Anti-pattern gallery¶
The same data, shown wrong (left) and right (right). Every pair uses the data of the cases, so the honest version is also the one used on the case pages. The misleading versions are not strawmen - each one is a default setting or a habit seen in real reporting.
1 · The axis that does not start at zero¶
✕ Misleading
✓ Honest
Why the left one misleads. Bar length encodes the value. Starting the axis at 90% turns a difference of a few points into bars that look several times longer, and the rainbow colours suggest the queues are different kinds of thing. What the right one does: zero baseline, one colour, the target as a reference line, and a title that says what the reader should conclude. (Lines and dots may zoom, because they encode position, not length - and should say so.)
2 · The average that hides the tail¶
✕ Misleading
✓ Honest
Why the left one misleads. Repair times are right-skewed: most repairs are short and a few wait days for spare parts. The mean of 'Tool wear' (6.3 h) describes almost no real repair - its median is 1.6 h. Ranking modes by the mean ranks them by their outliers. What the right one does: shows the distribution, so the typical repair and the tail are both visible - and points the decision at the tail (spare parts), not at the technicians. MTTR stays a mean in the catalog for the availability identity, and is always shown next to the median.
3 · Too many series¶
✕ Misleading
✓ Honest
Why the left one misleads. Ten colours exceed what anyone can match to a legend, several hues are indistinguishable for colour-blind readers, and no line is more important than another - so the chart has no message. What the right one does: one highlighted series carries the message, the others stay as grey context, and the label sits on the line instead of in a legend. If every series matters, use small multiples instead.
4 · The pie with too many slices¶
✕ Misleading
✓ Honest
Why the left one misleads. People compare angles and areas poorly; with 5 slices, the ranking of the middle modes is guesswork and the small slices vanish. What the right one does: sorted bars on a common baseline make the ranking and the gaps exact. A pie is acceptable only for two or three parts of a whole (the NPS bar in case 4 is the only part-to-whole chart on this site).
5 · The traffic light without a target¶
✕ Misleading
✓ Honest
Why the left one misleads. The colours compare each machine with the fleet average (95.9%), so about half of the fleet is red by construction, even if every machine met the target - and red/green dots carry no value, no target and no meaning for colour-blind readers. What the right one does: the status comes from the catalog's target and thresholds, the value is printed, the target and the amber band are drawn, and every colour travels with an icon (✓ ! ✕).
6 · Two y-axes on one chart¶
✕ Misleading
✓ Honest
Why the left one misleads. The two scales are independent, so where the line crosses the bars is an arbitrary choice of axis ranges - it invents a correlation (or hides one). Total volume also mixes the queue that changed with three that did not. What the right one does: one axis, both series indexed to the same base, and the comparison that explains the SLA drop (case 1) made explicit. Show the SLA trend in its own chart next to it.
The rules behind the honest versions are listed on the method page; the case pages apply them to every chart.