Dealing with multiple local modalities in latent class profile analysis

Authors:
Hsiu-Ching Chang;Hwan Chung
Affiliations:
-;-
Venue:
Computational Statistics & Data Analysis
Year:
2013

Citing 8
Cited 0

Deterministic annealing EM algorithm

Neural Networks
Choosing starting values for the EM algorithm for getting the highest likelihood in multivariate Gaussian mixture models

Computational Statistics & Data Analysis
Learning mixture models via component-wise parameter smoothing

Computational Statistics & Data Analysis
Monte Carlo Statistical Methods

Monte Carlo Statistical Methods
Inference in Hidden Markov Models

Inference in Hidden Markov Models
Root selection in normal mixture models

Computational Statistics & Data Analysis
Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images

IEEE Transactions on Pattern Analysis and Machine Intelligence
Bayesian approaches to the model selection problem in the analysis of latent stage-sequential process

Computational Statistics & Data Analysis

Quantified Score

Hi-index	0.03

Visualization

Abstract

Parameters for latent class profile analysis (LCPA) are easily estimated by maximum likelihood via the EM algorithm or Bayesian method via Markov chain Monte Carlo. However, the local maximum problem is a long-standing issue in any hill-climbing optimization technique for the LCPA model. To deal with multiple local modalities, two probabilistic optimization techniques using the deterministic annealing framework are proposed. The deterministic annealing approaches are implemented with an efficient recursive formula in the step for the parameter update. The proposed methods are applied to the data from the National Longitudinal Survey of Youth 1997 (NLSY97), a survey that explores the transition from school to work and from adolescence to adulthood in the United States.