All lessons

Module 1: Describing data · Lesson 1.1 · about 15 min

Why analysts need statistics

What statistics adds to dashboards and SQL: describing what happened, and judging whether a change is real or just noise. The two halves of the subject and how this course is laid out.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 1 · about 15 minutes · no maths yet. You’ll be able to say what statistics adds to an analyst’s work, tell descriptive from inferential statistics, and know how this course is laid out.

Every number from data is the truth plus some noise; statistics measures the noise so you can tell a real change from luck.

In everyday terms: your bathroom scale reads a little differently each time you step on it. You would not panic over 0.3 kg on one morning, but you would notice a steady climb over a month. Statistics is the rulebook for telling the two apart.

1. The question behind every report

An analyst’s day is full of questions like these:

  • Sales are up 3 % on last week. Is that the new promotion, or an ordinary week?
  • The new checkout page converts 4.4 % of visitors, the old one 4.1 %. Should we switch?
  • Store A has more complaints than store B. Is store A worse, or just bigger and busier?
  • A survey of 300 customers says 62 % like the new menu. How sure can we be about all customers?

SQL and dashboards answer what happened in these rows. Statistics answers the next question: how much of that would change if we had different rows, such as another week, another sample of customers or another month of visitors. Without it, every random wiggle in a chart looks like news.

The analyst's loop 1. QuestionDid the promo lift sales? 2. DataWhich rows, which period? 3. DescribeAverages, spread, charts 4. UncertaintyReal, or just noise? 5. DecideKeep, stop or test more next question fibub.com
Statistics lives in steps 3 and 4. Describing data says what happened in the rows you have; measuring uncertainty says how much of that would change with different rows. Most bad business decisions skip step 4.

2. The two halves of statistics

Descriptive statistics summarise the data you have: averages, spreads, percentages, charts. “The median delivery took 31 minutes” is descriptive. Module 1 covers it.

Inferential statistics reason from a sample to something bigger: all customers, all future visitors, the true effect of a change. It always comes with a statement of uncertainty, such as a margin of error, a confidence interval or a p-value. “Between 58 % and 66 % of all customers like the new menu” is inferential. Modules 4 to 7 cover it.

Between the two sits probability (Modules 2 and 3): the language for chance, which inference is built on.

3. Noise is normal

Imagine a shop whose true average is 500 sales a day. Even if nothing changes, some days bring 470 and some 530, because customers decide independently. Weekly totals wobble too. A rise of 3 % in one week means nothing until you know how big the usual wobble is. This one idea, compare a change with normal variation, runs through the whole course:

  • a z-score compares one value with the usual spread (Lesson 1.5);
  • a standard error measures how much a sample average wobbles (Lesson 4.4);
  • a confidence interval gives the range the truth probably sits in (Lesson 5.1);
  • a hypothesis test asks whether a difference is bigger than chance would produce (Module 6);
  • an A/B test applies all of this to a product decision (Module 7).
Descriptive statistics
Summaries of the rows you have: averages, percentages, charts.
Inferential statistics
Claims about something bigger than your rows (all customers, next month), always with a statement of how sure you are.
Noise (variation)
The ordinary up-and-down that happens even when nothing has changed.
Physical meaning
A sales figure is like a photo taken in wind: the building (the true level) is still, but the trees (the noise) move. Statistics tells you how much the trees usually move, so you can say whether the building moved.

“The number went up, so something changed.” Numbers go up and down every week with no cause at all. A change is only news when it is bigger than the usual wobble, and the first job is always to measure that wobble.

Watch it happen, one step at a time:

The claim. A café usually sells 200 coffees a day. After a new sign goes up, Monday brings 215.
Measure the usual wobble. Looking back over past days, sales typically move about 20 coffees from the average (the standard deviation, Lesson 1.3).
Compare. 215 is 15 above average: 15 ÷ 20 = 0.75 of a usual wobble.
How rare is that? Days at least that good happen about one day in four or five with no sign at all (Lesson 3.4).
Decide. One good Monday is not evidence. Two weeks after against two weeks before would be (Lesson 6.3).
Maths and principle

Principle: every number from data is the truth plus noise. Statistics measures the size of the noise so you can see whether a pattern rises above it.

Equation (in words): what you measure=true value+random noise. Much of this course is about the size of that noise term and how it shrinks with more data.

Analogy: a bathroom scale that reads 0.5 kg differently each time you step on it. One reading up 0.3 kg proves nothing; a steady rise over many readings does.

4. Worked example

Made-up numbers. A café sells a mean of 200 coffees a day with a typical day-to-day difference (standard deviation) of 20. After a new sign goes up, Monday brings 215 coffees.

  • 215 is 15 above the mean, which is 15 ÷ 20 = 0.75 standard deviations.
  • Days that far above average happen often by chance (roughly one day in four or five, as Lesson 3.4 will show).
  • So one good Monday says almost nothing about the sign. Two weeks of data, compared with two weeks before, would say much more (Lesson 6.3 shows how).

5. How this course works

Every lesson has a picture or a simulation to play with, a Maths and principle box with the formula explained symbol by symbol, a worked example with small numbers, practice questions with answers, and a three-question quiz. The maths is school level, and each formula is explained the first time it appears.

The machine learning course From a straight line to an LLM covers fitting models; this course covers judging data and decisions. They link to each other where they meet. You finish with a capstone (Lesson 9.1): an experiment readout for your portfolio.

6. Practice

  1. Label each as descriptive or inferential: (a) “Average basket was $38 in October.” (b) “The loyalty scheme raises spend by $2 to $5 per visit, with 95 % confidence.” (c) “42 % of survey respondents were students.”
  2. A manager says a store “got worse” because its rating fell from 4.5 to 4.3 stars in a month with 12 reviews. What would you ask first?
Answers
  1. (a) Descriptive. (b) Inferential: it is about the true effect and states uncertainty. (c) Descriptive, as long as it only describes those respondents; if it is used to claim 42 % of all customers are students, it becomes inference.
  2. How much does a monthly average of only 12 ratings normally move? With so few reviews, one or two unhappy customers can shift it by 0.2.

Try it on your own data: the CSV profiler gives a first descriptive summary of any CSV file, all in your browser.

New words: descriptive statistics, inferential statistics, sample, noise, variation, uncertainty.

  • SQL and dashboards say what happened in these rows; statistics says how much would change with different rows.
  • Descriptive statistics summarise the data you have; inferential statistics reason beyond it and state uncertainty.
  • Compare every change with normal variation before calling it real.
  • Probability (Modules 2-3) is the language inference (Modules 4-7) is built on.

Check yourself

1Sales were 3 % higher this week than last week. What is the first statistical question to ask?
2Which of these is inferential statistics rather than descriptive?
3Why can a dashboard without statistics mislead?

Sources

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Numbers in the examples are made up to show the idea unless a source is given. Simulations run in your browser; nothing you type is sent anywhere.