Monday, October 15, 2012

Which poll do you believe?

I went to a timely seminar the other day given by Adam Berinksy, Professor Political Science at MIT.  His topic was "Opinion Polls and American Politics."  Offered by MIT Hillel, the subtitles of the event were, "How are polls conducted in America? How do you tell a good poll from a bad poll? And what should you pay attention to as the 2012 election draws near?"

Demonstrating the issue of sampling error, Adam hearkened back to one of the most notorious inaccurate polls, that conducted in 1936 by The Literary Digest.  Ten million "ballots" were sent out to people, and about 2.4 million were returned.  These demonstrated that Franklin Roosevelt would get 41 percent of the popular vote. When the election occurred, he received 63 percent.  The problem:  A non-random sample.

Sampling, though, has gotten more difficult, as seen in this summary by the Pew Center, with a smaller percentage of people being contacted, and fewer still cooperating and responding.


I asked Adam about a theory that I had heard several weeks ago, that presidential elections turn on the rate of consumer confidence in the summer before the November election.  Below a certain number, and the incumbent loses. Above that number, he wins.  Adam replied that this hypothesis is based on a very small sample of elections, 12 in all, and therefore should not be considered dispositive of the issue.

So, this is either bad news or good news for Mr. Obama.  And it is either good news or bad news for Mr. Romney.  That is a sure thing!

No comments: