Meta-Regression

Updated October 2026 · reviewed for today

In short

Meta-regression is a method used in meta-analysis that relates the results of many studies to features of those studies, such as the population, setting or method, to explain why findings differ. The outcome is each study’s effect size; the explanatory variables are study characteristics. T. D. Stanley and Stephen Jarrell proposed it for economics in 1989, and it is now standard in medicine through guidance such as the Cochrane Handbook.

Table of Contents

Toggle

How it works

01

Define the effect

Choose one comparable effect measure, such as a mean difference or odds ratio, across studies.

02

Pick moderators in advance

Specify the study features you will test before seeing the results, to avoid data dredging.

03

Weight the studies

Weight by precision and allow for unexplained differences between studies, as Thompson and Higgins recommend (Thompson and Higgins, 2002).

04

Interpret with care

Associations across studies are observational and can differ from what holds within each study.

Where you see it today

When humans and AI work together

A meta-analysis of human-AI experiments published from 2020 to mid-2023 found that, on average, human-AI combinations did worse than the best of humans or AI alone (Vaccaro et al., 2024). Moderator analysis showed losses in decision tasks and larger gains in content-creation tasks, which is the kind of finding meta-regression is built for.

Why AI productivity studies disagree

In one field study of customer-support agents, an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and low-skilled workers (Brynjolfsson et al., 2023). Worker experience is a moderator; pooling such studies without coding it would hide the pattern.

Reading the AI evidence base

Studies of AI tools differ by task, model, user skill and how success is measured. A meta-regression across them can show which of these features drive the gains, instead of quoting a single average.

Guardrails from medicine

The Cochrane Handbook advises against meta-regression with fewer than ten studies and warns about aggregation bias, also called the ecological fallacy. The same limits apply to business and AI research.

Meta-regression vs related methods

Meta-analysis

What it doesPools effect sizes into an average

Main limitationCan hide why studies differ

Subgroup analysis

What it doesCompares pooled effects across categories of studies

Main limitationOne factor at a time; many tests raise false positives

Meta-regression

What it doesRelates effect sizes to several study features

Main limitationNeeds many studies; results are observational

Individual participant data meta-analysis

What it doesRe-analyses raw data from each study

Main limitationRequires access to original data

Narrative review

What it doesSummarises studies in words

Main limitationHard to weigh conflicting results

How to use it

• Before quoting an average effect, ask how much the studies disagree and why.
• List likely moderators, such as task type and user experience, before you look at the results.
• Do not run a meta-regression on fewer than about ten studies.
• Treat cross-study patterns as hypotheses to test, not proven causes.
• When vendors cite AI productivity numbers, ask which tasks and workers they came from.

The Business Engineer lens

AI productivity evidence varies widely by task and worker, and decisions built on one headline average misread it. The Business Engineer’s view of the productivity spectrum is, in effect, a moderator analysis: where AI helps, for whom, and under what conditions.

Read: AI & The Productivity Spectrum →Explore the platform →

Related: Meta-analysis · Simpson's paradox · Selection bias · Lurking variable · Cherry picking

Frequently asked questions

What is meta-regression?

A regression in which each data point is a study: the outcome is the study's effect size and the predictors are features of the study.

How is meta-regression different from meta-analysis?

Meta-analysis estimates an average effect across studies; meta-regression explains why effects differ between them.

How many studies do you need for meta-regression?

The Cochrane Handbook advises against it when a meta-analysis has fewer than ten studies.

What is aggregation bias in meta-regression?

A relationship seen across study averages may not hold, or may even reverse, within individual studies; this is also called the ecological fallacy.

Sources

• Stanley and Jarrell (1989), Meta-Regression Analysis: A Quantitative Method of Literature Surveys, Journal of Economic Surveys
• Thompson and Higgins (2002), How should meta-regression analyses be undertaken and interpreted?, Statistics in Medicine
• Cochrane Handbook for Systematic Reviews of Interventions, Chapter 10 (current version)
• Vaccaro, Almaatouq and Malone (2024), When combinations of humans and AI are useful, Nature Human Behaviour
• Brynjolfsson, Li and Raymond (2023), Generative AI at Work, NBER Working Paper 31161

The full guide

The complete explainer, cleaned up and reformatted for mobile.

Meta-regression is a powerful statistical technique used in the field of meta-analysis to explore and quantify the relationships between study characteristics or covariates and the effect sizes observed in multiple studies. It allows researchers to investigate how various factors may influence the overall results of a meta-analysis, providing a more nuanced understanding of the underlying associations.

The Foundations of Meta-Regression

To comprehend meta-regression fully, one must grasp several foundational concepts and principles:

1. Meta-Analysis: Meta-analysis is a statistical technique that combines the results of multiple independent studies to estimate an overall effect size or summary statistic. It is commonly used to synthesize evidence from various sources and draw more robust conclusions.
2. Effect Size: In meta-analysis, effect sizes are standardized metrics used to quantify the magnitude of an effect or association. Common effect size measures include Cohen’s d for mean differences and odds ratios for categorical data.
3. Heterogeneity: Heterogeneity refers to the variability in effect sizes across studies included in a meta-analysis. It can be caused by various factors, including methodological differences, population characteristics, or other sources of variation.
4. Moderation: Moderation refers to the influence of one or more variables (covariates) on the relationship between the independent and dependent variables. In meta-analysis, meta-regression examines how these covariates moderate the effect sizes across studies.

The Core Principles of Meta-Regression

To effectively conduct meta-regression, researchers should adhere to core principles:

1. Covariate Selection: Identify relevant covariates or study characteristics that may influence the effect sizes. These covariates can be continuous or categorical variables.
2. Meta-Regression Model: Specify a meta-regression model that describes how the covariates are related to the effect sizes. This model can take various forms, including linear, nonlinear, or mixed-effects models.
3. Effect Size Transformation: Transform the effect sizes and their variances as needed to meet the assumptions of the meta-regression model. Common transformations include log transformations for odds ratios or correlation coefficients.
4. Assumption Checking: Evaluate the assumptions of the meta-regression model, such as the linearity of relationships and the homoscedasticity of residuals.

The Process of Implementing Meta-Regression

Implementing meta-regression involves several key steps:

1. Data Collection and Study Selection

• Collect Studies: Gather a comprehensive set of studies that address the research question of interest. These studies should provide effect sizes and relevant covariate information.
• Study Selection: Apply inclusion and exclusion criteria to select studies that meet the criteria for the meta-analysis.

2. Effect Size Calculation

• Compute Effect Sizes: Calculate effect sizes for each study using the appropriate metric (e.g., Cohen’s d, odds ratios). Effect sizes should represent the relationship of interest.
• Variance Estimation: Estimate the variances or standard errors of the effect sizes. This step accounts for the precision of each study’s effect size estimate.

3. Meta-Regression Model

• Covariate Specification: Specify the covariates that will be included in the meta-regression model. These covariates can be study-level characteristics or external variables of interest.
• Model Estimation: Fit the meta-regression model, which relates the effect sizes to the covariates while considering the within-study variances.

4. Model Assessment

• Assumption Testing: Evaluate the assumptions of the meta-regression model, such as linearity, normality of residuals, and homoscedasticity.
• Model Fit: Assess the overall fit of the meta-regression model, including the goodness-of-fit statistics and the explained variance.

5. Interpretation and Reporting

• Interpret Coefficients: Interpret the coefficients of the covariates in the meta-regression model. These coefficients indicate the strength and direction of the associations.
• Heterogeneity: Examine whether the inclusion of covariates explains some of the observed heterogeneity across studies.
• Publication Bias: Address potential publication bias and its impact on the meta-regression results.

Practical Applications of Meta-Regression

Meta-regression finds practical applications in various fields:

1. Medicine and Healthcare

• Clinical Trials: Investigate how study characteristics (e.g., sample size, study design) influence treatment effects in clinical trials.
• Epidemiology: Explore how covariates (e.g., age, gender, comorbidities) moderate the associations between risk factors and health outcomes.

2. Social Sciences

• Education Research: Analyze how teaching methods or interventions interact with student characteristics to affect academic outcomes.
• Psychology: Investigate the factors that moderate the effectiveness of psychological interventions or therapies.

3. Environmental Sciences

• Environmental Impact: Examine how environmental factors and pollutants interact to influence health outcomes or ecological effects.
• Climate Change: Investigate how climate variables and human activities impact ecosystems and biodiversity.

4. Business and Economics

• Financial Markets: Explore how economic indicators and external events moderate the relationship between financial variables.
• Consumer Behavior: Analyze how demographic variables and marketing strategies interact to affect consumer preferences and choices.

The Role of Meta-Regression in Research

Meta-regression plays several critical roles in research and decision-making:

• Covariate Exploration: It allows researchers to explore the impact of covariates and study characteristics on the observed effect sizes, providing insights into potential sources of heterogeneity.
• Hypothesis Testing: Researchers can use meta-regression to test specific hypotheses about the relationships between covariates and effect sizes.
• Quantification of Effects: Meta-regression provides quantitative estimates of how changes in covariates are associated with changes in effect sizes, helping researchers understand the strength of these associations.
• Subgroup Analysis: Meta-regression can be used to conduct subgroup analyses, identifying subpopulations or conditions where the effect sizes differ significantly.

Advantages and Benefits

Meta-regression offers several advantages and benefits

1. Exploratory Power: It allows researchers to explore the role of covariates and study characteristics in explaining heterogeneity across studies.
2. Quantitative Insights: Meta-regression provides quantitative estimates of associations, offering a more precise understanding of how covariates influence effect sizes.
3. Enhanced Interpretation: It enhances the interpretation of meta-analysis results by considering the impact of covariates, making the findings more applicable and informative.
4. Heterogeneity Assessment: Meta-regression can help account for and explain some of the observed heterogeneity, leading to more accurate conclusions.

Criticisms and Challenges

Meta-regression is not without criticisms and challenges

1. Data Availability: Availability of data on relevant covariates may be limited, restricting the scope of meta-regression.
2. Ecological Fallacy: Meta-regression provides associations at the study level, and caution should be exercised when generalizing to individual-level relationships.
3. Risk of Overfitting: Including too many covariates in the model can lead to overfitting, reducing the model’s generalizability.
4. Publication Bias: Meta-regression may be affected by publication bias if studies with certain characteristics are more likely to be published.

Conclusion

Meta-regression is a valuable statistical technique that enhances the capabilities of meta-analysis by allowing researchers to investigate how various factors and covariates may influence the overall results of multiple studies. It provides a quantitative framework for exploring associations, testing hypotheses, and understanding the sources of heterogeneity. While it requires careful consideration of covariate selection and model assumptions, meta-regression remains an essential tool in research fields ranging from medicine to social sciences, facilitating a deeper understanding of the relationships within meta-analyzed datasets.