Nate Silver suspects that PPP's original Michigan poll was an outlier and that part of the tightening in Michigan is due to regression to the mean. How he sees the race:

It would certainly not be out of line to characterize Michigan as a tossup. As my colleague Michael Shear notes, the Republican debate on Wednesday could prove decisive in Michigan if either Mr. Romney or Mr. Santorum emerges with a clear win. The same could hold in Arizona, where the debate is being held. Our model now gives Mr. Romney an 81 percent chance of winning in Arizona. But that does not account for a new Public Policy Polling survey, to be published on Monday night, which is likely to show a tighter race there.

 Ed Kilgore adds:

[O]ne popular idea—that Michigan is the new Florida, and Santorum’s eventual decline and fall in the former state is as inevitable as Newt Gingrich’s was in the latter—isn’t really based on empirical evidence, so far. Perhaps Mitt or his Super PAC will go savagely negative on Santorum in Michigan between now and February 28, and perhaps it won’t backfire. But that all remains to be seen.