Did Health-Insurance Reform Affect the Election?     Permalink

The DNC's communications director, Brad Woodhouse, looked at last Tuesday's exit polls and concluded that health-insurance reform was not a major factor in the election, with only 18% saying it was the biggest issue facing Congress (vs. 62% who said the economy was). Bill McInturff, one of the leading Republican pollsters (and John McCain's lead pollster in 2008), claims to have other data that contradict the exit polls. He said that in the 100 most-target districts, 51% called their vote a message in opposition to the law while just 20% said their vote was a sign of support for it. But this conclusion is so at variance with the exit polls it should be regarded with some skepticism.

What's the Matter with Nevada?     Permalink

For the most part, the public polling was reasonably good. Our projection election day had 51 Democrats and 49 Republicans (counting Lisa Murkowski) in the Senate. The final tally was 53 to 47. We had Ken Buck (R) up 2% on Sen. Michael Bennet (D-CO). Bennet won by 1%, so that is within the margin of error.

But Nevada was a different story altogether. We had Sharron Angle (R) winning by 2%. In the end, she lost by 5%. That is a 7% spread, technically within the margin of error (which is about 4% for each candidate), but still a lot. Josh Marshall at TPM has a piece on this based on an analysis by pollster Mark Blumenthal. So what happened? As a rule, internal polls are much better than public ones (only they are selectively released and who knows whether the figures released as the same as the internal ones). The pollsters working for campaigns have access to internal data and spend much more time on each poll that the firms that try to do half the country. A key issue here is making sure the sample is random. What the robopollsters (like, Rasmussen, SurveyUSA, and PPP) do, is have the computer pick phone numbers at random within a given exchange, for example, 914-949-xxxx (are supposed) to be phones in lower Westchester County, in New York state. By generating and calling, say, 1000 xxxx values and calling them, in theory you get a random sample of a territory that largely overlaps NY-18. By picking the right exchanges, the pollster can cover a congressional district or state.

That's the theory. The problems are manifold, though. For one thing, although exchanges don't cross state lines, they do cross congressional district boundaries, so the voter has to be asked if he or she lives in, say, NY-18. More important though, is that many people nowadays have a telephone number that does not correspond to where they live. For example, a New York-based adult child of an elderly woman might choose a phone number within the free calling area of his mother in Florida. So actually calling people who are potential voters in the election being polled is tricky. The people running the internal polls know the ins and outs of their territories much better than national pollsters who just show up, poll, and leave, once a month.

But even more important than geography is persistence. What happens when you call and nobody answers (which happens a lot)? The robopollsters just try a couple of times at most and then give up. More serious pollsters will try half a dozen times to nail their quarry. The problem with giving up easily is that you end up polling easy-to-reach people, and for various reasons, these are more Republican than the electorate as a whole because young people, single women, and minorities tend to be hard to reach and will be undersampled unless great care is taken to include them in the correct numbers. In Nevada this year in particular, Latinos, especially poorer ones, were hard to reach and seriously undersampled. And they are heavily Democratic. Much of the error in Nevada can be ascribed to missing Latino voters.

Another critical factor is the demographic model being used. Virtually all pollsters try to weight their samples to include the right number of men and women, young and old, rich and poor, Republicans and Democrats, and other factors. But what is the right number? If your model says 39% of the electorate consists of Republicans and the exit polls show it was 36%, then your polls were probably biased in favor of the Republicans. Ditto for all the other factors. So a good model--which is hard to come by--makes a big difference.

Another issue--especially in a low-turnout election--is likely voter models. Every pollster has a secret sauce for figuring out who will vote. Some pollsters just ask people a few questions about their likelihood of voting, but others get lists of who actually voted in 2008 and 2006 (which are public records in some states) and try to correlate these data to phone numbers, but it is an error-prone process.

Now throw in the fact that younger voters, who skew heavily Democratic, often do not have a landline. Robopollsters are not allowed to call cell phones (although traditional human pollsters are), so they miss a sizeable number of Democrats. Again, this can be corrected for if there is a good demographic model available, but the combination of all these factors means that just calling random numbers until you get 600 people to complete the survey isn't enough and sometimes these problems result in bad polls, this year especially in Nevada.

Republicans Control Redistricting     Permalink

Although under the national radar, elections to state legislatures also took place last week, with Republicans picking up 680 seats nationally, twice what the Democrats got in 2008. As a result of this and also the half dozen or so governorships the Republicans picked up, they will have control over redistricting in 195 congressional districts. The Democrats have total control over just 45. The rest have either split control or are drawn by independent commissions. The upshot of this is that the Republicans will be able to gerrymander many districts and increase their representation in the House for the next decade.

When states lose House seats, as is expected in Illinois, Iowa, Louisiana, Massachusetts, Michigan, Missouri, New Jersey, New York, Ohio, and Pennsylvania, they are forced to redistrict. Similarly, in the states that will gain one or more House seats--Arizona, Georgia, Nevada, South Carolina, Utah, and Washington will gain one each, Florida will gain two, and Texas will gain four--a new map is needed. In the other states, a new map is not required, but if one party controls the trifecta of state house, state senate, and governorship, they may do it anyway.

If you like this Website, tell your friends. You can also share by clicking this button