Flexible distributional models for meta-analysis of reading fluency outcomes from single-case designs: An examination...

Area of Science:

• Statistics
• Educational Psychology
• Behavioral Research Methods

Background:

• Growing interest in statistical modeling for single-case design (SCD) research.
• Existing methods like HLM and GLMM have limitations for SCD data, particularly curriculum-based measures.
• Need for flexible statistical models to accommodate the unique characteristics of SCD data.

Purpose of the Study:

• To demonstrate the utility of generalized additive models for location, scale, and shape (GAMLSS) for SCD research.
• To evaluate different distributional families and modeling specifications for reading fluency data from SCD studies.
• To assess the variability of outcome dispersion across studies and participants.

Main Methods:

• Utilized generalized additive models for location, scale, and shape (GAMLSS) with Bayesian methods.
• Evaluated normal (Gaussian), Poisson, and negative binomial distributional families.
• Employed graphical posterior predictive checks to assess model fit and dispersion variability.

Main Results:

• Negative binomial models with heterogeneous dispersions demonstrated superior fit compared to other families.
• The chosen models closely reproduced key features of the observed reading fluency data.
• Dispersion (outcome variability) was found to vary significantly across studies and participants.

Conclusions:

• GAMLSS provides a flexible and powerful tool for analyzing SCD data, especially for academic outcomes.
• The negative binomial distribution with heterogeneous dispersions is recommended for modeling reading fluency in SCD interventions.
• Future meta-analytic models for SCD data should consider a wider range of distributional families and varying dispersion.