Fewer than half of people land within 10 percent of their real burn#
Here is how accurate TDEE calculators really are, measured rather than argued. Researchers gave 56 healthy adults doubly labelled water, tracked them for 14 days of ordinary life, and then ran ten published prediction models against what the isotopes said they had actually spent. Only 42.86 percent of individuals were predicted to within ±10 percent. Every one of the ten models had a root-mean-square error above 10 percent, which the authors described plainly as sizable errors at the individual level, and all of them underestimated total expenditure across the full sample. The best performer overall, one of the Plucker equations, still averaged 195 kcal away from measured with a 20.68 percent RMSE1.
Round that off and it is a coin flip with slightly bad odds. Ask a calculator what you burn and there is a bit under a 50 percent chance the answer is within 240 calories of the truth for someone on a 2,400-calorie day. That is a genuinely useful tool for orienting yourself and a poor one for setting a 300-calorie deficit, and the distinction between those two uses is most of what there is to say here. TDEE explained covers what the number is made of. This is the audit.
Two very different things get called "accuracy"#
Nearly every accuracy figure quoted online is about half the calculation.
The well-studied half is resting metabolic rate. Those equations have been checked against indirect calorimetry in thousands of people, and the field has a clear winner for general adults, with a published error range and known population blind spots — the head-to-head is in Mifflin-St Jeor versus Harris-Benedict, and the systematic review behind the consensus is Frankenfield et al., 2005. If someone tells you a calculator is "70-something percent accurate," that number is almost always about this stage.
The other half is the activity term, and validating it requires measuring free-living expenditure over days, which means isotopes. So whole-TDEE validations are rare, small and expensive, and their numbers are consistently worse than the resting-stage numbers. Prado-Nóvoa's 42.86 percent is one such study. Another approach is to skip prediction equations entirely and fit a new equation directly to doubly labelled water data: 93 healthy adults aged 18 to 81 were measured that way, and the resulting equation using only age, weight, height and sex — the exact four inputs a consumer calculator collects — carried a standard error of estimate of 1.80 MJ per day2. That converts to roughly 430 kcal per day (our conversion, at 239 kcal per MJ).
| What is being predicted | Validated against | Typical individual accuracy |
|---|---|---|
| Resting metabolic rate | Indirect calorimetry | Best equations put ~70-80% of general adults within ±10%3 |
| Total daily expenditure | Doubly labelled water, 56 adults, 14 days | 42.9% within ±10%; every model RMSE > 10%1 |
| Total daily expenditure, equation fitted directly to isotope data | Doubly labelled water, 93 adults | Standard error ≈ 1.80 MJ/day, about 430 kcal2 |
Read the table downward and the arithmetic of the error becomes visible. Predicting how much you burn lying still is a solved-enough problem. Predicting what you then do with your day is not, and the second problem is where the calculator's error mostly comes from.
The calculators fail worst on the people most likely to use one#
The error in Prado-Nóvoa's data was not spread evenly, and the pattern is unwelcome.
Accuracy fell as activity rose: the error of the estimations was generally higher for more active participants. Split the sample at a physical activity level of 1.89 and the less active half did noticeably better — around 50 percent of them predicted within ±10 percent, with the best model for that subgroup averaging a 44 kcal miss and a 12.54 percent RMSE, against 20.68 percent for the best model in the whole sample. The equations also performed better in males than in females, and the error grew with the size of the expenditure being predicted1.
That last point deserves care, because it is easy to misread. Heteroscedastic error means the absolute miss grows as expenditure grows; it does not mean active people are being systematically shortchanged by a fixed amount. What it does mean practically is that a marathon-training 90 kg man is being handed a number with wider real error bars than a sedentary 60 kg woman receives, printed to exactly the same four digits.
What a 10 percent miss actually costs#
Ten percent sounds like a tolerance. Priced in calories it is a whole intervention.
Take a 2,400-calorie estimate. Ten percent is 240 calories a day, which is within touching distance of an entire moderate deficit. If the estimate runs high by that much, the deficit you believe you set is roughly half the size you think — and over a month the accumulated mismatch between plan and reality is about 7,200 calories, which is our arithmetic on that daily figure rather than a measured result. Nothing on the scale tells you which way you erred for the first fortnight, because normal weight noise is bigger than the signal over that window.
This is why the practical failure of TDEE calculators is almost never that they are wildly wrong. It is that they are plausibly wrong, in the same direction, for weeks, and the feedback arrives too slowly to attribute.
Why the whole number is less reliable than either half suggests#
Two stages, two errors, and they multiply rather than average.
The resting stage carries a real individual error even when you pick the best-validated equation for your population. The activity stage carries a larger one, because nobody measured it — you assigned it to yourself from a list of adjectives, and one step along that list usually moves the total more than swapping one resting equation for another does, which is the argument in activity multipliers explained. Stack them and you get exactly what the isotope studies find. The full account of why this cannot be engineered away — including what the best equation ever built publishes as its own margin — is in why every TDEE calculator is an estimate.
Wearables are not the escape hatch, either. They replace an assumed activity level with a sensor-derived one, which sounds like an upgrade and behaves like a different error; the device evidence is in how accurate fitness tracker calories are.
The number that stops being an estimate#
There is one figure in this whole area that is not a prediction, and it is worth being precise about why.
Hold your intake steady and logged for a fortnight or three, read off the weekly weight average, and the maintenance figure that falls out is not a better estimate of the population — it is a different kind of quantity altogether. It is a measurement of your own body under your own habits, including the parts of your day no equation can ask about. It carries its own limitation, which is that it is expressed in the units of your logging rather than in true calories, so it stays valid only while you keep logging the same way. The method, and the traps in reading it too early, are in finding your maintenance calories.
Set the two side by side and the choice is not close. One approach is right within 10 percent for fewer than half of people and cannot tell you which half you are in. The other takes a fortnight and answers the actual question. Use the calculator to pick where to start, then stop asking it anything.
FAQ#
How often do TDEE calculators land within 10 percent of measured burn?#
Less than half the time, on the best available evidence. When ten published prediction models were tested against 14 days of doubly labelled water measurement in 56 adults, 42.86 percent of individuals were predicted to within ±10 percent, and every model had a root-mean-square error above 10 percent. Accuracy figures you see quoted in the 70 to 80 percent range are usually describing resting metabolic rate equations only, which is roughly half of the calculation.
Why are TDEE calculators least accurate for active people?#
Because the activity term is the part nobody measured, and the more of your day it accounts for, the more of your total it can get wrong. In the isotope validation, error was generally higher in more active participants, and splitting the sample at a physical activity level of 1.89 raised the hit rate in the less active half to about 50 percent while lowering the best model's error to a 12.54 percent RMSE. If you train hard, expect a wider real margin than the four digits on screen imply.
Is a fitness tracker's daily calorie number more accurate than a calculator's?#
It is a different estimate rather than a reliably better one. A calculator asks you to classify your own activity; a tracker infers it from motion and heart rate, which removes your self-assessment and introduces the device's own error instead. Neither has access to the individual variation that no equation can name. Whichever you use, the number worth trusting is still the one your own intake and weight trend produce over a couple of weeks.
Sources#
- Prado-Nóvoa O, Howard KR, Laskaridou E, et al. Validity of predictive equations for total energy expenditure against doubly labeled water. Sci Rep. 2024;14:15754.
- Vinken AG, Bathalon GP, Sawaya AL, Dallal GE, Tucker KL, Roberts SB. Equations for predicting the energy requirements of healthy adults aged 18-81 y. Am J Clin Nutr. 1999;69(5):920-926.
- Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc. 2005;105(5):775-789.
- Berman ESF, Swibas T, Kohrt WM, et al. Maximizing precision and accuracy of the doubly labeled water method via optimal sampling protocol, calculation choices, and incorporation of 17O measurements. Eur J Clin Nutr. 2020;74(3):454-464.
- Bajunaid R, Niu C, Hambly C, et al. Predictive equation derived from 6,497 doubly labelled water measurements enables the detection of erroneous self-reported energy intake. Nat Food. 2025;6(1):58-71.
