Here is what my “traditional” Senate model now says, updated with new polling from PPP and the potentially dubious firm known as CrossTarget.
Scenario 1:
Under its original assumptions, the model now projects a very slight Brown edge, 49.3-48.7, which maps to a 55 percent chance of winning. Earlier today, it had given Coakley a 57 percent chance of winning. However, because the odds are under 60 percent, we still call this race a “toss-up” per our nomenclature, as we did before.
But what if we made different assumptions?
Here’s one that Democrats will like: let’s remove the CrossTarget poll and keep everything else the same. This is basically the best defensible Democratic case.
Scenario 2:
That brings Coakley’s odds up to 68 percent.
But now, let’s do a couple of things that Republicans will like instead. We’ll put the CrossTarget poll back in the mix. We’ll place more weight on more recent polls, reducing the half-life of a poll from 14 days to 7. And we’ll get rid of the result from the regression model (the reason you might not want to do that is explained here.)
Scenario 3:
Coakley drops all the way down to 17 percent!
Next, let’s make a middle-ground set of assumptions that I’m personally somewhat inclined toward. We’ll keep the half-life at 7 days to place more weight on the most recent polls, but leave the CrossTarget poll out and the regression result in. We’ll also increase by 50 percent the uncertainty parameter, owing to the fact that special election polling is harder than regular Senate polling. Coakley bounces back up to 58 percent under these assumptions:
Scenario 4:
However, even if we’re emphasizing the more recent polls, Coakley is still benefiting from sort of middle-aged Rasmussen and Research 2000 polls, when some pollsters have observed a further decline in her numbers since those surveys came out. Suppose that both pollsters come out tomorrow (as both are rumored to be in the field) with a new poll showing a 5-point decline in her numbers. That would mean Rasmussen shows Brown ahead by 3 points (rather than trailing by 2), and R2K shows Coakley ahead by 3 points (rather than 8). Warning: the following two charts contain hypothetical polls that don’t yet exist. I have no idea what the Rasmussen and R2K polls might say, or whether they’re in the field at all.
Scenario 5:
Now that’s really a toss-up: Coakley at 52 percent to win and Brown at 48.
Finally, let’s restore the original assumptions of Scenario 1, but include the hypothetical polls that we introduced in Scenario 5.
Scenario 6:
Coakley down to 41 percent to win, Brown up to 59, putting the contest just on the fringe of “toss-up” and lean GOP.
To state the obvious, one’s assumptions matter a lot! Any of these are reasonable and defensible sets of assumptions. And I’m sure that you some the more creative among you could come up with other wholly reasonable and defensible sets of assumptions, including some that fall outside the goalposts of the scenarios contained herein.
On the heels of the PPP poll, the consensus of other analysts is liable to be that Scott Brown is favored (which I might agree with in the most literal sense), and favored by a large enough margin to characterize the race as something other than a toss-up (which I don’t yet agree with.) That’s fine; I can see how they get there. The only thing I’d really caution against is that, because our minds are wired to detect patterns, and the story of this race has been Brown! Momentum! Rawwr! it’s perhaps easy to forget about some of the polls that did show Coakley ahead, like the Research 2000 poll (which is no less recent than the Suffolk or ARG polls), the Rasmussen poll (at least until they come out with a fresh one), and the Boston Globe/UNH poll, which is definitely old but showed a 17 (!) point lead. It’s also easy to forget that all of these polls have their hitches: with the possible exception of Ann Selzer’s polling in Iowa, there’s no poll anywhere that should be thought of as the gold standard.