Wednesday, May 29, 2019
Pacific D/L (Hyatt Regency San Francisco)
* noted as presenting author
The present model aims to elucidate the association between individual-level variables and contextual predictors.
This model is best suited when the hypothesis predicts that a given association will vary across different groups defined by ethnicity, gender, age, educational and workplace roles. The model has been derived to identify these cross-sectional clusters of demographic attributes that moderate the effects of contextual predictors on individual-level characteristics.
In this study, we model the relationship between contextual predictor variables and outcomes by conditioning this relationship on latent demographic classes revealed by the analysis. In this model, the latent classes (demographic profiles) will determine the parameters of a logistic regression from the contextual variables to the outcomes. We refer to this type of model as a conditional latent class regression (CLCR). Its parameters can be fit to the data using maximum conditional likelihood estimation (MCLE).
As far as we know, no commonly used statistical model captures the type of interaction we hypothesize. We propose a new model to capture the class-conditional association between predictors and outcomes and discover the latent classes along the way. In the proposed model, which we call conditional latent class regression (CLCR), the latent classes determine the parameters of a logistic regression from the contextual variables to the outcomes. This model is a straightforward combination of logistic regression and latent class regression.