Numerous models forecast general election outcomes by employing a variety of economic and political measures to make accurate predictions about whether the party in control of the White House will retain or lose the presidency (for an overview see Campbell Reference Campbell2012). In many ways forecasting presidential nominations presents a more challenging task. Important individual-level cues such as partisanship or systemic-level factors such as economic growth or the popularity of the incumbent are helpful in understanding why a voter might choose Bill Clinton over George W. Bush in 1992. Unfortunately, they are not useful in explaining why the same individual picked Paul Tsongas over Bill Clinton or Tom Harkin nine months earlier in the New Hampshire primary (Steger, Dowdle, and Adkins Reference Steger, Dowdle, Adkins, Mayer and Bernstein2012).
While the McGovern-Fraser reform movement of the early 1970s created a new system of presidential nominations designed to increase the role of voters in picking party nominees, a period of stability in the nomination process of both parties’ emerged by the end of the 1980s (Barilleaux and Adkins Reference Barilleaux, Adkins and Nelson1993). As these contests became more routinized, a number of scholars attempted to forecast the results of the presidential primary season by utilizing factors such as polling, financial resources, and elite support (Adkins and Dowdle Reference Adkins and Dowdle2000, Reference Adkins and Dowdle2001a, Reference Adkins and Dowdle2001b, Reference Adkins and Dowdle2005; Mayer Reference Mayer and Mayer1996; Steger Reference Steger2000; see Steger Reference Steger2008 for a comparison of the forecasts generated by the different models). Momentum from performing well in early primaries was also found to play an important role in determining nomination outcomes (Bartels Reference Bartels1988), though there is some controversy about the precise effect of particular contests (Adkins and Dowdle Reference Adkins and Dowdle2001a; Christenson and Smidt Reference Christenson and Smidt2012; Hull Reference Hull2008).
At first glance, current events appear to have altered this equilibrium in at least two important ways. First, super PACs, a relatively new type of political committee that arose from the Speechnow v FEC and Citizens United v FEC court decisions in 2010, should alter the impact of traditional sources of campaign finance (Dwyre and Braz Reference Dwyre and Braz2015). Second, the Republican elite has arguably fragmented in recent years, which should affect elite support on the process (Steger Reference Steger2015). Since traditional forecasting models encountered difficulty predicting the 2004 Democratic nomination correctly (Steger Reference Steger2008), these new factors should make predicting recent nomination outcomes even more challenging.
To forecast presidential nomination outcomes this research employs two OLS regression models that use the “open” presidential nomination contests from 1980–2012 and then applies the estimates to the 2016 Democratic and Republican presidential nomination contests to create forecasts for each.
MODEL SPECIFICATION
To forecast presidential nomination outcomes this research employs two OLS regression models that use the “open” presidential nomination contests from 1980–2012 and then applies the estimates to the 2016 Democratic and Republican presidential nomination contests to create forecasts for each. Footnote 1 The models examine Democratic and Republican contests from 1980 to 2012 inclusive, with the exception of the 1980, 1996 and 2012 Democratic races and the 1984, 1992 and 2004 Republican contests. Generally, nomination races with sitting presidents are foregone conclusions and bias the predictions by inflating the R-square statistic and thereby skewing the model’s results (Adkins and Dowdle Reference Adkins and Dowdle2000). Footnote 2
The models use a number of preprimary indicators (e.g. a candidate’s standing in the preprimary Gallup preference polls, percentage of party endorsements, and fundraising success) and indicators of early campaign success (e.g. finishes in the Iowa caucuses and New Hampshire primary) to generate two forecasts: a preprimary forecast and post-New Hampshire primary forecast. Footnote 3 In each model the dependent variable is the APV (percentage of the aggregate primary vote) received by each candidate for the Democratic and Republican presidential nominations, excluding the results of the New Hampshire primary.
Poll Results
Scholars have found a strong relationship between poll standing in the preprimary season and predicting presidential nomination outcomes (Mayer Reference Mayer and Mayer1996).
H 1 : The greater a candidate’s standing in the preprimary preference polls, the higher the percentage of the aggregate primary vote that a candidate will receive.
The Poll Results variable represents each candidate’s support among self-identified partisans in the average of preference polls taken for their party’s nominee for the 1980–2012 races during the fourth quarter of the year prior to the start of the nomination (e.g., Gallup poll averages for October 1–December 31, 1979 for the 1980 Republican nomination contest). Footnote 4
Campaign Expenditures
Candidates who raised the most money in the preprimary period typically won their party’s nomination, but in some instances candidates do underperform (Adkins and Dowdle Reference Adkins and Dowdle2002; Reference Adkins and Dowdle2008). To better measure the differential effects of money spent during the preprimary period and the start of the primary season, and to lessen problems with multicollinearity, we separated total fundraising into two variables: campaign expenditures and cash reserves.
H 2 : The larger the amount of money spent in the preprimary season relative to their opponents, the higher the percentage of the aggregate primary vote that a candidate will receive.
The Campaign Expenditures variable is calculated as a percentage of the campaign funds that each candidate spent during the preprimary period, relative to the total raised by all candidates during that time. Footnote 5
Cash Reserves
Cash reserves represent unrealized potential of the campaign to affect the candidate’s performance in the future (Adkins and Dowdle Reference Adkins and Dowdle2001b).
H 3 : The larger the amount of unspent campaign funds at the end of the preprimary period that candidates have relative to their opponents, the higher the percentage of the aggregate primary vote that a candidate will receive.
While early spending often has a tenuous relationship with nomination outcomes, a number of studies (Adkins and Dowdle Reference Adkins and Dowdle2000; Steger Reference Steger2002) find a strong positive relationship between the amount of financial reserves a campaign has at the end of the preprimary and success during the primary season. In order to control for both inflation and the context of individual election cycles, Cash Reserves are calculated as a percentage of the unspent money that each candidate has available at the end of the fourth quarter of the year prior to the election, relative to the cash reserves of the entire nomination field at the end of that same period. Footnote 6
Endorsements
Despite the changes following the McGovern-Fraser reforms, party elites still manage to play a crucial role in shaping nomination outcomes (Cohen et al. Reference Cohen, Karol, Noel and Zaller2008; Steger Reference Steger2007).
H 4 : The greater number of elite endorsements that candidates have relative to their opponents by the end of the preprimary season, the higher the percentage of the aggregate primary vote that a candidate will receive.
Endorsements represents the endorsements, defined as the unweighted total of House, Senate and gubernatorial endorsements, a candidate has as a percentage of the total endorsements in that contest by December 31 of the year prior to the election. Footnote 7
Iowa
The Iowa caucuses and New Hampshire primary are important early tests of candidate strength. Since Carter’s victory in 1976, many candidates spend resources disproportionate to the numbers of convention delegates awarded trying to win support of voters candidates in these two states or at least to try to beat popular expectations (Steger, Dowdle, and Adkins Reference Steger, Dowdle and Adkins2004). The first variable to measure candidate strength in Iowa represents whether candidates won the caucuses. Buell (Reference Buell and Mayer2000) contends that recent winners there and in New Hampshire receive a “bounce” going into the next round of primaries and caucuses.
H 5 : The winner of the Iowa caucuses will receive a higher percentage of the aggregate primary vote than other candidates in the field.
Iowa Win takes the form of a dummy variable with the winner coded as a “1” and the remainder of the cases coded as “0.” The second measure is the candidates’ share of the vote in the Iowa caucuses, which reflects the variation in candidate performance.
H 6 : The higher percentage of the vote that a candidate receives in the Iowa caucuses, the higher the percentage of the aggregate primary vote that a candidate will receive.
New Hampshire
Comparing the effects of Iowa and New Hampshire in nomination forecasts, Adkins and Dowdle (Reference Adkins and Dowdle2001b) found that the results of the New Hampshire primary produced a statistically significant impact on predictive capacity, but the Iowa caucuses did not.
H 7 : The winner of the New Hampshire primary will receive a higher percentage of the aggregate primary vote than other candidates in the field.
New Hampshire Win takes the form of a dummy variable where the winner coded as a “1” and the remainder of the cases as “0.” The second measure is the candidates’ share of the New Hampshire primary vote, which reflects the variation in candidate performance.
H 8 : The higher percentage of the vote that a candidate receives in the New Hampshire primary, the higher the percentage of the aggregate primary vote that a candidate will receive.
DATA ANALYSIS
Despite the changes that occurred in the nomination process over the past few years, the traditional dynamics of models that forecast presidential nomination outcomes persist. The results of the two OLS models are presented in table 1. The Poll Results and the Endorsements indicators are significant in both models, which echoes the results of previous works on the topic (see Steger Reference Steger2008). Cash Reserves, which had been significant in prior studies (Adkins and Dowdle Reference Adkins and Dowdle2001a; Reference Adkins and Dowdle2001b), is not significant in the preprimary model but is in the post-New Hampshire model. Campaign Expenditures is not significant in either model, which suggests higher levels of spending in the preprimary period do not correlate with winning more primary votes when other factors are accounted for in a multivariate model. This finding reminds us that some hopefuls (such as Phil Gramm in 1996) who perform poorly in early polls continue to fare poorly when voters begin to cast ballots in spite of spending large amounts of money in the preprimary period. Footnote 8
Early popular support, preprimary elite endorsements and a large campaign war chest entering the formal campaign remain strong predictors of successful nominees.
Table 1 OLS Forecasting Models of Aggregate Primary Vote, 1980–2012
Note: Coefficients are unstandardized ordinary least squares (OLS) coefficients; t scores are in parentheses ( ), standardized beta coefficients are in brackets [ ], SEE=standard error estimate. Significant at *p < .05, **p < .01.
The second model in table 1 also measures whether a candidate won Iowa or New Hampshire along with the impact of the votes that each candidate received in each contest. The positive effects of winning Iowa bolsters Hull’s (Reference Hull2008) claim that Iowa does play an important role in nomination outcomes, but the actual vote percentage from Iowa is not significantly correlated with the overall primary results, which is consistent with prior research (Adkins and Dowdle Reference Adkins and Dowdle2001a). On the other hand, both ordinal and interval-level finishes in New Hampshire are positively correlated with nomination outcomes as previous studies have indicated (Steger, Dowdle, and Adkins Reference Steger, Dowdle and Adkins2004).
In table 2 we also demonstrate the two models’ abilities to make ordinal predictions for the open races from 1980–2016. The preprimary model correctly predicts 10 of the 12 primary vote winners and all eight of the winners from 1980–2000. It misses Kerry in 2004 by a wide margin and predicts that McCain will finish third in 2008. Further, while it is technically correct that Clinton won the primary vote in 2008 and that Obama finished second, it did not predict a particularly close contest between the two. The post-New Hampshire model only makes one mistake, picking Tsongas to win in 1992 and Clinton to be the runner-up (likely because of his inability to win either Iowa or New Hampshire that year). On the other hand, the two models predict a decisive Romney victory in 2012. Overall, the forecasts are less accurate for second- and third-place finishers even though the post-New Hampshire model correctly predicts 8 of the 12 runner-ups.
Table 2 Combined Model Predicted and Actual Finish, 1980–2016
Note: Underlined names indicate a correct ordinal forecast in terms of percent of primary vote. Manuscript submitted before end of 2016 primaries so no final results are available yet.
In applying the model to the 2016 election cycle, both models forecast victories for Hillary Clinton and Donald Trump. Clinton receives 76% and 58% of the aggregate primary vote in the preprimary and post-New Hampshire primary models, respectively, with Bernie Sanders’ New Hampshire victory tightening the race. Trump barely beats Rubio (21 to 17%) in the preprimary forecast of a crowded Republican field, but easily surpasses Cruz (41 to 21%) in the post-New Hampshire model because of Trump’s strong finish there.
FINDINGS
Early popular support, preprimary elite endorsements and a large campaign war chest entering the formal campaign remain strong predictors of successful nominees. The results of the Iowa caucuses and the New Hampshire primary also influence both the eventual nominees and the margins of victory. Clearly, in spite of recent changes such as the increase in outside money and the fragmentation of support among party elites, the traditional models of forecasting presidential nominees are sufficient if they can correctly forecast a conventional insider like Hillary Clinton and an unorthodox outsider like Donald Trump.
ACKNOWLEDGMENTS
The authors would like to thank Dino Christenson, Tom Cronin, Wayne Steger, and two anonymous reviewers for their comments. A previous version of this manuscript was presented at the 2016 Western Political Science Association Annual Meeting.
