Tuesday, October 21, 2008

Presidential Polls Are Not Quite Useless

With two weeks to go in the election many people are carefully watching the presidential polls, particularly the state-by-state ones that will determine victory in the Electoral College. Political polls are based on models of people who will most likely vote, or people who voted in the last election, or some combination of both of them. This model has flaws that are becoming greater as time goes by. Foremost among them is that telephone polls don't include people who don't have telephones. Not many you say. Wrong! Many young adults, even professional ones don't own a landline. A significant number of "poor" people don't have landline to either. So these subpopulations cannot be included in telephone surveys.

Telephone polls also encounter the language problem. There are many American citizens who speak primarily or only Spanish, or another language. These people are excluded from telephone polls and become an important population subset that is not represented in the findings. The fact that these people tend to vote Democratic skew the results towards Republicans.

This year there is a special subpopulation that isn't accounted for in the sampling procedure used by telephone surveys. The Democratic Party has enrolled millions of new voters across the nation, and particularly in key so-called battleground states. If the Democratic Party enrolled these people, it is logical to assume that a significant majority of them will vote Democratic.

Telephone surveys never include people who refuse to talk to the pollsters. So by definition, telephone poles leave out a population subset that is hard to replicate.

If you are a McCain supporter hoping that he will close the gap, don't rely on poll information to support your belief that the race is getting closer. If you are an Obama supporter, you can't be sure whether all your supporters are included.

People have all too often said that bad data is better than no data at all. That is one of the most shortsighted, ignorant statements possible.

No comments: