Publication Bias in Meta-Analysis: Prevention, Assessment, and Adjustments

Publication bias occurs when studies with statistically significant positive results are more likely to be published than studies with null or negative findings, and this asymmetry can distort the conclusions of any meta-analysis that relies on the accessible literature. This article explains the mechanisms behind publication bias, describes practical methods for detecting it, reviews statistical adjustments including trim-and-fill and selection models, and provides a decision framework for choosing among assessment approaches. The guidance is written for researchers, students, and life-science professionals who conduct or appraise meta-analyses and need concrete steps for bias assessment and reporting.

Understanding Publication Bias and Its Consequences

Publication bias arises from a simple but consequential pattern in the research enterprise. Studies that produce statistically significant results are more likely to be submitted for publication and accepted by journals than studies with null or nonsignificant outcomes. This phenomenon has been recognized across biomedical and psychological research and poses a direct threat to the validity of meta-analytic conclusions. When researchers combine only the published studies they can identify, the pooled estimate may be incorrect, usually in an overly optimistic direction. The Systematic Analysis of Publication Bias in Neurosurgery Meta-Analyses confirms that statistically significant positive results are more likely to be published than negative or insignificant outcomes, and this pattern can skew the interpretation of meta-analyses.

The consequences extend beyond individual meta-analyses. Publication bias can lead to overestimated treatment effects, suggest the existence of effects that do not actually exist, and undermine the credibility of entire research literatures. A meta-meta-analysis examining publication bias in psychology and medicine found that publication bias yields overestimated effects and may suggest the existence of non-existing effects. The same study noted that overestimation was minimal but statistically significant in the homogeneous subsets examined, with evidence of publication bias appearing similar in both psychology and medicine.

The problem is also theoretical. Reviews of published meta-analyses consistently show that many researchers fail to assess publication bias at all. In a review of meta-analyses in neurosurgery journals, 56.8% of 190 articles assessed publication bias, but only 27.5% of those that found evidence of bias made corrections using the trim-and-fill method. This means 58.4% of the meta-analyses either did not assess for publication bias or, if they detected it, did not adjust for it. Similarly, a review of meta-analyses in plastic surgery journals found that only 46% of eligible meta-analyses conducted a formal assessment, and only 6.3% both assessed and attempted to correct for publication bias.

The pattern is not new. An assessment of publication bias in cardiovascular disease meta-analyses published in 2005 found that publication bias was assessed in only 11.1% of 225 reviews published between 1990 and 2002. Assessment rates increased over time, from 3.4% before 1998 to 19.0% in 2002, but remained low overall. The study identified several factors associated with higher rates of bias assessment, including more than seven primary studies, searches of four or more databases, meta-analyses of observational studies, and more recent publication year.

At a Glance: Publication Bias Assessment Methods

The table below summarizes the main methods for assessing publication bias, their primary uses, and their limitations. This comparison can help researchers select appropriate methods for their specific meta-analysis context.

Causes and Mechanisms of Publication Bias

Selective Submission and Editorial Decisions

The primary driver of publication bias is the behavior of researchers and journal editors. Studies with statistically significant results are more likely to be submitted for publication, and journals are more likely to accept them. This creates a systematic gap between the complete set of conducted studies and the set of published studies available for meta-analysis. The review of publication bias modelling approaches describes this as the bias induced by the fact that research with statistically significant results is potentially more likely to be submitted and published than work with null or nonsignificant results.

Selective Reporting Within Studies

Beyond the decision to publish an entire study, researchers may selectively report outcomes within a study. This phenomenon, known as selective reporting bias, occurs when the likelihood that a result is reported depends on the magnitude or statistical significance of the effect. The tutorial on addressing selective reporting bias in meta-analysis of dependent effect sizes explains that when applied to selectively reported data, conventional meta-analytic models can produce systematically biased parameter estimates. This form of bias is particularly challenging because it operates within studies that are otherwise published and accessible.

Small-Study Effects

Small studies are more susceptible to publication bias than large studies for several reasons. Small studies have less statistical power, making significant results less likely unless the true effect is large. When small studies do produce significant results, they may be more likely to be published because the findings appear noteworthy. Small studies also tend to have more variable effect estimates, which can create asymmetry in funnel plots even without publication bias. The review of publication bias in clinical meta-analyses notes that clinical evidence syntheses frequently rely on small or selective sets of trials, making them particularly vulnerable to these effects.

Prevention Strategies Before Data Collection

Protocol Registration and Prospective Design

The most effective approach to publication bias is prevention through transparent research practices. Registering a meta-analysis protocol before data collection begins establishes a public record of the planned methods, search strategies, and analysis plans. This makes it more difficult to selectively report outcomes or change methods after seeing results. The EQUATOR Network provides reporting guidelines and resources that support transparent and complete reporting of health research, including systematic reviews and meta-analyses.

Comprehensive Literature Searching

A thorough search strategy reduces the risk of missing unpublished or difficult-to-find studies. The review of systematic review methods emphasizes that the selection of studies involves searching in web repositories, and more than one should be consulted. A manual search in the references of articles, editorials, and reviews is mandatory. Searching multiple databases, including PubMed and other NCBI Literature Resources, increases the likelihood of identifying relevant studies. The cardiovascular disease meta-analysis assessment found that searching four or more databases was associated with significantly higher rates of publication bias assessment, suggesting that comprehensive searching is linked to more rigorous methodology overall.

Inclusion of Grey Literature

Grey literature includes conference abstracts, dissertations, preprints, and other materials not indexed in traditional databases. Including these sources can reduce publication bias by capturing studies that may not have been published in peer-reviewed journals. However, grey literature must be evaluated carefully for quality and completeness. The review of publication bias in clinical meta-analyses discusses emerging challenges related to preprints and AI-assisted evidence synthesis, noting that the dissemination landscape is becoming more complex.

Study Quality Assessment

Assessing the quality of included studies is essential for interpreting publication bias results. The systematic review methods review recommends that data collection on quality of the selected reports should apply validated scales and include specific questions on the main biases that could negatively impact the research question. Quality assessment should be conducted by two researchers independently to reduce bias in the assessment process itself.

Assessing Publication Bias: Methods and Procedures

Funnel Plots

The funnel plot is the most widely used visual tool for assessing publication bias. It plots the effect size of each study against a measure of study precision, typically the standard error or sample size. In the absence of bias, the plot should resemble a symmetrical inverted funnel, with more precise studies clustering near the pooled estimate and less precise studies spreading more widely. Asymmetry in the funnel plot suggests the presence of publication bias or other small-study effects.

The neurosurgery meta-analysis review notes that visual inspection of funnel plots is one of the statistical approaches developed to assess and account for publication bias. However, funnel plot interpretation is subjective, and the plastic surgery meta-analysis review found that 49% of meta-analyses that conducted a formal assessment used subjective inspection of funnel plots alone, without any statistical testing. This reliance on visual inspection alone is concerning because funnel plot asymmetry can arise from sources other than publication bias, including true heterogeneity, poor study quality, and chance.

Egger's Regression Test

Egger's test is a statistical approach that quantifies funnel plot asymmetry by regressing the standardized effect size on precision. A significant intercept indicates asymmetry and potential publication bias. The Biometrics article on quantifying publication bias notes that Egger's regression intercept may be considered as a candidate measure for quantifying publication bias, but it lacks an intuitive interpretation. The test is widely used and is mentioned in the neurosurgery meta-analysis review alongside the Begg test as a standard approach for assessing publication bias.

Begg's Rank Correlation Test

Begg's test examines the rank correlation between effect sizes and their variances. It is a nonparametric test that is less sensitive to outliers than regression-based approaches but generally has lower statistical power. The neurosurgery meta-analysis review lists Begg's test alongside Egger's test as a standard method for assessing publication bias. Researchers may choose between these tests based on the characteristics of their data, including the number of studies and the presence of outliers.

Fail-Safe N

The fail-safe N, also known as the file drawer analysis, estimates the number of unpublished studies with null results that would be needed to overturn the meta-analytic conclusion. This method provides an intuitive measure of the robustness of findings to publication bias. However, the fail-safe N has important limitations. It assumes that unpublished studies have null effects, which may not be realistic, and it does not provide an adjusted effect estimate. The review of publication bias modelling approaches discusses methods that can be employed as sensitivity analyses to assess the likely impact of publication bias, and the fail-safe N is one such approach.

Skewness-Based Measures

A newer approach to quantifying publication bias uses the skewness of the standardized deviates of the collected studies. The Biometrics article introduces this measure as a way to describe the asymmetry of the collected studies' distribution. This measure can serve as a characteristic of a meta-analysis and permits comparisons of publication biases between different meta-analyses. A new test for publication bias is derived based on the skewness, and its performance is illustrated using simulations and case studies.

Statistical Adjustments for Publication Bias

Trim-and-Fill Method

The trim-and-fill method is a nonparametric approach that adjusts the meta-analytic estimate for publication bias. The method works by trimming the asymmetric studies from one side of the funnel plot, estimating the true center of the plot, and then filling in missing studies on the other side to create symmetry. The adjusted estimate is then calculated from the complete set of observed and imputed studies.

The neurosurgery meta-analysis review found that trim-and-fill was the method used in the 11 meta-analyses that made corrections for publication bias. The Biometrics article classifies trim-and-fill as a funnel-plot-based method, alongside visual examination, regression and rank tests. While trim-and-fill is widely used and relatively easy to implement, it has limitations. The method assumes that the funnel plot would be symmetric in the absence of publication bias, which may not hold when true heterogeneity exists.

Selection Models

Selection models use weight functions to adjust the overall effect size estimate based on assumptions about the probability of publication given the study results. These models are typically employed as sensitivity analyses to assess the potential impact of publication bias. The Biometrics article distinguishes selection models from funnel-plot-based methods, noting that selection models use weight functions to adjust the overall effect size estimate.

The review of publication bias modelling approaches provides a comprehensive review of selection model methods that can be employed as sensitivity analyses. These methods require explicit assumptions about the selection process, and different assumptions can lead to different adjusted estimates. The review of publication bias in clinical meta-analyses emphasizes that publication bias analysis methods are prone to over-interpretation and may yield conflicting conclusions, and they should be understood as an inferential process that links detection, model-based adjustment, and interpretation under explicit and unverifiable assumptions.

Regression-Based Adjustments

Regression-based methods relate the effect size to study precision and adjust the estimate based on the fitted relationship. The precision effect test (PET) and the precision effect estimate with standard error (PEESE) are examples of this approach. The study estimating changes in meta-analytic effect size estimates after publication bias adjustment reanalyzed 433 data sets from 90 articles and found that the downward adjustment from these methods was minimal, with the greatest identified attenuation for the precision effect test. The study also noted that some methods tended to adjust upward, especially for data sets with fewer than 10 primary studies.

Robust Bayesian Meta-Analysis

Robust Bayesian meta-analysis (RoBMA) represents a more recent development that model-averages across complementary publication bias adjustment methods. The article on robust Bayesian meta-analysis explains that no single method consistently outperforms others across a wide range of conditions, so model-averaging across selection models and models of the relationship between effect sizes and their standard errors can provide more reliable estimates. The resulting estimator weights the models with the support they receive from the existing research record.

The multilevel extension of robust Bayesian meta-analysis extends this approach to handle dependent effect sizes within studies, simultaneously addressing within-study dependencies, model uncertainty, heterogeneity, moderators, and publication bias. This method is implemented in the RoBMA R package and JASP.

Meta-Analysis of Non-Affirmative Studies

A simpler approach to assessing robustness to publication bias is the meta-analysis of non-affirmative studies. This method includes only studies with nonsignificant P values or point estimates in the undesired direction. The BMJ article on this method explains that the resulting estimate corrects for worst-case publication bias, a hypothetical scenario in which affirmative studies are almost infinitely more likely to be published than non-affirmative studies. If this estimate remains in the same direction as the uncorrected estimate and is of clinically meaningful size, the meta-analysis conclusions would not be overturned by any amount of publication bias favoring affirmative studies.

This method has several advantages. It accommodates small meta-analyses, non-normal effects, heterogeneous effects across studies, and additional forms of selective reporting including P-hacking. It complements an uncorrected meta-analysis and other publication bias analyses.

Decision Framework for Selecting Assessment Methods

Step 1: Assess the Number of Primary Studies

The number of primary studies is a critical factor in selecting publication bias assessment methods. The cardiovascular disease meta-analysis assessment found that having more than seven primary studies was associated with significantly higher rates of publication bias assessment. However, the study on publication bias adjustment methods cautions that these methods may not be able to combat bias in small samples with fewer than 10 primary studies. When the number of studies is small, visual inspection of funnel plots and formal statistical tests have limited power, and researchers should interpret results cautiously.

Step 2: Evaluate Heterogeneity

Publication bias methods do not have good statistical properties when the true effect size is heterogeneous. The meta-meta-analysis from psychology and medicine assessed publication bias on homogeneous subsets of primary studies because publication bias methods do not perform well with heterogeneous true effects. When substantial heterogeneity exists, researchers should consider subgroup analyses or meta-regression to explore sources of heterogeneity before interpreting publication bias assessments.

Step 3: Choose Primary and Sensitivity Analyses

For meta-analyses with at least 10 studies, a combination of visual funnel plot inspection and at least one statistical test is recommended. Egger's regression test is widely used and provides a numeric test statistic, while Begg's rank correlation test offers a nonparametric alternative. The neurosurgery meta-analysis review lists both tests as standard approaches.

For sensitivity analyses, researchers should consider trim-and-fill, selection models, or meta-analysis of non-affirmative studies. The BMJ article recommends that meta-analysis of non-affirmative studies complements an uncorrected meta-analysis and other publication bias analyses. The review of publication bias in clinical meta-analyses emphasizes that applying different publication bias adjustment methods to the same evidence base can yield divergent adjusted effects, highlighting the assumption dependence of these methods.

Step 4: Consider Dependent Effect Sizes

When primary studies contribute multiple effect sizes, standard publication bias methods may not be appropriate. The tutorial on selective reporting bias in meta-analysis of dependent effect sizes reviews several recently developed methods including a regression-based adjustment technique, a step-function selection model, and a sensitivity analysis approach. The multilevel robust Bayesian meta-analysis article presents a three-level approach that simultaneously handles within-study dependencies, model uncertainty, heterogeneity, moderators, and publication bias.

Reporting Publication Bias Analyses

What to Include in the Methods Section

The methods section of a meta-analysis should describe the publication bias assessment plan before presenting results. This includes specifying which methods will be used, the criteria for interpreting results, and the planned sensitivity analyses. The EQUATOR Network provides reporting guidelines that support complete and transparent reporting of health research. Following these guidelines helps readers understand the methods used and the limitations of the analysis.

What to Include in the Results Section

The results section should report the results of all planned publication bias assessments, including visual funnel plot inspection and statistical tests. When publication bias is detected, the results should include both the uncorrected and adjusted estimates from sensitivity analyses. The review of publication bias in clinical meta-analyses emphasizes that publication bias analysis methods should be understood as stress tests of model adequacy instead of as definitive detectors of publication bias.

What to Include in the Discussion Section

The discussion should interpret the publication bias results in the context of the study's limitations. If publication bias is present, the discussion should acknowledge that the pooled estimate may be overestimated and should consider the range of plausible estimates from sensitivity analyses. The study on publication bias adjustment methods proposes that researchers should seek to explore the full range of plausible estimates for the effects they are studying.

Common Failure Patterns in Publication Bias Assessment

Failure to Assess Publication Bias

The most common failure is not assessing publication bias at all. The neurosurgery meta-analysis review found that 43.2% of meta-analyses did not assess publication bias. The plastic surgery meta-analysis review found that only 46% of eligible meta-analyses conducted a formal assessment. This failure is particularly concerning because publication bias can confound the estimated therapeutic effect in meta-analyses.

Reliance on Visual Inspection Alone

Among meta-analyses that do assess publication bias, many rely on subjective inspection of funnel plots without any statistical testing. The plastic surgery meta-analysis review found that 49% of those who conducted an assessment used subjective inspection of funnel plots alone. Visual inspection is subjective and may miss subtle asymmetry, particularly when the number of studies is small.

Detecting Bias Without Adjusting for It

Another common failure is detecting publication bias but failing to adjust for it. The neurosurgery meta-analysis review found that among the 40 meta-analyses that found evidence of publication bias, only 11 made corrections using the trim-and-fill method. This means that 72.5% of meta-analyses that detected publication bias made no correction.

Applying Methods Without Regard to Conditions

The study on publication bias adjustment methods notes that previous studies tended to apply publication bias adjustment methods without regard to optimal conditions for each method's performance. Each method has conditions under which it performs well, including sample size, level of heterogeneity, population effect size, and the level of publication bias. Applying methods outside these conditions can produce misleading results.

Over-Interpreting Adjustment Results

The review of publication bias in clinical meta-analyses cautions that statistical adjustment cannot recover information that was selectively generated, reported, or disseminated. Publication bias analysis methods are prone to over-interpretation and may yield conflicting conclusions. Researchers should treat adjusted estimates as part of a sensitivity analysis instead of as definitive corrections.

Limitations of Publication Bias Assessment Methods

Statistical Power Limitations

Publication bias tests generally have limited statistical power, particularly when the number of primary studies is small. The meta-meta-analysis from psychology and medicine found that the median number of effect sizes in homogeneous subsets was 6, which created challenging conditions for publication bias methods. The study on publication bias adjustment methods notes that these methods may not be able to combat bias in small samples with fewer than 10 primary studies.

Assumption Dependence

All publication bias adjustment methods rely on assumptions about the selection process or the relationship between effect sizes and precision. Different assumptions can lead to different adjusted estimates. The review of publication bias in clinical meta-analyses demonstrates how applying different publication bias adjustment methods to the same evidence base can yield divergent adjusted effects, emphasizing their assumption dependence.

Confounding with Heterogeneity

Funnel plot asymmetry can arise from sources other than publication bias, including true heterogeneity, poor study quality, and chance. The meta-meta-analysis from psychology and medicine assessed publication bias on homogeneous subsets because publication bias methods do not have good statistical properties if the true effect size is heterogeneous. When heterogeneity is present, publication bias tests may produce false positives.

Inability to Recover Missing Information

The review of publication bias in clinical meta-analyses makes a fundamental point: statistical adjustment cannot recover information that was selectively generated, reported, or disseminated. Publication bias adjustment methods should be understood as an inferential process that links detection, model-based adjustment, and interpretation under explicit and unverifiable assumptions.

Professional Escalation Criteria

When to Consult a Statistician

Researchers should consult a statistician or methodological expert when the number of primary studies is small, when substantial heterogeneity is present, when dependent effect sizes are included, or when different publication bias methods yield conflicting results. The review of publication bias in clinical meta-analyses provides a framework for evaluations and discusses common misuses of publication bias methods.

When to Consider Additional Searches

If publication bias is detected or suspected, researchers should consider additional literature searches to identify unpublished studies. This may include searching trial registries, contacting experts in the field, and searching for preprints. The systematic review methods review emphasizes that a manual search in the references of articles, editorials, and reviews is mandatory.

When to Reconsider the Research Question

If publication bias is substantial and adjustment methods produce widely divergent estimates, researchers may need to reconsider the research question or the scope of the meta-analysis. The review of publication bias in clinical meta-analyses aims to help align the strength of clinical conclusions with the robustness or fragility of the underlying data.

Quality Controls and Documentation

Pre-Specification of Methods

Pre-specifying publication bias assessment methods in a protocol or registration document reduces the risk of selective reporting of analyses. The EQUATOR Network provides resources that support transparent and complete reporting of health research. Pre-specification also helps readers distinguish between planned analyses and post-hoc explorations.

Independent Assessment

Publication bias assessment should be conducted by at least two researchers independently, consistent with the recommendation for study selection and data collection in the systematic review methods review. Disagreements should be resolved through discussion or by a third reviewer.

Documentation of Decisions

All decisions related to publication bias assessment should be documented, including the methods used, the criteria for interpretation, and the sensitivity analyses planned. This documentation supports transparency and reproducibility. The Research Data Framework from the National Institute of Standards and Technology provides a framework for managing research data throughout its lifecycle, which can support documentation of analysis decisions.

Use of Reporting Guidelines

Following established reporting guidelines improves the quality and completeness of meta-analysis reporting. The EQUATOR Network is an international initiative that provides reporting guidelines for health research. The Experimental Design Assistant from the NC3Rs provides support for experimental design, which can help prevent bias at the study design stage.

Frequently Asked Questions

What is the difference between publication bias and selective reporting bias?

Publication bias refers to the phenomenon where studies with statistically significant results are more likely to be published than studies with null or nonsignificant results. Selective reporting bias operates within studies, where the likelihood that a specific result is reported depends on the magnitude or statistical significance of the effect. The tutorial on selective reporting bias in meta-analysis of dependent effect sizes explains that selective reporting bias arises when findings from primary research studies are incompletely reported. Both forms of bias can distort meta-analytic estimates, and the review of publication bias in clinical meta-analyses notes that statistical adjustment cannot recover information that was selectively generated, reported, or disseminated.

How many studies are needed to assess publication bias reliably?

Publication bias assessment methods generally require at least 10 primary studies for reliable performance. The study on publication bias adjustment methods notes that these methods may not be able to combat bias in small samples with fewer than 10 primary studies. The meta-meta-analysis from psychology and medicine found that the median number of effect sizes in homogeneous subsets was 6, which created challenging conditions for publication bias methods. When the number of studies is small, researchers should interpret publication bias assessments cautiously and consider methods like meta-analysis of non-affirmative studies that accommodate small meta-analyses.

What is the fail-safe N and how should it be interpreted?

The fail-safe N estimates the number of unpublished studies with null results that would be needed to overturn the meta-analytic conclusion. It provides an intuitive measure of the robustness of findings to publication bias. However, the fail-safe N has important limitations, including the assumption that unpublished studies have null effects. The review of publication bias modelling approaches discusses methods that can be employed as sensitivity analyses to assess the likely impact of publication bias. The fail-safe N should be reported alongside other publication bias assessments instead of as a standalone measure.

What are the main problems with meta-analysis related to publication bias?

The main problems with meta-analysis related to publication bias include overestimated effects, the suggestion of non-existing effects, and reduced credibility of research findings. The meta-meta-analysis from psychology and medicine found that publication bias yields overestimated effects and may suggest the existence of non-existing effects. Additionally, many meta-analyses fail to assess publication bias at all, and many that detect bias fail to adjust for it. The neurosurgery meta-analysis review found that 58.4% of meta-analyses either did not assess for publication bias or, if assessed to be present, did not adjust for it.

How does the trim-and-fill method work?

The trim-and-fill method is a nonparametric approach that adjusts the meta-analytic estimate for publication bias. It works by trimming the asymmetric studies from one side of the funnel plot, estimating the true center of the plot, and then filling in missing studies on the other side to create symmetry. The adjusted estimate is calculated from the complete set of observed and imputed studies. The Biometrics article classifies trim-and-fill as a funnel-plot-based method, and the neurosurgery meta-analysis review found that trim-and-fill was the method used in the meta-analyses that made corrections for publication bias.

What is the difference between Egger's test and Begg's test?

Egger's test is a regression-based approach that quantifies funnel plot asymmetry by regressing the standardized effect size on precision. Begg's test examines the rank correlation between effect sizes and their variances. The Biometrics article notes that Egger's regression intercept may be considered as a candidate measure for quantifying publication bias, but it lacks an intuitive interpretation. The neurosurgery meta-analysis review lists both tests as standard approaches for assessing publication bias. Begg's test is nonparametric and less sensitive to outliers but generally has lower statistical power.

What are selection models in publication bias analysis?

Selection models use weight functions to adjust the overall effect size estimate based on assumptions about the probability of publication given the study results. The Biometrics article distinguishes selection models from funnel-plot-based methods, noting that selection models use weight functions to adjust the overall effect size estimate and are usually employed as sensitivity analyses. The review of publication bias modelling approaches provides a comprehensive review of selection model methods. These methods require explicit assumptions about the selection process, and different assumptions can lead to different adjusted estimates.

How should publication bias results be reported in a meta-analysis?

Publication bias results should be reported in the methods, results, and discussion sections of a meta-analysis. The methods section should describe the planned assessment methods and criteria for interpretation. The results section should report the results of all planned assessments, including visual funnel plot inspection and statistical tests. When publication bias is detected, the results should include both uncorrected and adjusted estimates from sensitivity analyses. The discussion should interpret the results in the context of the study's limitations. The EQUATOR Network provides reporting guidelines that support complete and transparent reporting of health research.

Related Articles

• Statistical Analysis Methods
• Statistical Analysis Methods
• Statistical Analysis Methods
• Statistical Analysis
• Statistical Analysis

References and Further Reading

• Research Data Framework. National Institute of Standards and Technology.
• EQUATOR Network. EQUATOR Network.
• Experimental Design Assistant. NC3Rs.
• NCBI Literature Resources. National Center for Biotechnology Information.
• PubMed. National Library of Medicine.
• The World Federation of ADHD International Consensus Statement: 208 Evidence-based conclusions about the disorder.. Neuroscience and biobehavioral reviews, 2021.
• Systematic Analysis of Publication Bias in Neurosurgery Meta-Analyses.. Neurosurgery, 2022.
• Quantifying publication bias in meta-analysis.. Biometrics, 2018.
• Publication bias examined in meta-analyses from psychology and medicine: A meta-meta-analysis.. PloS one, 2019.
• Modelling publication bias in meta-analysis: a review.. Statistical methods in medical research, 2000.
• Assessing robustness to worst case publication bias using a simple subset meta-analysis.. BMJ (Clinical research ed.), 2024.
• Assessment of publication bias in meta-analyses of cardiovascular diseases.. Journal of epidemiology and community health, 2005.
• A Review of Meta-Analyses in Plastic Surgery: Need for Adequate Assessment of Publication Bias.. The Journal of surgical research, 2024.
• Estimating the change in meta-analytic effect size estimates after the application of publication bias adjustment methods.. 2023.
• Robust Bayesian Meta-Analysis: Model-Averaging Across Complementary Publication Bias Adjustment Methods. 2021.
• Correcting what cannot be corrected: rethinking publication bias analysis methods in clinical meta-analyses.. 2026.
• Robust Bayesian multilevel meta-analysis: Adjusting for publication bias in the presence of dependent effect sizes.. 2026.
• Addressing selective reporting bias in meta-analysis of dependent effect sizes: A tutorial in R.. 2026.
• Publication bias in meta-analysis: Prevention, assessment and adjustments. Psychometrika, 2007.
• Systematic review and meta-analysis.. Medicina Intensiva, 2017.
• Global incidence and characteristics of spinal cord injury since 2000-2021: a systematic review and meta-analysis. BMC Medicine, 2024.
• Assessment of funnel plot asymmetry and publication bias in reproductive health meta-analyses: An analytic survey. Reproductive Health, 2007.
• The prevalence and effect of publication bias in orthopaedic meta-analyses. Journal of Orthopaedic Science, 2011.

This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.