If (as I'm sure you do) you have the times along with the votes it would be cool to see if there was actually a bias when comparing the before response and after response datasets.