Speculation or Specification? A Note on Flanigan and Zingale

Abstract

Flanigan and Zingale's "Alchemists' Gold" reviewed the classic problem of inferring individual-level relationships from aggregate data, attacked the specification error approach to it, and sought to replace Goodman's ecological regression with an informal and largely untestable procedure first proposed by Shively in 1974. They illustrated the Shively strategy by comparing evidence from the national CPS/SRC survey to state-level aggregate data from the 1968 and 1972 elections. In this note, I seek to show that their criticisms are misconceived and that ecological regression can, in many typical circumstances, lead to acceptably precise and much less vague judgments about the individual relationships than the method that Flanigan and Zingale promoted. I spell out, apparently for the first time in the cliometric literature, the mechanisms for using data on other variables in order to improve the estimates of the parameters of principal interest, and employ these techniques to recalculate the way 1968 major party voters behaved in the 1972 U.S. presidential election. The multivariate ecological regression figures come pleasingly close to reproducing the "true" percentages of loyalists and defectors drawn from the survey. Making use of all the available data and of well-developed and powerful statistical tests, ecological regression is far preferable to the Shively tactic.