Causal inference paper on voting in LA County

Recently I wrote about papers being presented at the political methodology conference this week at UC Davis. The paper that was previously unavailable, by Henry E. Brady and Iris Hui, is now available in electronic from the conference website. Here’s the paper abstract:

Two seemingly unrelated approaches to quantitative analysis have recently become more popular in social science applications. The first approach is the explicit consideration of counterfactuals in causal inference and the development of various matching techniques to choose control cases comparable to treated cases in terms ofsome predefined characteristics. To be useful, these methods require the identification of important characteristics that are likely to ensure that a statistical condition called “conditional independence” is met. The second trend is the increased attention given to geography and the use of spatial statistics. Although these two approaches have found their ways into the social science research separately, we think that they can be fruitfully combined. Geography and Geographic Information Systems (GIS) can improve matching and causal inference. Geography can be conceptualized in terms of “distance” and “place” which can provide guidance about potentially important characteristics that can be used to improve matching. After developing a conceptual framework that shows how this can be done, we present two empirical examples which combine counterfactual thinking with geographical ideas. The first example looks at the cost of voting and demonstrates the utility of matching using zip codes and distance to polling place. The second example looks at the performance of the InkaVote voting system in Los Angeles by matching precincts in LA with geographically adjacent precincts in surrounding counties. This example demonstrates the strengths and weaknesses of geographic proximity as a matching variable. In pursuing these examples, we also show how recent progress in GIS techniques provides tools that can deepen researchers’
understanding of their idea.

This is really a paper that is meant for a methodologically-sophisticated audience, as they really are dealing with the development of methodologies for studying many different types of social and political behavior. But for those who do work through the paper, there are two substantive applications that would be of interest to readers of Election Updates. The first is one that Brady has written about before, and that is precinct consolidation in LA County. The second is a newer line of research, looking at the relative effects of voting technologies on voter behavior.

It is this second line of new work that might be of most interest to students of voting technologies. Here they examine precincts in Southern California counties that can be “matched” with other similar precincts, and then study the differences in residual vote rates across the “matched” precincts. One way to do the matching is by using the matching methods currently in vogue in statistics (and increasingly in political science, as I’ve written about before). Another way, which Brady and Hui write about extensively in their paper, is to use geographic methods, essentially concentrating on precincts that are right on the border of say LA and Ventura county, to do the “matching”. A third approach, which is what in the end Brady and Hui are trying to advocate in their paper, is to combine the typical statistical matching algorithm with geographic matching. Their substantive conclusion, based on a variety of the matching procedures, is that the Inkavote system used in LA County seems to underperform in terms of residual vote relative to the voting systems used in other Southern California counties. This result, of course, is one that Brady discussed at a conference last fall of the effect of voting technologies on residual vote rates across California counties, and thus that this result seems confirmed by these more sophisticated methodologies may not be terribly surprising.