual generation.Footnote ⁶⁹ Consequently, the results might be cohort specific.

To assess the generalizability of the findings, I provide two tests from different cohorts. The first test uses data on a younger generation obtained in the last wave of the Political Socialisation Study. In 1997 the children of the initial youth sample – then aged fifty – were also interviewed (N = 769). Members of this third generation were raised in a rather placid political period that did not resemble the exceptional circumstances of the 1960s.

These offspring differ from their parents in that they do not constitute a school-year cohort, that is, they vary in terms of age. Although this might seem an inconvenience, it can be helpful for the present purposes.Footnote ⁷⁰ I divide the third-generation sample into two categories. The first includes offspring between twenty-three and twenty-nine years old, with a mean age of twenty-five (N = 370). This is approximately the age of the 1965 cohort in 1973. The second group comprises those between sixteen and twenty-one years old (N = 203), with a mean age of eighteen –the same age as the original youth sample in 1965. Although the third generation was only interviewed in 1997, and thus a truly dynamic empirical model is impossible, we can compare the younger and older third-generation cohorts. The relationship between parental PID and child PID for those with an average age of eighteen would correspond to the 1965 parent-child similarity of the previous analysis. Here we would expect a positive relationship between parent politicization and parent-child similarity, reflecting the consequence of the former on parent-child transmission during the pre-adult years. The relationship between parental PID and child PID for those with an average age of twenty-five would correspond to the level of parent-child similarity in 1973. Here we would expect a reversal in the relationship, with those coming from highly politicized homes holding views that are less similar to their parents than those from apolitical homes – a pattern driven by the former group's greater political learning and engagement during their ‘impressionable years’.

To examine this, I regress child PID (on the 0 to 6 scale) on parental PID, parental politicization (measured in the same way as in 1965 and again ranging between 0 and 8) and the interaction of the last two terms. The analysis is implemented separately for each of the two age cohorts. The marginal effect of parental PID conditional on parental politicization is presented in Figure 5. For the younger group, the level of parent-child partisan similarity is enhanced when parents are engaged in politics. For the older group, the pattern reverses, as expected. Partisan similarity drops as a result of parental politicization. The coefficient of similarity between parental and offspring PID doubles (from 0.3 to 0.6) among politicized families for the young group and drops to almost half for the old group (from 0.5 to almost 0.25). The small number of observations creates high levels of uncertainty accompanying these estimates. However, we see that even when a different cohort with completely distinct political experiences is analysed, the pattern identified earlier is repeated.

Fig. 5

The marginal effect of parental PID on child PID for two age cohorts of the 3rd generation (wave of 1997) across all levels of parental politicization. Note: the solid line presents the marginal effect of ParentPID₉₇ on OffspringPID₉₇, both of which are measured with the same 0 (Strong Democrat) to 6 (Strong Republican) scale. The histogram in both graphs shows the distribution of parental politicization for ParentalPID ≥ 3 and the black solid curve displays the density curve for ParentalPID < 3.

The second test relates to British youth during the 1990s and uses information from the British Household Panel Study (BHPS), one of the very few datasets that allows us to assess the level of parent-child partisan similarity in a longitudinal perspective while providing independent measures of parental characteristics as well.

The BHPS began in 1991 with a sample of some 5,500 households and 10,300 individuals drawn from 250 areas of Great Britain. All household members aged sixteen or more were re-interviewed in successive yearly waves and, if they split off from original households, all adult members of their new households were also interviewed. To gauge the trajectory of parent-offspring partisan similarity, I focus on respondents aged between sixteen and twenty years old in 1991 who still lived with one or both of their parents (N = 710). The analysis traces these individuals from 1991 (median age: eighteen) to 2008 (median age: thirty-five).

Only a few questions in the BHPS are devoted to political issues and attitudes. Parental politicization is captured only through parents’ self-evaluation of their level of interest in politics (ranging from ‘not at all interested’ [1] to ‘very interested’ [4]). Given that the British party system has more than two parties, a unifying PID measure equivalent to the American PID scale used in the previous sections is not appropriate.Footnote ⁷¹ The dependent variable is constructed by first creating a measure of concordance in party support between parents and children in 1991. If parents and children show support for the same party they score 1 (0 otherwise). Following the logic in Hypothesis 1, the resulting measure of party concordance in 1991 is used as a baseline to denote the impact of parental socialization during adolescence. For 1992 this measure is then subtracted from the degree of concordance between children in 1992 and parents in 1991. The same logic is followed until the 2008 wave of the study. The analysis includes respondents between sixteen and twenty years old in 1991, thus on average of equal age as the class of 1965 in that year. The generic form of the model used to capture the trajectory of parent-child convergence according to the level of parental politicization is:

where t = 1992, 1993,…,2008. The subscript f denotes error clustering at the family level.Footnote ⁷² The dependent variable takes values between −1 and 1. A score of -1 means that parents and children agreed in their party preference in 1991 but disagreed in t. A score of 1 implies that although there was no agreement in 1991, parents and offspring agree in t. A score of 0 implies that members of the two generations either agreed at both time points or disagreed at both time points. The validation of Hypothesis 1 in the British context requires that b_t become negative as the offspring enters into early adulthood (as t becomes larger). Moreover, if the long-term pattern found above also holds in Britain, b_t should remain stable once the offspring reach the age of mid- or late twenties.

Figure 6 summarizes the results. The scatterplot represents the coefficients of parental politicization across all yearly waves from 1992 to 2008. The solid line denotes a local regression curve (loess) fitted into the scatterplot. The two dashed curves display the point-wise 95 per cent confidence intervals of the loess estimates. The overall pattern seems to confirm prior expectations. The downward trend indicates that offspring from more politicized homes are more likely to deviate from their parents’ partisan influences. This trend stabilizes after 2000, when the average offspring age is twenty-eight. This is exactly what we should expect if the findings of the previous sections denoted a more general tendency that is not confined only to the ‘Vietnam generation’ in the United States.

Fig. 6

The effect of parental political interest on the longevity of parent-child partisan similarity. Note: the solid curve summarizes the yearly coefficients denoting the effect of parental political interest on parent-child partisan similarity. The dashed curves denote the point-wise 95 per cent confidence bands of these point estimates. BHPS data, 1991–2008.