Thursday, December 09, 2004

Is That Poll Real, or is it Memorex?

Can't let this pass me by. A couple hours ago, Slate posted this posthoc analysis of election polls. So, which polls were most accurate, and why? They found three key areas:

1. Party Identification: Historically, Democrats bring more people to the polls, but this time it split close to 50/50 (roughly 37% of the electorate came out for both the GOP and Democrats evenly). Polls that used historical data to weigh responses were wrong.

2. Undecided Voters: Traditionally, these voters break for the challenger, but in 2004 they broke only slightly for Kerry.

3. Automation: The most interesting finding is that the polls using automation (i.e., Rasumussen and SurveyUSA) were the best predictors of the winner in the bit battleground states. Why will no doubt launch a thousand poly sci dissertations.

No comments: