Beschreibung
The measurement of unobservable variables has gained increasing attention in marketing and social sciences. One important issue of contention in structural equation modeling is the detection of model (mis-)specification because there are contradictory results on how well model fit criteria capture model specification. In our simulation study, we examine how well currently recommended model fit criteria, construct validity criteria, and their combinations detect misspecification. We consider two types of misspecification simultaneously: misspecification severity and location. Furthermore, we include several facets of model size as distinct experimental factors. We show that fit criteria, construct validity criteria as well as recommended combinations of them only marginally cover model specification because they still include many misspecified models or drop too many correctly specified models. Therefore, we construct three new combinations of fit criteria and construct validity criteria by optimizing only the type I error, only the type II error, and both of them simultaneously. The resulting combination of the type I error optimization turns out to be superior in discriminating between correctly specified and misspecified models. The power to discriminate can be enhanced even further if it is combined with the result fromthe type II error optimization. This way, researchers can better classify their model correctly.| Zeitraum | 20 Juni 2019 → 22 Juni 2019 |
|---|---|
| Ereignistitel | INFORMS Marketing Science Conference |
| Veranstaltungstyp | Keine Angaben |
| Bekanntheitsgrad | International |