Eleven independent samples, drawn a year apart, across a decade that included a pandemic, a platform migration to TikTok, and a full turnover in what the phrase "social media" even denotes. The signal showed up every time, in the same direction, at a similar magnitude.

Replication across independently drawn populations is the part of this paper that earns its publication. Most of the field is one dataset analyzed forty ways. This is one analysis run on eleven datasets, which is the harder and less fashionable version.

Depression was captured as a single yes or no item on a list of 31 health conditions. The list ran from high blood pressure to corrective eyeglasses. No PHQ-9, no structured interview, no clinician anywhere in the loop.

The authors are honest about this in the paper, noting that some respondents may have subclinical symptoms while others may be diagnosing themselves. That caveat did not survive contact with the press release. What the model is classifying is the willingness to tick a box on a media survey, which correlates with depression and is not the same object.

There is a pleasing symmetry here that I want to flag before someone else does. The most cited skeptical finding in this literature, Orben and Przybylski's 2019 specification curve analysis, concluded that digital technology use explained at most 0.4 percent of the variance in adolescent wellbeing, an association its authors were happy to compare to ordinary household variables. Eyeglasses turn up on both sides of this argument, once as a punchline and once as a neighbor on the checklist.

A SHAP dependence plot shows how a feature's contribution to a model's output varies across the range of that feature. Where the curve crosses zero, the variable stops pushing predictions down and starts pushing them up. In these models, that crossing sat near 150 minutes.

A zero crossing on a SHAP plot is a property of the model and the distribution it was fit to. It is not a biological inflection point, it is not a dose, and it is not a number you can hand a patient. Move the sample, move the crossing. The authors say as much, calling it not an established clinical threshold and asking for longitudinal and experimental work. Every outlet that ran the number dropped the sentence that followed it.

The tell is the roundness. Real thresholds in medicine are ugly numbers. 150 minutes is two and a half hours, which is what a number looks like when it has been rounded toward memorability.

Eleven repeated cross sections, total N around 183,000. Different people each year. Nobody was followed. The paper says "repeated cross-sectional" in the second sentence of its abstract.

"Researchers Studied 183,000 Adults for 11 Years" describes a cohort study that nobody ran. The distinction is the whole ballgame, because a cross section cannot tell you whether heavy social media use precedes depression or follows from it. Depressed people spend more time alone with a phone. That explanation fits this data exactly as well as the one in the headline, and the design cannot separate them.