Wednesday, October 8, 2008

More Food For Thought About Google And Exit Polls

A couple of months ago I conjectured that Google might be able to produce a very accurate exit poll for the '08 Election. Today's post on the Official Google Blog shows some information that's very close to what I suspected they could generate - very cool. Particularly interesting is analysis of traffic from swing states. I'm sure they could do much more here - trending by counties, for instance.

In that previous post of mine, I guessed that they won't publicly release a prediction, but bits and pieces of information and trends like these are a good stand-in for a full blown exit poll.

1 comment:

Timothy J. said...

I agree that trend analysis would be interesting, but I doubt it would be "very accurate" because of two main problems, voting intention and selection bias.

First, how would you predict voting based on searches? You cannot count a search as a vote. Some people are researching the candidate they oppose, and many are probably researching both. They are using Google to fill information gaps, not to register voting preference. The high number of searches on Sarah Palin, for example, has more to do with her being relatively unknown than with her level of support.

Next, selection bias. You are only looking at the subset of people who doing Google searches on candidates. You don't know whether these people registered or likely voters or what percentage of the electorate to they make up. People under 40 are more prediesposed to use Google then people over 40, and to the extent their voting habits are different, it will skew your estimate. The bigger problem is that you know nothing about the other (presumably larger) subset of voters who have done no searches. In theory, these people could be 100% behind one candidate or the other.