(photo by Gage Skidmore)
This article on a new study about the unusual demographics of Donald Trump’s supporters struck me as conceptually parallel to gaining access to a previously unobservable variable in a choice model. Many attempts to discern voter preferences are foiled by the lack of dimensions along which voters can express differential preferences, given the strong incentives for voters to cast a dichotomous vote for a Republican or a Democrat. In turn this limits the preferences a researcher may observe.
The determinants of this particular binary choice are so hard to study because of the large amount of preference bundling implicit in any given set of election results. Bayer and McMillan (2005) look at neighborhood sorting by race through this lens and use a structural approach to consider the bundling of inframarginal undesirable neighborhood characteristics (lower quality schools, higher crime, etc.) implicit in the decision of some blacks to live in black neighborhoods. Their key observation is that in many areas, there does not exist a large enough black population to offer meaningful choices across neighborhood characteristics conditional on having majority black residents. This leads to poor estimation of the magnitude of preferences less decisive than the marginal characteristic of race composition (leaving aside issues of actual de facto restrictions on inter-racial neighborhood sorting, which have been historically binding in many areas). In other words, when there aren’t enough different choices bundled together with your primary preference, all one can infer is information about your primary preference. This is often the case in our voting system. Interesting recent work by Kuziemko and Washington (2015) on this sort of problem in voting uses regularly collected Gallup survey data on racial attitudes as a new variable to shed light on the election results that led to the southern transition from the Democratic to the Republican party.
The Donald seems as if he may prove to be another dimension altogether in the election space this cycle. The Times article reports on an unusually large and more detailed survey of the (statistically weighted) opinions of Republican-leaning voters on Trump. Among the odd takeaways are that Trump’s “best voters are self-identified Republicans who nonetheless are registered as Democrats.” It goes on
…Mr. Trump has broad support, spanning all major demographic groups. He leads among Republican women and among people in well-educated and affluent areas. He even holds a nominal lead among Republican respondents that Civis estimated are Hispanic, based on their names and where they live.
Mr. Trump’s best state is West Virginia, followed by New York. Eight of Mr. Trump’s 10 best congressional districts are in New York, including several on Long Island. North Carolina, Alabama, Mississippi, Tennessee, Louisiana and South Carolina follow.
This candidate’s offered set of characteristics certainly seems to be fostering some very counter-intuitive groupings of voters. He may turn out to be just the sort of omitted variable that can cast a bit more light on the strange agglomerations of voters that end up deciding many US presidential elections. It appears that Trump’s breakout is based on his giving voice to some typically “left-unsaid” opinions about race and “otherness” but this is exactly what may make his candidacy so useful as a proxy for typically unobservable voter preferences.