This interview with this year’s Nobel laureate in economics, Angus Deaton, is currently making the rounds in social media. It is from a forthcoming volume from MIT Press called Experimental Conversations.
There is lots to chew on here, but this is the quote which I find most meaningful.
One of the first things one learns in statistics is that unbiasedness is something you might want, but it’s not as important as being close to the truth. So a lexicographic preference for randomized control trials-–the “gold standard” argument–-is sort of like saying we’ll elevate unbiasedness over all other statistical considerations. Which you’re taught in your first statistics course not to do.
I can confirm that in the only graduate level statistics course that I took, the idea that we should privilege unbiasedness over, say, mean squared error would have been met with some puzzlement. This was back in 2004, though, so things might have changed. It is rather amazing, when you think about it, the extent to which unbiased estimation has become the objective of empirical work in the past twenty years, and with relatively little commentary about other things besides bias that empirical research might seek to minimize.
(To be very clear, I basically share this view myself, but I can’t help but wonder how we got this way. And if you are reading this as a criticism of experimentalism in political science and related disciplines, sorry, but that’s not my intent here. It’s perfectly reasonable to stipulate that your objective is to minimize bias and to invite your reader to judge you on that, I just wish we’d see more about other types of objective functions too.)
I also quite appreciated Deaton’s references to Heckman and Manski and others who were quite aware of the problems with naive regressions for policy work. You can read my own reference to Deaton as a critic of experiments here.