Six rules that every chart page in these docs follows.
Find the question your reader is asking, then pick from the charts that answer it.
How far is each value from a reference: a target, an average, zero?
Diverging bars, a line with a reference line, a surplus and deficit area.
Do two measures move together?
Scatter, bubble (a third value as size), connected scatter, an XY heatmap for dense data.
Which item is first, and by how much?
Sorted horizontal bars, a dot plot, a lollipop chart.
Avoid: Unsorted bars. Sort by value unless the categories have a natural order.
How are the values spread, and where do they cluster?
Histogram, box plot, violin, a strip or dot plot for small samples.
Avoid: A bar of averages alone, which hides the spread.
How did a value move across days, months or years?
Line, columns for a few periods, area for one series, candlestick for prices, a calendar heatmap for daily data.
How big is each value, compared with the others?
Bars or columns, grouped bars for two measures, a dot plot when the values sit far from zero.
Avoid: Bubbles or areas when the reader has to compare sizes.
What share of the total does each part hold?
A stacked bar (100% when only shares matter), a pie or donut with a few parts, a treemap for many or nested parts.
Avoid: A pie with many slices or similar values. Use sorted bars.
Where does it happen?
A choropleth for rates, proportional symbols for counts, flow lines between places.
Avoid: A map when location is not the point. Large, empty regions outweigh small, dense ones, so a ranked bar often reads better.
How much moves from one state or place to another?
Sankey, chord, a network graph, a funnel for steps that only lose volume.
How well readers compare values, from most to least accurate.
Put the comparison that matters most near the top of the list. Area and color intensity suit patterns and outliers, not reading exact values. This ranking comes from Cleveland and McGill's studies of graphical perception.
A chart is worth its space when the shape of the data is the message.
For the pitfalls of individual chart types, the data-to-viz caveats go deeper.