Sunday, July 17, 2016

A PUZZLE

As everyone knows who reads this blog, I am obsessed with national politics, and in particular with the outcome of the present election, which I consider truly fateful for this country.  Lately, I have been puzzling over what seems to me to be a contradiction between established fact and a belief so widespread and plausible that it seems absurd to doubt.  Let me explain.

In recent years, several sophisticated statisticians have had extraordinary success in predicting election outcomes by amalgamating the results of polls that they themselves do not conduct.  Nate Silver at 538.com and Sam Wang at Princeton Election Consortium simply take all the poll results that the rest of us read as they appear – Ipsos, Rasmussen, Gallup, Quinnipiac, et al. – and generate extraordinarily accurate predictions.  Sam Wang got 50 out of 50 states right in 2012, and predicted most of the senate races as well.  Silver adjusts his calculations by taking into effect non-poll data [popularity of the sitting President, unemployment rate, etc.] whereas Wang adopts a “polls only” approach.  And Wang’s record of prediction is even better than Silver’s.

This is the established fact – that amalgamation of polls, all by itself, is a spectacularly accurate predictor of election outcomes.

Now for the widespread and plausible belief.  Everybody who comments on national politics agrees that ground game, organization, outreach, use of social media, paid advertising and the rest – the nuts and bolts of electoral politics – make a significant difference in election outcomes.  Barack Obama’s game-changing get out the vote effort in 2008 and again in 2012 is credited by everyone with the strength of his success.  I put in my time in 2008 and 2012 here in Chapel Hill walking door to door, registering new voters at supermarkets, entering data in the headquarters office, and I can attest to the ground level sophistication of the Obama campaign.

But these two – the established fact and the widespread belief – are in flat out contradiction with one another, or so they seem.  The polls simply ask the chosen sample how they intend to vote [and other things as well, of course].  As the election grows closer, the polling organizations move from polling all registered voters to polling “likely voters,” a rather shaky and shifting classification, to be sure.

If Sam Wang can predict 50 out of 50 state results on the basis of those polls, then what earthly difference does the ground game make?  So far as I can tell from a Google search, the identification of who is a likely voter has nothing to do with the impact of the ground game.


I await the greater wisdom of my readers.

10 comments:

  1. I think the conventional wisdom among political scientists is that each side's ground game cancels the other out. That's only true if both sides are actually organized. I think HRC learned this the hard way in 2008.

    ReplyDelete
  2. Let's say that we are actually interested in sprinters.

    We discover that by taking an average of the last ten times of each runner, we can predict with great accuracy the outcome of a particular race.

    Your question seems akin to asking why the runners should bother with their training and pre-run preparations, since they won't have any effect on the outcome.

    Polls in part measure the results of the ground-game, and ground-game also helps to actualize the potential identified by polls.

    ReplyDelete
  3. Marc Hamman, that occurred to me, but while it is pretty clear how a runner's training affects her performance, it is not [to me] clear how the ground game affects the results of the polls, since the ground game is focussed on getting out the vote, not merely on persuading voters [which would, obviously, affect the polls.] Does it make sense to put all one's money into voter persuasion and none into getting voters to the polls? Nobody who follows politics thinks so, but that seems to be the implication of the success of the Sam Wang approach to prediction.

    ReplyDelete
  4. There are two things that are measured by polls: preference and intensity of preference.

    Someone who is committed to going out to vote for a candidate has a stronger preference than someone who merely prefers that candidate if asked over the phone.

    Furthermore, as salesmen (and psychologists) know, getting someone to commit to an action related to a weak preference can actually persuade them to intensify their preference into a strong one.

    The very act of getting people to tell you that they will go out and vote for your candidate increases their intensity of support and makes it more likely they will give that support in the end.

    ReplyDelete
  5. Actually, the sophisticated conventional wisdom is that ground game can make the difference in a close election, but that it's not going to affect the result by more than a few percent. See, e.g., http://talkingpointsmemo.com/edblog/some-christmas-eve-morsels

    ReplyDelete
  6. Just guessing here: Democrats tend to have more support but supporters are less likely to go to the polls and be able to vote. When the democrats have a strong ground game, as in the past couple of presidential elections, it drives the voting numbers pretty close to the polling numbers. In the absence of a good ground game, polls would be less accurate, skewing more Democrat than the actual vote.

    ReplyDelete
  7. This is a bit like asking if human beings have free will. If pollsters can predict the outcome, do we really need to vote? I'm a compatibilist on both counts. Our minds behave mechanistically (determinism in a quantum and chaotic universe isn't quite the same as predictable), but our experience of decision-making is part of that mechanistic evolution. Likewise with voting: likely voters must be converted to actual voters. That's the ground game.

    Btw: I've always found Marxism paradoxical along similar lines: if it is historically inevitable why organize?

    ReplyDelete
  8. I don't think that there is a "flat out contradiction" between the polling data and the widespread belief. Suppose that there were a strong causal connection, of a sort that could not easily be upset, between what people say in response to polls and how they vote, Then, there would be a good explanation for the accuracy of the polls.

    But the widespread belief, if it is true, shows that our supposition is false. The election results seem dependent on voter turnout, and this seems independent of how people answer questions about how they intend to vote.

    This does not suffice to show a contradiction, though. What follows is that if the widespread belief is true, and the predictions based on the polls are nevertheless accurate, we lack a good explanation of why the results are accurate. Only if the polls could not be accurate unless the causal dependence view were true would there be a contradiction.

    The point is I think clearer if we suppose that someone was able to predict elections by reading tea leaves. This would be extremely odd, but the supposition is not contradictory, even if the patterns in the tea leaves play no causal role at all in producing the election results.

    ReplyDelete
  9. +Marc Hamann

    I came up with the same response as you did after reading Proff. Wolff's answer to your first comment. It would be a good guess to say that "if one is already willing to participate in political matters, it is more convenient to talk to them about changing their minds if such a change is necessary, than <> someone out of their house to vote when they did not originally intend to get involved". I would like to add that instead of spending money trying to get the latter to vote, it would be much more useful to spend this money trying to get them to know the value of participating and educating them in political matters, debates and so on in an election-free period. By doing so it will make that person deem it as a value that they should hold in general, rather than just show up in the ballot every four years.

    ReplyDelete