Abstract
We evaluate the empirical relevance of the Job Characteristics Model of Hackman and Oldham in the modern organizational environment using unique, nationally representative data from a survey of British establishments. The data contain information on a large number of establishments and multiple workers within each establishment. The results generally support the Job Characteristics Model's predictions that task variety and worker autonomy are positively associated with labour productivity and product quality and that autonomy is positively associated with worker satisfaction. In contrast to previous studies, we find the results for task variety are stronger for the performance-related outcomes than for worker satisfaction. The theoretically predicted moderating effect of context satisfaction is largely unsupported in the data.
Acknowledgements
The Department of Trade and Industry, the Economic and Social Research Council, the Advisory, Conciliation and Arbitration Service and the Policy Studies Institute are acknowledged as the originators of the Workplace Industrial Relations Survey data, and the Data Archive at the University of Essex as the distributor of the data. None of these organizations bears any responsibility for the authors' analysis and interpretations of the data. We are also grateful to Mike Gibbs, Martine Haas and an anonymous referee for helpful comments.
Notes
1 The WERS does contain objective information on absenteeism and turnover, although not in a form that is useful for testing the predictions of the JCM. The absenteeism information concerns absences resulting directly from an injury or from ill health as opposed to absences resulting from direct influences of core job characteristics on worker behaviour. Similarly, the information on turnover does not capture the voluntary turnover behaviour on the part of workers (i.e. quits) that is most relevant to the JCM.
2 Given the nature of the data, standard linear regression analysis would be inappropriate. A regression treats the difference between a ‘1’ and a ‘2’ in the dependent variable the same as the difference between a ‘2’ and a ‘3’, whereas in fact these are only labels reflecting a ranking. The linear regression cannot be used to produce the predicted probabilities that the outcome variable will take each of its particular discrete values.
3 This method is directly analogous to ‘interactive’ or ‘moderated’ regression, which is the standard way of testing for moderating effects in a linear regression model. In either case, the goal is to test whether the effects of interest are larger when allowing for the moderating effect than when not allowing for it. In the case of moderated regression, this is accomplished simply by observing the sign and statistical significance of the estimated coefficient on the interaction term. In the present context of ordered probit (which, unlike the linear regression model, is nonlinear in the parameters) we compare the changes in the predicted probabilities implied by Specification 2 (which includes the interaction terms) to those implied by Specification 1 (which omits the interaction terms). If the implied changes in the probabilities computed from a model that includes the moderator (i.e. Specification 2) are more supportive of the JCM than are those computed from a model that omits the moderator (i.e. Specification 1), then the variable moderates as predicted.
4 These results are available from the authors upon request.