We therefore add to our regression specification these tax variables, as well as their own interactions with predicted smoking.
In column, we see the effect of the beer tax, in column 2, the effect of the gas tax, in column 3 the effect of the sales tax and in column 4, the effect of total state revenues. In all four cases and in both countries, we see that the inclusion of these variables does not much affect the initial estimate of the cigarette tax. Moreover, the new interaction terms with other taxes themselves are never negative and significant, although, for the U.S., there is a marginally significant negative effect of revenues per capita.
For Canada, the interactions with gas and sales taxes are actually positive and significant, suggesting that higher tax rates on those items raise unhappiness among predicted smokers. This may reflect the fact that these regressive taxes are targeted to those low income persons most likely to smoke. But, if anything, they suggest a bias against our finding for cigarette taxation. Thus there is little evidence that it is spending of tax money (rather than the tax itself) that is affecting smoker happiness.
Yet another possibility is that cigarette taxes are somehow spent differently than other kinds of taxes, so that there remains a happiness effect through the revenue side. We have investigated this possibility by gathering data on the composition of public spending over the 1977-1999 period, decomposing total spending into spending on: educational services, social services, transportation, public safety, environment and housing, government administration, utility expenditures and other spending. We then regressed each of these spending categories on the different taxes to determine whether the marginal effect of cigarette taxes was different than the other taxes we have studied. No significant pattern was found. This suggests that differential spending of cigarette tax revenues does not drive our results. this
These results so far are consistent with a time inconsistent model. But could they also be consistent with the time-consistent model? On the surface they are not, but with some reinterpretations they can be. One possibility is to argue that it is not smokers who are made happier but instead the spouses and relatives of smokers. Since our identification strategy compares predicted smokers to predicted non-smokers, our estimates would also include this externality effect if spouses and relatives have similar background characteristics. They would then also appear to be predicted smokers.
Of course, if higher taxes made family members better off, then this would indicate another potential failure of the standard model: imperfect family utility maximization. That is, by the same logic that shows that time consistent smokers cannot be made better off by a higher tax, families of smokers cannot be made better off by a higher tax if the smoker was maximizing family utility.