« Krosnick book conference | Main | Happy holidays »

21 December 2005

Looking for variance

The old Michigan triad of partisanship, issues, and candidate evaluations as an explanation for vote choices is proving less useful in recent days. The main reason is that party identification and the vote are practically one and the same. In the 2000 and 2004 NES data, better than 90% of partisans voted for the presidential candidate of their party. In 2004 only 40 respondents (7% of partisans) voted against their stated party identification. The last two presidential contests look strikingly like what the Michigan model would call "normal" elections.

For reference I checked out the 1960 NES since it too was a close election with high turnout. Roughly 95% of Republicans voted for Nixon but just 84% of Democrats voted for Kennedy. A larger base of Democrats back then obviously helped put Kennedy over the top. Over the past 45 years Democratic loyalty has almost caught up to Republican loyalty as the Democratic camp has become smaller and presumably more homogeneous.

Extremely high correlations between party identification and vote choice leave little room for other variables to influence the vote. There are at least three consequences of this for researchers. First, we ought to back the research question up a step (or two) and ask what determines party identification. Second, we ought to consider -- and test -- whether issues and candidate evaluations are already in party identification. Third, attention should shift to studying independents rather than partisanship since they broke almost evenly in 2000 and 2004.

Posted by Barry Burden at December 21, 2005 10:59 AM