Title: The structured approach to competing hypotheses regarding an association over the life course.

Abstract:

Epidemiologists are often interested in different hypotheses for how an exposure measured repeatedly over the life course relates to later-life outcomes. Sometimes, expert knowledge or existing literature points to competing hypothesized associations for a particular exposure-outcome pair. The ‘structured approach’ to such hypotheses has been developed to allow these hypotheses to be compared, and to identify the hypothesis that is most supported by observed data.

This seminar will introduce various examples of hypothesized associations over the lifecourse. For instance, a critical period hypothesis proposes that the outcome depends on the exposure only during a defined period of the life course. Alternatively, an accumulation hypothesis proposes that it is the amount of exposure over the life course, regardless of timing, that affects the outcome.

All of these hypotheses can be written as simple linear models. Identifying the hypothesis that is most consistent with observed data can therefore be thought of as choosing the best-fitting model. Two statistical methods have been proposed for finding this model for repeatedly-measured exposures. The first compares the model to a saturated model using an F test, and is appropriate for binary exposures. A more advanced technique with wider applications, based on least angle regression (LARS), will be discussed.