RALEIGH – In the nearby links of election stories, I purposely left off a story in the Sunday News & Observer I recommend you read and consider (see here:http://www.news-observer.com/front/story/1713219p-1729085c.html). Written by chief political correspondent Rob Christensen, the piece makes a number of observations about the U.S. Senate primaries on Tuesday, as does Rob’s companion column on how the race reminds him of the 1950s (see here: http://newsobserver.com/news/columnists/story/1713050p-1729024c.html).
Although I won’t come across it until about halfway down, the real news of the news piece is the revelation of a poll by ABC affiliates that shows Erskine Bowles widening his lead in the Democratic primary at 38 percent, followed by Dan Blue at 26 percent and Elaine Marshall at 20 percent. The latter two results show little change at all, but the Bowles number is a four-point gain from a previously released poll. Equally important, the ABC poll shows that Elizabeth Dole’s lead over Bowles is only 52-42, while her more sizable leads remain in hypothetical matchups with Blue and Marshall.
Can these results be believed? Is Bowles surging?
Possibly. But also keep in mind a few technical points about the polls. Both the latest one and the previous one were conducted using automated phone dialers. I don’t have any deep-seated aversion to the technology; indeed, we have commissioned polls in the past that use auto-dialers, which are relatively cheap and thus allow for larger sample sizes that reduce the margin of sampling error.
The problem comes in interpretation. Some pollsters using automatic dialing haven’t adjusted their methodology for weighting the results on the basis of race, income, sex, etc. You probably can’t get away with that. All poll respondents are starting to skew Democratic, given the slightly greater propensity of Republican-leaning voters not to answer the phone, to employ Caller ID, or to use other technologies to screen calls. In order to adjust for this trend, pollsters must weight the results using a model that they believe best represents who will show up on Election Day.
Of course, you can take this too far. One pollster I know was too aggressive in weighting his outcomes, and thus significantly overstated Republican turnout in 2000. This is more an art than a science, and experience with automatic dialing is still so limited that a would-be polling artist has little to draw from.
More problematic this year is that North Carolina is holding its primary in September, not in the usual May. So most historical assumptions about partisan turnout aren’t applicable this go round.
I do believe that Bowles is at least maintaining his small lead going into Tuesday. I also predicted last week on “N.C. Spin” that he would eke out a victory. But anything could happen. Oh, and one last thing: the previous poll that showed Bowles at 34 percent had a smaller sample size. Its margin of error was such that the new result may be just within in it, meaning that there hasn’t been any recent trend towards Bowles.
Basically, polls are useful and entertaining – but don’t get too fixated on them.