Considerations

Most analyses of voting records begin by looking at the bills, and only looking a "key" votes (there are a lot of procedural moves in the roll call, as well as nonsense votes). So, by careful selection, the National Journal is able to "show" that Barack Obama has the most liberal voting record in the Senate (more liberal than Bernie Sanders, the socialist), just as they did for John Kerry in 2004. Conversely, Progressive Punch gives Clinton a 91.5 to Obama's 89.28. One of the ratings sites used to call Obama a "Rank and File Democrat" and Clinton a "Radical Democrat", but I can't find it, so maybe they're in the process of reversing the designations.

Furthermore, there's really no way for an outsider to know by looking at the write-up on Thomas what a particular vote really means (is an "Aye" vote on a procedural move the same as a "Nay" vote on the underlying bill? Only sometimes.) The only people who can possibly know what a vote actually are the people in the Senate, so I decided to avoid classifying bills or roll-calls at all. Let the Senators classify them, by their votes.

Methodolody

GovTrack.us has rdf yearly files of roll call votes here, which list by bill who voted which way. Using a simple xml parser, I invert these so I have a list for each senator of how they voted on each bill. That makes it easy to compare each senator's voting record for a year.

I started by producing two lists: Obama's voting record compared to everybody else's, and Clinton's voting record compared to everybody else's. These are sorted by descending order of percent of agreement. For example, here's the top of Obama's 2007 list (percent agreement, number of comparable votes, senator):

and here's the top of Clinton's 2007 list:

So Clinton and Obama agreed 96.3% of the time, and both were in close agreement with Durbin, Schumer (the senior Senators from their states), Lautenberg and Menedez. But I noticed that sometimes there was a fairly large gap between where a Senator fell on the two lists. So I combined the two lists by subtracting Clinton's percent of agreement from Obama's for each Senator. A negative number therefore means that Clinton agreed with that Senator more often than Obama, while a positive number means Obama agreed with that Senator more often. When I ran that analysis, the results were startling. Where they differed, (which, remember is really not all that often), Clinton was far more likely to side with a Republican.

Here are the top 7 senators in 2007 with whom Clinton was more inclined to agree than Obama (senator, net percentage):

And the bottom of the list (the 7 senators with whom Obama was more inclined to agree):

Notice there are 6 Rs and 1 D on Clinton's list, and 3 Rs and 4 Ds on Obama's. But we're talking about small numbers here. A few examples is not enough. We need an unmistakable trend. And here it is.

Where Hillary Clinton and Barack Obama differ, Hillary Clinton is far more likely to take the Republican side.

If you want the (python) code that produced these results, PM me with your email. And if you're wondering why the numbers don't seem to balance, so did I. Until I realized I could produce an example of 4 voters and 3 bills where the results wouldn't "balance", either.