Twenty studies test the same relationship; some find strong effects, some weak, one finds nothing. What does the evidence as a whole say? Meta-analysis answers that by statistically pooling the effect sizes from comparable studies into one weighted overall estimate — the closest social science gets to a verdict.
It starts with a systematic review
A meta-analysis is only as credible as the search behind it. You first run a full systematic review — protocol, databases, PRISMA screening — and only then pool the studies that report comparable effects. Pooling a convenience sample of papers isn't meta-analysis; it's averaging your own bias.
The currency: effect sizes
Every included study must yield a comparable effect size — commonly Cohen's d or Hedges' g for group differences, the correlation r for relationships, or odds/risk ratios for binary outcomes. Each estimate is weighted (larger, more precise studies count more) and combined.
Fixed effect vs random effects
The fixed-effect model assumes every study estimates one identical true effect — plausible only for tight replications. The random-effects model assumes true effects vary across contexts and populations, and is therefore the default in management and social-science meta-analyses. The choice changes weights, confidence intervals and interpretation, so justify it explicitly.
Heterogeneity — the interesting part
I² describes how much of the variation between studies reflects real differences rather than chance (25% low, 50% moderate, 75% high). High heterogeneity isn't failure — it's a research question: moderator analysis (subgroups, meta-regression) then probes why effects differ, often producing the paper's real contribution.
The two signature plots
- Forest plot — every study's effect and confidence interval, with the pooled diamond at the bottom; the single most information-dense figure in evidence synthesis.
- Funnel plot — effect size against precision; asymmetry suggests publication bias (small negative studies missing), probed further with Egger's test and trim-and-fill.
Software and reporting
R's `metafor` and `meta` packages are the research standard; CMA, JASP and RevMan are friendlier menu-driven options. Report to PRISMA plus meta-analytic essentials: model choice, weights, pooled estimate with CI, I², moderator tests and bias analyses. Our meta-analysis mentoring coaches the full pipeline from search to forest plot.
Pooling studies that aren't conceptually comparable — different constructs, incompatible designs — produces a precise estimate of nothing. Inclusion criteria and construct definitions decide whether your meta-analysis means anything at all.
Frequently asked
How many studies do I need for a meta-analysis?+
Technically two; credibly, most published meta-analyses pool 10+ effect sizes, and moderator analysis needs more. Fewer studies mean wider intervals and weaker claims — sometimes a systematic review without pooling is the honest choice.
Can I do a meta-analysis for my PhD in management?+
Yes — meta-analyses are increasingly common as thesis chapters and standalone papers in management research, especially where a relationship has 15+ empirical studies with extractable effect sizes.
phdguide's mentors are senior academics, former supervisors, statisticians and publication specialists with 25+ years of combined experience guiding MBA, MPhil and PhD scholars from topic to viva.
Ethical, compliant guidance: We provide academic support, mentoring, analysis, editing and structuring — not authorship. Your work stays compliant with university policies.