When surveyors rigorously select individual addresses in advance (by GPS co-ordinates for example), field teams are in a strong position to explain why this house and not the one next door was selected. But when the criteria for selecting houses is itself open to interpretation, and selection is being carried out on the spot, with residents looking on, a survey team must be put in a nearly impossible position.

The Lancet authors have said that their surveyors selected the 40 houses by picking one house and then moving to what has been described as the “adjacent” or “nearest” dwelling. But one dwelling will tend to have several other dwellings close by. Thus there will always be a range of choices for which dwelling to pick next. It is entirely unclear from the Lancet report or any subsequent commentary what specific protocols existed to handle these choices. There is also no mention of any protocol to deal with the almost inevitable disappointments and outrage among people with genuine tragedies who were informed beforehand of the teams’ intention to do a good deed in recording their losses, but were then refused inclusion. Some people do not take such affronts lightly, especially in the suspicious and volatile Iraq of July 2006.

The one solution which would have offered itself to the survey team is, of course, not to refuse people who had such experiences to relate, regardless of whether or not they were technically qualified for inclusion.

That difficulty, we contend, may well partly explain what looks to be a massive over-estimate.

Finally, it should be noted that the only cluster in the first Lancet survey which reported such an enormously high number of deaths that it had to be discarded as an obvious outlier, was also the only cluster that used a field-team judgement-reliant method of selecting interviewed homes instead of GPS units and a grid. This should have been seen as a warning of the pitfalls of this approach, but appears to have been taken as an endorsement of it.