One of these Polls is Not Like the Others...

I awoke this morning to discover-- thanks to the Sunday Boston Globe -- that the Gubernatorial race here in Massachusetts is now looking like a toss-up (5 weeks or so from election day).

I must confess to being a bit surprised....after all, last week the typically reliable Suffolk Poll had Governor Patrick leading by 7 percent (outside of the poll's margin of error) while the Globe Poll (conducted by Andrew Smith at the University of New Hampshire) has the Governor leading by a single point (a statistical tie given that it is well within the poll's margin of error). 

As I have discussed previously, up until today the preponderance of the polling evidence has indicated that Governor Patrick has a relatively decent lead and a high probability of victory.  So what, if anything, has changed and what possible explanations can we offer for the apparent change in the Governor's political prospects.
 
A closer look at the technical details and findings of both the Suffolk and the UNH/Globe Poll is revealing:

1) Even though they were released several days apart, these two Polls were actually in the field at the same time:

The Suffolk Poll- Conducted September 16th-19th (Source: Boston Herald)
UNH Poll- Conducted September 17th-22nd (Source: Boston Globe)

This begs the question of whether there was a major sea change in public opinion during the three days last week when the UNH poll was in the field and the Suffolk Poll was not.  And if so, why?

UNH reports that they interviewed 522 adults (471 of which were likely voters) and so they averaged about 87 interviews per day over the six day period.  So, for the sake of discussion, let's assume that 261 (or 87*3) of the 522 respondents were answering the UNH questions during the three day period in question.

In order for the data collected during those three days to have a large enough impact to explain the difference between the two polls, there would have had to have been a pretty large shift in sentiment that UNH was able to capture and Suffolk missed. 

So, what, if anything, happened between September 20th to the 22nd that might account for this change?  A few possibilities come to mind:

1) We have had some debates....maybe they made a difference?

The latest debate was broadcast on the evening of September 21st and so the UNH/Globe Poll did interview somewhere between 87 and 174 of its respondents during or after the debate.  Could this have made a difference?  Sure.  Could it explain the entire difference?  Seems unlikely to me.

In order for the debate to have made a difference of this magnitude, it would have had to have had a huge audience and there would have to be a very large proportion of non-Baker voters whose minds were changed by the debate. 

2)  Didn't Tim Cahill's campaign manager and senior advisor just quit?  Couldn't this have led to some of his supporters defecting to Charlie Baker?

Maybe, but any impact could not have been reflected in the UNH/Globe poll.  According to the Herald,  Cahill's campaign manager quit the evening of the 23rd (his campaign manager quit shortly thereafter), after the UNH/Globe Poll had completed its interviews. 

3)  Maybe the undecided voters decided en masse last week and they are disproportionately declaring their intention to vote for Baker?

Nope.  The Suffolk Poll identified 10% of the voters as undecided, the UNH/Globe Poll reported that 14% of voters remain undecided.

At this point some of you may be asking yourselves, "Isn't it possible that the Suffolk Poll was wrong?"

Indeed it is. However, since the Suffolk Poll is consistent with the previous polling data, it is the UNH/Globe poll that, as the folks from Sesame Street so nicely put it, does not look like the others and therefore is the "data point" that requires an explanation.

So, absent some missing "real world" event that changed voters minds, one has to wonder whether the UNH/Globe Poll is an outlier, a statistical anomaly that occasionally and unavoidably throws a wrench into the best laid plans of survey researchers to conduct a valid poll.

If, as is common practice, the estimates contained in the UNH Poll are offered with a 95% Confidence Interval this could simply be a case of bad luck for UNH's Dr. Smith.  In this context, what a 95% Confidence Interval means is that if the Poll were conducted 20 times, one of those times the estimates would fall outside of the margin of error, perhaps even well outside. In other words, any given poll has a 1/20 chance of being wrong, perhaps even wildly wrong.

I can't help but wonder whether this is one of those times....