EARLY VOTING AND EXPECTATIONS

These strands of literature allow us to understand how, and why, early election events have critical effects on the final outcomes in presidential nomination contests. Specifically, how (and why) do the Iowa caucuses and New Hampshire primary affect subsequent results in the nomination process, even when these two states play a trivial direct role in allocating convention delegates? Put differently, how do early events in small states contribute to candidate momentum in sequential nomination contests?

We propose a model of outcomes where early events matter, in part, because news about outcomes in these states serve as a major source of information about candidate viability in a relatively low-information choice setting. Early nomination events receive disproportionate media attention (relative to their share of delegates), and much of that media attention relates to expectations about a candidate's performance in early contests. The former claim here is uncontroversial, and the latter has been noted elsewhere (Brady and Johnston Reference Brady, Johnston, Orren and Polsby1987). In this model, the role of the media can be seen as somewhat analogous to the process where share-market analysts set corporate earnings expectations. In share markets, when a firm exceeds its earnings expectations, its share price may rise. If it fails to meet expectations, its share price may fall. Likewise, more media attention may be earned by candidates who exceed expectations. Those who fall short of expectations may see their share of news coverage shrink.

This model also grants the media substantial discretion in setting and adjusting expectations. Reporters, editors, and pundits define the criteria for determining whether a candidate scored an “easy win,” managed an “upset,” was “far behind,” or suffered “defeat.” There is substantial discretion in framing whether 25% is a “Comfortable Second” (Bill Clinton in New Hampshire in 1992), or 23% is a “Strong Second” (Pat Buchanan in Iowa in 1992); or if 26% is a “Flat Tire” (Bob Dole in New Hampshire in 1996) or 26% is an “Overwhelming Defeat” (Howard Dean in New Hampshire in 2004).Footnote ³

Horserace coverage of campaigns involves handicapping the candidate pool—with a substantial proportion of coverage focusing on who the frontrunners are expected to be, who the underdogs are, and who beats or fails to meet expectations. Initially, the decision to even report on one particular candidate rather than another, and the amount of attention granted, can be seen as the expression of media expectations. We can assume that candidates who are not expected to be players in a contest will receive less media attention—if for no other reason than media resources (column inches, minutes of news time, etc.) are finite. Attention must be rationed in favor of candidates who are expected to place relatively high.

Voters thus receive substantial information about the media's expectations of candidate viability, and of the media's interpretation of whether candidates met expectations, exceeded expectations, or failed to meet expectations. If some voters make choices on the basis of expectations about who is viable or electable, election results from early contests and subsequent changes in the media's treatment of candidates are likely to be a major source of readily available information for voters in later contests. Although this argument is not wholly original, few (if any) studies have estimated outcomes in U.S. presidential nomination contests as a sequential process that includes adjustments for media expectations associated with results from initial contests.Footnote ⁴ Conventional accounts of outcomes in nomination contests emphasize the role of: (1) candidates' national opinion standings at the start of the process, (2) candidates' financial resources at the start, and (3) home-state advantages (Norrander Reference Norrander1993). Previous studies do not account for how the media sets its expectations, nor have many previous studies considered how alterations in the media's attention to a candidate because of an early outcome affect the candidate's prospects in subsequent contests.

DATA

Data from nomination contests from 1976 to 2008 are used to model initial press attention to candidates, candidate performance in Iowa and New Hampshire, and aggregate performance, respectively. These data include Gallup opinion measures of each candidate's national poll standing prior to the Iowa event, and measures of candidate fundraising in the year prior to the first nomination event (Iowa). Measures of national media attention to candidates include attention to candidates two weeks before and the days immediately after the Iowa caucus, and the week prior and immediate days after the New Hampshire primary. A total of 91 candidacies are in the dataset. Two incumbent presidents who had serious primary challenges (Gerald Ford in 1976 and Jimmy Carter in 1980) are included, while other incumbents who lacked serious challenge (George H.W. Bush in 1992, Bill Clinton in 1996, and George W. Bush in 2004) are excluded.Footnote ⁵

MODELING EXPECTATIONS

Given that we have no readily available measure of media expectations about front-running candidates, we make use of a relatively straightforward surrogate. Information about media expectations for each candidate is represented by the number of times a candidate's name appeared in New York Times stories about Iowa and New Hampshire that ran prior to voting in each state, and in stories that ran after voting.Footnote ⁶ We coded mentions of the candidates' names in campaign related stories, while we omitted mentions of candidates in stories about governing. We assumed that candidates being mentioned most frequently across several stories were expected to be frontrunners.

Total attention to the Iowa and New Hampshire events is uneven across time. We measured media attention to individual candidates in a manner that is comparable across time by calculating the proportion of all candidate mentions of each Democrat and each Republican, respectively. Table 1 lists the top candidates on this measure based on New York Times stories that ran two weeks before Iowa, for both parties. Thus, Table 1 illustrates who received the most press attention prior to the Iowa caucuses, which we assume to reflect initial (pre-Iowa) media expectations of candidate viability.

Table 1.

Most Frequently Mentioned Candidates in Iowa Stories (Pre-caucus)

Note: Cell entries are the candidate's proportionate share of all candidate mentions, per party.

We also calculate how media attention to these candidates shifted in the days immediately after Iowa and New Hampshire, respectively, by comparing initial press attention prior to voting to attention in articles after results were known. Table 2 lists the candidates with the largest net changes by how often they were mentioned in stories about Iowa before the vote, and then after. Table 3 lists the same information for New Hampshire. For example, Pat Robertson was mentioned quite infrequently in stories about Iowa prior to the 1988 vote, but his proportionate share of references to all GOP candidates increased by 21 percentage points (from just 10% to 31%) in stories about Iowa published in the days immediately after his second-place finish.

Table 2.

Largest Change in Press Attention to Candidate: 1976–2008; Before and After the Iowa Caucus

Table 3.

Largest Change in Press Attention to Candidate: 1976–2008; Before and After the New Hampshire Primary.

Note: Percent change in candidate's share of references among candidates from the same party.

This measure of change in media attention serves as a surrogate measure of how media expectations of candidate viability adjust after Iowa votes. Prior to the result of the 1988 Iowa caucus, expectations (and attention) for Dole, Bush, and Kemp were higher; after Iowa caucused, expectations about Robertson shifted, and he enjoyed greater media attention prior to New Hampshire. As another example, Gary Hart received relatively little notice prior to Iowa (10% of Democratic candidate mentions in 1984). However, after posting a surprising second-place finish in Iowa (with just 16%, 32 points behind Walter Mondale), his share of media attention in post-result coverage of Iowa more than doubled (increasing from 9% to 19% overall), while Mondale's share of press attention declined relative to that given his rivals. Hart's 1984 victory in New Hampshire corresponded with another 27% bounce in attention; Buchanan enjoyed a similar phenomenon after collecting a mere 22,000 Iowa caucus votes in a surprise second-place finish in 1996. In terms of beating initial media expectations, Hart (1984, 37%), Reagan (1976, 36%), B. Clinton (1992, 35%), Buchanan (1996, 28%), Tsongas (1992, 23%), Carter (1976, 20%), Huckabee (2008, 19%), H. Clinton (2008, 18%), Gore (2000, 17%), Kerry (2004, 17%), Robertson (1988, 17%), and Obama (2008, 16%) rank highest in the net gain in attention from before Iowa to immediately after New Hampshire.Footnote ⁷

How then, are initial media expectations set, and how might they predict voting in early events? More important, how do changes in expectations produced by the Iowa results affect voting in a subsequent nominating event (New Hampshire)?Footnote ⁸ Conventional wisdom and logic suggest several factors that drive the press to give some candidates more early attention: fundraising, poll standing, incumbency,Footnote ⁹ and home-state advantages.Footnote ¹⁰ We expect candidates who raised more money prior to Iowa, those with higher national poll standings, those from Iowa, and incumbents to receive more initial media attention. We measured campaign fundraising as total funds raised the year prior to Iowa in terms of inflation-adjusted (to 2000) dollars. These factors are used to estimate a candidate's share (proportionately) of total news mentions of candidate names prior to Iowa. Although some of these items are well correlated, the correlations are by no means perfect.Footnote ¹¹

Table 4 reports results of estimates of candidate share of press attention in the weeks before Iowa. We find that about 70% of variance in candidate share of press attention (our surrogate for expectations) can be explained by fundraising, poll standing, and the two candidate-specific factors. Each additional 10% in opinion standing is associated with 5% greater media attention, and $10 million adds an additional 4.8% share. These results are not surprising, but they do illustrate that money and poll numbers are not perfect predictors of media attention. Part of press coverage likely involves setting expectations by interpreting if a less-known but well-financed candidate is deserving of as much attention as a well-known officeholder. Indeed, these nomination contests are frequented by well-financed candidates who gain little traction with voters (John Connelly, $19 million in 1980; John Glenn, $11 million in 1984; Phil Gramm, $22.3 million in 1996; Rudy Giuliani, $51 million in 2008) and well-financed candidates who were relatively unknown quantities early on (Robertson, $24 million in 1998; Steve Forbes, $20 million in 1996; Mitt Romney, $74 million in 2008).Footnote ¹² What then are the potential effects of media attention/expectations, independent of candidate poll standing and fundraising? Or, forgetting pretense to causal arguments, does media attention predict something that fundraising and poll standing might not?

Table 4.

Estimates of Media Attention to Iowa Candidates 1976–2008, Pre-caucus

Note: Dependent variable equals percent of all references to candidate. OLS coefficients reported, with errors in parentheses. All models estimated with dummies for 1976, 1980, 1984, 1988, 1992, 1996, 2000, and 2008 (coefficients not reported).

Table 5 reports estimates of Iowa caucus results from 1976–2008, using the standard variables included in models estimating nomination outcomes (Norrander Reference Norrander1993; Mayer Reference Mayer and Mayer1996; Reference Mayer2003). When standard forecasting variables are used (column 1), money and poll standing appear to have substantial power to predict results in Iowa. In contrast, when press attention to candidates is used to estimate results, the effects of money are eliminated, and the effects of poll standing disappeared (when vote percent is modeled—but not when place of candidate finish is estimated). The Iowa vote share is also estimated here with an instrumental variable, where press attention predicted from the model reported in Table 4 is used to predict the Iowa vote share. Again, we see that press coverage of a candidate (predicted by the candidates' fundraising and polling numbers) outperforms models that use only polling and finance to predict outcomes in Iowa. How should these results be interpreted? Why would media attention better predict (or predict as adequately) as the direct measures of money and poll status?

Table 5.

Estimating Iowa Caucus Results, 1976–2008

Note: OLS estimates reported. All models estimated with dummies for 1976, 1980, 1984, 1988, 1992, 1996, 2000, and 2008 (coefficients not reported).

+ Instrument generated from Table 4, without dummies for year.

* Same substantive results with ordered probit.

Clearly, media attention to candidates covaries with fundraising, and there is no way to clearly sort out the alternate causal processes that may be at work here. Reporters and editors may be particularly savvy at using information beyond poll numbers to anticipating who will succeed in Iowa, and thus direct more of their attention to those candidates. That said, these results are consistent with a process where candidates who receive more media attention gain an electoral advantage beyond that associated with their fundraising and national standing in opinion polls.

Iowa's potential effects on nomination contests in subsequent states are a more important matter. As good as reporters, editors, and pundits may be at anticipating outcomes in Iowa, they often find their initial expectations were off. One of the primary political functions of the news media is interpreting and framing events—that is—defining the meaning of such things as victory, second place, or 26%. Expectations are then adjusted, with increased attention directed at candidates who exceeded initial expectations (Hart in 1984; Robertson in 1988; Buchanan in 1996; Kerry in 2004) or were not expected to do well anyway (Bill Clinton and Paul Tsongas in 1992). Table 6 reports estimates of the New Hampshire primary results from 1976 to 2008. Candidate vote shares (and places) are estimated as a function of the standard variables (early poll standing, finances, state of residence), with two independent variables representing the potential effects of Iowa: the candidate's vote share in Iowa and the change in media attention directed at the candidate immediately after Iowa.Footnote ¹³ Again, the underlying assumption here is that some voters opt for candidates they expect to be more viable, and that they make use of election results, and the media's interpretation of results, to assess viability.

Table 6.

Estimates of New Hampshire Primary Results, 1976–2008

Note: All models estimated with dummies for 1976, 1980, 1984, 1988, 1992, 1996, 2000, and 2004 (not reported).

* Logit estimates.

+ Same substantive results via ordered probit.

Results in Table 6, albeit estimated with aggregate data, are consistent with such a process. We see a robust association between a candidate's performance in Iowa and New Hampshire. The Iowa vote share, and the Iowa place of finish (not shown), are significant predictors of the New Hampshire vote share, the likelihood of winning in New Hampshire, and the place of finish in New Hampshire. This result holds when we control for the candidate's fundraising and initial standing in national polls. Independent of these effects, we also see that change in media attention toward a candidate post-Iowa also has a significant relationship with support in New Hampshire. Candidates like Hart, Robertson, Buchanan, and Kerry may have had an additional edge in New Hampshire because of the shift in media attention they earned from their surprise finishes in Iowa. Although there is no relationship between the shift in media attention toward a candidate and winning New Hampshire, the potential importance of the media bounce coming out of Iowa on the overall nomination contest should not be underestimated.

As Table 7 shows, performance in New Hampshire is a strong predictor of the aggregate primary vote (and thus delegate share), with Iowa having a more muted effect (depending on specification). But results in Table 7 demonstrate that change in media attention after Iowa, and after New Hampshire, have important substantive effects on how much support a candidate receives throughout the nomination contest. Increased attention to a candidate immediately after Iowa, and immediately after New Hampshire, has a significant relationship with increased vote share across the nomination contests. When Table 5 and Table 6 are considered together, the results suggest that changes in news about candidates due to results in Iowa affect how well a candidate does in New Hampshire. Candidate performance in New Hampshire then produces additional adjustments in media attention to candidates (and expectations about viability), and this is associated with how well a candidate fares overall. One need not win Iowa to win New Hampshire, nor must one win New Hampshire to win a nomination (although it clearly helps). However, additive models in Table 7 suggest performance in Iowa had less effect on overall primary vote share than performance in New Hampshire did; but it is important to remember that the nomination process is sequential. Success in New Hampshire corresponds with earlier electoral success in Iowa and the media bounce associated with beating expectations in Iowa. Beating early expectations may determine whether a candidacy ends quickly or whether it lasts longer.

Table 7.

Estimates of Aggregate Primary Vote and Nomination Outcomes

* Logit estimates.

Note: OLS estimates unless noted otherwise. All models estimated with dummies for 1976, 1980, 1984, 1988, 1992, 1996, and 2000 (not reported).

We also produced estimates from the first model in Table 7 using data from the 1976–2004 elections (omitting 2008) in order to test how the model predicted the 2008 outcomes. These predictions are listed in Table 8. Table 8 illustrates that the model performs fairly well in predicting the overall vote share for the main candidates, although it under predicts Obama's and McCain's vote share. Of course, vote share is not the same as winning the nomination. Results in Table 7, and the predictions in Table 8, suggest that a model based on pre-Iowa poll standing, results from the earliest contests, and shifting media attention do a good job explaining a candidate's vote share (which corresponds highly with how long a candidate remains in the contest). These results also demonstrate that such models have less utility in distinguishing which candidates actually win.

Table 8.

Actual and predicted 2008 Results, From Table 7, Column 1

Nomination contests are sequential. This sets the stage for early events to have important effects that cascade over time. Early results can alter media assessments of a candidate's viability, with the change in news attention breathing new life into some candidacies while leaving others all but forgotten. From 1976–2008, media expectations about which candidates were viable were set before Iowa voted. Iowa's results then led to altered media expectations about who the frontrunners were. New Hampshire results further altered media expectations about candidate viability, and these shifts in press attention to candidates then shaped the context voters faced in subsequent states.