Friday, August 20, 2004

Polling Statistics

No Right Turn comments on Kevin Drum's criticism of the tendency of journalists to say things such as "statistical tie" when polling results for candidates are within the poll's margin of error - usually the 95% confidence interval. This is further discussed over at Matt Yglesias which adds some expert opinion from a few statisticians.

I don't quite agree. Although Drum's analysis is technically sound when considering a particular election another way of looking at the problem is to define the sample space as elections in general. In other words, in what percentage of elections where polls have shown, say, a 3% lead for a candidate, has that candidate gone on to win?

I think that is what journalists are implying when they talk of a "statistical tie". We have all seen elections where polls have shown a significant lead for one candidate but victory go to their opponent. So it is not really symptomatic of lazy journalism at all, rather a reflection of peoples' experience of polls and elections over time - a different sampling space.


Post a Comment

<< Home