Common Errors in Statistics (and How to Avoid Them), 4th Edition

THE ECOLOGICAL FALLACY

The Court wrote in NAACP v. City Of Niagara Falls, “Simple regression does not allow for the effects of racial differences in voter turnout; it assumes that turnout rates between racial groups are the same.”² Whenever distinct strata exist, one ought develop separate regression models for each stratum. Failure to do so constitutes the ecological fallacy.

In the 2004 election for Governor of the State of Washington, out of the over 2.8 million votes counted, just 261 votes separated the two leading candidates, Christine Gregoire and Dino Rossi, with Mr. Rossi in the lead. Two recounts later, Ms. Gregoire was found to be ahead by 129 votes. There were many problems with the balloting, including the discovery that some 647 felons voted despite having lost the right to vote. Borders et al. v. King County et al. represents an attempt to overturn the results, arguing that if the illegal votes were deducted from each precinct proportional to the relative number of votes cast for each candidate, Mr. Rossi would have won the election.

The Court finds that the method of proportionate deduction and the assumption relied upon by Professors Gill and Katz are a scientifically unaccepted use of the method of ecological inference. In particular, Professors Gill and Katz committed what is referred to as the ecological fallacy in making inferences about a particular individual’s voting behavior using only information about the average behavior of groups; in this case, voters ...