Nick Rowe has an excellent, easy to understand post on the difference between economic analysis and economic forecasting that anyone that's ever waxed poetic about economics not being a science because it's not the greatest forecaster should read. It's very basic stuff, but important stuff.
When the Romer-Bernstein unemployment forecast ended up being wrong it barely registered with me. I could care less. The fact is, forecasting has almost nothing to do with the impact estimates you see when you read about policy analyses.
Daniel,
ReplyDeleteThese are very interesting points, and I need to concede that I hadn't thought all of this through completely when I argued with you about this some weeks ago. You're right. It IS a bad oversimplification to argue that, for example, the inaccuracy of the Romer-Bernstein unemployment forecast implies failure of the policies that the forecast suggested should be implemented. I see more clearly now what you were getting at when you said things like, just because the weatherman gets it wrong doesn't mean we shouldn't take his advice about whether to carry an umbrella on a particular day.
I do worry, though, about actually obtaining the counterfactual against which to compare the effectiveness of a policy. AND, even if the counterfactual is obtained, the analysis of effectiveness is still complicated by the large number of relevant variables.
I know, I know. That's why it's so important to employ sophisticated statistical methods. But even there, I end up worrying about things like what's a 'low' p-value. There was an article in the New Yorker recently by Jonah Lehrer that discusses the weakness of scientific work that depends heavily on finding statistical correlations (at least, that was my take-away from it).
Wish it was mine, but it was written by my co-blogger, Stephen Gordon!
ReplyDelete