DenBeste on Election Polling
Steven DenBeste has GAFIAted. His blog is still up and is very much worth perusing for the library of well-researched entries.
Yesterday, a single-line entry at the top
points to a very interesting inspection of political poll results presented in an easy to read format.
Steven:
bq. Looking at the
"RealClear Politics" plot of the presidential polls, I see two long term trendlines, punctuated by a hell of a lot of what I would refer to as "experimental error":
bq. I don't believe that public opinion has been changing as much as these polls seem to suggest. The variation we see up through July looks like what engineers call "sample aliasing" or "jitter". Note that it falls well within the oft-claimed ±4 points of error. This is typical for data taken in noisy sampling environments; I've seen this kind of thing many times.
bq. August and September are different. I've seen that kind of thing, too.
bq. In my opinion, the polls were being deliberately gimmicked, in hopes of helping Kerry. In early August it looks as if there was an attempt to engineer a "post-convention bounce", but it failed and was abandoned after about two weeks. But I'm not absolutely certain about that.
bq. The data for September, however, is clearly an anomaly. The data is much too consistent. Compare the amount of jitter present before September to the data during that month. There's no period before that of comparable length where the data was so stable.
bq. The September data is also drastically outside of previous trends, with distinct stairsteps both at the beginning and at the end. And the data before the anomaly and after it for both Kerry and Bush matches the long term trendlines.
bq. If I saw something like that in scientific or engineering data, I'd be asking a lot of very tough questions. My first suspicion would be that the test equipment was broken, but in the case of opinion polls there is no such thing. My second suspicion would be fraud.
bq. In September, I think there was a deliberate attempt to depress Kerry's numbers, so as to set up an "October comeback". Of course, the goal was to engineer a bandwagon.
bq. Public opinion isn't usually as ephemeral as these polls suggest that it is. But there can be long-term trends, and I find it interesting that such a thing actually does show through. It's quite striking how close some of the data falls to the long term trendlines which I've drawn in.
bq.
The reason the Democrats and the MSM are getting frantic is that they're losing.
Emphasis mine.. Heh...
Posted by DaveH at October 17, 2004 11:44 AM