The first model we fit is an unconditional model, which is a unidimensional factor model without predictors. Unlike traditional factor analysis, the relationships between the factor and the items are not necessarily linear and depend on the scales of the indicators. For instance, for the binary symptom indicators in our example, we specified a logistic function, so that the probability of endorsing any given symptom is bounded between zero and one. The latent variable is viewed as a common cause of all of the symptoms (i.e., someone high in latent depression is more likely to endorse multiple symptoms compared to someone low in depression) and is assumed to account for all associations among the symptom indicators. This is a standard assumption of many latent variable models and is often referred to as local independence. Additionally, because depression is a latent variable, we must set its scale, and we do so by setting the mean to zero and variance to one. Thus the factor scores we generate from this model are on a standard normal metric (see sidebar 2). The resulting
