Wednesday 13 February 2008

Cod equations again

From the BBC: How to make better decisions by Garth Sundem. This is presumably to hype yesterday's accompanying Horizon program (and Sundem's Geek Logik book - he has a couple of blogs too, here and here. If Sundem's BBC article is anything to go by, there's nothing to see , folks.

At your local pub, you have many beers to choose from. Which is best? If you are like most human beings, the answer is "the most expensive one."

Bollocks. Here's a sample beer tariff from a menu I did a while back: Carling £2.30 / 4X £2.10 / Guinness £2.60 / Stella £2.60 / Boddingtons £2.30 / Otter £2.40 / London Pride £2.40 / Abbot £2.40. By Sundem's theory, I should think Guiness or Stella to be best. No. I don't like stout, and generally think London Pride or Otter to be best because of the flavour, but I sometimes drink the least expensive, Carling, on hot summer days. It's simplifying to the point of nonsense to reduce a complex decision of taste, expense, time of year, and abv (e.g. one might like a strong beer for a single pint, or prefer a weaker and less expensive "session beer" if out for an evening) to "most expensive".

As described by the Heisenberg Uncertainty Principle, by introducing tools to measure a situation

Nooo. This is a classic handwaving misapplication of the Uncertainty Principle, which is a precise mathematical relation between specific measurements at subatomic level. It is not a generic rule for macroscopic phenomena.

To add a layer of mathematics, if one factor on the list is more important than the others, we might multiply it by two. If it is very important, we could even square or cube it.

Again, handwaving. Even if an equation can be constructed, it is meaningless as an accurate description of the phenomenon modelled unless the dependence on different variables is actually measured and the precise power law investigated. Flinging an ad hoc multiple or power at a variable won't do that. This is even assuming that subjective variables such as R = amount of current relationship are meaningful quantities.
      Unlike most of the PR and news story creators of cod equations, Gareth Sundem has sufficient mathematical nous to make equations that are mathematically well-formed and create plausible output. But they are still oversimplifications with no real justification for the variables chosen or the coefficients/powers applied to them. For example:



is supposed to describe a man's chances of success with approaching a woman. On what basis are the variables chosen? (W=Wittiness, G= Aggression, A = attractiveness, R = "amount" of relationship). Why the different coefficients 2 and 3 on male and female attractiveness? Why square R and divide it by 20? Not to mention the appallingly sexist assumption that unattractive women will be "easy".
      In short, it's a bit of psychological candyfloss framed as mathematics. No harm in that; but the BBC shouldn't be presenting it as a serious scientific approach to decision-making.

1 comment:

jdc324 said...

Good stuff Ray. Via your first link, I clicked on the "should I apologise" link. They have drop-down bars giving you options from 1-10. If D=1 (how big a deal it is) and Ra=1 (actual responsibility) and P=1 (how pissed off the other person is), but Ra=10 then A=11 and I should apologise. (For something that: isn't a big deal; isn't my fault; and hasn't pissed the other person off).