Sunday, October 21, 2012

Rasmussen's Weighting

As it happens, Mark Blumenthal wrote about Rasmussen's weighting yesterday just around the same time I was writing my blog post. It was a piece about the national polls generally and Blumenthal noted Rasmussen's importance as the one firm that has done state-level snapshots before and after the second debate and the fact that they show slight gains for Romney. But Blumenthal explains these gains are related to the "dynamic" party weighting Rasmussen is using:
Rasmussen is one of the few pollsters to routinely weight its samples so they match predetermined targets for the percentage of likely voters that identify as Democrats or Republicans. The catch, as Rasmussen Reports confirms to The Huffington Post, is that its weighting targets are now adjusting on a weekly basis to match the average party identification for likely voters measured on their last six weeks of calling (after weighting for demographics, but not for party). So the party weights for the past recent week may be slightly different than the party weights the week before.

More important, the weight targets for Rasmussen's national samples grew slightly more Republican in mid-October. Although the data are published on pages available to paid subscribers only, Rasmussen indicates that the national interviews for the week of Oct. 8 to 14 gave Democrats a 1-point edge over Republicans (38 to 37 percent). The party balance for the two prior weeks, Oct. 1 to 7 and Sept. 24 to 30, was a 3-point Democratic advantage (39 to 36 percent).

In the past, Scott Rasmussen has explained that the state-level party weighting targets are derived, in part, from national numbers and the "national shifts appear to provide a good indicator" of mid-year changes at the state level.

A 3-point shift toward a more Republican identification would more than explain the one-point shift to Romney in the five states Rasmussen surveyed this week.
Now, here's the thing. IF you're going to weight by party, the shift that Rasmussen appears to have made does make some sense. There was reason to believe, in the wake of the first debate, that a poll should have a few more Republicans in it and a few less Democrats than before. But then, this is why weighting by party ID is a bad idea in the first place.

Anyway, what this tells us is that Rasmussen is shifting their party ID weights slightly over time but they are always doing so in response to last week's polling. And that's problematic. If you are looking for a snapshot of where the public is today (the point of a poll), using their mood from last week as a determinant is really problematic.

No comments: