|Saturday, February 22|
|CS23 Modeling Techniques||
Sat, Feb 22, 10:45 AM - 12:15 PM
Fractional polynomials: flexible, interpretable, and an alternative to splines. (302751)*Michael D. Regier, West Virginia University, Department of Biostatistics
Keywords: Fractional polynomials, splines, smoothing, interpretation, modern regression methods
Fractional polynomials are a powerful smoothing technique that has been implemented in statistical packages such as R and Stata, and has been applied to medical administrative data, force of infection in animal husbandry, and epidemiological exposure-response relationships. Research involving fractional polynomials continues to grow as there are many open questions concerning their properties and utility. In this presentation, we review fractional polynomials, their use, implementation and interpretation. We demonstrate this regression technique with an ecological public health data set and a biomarker dose-response data set. For both sets of data, we will compare fractional polynomials with splines, emphasizing the differences between the two approaches with respect to the modeling objectives. Finally, we will discuss the problem of model selection for fractional polynomials and compare current model selection techniques.