The very insightful Carl Beijer, who seems to understand the underpinnings of rational choice political science far better than many self-claimed rational choicers, has three brief but excellent posts that are worth a bit of discussion.
The first concerns seemingly an old debate, whether Clinton is any more electable than Sanders. To those accustomed to thinking in spatial model terms, it is fairly “obvious” that Clinton is more “moderate” than Sanders, as is the average Clinton voter, compared to the average Sanders voter. Thus, it is intuitive, by the “median voter” logic, that Clinton should have an edge. The fallacy of this argument is that the spatial underpinnings of alleged “ideology” is itself built on shaky foundations. Given the Hotelling logic, spatial “distance” is simply a stand-in for preference. The median is represented, in the spatial logic because she is not committed to one side or the other. As Beijer points out, in order for Clinton to be more electable versus Trump, that must mean that the Sanders voter would not vote for a Trump but a Clinton voter might. In practice, this is exactly the opposite: Sanders represents a diverse coalition that included a large minority of discontented working class whites. Even if they may not be the “average” Sanders voter, these are the voters who might under some conditions, be lost to Trump. The core Clinton voters, on the other hand, are committed Democrats who would not vote for any Republican, Trump or anyone else. Sans this “swingability,” they are not exactly “moderate.” What does wind up depicting them is the mistaken use of DW Nominate type scores: Clinton’s allies in Congress are moderate because they are voting with the Republicans on some issues: most notably, free trade and foreign affairs in recent years. Their willingness to defect to the Republicans on these issues does not imply that they are “moderate” on other issues, especially those concerning socio-cultural matters. That they are willing to side with the Republicans in support of interventions abroad and promoting free trade, rather than making them “moderate” vis-a-vis Trump, actually makes them far more stridently anti-Trump.
Of course, this reveals an important sense that Clinton and her backers are “moderate”: they can draw the support of the regular Republicans (or, at least they hope they do) which, realistically, Sanders would not be able to. But, taken together with the greater “swingability” of the Sanders voters, or, at least, a large minority thereof, this sets the stage for another point that Beijer raises: is there such a thing as a Democratic coalition? Clinton and her allies do not care to retain the support of the Sanders supporters. They are too busy courting the regular Republicans whose support they are (too) eager to capture. While “rigged” may not be the most accurate description, DNC emails do illustrate what every “institutions” person in political science should be (too) familiar with: that most institutions are rigged because their keepers put their hand firmly on the scale, tilting them to their advantage. The problem that Beijer raises, though, is the corollary to this problem that is rarely raised: yes, on average, the institutions are rigged, but, if so, and if every outcome is tilted in favor of the institutional insiders, why should the outside faction cooperate by playing through them? I would not go so far as to say that the institutions and the process need to be “fair,” as Beijer does, but that there needs to be high enough probability that the outsiders would win, even if, on average they might lose. This uncertainty, deliberately inserted into the game and maintained assiduously, is essential for keeping the institutions stable (if you talked to me about correlated equilibria in game theory, you would have come across a variation on this theme before). Take away this uncertainty and make the game both rigged AND low variance, you are asking for trouble.
This sets up Beijer’s third point: does it make sense for the voters to vote strategically, for the lesser of two evils? It does not, but perhaps not necessarily for the reason Beijer brings up. Clinton represents a low variance candidate whose mean is not very satisfictory. Trump represents a high variance candidate whose mean is farther than Clinton’s. Yet, both are sufficiently far from the voters away from the “middle” that the difference in their means may not mean much, and with sufficiently high variance, the conditional probably of getting a “better” outcome at a given point from a “worse” mean, but bigger variance distribution is greater than one from a “better” mean, but a lower variance. Now, a spatial modeller would say that this would be washed out by the higher probability of even worse outcomes, but not necessarily if the utility for far away outcomes are discounted–while I can’t work out the math at the moment, for the suitably large variance, as long as the differences are small enough, the voter is actually better off choosing a high variance candidate whose mean is far away. So, somewhat ironically, it is possible that voting for the seemingly greater evil (but with high variance) might be fully rational!