Institutional members access full text with Ovid®

Share this article on:

Bayesian Multilevel MIMIC Modeling for Studying Measurement Invariance in Cross-group Comparisons

Bruyneel, Luk MSc; Li, Baoyue MSc, PhD; Squires, Allison PhD, MSN; Spotbeen, Sara MSc; Meuleman, Bart PhD, MSc; Lesaffre, Emmanuel PhD, MSc; Sermeus, Walter PhD, MSc

doi: 10.1097/MLR.0000000000000164
Applied Methods

Background: Recent methodological advancements should catalyze the evaluation of measurement invariance across groups, which is required for conducting meaningful cross-group comparisons.

Objective: The aim of this study was to apply a state-of-the-art statistical method for comparing latent mean scores and evaluating measurement invariance across managers’ and frontline workers’ ratings of the organization of hospital care.

Methods: On the 87 nursing units in a single institution, French-speaking and Dutch-speaking nursing unit managers’ and staff nurses’ ratings of their work environment were measured using the multidimensional 32-item practice environment scale of the nursing work index (PES-NWI). Measurement invariance and latent mean scores were evaluated in the form of a Bayesian 2-level multiple indicators multiple causes model with covariates at the individual nurse and nursing unit level. Role (manager, staff nurse) and language (French, Dutch) are of primary interest.

Results: Language group membership accounted for 7 of 11 PES-NWI items showing measurement noninvariance. Cross-group comparisons also showed that covariates at both within-level and between-level had significant effects on PES-NWI latent mean scores. Most notably, nursing unit managers, when compared with staff nurses, hold more positive views of several PES-NWI dimensions.

Conclusions: Using a widely used instrument for measuring nurses’ work environment, this study shows that precautions for the potential threat of measurement noninvariance are necessary in all stages of a study that relies on survey data to compare groups, particularly in multilingual settings. A Bayesian multilevel multiple indicators multiple causes approach can accommodate for detecting all possible instances of noninvariance for multiple covariates of interest at the within-level and between-level jointly.

*Department of Public Health and Primary Care, Katholieke Universiteit Leuven, Kapucijnenvoer, Leuven, Belgium

Department of Biostatistics, Erasmus University Rotterdam, Rotterdam, the Netherlands

College of Nursing, New York University, New York, NY

§Department of Sociology, Katholieke Universiteit Leuven, Leuven, Belgium

The authors declare no conflicts of interest.

Reprints: Luk Bruyneel, MSc, Department of Public Health and Primary Care, Katholieke Universiteit Leuven, Kapucijnenvoer 35, Leuven 3000, Belgium. E-mail: luk.bruyneel@med.kuleuven.be.

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.