This is a supplemental application to significantly expand the scope our currently funded study of schooling and smoking. In the currently funded parent grant, we are investigating the relationship between schooling and smoking as a case study to shed light on more general questions about the links between schooling and health. One causal pathway is through health information, where general schooling helps consumers respond to new information about the health consequences of behaviors such as smoking. Alternatively, the association between schooling and health may not be causal, but may instead reflect hard-to- observe differences between people with different levels of schooling. In the parent grant's Study 2 we are using the econometric method of instrumental variables (IV) to identify the causal impact of schooling on smoking. The overarching goal of this supplemental/revision application is to enhance the econometric analysis in the parent grant's Study 2. The goal is motivated by recent simulation studies that show that failing to account for inherent nonlinearity in regression contexts involving endogeneity can result in substantially biased estimates of causal effects. To meet its goal, the work proposed has three specific aims.
Specific Aim 1 is to develop a robust conditional mean regression model of the duration of smoking among former smokers to include ordinal endogenous schooling dummies. This model will extend the econometric methods being used in the currently funded grant in several ways. We will implement this estimator in the Stata 10 programming language and disseminate the new software.
Specific Aim 2 is to further develop the model of the duration of smoking among ever smokers. In data on ever smokers, it is necessary to correct for bias due to right censoring of observed durations. About half of ever smokers are former smokers, and the other half are current smokers who by definition continue to smoke beyond the time of the interview so their durations are right censored. We will extend the model and method developed for Specific Aim 1 to account for potential bias due to the presence of right-censored duration observations in the analysis sample (the current smokers). Here, as under Specific Aim 1, corresponding Stata 10 software will be produced and made publicly available.
Specific Aim 3 is to evaluate the performance of the new estimators to be developed under Specific Aims 1 and 2, using simulated and real data.
The proposed study focuses on the link between schooling and smoking. Smoking contributes to more than 400,000 deaths annually in the US, and has been called the leading preventable cause of death. If we find more evidence of a causal link between schooling and smoking, we will strengthen the case that investments in schooling also improve public health. ? ? ? ?
Bar, Haim Y; Lillard, Dean R (2012) Accounting for heaping in retrospectively reported event data - a mixture-model approach. Stat Med 31:3347-65 |
Cheng, Kai-Wen; Kenkel, Don S (2010) U.S. Cigarette Demand: 1944-2004. B E J Econom Anal Policy 10: |