November 06, 2012
Michael Gerson used his column today to make a bizarre attack on the NYT’s polling analyst Nate Silver. He complains to readers:
“Silver’s prediction is not an innovation; it is trend taken to its absurd extreme. He is doing little more than weighting and aggregating state polls and combining them with various historical assumptions to project a future outcome to project a future outcome with exaggerated, attention-grabbing exactitude. His work is better summarized as an 86.3 percent confidence that the state polls are correct.”
Actually Silver is doing nothing more than weighting and aggregating state polls and combining them with various historical assumptions. He is very clear on this. Gerson can go to Silver’s website and find in great detail the methodology that Silver uses for weighting various polls based on their past track records. Gerson apparently thinks this is an indictment, complaining about Silver’s precise 86.3 percent probability estimate.
The real problem here is simply that Gerson does not understand what Silver is doing. Silver’s 86.3 percent prediction is premised on the assumption that the polls do not contain a systemic bias and that there is not some event(s) that radically shifts the attitude of the electorate between the last round of polls and the election. With these assumptions we can treat the polls as comparable to the draw of white and black marbles out of a huge jar.
If we take enough draws of 1000 balls (you’re welcome to use a different number) and the average of each of these draws is that 50.5 percent of the marbles are black, we can begin to say with great confidence (which can be specified with many decimal points) that the majority of the marbles in this huge jar are in fact black. This is what Nate is doing. He does adjust the draws — some polls consistently find more white or black marbles than the average of the other polls. Unless these polls have proved to be accurate in these divergences, Nate makes an adjustment for their tendency to find too many white or black balls. This process is the value-added that Nate provides over a simple averaging of the various polls.
This doesn’t guarantee that Nate will prove right. There could be some systemic bias in the polls. This would be comparable to a situation where the black balls are heavier and therefore fall to the bottom and are less likely to be in the draw. The way to argue this case is to present a reason for why the polls could be biased. There are possible stories here: voters with only a cell phone may be undercounted, the assumptions about who is a likely voter may prove wrong, or the polls may be undercounting Spanish speaking voters.
These factors, and others, could lead to a systemic bias in the polls. But if Gerson, or anyone else, thinks this is the issue now, then it is incumbent on them to make the case, not get angry at Silver for using statistical methods.
The real problem is that Gerson just seems to have difficulty with numbers. He concludes his piece by telling readers:
“And so, at the election’s close, we talk of Silver’s statistical model and the likely turnout in Cuyahoga County, Ohio, and relatively little about poverty, social mobility or unsustainable debt.”
Yes, it would be great if we had more discussion of poverty and social mobility throughout the campaign and beyond. It’s hard to blame Silver for the lack of such discussion. The pre-Silver elections were not exactly dominated by serious discussions of major national issues. I recall in the 1988 presidential election when the big issues in the race between George H.W. Bush and Michael Dukakis were Willy Horton and the pledge of allegiance.
As far as the third item on Gerson’s list, unsustainable debt, this is where his knowledge of math again fails him. Here is the ratio of interest payments to GDP over the last four decades:
Source: Congressional Budget Office.
As can be seen, the debt burden is very far from unsustainable, the interest burden is near its post-war low. In other words Gerson is angry because he thinks that somehow Silver’s polling analysis has diverted the country from discussing a non-existent problem.
Comments