Mean Difference and Standardized Mean Difference

When the outcome is continuous, such as blood pressure, a pain score, or weight, the effect size is a difference in means, not a ratio. The raw mean difference is the treated-group mean minus the control-group mean, expressed in the original units, for example 'the drug lowered systolic blood pressure by 8 mmHg'. Because it is in familiar clinical units, it is easy to interpret, and you report it with a 95% confidence interval; an interval that includes 0 is not statistically significant. But sometimes you need to compare or combine studies that measured the same underlying concept on different scales, for example depression measured by three different questionnaires. You cannot simply average '5 points' from incompatible scales. The standardized mean difference (SMD) solves this by dividing the mean difference by the pooled standard deviation, turning it into a unitless number of standard deviations. Cohen's d is the best-known SMD, with rough benchmarks of 0.2 (small), 0.5 (medium), and 0.8 (large). This is why meta-analyses of continuous outcomes so often report an SMD: it is the common currency that lets results from different scales be combined. Two cautions are important. The SMD's small/medium/large benchmarks are arbitrary and context-free; a 'small' SMD can be clinically important and a 'large' one trivial, depending on the outcome. And because the SMD depends on the standard deviation, a study with an unusually small SD can show a large SMD for a clinically tiny difference. So prefer the raw mean difference, judged against the minimal clinically important difference, when a single familiar scale is used. In short, reach for the raw mean difference whenever one well-understood scale applies, because it speaks the language of practice, and reserve the standardized mean difference for the specific job of combining or comparing across different measurement scales.