The time frame at which presidential nominations can be forecast accurately provides us with a clue about when a winning nominating coalition coalesces in a political party, which in turn provides insights about political power in the nomination process. This article uses updated forecasts of the contested vote in presidential primaries to assess competing hypotheses from different theoretical approaches to the study of presidential nominations. One theoretical perspective holds that presidential nominations are largely determined during the invisible primary (e.g., Cohen et al., 2003, 2008; Hadley 1976; Haynes et al. 2004; Mayer 1996; Steger 2000). In this scenario the caucuses and primaries play a plebiscitary role, confirming the results of the campaign occurring before the primaries when party elites, activists, donors, and groups evaluate candidates and coordinate among themselves in order to unify behind a presidential candidate--before the mass membership of the party weigh in on the selection of the nominee. The other theoretical approach focuses on candidates' campaign momentum during the caucuses and primaries (e.g., Aldrich 1980; Bartels 1988; Norrander 1993, 2006; Popkin 1991). We can get a sense of which argument is correct by assessing the accuracy of presidential primary vote forecasts that use information from different points in time.
Forecasts of the presidential primary vote differ from general election forecasts and other kinds of static explanatory models in a critical respect. Since presidential nominees are chosen through a sequential process that begins years before and continues through a series of caucuses and primaries, the process can be thought of as a Bayesian updating model (e.g., Morton and Williams 1999). In Bayesian terms, the nomination campaign leading up to the caucuses and primaries establishes an expectation for the outcome of the campaign, and this baseline expectation is updated through successive caucuses and primaries. As such, it is possible to use forecast models to assess the relative impact of conditions and events at different points in time to determine which phases of the campaign are critical to the determination of the outcome. This article estimates forecast models using information known at the end of the invisible primary period and compares these with forecast models using information from the earliest caucuses and primaries--the events considered to have the greatest impact on subsequent primary vote. If the nominee can be predicted accurately using information from the invisible primary, then the caucuses and primaries would appear to confirm the results of the earlier processes. However, if pre-primary forecasts have substantial error and that error is reduced by information about the results of the earliest nominating elections, then the caucuses and primaries can be said to have an independent influence on the outcome and thus campaign momentum affects the outcome.
Rather than affirming one perspective or the other, this study shows that both patterns exist in different presidential nomination campaigns. One pattern is evidenced by substantial coalescence of party elites and mass partisans during the invisible primary behind a "front-runner" who goes on to win the nomination. Party elites, activists, and groups, however, do not always unify behind a candidate before the caucuses and primaries. These campaigns are distinguishable in forecasting models by larger errors in the prediction of candidate vote shares. In these nomination cycles, models incorporating information from the results of early caucuses and primaries substantially improve predictive accuracy. The occurrence of both patterns raises an important question about why political party elites and mass identifiers unify during the invisible primary period in some election cycles more than during others.
The next section briefly reviews models of presidential primary vote forecasts. …