peer_reviewed_pete@peer_reviewed_pete"Correlation is not causation" gets quoted a lot. Here's the underrated follow-up: how scientists actually try to establish causation.
Everyone can recite "correlation isn't causation." Fewer people know what the field does about it, so the phrase ends up as a conversation-ender instead of a starting point. The honest answer is that establishing causation is hard work with a real toolkit. The gold standard is the randomized controlled trial: you randomly assign who gets the treatment, so on average the two groups differ only in the thing you're testing. Randomization is what breaks the link between the treatment and all the confounding variables you didn't think of. That's the whole magic — it's not that the groups are identical, it's that any difference is left to chance and you can quantify the chance. When you can't randomize — you can't ethically assign people to smoke, for instance — epidemiologists lean on a set of considerations often traced back to Austin Bradford Hill in the 1960s: things like a dose-response relationship, consistency across independent studies, a plausible mechanism, and correct time order (cause before effect). No single one proves causation; together they build a case. The smoking-and-lung-cancer link was established exactly this way, without a single randomized trial on humans. So "correlation isn't causation" is true, but it's the beginning of the scientific question, not the end of it.
Join the conversation
Facet is free to read. To reply you need an account: one private root identity, and up to ten public personas that can never be linked to each other or to you.
Create an account
thesispending@thesispending· 7/5/2026, 2:43:44 PMThe randomization point deserves underlining because it's so counterintuitive. People assume a good study means carefully matching the two groups on every variable. But you can only match on variables you know about. Randomization handles the confounders you never even thought to measure — that's the thing a matched design can't promise.
sample_size_of_one@sample_size_of_one· 7/5/2026, 4:55:16 PMMy whole handle is a joke about this, so obviously I'm here for it. The Hill considerations are underrated precisely because they're not a checklist that spits out a yes/no — they're closer to how a careful person weighs evidence. Dose-response is the one that convinces me fastest: if more of the thing reliably means more of the effect, that's hard to hand-wave away.
orbital_decay@orbital_decay· 7/5/2026, 8:47:04 PMWorth noting the same logic runs in fields that can't run trials at all. Astronomers can't randomly assign stars to conditions, so we lean hard on the consistency-and-mechanism side — independent observations converging, plus a physical model that predicts what we see. Different domain, same epistemic muscles.
peer_reviewed_pete@peer_reviewed_pete· 7/5/2026, 9:28:42 PMExactly. Observational sciences are proof that "you can't randomize" doesn't mean "you can't know." It means you work harder for the same confidence, and you stay louder about your uncertainty. Which, honestly, more fields could stand to do.
orbital_decay@orbital_decay· 7/9/2026, 4:04:21 PMBeen chewing on the causation thread for a couple days. The Hill considerations quietly describe how I evaluate almost any surprising claim now, science or not: is there a dose-response pattern, do independent sources agree, is there a plausible mechanism, and does the timeline actually work. It's a decent filter for a noisy world.