And many scientists -- including four experts who anonymously peer-reviewed an article for the prestigious medical journal, The Lancet -- insist that the 654,000 figure, a measurement of what demographers call "excess mortality," is derived by a scientifically valid methodology from a statistically valid sample.
The U.S. military's estimates, buried in a little-noticed recent report to Congress, are drawn from a daily tabulation of "significant activity reports," about "incidents observed by or reported to U.S. forces," known as the SIGACT database. do not distinguish deaths from injuries, nor between Iraqi civilians and members of the army, police or other government security forces.
The estimates "are derived from unverified initial reports submitted by Coalition elements responding to an incident; the inconclusivity of these numbers constrains them to be used for comparative purposes only," says the report, titled "Measuring Stability and Security in Iraq."
But the comparisons they enable show that average casualty rates for Iraqis have sky-rocketed from just over 20 per day in the first quarter of 2004, to nearly 120 per day between May and August of 2006.
The 654,000 estimate of deaths caused by the invasion and resultant insurgency was made by a team of public health statisticians based at Johns Hopkins University. The figure is actually the midpoint in a statistical projection of "excess" deaths -- the number of people who have died as a result of increased mortality rates since March 2003.
The projection says that number of excess deaths can be said, with 95 percent confidence, to range between 393,000 and 943,000. It said that 92 percent of the excess deaths were from violence, mainly gunfire, but also explosions. In those instances where respondents allocated
By way of comparison, Human Rights Watch has estimated Saddam Hussein's regime killed 250,000 to 290,000 people over 20 years.
Researchers from the al-Mustansiriya University in Baghdad gathered information on deaths from more than 1800 households in 47 clusters scattered across all but two of Iraq's 18 provinces. Clusters in the provinces of Duhuk and Muthanna were discarded from the study because of problems with the data.
The figures they gathered showed that the mortality rate before the U.S. invasion in March 2003 was 5.5 per 1,000. By July 2006, it was 13.3 per thousand.
"I don't consider it a credible report," said President Bush, the day it was published, "Neither do Iraqi officials."
"This is the standard way of measuring death rates in developing nations," retorts one of the study's authors, Les Roberts, who has left Johns Hopkins for Columbia University in New York since it was completed. "The U.S. government had no problem with this exact same methodology when it was used to estimate death rates in the Kosovo conflict."
But some statisticians and other scientists have also criticized the study, saying that while the statistical methodology may be sound, there are problems with some of the ways the underlying data was collected and/or analyzed.
One of the critics is Scott Gilbreath, who works for the Yukon State Bureau of Statistics in Canada, and blogs as "StatGuy," a "perpetually perplexed Christian statistician."
Gilbreath writes that the methodological account of the survey leads him to believe that the clusters were in urban areas, where the death rates are likely to have been much higher. Moreover, within each cluster, the households visited were selected at random, but changes could be made to the randomly generated targets, if the surveyors felt the security situation demanded it.
"The article doesn't say how often the interview team exercised its discretion to change to an alternate location," writes Gilbreath, adding "To me, that is a serious omission ... the article's description of survey field operations is, in the absence of further supporting documentation, highly problematic."
Roberts says that Iraq is a highly urbanized country, but adds that the clusters were allocated to provinces, and to the administrative units within them, in exact proportion to the distribution of population. "Some of the clusters were in very rural areas," he said, "A rural person and an urban person had exactly the same chance of being surveyed."
Steven Moore, a Republican political consultant who has trained public opinion researchers in Iraq says that 47 clusters is a very low number, especially in a country the size of Iraq.
Moore says that the United Nations Development Program used 2,200 cluster points of 10 interviews each for a total sample size of 21,688, in its 2004 survey of Iraq.
"What happens when you don't use enough cluster points in a survey?" asks Moore, "You get crazy results."
Roberts retorted that 30-cluster surveys are regularly used for countries large and small by U.S. and U.N.-funded efforts to determine mortality in conflict zones in developing countries. "This is the standard way to do this," he said.
Finally, the Iraq Body Count, a non-profit that tallies media reports of casualties, points out that the survey says death certificates were produced in more than 90 percent of cases, which would mean that about 550,000 had been issued for violent deaths since March 2003. But the Iraq Ministry of Health's own figures say that only about 50,000 such certificates have been issued.
"If the Lancet estimate is correct," says the group in a statement, "then it follows that ... 500,000 documented violent deaths, for which certificates were issued, have somehow managed to completely disappear without a trace to Iraqi officials or the international media."
Roberts says that is exactly what happened. "If you look at non-violent deaths, there have been at least 400,000 since the invasion... Yet the health ministry has counted only about (40,000) certificates." He said most certificates were written by local doctors or health officials, but lacked any kind of "paper trail back to Baghdad."