Dance for Depression: The Mislabelled Effect Size, and What the Review Actually Reports

The Cochrane review of dance movement therapy for depression includes 3 studies and 147 participants, reports a whole sample result of SMD -0.67 (95% CI -1.40 to 0.05) at very low quality, and concludes in its authors’ own words that the evidence “does not allow any firm conclusions to be drawn regarding the effectiveness of DMT for depression”. The figure that circulates in place of that one, usually written as SMD -7.33, is not an effect size at all, and this page is mostly about what it actually is.

I met the number in a funding bid. A partner organisation sent me a draft to comment on, and the evidence paragraph contained a sentence about a large effect on depression with the figure attached and no citation. I asked where it came from, expecting a report, and got back a link to another organisation’s website, which cited a third. It took an afternoon to reach the review itself, and when I did, the number was there in the abstract, in exactly the form everybody was quoting, with a sentence immediately after it that nobody had passed on. The person who sent me the draft had done nothing wrong. She had found the number in a source that looked reliable, because it was reliable, and had trusted the label printed beside it.

Where a claim about effect appears below it carries a tier (Tier 1 supported, Tier 2 promising but limited, Tier 3 practice and anecdote), explained on reading arts in health research.

What the review reports, whole sample first

Three trials, 147 participants, no reliable effect on depression, very low quality evidence.

The review covers 3 studies totalling 147 participants, of whom 107 were adults and 40 were adolescents. Across the whole sample it reports SMD -0.67 (95% CI -1.40 to 0.05) at very low quality, and describes that as no reliable effect on depression¹.

Take that interval seriously, because it is doing more work than the point estimate. It runs from -1.40, which would be a large benefit, to 0.05, which is a very slight harm. The data are compatible with all of that. A confidence interval of that width, from three small trials, at very low quality, does not settle the question in either direction. It tells you that the studies done so far are not big enough or good enough to detect what is going on.

The authors’ own conclusion is the sentence to carry away: “The low-quality evidence from three small trials with 147 participants does not allow any firm conclusions to be drawn regarding the effectiveness of DMT for depression.” They go on to call for larger trials of high methodological quality, with economic analyses and acceptability measures, across all age groups¹.

The review has not been updated. It is still at its 2015 version, so this is the current Cochrane position on the question rather than a superseded one. That is worth checking rather than assuming, because the opposite trap is common in this field: the dementia music review was updated in 2025 and reached materially different conclusions from its predecessor, so any source quoting the older numbers is out of date. See music therapy for dementia.

The number in circulation, and what it really is

The figure quoted almost everywhere is -7.33, and it is quoted as a standardised effect size. It is not one.

Here is where it comes from. The review’s abstract contains a subgroup analysis of adults only, across 2 of the 3 studies and 107 participants, and reports -7.33 (95% CI -9.92 to -4.73) under a standardised mean difference label. In the same sentence, the abstract states that this failed to meet clinical significance¹. So the number is real, it is printed in a Cochrane abstract, and the caveat sitting immediately beside it is the part that never travels with it.

Now the arithmetic problem, which is the useful thing to take away from this page. A standardised mean difference expresses the gap between two groups in standard deviations. The conventional rough reading is that 0.2 is a small effect, 0.5 moderate and 0.8 large, and values much above 1.5 are unusual in psychological research. A standardised mean difference of -7.33 would mean the two groups differed by more than seven standard deviations. Nothing in psychological or behavioural research produces that. If it did, the effect would be so overwhelming that no trial would be needed to see it.

What the number actually is: a raw mean difference on a questionnaire subscale, specifically the depression subscale of the symptom checklist used in those adult studies, expressed in points on that scale rather than in standard deviations, and printed with the wrong label. The interval, -9.92 to -4.73, is in the same raw units. Once you know that, the figure stops being astonishing and becomes ordinary, which is exactly what it should have been all along.

So the instruction for this site, and the suggestion for anybody else: name the -7.33 as a mislabelling, and never repeat it as an effect size. It cannot be corrected by adjusting the wording around it, because the number itself belongs to a different scale from the one its label implies. What replaces it is the whole sample figure, which is SMD -0.67 (95% CI -1.40 to 0.05), very low quality, three studies, 147 participants.

Whole sample first, subgroup second

Even setting the mislabelling aside, quoting an adults only subgroup ahead of a null whole sample result reverses the honest order.

This is a house rule on this site and it is not a stylistic preference. When a trial or a review reports a null overall and a positive subgroup, leading with the subgroup can be accurate sentence by sentence and still leave a reader with an entirely false picture of what was found. Subgroups are smaller, so they are noisier. They are more numerous than people realise, because several can be examined from one dataset. And they are selected after the data are in far more often than reports admit.

The cleanest illustration of the rule in this field is not about dance. The 2018 three arm trial of singing for postnatal depression found no significant effect across the whole sample (p=0.16). The result that gets quoted is a moderate to severe subgroup at week 6, which was no longer significant by week 10, and singing did not beat the trial’s active comparator of creative play². The fair summary is faster recovery among the more severely affected rather than a better endpoint, which is Tier 2 on the subgroup and early timepoint criteria. Notice how much less exciting that sentence is than the one in circulation, and how much more useful.

The other outcomes in the review

Little that supports a claim, and worth listing so that nobody has to take my word for the summary.

• Social functioning: one study showed a large positive effect, at very low quality evidence.
• Quality of life: no significant effect.
• Self esteem: no significant effect.
• Dropout: one adult study reported a non significant difference, at low quality evidence¹.

Each of those rests on one or two small trials. A single very low quality signal on social functioning is a reason to run a larger trial; it is not a claim to put in a leaflet, and it certainly is not a basis for telling somebody that dance improves their social life. It belongs in the same category as the rest of the review: an indication of where to look next.

How to grade this, and which kind of Tier 3 it is

Tier 3, and specifically the kind where nobody has looked properly, not the kind where a good trial reported no difference.

This distinction runs through the whole site and it carries most of the meaning of the label. Tier 3 covers two very different situations. In the first, a well conducted controlled study exists and found nothing, which is a positive finding of absence: that is the position for group art therapy in schizophrenia, described on art therapy for psychosis, and for music and agitation in dementia, where the 2025 review reports SMD -0.05 (95% CI -0.27 to 0.17) at moderate certainty³, a narrow interval around zero from evidence the reviewers trusted. In the second, the studies are too few, too small and too weak to detect anything, which is an empty file.

Dance and depression is the second. Three trials, 147 people, very low quality, and an interval spanning a large benefit and a slight harm. A Tier 3 label here is not a verdict that dance does not help depression. It is a statement that the research needed to answer the question has not been done, and the reasons for that have little to do with whether the activity helps: dance is impossible to blind, the funding runs in short cycles, the outcomes people care about are hard to measure, and the sector’s money goes into running classes rather than into trials.

For calibration, this is what a small real effect looks like when it has been properly measured. The same 2025 dementia review reports depressive symptoms at SMD -0.23 (95% CI -0.42 to -0.04) at moderate certainty from 30 studies and 1,720 participants randomised³, which is Tier 1 and is roughly a fifth of a standard deviation. A genuine, checkable, modest finding sits at about a fifth of a standard deviation. The mislabelled dance figure was supposedly at seven.

What would change this page

Stated so it can be held against, in the same way the site asks of everybody else.

• A larger trial, or several. The review’s own authors ask for exactly this, and 147 participants across three studies is not enough to detect the size of effect that would plausibly be there.
• An active comparator. A dance intervention compared against another supervised group activity of similar length and sociability, rather than against a waiting list, so that the dancing can be separated from the attention, the company and the weekly structure.
• Consistent outcome measures, so that trials can be pooled without straining, and reported in the units the instrument uses with the label that matches.
• A certainty rating that gets above very low, which mainly means blinded outcome assessment and adequate size.

If those arrive, the tier on this page changes and the change gets recorded on the page rather than made quietly. Until then, the figure quoted here is SMD -0.67 with its interval, and the -7.33 appears only as an example of a mislabelling.

For anybody weighing up a class

Nothing on this page argues against dancing, and it would be a poor reading of it to take it that way.

Weak evidence is a statement about the research literature. It is not a statement about whether an hour of movement, music and company is worth your Tuesday. People who go to these classes generally keep going, which is a fact worth something even though it is not evidence of an effect on depression. What the evidence does not support is treating a class as a treatment, or as a reason to change anything about the treatment you already have.

Two practical points. If you want the clinical version rather than the community version, the profession and how to check a practitioner are covered on dance movement psychotherapy, and the accredited register is the place to search⁴. If you want the community version, finding a dance for health class sets out how these are organised and what to ask, and dance and health covers why exercise, therapy and a social night out keep getting treated as one thing. The equivalent question on the music side, where the evidence is differently shaped but not stronger for adult mental health, is on music therapy for depression and anxiety, and the one movement finding in this field that reaches Tier 1 is rhythmic auditory stimulation, at 11.34 m/min of gait velocity from 9 trials and 268 participants⁵, which is about walking after stroke and nothing to do with mood.

One more caution, and it is the one that matters most on a page about depression. If your mood is getting worse, or you are having thoughts of harming yourself, that belongs with your GP or your mental health team today. Nothing here is a reason to alter, delay or decline any part of anybody’s treatment, and no arts activity substitutes for one.