Couldn't you encode the sizing with the size of the pie? ie, if Independents are the smallest group, they should be a smaller circle. In this manner, area allows for accurate comparison (it could be that more Republicans disapprove of direct voting than Independents approve for example).
You could do that, but then you compound the problem of comparing the size of slices with an additional size of circle comparison. Humans aren't good at comparing size between these kind of shapes. We're much better at comparing rectangular shapes. You'd be much better off splitting the visualization into two bar charts - one for the % of group calculation and one for the total count of people calculation.
