ABSTRACT
Often researchers are faced with analyzing data from a biased sample that is unrepresentative of the population of interest. Studies that lead to samples that are biased-by-design are only useful if statisticians have a way to compensate for biased procedures that would result from na¨ıve use of such non-representative data. This chapter tells the story of how we encountered biased survival data in a major Canadian study of dementia in the elderly. We relate how we overcame the problem of bias and developed methods to answer questions about dementia. Although the story is woven around the Canadian Study of Health and Aging, there are many other areas in which the type of bias that we encountered also arises.
