Like everyone else, I find the polling picture very confused and confusing (and that's caused me to hesitate in composing a new Senate race ratings diary), but I still would like to make new predictions, for whatever they're worth. In making these predictions, I am considering poll results, dreaminonempty's post on Senate polling error patterns, and ultimately, a bit of guesswork. I will reference previous ratings in my 8/19/2012 diary where they are different, as well as giving new ones.
Safe = No path to victory for the other candidate, absent some unforeseen disaster to the leading candidate.
Likely = Perhaps a 10-20% chance of victory for the trailing candidate.
Lean = Slight edge to one candidate, with the other having about a 25-35% chance of winning.
Tossup/Tilt = A very tenuous edge to one candidate
Arizona: Tossup/Tilt D (was Tossup/Tilt R) - I am uncomfortable with this rating, but I will not take the easy way out and call this a pure tossup. And the fact is, Carmona's polling numbers have clearly improved against Flake, who has proven to be a much less formidable and much more problematic candidate than I and most other observers expected. Lately, most polling has been showing Carmona with a small lead. dreaminonempty theorizes a 2-point difference between the polling average and the final result, but I think that he is likely to win, anyway. The thing that bugs me about this prediction is that it is still uncertain whether Carmona's ill-advised joke about Candy Crowley's looks might yet hurt him. But despite the mild Republican lean of the state of Arizona and great unlikelihood, at this point, of Obama carrying the state, I believe that a combination of a possibly significant number of Republicans and right-of-center independents voting for a really compelling candidate, the likelihood of a humongous Latino turnout for a Latino candidate and against a Republican Party that's made Latinos and immigrants scapegoats for everything they consider wrong with the US, and some pretty serious knocks against Flake's record are likely to propel Carmona to a narrow victory.
California: Safe D.
Connecticut: Likely D. Polling is showing this to be a close race, but it seems like McMahon gets worse as the campaign continues, and while her liberal spending gave her a fighting chance, her mouth will sink her. The margin may be under 10 points, but I don't see an obvious path to victory for her.
Delaware: Safe D.
Florida: Likely D (was Lean D). As with the Connecticut race, I don't see an obvious path to victory for Connie Mack IV, whose campaign has to be deeply disappointing to Republicans around the country. Liberal Democrats will never be passionate supporters of Senator Bill Nelson, but they will vote for him, along with quite a number of independents and some Republicans. Regardless of whether Romney narrowly wins the state or narrowly loses, a victory by Mack would be quite surprising.
Hawaii: Safe D. Hawaii polling sucks, but it's clear that Mazie Hirono will romp in this election.
Indiana: Tossup/Tilt R (was Tossup/Tilt D). I don't know what to think about this race. The final result will depend on what disaffected Lugar voters do. If they split more toward Donnelly and the Libertarian candidate than toward the extreme Republican candidate, Mourdock, Donnelly will win. I just don't feel confident in making a prediction, but since I'm making myself do so, in a very close election, I'll go with my head and not my heart. I hope my prediction proves incorrect.
Maine: Likely I (was Safe I). This race has gotten closer than seemed possible a few months ago, but in spite of lots of spending by Karl Rove's organization and so forth, it is hard to imagine King really losing this. If polls get closer, I'd revise this to Lean I, but right now, I think Likely is a reasonable slot for this. And of course, if elected, King would caucus with the Democrats.
Maryland: Safe D.
Massachusetts: Lean D (was Tossup/Tilt D). Elizabeth Warren made some early mistakes in her advertising strategy, but the strong Democratic lean of her state gave her the room to revamp her strategy, and overall, she has run a smart, authentic campaign and showed the limits of Scott Brown's talents. Brown has been unmasked as a small, bitter man who's likely to lose and lose ugly. I'm not rating this Likely D yet, but it's not easy to see how Brown could come back and win this one.
Michigan: Likely D (was Safe D). I think my rating may have been a bit enthusiastic in August. Some polling has shown this race within single digits. However, the prospect of "Pete Spend-It-Not" actually winning this race definitely seems a bridge too far. The only way he wins it is if President Obama somehow implodes in the next 2 weeks, and I just don't see that.
Minnesota: Safe D.
Mississippi: Safe R.
Missouri: Lean D (was Lean R, updated to Tossup/Tilt D after Akin's "legitimate rape" remarks). I'm almost tempted to rate this as Likely, because Akin lacks money and keeps on hurting himself further with the nonsense that comes out of his mouth. Unlike a lot of other Republican politicians, he really does believe all that crap.
Montana: Tossup/Tilt D. This is another really difficult rating. Polling has tended to show a small lead for Tester, but an Obama win is nearly impossible, and dreaminonempty estimates a 1 to 2-point difference in favor of the Republican senate candidate between the polling average and the actual results. Either way, this will probably again be one of the closest Senate races in the country.
Nebraska: Likely R (was Safe R). Kerrey recently released an internal poll result, showing him with a 5-point deficit. If that is accurate, it's barely conceivable that it could close further, but it's very hard to see an actual win for him. I'd give him perhaps a 5% chance of winning, so Likely is a stretch here.
Nevada: Tossup/Tilt D. Same reasoning as in my previous Senate race ratings diary: A slight opinion poll lead for the Republican is likely to evaporate in actual election returns, given the recent history of Nevada polling. Plus, absentee and early voting so far in Nevada have been highly Democratic.
New Jersey: Safe D (was Likely D). Menendez may never become that popular, but I don't see any path to victory for Kyrillos, and polling lately has mostly not been close.
New Mexico: Likely D (was Lean D). Polling has shown a consistent lead for Heinrich, and Obama will probably win the state by double digits.
New York: Safe D.
North Dakota: Tossup/Tilt R (was Tossup/Tilt D). If I go with my heart, I pick Heitkamp - to whom I sent money - to win. Going with my head, I have to take some of the recent polling seriously. A tie in a public poll of a state that clearly leans Republican is a danger sign for the Democratic candidate. The voters have to pick between an apparently squeaky-clean candidate who's run one of the best races in the country and is known to be very likeable in a state where retail politics are still important and a not very popular, somewhat damaged candidate from the party that will win the state's Electoral College votes by 12 or so. Heitkamp has a great chance, but I would need to see more positive polling to feel more confident of her winning.
Ohio: Likely D. Some of the recent polling has been close, but I really don't see Brown losing.
Pennsylvania: Likely D (was Safe D). The fact that the DSCC has to spend money on this race shows that it hasn't been put away. But I would still rate Casey's chance of victory at 90%.
Rhode Island: Safe D.
Tennessee: Safe R.
Texas: Safe R.
Utah: Safe R.
Vermont: Safe I (caucusing with the Democrats).
Virginia: Tossup/Tilt D. According to dreaminonempty's calculations, the expected result of a Senate race in Virginia should be approximately 2 points toward the Republican side from the polling average, but I think Kaine will win, anyway (and I also expect Obama will carry the state).
Washington: Safe D (was Likely D). I'm not sure many Republicans thought Baumgartner would actually win this race, but they have to be disappointed with his performance.
West Virginia: Safe D.
Wisconsin: Lean D (was Lean R). If Heitkamp may have run the best Senatorial campaign of the year, Thompson may have run the worst of any candidate in a competitive race. In a word, he has been awful. He should have stayed retired.
Wyoming: Safe R.
Safe D: CA, DE, HI, MD, MN, NJ, NY, RI, VT (I), WA, WV
Likely D: CT, FL, ME (I), MI, NM, OH, PA
Lean D: MA, MO, WI
Tossup/Tilt D: AZ, MT, NV, VA
Tossup/Tilt R: IN, ND
Likely R: NE
Safe R: MS, TN, TX, UT, WY
Race changes in favor of the Democrat (8): AZ (Tossup/Tilt D from Tossup/Tilt R), FL (Likely D from Lean D), MA (Lean D from Tossup/Tilt D), NE (Likely R from Safe R), NJ (Safe D from Likely D), NM (Likely D from Lean D), WA (Safe D from Likely D), WI (Lean D from Lean R).
Race changes in favor of the Republican (5): IN, (Tossup/Tilt R from Tossup/Tilt D), ME (Likely I from Safe I), MI (Likely D from Safe D), ND (Tossup/Tilt R from Tossup/Tilt D), PA (Likely D from Safe D).
I'll be interested in reading everyone else's take on these races in the comments.