In this article, we ask how types and antitypes in Configural Frequency Analysis (CFA) can be explained at the level of effects in log-linear models. Two strategies are proposed for the search for the effects that cause types and antitypes to emerge. The first strategy is ascending and inclusive. Beginning with the CFA base model, effects are systematically added until the most parsimonious model is found that meets the following two conditions. The first condition is that all types and antitypes disappear. The second condition is that no new types and antitypes emerge. The second strategy is descending and exclusive. Beginning with the saturated model, effects are systematically excluded until (1) the effects are identified that would make all types and antitypes disappear, and (2) no new types and antitypes emerge. A first comparison of the two models suggests that the ascending, inclusive strategy yields more parsimonious models. However, results from the descending, exclusive strategy have a more clear-cut interpretation. Data examples are given that describe the development of severe domestic violence.
|Pages (from-to)||35 - 52|
|Journal||Bulletin de la Société des Sciences Médicales du Grand-Duché de Luxembourg|
|Publication status||Published - 1 Nov 2008|