Real-Life Probability

GE used to have a human resource policy that they would let go the bottom 10% of their workforce. Assuming the performance review results fit a bell curve (normal distribution), that means you couldn't be more than 1.28 standard deviations below the mean. If you remember anything about results from a stable, controlled process, it doesn't take much to find yourself 1.5 or 1.8 standard deviations from the mean.

We would say the performance review process and the human resource system is giving us the results we expect unless we have a result more than 3 standard deviations from the mean, or two out of 3 beyond 2 standard deviations. Deming, who famously exhorted us leaders to drive fear out of the workplace, would have gone ballistic with this system. Getting a low rating would have seemed capricious and random, and getting a high enough rating would have seemed indeterminable and unearned.


Lately a financial advisory company has been showing a commercial that has people creating a histogram from the age of the oldest person they know. It purportedly demonstrates that we're living longer and therefore we need more to carry us longer into our old age. However, it's not a distribution of the average age at death. It's not a picture of the average age of people we know. It's not a picture of the average life past retirement. It's a skewed question giving us skewed results. It's only a picture of the right-hand tail of an age distribution. And not a very good one...

The commercial assumes people are randomly selected and not from the same neighborhood or family groups. The results can be skewed because everyone knows the same old man or old woman. In one community, we asked people if they had a drug or alcohol problem. The results were really low. The answer to the question of whether they knew someone else who had a drug or alcohol problem came back ten times the answer to the other question. Either people lied about themselves or they all know the same 'town drunk'.

When we look at business results, we need to understand the likelihood of the outcomes. We all remember decision trees from the business school class on decision-making. By knowing the probabilities of success for a range of strategies or the probabilities of occurrence for a range of events and the monetary results given those circumstances, we can calculate the expected results for the organization. How often do we do this? I have yet to see a company do it, or hear of a company doing it. Yet, it's good information when analyzing the  year-end results, and knowing if we induced a 'special cause' and got extraordinary results...or we just saw normal variation and normal results though higher or lower than the year before.