Sunday, September 10, 2006

What to eat, who to marry!

Tim Harford writes on about about the problem of deciding what to eat from a buffet:

If the buffet offers you every choice simultaneously, your best strategy is to try a little of every plausible dish so that you can decide what you would really like to eat. Then go back and get properly stuck in: to your favourite dish if you have no taste for variety, otherwise to your favourite two or three.

If the dishes are presented sequentially, then you will have to take more risks. There is always the chance that you will take a too-small portion of what later turns out to have been much the best course.

This is similar to the problem of deciding who to marry. And there is a solution from Group Theory that says if you would test a sufficiently large (but finite) population, test the first 37% and then settle on the one that is better than anything you've seen before.

Why am I blogging about it from BlogCamp? The same problem: which platform to settle for the blog, or the mobile phone or, the person to chat with at lunch in a little while.

Tags: , ,

PS. I didn't read Mr Harford's full argument because it is available only for pay.


Nanyaar? said...

Seems Interesting can you please be kind enough to give me the link?


Sunil Bajpai said...

Can I find a link to this rule? Would try and locate.

But .37 is approximately 1/e, where e = base of natural logs. What the rules says is that you select the best of first 37% samples as a standard to beat in the remaining sample population. Any less would give reduce the chance of finding a good representative to aim for. Any more would diminish the probability of beating the selected standard. I'm afraid this explanation would have to do till either you or I can find the url for a rigorous proof.