William ‘Boss’ Tweed captured the importance of nominating candidates when he said he did not care who ‘did the electing’ as long as he ‘got to do the nominating’. In contemporary American elections, voters – not party bosses – do the nominating. This raises a host of questions about the virtues and vices of primary elections. Critics of primary elections have long questioned whether primary voters are representative of the broader party.Footnote 1 Polsby argues that ‘a lack of demographic representativeness in a primary electorate may produce significantly different results in the types of candidates chosen to lead the party’.Footnote 2
Two particular concerns among commentators and some scholars are that primary voters are ideologically extreme, and that their influence over the nominating process produces ideologically extreme candidates and more polarization between the parties in Congress and state legislatures. For this reason, some advocates argue that the primary process should be reformed in order to reduce polarization.
However, despite consistent skepticism about the representativeness of primary electorates, early research challenges this view. Drawing on data from the 1976 and 1980 elections, Geer and NorranderFootnote 3 found that a party’s primary voters were not more ideological or partisan than general election voters who identified with that party or voted for its presidential candidate – what Geer called the ‘party following’ – or than general election voters in the party who did not vote in the primary. Norrander concludes: ‘Fears about extremist primary voters selecting extremist candidates unpalatable to the more moderate general election voters are unsupported. Primary voters just are not more ideologically extreme.’Footnote 4
Since this research was conducted, however, primary turnout has declinedFootnote 5 and the parties have become more ideologically sorted.Footnote 6 These changes in the composition of the parties raise the possibility that the primary electorate is no longer representative of rank-and-file partisans, and a new scholarly debate has emerged. Prominent scholars have argued that differences between the primary electorate and the party rank and file are large and important. For example, Mann argues that ‘Since primary electorates are skewed toward each party’s ideological pole, the appearance of a credible threat in the primary election will push the incumbent in the same direction.’Footnote 7 And Fiorina and Abrams argue that these differences are particularly large in congressional primaries, saying ‘Some studies of presidential primary voters have concluded that the primary voters are not as unrepresentative as popular commentary assumes, but when we are talking about a sixth to a tenth of the electorate voting in a subpresidential primary – often split between the two parties – the likelihood is that we are talking about a primary electorate composed disproportionately of hard-core wing-nuts in the two parties.’Footnote 8
Some recent evidence is mixed, however. Based on an analysis of state exit polls from the 2000 presidential primaries and the 2004 Democratic presidential primary, AbramowitzFootnote 9 sides with the earlier research, arguing that ‘the differences in ideological identification between primary and general election voters were very small’. But two more recent studies argue the opposite. Drawing on the 2010 Cooperative Congressional Election Study (CCES), Jacobson finds that self-reported Republican primary voters in that election year were more ideologically extreme than self-reported general election voters who said they did not vote in those primaries. He writes of ‘primary electorates in which ideological extremists are overrepresented’.Footnote 10 Hill draws on the 2010 and 2012 CCES, but uses validated rather than self-reported turnout data. His findings are mixed: raw estimates show few differences between primary voters and the broader party, but estimates from a hierarchical model of policy attitudes suggest larger differences. He argues that ‘primary voters and primary electorates are less centrist […] than party voters in the general electorate’.Footnote 11
We are able to improve on these recent studies. First, our evidence is broader. We study more elections (four, compared to one or two), which means that we have multiple observations for each party primary and both presidential and congressional primaries. Our results are not the fluke of a single election cycle. We examine a broader set of indicators than existing work, ensuring that our results are not due to the idiosyncratic nature of a particular political issue or measure. The five surveys we analyze have large enough samples to encompass more states than are typically present in the state exit polls. Our focus on the national level combined with very large samples in every election gives us enough power to detect differences across groups without the assumptions needed in a hierarchical model, as in Hill.Footnote 12 Finally, we can rely on validated turnout rather than self-reported turnout, as in Jacobson,Footnote 13 which has been shown to be an unreliable measure of turnout (Ansolabehere and Hersh, Reference Ansolabehere and Hersh2012; Vavreck, Reference Vavreck2007).
We show that primary voters are not demographically distinct or ideologically extreme compared to those who identify with the party or who voted for its presidential candidate in the general election, or than those who identify with the party and voted in the general election but not in the primary. The only substantial difference is that primary voters report more interest in politics. These patterns emerge in both presidential and midterm years.
In contrast to other recent studies,Footnote 14 our findings suggest that the ideological extremity of primary voters has not changed in the three decades since the early studies by Geer and Norrander.Footnote 15 Of course, our findings cannot speak to the impact of simply having a party nomination process prior to the general election. Nor can we compare primary elections to other types of nomination processes – such as conventions, caucuses or smoke-filled backrooms. Nonetheless, we provide considerable evidence that primary electorates are not ideologically unrepresentative of the broader party. This implies that reforms to the primary process are unlikely to influence polarization.
RESEARCH DESIGN
We use data from five large surveys of the American public: the 2008 Cooperative Campaign Analysis Project (CCAP)Footnote 16 and the 2008–14 CCES. After weighting, CCAP respondents are representative of registered voters and CCES respondents are representative of the American public.Footnote 17 For each dataset, the survey provider matched respondents to voter file data that contain validated primary and general election turnout.Footnote 18
These data offer four main advantages. First, they encompass two presidential and two midterm elections and allow us to separate presidential and congressional primary voters in states that hold presidential and congressional primaries on different dates in presidential election years. Secondly, they contain large enough samples to estimate the impact of primary rules, which vary across states. Thirdly, they feature many measures of political attitudes. Finally, these data allow us to rely on validated turnout rather than potentially biased self-reports. The validated turnout data reveal substantial overlap in the primary and general electorates. In the 2008 CCAP, 68 per cent of validated general election voters also voted in their state’s primary.Footnote 19 The overlap between the two electorates means that roughly a third of 2008 general election voters voted ‘only’ in the general election and not in the primary. Any differences between the primary and general electorates must therefore manifest themselves in this relatively small group of voters.
We compare primary voters to two definitions of ‘the party’ found in previous literature: (1) general election voters who self-identify with a party or voted for that party’s candidate in the general election; and (2) a smaller subset of those voters who only voted in the general election but not in the primary. Following Geer,Footnote 20 we call the former the ‘party following’Footnote 21 and the latter ‘general-only voters’.
COMPARING PRIMARY AND GENERAL ELECTION VOTERS
Tables 1 and 2 show the results of our analysis. They compare the demographic and ideological profiles of primary voters with voters in the party following and general-only voters in presidential and congressional elections, respectively. We omit tests of statistical significance, given that each survey contains tens of thousands of respondents and so the quantities presented here are very precisely estimated.
Table 1 The Characteristics of Primary and General Electorates in Recent Presidential Elections
Table 2 The Characteristics of Primary and General Electorates in Recent Congressional Elections
Because primary voters are frequently characterized as political activists, we might expect them to be older, better educated and more interested in politics. But although primary voters were about 6–8 years older than those who voted only in the general election, they were only 1–4 years older than the average in the broader party. Primary voters were also only a few points more likely to have a college degree than those who voted only in the general election or than the party following. Larger differences emerge with regard to campaign interest and campaign donations. For example, 63 per cent of Democratic presidential primary voters in the 2008 election said they were very interested in the campaign, compared to 44 per cent of those who voted in the general election but not the primary. This gap appears among Republicans and in other elections as well.
But differences in campaign or political interest do not translate into large differences in three different measures of political attitudes. In these elections, the average Democratic primary voter’s self-reported symbolic ideology on a five-point scale from very liberal to very conservative was only slightly to the left of Democrats who voted in the general election but not in the primary. Indeed, in the 2008 elections, Democratic presidential primary voters actually identified as more conservative than the party following, on average (top left panel of Table 1). The differences among Republicans were slightly larger but still small in absolute terms. On average, Republican primary voters in the 2008 election were 0.17 points more conservative than Republicans who voted in the general election but not the primary (4.13 vs. 3.96). But this difference only represents 2.5 per cent of the five-point scale and 11 per cent of the gap between the followers of each party. The difference between Republican primary voters and the Republican party following was even smaller. There were similarly modest differences in congressional elections (Table 2).
There are also small differences in issue positions among these groups, and not always in the direction that the conventional wisdom would presume. For instance, Democratic primary voters in 2008 were slightly less supportive of civil unions than the broader party following, and the differences on other issues were generally in the single digits. The lone exception was that 2008 Republican primary voters were fifteen points less likely than general-only voters to favor raising taxes on the wealthy. However, this difference was dwarfed by the forty-six-point difference between the parties.
Finally, there are few notable differences in the one-dimensional ideal points of these different groups based on their responses to a larger set of issue questions.Footnote 22 For example, Figure 1 shows that primary voters were only a bit more ideologically extreme than party followers in the 2008 presidential and 2010 congressional elections. In 2008, Democratic primary voters were slightly less extreme than the party following.
Fig. 1. Ideal points of primary voters and the party following in 2008 and 2010. Note: the dashed lines are the mean ideal points of primary voters in each party and the solid lines are the mean ideal points of the party following in the general election. Sources: 2008 CCAP and 2010 CCES.
To illustrate the very modest magnitude of these differences, it is useful to compare them to ideological differences among US senators using DW-Nominate scores. For example, the difference between Democratic primary voters and the party following in the 2012 congressional primary election – 0.15 standard deviations – is the largest one among Democrats in these elections. This is almost exactly the difference between Senators Jeff Merkeley and Ron Wyden, both Democrats from Oregon; Senator Merkeley is the more liberal of the two. The difference between California Senators Barbara Boxer and Dianne Feinstein is more than two and a half times as large. Among Republican voters, the greatest difference (0.2 standard deviations) also occurs in the 2012 congressional election. This is essentially the difference between Senator John McCain of Arizona and Senator Kelly Ayotte of New Hampshire, where McCain is the more conservative. The difference between Senators Mitch McConnell and Rand Paul, both of Kentucky, is more than five times as big.Footnote 23
In general, the 2008–14 data show that primary electorates in recent congressional elections are not substantially more ideologically extreme, relative to the party following, than the primary electorates in presidential elections. Similarly, primary electorates in congressional elections do not appear to be more extreme in midterm election years than in presidential election years. This runs contrary to fears that a smaller turnout in midterm elections enhances the power of the ideological extremes.
Why do these results differ from those of Jacobson and Hill,Footnote 24 even though we are analyzing some of the same surveys? In contrast to Jacobson, we use validated turnout data. As we describe in the online appendix, self-reported turnout produces larger differences between primary voters and the party following. And unlike Hill, we rely on simple disaggregated means and very large sample sizes, rather than a hierarchical model.Footnote 25
Even though there were few substantive differences between primary voters and the party following, larger differences might emerge in particular types of primaries. Table 3 compares the mean ideal points of people who voted in the congressional primary to those of the party following in closed, semi-closed and open primaries – pooling observations from the 2008–14 CCES (four elections) across states and years. The differences between primary voters and the party following are not much greater in closed primaries than in open primaries, even though closed primaries are thought to create larger differences by limiting the primary electorate to registered partisans. This null effect of primary rules confirms previous research.Footnote 26 As Kaufman and colleagues conclude: ‘[…] the key to greater ideological representativeness is not the rules alone’.Footnote 27
Table 3 Association between Primary Type and Ideal Points
CONCLUSIONS
In 1956, V.O. Key wrote skeptically about the primary system:
The elevation of such minorities to power within the nominating process through the smallness of total participation and bias may […] throw into office the most improbable sorts of characters who have won nominations through the vagaries of primaries.Footnote 28
That skepticism has persisted for many years. After the 1984 presidential election, a supporter of Jack Kemp said ‘The Republican presidential primary process remains a right-wing orgy.’Footnote 29 After the 2012 election Republicans worried that primaries ‘push their presidential nominees far to the right’ and ‘produce lackluster Senate candidates’.Footnote 30 In 2016, the success of Donald Trump led conservative columnist George Will to blame open primaries for including less faithful, and presumably more moderate, Republicans.Footnote 31 Similar debates about primary rules took place among Democrats.Footnote 32 Meanwhile, reformers concerned about polarization advocate reforms to primary elections. Phil Keisling, formerly Oregon’s Secretary of State, writes: ‘Want to get serious about reducing the toxic levels of hyper-partisanship and legislative dysfunction now gripping American politics? Here’s a direct, simple fix: abolish party primary elections.’Footnote 33
Clearly there is a recurring debate among political scientists and commentators about whether primary electorates are representative of their parties. Our evidence does not confirm repeated claims that the primary electorate is ideologically extreme or otherwise distinctive – even in the context of today’s polarized politics. In 2008, 2010, 2012 and 2014, primary voters were ideologically representative subsets of the broader party following. Moreover, the ideological composition of primary electorates did not depend very much on primary rules or the type of office. Our findings confirm Norrander’s review of past work that ‘rather than being a more ideologically extreme proportion of the electorate, presidential primary voters are more aptly described as the slightly more interested and knowledgeable segment of the electorate’.Footnote 34 To be sure, our claim is not that primaries have no consequences for the candidates who run or the candidates who win. Moreover, primaries might be problematic for other reasons, such as that they do not provide sufficient deliberation within the party or a thorough enough review of each candidate’s qualifications.Footnote 35 Nevertheless, our findings should serve to allay one concern about primary elections: that they empower ideological extremists within the parties.
