Looking next at the relationship between change in candidate preference and change in party ID, we found a fair amount of party ID change even among those who did not change their candidate preference between September and October 2004. Table 5 shows that 15 percent of those who supported the same candidate as last time (or remained undecided) nonetheless reported a different party ID.

Those who switched candidate preferences, however, were about twice as likely to switch their party ID as those who kept the same candidate preference (31 percent vs. 15 percent). Those who switched both candidate preference and party ID were more likely to move in a particular direction. Those who changed party ID but not candidate preference were about equally likely to move toward the Republicans as the Democrats (7 percent vs. 8 percent); those who changed both were more likely to move toward the Democrats (18 percent vs. 13 percent toward the Republicans).

It is worth noting that the largest group of switchers in candidate preference were those who identified themselves as Independents or otherwise not aligned with a major party in both September and October (40 percent), while about three in ten (28 percent) identified with the same major party both times, and another three in ten (31 percent) changed their party ID along with their candidate preference.

What can we conclude from the analysis shown in our three-part series on party identification? Building on an initial study in Newsweek polls conducted from 1993 to 2001, our new study provides further evidence that party ID is a soft measure—that is, an attitude rather than a fixed demographic characteristic appropriate to use in weighting survey data. Yearly averages from Newsweek poll data, which reduce sampling error to a minimum, show a significantly more Republican party ID distribution during George W. Bush’s first term (2001-04) than during Bill Clinton’s second term (1997-2000). Pollsters who used the five-point Democratic advantage in party ID from the 2000 national exit poll as a weighting parameter either ignored this evidence or viewed it as irrelevant. Even if their final poll results were reasonably close to the actual results, their performance was clearly not improved by flawed assumptions about party ID.

Quarterly averages from the Newsweek poll show that party ID did not change gradually in recent years, but at times shifted rather suddenly. For example, between the last quarter of 2002 and the first quarter of 2003, party ID moved from a four-point Democratic advantage to dead even. With the benefit of hindsight, it is easy to link shifts in party ID found in surveys to real-world events and changes in the political climate. But in real time, this movement is often not so obvious.

Taken as a whole, the post-9/11 period leading up to election year 2004 was associated with gains for the Republicans, but the quarterly averages did not move in a straight line for long. They changed direction four times between 2001 and 2003. After Saddam Hussein’s capture in December 2003, one might have expected 2004 to be a good year for Bush and the Republicans. In the end, it was—but not in an entirely predictable manner. The Democrats bounced back to regain their “normal” five-point party ID advantage in the second quarter of 2004, but the year ended with the GOP cutting the margin to one point in polling over the final quarter of the year.