Ecological Inference in the Law, Part III 

 In two previous posts  here  and  here , I discussed the ecological inference problem as it relates to the legal question of racially polarized voting in litigation under Section 2 of the Voting Rights Act.  In the latter of these two posts, I suggested that this field needed greater research into the case of R x C, as opposed to 2 x 2, tables. 

 Here's another suggestion from the courtroom:  we need an individual level story. 


 The fundamental problem of ecological inference is that we do not observe data at the individual level; instead, we observe row and column totals for a set of aggregate units (precincts, in the voting context).  This fact has led to some debate about whether a model or a story or an explanation about individual level behavior is necessary to make ecological inferences reliable, or at least as reliable as they can be.  On the one hand, Achen & Shively, in their book Cross-Level Inference, have argued that an individual level story is always necessary to assure the coherence of the aggregate model and to assess its implications.  On the other hand, Gary King, in his book  A Solution to the Ecological Inference Problem , has argued that because we never observe the process by which ecological data are aggregated from individual to group counts, we need not consider individual level processes, so long as the row counts (or percentages) are uncorrelated with model parameters.  

 From a social science point of view, this question is debatable.  From a legal point of view, we need an individual level story, regardless of whether such a story produces better statistical results.  When judges and litigators encounter statistical methods in a litigation setting, they need to understand (or, at least, to feel that they understand) something about those methods.  They know they will not comprehend everything, or perhaps even most things, and they have no interest in the gritty details.  But they will not credit an expert witness who says, in effect, "I ran some numbers.  Trust me."  What can quantitative expert witnesses offer in an ecological inference setting?  The easiest and best thing is some kind of individual level story that leads to the ecological model being used.