Wednesday, October 11, 2006

Let's Talk About Polls

Do we have an expert on polling or at least interpreting polls in the audience?

I ask because the SurveyUSA poll of Maryland voters on the senatorial and gubernatorial race is making me scratch my head. Maybe you can help.

So, it's not a big surprise that whites favor Ehrlich but maybe a bit that they tilt toward Steele at the same rate. And Steele does significantly bettter among African Americans. The numbers that threw me were that among Hispanics Ehrlich beats O'Malley 50 to 17%, and Steele bests Cardin 61 to 37%. This isn't what most of us expected. What's the deal? Anyone want to take a look at the methodology to take a crack at whether these numbers make sense?

5 Comments:

Blogger Unknown said...

It is odd. Usually, SUSA is one of the best polls, but the Senate poll is a real outlier.

10/11/2006 01:09:00 PM  
Blogger Andrew Kujan said...

Not to mention, if Catholicism had anything to do with this, O'Malley should be scoring big. Baltimore just demolished one of it's most beautiful historic buildings, the Rocheambeau Hotel, to make way for a private Catholic prayer garden.

10/11/2006 02:59:00 PM  
Blogger MoCoPolitics said...

Not a polling professional, but a poll geek of long standing.

Adding to what others have said:

There also was an oversample of Republicans. The sample was 50% Democratic, 34% Republican and 16% other. Current MD voter registration figures are 54.9% Democratic, 29.0% Republican, and 16% other.

I also agree with Mike Raia that the subsample of Hispanics is so small as to be statistically meaningless. Looking at the breakdown, I believe the actual number of Hispanics was 12 rather than 10, which would be 2.29% of the sample. Half (6) supported Ehrlich, 17% (or 2) supported O'Malley and the rest were undecided. The 1% for Driscoll is clearly a typo as there is no way to get 1% of 12 voters -- even 1 vote out of 12 would be about 8%.

Even with respect to the black vote, the number of African-Americans in the subsample was in the neighborhood of 131 -- the margin of error here is likely 10-12% or more.

If you want to know about blacks or Hispanics or Asians or other groups, it is not a good idea to look at subsamples of broader polls. It is necessary, particularly in the case of groups such as Hispanics with language and other cultural barriers, to have a poll specifically geared to that group, and to have a sample, as Mike suggests, of at least 600-700 responders. I would submit that a figure of 1000-1500 would be far superior.

10/11/2006 03:48:00 PM  
Blogger Kevin Gillogly said...

MoCo,

I think their sample size by party is spot on. Unaffiliateds tend to vote at fifty percent; Dems around high fifties to low sixties; and Reps slightly higher.

Add in the fact that the GOP is making a concentrated effort to vote absentee and the stories in general about faulty machines, lack of judges, disenfranshisement, and a possible lower turnout among PG African-Americans makes their universe plausible.

As for the sample size based on race, I will leave for others.

10/11/2006 06:13:00 PM  
Blogger Kevin Gillogly said...

Mdman,

I realize that. Both sides do in every election. However the extent of the GOP absentee ballot program is more intense than the Dems. The GOP has the "bully pulpit"; the dems do not. GOPers have a sent out more absentee forms than the dems have. I have seen twice as many mailings for the GOP absentee program than I have for the Dem program.

Again I do not deny that dems are doing it; but I believe that the GOP is doing it better.

10/12/2006 04:24:00 PM  

Post a Comment

<< Home