Problems with polls

One of the problems with opinion polls in elections is that the margins of error are often greater than the majority for either side.

This can be seen most graphically in the Electoral Vote Predictor for last November's U.S. presidential election.

On election day this map shows what the editors of Electoral Vote Predictor believed the result would be.

Apart from getting the overall result wrong by predicting a john Kerry victory, no fewer than five states were won by the opposite party, and the errors were not all to the Democrats. New Jersey was also called an "exactly tied" state. I note Virginia listed as "Barely Bush" when it was won by about 9 percentage points.

The fault is not with the website editors, although they could have disregarded the less credible Zogby polling.

No comments: