engleski

A simulation study of parcelling in confirmatory factor analysis

Parcelling was recommended as a remedy for problems of asymmetric distributions and small sample size in confirmatory factor analysis (CFA). It was also criticized for arbitrary allocation of items to parcels, and lack of interpretability of new items. Research goal was to assess sensitivity of CFA model fit to item- parcel allocation. Two simulation studies were performed. The first one was based on empirical data from the Survey of Attitudes Towards Statistics (SATS36) administered to two generations of students. Data for the second one were simulated using a model of SATS36 structure with parameter estimates taken from literature. Variances of model fit indicators (CFI, TLI, rmsea) were similar in both studies. For empirical data, sizable proportions of simulations did not achieve acceptable fit (CFI, TLI>0.9 for 70% and 42% simulations respectively, 0.05<rmsea<0.09). For simulated data, fit criteria were always met (CFI, TLI>0.90, rmsea<0.055). CFA model fit is sensitive to item allocation.

CFA, parcelling, simulation, SATS36.

nije evidentirano