Sunday, March 22, 2009

academia

I was fooling around with the ANES 2008 time series data when I noticed a problem - the sample reported voting 65.5% for Obama and 32.9% for McCain.

Granted that it's a preliminary release, but you'd think somebody might have noticed that.

Being post-election data, there's a possibility of some winner's bias. Need to check back with previous data sets before I get too riled up.

2 comments:

Anonymous said...

I don't understand what you are talking about here - would be interested to hear more

- Matt

JoeCollins said...

The sample is supposed to be representative of the general population. It's over 10% off on the Presidential ballot question, so I'm questioning how much the sample really reflects the population.

Winner's bias : The study has two components, pre and post election. The question of "who did you vote for" is asked after the respondent knows the outcome of the election and has been subjected to the resulting media fanfare, and may lie to the interviewer about who s/he voted for.

If previous studies also show some winner bias in the post-election side, it may actually be reflective of the population... including the part of the population that lies to interviewers. However, those respondents who "remembered" their 2004 vote also overwhelmingly voted for Kerry, which will eventually lead me to look at the 2004 study when I get around to it.

It's a really lengthy study, so there's lots of other good stuff in there for researchers to work with, but either way, researchers should be cautious about what conclusions they draw from the Prez ballot question. I'm sure a lot of correlations will generally hold, but those of us who are just trying to count to 50% + 1 are looking in the wrong place.