Benchmark for Publication Bias Correction Methods

Implements a unified interface for benchmarking meta-analytic publication bias correction methods through simulation studies (see Bartoš et al., 2025, ). It provides 1) predefined data-generating mechanisms from the literature, 2) functions for running meta-analytic methods on simulated data, 3) pre-simulated datasets and pre-computed results for reproducible benchmarks, 4) tools for visualizing and comparing method performance.

README

PublicationBiasBenchmark

PublicationBiasBenchmark is an R package for benchmarking publication bias correction methods through simulation studies. It provides:\

• Predefined data-generating mechanisms from the literature\
• Functions for running meta-analytic methods on simulated data\
• Pre-simulated datasets and pre-computed results for reproducible benchmarks\
• Tools for visualizing and comparing method performance

All datasets and results are hosted on OSF: https://doi.org/10.17605/OSF.IO/EXF3M

For the methodology of living synthetic benchmarks please cite:

Bartoš, F., Pawel, S., & Siepe, B. S. (2025). Living synthetic benchmarks: A neutral and cumulative framework for simulation studies. arXiv Preprint. https://doi.org/10.48550/arXiv.2510.19489

For the publication bias benchmark R package please cite:

Bartoš, F., Pawel, S., & Siepe, B. S. (2025). PublicationBiasBenchmark: Benchmark for publication bias correction methods (version 0.1.0). https://github.com/FBartos/PublicationBiasBenchmark

Overviews of the benchmark results are available as articles on the package website:

• Overall Results
• Stanley (2017)
• Alinaghi (2018)
• Bom (2019)
• Carter (2019)

Contributor guidelines for extending the package with data-generating mechanisms, methods, and results are available at:

• How to add a new data-generating mechanism
• How to add a new method
• How to compute method results
• How to compute method measures

Illustrations of how to use the precomputed datasets, results, and measures are available at:

• How to use presimulated datasets
• How to use precomputed results
• How to use precomputed measures

The rest of this file overviews the main features of the package.

Installation

Install the released CRAN version

install.packages("PublicationBiasBenchmark")

Install latest development version from GitHub

remotes::install_github("FBartos/PublicationBiasBenchmark")

Versions

Addition or modification of a method or data-generating mechanisms is always reflected in a minor version update. Minor changes to infrastructure etc are reflected in patch updates.

Usage

library(PublicationBiasBenchmark)

Simulating From Existing Data-Generating Mechanisms

# Obtain a data.frame with pre-defined conditions
dgm_conditions("Stanley2017")

# simulate the data from the second condition
df <- simulate_dgm("Stanley2017", 2)

# fit a method
run_method("RMA", df)

Using Pre-Simulated Datasets

# download the pre-simulated datasets
# (the intended location for storing the package resources needs to be specified)
PublicationBiasBenchmark.options(resources_directory = "/path/to/files")
download_dgm_datasets("no_bias")

# retrieve first repetition of first condition from the downloaded datasets
retrieve_dgm_dataset("no_bias", condition_id = 1, repetition_id = 1)

Using Pre-Computed Results

# download the pre-computed results
download_dgm_results("no_bias")

# retrieve results the first repetition of first condition of RMA from the downloaded results
retrieve_dgm_results("no_bias", method = "RMA", condition_id = 1, repetition_id = 1)

# retrieve all results across all conditions and repetitions
retrieve_dgm_results("no_bias")

Using Pre-Computed Measures

# download the pre-computed measures
download_dgm_measures("no_bias")

# retrieve measures of bias the first condition of RMA from the downloaded results
retrieve_dgm_measures("no_bias", measure = "bias", method = "RMA", condition_id = 1)

# retrieve all measures across all conditions and measures
retrieve_dgm_measures("no_bias")

Simulating From an Existing DGM With Custom Settings

# define sim setting
sim_settings <- list(
  n_studies     = 100,
  mean_effect   = 0.3,
  heterogeneity = 0.1
)

# check whether it is feasible
# (defined outside of the function - not to decrease performance during simulation)
validate_dgm_setting("no_bias", sim_settings)

# simulate the data
df <- simulate_dgm("no_bias", sim_settings)

# fit a method
run_method("RMA", df)

Key Functions

Data-Generating Mechanisms

• simulate_dgm(): Generates simulated data according to specified data-generating mechanism and settings.
• dgm_conditions(): Lists prespecified conditions of the data-generating mechanism.
• validate_dgm_setting(): Validates (custom) setting of the data-generating mechanism.
• download_dgm_datasets(): Downloads pre-simulated datasets from the OSF repository.
• retrieve_dgm_dataset(): Retrieves the pre-simulated dataset of a given condition and repetition from downloaded from the pre-downloaded OSF repository.

Method Estimation And Results

• run_method(): Estimates method on a supplied data according to the specified settings.
• method_settings(): Lists prespecified settings of the method.
• download_dgm_results(): Downloads pre-computed results from the OSF repository.
• retrieve_dgm_results(): Retrieves the pre-computed results of a given method, condition, and repetition from the pre-downloaded OSF repository.

Performance measures And Results

• bias(), bias_mcse(), etc.: Functions to compute performance measures and their Monte Carlo standard errors.
• download_dgm_measures(): Downloads pre-computed performance measures from the OSF repository.
• retrieve_dgm_measures(): Retrieves the pre-computed performance measures of a given method, condition, and repetition from the pre-downloaded OSF repository.

Available Data-Generating Mechanisms

See methods("dgm") for the full list:

• "no_bias": Generates data without publication bias (a test simulation)
• "Stanley2017": Stanley et al. (2017)
• "Alinaghi2018": Alinaghi & Reed (2018)
• "Bom2019": Bom & Rachinger (2019)
• "Carter2019": Carter et al. (2019)

Available Methods

See methods("method") for the full list:

• "mean": Mean effects size
• "FMA": Fixed effects meta-analysis
• "RMA": Random effects meta-analysis
• "WLS": Weighted Least Squares
• "trimfill": Trim-and-Fill (Duval & Tweedie, 2000)
• "WAAPWLS": Weighted Least Squares - Weighted Average of Adequately Power Studies (Stanley et al., 2017)
• "WILS": Weighted and Iterated Least Squares (Stanley & Doucouliagos, 2024)
• "PET": Precision-Effect Test (PET) publication bias adjustment (Stanley & Doucouliagos, 2014)
• "PEESE": Precision-Effect Estimate with Standard Errors (PEESE) publication bias adjustment (Stanley & Doucouliagos, 2014)
• "PETPEESE": Precision-Effect Test and Precision-Effect Estimate with Standard Errors (PET-PEESE) publication bias adjustment (Stanley & Doucouliagos, 2014)
• "EK": Endogenous Kink (Bom & Rachinger, 2019)
• "SM": Selection Models (3PSM, 4PSM) (Vevea & Hedges, 1995)
• "pcurve": P-curve (Simonsohn et al., 2014)
• "puniform": P-uniform (van Assen et al., 2015) and P-uniform* (van Aert & van Assen, 2025)
• "AK": Andrews & Kasy selection models (AK1, AK2) (Andrews & Kasy, 2019)
• "RoBMA": Robust Bayesian Meta-Analysis (Bartoš et al., 2023)
• "MAIVE": Meta Analysis Instrumental Variable Estimator (Irsova et al., 2025)
• "MAN": Meta-Analysis of Nonaffirmative Studies (Mathur & VanderWeele, 2020)
• "RTMA": Right-Truncated Meta-Analysis (Mathur, 2024)

Available Performance Measures

See ?measures for the full list of performance measures and their Monte Carlo standard errors/

DGM OSF Repositories

All DGMs are linked to the OSF repository (https://osf.io/exf3m/) and contain the following elements:

• data : folder containing by-condition simulated datasets for all repetitions
• results : folder containing by-method results for all conditions * repetitions
• measures : folder containing by-measure performance for all methods * conditions
• metadata : folder containing the following information:
  • dgm-conditions.csv : file mapping of all conditions and the corresponding settings
  • dgm-generation.R : file with code for exact reproduction of the pre-simulated datasets
  • dgm-sessionInfo.txt: file with reproducibility details for the pre-simulated datasets
  • dgm-session.log: file with reproducibility details for the pre-simulated datasets (based on sessioninfo package)
  • results.R : file with code for exact reproduction of the by method results (might be method / method groups specific)
  • results-sessionInfo.txt: file with reproducibility details for the precomputed results (might be method / method groups specific)
  • pm-computation.R : file with code for computation of performance measures

References