Conferences

Since the early years of LISREL, the number of fit measures to evaluate global model fit has grown with an average rate of almost 1 per year. Nowadays, over 40 fit measures are avail-able (Marsh, Hau, Grayson, 2005) and when they are combined (Hu & Bentler, 1998) the ways to evaluate global model fit are even more numerous. All these possibilities would have never been investigated, if any (or a combination) of the 40 fit measures performed as required from a normal model evaluation procedure. That is, any model that is correct for all practical purposes should be accepted, while any model with large misspecifications should be rejected. Of course if we take sampling fluctuations into account, one can expect that errors (known as type I and type II errors) are made in this respect. This is no problem, as long as we are able to control these errors. The current fit measures have serious issues in controlling for both type I and type II errors. These issues with the current procedures have lead to the development of an alternative way to evaluate models: 'the detection of misspecifications’ (Saris, Satorra, Van der Veld, 2009). The procedure is build upon the traditional procedure that is used for model improvement, i.e. by means of the Modification Index (MI), the Expected Pa-rameter Change (EPC), and theoretical understanding. An EPC with a significant MI is usu-ally introduced in the model. if it makes theoretical sense. However, the Modification Index, suffers from a lack of control of type II errors. Therefore, if the power is very high, even irrelevant misspecifications could lead to a significant MI. Our new procedure, however, to 'detect misspecifications’ takes into account the power of the test. This procedure has been implemented in a software package, called JRule. This program makes the detection of misspecifications in any model very simple. In this session we will explain the shortcomings of the popular global model evaluation procedures and show that JRule does a better job. In addition, we present substantive research that used the JRule software to evaluate whether models are misspecified, i.e. whether theories under evaluations are correct for all practical purposes.