Sunday, November 18, 2012

The Shangri-La Diet Needs Our Help

A while ago, Eliezer Yudkowsky, a man whose thinking and writing I very greatly respect,
mentioned a crazy-sounding diet in a positive light. He said effectively: "It sounds like it should work, but it doesn't work for me." (

I decided that I'd try it out on myself. And to my surprise, it does seem to work.

It seems that Seth Roberts has found a hack into the process that controls how much we eat, and (he says) that it works for most people. I am reasonably confident that it works for me. There are a fair number of experiments on rats that show that the various mechanisms work in rats.

If so, then Dr Roberts has found a cure for one of the great scourges of life in the first world.

Obesity is a great problem for many people, and made worse by the incredible lack of sympathy shown by people who never have the slightest trouble with their weight.

But anyone who believes that anecdotal evidence is more than a suggestion of a good place to look, will assuredly end up going to aromatherapy sessions to cure whatever problems they have. And that may even work.

So it seems that we need to do a randomized controlled experiment to find out whether this hack works, and also from a less practical but more curiosity-satisfying point of view, whether it works for the reasons Dr Roberts says it does or not.

I have been swapping e-mails with Seth Roberts, and he reports that there is barely any interest in academic circles in testing his theory.

I quote with permission:
I disagree that a trial would be dead easy. Let me explain some of the difficulties. 1. Who would do it? Not any grad student: after you get a Ph.D. associated with a quack diet, who would hire you? The only people I could hire would know zero about this sort of research, which is very rare. 2. Where? 3. What comparison group? Nothing, placebo control, or best alternative treatment? 4. If placebo control what would it be? 5. Enormously difficult to measure compliance and equate compliance between the two groups. 6. There are maybe 10 labs in the whole world experienced in doing this sort of expt. Unclear why they would want to do this. I would need a lot of money to convince them.

This is clearly insane. There is a very great benefit to be had if this is true, so the fact that it looks a priori rather unlikely is no reason not to test it, if the test is cheap by comparison to the obesity crisis.

But I have also been talking to academic friends, and they find it all too plausible that what Dr Roberts says about nobody being interested in testing his theory is true.

From which we can conclude that academia is very badly broken.

So we need to do it ourselves.

I would like to come up with a way of testing this, that will not cost me very much money, and yet will provide some more reliable evidence than 'try it yourself and see if it works'.

But of course, I have absolutely no idea how to go about this.

My initial intuition is that we should make a vast number of little gelatin capsules, some containing 15g of extra-light olive oil, and some identical-looking ones containing something harmless with no calories in it.

And we should get around twenty volunteers to promise to take one first thing every morning, and not to eat or drink anything for an hour afterwards, and to record every day their weight at the time they got up, and whether they felt hungry or not during the day, and to keep this up for a couple of months.

We should send each volunteer a selection of these capsules, with a pattern like 'one week of olive oil, then one week of placebo, then one week of olive oil, then one week of placebo'.

And each volunteer should get a different pattern, which pattern is recorded secretly by me.

If Seth's theory is correct, then the secret patterns should be reflected back clearly in the data, with perhaps some sort of time-lag.

I have no idea whether this is a sensible sort of experiment to run, or how to analyse the data. I have just made all this up.

And I wonder if anyone else can make suggestions, or explain why this sort of experiment is so badly flawed that we will learn nothing useful from it.

One suggestion that has been made is that there should be a third group, or a third part of the random patterns, where no capsules are taken, but the same data is recorded.

That's probably a good idea, but of course it makes the trial half again as much work. So how do I tell whether the extra information from that is worth the extra effort?

Anyone know how to do this sort of thing?

1 comment:

  1. I've been occasionally reading your posts with interest - well done on explaining Bayesian math to kooky diet-lovers ;)

    For your randomised trial, I can see an issue : anything that feels like oil will probably contain calories, or be harmful, so people will quickly work out which is oil.

    I recall you mentioning the use of sugar-water as an alternative to oil. If this is possible, how about using a calorific (but non-sucrose) sugar vs a noncalorific synthetic sweetner? The reason for using non-sucrose is that people won't readily identify e.g. dextrose as "more natural" in taste than a synthetic sweetner.

    Of course you have the problem that in both cases your mind can associate the taste with calories (or lack thereof), so one possible mechanism for the effect is not being tested. However most of the other mechanics should be testable this way.

    Also make sure you have long enough runs with each pill type to see any impact - a couple of weeks sounds from your experience like not enough, perhaps two months at a time would be better?