Response to the VTP criticism of the study by Dopp/Liddle

The Caltech/MIT Voting Technology Project has posted a critique of the analysis by Dopp and Liddle of Florida county voting data in the recent election.  The point of this work is that the disparity in party shift across counties is entirely explained by cultural differences across the state.  In particular, the "Dixiecrat" phenomenon is invoked as a full explanation of the effect of voting machine type that Dopp and Liddle had reported.

In support of this conclusion, the authors offer a two-variable regression, in which they show that the 2000 vote for Gore/Bush is sufficient to predict the county results in 2004.  Adding a variable for voting machine type (which is 1 in DRE counties and 0 in Op-scan), they claim that the regression coefficients for the voting machine variable are insignificantly different from zero.

In fact, much of this criticism is already mooted by Liddle's analysis, limiting the regression to 26 mid-size counties.  When I repeat the VTP analysis in just these 26 counties, I find that the 2000 party shift is indeed a good predictor of the 2004 party shift; however voting machine type also has a significant coefficient (p=0.01).  ANOVA for this 2-variable model shows that 3.3% of the variance is explained by voting machine type, even as 2000 results explain 91.4%.

-Josh Mitteldorf

The analysis is based on the 26 mid-size counties with registered voters in the range 80,000 to 500,000.  8 of these counties use touch-screen machines (machine type=0) while 18 use DREs (machine type=1).  "Vote Shift" is defined in terms of the difference between Republican registration and Republican voting.  In particular:  Registration percentage was defined as R/(R+D), (independent and third-party registrants were not part of the measure). Similarly, voting percentage was defined as Bush/(Bush+Kerry). "Vote Shift" was then defined as the difference between Republican voting percentage and Republican registration percentage.