Wednesday, May 10, 2006

Polls Coming Fast and Furious

CBS4 released a new Survey USA poll yesterday which showed Lieutenant Governor Kerry Healey trailing all three Democratic candidates in the gubernatorial election. Here are the results, along with the March Results.

Tom Reilly (D)37%36%
Kerry Healey (R)31%31%
Christy Mihos (I)18%22%
Deval Patrick (D)34%30%
Kerry Healey (R)32%35%
Christy Mihos (I)17%20%
Chris Gabrieli (D)37%NA
Kerry Healey (R)32%NA
Christy Mihos (I)16%NA

CBS4's Jon Keller has a few thoughts on the poll, noting that Healey doesn't seem to be doing well with women (true), that Mihos has a stronger showing in this poll than others (also true) and that Deval Patrick's momentum seems to be slowing (debatable). Patrick is ahead in this poll where he was behind in the same poll in March. I'm not sure what Keller is looking for in terms of momentum, but a seven point swing against the sitting Lieutenant Governor is pretty good news for a candidate is still not widely known.

As far as the Mihos question goes, the difference there is explained by the difference in the polls themselves. You have to be careful when comparing two polls by different polling outfits as they are almost guaranteed to have differences in methodology that will manifest themselves when looking across the data sets. Polls like the State House News poll for example, do not screen out non-registered voters. As you drill down from the universe of all respondents to just registered voters, to just registered Democrats, the size of your population gets smaller and smaller, meaning that the confidence interval (your poll's margin of error) gets bigger and bigger. This difference can partly explain a small but significant swing in a candidate's support when compared to a poll that starts out with registered voters (as the CBS4/SurveyUSA poll does). The other point to make on this is that registered voters have taken a step to show they are at least moderately interested in politics -- they've registered themselves to vote. As such, they're more likely to be following at least the high-profile races and more likely to have heard of the low-name-recognition candidates.

This brings me to another point I wanted to make about polls. People seem to equate them with election night returns. On election night a 1% difference between the candidates is a big deal. In a poll (especially a poll in May) this is just statistical noise. In addition, the predictive power of polls this early out is not very good. Only one candidate has hit the airwaves. The poll numbers will change as the electoral environment does. If you've been following them for a while, you'll notice they exhibit kind of a punctuated equilibrium. They stay the same for a while, then some event (the St. Fleurasco, Patrick's caucus wins, Mihos entering the race, Gabrieli's ads) causes them to shift, and the candidates stay roughly at their new levels until the next event. No one thinks that the campaigns are going to rest on their laurels for the next four to six months.

(hat tip to Kerry Healey -- Out of Touch)