Donivan J. Watley

The Effectiveness of Variables for Predicting Academic Achievement for Business Students

A PREVIOUS study by Watley and Martin (2) in vestigated the effectiveness of a large number of in tellectual and nonintellectual variables for predict ing academic achievement for male freshmen in a college of business administration. Five predictors were identified as the best combination of variables for this purpose. These were the mathematics and verbal scores of the Scholastic Aptitude Test (SAT) of the College Entrance Examination Board, the Re straint and Thoughtfulness trait scores of the Gu il ford-Zimmerman Temperament Survey (GZTS), and high school rank (HSR). The multiple c o r r e la tion coefficient between these five predictor varia bles and the criterion (first-year grades) was .82. Cross-validation of these results produced a validity coefficient of .52. The identification of personality traits as useful predictors of achievement is unusual. The inclusion of these two GZTS variables did, however, increase significantly the size of the obtained multiple corre lation. The purpose of this study was to analyze these data further by a method presented by Merwin (1). This method studies the relationship between each of these five predictors and success as a freshman business student. Increasing concern with the use of cutting scores and expectancy tables in student selection suggested analysis of the results using this technique. Scholastic ally successful students (C av erage or better) were compared with unsuc ees sful students (less than C average) on the basis of these five variables.

The Relationship of Parental Education to Achievement Test Performance of Girls vs. Boys

High scorers on the National Merit Scholarship Qualifying Test were crosscategorized by levels of fathers and mothers education. The relative probability of high achievement for males as compared with females of the same background is then indicated by the ratio of the number of males to the number of females in each category. This ratio was found to decrease with increasing fathers and mothers education; however, the association of fathers education with this ratio disappeared when mothers education was controlled. Examination of subtest scores revealed the same findings on the Mathematics Usage, Social Studies Reading, and Natural Sciences Reading subtests; no trends were observed on the English Usage and Word Usage subtests.

An Attempt to Improve Prediction of College Success by Adjusting For High School Characteristics1

Studies reported periodically over the past fifty years have been concerned with the relationship between high school characteristics and college achievement. The findings of much of this research have been markedly inconsistent. Although Pittinger (1917) reported that graduates of large high schools obtained the best grades in college, other studies (Seyler, 1939; and Saupe, 1941) have obtained results favoring the graduates of small schools. More recently, Hoyt (1959) found a trend for students from smaller high schools to earn lower college grades when grades were adjusted for high school rank. Watley (1964) assessed the effects of type (public or private), location, and size of high school in relation to academic achievement in an engineering college and found that predictive efficiency was improved using predictions computed specifically for graduates of large public high schools located in urban areas and for graduates of private high schools. Major efforts have been made recently to improve predictive efficiency from scaling methods designed to adjust high school grades on the basis of grades earned in college and then in turn correcting college grades on the basis of high school grades. Bloom and Peters (1961) reported that by adjusting both high school and college grades for institutional variation they were able to increase the over-all correlation between school and college grades from about .50 to .77. For a sample of 23 high schools they found a median within-school correlation of .54 for unscaled high school and college grades, and with scaled grades this was raised to .77. For 13 colleges the median within-college correlation of .57 for unscaled school and college grades was raised to .68 for adjusted grades. Although these findings are important and deserve close attention, several aspects of the design of the study make the

A Note on the Problem of Determining the Most Effective Predictor Variables in Multiple Regression

A PROBLEM that often troubles educators is de termining which variables most effectively predict a certain criterion such as grades. Oftentimes a number of predictor variables are considered and the primary aim is to select as few of these as pos sible for prediction purposes without losing too much predictive power. Several possible approach es to resolving such a problem do not always lead to the same solution. A convenient solution is sometimes found through logical analysis, while other times variables are selected that are most practically feasible. When the problem is consid ered statistically, it is a matter of determining the least number of predictor variables that account for most of the variance in the criterion variable. Fre quently this becomes a question of determining which combination of perhaps two or three predic tor variables appears to be the most useful combi nation of the original six or seven predictors con sidered. Although procedures for handling such a problem statistically are described in many statistics text books, several important points in relation to the application of these procedures seem to be in need of clarification. Specifically, application of such procedures can lead to conflicting results unless caution is exercised in defining the problem. The oretically the problem could be stated in a number of ways, but for illustration purposes this paper will concentrate upon only two of these. First, a solution might be sought by eliminating predictor variables from the total multiple corre lation coefficient for all variables included in the problem, with the aim being not to reduce the size of this R significantly (or the amount of variance accounted for). Those variables that remained af ter this elimination process would be used for pre diction purposes. A second solution to the same multiple regression problem could be found by start ing with a single variable and adding other variables that contribute significant amounts of variance until the best combination is identified. An example will illustrate the divergent outcomes that can be obtained due to the different definitions employed. In solving for these solutions, letus use a set of hypothetical multiple correlation coefficients for eight predictor variables as shown in Table 1.

Book Reviews : The Use of Academic Prediction Scales for Counseling and Selecting College Entrants by Benjamin S. Bloom and Frank R. Peters. New York: The Free Press of Glencoe, Inc., 1961. Pp. 145

very little systematized knowledge or progress worthy of expression. The Sixty-second Yearbook of the National Society for the Study of Education may never be regarded as &dquo;A Sure Guide to the Use of Tests,&dquo; as the editors for the Society (or Dilworth and Fenning) might imply; moreover, it does not present a particularly outstanding historical anthology of the forces and movements that have exerted powerful influences over school testing programs in our time. However, it does provide a convenient and authentic source of generalizations, recommendations, and suggestions that are based on the very best of current measurement theory and practice. There is no doubt that this volume will make its own impact and improvement, becoming a &dquo;pearl of great price&dquo; in the hands of those responsible for the future development and promotion of school testing programs. NORMAN C. MABERLY

Type, Location, and Size of High School and Prediction of Achievement in an Institute of Technology

ASSESSING the predictive efficiency of measures used in making educational decisions is crucial if counseling and selection procedures are to be effective. Many studies have been conducted at the University of Minnesota related to the development and evaluation of predictive measures used in the Institute of Technology (Berdie, 1944, 1951; Berdie and Sutter, 1950; Layton, 1951; Layton and Swanson, 1957; and Swanson and Berdie, 1961). The study by Swanson and Berdie (1961) presented predictive validities for a large number of variables. Of those variables investigated, the Institute of Technology Mathematics Test (ITMT) correlated highest with first-quarter grade averages (GPA), consistently around .60. In addition to the ITMT, percentile rank in high school graduating class (HSR) has also proven consistently to be among the most effective predictors of scholastic performance for IT students. Combined with the ITMT, it produces multiple correlations around .65. These are the two measures currently used by the Institute of Technology in determining admission. The present study investigated the differential predictability of the ITMT and HSR when type, location, and size of high school class were considered. The aim was to determine whether predictive efficiency can be increased by taking these three factors into account. With students appropriately classified into groups according to high school attended, the predictive power of each of these variables