Computational Tools for Bayesian Inference for Finite Mixtures Models Under Model Specification Uncertainty

2006 ◽  
pp. 125-167
Keyword(s):  
Bayesian Inference ◽  
Model Specification ◽  
Finite Mixtures ◽  
Computational Tools

scholarly journals Bayesian inference for finite mixtures of univariate and multivariate skew-normal and skew-t distributions

Biostatistics ◽  
2010 ◽  
Vol 11 (2) ◽  
pp. 317-336 ◽  
Author(s):  
Sylvia Frühwirth-Schnatter ◽  
Saumyadipta Pyne
Keyword(s):  
Bayesian Inference ◽  
Mixture Models ◽  
Random Effects ◽  
Data Augmentation ◽  
Heterogeneous Data ◽  
High Dimensional ◽  
Finite Mixtures ◽  
Truncated Normal ◽  
T Distribution ◽  
Skew Normal

Abstract Skew-normal and skew-t distributions have proved to be useful for capturing skewness and kurtosis in data directly without transformation. Recently, finite mixtures of such distributions have been considered as a more general tool for handling heterogeneous data involving asymmetric behaviors across subpopulations. We consider such mixture models for both univariate as well as multivariate data. This allows robust modeling of high-dimensional multimodal and asymmetric data generated by popular biotechnological platforms such as flow cytometry. We develop Bayesian inference based on data augmentation and Markov chain Monte Carlo (MCMC) sampling. In addition to the latent allocations, data augmentation is based on a stochastic representation of the skew-normal distribution in terms of a random-effects model with truncated normal random effects. For finite mixtures of skew normals, this leads to a Gibbs sampling scheme that draws from standard densities only. This MCMC scheme is extended to mixtures of skew-t distributions based on representing the skew-t distribution as a scale mixture of skew normals. As an important application of our new method, we demonstrate how it provides a new computational framework for automated analysis of high-dimensional flow cytometric data. Using multivariate skew-normal and skew-t mixture models, we could model non-Gaussian cell populations rigorously and directly without transformation or projection to lower dimensions.

scholarly journals Computing Bayes factors to measure evidence from experiments: An extension of the BIC approximation

2018 ◽  
Vol 55 (1) ◽  
pp. 31-43 ◽  
Author(s):  
Thomas J. Faulkenberry
Keyword(s):  
Bayesian Inference ◽  
Closed Form ◽  
Analysis Of Variance ◽  
Degrees Of Freedom ◽  
Bayes Factor ◽  
T Test ◽  
Bayes Factors ◽  
Computational Tools ◽  
Testing Hypotheses

Summary Bayesian inference affords scientists powerful tools for testing hypotheses. One of these tools is the Bayes factor, which indexes the extent to which support for one hypothesis over another is updated after seeing the data. Part of the hesitance to adopt this approach may stem from an unfamiliarity with the computational tools necessary for computing Bayes factors. Previous work has shown that closed-form approximations of Bayes factors are relatively easy to obtain for between-groups methods, such as an analysis of variance or t-test. In this paper, I extend this approximation to develop a formula for the Bayes factor that directly uses information that is typically reported for ANOVAs (e.g., the F ratio and degrees of freedom). After giving two examples of its use, I report the results of simulations which show that even with minimal input, this approximate Bayes factor produces similar results to existing software solutions.

Forced Zero Cross-Loading Misspecifications in Measurement Component of Structural Equation Models

Methodology ◽  
2014 ◽  
Vol 10 (4) ◽  
pp. 138-152 ◽  
Author(s):  
Hsien-Yuan Hsu ◽  
Susan Troncoso Skidmore ◽  
Yan Li ◽  
Bruce Thompson
Keyword(s):  
Structural Equation ◽  
Structural Model ◽  
Structural Equation Models ◽  
Equation Model ◽  
Model Fit ◽  
Model Specification ◽  
Simulation Design ◽  
Measurement Models ◽  
Fit Statistics ◽  
Path Coefficients

The purpose of the present paper was to evaluate the effect of constraining near-zero parameter cross-loadings to zero in the measurement component of a structural equation model. A Monte Carlo 3 × 5 × 2 simulation design was conducted (i.e., sample sizes of 200, 600, and 1,000; parameter cross-loadings of 0.07, 0.10, 0.13, 0.16, and 0.19 misspecified to be zero; and parameter path coefficients in the structural model of either 0.50 or 0.70). Results indicated that factor pattern coefficients and factor covariances were overestimated in measurement models when near-zero parameter cross-loadings constrained to zero were higher than 0.13 in the population. Moreover, the path coefficients between factors were misestimated when the near-zero parameter cross-loadings constrained to zero were noteworthy. Our results add to the literature detailing the importance of testing individual model specification decisions, and not simply evaluating omnibus model fit statistics.

Sign in / Sign up

Export Citation Format

Share Document