UW STAT 220 - Estimating Mortality in Civil Conflicts

Unformatted text preview:

CoverIraq Mortality Ver 111 June, 2007 Triangulating different types of mortality data in Iraq Debarati Guha-SapirOlivier Degomme Estimating mortality in civil conflicts: lessons from Iraq Centre for Research on the Epidemiology of Disasters University of Louvain School of Public Health 30 Clos Chapelle aux Champs 1200 Brussels www.cred.be For paper copies,contact Regina Below ([email protected]) CREDCRED WORKING PAPER JUNE 2007 Estimating mortality in civil conflicts: lessons from Iraq 1.0 Introduction The civil and political conditions in Iraq have steadily degenerated since the military invasion of March 19th, 2003. Civilians are increasingly bearing the brunt of relentless violence. Extremely poor security conditions are disrupting the lives and livelihoods of millions and the end is not in sight. Recent studies on the human costs of war have focused on measuring deaths due to violence [1-3]. The debate around the numbers of excess deaths has opened a Pandora’s Box of methodological issues related to measuring mortality in conflicts, underlining the political sensitivities that accompany such exercises. The credibility of mortality estimates from conflict situations depends on three factors: - Political objectivity of the originators - Soundness of methodological foundations - Transparency of declared limitations In most studies of mass mortality, estimates are derived from multiple sources and are expressed in ranges [2, 4-6]. A single precise number gives a sense of accuracy that is deceptive in survey type sources. Recent estimates of war-related deaths in Iraq vary widely and have been the subject of much discussion (Table 1). But even before the invasion and the ensuing chaos, data from Iraq presented challenges. For instance, UNICEF’s Multiple Indicator Cluster Survey (MICS) 2000 and 2006 analyses [7, 8] shows a substantial reduction in child mortality rate (U5MR) between two contiguous pre-invasion 5-year periods, which requires further examination. Similarly, estimates from the post-invasion period range from 43,771 to 15 times that number. 1CRED WORKING PAPER JUNE 2007 Table 1: Mortality estimates from surveys in Iraq 1994-2006 period covered sample size (hh) U5MR (95%CI) CMR (95%CI) war-related deaths (95%CI) area covered source 1. 1994-1999 24,000 131 [127;135] - NA Iraq excl. Kurdish gov Ali & Shah (2000) 2. 1994-1999 16,000 72 [68;76] - NA Kurdish gov Ali & Shah (2000) 3. 1999-2003 21,668 40 - NA Iraq Iraq Living Conditions Survey (2004) 4. 2001-2006 18,144 41 - - Iraq Multiple Indicator Cluster Survey (2007) 5. 2002-2003 988 - 5 [3.7;6.3] NA Iraq Roberts et al (2004) 6. 2002-2003 1,849 - 5.5 [4.3;7.1] NA Iraq * Burnham et al (2006) 7. 2002-2004 21,668 - 23,743 [18,187; 29,299] Iraq Iraq Living Conditions Survey (2004) 8. 2003-2004 988 - 7.9 [5.6 ;10.2] 98,000 [8,000; 194,000] Iraq † Roberts et al (2004) (including violent and non-violent deaths) 9. 2003-2006 (June 30)1,849 - 7.2 [5.2 ;9.5] 601,027 [426,369; 793,663] Iraq * Burnham et al (2006) 10. 2003-2006(June 30) - - - 43,771 [41,441; 46,101] Iraq Iraq Body Count (2006) * excl. Dahuk, Muthanna † excl. Anbar We re-estimate mortality in Iraq using data from multiple sources based on their methods and coverage. Burnham [9] and IBC [10] provided raw data and were the most complete data sources on war-related mortality. We also used the report from the Iraq Livelihoods Condition Survey [11]. The strengths and weaknesses of each are described below. We triangulate the findings of these three data sources offsetting strengths against weaknesses. We also draw methodological lessons for future mortality studies in conflicts. 2.0 Summary analysis of the Burnham survey dataset The Burnham survey estimated deaths caused by the US led invasion using a national, cross-sectional mortality survey. The survey itself, undertaken between May – July 10th, 2006 covered 47 clusters and a recall period from January 1st, 2002 to June 30th, 2006. The sample estimates were extrapolated to the 2004 population of all governorates (UNDP Census). Thus each survey sample death represented about 2000 population-based deaths. The authors presented a point estimate of 654,965 deaths due to the 2003 invasion, an average of 594 civilian war casualties everyday over 3 years and 4 2CRED WORKING PAPER JUNE 2007 months. Of these, nearly 92% were attributable to war related violence based on a pre-invasion baseline of 2 violence related deaths. The study has received wide publicity in both academic as well as political circles, underlining as it did, the devastating effects of the war on civilian populations [12-14]. We checked the data for accuracy and validity of the numerator, denominator and sampling weights of governorates. The following errors and methodological weaknesses were identified. a) Validity of numerator: Eleven deaths were removed from the numerator because the households in which they occurred were excluded from the denominator due to missing household size. In addition, since the recall period ended on June 30th [9, 15], 24 car bomb deaths that occurred in cluster 33 in July 2006 were eliminated. b) Sampling distortions: Three governorates with the highest violence-related mortality rates were over sampled and 3 governorates with the lowest violence-related mortality rates were under sampled (Table 2). In order to correct this, we applied weights to each governorate according to its population to re-estimate nationwide mortality rates. Table 2: Population and sample distribution by governorate 3CRED WORKING PAPER JUNE 2007 c) Design Effects: We calculated the design effects (DEFF) for the overall and cause-specific violence-related mortality rates1. The DEFF for all violence-related deaths was 5.4 and for some specific causes, 9.6. This indicates that the violent death phenomenon was highly clustered over time or space, making it difficult to reliably generalise to the entire country. d) Validity of denominator: Recalculating household size from each record, we obtained 66 households with less than 1 member per household and 33 of those had a negative (<0) household size in 2002. More importantly, the authors did not take population migration into account for the denominator, the implications of which can be sizeable. A household member who enters early in the recall period (e.g. Feb. 2003) is at risk


View Full Document

UW STAT 220 - Estimating Mortality in Civil Conflicts

Download Estimating Mortality in Civil Conflicts
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Estimating Mortality in Civil Conflicts and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Estimating Mortality in Civil Conflicts 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?