Cluster sampling, in which researchers interview families from a few representative segments of society and then extrapolate out to arrive at an estimate, is how most surveys are conducted. It's how exit polls are run. It's also the method by which we've come to the figure 400,000 for the number of people killed in Darfur, a measurement that has allowed the media to call what is happening there a genocide.
In this study, 50 clusters were randomly selected from 16 of Iraq's Governorates, with every cluster consisting of 40 households. And out of the 1,849 households, there were 1,474 births and 629 deaths in the 40 months post-invasion. Whereas the mortality rate before the war was 5.5 per 1,000 people per year, since March 2003 it has been 13.3 per 1,000 people per year. Multiply that out for the whole country and they arrive at an average of 654,965 for the whole of Iraq, with a margin of error between 426,369 and 793,663.
There are two reasons for thinking the survey might be more accurate than has been portrayed, both of which were not mentioned much yesterday. First, the researchers were able to duplicate, with different households, the results of a survey they conducted two years ago (which was also widely disputed) that put the death toll then at 100,000. And secondly, the pre-invasion mortality rate of 5.5 per 1,000 people per year, found in both surveys, is similar to the estimate used by the CIA and the U.S. Census Bureau.
...
There seems just as much reason to trust these numbers as to doubt them. It should also be remembered that though the White House or Iraqi authorities may be unhappy with this news, it is not as if they are offering alternative figures. As the New York Times reminded us in its article, the Iraqi government recently barred the central morgue in Baghdad and the Health Ministry from releasing information about civilian deaths. The American military won't even give numbers, preferring to stick to percentage comparisons, citing, for example, a 46 percent drop in the murder rate in Baghdad in August from July as proof that the recent crackdown has been effective.
http://www.cjrdaily.org/politics/debating_the_body_count_in_ira.php