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

LONGITUDINAL MEASUREMENT NON-INVARIANCE ON GROWTH PARAMETERS RECOVERY AND CLASSIFICATION ACCURACY IN GROWTH MIXTURE MODELING

Zhang, Caiyan (2016) LONGITUDINAL MEASUREMENT NON-INVARIANCE ON GROWTH PARAMETERS RECOVERY AND CLASSIFICATION ACCURACY IN GROWTH MIXTURE MODELING. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img] PDF
Primary Text
Restricted to University of Pittsburgh users only until 22 January 2021.

Download (2MB) | Request a Copy

Abstract

First-order growth mixture model (1-GMM) has received increased attention over the past decade. It models class-specific latent growth trajectory and individual classification using composite scores computed over items of the same scale across multiple time points. By default, using composite scores assumes identical item-to-construct relationship over time (longitudinal measurement invariance; L-MI), which is not necessarily the case in research practice.

Violation of L-MI assumption has been studied using latent growth curve modeling where subjects are assumed to be sampled from one latent class. Deviation from L-MI assumption impacted the growth characteristics, thus producing invalid conclusions on the pattern of change. This study extends the prior research on the impact of L-MI violation to the situation where multiple latent classes exist. A Monte Carlo study was performed to examine how systematically varied measurement non-invariance impacted class-specific growth factor parameter recovery and classification accuracy. Five factors were systematically manipulated in studying the impact of L-MI assumption violation: directional change in non-invariant item intercepts, patterns of item loadings and item intercepts, percent of items containing a set of non-invariant item parameters, presence of time-adjacent within-item correlated measurement error, and latent class distances. Additionally, three GMMs were compared to assess their robustness against longitudinal measurement non-invariance, including 1-GMM, second order GMM with constrained measurement invariance, and second order GMM with freely estimated item factor loadings and item intercepts.

Accuracy, precision, Type I error, and power were examined on the slope factor parameter estimates. Additionally, mixture proportion and individual classification were assessed. Results show that the second order GMM with freely estimated item loadings and item intercepts was robust under various violation of L-MI and able to produce accurate estimates of slope factor parameters. Performance of the second order GMM with constrained measurement invariance on slope factor parameters recovery depended on the specific generating measurement non-invariance configuration. 1-GMM, on the other hand, was not able to recover the slope factor parameters with deviation from the L-MI assumption. With extremely unbalanced mixture proportions, class membership assignment was found not satisfactory regardless of simulated measurement non-invariance condition and analysis model.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Zhang, Caiyancaz18@pitt.eduCAZ18
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairYe, Feifeifeifeiye@pitt.eduFEIFEIYE
Committee MemberLane, Suzannesl@pitt.eduSL
Committee MemberStone, Clement A.cas@pitt.eduCAS
Committee MemberWright, Aidan G. C.aidan@pitt.eduAIDAN
Date: 22 January 2016
Date Type: Publication
Defense Date: 19 November 2015
Approval Date: 22 January 2016
Submission Date: 20 January 2016
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 156
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: Statistics; Quantitative psychology and psychometrics; Developmental psychology
Date Deposited: 22 Jan 2016 18:34
Last Modified: 15 Nov 2016 14:31
URI: http://d-scholarship.pitt.edu/id/eprint/26754

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item