Using Synthetic Data in Communication Sciences and Disorders to Promote Computational Reproducibility and...

Area of Science:

• Communication Sciences and Disorders
• Data Science
• Scientific Reproducibility

Background:

• Data sharing is crucial for scientific reproducibility but is uncommon in Communication Sciences and Disorders (CSD) due to privacy concerns.
• Synthetic data offers a solution by creating artificial datasets that mimic original data's statistical properties without compromising privacy.

Purpose of the Study:

• To explore the feasibility and preliminary utility of synthetic data generation for promoting transparency and reproducibility in CSD research.
• To assess if synthetic data can preserve statistical properties and inferential relationships from original CSD datasets.

Main Methods:

• Ten open CSD datasets from the American Speech-Language-Hearing Association
• Big Nine
• domains were utilized.
• Synthetic data were generated using the synthpop R package.
• General utility was assessed using visual inspection and the standardized ratio of propensity mean squared error (S_pMSE); specific utility was evaluated by comparing inferential statistics (model fit, coefficients, p-values) between original and synthetic data.

Main Results:

• All generated synthetic datasets demonstrated strong general utility, preserving univariate and bivariate distributions.
• Six out of nine synthetic datasets successfully maintained inferential relationships, indicating strong specific utility.
• Three datasets with hierarchical structures showed low specific utility, suggesting limitations for complex data.

Conclusions:

• Synthetic data effectively preserves statistical properties and relationships in non-hierarchical CSD data, supporting its use for enhancing reproducibility.
• Further research is required to develop effective methods for generating synthetic data from hierarchical structures.
• Researchers should evaluate the utility of synthetic data for their specific use cases to ensure accurate preservation of results.