Survey Details

Why our Methodology is More Appropriate for a Primary Election

Academics and pollsters continue to debate the relative merits of samples utilizing voter lists, which contain information about past voting history, versus samples generated by random digit dialing (RDD), a technique whereby computers generate random phone numbers within a pre-determined set of exchanges.

In conducting the poll, we chose the voter list method because we sought to interview likely voters about the upcoming GOP gubernatorial primary. In last year's hotly contested GOP primary for U.S. Senate, less than seven percent of New Jersey's more than 4.5 million registered voters actually participated. Even in this year's general election for governor, less than 40 percent of those eligible to vote are likely to actually cast a ballot. The PublicMind sample included only respondents who voted in the June 2000 Republican Senate primary and who indicated a firm intention to vote in the June 2001 Republican gubernatorial primary. The advantage of sampling from voter lists is that these lists give analysts confidence that they are interviewing actual likely voters.

We chose not to use an RDD sample because such samples do not allow researchers to determine accurately which respondents are likely voters. Lacking any information about respondents' past voting histories, researchers using RDD samples are forced to determine an individual's likelihood of voting by asking each respondent if he or she is registered, what past elections he or she has voted in, and whether or not he or she is likely to vote in the upcoming election. This self-reported information is highly unreliable, since many people, socialized to believe that they should vote, consistently "over-report" their voting practices when talking to survey interviewers. In short, the problem with using RDD samples to measure voter behavior is that they inevitably end up measuring a substantial amount of nonvoter behavior.

Critics of voter list samples point out that, unlike RDD samples, voter list samples eliminate respondents with unlisted phone numbers. However, this disadvantage may be less important in surveys of likely voters. Individuals with unlisted phone numbers fall into two groups: those on voter lists and those not on voter lists. Demographically speaking, individuals with unlisted phone numbers tend to be younger, more transient, and less educated. As such, they tend to look a lot more like nonvoters than voters. Thus, while critics of our survey methodology noted that 37 percent of Hudson county residents have unlisted numbers, the more important issue is how many of these unlisted individuals are actually voters. Given the demographic research on individuals with unlisted numbers, we suspect relatively few. As for individuals on voter lists who have unlisted telephone numbers, there is no evidence that we know of to suggest that their political behavior is significantly different from individuals on voter lists who have listed phone numbers.

Neither RDD nor voter-list samples is inherently better. Each has its advantages and disadvantages, and each is more appropriate in certain circumstances. Because we sought to examine voter behavior in our recent surveys, we chose the methodology most likely to put us in touch with actual voters.