November 18, 2004

Views of the Berkeley Study of Florida's Electronic Votes

Wisconsin political scientist Ken Mayer writes:

    Count me as skeptical; there are many assumptions built into this analysis, and it excludes a number of variables that would clearly affect vote percentages -- Nader's lack of presence in 2004 vs. 2000, the problems with the Palm Beach ballot (which mean that a key vote percentage from 2000 is probably off), the number of registered voters, African Americans, etc. The specification, which includes both squared and interaction effects, doesn't strike me as particularly realistic.

    Moreover, a more sophisticated analysis would look not at the county level, but at precincts, which give you many more independent variables and much more variation. Finally (although this introduces some ecological inference problems), the key is not the percentage of the vote, but the number of votes. And I'd be far more convinced if a similar analysis of the Senate vote failed to show the same sorts of movement (if it did, that would clearly indicate that the assumptions of the presidential analysis were off, or that the conspirators were so stupid that they'd risk cheating on two elections at the same time).

    Finally, it always sets my spider sense atingling when a quantitative social science paper uses more significant digits in the results than you have in the underlying variables. You can't get 7 or 8 meaningful significant digits (table 3) when the original data only have 1 or 2 significant digits. That implies a level of precision that you just don't have.

More thoughts from Dan Tokaji and The Mystery Pollster.

Posted by Rick Hasen at November 18, 2004 09:04 PM