Sampling weights are recommended to be incorporated in surveys to compensate for the disproportionality of the sample with respect to the target population of interest. This report presents how to develop sampling weights for a population-based study where a sample was randomly selected and demonstrates the process of developing such sampling weights. We exemplify the development of sampling weights with a real research project entitled School Physical Activity and Nutrition (SPAN) project. In this report, we first introduce the probability-based survey and related key concepts, such as sampling design, sampling frame and sampling weights. Then we discuss the sampling design and the construction of the sampling frame for the SPAN project. We next demonstrate the method and the process of developing the sampling weights for the SPAN project. Lastly, we present the results with an example. / text

In September 2002, the Canadian Institute of Health Information, Health Canada and the National Task Force on Health Information created the Canadian Community Health Survey (CCHS) whose objective was to determine the status of the Canadian health care system and the health status of Canadians themselves. The CCHS was divided into two cycles (Cycle 1.1 and Cycle 1.2) of which CCHS Cycle 1.1 was general population health survey designed to provide information for 136 health regions covering all provinces and territories. For this particular report, data from Cycle 1.1 was analyzed in order to determine if Work stress, Type of Smoker, Body Mass Index (BMI) and Household Income were significant factors contributing to the general health of the population of Ontario. Each of the above mentioned variables was considered individually for a given age, gender and marital status of the respondents. Logistic Stepwise Regression was used to determine if these variables were significant predictors of general health and all the possible two way interactions were explored at the five percent level. Furthermore, a diagnostic check of the fitted models was conducted and the validity of the models was assessed once again after removing the influential points. In addition, the linearity of all the continuous variables was tested in the logit models followed by a comparison of the fitted models using weights. Overall, it was found that all of the potential predictor variables of concern were significant predictors of general health. The majority of two way interaction terms were included in the fitted models, however, the overall fit of some models was found to be poor. In some cases, upon removing the influential points, the overall fit improved significantly, while for others, the fit did not improve by much. Also, it was found that Household Income and Work stress were not linear in the logit model while Age was found to be linear. Upon comparing models with and without sampling weights, it was found that the model that included sampling weights consisted of the same main effects as the model without sampling weights, along with additional interaction terms which were not present earlier. KEY WORDS: CCHS, Sampling Weights, Logit Models, Fitted Models, Interactions, Stepwise Regression / Thesis / Master of Science (MS)