Organizational Commitment, Job Involvement, and Turnover: A Substantive and Methodological Analysis

This study was conducted to examine the hypothesis that organizational commitment and job involvement interact in the prediction of turnover (Blau & Boal, 1987). Prior work in this area has not incorporated a sufficiently broad definition of commitment, has omitted relevant covariates, and has utilized inappropriate estimation procedures (ordinary least-squares regression [OLS]). The presence of a commitment–involvement interaction was tested in three estimation models with data obtained from 138 supervisors. Models estimated with OLS replicated prior work (Blau & Boal, 1989) irrespective of whether additional covariates were included. Identical models estimated with logistic regression provided no support for the presence of a commitment–involvement interaction. It is concluded that results obtained with linear techniques are a function of an inappropriate estimation procedure when the dependent variable is binary. The potential impact of the widespread use of linear estimation procedures in turnover research is discussed. Organizational commitment and job involvement have been major themes in the organizational literature, especially with regard to the prediction of organizational outcomes, such as turnover. In fact, it is the potential influence of these variables on turnover that represents a particular methodological challenge for this line of research, namely the estimation of statistical models with binary dependent variables. In this article, we report the results of a study that extends prior work on organizational commitment, job involvement, and turnover, and we illustrate how the use of inappropriate estimation procedures raises significant questions about the validity of earlier research. The focus of this study, therefore, was twofold: one substantive, the other methodological. The substantive research question considered was a test of Blau and Boal’s (1987) hypothesis that organizational commitment and job involvement interact to influence turnover. The methodological issue (that nonlinear logistic regression is the appropriate estimation procedure for models with dichotomous dependent variables) is applicable to turnover research generally Although this issue is discussed in many statistics texts and has largely been resolved in practice in other disciplines, industrial psychologists and organizational researchers seem to have been reluctant to adopt the appropriate procedures. In this article, we review these procedures and graphically illustrate how the choice of estimation technique can influence both the pattern and significance of results.