Assessing and tracking body composition seems to be a mild obsession in the fitness community. On one hand, this preoccupation is at least somewhat understandable – if you’re aiming to lose weight, you’re probably more interested in losing fat than muscle mass, and if you’re aiming to gain weight, you’re probably more interested in gaining muscle mass than fat. On the other hand, I’m concerned that we’ve gotten the cart before the horse.
If you’re going to assess an outcome (any outcome) for the purpose of evaluating progress toward a goal, generating training or nutrition recommendations, or measuring the effects of a particular training program or dietary strategy, it’s worth asking how well you can assess the outcome of interest. How accurately can you measure the outcome? How long does it take to reliably detect changes of a reasonable magnitude? How straightforwardly can you interpret the results of your measurements? Are there alternative outcome measures that are more useful for the goal(s) you’re pursuing?
In this article, I’ll discuss why individual-level body composition assessments are far less useful than most people realize and, by extension, why body composition data is used sparingly in MacroFactor.
First, though, we need to cover some basic background information about body composition assessment. And, when I say basic, I do mean basic.
The first thing to know is that there’s no way of measuring body composition in living, breathing humans. There is exactly one way to measure body composition: carefully dissecting a cadaver, and weighing each component of the cadaver after completing the dissection. That’s certainly a morbid fact to ponder, but it’s the bedrock of body composition assessment.
If you’ve never been dissected – since you’re reading this article, I assume you haven’t – you’ve never had your body composition measured. You’ve only had your body composition estimated. Understanding this fact is important for two reasons. First, it invites a simple question: “what IS actually being measured in order to estimate body composition?” Second, it invites us to consider the validity and reliability of body composition estimates.
What is actually being measured?
There are five types of measurements commonly used to estimate body composition. I’ll briefly explain each measurement, and the methods of body composition analysis that use each measurement.
Body density
Two popular methods of body composition analysis – underwater weighing and the Bod Pod (air displacement plethysmography) – estimate body density in order to estimate body composition. Density = Mass ÷ Volume. With the BodPod, body volume is estimated based on air pressure changes within an enclosed chamber. With underwater weighing, density is estimated based on the Archimedes principle.
These methods of body composition analysis rely on one simple fact: different body tissues have different densities. For example, muscle is about 15% more dense than fat – it weighs about 15% more per unit of volume. So, if two people weigh the same amount, but the total volume of Person A’s body is smaller than the total volume of Person B’s body, Person A would have a greater density than Person B. Therefore, all else being equal, Person A should have more muscle mass and less body fat than person B.
Subcutaneous fat thickness
Your body stores fat in quite a few different places, but the two largest compartments of fat storage are the subcutaneous compartment – fat stored between your skin and muscles – and the visceral compartment – fat stored around the organs in the abdominal cavity.
In general, higher amounts of subcutaneous fat are correlated with higher amounts of total body fatness. Estimating body composition using calipers takes advantage of this relationship. You can’t easily measure the amount of fat within your abdominal cavity, but subcutaneous fat is far more accessible. A trained assessor uses calipers to measure the thickness of your subcutaneous fat, in order to estimate the total amount of fat on and in your body. Once these subcutaneous thicknesses are obtained, they’re entered into formulas that lean on population-level observations about the typical relationship between subcutaneous fat storage and total body fat percentage.
A similar but less accessible method of body composition assessment uses ultrasound (in place of calipers) to accomplish the same purpose.
Impedance of electrical currents
Your body tissues conduct or impede electrical currents to different degrees. In general, tissues with more water are better conductors of an electrical current.
One popular method of estimating body composition takes advantage of this fact: bioelectrical impedance analysis (BIA). Bathroom scales and small handheld devices that estimate your body fat percentage are using BIA.
These devices pass a weak electrical current between electrodes, and measure how long it takes for the electrical current to pass through your body’s tissues. Since muscle has a greater water content than fat tissue (~75% versus ~10%), the electrical current will travel a bit faster through leaner people. In essence, BIA is estimating the hydration level of the tissues an electrical current is passing through, in order to estimate the amount of each type of tissue (fat versus lean) the current is passing through.
Bioelectrical impedance spectroscopy (BIS) is a similar (but strictly superior) method of body composition assessment that relies on this same relationship. BIS uses a wide array of electrical currents across a broad spectrum of frequencies to differentiate between intracellular and extracellular water – this differentiation allows for more nuanced body composition estimates compared to relying on total body water alone.
Body geometry
There are a few different methods for estimating body composition that rely on measurements of body shape – typically height and various circumference measurements. The most popular formulae in this category are the US Navy formula and waist-to-height ratios (though there are certainly others). 3D body scanners are a high-tech variation on the same theme.
In essence, methods of estimating body composition based on body geometry rely on the fact that most humans have characteristic fat storage patterns. People store most of their body fat around their waist and hips. More often than not, if two people are the same height, the person with the larger waist will be carrying more body fat.
Gray pixels
Savvy readers have probably noticed that I haven’t mentioned DEXA (dual-energy x-ray absorptiometry) scanners yet. I think people tend to overestimate the accuracy and reliability of DEXA scans for estimating body composition, but DEXA scans are probably the best method of estimating body composition that’s somewhat accessible.
The way DEXA scanners estimate body composition is pretty interesting, though – they use a combination of very high-tech and very simplistic techniques.
A DEXA scanner passes x-rays (with a relatively low radiation dose) through your body, and measures how much of the energy each part of your body absorbs – bone absorbs the most energy, muscle absorbs less energy, and fat absorbs the least energy. This creates an image of your body where bones are nearly white, muscle tissue is a bit darker, and fat tissue is darker yet. From there, a computer program essentially just counts the number of pixels of different colors. If you have more dark (fat) pixels than semi-dark (muscle) pixels, the scanner will say your body fat percentage is higher (and vice versa).
Due to the relatively expensive and high-tech nature of these scanners, people often assume that their body composition estimates are nearly perfect. However, there are some noteworthy limitations that are rarely acknowledged outside of academic journals. For example, DEXA scanners are unable to directly measure tissue thickness, so they are forced to derive composition estimates of a 3-dimensional body from a 2-dimensional scan. Further, the devices are only able to estimate soft tissue composition (i.e., the relative content of lean tissue and fat tissue) in pixels that contain no bone mass. For most scans, about 40% of the total pixels will contain bone, so nearly half of the scan is essentially uninterpretable (without assumption-driven calculations) for the purpose of body fat estimation.
Accuracy of Body Composition Assessments
Since all popular methods of assessing body composition aren’t actually measuring body composition – they’re measuring things that are proxies for body composition – it’s worth asking a simple question: How good are the estimates produced by these technologies?
The answer to that question largely depends on the reason you’re interested in assessing body composition in the first place.
If you’re a researcher, and you’re interested in characterizing the body composition of a group of subjects, or you’re interested in assessing how a particular training or nutrition intervention affects average changes in body composition over time, all of the methods above work pretty well (except for BIA, arguably).
There aren’t a ton of human cadavers lying around (though animal cadavers are sometimes used), so when a new technology for assessing body composition hits the market, it’s usually compared against body composition estimates derived from a 4-compartment model (which is considered the practical gold standard for estimating body composition) 1A full 4-compartment body composition assessment combines various techniques of assessing body composition in order to estimate fat mass, total mineral mass (mostly from bone), total protein mass, and total body water. You generally won’t find all of the necessary equipment to do a 4-compartment body composition assessment outside of a research laboratory. As such, while full 4-compartment body composition assessments are quite accurate, they’re rarely accessible for individuals interested in monitoring their body composition. Most methods of body composition analysis produce group-level estimates of lean mass and body fat percentage that differ from the gold standard analytic method by ~1-4%, on average. Furthermore, estimated changes in body composition tend to be pretty similar when comparing the aforementioned technologies to estimates derived from a 4-compartment model – if a 4-compartment model says a group of subjects decreased their body fat percentage by an average of 5%, the technologies above will generally estimate that body fat percentage decreased by ~3-7%.
So, in general, if you’re interested in assessing average body composition for a group of people, or tracking average changes over time, most popular methods of body composition analysis work pretty well. Some certainly work better than others – if DEXA and BIA are both available to you, I can’t think of a good reason to choose BIA over DEXA – but they all typically do a good-to-excellent job of estimating the body composition of groups of people.
However, I suspect that most people reading this article aren’t particularly interested in assessing the average body composition of a group of 50 subjects. I suspect you’re interested in estimating your body composition, and tracking your body composition over time.
If that’s the case, I’ve got some bad news for you: There’s really not a particularly good method for estimating body composition (assuming you’re interested in having accurate, precise data).
For this point, I’ll refer you to an excellent series of articles by James Krieger, if you’d like to do a deep dive into this topic (one, two, three, four, five, six, seven). However, the basic point is pretty straightforward: You shouldn’t assume that group-level errors generalize to individuals.
A method of assessing body composition could underestimate body fat percentage by 1% on average, while still producing massive underestimates and overestimates for individuals. To illustrate, an underestimate of 11% and an overestimate of 9% equate to an average underestimate of just 1%.
The figures below demonstrate this point with illustrative synthetic data. The scatterplot shows the relationship between actual body fat percentage on the x-axis, and estimated body fat percentage on the y-axis. Overall, the estimated body fat percentage data looks quite good – estimated values are highly correlated with actual values (r = 0.89), and the summary statistics bear this out. The average error is only 0.5% (25.3% vs. 25.8%), and the average absolute error is just 2.7% (which is similar to DEXA). However, the histogram tells a slightly different story. Less than half of the individual errors were smaller than 3%, around a third of the errors were between 3% and 6%, and about 1/6th of the errors were larger than 6%, topping out at 10.1%.
For both estimating body fat percentage at a single point in time, and for estimating changes in body fat percentage over time, the methods of body composition analysis listed above (even DEXA) can produce individual errors of up to ~4-5% at best, and errors exceeding 10% at worst.
In other words, if your body fat percentage is estimated to be 20%, that actually means your body fat percentage is somewhere between 15-25% if you feel particularly optimistic about body composition estimation; more realistically, it means your body fat percentage is somewhere between 10-30%. To say that such a wide range is minimally informative would be an understatement.
Similarly, if you track your body composition over time, and a particular method for assessing body composition suggests that your body fat percentage has decreased by 5%…that optimistically means it has decreased by 0-10%, and more pessimistically means your body fat percentage has either increased by up to 5%, or decreased by up to 15%. I suspect that virtually everyone would be able to reach the same conclusion (probably with an even higher degree of precision) by just looking in the mirror or assessing how their clothes fit.
In short, assessing body composition can be very valuable in a research context. If we want to see how a particular intervention affects changes in fat or lean mass for a group of subjects, we have plenty of tools that are well-suited to the job. However, I’d argue that assessing body composition offers virtually no utility whatsoever for individuals. On the group level, we can estimate body composition at a single point in time quite well, and track changes in body composition over time quite accurately. On the individual level, estimates of body composition at a single point in time are too imprecise to be particularly informative, and estimated changes over time are too imprecise to be particularly useful.
As one final illustration, let’s assume that you recently completed a diet, and your body mass decreased from 100kg to 80kg. You got DEXA scans before and after your diet, and DEXA said you went from 30% body fat to 20% body fat.
If you take those numbers at face value, you’d assume that you lost 14kg of fat mass and 6kg of lean mass. Depending on your perspective, you may consider that to be a pretty successful diet (your body fat percentage decreased by 10%), or you may consider it to be a failure, especially if you engage in resistance training (6kg is a lot of lean mass to lose). However, if we optimistically assume that DEXA produces individual errors of up to “just” 5% for tracking changes in body composition over time, this diet could tell two very different stories.
For instance, if your body fat percentage decreased by 15% instead of 10%, now the diet looks like an unqualified success. You lost 18kg of fat and just 2kg of lean mass – I think virtually anyone would be stoked about that outcome. However, if your body fat percentage decreased by just 5% instead of 10%, the outcome of the diet would appear quite dire. That would mean that half of the weight you lost was lean mass – 10kg of fat mass, and 10kg of lean mass.
To be clear, that’s how you should interpret the results of those DEXA scans. If DEXA says you lost 14kg of fat and 6kg of lean mass, that actually means you lost somewhere between 10-18kg of fat, and somewhere between 2-10kg of lean mass. In other words, DEXA is telling you that the outcome of your diet was somewhere between “unambiguously good” and “catastrophically bad.”
If you think you’d be able to figure that out for yourself by simply looking in the mirror from time to time, then congratulations – you probably don’t need to worry about assessing your body composition anymore. 2The prior sentence is probably painting with slightly too broad of a brush – when most people hear “body composition,” they’re generally thinking about estimates of fat mass, lean mass, and body fat percentage. I’m writing with that particular connotation of the term in mind. However, assessments of bone mineral density and bone mineral content also fall under the umbrella of “body composition,” and DEXA scans can actually estimate BMD and BMC quite accurately – if you have osteoporosis or osteopenia, or if you’re at risk of osteoporosis or osteopenia, BMD and BMC assessments may be quite useful and informative. That’s a conversation to have with your doctor, and it goes beyond the scope of this article. Also, keep in mind that this is a pretty optimistic illustration. Rather than using DEXA, most people track their body composition over time using BIA scales, which produce even larger errors (both for assessing body composition at a single point in time, and for tracking changes in body composition over time).
How MacroFactor Uses Body Composition Data
As the prior section should make clear, we believe the evidence suggests that body composition data is generally too imprecise and inaccurate to be useful for individuals. However, if you’re a MacroFactor user, I’m sure you noticed that we ask for a rough estimate of your body fat percentage during the onboarding process, and I’m sure you’ve noticed that you can track your body fat percentage in-app. So, how does MacroFactor use body composition data?
We use your profile-level body fat percentage estimate for two things. Your profile-level body fat percentage estimate is what you entered during onboarding, and you can change it at any time by going to “More” → “Profile” → “Body Fat %”.
First, it’s used to generate an initial total daily energy expenditure estimate. We use the Cunningham formula to estimate your BMR, and lean body mass is the primary input in the Cunningham formula (estimating body fat percentage and estimating lean body mass are two sides of the same coin).
Second, it’s used to generate protein recommendations. Protein needs generally tend to scale with lean body mass, so we need a rough estimate of your lean body mass to give protein recommendations for users on Coached macro programs.
If you track your body fat percentage day-to-day along with your weight, we don’t use that data for anything. We simply allow users to track it for their own purposes.
This section will explain why we do use body composition data for the two purposes listed above, and the next section will explain why we don’t use day-to-day body composition estimates for the purpose of making program adjustments.
Initial Total Daily Energy Expenditure Estimate
As discussed in a previous article, total daily energy expenditure estimates from static formulae are always pretty rough estimates. That’s one of the primary reasons MacroFactor exists in the first place – a continuous stream of weight and nutrition data allows you to estimate your total daily energy expenditure (for the purpose of setting calorie intake targets to gain, lose, or maintain weight) far better than any static formula could. With that in mind, we opted to use the Cunningham formula for two primary reasons.
First, we know that lean body mass is the primary predictor of energy expenditure, so we figure it makes sense to just jump straight to the source. Other formulas (like the Harris-Benedict equation, for example) just use a mix of variables that happen to be predictive of energy expenditure because they tend to correlate with lean body mass (sex, age, weight, height, etc.).
Second, a pretty good chunk of MacroFactor users engage in resistance training, and the Cunningham equation does a particularly good job of estimating energy expenditure for lifters. So, the Cunningham equation should produce energy expenditure estimates that are about as good as any other formula for the non-lifters who use MacroFactor, and better energy expenditure estimates than other formulas for the lifters who use MacroFactor.
Ultimately, we view the decision to use body composition data for this purpose – providing an initial energy expenditure estimate – to be a relatively minor decision. We are well aware that some users will overestimate their body fat percentage, and other users will underestimate their body fat percentage during onboarding, but a) errors in initial energy expenditure estimates are unavoidable, b) we allow users to override the app’s initial estimate if they think it’s too high or too low, and c) after about three weeks of consistent weight and nutrition tracking, the impact of this initial energy expenditure estimate will be fully washed out. Thus, if error is introduced due to poor body composition estimation, the effect will be relatively minor and short-lived.
Furthermore, data suggests that when people misestimate their body fat percentage, they’re most likely to think they’re leaner than they actually are. And ultimately, if MacroFactor misestimates energy expenditure for a user during onboarding, I’m far more comfortable with overestimates than large underestimates.
For example, imagine someone has a total daily energy expenditure of 3000 calories, and they’re aiming to lose two pounds per week (which would require an average daily energy deficit of approximately 1000 calories). If we initially estimated that their daily expenditure was 3500 calories, and recommended that they consume 2500 calories per day, their initial rate of weight loss would be slower than intended (about 1 pound per week instead of 2 pounds per week), but they certainly wouldn’t feel like they were being thrown into the deep end with an unsustainable diet, and MacroFactor’s algorithms would be able to gradually reduce their calorie recommendations until they were losing two pounds per week.
Now, let’s run this scenario back, but let’s assume that we initially estimated that their daily energy expenditure was 2500 calories per day – a 500-calorie underestimate instead of a 500-calorie overestimate. In this scenario, their initial calorie recommendation would be just 1500 calories per day (consistent with losing three pounds per week, given their true energy expenditure of 3000 calories per day). That’s a really steep deficit. And sure, MacroFactor’s algorithms would be able to gradually increase calorie recommendations until their rate of weight loss decreased to two pounds per week, but there would be a much greater risk of the user (understandably) viewing the diet as unsustainable and losing motivation within the first couple of weeks.
Ultimately, we always strive to provide the most accurate recommendations possible, so we certainly don’t aim to overestimate users’ initial energy expenditures. However, in most circumstances, initial overestimates are preferable to initial underestimates. So, if there’s a tendency for users to underestimate their body fat percentage (or overestimate their activity levels) during onboarding, and subsequently eat a few extra calories during their first few weeks using MacroFactor, I’m very comfortable with that.
Protein recommendations
Body composition estimates are used to inform protein recommendations to ensure that MacroFactor’s nutrition recommendations will be tolerable and appropriate for users across the full spectrum of human body composition.
In the evidence-based fitness community, protein recommendations are generally provided in terms of grams of protein per kilogram of total body mass. The go-to reference is a meta-analysis by Morton and colleagues, finding that about 1.6-2.2 grams of protein per kilo of total body mass (0.73-1g/lb) was likely sufficient to maximize lean body mass.
Realistically, this 1.6-2.2g/kg range is an excellent heuristic for a lot of people, but it’s not perfectly precise or universally generalizable. For example, several studies with protein intakes below 1.6g/kg have reported similar gains in lean mass as studies with much higher protein intakes. More importantly, it can result in some unrealistically high protein recommendations for people who are quite heavy. If you weigh 150kg and have a reasonably high body fat percentage, there’s probably no reason you need to consume 330 grams of protein per day.
Instead, MacroFactor’s protein recommendations are scaled to lean body mass. Subjects in the types of studies included in the Morton meta-analysis are generally in the neighborhood of 20% body fat, allowing us to scale the protein recommendations from that meta-analysis (provided in terms of grams of protein per kilogram of total body mass) to protein recommendations per kilogram of lean body mass easily enough. So, that range of 1.6-2.2 grams of protein per kilogram of total body mass corresponds to a range of about 2-2.75 grams of protein per kilogram of lean body mass. This calculated range is reinforced by additional mechanistic research indicating that the estimated average protein requirement for male bodybuilders is around 2 grams of protein per kilogram of lean body mass.
In our view, scaling protein recommendations to lean body mass comes with a clear benefit and minimal drawbacks, even if very rough estimates of body fat percentage are used to inform our estimate of lean body mass.
The key benefit, again, is that scaling protein recommendations to lean body mass ensures that protein doesn’t “crowd out” too much dietary fat and carbohydrate for users with higher body fat percentages. In a vacuum, there’s not necessarily anything wrong with eating more protein than you strictly need for the amount of lean mass you have. However, diets tend to be more flexible, tolerable, and hedonically pleasing when protein-rich foods don’t crowd out most other food groups.
The only drawback is that users could be recommended higher or lower protein targets than would be theoretically optimal if they misestimated their body fat percentage. However, as previously mentioned, this is a minor drawback.
To illustrate, let’s assume that my body fat percentage is 25%. I currently weigh about 225 pounds (102kg). If I set up a coached program in MacroFactor and selected a moderate protein level, my protein intake recommendation would be 192g if I said I was 18-23% body fat, 177g if I said I was 24-30% body fat, and 165g if I said I was 30-34% body fat.
Of these three potential recommendations, 177g/day is “right,” but eating 12 fewer grams of protein per day isn’t going to materially impact my results, and eating 15 more grams of protein per day isn’t going to crowd out that much dietary fat and carbohydrate from the rest of my diet. In other words, your protein recommendations are always going to be solid, unless you misestimate your body composition to a pretty astronomical degree. Furthermore, the user is always in control – if they want a higher or lower protein target, they can always select one (or just set up a collaborative program, where protein targets can be as high or low as they want).
In short, your estimated body composition doesn’t impact much about your experience with MacroFactor. In the two instances where body composition data is used, we believe the clear upsides outweigh the potential downsides, even when acknowledging the potential for erroneous body fat percentage estimates.
Why MacroFactor Doesn’t Use Day-To-Day Body Fat Estimates To Inform Recommendations
If you’ve made it to this point in the article, the primary reason MacroFactor doesn’t use daily body fat estimates to inform nutrition recommendations should be fairly obvious – the quality of the data would be too low to be particularly useful.
Biological data is inherently noisy, because biological system (including the human body) are messy. When evaluating whether a particular data source can be used to provide actionable insights, you need to consider whether you can cut through the noise to pick up the signal in an underlying dataset in a timely manner.
In the case of weight data, there’s plenty of noise, but there’s also plenty of signal. If your weight is up or down by 2-3 pounds on a particular day, that may not mean much. However, if your weight is consistently trending up or down over a period of weeks, that provides you with a very strong indication that your body weight is truly increasing or decreasing. Furthermore, there’s enough signal in weight data to reliably pick up on trends in a timely manner – over a period of days-to-weeks.
You can’t say the same about body composition data because body composition estimates have far greater day-to-day variability (relatively speaking) than weight measurements. So, reliably picking up on trends would take weeks-to-months, instead of days-to-weeks. Even if you could pick up on trends, you wouldn’t be able to use those trends to accurately inform nutrition recommendations in a timely manner.
Furthermore, weight data is particularly useful because, with few rare exceptions (for example, lymphedema), weight trends reflect changes in stored chemical energy over the short-to-medium term. The link between stored chemical energy and body mass is the reason that the calories in/calories out model of weight regulation works. Sure, water weight fluctuations can impact the number on the scale day-to-day, but if you’re 20 pounds lighter today than you were three months ago, that almost necessarily means you’ve truly lost weight, and there’s less chemical energy stored in your body than there was three months ago.
The same can’t necessarily be said about day-to-day body composition data.
For tracking changes in body composition over the short term, most people rely on either BIA or a formula that estimates body composition from circumference measurements (most people aren’t getting daily DEXA scans). Beyond the general accuracy and reliability issues associated with these methods of assessing body composition, there are clear circumstances where these methods of assessing body composition will provide completely erroneous (not just noisy) “insights” over a pretty long period of time.
Further Problems with BIA
Starting with BIA, refer to my previous description of how this technology works: “These devices pass a weak electrical current between electrodes, and measure how long it takes for the electrical current to pass through your body’s tissues. Since muscle has a greater water content than fat tissue (~75% versus ~10%), the electrical current will travel a bit faster through leaner people. In essence, BIA is estimating the hydration level of the tissues an electrical current is passing through, in order to estimate the amount of each type of tissue (fat versus lean) the current is passing through.”
Particularly astute readers may have had some questions when encountering that description: “Sure, fat may impede the electrical current to a greater extent than muscle, but don’t all tissues impede the electrical current to some degree? So, would a BIA device think you had more body fat if the electrical current simply had to flow through more total tissue (including more muscle tissue)?”
The answer to both of these questions is a firm “yes.” In other words, if your body composition doesn’t change at all, but you gain a significant amount of muscle, that additional muscle will further impede the electrical current, so a BIA device will think you gained at least some fat.
If you have a BIA scale that lets you specify whether you’re an athlete or not, you can test this out for yourself. As of the time of writing, my BIA scale says I’m 19.8% body fat if I say I’m an athlete, and 30.3% body fat if I say I’m not an athlete. That’s a pretty huge divergence.
The reason for this divergence is that athletes tend to have considerably more lower-body muscle mass than non-athletes. If you tell your BIA scale that you’re an athlete, it will anticipate greater impedance of the electrical current, because it will assume that the current is flowing through more total muscle tissue.
In other words, if a non-athlete at 30.3% body fat had an astoundingly successful period of body recomposition, they could lose a ton of fat and gain a ton of muscle at the same body weight, get down to 19.8% body fat, and their BIA scale wouldn’t be able to tell the difference. In my case, that would mean that I could lose nearly 24 pounds of fat and gain nearly 24 pounds of muscle without my “smart scale” being any the wiser.
This same arithmetic works in reverse – an athlete could stop working out, lose a ton of muscle, gain a ton of fat, and still be told by their BIA scale that their body fat percentage was unchanged.
The only way the scale would know to interpret the impedance values differently would be if the user changed the athlete/non-athlete toggle at some point in the process – at which point their day-to-day estimate of body fat percentage would increase or decrease by >10% in a single day. In other words, changes in body composition estimates from BIA scales could differ from true changes in body composition for a period of months or even years. This potential long-term tracking error compounds the issues of poor accuracy and reliability. As such, it doesn’t seem prudent to use day-to-day body composition estimates from BIA to inform MacroFactor’s algorithms.
Further Problems with Body Composition Estimates from Circumference Measurements
The most popular circumference-based formula used to estimate body composition is the US Navy’s formula.
The Navy’s formula requires two circumference measurements for men – waist and neck circumference – and three circumference measurements for women – waist, hip, and neck circumference.
In this formula, waist and hip measurements are assumed to positively correlate with body fat percentage, since males tend to store fat around their waist, and females tend to store fat around their waist and hips. Furthermore, neck circumference measurements are assumed to negatively correlate with body fat percentage. It’s assumed that neck circumference is a general proxy for muscularity and the size of an individual’s frame. So, if two people have the same waist circumference, but one person has a larger neck, the Navy formula will assume that the large-necked individual is leaner.
The main drawback to this approach for estimating changes in body composition over time is that changes in neck circumference don’t really reflect changes in general muscularity or the size of one’s frame over time.
So, for example, if you started doing some resistance training for your neck musculature, the Navy’s formula would say that you were getting substantially leaner, even if your body composition wasn’t changing. More worryingly, some people store a decent amount of fat around their neck – for those individuals, decreases in neck circumference are a positive indicator of fat loss, but the Navy formula would interpret the decrease in neck circumference as an indicator of fat gain. The inverse is also true – if you happen to store a decent amount of fat around your neck, you may be able to gain a significant amount of fat without the Navy formula suggesting that your body fat percentage was increasing very much, because increases in waist and hip circumference would be partially (or entirely) “offset” by increases in neck circumference.
As with BIA, changes in body composition estimates derived from circumference measurements could fully conflict with actual changes in body composition over reasonably long time scales.
In short, the two most feasible methods for frequently estimating body composition have the potential to be misinformative – not just uninformative – over periods of weeks to months. At best, daily body composition estimates require a relatively long observation window to separate the signal from the noise. At worst, daily body composition estimates can provide you with a completely false signal. As such, we don’t think it would be prudent to incorporate this data into MacroFactor’s coaching algorithms, or use it for the purpose of goal-setting.
Even if a user is primarily using MacroFactor to change their body composition (rather than simply to gain or lose weight), we believe that it makes the most sense to focus on the weight-related goal that generally accompanies the desired change in body composition (weight loss for people aiming to lose fat, weight maintenance for people aiming to recomp, and weight gain for people aiming to gain muscle), along with the behavioral factors that make a successful body composition-related outcome more likely (activity levels, adequate sleep, resistance training, self-monitoring, etc.). This approach is more likely to lead to a successful outcome than directly focusing your attention on changes in estimated body composition.
How Does Recomping Affect the Performance of MacroFactor’s Algorithms?
Since MacroFactor doesn’t use day-to-day body composition estimates for program adjustments, some users may wonder how MacroFactor handles periods of body recomposition when a user is simultaneously losing fat and gaining muscle.
Fat tissue is more energetically dense than lean tissue – lean tissue stores about 1800kcal per kilogram, while fat tissue stores about 9400kcal per kilogram, on average. So, in theory, you could be in a non-trivial energy deficit without a change in weight if you were simultaneously losing fat and gaining muscle. This principle also applies in reverse – if you lost a kilogram of muscle while gaining a kilogram of fat, that would represent a net surplus of approximately 7600kcal without a change in total body weight.
So, how would body recomposition affect your energy expenditure estimate and nutrition recommendations in MacroFactor?
I’ll illustrate using an uncharitable scenario of considerably better-than-average body recomposition. In other words, this scenario would essentially be the worst-case scenario for MacroFactor’s algorithms, and represent a practical upper limit on the degree to which body recomposition could “confuse” MacroFactor’s estimate of your total daily energy expenditure.
Over about 10 weeks of resistance training, untrained men gain about 1.6kg of lean body mass, on average. For this scenario, we’ll assume that an individual is gaining lean mass at twice that rate – 3.2kg of lean mass over 10 weeks. Typically, such a gain in lean body mass would accompany a gain in total body mass, but we’ll assume that this individual is not only gaining lean mass at twice the typical rate; they’re also fully recomping. Thus, they don’t gain any weight at all, but they lose 3.2kg of fat mass over 10 weeks. This is truly a pie-in-the-sky dream scenario for virtually any lifter – an extreme amount of body recomposition, in a relatively short period of time. Finally, although it’s not too important for this particular illustration, we’ll assume that this individual is consuming 3000kcal per day.
In this scenario, MacroFactor would see that this individual is consuming 3000 calories per day, and not experiencing any change in body mass. Therefore, their estimated energy expenditure would be 3000 calories per day. Simple enough.
However, we know that MacroFactor’s estimate of this individual’s energy expenditure is incorrect – since they’re losing (calorically dense) fat tissue and gaining (less calorically dense) lean mass at the same rate, they must be in a net energy deficit. In other words, we know that they’re expending more than 3000 calories per day.
We can calculate this individual’s weekly energy deficit easily enough. They’re gaining 0.32kg of lean mass per week, and losing 0.32kg of fat mass per week. Gaining 0.32kg of lean mass means this individual is storing a net of 576 calories per week (0.32kg * 1800kcal/kg) in their newly synthesized lean tissue, and expending a net of 3008 (0.32kg * 9400kcal/kg) from their catabolized fat tissue. Thus, each week, they’re in a net caloric deficit of 3008 – 576 = 2432kcal. A weekly deficit of 2432kcal represents a daily deficit of about 347 calories. Therefore, this individual’s actual energy expenditure would be 3347kcal/day (not 3000kcal/day).
To put things in perspective, that magnitude of error (347 calories per day, or approximately 10%) represents an extreme edge case. With a more typical rate of body recomposition – for example gaining 1kg of lean mass while losing 1kg of fat over 10 weeks – the error would only be about 108 calories per day (or about 3.5%). Furthermore, this worst-case scenario for the accuracy of MacroFactor’s algorithms still compares favorably to other methods someone might use for estimating their daily energy expenditure. Wearable devices under- or over-estimate total energy expenditure by more than 10% around 82% of the time, and static equations used to estimate total daily energy expenditure have an average error of ~325 calories per day.
Thankfully, this inaccuracy doesn’t actually have a negative impact for anyone pursuing a weight-related goal – or even a goal related to body recomposition – when using MacroFactor.
For example, if someone is aiming to lose a pound per week (which typically requires someone to consume about 500 fewer calories per day than their maintenance intake), it makes the most sense to calculate their calorie intake targets based on the level of calorie intake that results in weight maintenance, rather than the individual’s “true” total daily energy expenditure.
In the scenario above, a daily caloric intake of about 2500 calories would result in about a pound of weight loss per week, based on the fact that an intake of 3000 calories per day results in weight maintenance. If we used an energy expenditure figure of 3347 calories per day to recommend a calorie intake target for losing a pound per week, the resulting intake target – 2847 calories per day – would almost certainly be too high for the user to lose weight at their desired rate.
Similarly, if someone is specifically aiming for body recomposition – targeting weight maintenance, a very slow rate of weight gain, or a very slow rate of weight loss to simultaneously gain muscle and lose fat – baking in the energy expenditure estimation “error” resulting from body recomposition is actually preferable when calculating energy intake targets.
In the scenario above, the user was simultaneously building muscle and losing fat while consuming 3000 calories per day. If they want to keep gaining muscle and losing fat while maintaining their current body weight … they should keep eating 3000 calories per day. In other words, the expenditure estimation error is the result of body recomposition. There’s no need to correct that error, because the energy expenditure estimation error is the outcome of achieving your desired body recomposition, and the result of the fact that your current intake targets are completely appropriate for your goals. The old adage “don’t fix what isn’t broken” perfectly describes this situation.
What you should measure if you’d like to track body composition
Although it’s a challenge to accurately measure your precise body composition, you may still be interested in tracking some sort of quantitative metric(s) to gauge the relative success of a weight gain or weight loss attempt.
With that in mind, these would be my recommendations:
1. Waist circumference
Though waist circumference isn’t a perfect predictor of body fat percentage, it is a useful predictor of central adiposity and visceral fat accumulation, which are associated with a lot of negative cardiometabolic health outcomes. If you’re aiming to lose weight, consistent decreases in waist circumference are a good indicator that you’re losing fat around your abdomen; if you’re gaining weight, no increases in waist circumference (or very slow increases in waist circumference) are a good indicator that your attempt to gain muscle isn’t resulting in a ton of indiscriminate fat gain.
2. Skinfold thicknesses
While I wouldn’t necessarily recommend converting skinfold thicknesses to a precise estimate of body fat percentage, skinfold thicknesses are a direct measure of subcutaneous fat thickness. In other words, if your sum of skinfold thicknesses decreases from 150mm to 130mm, you can’t say that your total body fat percentage decreased by exactly 2.1% (for example), but you do know that you have less total subcutaneous fat than you used to have.
I’d also recommend interpreting these measurements carefully, and making comparisons at representative points in time over successive attempts to gain or lose weight. In particular, waist circumference can be a somewhat noisy measurement, since it’ll be impacted by general bloating and the amount of food you’re presently digesting. For example, if you measure your waist circumference at the very end of a weight loss attempt, and then track changes in waist circumference as you attempt to gain weight, you’ll likely find that your waist circumference increases by an inch or two over the first couple of weeks when caloric intake is higher. This increase doesn’t necessarily mean that you’ve rapidly accumulated visceral fat – it just means that there’s more food in your digestive tract.
However, comparisons between representative points in successive diets can be informative. For example, at the end of your last weight gain phase, your waist circumference may have been 36 inches and your sum of skinfold thicknesses may have been 140mm at a body weight of 190 pounds. If you measure at the end of your next weight gain phase, and your measurements are the same (a 36-inch waist and a 140mm sum of skinfold thicknesses) at a body weight of 197 pounds, that’s an excellent indicator that your recent attempts to lose fat and build muscle were successful. You could also compare measurements at the end of successive weight loss attempts. If you wound up at a body weight of 160 pounds after two successive weight loss phases, but your waist circumference was 2 inches smaller and your sum of skinfold thicknesses was down by 10mm, that’s an excellent indicator that you now have more muscle and less fat at the same body weight.
Why worry about body composition in the first place?
By this point in the article, I wouldn’t be surprised if some readers feel unmoored and adrift in a sea of uncertainty. If social media is any indication, it seems that many people put a lot of stock in precise estimates of their body fat percentage. People on Instagram like to brag that they got down to 6% body fat for a physique contest, or that they “walk around at 12% year-round.” Nutrition-related Facebook groups are awash in posts from people asking group members to estimate their body fat percentage from photos, or posts trumpeting the success or bemoaning the failure of their last diet based on DEXA scans pre- and post-diet. Clearly, many people think that there’s a lot of value in attempting to precisely quantify their body composition.
However, I’d like to push back against that impulse.
I suspect that you don’t actually care very much about your exact body fat percentage or body composition for its own sake. I suspect that, if you stop and think about it, you’ll realize that you primarily care about your body composition either as a means to an end, or as a proxy measure for something else that actually matters to you.
For example, if you’re an athlete, you may want to maintain muscle mass while losing fat for the purpose of improving performance in your sport. If that’s the case, I’d suggest that monitoring sport performance (or more direct proxies for sport performance, such as sprint speed, jump height, aerobic endurance, maximal strength, etc.) provides you with far more useful information than a body composition estimate would. If you successfully lose fat while maintaining muscle mass, but your sport performance suffers because you were underfueled for workouts and practices, was your diet truly a success? Conversely, if you experienced the desired increase in sport performance, does it really matter whether or not you achieved the change in body composition that you were pursuing for the purpose of improving your sport performance?
Similarly, if your goals are primarily related to aesthetics – you’re aiming to gain muscle or lose fat for the purpose of achieving a certain appearance – I’d posit that progress photos are dramatically more valuable than body composition estimates. If you think you look better than you used to, do you really need to know the precise amount of fat you’ve lost and muscle you’ve gained? Your desired outcome is a visual outcome, after all.
Perhaps your goals are related to health and graceful aging – you’d like to lose some fat to reduce your risk for heart disease, and build or maintain strength into older age. In that case, I’d posit that changes in blood pressure, blood lipids, and biomarkers for inflammation are far more informative than knowing whether you lost precisely 5kg versus 8kg of fat mass, and I’d further posit that more direct assessments of muscular strength (i.e. performance in the gym) are far more informative than knowing whether you gained precisely 2kg versus 4kg of muscle mass.
In short, changes in body composition are rarely the actual goal someone is pursuing. Rather, changes in body composition are assumed to be associated with the goal someone is pursuing. More often than not, you can assess progress toward the actual goal more directly and more precisely than you can assess body composition. So, while it might be nice to be able to accurately and reliably assess body composition for individuals, we don’t actually lose much by acknowledging that we can’t accurately and reliably assess body composition for individuals.
Wrapping up
Just to briefly recap, here are the take-home points of this article:
- You can’t measure body composition. You can only estimate body composition.
- Group-level estimates of body composition are generally quite good. Individual-level estimates of body composition are generally too imprecise and inaccurate to be particularly informative or actionable.
- MacroFactor uses a rough estimate of your body composition to inform your protein targets and our initial estimate of your energy expenditure. More precise estimates wouldn’t meaningfully affect or improve our nutrition recommendations.
- MacroFactor doesn’t use day-to-day body composition estimates to inform its recommendations for two reasons: 1) body composition estimates are a lot noisier than weight measurements, so it would take too long to separate the signal from the noise, and 2) practical methods for estimating body composition can be misinformative – not just uninformative – over reasonably long time scales.
- More often than not, quantitative body composition goals serve as proxies for the individual’s true goal, and the pursuit of body recomposition underpins other, more meaningful goals. Thus, you don’t actually miss out on much by acknowledging that body composition can’t be accurately and reliably measured at the individual level.
Footnotes
- A full 4-compartment body composition assessment combines various techniques of assessing body composition in order to estimate fat mass, total mineral mass (mostly from bone), total protein mass, and total body water. You generally won’t find all of the necessary equipment to do a 4-compartment body composition assessment outside of a research laboratory. As such, while full 4-compartment body composition assessments are quite accurate, they’re rarely accessible for individuals interested in monitoring their body composition.
- The prior sentence is probably painting with slightly too broad of a brush – when most people hear “body composition,” they’re generally thinking about estimates of fat mass, lean mass, and body fat percentage. I’m writing with that particular connotation of the term in mind. However, assessments of bone mineral density and bone mineral content also fall under the umbrella of “body composition,” and DEXA scans can actually estimate BMD and BMC quite accurately – if you have osteoporosis or osteopenia, or if you’re at risk of osteoporosis or osteopenia, BMD and BMC assessments may be quite useful and informative. That’s a conversation to have with your doctor, and it goes beyond the scope of this article.
