Every drop of ink should carry data.
INKRATIO is a school for the craft of charts. Choose the right form, take out everything that is not data, keep the axes honest, and spend your one bright colour on the point that matters. The meter in the header measures the data-ink ratio of whatever chart you are looking at. Watch it climb.
Erase the junk.
- 0The chart as found
- 1Take down the grid
- 2Drop the frame and shadows
- 3Label directly, drop the colours
- 4Thin the bars, lighten the axis
- 5Make one point matter
Six numbers, buried. A gradient panel, a heavy grid, fake depth, drop shadows, a boxed legend and six colours that mean nothing. The meter reads the share of ink that shows data.
Gridlines are scaffolding. Once the reader can see the bars, take the scaffolding down.
The border, the grey panel and the shadows describe the software, not the rainfall. None of it is data.
A legend makes the eye travel to decode. Put each name where its bar is. Once names sit under bars, six colours are redundant too.
The fake third dimension goes, the bars thin, the heavy axis becomes a hairline, and the values move onto the bars so the scale can leave.
Now spend the one accent. The record month at Harrow Ridge turns crimson and a short note says why it matters.
Same data, five charts.
- aBar chart
- bDot plot
- cLine chart
- dSlope chart
- eSmall multiples
One value per outlet, for 2026 only. Lengths from a zero baseline are the easiest thing for the eye to compare.Works when: one number per category
The same comparison with a fraction of the ink. Position carries the value, so the scale is free to start anywhere.Works when: many categories, close values
Six years appear and the story becomes direction. Social feeds fall every single year while journals climb.Works when: change over continuous time
Keep only the first and last year. Steepness and crossings show who changed most, in one glance.Works when: two moments, before and after
One panel per outlet on a shared scale. Each line is read on its own and nothing tangles.Works when: lines overlap, or there are many groups
Lying axes.
1 The size of an effect shown in the graphic divided by the size of the effect in the data. Honest charts sit near 1. Bars encode value by length, so a bar chart must start at zero; a dot or line chart may not need to.
The eight plates.
Eight lessons, one specimen each. Every specimen can be stripped to the data or bloated with a frame, a panel, a grid and shadows. Watch the meter as you do.
Now make one better.
The makeover clinic
Load one of six bad charts and switch on the fixes one by one: remove gridlines, label directly, start at zero, sort, highlight, reduce colours, annotate. Export the cleaned chart as SVG.
Pick the chart
A data question appears. Choose from six chart forms and see why the right one works. Then try Spot the lie: six charts that mislead, and you name the trick.