Abstract
Background/Aims Clinical researchers, especially in the area of mental health, have noticed strength of some psychometric methods. Clinical researchers, often coming from a background of biostatistics or mathematical statistics, had little training in psychometrics using item response theory. In recent years, Food & Drug Administration’s requests on pharmaceutical companies to report item fit statistics have echoed such phenomenon.
Methods To demonstrate how item-specific type of analysis can expand our understanding in causes of depression, we use demographic variables to predict patients’ levels of depression through a traditional regression and a latent regression (i.e., an explanatory Item Response model).
Results of these two models on the same data are compared. When using the general regression model, level of depression is sum of 15 items from Short-Version Geriatric Depression Scale. When using the latent regression, a distribution of propensity for endorsing each GDS item is generated. As a result, some demographic variables prove to be no effective predictors for level of depression in the general regression model, but an effective predictor in the latent regression.
Discussion The current paper aims to explain how analyzing clinical data on item levels can contribute to our understanding in causes of depression, especially when clinical data contain information of specificity for different aspects of depression.
