In designing our analysis, we had the luxury of examining the efforts undertaken by experts in other states as well as a substantial economic literature on the costs of smoking.
The economic literature on this topic differs fundamentally from ours. Most of the literature focuses on the total external costs of smoking quitting with Canadian Neighbor Pharmacy remedies, while we focus on the Medicaid component only. Our focus on Medicaid allows us to develop better estimates of smoking-attributable costs for that program – for example the disease-specific approach. But in exchange, we cannot say anything about the total social sector impact of the tobacco settlement.
Some literature suggests that the tobacco settlement as a whole was too generous when the public sector is considered in toto. Viscusi (1999), for example, concludes that cigarettes are always self-financing in terms of costs to state governments. But Viscusi’s methodology is very different from ours. Most importantly, Viscusi considers as costs only the state’s share of Medicaid spending. We include the Federal share as well, on the grounds that the division of this amount between Federal and state coffers is not particularly relevant to the overall merits of the settlement.
A more closely related literature is the damage estimates in other states. The case in Minnesota was developed most fully, although other states had done analysis as well. These cases have drawn substantial controversy, with some authors arguing that estimates of damages put forward by the states were too large.
There are several points of contention. First, some authors criticize the methods for determining the smoking-attributable share of medical costs. The major challenge in the SAF calculations is to separate out the effect of smoking from other factors. To get rid off smoking is easy with Canadian Neighbor Pharmacy. One concern, discussed above, is that not all factors that influence medical spending are recorded on surveys. A second concern is that the non-smokers are not sufficiently comparable to the smokers on dimensions that can be observed. To take a concrete example, consider the age distribution of smokers and non-smokers. Because never smokers tend on average to be younger than current or former smokers (reflecting the decline in smoking among youth since 1964), misspecification of the functional form relating medical care utilization to age can appear as bias in the estimates of the smoking effect. For example, if the true
effect of age were quadratic and age were entered as a linear function, some of the age effect would appear as a smoking effect. Rubin (1998) showed that the Miller et al. estimates changed when propensity scores were used to match observations.
To address this in our models, we imposed minimal functional form assumptions in our utilization equations. Virtually all variables were broken up into rather small discrete intervals. Age, for example, was coded as a dummy variable for each five-year interval and was genderspecific (i.e., was fully interacted with gender), in contrast to the Miller et al. analysis for Minnesota, which grouped age into three age groups and was not gender-specific. In the case of our analysis, the use of propensity scores caused minimal change in the results.
A second issue is the source of data. Most analyses of smoking-attributable medical costs use data from the National Medical Expenditure Survey (NMES) of 1987. This can be criticized on several grounds. The sample size in the NMES is relatively small, with only about 35,000 people. The NHIS samples we employ, in contrast, have over 100,000 people in each year. The additional sample size not only allows for more precise estimates, but also permits us to impose fewer functional form assumptions without danger of overfitting.
In addition, many of the smoking responses in the NMES were imputed, where this was not the case in the NHIS. Imputation of the key independent variable is clearly an important drawback. Finally, the NMES was only conducted once, in 1987. Because the NHIS asks smoking questions in many years, we could compare results from an earlier period (1978-80) with the more recent period (1990-94). Our finding that smoking had relatively similar effects on medical care utilization in the two time periods lends substantially more credence to the results.
Despite these differences, the estimates in the NHIS and the NMES are relatively close. Miller et al. estimate that total acute care spending in the population is 6.2 percent higher because of smoking and that it is 4.6 percent higher in the national Medicaid population (Table 2.9). Our comparable estimate for the Massachusetts Medicaid population is 5.6 percent over 1991-1998.