Why Your Watch Overestimates How Many Calories You Burned

Your watch says the run burned 620 calories. A MET-based estimate says about 400. One of them is wrong, and it is usually the watch.

This matters because a lot of people treat that number as a budget. Understanding where it comes from explains why it should not be spent.

What a wearable actually measures

Two things, both well: heart rate, via an optical sensor reading blood flow at the wrist, and motion, via an accelerometer.

It does not measure energy expenditure. It cannot. The laboratory method, indirect calorimetry, works by measuring oxygen consumed and carbon dioxide produced, because that is what energy metabolism physically does. A watch has no access to your gas exchange.

So it infers. It takes heart rate and motion, adds your age, sex, height and weight, and runs a proprietary model that maps those onto an estimated burn.

Every step of that inference introduces error, and the errors compound in the same direction.

Why heart rate is a poor proxy

Heart rate correlates with effort. It is not the same as effort, and it rises for many reasons that involve no additional energy expenditure at all:

• Heat. The same run on a hot day produces a substantially higher heart rate because you are moving blood to the skin to cool down.
• Dehydration. Lower blood volume means more beats to move the same oxygen. Cardiac drift during a long session can add 10 to 20 beats per minute at unchanged pace.
• Caffeine, stress, poor sleep and illness. All raise heart rate. None burn meaningful extra calories.
• Fitness. A trained heart moves more blood per beat, so a fit person’s heart rate is lower at the same workload. Devices frequently read fitter people as working less hard than they are.

There is also the sensor itself. Wrist optical readings degrade with movement, cold hands, tattoos, darker skin tones and loose straps. Weightlifting and rowing are particularly bad cases, because the wrist is flexing constantly.

The size of the error

Studies comparing consumer devices against indirect calorimetry have found error rates in energy expenditure commonly in the 20% to 90% range, with the direction almost always upward.

A widely cited Stanford study of seven popular wrist devices found that most were reasonably accurate on heart rate, with median errors under 5%, but that not one came close on energy expenditure. The best was off by about 27%; the worst by well over 90%.

That pattern has held up across later work. The sensor is good. The calorie model built on top of it is not.

Why estimates always seem to be generous

Some of it is technical. Some of it is not.

A device that tells you the run was worth 400 calories is less satisfying than one that says 620. Manufacturers are not required to publish their models, are not audited on accuracy, and compete on engagement. There is no commercial pressure toward the lower number and some pressure away from it.

The gross-versus-net distinction also inflates things quietly. Almost all devices report gross calories: everything your body used during the session, including the calories you would have burned lying on the sofa for the same period. For a 30 minute session that resting component is around 30 to 40 calories. Over a day of tracked activity it adds up to a meaningful overcount of what the exercise itself contributed.

What MET values do differently

The alternative approach uses metabolic equivalents. One MET is resting metabolism, about one calorie per kilogram of body weight per hour. Every activity is assigned a multiple of that, measured in a laboratory and published in the Compendium of Physical Activities.

calories = MET × weight in kg × hours

Running at 8 km/h is about 8.3 METs. For a 70 kg person over 30 minutes that is roughly 290 calories.

This method has real limitations. It uses population averages, so it does not know your individual efficiency. It cannot account for hills, wind, heat or terrain. It scales with total body weight, which slightly overstates the burn for people carrying more fat.

But it is derived from direct measurement rather than inference, it is published and checkable, and it is not systematically biased upward. For estimating energy cost, that combination beats a proprietary model with a commercial interest in a bigger number.

Use the watch for what it is good at

None of this means wearables are useless. They are genuinely good at several things:

• Heart rate accuracy for steady-state activity, which is what training zones need.
• Consistency over time. Even a biased estimate is useful for comparison if the bias is stable. This week versus last week is informative; the absolute number is not.
• Adherence. People who track move more. The behavioural effect is real and probably outweighs the measurement error.
• Resting heart rate and heart rate variability trends, which are among the better available signals for fatigue and illness.

What they are not good at is producing a number you can eat.

The practical rule

If you are using calories burned to inform eating, discount whatever your device says by something like a third, and do not eat all of it back regardless.

The estimate is imprecise, it includes resting calories you would have burned anyway, and appetite tends to rise after exercise by more than the deficit created. Eating back roughly half of a conservative estimate is a reasonable compromise that avoids both under-fuelling and quietly cancelling the session.

And if the number is the reason you go out, keep the number. A slightly optimistic figure that gets you running four times a week is doing more for you than an accurate one that does not.