Why You Should Ignore Most Results from “Scientific” Studies in Sport Science

Fooled by non-ergodicity

A major problem with scientific studies in Sports Science—which then set trends and guidelines for most practitioners—is their bias toward favoring higher-risk interventions if they yield better short-term results.

In summary, we are treating as ergodic processes that are by no means ergodic.

To explain this, I’d like to bring in the concept of ergodicity. We say a system is ergodic when the average outcome of repeating an intervention many times in one individual is equal to the average outcome of applying that intervention across a group. For example, if 100 people roll a die 100 times each, we can expect the average result for each person to be similar to the group average on the first roll. This would be an ergodic system.

When a system is ergodic, we can predict its future trend simply by looking at a few examples. Knowing the average of the first roll (±3.5) is enough to know what the average will be after 100 rolls for each person. This means the result of one action doesn't affect the next.

Ergodic processes are indifferent to history, but non-ergodic ones are not. In non-ergodic systems, there is a risk of ruin—the risk of losing everything—and if you play indefinitely, the expected return is zero. For example, imagine a very brave cyclist who has a 20% chance of winning a stage and a 10% chance of getting injured because he takes many risks. If he races 10 stages, how many wins can we expect?

The typical answer would be 2 wins, but that’s incorrect. If he crashes in the first stage, he won’t race the next ones. There’s only a 90% chance he’ll race the second stage, and an 81% chance for the third. Therefore, the probability of winning the second stage is just 18%, and 16% for the third. Over the course of 10 stages, taking this into account, we realize that the expected return is 0.71 wins, not 2.

Well, it might surprise you to learn that scientific studies and training programs treat non-ergodic processes as if they were ergodic, leading to incorrect decisions—and hardly anyone seems to notice.