Using Multilevel Factor Analysis With Clustered Data: Investigating the Factor Structure of the Positive Values Scale

Advances in multilevel modeling techniques now make it possible to investigate the psychometric properties of instruments using clustered data. Factor models that overlook the clustering effect can lead to underestimated standard errors, incorrect parameter estimates, and model fit indices. In addition, factor structures may differ depending on the level of analysis. The current study illustrates the application of multilevel factor analytic techniques using a large statewide sample of middle school students (n = 39,364) from 423 schools. Both multilevel exploratory and confirmatory factor analyses were used to investigate the factor structure of the Positive Values Scale (PVS) as part of a school climate survey. Results showed that for the PVS, a two-correlated factor model at Level 1 and a one-factor model at Level 2 best fit the data. Implications and guidance for applied researchers are discussed.