Scott, Paul Wesley
(2018)
DISAGGREGATING BETWEEN AND WITHIN-PERSON EFFECTS IN AUTOREGRESSIVE CROSS-LAG MODELS.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
There is often interest in evaluating bidirectional relationships amongst processes over time. Random Intercept Cross-lagged Panel Model (RI-CLPM) and Latent Growth Curve Model with Structured Residuals (LGCM-SR) are two models developed to disentangle within- from between-person effects. These models have shown to out-perform Cross-Lagged Panel Model (CLPM) that confounds between- and within-person effects.
This study uses both empirical and simulated data to compare the performances of these models in assessing the bidirectional relationship between two developmental processes. Data from the Longitudinal Study of American Youth were used to explore bidirectional relationships between student math self-concept and task value from grades 7 to 12. The CLPM indicated Self-Concept dominated Task-Value, while the RI-CLPM and LGCM-SR indicated Task-Value dominated Self-Concept, suggesting that the CLPM’s confounding of between- and within-person effects leads to substantively different conclusions than RI-CLPM and LGCM-SR.
A Monte Carlo study was conducted to compare RI-CLPM and LGCM-SR. The RI-CLPM fits time-specific means to capture the functional form of the trajectory, but does not capture variation around the trajectory as a LGCM-SR would. Data were simulated from both models under different causal dominance conditions. For LGCM-SR models, data were generated with different slope variance, covariance, and trajectory shape. Fitting LGCM-SR models to RI-CLPM data results with negative variances of growth factors, suggesting over-parameterization. Fitting RI-CLPM to LGCM-SR data results with underestimation of cross-lagged path coefficients, and the bias is larger in non-dominance conditions and increases with larger slope variances, suggesting the necessity to consider the slope heterogeneity if present. For the nonlinear LGCM-SR data, as RI-CLPM was estimable while linear LGCM-SR always had negative slope variances, capturing the functional form might be more important than capturing variability in slope. Relative fit indices performed well in selecting the correct model between RI-CLPM and LGCM-SR. BIC proved superior at correctly choosing the linear LGCM-SR over the unspecified LGCM-SR.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
| Creators | Email | Pitt Username | ORCID  |
|---|
| Scott, Paul Wesley | pws5@pitt.edu | pws5 | |
|
| ETD Committee: |
|
| Date: |
30 January 2018 |
| Date Type: |
Publication |
| Defense Date: |
2 October 2017 |
| Approval Date: |
30 January 2018 |
| Submission Date: |
16 January 2018 |
| Access Restriction: |
1 year -- Restrict access to University of Pittsburgh for a period of 1 year. |
| Number of Pages: |
166 |
| 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: |
multilevel models; intensive longitudinal data analysis; autoregressive cross-lag; cross-lag panel; structural equation modeling |
| Date Deposited: |
30 Jan 2018 17:29 |
| Last Modified: |
30 Jan 2019 06:15 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/33704 |
Metrics
Monthly Views for the past 3 years
Loading...
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}
