Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form


zhou, li (2016) IGNORING MULTILEVEL DATA STRUCTURE IN COMFIRMATORY FACTOR ANALYSIS WITH ORDINAL ITEMS. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Download (1MB) | Preview


This study used the Monte Carlo method to compare multilevel and single-level models in confirmatory factor analysis (CFA) of ordinal items with clustered data. Specifically, model fit indices, estimates of factor loading and standard error were compared among three models, two-level CFA, single-level CFA with adjusted standard error, and single-level CFA with normal standard error. Two different factorial structures were considered, 2 factors at both the within- and between-level (W2B2) and 2 factors at the within-level and 1 factor at the between-level (W2B1).
All model fit indices indicated that the two-level CFA model fitted the clustered data well. The model fit of the two-level CFA model was better than that of the single-level CFA model with adjusted standard error, which was better than that of the normal single-level CFA model. Chi-square p value and RMSEA were not as sensitive as CFI and TLI in the small sample size to the misspecification of factorial structure. When factor loadings across levels were the same in the true model, factor loadings estimated from the single-level models were acceptable. The standard error of the within-level factor loading estimated by the normal model was significantly smaller than the complex model, which was smaller than the two-level model, suggesting that standard errors are underestimated when the single-level model is used to estimate the two-level data. The effect of design factors on the relative bias of the factor loading and standard errors between W2B2 model and W2B1 model were similar in most conditions.
These results suggest applied researchers consider the interest of the study first when selecting CFA models of clustered data. The single-level CFA with adjusted standard error is preferred when the interest of the study is at the individual level, while multilevel CFA is recommended when the interest is at the cluster level. However, in either case, the recommendation is to compare both models to prevent the spurious clustering effect. If there truly exists a multilevel data structure, standard errors estimated from the two-level CFA model are expected to be significantly larger than adjusted standard errors in the single-level model.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
zhou, liliz63@pitt.eduliz63
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Thesis AdvisorYe,
Committee MemberStone,
Committee MemberLan,
Committee MemberTerhorst, ,
Date: 20 December 2016
Date Type: Publication
Defense Date: 14 November 2016
Approval Date: 20 December 2016
Submission Date: 18 December 2016
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 137
Institution: University of Pittsburgh
Schools and Programs: School of Education > Psychology in Education
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: ignoring multilevel confirmatory factor analysis ordinal items
Date Deposited: 20 Dec 2016 19:02
Last Modified: 20 Dec 2021 06:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item