This area does not yet contain any content.
This area does not yet contain any content.
This area does not yet contain any content.
This area does not yet contain any content.
This area does not yet contain any content.

Entries in elections (2)


Dick Morris, Oversampling Democrats & Gallup

This year, you've heard a lot of Republicans claim that mainstream pollsters are oversampling Democrats. Polls that show Obama winning always have more Democrats sampled than Republicans, when pollsters ask about party ID. This is important because around 90% of the identified partisans vote the way you would expect. If you increase the number of people in the poll who say they are Democrat, you markedly increase the proportion of votes going to Obama. When pollsters make calls, they begin with the principle of randomness: whoever they call, they call. But then, they adjust their sample for things like age, gender, race, etc., so that the end result looks like a true cross section of the public.

Republicans are arguing that they need to adjust the samples for Party ID as well. Their eyes always focus upon something they call the "D-plus" factor. For example, if 36% of the respondents call themselves Democrat, but only 29% say Republican (35% Independent), it would be called "D+7" in the current lingo. This is because Dems have a 7 point advantage. Republicans object to this, citing 2010 exit polls and Gallup's massive voter ID studies, both of which claim to show a very competitive electorate, both nationally and in various swing states. Dick Morris is the poster child for this view:

Here's the problem: the argument is seriously flawed. First, it isn't the case that the Gallup data shows an equal electorate. In September, Gallup shows a D+8 and +4 for the final two surveys, and D+7 and +8 if you include leaners: Besides, those numbers tend to fluctuate a lot. If Democrats are about to win, you'll see a high plus factor in the polls. If Republicans are about to win, you'll see different numbers. 

So where is Dick Morris getting this a-priori (fixed) picture about the number of Republican hearts that belong in every poll? It appears the mischief comes from two sources. One is from the exit polls from the midterms, which we know is yesterday's news, not today's. But the other comes from none other than the treacherous likely-voter screen used by Gallup, a constant irritant in this race. Gallup pretty much gives people an exam before they can be classified as a likely voter. They administer a 7-point test. The questions ask if you know where the polling place is, have voted in the precinct before, and if you "nearly always" vote in the past. Wrong answers here get your vote thrown out. I'm not a likely voter according to Gallup. Had no idea where the precinct was until the day I went (Saturday). And I hadn't voted in the midterms. I also only voted in 1 Ohio election in my life (Obama!) because I just moved here about 5 years ago. 

Anyway, here is the point: Gallup's likely voter screen is what causes the electorate to look more competitive than it is. After you throw out the people who fail the test, you end up with Republicans actually leading 36-35 (D-1). See here:  And this is why, of course, Gallup has been widely off in this election. It's had Romney +5 for pretty much the entire month of October when no other poll had those trends. I think Gallup is scared about this, because: (a) it shut the poll down after Sandy; and (b) opened it back up just in time to measure the hurricane effect in their final sample: . This made the poll look like it wasn't a loser, because now it shows a 1-point race. Under the Morris logic, it shows this only among a Republican-biased sample, thanks to their 7-point screen.

Truth is, in all the polls, D+ is naturally high on a random sample of people who say they are certain to vote, because Obama is about to win. That's what happens to party ID. It comes and goes with the tide.        

Dr. Sean Wilson, Esq.

My New Book:


My Election Map