Reading Numbers: Averages, Samples & Signal (step by step)
P1.stats-literacy.04 · Audience: guest, it-ml, language-pro · Prerequisites: Probability Traps & Fallacies (step by step)
Take it one idea at a time: why one 'average' number can hide the real story, why a huge survey can still point the wrong way, and how to tell a genuine trend from normal wobble. Each step has a worked example. No maths required.
One mental model. A single number is a summary, and every summary drops information. Before you act on a number, ask what it left out: the spread, who was counted, and whether a move is bigger than the usual wobble.
An average (the mean) adds everything up and divides — so a few extreme values drag it far from where most people sit. The median — the middle value — is more robust for skewed data like incomes, response times, or deal sizes.
Worked example — the 'average' salary nobody earns
Ten people sit in a room. Nine earn €30k; one is the founder on €300k.
| Measure | Value | Who actually experiences it |
|---|---|---|
| Mean (the 'average') | €57k | nobody |
| Median (the middle person) | €30k | 9 of the 10 |
The mean of €57k describes no one in the room — one outlier dragged it up by €27k. The median of €30k is what a typical person actually earns.
The 'typical' customer implied by an average may not actually exist.
When someone shows you one average, ask for the median and the range — not just the mean.
⚡ Interview Ref — the quick-scan Reference face
Reading numbers: averages, samples & signal. In plain language: why one 'average' number can hide the real story, why a huge survey can still point the wrong way, and how to tell a genuine trend from normal wobble. No maths required — just better questions to ask when someone shows you a number.
💡 One mental model. A single number is a summary, and every summary drops information. Before you act on a number, ask what it left out: the spread, who was counted, and whether a move is bigger than the usual wobble.
1 — What an average hides and when it misleads (b3)
An average (the mean) adds everything up and divides — so a few extreme values drag it far from where most people sit. One customer spending a fortune, or one support ticket that took days, can pull the average to a number almost nobody experiences.
It also hides two things you usually care about:
- the spread — are values tightly clustered or all over the place?
- subgroups — one region or plan may behave completely differently from another.
The median — the middle value, where half are above and half below — is more robust for skewed data like incomes, response times, or deal sizes, because outliers can't drag it around.
The 'typical' customer implied by an average may not actually exist.
When someone shows you one average, ask for the distribution, the median, and the range — not just the mean.
2 — Sampling bias: why large surveys still mislead (b4)
A big number of responses feels reassuring — but size doesn't fix a crooked sample. If the people who answered aren't representative of who you care about, more responses just give you a more confident wrong answer.
A self-selected or convenience sample bakes in sampling bias from the start:
- only the very happy or very angry customers bother to reply,
- the channel skews who you even reach (an email survey misses phone-only users),
- the easy-to-reach group stands in for everyone, even if it isn't like everyone.
Big but biased beats small but representative only in size — not in truth.
When you see a survey, don't ask 'how many answered?' first. Ask who was included and who was missed — a small, carefully representative sample can beat a huge one riddled with sampling bias.
3 — Telling a real trend (signal) from noise (b12)
Numbers wobble on their own. Normal variation makes almost any metric move up and down week to week even when nothing has really changed. A single up-tick or down-tick is usually noise, not signal.
Before you declare a trend, check three things:
- the normal range — how much does this number bounce around in a quiet period?
- a longer time window — does the move survive over months, not just one week?
- size versus usual variation — is this move bigger than the wobble you'd see anyway?
A real signal keeps showing up across time and is larger than ordinary noise. One data point is a mood; a sustained move is a trend.
Reacting to noise as if it were signal leads to whiplash. Wait for the move to clear the normal range before you act on it.
Interview one-liners
- An average is dragged by outliers and hides the spread — ask for the median and the range, not just the mean.
- A big survey with sampling bias gives a confident wrong answer — ask who was included, not how many replied; a representative sample wins.
- One up-tick is usually noise, not signal — look at the range and a longer window before calling it a trend.
📚 Go Further (from the quick-scan note)
Plain-language explainers on reading numbers without being misled.
| Type | Resource |
|---|---|
| Book | How to Lie with Statistics — Darrell Huff, on misleading summaries |
| Book | The Signal and the Noise — Nate Silver, on separating trends from wobble |
| In-app | Evidence & evaluating claims, for judging the numbers you're shown |
📚 Go Further
Plain-language explainers on reading numbers without being misled.
| Type | Resource |
|---|---|
| Book | How to Lie with Statistics — Darrell Huff, on misleading summaries |
| Book | The Signal and the Noise — Nate Silver, on separating trends from wobble |
| In-app | Communicating Uncertainty, for judging the numbers you're shown |
Ask the mentor about this module
Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.
Where next?
Later in Data & Statistics Literacy