Bias and Precision of Parameter Estimates in Structural Equation Modeling and Multiple Regression

Master's Thesis
Thumbnail

Abstract

SEM is often preferred over multiple regression due to the many statistical and conceptual advantages. One of these is its ability to account for measurement error and obtain unbiased estimates of the relationships between latent variables. This decrease in bias does not come without cost as SEM suffers from decreased precision of parameter estimates as compared to multiple regression. Reduced precision is a consequence of increased model complexity as well as the increased effect of collinearity due to the dissattenuation of the correlation between latent predictors. This paper examines the bias, precision, accuracy, and confidence interval coverage of parameter estimates in SEM and multiple regression. Results show that with small sample sizes multiple regression can often produce parameter estimates that are more accurate than SEM even though the estimates are biased. Multiple factors that affect the accuracy of estimates are explored and some suggestions are provided for researchers.

Attributes

Attribute NameValues
URN
  • etd-12112009-133431

Author Robert Anthony Perera
Advisor Scott E. Maxwell
Contributor Ke-Hai Yuan, Committee Member
Contributor Guanjian Zhang, Committee Member
Contributor Scott E. Maxwell, Committee Chair
Degree Level Master's Thesis
Degree Discipline Psychology
Degree Name MA
Defense Date
  • 2009-12-04

Submission Date 2009-12-11
Country
  • United States of America

Subject
  • Structural Equation Modeling

  • Accuracy

  • Bias

  • Latent Variables

  • Precision

  • Multiple Regression

  • Confidence Interval Coverage

Publisher
  • University of Notre Dame

Language
  • English

Record Visibility Public
Content License
  • All rights reserved

Departments and Units

Files

Please Note: You may encounter a delay before a download begins. Large or infrequently accessed files can take several minutes to retrieve from our archival storage system.