Conferences

Recently, an alternative procedure has been suggested to evaluate structural equation models. This procedure was necessary because the popular procedures to evaluate model fit have a tendency to find very small and irrelevant misspecifications when a model under consideration is actually a very good model in terms of measurement error and explained variance. On the other hand, for not so good models, even large misspecifications go undetected. This is an awkward situation, good models are rejected (too often) while bad model are accepted (too of-ten). In order to change this situation we should control for type II errors. Recently Saris, Satorra, and Van der Veld (forthcoming) have suggested an alternative procedure, ‘the detection of misspecifications’, that will do just that. This procedure is quite laborious, even for fairly simple models. Therefore, we have developed a software program called JRule, which for each restricted model parameter judges whether it is [1] not misspecified (high power, nonsignificant MI), [2] misspecified (low power, significant MI), [3] maybe misspecified (high power, significant MI), or whether [4] we cannot say anything at all (low power, non-significant MI). JRule makes these judgements – 1, 2, 3, or 4 – based upon the output of standard SEM software. The advantage of JRule, is that we immediately look for misspecifica-tions in the model and also control for type I and type II errors. Moreover, in line with Steiger (1990), Browne & Cudeck (1993), and MacCallum, Browne & Sugawara (1996) we also believe that models are always misspecified to a certain level, that is why in JRule it is also possible to define the size of the misspecifications one believes is irrelevant for all practical pur-poses. This prevents models being rejected, due to very small (irrelevant) misspecifications. In this presentation we will shortly discuss the shortcomings of the popular global model evaluation procedures and introduce JRule as an alternative way to test structural equation models.