An examination of the turnover literature reveals several variables which have statistically significant relationships to turnover, yet explain only a small amount of the variance. Since aggregating individual responses into groups actually changes the structure of the variance, a possible explanation of the weak results is inadequate grouping. Four conceptual models of turnover are replicated to provide a data base to use to examine the effects of groups. The models are an expectancy model, a role conflict/ambiguity model, an exchange/investment model, and a sociological model. Two theoretically relevant methods of grouping respondents (by psychological type as defined by the Myers-Briggs Type Indicator, and hierarchical level) are compared in this study by examining their effect on the results of tests of the four models. The subjects are professional staff accountants in a "Big 8" accounting firm. The results provide support for the existence of the effect of groups on the analysis of turnover data. Although the power of the design is low, all models except the role conflict/ ambiguity model had results that varied between grouping methods. The expectancy model showed the strongest effect. This indicates that exploring the existence of groups should be a part of the development of conceptual models of employee turnover. This study demonstrated the need for further research on the effect of groups on the structure of measured variance in empirical research in accounting. / Ph. D.