Thursday, July 5, 2012

Shangri La Experiment Prior Beliefs and Biases


As a good Bayesian, I should work out what my priors and predictions are. 

As a good sceptic, I should work out what I want to be true so that I can be suspicious of it.

Models

Shangri-La: Appetite will collapse. Weight will drop significantly. Over a month belt notch will move to 3 or 4.

Willpower: Appetite won't change. Weight will increase. Belt will move to last notch and probably become unwearable in the evening.

Helplessness: Appetite will decline very slightly to compensate for extra calories (probably unnoticeably). Weight will stay pretty much the same. Belt notch will stay as it is.

Call these S,W,H for short

Current beliefs

S: fad diet. lots of people think it works, but anecdote only. You get the same evidence for chiropractic, aromatherapy, homeopathy, prayer, acupuncture, etc, etc.

No rational reason that this should be true except that it's got a story that sounds good to me.

Give it a 1% chance. There's a stack of crap in the world, and this is just some that happens to appeal to me.

W: Very much the standard model. It's obviously and uncontroversially true that weight loss = calories in - calories out. The bit I don't like is that 'eat less and exercise more' doesn't seem to be good advice. The world is full of skinny people eating whatever they like and hopeless fatties trying to live on lettuce and rice and making themselves completely miserable. It's probably true that if you eat less and exercise more you'll lose weight, I just don't think it's humanly possible to defy your basic drives by exercise of willpower.

Nevertheless, this is what most people believe, and that gives it respect. I'll give it a half chance of being the truth. (i.e. I notice my confusion and make no prediction)

I also notice in passing that I dislike this model because it means that all the smug types who think that fatties deserve it and that anorexics should just stop starving themselves to death for no reason are right.

And I also note that I dislike the 'soft sciences' and medicine because even though their subject is very hard they pretend to scientific infallibility as if they were physicists. So I love it when they're wrong even though I'd like them to find out the truth (and I accept that they are trying!)

So I'm really badly biased against this model, and I should watch myself for evidence that I'm frigging my own experiment in order to attack it.


H: I hate this model but it looks more true. It says that your weight has a set point that you can do almost nothing about. Fatties stay fat, the scrawny stay feeble. By immense exercise of will you can change yourself slightly but once you stop with the willpower, you'll go back to normal.

This is what I think is true and I've got personal and anecdotal evidence that it is. I'll give it a prior of 90% truth. I'd be surprised if it was wrong, but not nearly surprised enough to disbelieve the evidence.

I notice that I'm baised towards this model because it's mine and I've advocated it publically. I'm also biased against it because I don't like helplessness. I'd prefer either of the others to be the truth.

Immediately I notice that my priors add up to 141% (so what the hell is my brain thinking).

I'll assume that there are only three possibilities and normalize (roughly) to:

H60:W39:S1

Irrelevances

I also note that I believe the following things, and that they make no sense in terms of model W, but that none of them are relevant to this test so I'm ignoring them:

Smoking moves your set point downwards. Unnaturally fast carbohydrate foods (rice, wheat, sugar, potatoes, etc) screw with the system and make you more hungry than you should be, leading to steady weight gain and eventually morbid obesity. The Lord alone knows what is going on with anorexia but it's probably something to do with long term voluntary starvation somehow moving your set point unhealthily low.

No comments:

Post a Comment

Followers