At Mike Dunford's science blog, I asked whether the streets/intersections method of selecting a sample in fact led to an urban bias in the John Hopkins mortality study. A Kevin Donoghue made the following response, which I pass along to you:
Pithlord, the clusters were assigned in such a way that, at the outset, as far as possible every household in Iraq had an equal probability of inclusion. Admittedly an isolated farm or village might have had no chance, but no survey really lives up to the textbook ideal of a random sample. I can't see that a household in a small town containing a hundred homes or so would have been any less likely to get included than a household in Baghdad.
If isolated farms and small groups of houses are very safe/(unsafe) that would bias the mortality estimate upward/(downward), by an amount which depends on the proportion of Iraq's population living in such places. AFAIK that proportion is very small.
I suspect Mike's main reason for de-emphasising that concern is he doesn't know which way the bias (if any) goes. Most likely nobody does.
Update Via Deltoid, here is an interesting take from UK Polling Report.