En lisant le gros papier de Tim Harford dans le FT de ce WE, un passage me sidère un peu. Il décrit une expérience réalisée en Afrique du Sud par Dean Karlan, unprof de Yale:

In South Africa, in autumn 2004, Karlan and Zinman persuaded an anonymous consumer finance company that we’ll call “ZaFinCo” to participate in an unusual experiment. Ordinarily, almost half of ZaFinCo’s borrowers would have been turned away as a bad credit risk. But for two months, ZaFinCo loan officers were instructed to identify marginal applicants who had narrowly failed to pass credit checks. From this pool of near-customers, a computer selected almost half and requested that the branch manager reconsider and offer a loan anyway. This procedure emulated the way that new medicines are tested, using randomised trials. These trials are a gold standard for evaluation: after all, a more typical, non-random comparison of borrowers versus non-borrowers would not be able to tell whether borrowers were doing well because they had access to loans, or because they were confident, risk-taking, entrepreneurial people. Karlan and Zinman wanted to know what value there might be in expanding access to credit. ZaFinCo was no dewy-eyed social business, but a hard-nosed, profit-minded company, charging 11.75 per cent per month on a four-month loan, or 200 per cent APR, much more than Compartamos was generally judged to have been charging. Despite the high rates, the results were astonishing. “We expected to see some good effects and some bad,” explained Karlan, who checked in with the experiment’s participants six to 12 months after they had filed their initial loan applications. “But we basically only saw good effects.”

Most strikingly, those “treated” by the experiment – that is, those for whom the computer requested a second chance at a loan – were much more likely to have kept their jobs than the control group. They were also much more likely not to have dropped below the poverty line, and were less likely to have gone hungry. All these outcomes were recorded well after the loan had been taken out and (usually) repaid, so this was not measuring a temporary debt-funded binge.

Donc ils auraient pu aussi bien supporter des taux d'intérêt de 500% ou 1000% ? J'ai du mal à comprendre les "good effects", sauf pour le prêteur à qui on rembourse. La question, qui n'est pas traitée, reste : ces prêts ont-il ou non enrichi les emprunteurs ? S'ils leur ont juste servi à garder leurs emplois salariés, ça veut dire que le emprunteurs mettront des années à rembourser ; ils sont liés par une nouvelle forme de "servitude de dette" dont Karlan le parle pas...

L'autre aspect eset celui de la méthode d'évaluation. Mon impression est qu'il y a une naïveté effarante dans ce type d'expérimentation (copiée, en effet, sur les comparaison placébo - médicament dans les études des nouvelles molécules) : dans les sciences sociales , l'observateur modifie l'observation : c'est un axiome de base. J'aimerais bien en savoir plus sur la méthode !

(Quant au groupe de clients comparé aux autres, on peut aussi rappeler aux "scientifiques" que le placebo a un effet ! Inférieur à celui du médoc, dans les meilleurs cas, mais c'est la définition même d'un effet placebo : il y a un effet ! !)