Fear of Clowns

"Faith may be defined briefly as an illogical belief in the occurrence of the improbable."
- H. L. Mencken

Saturday, September 18, 2004

Likely voters, flawed methodologies and the anybody but Bush phenomenon 

A particular observation regarding Gallup polls from The Left Coaster has been rather popular among those who don't like the results. In a nut-shell, the argument is that in the 2000 general election, exit polls had 39% identifying as Democrats, 35% as Republicans and 27% as independents, but the Gallup group of "likely voters" in their most recent polling for the 2004 election, 40% identify as Republicans, 33% as Democrats and 28% independents. The conclusion inferred by many is that Gallup is intentionally oversampling Republicans.

That, however, is not necessarily what is being reflected. Which political party the subjects of the poll identify with is a question asked in the poll. There is no way of telling what party the population polled identifies with unless you ask them - many states, such as Texas, do not register voters by party. So when it's reported that 40% of likely voters identify as Republican, the result is due to the way Gallup determines the likely voting population and is subject to the poll's margin of error and confidence level.

The art in the methodology of political polling is coming up with a group of voters who are likely to vote. You have to figure out some way to eliminate a potion of the population polled because voter turnout is always less than the number of registered voters - in 2000, for instance, national turnout was 67.5% of registered voters. A theoretical pollster who perfectly determined "likely voters" would have used a methodology that found 675 out of 1000 people polled were "likely voters".

A simplified example of how a "likely voter" may be defined could be someone who,

Moving through those rules to find "likely voters" in a polled population of registered voters,

You're left with 725 registered voters that according to your formula are "likely voters". And among those 725, a certain percentage say they're Republican, a certain percentage say they're Democrat and a certain percentage say they're independent. A pollster may indeed use party identification in their determination of a "likely voter", but it would only be part of the formula.

Although the example methodology given is just an illustrative example, the types of determinations mentioned are in fact the types of things pollsters use to define the "likely voter" population. Formulas which may be accurate in "normal" elections may be flawed for this one: many people will be going to the polls with the sole intention of kicking Bush out of the White House.

I would suggest that pollsters should include everybody who would answer "yes" to "Do you despise George Bush so much that you will vote against him no matter what?" Some of those people are eliminated based on the strength of support they have for their candidate, voting history, interest in election and for any other reason not to include them as "likely voters". But they will still vote.

I've previously noted that FOX's polls have been among the few where Kerry has done better among the pollster's definition of likely voters than among registered voters. This is what you'd expect to see when the electorate is motivated to vote for a candidate in order to oust the incumbent.

Looking at the internals of the most recent ABC/Washington Post poll, we find that 55% of Kerry supporters said they're voting against Bush, whereas only 14% of Bush supporters say they're voting against Kerry.

The poll showed Kerry has whopping 41 point advantage in these areas: among people who are angry enough to vote for the first time in years, that will be voting despite paying little attention to the campaigns, and even despite not being overly excited about their candidate. To counter-balance the anger against Bush, his campaign has tried to decrease Kerry supporters' desire to vote for Kerry - so far they've only succeeded in creating a 15 point advantage for Bush in the area of strength of support: In the same poll, Bush showed 90% strong support and Kerry showed 75% strong support.

In conclusion, many of the polls we're seeing may indeed be using flawed methodologies, but it's not because they're choosing to include more Republicans than Democrats. It's because they're eliminating people from "likely voters" for reasons that do not apply when a large part of the electorate will be going to the polls for the main purpose of getting Bush out of office.

Post a Comment



Post a Comment

This page is powered by Blogger. Isn't yours?
Listed on BlogShares