Publishing in the journal Obesity Science & Practice, researchers have analysed a specific nutrition survey and report that consumption of fast food or soft drinks was not correlated with body mass index (BMI).
Prof. Kevin McConway, Professor of Applied Statistics, The Open University, said:
“It’s an interesting paper, but there are aspects of the statistical analysis, and the conclusions from it, that I just don’t understand.
“Many of the conclusions seem to be based on the last two columns of Table 1, giving what are called p-values, for analyses omitting the most extreme people (underweight, or morbidly very obese – second last column), and for analyses including everyone (last column). My first concern is that the paper’s title says (among other things) that candy intake is unrelated to body mass index (BMI) for 95% of American adults, that is, there’s no relationship if we leave out the 5% of people with the most extreme BMI values (high or low). But the main information on candy eating seems to be in the row of the table called “Sweet snacks”. In this row, the two p-values are given as <0.001 for both analyses. Such a (very small) p-value is, in the jargon, statistically highly significant – that is, it represents a relationship that is too strong to be attributed to chance variation. But that’s the case whether or not the two extreme BMI groups are omitted. I can’t see any support therefore for the assertion that candy intake is not related to BMI if the 5% of people with extreme BMI are left out. Perhaps the authors are basing this conclusion on the correlations they report elsewhere in the paper, but a puzzle here is that they say the correlations are negative but the actual numbers they report are positive. That could simply be a typo, but even if it is, the correlations are reportedly still statistically significant, which is again not in line with the overall assertion that candy intake isn’t related to BMI.
“There’s something else strange about these p-values too. A p-value is a kind of probability, and probabilities have to lie between 0 and 1, so, for example, they can’t be negative. But many of the p-values given in Table 1 are negative. I have no idea what that means, and the paper doesn’t explain.
“The paper does raise some intriguing questions, particularly the observation (shown in Figure 1) that people who are underweight, as well as those who are extremely obese, eat more fast food meals per week than do the great majority of people in between. I’d like to know why that is, and whether it’s the case in other countries too. But I’d need to learn more about the details of the statistical analysis in order to take more notice of this study’s other detailed findings.”
Ms Catherine Collins, Principal Dietitian at St George’s Hospital NHS Trust, said:
“This is an interesting study from food psychologist Brian Wansink and colleague, whose work from Cornell Food and Brand lab has helped shape the ‘nudge nutrition’ approach to food choice and selection. On two separate occasions, healthy people across a wide range of body weights were questioned about their diet in the previous 24 hours.
“The results showed obese people ate high calorie foods a similar number of times each week compared to slimmer subjects, and so the authors call into doubt the usual advice to cut out takeaway meals, sugary drinks and confectionary as not having much effect on weight loss.
“However, the study ignores two cardinal rules of dietary assessment – accuracy of reporting, and portion sizes. As a dietitian I know that it’s common for people to under-report their actual intake especially as body weight increases, so frequency of consumption might not reflect actual intake. Secondly, portion sizes are highly variable. Sugary drinks range from 200ml to 2 litre bottles, and a takeaway meal may be a modest calorie chicken shish kebab, or a diet-busting doner kebab with fried onions and fries on the side. Chocolate can range from a 45g (240 kcal) ‘single serve’ bar, to a 200g (over 1000kcals) large block – but using the same score as the authors used this would count as ‘one serving’.
“To be obese requires a surfeit of calories in relation to body needs, and calorie intake is influenced by food choice, frequency of eating, and food portion sizes. The authors considered the first two, but not the third. The important take-home message from this? It’s better to downsize than to dismiss particular foods from the diet – unless it’s possible to swap to a zero calorie option, such as diet instead of standard drinks.”
‘Fast food, soft drink and candy intake is unrelated to body mass index for 95% of American adults’ by David R. Just and Brian Wansink was published in Obesity Science & Practice on Thursday 5 November 2015.
Prof. Kevin McConway: “I have no relevant conflicts of interest.”
Ms Catherine Collins declares that she has no relevant interests.