Saturday, November 5, 2016

Annals of Derp: Handicapping the 2016 race

Hi MBRU Cowpokes, and thanks as ever, Blogenfreude!

Clara Bow in William Wellman's Wings (1927), from somebody's Pinterest.

Watching all day in wonder as the Twitter reports extraordinary volumes of early voting in Florida, Nevada, and North Carolina (contrary to Elaine's tweet there), with the implications of higher than expected participation by women, Latinos, and African Americans and concomitant higher hopes for Democrats in those states, and FiveThirtyEight continues not to reflect this sort of thing, by design, as Silver explains:
some of the factors that the model doesn’t consider: the disagreement in the polls, the unusual nature of Trump’s candidacy and the demographic changes it is producing, Clinton’s superior turnout operation, the possibility of “shy Trump” voters, the fact that the news cycle is still somewhat fluid headed into the final weekend, the declining response rates to polls, and the substantial number of high-profile polling misses around the world over the past few years. We think this is a good year for a forecast that calls for more caution and prudence.
I think everybody needs to stop getting mad at him over this, as if he were deliberately trying to frighten us. Every well-conducted forecast is a scientific test of its model, even the idiotic Los Angeles Times/USC tracking poll, and if you start messing around with your model to make it fit reality better then you wreck its scientific value, since everything you truly learn from an experiment is from where it goes wrong.

At the same time, we need to stop allowing it to frighten us. It is not the case that if we were to hold 1000 presidential elections between now and next Friday Clinton would win 647 of them and Trump 352 (with the other one going to Gary Johnson, I believe, even though Evan McMullin does better on Silver's Electoral College forecast). Or that out of every 1000 possible universes in which this election is actually being held on Tuesday Clinton actually wins just 647. What is coming on Tuesday is going to happen just once in just one world we have access to; it is an entirely unique event, which will in fact end in only one way; you can never test the accuracy of the prediction the way you can with a basketball game (where the Knicks will in fact play the Celtics x many times in a season and neither team is going to win all of them).

If you focus your attention on 538's electoral map, you can see Clinton doesn't have very many ways to lose, in that the states she is virtually assured to win in give her around 290 votes total, or 20 more than necessary.

You can imagine a scenario where she doesn't get all of them (John Podesta confesses that his brother's apparent mild interest in famous performance art practitioners is really a beard for his own passionate commitment to Satanist ritual, I'll give you 500 to 1 on that; or the SEPTA strike in Philadelphia really makes voting too difficult in the city and allows Pennsylvania to go to Trump, which I'm a little honestly worried about, but the Building and Construction Trades Council is proposing to get everybody to the polls in that event, and those people have serious skills which will not be aimed at giving Republicans an advantage, let's just say), but you can't take it any more seriously than the scenario in which Florida, Nevada, and North Carolina go to Clinton because there is empirical evidence that that is what's happening. In real human life (as opposed to coin tosses), our inability to predict what is going to happen on an individual scale obviously maps on to a formal inability to predict what will happen on a mass scale, but that's not even what makes elections hard to forecast. What does that is that we don't know how to rank the importance of all the factors, from the economic situation to the weather, that play a role.

What Silver is really saying is that knowing the factors whose relevance he is certain about makes him 64.7% certain that Clinton will win. He knows about the polling data at every given moment, which gives a kind of baseline range of values for how people think they plan to vote (that Silver can weight according to his knowledge of how successful those particular pollsters have been in the past), and he has objective markers for the economic "fundamentals" that can influence voters in ways of which they themselves are not completely aware; and he has the history of presidential polling to tell him how important these two factors have been in the past. That takes him 64.7% of the way to knowing that Clinton is going to win, if you will, as far as history can tell him, and the 35% balance is the importance of the other factors whose influence history doesn't adequately account for.

And as we know FiveThirtyEight's aggregation of many polls has a very good track record, though not exactly perfect. But the outcome in the end will be the one that would have been pretty close to 100% if we had had a clear enough picture of all the factors. So if you think you know something that Silver doesn't allow himself to know, you might well be right. Or you could be wrong like those idiots at the top of the post.

If you want odds, go to a bookmaker (2/7 for Clinton at the moment at Coral UK, and 5/2 for Trump; when there's a favorite on that scale, smart money leaves the race alone).

That "Bill Mitchell", the remarkably prolific, presumably Russia-financed Trump troll, isn't working as hard as we are:

Nor is this guy from the party's theocratic wing:

No comments:

Post a Comment