Thursday, March 25, 2010

Friday quick links

1. Most findings of statistical research are false, and can be easily demonstrated to be so.  If I haven't convinced you to scrutinise statistics carefully, then this may. Warning: the linked paper is a little nerdy and mathematical.

2. Is prescribing a placebo a good idea?

3. One laptop per child and a computer on every student's desk - some evidence that computers help children learn computer skills, but detract from their learning of other more basic skills such as maths and English. 

4. My interest rate bet looks shaky - straight from the horse's mouth.

5. Moral self-licensing is when doing something good in one part of your life helps you justify doing something bad in another part.  This 'green' consumer experiment is a classic - shoppers, however, earned on average 36¢ more, showing that they had lied to boost their income.

I must say that in moments of raw self-reflection I can see myself issuing a subconscious (sometimes conscious) moral licence.  'I've been good for a while, now I can justifiably do something bad" 

Maybe it has something to do with our upbringing.  I know that I often reward my son with otherwise 'bad' foods (he loves Jatz crackers) when he has behaved well.  It would be nice to conduct a cross-cultural comparison on this topic. 

It is also a example of actively reverting to the mean.  People think they are at the extremes of socially normal behaviour, so they do something that is at the other end of the spectrum to keep themselves in line with others.

1 comment:

  1. Ah, the Ioannidis paper.

    I think he's being a bit cheeky with his title, because what he's claiming is trivially true.

    However, I think most that most scientists worthy of the name would never make the mistake of assuming that the results of one study was enough to make a completely robust statement WRT to falsification of a hypothesis.

    Even a P-value of 0.001 means there's still a 1-in-1000 chance of a Type 1 error, which although, unlikely, is not impossibly so...