On Friday, the Herald reporter/blogger Kevin Rothstein posted the results of a Zogby Interactive/WSJ poll that showed Deval Patrick winning over Kerry Healey by a larger margin than would attorney general Tom Reilly. Rothstein asked if this was the first time such a phenomenon was noted in the polls. This is not the case, because Zogby reported similar numbers in January. In any case, these poll results were then picked up by Blue Mass Group and other blogs. You can see them yourself, along with a number of other races at the Wall Street Journal website. Here are the results for the Massachusetts gubernatorial election, for what they're worth.
| Deval Patrick: | 53.0% |
| Kerry Healey: | 31.5% |
| Tom Reilly: | 48.0% |
| Kerry Healey: | 33.1% |
In any event, every time I've put out the results of a Zogby Interactive poll, I've tried to do so with a disclaimer that the methodology behind these polls is suspect. Zogby has described the methodology in detail, but the short version is that the sample in these polls consists of people who fill out a form on Zogby's website saying that they're interested in taking polls. Zogby then randomly selects from that pool when conducting an opinion poll. In some cases for these WSJ polls, though not in Massachusetts, he supplemented this with more traditional phone calling. Zogby then crunches the raw data, weighting it by demographic factors, including gender, region, age and party. In Massachusetts weighting polls by party is particularly problematic since there are just so few registered Republicans in the state compared to all other parties. The result of all this is that these Zogby Interactive polls should not be compared with other, more traditional polling, since the methodology is so different. If, for example, Patrick polls closer to 43% in a poll released tomorrow, it's not because he lost ten points over the weekend, but it's just an artifact of the different ways those numbers are calculated.
Note that Zogby International (also run by pollster John Zogby) does more traditional phone-based polling, and polls done under that name are usually more theoretically robust.
|