One day Ogg, the Cro-Magnon design engineer, was invited by Snrg, the marketing manager, to play goof (an early form of golf) at the Bison Valley Country Club. Snrg was quite good at goofing because he worked in marketing, where this, like eating in fine restaurants, was classified as work. On the other hand, Ogg, like most design engineers, was genetically superior, thus evening the odds. Snrg was using his new ironwood club, carved by the tribe shaman. "It cost me a lot, but its magic powers will improve my game," boasted Snrg, beginning a tradition which endures today in golf stores.
Snrg lifted his club high over his head and whipped it toward the small stone ball. Crack went the club, and the goofball sailed down the fairway. Quite by coincidence, at this very moment, a total eclipse blocked the light of the sun. "Hmm," said Snrg, "This ironwood club is much more powerful than I thought, if it can stop the sun from shining."
One amazing thing about human learning is how we frequently learn the wrong things. The psychological term for this is "superstitious learning," which is making incorrect attributes of causality. This ability to do so is hard-wired into our brain. Such quick judgments of causality aren't necessarily bad. It would be a smelly world if we needed a sample size of 200 before we decided it's a bad idea to play tag with skunks. Still, these automatic attributes of causality can get us in trouble.
A classic error is to automatically assume that one thing causes another if they occur closely in time or space. Though this is often an accurate assumption, it can be incorrect for complex systems in which cause and effect can be separated in both time and space. This complexity makes it much harder to decipher causality within the system.
Let's look at an example. The engineering department is unhappy with the support from purchasing, so they decide that they need their own buyer. It works. They receive parts faster and with less paperwork. The group concludes that a separate buyer for engineering improves product development.
In reality, a separate buyer increases the chance of using vendors who cannot meet the manufacturing cost, delivery, and quality requirements. The true root cause of unresponsive purchasing is usually that "responsiveness" isn't defined or measured. When we begin to measure purchasing on the basis of response time, the problem goes away.
How can we avoid superstitious learning? First, obtain more data. Look for contradictions, situations in which the effect appears without the cause or the cause without the effect. Also, try to understand why the cause produces the effect. This will help you find the true root cause. For instance, a senior management champion will accelerate a development project, but the true cause of acceleration is receiving priority within queues not the sponsorship of senior management.
Last, be wary of time lags. It's easy to overlook delayed effects. For example, buyers transferred to engineering will initially tend to protect the interests of manufacturing because they are used to doing this. It typically takes a year or two before they go native and start causing problems for manufacturing.
In the end, you must acknowledge the possibility that you are learning the wrong thing, and strive for deeper understanding of the system.