A scientific result often arrives with a small Greek letter attached: 3σ, 4σ, 5σ. It can look like a stamp of certainty. It is not.

Sigma is a way of asking a narrower question: if there were only the expected noise here, how surprising would this signal be? That is a useful question. It is not the same as asking whether the scientific interpretation is true.

The short version

In many fields, especially physics and astronomy, sigma measures how far an observed signal sits from the noise or background the researchers expected.

  • Low sigma means the signal is not far from ordinary noise.
  • Higher sigma means the signal is harder to explain as a random fluctuation.
  • 5 sigma is a very strong conventional threshold: under the statistical model being used, a random fluctuation that large would be very rare.

That last phrase matters: under the statistical model being used. A high-sigma signal can still be affected by calibration, foreground contamination, model choices, selection effects, or a wrong physical interpretation. Sigma helps answer “is there probably something here?” It does not answer “what is it?”

Noise first

Imagine taking a picture of a dark sky. Even if no real source is present, the image is not perfectly blank. Detectors have noise. The sky has background. Data processing leaves small wiggles. If you measure enough empty patches, some will look a little bright just by chance.

So the first job is not to ask whether a bright spot is exciting. It is to ask what ordinary empty sky looks like. Researchers estimate that background, measure how much it varies, and then ask how far the candidate signal rises above it.

That distance is what sigma counts.

What sigma measures

One sigma is one standard deviation: one typical unit of variation in the noise. A 5σ signal is five of those units away from the expected background.

The exact probability depends on assumptions about the noise and on whether the test is one-sided or two-sided, but the practical reading is simple enough: a 5σ signal is not a casual bump. If the background model is right, noise alone should almost never make something that extreme.

This is why a paper may say a source is detected at 5.2–5.3σ. It means the measured signal is about five standard deviations above what the authors expect from background fluctuations. It does not mean there is a 99.9999% chance that the authors’ explanation is true. That is the common mistake.

Why five sigma became a line

Different fields use different conventions. In particle physics and much of astronomy, 3σ often reads as evidence: interesting, worth attention, not enough by itself. A 5σ result is often treated as a discovery-level detection.

The high bar exists for a blunt reason: scientists look at many noisy things. If you search enough places, something will eventually look unusual by chance. A stricter threshold reduces the risk of celebrating a random bump.

The threshold is a convention, not a law of nature. A 4.9σ result is not worthless; a 5.1σ result is not magically immune to error. The number is a tool for discipline, not a sacrament. Statistics has enough robes already.

Local vs global

There is one more trap: where did you look?

If researchers test one pre-specified place — one wavelength, one position, one signal shape — then the sigma can be read as a local significance: how surprising the signal is at that exact place.

But if they scan thousands of places, many wavelengths, many cuts of the data, or many possible signal shapes, the question changes. Somewhere in that search, noise is more likely to produce a fluke. After accounting for all those chances to be fooled, the result may have a lower global significance.

This is the look-elsewhere effect. It is not a technicality. It is the difference between “I found a strange mark exactly where I said I would look” and “I searched the whole wall until one stain looked like a face.”

What sigma does not prove

A high-sigma detection can be real and still be misread.

It can show that there is a signal in the data, while leaving open:

  • whether the signal comes from the object the authors think it does;
  • whether a foreground source or contaminant is involved;
  • whether the instrument calibration is fully under control;
  • whether the background model was the right one;
  • whether the physical interpretation is unique.

That is why careful papers do more than quote sigma. They check empty regions of sky. They test for foreground interlopers. They compare instruments or filters. They ask whether the same result appears under different assumptions. Sigma is the entrance exam. It is not the whole degree.

How to read a sigma claim

When an article says a result is 5σ, read it as a strong statement about the data, not as a final verdict about the story.

Ask four questions:

  1. What is the noise model? What did the authors count as ordinary background?
  2. Was the test local or global? Did they look in one place, or search many?
  3. What checks rule out contamination or artefacts? A high number cannot do that alone.
  4. What interpretation is being attached to the signal? Detection and explanation are separate steps.

If those answers are good, a high-sigma result deserves confidence. If they are missing, the number may still be impressive — but it is doing more work than it should.

About this guide

This is an evergreen explainer, not coverage of a single paper. It is prepared with AI assistance and human editorial review and revised over time; the date above is when it was last checked. It teaches how to read the numbers — it is not medical or statistical advice.