Political Behavior

, Volume 34, Issue 4, pp 689–718

Primary Voters Versus Caucus Goers and the Peripheral Motivations of Political Participation

Authors

    • Department of GovernmentHarvard University
Original Paper

DOI: 10.1007/s11109-011-9175-8

Cite this article as:
Hersh, E. Polit Behav (2012) 34: 689. doi:10.1007/s11109-011-9175-8

Abstract

Depending on their state of residence, Americans can participate in Presidential nomination contests either by voting in a primary or by attending a caucus. Since caucus participation requires more time and effort than primary voting, it has long been thought that caucuses must attract a more partisan, activist, and politically extreme cohort of citizens than primaries. This paper challenges the view that more burdensome electoral institutions necessarily ought to attract more politically engaged citizens. I propose a theory of peripheral motivations that predicts caucus goers and primary voters will not differ in terms of their political attitudes or interest, but they will differ in their levels of community engagement. The key insight is that many of the reasons why citizens choose to participate or abstain from politics actually have little to do with politics. Analysis of two surveys from the 2008 Presidential election substantiates the theoretical expectations.

Keywords

Presidential nomination contestsComparative electoral institutionsVoting

Depending on their state of residence, Americans can participate in Presidential nomination contests either by voting in a primary or by attending a precinct caucus. Most states use a primary system, which from the point of view of voters is procedurally identical to typical general election voting. Over a quarter of the states, however, nominate Presidential candidates through a caucus system, and a caucus is anything but typical voting. A caucus is a lengthy neighborhood event: attendees must show up at a particular time on a particular date for an evening of debate, speeches, active persuasion, and public voting. Attending a caucus is arguably the most burdensome form of election participation available to an American citizen.

Any careful observer looking at the caucus procedure in comparison with the primary procedure will no doubt suspect that the institutional design of the electoral method must have some bearing on the types of candidates who win and lose. After all, in every domain of politics, the rules governing institutions affect outcomes therein produced. On account of the rules, the electorate of caucus attendees is expected to be, in some important way, politically distinct from the electorate of primary voters.

Conventional wisdom, corroborated by scholarly research, suggests that caucuses are dominated by activists with extreme views (e.g. Stone et al. 1992; Mayer 1996; DiClerico 2000; Panagopoulos 2010), who are likely to favor candidates whose political views are outside the mainstream. This position has been cause of disdain for the caucus. The late David Broder of the Washington Post wrote disparagingly that the “[caucus] system empowers the activists and those with built-in organizational ties who can mobilize people to leave their homes for a couple of hours on a weeknight”. 1 More serious than the complaints of journalists are claims by Presidential campaigns that this electoral system is actually unfair on account of the supposed unrepresentativeness. In the closely contested 2008 Democratic nomination battle, the Hillary Clinton campaign openly questioned the caucus’ legitimacy. At the height of campaign, Pennsylvania Governor Ed Rendell, acting as a Clinton surrogate on Meet the Press, stated, “Caucuses are undemocratic \(\ldots\) so we want primaries. That’s the way we elect Presidents.” 2 The effect of caucus rules versus primary rules is thus no small matter; the legitimacy and fairness of U.S. Presidential elections is at stake in the difference between the two nomination procedures.

The conventional view that caucuses attract more extreme voters than primaries stems from an obvious cost-benefit rationalization: compared to primary voting, caucus participation is a costly endeavor, and as costs increase, only the most committed political activists will take part. Just as burdensome registration requirements will keep politically disengaged citizens from participating in elections (Highton 2004), so too should a caucus format keep mildly interested citizens at home.

But there is an alternative, equally plausible, view that suggests that burdensome elections should not attract a more politically engaged cohort of citizens. In this article, I develop an argument that the peripheral motivations an individual may have either for participating in political activities (e.g. an opportunity for social interaction) or for abstaining from political activities (e.g. work or family conflicts) are sufficiently orthogonal to one’s level of political engagement that a seemingly costly activity like a caucus will actually not attract a more concentrated group of politically active voters than a primary election would attract. The argument is an extension of Berinsky’s (2005) explanation for why voting reforms, such as vote-by-mail, do not reduce participatory bias in elections.

The goal of this research endeavor is to juxtapose the conventional cost-benefit argument with the conflicting perspective of peripheral motivations, and to test the theories’ opposing predictions with data. I first show that the 2008 caucuses did not attract a concentrated group of political activists as compared to primaries, a finding consistent with the peripheral motivations model. This model also yields a second important prediction about Presidential caucuses. Because of the public nature of the caucus voting experience, social engagement is a key peripheral motivation for participating in this kind of political act (Grose and Russell 2008). Because the caucus is in essence a neighborhood meeting, socially engaged voters are also expected to be prime targets of mobilization. I test the hypothesis that, compared to primaries, caucuses attract participants who are particularly engaged with their communities, regardless of their level of political engagement.

The prediction that caucuses will attract citizens who are engaged with their communities is consistent with the dominant theories of participation and mobilization, such as those offered by Verba et al. (1995) and Rosenstone and Hansen (1993). What is unique in this theoretical angle is an explanation for why community engagement may distinguish caucus attendees from primary voters even though measures of political engagement do not distinguish these sets of participants. Typically, citizens are expected to need not only social, or solidary, benefits in order to overcome participatory costs, but also to have a heightened engagement with politics. The theoretical contribution highlights that even as a political act becomes more intense, it need not be the politically engaged citizens who dominate. Paradoxically, interest in politics may be less central to participation in politics than we had thought.

To test the key hypotheses, I report evidence from two public opinion surveys from the 2008 Presidential election. I start with a basic methodological approach, and then re-estimate the basic models while accounting for several possible sources of bias. The propositions of the theory are largely confirmed in the data. However, the data presented are observational and come exclusively from the 2008 election cycle, and so broad conclusions are modestly drawn. The unusual circumstances of the 2008 nomination battle, in which both parties held competitive contests in most states, is too alluring of a research opportunity to be hindered by suboptimal data resources. The insights gained here are thus exploratory and tentative, but they do serve to allay some concerns about the fairness of Presidential caucuses and they reinforce the point that disengaged citizens will be incorporated into the political system through civic engagement rather than through institutional reform.

Election Rules and Representativeness

In the wake of the McGovern-Fraser Commission recommendations to modernize the Democratic Party’s nomination process following the 1968 Presidential election, scholars studied the effects of the new voting system on participation and on nomination outcomes. Much of the research focused on voting rules. This body of work asked how rules, such as delegate allocation schemes, might favor some types of candidates over others (e.g. Lengle and Shafter 1976; Geer 1986; Parent et al. 1987; Gerber and Morton 1998). A separate and distinct line of inquiry was also pursued, in which researchers asked how representative, demographically and attitudinally, primary and caucus voters are of primary and caucus abstainers (e.g. Ranney 1972; Geer 1988; Norrander 1989). 3

Whereas nomination rules were broadly heralded as important to election outcomes, the representativeness literature produced quite mixed results. When compared to citizens who vote in general elections but not in primaries or caucuses, primary/caucus voters were generally found to be unrepresentative with respect to age, education and other demographic features, but research shows conflicting evidence about how attitudinally distinct primary/caucus voters are from non-participants (Norrander 1996). The conflicting evidence is puzzling since an obvious mechanism for how election rules might affect outcomes is through the kinds of voters who are attracted to some kinds of elections but not to others.

Integrating these two agendas and asking how election rules in the nomination contests might affect the kinds of voters who participate has produced similarly mixed results. Mayer (1996) shows 1988 caucus goers to be more politically extreme than primary voters. Stone et al. (1992) show that 1984 Iowa caucus goers were more ideologically committed than non-participating partisan identifiers as well (see also Stone et al. 1989). Panagopoulos (2010) finds that in 2008 caucus goers were distinct from primary voters in their policy views. On the other hand, Panagopoulos (2010) does not find ideological differences between primary voters and caucus goers. Redlawsk et al. (2008) find that compared to non-participants, 2008 caucus goers held similar policy positions, and Marshall (1978) finds that the 1972 caucus goers were not more extreme with regard to their political attitudes than primary participants.

Not only is the evidence mixed that caucuses attract a distinct set of voters from primaries, but before the 2008 election, any differences between caucus and primaries seemed to be of very little consequence to election outcomes. Looking back at the nomination contests from 1972 through 1992, as Mayer (1996) has done, or even through 2008, as Squire (2008) has done, caucuses generated about the same results as primaries. Mayer shows evidence that more ideologically extreme candidates (like Jesse Jackson and Pat Robertson in 1988) have done better in caucuses than primaries, but the general finding is that this advantage has made no difference at all in determining the party nominees. Squire (2008) shows that in all but a few election years, the preferences of Iowa caucus goers tracked with national public opinion.

Finding only minor institutional effects, few have attempted to make sense of this form of political participation or to offer theories about how and why caucuses are different from primaries in regard to the voters they attract. But the 2008 nominating process has demonstrated a need to explain the difference caucuses make. For the first time, the caucus-primary distinction may have influenced the nomination outcome in both parties. Consider the 2008 Republican nomination. The eventual nominee, John McCain, won 30 out of 38 primaries but won only two out of 11 caucuses. The 2008 Democratic nomination contest was similarly asymmetric. The Democrats faced off in 14 caucuses and 39 primaries. Barack Obama won 13 of the 14 caucuses but won fewer than half of the primaries. 4

What explains the difference in outcomes between these two types of elections in 2008? Is this simply a regional pattern, with caucus states clustered in one part of the country? No. Caucuses are held in every region of the United States, with similar results across the board. What makes caucuses different is not where they are, but what they involve. And what they involve, as one Iowa county supervisor once put it, is “[turning] folks out for four hours on the coldest night of the year to fight with their neighbors about politics (qtd. in Winebrenner 1998, p. 81).” If caucuses are to generate different outcomes than primaries, and the supposed mechanism causing this difference is the composition of participants, then it is imperative to understand the ways in which caucus goers do and do not differ from primary voters. That is the goal of this project.

Who Can Overcome Participatory Costs?

The Conventional View

The conventional view of caucuses entails that to a greater extent than primaries, caucuses attract citizens who are outliers in terms of their ideological disposition and in their political activism. The rationale is that because the caucus is a much more demanding form of participation than primary voting, only the most committed activists are likely to attend.

Consistent with this view, Mayer (1996) shows that participants in 1988 caucuses are wealthier and more educated, more partisan and ideological than primary voters. However, two points are worth noting. The first is that while Mayer shows that caucus goers in 1988 are both demographically distinct and ideologically distinct, these two findings are not necessarily linked (Norrander 1989). Indeed, the wide gaps in income and education between voters and non-voters have been found time and again not to translate into similar gaps in attitudes (see, for example, Wolfinger and Rosenstone 1980; Citrin et al. 2003; Ansolabehere and Hersh, n.d.). Secondly, what was true about caucuses in the 1970s and 1980s is not necessarily true anymore. In the earlier years of the modern Presidential nomination era, caucuses might very well have been dominated by party elites and activists. As technology and campaign strategy have evolved, caucuses may no longer be the smoke-filled rooms they once were.

The presentation of data that is the core of this article shows that in the 2008 nomination contest, in contrast to the conventional view, caucuses did not attract more politically active, politically knowledgeable, or politically extreme voters than primaries; however, caucuses were dominated by citizens engaged in community endeavors. The community engagement finding is not particularly surprising given the literature on participation and mobilization. Indeed, we expect a time-intensive, community gathering like a caucus to attract those who are already engaged in their communities, both because these are the kinds of citizens who derive social benefits from involvement and because they are the kinds of people at the center of social networks likely to be mobilized to participate. What is surprising is that levels of political engagement do not differ when comparing caucus participants and primary voters.

What can explain the finding that a higher cost political activity will not demand a higher level of political engagement from participants? The proposition offered here is that the reasons for participating and abstaining from political activities may be less tied to a citizen’s interest in politics than is often assumed. This perspective extends the work of Berinsky (2005), which though unrelated to Presidential nominations, offers an instructive parallel in which a high-cost voting environment is compared to a lower-cost voting environment but the difference in cost does not affect the correlates of participation as one might expect.

Berinsky (2005) considers recent election reform efforts such as no-excuse absentee voting and early voting. These reforms, he explains, are intended to reduce voting costs and make voting easier for everyone, especially those who are chronic non-voters. Lower-income individuals, for example, might not be able to leave their jobs to vote on Election Day, but by opening the polls for a period of several days or weeks prior to Election Day, more of these citizens can participate.

But it turns out, Berinsky finds, that when voting is made easier through such reforms, those who take advantage of the expanded voting opportunities seem to be individuals who are already “rich in politically relevant resources,” but who, due to sickness, a recent move, a business trip, or some other peripheral factor, might have missed voting on Election Day. Thus, instead of increasing participation among the disengaged, the reforms exacerbate representational bias. The theoretical perspective articulated here is an extension of this insight: lower participation costs (as in primaries) will not attract less politically engaged citizens.

The Peripheral Motivations View

A caucus is a high-cost voting experience. Attendees must show up at a particular hour on a particular day for what is quite an active political experience. Who is able to bear these costs? To answer this question, it is useful to separate participatory costs into two categories: central costs and peripheral costs. Costs are central if they are integral to the participatory act itself. For example, anxiety about a public voting environment and the burden of arguing with one’s neighbors about politics are central costs to caucus participation. On the other hand, peripheral costs are logistical: examples include having to set aside four hours at a particular time of day, the opportunity-cost of missed work, illness, and childcare. The benefits of participation can similarly be divided into central benefits and peripheral benefits. The central benefits of caucus participation are the chance to deliberate about politics and participate in a highly engaging political exercise. The peripheral benefits of participation are the non-political aspects of socializing with neighbors. Peripheral benefits are equivalent to what Clark and Wilson (1961) referred to as solidary incentives.

Peripheral Costs

First, consider the peripheral costs of a precinct caucus. Concerns such as a work schedule, a business trip, or a child at home might prevent an otherwise interested citizen from participating in a caucus. But suppose that election costs are suddenly lowered, say by converting the caucus into a primary, as many states did during the nomination reform era (Atkeson and Maestas 2009). Now, the otherwise interested voter who works late at night is able to vote in the morning, and the citizen who can spare thirty minutes of her day but cannot spare four hours can cast a ballot and move on to other things. Does this reduction of voting costs decrease representational bias, as the conventional view of caucuses predicts, or increase representational bias, as the Berinsky model might predict?

For the conventional theory to be coherent, it must assume that there is a positive relationship between one’s willingness to overcome peripheral costs and one’s level of political engagement. Thus, a politically engaged citizen will leave work early if necessary to get to the caucus, but a less politically engaged citizen will not do the same. This assumption is underlying in Verba et al. (1995) Civic Participation Model as well as in most other models of participation. Verba et al. (1995) argue that increasing intensity of political participation ought to be positively correlated with political information, partisan commitment, political efficacy and political interest (see Chap. 12).

Yet, Berinsky’s insight suggests that the relationship is more complicated than this. Perhaps the politically active lawyer who works very late must skip a nominating caucus but the less engaged worker who gets off at five is able to attend. Surely, if politically engaged voters are wealthier they might be more equipped to overcome some costs (like paying for a baby-sitter), but they might also be more likely to work late into the evening.

The assumption here is that overall, peripheral costs are likely to balance each other out among the more politically active and less politically active types and thus result in an orthogonal relationship with political engagement. If this assumption is true, then the result would be a lower turnout in caucuses than in primaries (since the costs are higher) but a similar distribution of political engagement among participants in the two election forms (since the costs do not disproportionately affect one type of citizen). Thus, if we imagine a state shifting from a caucus to a primary, turnout would increase at the same rate among more and less politically active citizens. Conversely, the assumption of the conventional model is that since the ability to pay peripheral costs is positively correlated with political engagement, caucuses should have lower turnout than primaries and those who do turn out should be a concentrated group of politically engaged citizens.

Peripheral Benefits

To determine whether politically engaged citizens are more likely to overcome central costs (like the anxiety of public voting) than less engaged citizens, it is necessary to explore the benefit side of participation that might counteract such costs. What kind of citizen would be willing to bear the cost of an evening of political debate? There are two kinds. Politically engaged voters will be interested in the political activity because it satisfies their political interest. They derive a benefit that is central to the political act itself. But another set of people derive a peripheral benefit from caucus participation. These are individuals who value community gatherings and are simply willing to stomach four hours of neighborhood politics to receive such a benefit. 5

The peripheral, community-based, rationale for participation can alternatively be viewed through a mobilization framework (See McAdam et al. (2001), Chapter 5). Individuals who are positioned in community networks such that they are asked by peers to attend the caucus, may attend the caucus in order to comply with the community request rather than on account of political engagement. If an individual deeply invested in community activities is particularly likely to attend a caucus, it may be because of his or her own interest in the community event or it might be on account of mobilization efforts. Though these reasons are distinct, they are conflated in this analysis for two reasons. First, both of these reasons stand in contrast to participating on account of an ideological, partisan, or general political commitment. For another, a personal social motivation and a mobilization-based social motivation cannot be disentangled with the available data.

The social, community-oriented, or solidary incentive to participate is widely discussed in the literature on representation and mobilization. However, this rationale for participation is typically treated as one of many reasons why individuals may overcome participatory barriers. For instance, Rosenstone and Hansen (1993) and Verba et al. (1995) emphasize the importance of social rationales for participation, but to bear costs of high-intensity political activities, a person is not just presumed to derive these solidary benefits, but also to be particularly engaged, interested, and knowledgeable about politics. Or, as Verba et al. (1995) write, “Political participation is the result of political engagement and resources (p. 354, emphasis in the original).” The modest contribution of this discussion of central and peripheral costs and benefits is to tease out the conditions under which a person may participate in a high-cost political activity but not be drawn to do so by extra-ordinary political interest. The point is to emphasize that the solidary benefits of political participation may not be just one necessary rationale among many that lead citizens to take part in activities like caucuses, but may be their dominant reason for doing so (see Anderson 2009). Caucuses are high-cost activities, but overcoming the associated costs need not require an intense interest in politics, but could alternatively require the fulfillment of other, less political interests.

On the other hand, the plausible alternative view holds that the political content of the caucus drives away all but the most politically engaged, even citizens who otherwise participate actively in community endeavors (see Mayer 1996; Trish 1999; and Mendelberg 2002). Berry et al. (1993) sum up this general viewpoint well when, after investigating deliberative politics in America’s cities, write:

People are not unwilling to devote time to their communities... But face-to-face democracy can be intimidating. Even though these systems of citizen participation are centered in the neighborhood, where people may know other participants, the give-and-take of small group politics remains an unattractive way to spend an evening for the vast majority of people (p. 97).

While it may be true that most people are turned off by the prospect of a neighborhood political meeting, it might be the case that the social aspect of the caucus is the prime motivator for citizens who do participate. A modest interest in the political act might be a prerequisite for attending a caucus, but for many caucus participants perhaps it is the social aspect that tips the rational voter equation towards the benefit side and stimulates their participation. For whom is this social benefit most significant? It is those who are already community participants, who know their neighbors, and who are accustomed to civic meetings for whom this benefit might actually draw them into such a task as a caucus.

There are three reasons why citizens who are involved in community endeavors are more likely to attend a caucus than those who are uninvolved, regardless of their level of political engagement. The first reason is that caucuses provide citizens with an opportunity to reinforce their social ties. A caucus is a time when people can catch up with their neighbors and hear about what is going on in the community. Those who are uninvolved in community endeavors will not derive this kind of benefit, for they will be in a room of strangers.

The media-portrayed image of a caucus as a highly contested screaming match seems like an inopportune place to reinforce social ties for anyone. On the contrary, it may seem like a more likely venue to make enemies. However, this image of a caucus as a highly polarizing event is overstated. Caucus-goers often organize potluck dinner or snack tables in order to make the experience more enjoyable and social. Typically, the most highly contested portion of a caucus event is the debate on issue platforms that follows the vote on candidates and this portion of the meeting is not attended by participants who rather not debate issues. Finally, because most caucuses take place in rural states with small close-knit communities, even participants who are not at all politically active often know many of the other people in attendance. While the caucuses are certainly animated events in which politics is discussed, sometimes heatedly, and in which votes are held in public format, they may indeed also be venues in which neighbors can reinforce social ties. 6

The second reason why community activists may be more likely to participate in a caucus is that being active in associational life provides citizens with civic skills that make participation in a neighborhood meeting less intimidating (Putnam 2000; Verba et al. 1995). If citizens are not anxious about the content of the meeting, if they know what to expect from a community discussion—including a familiarity with, and tolerance for, the procedural boredom of a formal meeting and those few insufferable personalities who will no doubt be present there—then they will be more likely to attend.

The third reason might more appropriately be thought of as a cost of non-participation. For those who know their neighbors and participate in the community generally, social pressure may exert a penalty on non-participation (Gerber et al. 2008; Knack and Kropf 1998). Indeed, at the heart of research on voter mobilization is what Dale and Strauss (2009) call Social Occasion Theory, or the insight that the more personal the appeal to citizens the more effective the appeal is at stimulating participation. The community-centered and public voting format make the caucuses a particularly good platform for grassroots, social network mobilization, as Busch (2008) points out. Citizens who are active in the community but not in politics will likely be asked by their more political neighbors to attend, and since in a caucus attendees know who shows up and who stays home, active citizens will likely face serious pressure to be present. Of course, citizens who are detached from community life will feel less of a social penalty should they choose to abstain.

Summary of Theoretical Contribution

The question confronted here is why might caucuses, which are undeniably a more burdensome form of a political participation than primaries, not attract a more politically extreme cohort of participants. The proposed answer has two key elements. First, many of the participatory costs associated with caucuses, such as that caucuses are at a set time of day and last several hours long, will not necessarily be any easier to overcome for someone who is very interested in politics than by someone who is only somewhat interested. And second, overcoming costs of political participation does not necessarily require a person to have a extra-ordinary commitment to politics, but may simply require that he or she benefit from the non-political features of the activity, such as the social aspects.

Political Engagement and Participation

I first test the proposition that caucus-goers are distinctive in their political attitudes and activism from primary voters. I turn to two public opinion surveys from the 2008 election. While a study of a single election year is not without its perils, 2008 presents a unique opportunity to study primaries and caucuses. The opportunity stems from two facts about this particular election cycle: (1) both parties held competitive contests (i.e. no incumbent President or Vice-President was seeking a nomination) and (2) whereas in some years only the states that hold early primaries and caucuses are competitive, in 2008 the contests in both parties were competitive in most states. Because so many citizens across the country had an opportunity to participate in a meaningful election, national political surveys can be employed in a way that has been more difficult in earlier election cycles. Yet, there are valid reasons to be concerned about the generalizability of evidence from the 2008 election. These reasons are addressed in a section below in which I confront several possible sources of bias.

The survey questions analyzed here come from the 2008 American National Election Study (Krosnick and Lupia 2009) and the 2008 Cooperative Congressional Election Study (Ansolabehere 2009a). With respect to this analysis, these two surveys each have strengths and weaknesses. The strength of the National Election Study (NES) is the broader range of survey questions relating to community engagement, which will be the focus of the latter portion of the data analysis. The strength of the Cooperative Congressional Election Study (CCES) is that with 32,800 respondents, its sample size is nearly sixteen times that of the NES. Because caucuses are only held in about a quarter of the states and because compared to primaries, fewer citizens participate in caucuses, the larger sample is quite beneficial. This is particularly true in this first part of the analysis, in which the goal is to show that two groups are not distinct. A large sample size places a much greater burden on a null hypothesis, as the confidence intervals surrounding estimates are small, and even slight differences between groups can emerge. Finally, because the CCES and NES surveys are conducted very differently—the first done on the Internet, the second in person—if both surveys show similar results, this should increase confidence in the robustness of the findings. 7

Survey respondents are divided into those residing in caucus states and those residing in primary states. Residents of six states are omitted from this analysis because the states do not fall neatly into a caucus state-primary state classification. 8 To understand why these states are excluded, consider two examples. In Washington State, both parties hold caucuses as well as primaries. 9 In a state like Washington, citizens have the option of attending either a primary or a caucus or both. The NES survey question on primary/caucus turnout in states like these is not specific enough to differentiate caucus goers from primary voters. A somewhat different case is a state like Idaho, where the Democrats hold an open caucus and the Republicans hold an open primary. In this case, there is not an identifiable group of voters who could have participated in a caucus (primary) and decided to participate or not participate. These six idiosyncratic states must be considered separately, a task I leave for future work.

After excluding these states, we are left with a total of forty-four states plus Washington DC, including the ten caucus states of Alaska, Colorado, Hawaii, Iowa, Kansas, Maine, Minnesota, Nevada, North Dakota and Wyoming. While all of these states are included in the CCES analysis, some are missing from the NES analysis because the NES survey is not conducted in all states. The NES analysis focuses on the twenty-seven primary states and five caucus states in which citizens were surveyed. The caucus states in the NES analysis include Colorado, Kansas, Minnesota, North Dakota, and Nevada. 10

In consideration of the political extremism hypothesis, I scoured the two surveys and found a combined nineteen survey items along which caucus goers might be expected to be outliers. 11 Before turning to multivariate regression models, I begin with a simple analysis of differences of means for each of the nineteen survey items. Along each measure of political activism, I compare how caucus-goers differ from caucus-abstainers and compare how primary voters differ from primary abstainers. I then compare the two differences. If the difference between the caucus states and primary states is large, then this would be consistent with the theory that the caucus attracts an especially unrepresentative subset of citizens.

The reason for using this measurement approach is that it accounts for average differences between states that hold primaries and states that hold caucuses. As will be discussed in greater detail later in this paper, the most difficult challenge in comparing the institutions of caucuses and primaries is that they are not assigned randomly to states or to voters. Cultural and historical reasons are at the root of these institutional choices. Caucus states and primary states are different, and so to compare them, it is important to account for the state differences to the greatest extent possible. As an illustration of the problem, consider ideology. According to the NES, caucus participants are more ideologically extreme than primary participants. But, caucus abstainers are also more ideologically extreme than primary abstainers. Thus a direct comparison of behaviors between primary voters and caucus goers without accounting for the average amount of that behavior in the different states will lead us astray in comparing the effect of the institutions on participation. Throughout the analysis, I therefore compare differences between participants and non-participants and see how those differences vary by institution.

Figure 1 compares voting participants and non-participants on political items, with the NES data displayed on top and the CCES data displayed on bottom. 12 Five of the NES items displayed are indicator variables: (1) whether the respondents wear campaign pins or post signs in their homes or on their cars, (2) whether they have contributed to a political campaign, (3) whether they have contacted a public official to express their opinions, (4) whether they have ever participated in a protest, and (5) whether they have ever signed an online petition. The four remaining items—partisanship, ideology, one’s interest in politics generally, and in the presidential campaign specifically—have a range of values and so for the purposes of display here are standardized to have mean of zero and standard deviation of one.
https://static-content.springer.com/image/art%3A10.1007%2Fs11109-011-9175-8/MediaObjects/11109_2011_9175_Fig1_HTML.gif
Fig. 1

Caucus goers versus primary voters: who is more politically engaged? Note: Each point on the graph represents the mean value for participants minus the mean value for non-participants, in caucuses (black color) and in primaries (gray color). 95% confidence intervals from two sample difference of means tests are indicated (unequal variances assumed)

To interpret the values in Fig. 1, consider the Contact item in the NES data as an illustration. Thirty-three percent of caucus participants reported having contacted a public official to express their view in the prior twelve months. This compares to only 13% of caucus abstainers who reported this behavior. The mean for caucus goers minus the mean for caucus abstainers then, is 33 − 13 = 20%, the value shown in the figure. A two-sample difference of means test (unequal variances assumed) is performed to generate confidence intervals around the difference. The same procedure is performed for primary voters and primary abstainers. In the case of the Contact survey item, the difference between primary voters and abstainers is also roughly 20%, and so we conclude that caucus goers are no more or less likely to have contacted an official than primary goers.

The bottom half of Fig. 1 contains a similar battery of survey items, but these originate from the CCES survey. Again, all variables other than the partisan item, the ideology item, and the two measures of political interest are binary. Partisan ranges from independent to strong partisan, and Ideology ranges from moderate to very liberal or very conservative. With much smaller confidence intervals than in the smaller sample NES, there is some divergence on several of the CCES items. Notably, on the measures of political interest and attention to news, it is primary voters who are somewhat more engaged than caucus goers. On the other hand, caucus goers appear slightly more likely to have worked on a campaign and are slightly more ideological.

Looking at Fig. 1, there is scant evidence that caucus goers are political outliers as compared to primary voters. On some items primary voters are slightly more active than caucus goers, on others caucus goers are slightly more active, and on most items they are statistically indistinguishable. This pattern is true not only with regard to the NES data, which has a rather small sample of caucus state residents, but also with the larger-N CCES data. Thus in a highly contested Presidential nomination, in which the caucuses and primaries produced vastly different outcomes for the candidates, the hypothesis that caucus goers represent extreme or hyper-engaged voices has little support.

Looking at the question items in Fig. 1, these variables fall into three categories, each of which picks up on a separate aspect of political engagement. The first two items, Ideology and Partisanship, relate to the extremity of one’s political attitudes. The two items about attention to the campaign and to political news relate to the extremity of one’s political interest. The remaining items relate one’s political activism. I thus combine the survey items into these three categories: Attitude, Interest, and Activism. Within the categories, as I have defined them, I use principal components analysis to generate summary measures. 13

I build a logistical regression model where turnout in either a primary or caucus is the dependent variable. The independent variables of interest are the three measures of political engagement and the interaction of these terms with an indicator of the respondent living in a caucus state. Again, this method accounts for heterogeneity in average levels of political engagement between caucus states and primary states. The key test here is the significance of the coefficients on the interaction terms: large, positive interaction terms mean that in caucus states political engagement is more related to participation than in primary states.

Table 1 shows four logit models that test this relationship. Models 1 and 3 include the main effects and interactions with no additional controls. Models 2 and 4 include demographic controls for race, income, education, age, marital status, gender, length of residence in the community, an indicator for whether one has children at home, and indicators for Democratic and Republican identification. The first two models are generated from the NES data, the second two are from the CCES data. All four models tell the same story: the probability of turning out is higher in primary states than in caucus states and is higher for those with stronger political attitudes, greater political interest, and higher levels of activism. However, it is not the case that political engagement is more relevant to caucus participation than primary participation. 14
Table 1

Caucus goers versus primary voters on political engagement

Dep var: turnout

Model 1

Model 2

Model 3

Model 4

NES data (reduced)

NES data (full)

CCES data (reduced)

CCES data (full)

Independent vars:

β

SE

β

SE

β

SE

β

SE

Intercept

−0.28**

0.06

−3.72**

0.45

1.00**

0.02

−1.64**

0.12

Political attitude

0.37**

0.06

0.24**

0.08

0.30**

0.02

0.26**

0.02

Political interest

0.51**

0.07

0.43**

0.08

0.53**

0.02

0.41**

0.02

Political activism

0.59**

0.07

0.60**

0.08

0.60**

0.02

0.57**

0.02

Caucus state

−1.00**

0.20

−0.91**

0.22

−0.98**

0.05

−1.00**

0.06

Atittude × caucus state

−0.02

0.20

0.06

0.21

0.04

0.05

0.09

0.06

Interest × caucus state

0.05

0.24

0.16

0.26

−0.11

0.07

−0.15**

0.07

Activism × caucus state

−0.17

0.18

−0.23

0.19

0.06

0.06

0.09

0.07

Control vars.

 Black

  

0.42**

0.16

  

0.21**

0.06

 Hispanic

  

0.23

0.20

  

0.11

0.07

 Education

  

0.19**

0.06

  

0.12**

0.01

 Income

  

0.02

0.01

  

0.03**

0.01

 Female

  

0.30**

0.12

  

−0.09**

0.04

 Age

  

0.03**

0.00

  

0.02**

0.00

 Married

  

0.30**

0.14

  

0.12**

0.04

 Children

  

0.18

0.14

  

0.04

0.04

 Residence

  

0.08*

0.05

  

0.15**

0.01

 Republican

  

0.31

0.28

  

0.17**

0.07

 Democrat

  

0.54**

0.26

  

0.49**

0.07

Observations

1,687

 

1,576

 

21,012

 

19,584

 

Log Likelihood

−960

 

−846

 

−10,919

 

−9,845

 

Note: The above table presents estimates from logistic regressions in which turnout in either caucuses or primaries is the dependent variable. ** Denotes significance at the p < 0.05 level, * Denotes significance at the p < 0.10 level. The lack of statistically significant positive coefficients on the interaction terms suggests that measures of political engagement are no more correlated with caucus participation than primary participation

Figure 2 provides an illustration of the relationship between political engagement and turnout in caucuses and primaries. Expected probabilities are generated for several hypothetical individuals using models 2 and 4 from Table 1. 15 In each case, the demographic control variables are set at their means. For the ‘low type’ hypothetical, the three measures of political engagement are set one standard deviation below their means; for the ‘middle type’, these three variables are set at their means; and for the ‘high type’, the variables are set one standard deviation above the means. Figure 2 indicates the expected probability of participating in either the caucus or the primary for each ‘type’ of voter. The figure demonstrates that the relationship between political engagement and turnout, as measured by the slopes of the lines in the figure, is quite similar in caucuses and primaries.
https://static-content.springer.com/image/art%3A10.1007%2Fs11109-011-9175-8/MediaObjects/11109_2011_9175_Fig2_HTML.gif
Fig. 2

Expected probabilities for caucus/primary participation by level of political engagement. Note: This graph shows expected probability estimates with 95% confidence intervals generated from Models 2 and 4 in Table 1. With control variables set at their means, the graph indicates how expected caucus and primary participation increases with increasing levels of political engagement. With higher levels of political engagement, participation in the two election forms increases at similar rates

In a quest to examine the hypothesis that caucus goers are more politically engaged than primary voters, I have enlisted nineteen survey items from two national surveys, and I have shown three distinct methodological angles at the evidence: difference-of-means tests for the individual survey items, logistic regression estimates, and expected probabilities for hypothetical citizens calculated from the logit models. There are small pieces of evidence that caucus goers are actually less politically engaged than primary voters (i.e. political interest variables in Fig. 1 and NES results from Fig. 2), and also small pieces of evidence that caucus goers are more political engaged (i.e. a few of the items in Fig. 1 and the CCES results in Fig. 2). In all, the null hypothesis that 2008 caucus goers and primary voters are identical in their level of political engagement certainly cannot be rejected.

This finding is quite surprising in light of conventional expectations about the relationship between participatory costs and participatory bias. Political commentators like David Broder and political scientists like Verba et al. (1995) and Mayer (1996) assert that a higher-cost activity like a caucus ought to attract individuals who are more interested or invested in politics. The result here is contrary to that view, but is consistent with the peripheral motivations model in that lower election costs do not seem to reduce the concentration of politically engaged citizens among election participants.

Community Engagement and Participation

The 2008 National Election Study asks a battery of questions related to community involvement. Respondents were asked if they had done any community work in the previous year, if they had attended a school or community meeting in the previous year, if they had done any volunteer work in the previous year, and if they are members of any organizations. Together, these four survey items should capture the level of participation in community activities. According to our theoretical expectations, it is the respondents who answer yes to these kinds of questions who would be expected to receive a high social benefit from participating in a caucus, irrespective of their level of political engagement.

Figure 3 shows the proportion of respondents answering affirmatively to each of these four questions. Respondents are divided into caucus participants, caucus abstainers, primary participants and primary abstainers. Notice that the overall rates of community involvement appear to be higher in the caucus states than in primary states. That is, caucus goers are more involved than primary goers and caucus abstainers are more involved than primary abstainers. More importantly, take note of the gap between participants and non-participants. In the first two items, and to a lesser extent in the third item, the gap between caucus-goers and abstainers is greater than the gap between primary voters and abstainers. Only in the fourth item is the caucus gap greater than the primary gap.
https://static-content.springer.com/image/art%3A10.1007%2Fs11109-011-9175-8/MediaObjects/11109_2011_9175_Fig3_HTML.gif
Fig. 3

Community activity of caucus/primary participants and non-participants. Note: This graph summarizes how caucus and primary participants and non-participants differ in their levels of community engagement, as measured with four survey items from the 2008 National Election Study. 95% confidence intervals are indicated. The difference between caucus participants and non-participants is larger than the difference between primary voters and non-voters on most items

Figure 3 suggests that it may indeed be true that those involved in their community are more likely to attend a caucus, but not more likely to vote in a primary. To test this proposition, I return to logistic regression. Participation in either a caucus or primary is the dependent variable. As the independent variable of interest, I generate a measure called Community that combines the four community involvement items into a single variable. Community is the sum of the four root variables and ranges from zero (answers ‘no’ to all four questions) to four (answers ‘yes’ to all four questions). As with the political extremism model, this model includes both a main effect and an interaction with an indicator for residence in a caucus state. A statistically significant positive coefficient on the interaction would imply that the community variable is more relevant to caucus turnout than to primary turnout.

In Table 2, two models are shown. In both models, all of the control variables that were included in Table 1 are estimated. Model 5 includes the community measure and interaction term. Because it is clear both from theory and from evidence shown in Table 1 that political engagement is relevant to participation generally (though not relevant to caucuses especially), Model 6 includes the three measures of political engagement, the three interactions of political engagement with caucus state residence, as well as the community measures. Model 6 tests the crucial hypothesis here: does community involvement matter more in caucuses than in primaries, holding constant one’s political involvement?
Table 2

Caucus goers versus primary voters on community engagement

Dep var: turnout

Model 5 (NES data)

Model 6 (NES data)

Independent vars:

β

SE

β

SE

Intercept

−4.95**

0.41

−3.76**

0.45

Community

0.33**

0.05

0.13**

0.05

Caucus state

−1.38**

0.33

−1.58**

0.36

Community × caucus state

0.24*

0.14

0.38**

0.16

Political attitude

  

0.23**

0.08

Political interest

  

0.41**

0.08

Political activism

  

0.53**

0.08

Atittude × caucus state

  

0.09

0.22

Interest × caucus state

  

0.19

0.28

Activism × caucus state

  

−0.42

0.21

Control vars.

    

 Black

0.47**

0.15

0.40**

0.16

 Hispanic

0.27

0.19

0.26

0.20

 Education

0.29**

0.06

0.16**

0.06

 Income

0.03**

0.01

0.02

0.01

 Female

0.15

0.12

0.27**

0.13

 Age

0.03**

0.00

0.03**

0.00

 Married

0.23*

0.13

0.29**

0.14

 Children

0.04

0.14

0.14

0.15

 Residence

0.05

0.04

0.08*

0.05

 Republican

0.84**

0.24

0.25

0.28

 Democrat

1.18**

0.22

0.54**

0.26

Observations

1,588

  

1571

Log Likelihood

−914

  

−835

Note: The above table presents estimates from logistic regressions in which turnout in either caucuses or primaries is the dependent variable. ** Denotes significance at the p < 0.05 level, * denotes significance at the p < 0.10 level. Model 6 suggests that community engagement is more associated with caucus attendance than primary participation, even when controlling for levels of political engagement; however, political engagement is similarly related to the two forms of election participation

The estimates in Table 2 tell an important story. Community involvement, like political engagement, is relevant both to turnout in primaries and turnout in caucuses. Whereas political engagement seems to have no greater bearing on caucus participation than primary participation, community involvement does matter more for caucuses. Consistent with the theory of community involvement articulated above, the evidence in Table 2 suggests that it is community involvement, not political involvement, that distinguishes caucus participants from primary voters.

As an indication of the marginal effect of community involvement on participation, Fig. 4 shows expected probabilities for turning out generated from Model 6. Here, a hypothetical person who is average on all political variables and control variables but answered ‘no’ to all community involvement items is estimated to have a 12% probability of caucusing, and a 39% probability of primary voting. The hypothetical average person who answered ‘yes’ to all community involvement items has a 50% probability of caucusing and a 52% probability of primary voting. The implication here is that a person who is deeply involved in community endeavors is essentially as likely to go to a caucus as a primary in spite of the much higher costs associated with a caucus. Figure 4 shows clearly that community involvement seems to have a much larger effect on caucus participation than on primary participation.
https://static-content.springer.com/image/art%3A10.1007%2Fs11109-011-9175-8/MediaObjects/11109_2011_9175_Fig4_HTML.gif
Fig. 4

Expected probabilities for caucus/primary participation by level of community engagement. Note: This graph shows expected probability estimates with 95% confidence intervals generated from Model 6 in Table 2. With control variables set at their means, the graph indicates how expected caucus and primary participation increases with increasing levels of community engagement. With higher levels of community engagement, participation in caucuses increases more than participation in primaries

Confronting Four Sources of Bias

There are four possible sources of bias that I have not, to this point, considered. In this section, I will address these issues and re-estimate models from Tables 1 and 2 under more restrictive assumptions. While I will not here be able to fully accommodate these important concerns because of data constraints, I seek to acknowledge the limits of the analysis and to point out directions for future work.

The Calendar

First, the tests above do not account for the nomination calendar. Primaries and caucuses are held over several months, and some positions in the calendar may draw a different crowd of voters than other positions. Of greatest concern here is that the national Republican contest ended before the Democratic contest, and perhaps the Republican contests that took place after John McCain clinched the nomination (almost all of which were primaries) attracted a particularly active cohort of voters, thus biasing the results. I re-run the models for the truncated set of states in which both parties’ nominations were actively contested (i.e. no candidate had yet clinched a party nomination). Models A and B in Table 3 in the Appendix mimic Models 2 and 4 in Table 1, but exclude contests that took place after March 4th, 2008, when the Republican contest was no longer competitive. Appendix Table 5 addresses issues of bias in models testing the community engagement hypothesis, and Model I in that table estimates coefficients for contests held on or before March 4th. In all of these cases, the results stay substantively the same as in Tables 1 and 2.

To further allay the concern that results may be biased due to state contests occurring at different points in time, I again truncate the data to include only those contests occurring on a single day: “Super Tuesday.” Super Tuesday is the day, early in the election season, on which more states hold elections than on any other day of the primary calendar. Once again, the results for the political engagement models, identified as Models C and D in Appendix Table 3, remain substantially the same. These models suggest that even if one looks only at the competitive contests, even those occurring just on one day, it is still the case that caucus goers and primary voters look similar with respect to political engagement. As for the community engagement hypothesis, we see that in the Super Tuesday-only model, identified as Model J in Appendix Table 5, the coefficient on the Community × Caucus State interaction term remains positive, but is no longer statistically significant at conventional levels. Of course, as the dataset is truncated more and more, the small sample size makes it difficult to identify clear effects.

The Election Year

It is worth lingering on whether 2008 was a unique election year in a sense that findings identified here should not be expected to apply to other contexts. Perhaps the 2008 election mobilized many more citizens than typical Presidential elections, and so extreme political activists made up a smaller proportion of the electorate than usual. It is possible that the heightened participation of ordinary citizens makes the pool of caucus and primary participants more similar than in other years, leading to the null finding in Table 1. While acknowledging this possibility, it nevertheless seems inconsistent with conventional cost-benefit rationalizations that an election mode involving very high costs would attract citizens with similar levels of political interest as an election mode with comparatively lower costs. Furthermore, the fact that previous studies of other election years also found similar levels of political engagement among caucus goers and primary voters buttresses the finding here (e.g. Marshall 1978).

A related concern, however, is that because the contests in the 2008 nomination fights were so unusually competitive, the results here are not likely to apply broadly to contests in the future or even in the past. Indeed, if more election years in the past witnessed competitive nomination contests like 2008, data would more likely be available to test the hypotheses articulated here for multiple years. To this concern, I reiterate that conclusions drawn from one election year are always tentative and future research should focus on gathering samples in states that will hold competitive contests, such as the states that typically host the early primaries and caucuses. 16

State Selection

A third concern relates to state selection in holding a caucus or a primary. The estimation methods used here accommodate differences in the average level of political engagement and community engagement in caucus states and primary states; however, the methods do not acknowledge that states are not randomly assigned a caucus system or a primary election system. One may wonder if states that hold caucuses are also states that attract a particularly engaged cohort of citizens. 17 Estimates in Tables 1 and 2 would be biased if there is something about caucus states that affect attitudes and behaviors of caucus participants differently than it affects attitudes and behaviors of non-participants. This concern is especially valid with respect to the latter part of this essay in which the positive relationship between community engagement and caucus participation is observed.

I attempt to accommodate the concern of state selection first by excluding primary states that are very different from caucus states. In particular, most Americans live in a few large states, none of which hold caucuses. The ten largest states contain the majority of citizens and all of them hold primaries. For the political engagement hypothesis, I run the analysis while excluding the ten most populous states, and results are displayed in Models E and F in Table 4 in the Appendix. Model K in Table 5 is the equivalent test for the community engagement hypothesis. The results remain the same, with the Community × Caucus State still positive and statistically significant, though now at the 10% level.

Models G, H, and L are even more restrictive. For each caucus state in this analysis (excluding Hawaii and Alaska, for lack of neighbors), I select a neighboring primary state. 18 Models G, H, and L test the effect of political and social engagement on caucus and primary turnout just for these matched states. In the CCES models in Table 4 some of the coefficients on the interaction terms reach statistical significance, though in two of the three cases the significant coefficients are negative, indicating a larger effect of political engagement on turnout in primaries than in caucuses. For the community engagement hypothesis in the matched-state estimation, only 322 individuals are left in the NES sample, and the key coefficient of interest falls to nil.

Though the attempts in Appendix Tables 3, 4, and 5 to confront issues of the calendar and of state selection do not wholly diffuse concerns of biased estimates, to the greatest extent allowable with these data, they address important measurement issues, and the core insights stand. More data must be collected to provide further confidence in the truth of the claim that community engagement matters more for caucus participation than for primary participation. The estimates shown here provide compelling, but tentative, support.

Party Differences

Aside from controlling for party identification in the logistical regression models, I have not accounted for differences attributable to party. It may be that differences between primary/caucus participants and non-participants vary by party, such that the effects of political engagement are concentrated in one party. There are two reasons why it is difficult to accommodate this concern. First, doing so would necessitate determining the pool of citizens who decide to attend or not attend a Democratic (Republican) caucus or primary. One might do this by observing the pool of Democratic identifiers, for example, and comparing those who participate in the caucus or primary and those who do not. Such a design would ignore that political independents can participate in party nominations in some states and that some states open their primaries and caucuses to members of the opposite party. In the CCES and NES surveys, 7–10% of respondents identifying as Republican claimed to have participated in a Democratic primary or caucus.

A second reason why it is difficult to incorporate party differences into these models is that in order to capture the nuances of open versus closed contests, the data would have to be sliced too thin. Studying caucuses using national surveys such as the NES requires a degree of generalization on account of the sample sizes. Without firm reason to believe that the issue of party is biasing results in favor of my hypothesis, I submit the findings here as average effects with respect to party, and I leave a more nuanced analysis to future research.

Conclusion

Standard models of political engagement posit that as participatory costs rise, citizens must derive explicitly political benefits to overcome such costs. Interest in politics, knowledge of the political system, and ideologically extreme views ought to be correlated with overcoming participatory costs. Such a position is found in classic models offered by Verba et al. (1995) and Rosenstone and Hansen (1993). Here, I simply aim to ask the question: how explicitly political must the benefits/resources be in order for citizens to overcome participatory costs? On measures of political engagement ranging from partisanship to political donating, from petition signing to bumper-sticker displaying, caucus participants and primary participants are very much alike. This is not what one might expect from a model that closely ties participatory costs to political resources.

In this essay, I have suggested a theory for why, in spite of the high costs of caucus participation, participants are politically similar to primary voters. The reason is that citizens may not participate in political activities like caucuses strictly for political reasons. Some people attend caucuses because they want to see their neighbors, or because they are under social pressure from their neighbors to attend, or because they want to be part of a community event, or because they want to see and be seen. The broader lesson for political scientists is that viewing political activity through too narrow a political lens might lead to faulty inferences. In other words, ‘politics’ is not always the reason that people participate in politics.

This finding adds a wrinkle to the debate about the distinctiveness of caucus goers. The efforts of Marshall (1978), Redlawsk, Bowen and Tolbert (2008), Panagopoulos (2010), Mayer (1996) and others have generated a mixed set of results about how different caucus goers are from primary voters. This research contributes the finding that while caucuses do not seem to be the domain of the unusually politically oriented, they do seem to be the domain of the community oriented. Whether this result is robust to other electoral environments and whether it affects candidate selection are tasks left to future work.

There are several key lines of inquiry that ought to pick up where this essay leaves off. First among them is an investigation into the relationship between community involvement and candidate choice. Are certain candidates more attractive to citizens who are involved in community activities? Do candidates who come across as more community-oriented do better in caucuses than less neighborly candidates? Can Obama’s success in the 2008 caucuses be attributed to the community activists who were present there?

A second line of inquiry is the mitigating effect of party. In this work, I have compared the institutional design of caucuses with that of primaries and suggested how the differences in the rules affect who participates. But there is a second dimension to caucuses in that Democratic caucus rules differ somewhat from Republican caucus rules. On the Republican side, the actual act of voting typically remains private. On the Democratic side, not only is voting public but in many cases participants are invited to persuade their neighbors to change their vote choice. For this reason, the social benefit of caucuses may be different in Republican caucuses than Democratic caucuses, though this is not a proposition that can be tested well with the present data. Nevertheless, this would be a valuable research path to pursue with other sources of data.

Though the caucuses produced vastly different outcomes in both party nominations than the primaries in the 2008 election, it is not the case that the difference is attributable to the caucuses being overrun by ideologues and political activists. Perhaps this evidence should give pause to those who claim that the caucuses are undemocratic and unrepresentative. If through reform efforts all of the 2008 caucuses were suddenly turned into primaries, the costs of participation would be dramatically reduced, but the implication of this research, paralleling Berinsky (2005), is that such a change would not lead to a larger proportion of marginally active citizens taking part in the electoral process. Reducing representational bias requires something different from institutional reform; it requires civic and community engagement.

Footnotes
1

David Broder, “Wait for New Hampshire”, Washington Post, 3 January 2008.

 
2

“Transcript for March 9 2008,” Meet the Press, MSNBC 9 March 2008. See also, Julie Bossman and Jeff Zeleny, “Clinton Works Wyoming to End Caucus Streak,” New York Times, 8 March 2008.

 
3

Consult Norrander (1996) for a comprehensive review of the literature on “post reform-era” nomination research. See also Shafer and Wichowsky (2009).

 
4

The number of caucuses and primaries for each party do not add up to 51 (for states plus Washington, DC). Parties in some states have both primaries and caucuses. In a few states, the Republican Party does not have either primaries or precinct-level caucuses. In these states, party leaders at county or state conventions nominate delegates.

 
5

Consider the Addonizio et al. (2007) study in which researchers organized carnivals at voting precincts. Precincts that hosted carnivals had higher levels of turnout than control precincts, apparently due to a purely peripheral participatory benefit.

 
6

Because precinct caucuses are not institutions exhaustively examined in the literature, I conducted 30 in-depth interviews with Iowans in Sioux County and Louisa County in April 2008, following the precinct caucuses, in order to gain some insights into the caucus experience beyond reports in the news media. Interviewees were selected randomly from the voter registration files in these counties, and were conducted by telephone.

 
7

Note that both the CCES and NES surveys are fielded in the Fall, just before and after the November general election. Primaries and caucuses are held up to ten months earlier than the general elections, which requires that we assume a degree of accuracy in respondent recall with respect to their election participation. See Atkeson (1999) for a study of misreporting in primaries.

 
8

The six states include Idaho, Montana, Nebraska, Texas, Washington, and West Virginia.

 
9

Washington State Democrats, however, do not use the results of the primary in the selection of their party delegates.

 
10

The NES respondents from caucus states are distributed fairly evenly across these five states, with 32% coming from Colorado, 11% from Kansas, 16% from Minnesota, 30% from North Dakota, and 11% from Nevada.

 
11

Appendix 2 section details variable sources and coding decisions.

 
12

Survey weights are used in analyzing NES and CCES data throughout this study. In both cases, the weights are intended to make the sample more representative of the population. For details, consult Lupia et al (2009) and Ansolabehere (2009b).

 
13

The first two summary measures merely combine two variables into one, which is essentially equivalent to averaging the two. The summary measure for political activism is based on summarizing 5–6 variables, and in both the CCES and NES, the variables load onto single factors. In the CCES, the first factor has a Eigenvalue of 2.30, with the next factor having a Eigenvalue of 0.86. In the NES, the first factor has an Eigenvalue of 1.93, with the next factor having an Eigenvalue of 0.92.

 
14

As a robustness check, each of the three political engagement variables can be included without the other two in the model. In no case do any of the interaction terms show a statistically significant positive effect.

 
15

Expected probabilities and first differences are generated using the Zelig package in R (Imai et al. 2007).

 
16

A similar concern is whether the results are generalizable beyond Presidential contests. Given the heightened salience of Presidential contests, it is unclear whether a caucus-primary study of down-ballot races would be comparable to this study. Niven (2001) finds that mobilization of unengaged voters is particularly difficult in state house primaries. In down-ballot races, the peripheral benefits might be so limited that they alone could rarely overtake the costs of participation.

 
17

For a related concern, see Kenney and Rice’s (1985) discussion regarding states that were early adopters versus late adopters of primary elections.

 
18

For Colorado, the matched state is New Mexico; for North Dakota, South Dakota; for Minnesota, Wisconsin; for Kansas, Oklahoma; for Iowa, Missouri; for Maine, Vermont; for Wyoming, Utah; for Nevada, Arizona. Note that not all of these states are included in the NES.

 

Acknowledgements

The author thanks Stephen Ansolabehere, Gabe Lenz, Brian Schaffner, and participants in Harvard’s Political Psychology and Behavior Workshop for their helpful comments. Additional thanks to Greg Distelhorst, Jennifer Hochschild, Orit Kedar, David Mayhew, Robert Putnam, and Patrick Warren for their advice on a course project from which this paper emerged.

Copyright information

© Springer Science+Business Media, LLC 2011