Brendan O'Neill's comments about sampling of Iraqis for the ORI survey show a profound misunderstanding about probability sampling. The requirements for a random sample are that each individual have a known, non-zero chance of selection. The estimate, to be unbiased, must use the reciprocal of these probabilities as weights. Anything beyond those two principles in the design of the sample and the estimation is to lower the sampling error. That is what demographic weighting is supposed to accomplish. Unfortunately it also can increase the sampling error under the wrong circumstances.
The point here is that one need not know the demographic characteristics of the country, as O'Neill seems to think, in order to present a reliable set of sample data. Given his questions about the quality of the last Iraqi census it would seem better not to use census characteristics for weighting.
The ORI approach, from what I can tell, seems reasonable and certainly much more defensible than the Zogby approach a few months ago.
warren mitofsky