Dave's Methodology Corner

>> Tuesday, September 08, 2009

I usually find Nate Silver's stuff over at 538 (and especially at BP) careful and sound, if often atheoretical.  On balance his foray into politics and political science in the blogosphere has been excellent, and we're all better off for it.

However, this is problematic.  Not for the findings per se, because they're irrelevant -- nobody should read much of anything into a bivariate analysis based on a small handful of data points (and with an R2 of .35 or so), but for the method to begin with.  What we have here are what appear to be a bit over 30 data points, plotting unemployment as the independent variable against support for unions as the dependent variable.  I'm surprised he put it up at all, since a) he is quite adept with multi variate modeling, and b) union support almost certainly is a function of a bunch of variables, of which unemployment may (or may not) be just one.  Furthermore, I strongly suspect that there is a time element involved -- that union support varies by generation, and this contextual effect may mediate the explanatory power of any single independent variable.  To wit, even if we accept that there is a causal relationship between the unemployment rate and union support, this is probably a stronger relationship when union membership is a smaller percentage of the work force . . . which also calls into question the assumption that the relationship is linear.

Other equally troubling issues with this analysis:  As one commenter points out, there are additional data points in the Gallup data that are not included in the analysis.  Additionally, as another commenter correctly suggests (and many more probably considered), with a small, small N, the effect of any outlier might just warp the analysis.  There is no doubt in my mind that if the data point in the lower right hand corner is dropped -- and that data point appears to be today -- the relationship between the two variables almost disappears entirely.  

But hell, at least a lot if not most of the comments are of the "unions suck and are evil communists!" vs. "no, big business sucks and are greedy capitalist pigs!" genre, which I always find comforting and amusing.


Post a Comment

About This Blog

  © Blogger template Simple n' Sweet by Ourblogtemplates.com 2009

Back to TOP