Effect Size Types in Meta-Analysis Explained

Before you can pool anything in a meta-analysis, you need to decide what number you're actually pooling. Different outcome types call for different effect size types in meta-analysis, and choosing the wrong one, or switching inconsistently between studies, is a foundational mistake that undermines everything built on top of it.

Binary outcomes: odds ratio, risk ratio, and risk difference

When your outcome is binary (an event either happened or it didn't, such as recovery, mortality, or complication), you have three common effect size choices.

**Risk ratio (relative risk)** compares the probability of the event occurring in the intervention group versus the comparison group. It's generally the most intuitive to interpret, a risk ratio of 0.75 means the intervention group had 75 percent the risk of the comparison group, and it's usually the preferred default for randomized trials with a clearly defined follow-up period.

**Odds ratio** compares the odds of the event, not the probability, between groups. It's the standard choice in case-control studies, where risk ratios often can't be calculated directly due to how cases and controls are sampled, and it's also the output most logistic regression models naturally produce. The common pitfall here is treating an odds ratio as if it were a risk ratio when the event is common; the two diverge substantially as baseline event rates rise above roughly 10 percent, and conflating them overstates or understates the real effect.

**Risk difference (absolute risk reduction)** reports the raw difference in event probability between groups, in percentage points rather than a ratio. It's less commonly used as the primary pooled measure, but it's valuable for communicating clinical or practical significance, since a risk ratio of 0.5 means something very different in absolute terms depending on whether the baseline risk is 40 percent or 0.4 percent.

Continuous outcomes: mean difference and standardized mean difference

When your outcome is continuous (a measured quantity like blood pressure, pain score, or test performance), the choice depends on whether included studies used the same measurement instrument.

**Mean difference (MD)** is used when all included studies measured the outcome on the same scale, for example, if every study used the identical validated questionnaire. It reports the pooled difference in that scale's own units, which keeps the result directly interpretable.

**Standardized mean difference (SMD)**, commonly Cohen's d or Hedges' g, is used when studies measured the same underlying construct but with different instruments or scales. SMD expresses the effect in standard deviation units instead of the original scale, which allows pooling across studies that wouldn't otherwise be directly comparable, at the cost of a less intuitive interpretation for non-statistical readers. Conventional benchmarks (roughly 0.2 as small, 0.5 as moderate, 0.8 as large) are commonly cited but should be applied cautiously, since their appropriateness varies by field and outcome.

Time-to-event outcomes: hazard ratio

When your outcome is time-to-event (how long until death, recurrence, or another event occurs, accounting for varying follow-up periods and participants who leave the study before the event happens), the hazard ratio is the standard effect measure, typically extracted from a Cox proportional hazards model reported in the primary studies.

Getting this right in a mixed-outcome review

A single meta-analysis sometimes needs to report multiple effect size types across different outcomes, a risk ratio for a mortality outcome and an SMD for a quality-of-life outcome in the same review, for instance. Each needs its own forest plot and its own model decision (fixed-effect versus random-effects), and conflating them into a single narrative that treats all effect sizes as directly comparable is a common and avoidable error.

**[Get a quote](/systematic-reviews#meta-analysis-support)** for help selecting and pooling the right effect size types for your specific outcomes.