Fear of Clowns

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

Wednesday, September 29, 2004

How accurate is Gallup? 

MoveOn.org recently ran a full-page ad in the NY Times noting "Gallup's methodology has predicted lately that Republican turnout on Election Day is likely to exceed Democrats' by six to eight percentage points." This is true.

As I've noted before, many have the misconception that Gallup first decides how many Republicans and Democrats are going to show up and then figures out how they'd vote. The actual problem has been that the questions used to determine what a "likely voter" are biased towards people who will vote for Bush - they don't take into account the "anybody but Bush phenomenon".

MoveOn.org's ad doesn't stretch anything other than perhaps calling it a "long standing problem." Indeed, two days before the 2000 election CNN/USA Today/Gallup had Bush ahead of Gore by 5 points - they were 5.5 points off in the spread.

But two days ahead of the 1996 election, the same poll had Clinton 16 points ahead of Dole, Clinton ended up 8.5 points ahead of Dole on Election Day. So the "long standing problem" would go back no further than 2000.

Becoming curious if there was a long-standing bias of some type, I looked into pre-election Gallup polls back to 1996, which Gallup conveniently compiles in two pages. (Note: a common confusion is that the "Gallup" and "CNN/USA Total/Gallup" polls are the same poll - they are not. They use the same data, but different methodologies).

Suspecting that it might be a bias towards the incumbent. I graphed the inaccuracies over a background indicating who the incumbent party was as well as if an incumbent is running for re-election. Here is what it looks like:

Gallup has ...

So it appears that there is a bit of a long-term bias towards Republicans.

What about incumbency? Gallup has,

What about wars? During the Vietnam War, Korean War and WW II, Gallup has,

There might be some biases toward Republicans exhibited by Gallup's methodology, but they would be small. I continue to believe that no poll or methodology can accurately predict how "likely" it is that "anybody but Bush" voters turn out - other than the one on November 2.

Post a Comment

Comments:

 

Post a Comment

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