Wednesday, September 24, 2008

Biased Polls: Party Affiliation

In a continuing series of posts about biased polls, I wanted to take a little time to talk about party affiliation. There are some subtleties involved in looking at poll internals and it is difficult to understand the bias in some cases.

Quite simply, many pollsters are playing games with party affiliation and it is destroying the accuracy of the results and the credibility of the pollsters.

For each voter, there are two characteristics that are separate and distinct: (1) who you are voting for and (2) what your party affiliation is. The probability for a person to vote for someone "on the other side" in terms of party in one particular election is much higher than the chance that they change party affiliations.

Consider the term "Reagan Democrats". Who were these people? They were people that were lifelong Democrats that chose to vote for a Republican candidate--Ronald Reagan. More important is who they were not. They were not people who, after Reagan, became affiliated with the Republican party. They were still Democrats.

People change party affiliations very very rarely. From my readings, it is estimated that changing once in a lifetime is rare; twice is very rare. In someways political affiliation is like religion. Some suggest that the best indicator of your party affiliation is your parents party affiliation.

If you look at the current pollsters, they are changing their party affiliation numbers day by day. That's ludicrous. It is certainly possible, after a major event like a stock market crash, that support will shift from one candidate to another. Pollsters should try to capture that change of public opinion. But is absurdly unlikely that one week there the electorate is 35% Republican and the next it is 29% Republican and the next 37%. People don't change parties like that. Support for a candidate can waver--party affiliation does not.

A majority of the biases being seen in current polling can be attributed to this mistake by the pollsters. Obama releases a new ad, a poll is conducted, and his support has jumped by 10%. That in alone is suspect. But then you look at the internals and they suggest that 7% of the population changed political parties because of that ad. That wouldn't be a simple ad--it would be a form of mind control. It just doesn't happen.

So try to remember this when you are panicking or cheering some poll results. If the internals suggest that everyone is abandoning one party for another on whim, the best thing to do it ignore the results and move on.

No comments: