Every polling organization has its own way of determining likely voters. This usually entails some combination of self-professed likelihood to vote, past voting behavior, and interest in the election.
Political pollsters tend to use samples drawn from registered voter lists that include actual voting history, but these lists miss many voters (due to errors in matching telephone numbers, accurate record keeping, etc.). List quality is a definite problem with New Jersey records. Public pollsters on the other hand, tend to rely on random digit dial samples that encompass more households, but rely on respondents’ self-reports of their past voting behavior, which they tend to overstate.
There are pros and cons to both sampling methods, and I have personally used both depending on the type of election, as I discussed here and here. Just as private pollsters have techniques to adjust for list quality issues, public pollsters have techniques to adjust for false self-reports of past voting (e.g. asking the respondent to identify his or her polling place). However, neither approach is fool-proof.
Regardless, both methods of determining likely voters rely to some extent on a self-professed interest to vote in the upcoming election. In October, that’s usually not a problem. In August, that report can be highly suspect, especially in a state like New Jersey where voters are notoriously late to engage. The summer likely voter, like Thomas Paine’s summer soldier, can be unreliable.
[This is also one of the reasons why I tend to be dismissive of polls in non-election years that purport to measure policy opinions among “likely voters.” Likely to vote in what? “Most” elections? Without a specific election to anchor it, the determination of likely voters has to be highly arbitrary on the part of the pollster.]
Moreover, the “horse race” question in summer polls is the indicator most prone to volatility, which is compunded by the instability of the likely voter model. Seasoned campaign observers view horse race results as ballpark numbers (i.e. any way you slice it, Chris Christie has held a consistent lead). They focus more on the candidates’ favorability ratings and issue advantages, which tend to be harder to change once voters’ opinions are formed.
Even with this potential for volatility, we usually don’t see a wide divergence between registered and likely voters in the horse race numbers. Unfortunately, that is partially due to the fact that most private campaign polls, and many public polls, don’t provide results for both groups of voters. So the bottom line is that we don’t really know how often this divergence occurs.
In the poll we released today, Republican Chris Christie has built support among groups that nearly always vote. Moreover, the recent corruption arrests in New Jersey have had the ancillary effect of increasing turnout likelihood among voters who desire a change for the state. On the other hand, President Obama’s campaign stop last month helped incumbent Jon Corzine with minority voters, but unfortunately they continue to be much less likely to vote. The end result is that Corzine has been able to keep the RV number relatively close while falling behind in the LV number.
[As an interesting side note: I wrote yesterday (scroll down) that the Obama visit did not yield the hoped-for impact according to our poll, but we should wait until he started appearing in Corzine ads to judge. Coincidentally, a few hours after I wrote that, the Corzine campaign put up their first ad featuring the President’s Garden State appearance.]
As I stated in our press release today, the role of the Monmouth University Polling Institute is not to predict outcomes, but to explain where an electorate stands and why. At this early stage, understanding divergent views among both likely and unlikely voters indicates the potential for change as the race heads into high season.
For more on likely voter modeling, please visit Pollster.com.
Addendum 8/07: To consider why both the RV and LV numbers are worth looking at this early in a race, it might help to put it in the context of raw numbers. There are about 5 million registered voters in New Jersey. By examining the voter rolls, you can probably (and these numbers are hypothetical) identify about 1.5-2.0 million who always or nearly always vote and eliminate about 1.5-2.0 million who never vote or only vote in presidential elections. That still leaves you with at least one million voters whose likelihood to vote this year is still up in the air, regardless of whether you use list-based or random dial sampling.