Place · Level 3 · Measurement
The Number on the Body-Fat Scale
Last updated
Story path
Chapter 1
It does not measure fat
It does not measure fat
You step on, and it reads 22.4%. One decimal place. It looks like a measurement.
It is not a measurement. It is an inference.
A body-fat scale does exactly one thing: it passes a very small current up through your feet (and, on models with handles, your hands too) and measures how much resistance that current meets on the way. That resistance is called impedance.
It is not measuring fat. It is measuring how easily electricity travels.
And how easily electricity travels through you depends mostly on water. Muscle, blood, and organs are water-rich and conduct well; fat holds little water and conducts poorly. So the raw datum a body-fat scale actually obtains is a number tightly related to how much water is in you.
Between that number and the 22.4% on the display sit two more inference steps — and fat itself is never measured, not even once.
It is not a measurement. It is an inference.
A body-fat scale does exactly one thing: it passes a very small current up through your feet (and, on models with handles, your hands too) and measures how much resistance that current meets on the way. That resistance is called impedance.
It is not measuring fat. It is measuring how easily electricity travels.
And how easily electricity travels through you depends mostly on water. Muscle, blood, and organs are water-rich and conduct well; fat holds little water and conducts poorly. So the raw datum a body-fat scale actually obtains is a number tightly related to how much water is in you.
Between that number and the 22.4% on the display sit two more inference steps — and fat itself is never measured, not even once.
Mechanism · why fat conducts poorly
Current does not travel through the body on electrons; it travels on ions — sodium, potassium, chloride, charged particles dissolved in water. Wherever there is more water and more ions, there is an easier path.The inside of a fat cell is almost entirely occupied by one large droplet of triglyceride, with very little water or ions. To a current, that is a stretch of road with almost no charge carriers. Muscle, blood, and organs are the opposite extreme: high water content, stable electrolyte concentration — the body's natural wiring.
So high impedance ≈ less water-holding tissue is a real relationship with a genuine physical basis. It is not marketing, and the whole design rests on it.
The problem is not this step. The problem is that getting from *how much water-holding tissue there is* to *how much fat there is* takes two further steps, and each carries an assumption that may not hold for you. The next scene takes that chain apart.
Chapter 2
Fat is the leftover
Fat is the leftover
Between impedance and that percentage sits a three-step inference chain, and each step rests on an assumption.
Step 1 · impedance → total body water. Assumes the body is a conductor of uniform cross-section. It is not. Your limbs are long and thin, and nearly all the resistance lives there; your trunk is short and wide, carries most of your mass, and contributes almost none. A current running foot-to-foot is largely measuring your legs.
Step 2 · total body water → fat-free mass. Assumes the water fraction of fat-free mass is a fixed constant (about 73%). That fraction varies between people, and drifts with your state.
Step 3 · fat-free mass → fat. This step is subtraction: body weight minus fat-free mass, and whatever is left is called fat.
Step 3 is the crux. Fat is never measured at any point — it is the remainder. Every error from the first two steps lands, in full, on it.
Step 1 · impedance → total body water. Assumes the body is a conductor of uniform cross-section. It is not. Your limbs are long and thin, and nearly all the resistance lives there; your trunk is short and wide, carries most of your mass, and contributes almost none. A current running foot-to-foot is largely measuring your legs.
Step 2 · total body water → fat-free mass. Assumes the water fraction of fat-free mass is a fixed constant (about 73%). That fraction varies between people, and drifts with your state.
Step 3 · fat-free mass → fat. This step is subtraction: body weight minus fat-free mass, and whatever is left is called fat.
Step 3 is the crux. Fat is never measured at any point — it is the remainder. Every error from the first two steps lands, in full, on it.
Deeper · whose 73% is it
That constant in step 2 deserves its own look, because it is the quietest assumption in the chain.It is a population average. Fat-free mass is not one homogeneous substance; it is a mixture of muscle, bone, organs, blood, and connective tissue, and those components differ widely in water content. Change the proportions in a given person and the overall water fraction moves with them.
The standard methodological reference for BIA (Kyle 2004, ESPEN Working Group) puts it more firmly: BIA can determine fat-free mass and total body water only in people without significant fluid and electrolyte abnormalities, and only when the equation used is specific to that population, age, or pathology. Which means a consumer scale applying one equation to everybody is already, methodologically, outside those conditions.
Who drifts furthest from the average:
People with very high or very low muscle mass — off in both directionsOlder adults — body composition shifts with ageAnyone with oedema or fluid retention — more water, but that is not more fat-free massChildren and adolescents — the developing body's water fraction differs from an adult's by default
Note the counter-intuitive consequence: someone retaining fluid has more body water → a higher estimated fat-free mass → and after the subtraction, less fat. The scale will tell you your body fat fell because you were swollen.
This is why 'step 3 is subtraction' matters so much: it turns every upstream error into a confident-looking conclusion about fat, with nothing on the display to warn you.
Chapter 3
How far off is it
How far off is it
So how far off does this chain end up?
Answering that needs a better ruler first. Research uses the four-compartment model: measure body mass, volume, total body water, and bone mineral all at once, split the body into four parts. It is the accepted closest approach to ground truth.
A 2026 systematic review pooled 12 such validation studies and reported two numbers:
Mean bias: −3.5% to +4.4% (some devices run systematically high, others low)Limits of agreement: typically spanning 15 to 20 percentage points
The second number is the one that matters. It means that when the scale reads 25%, your true body fat sits somewhere around 17% to 33%.
That is not a body-fat percentage. That is an interval — and one wide enough to cross several category boundaries.
Answering that needs a better ruler first. Research uses the four-compartment model: measure body mass, volume, total body water, and bone mineral all at once, split the body into four parts. It is the accepted closest approach to ground truth.
A 2026 systematic review pooled 12 such validation studies and reported two numbers:
Mean bias: −3.5% to +4.4% (some devices run systematically high, others low)Limits of agreement: typically spanning 15 to 20 percentage points
The second number is the one that matters. It means that when the scale reads 25%, your true body fat sits somewhere around 17% to 33%.
That is not a body-fat percentage. That is an interval — and one wide enough to cross several category boundaries.
Reading it · bias and limits of agreement differ
These two numbers often get used interchangeably, but they answer completely different questions — and telling them apart is what tells you what the scale can do.Mean bias answers: does this device run high or low overall? It is the average error across a group of people. A device with +3% bias overestimates body fat by three percentage points on average — and that part is systematic, drifting the same direction each time for the same person.
Limits of agreement answer: for one specific person, how far off can a single reading be? It describes spread, not average.
Why the second matters more: you are not a group, you are one person standing on a scale wanting your own number. Mean bias can in principle be subtracted out afterwards; spread cannot.
Kyle 2004 supplies the other half of the same conclusion from the methodological side: clinical BIA has long lacked a standardised method and quality-control procedures. This is not one manufacturer cutting corners; the method as a class was never standardised.
One practical conclusion falls straight out of this: the review emphasises that device performance varies widely, and agreement figures from one analyser must not be generalised to another. So a reading from the gym machine, one from your home scale, and one from a health check are not comparable — they are not three marks on one ruler, they are three different rulers.
Chapter 4
Wrong, but not for that reason
Wrong, but not for that reason
Once people learn it is inaccurate, the next sentence is usually: so if I skip water, skip food, and skip training beforehand, does it become accurate?
Here the result is counter-intuitive.
A 2021 study set out to deliberately violate those preparation rules: 40 people, three consumer body-fat scales, measured while dehydrated, after exercise, just after drinking water, just after a meal, and with a full bladder — each against a standard control condition.
The result: no condition differed significantly from control. 86% of tests landed within 2 percentage points of the control value, and 97% within 5. The largest errors came from dehydration, and women were more likely than men to exceed a 2-point difference.
In other words — it really is inaccurate, but not because of that glass of water.
The problem lives in the inference chain from the previous scene, not in the two hours before you stepped on.
Here the result is counter-intuitive.
A 2021 study set out to deliberately violate those preparation rules: 40 people, three consumer body-fat scales, measured while dehydrated, after exercise, just after drinking water, just after a meal, and with a full bladder — each against a standard control condition.
The result: no condition differed significantly from control. 86% of tests landed within 2 percentage points of the control value, and 97% within 5. The largest errors came from dehydration, and women were more likely than men to exceed a 2-point difference.
In other words — it really is inaccurate, but not because of that glass of water.
The problem lives in the inference chain from the previous scene, not in the two hours before you stepped on.
Boundaries · what this study did and did not show
This scene is unusually easy to read backwards, so here are its boundaries.What it shows: violating the preparation rules over a short window barely moves the reading. That is, the device's reproducibility is better than gym folklore assumes.
What it does not show: that the reading is correct. Reproducibility and accuracy are different things — a ruler with mis-printed markings gives you the same wrong number for the same table every time. It is extremely stable, and it is always wrong.
The systematic review in the previous scene measures exactly the other half: against the four-compartment model, limits of agreement span 15 to 20 percentage points. Put the two studies together and the picture is stable but offset — which happens to be the most deceptive failure mode there is, because *stable* looks so much like *accurate*.
One detail worth keeping: the largest errors did come from dehydration. So the manufacturer's line about not measuring while dehydrated is not wrong — it has simply been inflated into an entire ritual, and the ritual does not touch the real problem.
This also sets up why the next scene recommends watching the trend: a stable offset partly cancels when you subtract two readings, whereas the absolute error it causes cannot be prepared away.
Chapter 5
So how should you use it
So how should you use it
So the conclusion is not to throw the scale away. It is to ask it only the questions it can answer.
What it cannot answer: what is my body fat right now? The uncertainty interval on that absolute value is too wide to place you in a category.
What it can answer: has the direction changed? Same device, same time of day, same state, read over several weeks — the systematic offset is roughly the same amount each time, so part of it cancels when you subtract. The trend is far more trustworthy than the level.
The better move is to demote it to one signal among several:
Waist circumference: a cloth tape, cheap, repeatable, and more directly related to visceral fat than a body-fat percentage isPhotographs taken under the same conditionsStrength and girths: your bench going up while your waist comes down carries far more information than 22.4% becoming 21.8%
What it cannot answer: what is my body fat right now? The uncertainty interval on that absolute value is too wide to place you in a category.
What it can answer: has the direction changed? Same device, same time of day, same state, read over several weeks — the systematic offset is roughly the same amount each time, so part of it cancels when you subtract. The trend is far more trustworthy than the level.
The better move is to demote it to one signal among several:
Waist circumference: a cloth tape, cheap, repeatable, and more directly related to visceral fat than a body-fat percentage isPhotographs taken under the same conditionsStrength and girths: your bench going up while your waist comes down carries far more information than 22.4% becoming 21.8%
Comparison · where the better methods sit
If you genuinely need the most accurate absolute value you can get, here is where each method sits — but the conclusion first: most people do not need one, because the decisions in front of you (what to eat, what to train, how long to keep going) do not change depending on whether that number is 22% or 26%.Four-compartment model: the reference standard in research, requiring several measurements at once. It essentially exists only in laboratories; it is not a service you can book.DEXA: common in clinics and research, and it also reports regional distribution and bone density. Far more accurate than a body-fat scale, though it carries its own error and needs equipment plus a low-dose X-ray.Volume-based methods (BodPod) and underwater weighing: these measure body density and infer composition from it — note that this too is an inference, just one resting on fewer assumptions.Skinfold calipers: cheap, but heavily dependent on the operator's technique; the same person measured by someone else gets a different answer.
One thing becomes visible reading down that list: every method is indirect. The only way to measure body fat directly is dissection, so there is no ground truth available here — only inferences with different error bars.
That is not nihilism; it is the point of this whole island. Knowing how a number was produced is what tells you how much conclusion it can carry. What a body-fat scale can carry is a trend, not a verdict.
References · 3
- Kyle, U. G., Bosaeus, I., De Lorenzo, A. D., Deurenberg, P., Elia, M., Gomez, J. M., Heitmann, B. L., Kent-Smith, L., Melchior, J. C., Pirlich, M., Scharfetter, H., Schols, A. M., & Pichard, C. (2004). Bioelectrical impedance analysis - part I: review of principles and methods. Clinical Nutrition, 23(5), 1226-1243. ESPEN Working Group methodological review. States that BIA use is widespread but suffers from a lack of standardised method and quality-control procedures, and that BIA allows determination of fat-free mass and total body water only in subjects WITHOUT significant fluid and electrolyte abnormalities, and only when using population-, age- or pathology-specific BIA equations together with established procedures. It also states that the use of segmental BIA, multifrequency BIA or bioelectrical spectroscopy in altered hydration states requires further research. 10.1016/j.clnu.2004.06.004
- Oliver, C. J., Del Vecchio, L., Minehan, M., Climstein, M., Rosic, N., Myers, S., & Tinsley, G. (2026). The validity of bioelectrical impedance analysis compared to a four-compartment model in healthy adults: A systematic review. Journal of Functional Morphology and Kinesiology, 11(1), 65. Twelve validation studies in healthy adults against the four-compartment criterion model. Mean bias for percentage body fat ranged from -3.5% to +4.4%, and limits of agreement typically spanned 15 to 20 percentage points. Conclusion: BIA estimates were overall not equivalent to the 4C model, performance is device-specific so agreement from one analyser must not be generalised to another, and alternative criterion methods such as MRI are recommended. 10.3390/jfmk11010065
- Randhawa, A. K., Jamnik, V., Fung, M. D. T., Fogel, A. S., & Kuk, J. L. (2021). No differences in the body fat after violating core bioelectrical impedance measurement assumptions. BMC Public Health, 21(1), 495. Within-subject study, n=40 across three consumer BIA devices, measuring percent fat mass under control, dehydration, exercise, water intake, meal intake and non-voided bladder conditions. No condition differed significantly from control, range -1.9 to 0.4 percent; 86 percent of tests fell within 2 percentage points of control and 97 percent within 5, despite deliberately violating the manufacturer preparation rules. Errors were largest with dehydration, and women were more likely than men to exceed a 2-point difference. 10.1186/s12889-021-10552-y