Assessing strategic voting in the 2008 US presidential primaries: the role of electoral context, institutional rules, and negative votes

Others have used the term negative strategic voting to refer strictly to partisans who cast a ballot for an opposing party candidate they do not sincerely support (Southwell 1991). In contrast, our definition does not factor in the party affiliation (or registration) of the voter or candidate. As we show, there is a sizeable number of independents and in-partisans who vote for a candidate in the primary they intend to vote against in the general election.

Cain and Gerber (2002) have a terrific edited volume that offers a comprehensive analysis of hedging versus mischievous voting in California.

The operationalization of crossover voting actually varies widely in the literature. Some include Independents or unaffiliated voters as crossover voters, while others consider only partisans. Some use measures of party registration, while others use partisan self-identification. See Sides et al. (2002) for further discussion of this point.

There are a handful of notable exceptions (Geer 1986; Southwell 1989; Alvarez and Nagler 2000).

The codebook notes that two thirds of respondents were interviewed in the week following Super Tuesday when both the Democratic and Republican contests were competitive. And in the classic work on the topic, Abramson et al. (1992) construct within party rankings so there is no consideration of preferences that might cross party lines.

Geer (1986) argues that if someone really wants to vote in a primary, she just has to figure out what is necessary to participate. While this is surely true, it also is the case that the competitive context that motivates strategic voting might not emerge until after party registration deadlines have passed. Using voting experiments, Cherry and Kroll (2003) find that negative strategic voting is more likely in open than closed rule conditions.

Although these data limit our ability to explicitly test for the role of gender and racial attitudes, a cursory analysis suggests that among the Democrats who cast negative strategic votes for Clinton (i.e., they preferred McCain to either Clinton or Obama), only 36 % said they had a favorable view of blacks, compared to 56 % of those who cast a sincere vote for Clinton, 53 % of Democrats generally, and even 43 % of sincere McCain voters.

Moreover, given the fragmented information environment, coordination might come from social networks, blogs, or message boards rather than political elites. Scholars have challenged the conclusion that Limbaugh was responsible for the increased rate of crossover voting because the trend started even before his announcement (Stephenson 2011; Donovan 2008). Although we are not able to explicitly test for elite/media effects in our data, we do find that negative strategic Clinton voters were more likely than sincere Clinton voters and even sincere McCain voters to say they often watched Fox News. See footnote 36 for more on this point.

The study was a collaboration between the Associated Press and Yahoo Inc., with support from Knowledge Networks. The KnowledgePanel(R) panel members are chosen via a probability-based sampling method and using known published sampling frames that cover 96 % of the US population. Sampled non-Internet households are provided a laptop computer or MSN TV unit and free Internet service. The wave 1 survey (baseline) was fielded on November 2 2007 to a sample of 3,548 panel members of age 18 years or older who represented a general population sample. The total number of completed interviews at the baseline was 2,714, for a 76.5 % cooperation rate. Using the formula specified in Callegaro and Disogra (2008) to take into account initial recruitment into the KnowledgePanel, this represents a cumulative response rate (CUMRR1) of 11.2 %. This rate is a multiplicative combination of the panel recruitment response rate (AAPOR3), the household profile rate and the survey completion rate, but excludes the household retention rate. The study attempted to re-interview each of the baseline cases for a total of 11 waves. Kruse et al. (2009) and Deng et al. (2013) find little bias in these data from panel attrition and panel conditioning.

Demographics are measured in a profile survey gathered separately from the panel study. Self-reported vote choice in the primary was asked in either wave 4 or 5 depending on the date of the election in the respondent’s state.

In all waves, more than 75 % of the sample completed the survey within the first week of being fielded.

The question was worded as follows: “For each of the following individuals, please select if you have a favorable or unfavorable impression of that person. If you dont know enough about the person to have an opinion, you can say that too” [very favorable; somewhat favorable; somewhat unfavorable; very unfavorable; don’t know enough to say].

Table 1 shows which candidates were ranked in each wave. Once a candidate dropped out of the race the candidate was not ranked in the subsequent wave. Additionally, Huckabee, who rose to prominence only just before the Iowa Caucus, was not included as a candidate in wave 1 (November 2007). A handful of respondents—less than 4 % in waves 1–3 and 5 % in wave 4—rank all candidates equally.

We use the standard party identification question from wave 1 of the survey to help avoid conflation with primary voting. Independent leaners are included with partisans.

We must be cautious in focusing on any particular trend across waves since the number of candidates ranked (and thus the worst ranking possible) varies across waves. We find similar patterns if we instead look at the percentage of voters who rank each candidate first by party—the measure used in subsequent analysis—but we present the full rankings here to clearly illustrate the structure of our key measure.

Again, individuals could have more than one candidate ranked first.

Each candidate was evaluated according to the following traits: likeable, decisive, strong, honest, experienced, compassionate, refreshing, ethical, and attractive. The exact question wording is as follows: “Please tell me how well each word describes [INSERT NAME OF CANDIDATE] (not at all well, slightly well, somewhat well, very well). Increasing the number of items used to measure candidate preference helps to reduce the measurement error (Ansolabehere et al. 2008). As an additional robustness check on the effects of measurement error, we reestimate our models, where possible, just for the subset of politically interested respondents (on the assumption that they should have less measurement error in their preference rankings) and find similar patterns in the data.

It is interesting to note that if we look into the general election, just 8.7 % of general election voters ultimately cast a ballot for the candidate (McCain or Obama) they ranked the highest in November 2007.

Interestingly, the same is not true for John Edwards voters. This may reflect the fact that those voting for Edwards actually preferred Obama or Clinton but feared a woman or African-American nominee might not be electable in the general election.

For roughly two-thirds of respondents this was wave 3, thus typically capturing their vote preference before Super Tuesday, when Giuliani, Romney, Huckabee, and McCain were all running at double digits in the polls.

Given the added complexity and limited payoff of coding all eight candidates (since there was no incentive to cast a negative strategic vote for any of them), we focus here only on the rankings of the three always competitive candidates—McCain, Obama, and Clinton—since these were the candidates who were competitive when all voters were casting ballots. Seventy-five percent of voters cast a ballot for one of these three candidates.

Although some previous research strictly considers second choice preferences, this seemed too restrictive because so many respondents ranked uncompetitive candidates both first and second.

To be clear, this is distinct from previous measures of mischievous voting that consider only crossover voters—we deliberately do not consider the party of the candidate or voter in determining if a voter cast a negative strategic or even a mischievous vote.

For example, Abramson et al. (1992) found that 13 % of primary voters cast (positive) strategic ballots.

For the sake of comparison with previous research, we looked at negative strategic voters by party identification. We find that 63 % of negative strategic voters were crossover voters, while 37 % were partisans (including leaners) who nonetheless cast a negative strategic ballot for a candidate of their same party. These were, for example, self-identified Democrats who voted for Clinton even though they ranked McCain higher. This suggests that previous research that focuses on crossover voters captures the majority of negative strategic voters, but—at least in 2008—this more restrictive definition would miss a notable subset of voters who cast a negative strategic ballot for a candidate of their own party.

We relied on actual vote totals as reported by David Leip’s Atlas of US Presidential Elections.

For example, in her concession speech on 4 June 2008, Clinton remarked that “Nearly 18 million of you cast your votes for our campaign, carrying the popular vote with more votes than any primary candidate in history.”

However, there is little evidence that Clinton primary voters were especially likely to stay home on Election Day: 5 % of Clinton primary voters did not vote, compared to 4 % of Obama primary voters and 3 % of McCain voters. See Henderson et al. (2010) for a more extensive analysis of the link between primary and general election behavior in the 2008 campaign.

One interesting side note comes from the comparison of expected and reported behavior of sincere Clinton voters. Although these Clinton voters were no less likely than sincere Obama voters to report voting for McCain by November, in the heat of the competitive primary in April, they were significantly more likely to say they planned to vote for McCain than sincere Obama voters. These results are consistent with other analyses finding that supporters of losing candidates eventually rally behind the party nominee (Henderson et al. 2010). And to the extent that Obama did not win over all Clinton voters, it seems it might be because many never planned to vote for a Democrat in the general election.

A handful of states have different rules for each party within the state. Given our expectation that negative strategic voting is more common in the Democratic primary, our analysis codes states based on the Democratic party rules. Thus, we code as open any state that allows Republicans to vote in the Democratic primary, has no party registration requirement, allows Independents to vote, or has same day party registration (which can differ from voter registration rules in some states).

Although McCain did not formally reach a majority of delegates until the March 4 primaries, the media began calling McCain the “presumptive” nominee after Romney dropped out (Helman and Issenberg 2008). Although rather blunt, this measure captures the key transition point in the competitive environment.

Admittedly, this measure is far from ideal. A preferable measure of electability might ask about the perceived chances of each candidate in each wave. Although the study did ask a similar question about each candidate in wave 3, it was asked only of the subset of respondents who said they would vote in that candidates’ primary—making it inappropriate for our measure of strategic behavior. Even this standard measure of electability has been the subject of considerable criticism since it is so susceptible to projection (Bartels 1987). These are two additional reasons we later focus on the always-competitive candidates.

All model first differences and predicted probabilities are calculated holding all other variables constant at their means or modes for indicator variables.

We reestimated the model using polling numbers as a more nuanced measure of competition with similar conclusions. The use of polling numbers, however, is complicated by the fact that competition is less easily characterized in a mulitcandidate than in a two-candidate race (where margin can be used) and there were no polls on the Republican side after March 4. However, replicating the model using McCain and Obama’s absolute poll numbers (using 100 % for McCain after March 4) finds a positive and statistically significant coefficient for McCain’s poll numbers and no relationship with Obama’s poll numbers.

While this last variable is perhaps surprising, the measure is far from ideal—it is notoriously difficult for survey respondents to articulate the motivations underlying behavior and the other competition measures (support for trailing candidate and the time of election variable) are related to objective electability.

Although a more extensive analysis is beyond the scope of the article, we do find interesting variation in individual characteristics of those casting negative strategic votes for particular candidates. For example, among negative strategic Clinton voters, 42 % say they watch Fox News “quite a bit” or “a lot” (another 38 % say “some”), compared to 22 % of sincere Clinton voters and even 36 % of sincere McCain voters. Among the Democrats who cast negative strategic votes for Clinton (i.e., they preferred McCain to either Clinton or Obama), only 36 % said they had a favorable view of blacks, compared to 56 % of those who cast a sincere vote for Clinton, 53 % of Democrats generally, and even 43 % of sincere McCain voters. The numbers are obviously small and the analysis cursory, but this is at least suggestive of some of the motivations for negative strategic voting for Clinton.

Although the number of Democratic voters alone was not decisive in the Michigan Republican primary, exit polls reported that 53 % of Democrats cast their ballot for the more ideologically extreme Rick Santorum.