Estimating excess deaths in Iraq since the US–British-led invasionNo-one knows how many Iraqis have died as a result of the invasion of their country. Surveys published in The Lancet in 2004 and 2006 estimated the number at 10 times figures previously given. Predictably, they raised a storm of protest. Politicians attacked the surveys; public media gave their own spin. Here Scott Zeger and Elizabeth Johnson, statistical analysts of the surveys, discuss these and other attempts to count the dead, and how statistical messages should be best communicated. Targets of inferenceEstimation to speculationTh ere are many targets for inference—things we would like to know—when attempting to quantify the loss of life attributable to the invasion. Th ey can usefully be ar-ranged along a continuum from estimation to extrapo-lation to speculation. At the extrapolation–speculation end is inference about the counterfactual quantity: how many more Iraqis have died since the Iraq invasion than would have died had the invasion not been launched. Th is “invasion eff ect” has two components: the number that have died since the invasion and the number who would have died otherwise. Inference about this latter quantity is, at a minimum, an extrapolation since there is no direct evidence about deaths absent the invasion, i.e., those that would have occurred had the invasion not happened. An original, if now discredited, US justifi cation for the invasion was to pre-empt develop-ment of Iraqi nuclear weapons. Had they been devel-oped and used, the retribution on the Iraq population In the predawn hours of Th ursday, March 20th, 2003, the USA and the UK led a coalition of forces that began an invasion of Iraq with a bombardment of Baghdad. 4 years on and roughly 150 000 American and British sol-diers continue the war. Th e toll on human life has been substantial. For example, CNN reported that, between March 20th, 2003, and June 6th, 2006: “Th ere have been 2699 coalition deaths, 2475 Ameri-cans, two Australians, 113 Britons, 13 Bulgarians, three Danes, two Dutch, two Estonians, one Fijian, one Hun-garian, 30 Italians, one Kazakh, one Latvian, 17 Poles, two Romanians, two Salvadoran, three Slovaks, 11 Spaniards, two Th ai and 18 Ukrainians…”(http://www.cnn.com/SPECIALS/2003/iraq/forces/casualties/index.htmlTh e precision of the military fatality counts is no-table. In contrast, we know much less about how many Iraqis have died as the result of the invasion. Th e objective of this article is to discuss recent sur-veys to quantify the excess deaths caused by the invasion and a variety of statistical issues related to estimating and communicating about this quantity.There have been 2699 coalition deaths. The precision of military fatality counts is notable. In contrast, nobody knows how many Iraqis have died as a result of the invasion.54june20074(2)_02 Zeger&Johnson_IraqiDeaths.indd 54 10/05/2007 11:48:2155june20074(2)_02 Zeger&Johnson_IraqiDeaths.indd 55 10/05/2007 11:48:5856june2007might have been substantial. If one argues that these deaths were avoided by the invasion, they would be counted as lives saved in a counter-factual estimator of Iraqi deaths caused by the invasion. Because this “estimate” of the excess deaths depends critically upon unverifi able as-sumptions, even speculations, about the coun-terfactual world, we might more accurately call it an “extrapolate” or “speculate”. On the other hand, the fi rst quantity in the “invasion eff ect”, the number of deaths that have actually occurred, would be directly observable (with some degree of error) if the Iraqi population was covered by a functioning, accurate death registry. Since the death registry is unlikely to be accurate, the total number of deaths that have occurred must be estimated. However, this is a standard problem of infer-ence. So too is estimating the diff erence in the rates or numbers of deaths for periods before and after the invasion, as has been done by some studies. Iraq mortality surveys since the invasionTotal mortality in IraqTh ere are several independent estimates of to-tal mortality in Iraq, including two published by branches of the US government and one from the United Nations. Th e World Factbook, published by the Central Intelligence Agen-cy (https://www.cia.gov/cia/publications/factbook/geos/iz.html), reports a 2006 estimate of the crude death rate (total deaths in a year divided by total population) as 5.38 deaths per 1000 people per year on a population of 26.78 mil-lion. Th is corresponds to roughly 144 000 deaths per year. Th e US State Department publishes its own international demographic data in the Population Reference Bureau’s 2006 world data sheet (http://www.prb.org/pdf06/06WorldDataSheet.pdf). In 2006, it reported a crude mortality rate of 10 per 1000 per year on a population of 29.6 million for total deaths of 296 000, more than twice the CIA estimate. Th e United Nations Population Fund (http://www.unfpa.org/profi le/compare.cfm) reports a population of 25.8 million and a mortality rate of 8.8 per 1000 per year for a total of 227 000 deaths per year. None of these reports provide methods of estimation, uncertainties or detailed informa-tion about the sources of data on which they are based. Unfortunately, they diff er by a factor of 2.Deaths caused by the invasionTh ree independent groups have provided em-pirically-based estimates of the number of ex-cess deaths attributable to the US–British-led invasion, while also presenting the methods used. Each is briefl y reviewed in turn.Iraq body countTh e Iraq body count (IBC) maintains a run-ning total of “casualties” that are “derived from a comprehensive survey of online media re-ports from recognized sources” (http://en.wikipedia.org/wiki/Iraq_Body_Count_project#Method). Th ey report individual violent deaths that total approximately 60 000 civilians from the inva-sion in March 2003 to present, or about 1000 violent deaths per month. Th e IBC is explicit that theirs is a partial list of violent deaths, i.e., the ones covered in the media or handled by the major morgues. Assuming a pre-invasion baseline rate of 5 deaths per 1000 per year for a total of 150 000 deaths per year by non-vio-lent causes, the IBC partial list represents an increase in the crude mortality by a factor 1.08 or about 8%.Iraq living conditions surveyBetween March and August 2004, the United Nations