If family utility was being maximized, and family members were better off with less smoking, then smoking would have already fallen. But this is a very different type of failure than that discussed earlier, so it is important to distinguish whether this is driving our results.
We investigate this possibility in two ways in Table 5. First in columns through, we separately examine the effect by marital status and gender. If our effects are due to individual internalities, there is no a priori reason to believe the self-control problem ought to be greater for any particular group (holding constant the predicted level of smoking). If, on the other hand, our effects were due to intra-family externalities, one would expect differences. Specifically, one would expect married people to show bigger effects since they are more likely to experience the externalities of smoking. Moreover, since men smoke more, wives should experience a bigger externality than husbands.
In Table 5, we therefore separately estimate our baseline model for married versus single people and then for four different groups: married men, married women, single men and single women. In the US data, there is some evidence that married people show a bigger effect, and that the effect is indeed largest for married women. In the Canadian data, however, the largest effect is for single men and the effects for singles are much larger than the effects for marrieds. Thus, the variation across groups seems essentially random and unrelated to the externalities story. in detail
In columns through, we examine this possibility in a different way. In the U.S. data, which has information on spousal education and labor supply, we estimate spousal predicted smoking as a function of the same set of covariates as above, but using the spouse’s education and labor supply in place of the respondent. This allows us to separately include the predicted smoking of the spouse and it’s interaction with the tax; unfortunately, this exercise cannot be carried out in Canada due to the paucity of information on spousal characteristics.
In fact, we do find some weak evidence for a role for spousal smoking in determining own happiness. The interaction of spousal predicted smoking and the tax is negative and marginally significant for all married persons; that is, married couples where the spouse is more likely to smoke are also made marginally better off by the tax. But the inclusion of this term has no effect on our key estimated interaction of respondent predicted smoking with the tax. Thus, within-family externalities appear unable to explain our results.
There is a further complication with interpreting these results as evidence for a time inconsistent model, however, which is much more fundamental. The key differential prediction between the time consistent and inconsistent models is over the immediate impact of taxation on the present discounted value sum of utility.