Contents
Academy · Research Data, Analytics & AI · Interpreting SMDs
Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap
Cohen’s conventional thresholds — an SMD of 0.2 as small, 0.5 as medium and 0.8 as large — correspond to concrete, verifiable quantities of approximately 15%, 33% and 47% non-overlap between two distributions. This calculation underlies the rule of thumb but does not by itself establish the practical importance of an effect.
A standardized mean difference (SMD) is a mean difference between two groups divided by a pooled standard deviation, which rescales the comparison into standard-deviation units so that studies measuring the same construct on different original scales can be combined1. The conventional rule of thumb for interpreting the resulting number, introduced by Jacob Cohen, labels an SMD of around 0.2 as a small effect, around 0.5 as medium, and around 0.8 as large2. Understanding what the number represents geometrically, in terms of distributional overlap, is a separate and more useful skill than memorizing the three labels or applying them mechanically, because it allows a reader to translate an abstract SMD value into an intuitive sense of how different two groups are.
Learning objectives
- The numerical values Cohen assigned to small, medium and large SMDs, and their origin
- How an SMD value corresponds to the percentage of non-overlap between two distributions
- The practical limitations of applying this rule of thumb mechanically without field-specific context
Common misconceptions
- “An SMD rule of thumb is merely a label; the exact value behind ‘small’, ‘medium’ and ‘large’ is of little importance.” Cohen's own derivation ties each label to a specific, calculable quantity: an SMD of 0.2 corresponds to roughly 85% overlap between two normal distributions (about 15% non-overlap), an SMD of 0.5 to roughly 67% overlap (about 33% non-overlap), and an SMD of 0.8 to roughly 53% overlap (about 47% non-overlap), so the labels are shorthand for a concrete geometric relationship, not an arbitrary naming convention.
Origin of the three numerical benchmarks
Cohen's 1988 text, Statistical Power Analysis for the Behavioral Sciences, introduced 0.2, 0.5, and 0.8 as conventional small, medium, and large benchmarks for a standardized mean difference, derived originally by considering what magnitude of difference researchers in the behavioral sciences at the time typically observed, and by extension, what sample sizes would be needed to detect each magnitude with reasonable statistical power2. Cohen himself described these specific numbers as a necessary convention for researchers who had no better basis for choosing a target effect size when planning a study, not as a scientifically derived boundary between categories2. The Cochrane Handbook's own discussion of standardized mean difference references these conventional thresholds as a familiar interpretive aid available to review authors, while also noting that interpretation should ultimately be grounded in the specific clinical context of the outcome being measured1.
Expressing an SMD as distributional overlap
An SMD has a direct geometric interpretation when the two groups being compared are each assumed to be approximately normally distributed with equal variance: it describes how far apart the two distributions' centers are, measured in shared standard-deviation units, which determines how much the two distributions overlap. Cohen's own tabulated calculations, reproduced in later effect-size reference guides, show that an SMD of 0.2 corresponds to about 85% overlap between the two distributions (roughly 15% non-overlap), an SMD of 0.5 corresponds to about 67% overlap (roughly 33% non-overlap), and an SMD of 0.8 corresponds to about 53% overlap (roughly 47% non-overlap)23. A related and often more intuitive translation is the “probability of superiority”, the probability that a randomly selected person from the higher-scoring group scores higher than a randomly selected person from the other group, which is roughly 56% for an SMD of 0.2, 64% for 0.5, and 71% for 0.8, compared with the 50% baseline expected if there were no difference between groups3. The small, medium and large labels are shorthand for these overlap and superiority figures, which show why even a “large” conventional effect leaves considerable overlap between the two groups’ distributions rather than implying that one group categorically outperforms the other.
Limitations of applying the rule of thumb
Lakens’s practical primer on reporting effect sizes cautions that, despite their mechanical precision, these benchmarks were never intended as a universal yardstick for clinical or practical importance, and that researchers are better served by interpreting a given SMD against previously observed effects within the same research area than against Cohen’s generic scale4. The overlap and superiority-probability translations described above remain useful regardless of field, because they describe exactly what the number represents geometrically; what changes by field and outcome is whether that degree of overlap or superiority corresponds to something a patient, clinician, or decision-maker would consider meaningful. In practice, careful application of the rule of thumb involves reporting the SMD and its confidence interval, translating the number into overlap or superiority-probability terms where this aids a reader’s understanding, and reserving the final judgment of practical importance for a field-specific comparison rather than the generic label alone.
Worked numerical example
Consider a meta-analysis that pools a treatment-arm mean anxiety score of 12.0 against a control-arm mean of 15.0, with a pooled standard deviation across both arms of 6.0. The SMD is calculated as the mean difference, -3.0, divided by the pooled standard deviation, 6.0, giving an SMD of -0.5, which falls on Cohen’s conventional “medium” benchmark2. Using the overlap translation described above, this -0.5 SMD corresponds to roughly 67% overlap between the treatment and control groups' anxiety-score distributions, and a probability of superiority of about 64%, meaning a randomly selected treated patient has roughly a 64% chance of scoring lower on anxiety than a randomly selected untreated patient3. Reporting the result in this form — “a medium effect (SMD -0.5), corresponding to roughly 64% probability that a treated patient scores lower than an untreated patient” — conveys considerably more to a non-statistical reader than the bare label “medium effect”, while remaining fully consistent with the same underlying number.
Interpreting negative and near-zero SMDs
The sign of an SMD depends entirely on which group is subtracted from which and which direction on the outcome scale represents improvement, so a negative SMD is not inherently a "worse" or "better" result; it is only interpretable once a review states, in its methods, which direction favors which group1. A review pooling a symptom-severity scale, where a lower score is better, might report a negative SMD favoring treatment, while a review pooling a quality-of-life scale, where a higher score is better, would report a positive SMD for the same direction of clinical benefit; Cochrane Handbook guidance recommends stating this orientation explicitly in the methods and again in any plain-language summary, since a reader unfamiliar with the specific scale cannot otherwise tell whether a negative number is good or bad news1. An SMD close to zero, meanwhile, should not automatically be read as proof of no effect; as with any point estimate, its confidence interval should be examined to determine whether a clinically important effect in either direction remains statistically compatible with the data, the same caution that applies to any near-null point estimate regardless of the effect measure that produced it. For the same reason, a results table or forest plot that reports an SMD should state the outcome scale’s direction of benefit in a footnote or legend wherever space allows, since a reader moving quickly through several forest plots in the same review, each measuring a different outcome, cannot be expected to remember which direction favors treatment for every scale without an explicit reminder each time. The same principle applies when an SMD is converted into the overlap or probability-of-superiority terms described earlier in this article: the conversion itself does not carry directional information, so a stated “64% probability that treatment scores better than control” must specify what “better” means on that scale as plainly as the SMD’s own sign does; otherwise, a translation intended to aid understanding may introduce a second source of confusion rather than resolving the first. A short glossary line attached to the first forest plot in a review, stating once which scales run in which direction and which outcomes were recoded for consistency before pooling, is a low-cost means of establishing this context for an entire report rather than re-explaining direction for every individual outcome, provided that the same orientation is then maintained throughout the document. Checking this consistency is a quick, mechanical task that should form part of the final review of any completed report: every forest plot should be scanned in order to confirm that an improvement always sits on the same side of the null line, and any outcome pooled with the opposite orientation should be recoded before the report is released, since a single inconsistently oriented outcome among a dozen correctly oriented ones is precisely the kind of error a reader is unlikely to detect but a careful final check will identify.
Key points
Key points
Cohen’s small, medium and large SMD benchmarks (0.2, 0.5 and 0.8) correspond to specific, calculable degrees of distributional overlap of approximately 85%, 67% and 53%. This correspondence is useful for developing an intuitive understanding of what an SMD represents, but the labels were never intended as a universal measure of practical or clinical importance.
Practical application
Practical application
- Report the SMD and its confidence interval, not only the categorical label. The value itself, with its interval, conveys more information than conversion to ‘small’, ‘medium’ or ‘large’ alone.
- Express the SMD in terms of overlap or probability of superiority for a non-specialist audience. These translations make the magnitude concrete and remain valid regardless of field, unlike the practical meaning of the small, medium and large labels.
- Ground the final judgment of importance in field-specific comparisons. The SMD should be compared with previously observed effects for the same outcome and population rather than judged against Cohen’s generic convention alone.
Frequently asked questions
Approximately 0.2, which Cohen's own calculations show corresponds to roughly 85% overlap between the two compared distributions.
No. An SMD of 0.8 still corresponds to roughly 53% overlap between the two distributions and a probability of superiority of about 71%, meaning a sizable share of both groups' distributions still coincide.
These translations describe exactly what the SMD number represents geometrically and hold across fields, whereas the practical or clinical meaning of the same label can differ substantially from one outcome to another.
No. Cohen described them as a necessary convention for researchers lacking a better basis for choosing a target effect size, not as empirically derived category boundaries.
References
- Higgins JPT, Li T, Deeks JJ. Chapter 6: Choosing effect measures and computing estimates of effect. Cochrane Handbook for Systematic Reviews of Interventions. read 10 October 2026. View on training.cochrane.org
- Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Lawrence Erlbaum Associates; 1988. read 10 October 2026. View on routledge.com
- Magnusson K. Interpreting Cohen's d effect size: an interactive visualization. read 10 October 2026. View on rpsychologist.com
- Lakens D. Calculating and reporting effect sizes to facilitate cumulative science: a practical primer for t-tests and ANOVAs. Frontiers in Psychology. 2013;4:863. read 10 October 2026. View on Frontiers
- Sullivan GM, Feinn R. Using Effect Size-or Why the P Value Is Not Enough. Journal of Graduate Medical Education. 2012;4(3):279-282. read 10 October 2026. View on JGME
- Deeks JJ, Higgins JPT, Altman DG. Chapter 10: Analysing data and undertaking meta-analyses. Cochrane Handbook for Systematic Reviews of Interventions. read 10 October 2026. View on training.cochrane.org
- Hedges LV, Olkin I. Statistical Methods for Meta-Analysis. Academic Press; 1985. read 10 October 2026. View on elsevier.com
- Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. John Wiley & Sons; 2009. read 10 October 2026. View on wiley.com
- Schünemann HJ, Higgins JPT, Vist GE, et al. Chapter 15: Interpreting results and drawing conclusions. Cochrane Handbook for Systematic Reviews of Interventions. read 10 October 2026. View on training.cochrane.org
- Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924-926. read 10 October 2026. View on BMJ
Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 10). Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap. Pubrica. https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/
Pubrica Academy. “Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap.” Pubrica, 10 Oct. 2026, pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/.
Pubrica Academy (2026) Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap. Pubrica. Available at: https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/ (Accessed: 15 October 2026).
Pubrica Academy. “Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap.” Pubrica. October 10, 2026. https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/
Pubrica Academy, ‘Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/> accessed 15 October 2026.
Pubrica Academy. Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/
Pubrica Academy, “Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/
@misc{/la_smd_rules_of_thumb_interpretation_2026,
author = {{Pubrica Academy}},
title = {Interpreting Standardized Mean Differences (SMDs): Cohen’s Rules of Thumb and Distributional Overlap},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/understanding-standardized-mean-differences-smds-effect-sizes/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
SMD Interpretation Note
This tool supports the interpretation of an SMD result for a non-specialist audience.
[ ] Reported the SMD with its confidence interval [ ] Noted the Cohen's small/medium/large label as a reference point only [ ] Translated the SMD into percentage overlap or probability-of-superiority for intuitive framing [ ] Compared the SMD against previously observed effects for this specific outcome/field [ ] Avoided presenting the categorical label as a measure of clinical importance on its own
Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica’s biostatistics team expresses a standardized mean difference in terms of overlap and field-appropriate importance, so that a report does not leave readers with an unexplained small, medium or large label.