The Project Packaging Method Is Used In The Potential Category/profile Analysis: A Simulation Study Of Feasibility And Influencing Factors

Latent Class Analysis(LCA)and Latent Profile Analysis(LPA)have become popular methods for modelling latent population heterogeneity in psychological,medical and management science.Due to the benefits of using item parcels as manifest variables in structural equation models,researchers use parcels in LCA/LPA models.However,there had been no study to explore the feasibility of using parcels as indicators in LCA/LPA models.Via applying Monte Carlo simulation methods,we generated artificial data to compare the performance of parcels with items under different LCA/LPA modeling conditions.Each simulation composes of these combinations: sample sizes,number of parcels and number of items in each parcel.Results found that latent class distance plays an important role in class enumeration when using parcels as latent variable indicators.When analyzing a unidimensional scale in LCA/LPA model which class distance is large,the class enumeration of parcels is consistent with the results of items.When the model analyzes more than one unidimensional scales with large class distance,the model evaluation is influenced by sample sizes,number of parcels and number of items in each parcel.Parcels perform well with more number of items in each parcel across sample size and number of parcels.In addition,parcels perform well with more number of parcels.However,when the latent class distance of the un-parcel model is small,making the right judgement of model evaluation is much smaller with parcels as indicators.Also,it is hard to achieve coincident results in consideration of sample sizes,number of parcels and number of items in each parcel.The simulation indicates that item parcels can be used under large latent class distance and a single unidimensional scale condition.When the model composes of more than one unidimensional scale with large latent class distance,the combinations of sample sizes,number of parcels and number of items in each parcel should be considered.It is unsuitable to use item parcels when the latent class distance of the un-parcel LCA model is small.