In depression, likes are more reinforcing, not less

Design

Cross-sectional analysis of longitudinal posting behaviour across three datasets, one preregistered

Population

7736 social media users contributing over 17 million posts

Primary outcome

Association between previous-day likes and current-day posting, by depression status or severity

Effect

Greater reinforcement by likes with depression in all three datasets (beta 0.013, P = 0.002; 0.026, P = 0.03; 0.008, P = 0.02)

Laboratory reward-learning tasks consistently show blunted reinforcement in depression. This study asked whether the same holds for a real social reward in the real world, using more than 17 million posts from 7736 social media users across three datasets: passively scraped accounts of 1045 users who had disclosed a depression diagnosis alongside 5001 randomly sampled users, and two recruited samples of 601 and 1089 people with measured depression severity.

Reinforcement tendency was estimated as the relationship between the previous day's likes and the current day's posting. In all three datasets, people with a depression diagnosis or higher depressive symptoms were more reinforced by likes, not less (beta 0.013, P = 0.002; 0.026, P = 0.03; 0.008, P = 0.02 in the preregistered dataset). In the third dataset the effect mapped specifically onto a transdiagnostic anxious-depression factor, while a separate behaviour - persistence in posting regardless of response - mapped onto compulsivity and intrusive thought.

The replication across three samples with different recruitment and different depression measures is what makes this worth attention, and the direction contradicts the laboratory literature. The obvious clinical reading - that social media therefore worsens depression - is not what a cross-sectional design can support; heightened sensitivity to social reward could equally be a feature of depression that predates any account. What it does give you is something concrete to ask about, because a patient whose posting is driven by yesterday's likes is describing a loop they can usually recognise when it is named.

• Ask specifically whether posting is driven by the response to the last post; the pattern is recognisable to patients.
• Do not tell a patient that social media caused their depression on the strength of this; the design cannot show direction.
• Note that heightened social-reward sensitivity contradicts the blunted reward learning seen in laboratory tasks.
• Distinguish reinforcement-driven posting from compulsive posting; they mapped onto different symptom dimensions.
• Behaviour on one platform, in largely Western samples, may not generalise to the platforms your patients use.

The statistics, in plain English

The beta values are tiny in absolute terms - 0.008 to 0.026 - which reflects the fine granularity of day-to-day posting rather than a trivial phenomenon, but it does mean the effect is small at the level of any single day. What carries the finding is replication in three samples measuring depression three different ways, including one preregistered. Being cross-sectional, none of them can say whether the reinforcement sensitivity precedes, follows, or maintains the depression.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.