I’m starting this page to serve as a central repository of significant concrete facts about obesity. I’m making this a page because I anticipate frequently updating it as necessary. There will be no speculative musings or inferences based on partial evidence, everything here will be solid facts as per the research.
Most people who comment on this topic – especially the most vocal – tend to be quite clueless on these matters. This should be a one-stop spot to bring everyone up to speed.
Individual heritability of obesity
Obesity is known to be highly heritable. A meta-analysis of twin and family studies (which included the standard MZ together vs. DZ together – MZT-DZT – in addition to MZ apart, MZA), which looked at a combined total of 140,525 individuals found a heritability of body mass index (BMI) in the range of 0.75 to 0.82. The countries observed included the U.S., Britain, Australia, Denmark, Sweden, Norway, Italy, Finland, South Korea, Taiwan, China, and Japan. The results are highly consistent with little variability across the studies.
If we were to rely solely on estimates drawn from MZT-DZT studies, the matter of whether some sort of environmental factor(s) that serves bias upward heritability estimates would remain. This is addressed by the smaller MZA & DZA studies (included in the analysis). One in particular had a sample of 93 pairs of MZA and 218 pairs of DZA. The correlation was for the MZA (for men) was 0.74 (n pairs = 49) and 0.66 for women (n pairs = 44). All told, the study estimated a broad sense heritability of 0.74 for men and 0.69 for women, with much of the variance being due to non-additive genetic effects.
This is additionally supported by adoption studies. One review of twin and adoption studies (n adopted ~ 1,400) found significant correlations between adopted children and their biological relatives and no correlation between adoptive relatives, at least not in late adolescence and adulthood. In other words, in the adoption data, the heritability was found to be high and the shared environment impact was found to be zero, in agreement with twin studies.
The high heritability of BMI was also confirmed by a genome-wide complex trait analysis (GCTA), which used directly measured genomic similarity to assess heritability. One analysis with 20,240 sibling pairs found that captured SNPs could explain 42% of the variance in obesity. This estimate is a lower bound of the additive heritability only, and then only the heritability explained by common SNPs. This is in agreement with the apparent additive heritability of obesity (0.3-0.4), which appears to be in that range from MZA studies.
Another GCTA, from Iceland, which looked at 38,167 individuals and estimated heritability from both relatives and non-relatives, found a lower-bound broad-sense heritability estimate (additive + non-additive genetic variance) of BMI of 0.471. GCTA results are lower bounds estimates, because they are limited by genetic variants captured by the analysis.
The bottom line is that genes heavily influence individual variation in obesity. It seems some people are incapable of understanding this, but it should go without saying that genetic differences are, however, not involved in the change in obesity rates over time. Genes haven’t changed significantly during the past few decades, so some “environmental” factor(s) must be in play. Please see my post Why HBD for a description of how gross environmental change can lead to gross phenotypic change without any genetic change.
It is also important to note that the shared environment contribution is reliably zero. This means, as is the case with behavioral traits, parents and the family environment have no effect on adult obesity. One cannot blame adult body weight on parental choices etc.
Finally, there is a marked misconception that there is some sort of “debate” about whether body weight is genetic (which it heavily is) or stems from “choices” (i.e., behavior). This is one of those things that’s not even wrong. There can be no dichotomy between “genes” and “choice.” As HBD Chick would put it, where do choices come from? As readers of this blog know, the First Law of behavioral genetics is that all human behavioral traits are heritable. Genes (i.e., specifically, genetic differences) impact all human behavior (i.e., behavioral differences) to some degree. Indeed, often considerably so. More fundamentally, the phony dichotomy between genes and choice stems from a key misunderstanding: the failure to realize that the universe is deterministic. All events, including human behaviors, have causes. (This is even considering quantum mechanics; quantum indeterminacy is merely another “deterministic” force. Random interplay is just another causal agent.) See my post No, You Don’t Have Free Will, and This is Why.
To the extent that behavior (again, specifically, behavioral differences) lead to differences in body weight – whatever extent that is – those behaviors are themselves quite heritable.
Indeed, in the case of “self-control” (which may or may not be causally related to obesity), it too appears to be largely heritable and not affected by the shared environment. One recent study found that 76% of the variance could be attributed to additive genetic factors.
Group-level heritability of obesity
As I’ve discussed previously, there is not much by way of evidence that the underlying nature of differences between groups is all that different from the nature of differences between individuals within a group – with perhaps the effects of highly deprived environments (such as encountered in sub-Saharan Africa and other highly impoverished areas) excepted. Yet there are significant group-wide differences in the obesity rate. As seen in my post A Fat World – With a Fat Secret?, we see a pattern within nations and between them.
As discussed in the above post of mine, overall rates of obesity cluster in related groups, even when those groups live together in the same nations. As seen in the U.S., the pattern of ethnic clustering is visible among American Whites, as seen by this estimate of county-level White obesity rates by Razib Khan (blue = higher; red = lower):
The obesity rate heavily tracks area of Scot and Scots-Irish ancestry – and secondarily, German ancestry, in line with the high obesity rates in Scotland and Germany, respectively. See my posts HBD is Life and Death and More Maps of the American Nations.
The evidence strongly indicates that group-level differences in obesity are highly heritable, just as is individual variation in obesity – founder effects and selective migration (and, in the developing world, extreme deprivation and dietary incompatibility) notwithstanding.
Ineffectiveness of obesity treatments
Obesity is very difficult to impossible to treat. The most common prescription, and indeed the prevailing conventional wisdom, is that “lifestyle” changes are the best solution. This typically means diet and exercise. However, this has been extensively studied. Across the population, diet and exercise, each individually and in tandem, are completely useless to treat obesity, in the long term.
In the case of exercise, randomized controlled trials (RCTs) don’t even show a short-term benefit. One 2007 meta analysis by Franz et al looked at the results of all sorts of different interventions. For exercise-alone prescriptions, it found that the treatment groups lost no weight at 6 months (well, less than 2 kgs, but even this number comes only when you look at those who remained in the study). Indeed, after a year, the control groups actually lost more weight than the treatment groups. The total weight change was small and close to zero throughout.
In the case of diets, particularly the most common low-fat and low-calorie diets, a very large meta-analysis of RCTs with a combined N > 60,000 (of which ~48,000 came from a single mammoth trial) and a study duration of 2.5 – 10 years, found that diet was completely ineffective for weight loss. The subjects showed no aggregate permanent weight loss at the end of the study period. The largest of these studies, the one by Howard et al (2006) found little change, a total loss (over 3 years) of less than 1 kg (and a difference between control and treatment groups of 1.29 kg, favoring treatment).
As for diet and exercise combined, several studies in both previous meta-analyses look at trials which tested both together. The result was the same: little to no significant aggregate weight loss, especially after longer periods of time.
This is true of low-carbohydrate diets as well. One meta-analysis looked at RCTs. Each of the trials were individually small (n = 11 – 153), but all told there were 712 subjects in the low-carb trials. The duration of studies ranged from 12 to 24 months. The total average weight lost with the low-carb diet groups was on the order of 4 kg! And that’s with considerable attrition in the studies. Low-carb diets don’t work much better, either.
A new randomized comparison trial (Bazzano et al, 2014) of a low-carb vs a low-fat diet (N = 148) found only a weight loss of 5.3 kg after 1 year with the low-carb diet, but only 1.2% change in body fat percentage.
The results are visualized in this graphic from Franz et al:
Drugs designed to treat obesity often don’t fare much better in trials, as seen with the drugs examined here.
Randomized controlled trials are necessary to judge the efficacy of obesity treatments, especially when crafting broad prescriptions for treatment. Popular conceptions of weight-loss – even in the minds of medical professionals – are biased by odd examples of individuals with impressive short-term weight loss or rare individuals who managed to maintain a lowered weight for a long period of time. This is obviously improper, because it ignores the fact that these individuals are exceptional. The efficacy of any treatment, especially one that is to be recommended as a general prescription to treat any condition, needs to be subjected to clinical trials to judge their effect across the population to which the treatment is intended. This is what we do for new drugs.
However, even these have their limits. RCTs of lifestyle modifications are plagued by considerable non-compliance, even for those who remain in the study. This is especially acute towards the later parts of the longer trials. Some commenters correctly point out that this weakens any conclusions we can draw because we don’t know what the results would be if the subjects stuck to the trial.
However, I argue that this point, as technically true as it might be, is ultimately wholly academic. Patient non-compliance is an important part of the study, because these people aren’t going to live their entire lives in a laboratory under tightly controlled conditions; they are going to live their lives in the real-world feeding and exercising at their own accord. If dietary changes and exercise fail thanks to non-compliance – a point which is not at all clear at the moment, mind you – then they are useless as prescriptions to treat obesity en masse. It does no good to give advice that most people can’t stick to, assuming that non-compliance was the point of failure.
There has been quite a bit of in-lab based obesity experimentation. Among the most interesting studies out of these are overfeeding experiments. These are experiments where subject are deliberately overfed for some period of time. The best of these confine subjects to a laboratory for the duration of the experiment. The subjects are monitored extensively during that time. Because of the expensive nature of these in-patient trials, they always have tiny samples. A review such studies can be found here:
One of the best studies used pairs of identical twins (Bouchard et al, 1990). Twelve pairs of identical twins were kept in a lab and overfed for three months. All the twins gained some weight. However, there was a marked variation in how much weight each subject gained. But, more strikingly, there was a tight correlation between co-twins in weight gain:
Both the total weight gained and the distribution of that weight was strongly correlated between twins. Despite the small sample, this is a poignant demonstration of both the great degree of individual variation in propensity to gain weight – even in the same broad conditions – and the genetic underpinnings of such variation.
Even more interesting is the follow-up of these twins (Bouchard et al, 1996). As is the case with most overfeeding experiments the subjects returned to their pre-study weight following their release (however, there was some attrition in the follow-up, so it’s unclear how much that affects these results):
That change was also genetically directed:
Overstated impact of obesity per se on health
Epidemiology has shown that obesity is associated with a myriad of adverse health outcomes. Epidemiologists have thus proclaimed that obesity is directly causal to those negative health outcomes. This conclusion is faulty for a key reason. As described by Gary Taubes (emphasis mine):
Another problem endemic to obesity and nutrition research since the second world war has been the assumption that poorly controlled experiments and observational studies are sufficient basis on which to form beliefs and promulgate public health guidelines. This is rationalised by the fact that it’s exceedingly difficult (and inordinately expensive) to do better science when dealing with humans and long term chronic diseases. This may be true, but it doesn’t negate the fact the evidence generated from this research is inherently incapable of establishing reliable knowledge.
The shortcomings of observational studies are obvious and should not be controversial. These studies, regardless of their size or number, only indicate associations—providing hypothesis generating data—not causal relations. These hypotheses then have to be rigorously tested. This is the core of the scientific process. Without rigorous experimental tests, we know nothing meaningful about the cause of the disease states we’re studying or about the therapies that might work to ameliorate them. All
we have are speculations.
The suspicion dawned on me that the connection between mortality and obesity could be mostly, if not entirely, a result of IQ.
And sure enough, I found something that strongly suggested this. It turns out that the venerable Satoshi Kanazawa did a study that found, in a White British sample, IQ measured in childhood predicts obesity at age 51. I discussed this in a post, that is currently experiencing decent readership thanks to the Geoffrey Miller fiasco: Obesity and IQ
The next logical step was to ask the question of how well IQ correlates to shortened lifespan. And I did that with my 99th post, IQ and Death. Looking at a meta-analysis of several studies of IQ and mortality, it was found that IQ is associated with longer lifespan. Indeed, at least one study in the meta-analysis did look at other possible attenuating factors. It found that IQ was by far the strongest predictor of death. Indeed, “marital status, alcohol consumption, systolic and diastolic blood pressure, pulse rate, blood glucose, body mass index, psychiatric and somatic illness at medical examination) was negligible (10% attenuation in risk)!”
The association between obesity and shortened life, and perhaps most health problems, is mostly, and perhaps entirely, a result of obesity’s association with IQ. As I have noted, and as I have been embroiled in a little controversy over, the “conventional wisdom” on diet, exercise, obesity, health, and death is pretty much bullshit.
Bolstering this point, and much more important from a practical standpoint, is the failure of interventions targeted obesity to produce concrete improvements in health or extension of life. As reported in Tomiyama, Ahlstrom, and Mann:
The diets did not appear to meaningfully lower lipid levels, and accordingly, improvements in coronary morbidity/mortality and stroke were minimal. In all five studies that reported on these outcomes (Hanefeld et al., 1991; Howard et al., 2006; Miettinen et al., 1985; Sone et al., 2010; Whelton et al., 1998), the diets did not lead to significant reductions in coronary morbidity or mortality. Furthermore, in only two (Miettinen et al., 1985; Sone et al., 2010) of the five studies did the diet lead to significant reductions in stroke, and the researchers for one of these studies (Sone et al., 2010) noted that the significant finding should be treated with caution, as there were no group differences on most of the risk factors
Overall, there were only slight improvements in most health outcomes studied. Changes in diastolic and systolic blood pressure, fasting blood glucose, cholesterol, and triglyceride levels were small, and none of these correlated with weight change. There were also very small effects of these diets on lipid-lowering medication use and coronary morbidity and mortality.
The much hyped PREDIMED trial from Spain (Estruch et al, 2013) of the “Mediterranean diet,” which claimed to find a reduction in cardiac deaths in the treatment group, failed to make a significant reduction in all-cause mortality. This is addition to the many other methodological problems with this trial (like stopping early).
The study randomly assigned 5,145 overweight or obese people with Type 2 diabetes to either a rigorous diet and exercise regimen or to sessions in which they got general health information. The diet involved 1,200 to 1,500 calories a day for those weighing less than 250 pounds and 1,500 to 1,800 calories a day for those weighing more. The exercise program was at least 175 minutes a week of moderate exercise.
But 11 years after the study began, researchers concluded it was futile to continue — the two groups had nearly identical rates of heart attacks, strokes and cardiovascular deaths.
But the outcome is clear, said Dr. David Nathan, a principal investigator and director of the Diabetes Center at Massachusetts General Hospital. “We have to have an adult conversation about this,” he said. “This was a negative result.”
The nature and implications of these findings, which is rather damning considering the conventional prescription for heath, is captured well by Tomiyama, Ahlstrom, and Mann:
We believe the ultimate goal of diets is to improve people’s long-term health, rather than to reduce their weight. Our review of randomized controlled trials of the effects of dieting on health finds very little evidence of success in achieving this goal. If diets do not lead to long-term weight loss or long-term health benefits, it is difficult to justify encouraging individuals to endure them.
Pretty much says it all.
Well, there you have it. The matter of obesity is one that is deeply personal and brings out powerful emotions in people. This is to the point that it clouds people’s reasoning and ability to look at the matter rationally. The overwhelming weight of the evidence speaks to considerable genetic involvement in obesity, both on the individual and group level; to the lack of a causal effect on health or lifespan; and to the total failure of interventions – especially the two most commonly prescribed ones – diet and exercise – to significantly impact it. As well, even as such, the evidence makes clear there is little impact on health from these inventions, which – considering the material and emotional cost – makes them less than useless.
Indeed, this video (thanks to a slick commenter) from College Humor perfectly captures the situation:
In this edition of this page, I have forgone talking about surgical interventions. For now, I will say the evidence isn’t looking too good there, either. Future updates will review the evidence on bariatric surgery, as well as new findings as they come to my attention.
Note: Critical commenters that make it obvious that they did not read the page face the prospect of being permanently banned on the spot. Please read the relevant section of this page carefully before raising your objection