The google data science blog had a recent post, “Uncertainties: Statistical, Representational, Interventional.”
Statistical Uncertainty is the fact that the true value may be different from what is estimated. Representational Uncertainty is the fact that what you measure way not fully represent what ultimately want to impact. Interventional Uncertainty is that the full impact is often different from what you measure.
I think a great example of these ideas has to do with health goals. There are multiple dimensions of health. The broadest being mental and physical. Losing weight can improve both and it is common for people to choose weight as a goal. Weight is representing health, but it is not capable of fully representing health.
Weight is a highly volatile metric since it depends on what and how much we consume. I know my weight throughout the day can vary 3lbs – 5lbs, but I do not view myself as being heavier or lighter in terms of health based on this variation. Drivers of the statistical uncertainty are from the scale (usually small), from recent consumption, and from water retention.
The main driver of short-term weight loss is calorie deficits. This can be done through a combination of reduced consumption and or increased exercise. All combinations lead to an increased cognitive load from a new behavior and can affect recovery, sleep, and mood.
The intervention of diet and exercise to meet the goal of better health as measured by weight has a negative impact on how we feel during the intervention. Probably why so many people quit – it sucks and it is unclear if it is working.