The work environment of Army hospital nurses: Measurement and construct validity

Abstract

Statement of the problem. Investigators interested in examining organizational factors related to the hospital work environment have gravitated toward the use of several widely used measures. These instruments were developed for use with civilian nurses, not military nurses. The problem is that investigators examining the military hospital work environment use these same instruments assuming construct validity and measurement properties are equivalent across military populations. Aim. The aims of this study were: (1) to test the validity and measurement properties of the Practice Environment Scale (Lake, 2002) and Maslach Burnout Inventory (Maslach, Jackson & Leiter, 1996) and (2) to examine a proposed intent to leave model among nurses who work in U.S. Army hospitals. Methods. A non-experimental design using secondary data formed the basis of this study. The sample consisted of U.S. Army nurses (n = 333) and Department of the Army civilian nurses (n = 408) who were administered the Practice Environment Scale and Maslach Burnout Inventory. Covariance matrix models were tested using structural equation modeling. Model testing included confirmatory factor analysis, multi-sample modeling and multilevel modeling. Results. The Practice Environment Scale model fit the data for both Army and Department of the Army civilian nurses when modified with error covariances. Additionally, multi-sample modeling provided evidence of construct validity. Equivalence of measurement properties was mixed. For example, measurement error and unanalyzed factors were not equivalent. This suggests that the model was mis-specified or problematic. For the Maslach Burnout Inventory, a three-factor model fit the data poorly in both groups. However, a stand-alone emotional exhaustion subscale model fit the data and was also invariant across groups. For the intent to leave model, while there were strong correlations between variables, structural equation modeling did not demonstrate causal relationships. Therefore, the model was rejected.