Model Interpretation from the Additive Elements of the Likelihood Function

The interpretation of a fitted statistical model such as the classical linear or the generalized linear model Is substantially clarified by a full partitioning of the maximized log‐likelihood ratio test statistic Into additive elements. This method generalizes the regression elements of Newton and Spurrell (1967) used to aid interpretation of regression equations. The primary elements measure the unique contribution of each explanatory variable whereas the secondary and higher order elements measure the effective balance In the observed design. It Is remarked herein that the elements correspond to the parameters of a saturated factorial model fitted to the likelihood function. This permits a coherent computational procedure. Examples are taken from some well‐analysed data sets Illustrating the interpretation of regression and log‐linear models.