In 1925, a British statistician suggested 1 in 20 as a handy cutoff for spotting real results, and science turned his casual idea into one of its most rigid rules.
That number is 0.05. If you have ever read a headline about a “significant” study, it almost certainly passed through this tiny gatekeeper. Yet the man who proposed it never wanted it to be carved in stone.
The number that was never meant to be a law
The idea behind it is older than most people think. In the 18th century, John Arbuthnot and Pierre-Simon Laplace were already calculating early versions of what we now call p-values, each working on the ratio of boys to girls at birth.
But the famous threshold came from Ronald Fisher. In his 1925 book Statistical Methods for Research Workers, he offered a probability of one in twenty as a “convenient cutoff level.” A practical rule of thumb, nothing more.
Fisher never intended 0.05 to become a universal, untouchable standard. Decades later, that modest suggestion had turned into an unquestioned dogma across research worldwide.
Two statisticians turned a hint into a rule
In 1933, Jerzy Neyman and Egon Pearson (son of the statistician Karl Pearson) gave the cutoff an official name: the “significance level,” written with the Greek letter alpha.
They added one crucial instruction. The level should be chosen before any data is collected, not adjusted afterward to fit the results. Today, alpha is usually set at 5% or lower, depending on the field.
What 0.05 actually tells you, and what it does not
Here is where things get slippery. Many people believe that a result with a p-value of 0.05 or less means there is a 95% chance the researchers are right.
That is not what it means at all. The p-value answers a different question: if there were really no effect (what statisticians call the null hypothesis), how likely would it be to see a result this extreme, or even more extreme?
Those are two inverse probabilities, and mixing them up is one of the most common errors in reading science. A low p-value says the data would be surprising under “nothing is happening.” It does not say how likely your own theory is to be true.
There is a second trap. Statistical significance is not the same as practical importance. A result can clear the 0.05 bar and still matter very little in real life. That is why researchers also report the effect size, which measures how big the effect actually is.
Why particle physicists demand 1 in 3.5 million
Not every science settles for 1 in 20. When physicists announced the discovery of the Higgs boson in 2012, they relied on a far stricter standard known as “5 sigma.”
In plain terms, that corresponds to a p-value of roughly one chance in 3.5 million. Compared with that, the 0.05 threshold common in many social sciences looks remarkably relaxed.
The backlash against a single number
Over time, the cracks started to show. Researchers who focused only on statistical significance often reported results that were thin and hard to reproduce. And every failed replication makes it more likely that the original finding was just a false positive.
In 2015, the journal Basic and Applied Social Psychology went as far as banning p-values from its pages entirely. Critics replied that the tool itself was not the problem. The widespread misuse of it was.
The criticism kept growing. In 2016, the American Statistical Association published an official statement warning against the automatic use of the p ≤ 0.05 rule. In 2019, more than 800 statisticians signed a statement calling for the term “statistically significant” to be dropped altogether.
Tightening the rule is not a magic fix
A tempting solution would be to simply make the bar stricter, say from 0.05 down to 0.005. But lowering alpha has a cost: it raises the risk of false negatives, meaning real effects that go completely undetected.
Every threshold is a trade-off between two opposite mistakes, seeing something that is not there or missing something that is. Fisher understood that when he called his number a convenient cutoff. Perhaps the real lesson of 0.05 is that science works best when its rules of thumb are treated as exactly that.

