The folly of the mean – Why do they differ so?

A few years ago I did a small study where we combined a ketogenic diet with resistance exercise [1]. After 10 weeks the 8 women in the diet group experienced a mean weight loss of 5,6kg. DEXA-scanning (Dual Energy X-ray Absorptiometry) showed that the mean fat mass loss was 5,6kg and that the fat free mass was unchanged. But, it would be wrong of me to say that this strategy conserves lean body mass, because the truth of the matter is that it all depends. Because 4 of the women in the diet group increased lean body mass, while 4 lost some despite the 10 weeks of resistance exercise. The individual results from the study are illustrated below.

As you can see, there are large individual variations. But individual variations are often not presented in scientific articles. Instead we are given results as means, and the means do not tell us what happens at the individual level. Of course, results also commonly include a measure of variation such as standard deviation (SD) or confidence interval (CI), but that still does not tell us what the true variation was and exactly how the different individuals responded. And this is the trouble with presenting study results as means; there are always individual variations and the funny thing is that it is usually these very variations we really want to understand.

Exercise studies are good examples of how misleading averages and means can be. Generally speaking we have to say that exercise at best is a very poor weight loss strategy. The reason is that exercise studies rarely yield weight loss of any significance. Even if we didn’t have exercise studies we would still know that exercise does not make us lean, because we all know that our body weight does not change with changes in exercise volume. My body weight is practically constant, no matter how much or how little I exercise, and most people seem to be like me. Of course, if we do resistance or strength training we will build muscles, and often gain weight, but there is just no obvious link between energy expended during exercise and concomitant weight loss.

However, if we look closer at exercise studies it may seem that it is possible to lose weight from exercising, but that it doesn’t happen to all of us. Take for example a study lead by Neil A. King at Queensland University of Technology [2]. Fifty-eight overweight and obese men and women completed 12 weeks of supervised exercise in a laboratory. The exercise sessions were designed to expend 2500 kcal/week and involved exercising at 70% of each individual’s maximum heart rate for 5 days a week. The aim of the study was to assess the effects of 12 weeks of mandatory exercise on appetite control.

After the 12 weeks of exercise the mean weight loss was 3,6kg. Now, we cannot conclude that this loss was caused by the exercise itself. Usually when people are recruited to studies such as these, they tend to change their behavior towards a more healthy lifestyle. Unless we can control for eating behavior, stress, alcohol intake or any other factor known to influence weight loss we cannot say that exercise is a causal factor. Anyway these are the individual results from the study:

About 22 of the participants, or roughly half, lost more weight than the mean. And 10 participants gained weight. So the really interesting question is; what is the cause of the difference in the individual responses? Although the mean weight loss was small and likely affected by non-exercise factors the above results do not exclude the possibility that some of these people lost weight simply by exercising and without significant changes in other lifestyle factors. In fact the very statement that I usually make, that exercise does not cause significant weight loss, is based on results given as means. But what if there’s always responders and non-responders equaling each other out?

Still, we need to remember that if exercise by itself caused some of the people in the King study to lose weight, it is more likely because of factors such as reduced insulin resistance, reduced glycogen stores or improved fat metabolism, than because of increased energy expenditure.

Diet studies are also hard to interpret based on results presented as means. The Look AHEAD (Action for Health in Diabetes) [3,4] documented the effect of a traditional lifestyle intervention on overweight and type 2 diabetics. More than 5000 overweight men and women were randomized into an intensive lifestyle intervention group (ILI) or a control group that only received information and support. The study lasted 4 years (this is a gigantic study and one of very few randomized controlled trials of this size) and the goal of the ILI group was to achieve a 7% weight loss and to maintain the loss throughout the 4 years. The participants in the ILI group were asked to eat 1200-1800 kcal per day of which less than 30% was to come from fat. In addition the goal was to exercise 175 minutes per week, and they participated in regular group and individual counseling. All in all, the researchers did all they could to make sure the participants lost weight.

After 4 years of dieting the mean weight loss in the ILI group was 4,9kg or 4,7% of baseline body weight. The below graph illustrates that many achieved a great weight loss after 1 year, but as the study progressed, the participants gradually regained their lost weight, and if the study had lasted any longer the mean weight loss would probably have been even smaller.

But once again the mean results don’t really tell us much. Wadden and coworkers reveals that only 74% of the participants lost weight and that the remaining 26% gained weight. Only 46% lost more than 5% of initial body weight (which was 95kg in women and 109 in men, so roughly 5kg), and only 35% lost more than 7% of initial weight. With so much effort in so heavy people these results are a strong indication that traditional dieting simply does not work. But once again we need to ask what the difference between the participants who lost a lot of weight and those who lost little was. This is the really important question, and a question that is asked to rarely. Those who continue to cling to the old dogma might say that those who lost the most weight probably were those who followed the given advice the most and that those who lost little did not do as told.

And they might be right. But we need to know. We can be pretty sure that many of the participants did not do as told and that many did more. That’s just how people are and had this been a smaller and more tightly controlled study, anyone who did more or less than asked would have been excluded from the analysis. But there might also be large individual variations in response to the same intervention, and once again it is these variations – the reason we respond differently to the same stimuli – we truly want to understand.

One way we sometimes try to shed light on some of the individual variations is to do correlation analyses. For example, if in the above study, there was a strong correlation between protein intake and weight loss, then differences in protein intake was probably an important reason for the individual variations. But the ting is that we rarely find such strong correlations in weight loss studies and so we are left in the dark when it comes to understanding what biological mechanisms are hidden in the mean.

Even though studies fail to elucidate the reason for individual variations, I would still like the individual results to be presented more often, because this acts as a strong reminder that we cannot truly understand the world if we cannot understand why we differ so.


1. Jabekk PT, Moe IA, Meen HD, Tomten SE, Hostmark AT: Resistance training in overweight women on a ketogenic diet conserved lean body mass while reducing body fat. Nutr Metab (Lond) 2010, 7: 17.

2. King NA, Caudwell PP, Hopkins M, Stubbs JR, Naslund E, Blundell JE: Dual-process action of exercise on appetite control: increase in orexigenic drive but improvement in meal-induced satiety. Am J Clin Nutr 2009, 90: 921-927.

3. Wing RR: Long-term effects of a lifestyle intervention on weight and cardiovascular risk factors in individuals with type 2 diabetes mellitus: four-year results of the Look AHEAD trial. Arch Intern Med 2010, 170: 1566-1575.

4. Wadden TA, Neiberg RH, Wing RR, Clark JM, Delahanty LM, Hill JO, Krakoff J, Otto A, Ryan DH, Vitolins MZ: Four-year weight losses in the Look AHEAD study: factors associated with long-term success. Obesity (Silver Spring) 2011, 19: 1987-1998.

8 kommentarer om “The folly of the mean – Why do they differ so?”

  1. There are some great points here. It seems that too often, people analyze studies as though the variations across individuals were nothing more than random chance, yet there are of course concrete (maybe just not known) explanations.

    Many give weight loss advice based on shallow «tricks» to simply reduce calorie intake (or increase expenditure–«take the stairs»). However, the individual variation in overfeeding, exercise, underfeeding, etc *implies* more complexity and a need to have a deeper understanding. Unfortunately that goes largely unrecognized.


  2. There are weaknesses inherent to all studies and I think stats are the biggest!!

    Personally I think hormonal havoc is the reason for unintentional weight gain and obesity… (having experienced this myself multiple times)

    What are your thoughts on scientists who subdivide subjects in the analysis by responders v. non-responders?
    Insulin secretion and IR are the signs and symptoms that can divide those who lose the most weight with low GI diet versus low fat (high GI).

    I think it is neat because we have so many reasons for IR (insulin resistance) like to begin with despite perfect paleo diets — pesticides, plastics, parabens, xenoestrogens, hypothyroid conditions, adrenal dysfunction, heavy metals (from batteries, dental amalgams, municipal sources, aluminum, etc), phthalates, PCBs, aromatic hydrocarbons, flame retardants (mostly Calif), fungicides (tributyltin, food preservatives, etc), pharmaceutical obseogens, blah blah blah… Imperfect paleo diets like VLC and ketosis in susceptible individuals also causes physiological IR by triggering adrenal dysfunction and hypothyroid states…

    Have you seen this gem? Ebbeling, Ludwig et al JAMA 2007, which was cited in the (my favorite) Carnivore Connection Revisited article.

    Living on Earth — chemicals that make us fat


  3. Dr. B G, I agree stats is an inherent weakness in biological studies.

    I think that usually there is nothing to gain by dividing subjects into responders and non-responders. As you say, there are just some many influencing factors (confounders). Most likely many of these confounders will change with diet, lifestyle, toxins, stress and of course hormonal milieu and so those who are non-responders may not be non responders all the time. Our biology is not static.

    Insulin resistance is far more complex than carbs and insulin (or paleo vs western diet). And it is important to remember that IR is also a protective mechanism and that there are many ways to measure IR and that the most common method only measures hepatic IR.

    I loved the carnivore connection hypothesis article (think I might have written about it on the blog once), but I don't think I've read the Ebbing et al. There's some interesting stuff there. Going to have to read it more thoroughly though.


  4. Here's er BBCs Horizon on health, diet, exercise and 11 genes:

    My guess is that our ability to respond to exercise and diet are very much dependent on which genes we have, and which are switched on or off. The latter is of course partly dependent on our behavior and our diet (and environment). The (far?) future is indivudually tailored diets and exercise programs to let us maximize quality of life and health, within the framework of our genes.

    Just my 2c 😉


  5. Thanks Pål! I concur with the thoughts on this post and particularly your statement 'IR is also a protective mechanism' because without it we lose energy too readily and do not store fat efficiently in the appropriate depots…for the brain, fertilty, body composition, etc. All of our hormones actually affect IR too.

    Ja. *ha aha!* you did discuss in Jan!

    Have you heard of Björntorp? He's done a lot of cool stuff with insulin and hormone body fat 'mapping' research. Here — he restores normal insulin sensitivity to oophorectimized rats with E2 and P (estradiol, progesterone). P alone does not work; requires E2 in the symphony of repairing hormone havoc.
    Fasting doesn't work for a lot of girls (like me and HIIT mama, Meredith who commented). I tried to post here with Stefani but got spammed out apparently. For females, the animal studies and human anecdotes (like me) appear to show that cortisol gets knocked out the ceiling with intermittent fasting or starvation. I would put ketosis diets into here as well for adversely raising cortisol and IR, and thus prevently body fat loss.

    The gender differences also are not highlighted in many studies which is a shame because hormonally we are VASTLY different. Females may also hoard environmental toxins differently than men since our body fat % is ~DOUBLE.


  6. One more thing… as you know I loved your published study. It creates more questions than answers! Besides including blood and salivary hormone testing in your NEXT study, I would not mind seeing a breakdown of DEXA fat loss among fat metabolizers v. non-fat metabolizers. The metabolic flux, substrates, fat burning responses to exercise and EPOC are fascinating. This study divided the responders by RQ, respiratory quotient (low=higher fat burning which corresponded better with negative energy balance with mod intensity extended exercise).


  7. B.G.

    I agree we have to be careful of assuming results relating to IR in men also apply in women. The example with fasting is a good one. Hormones matter.

    Interesting Bjørntorp article. I know him from other research, but had not seen that particular article.

    Noticed this today about some sex differences in fat storage:

    Assume you've seen this, but your link made me think of it:


  8. P–

    The articles are neat. I haven't seen either so I appreciate both. The last one on RQ was particularly revealing. The authors discussed how certain observations are contrary to expected but are aligned with not only my experiences but to the great majority of cases I see/hear about:

    (a) 'On the other hand, short-term studies conducted in humans
    with dietary manipulation of muscle glycogen stores have
    produced inconsistent, mostly negative, effects on food intake' — when I'm exercising well (aerobic endurance + anaerobic HIIT/weights, my appetite indeed SHOOTS DOWN. There's a threshold ( less than 40-60min/day isn't enough)

    (b) 'association between fat mass and DEI [daily energy intake] in the full model in Table 3. The explanation for this is not clear, but perhaps fat mass, after other metabolic variables are accounted for (fat-free mass in particular), acts as a brake on increased food intake.' I think sarcopenia should be a disease state since it may be able to predict many things including mortality, insulin resistance, hormone status, RQ and particularly LEAN BODY MASS… We have leptin for long term fat mass signalling, but what is the signal for long term lean body mass…??? AMPK? PPAR activation? Glut 4? hormones? Insulin resistance (liver v. muscle v. adipose (BAT v. WAT v. VAT)???


Legg igjen en kommentar

Fyll inn i feltene under, eller klikk på et ikon for å logge inn:

Du kommenterer med bruk av din konto. Logg ut /  Endre )


Du kommenterer med bruk av din Google+ konto. Logg ut /  Endre )


Du kommenterer med bruk av din Twitter konto. Logg ut /  Endre )


Du kommenterer med bruk av din Facebook konto. Logg ut /  Endre )


Kobler til %s