What the base rate is
The base rate is simply how common something is in the group you are looking at, before you consider any particular clue. If one player in 200 is cheating, the base rate of cheating is 0.5%. If half of all emails are spam, the base rate of spam is 50%.
Base rate neglect is the habit of ignoring that number and focusing only on the clue in front of you. A detector flags a player, a test comes back positive, a hand looks suspicious, and the mind jumps to the accuracy of the detector instead of asking how many cheaters there were to find in the first place.
An illustration with a cheat detector
The numbers here are invented to show the mechanism, not taken from any real system.
Imagine an online game with a cheat detector. It correctly flags 95% of cheaters. It also wrongly flags 5% of honest players. That sounds like a good detector. Now suppose 1 player in 200 is a cheater, and a player has just been flagged. How likely is it that they are really cheating?
Use a population of 20,000 players, which makes every count a whole number:
- Cheaters: 20,000 / 200 = 100. The detector flags 95% of them, so 95 are flagged.
- Honest players: 20,000 − 100 = 19,900. The detector flags 5% of them, so 0.05 × 19,900 = 995 are flagged.
Total flagged: 95 + 995 = 1,090. Of those, 95 are real cheaters. The chance that a flagged player is a cheater is 95 / 1,090, which is about 8.7%.
So a flag from a “95% accurate” detector is wrong more than nine times out of ten. The detector is not broken. There are simply far more honest players than cheaters, so even a small error rate applied to the large honest group produces more false alarms than the large hit rate applied to the small cheating group produces real catches.
The same detector on a common problem
Change only the base rate. Suppose half of the players in some group are cheaters, and the detector behaves identically.
In a group of 10,000, there are 5,000 cheaters and 5,000 honest players. The detector flags 95% of 5,000 = 4,750 cheaters and 5% of 5,000 = 250 honest players. Total flagged is 5,000, and 4,750 of them are cheaters, so a flag now means a 95% chance of cheating.
The detector’s accuracy did not change. What changed is how often there was anything to find. This comparison is the heart of the matter: the meaning of a positive result depends on both the test and the setting where it is used.
Why the mind slides past it
People tend to treat a vivid, specific piece of evidence as more informative than a dull statistic about how common something is. “This player was flagged” feels like a story. “One in 200 cheats” feels like background. Yet the background number often decides the answer.
A second reason is a mix-up of two different questions. “If the player is a cheater, how likely is a flag?” is 95%. “If the player is flagged, how likely are they a cheater?” is about 8.7% in the first scenario. These are conditional probability questions that read alike but are not the same, and swapping them is one of the most common slips in probability. Its courtroom version is the prosecutors fallacy.
Where it appears in games and sport
Base rate thinking helps with a range of ordinary situations, none of which involve guaranteed outcomes:
- Streak spotting. A player who seems suspiciously lucky is one of many players, and with enough players, some will always look unusual. The base rate of “someone somewhere has a remarkable run” is high even when each individual run is unlikely.
- Scouting and rankings. A rare trait, such as a player who becomes a star, stays rare even among players who show early promise. Early signs raise the chance, but the starting rarity still weighs on the result.
- Reading an opponent. If a particular strong holding is uncommon, a single suggestive action does not make it likely. How often opponents have that holding in the first place needs to be part of the estimate.
Small samples add to the confusion. A handful of observations can make a rare event look common, which links base rate neglect to the broader issue of small sample sizes.
A habit that helps
When you meet a claim of the form “the test is X% accurate”, ask three questions before drawing a conclusion:
- How common is the thing being searched for in this group?
- How often does the test flag things that are not there?
- Out of a round number of cases, such as 10,000, how many true flags and how many false flags would there be?
The third question is the practical fix. Counting cases makes the imbalance obvious in a way that percentages hide. This is the approach behind natural frequencies, and it is the same bookkeeping that Bayes theorem formalises.
A note on gambling
None of this offers a way to beat a game of chance. Casino games have a house edge, outcomes are independent from one spin or shuffle to the next, and past results do not alter future odds. Base rate reasoning helps you interpret information correctly; it does not make a losing proposition profitable. Gambling can cause real harm, and anyone struggling with it should consider reaching out to a local support service.