skip to main content skip to footer

The "Ruled Surface Regression " as a Starting Point in the Investigation of "Differential Predictability"

Saunders, David R.
Publication Year:
Report Number:
ETS Research Memorandum
Document Type:
Page Count:
Subject/Key Words:
Interest Inventories, Predictive Measurement, Predictor Variables, Regression (Statistics), Ruled Surface Regression


There are many examples of situations in which the predictive validity of some psychological measure varies systematically in accord with some other independent psychological variable. For example, the prediction of freshman engineering grades from appropriate scales of the Strong Interest Test is lower for groups thought to be compulsive than for groups thought to be non- compulsive (RM-52-13). Analysis of covariance provides a statistical method for studying situations in which the degree or mode of predictability is thought to vary as a function of membership in one or another of designated groups, which are presumed to be distinct and homogeneous. On the other hand, the "ruled surface regression" provides a simple generalization to the case in which the basic parameter is not membership in some group, but score on some continuous variable. The "compulsiveness" of the example cited above ought to be one illustration of such a continuous variable. The amount of prediction obtainable from the Strong Engineer Scale Score should vary continuously with the score on compulsiveness, and should not jump from one value to another at some arbitrary level of compulsiveness as required by the analysis of covariance. Evidence that this is indeed the case is provided.

Read More