Site Meter

Monday, November 30, 2009

I am puzzled by this Poll in the Washington Post.

I guess this is a bleg. Can anyone explain it to me ?

I care because there is were very interested results of question 14

14. Overall, would you say most of your friends and family think of themselves as (Republicans), most as (Democrats) or most as independents?
Other No
Republicans Democrats Independents (vol.) opinion
11/23/09 All adults 36 33 19 8 5

36% is much higher than the fraction of adults in the USA who self identify as Republicans. It seems from this question that self reported patisan affiliation and perceived affiliation of "most friends and family" are systematically different. I would find this result very interesting if I could understand the sampling design of the poll, but I can't.

This Washington Post poll was conducted by telephone November 19-23, 2009, among a random national sample of 1,306 adults, including additional interviews with randomly-selected Republicans and Republican-leaning nonpartisans for a total GOP sub-sample of 804. Interview were conducted on both conventional and cellular phones. The results from the full survey have a margin of sampling error of plus or minus three percentage points; four points for the sample of 485 Republicans and five points for the sample of 319 GOP-leaning nonpartisans.

Basically the phrases "additional interviews" and "sub-sample" seem to contradict each other.

One possible interpretation is that they polled 1,306 adults and then asked additional questions of those who leaned Republican and called those additional questions an "additional interview". I think this is not what happened, as there are too many Republicans for a random sample 485 is more than 36% of 1,306 and a much smaller fraction self identify as Republican in all recent polls.

Another possibility is that "sub-sample" is simply incorrect. That the "additional interviews" were additional phone calls and in the additional sample only people who leaned Republican were asked further questions. This means there would be two overlapping samples one of the general public and the other of Republican and Republican leaning independents. This is my current guess as to what was done. If so, the answers on question 14 show a gap between self identification and the perceived self identification of "most ... friends and family"

Finally a totally crazy conceivable possibility is that the pollster, TNS of Horsham, Pa., deliberately oversampled Republicans and then reported results for the skewed sample as results for all adults. They didn't do that. Approval of Obama administration policies is 49%. The wording isn't exactly standard but the number is similar to Obama's job approval in other polls. The sample labled "all" isn't grossly unrepresentative. I'm confident that the poll didn't use the totally crazy approach.

Under the second interpretation of the sampling, the result that the median American perceives himself as being to the left of the median American holds (because the all sample is a random sample). That would be interesting. I wish some poll with a clearly described sampling strategy included question 14.

No comments: