Thursday, September 11, 2008

Biased Polls: How It Is Done

Today Public Policy Poling (PPP) released a poll on Colorado with the results showing Obama 47%, McCain 46%. Sounds fair enough. Colorado is known to be a battleground state this election. The DNC was held in Denver, so that should give Obama a boost relative to national numbers which show him slightly behind.

But let's look at the internal numbers just a little. In terms of weightings, they used the following breakdowns:
Republican: 37%
Democrat: 36%

Hispanic: 14%
White: 77%
A commenter over at HedgehogReport pulled up the exit poll numbers from the 2004 election:
Republican: 38%
Democrat: 29%

Hispanic: 8%
White: 86%
So in this case PPP is predicting that in 2008 Democrats will see a 7% higher turn out and Hispanics a 6% higher turn out. The question is why? The Hispanic population in Colorado is indeed growing but it hasn't grown by that much since 2004. Obama has certainly energized Democrats but the same can be said for Palin and Republicans. Both are pulling for independents and to suggest 7% of the population will change from independent to Democrat is a rather large assumption.

Of course no one can predict the future; the breakdown of party lines and race might be exactly as PPP is predicting. But the point is they really have no scientific basis for the new numbers. The new numbers could also be the result of over-polling the Denver metropolitan area. If you adjust the number to a less severe weighting change, McCain has a 5% advantage.

I'm not claiming this bias is intentional. It just serves as an example how careful you have to be when conducting a poll (and analyzing poll results), especially in a battleground state.

No comments: