cartographie des affaires de corruption

Sweetener, dash, kick-back, pot-de-vin, brown envelope, whenever you want something done you can find someone who will expedite your case if you grease the wheels of the bureaucracy. I'm not sure if I have actual readers in France, rather than a particularly active [deranged] scan-bot, but my page-views from that country currently out-number the rest of the world. Alors, mes gars, I thought you'd be interested in this graphical display of cartographie des affaires de corruption in the sea-green incorruptible [a reference to Robespierre the revolutionary with rectitude] Republic. Because you never know when you might need a little help.  According to the perception of corruption survey 2016, France is up 2 places to #23 nudging out Chile but still behind Estonia. Despite my pal Pepe's proud assertion Spain is less corrupt than Syria or Sudan.

You're out of luck if you live in, or are thinking of moving to, Ardennes [77], Cantal [97], Cher [73], Creuse [100], Gers [90], Loir-et-Cher [71], Nièvre [87], or  Sarthe [46]. These departments have 0 [nul nix nada zonders] cases of corruption. The numbers [N] are the position in the list of the 101 départements when sorted by order of population.  You will note that only one of these corruption-free zones is in the top half of the country as regards number of people. Sarthe is located between Paris and Brittany. Its biggest centres are Le Mans, La Flèche, and Sablé-sur-Sarthe: it's okay if you've not heard of the other two because they muster rather less that 30,000 inhabitants [all scrupulously honest] between them.  The first case cited in the Big Think study is that of Roland Dumas who ripped off the family of Giacometti - but I scooped that story earlier this month.

I'm currently reading Thinking, Fast and Slow by Daniel Kahneman which looks under the stones to find the basis of bias: some of the case-studies expose the fallability of our minds but some are just errors due to statistics: such as the Law of Small Numbers. If corruption - or whatever you're having yourself - is rare then you're unlikely to find any cases in a small population. If you're formulating policy, it is clearly more sensible to report the thing you're interested in as a rate [count/population] rather than a raw count. But even if you do that, the small number problem doesn't go away; indeed it tends to get magnified because of the rule of extreme events.

If you have a friend you can carry out an experiment to show this:

1. get an urn
2. throw out your grand-father's ashes
3. put 100 red and 100 white marbles in the urn
4. you draw out 4 marbles and note their colours
5. your oppo draws out 7 marbles and note their colours 
6. both of you note and count the cases where you get all one colour

Q1. Who will get more extreme [all red or all white or NO cases of corruption] events?
Bonus Q: How much more?

Turns out that the 4 marble case will have 8 times ! the number of all-same events compared to the 7 marble draw. You do the math after you have recovered from your surprise. One example cited by Kahneman is the hot spots of kidney cancer rate in US counties. Educated epidemiologists were concerned and surprised when it transpired that cancer rates were disproportionately high in small rural counties [example for most dangerous counties see R from] but they manfully found just-so explanations about exposure to pesticides because farmers are thick as pig-dribble and won't wear protective clothing. Until someone flipped the data and showed that small rural counties also headed the pack for areas which had the lowest rates of kidney cancer; which, of course, you can explain with a story about fresh vegetables, clean water and attendance at church.

Except that there is nothing to explain because the rate is going to fluctuate wildly in small communities.  These data are just data until you start to sink money into some cunning plan on the basis of the numbers. As did the Gates Foundation on foot of a finding that small school were disproportionately represented in the top 'successful' schools; they invested $1,7 billion in dismembering mega-schools into smaller units until someone pointed out that small schools were also packing the opposite tail of the distribution. One smart kid, from a class of 20 high-school graduates, making it to Harvard will rocket her school into the elite category. Next year the school's success will have rapidly regressed to the mean.

Interesting [with maps!] investigation of the Law of Small Numbers using Canadian and US data and suggesting that Bayesian [bloboprev] methods can be used to calculate 'surprisingness' as a second order function from the raw counts. I have to work really hard to get my head round conditional probability and Saturday is my Sofa-time, so I'll leave the heavy lifting to you cher lecteur.