What the board room can learn from the shop floor
or, how to avoid being fooled by percent changes
By Ben Joneshttp://dataremixed.com
Meet Jim
(Jim says Hi)
Jim runs a company selling…tractors.
Tractor Sales are down 3.6% this year
-3.6%
Jim wants answers…
Meet Jim’s two sales guys, Joe and Larry
Joe Larry
Larry’s numbers are up 2.9%…
Larry
+2.9%
Joe’s numbers are…not.
Joe-8.3%
What should Jim do?
(Jim is polishing up his resume…)
but what if it was all just a big misunderstanding?
Instead of just looking at % change “Year
Over Year”…
-3.6%
…what if we plotted units sold by month?
and what if we applied some brand new
techniques?
http://amzn.to/dOWKBwWalter Shewhart written in 1939
to understand the variation in the data:
x-bar (or average)
UCL – Upper Control Limit
LCL – Lower Control Limit
“Common Cause” variation
“Special Cause” variation
“Special Cause” variation
So wait, only LAST month is a “down” month…
…and WHO do you think caused THAT?
Joe Larry
Probably Joe, right?
Joe-8.3%
But if we look at Joe’s numbers over time…
-8.3%
2009: 96 Units
But if we look at Joe’s numbers over time…
-8.3%
2010: 88 Units
Joe
We see no statistically significant change!
“Common Cause” variation
But Larry can’t be the reason, his numbers are
up, right?
Larry
+2.9%
Let’s look at Larry’s numbers over time…
2009: 69 Units
Let’s look at Larry’s numbers over time…
2010: 71 Units
Looks like Larry has some explaining to do about
December…
Larry“Special Cause” variation
These two approaches tell two very different stories…
Percent Change “YOY”
Control Charts
Blaming Joe would be mistaking NOISE for a
SIGNAL
Rewarding Larry would be missing the opportunity
to identify a SIGNAL
Either way, Jim’s business loses an opportunity to
LEARN and GROW…
…and the fate of our sales guys’ careers hangs
in the balance…
Joe Larry
The Morals of the Story:• DON’T just compare % change
• DON’T assume a % change is statistically significant just because you don’t like it.
• DO use a Control Chart to filter out noise in a data set and identify signals
• NEVER adjust the y-axis to cross the x-axis at anything other than 0 (did you catch that trick? The column charts grossly over-exagerate the changes)
For bar charts, always set the y-axis to 0
y-axis starts at 157 y-axis starts at 0
Appears to be 4X less
Got all that, Jim?
More about Control Charts (in business)
• Understanding Variation– Donald J. Wheeler
Presentation by Ben Joneshttp://dataremixed.com