Voting as a “Habit” and Path Dependence

Virtually all major works on turnout have concluded that voting behavior is, in part, a gradually acquired habit. Milbrath explicitly invokes psychological concepts of reinforcement and “habit strength” to characterize habitual voting (1965, 31; see also Verba and Nie 1972, 148; Miller and Shanks 1996, 17, 62). In short, although the terminology varies, there is a longstanding agreement that voting behavior is habitual. Thus, it is useful to think about civic development as a largely path-dependent process.

The process begins as young citizens confront their very first election. Each has a latent probability of voting resulting from parental, demographic, and personal factors. Some will vote and become habitual voters, but most will not and are likely to remain nonvoters in subsequent elections. That is, most new citizens show evidence of inertia. Inertia is rather different from persistence, a central organizing concept in political socialization research (Sears 1990; Searing, Schwartz, and Lind 1973). The notion of persistence suggests that one can find the origins of adult attitudes and ideology in events and influences that occurred many years earlier. In contrast, inertia suggests that the roots of current turnout can be found in voting behavior in the previous one or two elections.

Almost all habitual nonvoters gradually increase their probability of voting and eventually make the transition to being habitual voters. Some do so almost immediately, but others remain habitual nonvoters for many elections. From this perspective, the key explanatory task is identifying the factors that influence how quickly different voters make the transition from one inertial state to another. Of course, habitual voters are sometimes thwarted from voting in a particular election. But failing to vote once does not shift them back to nonvoting. In most cases, the disruption is short-term and they return to their habitual behavior. Once voters are on the path of regular voting, factors that were once important (parental influences, educational attainment) diminish.

Starting Levels of Turnout

As young citizens confront their first election, all of the costs of voting are magnified: they have never gone through the process of registration, may not know the location of their polling place, and may not have yet developed an understanding of party differences and key issues. Moreover, their peer group consists almost entirely of other nonvoters: their friends cannot assure them that voting has been easy, enjoyable, or satisfying.

Young people also lack many of the resources that can promote participation. Because they have little disposable income, they are not attractive targets for parties seeking campaign contributions or for interest groups mounting direct mail campaigns. Few of them own homes, have stakes in community politics, or have completed college. Thus it is not surprising that few young citizens are registered to vote (Timpone 1998) and their turnout is relatively low.

Nevertheless, many young citizens do vote—presumably because “the start-up costs of voting are not borne equally by all young people” (Wolfinger and Rosenstone 1980, 60). A few have resources that allow them to overcome the high costs of first-time registration and first-time voting. These resources can come from their parents or from their own achievements. Verba, Schlozman, and Brady (1995, chap. 15) distinguish between parental socioeconomic status and parental political involvement. Parental education promotes offspring education and offspring political knowledge, including such mundane information as how to register and where to vote. Thus political knowledge is an asset that can directly offset costs of initial turnout. Parental political involvement can provide both behavior to model and campaign-relevant information that children rarely get from formal schooling.

Parental resources combine with young citizens' own achievements before their first election. Although they may not have yet completed their formal education, they may have acquired some political knowledge and may have developed an interest in public affairs (reflected, e.g., in reading newspapers or weekly news magazines). They may also have some partisan attachments, though these are likely to be malleable over the long term and may therefore have weak or transient effects (Jennings and Niemi 1981).

In sum, a variety of factors—all previously identified by other authors—should be expected to influence starting levels of turnout. However, it is unclear whether any of these factors will matter in the future or whether their effects will dissipate because of the inertia that typically follows.

The Nature of Inertia and the Habits of Turnout

A citizen's voting history is a powerful predictor of future behavior. This is easily illustrated by using self-reports of turnout from those who completed the 1972–1976 National Election Studies (NES) panel study (Center for Political Studies 2000). Of the 516 respondents who reported voting in both 1968 and 1972, only 3% missed voting in both the 1974 and the 1976 elections. In contrast, of the 57 respondents who reported missing two consecutive elections (either 1968–72 or 1972–74), more than two-thirds did not vote in the next election.

Unfortunately, the nature of inertia is undertheorized. Here, I offer some speculations, the specifics of which will remain for future research to fill in. First, the two inertial states are very different. The stability of habitual nonvoting is one that most citizens “outgrow” and is, therefore, weaker. Citizens can outgrow the nonvoting habit if the costs of voting remain relatively constant over the life course but the resources available to overcome those costs increase. Income, for example, tends to increase in young adulthood, and this may gradually increase the likelihood of being recruited into electoral politics (Verba, Schlozman, and Brady 1995, chap. 5). Second, cognitive resources—especially political knowledge—should increase with age (Wolfinger and Rosenstone 1980; Strate et al. 1989; delli Carpini and Keeter 1996). Third, life events such as marriage, home ownership, and having children in school often increase community ties, enhance the relevance of local elections, and increase the perceived stakes of electoral outcomes (Verba and Nie 1972). Fourth, even constant activities can have cumulative impacts. For example, young voters who attend weekly religious services may gradually acquire politically relevant resources [information from politicized sermons, opportunities for leadership and public speaking, etc. (see Verba, Schlozman, and Brady 1995)].

Overall, many resource (e.g., socioeconomic status, cognitive assets, political engagement, social networks) and life-cycle variables (e.g., marriage, home ownership) act on habitual nonvoters by increasing their probability of voting in subsequent elections. Each resource variable should be expected to speed the transition from nonvoter to habitual voter.

The Inertia of Habitual Voters and Temporary Disruptions

Both habitual nonvoters and habitual voters display inertia, but the nature of that inertia differs. Nonvoters must overcome bureaucratic barriers (most notably registration) and costs of various types (e.g., information costs). These same factors, however, contribute to stability once a citizen has crossed the threshold and voted once. Although residential mobility is common, most voters do not move in a four-year period. So once they register and vote, the barrier of registration is unlikely to come into play in the next election or two. Similarly, once voters attain a certain degree of political knowledge—for example, finally understanding some key differences between the parties—they are unlikely to forget it or find that their knowledge has become obsolete in the next few elections.

Norms also contribute to inertia. Voting for the first time is likely to bring positive reenforcement from friends, family, and co-workers. As voters age, their peer cohort has increasingly higher participation rates and they eventually move out of age-homogeneous settings (e.g., college residences) and into others (workplaces, community organizations) where the average levels of political knowledge and turnout are higher. Thus, as young citizens take on more adult roles, they are likely to be subject to more and more intense participation norms.

For all of these reasons, after young citizens vote once, they should only occasionally experience an increase in the costs of voting relative to their resources and social pressures. A reversal may occur as a result of a major life event that creates immediate demands for one's time and attention. For example unemployment has a noticeable effect on voter turnout but only if the onset of unemployment was very close to the election; if more than a few months pass, voters seem to rebound (Rosenstone 1982). This pattern of a “rebound” raises the question, “Rebound to what?”

A rebound implies that severe economic dislocations will show up as major factors discouraging habitual voters, but severe disruptions may be largely irrelevant to nonvoters because most would not vote anyway. At most, hardships are just one of many barriers facing habitual nonvoters or newly eligible citizens. Therefore, smaller hardship effects should be evident among the very young because so few are habitual voters.

The central disruption factor that has been studied to date is residential mobility (Squire, Wolfinger, and Glass 1987). Habitual voters who move from one precinct to another usually must reregister to vote and discover the location of a new polling place. Thus, residential mobility may temporarily disrupt their regular pattern. Like other disruptive factors, residential mobility is primarily a factor affecting habitual voters, and its effects should be considerably less evident among younger citizens than their older counterparts.

To summarize, voter turnout is developmental and characterized by modest inertia for nonvoters and strong inertia for voters. Key explanatory variables from previous research can be characterized in terms of whether they affect starting levels or the speed of transition or whether they function primarily to disrupt established patterns. This relatively simple framework provides the basis for more precise hypotheses (e.g., that the effects of parental characteristics dissipate rather than persist and that disruptions such as mobility and parenthood are likely to have increasing explanatory power as citizens age).

Before moving putting these expectations to the test, it is useful to examine the methodological challenges involved in analyzing turnout via cross-sectional survey data and to introduce an analytic approach that has not previously been exploited in research on turnout.

Block I. Demographic and Parental Characteristics

Block I variables include sex, race, and three indicators of parents' socioeconomic status: educational attainment, reported family income, and the Duncan Occupational Prestige score for the head of household.

Five additional variables measure the political environment created by the parents. The first is a standard question about the parents' interest in politics; the second is the parents' score on a political knowledge quiz consisting of six factual items. Additional measures are scores on a three-item scale of political trust, parents' strength of party identification, and parents' self-reported vote in the 1964 presidential election.⁵

Unless otherwise noted, all predictors that are numeric scales (e.g., family income and years of education) or have arbitrary metrics (attitude measures with agreedisagree responses) are standardized (mean of 0 and standard deviation of one), while dummy variables remain in their natural metric. This coding ensures that all predictors have a meaningful value of 0, aiding in interpretation of the intercept.

Block II. Youth Characteristics at the Completion of High School

This block includes resource measures based on interviews three years before the seniors became eligible to vote. I use self-reported high school GPA as an indicator of educational achievement. An index of activity in student politics and student organizations addresses activities whose recall was found to be important by Verba, Schlozman, and Brady (1995, chap. 14); here they are measured as a scale composed of indicators of having run for student office and of a holding leadership role in any student organization. A measure of church attendance serves as an indicator of community-based resources.

Three variables tap into psychological resources and political engagement. The first is a three-item engagement in politics scale, which is a standardized composite of expressed interest in politics and the frequency of both newspaper and newsmagazine readership.⁶

Block III. Events and Achievements Between High School and the First Election

Although the youth were not interviewed between 1965 and the 1968 election, the 1973 interview was sufficiently detailed to enable construction of measures for four key life transitions during this period. Two are potential resources and two are presumed disruptions. Dummy variables indicate whether each young adult had attended college (63% of the sample did so), was married in 1968 (28%), or became a parent before the 1968 election (15%).

A simple measure of geographic mobility was impossible to construct because the questionnaire did not ask when each move between 1965 and 1973 occurred. However, respondents indicated if and when they moved to another state, permitting the computation of a dummy variable indicating if a respondent moved across state lines between 1965 and 1968.

Block IV. Life Events and Achievements After the First Election

The last causal block includes a wide array of resources and disruptive factors. The resources include both events/achievements (reports of final educational level, economic status, marriage, parenthood, and community ties) and psychological/subjective measures (political engagement and political knowledge).

The reports of events and behaviors include most of the usual variables suggested in life-cycle research. All are measured as of 1973—less than a year after the 1972 election and prior to subsequent elections in 1976 and 1980. Thus, they capture events during the first five years of voting eligibility for each respondent. Three variables tap various aspects of socioeconomic status. An ordinal measure of “final” educational achievement is coded −1 for those who did not attend college, 0 for those with some college attendance, and +1 for those who completed a college degree. Family income is measured by an 18-category variable that has been standardized. Income differences tend to widen with age and may be a poor indicator of one's position in the stratification system at age 25. At that age, young attorneys and management trainees earn less than many skilled blue-collar workers even though the former understand that they are on a track toward an upper middle-class standard of living. I therefore include a dummy variable denoting whether the head of household has a professional, technical, or high managerial position.

Three variables tap aspects of community involvement and social ties: a dummy variable indicating home ownership, a four-category measure of frequency of church attendance (also standardized), and marital status (measured by two dummy variables indicating whether the respondent was married or divorced; never married is the omitted category). Divorce can be thought of as a disruption that can lower turnout.

Additional disruption measures include the number of residential changes since 1965 (standardized) and parenthood, as measured by two dummy variables. The first indicates the presence of school-age children (age six or older), and the second is coded 1 if the respondent has one or more children under six (but none older). Childless respondents are the omitted category.

Limitations of the Data

The Jennings and Niemi study constitutes the best (perhaps the only) data set that will permit estimating latent growth models of turnout. However, certain features of the data set bear on generalizibility. Most obviously, the period between 1965 and 1973 encompasses the escalation of the Civil Rights, antiwar, and campus free speech movements, along with related movements and political events. Undoubtedly, some of the parameter estimates would differ if a different cohort of high school seniors were studied (although it is difficult to anticipate which these might be).

Second, the study's design eliminates those who dropped out of high school prior to their senior year. In 1965 this was roughly 27% of the cohort. Thus, the restricted range of education almost surely narrows the range of several variables including parent education and SES, parent and student turnout, and youth's 1973 income. These probably have the effect of attenuating some relationships, yielding fairly conservative hypothesis tests and downwardly biased slopes (and relatively high percentages of self-reported turnout). The restriction in range almost surely exaggerates the speed with which high-resource and low-resource segments of the population converge in their turnout rates as they age.

Finally, with only four elections observed for each respondent, the smooth curves suggested by latent growth curve models represent a simplification. With a longer series of observations, life events producing dramatic short-term increases or temporary disruptions could be modeled more precisely (indicators of shocks preceding particular elections could be added to equation two). In the analyses summarized below, shocks will be reflected in substantial changes in the steepness of estimated growth curves, but these curves will nevertheless be estimated as smooth, monotonic functions.

Issues of Model Specification

Previous research suggests that all variables in Blocks I, II, and III are plausible predictors of turnout in the 1968 election. However, if these contribute to each individual's underlying probability of turnout in their first eligible election, it is not clear whether they should be expected to have additional impact on the rate of increase. Consider, for example, the impact of having parents with high levels of education, income, and political involvement. On the one hand, a rich cognitive environment could have a lasting legacy that provides an additional impetus toward political interest and participation, reflecting the idea that education provides the basis for further learning throughout the life course. Thus, a cognitively rich and politically engaged parental environment in childhood and adolescence might result in an acceleration turnout growth, reflected in a hypothesized positive effect on β₁. Such a growth hypothesis will be tested for each predictor.

However, the effect of parental resources could diminish over time as the young citizen leaves home and is exposed to more recent effects of community transitions, socioeconomic achievements, and family formation. Thus, in the absence of strong theory or previous research, hypotheses of no effect on growth are at least as plausible as hypotheses of positive effects. Thus, the question of whether the β₁ side of the model is over-specified cannot be resolved on theoretical grounds.

The initial model, therefore, will start with all Block I variables modeled as predictors of β₀ and of β₁. My standing expectation, however, is that all effects on β₁ will be 0. For reasons of parsimony and to avoid potential problems of multicollinearity, insignificant predictors of β₁ will not be retained when models add additional variables further along in the causal chain. This strategy strikes a balance between concerns about collinearity and the weak theoretical bases for hypotheses concerning long-term growth.⁷

Table 1 lists the independent variables by temporal block and shows each variable's expected effect. The table has several notable features. First, the expected dissipation of effects of youth experiences is reflected in expected slopes of 0 for all pre-1968 measures on growth rate. Second, it is impossible for achievements occurring after 1968 to influence turnout in the 1968 election; therefore, all of these slopes are constrained to 0 a priori. Resources all have expected positive slopes. Disruptions occurring before the first eligible election are not expected to have any effects, but disruptions occurring later should lower turnout and this will be reflected in flatter growth rates (negative slopes).

Pre-1968 Factors: A Summary

Many events and social characteristics occurred or were fixed before the class of 1965 confronted its first election. Among these, having highly educated parents who voted in the 1964 election, being psychologically engaged with politics as high school seniors, and attending college boosted initial turnout, while mobility between 1965 and 1968 depressed it.

These findings supporting and clarify recent scholarship about education's role in political participation. Verba, Schlozman, and Brady (1995, 437–8) conclude that effects of parent's education are mediated largely by offspring's own socioeconomic status and interest in politics. Table 2 shows that parental education and political knowledge have no effect (neither direct nor indirect) on turnout growth. Their modest effects on initial turnout in the baseline model (γ's of 0.276 and 0.211) drop by only 25% (to γ's of 0.225 and 0.150, respectively) after accounting for offspring political engagement and college attendance (Table 4). Thus, the mechanism translating parental resources into children's initial political involvement is not captured by the variables in the model—perhaps because of poor specification (the transmission processes are captured by different variables entirely) or measurement error (high school grades and a simple indicator of college attendance may not adequately capture underlying variance in youth's educational achievements). One implication for cross-sectional studies, however, is that the effects of parental environment diminish over time and should, therefore, be modeled as interactions with age.

Among 21 pre-1968 variables, only parental strength of partisanship influenced turnout's rate of increase. Since partisanship is presumably a stable trait, the 1965 measure may be a reliable indicator of levels of partisanship many years in the future. Thus having partisan parents may accelerate the acquisition of knowledge and resources implicated in the lifelong learning process that underlies Rosenstone and Hansen's (1993) life experience approach.

Home Ownership and Mobility: Additional Considerations

Home ownership has long been regarded as an enhancer of turnout (e.g., Verba and Nie 1972, 139) and mobility a major barrier, but the estimates show that neither has an impact on growth rates. In each case the null findings could be (a) an artifact of the data and methods used here, (b) a peculiar characteristic of this particular cohort, or (c) a serious challenge to prior work.

The developmental model suggests that mobility effects will be greatest for habitual voters and, therefore, negligible for young voters and the estimates are consistent with that argument. Yet the possibility of an artifact is also high because the mobility measure is imprecise and stretches back to 1965. To get more leverage, I tried to replicate this finding using cross-sectional data from the American National Election Studies cumulative file (Miller and National Election Studies 1999). I pooled data from the same four presidential elections and estimated a model of turnout for four age groups. The youngest were 21–33 and correspond to the ages of those in the latent growth models. Table 6 reports the estimated slopes for geographic mobility, controlling for the year surveyed, age, race, education, sex, home ownership, family income, children in household, married, and folded party identification. The effect of mobility is substantively small and insignificant for young citizens and is large and significant for voters over 33 years old. These striking results provide strong corroboration that disruptions from mobility have an increasing impact throughout the life course. This conclusion suggests that an interaction with age may be the best way to specify the effects of mobility with cross-sectional data.

In contrast, pooled NES data (analysis not reported here but available on request) casts doubt on my estimates for the effect of home ownership; it has a robust effect at all age groups. After some exploration, I have concluded that the null effects here are probably due to a combination of factors—both substantive and methodological. First, home ownership is largely cumulative—most first time home owners continue to own homes for a long period thereafter—and therefore tracks closely with age and turnout. Thus, if unreliably measured, its true effects are likely to be absorbed by age or (in this research design) time.

Second, home ownership status changes very rapidly in the age range represented by our data. According to the Census's (1993) analysis of 1982 home ownership, only 19% of those under 25 owned a home but that number soars to 57% for those 30 to 34. Pooling Census data for a number of years (details available on request) shows the home ownership rate increases by 3.4% per year between age 21 and age 34. In 1973 38% of the sample reports owning a home and the Census data suggest that nearly 60% would own a home by the time of the 1980 election. Thus, the single measure has the benefit of being prior to voting outcomes but the corresponding disadvantage of being an extremely poor indicator of ownership status at the times of the last two elections. The null effects, combined with consistent and robust effects in cross-sectional studies, suggest that the positive impact of home ownership is probably immediate and therefore poorly captured by the developmental specification used here.

Family Life and Voter Turnout

The final set of objective variables measure aspects of family life: two dummy variables measuring marital status in 1973 and indicators of whether there are young children or school-age children in the household. The distinction between younger and older children derives from the idea that school-age children get parents involved in secondary associations such as the PTA and that schools provide a personal link to government policy. This implies that while very young children drain time and energy and lead to lower turnout, older children enhance networks that would result in greater political knowledge, mobilization, and an interest in outcomes. Taken together, these four variables paint a complex picture of the relationship between family life and turnout.

Although marriage when very young (most respondents were 22 or younger in 1968) had no effect on initial turnout, the same measure in 1973 had a substantial positive impact on turnout growth. In contrast, those who were married and divorced by 1973 (eight years after graduating from high school) had distinctively low growth rates (although the divorce measure does not achieve statistical significance, the two variables as a set improve the fit of the model). These growth rates, however, are higher than those of the never married—so divorce reverses only a portion of the marriage benefit. Yet, as Milbrath speculated, marriage is a mixed blessing for turnout because marriage is most often accompanied by children that erode about half the gain in turnout associated with marriage. Neither measure of the presence of children achieves significance and the two estimates are similar in magnitude. Given the similarity in effect estimates, I substituted a summary measure that simply indicates the presence of children. This did achieve significance (γ = 0.253, one-tailed p = 0.035) and is included in results reported in Table 7.

Psychological Engagement in Politics

Table 7 reports a model that adds the political engagement and political knowledge scales, both of which are significant predictors of turnout growth. Individually, the effects are modest but their combined impact can be substantial. Note also that the effect of education on growth is no longer statistically significant and the estimate has declined by about a third. Thus the two cognitive measures used here mediate about a third of the effect of higher education.