Zero observations and gender differences in cigarette consumption

Abstract

Censoring mechanisms and gender differences are investigated for cigarette consumption by individuals in the USA. The Gaussian single-hurdle model is proposed which generalizes the specifications of Cragg (1971) and Heckman (1979) and allows examination of the empirical relevance of the two censoring mechanisms in the existing double-hurdle model. The proposed model performs better than Cragg's and Heckman's models but not as well as the double-hurdle model and also produces different elasticity estimates. The hypothesis of equal consumption parameters is rejected and demand elasticities found to differ between men and women. Income does not play a role and age has conflicting effects on the probability and level of cigarette smoking. Older individuals are less likely to smoke but, conditional on smoking, consume more cigarettes than their younger counterparts. Education has negative effects on the probability and level of smoking and can be an effective tool to curtail cigarette smoking.

Acknowledgements

Research for this study was funded in part by US Department of Agriculture's Economic Research Service under Cooperative Agreements No. 43-3AEM-2-80063 and No. 43-3AEM-4-80052. The author benefitted from discussions with J. Scott Shonkwiler and comments by seminar participants at Academia Sinica, Rutgers University and Pennsylvania State University at an early stage of the study. The usual disclaimer applies.

Notes

 Failure to truncate the error distribution may have contributed to the frequently-reported empirical difficulty with the SS. Another approach to accommodating positive values in the dependent variable for the ‘selected’ observations is through transformation of the dependent variable (Cragg, 1971, Equations 12 and 13). The logarithmically transformed SS, for instance, is widely used in health-care demand modelling (see Jones, 2000).

 Elasticity formulae are available upon request.

 Elasticities from the pooled sample show less notable differences between the SH and DH but are very different from the segmented-sample estimates reported. All pooled-sample estimates are available upon request.

 Further studies might consider the potential endogeneity of body mass in cigarette smoking.