The prediction of voluntary attrition of college freshmen : a test of certain hypotheses from industrial-organizational psychology

This study has tested, in the prediction of voluntary attrition in college student populations, several hypotheses derived from empirical research in the field of Industrial- Organizational Psychology. An ancillary purpose was the integration and synthesis made of the existing literature concerning the prediction of voluntary attrition in college student populations, and the prediction of voluntary turnover in industry. This review of research on turnover and attrition pointed out a number of striking similarities between the two domains. In both domains, there are serious and similar criterion definition problems which confound the comparison and interpretation of studies.

In both education and the industry, studies using the individual as the unit of analysis have been more common than studies which use the organization as the unit of analysis. Satisfaction (both overall and facet-specific) has been more often focused upon in industrial research on turnover. Industrial models of voluntary withdrawal have focused primarily upon motivational components, however, those models in education have focused upon the interaction of the individual with the institutional environment in determining voluntary withdrawal.

The hypotheses tested in this research were primarily concerned with the improvement of the prediction of attrition. Questionnaires, which acquired biographic and other data such as attitudes and aspirations, were administered to two groups of entering college freshmen. Students were followed over a one-year period to determine whether they had persisted into their sophomore year, and performance data (first semester college GPA) were also gathered. The questionnaire and performance data were then used to predict attrition. The data from the first group were used in a cross-validation procedure with the data from the second group, and a significant cross-validated validity coefficient was demonstrated.

It had been hypothesized that the expressed intention to drop out would better predict an attrition criterion than would a predictor set composed of biographical variables. No support was found for this hypothesis. It was also hypothesized that attrition/persistence would relate to the extent to which expectations concerning grades were met in the college environment. It was expected that those students with large negative discrepancies (the actual reception of low grades and the expectancy of high grades) would be more likely to drop out. This study revealed that the met expectation variables did indeed relate to the criterion. It was also shown that no apparent disadvantage accrued with the use of this "difference" variable compared with the simple use of the two variables from which it was computed in multiple regression analysis.

Personality variables have not consistently been shown to relate to voluntary withdrawal. As an explanation of this phenomenon, it was hypothesized that personality variables relate in a curvilinear fashion to a withdrawal criterion. Support was found for this hypothesis in that those characterized as extremely high or extremely low on the Short Dogmatism Scale were more likely to drop out. Curvilinear relationships with the criterion were not shown for several other "personality" variables which had been created for exploratory purposes.

Finally, it was hypothesized that the mean differences between the persisters and dropouts on a given predictor variable were not necessarily the same for groups at high and low performance levels. Support was found for this hypothesis. The implication of this finding is that the relationship of a predictor variable with the attrition criterion will be different at different performance levels. In a statistical sense, there may be said to be an "interaction" of performance level with certain predictors of attrition.

The support for the latter two hypotheses generated the conclusion that the prediction of attrition might be substantially improved if "complex" terms (i.e., interaction and curvilinear terms) were entered into the prediction equations of voluntary withdrawal.