Romer’s claim is wrong. In fact, in my paper which Romer cites (first presented on January 4, 2009 at the AEA meetings and published in the American Economic Review), I explicitly state that one must take account of other variables. Here is a quote from that paper: “policy evaluation requires going beyond graphs and testing for the impact of the rebates on aggregate consumption using more formal regression techniques….an advantage of using regressions is that one can include other factors that affect consumption.” In that investigation, which focused on the 2001 and 2008 rebates, I used monthly data and included in the regressions monthly data on oil prices, which rose dramatically in the first half of 2008 and which would be expected to reduce consumption around the time of the rebates. Indeed, oil prices had a highly significant coefficient in the regressions, and yet I found no significant effect of the rebates as shown in Table 2 of the paper. In another paper published in the Journal of Economic Literature, (discussed in the blogosphere here and here) I used quarterly data to investigate the 2001 and 2008 stimulus packages and also the 2009 stimulus. With quarterly data, I also included a household net worth variable from the Fed’s flow of funds accounts, along with the quarterly average of oil prices. The net worth variable had a significant effect, and yet I still found no statistically significant impact of the temporary payments as shown in Table 1 of the JEL paper.
In sum, my research does consider the impact of third variables, contrary to what Romer claimed. And the results I reported are robust to adding such variables, contrary to what Romer conjectured.