First, most of Sanders’ performance in Michigan was really foreseeable in the qualitative sense. It is no surprise that Sanders’ campaign has specifically targeted two oft-overlooked populations from the beginning–he has been doing it EVERYWHERE: the young and the “missing white voters.” At least as far as Northern primaries are concerned, Sanders has been generally successful getting them out in disproportionate numbers: New Hampshire, Vermont, and Massachusetts. He has been only somewhat successful with them in Virginia, however, and not at all in the Deep South. Caucuses, by their low turnout nature, are an entirely different set of data altogether. From these observations, it is not that shocking that the young and the formerly missing white working class independents showed up in large numbers and that they should have broken overwhelmingly for Sanders: this is what they have done in every other Northern primary.
The rot sets in, however, when these observations are to be turned into actual quantitative predictions. The surplus turnout in New Hampshire, Vermont, and Massachusetts for these demographics, certainly, can be calculated, but only with the sample size of three. Any extrapolation therefrom would be, for all intents and purposes, would be 90% wishful thinking and 10% ass-pull. If I were a skeptic, i would not trust such projection at all for perfectly good reasons. As I keep repeating, science is not “right.” Science demands rigorous evidence, and often, there is just no rigorous enough data except old and outdated ones, and as such, science is biased in the direction of being wrong “conservatively,” i.e. in favor of the conventional wisdom. If the conventional wisdom is that the young don’t vote and the missing white voters are, well, missing, then it follows that the primary electorate in Michigan should be dominated by regular Democrats who did in fact vote roughly 60-40 in favor of Hillary Clinton, as was predicted. (Yes, one could conceivably think of creative ways to project how much support Sanders really had at the time of the poll by knowing that many voters who were not deemed “likely Democratic primary voters” supported Sanders. I did that myself. But I also know that that was an ass-pull that cannot be justified except post facto.)
Of course, Sanders’ electoral strategy is built on precisely this intuition, of drawing in the voters that don’t make sense given the conventional wisdom: this is thinking outside the box. For that exact reason, of course, it was a huge gamble that could not be counted on just based on “science” alone, even if it was informed by the same science. A clever and knowing combination of successful operation art and strategy of science, as von Moltke the elder described his maneuver at the Battle of Sedan (and von Moltke was himself paraphrasing Austrian Archduke Charles), with a good dose of good fortune thrown in–but it is not guaranteed to succeed.
This is a nifty illustration of what I’d been calling the law of variance: all theories are wrong, and statistics tells us how wrong they are. Even if “the theory” tells us that, on average, the young don’t vote and the missing white voters are missing or Republican, it does provide useful hints on how likely the young are to vote and the working class whites are neither missing nor Republican, whatever their “averages” might be. This can be used as the basis for a gambling strategy that could pay off with a respectable chance. This is how to use science properly, not just be used by “science.” Something to keep in mind.