Most people asking which exercise is best for their mental health are asking a question the research has already answered, and then discarding the part of the answer that would actually help them.
In 2024, a team led by Michael Noetel published a network meta-analysis in the BMJ pooling 218 randomised trials and 14,170 participants, all of them people meeting clinical criteria for major depression. The design let them compare exercise modes against each other and against usual care, psychotherapy and antidepressants, rather than testing one activity in isolation. It is one of the largest attempts anyone has made to answer the “which one” question directly.
Walking or jogging produced a moderate reduction in depression. So did yoga. So did strength training. Mixed aerobic exercise and tai chi landed close behind. Several very different activities clustered in roughly the same band, and across the board the effects scaled with how hard the exercise was prescribed.
That is usually where the coverage stops, and it is the least useful place to stop.
The finding underneath the finding
Noetel and colleagues also reported which modes people tolerated. Yoga and strength training came out best on that measure. Not the highest effect sizes. The ones participants were most willing to keep doing.
Tolerability is treated as a footnote in most write-ups because it sounds like a comfort variable rather than an outcome variable. It is not. An intervention that works on paper and gets abandoned in week two produces nothing. An intervention with a slightly smaller effect that survives three months produces the effect three months of times over. Adherence is not a soft consideration sitting next to the science. It is the multiplier on it.
This is where most movement plans quietly fail. The plan gets designed for the person you intend to be by autumn. Six sessions a week, real intensity, no missed days. Then a real week arrives, with its travel and its sick child and its deadline, and the plan lasts about nine days. And the collapse gets filed under character. Lazy. Undisciplined. Bad at follow-through.
The plan was never sized for the life it had to fit into. That is a design problem, and design problems have design fixes.
What this evidence does not say
Two limits deserve stating plainly, because a body of research this large invites overreach.
First, this was measured in people meeting clinical criteria for depression. It is evidence about what movement does in that population under trial conditions. It is not a prescription for any individual, and it is not a reason for anyone to change how they are managing a diagnosed condition without talking to the clinician who is managing it with them.
Second, the authors are candid about the quality of what they pooled. Only one of the included trials met the strictest standard for low risk of bias, and their formal confidence rating in most of the comparisons was low. The direction of the finding is well supported. The precise size of it is not settled, and anyone quoting a decimal place from this literature as though it were fixed is quoting further than the data reaches.
Both of those caveats make the practical takeaway more robust rather than less, because the takeaway never depended on the decimal place.
How to size a dose you will actually keep
The usable shape of this evidence is narrower than the headlines and far more actionable.
Intensity matters, so the version you choose should be genuinely effortful for you rather than nominally active. But mode appears to matter much less than the field’s marketing implies, which means you are free to pick on a different criterion entirely: what you will still choose on a bad day, in bad weather, at the end of a week that went badly.
That is the question worth sitting with. Not what is optimal. What is repeatable at your current capacity, in your current schedule, with your current energy. If the honest answer is twenty minutes of something unglamorous three times a week, that is a real intervention. If the honest answer is a programme you have restarted four times this year, that is not a plan, it is an intention with a calendar attached.
The point of understanding the mechanism is never the mechanism. It is that a person who knows their plan failed on tolerability rather than on willpower can fix the right thing. They stop recruiting more discipline for a plan that was mis-sized, and start sizing a plan they will not need to recruit discipline for.
The dose you repeat beats the dose you admire. That is most of the finding, and nearly all of the value.
If this is the kind of design question you are working through, there is more on how I approach it at adamgrimmcoach.com.
