scatter

When to use

• Two numeric measures, and the relationship is the point. Cost against value, price against adoption, risk against return, effort against impact — with real numbers on both. The slide's claim is that the two move together, or that one entity sits off the pattern. If only one measure carries the argument you have a ranking, not a scatter: use progress.
• The units matter and the reader may want to read a value off the chart. This is the line between scatter and quadrant. A scatter prints a tick ladder and a gridline on BOTH axes, in the author's own affix — $0 · $100k · $200k, 20% · 40% · 60% — so 'Atlas costs about $420k' is readable from the picture. A quadrant prints only the two extremes on a unitless canvas; you can say which box a thing is in and nothing more.
• Five to ten entities, each with a short name. Below four there is no pattern to see — state it in prose or a stats row. Past about ten the names start needing leaders and eventually get dropped rather than overprinted. Every dot carries its own name, so keep names to a word or two: a long one wraps to three lines and crowds its neighbors out of position.

When not to use

• A unitless 2x2 score. If the axes are unitless 1-to-10 judgments and the read is which of four named zones an item lands in, use quadrant.
• A trend line over a handful of points. scatter trend refuses a least-squares line under five points. Even at eight it says 'these move together', not 'this predicts'.
• Points closer together than the eye can separate. Four tools within four points are four dots inside one dot's width: the ring keeps the overlap visible, but the names travel.
• Time on the x axis. A series measured at successive dates is a line, not a cloud — the reader needs the connection between points. Use line.

Common mistakes

• Splitting a point's two numbers into one comma-separated pill, ` 4.2, 8.1 , the way quadrant` takes them. A scatter takes TWO separate pills — ` 4.2 8.1 — because each axis carries its own unit and its own affix ($420k on x, 18%` on y), which one shared pill cannot express. An item without two numeric pills is skipped rather than plotted, so the point disappears from the chart and from its description.
• Expecting the axes to start at zero. They do not, and that is deliberate. The domain is the data's own range plus about 8% of air. Two measures with narrow ranges — margin 38-44%, NPS 51-58 — forced to include zero collapse into one corner and the relationship disappears. A non-negative series still gets its air below zero — a bubble sitting at zero has to fit inside the plot — but its axis never prints a negative tick.
• Writing the axis names as a normal one-pill eyebrow, ` Cost vs value `. That is the chart eyebrow and it stays in the masthead; the plot then has unlabeled axes, which is the one thing a scatter cannot survive. Write the two captions as two inline-code spans in one paragraph.
• Assuming a name that does not appear on the plot was lost. A name with nowhere left to sit is dropped rather than painted through its neighbor — two overprinted names are two names lost, not one. The name still rides data-label on its dot, the mark-detail popover, and the <desc> a screen reader reads. Fewer points, or shorter names, brings it back.
• Encoding a third measure in the dot's RADIUS. bubble scales AREA, never radius, because radius-encoding overstates by the square: double the number and a radius-scaled dot looks four times the quantity. The area runs linearly from a minimum visible size, so the smallest value is still a circle you can see, and a point with no third pill is drawn at that floor and flagged rather than given a magnitude nobody typed. It also needs a size key, which is why the third inline-code span on the axis line names the measure. Past about ten bubbles the areas stop being comparable at all — split the slide.

Authoring

<!-- _class: scatter -->

`X measure` `Y measure`

## Two measures, one relationship.

- First entity `4.2` `62`
- Second entity `2.1` `38`
- Third entity `6.8` `81`
- Fourth entity `3.4` `55`

Slots

Data shape

• A point is a name plus TWO trailing inline-code pills — ` Atlas $420k 18% — x then y, in that order. A third pill is the bubble magnitude and is ignored without the bubble` class. Fewer than two numeric pills and the item is skipped, not plotted at zero.
• Magnitude suffixes SCALE: 1.2M is 1 200 000 and 800k is 800 000, so the two can share one axis. % is not a magnitude, so 12% is 12. A minus written outside the currency symbol (-$400k) is understood as negative.
• The affix is per axis and is adopted only when EVERY value on that axis agrees on it: six values in $ give a $ axis, and a mixed series ($4M, 12%) gets a bare one, which is the honest read of an axis that cannot describe itself.
• Both axes are linear. There is no log scale, so a series spanning several orders of magnitude will pile up at one end — take the log yourself before authoring and say so in the axis caption.

Anatomy

┌─────────────────────────────────────────┐
│  header                                 │
│   What we pay most for, nobody uses.    │
│                                         │
│       80% |            (o) Ember        │
│              |      (o) Delta           │
│            40% |  (o) Cirrus            │
│          | (o) Borealis  (o) Atlas      │
│       0  +-----------------------       │
│           $0k     $200k    $400k        │
│                   ANNUAL COST           │
│                                         │
│  footer                           1/19  │
└─────────────────────────────────────────┘

Variant decision rule

• default (no modifier) — Two measures per entity — the plain XY plot.
• bubble — A third numeric measure (seats, headcount, revenue) should scale each dot's AREA. Keep it to about ten entities, and name the measure in the third inline-code span so the size key has a caption.
• trend — The claim is explicitly that the two measures move together, there are at least five points, and the audience will read the line as a summary rather than a forecast.

Variants

<!-- _class: scatter bubble -->

`Annual cost` `Teams adopting` `Seats`

## bubble sizes each dot by a third measure.

- Atlas `$420k` `18%` `1200`
- Borealis `$310k` `24%` `640`
- Cardinal `$180k` `52%` `900`
- Dovetail `$95k` `61%` `310`
- Everline `$240k` `31%` `180`
- Fathom `$60k` `74%` `450`

<!-- _class: scatter trend -->

`Annual cost` `Teams adopting`

## trend draws the least-squares line through the cloud.

- Atlas `$420k` `18%`
- Borealis `$310k` `24%`
- Cardinal `$180k` `52%`
- Dovetail `$95k` `61%`
- Everline `$240k` `31%`
- Fathom `$60k` `74%`
- Juniper `$400k` `20%`
- Keystone `$88k` `66%`

Related

• quadrantthe axes are unitless scores and the read is which of four named zones
• matrix-2x2items are placed by category, not by coordinate
• radareach entity is rated on more than two measures
• progressone measure per entity, compared as lengths
• list-tabularthe exact numbers matter more than the shape of the relationship