A weekly report stacks five teams' ticket counts into one band each; leadership asks which team improved most — why does that chart fail them?

## What a shared baseline buys the reader The **mark** is what gets drawn for each row or group — a point, a bar, a line, a band. When several marks are magnitudes drawn from **one common baseline**, the reader compares lengths that all start in the same place. That is the single most accurate comparison a chart can offer: it needs no arithmetic, it survives small differences, and it works across a dozen groups at a glance. Almost every honest magnitude chart is built on that property, and almost every magnitude chart that misleads has given it up somewhere. A stack gives it up on purpose, and for a reason. ## What a stack actually asserts 1. **The bottom band is honest.** It runs from the baseline to its own value, so its length is a magnitude and its change across weeks is readable. 2. **Every other band is measured from a moving floor.** Its lower edge is the running sum of everything beneath it, so the reader can only judge its *thickness*, and thickness without a shared starting point is read far less accurately than length from a fixed zero. 3. **Movement propagates upward.** When a lower band grows, every band above it shifts up bodily. Its thickness has not changed and its count has not changed, but its position has — and position is what the eye tracks first. The reverse is just as bad: a band that genuinely improved can be pushed around until the improvement is invisible. Point 3 is what makes the stack actively dangerous for a change question rather than merely imprecise. The picture moves when the data did not. ## Which comparisons the stack serves | the comparison | does the stack serve it? | |---|---| | the total each week, and roughly how it is composed | yes — the top edge is a length from the baseline | | which component is largest at one point in time | partly — thicknesses compared across different starting heights | | which component changed most across weeks | no — only the bottom band has a fixed floor | | each component's share of the total over time | only if every stack is drawn to a full height, and then the total is gone | So the chart in front of us is not a bad chart. It is a good chart for the total, handed to someone asking about change in the components, which is the general failure this leaf is about: a mark that answers a question other than the one asked. ## What to draw for "which team improved most" - **One baseline per team.** One line per team in one frame gives every team the same floor and makes the change in each one readable, at the cost of crowding once there are many teams. - **Draw the answer itself.** If the question is genuinely "who improved most", the improvement is one number per team, and one bar per team from zero answers it directly — the shape of the weekly path is not part of the question. - **Keep the total, separately.** A stack is often defended with "but we need the total too". Give the total its own mark rather than making one picture carry both comparisons; a mark asked to serve two questions is usually read as neither. The stack earns its place when the reader's question really is about the whole: how big was the week, and roughly what made it up. It also holds up reasonably with **two** components, because the bottom band is exact and the second can be read as the difference between the total and the bottom. Past three or four components, nothing but the bottom band and the total remains legible. ## Where the designs differ What happens when a second column is bound to colour is not the same everywhere: some designs stack the resulting bands by default, some place the groups side by side sharing one baseline, and some draw them overlaid. The same intent expressed in two tools can therefore produce two different comparisons, and the author often never notices because both pictures look deliberate. **Check what was drawn, not what you meant** — the fastest check is to ask whether every mark you care about starts at the same place. Which order the bands appear in is a separate matter again and can also change how the picture reads. ## What an interviewer is listening for The phrase "common baseline" and a correct account of who has one. Then the propagation point, which separates people who have been misled by a stack from people who have only read about it: a lower band's movement displaces everything above it, so a flat component can look volatile and a real improvement can vanish. Then generosity about the stack — say what it is genuinely good at rather than condemning it — and finally an alternative with its cost stated, including what you would do about the total.