Yglesias takes aim at a familiar genre of argument for voting for a third party:

I’ve noticed that various anti-Obama pro-third-party arguments on the Internet proceed with an annoying two step. Usually the headline and the lede of the piece will be very focused on Obama, the evils of Obama, and the braindeadness of the Obamabots but then the argument will employ as a lemma something like it doesn’t matter who you vote for because your vote won’t make a difference anyway. I think that math is more contestable than people often realize but whatever you make of it, if your argument is that it doesn’t matter who you vote for then that’s an argument about voting not an argument about Obama. If it’s true that you shouldn’t feel constrained to choose a major party candidate on the grounds that your vote won’t swing the outcome anyway, then the exact same conclusion would hold even if Obama had cracked down on banks much harder or never bombed a soul or delivered single payer health care or whatever you like. The argument may be correct, but it’s an argument about an entirely different subject.

I also think that as an argument written for public consumption on a well-traffic blog or website (as opposed to simply offered over drinks at the bar) it’s an illegitimate form of argument.

“Why I’m Voting For Jill Stein” or “Why I’m Voting For Gary Johnson” is, qua article, an effort to persuade other people to do the same thing. A persuasive argument that takes as one of its premises its own failure to persuade is inherently problematic.

I don’t think this works. Consider the type of thought experiment you sometimes encounter in ethics: There are two groups of people in danger: Group A consists of 100 people who will die unless a rescue effort involving at least five rescuers is mounted, while Group B consists of 10 people in the same situation. Each potential rescuer can only join one effort. If you think you’re likely to be the marginal fifth rescuer, or are advising that person on what to do, then it seems pretty clear that (other things equal) you ought to join the mission to rescue Group A and advise others to do the same. (To make this more precisely analogous to voting, we should probably make the required mission sizes large enough that any individual is highly unlikely to make the marginal difference between success and failure, and also assume that we want people to engage in some degree of moral generalization lest everyone stay home. But let’s start simple.)

If everyone were certain to follow the rule of action you stipulated for this case, and there were no coordination or information about what others were doing, you’d similarly select the rule that says you should try to help Group A. However—and this is where the crudest form of a Kantian universalization test breaks down—it also makes sense to factor in what others are realistically doing, because ideally you’d like to see both objectives accomplished, and 110 people saved rather than just 100. If, for example, you see many dozens of people already mobilizing to save Group A, while only one or two are headed for Group B, what you ought morally to do is (again, I think pretty clearly) join the mission to save Group B and call for others to do likewise, provided that it is sufficiently improbable that you will persuade so many people that the initially oversubscribed mission to save Group A now fails.

Now, it’s true that this argument would have to “take as one of its premises its own failure to persuade.” If you knew that everyone would follow your advice—and there were no way to reliably coordinate the numbers in advance—then you would have to call for everyone to help Group A, sadly resulting in the death of the 10 who might otherwise have been saved. (Well, what you’d really do for a sufficiently large group is tell everyone to roll a die and join the B mission iff they roll a 6, and otherwise join the A mission, but again, let’s bracket that for the sake of simplicity.) This would be a highly unfortunate outcome!

Yet in the real world, we can often predict with great confidence that we will not persuade literally everyone to follow our course of action. It may, rather, be highly probable that we will persuade enough to make the mission to rescue Group B successful without any realistic risk of drawing off so many people that the effort to rescue Group A now fails. To be sure, in this case, the argument for publicly advocating that people join the B mission does take as one of its own premises the assumption that this advocacy will not be perfectly persuasive. But if that premise is true, and known at a very high level of certainty to be true, then this is not actually “problematic”: It is what allows all 110 people to be saved instead of only 100! Now, obviously, if you’ve calculated your probabilities wrong, you can imagine such a choice backfiring badly, but in the real world every aspect of the decision process (including the projected consequences of mounting each mission) is ultimately probabilistic, and you’ve got to make the choice you think will do the most good given the information available. So we can quibble about the numbers, but there are surely probability ranges in which it’s clear that what you morally ought to do is advocate that people join the B mission, or to heed that call given your knowledge about what others are likely to do.

Now add back in the consideration we left out for simplicity’s sake at the start. Suppose there are enough public spirited people around that even if you don’t join either mission, it’s almost certain that enough others will to make either mission succeed. In this case, you need not stir yourself at all to achieve the desired outcome—though this is only the case because most others don’t reason in this way. Most of us, I will assume, think that a decent person will often (though perhaps not in every instance) resist this apathetic logic as a kind of moral free riding that depends on others behaving differently. But it does not follow at all that, having accepted a duty not to engage in such free-riding, we must all collectively commit to ignoring considerations of marginal benefit entirely, so that Group B is always allowed to die. If you’re at all tempted to endorse this perverse scenario, it’s a good sign something has gone badly wrong with your moral reasoning.

The analogy to the voting situation should be obvious, but just to spell it out: Let’s assume there’s a pool of writers and voters whose first priority—call it Objective A—is that the major party candidate they regard as least bad wins. Because their less-bad candidate is still quite bad on a number of issues, however, they would also like to achieve Objective B: An unexpectedly strong electoral showing for a third-party candidate (perhaps their first preference in an ideal world, but with no realistic chance of victory) who is a vocal critic of the less-bad candidate on these very issues, and whose success at the polls may help focus attention on those issues, and minimize the impression of an overwhelming mandate for the victorious less-bad candidate. Here, again, we can quibble about the relevant numbers and probability ranges, but it seems pretty clear that there are highly plausible fact patterns under which, given well-founded beliefs about what others are likely to do, what one ought to do in order to achieve both objectives is advocate (or cast) a third-party vote.

Moreover, Yglesias glosses over the extent to which the argument for an Obama or Romney vote also depends on the very considerations he thinks are irrelevant when deployed to support a third-party vote. What, after all, is the argument for choosing between Romney and Obama if you actually believe Johnson or Stein would make the best president? Obviously it’s that your first choice can’t actually win, given what we know about how other people are likely to vote—which is just another way of saying that your individual vote (or, for a writer, the pool of people whose votes you’re likely to be able to influence) will not make the difference between victory and defeat for your most preferred candidate. The writer who’d actually prefer Stein or Johnson but makes the case for holding your nose and casting a pragmatic vote for the “lesser evil” major party candidate is also taking as a key premise the imperfect persuasiveness of his own argument—otherwise he’d be arguing for his first preference—but Yglesias clearly doesn’t find reliance on this premise “problematic” in such cases.

Against this background, perhaps we can say something about the link between the two types of arguments Yglesias identifies, and why they’re not really about “entirely different subjects.” Implicit in the case Yglesias makes is the notion that the “your vote won’t make the difference” argument is really an argument against voting at all, and that someone who’s prepared to cast a ballot for some candidate or another must already be prepared to reject that argument. So let’s consider some grounds on which people might do so.

One possible rationale is utilitarian: Even if the probability of casting a decisive vote is vanishingly tiny, it is not zero, and so has to be multiplied by the difference in social welfare that would accrue if it were decisive in order to yield an expected social value for the action. I won’t say much about this argument, mostly because I doubt that it’s what actually motivates anyone but Derek Parfit.

Another potential rationale is that people see voting as having expressive and symbolic value even when they’re sure it won’t be decisive. People with opposing political views, who know their votes will simply cancel each other out, could just make a friendly agreement to both stay home and save themselves the time and trouble—which would take care of the probabilistic utilitarian argument—but very few people actually do this, suggesting that most voters see a value to the act of voting beyond any possible effect on the outcome. Yet surely the symbolic and expressive value of voting depends in part on our not thinking of it as purely symbolic, but as a constituent of the process by which we together really do determine our shared political fate.

That brings us to what we might call the “folk Kantian” line of reasoning—also known as “what if everyone did that?”—which undergirds a lot of common-sense morality: Yes, it might not ultimately make a difference whether I individually vote (or litter, or keep a vegetarian diet), but if everyone reasoned that way, the system would fail to all our detriment. So we internalize a kind of moralized collective rationality: We each act in the way that it would be morally best for all rather than none of us who are relevantly similarly situated to act, even if it would be no worse morally (and perhaps even somewhat better individually) for some subset of us to defect.

The trick, of course—and one of the notorious difficulties in applying Kant’s test of universalizability—is figuring out who are the members of the “relevantly similarly situated” group following a rule of action at a given level of specificity. As the Rescue hypothetical (and common sense) shows, we often don’t want every member of humanity, or even a particular society, doing literally the same action, but rather to act according to a rule they’d want generalized across some relevant social subgroup. Yglesias, for example, doesn’t want people with the preference ordering {(1) Johnson, (2) Obama, (3) Romney} to vote in the way they’d will all Americans to vote, but rather to take into account the certainty that most Americans won’t vote that way, and instead act in the way they wish the subset who prefer Obama to Romney would vote given that the very large subgroup who prefer Romney to Obama will be doing the same.

This, I think, is the link Yglesias is missing between the “problems with Obama” thread of the argument and the “won’t be decisive” thread, which doesn’t just apply individually, but at least in certain states also collectively, to the subgroup who might be motivated to cast a protest vote in light of their strong reservations about Obama. It’s true that if someone accepts the “won’t be decisive’ thread as dispositive at the individual level, that’s an argument about voting as such. But the “problems with Obama” argument is actually doing double duty here: Yes, it’s an argument for why the reader should not want to signal unqualified support for Obama, should want dissenting voices to do well at the polls, and so on. But it’s also an attempt to construct an imagined community over which individuals generalize when they decide on a rule of action: The social subgroup of people for whom these concerns are especially salient, whose collective votes (unlike an individual vote) might well have significant signalling value without altering the outcome of the two-party contest. Yglesias can challenge the empirical, mathematical premises of this argument if he wants—and he does that too, in passing—but he can’t reject it as formally incoherent without implicitly indicting the case for a major-party vote in the same breath.

Addendum: Henry Farrell reminds me that Daniel “D-Squared” Davies, in a 2010 Crooked Timber post, made essentially the same point about the tension between folk-Kantian and instrumentalist arguments for casting a major-party ballot.