Meta-Analysis & Evidence Synthesis

Meta-Regression Services

When effects vary from study to study, the useful question is why. Meta-regression models how the effect size depends on study-level characteristics—year, dose, sample, setting, quality—turning unexplained heterogeneity into an explanation. We design and deliver meta-regressions that answer the “it depends” question rigorously.

Meta-regression is an extension of meta-analysis that models how the effect size varies with one or more study-level characteristics (moderators). Instead of a single pooled estimate, it estimates the relationship between a moderator and the effect—explaining part of the between-study heterogeneity and identifying the conditions under which an effect is larger or smaller.

Explains heterogeneity Continuous & categorical moderators Bubble plots & diagnostics Reproducible, journal-ready

Talk to a Strategist Back to Meta-Analysis

Sample output study (size = weight) fitted line

What meta-regression does

A meta-analysis produces one pooled effect; when studies genuinely disagree—when there is real heterogeneity—that single number hides more than it shows. Meta-regression addresses this directly by asking whether the differences between studies are systematic: does the effect get larger or smaller as some study-level characteristic changes?

It extends the meta-analysis model with one or more moderators—variables measured at the study level, such as publication year, average dose or intensity, sample characteristics, the setting or country, the design, or study quality. The output is a coefficient for each moderator (how much the effect changes as the moderator changes) and an estimate of how much of the between-study variance the moderators explain. In short, it turns “the effect varies” into “the effect varies because of this.”

When to use it

Use meta-regression when a meta-analysis shows substantial heterogeneity and you have a plausible, theory-driven reason for it that is captured by a study-level variable. It answers questions such as: has the effect changed over time? Is it stronger at higher doses? Does it hold in one setting but not another? Does study quality relate to the size of the effect?

It is closely related to subgroup analysis—in fact, comparing subgroups is meta-regression with a single categorical moderator—but meta-regression is more flexible: it handles continuous moderators, several moderators at once, and gives a modelled relationship rather than a bucketed comparison. When your moderator is continuous (dose, year, age), meta-regression is the right tool where subgroups would waste information.

At a glance

Subgroup analysis vs meta-regression

Two ways to explain heterogeneity
	Subgroup analysis	Meta-regression
Moderator type	Categorical (grouped)	Continuous or categorical
Multiple moderators	Awkward	Handled together
Continuous variables	Must be binned (loses info)	Modelled directly
Output	Effect per subgroup	Slope + variance explained
Best for	A few clear groups	Dose, year, age, quality, several factors

Methodology

Doing it credibly—and its limits

Meta-regression is powerful but easy to misuse, and honest practice respects its limits. The first is power: meta-regression needs a reasonable number of studies per moderator—fitting several moderators to a handful of studies overfits and produces unreliable coefficients. A common guideline is to be conservative about how many moderators the number of studies can support.

The second, and most important, is that meta-regression is observational at the study level. A moderator that predicts the effect is a correlation, not a proven cause—studies with a given characteristic may differ in other, unmeasured ways (this is aggregation or ecological bias). So a meta-regression finding is best read as “the effect is associated with this study-level feature,” not “this feature causes the effect to change.” And moderators should be pre-specified: testing many moderators after seeing the data is a form of p-hacking that manufactures false positives.

A significant moderator is an association, not a cause. Pre-specify your moderators, keep their number in proportion to the studies, and report meta-regression findings as study-level associations—not proof that the moderator drives the effect.

What we report

We deliver the moderator coefficients with confidence intervals, the residual heterogeneity and the proportion explained, a bubble plot for each continuous moderator, and the diagnostics needed to judge the model—all with reproducible code in R’s metafor.

How we work

How we deliver a meta-regression

Meta-regression sits within our wider meta-analysis and evidence-synthesis service, run on a full systematic-review workflow—so the moderators are analysed on a sound, reproducible review.

We begin with a registered protocol that pre-specifies the moderators and their rationale, a comprehensive search, and careful extraction of both effect sizes and the study-level moderator data. We then fit the meta-regression (mixed-effects), keeping the number of moderators in proportion to the evidence.

Reporting follows PRISMA standards, with moderators, model, and residual heterogeneity reported transparently.

You receive the moderator coefficients and intervals, the variance explained, bubble plots, residual-heterogeneity and sensitivity checks, and reproducible code and data—framed honestly as study-level associations. The result explains where and when the effect holds, defensibly.

Where we apply it

Meta-regression across Management & Allied Studies

Explaining why an effect varies is often the most valuable output of a synthesis—and we apply meta-regression across the disciplines we serve.

Management & Organisational Studies

Testing whether an intervention’s effect depends on firm size, industry, time period, or study design.

Economics & Public Policy

Modelling how a policy’s estimated effect varies with context, intensity, or evaluation period across studies.

Marketing & Consumer Research

Examining whether an effect strengthens with treatment intensity, medium, or sample characteristics.

Finance & Accounting

Relating effect sizes to market, period, or regulatory regime to explain cross-study variation.

Education & Learning Sciences

Testing dose-like relationships—programme length or intensity—and moderators such as grade level.

Health, Behavioural & Social Sciences

The classic setting for meta-regression on dose, duration, age, and study-quality moderators.

FAQ

Meta-regression: common questions

What is meta-regression?

Meta-regression is an extension of meta-analysis that models how the effect size varies with one or more study-level characteristics (moderators), such as year, dose, sample, or setting. Instead of a single pooled estimate, it estimates the relationship between a moderator and the effect—explaining part of the between-study heterogeneity and identifying when an effect is larger or smaller.

How is meta-regression different from subgroup analysis?

Subgroup analysis compares the pooled effect across categorical groups; it is effectively meta-regression with a single categorical moderator. Meta-regression is more flexible: it handles continuous moderators (dose, year, age), several moderators at once, and gives a modelled slope plus the variance explained—rather than binning a continuous variable and losing information.

Does a significant moderator prove causation?

No. Meta-regression is observational at the study level, so a moderator that predicts the effect is a correlation, not a proven cause—studies with a given characteristic may differ in other, unmeasured ways (aggregation or ecological bias). Findings should be reported as study-level associations, not as evidence that the moderator drives the effect.

How many studies do I need for meta-regression?

Enough to support the number of moderators you test. Fitting several moderators to a small set of studies overfits and yields unreliable coefficients, so the number of moderators should be conservative relative to the number of studies. With few studies, it is better to test one carefully chosen, pre-specified moderator than many.

Should moderators be chosen in advance?

Yes. Testing many moderators after seeing the data is a form of p-hacking that produces false positives. Credible meta-regression pre-specifies the moderators and their rationale in the protocol, keeps their number proportionate to the evidence, and reports all tested—so the findings reflect theory-driven questions, not a search for significance.

Heterogeneity you need to explain?

If your meta-analysis shows real between-study variation and you have a theory-driven reason for it, meta-regression can turn that heterogeneity into an explanation. We design and deliver it—pre-specified and honestly interpreted.

Scope a meta-regression Back to Meta-Analysis