Title: New resampling methods applied to latent class model fit assessment
Authors: Geert van Kollenburg - Tilburg University (Netherlands) [presenting]
Joris Mulder - Tilburg University (Netherlands)
Jeroen Vermunt - Tilburg University (Netherlands)
Abstract: The assessment of model fit is an important part of statistical analysis. The researchers interest may lie with specific aspects of a model, or in the global fit. Asymptotic $p-$values are not available for every conceivable statistic and even when they are available they may not be valid when sample sizes are not very large. To get more reliable $p-$values, researchers may resort to resampling methods. Some of these methods are time consuming, while others may provide $p-$values which are not uniform under the null-hypothesis. The most common resampling methods to test Latent Class model fit will be illustrated. A recently proposed calibration, the posterior predictive $p-$value, will be discussed. Finally a very fast resampling scheme is proposed where the statistics are based on data only, which requires that each model of interest is estimated only once.