Demographic change, political backlash, and challenges in the study of geography

Hill et al.’s study (1) of the relationship between Hispanic population growth and voters’ support for Donald Trump contributes to our understanding of the larger puzzle of diversity and politics. The relationship between diversity and reactionary politics should be considered one of the most important sociopolitical issues facing the world today—it is a near certainty that almost every developed country and many developing countries will be more diverse a generation from now than they are today (2). And, thus, if increasing diversity affects political outcomes, the relationship can point in two consequentially different directions: toward increased diversity liberalizing politics or toward increased diversity causing a reactionary backlash.

The different possible directions of this effect may point to profoundly divergent paths for the harmony of future societies. If demographic change causes a reactionary backlash, this has grave implications for the long-term success of diverse places: If people are compelled to “hunker down” in the face of demographic change (2) and cannot achieve political and economic cooperation with diverse peers (3, 4), then societies face a choice of trading short-term economic and political success for the long-term social, economic, and ethical benefits of increasing diversity (5). Indeed, the relationship between ethnic diversity and a host of socially undesirable outcomes, including poor economic growth (6), unstable politics (7), and even violence (8), paints a grim picture for the future of a diversifying world.

The election of Trump in 2016 is, by some accounts, a demonstration of this phenomenon: Following on his antiimmigrant campaign rhetoric, Trump has pursued the xenophobic policies that have slowed the flow of immigration into the United States—thus a reactionary backlash to demographic change has damaged the potential human-capital benefits that come with future diversity. And the antiimmigrant sentiment that brought Trump victory is, by some accounts, part of a larger phenomenon: Scholars have drawn parallels between Trump’s election and the seeming connection between increased diversity in some European countries and the rise of right-wing populism (9).

Hill et al. (1) ask the question of whether this backlash manifests locally, that is, whether it is the places where the Hispanic population has recently increased that we also saw a backlash in the form of increased votes for Trump. In speaking to this, Hill et al. (1) are engaging with a long-standing question in the social sciences, dating back, at least, to the mid-20th century when the concern was the reaction of white Americans to local African-American populations (10, 11), and scholars have noted that the phenomenon of localized backlash can be seen over longer stretches of time and space, including toward religious minorities in medieval Europe (12).

The potential for localized reactions to demographic change has policy implications: If antiimmigrant backlashes are localized to places experiencing immigration, then policy should focus on mitigating these local negative reactions to diversity, perhaps keeping the backlash in check while waiting for the forces of interpersonal contact to lead to long-term harmony (13). But if a backlash does not manifest locally, then the problem is more diffuse, perhaps because of economic or social conditions that spur antiimmigrant attitudes, and the nature of the remedy must be less localized. And, importantly, if backlashes do manifest locally, it means that places that now seem to be free of antiimmigrant reactionary politics may experience such politics in the face of increased diversity—in other words, everywhere has the potential for bigotry.

If this question is among the most important ones facing social scientists and policy makers today, it is also one of the most intractable areas of social science to study rigorously and definitively. The mixed, even contradictory, findings about the relationship between localized increases in Hispanic population and support for Trump demonstrate this difficulty. One may think that this relationship should be easy to study in 2016: The election happened recently, so the sociopolitical contours are relatively unambiguous (nearly everyone agrees that Trump was the antiimmigration candidate) and the data are recent and, presumably, readily available. But, as Hill et al. (1) note, different scholars studying this same question have arrived at different conclusions: some finding that local demographic change led to a local backlash that helped to propel Trump into the White House and others finding no relationship or, as with ref. 1, a negative relationship—indicating that if a local backlash did cause an increase in votes for Trump, this was outweighed by other forces that caused votes for his opponent. Or, perhaps, the backlash simply was not strong enough to reverse a relationship between the local Hispanic population and Democratic voting that may be caused by another lurking variable, such as local economic conditions.

Why are there contradictory findings despite the seemingly straightforward question? The relationship between local diversity and politics is difficult to study for many reasons, which are too numerous to cover in detail. These include a range of “problems of aggregation,” most notably the often-discussed modifiable areal unit problem, in which a continuous process must be discretized into geographic units; and, of course, as Hill et al. (1) recognize, the causal effect of local diversity on politics is very difficult to establish because diversity is endogenous to other sociopolitical processes. Moreover, the data for studying these questions are not as readily available as one may naively assume: Election data in the United States are not centrally administered and demographic data are infrequently collected and are often aggregated into large geographic units (often these aggregate units are different from the units used for election data, creating a need for interpolation across geographies).

Another challenge facing the study of diversity and politics is that, unlike many areas of scientific inquiry, there is no agreed-upon unit of analysis. Hill et al. (1) make an important contribution by using geographic units with a high degree of resolution (the voting precinct), which means that, compared to analysis using larger geographic units, we have a better idea of the relationship between local demographic change and the type of social processes that may cause this change to be related to politics. For example, if a geographic unit is as small as a precinct (perhaps just a few hundred people), it is reasonable to think that the social diversity of this geographic unit may have a real impact on the lives of individuals in that area and this impact may affect voting. This is to say that if a non-Hispanic person lives in a precinct that has, say, grown from 10% to 50% Hispanic over the course of a decade, it would be reasonable to assume that that person has more exposure to Hispanics in everyday life than that person did a decade ago. Prior research tells us that there is reason to believe such changes may shape voting (14) and, thus, in aggregate induce a relationship between growth in percent Hispanic and voting outcomes on the precinct level. On the other hand, when research is conducted using a less-detailed geographic resolution, for example the county, the relationship between local increase in diversity and the experiences of individual voters is less clear. This is because counties can be large enough that a non-Hispanic person can live in a county experiencing an increase in Hispanic population and still have no appreciable change in personal experience with Hispanics.

But using these high-resolution geographic units also presents a possible tradeoff of gaining insight into individual experience, while losing information about other geographic features of interest. For example, examining the relationship between Hispanic population growth in a precinct and votes for Trump is

Hill et al.’s study of the relationship between Hispanic population growth and voters’ support for Donald Trump contributes to our understanding of the larger puzzle of diversity and politics.

aspatial in that it examines the relationship without considering the larger picture of where this new Hispanic population lives compared to non-Hispanics. Observational research and experiments have demonstrated that proximity to a local immigrant population in the context of segregation between natives and immigrants may cause a backlash against immigrants (15) because segregation inhibits interpersonal contact and may change perceptions of group difference (16). If this is the case, then looking within counties, it is the people who are integrated with an immigrant population, e.g., people with a high percentage of immigrants in their precinct, that would have the most positive reaction to the immigrants, while simultaneously causing those in the same county but segregated in a different precinct, those who are “close but far,” to have a negative reaction. This phenomenon would induce the negative relationship between percentage of Hispanics and votes for Trump at the precinct level when the relationship is examined with county fixed effects, as Hill et al. (1) find in their statistical analysis, while the relationship between percentage of Hispanics and votes for Trump at the county level would be positive. Thus, the relationship between Hispanic population growth and Trump support will change direction, depending on the unit of analysis.

The opposite relationships between Hispanic population growth and votes for Trump when looking within counties and precincts are consistent with, more generally, the fact that at higher levels of aggregation, e.g., across countries, diversity is correlated with outcomes that may be the result of a reactionary backlash against diversity (e.g., ref. 6) while at lower levels of aggregation, e.g., within countries, the relationship can be reversed. This is not to say that the Hill et al. (1) analysis is wrong or misplaced—it is a valuable piece of a larger puzzle. But it does highlight a difficulty in studying this question: Different choices of geographic unit can lead to different inferences and there is no clear way to sort out which geographic unit is more important—it depends on the scientific, political, and policy questions that one is pursuing. For example, small geographies, such as precincts, are useful for answering the question of how some voters’ day-to-day experience with diversity can affect voting, but the long-term policy relevance of this analysis depends on long-term projections about segregation: If ethnic minority populations are segregated and those voters without day-to-day contact continue to react negatively against the minority group, then even a local positive correlation between diversity and liberal politics may not lead to long-term harmony for a society.