apples and oranges

I have great respect for Steven den Beste, pine for his return to blogging, and will defer to his judgement on matters pertaining to engineering, World War II strategy, and anime. However, his recent analysis of polling trends[1], and especially his conclusion[2], are I think a validation of Razib's basic thesis, that ferociously intelligent people can often be quite dumb when they wander off their intellectual reservation. Steven den Beste is, simply put, not a statistician.

(Of course, if you apply that thesis to me, it means that I'm out of my league either here or here. Preferably the former, otherwise the past six years will have been a remarkable waste of effort...)

[1] It is statistically meaningless to assume that polls with different question methodologies, likely voter models, weighting (or not) of party ID, and sampling distribution between Republicans and Democrats, can be "averaged" together the way that Steven's source material tries to do. Plus, there is a severe misunderstanding of what "margin of error" means - it's not as simple as "if the lead is within the MoE then there is no lead". It's actually a measure of probability. The abuse of statistics aside, these are mostly national polls being analyzed, which can only measure the popular vote, which does not determine the winner by any means. Finally, the polls do not take into account the enormous surge in new-voter registrations across the country, which usually portends poorly for an incumbent,m especially one with approval ratings below the 50% mark, from which no incumbent has ever prevailed. The engineering analogies that Steven invokes are strained, to put it charitably. This isn't engineering, it's social engineering, which is as far removed from the mathematical kind of Steven's expertise as genetics is from Intellgient Design.
[2] In general, conspiracy theories are the sign of a weak argument, especially when the data can be explained easily by long-known phenomena. Polling is neither an exact nor a new science. The kind of manipulation that Steven imputes to his results (themselves false, see point 1 above) would require that polling be much more.. engineering-like.. than it can ever be. I am no engineer myself, but even I know it's not a stochastic process. Perhaps another fallacy of Steven's is the old saw, "to a man with a hammer, every problem looks like a nail."

No comments: