The results of our initial estimation of the stratified Cox model (Model 1) are reported in the first column of Table 1. The independent variables are jointly significant in the statistical sense and most are individually significant.Footnote 25 Given the manner in which Cox models are parameterized, a positive coefficient estimate indicates that an independent variable increases the hazard of a candidate ending their candidacy at time t. Since the duration of a campaign is inversely related to the hazard of a candidate withdrawing on a given day, a positive coefficient indicates that an independent variable decreases the expected duration of a candidacy.
Table 1 Cox model of candidate withdrawal from presidential nomination campaigns, 1980–2008
Turning to the parameter estimates of interest, we argue that candidates will stay in the race longer if they have more to gain in terms of raising their profile. The coefficient estimate for Name Recognition in Model 1 is in the direction we hypothesize and is statistically significant. A candidate’s initial level of name recognition has a positive effect on the hazard of that candidate withdrawing at time t. Candidates with lower levels of name recognition are less likely to drop out on a given day and thus have longer candidacies, all else equal. It appears that candidates are motivated to stay in the race by the possibility of increasing their profile, with an eye towards improving their future political role or career. However, the estimate for Media Coverage is neither statistically significant nor in the expected direction.Footnote 26 Thus, while a candidate’s initial level of name recognition affects how long they remain in the race, recent media coverage of the candidate does not decrease his or her likelihood of withdrawing.
We also contend that candidates vary in how much they care whether staying in the race for a lengthy period of time causes any potential harm to their reputation within the party or eventual nominee’s success in the general election. The results of Model 1 are consistent with our theoretical expectations as it appears that candidates who are connected to their party terminate their campaigns more quickly than those less connected to their party’s mainstream. The positive and statistically significant estimates for Member of Congress and Other Major Office reveal that candidates who have held significant political positions have a greater hazard of exiting the race at time t than candidates who have not held any sort of major office and are thus presumably less connected to the party.
In addition, the result for Ideological Deviation indicates that candidates who deviate from their party’s mean ideological position in Congress are less likely to drop out at time t than candidates who are in the ideological mainstream of the party. As we argue above, this is the case because it is less costly for party outsiders to engage in lengthy, losing candidacies than it is for party insiders. Steger (2008) finds that ideologically extreme candidates get more primary election votes than would otherwise be expected. Our results suggest that these candidates also stay in the race longer, even once any competitive advantages are controlled for.
Turning next to the variables assessing the influence of competitiveness on candidacy duration, the results of Model 1 provide little initial support for the importance of retrospective indicators of competitiveness in shaping a candidate’s decision to remain in the race or withdraw. The estimates for Contest Distance and Delegate Count Distance are in the predicted direction (positive), but are not statistically significant.Footnote 27 It thus appears that recent electoral results and the allocation of delegates have surprisingly little influence on the decision to exit from a presidential nomination campaign. Below, we revisit the result for Contest Distance.
The results for Field Size and Poll Distance suggest that prospective evaluations of future competitiveness do influence a candidate’s decision to remain in the race. Consistent with Aldrich (1980), the larger the Field Size at time t the shorter the length of time the candidate will stay in the race, all else equal. The estimate for Poll Distance reveals that the further behind the leader a candidate is, the more likely it is that s/he will drop out of the nomination contest. Cash on Hand, however, does not exert a significant effect. This interesting result (one that we further discuss below) implies that being successful in raising money does not translate into longer candidacies, once other indicators of competitiveness and candidate motivation are controlled for in the model.Footnote 28
Not surprisingly, Certainty of Losing has a positive and significant coefficient estimate, revealing that as it becomes increasingly clear that trailing candidates will not ultimately win the nomination they become more likely to drop out. This result, in conjunction with the insignificant estimate for Delegate Count Distance, implies that candidates care about how close the frontrunner is to securing the nomination, not how far behind the frontrunner they are in the delegate totals. This result is compatible with Norrander’s (2000) conclusion that the competition for a nomination ends once the frontrunner has won a substantial proportion of the necessary delegates. This result also subsumes the commonly made argument that frontloading shortens the length of candidacies during the nomination phase of an election year. In a front-loaded campaign, the eventual losers become more quickly aware of their fates.
Reconsidering Contest Distance
One conventional wisdom regarding nomination campaigns is that the Iowa caucus and New Hampshire primary play a disproportionate role in determining nomination outcomes and the length of campaigns (e.g., Adkins and Dowdle 2001a; Steger et al. 2004). To test whether these two specific electoral contests might affect the decision to withdraw from the race, we estimated our model a second time including two interaction terms. The first consists of Contest Distance and a dummy variable indicating whether the most recent contest in question is the Iowa caucuses (Iowa). The second consists of Contest Distance and a dummy variable indicating whether the most recent contest is the New Hampshire primary (New Hampshire). These interaction terms allow us to assess whether these two events play a role in determining whether candidates drop out, even though Contest Distance generally does not appear to influence this decision.Footnote 29
The results from this model estimation (Model 2) are reported in the second column of Table 1. The coefficient estimates for the two interaction terms are positive and statistically significant, indicating that the results of the Iowa caucus and New Hampshire primary influence the decision to remain in the race or withdraw. While Contest Distance generally does not have any influence on candidacy duration, the results of these two specific contests do affect candidacy duration. The worse that a candidate does in either of these contests, as compared to the winner, the more likely it is that this candidate will withdraw from the race. The substantive inferences for the other independent variables are the same as in Model 1.
To get a better substantive feel for these effects, we generate predicted hazard rates for candidates dropping out of the race. Figure 1 presents predicted hazards (using Model 2) on the y-axis while Contest Distance varies on the x-axis. Three curves are plotted: the first for hazards when Iowa equals one (i.e., the most recent contest result is that of the Iowa caucus), the second for hazards when New Hampshire equals one (i.e., the most recent contest result is that of the New Hampshire primary), and the third for hazards when both these dummy variables equal zero (the most recent contest result is from a subsequent primary). For the Iowa hazards, the baseline hazard is set to its mean value for the week after this caucus. For the New Hampshire hazards, the baseline hazard is set to its mean value for the week after this primary. For the Other Contests hazards, the baseline hazard is set to its mean value for all other campaign days.Footnote 30
The plotted hazards associated with days following the post-New Hampshire contests (the Other Contests hazards) show that candidates are generally more likely to drop out after these subsequent primaries. However, the results of these later contests do not appear to affect the hazard of withdrawing (the hazards appear to decrease slightly as Contest Distance increases, but based on the insignificant coefficient estimate for this variable we cannot conclude that this slope is truly negative). In contrast, the worse a candidate does in the Iowa caucus or New Hampshire primary, the greater the hazard of withdrawing from the race. The relatively flat curve for the Iowa caucus results, however, suggests that candidates’ performances in Iowa relative to the winner do not exert much of a substantive effect on the exit decision despite the statistically significant coefficient for the Contest Distance × Iowa interaction. The steep increase in the hazard for the New Hampshire primary indicates that candidates’ performances in the Granite State have strong substantive effects on their hazard of withdrawing from the race.
Similarly, we can more clearly assess the substantive effects for party-related costs by generating predicted hazard rates (from Model 2) for candidates dropping out of the race as a function of variation in party connectedness and Ideological Deviation (implicitly conditioned by whether we have a measure of this variable for the candidate). Figure 2 plots three sets of hazard rates: one for candidates who served in Congress, one for candidates who have served in another major office, and one for candidates who have never held any kind of major office. The full observed range of Ideological Deviation is plotted on the x-axis.Footnote 31 As discussed above, Ideological Deviation does not vary for candidates who never held a major office and the hazards for this candidate type are therefore flat.
This figure reveals that those who have not held elective office (i.e., party outsiders) generally have low likelihoods of dropping out on a given day, all else equal. In contrast, candidates who have held office typically have a greater hazard of withdrawing from the race; reflecting the greater party related costs that these candidates may face for continuing an unsuccessful bid for their party’s nomination. At the same time, this effect is not constant. Rather, Ideological Deviation exerts a negative effect on the hazard rates for office holders. Indeed, office-holding candidates who are ideological outliers appear to have very similar hazards to those who have never held major office.
The Role of Money
While the results presented above are largely consistent with our theoretical expectations, these results diverge from some of the conventional wisdom regarding nomination campaigns, as well as the findings of prior research on the winnowing process. For instance, in her analysis of the length of time that candidates remain in the race for a nomination, Norrander (2006) focuses on the role of money and finds that it has a quadratic relationship with the length of candidacies. Well-funded and poorly funded candidates stay in the race for a long time while moderately funded candidates exit earlier. We find that cash reserves do not exert a simple linear effect on the duration of a candidacy, but have we misspecified our model in this respect?
We suspect the answer is no. Norrander’s results are likely a function of party outsiders often being poorly funded and, as we argue, less likely to drop out early since lengthy losing candidacies are less costly to them. Similarly, candidates with low levels of name recognition stay in the race longer, and are probably less successful in raising money. Since we account for the degree to which a candidate is connected to his or her party as well as name recognition, it seems there is little reason to expect money to exert a curvilinear effect. Nonetheless, we also estimated our model while including the square of Cash on Hand. Under this specification the estimates for both Cash on Hand and its square are statistically insignificant, supporting our suspicion that once profile-raising benefits and party-related costs are controlled for money should not exert a curvilinear effect on the hazard of withdrawing from the race.
The null finding regarding the role of campaign money is interesting in its own right, given the importance typically attached to fundraising. We are certainly not willing to go so far as to conclude that money is irrelevant to nomination campaigns. Indeed, one reason for our results may be that for many candidates fundraising, poll standings, and electoral performance are closely intertwined (Damore 1997). Thus, while candidates’ ability to raise money may be a precursor to other factors shaping candidates’ decision making, fundraising in and of itself may not directly influence how long a candidate remains active in a nomination campaign, all else equal (also see Cohen et al. 2008).Footnote 32
Is the Duration of a Candidacy Determined Before the Iowa Caucus?
Recent research (e.g., Adkins and Dowdle 2001b; Cohen et al. 2008; Steger 2008) suggests that activity during the invisible primary has significant effects on determining the winners and losers of presidential nomination campaigns. Does the same hold for the exit process? While our model does include a time-constant component (e.g., candidate characteristics indicating connection to the party), our argument regarding perceptions of competitiveness and the certainty of losing the nomination points towards a more dynamic (or time-varying, to use duration terminology) view of nomination campaigns. We find that the duration of a candidacy is explained by several variables that change in value as the campaign progresses. Specifically, Field Size, Poll Distance, and Certainty of Losing are all time varying and are all significant predictors of the length of a candidacy. In light of the performance of these variables, it appears that what occurs during the delegate-allocating phase of the nomination process plays a significant role in the length of time that candidates remain active. Or put differently, how long candidates choose to contest their party’s nomination is not determined by what occurs prior to the Iowa caucuses.
There is, however, the possibility that variation in Poll Distance and Field Size across candidates and election cycles, respectively, explains the hazard of exiting the race. In other words, it could be the case that the static components of these two variables have the explanatory power, not the dynamic components that change over time for a given candidate. To test this possibility, we estimated our model with two additional independent variables that do not vary over the course of a campaign: Initial Poll Distance and Initial Field Size at the time of the Iowa caucus. With these controls in place, the estimates for Poll Distance and Field Size remain statistically significant. These results indicate that static components of these two variables are not driving the coefficient estimates. The expected duration of a candidacy changes from day-to-day in a dynamic manner, instead of being fixed at the start of the campaign.