The Senate race in Connecticut is one that Democrats are becoming increasingly anxious about. Their candidate there, Attorney General Richard Blumenthal, once held leads of 20 to 30 points in most polls over Linda McMahon, the Republican nominee. Since then, the race has tightened considerably, and Ms. McMahon, the former chief executive of World Wrestling Entertainment, is within striking distance.
But has the race tightened further over the past several weeks? Some smart analysts, like The Washington Post’s Chris Cillizza, are suggesting so on the basis of a new poll from Quinnipiac University, which shows Mr. Blumenthal ahead by only 6 points, 51 to 45. Last month, a Quinnipiac poll showed Mr. Blumenthal with a 10-point advantage.
The new Quinnipiac poll, however, does not provide for an apples-to-apples comparison. Their August poll, and their previous polls of the race, was conducted among all registered voters. The new poll, on the other hand, was conducted among the narrower group of likely voters.
There is an abundance of evidence this cycle that likely voter polls are more favorable to Republicans than registered voter ones — a reflection of the superior enthusiasm among their voters this year. And that’s great news for Republicans, because likely voter polls are generally more reliable, especially in midterm elections.
FiveThirtyEight’s forecasting models include an adjustment for this, in fact. By analyzing polling from firms that have released both registered voter and likely voter results, we have found a gap between the two of about 4 points, on average, favoring the Republican. Therefore, a registered voter poll that shows an 8-point lead for the Republican is in fact treated by our model as showing the equivalent of a 12-point lead.
Now, let’s look again at those two polls in Connecticut. Quinnipiac’s August poll gave Mr. Blumenthal a 10-point lead among registered voters; their poll this morning gave him a 6-point lead among likely voters. In other words, we see a 4-point shift in the Republican candidate’s favor — exactly what we should have expected based our previous analysis of likely voter polls.
But the shift reflects the change in methodology: it provides no evidence that the race is tightening.
As it happens, there is a convenient counter-example in the form of Rasmussen Reports polling. Rasmussen also surveyed the Connecticut race last week, and they had also done so in August. Both polls were conducted among likely voters, so an apples-to-apples comparison is more appropriate.
But Rasmussen shows no tightening in the race; their August poll had given Mr. Blumenthal a 7-point lead; now, they show him with a 9-point lead. Since the sample size on both polls was fairly small, we should probably not make much of the modest movement in Mr. Blumenthal’s direction — but, obviously, there is also no trend toward Ms. McMahon.
This is something to be mindful of as other pollsters switch from registered voter to likely voter models, as tends to occur this time of year. If a poll shows strong numbers for the Republicans among likely voters, that is — unambiguously — good news for them. But it may not reflect further “tightening” or “momentum” toward the Republican, if the comparison point is with a poll of registered voters.