Cross-Loadings in Scale Development: Monte Carlo Study to Compare Diagnostic Methods With Small Samples

• In Event: AERA Poster Session 12
  In Poster Session: Cross-Loadings in Scale Development: Monte Carlo Study to Compare Diagnostic Methods With Small Samples (San Diego)

Abstract

Applied researchers can perform item analyses with small samples to verify that items contribute well psychometrically as they develop unidimensional scales for survey research. For structural validity of multidimensional scales, they need much larger samples for component/factor analyses, which can also identify cross-loaded items that are correlated with multiples subscales. Methods that allow diagnosis of cross-loading items with small samples would help scholars address these problematic items earlier in the scale development process, while items can more easily be revised before beginning to collect validity evidence. We present Monte Carlo simulation results that show samples of 50-80 may be sufficient in many circumstances to identify cross-loaded items with both a method we call Item-Total Components Analysis and PCA.

Authors

• Nina Adjanin, Ohio University

• James Sika Pokoo, Ohio University

• Gordon P. Brooks, Ohio University