Get started
Work

Evaluating Missing Data Recovery Techniques in Two Wave Planned Missing Data Designs for Estimating Latent Growth Parameters

Year: 2024
Type: dissertation
Abstract: Researchers in longitudinal studies often face the challenge of missing data. This challenge can be addressed through adoption of Planned Missing Data Designs (PMDDs) and modern missing data treatment methods in longitudinal research. These approaches improve data quality, reduce burden, and optimize resources. However, researchers must carefully consider the optimal sample size when implementing PMDDs. The proposed design incorporates intentional missingness across waves for all groups, offering increased flexibility in study design and resource allocation. By eliminating the requirement for a fully observed group, the proposed design effectively reduces resource demands. This strategy also alleviates the logistical and data collection burdens typically placed on researchers and participants in traditional designs that mandate a fully observed group. The proposed wave planned missing design employs modern methods for handling missing data. Researchers can then compare the outcomes of this new proposed design with those of the established wave planned missing design. This study employs Monte Carlo simulation to demonstrate the feasibility of the proposed wave planned missing data design. This study enables researchers to compare the performance of two types of wave planned missing data designs. It also examines the effectiveness of two modern methods for handling missing data under simulated conditions. This study aims to demonstrate that the proposed wave-planned missing data design can produce results like those from a complete dataset when modern missing data techniques are applied. Furthermore, the study aims to demonstrate that the proposed design performs similarly to the established wave-planned missing data design. It evaluates 24 conditions impacting the design's viability by comparing its performance for estimating growth curve parameters in longitudinal research. The study also introduces a robust visual technique, the ridgeline plot, to enhance the understanding of the performance of the two designs and the two methods for handling missing data. The study results show that the proposed wave-planned missing data design works well, even with small sample sizes of at least 300 per group. FIML consistently outperforms the PcAux missing data treatment method across most conditions, with a greater extent as sample size increases. The Absolute Relative Bias (ARB) of latent growth estimates is significantly influenced by the magnitude of the population parameter, a trend consistent across both designs. Relative efficiency indicates that PcAux becomes increasingly unreliable in handling missing data as sample size increases, with its performance deteriorating in terms of accuracy, bias, and efficiency in parameter estimation. In contrast, FIML shows increasing efficiency when sample size increases, often achieving relative efficiency greater than 60%. The study demonstrates that the newly proposed wave planned missing design with samples of at least 300 per group performs similarly to the established traditional wave planned missing data design, offering the advantage of reducing the research burden by using FIML missing data treatment method. This makes it a practical alternative, without significantly compromising accuracy, bias, or efficiency in parameter estimation. Overall, the study successfully demonstrates that the proposed design can return unbiased parameter estimates with sufficiently large sample size with FIML missing data treatment method.
Language: English
Open Access status: green
OpenAlex ID: W7019089156
Cites:
Cited by:
Related to: