Wednesday, November 23, 2005

Fraud in Ohio? Doubtful.

Several lefty bloggers are claiming evidence of voter fraud in this month's voter referenda in Ohio. The basic claim is that several Democratic-sponsored measured went down to landslide defeats, despite the fact that a pre-election mail poll by the Columbus Dispatch newspaper (which has a strong record of accuracy in previous elections) predicted that they would actually win by landslides. Mystery Pollster does a great job showing why there are good reasons to doubt the accuracy of the Columbus Dispatch poll, but that still doesn't get to the heart of the fraud allegations. In particular, those suggesting fraud claim that counties with new electronic voting machines, particularly those manufactured by Diebold (a Republican-affiliated corporation) are responsible for the fraud.

To check these claims, I did some quick regression analyses of the referenda results to see if there are any statistical anomalies in counties with different types of voting machines. Ohio uses four different types of voting machines: the Diebold AccuVote, optical scan, electronic, and punch cards. In addition to these variables, I also controlled for the percentage vote for Kerry in the county, the percentage of whites in the county, and voter turnout in the county. Results were weighted for the total number of registered voters in the county. Here are the results for each of the 5 propositions on the ballot. Statistically significant variables (p<.05) are marked with a *:

Measure 1: Bond Package
Adj R20.70
Coef.Std. Err.
Kerry Percent0.360.05*
Percent Turnout0.090.10
Optical Scan0.010.02
White Percent0.040.06

Measure 2: Vote By Mail
Adj R20.83
Coef.Std. Err.
Kerry Percent0.520.05*
Percent Turnout0.440.10*
Optical Scan0.020.02
White Percent0.040.06

Measure 3: Contribution Limits
Adj R20.78
per3Coef.Std. Err.
Kerry Percent0.410.05*
Percent Turnout0.360.10*
Optical Scan0.060.02*
White Percent0.140.06*

Measure 4: Redistricting
Adj R20.77
per4Coef.Std. Err.
Kerry Percent0.470.06*
Percent Turnout0.430.11*
Optical Scan0.050.02*
White Percent0.100.07

Measure 5: Secretary of State
Adj R20.81
per5Coef.Std. Err.
Kerry Percent0.390.05*
Percent Turnout0.270.10*
Optical Scan0.040.02*
White Percent0.210.06*

For Measure 1, counties with the AccuVote system did have a lower percentage of yes votes (compared to punch machines, the excluded variable) and this difference is statistically significant, but this was a business-sponsored bond measure supported by the Republican Taft administration. If the AccuVotes were rigged, it's highly unlikely that they would have been rigged in a manner to hurt Measure 1.

With Measure 2, there was no statistically significant impact for any machine type.

For Measures 3, 4 and 5, all three machine variables are statistically significant, but the coefficients are positive, indicating that each of these machines actually had higher levels of yes votes than punch card systems. This result runs counter to the fraud allegations.

This is, of course, a quick and dirty analysis, but nonetheless, there doesn't seem to be any evidence to support allegations of vote fraud in the 2005 Ohio election.


Kevin said...

I'd recommend also trying controlling for county population size. Counties with higher populations are likely to have longer lines and more misvotes, and are also likely to have more Democratic voters.

Philip Klinkner said...


I ran the analysis again, but this time weighting for population instead of registered voters. The results were almost exactly the same.

Mike Anderson said...

Dr K:

Nice regression analysis! Is your data set somewhere handy (either yours or the original) that I can download? I'd love to give this problem to my undergrads to analyze.

Philip Klinkner said...

I'd be happy to email the data to you if you contact me.

Kevin said...

Ah, my mistake -- I must have accidentally glossed over the weighting by registered voters. Thanks for the information though.