Ecological Regression Estimates: Alchemist’s Gold?

Extract

Those skeptical of ecological regression in voting behavior studies continue to suggest that problems in applying the technique severely limit its utility. But the cautionary offered in the Winter 1985 number of this journal by William H. Flanigan and Nancy H. Zingale (“Alchemist’s Gold: Inferring Individual Relationships from Aggregate Data,” Social Science History 9: 71-91) goes so far as to suggest that these problems are insurmountable—or virtually insurmountable. As a user, I was prepared to be devastated, but in fact find myself cheered (if a little puzzled).

Interested readers will recall that the centerpiece of the authors’ argument is a test involving this question: How did the voters of 1968 behave four years later in the presidential election of 1972? The test consists of comparing voters’ actual behavior, as determined by survey data, with ecological regression estimates of that same behavior. The tabulated results are alleged to be decisive in proving the authors’ point, but instead appear to prove just the opposite of what is intended, as a fresh look at the material reveals.