Run Charts

Adding the element of time will help clarify yourthem. Another way of doing this is to count the
understanding of the causes of variation in yournumber of times the run-line crosses the median, and
processes. A run chart is a line graph of data pointsthen add one. Compare the number of runs you count
organized in time sequence and centered on theto the accompanying chart.
median data value. The patterns in the run chart can• Numbers of runs outside the range shown for the
help you find where to look for assignable causes ofnumber of data points are statistically unusual.
variation.• Too few runs (below the lower limit) generally
What can it do for you?indicate that something cyclic is systematically shifting
Histograms or frequency plots can show you thethe process average.
general distribution or variation among a collection of• Too many runs could point to a problem of
data points representing a process, but one histogramconsecutive, over-compensating process adjustments
or one frequency plot can not show you trends or helpor indicate that the data points actually came from two
pinpoint unusual events. Sometimes, a normal-lookingsources with different process averages.
distribution will hide trends or other unusual data. To• Look for sequences of ascending or descending
spot those trends, the data must be considered in timevalues. Seven or more continuously increasing or
order. Plotting data on a run chart can help you identifycontinuously decreasing points indicates a trend that is
trends and relate them to the time they occurred. Thisshifting the process average. When counting points,
will help you in your search for the special causes thatignore any points that repeat the previous value.
may be adding variation to your process.Repeated values neither add to the length of the run
Run charts are especially valuable in the measure andnor break it.
analyze phases of Lean Six Sigma methodology.• Search for seven or more consecutive points on
How do you do it?the same side of the median line or 10 of 11 points or
1. Select a characteristic from one of your processes.12 of 14 or 16 of 20. (Ignore any points that are exactly
This characteristic could be presenting a problemon the median.) Such a sequence indicates that
because excessive variation often drives it outside ofsomething has occurred to shift the process average
specification limits, or it could be a cause of customerin that direction.
complaints.• A sequence of 14 or more data points alternating
2. Measure the characteristic over time intervals andup and down suggests a variation related to sampling
record the data. Note the time or the time period that(such as one reading early in the day and one reading
is associated with each data point.toward the end) or that the data is coming from two
3. Find the median data value. To do this, list the datasources with different process averages (such as
values in numeric order. Include each data point, even iffrom two machines making the same part.) In looking
it is a repeat value. If the number of data points is odd,for up-and-down alternation, ignore any points that are
the median is the middle value. If the number of dataexactly the same as the preceding point.
points is even, the median is halfway between the two• A sequence of seven or more points with exactly
values nearest the middle. For example, if the collectedthe same value usually should signal you to look for a
data points were: 5, 1, 18, 8, 12, 9, the ordered valuesspecial cause. While it is possible that your process
would be: 1, 5, 8, 9, 12, and 18. The middlemost valueshas improved to the extent that the existing
are 8 and 9. The median is the average of thosemeasurement technique is no longer sensitive enough
values, or 8.5. (Remember, the numerically-orderedto measure variation, it is usually more probable that a
data is only for determining the median. The data mustgauge is stuck or broken or that someone is making
be plotted in time order on the run chart to be of anyup the data.
value.)Now what?
4. Set up the scales for your run chart. The verticalRun charts can be very valuable in helping your search
scale will be the data values, and the horizontal scalefor sources of variation. They are easy to plot and
will be the time. Make the horizontal scale about two toeasy to interpret. The sampling is uncomplicated, and
three times the distance of the vertical scale.there are no statistical computations to make. They
5. Label the vertical scale so that the values will becan also be applied to almost any process or any
centered approximately on the median and so thedata.
scale is about 1 ½ to 2 times the range of theOn the other hand, they are not an instant indicator.
collected data.They are best used for spotting trends; short shifts in
6. Draw a horizontal line representing the median value.the process cannot always be detected with run
7. Plot the data points in sequence. Connect each pointcharts. In addition, special causes that produce general
to the next point in the sequence with a line.piece-to-piece variation will not be readily detected on
Some special cause variation reveals itself in unusualrun charts.
run-chart patterns. These clues can direct you in yourFinally, a simple run chart cannot establish the natural
search for causes. Count the number of runs. Runscapabilities of a process, so it isn’t possible to use
are sequences of points that stay on one side of,one to predict what specifications a process can
either above or below, the median line. One way ofactually meet. To do that, you need to create a control
counting the runs is to circle these sequences and tallychart, a run chart with statistical control limits.