A Replication of ‘‘Coveting Thy Neighbor’s Manufacturing: The Dilemma of State Income Apportionment’’ (Journal of Public Economics 2000)

Goolsbee and Maydew (2000) reported that lowering the weight on payroll in states’ corporate income tax apportionment formulae had the potential to raise manufacturing employment. Their analyses continue to be cited in academic articles and are still influential in the policy debate. I gather data and attempt to replicate their analyses and findings. I identify an apparent but inconsequential error in G&M’s sample, and I replicate the most widely cited result in the original paper. Other results are substantively but not quantitatively replicated. I show that G&M’s results are sensitive to relatively arbitrary choices about the sample that is used. I argue that the most cited result in the paper does not come from the most preferred econometric specification and that when the most preferred econometric specification is used G&M’s original paper found no statistically significant evidence that lowering the apportionment weight on payroll raises employment. Similarly, when I use this specification with data covering the period G&M studied (1978 to 1994), I find no statistically significant evidence for this hypothesis. When I modify the regression specification to separately include the payroll weight and the state corporate tax rate in addition to their product (i.e. state payroll burden), I find increased statistical significance when I use Huber/White standard errors. When standard errors are clustered by state, as is now common econometric practice, lowering the weight on payroll in states’ corporate income tax apportionment formulae has no statistically significant impact on manufacturing employment. I do a similar analysis using more recent data and obtain similar results. In summary, econometric evidence to support the hypothesis that changes in the payroll weight affected the distribution of manufacturing employment among U.S. states in the 1978 to 1994 period appears less strong than G&M asserted even when using G&M’s data and methods. More recent data also provide only weak econometric evidence in support of G&M’s main hypothesis.