Fluke poll no reason to say Angelle caught Vitter

So what does it mean when two different polls about the same political contest come up with differences beyond the marginal? It means that observers get a peek into the imprecise world of survey data and the impact it can have on larger perceptions.

The fourth poll in the past five months conducted by The Hayride website, in conjunction with the MarblePort agency, on Aug. 4 and 5 produced governor’s contest results consistent with the previous versions: Republican Sen. David Vitter led the pack, a little ahead of Democrat state Rep. John Bel Edwards, and then with about half as much support comes Republicans Public Service Commissioner Scott Angelle and Lt. Gov. Jay Dardenne. Since May, Vitter and Dardenne have drifted down a bit, Edwards has drifted up a bit, Angelle has made much greater progress, and the undecided vote has shed about a third.

This gives the results what is termed “face validity;” that is, it seems consistent with other indicators. Vitter had the most name recognition and ought to lose some support as other candidates become better known; Angelle has had the most advertising and ought to come up the most; the undecided slowly have peeled off as the election approaches; the results move slowly but steadily in certain understandable directions.

It also appears to have pretty decent external validity. This means survey results, taken from a sample, seem likely to reflect the real world. There can be small quibbles with the sampling frame – 55 percent female likely overweighs the actual voting population in two months (that’s close to the registrant total, but historically nationally the recent female/male gap when actually voting has been only about 5 percent) and 14 percent registered no party/other voters likely underweighs this group (that was their proportion of turnout in 2011, but since then their total registrations have increased 2.2 percent) – and that Edwards is getting just about two-thirds of the black vote, where of the undecided blacks a majority of them will vote for him with almost all of the remainder not voting, understates by a few points his support if the election were held today. Yet all in all it looks reasonable.

But another poll, by Market Research Insight, done at the end of July shows a dramatically different result. Commissioned by a group of wealthy individuals interested in politics and potentially willing to assist candidates, this one puts Vitter, Edwards, and Angelle all between a fifth and fourth of the intended vote, with Angelle actually leading the way. Redistributing the undecided black vote by giving 90 percent of it to Edwards gives him a realistic third of the vote and puts Vitter and Angelle in a tie (although in reality a significant chunk of these black undecided voters simply will not vote).

So how could such dramatically different assessments of Angelle and Vitter be present in polls taken just days apart, with the earlier of the two completely deviant of the trend subsequently recently enforced by the later one? The answer lies in the sampling frame. This one overweighed registered Democrats by 6 percent and contained only 12 percent no party/other. This is reflected in a self-identification question concerning partisanship, where 27 percent claimed independence, 31 percent Democrat (or leaning that way) and 39 percent Republican (including leaners).

In other words, the sample disproportionately picked up people registered as Democrats and those who registered under that label, probably decades ago, but who do not vote consistently Democrat, if even a majority of the time, and thereby perceive themselves as independent. That latter group has become a prominent phenomenon in the study of political behavior, part of a slowly detaching Southern Democrat electorate, over the past couple of a decades and for its members Angelle in particular would be a reasonable choice, having been a Democrat until the past few years and working under elected Democrats, even if he worked under GOP Gov. Bobby Jindal and now calls himself a Republican, accentuated by him being the most actively advertising candidate over the past quarter of the year.

In the final analysis, the poll set up a sample biased towards Angelle and against particularly Vitter. Unrepresentative samples happen from time to time despite the best intentions of the pollster (which will be assumed here), much less the random chance of getting an unrepresentative sample despite having it closely match population parameters (this one collected enough responses to adhere to the typical 95 percent standard with a four percent margin of error; that is, one out of 20 of these polls using the same sampling frame will produce statistics in reality outside four percent either way of their observed values).

As a result, just as the Edwards camp got all excited when it cherry-picked a selected statistic (interestingly, from a previous MRI poll)  to make their candidate appear to be doing better than he actually was, the Angelle camp should pay little heed to the aggregate numbers of this one. This contest with the candidate field as it is still has Angelle and Dardenne substantially behind Vitter and one needing to pass him in order to head to a runoff with Edwards. More data must appear inconsistent with this hypothesis in order to doubt it.