Charts that answer the question
A good chart answers one question so clearly that a busy person gets it in five seconds. This lesson shows you how to pick the chart from the type of question, write a title that states the finding, make the one line that matters stand out, index groups of different sizes to 100, and spot the tricks that make a chart lie.
By the end you can
- Pick a chart from the question: comparison, trend, distribution, relationship or part of a whole
- Write a title that states the finding and highlight the one series that matters
- Index weekly sales to 100 so groups of very different sizes fit on one chart
- Spot and fix misleading charts: cut bar axes, dual axes, 3D, sized bubbles, picked date ranges, cumulative lines and too many colours
The everyday picture
Think of the metro map in any big city. It is not an accurate picture of the city. Streets, parks and real distances are gone. Stations sit at even gaps, and lines run only straight or at 45 degrees. A satellite photo is far more complete, and nobody can use it to find their train.
The map works because its makers decided what question it answers: "which line do I take, and where do I change?" Everything that helps with that question stays. Everything else goes. One line can be drawn in bold when the station staff want you to notice it, for example a line that is closed this weekend.
A chart is the same kind of object. A spreadsheet with 54 stores and 8 weeks of sales is the satellite photo: complete and unreadable. Your chart is the metro map. It leaves almost everything out, on purpose, so the one thing that answers the question stands out.
Start from the question, then pick the chart
Most bad charts start in the wrong place: someone selects a table, clicks the chart button and accepts whatever comes out. Start from the question instead. In da-01 you turned "East city sales are down. Why?" into precise questions, and in da-03 your pivot tables found where the change sits. Each of those questions has a type, and each type has a chart that answers it best.
- Comparison ("which stores fell most?"): a bar chart, sorted from biggest to smallest. Use horizontal bars when the labels are long, such as store names.
- Trend over time ("when did it start?"): a line chart, with time running left to right at even steps.
- Distribution ("how are basket sizes spread?"): a histogram, or a box plot when you compare the spread of several groups. Describing data honestly, in the statistics course, goes deeper.
- Relationship ("do stores with more morning out-of-stocks lose more sales?"): a scatter plot, one dot per store.
- Part of a whole ("what share of sales is fresh?"): a stacked bar, or a 100% stacked bar when you compare shares across stores or weeks. A pie chart only works for two or three parts.
When the reader needs exact numbers to copy into a budget, use a small table instead of a chart. A chart shows shape and size; a table gives values.
Why bars and lines so often? In 1984 William Cleveland and Robert McGill tested how accurately people read different kinds of chart. People judged position along a common scale most accurately, then length, then angle, and area and colour shade worse still. A bar chart encodes values as position and length from a shared baseline. A pie chart asks you to compare angles and areas, which people do badly. That is why a pie with seven slices of similar size is almost impossible to rank, and the same numbers as sorted bars take a second.
Quick check
The head of stores asks "which of the 54 East city stores lost the most sales, comparing the last 4 weeks with the same weeks last year?" Which chart answers it best?
It is a comparison across items, so sorted bars win: the biggest fallers sit at the top and the store names fit on horizontal bars. Fifty-four lines would be a tangle, a pie shows shares and not change, and a scatter plot answers a relationship question.