Bayesian Estimation of Multivariate Latent Regression Models: Gauss Versus Laplace

Research output: Contribution to journalArticle

Abstract

A latent multivariate regression model is developed that employs a generalized asymmetric Laplace (GAL) prior distribution for regression coefficients. The model is designed for high-dimensional applications where an approximate sparsity condition is satisfied, such that many regression coefficients are near zero after accounting for all the model predictors. The model is applicable to large-scale assessments such as the National Assessment of Educational Progress (NAEP), which includes hundreds of student, teacher, and school predictors of latent achievement. Monte Carlo evidence suggests that employing the GAL prior provides more precise estimation of coefficients that equal zero in comparison to a multivariate normal (MVN) prior, which translates to more accurate model selection. Furthermore, the GAL yielded less biased estimates of regression coefficients in smaller samples. The developed model is applied to mathematics achievement data from the 2011 NAEP for 175,200 eighth graders. The GAL and MVN NAEP estimates were similar, but the GAL was more parsimonious by selecting 12 fewer (i.e., 83 of the 148) variable groups. There were noticeable differences between estimates computed with a GAL prior and plausible value regressions with the AM software (beta version 0.06.00). Implications of the results are discussed for test developers and applied researchers.

Original languageEnglish (US)
Pages (from-to)591-616
Number of pages26
JournalJournal of Educational and Behavioral Statistics
Volume42
Issue number5
DOIs
StatePublished - Oct 1 2017

Fingerprint

regression
student teacher
mathematics
school
evidence
Values
Group

Keywords

  • Bayesian Lasso
  • National Assessment of Educational Progress
  • multivariate generalized asymmetric Laplace distribution
  • multivariate regression
  • probit model

ASJC Scopus subject areas

  • Education
  • Social Sciences (miscellaneous)

Cite this

@article{7155cdba7caa41e48714d6da8f3cfcf6,
title = "Bayesian Estimation of Multivariate Latent Regression Models: Gauss Versus Laplace",
abstract = "A latent multivariate regression model is developed that employs a generalized asymmetric Laplace (GAL) prior distribution for regression coefficients. The model is designed for high-dimensional applications where an approximate sparsity condition is satisfied, such that many regression coefficients are near zero after accounting for all the model predictors. The model is applicable to large-scale assessments such as the National Assessment of Educational Progress (NAEP), which includes hundreds of student, teacher, and school predictors of latent achievement. Monte Carlo evidence suggests that employing the GAL prior provides more precise estimation of coefficients that equal zero in comparison to a multivariate normal (MVN) prior, which translates to more accurate model selection. Furthermore, the GAL yielded less biased estimates of regression coefficients in smaller samples. The developed model is applied to mathematics achievement data from the 2011 NAEP for 175,200 eighth graders. The GAL and MVN NAEP estimates were similar, but the GAL was more parsimonious by selecting 12 fewer (i.e., 83 of the 148) variable groups. There were noticeable differences between estimates computed with a GAL prior and plausible value regressions with the AM software (beta version 0.06.00). Implications of the results are discussed for test developers and applied researchers.",
keywords = "Bayesian Lasso, National Assessment of Educational Progress, multivariate generalized asymmetric Laplace distribution, multivariate regression, probit model",
author = "Culpepper, {Steven Andrew} and Park, {Trevor H}",
year = "2017",
month = "10",
day = "1",
doi = "10.3102/1076998617700598",
language = "English (US)",
volume = "42",
pages = "591--616",
journal = "Journal of Educational and Behavioral Statistics",
issn = "1076-9986",
publisher = "SAGE Publications Inc.",
number = "5",

}

TY - JOUR

T1 - Bayesian Estimation of Multivariate Latent Regression Models

T2 - Gauss Versus Laplace

AU - Culpepper, Steven Andrew

AU - Park, Trevor H

PY - 2017/10/1

Y1 - 2017/10/1

N2 - A latent multivariate regression model is developed that employs a generalized asymmetric Laplace (GAL) prior distribution for regression coefficients. The model is designed for high-dimensional applications where an approximate sparsity condition is satisfied, such that many regression coefficients are near zero after accounting for all the model predictors. The model is applicable to large-scale assessments such as the National Assessment of Educational Progress (NAEP), which includes hundreds of student, teacher, and school predictors of latent achievement. Monte Carlo evidence suggests that employing the GAL prior provides more precise estimation of coefficients that equal zero in comparison to a multivariate normal (MVN) prior, which translates to more accurate model selection. Furthermore, the GAL yielded less biased estimates of regression coefficients in smaller samples. The developed model is applied to mathematics achievement data from the 2011 NAEP for 175,200 eighth graders. The GAL and MVN NAEP estimates were similar, but the GAL was more parsimonious by selecting 12 fewer (i.e., 83 of the 148) variable groups. There were noticeable differences between estimates computed with a GAL prior and plausible value regressions with the AM software (beta version 0.06.00). Implications of the results are discussed for test developers and applied researchers.

AB - A latent multivariate regression model is developed that employs a generalized asymmetric Laplace (GAL) prior distribution for regression coefficients. The model is designed for high-dimensional applications where an approximate sparsity condition is satisfied, such that many regression coefficients are near zero after accounting for all the model predictors. The model is applicable to large-scale assessments such as the National Assessment of Educational Progress (NAEP), which includes hundreds of student, teacher, and school predictors of latent achievement. Monte Carlo evidence suggests that employing the GAL prior provides more precise estimation of coefficients that equal zero in comparison to a multivariate normal (MVN) prior, which translates to more accurate model selection. Furthermore, the GAL yielded less biased estimates of regression coefficients in smaller samples. The developed model is applied to mathematics achievement data from the 2011 NAEP for 175,200 eighth graders. The GAL and MVN NAEP estimates were similar, but the GAL was more parsimonious by selecting 12 fewer (i.e., 83 of the 148) variable groups. There were noticeable differences between estimates computed with a GAL prior and plausible value regressions with the AM software (beta version 0.06.00). Implications of the results are discussed for test developers and applied researchers.

KW - Bayesian Lasso

KW - National Assessment of Educational Progress

KW - multivariate generalized asymmetric Laplace distribution

KW - multivariate regression

KW - probit model

UR - http://www.scopus.com/inward/record.url?scp=85029076772&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85029076772&partnerID=8YFLogxK

U2 - 10.3102/1076998617700598

DO - 10.3102/1076998617700598

M3 - Article

AN - SCOPUS:85029076772

VL - 42

SP - 591

EP - 616

JO - Journal of Educational and Behavioral Statistics

JF - Journal of Educational and Behavioral Statistics

SN - 1076-9986

IS - 5

ER -