Academic ResearchObesityOther FoodsWomen's Health

Chocolate ‘may help keep people slim’ (not)!

On Tuesday 27th March 2012 the headlines started “Chocolate ‘may help keep people slim‘” announced the BBC. The Independent headline was remarkably similar: Chocolate ‘can help keep you slim‘” Australia reported “Chocolate keeps you slim” and the American Huffington Post went with “Eating chocolate could keep you slim, study finds

This is the original article in the Archives of Internal Medicine, entitled “Association Between More Frequent Chocolate Consumption and Lower Body Mass Index“. Was there a press release that claimed chocolate may help keep people slim? Or did the BBC run with that headline and other sheep follow? Whatever happened – the media headlines are some distance apart from the headline of the article, which was reasonably accurate except that the words “Extremely small” should have appeared before the word association.

You will have to pay $30 for the privilege of being able to view the article for one day, so I hope that this saves you some time and money. The article is very short – approximately 1,100 words and with one table. The details are as follows:


The study involved 1,018 men & women aged 20-85. 68% of these 1,018 people were male and 32% female. However, BMI was only available for 972 subjects and 975 completed a Food Frequency Questionnaire. There is no clarity given as to the number actually used in the study – presumably the overlap between subjects who completed a questionnaire and those for whom BMI was known.

The average (mean) age was 57. The average (mean) BMI was 28 (for 972 subjects) – i.e. overweight. The mean frequency of chocolate consumption was twice a week and the mean frequency of exercise was 3.6 times a week.


The participants were asked the question: “How many times a week do you consume chocolate?”


Two key statements in the article are:
– “Greater chocolate consumption frequency was linked to lower BMI.”
– “Causality in the observed association cannot be presumed.”

There is no abstract for the article. All you can see on the Archives of Internal Medicine web site without paying for the article is the first 150 words. This is not a useful abstract therefore – one where the headlines of the study and conclusions are presented up front. You cannot tell from the first 150 words what strength of association is being claimed and the article itself makes no claims – it merely makes the “greater chocolate consumption frequency was linked to lower BMI” statement quoted above. May I suggest that this is because the observed association is so insignificant as to be not worth declaring up front in a summary.

We need to look at the only table in the article to see the extent of the association between BMI and frequency of chocolate consumption, as reported by the subjects.

The Beta coefficient (β) is as follows:

For unadjusted data  β = -0.142
Adjusted for age/gender β = -0.126
Adjusted for age/gender/exercise/fat/fruit/veg intake/depression scale & calorie intake! β = -0.208

This is tiny. I appreciate that beta coefficients have been used because the units of the two variables are substantially different (BMI vs frequency of consumption). However, had the Pearson correlation method resulted in a correlation of -0.142 (r=-0.142), the r squared would be 0.02. We would interpret this r squared to mean that 2% of the variability in BMI may be accounted for by frequency of chocolate consumption. This hardly justifies world headlines.

Some comments on the study:

1) The headlines “Chocolate ‘may help people slim’” is not justified by this study. As the article notes – no causation can be presumed – in either direction. Does chocolate help people slim or does being slim help chocolate consumption? The latter is surely a far more logical direction of possible causation. I eat chocolate (85%+ cocoa content) frequently and in large quantities because I am slim. Were I not slim, I would hopefully have the good sense to curtail my chocolate consumption.

2) The article did not seek to find, or therefore find, or therefore claim that there is any relationship between any amount of chocolate consumed and BMI. The only question asked was about frequency of chocolate consumption and the only claim of an observed relationship was about frequency of chocolate consumption.

3) There is no definition of chocolate in the article. The European definition of chocolate is dependent on cocoa content. Anything with a low cocoa content is called confectionery. I suspect that this study is about confectionery – not dark, cocoa rich chocolate, but the article does not try to define or clarify this. There is mention in the opening paragraph of the article that “chocolate is often consumed as a sweet” (is this alluding to confectionery?) Later on the article notes “chocolate products are often rich in sugar and fat”. Chocolate products?! Are we talking chocolate cake, chocolate ice cream and/or chocolate biscuits now as well as confectionery? Dark chocolate is not rich in sugar, so can we presume that we are not talking about high cocoa content chocolate?

The final paragraph details a study done on rats where they were fed “cocoa-derived epicatechin”. Cocoa Powder is highly nutritious – exceptionally rich in several minerals such as magnesium, phosphorus, potassium, manganese, zinc, copper and iron. As an example, 100g of cocoa powder delivers 80% of the Recommended Dietary/Daily Allowance of iron. It is disingenuous, however, to mention a study of a cocoa derivative when one is making headlines about ‘chocolate’ consumption and neglecting to differentiate between chocolate and chocolate products, let alone confectionery and high cocoa content chocolate.

4) A study finding needs to be plausible and associations claimed should be shown to have possible rationale. This study fails this test. Claiming an association between frequency of consumption, but not quantity makes no sense. The explanation offered is as follows and please note that this occurs in the rat study paragraph where we are still talking about cocoa-derived epicatechin and rodents:

“Cocoa-derived epicatechin, specifically, is reported to increase mitochondrial biogenesis and capillarity, muscle performance and lean muscle mass and to reduce weight…” What does this have to do with the frequency of consumption of chocolate products with no measure of cocoa content, let alone epicatechin, in humans?

5) The study could have asked what colour socks the person wore or what American football team they supported.  There may also have been a tiny relationship between how frequently people ate chocolate and the colour of the socks they wore or BMI and the football team supported – any observation between whatever frequency of consumption of whatever was presumed to be chocolate and BMI is equally daft.

Would the headlines then have been “Blue socks ‘may help keep people slim'”!?

6 thoughts on “Chocolate ‘may help keep people slim’ (not)!

  • Nadine,
    I believe that fat is a substance. (Which it is!) The word ‘fat’ itself is a noun, and not an adjective. Obese is and adjective, to describe someone who carries too much body-fat. Because this distinction is ignored, it’s become an accepted “fact”, that fat makes us ‘fat’!
    As for chocolate being good or bad, surely it’s the same in theory as saying ‘It isn’t the beef-burger that’s the problem; it’s the bun!’ I.e., it isn’t the chocolate that’s the problem, it’s the sugar they put into it! :)

  • I also noticed this story on the BBC and I thought at the time there were so many holes. I didn’t imagine that they could have run so many headlines on the basis of such a tiny correlation coefficent, thanks for pointing that out!

  • Nadine, why does it always seems that all dietitians/nutritionist are accusing fat people of lying?
    The problem is carbohydrates. Eat 44 g fat (=400 kcal) and you are satified for at least 4 hours.
    Eat 100 g carbohydrates (=400 kcal) and 50 g are transformed into 10 g fat and stored. This means that after 2 h the blood glucose level will plummet and you are forced to eat another 100 g glucose to stay alive for another two hours and store another 10 g fat.

    So in the latter case you are forced by your own body to “overeat” those extra 100 g carbohydrates.

    By eating real animal fat you are always satified and the fat has less than 50% saturated fats, less than 50 % MUFA and normally 5 % PUFA. It’s the same composition of fats in our fat body stores of at least 10 kg animal fat. The perfect source of an essential nutrient as animal meat is the perfect source of other essential nutrients.

    Carbohydrates are toxic if more than 7 g in the blood of a 70 kg (=154 lb). More than 30-50 g glucose in the blood is lethal. So why should we eat nonessential and even toxic/lethal carbohydrates?

  • Hi Zoë,

    You write

    The Beta coefficient (β) is as follows:

    For unadjusted data β = -0.142
    Adjusted for age/gender β = -0.126
    Adjusted for age/gender/exercise/fat/fruit/veg intake/depression scale & calorie intake! β = -0.208

    This is tiny. I appreciate that beta coefficients have been used because the units of the two variables are substantially different (BMI vs frequency of consumption). However, had the Pearson correlation method resulted in a correlation of -0.142 (r=-0.142), the r squared would be 0.02. We would interpret this r squared to mean that 2% of the variability in BMI may be accounted for by frequency of chocolate consumption. This hardly justifies world headlines.

    You seem to be suggesting that the Beta coefficient for the unadjusted data would equal the Pearson’s correlation coefficient and allow an estimate of the variability in BMI attributable to the frequency of chocolate consumption.

    This isn’t right. The Beta coefficient is the slope of the best fit line through the scatterplot of BMI v. chocolate-eating frequency. Pearson’s, on the other hand, represents how tightly clustered along the best fit line the scatterplot points are. How well the points cluster along the line indicates what percentage of BMI variation can be accounted for by chocolate consumption frequency. The slope of the line doesn’t provide any information about this percentage.

    It’s possible to have a very large slope (Beta) and a correlation (Pearson’s) near zero, or vice versa – small slope but strong correlation. (The slope, but not Pearson’s, depends on units. If consumption were measured in times per month instead of per week, the slope would go down by a factor of four, but Pearson’s would not change at all.)

    The slope (Beta) simply indicates how big a decrease in BMI was observed for each additional time per week chocolate was consumed. It doesn’t, however, tell how reliable or consistent this association was – it’s only an average. (In this study, each additional time per week was associated with a drop in BMI of 1/5 of a point, on average, which is a very small change. Without more information, the range of BMI differences might have been wide, with some increases and some decreases, or it could have been consistently near 1/5.)

    I suspect you’re right that the effect here is not very meaningful, but you can’t use Beta to guess the percent of BMI variation due to chocolate-eating frequency.

  • Perhaps the fatter people underestimated their chocolate consumption?

    • Hi Nadine – Good point! I did a Radio 5 live interview on this on Tuesday and the doctor in the studio made the same point. I am quite happy to confess to my daily chocolate consumption, but then, with a BMI of 20-21, I am unlikely to be judged as fat and greedy for eating a bar of 85-90% daily. Would I be so keen to do this if I had a BMI of 29? Probably not. Overweight people already say they feel guilty and uncomfortable eating in public, as they feel that they are being judged. Would this impact declarations of consumption – quite probably.
      Nice one!
      Very best wishes – Zoe

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.