I did a “critic from the right” post at The New Republic on the stimulus debate, and Jon Chait has responded to it. I have a long post up replying in turn. Here is the key part of my response:
It is nerdy-sounding, but I believe critical to this discussion, to distinguish between measurement and knowledge. I made a very strong claim about measurement, and a very specific claim about knowledge.
I claim that we cannot usefully measure the effect of the stimulus program launched in 2009 at all. We can call this a “natural experiment” all day long, but in the absence of a control case, we cannot know what output would have been had we not executed the policy. Econometric models are not sufficient to estimate this counterfactual. Therefore, there is no achievable level of output in the United States in 2010, 2011, and so on that would enable a definitive answer to the question, “What was the effect of stimulus spending on output?” See, for example, in my original post, the response of leading economists when confronted by unemployment with stimulus that turned out to be higher than they projected unemployment would be without stimulus:
Ms. Romer famously projected in January 2009 that without government support, the unemployment rate would reach 9%, but with support the government could keep it under 8%. It’s 9.5% today.
Some Obama administration officials privately acknowledge they set job-creation expectations too high. The economy, they argue, was in fact sicker in 2009 than they and most others realized at the time. But they insist unemployment would have been worse without the stimulus.
All potentially useful predictions made about the output impact of the stimulus program are non-falsifiable. Failure of predictions can be simply justified by this sort of ad hoc explanation after the fact.
And pace Chait’s argument that private forecasters’ models all estimate a positive effect from the stimulus (implicitly because they all econometrically estimate a lower counterfactual than actually occurred), see Stanford Professor of Economics John Taylor’s analysis that adds to this list alternative economic models from the European Central Bank and Harvard that show no material effect of the stimulus. This argument will always degenerate back into endlessly dueling regressions, because there is no ability to adjudicate among them via experiment.
This does not mean that we have no knowledge about the potential effects of stimulus spending. It simply means that we have no scientific knowledge about this topic. Macroeconomic assertions about the effect of a proposed stimulus policy are not valueless, but despite their complex mathematical justifications, do not have standing as knowledge that can trump common sense, historical reasoning, and so on in the same way that a predictive rule that has been verified through experimental testing can.
When using stimulus to ameliorate the economic crisis, we are like primitive tribesmen using herbs to treat an infection, and we should not allow ourselves to imagine that we are using antibiotics that have been proven through clinical trials. This should not imply merely a different feeling about the same actions, but should rationally lead us to greater circumspection.