I sympathize with persons who have been taught that multiple regression is a means of “statistical control.” It took me longer than I care to think to see through the smoke and mirrors. Disbarring such terms would prevent much confusion. Disbarring such terms would focus concern on the central issue of causal analysis.
A second remedial measure would be to orient teaching toward development of extrastatistical judgment. Good choice of problem is the primary determinant of accomplishment. Extrastatistical judgment is needed to evaluate whether a given application of regression analysis is reasonable or ill-advised. Current texts concentrate on how to do regression analysis. Far more important is when—and when not (see also Note 16.2.2d).
Most texts do warn that substantive theory is generally prerequisite for process analysis with multiple regression. These warnings, however, are obscured under the mass of statistical detail, by such phrases as “statistical control, ” and especially by students' implicit assumption that what is being taught must be worth learning. At best, these warnings are little help because little is said about what constitutes adequate substantive theory.
One issue of substantive theory concerns the standard additive regression model. Experimental studies can assess the validity of this model, as illustrated in Chapter 20. This approach, however, is never considered in regression texts.
An alternative approach is to reorient learning–teaching toward the extrastatistical research judgment needed for effective use of multiple regression. To develop such judgment requires a collection of real-life applications, both well-taken and ill-taken, together with the extrastatistical considerations. The foregoing examples are a step in this direction, but systematic group effort is needed by experts in statistics and empirics of multiple regression.
Among texts on regression analysis, Pedhazur (1982) gives perhaps the best understanding of the pitfalls. Mosteller and Tukey (1977) is excellent though more advanced. Regression analysis by example (Chatterjee & Price, 1991) is a helpful guide to applications of regression analysis, emphasizing graphical methods with primary concern for revealing pattern in the data rather than significance tests. The exposition is built around numerous data sets, selected to illustrate particular techniques and problems.