Baked While You Sleep

I stayed up late last night—all the way to 10:00pm. At that point many of the important races had not been decided. I awoke to something which while still not technically a wave was pretty close to it, especially from the point of view of the Senate. I think that USA Today’s presentation of the data is a good one.

Several races surprised me. I didn’t see that horserace in Virginia coming and I suspect that the Democratic Senate re-election committee didn’t, either. As of this moment Warner is holding on by a hair. It’s definitely recount territory.

For the Senate elections to my mind the real story is how were so many polls so wrong? There were many races that were supposed to be closer that weren’t really close at all (Kentucky, Georgia, Kansas). What was going on there? Was the desire for a horserace story so overwhelming?

Here in Illinois Republican Bruce Rauner has been elected governor but incumbent Pat Quinn isn’t conceding:

Multimillionaire Republican Bruce Rauner proclaimed victory and “a new direction” as Illinois’ next governor Tuesday night, even as Democratic incumbent Pat Quinn refused to concede defeat.

“This is a historic time in Illinois,” Rauner told cheering supporters. “The voters have spoken. The voters have asked for divided government for the first time in many years.”

But a defiant Quinn declared, “I don’t believe in throwing in the towel” with votes uncounted.

“We will never, ever yield to a result until all the votes are in,” Quinn said, suggesting a complete count could take days

As I read the Illinois recount laws the results won’t be within the margin for a legitimate challenge and a judge would be wrong in granting a statewide recount but I wouldn’t be at all surprised if there were court challenges and there was a recount. I’m having a flashback to 2000 except without the reasonable doubt.

Interesting times.

13 comments… add one
  • TastyBits Link


    … how were so many polls so wrong? …

    The problem with mathematical or computer based upon datasets is that numbers are subjective. They may coincidentally work, but they may not necessarily work. The second is required for repeatability.

    This is the same problem staticians fail to take into account whenever creating their models. The housing market may respond according to their model, but it does not necessarily respond. The banks may respond to the Fed interest rate, but they do not necessarily respond.

    If you want accurate polling data, put Google to work.

  • PD Shaw Link

    Late last night, Nate Silver estimated that the polls had about a six-point bias in favor of Democrats in the most polled Senate races. Link (Not counting Virginia or Alaska, but however Virginia turns out, it looks like the polls were off more than six points)

  • Guarneri Link

    Does anyone know what in the world Quinn is thinking other than that the pre-printed Quinn ballots stored down near the sanitary canal in Stickney hadn’t made their way to the tabulators?

  • Andy Link

    The problem with polls is that they can’t know who will actually show up to vote.

  • PD Shaw Link

    Guarneri, last night when Quinn wasn’t conceding, there were still about 130 precincts, almost all in Democratic areas (Cook, St. Clair, Winnebego). I think now the concern might be that Republicans announced a vigorous poll-watching system, which might have created more challenges and provisional ballots than normal and they would skew Democratic. Thing is, that scenario seems more plausible in places like DuPage; poll-watchers don’t get to decide challenges. Looking at the results in Cook County, Rauner’s numbers are the best for a Republican at the top of the ticket since Ryan in ’98.

  • PD Shaw Link

    The dirty tricks played on the Republican poll-watchers is evidence of the concern.

  • Gee, you mean that heavily red states didn’t suddenly elect a bunch of Dems to back a very unpopular Dem president? Who’ld a thunk it!

  • Guarneri Link

    Heh. Here in duPage Cty the poll watchers looked more like my former grandmother, all but handing out cookies.

  • PD Shaw Link

    Nate Silver has updated his “bias” analysis to include all races — the polls had a 4.0 Democratic bias. But, if you focus on the most-polled races (those with more than four polls taken in the last three weeks), the Democratic bias is 5.1 percentage points. The greater the number of polls, the more “off” their aggregate results were, which is certainly counterintuitive.

  • Not that counter intuitive, PD. They kept pollibg till they got the results they wanted.

  • TastyBits Link

    The computer models work great until they do not. In the objective world, this means that they never worked. They were always worthless. Is anybody getting the picture?

    It does not matter how many times a person reads the goat entrails right. Goat entrails are not an objective modelling methodology. Voodoo is voodoo. It makes no difference if the goat entrails are digital.

    Need I bring up another set of electronic goat entrails that is failing? (To be honest, there are so many. I am not even sure which set I mean. insert The Hunting of the Snark reference)

  • PD Shaw Link

    @Tastybits, I think the assumption that an aggregation of different polls are going to be fairly accurate within a few percentage points still makes sense. But Silver is suggesting that pollsters improve the quality of their polls by checking with other polls and making some adjustments in their model. He calls it “herd” effect. If so, then the polls are not truly independent and the assumptions aggregation operates on are not accurate.

  • TastyBits Link

    @PD Shaw

    The number of significant variables needed for these models is well beyond anything the model makers imagine. For a niche area, a model maker may be successful, but few of these scale up or out. The housing market disaster, the financial market disaster, the peak oil bust, various population explosion busts, and the CO2 problem are all examples.

    Adjustments are not the problem. Missing variables are the problem. Without the variables, it is impossible to understand the way a system works. These missing variable may allow large distortions into the system, and there is no way to predict if or when they will manifest. The model works. Then, it does not, and nobody knows why. It was luck not science.

    None of this address the intentional uses of distorted polls. There is a substantial amount of disinformation. Polls can depress voting. If the opposition’s voters do not vote, there are fewer votes against your candidate. Politics is about hustling, and hustlers hustle.

Leave a Comment