Does Therapy Actually Work? What the Research Says

12 min read · Fact-checked

You’ve had the tab open a while now. A therapist directory, filtered down to four names, each of them $150 an hour, each with a waitlist. And the thought that keeps stopping you isn’t really about money or time. It’s simpler and more uncomfortable than that: does this do anything, or am I about to pay someone to agree with me once a week?

That’s a fair question, and most articles answer it badly. They quote one reassuring percentage and move on. So here’s the version with the caveats left in: what the outcome research shows, how much of it survives contact with real clinics, how many people get worse, and what any of it means for the decision sitting in your open tab.

What researchers mean when they say therapy “works”

Outcome research doesn’t measure happiness. It measures change on a symptom scale, usually against a comparison group, and it counts you as improved when your score moves further than measurement noise can explain. That threshold has a name, clinically significant change, and it’s deliberately conservative.

This matters, because “does therapy work” and “will I feel better” aren’t the same question. The research can tell you the odds that someone like you moves out of the clinical range on a questionnaire. It can’t tell you whether your particular Tuesday evenings will be worth $150.

The number most people have heard, and where it comes from

The foundational study is nearly forty years old and still the clearest thing we’ve got on how much therapy it takes. Howard and colleagues pooled 15 datasets covering 2,431 patients across more than 30 years, then fitted a curve to the relationship between number of sessions and probability of improvement (Howard et al., 1986).

By eight sessions, roughly 50% of patients were measurably improved. By 26 sessions, about 75%. By a year of weekly work, around 85%. The curve is steep early and flattens later, which is the part people usually miss: most of the movement happens in the first couple of months, and each extra session buys less than the one before it.

One finding in that paper is stranger and more useful than the headline. Between 10% and 18% of patients showed measurable improvement before their first session, simply as a function of having made the appointment. The authors put it down to the relief of knowing help is coming. If you’ve ever felt slightly better the day after booking something you’d been avoiding, that’s the effect, and it’s big enough to show up in the data.

Now the caveats, because they’re substantial

Howard’s samples were mostly psychodynamic and interpersonal therapy, from an era before most of the treatments people are offered today existed. None of the studies were behavioral. The curves also confound therapy with the simple passage of time, which the authors say plainly.

The modern equivalent is Cuijpers and colleagues’ 2016 meta-analytic update, covering 144 trials of cognitive behavioral therapy for depression and three anxiety disorders. Headline effects were large: g=0.75 for major depression, rising to g=0.88 for social anxiety disorder (Cuijpers et al., 2016).

Then the corrections start. Adjusting for publication bias pulled the depression effect down to g=0.65 and generalized anxiety to g=0.59. Nearly half the depression comparisons used waiting-list control groups, which reliably produce bigger effects than comparing therapy against normal care. The authors’ own conclusion is blunt: because so few high-quality trials exist, the true size of the effect is still uncertain.

That isn’t a debunking. An effect of g=0.65 is meaningful, comparable to plenty of treatments we accept without argument in physical medicine. It’s a correction to the idea that the evidence is settled and enormous.

What happens outside the trial

Here’s the finding that reframes this whole question. Hansen, Lambert and Forman compared the dose-response literature against what happens in ordinary clinics. In carefully controlled trials, between 57.6% and 67.2% of patients improved within an average of 12.7 sessions. In a national database of more than 6,000 people in real-world treatment, the average number of sessions attended was fewer than five, and the improvement rate was about 20% (Hansen et al., 2002).

Read those two numbers next to each other. The gap isn’t mainly a story about trial therapists being better. It’s a story about dose. People in routine care were leaving before they’d received the amount of therapy the research says the average person needs to shift.

Their review also puts the consensus estimate at 13 to 18 sessions for half of patients to improve, which is higher than Howard’s eight and probably closer to what you should expect walking into a clinic today.

So what’s doing the work?

If therapy helps, the obvious next question is which therapy. The evidence here is deflating for anyone hoping for a clean winner. Across decades of comparative trials, differences between established treatments stay small relative to what those treatments share.

The best-evidenced single ingredient is the working relationship. Flückiger and colleagues pooled 295 independent studies covering more than 30,000 patients and found a correlation of r=.278 between the quality of the alliance and how treatment turned out (Flückiger et al., 2018). That’s a moderate association, it held across theoretical orientations, and it showed no sign of publication bias. Pulling it to zero would take more than a thousand hidden null studies.

One caveat you won’t usually see mentioned: this is a correlation. Part of what looks like a good alliance producing improvement may be improvement producing a good alliance. Feeling better makes it easier to like your therapist.

Wampold’s contextual model gathers this and related findings, empathy, expectations, cultural fit, therapist differences, into the argument that what therapies share explains more of the benefit than the branded techniques do (Wampold, 2015). It’s still contested. But it’s why our guide to the major types of therapy opens by telling you the type matters less than you’d think.

Which therapist you get isn’t a rounding error

A systematic review of therapist effects across dozens of studies found the individual therapist accounts for about 5% of the variation in outcomes on average, with a range from 0.2% to 29% (Johns et al., 2019). Five percent sounds small until you notice it’s roughly what many specific techniques manage.

The review found something else worth acting on: therapist effects are larger when clients arrive with more severe difficulties. The harder your situation, the more the particular person across from you matters. That’s an argument for taking the search seriously rather than booking whoever has an opening, which is what our guide to finding the right therapist is built around.

The parts nobody puts in the brochure

About one in five people leave therapy before finishing. Swift and Greenberg’s meta-analysis of 669 studies covering 83,834 clients put the weighted dropout rate at 19.7% (Swift & Greenberg, 2012). Some of that is people who got what they came for and stopped. A lot of it isn’t.

And some people get worse. This gets discussed less than it should, partly because so few studies report it at all. Cuijpers and colleagues found that only 6% of trials comparing psychotherapy against a control condition reported deterioration rates in the first place. Among those that did, the median deterioration rate in therapy groups was 4%, and above 10% in some studies (Cuijpers et al., 2018).

The same analysis holds the reassuring half of the finding. Patients receiving psychotherapy had a 61% lower chance of deteriorating than patients in control groups, with a pooled risk ratio of 0.39. Twenty people need treating to prevent one case of deterioration. Getting worse happens in therapy. It happens more often without it.

The waiting problem

There’s one more number worth sitting with, because it reframes the question you’re actually asking. Wang and colleagues, working from the National Comorbidity Survey Replication, measured how long people wait between the first onset of a disorder and their first treatment contact. The median delay was 6 to 8 years for mood disorders and 9 to 23 years for anxiety disorders (Wang et al., 2005).

Almost nobody in that dataset was choosing between therapy and some rival intervention. They were choosing between therapy and years more of the same. That’s usually the real comparison, and it changes the arithmetic.

What the evidence supports you doing

Put together, the research points at a handful of unglamorous things.

Plan for a course, not a trial. If half of people need 13 to 18 sessions and the average person in routine care attends fewer than five, the most common way therapy fails is by stopping too early. Budget accordingly.

Judge fit around session three, not session one. The alliance is the most reliable predictor you have access to, and it’s the one thing you can assess without a symptom scale. First sessions are structurally awkward, which is why knowing what a first session normally looks like stops you misreading ordinary discomfort as a bad match.

Track something. Without a measure of whether you’re moving, you’re relying on memory, and memory is exactly what depression and anxiety distort. A number on a page every few weeks isn’t clinical, but it beats a vibe.

Don’t let format become the obstacle. The evidence on delivery is covered in our comparison of online and in-person therapy, and the short version is that the format you’ll consistently attend beats the format you’d theoretically prefer.

The takeaway

The most defensible answer to “does therapy work” is yes, moderately, for most people, if they get enough of it and land with someone they can work with. That’s a weaker claim than the marketing and a much stronger claim than nothing.

The honest uncertainty isn’t a reason to stay in the open tab. It’s a reason to go in with the right expectations: a few months rather than a few weeks, a decision point around session three or four about whether this person is right, and permission to change course if the answer’s no. Those expectations are what separate the people the research says get better from the people who attended four sessions and concluded therapy doesn’t work.

You’re not asking whether therapy works in general. You’re asking whether it’s worth trying. On the evidence, for most people, it is.

Frequently asked questions

Sources

1. Howard KI, Kopta SM, Krause MS, Orlinsky DE. (1986). The dose-effect relationship in psychotherapy. American Psychologist, 41(2), 159-164. (15 samples, 2,431 patients: about 50% measurably improved by 8 sessions, 75% by 26, 85% by 52; 10% to 18% improved before the first session.)
2. Hansen NB, Lambert MJ, Forman EM. (2002). The psychotherapy dose-response effect and its implications for treatment delivery services. Clinical Psychology: Science and Practice, 9, 329-343. (Trials: 57.6% to 67.2% improved in an average of 12.7 sessions. Naturalistic database of over 6,000 patients: fewer than five sessions attended on average, improvement rate about 20%.)
3. Cuijpers P, Cristea IA, Karyotaki E, Reijnders M, Huibers MJH. (2016). How effective are cognitive behavior therapies for major depression and anxiety disorders? A meta-analytic update of the evidence. World Psychiatry, 15(3), 245-258. (144 trials; g=0.75 for depression to g=0.88 for social anxiety, falling to g=0.65 and g=0.59 after adjusting for publication bias.)
4. Flückiger C, Del Re AC, Wampold BE, Horvath AO. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316-340. (295 independent studies, more than 30,000 patients: alliance-outcome correlation r=.278, no evidence of publication bias.)
5. Wampold BE. (2015). How important are the common factors in psychotherapy? An update. World Psychiatry, 14, 270-277. (The contextual model: alliance, empathy, expectations, cultural adaptation and therapist differences as drivers of benefit.)
6. Johns RG, Barkham M, Kellett S, Saxon D. (2019). A systematic review of therapist effects: A critical narrative update and refinement to Baldwin and Imel’s review. Clinical Psychology Review, 67, 78-93. (Weighted average therapist effect of 5%, range 0.2% to 29%, larger where client severity was higher.)
7. Swift JK, Greenberg RP. (2012). Premature discontinuation in adult psychotherapy: A meta-analysis. Journal of Consulting and Clinical Psychology, 80(4), 547-559. (669 studies, 83,834 clients: weighted dropout rate 19.7%.)
8. Cuijpers P, Reijnders M, Karyotaki E, de Wit L, Ebert DD. (2018). Negative effects of psychotherapies for adult depression: A meta-analysis of deterioration rates. Journal of Affective Disorders, 239, 138-145. (Pooled risk ratio 0.39; median deterioration 4% in therapy groups and above 10% in some studies; only 6% of trials reported deterioration at all.)
9. Wang PS, Berglund P, Olfson M, Pincus HA, Wells KB, Kessler RC. (2005). Failure and delay in initial treatment contact after first onset of mental disorders in the National Comorbidity Survey Replication. Archives of General Psychiatry, 62(6), 603-613. (Median delay to first treatment contact: 6 to 8 years for mood disorders, 9 to 23 years for anxiety disorders.)