Pick the chart from the question.
A chart is an answer to a question. Comparing groups, change in order, how values spread and how two numbers relate each have a chart that shows them best. Here they are, drawn from real numbers.
- Match a question to its chart
- Avoid the charts that mislead
- Say what a chart shows in one sentence
You need: Lesson 1.
- 1Statistics you use every day
- 2SQL that answers questions
- 3Charts that answer the question
- 4Power BI’s filter context
- 5A/B tests
- 6Guesstimates
-
Comparing groups: bars
Predict first
Revenue by payment mode: which chart?
Just show me
Bars, sorted: lengths are easy to compare, angles in a pie are not.
Bars start at zero, sorted biggest first, with the number on each. A pie only works for two or three parts of one whole, and even then bars read more easily.
UPI₹10,591Card₹3,998Cash on delivery₹2,499Delivered revenue by payment mode, the SQL guide’s shop
Quick check
Why start a bar chart at zero?
Bars are read by length, so a cut axis can make a 5% gap look like three times.
-
Change in order: lines
Predict first
Churn across tenure bands (first year, second year…): which chart?
Just show me
A line: the bands are in order, and the line shows churn falling from 13.4% in the first year to 2.6% in the sixth.
Lines are for things in order, usually time: the slope is the story. Don’t join categories that have no order (cities, products) with a line; the slope between them means nothing.
Share who left, by year of tenure (the ML guide’s churn table) Share who left Year 1 13.4% Year 2 10.2% Year 3 7.8% Year 4 3.3% Year 5 3.9% Year 6 2.6% Share who left, by year of tenure (the ML guide’s churn table)
Quick check
Monthly revenue for two years: line or bars?
Ordered time with many points: a line shows the shape.
-
How values spread: histograms
Predict first
How are 4,755 monthly charges spread? Which chart shows it?
Just show me
A histogram: values grouped into ranges, each bar’s height the number in that range.
Here the charges make a bell around ₹499, with thin tails, the impossible negatives from lesson 1 at the far left. A histogram shows the shape, the centre, the spread and the odd values in one look; a box plot shows the same more compactly when you compare groups.
Monthly charges in ₹50 ranges, 4,755 customers (the ML guide’s churn table, synthetic); below ₹0 marked
Quick check
Salaries in a company, to show how unequal they are. Best chart?
A question about spread needs a chart of the spread.
-
Two numbers together: scatter
Predict first
Does tenure relate to the monthly charge? Which chart?
Just show me
A scatter plot: one dot per customer, tenure across and charge up.
Here the dots make a shapeless cloud: the table’s code draws tenure and charge independently, and a scatter shows that honestly, where a single correlation number could hide a curve or clusters. The blue dots are customers who left: 15 of the 19 in this sample had stayed three years or less.
300 customers from the churn table, picked at random
Quick check
A scatter shows a clear upward cloud. What can you say?
Correlation, not causation, until an experiment or a careful method says more.
-
One sentence per chart
Predict first
What should a chart’s title say?
Just show me
The finding, not the topic: a busy manager should get it from the title alone.
Write the takeaway as the title, label the axes with their units, remove anything that doesn’t help (3D, heavy gridlines, rainbow colours) and use one highlight colour for the point you’re making. In interviews, explaining a chart in one clear sentence counts as much as building it.
Quick check
Which is not a good habit?
3D distorts sizes: it decorates instead of informing.
Lesson 3 done.
Next: Power BI’s filter context. In Power BI, a measure has no single value: it’s worked out again for every cell, under whatever filters apply there.
In the interview: charts that answer the question.
EasyWhich chart would you use for each question?
EasyWhich visual would you choose for each question?
EasyA regional sales manager asks for a dashboard. What do you build?
HardWalk us through a dashboard you built
Ask Prism about this lesson
Sources
- Cleveland and McGill, “Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods”, Journal of the American Statistical Association, 1984: why lengths read better than angles.
- Diez, Çetinkaya-Rundel and Barr, “OpenIntro Statistics”, 4th edition, 2019 (CC BY-SA): histograms, scatter plots and box plots.