Law of Small Numbers
🇳🇴Loven om små tallDefinition
The law of small numbers is the mistaken belief that small samples are as representative of reality as large ones. We draw strong conclusions from limited data and systematically underestimate how much randomness affects small samples. The name is an ironic twist on the 'law of large numbers' in statistics, which states that large samples yield reliable estimates – something small samples decidedly do not.
Real-world example
A hospital with few births might have 70% boys one year and 30% the next – pure statistical variation. Yet staff may develop theories about seasons or lunar phases. Research shows that small hospitals have far greater variation in gender ratios than large ones, but people intuitively expect the percentage to be similar everywhere.
In business, a startup might celebrate that 5 out of 5 demo meetings led to sales and assume they've found a winning strategy. But with such limited data, success could be due to luck, timing, or chance meetings with particularly interested customers. Only after 50–100 meetings does the pattern provide meaningful insight.
The same mistake occurs in hiring: A manager who has hired three people from the same university, all of whom performed well, may develop a preference for that school – without realizing that three cases are far too few to draw conclusions.
Supplementary perspective
The law of small numbers is closely related to the representativeness heuristic – we expect even small samples to 'look like' the population they come from. It's also connected to clustering illusion, where we see patterns in random data. Daniel Kahneman and Amos Tversky documented this bias extensively, showing that even experienced researchers underestimate how much randomness affects small studies.
Practical advice
Recognize
- —Always ask: 'How many observations is this conclusion based on?'
- —Be skeptical of percentages from small numbers (e.g., '3 out of 4 preferred...').
- —Notice when you or others generalize from personal experience with few cases.
Counteract
- —Demand larger samples before trusting patterns.
- —Ask for replication – has the finding been confirmed by others?
- —Use confidence intervals to visualize uncertainty.
- —Delay conclusions until the data foundation is sufficient.
Ethical use
- —Communicate uncertainty clearly when presenting data from small samples.
- —Avoid over-interpreting early results in pilot projects.
- —Encourage humility and continued data collection.