Don’t read too much into the title above – you wouldn’t want to extract meaning that isn’t there. I have been reading How We Decide by Jonah Lehrer, an excellent book that looks at the neuroscience behind the decisions we make. One key insight for market research professionals is that the human mind wants to find patterns and does a great job at finding them even when they are not present. The title of this post refers to an experiment where Yale college students and a rat were competing against each other in “finding the reward” in a maze. In the experiment, there was a single decision to be made, go left or right at the beginning of the maze. The reward was placed via random assignment with a probability of 60% for the left branch of the maze and a probability of 40% for the right branch. The rat quickly learned that the left branch was more fruitful and chose the strategy of always going left, despite the fact that he was unrewarded 40% of the time. Overall, counting the “learning time”, the rat received his reward just under 60% of the time. The Yale students continued to try and discern a pattern in the reward appearance. They never moved to a strategy of always choosing the left branch, and as a result received the reward only 52% of the time. The students resisted the simple explanation, that the reward was placed randomly, and so, they were outsmarted by the rat. There is a lot of wisdom in remembering the strategy of the rat when we design research studies. Occam’s razor – always choose the simpler explanation – is perhaps the most famous statement of the rat’s approach. W. Edwards Deming, the quality guru, provided the same advice. He used statistical process control to guard against making changes in a system too frequently, because humans would perceive a pattern that was really explained by randomness. Deming showed that reacting to these perceived patterns caused poor performance in manufacturing systems. Chaos theory, such as I understand it, brings this all together. In chaos theory, a large number of deterministic processes produce results that appear to be random. Therefore, we manipulate the small subset of these processes that we can find (or are easy to measure) and ignore the apparently random effects of all the processes we cannot find. As a result, we can generate poor predictions about future events and suboptimal decisions. So my advice, in any modeling effort: let’s not get outsmarted by a rat. Let’s make sure we look for the simplest explanation, which may be that the results are random.

Post Comment