Run chart

The points are your measure in the order it happened: a weekly rate, a daily count, a monthly average. The line across them is the median, the middle value, so half the points sit above it and half below when nothing is changing. Patterns that would be unlikely by chance tell you something has changed.

It is the chart to start with. It works with as few as ten or twelve points, needs no assumption about how the data is distributed, and anyone on a team can draw it by hand.

Use it whenever you are testing a change and want to know if it worked: a new discharge routine, a new booking system, a change to a form. Plot a baseline before the change, keep plotting after it, and read the rules. Health and care improvement teams use it as their everyday chart, and it suits any service where data is reported weekly or monthly.

It tells you whether something has changed, not whether the process is stable in the sense a control chart means, and it will not pick out a single unusual point as reliably. Once you have twenty or more points and need to know whether the process is predictable, move to a control chart. And as with any time chart, do not put two different processes on one line.

A shift: six or more points in a row all above or all below the median. Points that sit exactly on the median are skipped; they neither break nor add to the run.

A trend: five or more points in a row, each higher than the one before, or each lower. A point equal to the one before is skipped.

Too few or too many runs: a run is a series of points on the same side of the median. Count them. Published tables give the range of runs you would expect by chance for your number of points; fewer suggests a shift, more suggests the data is being mixed from two sources.

An astronomical point: one value so obviously different from the rest that anyone looking at the chart would agree it is unusual. This rule is judgement, so agree it as a team rather than alone.

A medical ward tracked the share of its discharges completed before noon each week. After 12 weeks it changed its routine: medicines to take home were ordered the afternoon before discharge, not on the morning itself. It kept plotting for 10 more weeks. The figures are illustrative, built to show the method.

The median of the 12 baseline weeks is 19.5%. The baseline itself shows no shift and no trend, so it is a fair picture of the old process. After the change, all 10 weeks sit above that median: a shift, well beyond the six points the rule needs. The change made a difference, and the chart shows it without any argument about whether one good week was luck.

Two cautions. A run chart shows that something changed at that point, not that the new routine caused it; check whether anything else changed at the same time. And the new level should become the new baseline: extend a fresh median from the weeks after the change, and watch it the same way.

The run chart needs only one calculation, the median, and then counting.

Put the dates in one column and the measure in the next. Find the median of the baseline period and add it as a column of the same value, so it draws as a flat line. Draw both as a line chart, mark the date of the change on it, and extend the baseline median across the later points so the shift can be seen against it.

Run charts belong to Measure, as the first look at how a measure behaves, and above all to Improve, where each test of a change is judged by its run chart. They are the backbone of the plan, do, study, act cycle used in health and care improvement.