Tuesday, October 25, 2005

Unpopular senators

SurveyUSA has their latest tracking poll of 100 US Senators. Roberts and Brownback are #71 and 87 respectively.

Brownback should be worried that he's less popular than so many other presidential contenders from the Senate, and that he's so weak among women and moderates. Self-identified moderates in Kansas are to the right of moderates in other states, probably even in Iowa, but certainly in New Hampshire.

Of course, this poll only asks about performance as a Senator. I'm sure you'd get very different results if you asked whether people would vote for him in the presidential primaries.

I also want to take a minute to beat up on FDM. Not for their weak analyses or patchy posting schedule, but for attempting to appear well-informed on polling methodology, and blowing it badly.

They critique these SUSA tracking polls, arguing essentially that they have a bias towards Democrats because of what FDM asserts is a bias toward self-identified Independents. FDM wants SUSA to impose a likely voter screen, a process described well by the Mystery Pollster. Basically, if you want to predict the outcome of an election, you ask "If the election were held today/On election day, who do you plan to vote for?" If the respondent isn't going to vote, you basically don't care who they'd vote for since they won't vote.

In contrast, if you want to know how people view a public figure, not how people will vote, an LV screen is inappropriate. Not unnecessary, inappropriate.

FDM notes that the polls from 2004 based on LV screens had about 15% Independents, while SUSA's polls show about 35% (it varies poll to poll). If you look at the statewide voter registration data, you see that Independents are about 27% of the registrations (including 3rd party registrations). The latest SUSA poll has 37% Independent.

Myster Pollster also has an important discussion of reweighting by party ID, and is fairly negative on the practice. He says:

The most important thing to remember is that Party ID is an attitude, not a demographic. People can change their views of political parties. They cannot change their age, gender, race, years of education and locale (unless they’ve moved).
It's common and uncontroversial to reweight by those demographic factors. Reliable census data exist on the actual ratios of all those things, and they don't change.

In contrast, there's reason to expect an interaction between approval ratings of public figures and party ID. As people come to dislike the Republican president, the Republican Congress, Republican appointed judges, a Republican state AG, etc. they will come to dislike the party, too, start calling themselves Independent rather than Republican, or even calling themselves Democratic. Reweighting is inappropriate because it makes it impossible to catch that dynamic, a dynamic that's very important in the long-term trends a tracking poll is meant to capture.

That's why the interesting story is not the precise numbers (52% approval for Brownback, 34% disapproval), but the fact that his approval had a small bump last month, but fell back to where it was, while his disapproval is climbing steadily. Whether there are methodological flaws or not (and SUSA is pretty reliable), that month to month trend will be the same.

FDM and their readers take some joy in ignoring inconvenient data, and that's fine. But that's a good way to lose. Polls are people speaking. Ignoring the ones who disagree is a guaranteed way to fail, and that's the only way to turn the third of Kansas that calls itself Independent into a sixth of your sample.