Application Research Of Models With Categorical Latent Variable In Heterogeneous Population

Objective In other fields of medicine, sociology and psychology, a largeheterogeneous Population divided into many small homogeneous subpopulation based onsome certain characteristics,implementing appropriate measures to deal with smallhomogeneous subpopulation to maximize the effect, which has an important significance inpractical applications. Sometimes, factors of population heterogeneity can be measureddirectly, and sometimes cannot be measured directly (ie, latent variable), and the need toreflect indirectly by measuring associated with a number of other items (ie, manifestvariables), the latent variable model can be measured. The purpose of this study was toexplore the latent categorical variable model in an invisible heterogeneous groupclassification.Methods For manifest continuous variables, the latent profile analysis can be used toclassify heterogeneous groups; for manifest categorical variables, we can use the latentclass analysis; if manifest variables have a multi-dimensional characteristics, we can use thelatent class factor model to classify the heterogeneity of groups from multiple dimensions.In this study, the latent profile analysis, latent class analysis and latent class factor modelapproach for model theory, parameter estimation, model evaluation, a posterioriclassification and graphical representation of in-depth discussions. Followed by datasimulation and case study methods, for continuous latent variable latent profile analysis andcluster analysis are compared on the effect of a heterogeneous group classification, and formanifest categorical variables with characteristics of more than one dimension, theapplication of the latent class analysis and latent class factor analysis and classificationresults effect are compared. The instance data are from Guangzhou residents’ healthknowledge survey and community health services satisfaction survey.Results Based on the parameter setting conditions of this study, data simulation resultsof the comparison of the latent profile analysis and cluster analysis suggest the average error rate latent profile analysis lower than the cluster analysis. when the original data classvariance is equal, variance between the specified class and inter-class variance rangingfrom latent profile analysis, the error rate quite; when the variance in the original dataclasses ranging, the specified class variance ranging from latent profile analysis, the errorrate is less than the specified class variance equal latent profile analysis. The instanceanalysis showed that latent class models can be clearly classified the crowd, however,cluster analysis results are poor.The comparison latent class analysis and latent class factor analysis of simulationresults, two-factor theoretical model of the two-level sampling conditions, the twodimensions of the original data is irrelevant or weakly correlated, latent class factor analysisto select the theoretical model as the optimal model the proportion of the proportion ofcorrect classification rate is higher than the latent class analysis to select the four categoriesof single-factor model; moderate when the two dimensions of the original data, withincreasing sample size, select two factor levels and relevant latent class factor modelgradually increased, and the correct classification rate is higher; when the two dimensionsare highly correlated, latent class factor models also tend to choose a factor model. Theinstance analysis results show that latent class factor analysis of heterogeneous populationsfrom multiple angles classification, the classification results are more accurate than thelatent class factor analysis provide more classified information, but can also explore theassociation between the factor.Conclusions Classification of latent profile analysis is superior to the system clustermethod, and it can be used as a powerful tool to solve classification problems of continuousvariables.Latent class analysis and latent class factor analysis can handle classification manifestvariable. The former classified observations from a single dimension, which combines theideas of factor analysis, classification from multi-dimensional observations.Comparing with latent class model, latent class factor models not only classifyheterogeneous crowd, but also investigate manifest variable potential factor structure, withcommon characteristics were degraded factor LCFA method also has the factor analysisfunctions and cluster analysis to classify observations have a wider variety of explanations,to provide a richer information than the latent class model, has a broader academic andapplication value.