Why 2016 is not like 1964 and 1972

Nadia Hassan writes:

I saw your article in Slate. For what it’s worth, this new article, “Ideologically Extreme Candidates in U.S. Presidential Elections, 1948–2012,” by Marty Cohen, Mary McGrath, Peter Aronow, and John Zaller, looks at ideology-based extremism and finds weak effects of ideology. Like the high end is 1980 and the authors estimate Carter got ~1.4 points of vote share from being appreciably closer to the median voter.

This seems consistent with what Rosenstone wrote in his classic 1983 book, Forecasting Presidential Elections, and what I wrote in my Slate article. It’s good to see the argument updated and presented in more detail.

Hassan continues:

Two other factors may have impacted in 1964 and 1972: approval and incumbent party tenure. LBJ had a sky high approval rating, perhaps bolstered somewhat by JFK’s death. The other is time in office. Parties seem to do better their first term in office than 2nd or later. LBJ and Nixon got a couple of points from that. Eisenhower had economic performance slightly weaker than 2004 in 1956 but he still won by double digits—his approval rating was over 60%, compared to Bush in the high 40s.

Yes, good points. Also since I’m linking here, let me just say that I don’t like the headline Slate gave to my article, “Trump-Clinton Won’t Be a Landslide.” The subheading, “Conventional wisdom is that fringe candidates get repudiated, à la 1964 and 1972. The story isn’t so simple,” is fine. But I don’t like going on the record with a deterministic prediction—especially a prediction that I never made!

My article’s fine, though, it’s just the headline that bothered me.

10 thoughts on “Why 2016 is not like 1964 and 1972

    • By the way, apropos of Andrew and David Rothschild’s article about failures of prediction markets: the PredictIt market for a Democratic landslide (defined as 370 electoral votes or more) has not moved all that much since at least early June. Other than a very brief dip at the end of July, the implied probability has never been outside the range 32% +/- 7%. That probability seemed very high to me in early June (33%), but that was before the FBI announced they weren’t recommending Hillary be indicted, and before Trump insulted and attacked a “gold star” father, and before a slew of prominent Republicans announced they won’t vote for Trump, etc. Surely those things were not inevitable, and should have moved the market more than a few percentage points. But the price today puts the probability around 30% somewhere. Of course 30% could be too high right now…but in that case 30% was way way way too high back at the end of May.

  1. I saw your article in Slate, also. There was a claim of superadditivity wrt economic growth (1964, 1972). Shouldn’t one be fitting one of those polynomials for economic growth as a predictor of popular vote spread? Oh, wait…

  2. Isn’t it pretty obvious that JFK sympathy vote must have had an appreciable impact back in 1964? He was a mythical young president with the highest approval ratings in modern times. His murder occurred less than a year before the election, too. That should be a major confound.

  3. By the way, how seriously should we take the years of research from political scientists in predicting events based on a sample of what is it, 10-15 observations? Does the economy really matter?
    I know some of the fits are quite impressive, but after years of research one wouldn’t expect anything else. They were guaranteed to find something. Plus, out-of-sample performance does not seem very impressive.

    The years of research is actually a bad sign. In small samples, either a simple heuristic works right away or you are done.

  4. Andrew,

    Years of research suggests lots of models being tried out, ergo lots of data mining. With so few observations, you go with as simple a model with as simple structure as possible and then just test it to see if it fits or not. You do not estimate anything nor do you try to improve the model. Data has nothing to teach you about the underlying structure: it can only tell you if there is a chance you were right in the first place about the “fundamentals”.

    • Krzys:

      Lots of intuition and survey data suggest that the economy matters to voters in U.S. presidential elections. Hibbs’s basic model using recent personal income growth is about as simple as you can get. It fits pretty well but didn’t do so well for 1952 and 1968, hence his addendum correcting for wars. This all seems reasonable to me.

      • Andrew,

        Yes, I agree. The model is very reasonable (and low dimensional) and even war additions are probably fine. You just have to be very careful not to try and learn from (very) small samples. So, after one or two passes, all those years of research are pretty much useless. They might have discovered something, but we will never know, since we cannot tell this process apart from data mining.

        The larger point is the confidence some economists/political scientists have in their “deep” understanding of the “fundamentals”. As opposed to idiot journos chasing random squirrels. However, this confidence is unearned. We simply don’t know and can’t tell.

Leave a Reply

Your email address will not be published. Required fields are marked *