Latent class analysis was accurate but sensitive in data simulations

AbstractObjectivesLatent class methods are increasingly being used in analysis of developmental trajectories. A recent simulation study by Twisk and Hoekstra (2012) suggested caution in use of these methods because they failed to accurately identify developmental patterns that had been artificially imposed on a real data set. This article tests whether existing developmental patterns within the data set used might have obscured the imposed patterns.Study Design and SettingData were simulated to match the latent class pattern in the previous article, but with varying levels of randomly generated variance, rather than variance carried over from a real data set. Latent class analysis (LCA) was then used to see if the latent class structure could be accurately identified.ResultsLCA performed very well at identifying the simulated latent class structure, even when the level of variance was similar to that reported in the previous study, although misclassification began to be more problematic with considerably higher levels of variance.ConclusionThe failure of LCA to replicate the imposed patterns in the previous study may have been because it was sensitive enough to detect residual patterns of population heterogeneity within the altered data. LCA performs well at classifying developmental trajectories