Survivorship bias

A recent discussion I had has prompted me to write this host about what I realize to be one of the biggest problems assailing not only the world of nutritional research, but science at large, the problem is, as the name of the title, survivorship bias. Now, what is survivorship bias? Let’s illustrate with an example. Say you have a certain model that is followed, say a certain way to do stock exchange. And then you poll a number of CEO’s, the actual number is not important, say 100 of them and it shows that 100 of them did use this model. This would then be reported as a 100% chance that by following this model you would become a CEO. That is the survivorship bias, because we are told absolutely nothing about either the CEO’s who did not follow this model or the people who followed the model and did not become CEO’s. We are told about one single variable about the past and are asked to generalize about it to the future. The future is by it’s very design unpredictable and lead to chance. A lot of things happen by randomness, which plays a huge part of our lives. The model is not always the reason for success. Boxers in the past ate a horrible diet, sometimes steak right before the match, abused their bodies without getting sufficient rest or sometimes even hydration. This does not make a good model for a boxer to follow, but by following only the survivors of the program we are led to think so.

If the survivorship bias is used negatively, that is that we know of something here, that could really be a problem. It’s only variable, but we would urge you to be cautious about it and maybe try to remove it from your life I don’t have as much problem with it. My issue is more when it is used in a positive fashion, positive here meaning, using the survivorship to justify continued use of something on the basis that these people did fine while using it, and maybe even improved on something, e.g. disease reversal. As long as we are told only one variable, doing these comparisons are EXTREMELY dangerous and irresponsible.

Advertisements