Effect size
Also known as Cohen's d, effect sizes
How big a difference or effect is, as a number research can compare.
At a glance
In plain words
Measured in spreads, not points, Small, medium, large are relative, Context over labels …
See all 4Examples from therapy research on this site
Almost nothing: d = 0.01, Very small: d = 0.04, At most small: d = 0.20 …
See all 6Connections
Common factors theory, Evidence-based practice in psychology and 6 others
See all 8
Also on this page: Key facts · Links · Sources · Related pages
- An effect size measures how large an effect is, and often its direction.
- Cohen's d, the most common one in therapy research, is the gap between two group averages divided by a standard deviation.
- Rough guide: 0.2 small, 0.5 medium, 0.8 large.
- Those labels are relative, so read a number in its context.
- Meta-analyses work by combining effect sizes across studies.
Why it matters
- Most claims about what works in therapy are stated as an effect size, usually d; reading one makes the research pages readable.
Key facts
- In short
- A quantitative measure of the magnitude and often direction of a phenomenon1
- Cohen's d
- The difference between two means divided by a standard deviation for the data1
- Rules of thumb
- Small 0.20, medium 0.50, large 0.80 (Cohen, 1988)1
- Extended scale
- Sawilowsky added very small (0.01), very large (1.20) and huge (2.0)1
- Used for
- Meta-analysis, which combines effect sizes from many studies1
In plain words
-
Measured in spreads, not points1
d says how many standard deviations apart two group averages are, so results on different questionnaires can be compared.
-
Small, medium, large are relative1
The terms are relative, not absolute; a small d can matter a lot in one field and little in another.
-
Context over labels1
Researchers are advised to judge an effect's real importance in a meaningful context; using Cohen's labels indiscriminately can mislead.
-
Another way to say it: the chance one beats the other2
The common language effect size is the probability that an observation from one group beats one from the other; for heights, a random man is taller than a random woman 92% of the time.
Examples from therapy research on this site
-
Removing a treatment's components in dismantling studies costs almost nothing, by the common factors evidence.
-
How closely a therapist follows the treatment manual barely predicts outcome.
-
The largest differences found between bona fide treatments, in the dodo bird debate.
-
How much outcomes depend on which therapist you see, in trials; in everyday settings it reaches 0.55.
-
How strongly the therapeutic alliance goes with outcome.
-
One session of chair work deepened emotional experiencing more than empathic responding alone, in a 2023 meta-analysis; it measured experiencing in the session, not symptoms.
Connections
Related
- Common factors theoryTopicits evidence is stated in effect sizes
- Evidence-based practice in psychologyTopic
- The alliance in adult psychotherapy: A meta-analytic synthesisPaper
- Comparative efficacy and tolerability of medications for attention-deficit hyperactivity disorder in children, adolescents, and adults: a systematic review and network meta-analysisPaperReports results as standardised mean differences
- The Efficacy of Cognitive Behavioral Therapy: A Review of Meta-analysesPaperReports results as small, medium and large effects
- How important are the common factors in psychotherapy? An updatePaperReports its evidence as Cohen's d
- A meta-analysis of the contribution of eye movements in processing emotional memoriesPaperReports its results as Cohen's d
- The efficacy of psychodynamic psychotherapyPaper
Links
Sources
Last checked against its sources today
Is something on this page wrong or out of date? Tell us
The line under a date fills as time passes: empty when new, then one equal share each for a day, a week, a month and a year. For episodes and other items it is full at 2 years; for "last checked", at a year. A striped line means the date is known only to the month or year.