Journal of Official Statistics, Vol.12, No.4, 1996. pp. 349–363
Correcting Unit Nonresponse via Response Modeling and Raking in the California Tobacco Survey
Charles C. Berry, Shirley W. Flatt, and John P. Pierce
Abstract:The California Tobacco Surveys are the principal data sources used to gather basic information about smoking behavior and to assess the California Tobacco Control Program – the largest program ever undertaken to reduce the impact of tobacco on society. These surveys use a two-phase design; an initial contact with a household – the ‘screener’ interview – gathers limited information about persons in the household, including their smoking statuses. Subsequent ‘extended’ interviews gather detailed information from all adult smokers and a sample of non-smokers. In 1992, unexpected results prompted investigation of unit nonresponse among these selected persons. Both nonresponse and smoking status vary according to education and sex, and nonresponse also depends on the age by smoking status cross-classification. A consequence of this and the sub-sampling of non-smokers is that the proportion of individuals interviewed in the ‘extended’ phase has a complicated pattern depending on age, education, sex, and smoking status. A generalized raking procedure was used to take these patterns into account and was implemented as a set of survey weights. This procedure seemed to reduce bias and variance in estimates of smoking prevalence and to slightly inflate the variance of estimates for variables which are not highly correlated with smoking status.
Keywords:Survey weights; auxiliary information; missing data; double sampling; generalized raking.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group