This paper addresses a simple but basic question: How many voters change their mind during the month preceding an election? The question is simple and basic, yet the vast and rich literature on voting and elections has somehow managed to avoid it. Despite all the talk about the impact of campaigns, I have not come across estimates of how many people vote for a party other than the one they intended to support at the beginning of the campaign.¹

The question is important. It makes a big difference whether “only” 5% of the voters change their mind in the month preceding an election or whether as many as 25% do. Our understanding of the meaning of the vote and of elections is bound to be affected by estimates of how many (or few) people change their mind during election campaigns or even on Election Day.

The purpose of the paper is to provide an estimate of the proportion of voters who vote differently from how they intended. The objective is not to explain why some voters are more likely to change their mind than others nor to explain why the vote is more stable in some countries than in others. The more modest goal is to get the “facts” straight. As indicated in the first paragraph, we simply do not know how many voters typically shift from one party to another in the month preceding the election, and we do not know whether the proportion varies substantially from one country to another. Establishing the facts, that is, the relative magnitude of vote change and its variation across space, should help us formulate more plausible theories about the factors that make voters more or less prone to change their mind.

To address this question, I have assembled all the election studies that I could find that included a campaign/post-election panel component; this allows for a comparison of vote intentions in the month (30 days) preceding the election and actual vote choice (reported in the post-election survey).

I have assembled a total of 27 election studies, conducted in five different countries: Britain, Canada, the Netherlands, New Zealand, and the United States (see the Appendix for some basic information on each study). A total of 27,820 respondents,²

An individual is defined as having changed her mind if she said in the postelection survey that she voted for a party other than the one she had indicated she intended to support in the campaign survey. And, of course, an individual is defined as having not changed her mind if she reported having voted for the party she had mentioned supporting in the campaign survey.³

All individuals who did not vote, did not indicate a vote intention in the campaign survey, or refused to say for which party they voted, are dropped from the analysis. I consider only people who voted and indicated a vote intention in the campaign. I am thus only concerned about vote choice, and not about the decision to vote or not to vote. And I focus on those who already had formed an intention during the campaign. I wish to estimate the proportion, among those with an opinion, who change their mind. There is an additional number of people who make up their mind after being interviewed; these people are not included in this study.

Vote intentions should firm as the campaign progresses and voters become more informed about party positions on important issues (Gelman and King 1993), so that by Election Day very few people should change their mind. I expect a quadratic relationship, that is, the propensity to move from one party to another should drop markedly in the last week of the campaign. The propensity to change one's mind is therefore regressed on the number of days before Election Day (DAY) and the same variable squared (DAY²).

I also wish to determine whether voters are more prone to change their mind in some countries than in others. Converse (1969) has argued that voters are more likely to develop strong and stable party loyalties when the party system is stable and has been so for some time. From that perspective, we might expect voters in Britain and the United States, where political life has been dominated by the same two major parties (for almost a century in the case of Britain and almost a century and a half in the case of the United States), to be less volatile than voters in the three other countries. I created four country dummy variables (Britain being the reference category) to capture overall differences among countries.

Table 1 presents the findings. They confirm that the propensity to change one's mind declines as the campaign progresses. The data support the hypothesis of a quadratic relationship, the drop being steepest in the last part of the campaign. They also indicate that Canadian, Dutch, and New Zealand voters are more likely to change their mind than their British counterparts, and that American voters are the most stable of all.

The implications are illustrated in Table 2, which indicates the predicted probability of changing one's mind in each country at different points in time. The median situation, represented by the Netherlands, is for 18% of the voters, about one voter out of six, to change their mind in the month preceding the election. Most people stick to their initial choice but the fraction that does shift from one party/candidate to another is far from being negligible. I also find vote shift to be much less frequent in the U.S. than in other countries. New Zealand is at the other end of the continuum. It would seem that almost one voter out of three changes her mind in that country.

Table 2 also shows that it takes time for vote intentions to firm up. The propensity to change one's mind starts decreasing only in the last two weeks of the campaign. The data suggest that some voters are still shifting from one choice to another by Election Day: about 4% in the U.S., 7% in Britain, 9% in the Netherlands, 10% in Canada, and 16% in New Zealand. These figures support the view that for some people the final decision is made at the very last minute.

It has been argued that the propensity to shift from one party to another has increased in more recent elections, as party attachments have weakened (Dalton and Wattenberg 2001). This raises the question whether more voters are now changing their mind than previously. The data set does not allow a clear test of the hypothesis because outside the United States, all the elections except one in the Netherlands were held after 1985. I did perform a logit estimation of the U.S. data set including a time variable and that variable proved to be significant. According to this estimation, the probability of changing one's mind in the month preceding an American presidential election moved from (slightly above) 6% in 1960 to (slightly above) 8% in 2000.

In short, about one voter out of six typically changes her mind during the month preceding the election in the five countries considered in this study. The amount of vote shift is, however, much smaller in the United States and much larger in New Zealand. And vote intentions firm up only in the last two weeks of the campaign. Even on Election Day somewhere between 5% and 10% of the voters typically change their mind.