Photo by rawpixel.com on Pexels.com
By Phil La Duke
Last week I wrote a post about my disdain for “predictive analytics” and a reader disagreed with it and told me I should read Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die By Eric Siegel, So I did. The book isn’t about safety per se, but a fair amount of companies are dumping millions into the attempt to predict how people will act in the context of safety. Don’t get me wrong, Predictive analytics, that is, the clues that tell us how a person is likely to behave in a given situation, has been around for a good long time. Market research is a great example, millions are spent on analyzing buying trends to determine whether people prefer Fruit Loops over Cheerios, or whether showing idiots amazed at the obvious will sell cars. But can this theory be applied to worker safety, and if it can be so what?
There is a couple of real big problems with predictions and human behavior is that people don’t usually act the way they do because of a predictable and rational reason. They tend to act this way because they are emotional. Brain researchers have determined that we make our decisions based, not on logic and careful analysis, but on emotion. I know that some of you are resisting this idea, and frankly so did I at first. But, the researchers also found is that people tend to make their decisions in the part of the brain that controls emotions, and then justify what they already want to do by seeking out only that information that supports it.
So if you read a book to which you have formed an emotional attachment that book becomes sacrosanct and nothing I can say, write, or do will convince you otherwise. So if someone says we can change the world using predictive analytics, I say “prove it” oh, but I won’t be spending millions letting you prove it.
Irrespective of the book and books like it, I still say bunk, and here’s why:
- Prediction is easier the closer it is to the predicted outcome. For example, if you see a person fall off a 30 story building you can probably predict his demise with amazing accuracy, but if you see a man sitting in a diner drinking coffee your prediction that he will die at 12:34 p.m. a month later. Now the advocates of predictive analytics would argue that given enough information, how often does he approach the edge of the building, how frequently is he on top of that building, how much expertise does he have, how much stress is he under, how much training has he had, how frustrated is he on the job, how much time has it been since his last bowel movement etc. they would be able to predict that he would fall and when it would happen, hell they could send flowers to the widow before he hit the ground! What a time saver!
There are companies who successfully use predictive analytics to determine which of their employees are most likely to leave the company. Okay, great. These companies then intervene and prevent the people from leaving, “at a great rate of success”. But how do they KNOW this is working? I guess one could say that the proof is in the pudding but I am dubious. First, I’d like to know how many would have left if they hadn’t intervened. I have worked at jobs I absolutely hated, but found it difficult to actively look for another job because recruiters tend to be lazy gits who only want to interview during the work week, which would cause me to take a day off and likely tip off my current employer. So while one could predict that I would be looking for another job, or would accept an offer, one could NOT predict my emotional state and therefore my ultimate decision. A 90% success rate is better than a 10% success rate to be sure, but in that 90% success rate is there the element of luck at play? I enjoy shooting craps. I know that certain bets are just foolhardy, but I make them anyway. Casinos, know that free drinks, bright lights, sleep deprivation, and other factors make it more likely that people will bet more foolishly and they put all those factors into play. They also know that if I am having a good time—win or lose—I will stay at the table longer and statistics predict that I will ultimately start losing. But there are many times that I will walk away from a table with a big score. I can assure you it wasn’t because I am a highly skilled gambler, but because I got lucky and won enough to satisfy me. Casinos might be able to predict that I will lose and how much time it will take me to lose, but there is chance that I will win, and their predictions will be wrong.
2. Why do we accept predictive analytics but reject forecasts? Many of us disregard the weather forecasts despite the fact that meteorologists using powerful computers have analyzed thousands of bits of data to create a fairly reliable same- or next-day forecast but when the weather forecast is five days out, it is darn near impossible to predict. We accept that long-range forecasts are little more than guesses, yet we have not compunction against dismissing these forecasts as not-so-educated guests.
3.The problem with prognostication. Whether we call it trend analysis, forecasts, or predictive analysis or whatever, is they base the prediction on a snapshot or a trend and ignore the fact that all the moving parts are going to keep moving and shifting and the more we try to master our ability to predict the future the bigger fools we become.