AN ANALYTICAL INVESTIGATION OF THE ROBUSTNESS OF THE RESTRICTION OF RANGE CORRECTION PROCEDURE.
Item
-
Title
-
AN ANALYTICAL INVESTIGATION OF THE ROBUSTNESS OF THE RESTRICTION OF RANGE CORRECTION PROCEDURE.
-
Identifier
-
AAI8515627
-
identifier
-
8515627
-
Creator
-
FLEISCHMAN, LYNN ELLEN.
-
Contributor
-
Alan L. Gross
-
Date
-
1985
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Psychology, Psychometrics
-
Abstract
-
The use of test scores for selection purposes is continually under legal scrutiny. An organization must be able to substantiate the validity of the test. Often this task is complicated by the problem of missing data, i.e., whereas test scores (x) are available for all applicants, criterion measures (y) are available only for selected cases. There is currently a statistical procedure that supposedly "corrects" for restriction of range, that is, it estimates the x-y correlation for total population. The correction for restriction of range yields a statistical estimate of the unrestricted correlation coefficient.;It is often assumed that this adjustment procedure is effective, i.e., will produce an improved estimate. However, this result is highly dependent upon a set of strong assumptions (linearity, homoscedasticity, and selection on the predictor alone). In practice, these assumptions are often violated. Furthermore, empirical research has shown that departures from these assumptions can lead to significant errors in estimating the unrestricted population correlation.;The primary goal of this research was to investigate analytically the robustness of the restriction of range correction procedure to violations in the assumption of linearity. This analytical investigation derived expressions for the bias, standard error, and expected mean square error of the squared correlation in the selected group and the squared corrected correlation where the regression of the criterion on the predictor is both linear and nonlinear.;The findings of the present investigation suggest that the correction formula is of limited value when sampling variability and violations of the linearity assumptions are considered. Only under certain conditions is it advantageous to correct for restriction of range. These cases occur for large sample sizes, liberal selection strategies, and high x-y relationships in the total group. Recommendations for both researchers and practitioners are discussed. In addition, potential areas for future research are suggested.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
-
Program
-
Educational Psychology
