News · Brain & Mental Health
Likes drove more posting in people with depression
Analysing 17 million posts from 7,736 Twitter users, researchers in JAMA Psychiatry found people with depression were more strongly reinforced by likes, the opposite of what laboratory reward tests usually show.
Based on a peer-reviewed analysis of 17 million posts in JAMA Psychiatry
- Over 17 million posts from 7,736 Twitter users across three separately assembled datasets.
- Reinforcement was measured as the link between yesterday's likes and today's posting.
- People with depression were more reinforced by likes in all three datasets.
- That reverses the usual laboratory finding of blunted reward learning in depression.
- The effect tied specifically to an anxious-depression factor, not to compulsivity.
Bring someone with depression into a psychology lab, offer them a small reward for a small effort, and they will typically work less hard for it than someone without. Blunted response to reward is one of the more reliable findings in the field, and it fits the clinical picture, since losing interest in favorite activities is among the defining symptoms.
A study in JAMA Psychiatry went looking for the same effect outside the lab, in the reward people actually chase all day, and found it running the other way.
Measuring reward where it happens
The researchers did not ask anyone how likes made them feel. They watched what people did next.
Reinforcement tendency was estimated as the association between previous-day likes and current-day posting. A user whose good days are followed by more posting scores high; one who posts at the same rate regardless scores low.
The scale is what makes it work. The 3 datasets contained over 17 million posts from 7736 Twitter users, assembled three different ways: passively scraped data, users recruited through ads on the platform, and users recruited through a research panel.
That construction is deliberate. Depression was measured differently in each, from self-disclosure on the platform in one to a validated questionnaire in another, so a finding surviving all three is not an artifact of one measurement choice.
The same answer three times
Individuals with a depression diagnosis or higher symptoms of depression tended more to be reinforced by likes in all datasets.
The coefficients were 0.013, 0.026 and 0.008. Those are small numbers describing a nudge in behavior, not a compulsion.
What lifts them above noise is the replication. The same direction appeared despite substantial differences between datasets in depression measurement and composition, and the third analysis was preregistered, meaning its predictions were filed before the data were looked at.
Not just any distress
The third dataset carried enough extra measurement to ask whether this is about depression specifically or about mental ill health generally.
Greater reinforcement tendency was specifically associated with a transdiagnostic anxious-depression factor, while a separate behavioral pattern, persistence, was linked to a compulsivity and intrusive thought factor.
So the effect sorted itself. The like-chasing tracked the anxious-depressive dimension; the doggedness tracked the compulsive one. Two different behaviors, two different underlying factors, which is a more informative result than everything loading onto one blob of distress.
The contradiction is the point
Set this against the laboratory literature and something has to give. The authors state the tension directly: their finding stands in contrast to typical laboratory-based findings of blunted reinforcement learning in depression.
There are readings that would reconcile the two. Social approval may not behave like the abstract points and tokens used in reward experiments. A person who has stopped enjoying most things might be more responsive to the one source of feedback still reaching them, not less. And a laboratory task compresses into an hour what a platform does over months.
None of those is established here. What the study establishes is that the lab result does not simply transfer to the setting where most people now encounter social reward.
What it cannot show
Direction is unresolved, and the paper says so. The framing offered is potential mechanisms linking social media use to worse mental health or vice versa, and the second half of that phrase deserves as much weight as the first.
The design is a snapshot of behavior over time rather than an experiment. Nobody was assigned to receive likes.
Everything also comes from a single platform. There are a variety of things that may cause depression, including genetic, biological, environmental, and psychological factors, and one product’s reward design sits somewhere inside that list rather than above it.
Why it is worth knowing
Depression is more than a feeling of being sad or irritable for a few days, and the argument about social media and mental health has largely been fought with screen-time totals, which measure exposure and nothing about what the exposure does.
This measures a mechanism instead: not how long someone was on a platform, but how much its feedback moved them. Whichever way the causal arrow eventually points, that is a more useful thing to be able to observe, and it can be observed at a scale no laboratory will match.
People also ask
What does reinforced by likes actually mean here?
It is a behavioral measure, not a feeling. Reinforcement tendency was estimated as the association between previous-day likes and current-day posting. If a good day of likes is followed by more posting the next day, that person scores higher. Nobody was asked how the likes made them feel.
Why is this the opposite of lab findings?
Reward-learning experiments in depression usually show blunted responses, meaning people work less hard for a payoff. Here the response to social reward was stronger, not weaker, in people with depression. The authors set that contrast out explicitly, and it is the reason the study is interesting rather than confirmatory.
How big were the effects?
Small in every dataset, with coefficients of 0.013, 0.026 and 0.008. What carries weight is not their size but that the same direction appeared three times over, despite substantial differences between datasets in depression measurement and composition. One of the three was preregistered, meaning the analysis was locked in before the data were examined.
Does this mean social media causes depression?
No, and the authors are careful about it. This is cross-sectional work on observational data, and they frame the result as potential mechanisms linking social media use to worse mental health or vice versa. The vice versa is doing real work in that sentence. Depression could just as easily change how someone uses a platform.
Does this apply to platforms other than Twitter?
Untested. All three datasets came from one platform with its own norms, reward design and user base, and depression was measured differently in each. Whether the same pattern appears where feeds and reward signals work differently is an open question.