The three methods share the same point of departure, namely that if the disorder is categorical, then each group has its own distribution (e.g. group-specific means or endorsement profiles, group-specific variances and covariance between symptoms). Meehl’s taxometrics limit the number of groups to two, a taxon and a complement group, and do not require many other assumptions to be met. Model-based clustering and LVMM can handle multiple groups but necessitate the choice of a specific distribution for the observed symptoms or items (e.g. multivariate normal distribution). For model-based clustering it is not necessary to specify how exactly the symptoms are related to the underlying disorder but for mixture modeling this is necessary.
