Sankhya: The Indian Journal of Statistics

1999, Volume 61, Series B, Pt. 1, pp. 91--105

AN ANALYSIS OF SURVEY DATA ON SMOKING USING PROPENSITY SCORES

By

MICHAEL D. LARSEN Harvard University, Cambridge

SUMMARY.Responses to questions on perceived health risks from smoking, amount of exercise, overall health, and desire to quit smoking vary across smoking groups. The smoking groups also have very different distributions of covariates. Categorizing respondents based on estimated propensities to smoke, as in blocking, creates groups with less measurable imbalance in distributions of covariates. Adjustment of results using propensity score classes lessens some effect estimates while increasing others and changes results of some significance tests. However, even after adjusting for covariates through the use of propensity scores, smokers report a higher risk of disease than do nonsmokers.

AMS (1991) subject classification.62D05, 62P99.

Key words and phrases. Logistic regression, subclassification,blocking, observational study.

Full paper (PDF)