A colleague asks for 'a chart of the sales data' — what must you settle before choosing between a bar, a line and a point?

## Start from the comparison, not the chart menu The **mark** is the thing actually drawn for each row or each group — a point, a bar, a line, a box, a tile. Choosing one is not a matter of taste, because a mark decides which comparison a reader can make *accurately* and which they have to guess at. So the first move is not to open a gallery of chart kinds; it is to write down the sentence the reader should be able to say out loud after looking for five seconds. That sentence always names a comparison, and nearly all of them are one of five: - **Magnitude** — which is biggest, and by how much. - **Change along an ordered dimension** — how a quantity moved across positions you can interpolate between. - **Distribution** — how one column's values are spread, where they cluster, how far the extremes reach. - **Relationship** — whether two measures move together across the same rows. - **Rank** — who is where in an order, and whether that order shifted. ## The five comparisons and what the mark must give the reader | the comparison | what the reader must be able to do | the mark that supports it | |---|---|---| | magnitude | lay lengths against one shared baseline | one bar per group, every bar from the same zero | | change along an ordered dimension | follow a path and read a slope off it | a line along that dimension, one per entity | | distribution | take in many values' positions at once | one point per row along a single axis, or a summary mark drawn from those values | | relationship | read two numbers per row jointly | one point per row, positioned by both measures | | rank | walk down an order and still see the value | ordered bars, or one mark per entity per period so order changes are visible | The table is a starting point, not a lookup. What makes it work is the middle column: name the perceptual job first, then pick a mark that makes that job perceptual rather than arithmetic. A reader who has to do subtraction in their head has been handed the wrong mark. ## Why swapping the mark changes the answer A mark is not decoration laid over a fixed set of numbers. The same numbers drawn three ways produce three different readings: - as **bars from a shared baseline**, the reader compares magnitudes confidently and sees no trend at all; - as a **connecting line**, the reader reads a path between the positions, and reordering them would change what they see; - as **segments stacked into one bar**, the reader reads the total accurately but can compare only the bottom segment, because every other segment floats on the ones beneath it. None of the three is wrong in general. Each is wrong for the comparisons it does not serve. That is why the comparison has to be settled before the mark is chosen, and why the most common charting failure looks entirely professional. ## When nobody can name the comparison 1. **Write the sentence out in full**, with the entity and the direction in it — "Region C took more than any other region last quarter", not "something about regions". A sentence you cannot finish is a chart you cannot choose a mark for. 2. **Check it needs only one comparison.** Two comparisons wanted at once is two pictures, or one picture plus a stated number; a single mark asked to serve both is usually read as neither. 3. **Then choose the mark**, and prefer the one that turns the reader's job into looking rather than calculating. ## What it costs to change your mind How cheaply you can try a second mark depends on the design in front of you, and the two families genuinely differ. Where a chart is **described as a base plus layers** — held as a value assembled from a data binding, marks and scales, and rendered only at the end — swapping the mark is one substitution and nothing else moves. Where drawing is **a session of calls against an implicit current target**, so that each call applies to whichever chart was drawn last and the order of statements is the program, each kind of chart is a different call, and the labelling, the limits and the legend are rebuilt by hand around it. Neither design is wrong, and this is not an argument for one of them. It matters because the second makes "let us look at that as a different mark" expensive enough that people skip the check — and a mark chosen in the first minute, never re-examined, is exactly the one that survives into the board pack answering the wrong question. ## What an interviewer is listening for That you ask what the reader is meant to conclude before you say a chart kind at all; that you can name more than two comparisons and say what each one needs perceptually; and that you treat "it looked better" as a reason to check rather than a reason to ship.