Matching the Mark to the Comparison

## What a connecting line asserts The **mark** is the thing actually drawn for each row or each group — a point, a bar, a line, a box, a tile. Most marks make a claim about one row at a time. A point says *an observation sat at this position*. A bar says *this group's quantity is this long, measured from a baseline every other bar shares*. A line mark is the odd one out: it is the only everyday mark that makes a claim about the space **between** two rows, and that claim has three parts. 1. The horizontal positions are **ordered** — one genuinely comes before the next, and the order is a property of the data rather than of the order the rows happened to arrive in. 2. Positions **between** the drawn ones exist and mean something, so a reader may put a finger halfway along and read a value off. 3. The quantity **passed through** those intermediate values on its way from one endpoint to the other, at the rate the slope shows. All three hold for a dimension you can interpolate along: a timestamp, a dose, a distance, a price point. None holds for eight unrelated product categories. There is no position between *Footwear* and *Garden*, nothing was measured there, and no fact of the world makes one the predecessor of the other. The line draws a path nobody travelled — and because the eye reads slope long before it reads tick text, that path is the first thing the reader takes away. ## The test that settles it in ten seconds Reorder the categories and redraw. **If the silhouette changes while every number stays identical, the silhouette was never in the data.** Alphabetical order gives one shape; largest-first gives a clean descent that reads as decline; the order the rows arrived in gives a third. An ordered dimension fails this test in the other direction: you cannot put March before January without lying about the data, and that is precisely why a line is allowed there. ## Ordered, unordered, and the case in between | the horizontal dimension | is there a meaningful point between two values? | what a connecting line then means | |---|---|---| | timestamp, dose, distance | yes, and the spacing is real | a rate of change the reader may interpolate | | an ordinal scale — small, medium, large | the order is real, the spacing is not | the direction is honest, the steepness is not | | unordered categories | no | a path that was never travelled | The middle row catches careful people. With an ordinal dimension the line is not lying about sequence, so nothing looks wrong — but the horizontal gaps were chosen, usually as equal steps, and the slope inherits that choice. Either say what the spacing means or drop the line. ## What to draw instead - **Bars from one common baseline**, one per category: length is read against a shared zero, the category positions carry no quantity at all, and reordering changes nothing a reader would take as a measurement. - **One point per category along a single line of positions**, when the labels are long or the categories are many — the same shared baseline, far less ink. - Choose the category order deliberately and make it legible, so the reader understands it as an editorial decision rather than mistaking it for something measured. - Keep the line for the moment the horizontal dimension becomes interpolable: the same categories measured every month, one line per category, each line tracing a single entity through time. There is one honest exception worth knowing. A chart that draws a line between two positions **to follow the same entity across two states** — the same team before and after, the same account at two dates — is using the line to carry identity rather than interpolation. The reader is meant to see which end connects to which, and the slope is a difference, not a rate. That reading works only because there is nothing in between to misread. ## Where the designs differ Charting interfaces do not agree about whether to stop you. Some treat a discrete horizontal dimension as genuinely discrete and will not connect marks until you state which rows belong to one series; the refusal is the design telling you the claim needs a justification. Others simply join the marks in the order the rows arrived and say nothing at all. **A chart that rendered without complaint is not a chart that was checked** — the absence of a warning tells you which tool you are holding, not whether the picture is true. ## What an interviewer is listening for Not "lines are for time series", which is a rule learned by rote and wrong at the edges. What earns the mark is naming the claim out loud — *a line asserts that the space between the marks exists and was travelled* — then applying it to the dimension actually in front of you, and finishing with what you would draw instead and why that alternative asserts less.