When training zones fail

Definition and purpose

Training zones are a categorical representation of an intensity metric like power, speed or heart rate. Zones can be both descriptive, allowing retrospective analysis, or prescriptive, providing an easy-to-interpret workout intensity to athletes. Training zone definitions are the result of a “binning”, “chopping” or “discretization” process where certain ranges of intensity are grouped together. Some examples of the prescriptive use of training zones are “1 hour in zone 1” and “6x6 minutes in your threshold zone”. Descriptive use of training zones could be “25% of your training in the last month was done in zone 2”.

Challenges with determining training zones

The first problem that arises when defining the training zones: What are the lower and upper limits of each zones? Should all zones be adjacent, or can zones even overlap? This are not trivial questions and have led to a plethora of different zone definitions over the years. Most often, zones are defined as percentages of a threshold metric (e.g. power, speed or heartrate). Assuming that an accurate threshold value is available (which is a big assumption in practice), this may work on a population level and for some individuals, but it is well-known that this does not work for most individuals (just do a search for “distribution of %VO2mas at threshold to get the answer). Basing training zones in this way entirely on (often roughly estimated) threshold values gives a false sense of accuracy that really does not benefit the athlete.

The binning problem

Secondly, there is a more technical problem with binning a continuous metric and this problem is especially relevant when using training zones for descriptive purposes. It is called the “binning problem”, where artificial boundaries (in this case the edges of the zones) can introduce interpretation errors. This problem is not specific to the sports domain but actually a common problem in data science.

An example of the binning problem is a workout executed at the upper limit of zone 1: Some of the data points of this workout will have values just below the zone 1 upper limit and some will have values just above it in zone 2. In absolute sense, the “load” of these datapoints is very similar, but they are weighted very differently when using discrete training zones.

In our sports domain, the binning problem is exaggerated when looking at historical training data over longer periods of time because then the zone edges not only shift, but they do so discontinuously (often when new lab tests are done).

Upper zone limit gravitation

There is another problem with zones when using them for prescriptive purposes that is not rooted in data, but in psychology: Ask a competitive athlete to run for 1 hour at 10-12km/h and he/she will most probably run for 1 hour hovering around the upper limit of the training zone range. I call this phenomenon the “upper zone limit gravitation” (but there is probably an existing term for it. Let me know if you know it!).

For some zones, especially low intensity zones, this does not have to be a big problem. However, for other zones (like the “threshold” zone that is upper limited by MLSS) hovering around the upper limit can significantly reduce the quality of the session by lowering the volume that you can accumulate (see the article on threshold training for more information).

Going back to the example, would we have asked the same athlete to run for 1 hour at 11km/h, he/she would have done exactly that (granted, some athletes might still have returned with 11.1 km/h…).

Metabolic demarcations

While I argue that training zones can oversimplify the complexity of athletic performance and can introduce errors in analysis and interpretation, I do appreciate that certain metabolic demarcations (like a carefully defined threshold intensity or (speed or power at) VO2max) are still valuable as metrics for monitoring or to drive workout intensity prescription. My only point is that we do not need to create training zones to benefit from these demarcations.

Moving beyond discrete training zones

In conclusion, I think that discrete training zones are not needed at all.

For descriptive purposes, a continuous representation of intensity metrics, if needed enriched/overlayed with metabolic demarcations, is all you need. In practice, this means that if you want to monitor or analyze training volume at different intensities over time, you would just use the volume at e.g. 11 km/h or at MLSS.
For prescriptive purposes, there is the principle of specificity: The human body doesn’t care if you call cycling at 200 Watt “zone 2”, “LIT” or “comfortable chatting zone”, it will respond to a training stimulus at that intensity with an adaptation that improves your performance at that intensity. Using a (quasi-)continuous metric like power, speed or heart rate directly will serve that purpose. Athletes will eventually control their intensity by that metric anyway.

GSSNS