Why your TDEE estimate can be off by 20%

You enter your details, the page returns a number, and it looks authoritative: 2,684 kcal. Three decimal places of confidence wrapped around an estimate that, for any single person, might be wrong by 400 calories in either direction.

That is not a flaw in the calculator. It is how population equations work. Understanding the size and direction of the error is more useful than pretending it does not exist.

What the equations actually predict

Mifflin-St Jeor and Katch-McArdle were both fitted to groups of people, not to you. They find the line that minimises error across hundreds of subjects. The line is excellent on average and mediocre individually.

For resting metabolic rate, measured against indirect calorimetry:

Method	Typical individual error	Notes
Mifflin-St Jeor	±10%	Best general-purpose option; tends to overestimate in obesity
Katch-McArdle	±8% when body fat is measured well	Depends entirely on the accuracy of the body-fat input
Harris-Benedict (revised)	±14%	Older cohort, still shipped by many apps

A ±10% band around 1,800 kcal of resting metabolism is 1,620 to 1,980 — a 360 kcal spread before you have even picked an activity level.

The activity factor is the bigger problem

Resting metabolism is the easy part. What you burn on top of it depends on how much you move, and most people describe their week badly.

Activity multipliers assume a stable pattern. Real weeks are lumpy: a desk job with three gym sessions, a weekend of hiking, a fortnight of travel where you walk 15,000 steps a day and then a fortnight where you barely leave the flat.

Non-exercise activity thermogenesis — fidgeting, standing, walking, chores — varies by up to 2,000 kcal per day between two people of the same size. It is also the first thing your body adjusts when you eat less, which is why a deficit that worked in week two stalls in week eight.

Treat the activity factor as a starting assumption, not a measurement. It is the field most likely to be wrong and the one worth revisiting first.

Three smaller sources of drift

Body-fat input. Katch-McArdle is only as good as the number feeding it. Home bioimpedance scales can swing 5 percentage points across a single day depending on hydration. Unless the figure came from DEXA or careful caliper work, treat it as a guess and prefer Mifflin-St Jeor.

Adaptive thermogenesis. When intake drops, output drops with it: less spontaneous movement, lower body temperature, cheaper digestion. The effect is real and typically a few percent, not the 40% some diet books claim.

Labelling and logging error. Studies of self-reported intake consistently find under-reporting of 20%–30%. If your records say 2,000 kcal and your weight is not moving, the records are the first suspect, not your metabolism.

How to narrow the error to something useful

You cannot fix a population equation with more population data. You fix it with your own.

1. Pick a starting number from the calculator and hold intake steady for 14 days. Log what you actually eat, including the oil in the pan.
2. Weigh yourself daily at the same time and average each week. Day-to-day noise from salt, glycogen and gut contents is far larger than fat change.
3. Compare the trend. If the weekly average is flat on a stated deficit, your true maintenance is lower than the estimate. If you are losing faster than planned, it is higher.
4. Adjust once, by 100–200 kcal, then repeat for another two weeks.

Two cycles of that gets most people within 5% of their real maintenance — closer than any equation will ever get from four inputs.

What the number is good for

An estimate with a ±15% band is still genuinely useful. It tells you whether your maintenance is nearer 2,200 or 3,200, which is the difference between a sane deficit and an accidental crash diet. It also gives you a defensible starting point instead of a guess built from a magazine article.

Use it as a hypothesis. The scale, over two weeks, is the experiment.