27. Nonrandomized Studies: Measurement Error Correction

  1. Bradley E. Huitema

Published Online: 14 OCT 2011

DOI: 10.1002/9781118067475.ch27

The Analysis of Covariance and Alternatives: Statistical Methods for Experiments, Quasi-Experiments, and Single-Case Studies, Second Edition

The Analysis of Covariance and Alternatives: Statistical Methods for Experiments, Quasi-Experiments, and Single-Case Studies, Second Edition

How to Cite

Huitema, B. E. (2011) Nonrandomized Studies: Measurement Error Correction, in The Analysis of Covariance and Alternatives: Statistical Methods for Experiments, Quasi-Experiments, and Single-Case Studies, Second Edition, John Wiley & Sons, Inc., Hoboken, NJ, USA. doi: 10.1002/9781118067475.ch27

Author Information

  1. Department of Psychology, Western Michigan University, Kalamazoo, Michigan, USA

Publication History

  1. Published Online: 14 OCT 2011
  2. Published Print: 14 OCT 2011

Book Series:

  1. Wiley Series in Probability and Statistics

Book Series Editors:

  1. Walter A. Shewhart and
  2. Samuel S. Wilks

ISBN Information

Print ISBN: 9780471748960

Online ISBN: 9781118067475

SEARCH

Keywords:

  • analysis of covariance (ANCOVA);
  • classical measurement error;
  • measurement error correction;
  • nonrandomized design;
  • randomized-group case

Summary

This chapter describes the effects of measurement error (in X and in Y) in randomized experiments, and the effects of measurement error in nonrandomized studies, and outlines of a measurement error correction procedure for analysis of variance (ANCOVA). The measurement errors referred to in the chapter are of the type associated with the “classical” measurement error model. Measurement error in the exposure variable and the covariates causes bias in the case of nonequivalent group designs; this bias can be serious. Measurement error corrected ANCOVA provides an estimate of the outcome that ANCOVA would yield if hypothetical true scores were employed as the covariate rather than observed scores. The reduction in bias is accompanied by an increase in error variance that results from the correction procedure and from uncertainty in the parameter estimates that are used to compute the corrections. Specialized measurement error software is recommended for estimating these models.

Controlled Vocabulary Terms

completely randomized design; measurement error