Allowing correlated error among similar depression items improves model fit.

Evidence

The CESD-R scale had a McDonald’s omega estimate of . (90% confidence interval [.95, .96]) and model fit indices for confirmatory factor analysis of the instrument were poor (root mean square error of approximation = .07; comparative fit index = .87; standardized root mean square residual = .05). Employing modification indices for similar items to covary with each other (insomnia, oversleeping, low self-esteem, suicidal ideation) resulted in improved model fit indices (root mean square error of approximation = .04; comparative fit index = .95; standardized root mean square residual = .04). — Naheed Ahmed, Measurement of Perceived Interpersonal…, 40

Categorization

• XREF: Directly relevant to anyone running confirmatory factor analysis on legacy instruments where wording overlap inflates error covariance.