ERIC Number: ED663513
Record Type: Non-Journal
Publication Date: 2024
Pages: 144
Abstractor: As Provided
ISBN: 979-8-3840-8902-5
ISSN: N/A
EISSN: N/A
Available Date: N/A
Applications of Planned Missing Data Designs in Latent Change Score Models
Ayse Busra Ceviren
ProQuest LLC, Ph.D. Dissertation, The Ohio State University
Latent change score (LCS) models are a powerful class of structural equation modeling that allows researchers to work with latent difference scores that minimize measurement error. LCS models define change as a function of prior status, which makes it well-suited for modeling developmental theories or processes. In LCS models, like other latent models, latent constructs require multiple observed variables, which may require a considerable investment in both time and financial resources. Planned missing data designs offer a valuable opportunity to make research more efficient in terms of time and cost with little or no loss of statistical power. Drawing on the benefits of planned missing data designs and the strengths of LCS models, I examined the planned missing data designs in the context of LCS models. Utilizing real-life data, different missing data designs and sample size conditions were tested on univariate LCS models with three indicators measured at two time points. An external Monte Carlo simulation study was conducted in each sample size condition (200, 500, and 1,000), in which 500 replications were generated for each planned missing data design. I assessed the convergence rate of LCS models, the relative bias of parameter estimates in LCS models, and the relative efficiency of parameter estimates in LCS models to determine if planned missing data designs have advantages over complete data designs. The convergence rates were 100% in all nine patterns and three sample sizes with no improper solutions. The results showed that the relative bias and relative efficiency in parameter estimates varied across planned missing patterns and sample size conditions. In patterns where planned missingness was set on only one indicator, 20% of missing data did not result in biased parameter estimates. However, setting higher levels of missing data (50% or 66%) on a single indicator resulted in bias in parameter estimates across sample sizes. Results on imposing planned missingness on one vs. two indicators indicated that when 93% of the complete data is available, the number of indicators with planned missing data did not make a difference in bias estimates. However, with only 83% or 78% of the complete data, only distributing the planned missingness evenly across two indicators (10%, 25%, or 33% planned missingness on each) at both time points resulted in accurate parameter estimates. Relative efficiency estimates suggested imposing the same pattern of planned missingness across time points. When the missingness pattern did not stay constant across time points, none of the models resulted in efficient parameter estimates. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com.bibliotheek.ehb.be/en-US/products/dissertations/individuals.shtml.]
Descriptors: Structural Equation Models, Error of Measurement, Statistical Bias, Monte Carlo Methods, Research Design, Data Analysis
ProQuest LLC. 789 East Eisenhower Parkway, P.O. Box 1346, Ann Arbor, MI 48106. Tel: 800-521-0600; Web site: http://www.proquest.com.bibliotheek.ehb.be/en-US/products/dissertations/individuals.shtml
Publication Type: Dissertations/Theses - Doctoral Dissertations
Education Level: N/A
Audience: N/A
Language: English
Sponsor: Institute of Education Sciences (ED)
Authoring Institution: N/A
IES Funded: Yes
Grant or Contract Numbers: R305A190029; R305E100030
Author Affiliations: N/A