send to a friend | print | comment

Skewed samples? Sort of (Updated)

Several readers have rightly pointed out that the Quinnipiac poll released this morning could have yielded somewhat innacurate results by including basically the same amount of Republicans and Democrats. I’m not sure if Quinnipiac accounted for this, but it doesn’t appear to be factored into the final tallies.

Democrats outnumber Republicans by 1.2 million voters in the state now, so the results could be off. I still think the poll is important to note, and it’s still a change from previous surveys.

I had noticed this in looking over the poll, but mistakenly failed to include that caveat in my original story.

Thank you to the alert readers who noticed the omission. I have updated the original post with the following paragraph.

The poll results could be skewed by the fact that an almost even number of Republicans and Democrats were included; Democrats now outnumber Republicans by 1.2 million voters in the state. But it was nonetheless a significant change from previous surveys using similar samples.

UPDATE, 12:24 p.m.—I just put in a call to someone at Quinnipiac who handles some of the methodology stuff. She said she wasn’t sure if or how the registration difference was taken into account, but she’s going to get back to me. Stay tuned.

UPDATE 12:55 p.m.—I just did a little math. If—and I emphasize if—the poll didn’t take into account party registration disparities, and if you plug in the same results through current registration data, you end up with about 49 percent of voters for Specter, 41 percent for Toomey and 9 percent undecided. I cannot overmphasize enough that this might not be the right way to interpret these numbers. The math is correct, but my interpretation of the original problem could be wrong still. It was just an exercise in curiosity while I wait for the folks at Quinnipiac to call back. So still, stay tuned.

UPDATE 2:30 p.m.—Someone at Quinnipiac just confirmed to me that the results were not weighted for party registration. Quinnipiac, like many pollsters, apparently prefers to use party identification, that is, which party people most identify with, as well as other demographics. They ask party registration simply to have accurate numbers for primary polls. However, my contact at Quinnipiac stresses that doing the math like I did above isn’t a totally kosher way to go about it, because the baseline numbers I used reflect party ID, not party registration. Also, the weighted results do reflect something different than just a 50-50 registration split, more along the lines of 45 percent Democrat and 40 percent Republican. Actual registration in the state is 51 percent Democrat and 37 percent Republican. When all is said and done, I think all of us would benefit from a little more caution when reading and interpreting these polls. But this is the way many polls are done. In the end, what matters is that Specter’s lead is narrowing signifcantly. Quinnipiac did the same kind of sample three months ago and Specter had a 20-point lead. Whether he’s still leading in reality or not, there’s no doubt that his edge has gotten smaller.

UPDATE, 3:01 p.m.—Polling guru Nate Silver trusts the survey.

July 22, 2009 at 10:57 am

--Dan Hirschhorn

comments

comments [8] | post a comment