Bivariate Density Estimation for Interval Censored Data from Complex Surveys

The nature of Statistics Canada's National Population Health Survey (NPHS) allows the use of event history analysis techniques to study relationships among events. The output from a series of health related questions to explore an association between pregnancy and smoking cessation was of especial interest. A methodology was proposed based on the notions of time order. In essence, the research team investigated whether one of the two events tended to precede the other closely in time. In this way, a causal interpretation of an association between these events would be more plausible. An important characteristic of the event times is that they are interval censored so that the exact times of occurrence were not known. Therefore the problem involved estimation of the joint density of the times to event taking into account the interval censoring and the complexities of the survey design. The result of the research was a new methodology for the analysis of interval censored data collected from complex surveys, which may have a variety of applications to research questions in the social sciences. Since the methodology was also extended to an application that considers an association between job loss and divorce using data from the Survey of Labour and Income Dynamics (SLID), the techniques that were developed may also be useful to other types of problems in the analysis of complex survey data. The team’s findings, which enable the study of relationships between important events, illustrated the value of complex surveys in general as well as the applicability of the survey's data in health and social program planning.

Faculty Supervisor:

Dr. Mary Thompson

Student:

Norberto Pantoja Galicia

Partner:

Statistics Canada

Discipline:

Statistics / Actuarial sciences

Sector:

University:

University of Waterloo

Program: