Nutrition 03: BMR, TDEE and the error around them

Short answer. A calculator's estimate of your daily energy expenditure can plausibly be off by a few hundred calories a day, which is the whole size of a normal deficit or surplus. Treat it as a starting hypothesis and correct it from your weight trend, not from a new calculation.

---

What it is

Basal metabolic rate (BMR) is what you burn at complete rest. Resting metabolic rate, measured under less strict conditions, is what most equations estimate, and the terms get used interchangeably. It is roughly half to seventy percent of a day's total.

Total daily energy expenditure (TDEE) is everything: resting metabolism, the cost of digesting food, deliberate exercise, and all the movement that is neither rest nor exercise.

Every calorie calculator does the same two steps. Estimate resting metabolism from an equation, usually Mifflin-St Jeor (Mifflin et al., 1990), then multiply by an activity factor. This piece is about how much error each step carries, because the error determines how you use the output.

The proposed mechanism

Bodyweight change follows from the difference between energy consumed and energy expended. If you can estimate expenditure, you can set intake to produce the change you want.

The logic is sound. Its usefulness depends on how precisely expenditure can be estimated for one person.

What has actually been tested

The best prediction equation is right within 10% for about seven people in ten. A systematic review of the four equations most used clinically found Mifflin-St Jeor the most reliable. It predicted resting metabolic rate within 10% of measured values in more individuals than the alternatives and with the narrowest error range, although individual errors remain (Frankenfield et al., 2005). A validation in women across body mass index categories put it within 10% in 71% of participants, concluding that clinically relevant errors can still be expected in about one third (Thom et al., 2020).

Turn that around. Roughly one person in three is off by more than 10% at the resting metabolism step alone. For someone whose true resting rate is 1,700 kcal, being 10% out is 170 kcal per day before any activity is added.

And the equations may share a systematic bias. A cross-validation in 3,001 adults attending nutrition clinics found all the common equations underestimated measured resting metabolic rate, Mifflin-St Jeor by a mean of 12.6%, with limits of agreement (the range most individual errors fell within) from +5.8 to -26.5% (Kfir et al., 2023). Other validations find it unbiased at group level. The disagreement is not noise. It is the sample. That cohort averaged a body mass index of 28.5 and was recruited clinically; the reviews reporting no bias pooled healthy non-obese adults. Which equation is unbiased depends on who you run it on.

The second step carries more error than the first. The activity multiplier is a self-reported category, and non-exercise activity thermogenesis (NEAT, the energy spent on all movement that is not exercise) is the most variable component of expenditure. It ranges from about 15% of the daily total in very sedentary people to 50% or more in highly active non-exercisers (Levine, 2003).

Expenditure also moves while you diet. A 10% reduction in body weight decreased total energy expenditure by 6 ± 3 kcal per kg of fat-free mass per day in non-obese participants and 8 ± 5 in those with obesity. A recent review reports those figures from a 1995 study (Theodorakis et al., 2024).

How much of that is adaptation beyond the weight loss itself is contested, and the individual spread swamps the average. Pooling 227 people who lost weight and then stabilized, about 55% showed reductions exceeding 40 kcal per day and roughly one in three finished with a resting rate above predicted rather than below (Magkos, 2020). In the larger of those two cohorts, adaptation averaged 92 kcal per day below predicted immediately after the deficit, with a standard deviation of 110, and more than halved to 38 within four weeks of eating at maintenance (Martins et al., 2020).

The group mean is not a person.

Our reading

The estimate is not wrong. It is imprecise, and the imprecision is large enough to change what you should do with it.

Stack the two steps. A resting metabolism estimate that misses by more than 10% for a third of people (Frankenfield et al., 2005) gets multiplied by an activity factor drawn from a self-reported category. That factor stands for a component ranging from about a seventh to half or more of the daily total. The plausible error on a personal TDEE estimate is a few hundred calories per day.

A few hundred calories per day is the entire size of a normal deficit or surplus. That means a calculator output can be, for some people, the difference between losing weight and gaining it.

No equipment removes this. Indirect calorimetry measures resting metabolism accurately and doubly labeled water measures total expenditure accurately. Neither tells you what your expenditure will be next month, because expenditure moves with bodyweight, training, sleep, how much you happen to move, and adaptation during a deficit. That is why correction is necessary, not optional. A measured number is a better starting hypothesis than a predicted one. It is still a hypothesis.

So the estimate should be treated as a starting hypothesis and the scale should be treated as the measurement. Set intake from the calculator, hold it steady for two to three weeks, and read your weight trend. The trend tells you your real energy balance at that intake, which no equation can.

Then correct from the trend rather than from a new calculation. If you are eating 2,600 kcal and weight is flat, your maintenance is approximately that, and the measured figure supersedes every prediction. Recalculating with a different formula is not an improvement; it is another estimate with the same error.

Daily bodyweight is noisy and the noise is solvable. It moves several hundred grams to a kilogram on water, food volume and glycogen, so a single weigh-in carries more noise than the weekly signal. Weigh daily and read the multi-day average.

And the target moves as you go. Expenditure falls as you lose weight, partly because a smaller body costs less and partly through adaptation. A deficit that worked in month one is a smaller deficit in month three. Not a broken metabolism. The same arithmetic with smaller inputs, which is why correction is periodic.

For a surplus the same logic applies with the sign reversed. The calculator cannot tell you how fast to gain either. The trend can.

Where the goals diverge

Losing fat. The estimate matters least and the trend matters most: a deficit that produces no weight loss is not a deficit. How fast to aim for is a separate question. (→ Nutrition 04: How fast to lose, how fast to gain)

Gaining muscle. The risk is a surplus larger than needed, and the feedback is slower because a small surplus produces a small trend that takes longer to distinguish from noise. Expect to need three to four weeks before the trend is readable. (→ Nutrition 01: Protein intake)

Maintaining. The easiest case and the one where a wide error band matters least, since small deviations self-correct across weeks. (→ Recovery category; Nutrition sub-hub)

Very lean or very heavy. Prediction accuracy is lower at both extremes, and the equations were validated primarily on Western populations. If you are unusually muscular, they treat you as your bodyweight rather than your composition. (→ Individual variation 01: Are you a non-responder?)

The case against

Weight is a noisy proxy for what you care about. It cannot distinguish fat, lean tissue, water and gut contents, and over a short window the non-fat components dominate. The trend measures energy balance, not body composition.

Intake is also estimated. Self-reported food intake tends to be under-reported, so this approach corrects the calculator's error using a second measurement with its own. It works because you are correcting a relationship rather than an absolute.

The validations disagree with each other. We have read the disagreement above as a population difference; it could equally be protocol.

And the adaptive thermogenesis literature is unsettled, with disagreement about magnitude, persistence and prevalence. An earlier version of this entry printed a range for it that we could not trace to a source, and we have removed it.

What is not known

How large the personal error is for any specific person without measuring it, which is the entire problem this article is about.

Whether adaptive thermogenesis persists. Studies have reported it at the end of weight loss and not at one and two years, and whether that reflects real recovery or measurement limits is not settled.

Whether trained lifters differ systematically from the populations these equations were built on.

Safety

Very low intakes carry real risks and are not covered here. If you are considering an aggressive deficit, have a history of disordered eating, or take medication affected by body weight, speak to a doctor or a registered dietitian.

---

OMNID starts from an equation estimate of your total daily energy expenditure, shown with its range when you set up your profile, then revises it from your own weigh-ins and food logs once there are enough of them. It also charts your bodyweight trend, so the correction described above can be read at a glance.