Friday, October 31, 2008

Why polls differ

Polls differ for a number of reasons: How representative is the sample? Do the number of Democrats and Republicans mirror the percentage in the population? Do the pollsters report raw data or used weighted factors to correct for skewed demographic samples? How do they identify "likely" voters? If they don't include cell phone calls, the youth vote is undersampled. How are the questions worded? How big is the uncommitted group? Do they push these people to make a choice? In a binary poll, with only two candidates listed, are third party candidates registered as uncommitted?

What's emerging in this race as a major factor is how the different pollsters decide on "likely" voters. Some pollsters simply ask if you plan to vote and, if so, report your choice. Others also consider your voting history. Have you voted before? With the millions of newly registered voters this time, it makes a difference.

Gallup reports both: that's why on RealClearPolitics the daily polls will have two Gallup results. For today, Gallup has Obama leading by 5% in its "traditional" poll (intend to vote and voted before) but by 7% in its "expanded" poll (intend to vote)

Another source of error is a sample that inaccurately reflects political party affiliation. The Fox News poll out today has Obama leading by 3%, but it's been criticized because they use the ratio of Democrats to Republican party registration from the 2004 election, but that has changed markedly since 2004 with a big shift toward the Democrats.

In contrast, the new CBS/New York Times poll has Obama 52% to McCain 41%. This poll identifies likely voters simply by their stated intention. They have made an estimate that 13% of registered voters will be voting for the first time; in the poll, those voters choose Obama by 2:1. If that holds up, it will be a big win.

**CAUTION** The only poll that counts is the count at the polls on election day.

Ralph

No comments: