Murray on Suffolk poll

Pollster Patrick Murray says that the turnout model on a Suffolk University poll released this morning is based on a 93% turnout in a race where less than half of the state's registered voters will turn out.  The poll, which included all twelve candidates for governor, has Gov. Jon Corzine leading Republican Christopher Christie 42%-33% among likely voters, with 7% for independent Christopher Daggett.

The full text of Murray's statement:

The Suffolk poll vote choice question is interesting in how it tries to mimic the ballot.  The 7% they get for Chris Daggett does match the "hard" support for Daggett that I found in our last poll.  However, there are a few things in their results that don't gibe with what other pollsters have been finding here in New Jersey.

The first issue is related to party identification.  The pollster weighted their party identification question to New Jersey's voter roll's party registration.  While that might work in some state's it does not correlate as well in New Jersey, as I explained in a blog post today (http://monmouthpoll.blogspot.com).  Even still, this should have a minor impact on the Suffolk poll results.  The vote choice split among their independent voters (35% Corzine to 31% Christie) is unlike every other New Jersey poll, which has Christie clearly ahead in this voter group.  Are they catching a major voter switch?  I'm not sure.

The bigger discrepancy seems to be with their likely voter model.  They use a relatively easy screen ("Are you very, somewhat or not very likely to vote?").  Only those who say they are "not very likely" are excluded from the sample.  Based on a conversation I had with the Suffolk pollster, 93% of registered voters made it through their likely voter screen.  That's in a race where turnout is expected to be below half of all registered voters.

I'm not a stickler for making turnout models exactly match turnout expectations.  A review of other research and my own experience indicate that over-stating turnout by about 10 to 15 points generally provides better poll estimates, because it includes those people who are not likely to vote, but will end up being prodded to the polls anyway.

Normally, likely voter models have little impact, as they tend to be only a couple of points different from the results for all registered voters.  However, that is not the case in this election. The Monmouth University/Gannett Poll has consistently found a larger than average divergence between the vote choice of registered voters and likely voters.  There are quite a few Democratic leaning voters out there who will vote for Corzine if you put a gun to their heads and march them to the polling place, but they won't go of their own volition.  Some of them will vote in the end.  The question is how many.  I doubt it'll anywhere near 93%.