Improving predictive accuracy for placement in entry level college mathematics courses using available student information

This study was designed to provide more information on predicting mathematics achievement for entering, developmental-level students in a community college. Particular attention was paid to evaluating predictions in a way useful to a college for planning placement, advising, and orientation systems for entering students.;Logistic regression and chi square tests were applied to assess predictive effects and accuracy of variables HS GPA, ASSET placement tests, writing sample, total hours, age, marital status, and employment status on mathematics achievement. Subjects (N = 584) were new college students taking one of three developmental level mathematics classes from Fall 1994 through Spring 1995.;Results showed little predictive accuracy in current college placement practices. Several single variable predictors achieved significant levels but added little accuracy. A number of multivariate models emerged from the analysis which added accuracy to predictions of mathematics achievement. Overall, accuracy gained by using multivariate models was not large. The best models added 7.7% to 9.3% accuracy to blanket predictions that all students would receive a satisfactory grade. There was no consistent pattern to the variables in the multivariate models although HS GPA and placement test scores appeared frequently. ASSET mathematics placement tests appeared in a limited number of multivariate models. There were indications of instructor effects on mathematics achievement. Although there were no differences in rates of awarding satisfactory and unsatisfactory grades, single variable predictors derived for instructor subgroups differed among subgroups and differed from the group as a whole. The best derived multivariate models also showed differences in accuracy across different instructor subgroups.;A major advantage of this analysis was results which could be used to make "probability of passing" forecasts for students and advisors. In addition, focusing on accuracy of prediction enabled value judgments to be made about placement information procedures.