Model, Choice of

Abstract

The choice of which statistical model to use in a given application is a complex matter, which draws not only on the technical expertise of the statistician but also on the nature of the application and the purpose of the analysis. Different purposes may require quite different models to be applied to the same dataset. The role of modeling in clinical trials, epidemiology, and general biostatistics is discussed. The relative contribution of expertise both in the area of application and in the relevant statistical issues is explored. The use of sensitivity analyses to ensure that substantive conclusions are not heavily dependent on model assumptions is stressed. A strong theme throughout the article is that the judgment of the biostatistician plays an inescapable role in an activity, which nonstatisticians too often see as wholly objective.

Prediction of gestational weight gain – a biopsychosocial model

“…Simultaneous entry regression was used for the final multivariable models, as the biopsychosocial model calls for biological, psychological and social variables to be examined simultaneously. Final model selection was guided by the authors' knowledge of the relevant literature, the conceptual framework and the exploratory findings of the stepwise regression models (45,46) . Only statistically significant variables (P < 0•05) were included in the final models.…”