NYT columnist says some positive things about election forecasting models

Nate Silver writes:

The forecasts made by Robert S. Erikson and Christopher Wlezien, for instance, are done very well and do a good job of accounting for pertinent information without resorting to data-dredging.

Unfortunately, Nate doesn’t get to that until paragraph 55 of a 58-paragraph article. I fear that casual readers will be misled by the tone of the article and miss the important point that thoughtful political scientists do know what they’re doing here. (On the plus side, Nate does start off with a link to my article with Gary, which I think still holds up well after nearly 20 years.)

11 Responses to NYT columnist says some positive things about election forecasting models

  1. pc March 26, 2012 at 10:49 pm #

    Am I the only one who thinks that Andrew Gelman frequently comes off badly in his comments on mainstream political reporting? Of course, much of said reporting is crap, which is surely worth pointing out and this site often does a great job in doing so, but in the heavily populated genre of Bai et al-smackdowns, this is literally the worst critique I’ve ever read. Readers of the NY Times are not looking for rundowns of the latest political science research. They are looking for news. The best a polisci prof can reasonably hope for is for the journalism to be informed by the relevant research. Silver delivers on that score more than 99 percent of his colleagues around the media, and in this case even links to one paper and references another, and yet Gelman still sees fit to ding him with the vague (to the point of meaninglessness) assertion that his readers would be misled by the tone of the piece. Give me a break.

    It’s also worth mentioning that had he tried to put the reference in the third graph, the editor would have surely pushed it way down or cut it entirely, and for all we know something like that happened.

    • Andrew Gelman March 26, 2012 at 11:08 pm #

      Pc:

      My blog entry above is not intended to be a smackdown. I think Nate is great; I’ve even published a couple of op-eds and a couple research articles with him!

      I just wanted to draw attention to an important paragraph which might otherwise have remained unnoticed. Nate is influential and I wanted to do my part to steer the take-home message away from “political scientists don’t know what they’re doing” to the more appropriate (to me) message that “some political scientists do know what they’re doing.” I put up my post right away in an attempt to affect the reception of Nate’s article, by pointing out where he says something unequivocally positive about political science research.

      • John March 27, 2012 at 12:54 pm #

        My understanding is that most political scientists do not create models to predict presidential elections in the United States. Attacking political scientists who engage in this particular pursuit is not at all the same thing as attacking all political scientists, or saying that all political scientists are incompetent.

  2. Eric L. March 26, 2012 at 10:59 pm #

    I agree with the Gelman critique here.

    That said, why “Dr. Fair” and “Ms. Vavreck”? I’m pretty sure Ms. Vavreck has a PhD, even if it isn’t in econ like Dr. Fair. Is that in the NY Times style manual?

    • Andrew Gelman March 26, 2012 at 11:13 pm #

      Eric:

      As a NYT writer myself, I can attest to that. The style guide is that physicists are called Professor, economists are called Doctor—whether or not they have a Ph.D.—while political scientists are called Mister or Mizz. But it’s even worse for sociologists: they’re just referred to by last name. And even worse for historians, who, according to the style guide, are to be referred to by first name, and worst for anthropologists, who are labeled by nickname. Thus a university committee might feature Prof. Greene, Dr. Fair, Ms. Vavreck, Baldassarri, Alan, and Meg.

      • Bernard Guerrero July 8, 2012 at 10:28 am #

        That’s phenomenal. There must be some path-dependent reason for this, no?

  3. Matt Jarvis March 27, 2012 at 12:11 pm #

    Sigh.

    Why take all the models? Why ignore that the models tell us something about causation, rather than a pure poll-based approach, which only confirms the Central Limit Theorem for us?

    I like what Nate does, really. But I think the tone of the piece is decidedly “I’m smarter than all these political scientists, what dolts!”

  4. History Grad March 27, 2012 at 3:46 pm #

    I agree with Matt. As someone in a more humanistic field, I’d love to say that I hate these models, but I don’t. Nate’s article makes some good points. But I don’t think the fact that they fail to predict outcomes perfectly negates their value. If a super-simple model like Hibbs’ can get within the same margin as a Gallup poll, doesn’t that just drive home the fact that economic fundamentals can get us very far towards understanding an election?

  5. Paul g March 28, 2012 at 1:28 am #

    I don’t know, I found Silver’s tone pretty dismissive. He describes his earlier critique as “theoretical” when there was not an ounce of theory there.

    Nate is a good statistician but he is at heart a data miner. His own “forecasts” on which he made his name were just weighted averages of polls. Golly, no surprise that you can predict an election accurately based on reams of pre-election polls. Even if the “fundamentals” models make errors (and btw I know of no political scientist who’d claim the “fundamentals” don’t include at least some measure of presidential approval) they still predict quite well based on six months old data.

    That says something about the fundamental forces in determining election outcomes. While it is true that what started out as an intellectual parlor game has turned into something more, but I never evaluated these models solely on the precision of their esimates. That’s the narrowly statistical perspective.

  6. Jim Bach March 28, 2012 at 8:41 am #

    I really don’t understand the point in using polling data (either presidential approval or vote intention) as a variable in predicting election outcomes, and I tend to dismiss models that use them. If one accepts the theory that economic conditions drive the aggregate vote, at least to some degree, if not exclusively, then the effort should be in finding the best economic indicator or indicators that allow one to most accurately predict the election result. If voters indeed are swayed by economic conditions, then polling responses are simply another indication of that and only add unnecessary complexity to a model.

  7. charles March 28, 2012 at 10:12 pm #

    I dont think the models do all that well. If you assume that 35% ( far less than McGovern) of the 2 party vote will go to each party; then you only have 30% of the vote left to explain. The models basically all missed Bush/Gore by 5% . 33% of the vote was miscalled! The polls were not off by very much. Most models don’t take into account the electoral cycle;

    A) 7 of the last 8 first term USA Presidential Parties have been re-elected since those 100 years old or less first voted ,
    B) The 100 year old Anglos (USA,CAN,AUS, NZ, UK) have an over 90% re-election rate for first termers.
    C) 100% re-election rate in the Full Democracies (Economist) of the first term parties on the last cycle after the last change. (To be broken by Greece?)

    3 (Heath,Clark,Carter) of the 5 Anglo defeats were immediately after Opec rises, one was an 80 year old Nash and Trudeau was a first termer but he and his party had been in power for more than 20 years (except for a few months) when the Party lost power.

    The models miss every election by more than the other methods, polling, gambling, expert predictions etc. In my view,we could predict just as well by doing a wisdom of crowds on line prediction.

    Right now, all prediction sources are predicting Obama- except the Hibbs type models. His is not modeled to 2012 reality -Obama is not getting blamed for the bad GDP nor the Afghan deaths and is unlikely to be on election day. Hibbs doesnt even believe his model and says to look at Intrade instead. Change on the first term is met by the voters saying- you just wait your turn, after all you just did?

    Americans are like everyone else-there is no magic-Obama won the 2012 election when the Republicans nominated a 72 +8=80 year old for president in 2008. No serious party except the Republicans puts up candidates like this as they have an atrocious record in the rich world. 65 year old Romney is past the limit too-time will tell.

    Models that ignore International elections totally are ill conceived. For first term re-election bids, I dispute that you need to know the name of the country to predict the election. You just need to know the type of country and whether there had been a NEW banking crisis or a NEW severe economic dislocation. America is an English speaking, rich, Anglo country and votes the same! The fact is that we know that the economy can help defeat a government- we have no idea how bad it has to be defeat a one term government. If the Republicans can win it would be an incredible victory and I might study Political science!