Hawthorne:
There have been a few questions floating in the back of my mind over time that I hope you can help me with.
Of what value is Rasmussen's strongly of approve/strongly disapprove numbers except to measure intensity? If they do measure intensity only, is there a way to equate that with turnout? Do we have any historical data on actual turnout vs. Rasmussen polls to give us an indication of how the numbers in these polls might shape the general approve/disprove numbers?
Is it true that he general approve/disapprove poll numbers are more predictive of election outcomes? Does that answer vary with the kind of election, for example, are primaries, midterms more reflected by strongly/approve numbers than by general approval numbers?
You referred to exogenous factors, do we have a way of relating events to poll numbers? Do we have an understanding of lag times? Do we have a way out distinguishing between a traumatic exogenous event, such as 9/11, and the slow grinding, sandpapering process conducted by the media?
I don't expect you to have these answers at your fingertips, I hope that you can advance the discussion.
>> Of what value is Rasmussen’s strongly of approve/strongly disapprove numbers except to measure intensity? <<
Great question!
Rasmussen says his “Index” (SA minus SD) has been predictive of changes in overall public opinion. I take him at his word, but I have no idea as to the time lags involved. It’s something that could be studied by using standard tools of statistical modeling like multivariate regression analysis. But I’m not aware that anybody has undertaken a serious study. If I had the mathematical chops, the proper software (SPSS, etc.) and the time, I’d love to try fitting Nerlovian lags to the data — but then . . . .
>> If they do measure intensity only, is there a way to equate that with turnout? <<
Another great question! If the RNC and the DNC are smart enough, each ought to have their own bevy of highly trained statisticians working on just that matter.
In fact, I seem to recall Karl Rove mentioned in his recent book that he had hired conservative professors of political science to do sophisticated statistical work for him. The relationship of intensity to turnout is just the sort of thing such researchers ought (IMO) to study. But if the parties or any political consultants ever sponsor such work, I think they would try to keep their results confidential.
>> Do we have any historical data on actual turnout vs. Rasmussen polls to give us an indication of how the numbers in these polls might shape the general approve/disprove numbers? <<
The data certainly do exist. If I were a professor of political science, I’d probably have several of my Ph. D. candidates writing dissertations using precisely the data in question!
>> Is it true that the general approve/disapprove poll numbers are more predictive of election outcomes? <<
I don’t see how it would ever be realistic to say the “general” numbers are more predictive than the “strongly” numbers. They both are important. And it’s not at all difficult to build statistical models that incorporate both. But then the question becomes: Which weights should one assign to the two components?
In the very simple statistical model I’ve built to assess the effects of intensity, I’ve simply made what I think are reasonable assumptions, for example, that Rasmussen’s “strongly” voters on both sides would turn out at 85% and his “somewhat” voters would turn out at 60%. But I’ve done a good bit of sensitivity analysis, to investigate the effects of varying these assumptions. Interestingly, what I’ve found is that when you vary the assumed turnout numbers over ranges of about ten per cent, the final “victory margin” usually stays within one per cent.
>> Does that answer vary with the kind of election, for example, are primaries, midterms more reflected by strongly/approve numbers than by general approval numbers? <<
I’ll bet the outcomes of primaries are a lot more sensitive
to the “intensity” measures than are the general elections. But then the question becomes, How much? The only way to know would be to do careful historical analysis. And again, it’s an area of research that ought to attract the interest of statistically-minded researchers in academic political science — as well as the interest of politicians and political consultants!
>> You referred to exogenous factors, do we have a way of relating events to poll numbers? <<
If you were doing something like a multivariate regression analysis to explain the relationship of “intensity” to electoral outcomes, then you’d probably try to isolate the effects of exogenous events by using a techique like dummy variables. So yes, there is a way to handle exogenous events. But there will never be a “general rule” as to how important any particular exogenous event will be. Every case will be different.
>> Do we have an understanding of lag times? <<
An excellent topic for a major research paper!
>> Do we have a way out distinguishing between a traumatic exogenous event, such as 9/11, and the slow grinding, sandpapering process conducted by the media? <<
I think it wouldn’t be difficult for a clever model builder to construct a reasonable set of equations that account for the differential effects you’ve posited. That conceptual step would probably be the easy part. But then would come the difficulty of fitting historical data to the model by deriving correlation coefficients, investigating alternative model formulations, doing statistical confidence tests on the various coefficients, and (finally!) testing the resulting models against independent data that weren’t used for deriving the models’ parameters.
At that final stage, the models derived might be useful — or they might be useless. The real-world utility of such analytical techniques just isn’t the sort of thing that can be predicted with confidence, at least not until there has been a lengthy process of trial-and-error revision.
In any event, thanks for a very thoughtful set of questions. You give us much to ponder. And I’m glad to discuss further!