Are sales really up?

by Phil Cohen

Variation appears in all business data: in sales figures, in prices, in production totals, in costs. The temptation is to try to make sense of every movement in these figures, but this is often simply a waste of time.

A famous teaching experiment by the Total Quality Management guru W Edwards Deming had a student use a wooden 'paddle' to pull a sample from a bucket containing a mixture of red and white beads. The student had no control over the number of red beads that were in the sample. Naturally, the number of red beads varied with each pull.

Deming continually asked the student questions like "You pulled 8 red beads that time, but last time you pulled 12 - why aren't you trying harder?". The point of the experiment is twofold: that some variation is inevitable, and that most variation usually comes from the 'process' (the design of the paddle and the beads) rather than from the person operating the equipment.

By analogy, there will always be some variation in monthly overhead spending, no matter how constant the facilities. Worrying about every monthly rise and celebrating every reduction just doesn't make sense. If managers are to manage from this kind of data, they have to know what is normal and what isn't.

Deming and a number of others provided a powerful tool for analysing the variation in a process (such as overhead spending). This tool, called a control chart, allows managers to differentiate between variation that is inherent ('common cause' variation) and variation that needs to be investigated ('special cause' variation).

There are a number of different types of control chart, but they all have three things in common. First, their construction is based on firm academic statistical foundations. Second, they are generally designed so that they can be used by non-numerate people after a little training. Third, they provide a reliable indication of when special cause variation is occurring.

The most commonly used type of control chart starts with a simple monthly (or other period) graph, and adds two horizontal lines. These lines mark the upper and lower 'control limits' for the graph. Any points on the graph which fall outside these control limits are almost certainly due to special causes, and not to the variation inherent in the process that produced the data.



If for example you constructed a control chart of overhead spending, you should find most of the points on the graph falling inside the control limits. Those points falling outside indicate months of unusually high or low spending, and are worth investigating. The variation within the control limits is ever- present, and can be ignored.

Calculation of control limits can be done with a financial calculator - they lie at the mean plus and minus three times the standard deviation of the data. This assumes that the data is (reasonably) normally distributed, which most financial data tends to be. There are a number of other considerations in spotting special cause data, and if you intend to use a control chart for real, we recommend the following text: "Guide to Qualicy Control", Dr Kaoru Ishikawa, Asian Productivity Organisation.


------------------------------------------------------------------------


This is one of a series of articles written by Phil Cohen and Onno van Ewyk, HCi . Most of the articles were also published in the Australian Financial Review. This article may be reproduced only with the permission of HCi Consulting (email HCi ). Copyright HCi , 1993-1998.