Item Meaning and Order as Causes of Correlated Residuals in Confirmatory Factor Analysis
Deborah L Bandalos
Abstract
In this study I examined the oft-made claim that correlated residuals result from similarities in item meaning. I also investigated the degree to which administering items with similar meaning or wording in a random, rather than a contiguous order would diminish the need for correlated residuals. Pairs of items were written to mimic item meaning similarities cited in the applied literature. Item order was experimentally manipulated such that pairs of items with similar meaning were presented either contiguously or randomly. Results indicated that items rated as most similar in meaning were more likely to exhibit correlated residuals, thus supporting the claim that meaning similarity is a driver of correlated residuals. However, randomly ordering items reduced the magnitude of correlated residuals and improved model-data fit relative to contiguous ordering. Implications of these results for those engaged in scale development and revision based on CFA results are discussed.