The interesting thing about this poll is that they use a [partisan turnout model based on the jungle primary. While this is basically never the case (both these exit polls are never right on and the electorate between the two is never the same), it does at least put them in a position to be able to give a sound rational for their demographic breakdown. The poll does however under-sample Women (based on the same exit poll) by 3%. She led women by 12 points over Cassidy and between Cassidy and Maness votes was within one point of their share put together on Women. That being said, the exit poll suggests that she recieved less of the vote than she actually got on election day. Therefore, this exit poll is likely underestimating Democrats that actually showed up to the polls.
This poll also under-samples African Americans by a couple points. A couple points in the comparison to the exit poll that under-sampled Landrieu.
This poll is still much more representative than the Gravis poll (yes the same firm that has a phantom Virginia poll that is hiding somewhere. I also own some swamp land containing magic beans I will sell you). The Gravis poll (in comparison to the Cassidy skewed exit polls) under-samples women by 4%, African Americans by 5%, but over samples Democrats by 9%. They also under-sample the under-50% crowd by 10%. Kind of scattered demographics. No wonder they retracted a poll this year after release.