Social Science Statistics Blog

« More on Cheating | Main | Replication is hard... »

8 March 2007

Is there a "best diet"?

Janet Rosenbaum (guest blogger)

The public often reports being confused by contradictory diet studies, and there is some effort to find the "best diet", but is that the right question to be asking? A study released today in JAMA compared four common diets in 311 overweight/obese women over a 12 month period.

Most weight loss occurred within the first two months, with no visible change for three of the four diets between months 2 and 6. Given the amount of weight that these women could lose, some have commented that the effect sizes seem to be
fairly small.

While I'm of course a fan of randomized controlled trials, I'm not sure that an RCT is answering the most salient question. An RCT is answering the question of how much weight will people lose on average on each diet. While understanding average behavior may have implications for our understanding of human biology, in practice the most important question for an overweight person and their health care provider is which diet will be best for them, given their assessment for why they are overweight, which diets have worked for them in the past, and their personal tastes.

People may differ substantially across these factors. Someone who eats 100 calories too much at every meal may need to employ different strategies than someone who eats a 500 calorie snack every other day, even though they have the same calorie surplus. Likewise, someone with a tendency to eat too much of a given food category needs to know whether moderation or total abstinence is the best long term strategy. My sense of the research is that there is quite a lot of psychological research on strategies for good short term outcomes, but no RCTs focus on the medical questions of long term outcomes.

Weight loss plans employ different strategies --- for instance, Weight Watchers tries for moderation, while Atkins advocates abstinence --- but studying the individual plans confounds the question of which strategies are best with other characteristics across which the plans differ, and it averages effects over groups of individuals with heterogeneous reasons for their overweight.

It seems to me that weight loss research needs to determine if there are in fact distinct groups of overweight, and focus studies more narrowly on these groups.

Studying more homogeneous groups on a more limited set of questions would answer the questions that are most relevant for clinicians and individuals, although it would be more expensive.

Posted by Mike Kellermann at March 8, 2007 11:53 AM