An Application of Unconditional Quantile Regression to Cigarette Taxes

Abstract

This study investigates heterogeneous response to state cigarette tax increases using unconditional quantile regression (UQR). We make two contributions to the empirical policy analysis literature. First, we argue that UQR provides more policy‐relevant information than conventional quantile regression in most empirical state policy analyses. Second, we document cigarette tax elasticity across a sample of adult smokers in the Current Population Survey Tobacco Use Supplements between 1992 and 2011. Our ordinary least squares regression show an imprecise negative relationship between taxes and cigarettes smoked in the past 30 days, while UQR reveals a U‐shaped relationship: Only moderate smokers reduce their smoking following a cigarette tax increase, and the magnitude of the effect is small. A $1.00 (135 percent) increase in the cigarette tax leads to a 3.5 percent reduction in the number of cigarettes smoked in the past 30 days among the most responsive smokers (implied tax elasticity = −0.03).

Heterogeneity in Smokers' Responses to Tobacco Control Policies

“…Two exceptions are Laporte et al (2010), who use quantile regression and panel data from Canada to estimate a rational addiction model of smoking, and Maclean et al (2014). Maclean et al (2014) estimate smokers' responses to tobacco control policies using data from the Tobacco Use Supplement to the Current Population Survey (CPS-TUS). Using unconditional quantile regression, they find that smokers in the middle of the smoking distribution, smoking between 10 and 20 cigarettes, are the most responsive to cigarette excise taxes, and they find little evidence that smoking bans reduce cigarette consumption at any part of the smoking distribution.…”

“…Maclean et al (2014) estimate smokers' responses to tobacco control policies using data from the Tobacco Use Supplement to the Current Population Survey (CPS-TUS). Although Maclean et al (2014) find statistically significant effects of cigarette excise taxes on cigarette demand, their estimated tax elasticities of demand are very small and not economically significant. Although Maclean et al (2014) find statistically significant effects of cigarette excise taxes on cigarette demand, their estimated tax elasticities of demand are very small and not economically significant.…”

Ethnic diversity in policing: An application of quantile regression to the New York City Police Department

“…In addition, the structure of the dataset affects the number of distinct quantile solutions (e.g., Davino, Furno, & Vistocco, 2014). A second drawback of QRMs is that the medians may be affected by atypical cases Maclean et al (2014) found that the conditional QR (CQR) model produced inconsistent estimates at the lower quantiles when examining the impact of cigarette taxes on smoking consumption. Another weakness of QRMs is that the interpretation of the coefficients changes when different sets of predictors (or covariates) are included in the model (Maclean et al, 2014).…”

“…A second drawback of QRMs is that the medians may be affected by atypical cases Maclean et al (2014) found that the conditional QR (CQR) model produced inconsistent estimates at the lower quantiles when examining the impact of cigarette taxes on smoking consumption. Another weakness of QRMs is that the interpretation of the coefficients changes when different sets of predictors (or covariates) are included in the model (Maclean et al, 2014). The interpretation of the coefficients also becomes more difficult when data transformations are performed on the raw scores (e.g., Hao & Naiman, 2007).…”

Education Curriculum and Student Achievement: Theory and Evidence

“…To address this concern, we also use the Recentered Influence Function (RIF) method recently developed by Firpo et al (2009), which explicitly relaxes the common distribution assumption. 2 More specifically, when the observed outcomes (in this case, test scores) vary monotonically with the unobserved variable (in this case, student preparation), RIF for the τ th quantile as: For our analysis, instead of examining students at the same quantile across states and years (as in the QDiD case), the RIF method compares students with the same test score and hence located at potentially different quantiles of the distributions across states 2 Given its flexibility, the RIF method has recently been applied to analyze a range of issues such as cigarette taxes (Maclean et al, 2014) and child care (Havnes and Mogstad, 2015).…”