Update: Diebold Effect explained.

Here’s a unique approach to understanding the Diebold effect: S.Walker has dealt with a potential multicolinearity problem between predictors by taking the principal components of a variety of demographic variables.

My brief rejoinder: the residuals of a logistic regression to predict the presence of Diebold machines based on Clinton Campaign presence, median age, % holding bachelor’s degrees, percapita income, and population density are themselves correlated with the residuals of a regression to predict Clinton’s votes based on the same predictors (R=.306, p<.001).

In other words, removing all of the variance due to demographic factors on Clinton’s votes and the vraiance due to demographic factors on the presence of Diebold machines still leaves a “leftover” Diebold effect, connecting the two.