systematic versus statistical uncertainty

Imagine weighing yourself on a bathroom scale. If you step on it ten times and get slightly different readings each time, that scatter is statistical uncertainty — the random jitter that shrinks as you average more measurements. But if the scale is simply mis-calibrated and reads two kilos too heavy every single time, no amount of repeating will reveal the error; that is a systematic uncertainty — a consistent bias built into the measurement itself. Every experimental result in particle physics carries both kinds, and they behave in completely different ways.

Statistical uncertainty comes from having a finite number of events. If you measure a rate from 100 events, the natural random spread is about the square root of 100, or 10 — a 10 percent uncertainty — and it falls as you collect more data: a million events gives a 0.1 percent statistical uncertainty. Systematic uncertainty, by contrast, comes from imperfect knowledge of your apparatus and methods: how well you know the detector's energy calibration, the luminosity, the efficiency of your selection, the background model, the parton distributions. Collecting more data does not shrink these; reducing them requires better calibrations, cross-checks, and understanding.

The distinction is central because it tells you where the limit of a measurement comes from and what to do about it. Early in an experiment's life, results are usually statistics-limited, so more data helps. Mature, high-precision measurements are usually systematics-limited, so the path forward is painstaking work on calibration and modelling, not just more collisions. The honest difficulty is that systematic uncertainties are hard to estimate — they require judgement, and an underestimated systematic is one of the classic ways a wrong result slips through, because it can make a measurement look more precise, or a bump look more significant, than it truly is.