Using the Latent GOLD Syntax for Generalized Latent Variable Modeling

Pre-conference workshop with Jeroen Vermunt, Department of Methodology and Statistics, Tilburg University, The Netherlands

20th July, 9.00 - 13.00, Castlereagh Room, Fisher Building

Jeroen VermuntLatent GOLD was originally developed as a simple latent class and finite mixture modeling program that would be easy to use for applied researchers. Over time additional functionalities and new models were included, such as mixture regression models (version 2.0), latent class discrete choice models (Latent GOLD Choice module, version 3.0), and models with continuous latent variables, multilevel latent class models, and complex sampling features (version 4.0). These features were always implemented using the point-and-click graphical user interface.

The main change in version 4.5 is that it contains a syntax module which eliminates certain limitations of the point-and-click GUI modules. The syntax offers a flexible system for generalized latent variable modeling comparable to that described in the book by Skrondal and Rabe-Hesketh. Models can be defined with latent variables of different scale types (nominal, ordinal, and/or continuous), with latent variables at multiple levels of a hierarchical structure, and with latent variables serving different purposes (true scores, latent traits, clusters, random effects, known classes, etc.). In addition to a flexible modeling framework, the syntax contains features such as simulation, multiple imputation, and resampling (jackknife and bootstrap) options, as well as facilities for using parameter estimates obtained with a training sample to obtain predictions for new samples.

The aim of this workshop is to introduce the Latent GOLD Syntax system. You will learn how to deal with different data structures (long files, wide files, and combinations of these), how to define the latent, dependent and independent variables to be included in the model, how set up the model equations, how to impose parameter restrictions, and how to use the various technical, output, and Monte Carlo options. As examples, I will use simple IRT and latent class models, as well as complex latent Markov and multilevel latent class models.