Multiple imputation strategies for multiple group structural equation models

Craig K. Enders, Amanda C. Gottschall

Research output: Contribution to journalArticle

34 Citations (Scopus)

Abstract

Although structural equation modeling software packages use maximum likelihood estimation by default, there are situations where one might prefer to use multiple imputation to handle missing data rather than maximum likelihood estimation (e.g., when incorporating auxiliary variables). The selection of variables is one of the nuances associated with implementing multiple imputation, because the imputer must take special care to preserve any associations or special features of the data that will be modeled in the subsequent analysis. For example, this article deals with multiple group models that are commonly used to examine moderation effects in psychology and the behavioral sciences. Special care must be exercised when using multiple imputation with multiple group models, as failing to preserve the interactive effects during the imputation phase can produce biased parameter estimates in the subsequent analysis phase, even when the data are missing completely at random or missing at random. This study investigates two imputation strategies that have been proposed in the literature, product term imputation and separate group imputation. A series of simulation studies shows that separate group imputation adequately preserves the multiple group data structure and produces accurate parameter estimates.

Original languageEnglish (US)
Pages (from-to)35-54
Number of pages20
JournalStructural Equation Modeling
Volume18
Issue number1
DOIs
StatePublished - Jan 2011

Fingerprint

Multiple Imputation
Structural Equation Model
Maximum likelihood estimation
Imputation
structural model
Software packages
Data structures
Maximum Likelihood Estimation
Group
Missing Completely at Random
Selection of Variables
Missing at Random
Structural Equation Modeling
behavioral science
Auxiliary Variables
Missing Data
Software Package
Estimate
Biased
Data Structures

ASJC Scopus subject areas

  • Modeling and Simulation
  • Decision Sciences(all)
  • Economics, Econometrics and Finance(all)
  • Sociology and Political Science

Cite this

Multiple imputation strategies for multiple group structural equation models. / Enders, Craig K.; Gottschall, Amanda C.

In: Structural Equation Modeling, Vol. 18, No. 1, 01.2011, p. 35-54.

Research output: Contribution to journalArticle

Enders, Craig K. ; Gottschall, Amanda C. / Multiple imputation strategies for multiple group structural equation models. In: Structural Equation Modeling. 2011 ; Vol. 18, No. 1. pp. 35-54.
@article{0edcfe8585cf437b81809043fba015cb,
title = "Multiple imputation strategies for multiple group structural equation models",
abstract = "Although structural equation modeling software packages use maximum likelihood estimation by default, there are situations where one might prefer to use multiple imputation to handle missing data rather than maximum likelihood estimation (e.g., when incorporating auxiliary variables). The selection of variables is one of the nuances associated with implementing multiple imputation, because the imputer must take special care to preserve any associations or special features of the data that will be modeled in the subsequent analysis. For example, this article deals with multiple group models that are commonly used to examine moderation effects in psychology and the behavioral sciences. Special care must be exercised when using multiple imputation with multiple group models, as failing to preserve the interactive effects during the imputation phase can produce biased parameter estimates in the subsequent analysis phase, even when the data are missing completely at random or missing at random. This study investigates two imputation strategies that have been proposed in the literature, product term imputation and separate group imputation. A series of simulation studies shows that separate group imputation adequately preserves the multiple group data structure and produces accurate parameter estimates.",
author = "Enders, {Craig K.} and Gottschall, {Amanda C.}",
year = "2011",
month = "1",
doi = "10.1080/10705511.2011.532695",
language = "English (US)",
volume = "18",
pages = "35--54",
journal = "Structural Equation Modeling",
issn = "1070-5511",
publisher = "Psychology Press Ltd",
number = "1",

}

TY - JOUR

T1 - Multiple imputation strategies for multiple group structural equation models

AU - Enders, Craig K.

AU - Gottschall, Amanda C.

PY - 2011/1

Y1 - 2011/1

N2 - Although structural equation modeling software packages use maximum likelihood estimation by default, there are situations where one might prefer to use multiple imputation to handle missing data rather than maximum likelihood estimation (e.g., when incorporating auxiliary variables). The selection of variables is one of the nuances associated with implementing multiple imputation, because the imputer must take special care to preserve any associations or special features of the data that will be modeled in the subsequent analysis. For example, this article deals with multiple group models that are commonly used to examine moderation effects in psychology and the behavioral sciences. Special care must be exercised when using multiple imputation with multiple group models, as failing to preserve the interactive effects during the imputation phase can produce biased parameter estimates in the subsequent analysis phase, even when the data are missing completely at random or missing at random. This study investigates two imputation strategies that have been proposed in the literature, product term imputation and separate group imputation. A series of simulation studies shows that separate group imputation adequately preserves the multiple group data structure and produces accurate parameter estimates.

AB - Although structural equation modeling software packages use maximum likelihood estimation by default, there are situations where one might prefer to use multiple imputation to handle missing data rather than maximum likelihood estimation (e.g., when incorporating auxiliary variables). The selection of variables is one of the nuances associated with implementing multiple imputation, because the imputer must take special care to preserve any associations or special features of the data that will be modeled in the subsequent analysis. For example, this article deals with multiple group models that are commonly used to examine moderation effects in psychology and the behavioral sciences. Special care must be exercised when using multiple imputation with multiple group models, as failing to preserve the interactive effects during the imputation phase can produce biased parameter estimates in the subsequent analysis phase, even when the data are missing completely at random or missing at random. This study investigates two imputation strategies that have been proposed in the literature, product term imputation and separate group imputation. A series of simulation studies shows that separate group imputation adequately preserves the multiple group data structure and produces accurate parameter estimates.

UR - http://www.scopus.com/inward/record.url?scp=78651317614&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=78651317614&partnerID=8YFLogxK

U2 - 10.1080/10705511.2011.532695

DO - 10.1080/10705511.2011.532695

M3 - Article

AN - SCOPUS:78651317614

VL - 18

SP - 35

EP - 54

JO - Structural Equation Modeling

JF - Structural Equation Modeling

SN - 1070-5511

IS - 1

ER -