Missing Data Techniques for Structural Equation Modeling.

Abstract

As with other statistical methods, missing data often create major problems for the estimation of structural equation models (SEMs). Conventional methods such as listwise or pairwise deletion generally do a poor job of using all the available information. However, structural equation modelers are fortunate that many programs for estimating SEMs now have maximum likelihood methods for handling missing data in an optimal fashion. In addition to maximum likelihood, this article also discusses multiple imputation. This method has statistical properties that are almost as good as those for maximum likelihood and can be applied to a much wider array of models and estimation methods.

The influence of maternal psychopathology on ratings of child psychiatric symptoms: an SEM analysis on cross-informant agreement

“…To use all available data for analysis, AMOS 8.0 estimates structural equation by maximum likelihood methods for handling missing data in an optimal fashion [5]. However, missing data may produce a sampling error, which puts the validity of findings in question.…”

Substance use outcomes 5½ years past baseline for partnership-based, family-school preventive interventions

“…The technique chosen for handling missing data was full-information maximum likelihood estimation (FIML). This method has been found to yield more efficient and less biased parameter estimates than traditional methods for dealing with missing data, such as pairwise or listwise deletion of cases (Acock, 2005;Wothke, 2000), and has become a preferred strategy for dealing with missing data (Allison, 2003;Schafer and Graham, 2002). 3 To test for risk moderation of intervention effects on frequency of substance use (AF, DF, CF, MF), MPU, and APU, participants were divided into higher-and lower-risk subgroups, based on their pretest levels of SII.…”

“…The control group appeared 3 Multiple imputation (MI) also was considered as an acceptable alternative to address missing data; however, MI utilizes all possible information as covariates for more accurate imputations, and because many of our outcomes (e.g., drunkenness, marijuana) have very low frequencies at pretest, MI has a higher potential for producing biased results (Schafer, 2003). FIML is an alternative that does not have that particular disadvantage, and is well-supported as a missing data technique (Allison, 2003;Schafer and Graham, 2002;Wothke, 2000). Supplemental FIML analyses utilizing an unrestricted sample that included all individuals who provided data at the pretest produced results similar to those presented.…”

Diverging Paths: Understanding Racial Differences in Civic Engagement Among White, African American, and Latina/o Adolescents Using Structural Equation Modeling

“…For descriptive statistics and Cronbach alpha reliabilities we used multiple imputation by chained equations (MICE), creating five sets of imputed variables (Royston & White, 2011). To impute missing data for the SEM models we used a full-information maximum likelihood approach (FIML) (Allison, 2003). The reason we used two different methods for imputation was because of software limitations.…”