The Variability Index
Aer the calculation of averages and, especially, the mode, Daniel asked,
“Now what’s missing to completely characterize each operator perfor-
mance. Remember, omas said that we needed two numbers. What
should be the other number?” omas explained that this second number
should normally measure the variance of the distribution of all the num-
bers collected. Daniel nodded and explained that although the standard
deviation would be the correct indicator, it might be complex to compute,
use, and explain on the shop oor. He then proposed a simple method
that consisted of counting the number of occurrences in the mode bar and
the two adjacent bars, the le and the right to the mode bar. is number
of occurrences is then divided by the total number of occurrences. He
called the result the Variability Index.
He did the calculation together for
the two groups and found for Group1: 31/37 = 84%, and 21/37 = 56%
for Group2 (Figures 7.1 and 7.2). He explained that this was completely
consistent with the assertion that the operator of Group 1 was the best
performer. As predicted, the best performer was the one with the least
variation. One of the participants noted that these numbers depended
highly on the width of the range, and underscored that for a range of three
for instance, the variability indices would not have been the
same. Daniel oered the following explanation. “You are right. is is a
question people have asked me very oen. Please let us keep things simple.
e author used the term Eciency for a variant of this indicator partly published in Patchong, A.,
T. Lemoine, and G. Kern. 2003. Improving car body production at PSA Peugeot Citroën. Interfaces
33 (1): 36–49. Goodyear and some Japanese companies (including Sumitomo Rubber Industries,
Ltd) use the terms Mode Ratio or Mode rate. We opted for the term Variability Index to underscore
that the indicator is variability-related. Also we adopted Japanese companies’ computation way,
which appeared to be less accurate but simpler and therefore easier to implement on the shop oor
than the one originally developed by the author.
Reminder: e current charts were done with a ve-second range.