Well, Princeton PhD student Ben Lauderdale has helped to explain why this is the case in a new paper (via Monkey Cage). Lauderdale uses roll call votes to estimate a "maverick score" for members of Congress. This is roughly a measure of how difficult it is to classify members' votes because their behavior is so unpredictable. Lo and behold, topping the list is our friend Russ Feingold:
John McCain was quite the maverick back in 2001-02, but today he doesn't even crack the top ten.
This strikes me as a very cool measure and a useful one. But I suppose the next question is, what exactly are we measuring? What does one's maverick score indicate? McCaskill and Bayh are both in moderate-to-conservative leaning states, so maybe they have to be a little mavericky to stay in office. But Feingold is quite safe in his seat, as is Bernie Sanders, number four on the list. So electoral safety/vulnerability doesn't look like the cause. (Lauderdale also notes that Rep. Barney Frank has had a pretty high maverick score throughout his career, with the exception of when he faced potential expulsion in the 101st Congress due to a scandal. In this case, his career incentive was to be as predictable as possible.) Is this just a personality type?