The demand for smokeless tobacco among male high school students in the United States: the impact of taxes, prices and policies

Abstract

Despite the deleterious effects of smokeless tobacco use, very little is known about the effects of tobacco control policies on smokeless tobacco demand. This paper uses data extracted from the 1995–2001 National Youth Risk Behavior Surveys (YRBS) augmented with tobacco taxes, prices and policies to estimate smokeless tobacco demand equations among male high school students. The estimates indicate that higher smokeless tobacco taxes would significantly reduce the number of male students who use smokeless tobacco and the number of days smokeless tobacco users use smokeless tobacco. Moreover, smokeless tobacco products and cigarettes were found to be economic complements in consumption.

Notes

¹ The use of a state tax as a percentage of wholesale price may pose a problem if smokeless tobacco manufacturers price discriminate by state at the wholesale level. Fortunately, research conducted on other tobacco products provides evidence against significant discrimination. Tennant (1950) was among the first to argue that cigarette wholesalers faced the same price, regardless of location. Keeler et al . (1996) empirically test for cigarette manufacturer price discrimination by state and find that state price discrimination is very small in magnitude compared to the final retail price. They further find that state cigarette taxes are slightly more than passed on to consumers (a 1 cent state tax increase results in a 1.11 cent price increase). Keeler et al . (1996) point out that cigarette price discrimination at the state level is even less likely to occur now than it was in the 1950s. They argue that there has been an increase in national chains that negotiate a single wholesale price for delivered cigarettes. They further point to the fact that interregional price discrimination on the part of manufacturers selling to wholesalers has been illegal since the Robinson-Patman Act was enacted in 1936.

² Where Y is the dependent variable, X is a matrix of explanatory variables, and ε is a random error vector with E(ε) = 0.

³ For a thorough discussion of prediction bias when employing a logarithmic transformed dependent variable see Mullahy, 1998; Manning, 1998; Manning and Mullahy, 2001; and Tauras, 2004.

⁴ If the raw-scale variance does not depend on the raw-scale prediction, λ₁ = 0, then use a Gaussian distribution; if the raw-scale variance is proportional to the raw-scale prediction, λ₁ = 1, then use the poisson distribution; if the raw-scale variance is quadratic in the raw-scale prediction, λ₁ = 2, then use the gamma distribution; if the raw-scale variance is cubic in the raw-scale prediction, λ₁ = 3, then use the inverse Gaussian distribution.

⁵Cohen (1988) suggests that correlation coefficients greater than 0.5 are strong in magnitude.

⁶ In addition to cigarette prices, we attempted to use cigarette excise taxes in the smokeless tobacco equations; however, Belsley, Ku, and Welsch (1980) collinearity diagnostics indicated that these estimates were confounded more by multicollinearity than were the models that used cigarette prices.