I ran a regression with change in the Democratic vote from Landrieu to Blanco as the dependent variable. As independent variables I used % white in the county, % for Bush in '00, and % for Duke in '91. Both the % white and the % for Bush are statistically insignificant, but the Duke variable is highly significant (p<.000). For each 10 point change in the Duke vote, Blanco did about 4.6 points better than Landrieu.