The importance of statistical modelling in clinical research : Comparing multidimensional Rasch-, structural equation and linear regression models for analyzing the depression of relatives of psychiatric patients.

BACKGROUND: Various studies have shown that caregiving relatives of schizophrenic patients are at risk of suffering from depression. These studies differ with respect to the applied statistical methods, which could influence the findings. Therefore, the present study analyzes to which extent different methods may cause differing results.

METHODS: The present study contrasts by means of one data set the results of three different modelling approaches, Rasch Modelling (RM), Structural Equation Modelling (SEM), and Linear Regression Modelling (LRM).

RESULTS: The results of the three models varied considerably, reflecting the different assumptions of the respective models.

CONCLUSIONS: Latent trait models (i. e., RM and SEM) generally provide more convincing results by correcting for measurement error and the RM specifically proves superior for it treats ordered categorical data most adequately.