The Line Someone Drew
Chapter 01·the premise
What was asked
- 01How strong is the screen-time association really, next to everything else measured on the same children?
- 02What has to be decided before “N times more likely to be depressed” can be written at all?
- 03Across every defensible pair of those decisions, what range of answers does this one dataset support?
- 04Is the association attached to any particular symptom, or spread across all of them?
What it found
Screen time is associated with depression scores here, and the association is tiny: +0.1344, accounting for 1.8% of why these teenagers differ from one another. Sleep quality accounts for 14.5% — 8 times as much — and holding hours slept constant absorbs 46% of the screen association. The larger finding is about how such a claim gets made. Saying anyone is depressed requires drawing a line across a smooth distribution, and saying anyone is a heavy user requires drawing a second one. Neither line is usually reported. Moving both across their plausible range yields 130 risk ratios from 1.15× to 2.97×, all arithmetically correct, of which only 92 can be distinguished from no difference at all. And breaking the questionnaire into its 21 items reveals no symptom to point at: the association is spread almost evenly across all of them.
You have read the headline. Screen time is linked to teenage depression. In this data, that is true — and it is also almost nothing.
4,810 teenagers, each with a screen-time measure and a 21-question depression questionnaire. The correlation between them is . Real, positive, and in a sample this size not a fluke.
But a correlation of that size accounts for of why these teenagers differ from one another in how they answered. Nearly all of the rest is something else. Screen time is not the story of teenage unhappiness. It is a rounding error that got a press cycle.
Here is what does better, in the same file, measured on the same children: how well they sleep, at — the explanatory power of screens. And nearly half of the screen-time association, of it, disappears the moment you account for how many hours they actually slept.
If percentages of variance mean nothing to you, here is the same gap in plain units. Both columns are average depression score, climbing as you go down. One staircase is shallow and one is steep.
| level | more screen time → | worse sleep → |
|---|---|---|
| 1 | 6.03 | 4.52 |
| 2 | 5.90 | 7.19 |
| 3 | 7.24 | 11.80 |
| 4 | 8.16 | 13.87 |
| 5 | 8.76 | 16.80 |
| 6 | 10.16 | 12.50 |
| swing | 4.26 points | 12.28 points |
Same 4,810 teenagers, same questionnaire, same scale. Going from the least to the most screen time moves the average by points. Going from the best to the worst sleep moves it by . Both climb steadily, so neither is noise — they are just very different sizes.
Komentar