US Exit Poll Data Confirms Net Suspicions ... or not

The release of complete exit poll data for the 2004 US presidential election has wags like this and this are peering into the discrepancy between exit poll data and actual poll and perceiving scandal (at the extreme) or anomaly:

I've a lot of respect for Lessig. His work in the copyright arena is thought provoking. Unfortunately, a keen understanding of copyright law doesn't translate to survey sample research. The creative presentation of the sampling discrepancy -- arrayed as it is, creates the illusion of some mysterious invisible hand at work. I believe the graph reveals little more than normal sample bias. If so, sampling bias is indeed on display here, this bar chart displays nothing unusual or untoward; just the artifact of random errors doing their thing to "bias" the estimate. That's why we call it an estimate, after all. End of mystery.

That said, it may be interesting to obtain similar data from past presidential elections and create similar displays. My hunch is that the observed patterns will be fairly consistent. Sure, the states will likely change places, but that's what one would expect from an analysis of sampling error ... unless, one could identify a factor that introduces systematic bias into the estimates.

Are these estimates outside what one would expect statistically? A question that could be answered definitively.

Again, exit poll data are estimates and should be regarded with all the cautions they are thus due.

No comments:

Post a Comment