OpenClustered: an R package with a benchmark suite of clustered datasets for methodological evaluation and comparison

Area of Science:

• Statistics
• Data Science
• Bioinformatics

Background:

• Clustered data, common in epidemiology and social sciences, exhibit correlations within groups.
• Existing data repositories lack comprehensive resources for clustered datasets.
• Traditional simulation studies for methodology evaluation can introduce bias.

Purpose of the Study:

• To develop an open-source data repository for clustered datasets.
• To facilitate methodologic comparison and benchmarking studies.
• To provide an alternative to potentially biased data simulation.

Main Methods:

• Development of the R package 'OpenClustered'.
• Inclusion of 19 diverse clustered datasets with binary outcomes.
• Creation of tutorials for data manipulation and analysis.

Main Results:

• The OpenClustered package offers 19 clustered datasets of varying sizes and compositions.
• Tutorials demonstrate dataset filtering, summarization, and benchmarking.
• A pilot study compared Frequentist and Bayesian generalized linear mixed models using the package.

Conclusions:

• OpenClustered serves as a valuable resource for benchmarking studies with open-source clustered data.
• The package promotes empirical methodologic guidance, enhancing research rigor.
• Future plans include expanding the dataset collection and user submission functionality.