In previous talks, I have described reasons for rejecting standard analytic methods as inadequate for inference from observational data. Briefly, those reasons are the failure of standard methods to reflect any source of uncertainty other than "random error", and the failure of accompanying discussions to adequately capture other sources of uncertainty and their interactions. Methods for multiple-bias modeling provide alternatives that can capture and integrate major sources of uncertainty, thus providing better input to research planning and policy analysis. I have concluded that multiple-bias modeling should be integrated into the core training of anyone who will be entrusted with the analysis of observational data. This presentation reviews these arguments, and illustrates some mechanics of multiple-bias modeling in a pooled analysis of case-control studies of residential magnetic-field exposures and childhood leukemia. The results highlight the diminishing returns from studies conducted after the early 1990s, and suggest strategies for maximizing the value of further research.
This work is supported by the Electric Power Research Institute (EPRI).
