Comparing Multiple Regression Procedures and Chi Square in MMPI Item Analysis

Abstract

Multiple regression procedures and chi square were used to analyze the MMPI item for a group of 110 black and white male college students. Most of the items identified by the multiple regression procedures were also identified by chi square. The added advantage of having R^2 as an estimate of the percentage of variance accounted for by race is discussed in relation to selecting items in empirical scale construction.

The Minnesota Multiphasic Personality Inventory (HHPI) consists of four
validity scales and 10 clinical scales. In addition to these 14 scales more than 400 special scales have been developed from the KMPI pool of 550 items (Duckworth & Duckworth, 1975). The method of empirical scale construction has been used in the development of many of the original and special MMPI scales.

In the empirical approach, MMPl items responses of two distinct groups (criterion and control) are contrasted to insure that scale items are selected which are empirically related to the characteristic being assessed. Researchers will usually select items for inclusion in the new scale on the basis of the level of significance, usually .05, from the results of a chi square test. Once the items have been selected, several other steps are carried out which include comparing the means of the criterion and control groups. If the purpose of the new scale is to classify individuals into dichotomous groups, (i.e., black-white, alcoholic - non-alcoholic) cutting scores are also established. The point of this general description of scale construction is to show that little, if any, direct attention is given to assessing the strength of the relationship between the item results and the criterion.

Leitner (1979) has recommended multiple regression procedures in analyzing R x C contingency tables. Chi square is usually applied to these type of data. The multiple regression approach provides the researcher with an estimate of the percentage of variance (R^2) accounted for as a way to better understand the strength of the relationship implied. The purpose of this study was to examine the results from item analyses conducted on a single scale using both chi square and multiple regression procedures.