use of the MI and SEPC in that it becomes simpler to identify parameters that will improve the fit of the model and also have large values (indicating their importance). Kaplan ( 1990) reviewed two studies supporting this practice, and at this point this approach seems to have merit for organizational researchers.
A final caution regarding specification searches based on MIs is related to the problem of equivalent models. As described by Lee and Hershberger ( 1990), during a specification search it may happen that two or more fixed parameters have the same MI value. As such, if any of these parameters are freed, the resulting models would have equivalent fit. When this happens, existing theory and prior empirical results become even more important in deciding which parameter will be included. Also, researchers should report that this has occurred and discuss the interpretation of the alternative model not chosen.
The use of causal models in OB research provides a powerful tool for testing theory in applied settings. This chapter has documented the increased use of this methodology over a 20-year period. During this time, there has been an evolution in the analytical approaches taken. Initially, simple correlation-based procedures were employed; this was followed by the use of a regression-based strategy in the form of path analysis. Currently, latent variable causal models are popular, because of the numerous advantages they possess. As these changes have occurred, OB researchers have faced the challenge of staying informed about which techniques are most desirable and how these techniques can be most effectively implemented in their substantive areas. Based on the reviews, it could be argued that they have been successful.
However, there is also evidence that technical advancements in latent variable methodology will continue to challenge the skills of OB researchers. The second part of this chapter provided some guidance for how the use of this methodology can be improved in future applications. Suggestions were provided for several stages of the research process. Only time will tell if there is as much progress in the effective use of causal modeling methodology in the next 20 years as was true of the past 20 years. It is hoped that this chapter will help ensure that this happens.
The authors would like to acknowledge the helpful comments of Mark Gavin, Gina Medsker, and Peg Williams on earlier drafts of this chapter, and the assistance of Julia Huffer in its preparation.
Anderson J., & Gerbing D. ( 1988). "Structural equation modeling in practice: A review and recommended two-step approach". Psychological Bulletin, 103, 411-423.