Wang, Hong
(2013)
A MONTE CARLO COMPARISON OF POLYTOMOUS ITEM ESTIMATION BASED ON HIGHER-ORDER ITEM RESPONSE THEORY MODELS VERSUS HIGHER-ORDER CONFIRMATORY FACTOR ANALYSIS MODELS.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Item response theory (IRT) and confirmatory factor analysis (CFA) are two statistical techniques that were originally developed from different disciplines, but they are closely related to each other. Both can analyze the relationship between item responses and underlying constructs. This research investigated the performance of the two statistical methods with a higher-order structure in estimating polytomous response data. The higher-order IRT or second-order CFA model formulates correlational structure of multiple domains through a higher-order latent trait. This study compared Markov chain Monte Carlo (MCMC) estimation under a higher-order IRT model to mean-and-variance adjusted weighted least square (WLSMV) estimation under a second-order CFA model. The accuracy of the two estimation methods in recovering item parameters, overall and domain-specific abilities, and their correlations was examined under varied conditions.
The results showed MCMC and WLSMV methods were comparable on the accuracy of item discrimination and threshold parameter estimations. Although both estimation methods were found to yield more accurate item discrimination estimates as the number of items in each domain increased, WLSMV method was more sensitive to the number of items. The study also showed both estimation methods performed equally well in estimating overall and domain abilities. The accuracy of ability estimation increased as the number of items, number of dimensions, and correlations between domains increased.
Beck Depression Inventory was analyzed using both estimation methods. Consistent with the findings in the simulation study, the results indicated the two estimation methods yielded quite comparable estimates for both item parameters and abilities at general and specific levels. The results also showed some variations in the item parameter estimates across different prior distributions used in MCMC, indicating the effect of priors on MCMC for the item parameter estimation. Furthermore, both estimation methods exhibited a convergence problem when the correlations between the general and specific factors were very high (i.e., r > .90).
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
10 January 2013 |
| Date Type: |
Publication |
| Defense Date: |
14 September 2012 |
| Approval Date: |
10 January 2013 |
| Submission Date: |
10 December 2012 |
| Access Restriction: |
5 year -- Restrict access to University of Pittsburgh for a period of 5 years. |
| Number of Pages: |
129 |
| 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: |
Higher-order IRT CFA Bayesian MCMC estimation |
| Date Deposited: |
10 Jan 2013 15:15 |
| Last Modified: |
10 Jan 2018 06:15 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/16876 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}