Thus far, we have looked at aggregated data to examine movement in party ID during election year 2004. Combining multiple datasets reduces the random sampling error that can introduce statistical noise to the data and allows for superior subgroup analysis. The results of individual polls, however, are needed to explore properly the relationship between party ID and candidate preference in presidential trial heats during an election year.

The strong relationship between party ID and candidate preference would seem to be the major reason party ID shifts have become such a big issue in the debate about the accuracy of preelection polls. Republican support for George W. Bush was typically reported at a level of 90 percent or more; Democratic support for John Kerry was typically in the mid- to high 80s.

The degree to which candidate preference and party ID tended to move together is shown in Figure 1, which indicates the margin between Bush and Kerry and that between Republican and Democratic party ID among registered voters in the Newsweek polls conducted between March 2004, when Kerry’s nomination was all but certain, and election day. The figure shows the two trendlines moving mostly in concert over this eight-month period. Both followed the same patterns of back-and-forth movement between July and early October, as the key televised events of the campaign—the conventions and candidate debates—took place.

This concurrent movement of candidate preferences and party ID has been interpreted in two ways: those who regard party ID as a constant tend to see deviations from the “correct” party ID distribution as primarily the result of an unbalanced sample that can distort the horserace in favor of one party’s candidate. Those who regard party ID as not a fixed characteristic see such movement as reflecting the same attitudinal shifts that produce changes in candidate preferences. If the first interpretation is correct, changes in party ID at the individual respondent level over the short term should be minimal. If the second interpretation is correct, individual respondents should change their party ID from week to week, or even day to day, in response to real-world events.

Since each Newsweek poll uses a fresh sample of adults/voters, our main database for this study cannot be used to address this issue. There is, however, another dataset that can. Last fall, the Pew Research Center conducted a callback survey of 1,001 registered voters from October 21 to 25 in order to study changes in candidate preferences. All respondents in this post-debates survey had previously been interviewed in three surveys conducted between September 9 and 26, after the GOP convention but before the first Bush-Kerry debate. Although the survey was designed to focus on swing voters who were undecided in September or still open to change, the dataset also included voters who were firm in their choice of candidates, and the data were weighted to be representative of all voters. Voters were re-asked the party ID question in October, so comparisons can be made to study changes in both candidate preferences and party ID at the individual respondent level.

Table 3 shows the overall distributions for candidate preference and party ID among this sample of registered voters as first reported in September, and then in October when they were recontacted. The findings are consistent with Newsweek poll data that did not use any callback sample. The horserace moved in Kerry’s direction during this period, as Bush’s lead decreased from seven to three percentage points. Party ID, which showed the Republicans with a two-point edge in September, closed to a tie in October.

Analysis at the aggregate level, then, suggests greater movement in candidate preferences than in party ID. But when the data are broken down to show movement at the individual respondent level, a different picture emerges.

As Table 4 shows, as many as 18 percent of voters—nearly one in five—reported a different party ID in October than they did in November. That figure is marginally higher than the 14 percent who switched their candidate preference during the same time period. In both candidate preferences and party ID, significant change is apparent in both directions. In party ID, 10 percent moved toward the Democrats, either switching to Democrat or switching from Republican to the “Independent/Other” category. Another 8 percent moved in the direction of the Republicans. These 2004 percentages are consistent with those reported by the Pew Research Center in their analysis of changing party ID over a short period of time in 1988 (16 percent) and 2000 (18 percent). This would suggest that it is normal for a large segment of the voters to switch their party ID over a relatively short time frame, at least in the context of a presidential election year.

Movement in party ID was not concentrated in certain demographic subgroups—the percentage who changed their party ID was not significantly different by gender, age, race, or region—although small differences were found by education: those with less than a high school education were significantly more likely to switch their party ID than college graduates (24 percent vs. 11 percent).