How might people have behaved in the 2006 midterm elections were it not for the ongoing conflict in Iraq? A casual observer of US politics might respond glibly, “Republicans would still control Congress!” Indeed they might. There is, however, a way to quantify this counterfactual, as we did in the 2006 New Jersey Senate race between Democrat Bob Menendez and Republican Tom Kean.
While Democrats in New Jersey had easily won every statewide race in the previous five years, there was reason to believe that the 2006 Senate race would be different. First, the Republican challenger was the son of a popular former governor, and thus had a level of name recognition far exceeding most statewide politicians. Second, he had no significant challenge in the primary election. Finally, the Democratic contender had serious problems to overcome. While technically the incumbent, Menendez had been appointed to the seat only a few months earlier and was little known outside of his home district in Hudson County. Hudson County, in turn, was mostly known for a truly remarkable level of corruption.
The only problem fly in the ointment was the Iraq war. While in the House of Representatives, Menendez had voted against the resolution to go to war, and often referred to it as one of the most important votes of his career. Kean, on the other hand, had been reluctant before the campaign to criticize either the President or the war in a state which had become very critical of both. Menendez would make this difference paramount in his campaign.
Testing the Iraq Counterfactual
In the end, Menendez won the race by a margin of 53 to 45: a victory widely attributed to his stance on the war. To test this attribution, we have to make two well supported assumptions. The first is that some percentage of respondents to the survey is thinking about the war in Iraq when we ask them questions about the Senate race, whether the question mentions it or not. The second is that respondents don’t necessarily know the causes of their own preferences – an assumption supported by the two major theories of the survey response, Zaller and Feldman’s “simple model” and Lodge and colleagues’ online processing model.
Because respondents don’t know what underlies their preferences for candidates or anything else, we can’t simply ask them about the effects of Iraq on their vote choice. What we can do is create an experimental condition in which one portion of the sample is prompted to think about the Iraq war, and compare the vote choice of that group with that of the control condition. If no one in the control condition is thinking about Iraq in reference to the election at hand, and everyone in the experimental condition is thinking about Iraq, the difference between the two groups will give us the true estimate of the impact of Iraq on the race. If, as is more likely, some people in the control condition are thinking about the war, and the priming fails to make some in the experimental condition do so, the difference between the groups will underestimate the true effect.
We do this by experimentally altering question order. Half of respondents in the four polls discussed here received questions about President Bush and the war in Iraq at the beginning of the survey, well before questions about the candidates and the candidate’s issue stances. The other half received questions about Bush and Iraq only after answering questions about the candidates in the Senate race. This experiment was carried in four RDD surveys of likely voters in New Jersey, between in July 2006 and the week before the election (average n: 547).
Results
When we look at the results of the primed versus unprimed conditions across all partisan groups, we can see a dramatic impact of the priming in the early months of the race and relatively little late in the race. In the early surveys, before the campaigns had gone into high gear, priming respondents to think about national issues marginally increased support for Menendez (by 3 percentage points in July and 5 in August), and substantially reduced support for Kean (by 5 percentage points in July and 8 in August).
Combined, these changes substantially altered the spread of the race. A dead heat in the unprimed condition in the July poll – similar to the results of other polls taken at the same time – turned into an eight point lead for Menendez in the primed condition.
Since we know that priming led to an eight point shift in July, we can say that if the entire electorate had been thinking about Iraq, Menendez would have gained at least that much. As evidenced by the decrease in Kean’s support in the priming condition, a substantial proportion of voters preferred him despite his views on the war. When the subject of the President and the conduct of the war were made salient to them, though, they ceased to support the Republican.
However, something interesting happens later in the campaign as voters begin to know more about the candidates, giving us different results in our last two surveys. By the end of September, with advertising and media coverage in high gear, voters were likely to know more about the candidates and issues, making the priming less potent. In other words, all voters are far more likely late in the campaign to be thinking about the President and the Iraq war when they come to the phone to be interviewed. We see that Kean’s support is higher in the primed than in the unprimed condition, and Menendez’s support is lower. The key to these results can be found by breaking the results down by the party identification of the respondent.
Note that in all of the surveys the priming condition slightly increases support for Kean among Democrats. This suggests that a small proportion of Democrats supported the war in Iraq without supporting Kean, and that reminding them of the war led to an increase in support for Kean.
In the two final polls, Kean gained overall support in the priming condition as the result of his slight uptick in support among those Democrats who had misgivings about Menendez’s strong anti-war stance. Also, reminding Republicans of the President and his Iraq policy in the final days of the election seems to have increased Republican support for the Republican candidate. The priming condition, it seems, reminded Republican voters that had been flirting with Menendez where their loyalties lay.
Conclusion
According to Zaller and Feldman’s “simple model” of the survey response, the primary cause of response instability is changing bundles of considerations. On one day, for instance, ethics might be a voter’s primary concern. On another, the war might be paramount. Priming has the advantage of holding these considerations relatively constant, and so, to the extent that we know what the major issues will be on Election Day, it allows us to predict those results well in advance of traditional polling.
The consistency of the primed condition is evident from Figure 1. Not counting those voters who were undecided or supported a third party candidate, Menendez had a four point lead in July. In the final poll, taken a week before the election, the primed condition predicted Menendez would win 52 to 48. On the day of the vote, Menendez won 53 to 45. This is not to say that there weren’t shifts in primed support, but that these shifts are far smaller than those in the unprimed condition, which fluctuated by almost 20 points. In the primed condition, unlike the unprimed condition, and unlike all of the other major polls taken of the race, Kean never had a lead. From day one, the primed condition made it evident that if Iraq were an issue, Kean was going to lose, and that’s exactly what happened. Absent the war in Iraq, the July results among the unprimed group indicate that Kean could have eked out a victory, but the primed condition suggested from the start it was never to be.
Thus, aside from the practical advantages of the experimental use of priming that we have laid out, priming allows us to move beyond anecdotal evidence in explaining counterfactuals. Those of us who carry out polls and report the results to the media and to the public are often called upon to explain what caused a swing, or how the race would be different if something were, or were not, an issue. No matter how well qualified we are to answer these questions, doing so moves us away from quantifiable, scientific polling to the realm of punditry. We can make educated guesses as to the effect of the war in Iraq or ethics scandals on a race, but there’s no way to attach a margin of error to our guesses, and the priming condition allows us to do just that.