Where Did the 'Sample Size of 30' Rule Come From? (And Why It Needs to Retire)

Somewhere in an American classroom right now, a student is writing down a rule that will follow them for decades: a sample size of 30 is enough.

It's stated with confidence. It shows up in intro stats notes, in business meetings, in newsroom debates about polling. And it is, at best, a half-truth that's been photocopied so many times nobody remembers what the original said.

The Origin Story Nobody Quotes Correctly

Like most stubborn myths, this one has a kernel of legitimate history. The number 30 traces back to the old days of the t-distribution. When your sample is small, the t-distribution (which accounts for extra uncertainty) has fatter tails than the normal distribution. But somewhere around 30 degrees of freedom, the t-distribution gets so close to the normal curve that statisticians figured the difference no longer mattered much.

Notice what that actually says: at n=30, you can approximate one probability distribution with another. That's it. That's the whole origin.

Somehow, over decades of textbooks summarizing textbooks, "the t-table converges around 30" morphed into "30 people is a sufficient sample." That's like concluding that because a ruler's markings stop being distinguishable at a certain distance, all objects are the same size.

Why 30 Can Be Way Too Small

Here's the part that surprises people: there is no universal sample size. The right number depends on three things — how big an effect you're hunting, how noisy your data is, and how certain you need to be.

Think of it like fishing. If you're trying to catch a whale in a pond, one cast might do it. If you're trying to detect a minnow in the Pacific, no reasonable number of casts guarantees anything. Statisticians call the whale a large effect size and the minnow a small one — and small effects demand enormous samples.

Psychology gave us a brutal lesson in this. For years, plenty of published studies ran 30 or 40 college students through an experiment, found a "significant" result, and moved on. Then large-scale replication projects tried to repeat the work with proper sample sizes — often in the hundreds or thousands — and a huge chunk of the findings evaporated. Some analyses suggested that studies relying on small samples were dramatically more likely to produce effects that couldn't be reproduced. When your sample is tiny, whatever you catch is more likely to be noise dressed up as discovery.

And Why 30 Can Be Way Too Big

Here's the twist that makes the myth even more misleading: sometimes 30 is overkill.

If a variable is extremely consistent — say, the temperature at which water boils at sea level — you don't need 30 measurements to characterize it. A handful will do. In manufacturing and engineering, quality control often works with far fewer units when the process is stable, because each additional observation adds diminishing value.

So the magic number fails in both directions. Too small when effects are subtle and variability is high (which is most of human behavior). Too large when the phenomenon is stable and well-behaved. A rule that's simultaneously insufficient and wasteful isn't a rule — it's a ritual.

What Actually Determines 'Enough'

Real sample-size planning — the kind researchers do before a clinical trial — comes down to a handful of honest questions:

• How big is the effect worth detecting? If a new drug reduces symptoms by 40%, you'll spot that with a modest trial. If it reduces symptoms by 4%, you need thousands of patients to distinguish that from a coin flip.
• How much do people vary? Blood pressure readings bounce around a lot between individuals. High variability means more subjects needed to see the signal through the static.
• How confident do you want to be? Demanding stronger evidence — smaller risk of a false positive, higher probability of catching a true effect — always costs more subjects.

These aren't academic niceties. Before a pharmaceutical company recruits a single patient, a biostatistician runs a formal power analysis that calculates the required sample from effect estimates and variability assumptions. Regulators at the FDA scrutinize those calculations. The number 30 appears nowhere in the process.

A Field Guide for Skeptical Readers

You don't need a stats degree to sanity-check the next headline you read. Ask four questions:

1. What's the claim? "Raises your risk by 80%" is easier to detect than "improves outcomes slightly." Match the sample size to the size of the promise.
2. How wiggly is the subject matter? Studies of human behavior, opinions, and health outcomes involve enormous individual variation — and need bigger samples than studies of, say, concrete strength.
3. Who's in the sample? Thirty students at one Midwestern university is not a sample of "Americans." Sample size isn't just a count; it's a question of who got counted.
4. Did they plan the number, or stop when it looked good? Researchers who keep collecting data until the result turns significant are essentially fishing until the pond gives up a fish. Pre-registered sample sizes are a good sign.

The Lesson Worth Keeping

The n=30 myth persists because it's comforting. A single number feels like mastery — a threshold you can carry in your pocket. But quantitative thinking is precisely about giving up the comfort of magic numbers and asking under what conditions, for what effect, with what variability.

So the next time someone tells you 30 is enough, ask them: enough for what? If they can't answer, they haven't done the math. Neither did the textbook.