Duration analysis in addictive behaviors: Cigarette smoking, alcohol drinking and illegal drug us

This thesis takes survival analysis for its basic framework, uses it to examine data on substance use, and focuses on the duration of first and regular use. However, the standard duration model is not suitable because under the conventional assumption, all individuals will eventually "fail" (become a substance user). The split population duration model, on the other hand, assumes that there is a possibility that not all individuals will "fail;" therefore this model is more appropriate for the analysis of human addictive behaviors. Furthermore, it not only provides the information regarding whether decide to use but also regarding when they begin.;In the second chapter, a multivariate probit model is first used to investigate the correlation between the individuals' decisions of first drug use. Based on its results, I develop a multivariate split population duration model with frailties to study the hazards of first drug use. The results show a stronger correlation between an individual's decisions regarding whether to use than when to use.;The third chapter discusses the hazard of becoming a regular substance user, but the group under study consists of individuals who had used tobacco and/or alcohol at least once at the time of the interview. A model similar to the one used in the previous chapter is applied but is adjusted for the difference in the group being studied. Model diagnostics reveal that the split population lognormal duration model fits better. The results show that the estimated hazards of becoming a regular tobacco/alcohol user increase in the first two years after first use and decrease afterwards.;In the last chapter, I apply probit models to investigate the decisions regarding alcohol use within cohabiting families, by type of relationship and by number of members. The results show that family members' decisions are positively correlated with other members'. Next, I develop a split population duration model integrated with a sequential choice model to analyze the duration of alcohol drinking. Under this model each individual makes decisions sequentially, i.e., first whether and if so, when to use, and then whether and if so, when to quit. The findings show that the demographic variables are more strongly associated with individuals' decisions than the economic variable.