Psychometrics and latent class model selection in complex public health survey data

Wide-spread availability of large public health survey data has been a boon to funders and researchers. Funders get more mileage out of the data they pay to collect. Researchers get access to samples previously unavailable to most. However, use of these data also raise analytic challenges and complexities. Psychometrics, essentially the study of the measurement of psychological constructs, developed as part of psychology, and was typically based upon small convenience samples. At it's foundation are measurement error modeling and the use of multiple indicators. Public health survey designers, on the other hand, typically avoid asking the same question in multiple ways in order to save money and reduce respondent burden. Other standard practices in public health surveys, such as filter items and data editing, can also increase the challenges encountered in psychometric modeling. Traditional psychometric models dealt with continuous latent variables, but increasingly latent class and mixture models are also being discussed as measurement models. Large public health surveys afford the capability to fit these sophisticated models, but difficulties are also often encountered. Challenges to the use of these models will be discussed.