A steady body fat reading can still be inaccurate. BIA and DXA studies show why device, body size, and testing conditions matter when judging changes.
You step onto the same body composition scale you used yesterday, but the body fat percentage has changed. Last night’s dinner comes to mind, and you wonder whether to eat less today. Before treating the display as evidence that you gained fat, it helps to understand what the instrument actually measures and how researchers assess its performance.
Comparisons in adults have found that a body composition device can give very consistent repeat readings while reporting a lower body fat percentage than a reference assessment. Repeatability and accuracy answer different questions.[1][15] A decimal place on the screen tells you how a value is displayed; it does not establish that the value is correct to that decimal place.
The research here compares methods of measuring body composition. These are not trials testing how much weight people lose through food or exercise. Recording frequency and the use of apps belong to other articles in this series. Here, the question is how much confidence to place in a measurement before using it to change your routine.
What a body composition scale actually estimates
Home body composition scales use bioelectrical impedance analysis (BIA) to estimate body composition. It estimates body composition using electrical properties such as resistance. One study examined equations that predict fat-free mass using resistance-related information alongside height, weight, age, and sex. Fat-free mass means the amount of the body that is not fat.[4] The displayed fat estimate is therefore a calculated result, rather than fat removed from the body and weighed directly.
That distinction matters when reading the different fields on a screen. Body fat percentage expresses a proportion; fat mass expresses an amount in kilograms; fat-free mass expresses the amount remaining after fat is excluded. They describe related aspects of body composition, but they are not interchangeable measurements.[4][6] A result for one field cannot automatically establish the accuracy of the others. Nor does a fat-free mass estimate mean that every kilogram in that category is muscle.
A frequent reference method in these papers is dual-energy X-ray absorptiometry, abbreviated DXA. Researchers use it to assess body fat percentage and fat mass. It would still be too strong to treat DXA as the single unquestionable true value under every condition. A study of 166 adults compared BIA, DXA, and air displacement plethysmography, another method of assessing body composition. DXA also showed deviations whose direction differed at higher and lower body fat percentages.[1]
The participants had body mass index, or BMI, values of 19–38 kg/m² and underwent repeat testing within a 5-day period. BIA had a strong relationship with DXA, with a coefficient of determination of 0.92. Yet BIA underestimated body fat percentage by about 2 percentage points on average, and the reported limits of agreement were wide, from −4.25 to 8.37 percentage points.[1] Limits of agreement describe how far differences between methods can spread. They reveal something that an average difference alone leaves hidden: the methods can disagree to different degrees for different people.
All the methods in that comparison had high repeatability, with repeat measurements differing by less than 0.2 percentage points.[1] This is the combination that can make a scale difficult to interpret. Measuring again may produce a reassuringly similar reading, while a different assessment method produces a different value. Consistency does not, by itself, remove an offset between methods. The instrument can be steady and still disagree with the reference.
The reported underestimate of about 2 percentage points is a difference on the body fat percentage scale. It does not mean that fat mass was about 2% lower.[1] Keeping percentage points, percentages, and kilograms separate prevents an apparently small change in wording from changing the meaning of a result. The studies also test particular instruments in particular participants. Multi-frequency BIA uses multiple frequencies, and its accuracy is assessed for the particular device and participants tested.[6][15]
初日の汗は、予約した人だけが持ち帰れる。
Strong correlation does not guarantee your value is correct
A claim that a device has a high correlation with DXA can sound like a promise that the values are interchangeable. Correlation instead asks whether people with higher readings by one method also tend to have higher readings by the other. Both methods can arrange people in a similar order while one consistently reports lower values. Strong correlations and meaningful average differences appeared together in the research.[4][6]
A cross-sectional comparison brought together 945 Brazilian adults and older adults. For both fat mass and fat-free mass, the correlation between BIA and DXA was 0.97. Nevertheless, BIA estimated an average of 3.1 kg more fat-free mass and 2.9 kg less fat mass than DXA. The standard deviations of those differences were 2.4 kg and 2.3 kg, respectively.[4] Standard deviation describes how much the differences spread around their average. It helps distinguish a group summary from a dependable correction for an individual.
The relative differences were +7.2% for fat-free mass and −13.0% for fat mass, and both average differences were statistically significant.[4] Although the absolute differences use the same kilogram unit, the underlying amounts of fat and fat-free mass are different, so the relative differences are different too. The researchers examined results by sex, age, and nutritional status classified using BMI. They also used a concordance measure, which assesses agreement, rather than relying on correlation alone. That coefficient was at least 0.93 for fat mass and fat-free mass.[4]
These findings are a useful example of why a favorable coefficient should be read alongside the actual differences. The study reported strong relationships and moderate to high agreement, yet it also found systematic differences in kilograms and percentages.[4] For someone deciding whether a small change on a home display represents a change in their body, the latter findings are part of the evidence. They cannot be discarded because the correlation looks impressive.
The average fat mass difference of 2.9 kg was a summary of the study population. To adjust predictions, the investigators developed equations for their population and tested those equations in other participants within the study.[4] Calibration is a process of building and checking a prediction, rather than simply choosing a convenient number from a published average.
Of the 945 participants, all aged at least 18, 611 were women. The researchers used 70% of the sample, or 659 people, to develop prediction equations and the remaining 30%, or 286 people, for validation. The new prediction used resistance-related information, sex, height, body weight, and age. After calibration, average fat-free mass did not differ significantly from the DXA measurements in the validation group.[4] This supports the usefulness of adjustment for that population. It does not establish an individual correction for a different device or a different group of users.
Another comparison assessed 1,000 healthy adults under conditions closer to everyday life. Meals, hydration, recent exercise, and time of day were not controlled. Among 667 men and 333 women, BIA body fat percentage was lower than DXA by an average of 4.2 percentage points in men and 2.8 percentage points in women. The standard deviations were 3.0 and 2.6 percentage points, respectively.[6] Even the average difference was not the same across sexes.
Fat mass and fat-free mass had their own differences. For men, BIA estimated 3.7 kg less fat mass and 3.4 kg more fat-free mass on average, with standard deviations of 2.6 kg and 2.8 kg. For women, the corresponding estimates were 1.9 kg less fat mass and 2.0 kg more fat-free mass, with standard deviations of 1.8 kg and 2.2 kg.[6] The authors described the method as moderately accurate for body composition under uncontrolled conditions, except for visceral fat. That is a qualified assessment of usefulness, not a guarantee that every small personal change is accurately detected.
The direction of error changes with the participants
It is tempting to turn the preceding results into a simple rule that BIA always reads low. A comparison of 132 healthy adults shows why that rule fails. Against DXA, BIA body fat percentage was lower by an average of 1.56 percentage points in the normal-BMI group, but higher by an average of 3.4 percentage points in the group classified as obese. There was no significant difference between methods in the overweight group.[19] Within the same study, the direction of the average discrepancy reversed.
The researchers also found that error between the methods increased as body fat percentage increased. Waist circumference was a significant predictor of systematic error. These were associations, and the authors called for further examination of body size and fat distribution in evaluating BIA accuracy.[19] It identifies a reason that validation should consider who is being measured.
There were also favorable findings. In 109 Korean adults, high-frequency BIA and DXA showed a coefficient of determination of at least 0.89 for body fat percentage. The authors judged the measurements to have good agreement. There were 55 men; average age was 43.4 years in men and 44.9 years in women, and average BMI was 25.5 and 24.0, respectively.[7] These details identify the people behind the conclusion and help define the setting in which the equipment was assessed.
That comparison examined lean mass in the arms and legs and total fat-free mass as well as body fat percentage. The coefficient of determination for total fat-free mass was at least 0.95.[7] The positive conclusion belongs alongside the less favorable comparisons. Together they show that the research is testing combinations of methods, devices, body composition measures, and populations. A single verdict on every BIA instrument would flatten those distinctions.
Height was relevant in a study of 121 community-dwelling older Mexican women whose mean age was 73.7 years. Across the whole sample, average BIA and DXA results differed significantly. Among women taller than 145 cm, however, there was no significant average difference. Mean body fat percentage was 40.3% with DXA and 40.7% with BIA, and the concordance coefficient was 0.814.[8] Restricting the group changed the assessment of agreement.
The study also compared 5 equations that predict body fat percentage from body measurements. One purpose was to see whether excluding shorter women improved agreement. In the subgroup taller than 145 cm, an equation using waist-to-height ratio had no significant average difference from DXA, but its concordance coefficient was 0.693 and its standard error of estimate was 3.37. BIA’s standard error of estimate was 2.62, and the authors evaluated its results favorably in that subgroup.[8] Similar average values therefore did not make every prediction method equally convincing.
A practical reading of these comparisons starts with the participants before moving to the headline coefficient. Were they younger adults, older women, or people across different BMI groups? Which body composition field was tested? The Brazilian calibration study, the Korean comparison, and the Mexican study each answer a defined question.[4][7][8] Their results provide reasons to examine a device’s supporting evidence carefully, rather than reasons to assume that a general claim of accuracy settles your own reading.
Numbers: body fat measurement
In 166 adults, BIA read about 2 percentage points lower on average, while repeat measurements differed by less than 0.2 percentage points.[1]
In 945 adults and older adults, BIA estimated an average of 2.9 kg less fat mass than DXA.[4]
In 1,000 healthy adults, average body fat percentage differences were −4.2 percentage points in men and −2.8 in women.[6]
In 132 healthy adults, BIA estimated body fat percentage 3.4 percentage points higher in the group classified as obese.[19]
In repeat tests of 14 healthy adults, body water differences were 0.0–0.2 L within a day and 0.0–0.5 L between days.[15]
初日の汗は、予約した人だけが持ち帰れる。
1
2