Replication of Goolsbee, Lovenheim and Slemrod's 'Playing with Fire : Cigarettes, Taxes and Competition from the Internet' (American Economic Journal: Economic Policy, 2010)

This study replicates the empirical findings of Goolsbee, Lovenheim and Slemrod (2010) ("GLS") and performs a variety of robustness checks. Using taxable cigarette consumption, real cigarette excise tax rates, wholesale cigarette prices, per capita income, and other state-level data for the period 1980-2005, GLS report that rising Internet penetration in the presence of cigarette taxes has a significant causal effect on the elasticity of demand for taxable cigarettes. I am able to exactly replicate GLS's findings. My robustness checks consist of three parts. First, I check sensitivity to the removal of certain outlier cohorts of states from the data set. Second, I use population unweighted state-year observations in place of GLS's population-weighted observations. Third, I probe the robustness of GLS's key interaction term (Internet penetration-cigarette taxes) by (i) adding Internet penetration interaction terms to all main effects in the model and (ii) performing an orthogonalization procedure (Balli and Sørensen, 2013) to prevent the estimate of the key interaction term from picking up the effect of the interaction of the included variables with cigarette taxes due to their correlation with Internet penetration. I found low sensitivity to the first two robustness checks. On the third, GLS's estimate was robust to the inclusion of additional Internet interaction terms but was not robust to the orthogonalization procedure. This raises the possibility that the effect identified by GLS is an artifact of spurious correlation between Internet penetration and cigarette taxes over time. In sum, GLS's data may have insufficient power to identify the stand-alone effect of Internet penetration on cigarette tax-sales elasticities in a fully-interacted model