The Application of Blocking in Regression Analysis

摘要

Occasionally the prediction equation obtained by conventional regression techniques is an unsatisfactory predictor because of its behavior over segments of the range of the independent variable(s). For such situations, a procedure is illustrated which has been found to yield a 'better fit' than that obtained by conventional regression analysis. The procedure consists of segmenting the levels of the independent variable(s) into blocks and separately fitting each block. The separate fits, however, are obtained simultaneously and the end result is a single prediction equation. Numerical examples are given typifying regression analysis problems encountered in which the proposed procedure yields a 'better fit'. In each example, the proposed procedure of blocking in regression analysis is compared with conventional regression analysis. Extensions in the application of blocking in prediction problems and in comparative problems are briefly discussed. (Author)