"They posted a $50k month screenshot, so the model works" — sampling on the dependent variable

"They posted a $50k month screenshot, so the model works" — sampling on the dependent variable

Income screenshots are the dominant evidence in creator advice and among the weakest.

Context. A screenshot is a single observation selected because it is large. This is textbook selection on the outcome — the methodological error that invalidates the inference.

Findings. The relevant figure is never the visible success but the unseen base rate: how many attempted the same model and earned little. Where denominators are estimable — course completion data, platform cohort analyses — the median outcome for most creator monetization paths is modest, with earnings following a power law where a tiny top tail produces the screenshots. The mean is dragged up by outliers; the median creator's experience is invisible.

Caveats. Screenshots can also be gross-not-net, single-month, or fabricated, and rarely show costs, refunds, or ad spend. None of this is verifiable from the image.

Implication. One large outcome tells you the ceiling exists, nothing about the odds. Decisions need the distribution, not its maximum.

What we still don't know: denominators are almost never published, so most creator-income base rates remain genuinely unknown.