But before very long, the self-examination ceased. Democrats were reassured by their friends in political science that they really had no problem with the working class and that they needn’t be concerned. With a few statistical sleights of handand enormous heaps of professional contempt for the laity, academics helped to shut down that debate.
I am ambivalent about this kind of attacks, both from political science on people like Frank, but also the kind engaged by Frank on political science. This kind of fight strikes me as a fight between blind men concerning whether an elephant is like a spear or a wall. They are both right and wrong at the same time. They are both seeing something that the other does not, but are both too full of hubris to see that neither has the complete answer, and that they can fill out a fuller picture if they start paying closer attention to what each other is really saying, that is, not what their conclusions are, but how they arrived at their conclusions, what assumptions they make, and what moving parts are involved in their thinking. As I was fond of saying about all the different theories about parties mattering or not in Congress, I don’t care if they say parties matter or don’t. How do they arrive at their conclusions, what do they think parties are supposed to do, and what evidence do they present? And how can all their answers be right at the same time? (except their conclusions, because they are all wrong–and that is a good thing.)
This, of course, is often the opposite of what people do in social sciences. Theories come first, not necessarily as a framework for thinking about problems (my approach), but as the stand-in for answers. So the important part is not why does this theory say parties are important and what evidence might there be for their importance, but the point is that parties are important…or not. Who cares what they do, as long as they are important, whatever that means? In context of elections, the indirect exchange between Frank and Gelman/Bartels on the “working class conservatism” is revealing: Frank is arguing that many working class voters are voting for conservatives, while Gelman and Bartels are saying that working class voters, on average, vote for liberals more than the conservatives.
Do they actually disagree with each other? NO! Many is not the same as most or even majority. Gelman and many others have shown without dispute that, on average, the wealth is an excellent predictor of the choice between Democrats and Republicans. However, they themselves point to a significant exception: the working class becomes more conservative in the more liberal states, or, as they call it, “what’s the matter with Connecticut?” The results of the 2016 election provide support for both sides: Trump made huge gains for Republicans among the working class in the North, not only in the Midwest, but also New Jersey and, yes, Connecticut. But the formerly solidly Republican wealthy voters in (formerly) solidly red states like Texas, Georgia, and yes, Kansas turned toward Clinton in largely numbers. Among the biggest gains for the Democrats came in Dallas, apparently. It is also worth noting that, while, for vast majority of the population, the tendency towards greater support for the Republicans with higher income holds, this sharply reverses itself with the highest income levels–it just so happens that these folks are not numerous enough to add much to the statistical findings, as shown below (from 2012 CCES data )
The lack of reliance on statistical data by folks like Frank does not mean that what they see does not exist. They are sharp observers–they are not making stuff up that don’t exist. Rather, they point to subtle patterns that slip through in statistical analysis using “standard” data that are lost in analyses by those who may know statistics and data, but not the context. If, say, a Jered Weaver is a successful major league pitcher who gets (got) excellent results (when his fastball was faster than 80mph, at leasat) despite very high FIPS, that’s a sign that there is something about what makes for a successful pitcher that FIPS is not capturing, not that Jered Weaver is a terrible pitcher because FIPS says so, even when the real data, the actual results, as opposed to a statistic summarizing it, i.e. the FIPS, says otherwise. Seen above, there are indeed such things as (very) rich liberals (I think they make up at least a near majority for those whose family income exceeds 250,000 dollars. Just that their presence is obscured by the pattern for vast of the population who make far less. The real problem for folks like Frank is that their arguments lack the sense of proportions: the near majority of liberals among the very rich does not exactly mean that the very rich are “liberals.”
For heresthetical politics, majorities who fit the “averages” are not especially important. When the full data becomes available and is analyzed in detail, I don’t doubt that very little changed in 2016 compared to 2012. The working class voters in the Midwest whose defection sealed the election for Trump made up very small percentage of the population. The Democrats voted for the Democrat and the Republicans voted for the Republican, while the low propensity voters stayed inactive, with relatively few deviations. But the forest is misleading: the 95% of the voters who stayed the same did not turn the election, but the few per cents who responded to the new information did. Again, hardly a new story: in US House elections, vast majority of voters always voted along the party lines: only about 10-15% of the voters, historically, did otherwise. But vast majority of those who did did so in favor of the incumbent, and 10% of the votes is a lot when they all go to the same side, even if they are but a tiny fraction of the whole.
The value added by careful observers like Frank is that they point to the subtle but important patterns likely to be overlooked when the focus is wasted on the big and obvious patterns. As per Admiral Rickover, the Devil may be in the details, but so is the salvation, and Admiral Rickover was a great man of science to have emphasized this. Elections turn on a few percents of voters who behave differently, respond to different campaigns, and believe in different things. The value added of Big Data is not that we can show big patterns in neat multicolored graphs, but that we can spot significant but understated details that would be lost in the noise if not for the sample size. Big Data can help place Frank in context, not bury him. Yet, too many people seem eager to abuse the Big Data to sweep away inconvenient details, not examine them further.