Blue Tit Nest

Last Friday afternoon, I was crossing the bridge over the dual carriageway near Farnborough North station, when I saw noticed a blue tit disappear into a gap in the tubular metal hand-rail.  After a few seconds it came out again and flew off.  then another blue tit with some grass in its beak flew in and disappeared into the hand-rail.  I would have thought that this was a rather exposed place for them to build a nest, but not many people walk over that bridge, so they might succeed.

Excuse

I was late home today.  I wandered into a bookshop and had to read my way out.

The Skin Cream Testing Paradox

Over at Bad Science, in a discussion of the recent BBC Horizon program on cosmetics, Ben Goldacre describes what I shall call the Skin Cream Testing Paradox:

But this is the astonishing thing. Other manufacturers don’t submit their creams to a university dermatology lab, or publish academic papers: so these aren’t so much the best results, they’re the only ones. Why? The tests cost just £15,000. Cellular Radiance Cream by La Prairie, for example, costs £340 a pot. Fifty pots would fund one study.

But those tests could bite any company right back: because if they show no effect, then your business is trash; but if they do show an effect, then busybodies wade in to regulate your pharmaceutically-active product. If it was my cosmetics company, I’d stick with the sciencey diagrams and hope for the best.

That is, if testing shows the creams to be effective then the manufacturers are subject to costly regulations, and if they are shown to be ineffective then the manufacturers have to admit that they are conning their customers.  Thus the manufacturers have no incentive to fund the testing of the efficacy of their own products.  They chose to keep their heads firmly in the sand.

However, what has not yet been pointed out is that each manufacturer has a strong incentive to fund tests on its competitor's products.  I wonder if there is a simple way to trigger an outbreak of such competitive cross-testing?

Corruption and Parking Tickets

Statistics can be a wonderfully subversive tool.  Take this discussion by Andrew Gelman of a paper by Ray Fisman and Ted Miguel of the correlation between unpaid parking tickets of United Nations delegates and the level of corruption in their countries.  I think someone should publish league tables!

Building Large Systems in C (and C++)

From Steve Yegge's old Amazon blog:

If you're trying to build a large system in C, you'll inevitably wind up making a choice between two alternatives:

1. Use the AlternateHardAndSoftLayers design pattern, in which you write most of your code in a high-level language, and the performance-critical parts in optimized C.
2. Try to make C a high-level language itself, which either leads to the madness of choosing C++, or the madness of implementing your own high-level language in C, without any actual syntactic support.

Most companies choose the second option, usually because they haven't learned the hard way that 90% of your code never needs to be optimized, and doing so won't produce any human-observable difference in runtime performance.

Indeed. The use of C++ for the whole of a large system can be seen as a 'premature optimization' in the sense of Donald Knuth.