11/08/2020

Causal attribution & election results

Many people on my timeline are sharing this excellent interview with Rep. Alexandria Ocasio-Cortez:
https://www.nytimes.com/2020/11/07/us/politics/aoc-biden-progressives.html

Here's a representative quote from her:

If the party believes after 94 percent of Detroit went to Biden, after Black organizers just doubled and tripled turnout down in Georgia, after so many people organized Philadelphia, the signal from the Democratic Party is the John Kasichs won us this election? I mean, I can’t even describe how dangerous that is.

After reading that, the philosopher part of my brain started wondering about how we should think through causal questions in the neighborhood. Did 94% of Detroit going to Biden win him the election, or did winning over the ‘John Kasichs’ (right-leaning centrists) of the swing-state electorate do it? (Let’s suppose, for the sake of argument, that there were a non-trivial number of John Kasichs in swing states who either voted for Biden or chose not to vote for Trump in the 2020 election; I am completely open to that latter being factually false.) The scenario I describe below is highly idealized, and may be massively disanalgous to what actually happened in the 2020 electorate, but the idealized scenario brings out something that MIGHT be going on in this causal debate about the 2020 results.

Suppose 9 people are voting on a proposal. Further imagine that the proposal passes, by a 5-4 vote. In a strict sense, all five of those ‘yea’ votes were necessary to bring about the effect of the proposal passing. (And we can even imagine that each of the 5 votes ‘yea’ for a different reason.) But we often speak of one (or more) of those 5 as THE cause, or at least the decisive cause, of the proposal passing. Relatedly, maybe one of those 5 voters is seen as especially responsible for the proposal’s passage (I recognize that responsibility and causation are not identical; but they are related). Often, this one is called the ‘swing vote.’

But all 5 of those votes was necessary to bring about the effect—so how can we pick out one as privileged over the 4 others? If we hold all votes but one of the 5 'yea's constant, and 'wiggle'/ intervene on that one 'yea', then the effect flips from passage to failure -- and that is equally true for all 5 of the 'yea' votes. Now one reasonable reaction to this is to simply reject the idea that one of the 5 yea-votes is in any way specical or privileged. People may think one of them is special, but they are wrong. Although this is a reasonable response, I am curious whether there might be anything salvageable or reasonable in ever causally privileging one 'yea' over the other 'yea's.

I am very much not an expert on the causal attribution literature, so I strongly suspect that someone has already said this. I couldn't find anyone saying it after a little googling, but if any readers know of someone who has already published this point, please let me know in the comments. Anyway, here’s my (probably not-new) hypothesis.

Out of a set of partial causes, each of which was necessary to bring about an actual effect, we privilege the cause that fails to hold in the closest possible world to our own.
This is why the swing voter is considered especially responsible for the proposal’s passage: out of all 5 ‘yea’ votes, the actual world would have to undergo the smallest change to flip a swing voter from yea to nay.

And this matches other causal attributions we make as well. We say that the match lit because it was struck, not because there is oxygen in the air, even though both those conditions are necessary for sustained burning to occur. On the hypothesis above, this is because the world in which I don’t strike the match is closer to our actual world than the world in which I am in a very low-oxygen environment.

So on this view, questions about whether Biden’s victory is caused by the John Kasichs of the electorate, or increasing turnout in Georgia, come down to the following question: Which is closer to the actual world, (a) the Kasichs of the electorate voting for Trump at roughly the same rate as in 2016, or (b) Black turnout in Georgia remaining at roughly 2016 levels? I genuinely have no idea.

As I think about it, the causal question seems actually not to matter for the political question of what the party should do, to win in the future – unless distance between possible worlds can be measured by money and other resources. The question, in terms of promoting future success, is not ‘What caused the Biden victory?’ (and then try to replicate that cause, next time around) but rather ‘What is the most cost-effective intervention to create more favorable vote margins?’. These are related, in that a possible world where I bought 1 more blueberry muffin than I actually did this morning is closer than the possible world where I bought 2 more blueberry muffins than I actually did. But it would be surprising if a cross-world metaphysical metric could be given by just tallying up dollars and cents. (Suppose in the actual world I bought 5 blueberry muffins today. Further suppose a muffin costs the same as an apple. Which is closer to the actual world: (i) the world in which I buy 1 apple in addition to the 5 muffins, or (ii) the world in which I buy 2 more blueberry muffins, in addition to the original 5?) That would make modality and causation very anthropocentric, it seems.

Another extremely important aspect of all this not addressed above is that there are also moral reasons to prefer one plan of action over another. When people’s ability to vote is being substantially suppressed or hindered, there is also a serious moral obligation to remove those obstacles, even if the dollar-per-vote-gained wouldn’t be as high as another TV ad targeting centrist voters who do not face significant obstacles to the ballot box. You don’t have to be an orthodox Rawlsian to think considerations of justice should outweigh considerations of efficiency, at least in most cases. This may be part of why Ocasio-Cortez says it would be so "dangerous" to focus future campaigns on flipping the John Kasichs of the electorate, instead of ensuring everyone is enfranchised in a substantive and meaningful way.

3 comments:

Kenny said...

One thing that the causal attribution literature sometimes talks about is whether some of these variables have a "default" value. If 4 of these 9 voters were by "default" going to vote yes, but the 5th wasn't, then we naturally say the 5th is the cause.

If we think that variables tend to have their "default" values at close worlds, then this is very similar to what you're saying. I think there's controversy about this (as about everything) in the literature on causation, but it basically leads to the same place you suggest (including the difficulty of attributing "defaults" in cases with multiple changes).

Noel said...

I think there is major factor in this question that you are missing: voter turnout. This adds another dimension to the arithmetic.

In Australia, voting is compulsory. You can spoil your ballot if you like and not vote for any party, but if you fail to exercise your democratic rights you will be fined. This makes puzzles like the above much simpler, as you can factor in the 10th or 11th person who did not vote for either party.

This year's presidential election throws the issue into stark relief, with an unprecedented voter turnout that muddies the mathematics. The "non-voters in every other election" were far more influential than swing voters this time and are not accounted for.

Greg Frost-Arnold said...

@Kenny--
Thanks for that comment, that's very helpful. I am ignorant of how the concept of 'defaults' works in the causal attribution literature. (I've tried to learn a little bit about defaults in default logic, but even there I'm still pretty ignorant.) But I am now VERY curious about whether a clear/ clean relationship can be stated about the relation between defaults and nearby possible worlds, as you suggested.

@Noel--
Non-voters (and in the US case, 3rd-party voters) are a very interesting piece of this, which I hadn't considered here. And it might actually create a problem for the hypothesis I presented here. Why? It would disprove my hypothesis if the following two things are true.
(1) People generally blame (/ attribute more causal power to) 3rd-party voters and/or swing voters than to non-voters. (This is my impression, ESPECIALLY with 3rd-party voters: Ralph Nader and Jill Stein voters were blamed for Gore losing in 2000 and Clinton losing in 2016, respectively; I'm hearing that the Libertarian candidate is being blamed somewhat for Trump's loss in 2020.)
(2) The possible world in which the people who didn't vote (in our actual world) show up to the polls in substantially larger numbers is closer to the actual world than the possible world in which the swing/ John Kasich voters who voted for Biden in the actual world voted for Trump instead.

(2) strikes me as plausible, but at least arguably not true. The engaged John-Kasich-type person is more likely to vote for the Democrat candidate, than a politically apathetic person is to go to the polling station. That's my current opinion, but I could definitely be talked out of it.