MerQuR+Statistics Glossary

Statistics Glossary → Standardized mean difference (SMD) and covariate balance

Standardized mean difference (SMD) and covariate balanceconcept

The difference between two group means expressed in standard deviation units; in propensity score analysis, a measure of how well balanced the groups are on a variable.

SMD = (x̄₁ - x̄₂) / s, where s is usually the square root of the average of the two group variances: √((s₁² + s₂²) / 2). For binary variables the proportions and their variances are used. The measure is unit-free, so imbalances in very different variables such as age, income and blood pressure can be compared on one scale. In meta-analysis, SMDs such as Cohen's d and Hedges' g serve as effect measures for combining studies that used different scales.

After propensity score matching or weighting, balance should be judged by the SMD, not by p values: the p value depends on sample size, and because matching shrinks the sample it can give a misleading impression that balance has been achieved. A common rule is |SMD| < 0.1. The SMD shows balance in the means only; the variance ratio and the shape of the distributions should be checked as well. It says nothing about unobserved confounders.

An everyday example

People who joined a smoking-cessation programme are compared with those who did not. Before matching, participants are on average 8 years older, and the standard deviation of age is about 12 years, so SMD ≈ 0.67. After propensity score matching the gap falls to 0.6 years (SMD ≈ 0.05); the two groups can be considered balanced on age.

In MerQur

'Balance threshold (|SMD|)' in Propensity Score Analysis: variables above this threshold are flagged and counted as imbalanced. In Meta-Analysis the standardized mean difference is used as an effect measure.

Analyses where this concept appears

Meta-AnalysisPropensity Score Analysis

Related concepts

Propensity score matching (nearest neighbour, caliper)