Fluctuating Parameters

Abstract

So far we have assumed that the underlying values of social policy variables are fixed; although sample values vary, the “true” values remain constant. This assumption has governed our selection of statistical inference methods. For many problems this assumption is unwarranted. Evaluation frequently involves monitoring the behaviors of large complex systems, such as workers compensation programs or nuclear power plants. The conditions of operation of these systems may alter in such a way as to change the outcomes with which evaluation is concerned. These changes may occur over time or place. In either case the “true effects,” about which evaluative inferences are to be drawn, may be fluctuating. Under this condition the methods of statistical inference presented in Chapters 5-8 are inappropriate. Rephrased in statistical terms, population parameters for these problems are changing; like sample statistics, these parameters can be thought of as having frequency distributions — the form in which inferences must be expressed.