Field note 04 · Model sensitivity
Why Two Energy Estimates Can Both Look Plausible
The same age, height and weight can produce a broad range of maintenance estimates when the activity assumption changes. Here is the calculation in full.
Separate the equation from the activity choice
The Mifflin–St Jeor equation is a population-based starting estimate for resting energy expenditure. A maintenance estimate then applies an activity assumption. That second step is not a direct measurement of your individual day. Nutrinaut Tools deliberately shows a range instead of treating one multiplier as the truth.
For a fictional 30-year-old at 70 kg and 175 cm using the published male equation branch, the resting estimate is 10 × 70 + 6.25 × 175 − 5 × 30 + 5 = 1,648.75 kcal/day. The equation’s male/female constants are published model parameters, not a complete description of identity or physiology.
Change only the assumption
| Activity band | Multiplier used | Arithmetic | Starting interval |
|---|---|---|---|
| Light | 1.40–1.69 | 1,648.75 × band | 2,308–2,786 kcal/day |
| Moderate | 1.70–1.99 | 1,648.75 × band | 2,803–3,281 kcal/day |
| Vigorous | 2.00–2.40 | 1,648.75 × band | 3,298–3,957 kcal/day |
The resting equation has not changed between rows. The activity classification alone moves the estimate. That is why a watch, a calculator and a food log can disagree without any of them revealing an exact personal requirement. Do not choose the row that best matches a target weight; choose the band that best describes the period being estimated, then treat it as a starting hypothesis.
What a two-week log can and cannot settle
A consistent log of intake and comparable morning weights may help you decide whether a starting interval is useful. Compare averages of days 1–7 and 8–14 rather than one reading. Training changes, fluid shifts, illness, travel and incomplete logging can overwhelm a short signal. Two weeks are a practical review window, not validation of your true expenditure.
The tool never turns a short change into a supposedly precise calorie correction. It does not set weight-loss deadlines, minimum intakes or medical diets. People with a clinical diet, eating-disorder history, pregnancy or unusually high training volume need individual support.
Make the estimate accountable
- Record the equation branch, units and activity band used.
- Keep the first estimate and its range instead of repeatedly adjusting it to fit a desired outcome.
- If tracking is appropriate, review a consistent period and note major changes.
- When the evidence is noisy, say “uncertain” and collect better data rather than adding decimal places.
Recalculate with visible assumptions or read the full method guide.
From a number to an auditable estimate
Which input actually explains the difference?
Start with a controlled comparison: keep every input fixed and change one assumption. The table above isolates activity. It does not show measured calorie use on three different days. The FAO categories describe habitual lifestyles, rather than assigning a whole category from one workout. FAO/WHO/UNU physical-activity categories.
Small input changes versus an activity choice
In the published equation, adding 5 kg while keeping height, age and equation branch fixed increases the resting estimate by 50 kcal/day. Increasing age by 10 years with everything else fixed reduces it by 50 kcal/day. These are properties of the equation, not predictions of what happens to a particular person over time.
For the fictional 70 kg example, moving the multiplier from 1.40 to 1.70 changes the unrounded calculation by 1,648.75 × 0.30 = 494.625 kcal/day. A large disagreement can therefore come from a hidden activity assumption, even when the personal inputs match.
The article’s table rounds reference arithmetic to whole calories. The calculator rounds maintenance boundaries to the nearest 10 kcal: for example, 2,308.25–2,786.3875 becomes 2,310–2,790 kcal/day. That display choice avoids implying precision; it does not make the model more accurate. Run the same inputs and compare the activity bands.
Use a two-week log as a consistency check
Synthetic log example: all 14 intake entries are 2,400 kcal. The mean weight is 70.2 kg for days 1–7 and 70.1 kg for days 8–14. The recorded intake mean is 2,400 kcal/day; the difference between the two weight means is −0.1 kg. The tool expresses that difference relative to the mean of both weekly means: 100 × (70.1 − 70.2) ÷ ((70.2 + 70.1) ÷ 2), or about −0.14%.
This describes that log. It does not establish that true maintenance is exactly 2,400 kcal, that the change is body fat, or that the next fortnight will behave the same way. The calculator’s ±0.25% trend band is an explicitly chosen planning heuristic, not a clinical threshold or a confidence interval.
- Count the observations: the optional tool check needs 14 weight entries and 14 intake entries. An unrecorded day is not zero intake; an incomplete series cannot pass that check.
- Keep conditions visible: note a change of scale, weighing time, travel or an incomplete food day. Do not repair missing observations by copying a convenient value.
- Compare like with like: retain the two weekly means, intake coverage and the original activity assumption. A single lowest weight is not a substitute for the series.
- Choose the next question, not an automatic correction: if the log is incomplete, improve coverage; if the period is atypical, keep that limitation. Do not convert a short weight change into an exact calorie adjustment.
Save the formula, inputs and assumptions alongside the result. That makes a later comparison useful: you can distinguish a changed estimate from changed evidence. The calculator’s public reference cases check implementation arithmetic, not the accuracy of the model for an individual.
Sources and scope
Sources support the general method or limitation, not the accuracy of an individual Nutrinaut result. The worked numbers on this page are labelled examples, not user data or clinical advice.
Found a factual or arithmetic error? Send the page URL to [email protected]. See our corrections policy.