Casualty Counts
I have two recent posts at Duck of Minerva about collateral damage statistics. The first post bemoaned the absence of clear-cut data disaggregating collateral damage victims from total civilian deaths in wars, and argued we need such data for conceptual and policy reasons and to clarify legal debates about war law. In the second post, I played around with some existing data on civilian deaths in interstate wars over time (Alex Downes') to create a conceptual example of what such a research agenda would look like and see what the numbers implied in existing data on civilian casualties might look like if so disaggregated. After twiddling Downes' numbers, I found that according to his data:
The ratio of collateral damage victims to war crimes victims has dramatically increased since the end of the Cold War. According to Downes' dataset, between 1823 and 1900, unintentional deaths constituted 17% of all deaths in war. Since 1990, that number has risen to 59%. But collateral damage is not only increasing as a percentage of all civilian deaths. The number of collateral damage victims is also increasing over time in absolute terms. Between 1823 and 1900, 84 civilians per year on average were the victims of collateral damage. Since 1990, the number is 1688 per year - a twenty-fold increase.
This post generated so many thoughtful comments at the Duck, Dan Drezner and Armchair Generalist that I thought I'd issue a follow-up, and since the follow-up is more than can be fit into comments threads on those blogs, I figured I'd post it here. I'm going to focus on critiques of the post rather than the validating comments because my aim here is to think through how to design a better way of getting at this understudied question rather than to defend the "findings" in the original post, which were (as I acknowledged originally) speculative and based on reading between the lines in data that was not actually designed to capture what I was looking for. (Alex Downes' dataset and coding manuals are here.)
Critical responses to the post fall into three categories. First, a few commenters dismiss the findings on the basis that Downes' data was faulty to begin with and can't be trusted. Second, some commenters critique the research question I'm asking on conceptual grounds. Third, some commenters have explanations for the variation in Downes' data. All of these critiques are extremely helpful in thinking about how to go about gathering genuine data on this question. I address them in turn below.
The Validity and Limitations of Downes' Dataset. The critiques fall into three categories: the types of wars on which he gathered data, the sources he used for casualty counts and the way he coded civilian victimization.
Excluded Wars. Downes gathered data on interstate wars between 1823 and 2003. As such, his dataset excluded a lot of bloodshed in civil wars across this time period. I don't disagree that any study on overall civilian deaths worldwide would have to look at civil wars as well, and I began with Downes' data simply because it was the only longitudinal dataset available on short notice organized in such a way that some collateral damage numbers are implied. As should be obvious given that I ripped this off as a blog post rather than an APSR article, this is a starting point and rasion d'etre for further investigation, not a definitive analysis.
Sources Used. Some commenters say Downes' numbers are useless because they come from official sources. Actually, he drew on many sources in constructing his casualty counts, and his dataset includes both low, medium and high estimates. As a shortcut, my numbers reflect his "medium" estimates for each war. It would be interesting to do it over using high estimates instead, or by averaging the three counts and creating a new column in the dataset. This doesn't address the critique that numbers are always going to be imperfect, and it doesn't address the moral uneasiness that some people have on counting corpses in the first place. I happen to be in the camp that says the perfect can be the enemy of the good when it comes to doing our best to study uncomfortable subjects empirically.
Coding Choices. The best critique of my use of Downes' data is that the way it's coded is not particularly amenable to the question I'm asking, and I don't deny this - in fact I explained it in the original post. The problem is, I haven't found any existing datasets that are so amenable, which is why I began by extracting some numbers from Downes'. (I am negotiating with the Uppsala people and with Ben Valentino to get additional data that may allow me to begin supplementing this analysis.) Downes' data does allow us to see how the numbers of civilian deaths for interstate wars where there is no evidence of intentional civilian victimization have changed over time, in absolute terms and as a percentage of all civilian death, and I think that's a start. However he's interested in civilian victimization, not collateral damage, so he codes all civilian deaths as victimization if there is any evidence that a country victimized some civilians intentionally in a given war. So actual collateral damage figures are likely to be much higher than what I extracted from his dataset. His approach made sense for his research and I don't think it's a critique of his data per se. I do think that it showcases my original point that existing data is not really getting at the question as I'm framing it.
In short, Downes' data is limited when it comes to this question - as are all datasets I know of so far, but I don't think that means his data is faulty. It means that he's not been asking an important question that I'm saying someone should ask. It also means that his dataset would need to be aggregated with other existing datasets before we could say anything decisive about this issue.
Conceptual Issues. This brings us to conceptual issues: if a study were constructed that were to either gather new data on the topic or try to aggregate implied data in existing datasets, what assumptions and definitions should drive such a study? Commenters are asking: what qualifies as collateral damage? What qualifies as a war crime? What about indirect, as well as direct deaths from military operations?
These are important questions. In terms of war law, by collateral damage we mean that civilians are not being targeted directly, nor are they being disregarded in targeting decisions - that there is evidence that militaries make some effort to avoid hitting them. So commenters who argue that Downes coded his data wrong because he included "indiscriminate bombing" as "civilian victimization" instead of leaving it in the "civilian deaths" category are mistaken, I think. Indiscriminate bombing is also a violation of humanitarian law, so victims of indiscriminate bombing would not be considered collateral damage as I'm defining it. I want to know how many people are dying from operations that are consistent with the laws of war, because that's the part of war law that is most unregulated and I'm interested in figuring out how urgent it is that that change.
The laws of war on targeting don't say anything directly about taking into account indirect deaths - like people on dialysis machines who will die in a hospital if you bomb the electric grid. But I think you could infer such a responsibility from the rules on targeting civilian objects, so I think it would make sense to include indirect deaths in the counts in a broader study. And I think there are some good emerging measures of this - the work of Bruce Russett and his collaborators for example and Neumayer and Plumper's. This would then imply that emerging legal norms should require states to consider those indirect effects in their calculations of proportionality, and for human rights organizations to report them alongside deaths and injuries from ordnance. I also think that if I were constructing a larger dataset on this I'd want collateral damage to be a continuous variable rather than a dichotomous one.
Explanatory Responses. A few of the commenters accept the findings on its merits, however speculative, but put forth hypotheses about how to explain the result. Some say it's not surprising, because of course collateral damage deaths as a percentage of all civilian deaths would be increasing as war crimes decrease. I already said as much in my post, but I'm with Drezner in thinking that the rise in absolute numbers is the surprising part - both because these numbers are only for international wars, which themselves are on the decline and because weapons are supposed to be getting more precise. I think that the commenters who argue that this data demonstrates a myth of precision targeting might be onto something, but I also agree that we can't know for sure until we do further analysis. Precision weapons are only as good as the intelligence and intent behind them. But I don't agree that collateral damage has only been an issue since the advent of aerial bombing. Ever since debates over the use of the longbow, there has been concern about removing the combatant from proximity to his target, precisely on these moral grounds. I think what we're witnessing is a moment in which the consequences are becoming so widespread that we may need to rethink some of the loopholes in war law.
I look forward to continuing to ponder this question in coming days and greatly appreciate the engagement of readers with questions of how this might be measured. If anyone knows of existing and available data where numbers to supplement Downes' might be found, please advise.
November 19, 2009 at 18:24
Reader Comments (3)
Charli, thanks for thinking about this stuff, and for posting this piece here at CTlab. This is important work, for all sorts of reasons; the one that I'm most interested in, and that was strongly suggested in an earlier Duck post, is how it all relates to LOAC provisions, specifically military necessity and proportionality. I'll be back with comments once I've read this piece in full.
"...I'm with Drezner in thinking that the rise in absolute numbers is the surprising part - both because these numbers are only for international wars, which themselves are on the decline and because weapons are supposed to be getting more precise. I think that the commenters who argue that this data demonstrates a myth of precision targeting might be onto something, but I also agree that we can't know for sure until we do further analysis."
I don't believe that it is a surprising result. I think that it is a spurious result due to substantial bias in measurement of unintentional deaths.
a) For wars coded as having the intentional targeting of civilians, you measure zero unintentional deaths. That undercounts unintentional deaths. A nation that typically fights with a callous disregard of civilians will nevertheless also kill civilians in circumstances that would be considered "legal" collateral damage. These latter "legal" collateral damage deaths are ignored in your analysis.
b) For wars coded as not having the intentional targeting of civilians, you measure 100% of the dead as unintentional. This overcounts unintentional deaths, since there were undoubtedly a few intentional war crimes in many or most of the post-1990 wars. These intentional war crimes are treated as collateral damage in your analysis.
c) There are a lot more wars coded as having intentional targeting of civilians in the past than post-1990. Further, these wars saw lots civilian dead.
d) As a consequence, you've generally undercounted unintentional civilian deaths in the past, and (likely to a much lesser extent) overcounted unintentional deaths post-1990.
e) This means that your comparison of absolute numbers of collateral damage deaths over time is biased and unreliable.
Put another way, the measurement error in the variable of interest (unintentional deaths) is correlated with the partitioning variable, time. I'm afraid that the renders the results biased and unreliable.
I'd like to note that, if starting today all combatants would simply start purposely targeting civilians going forward, this measure would show zero collateral damage deaths going forward. We'd see a large increase in civilian deaths, but a complete cessation of collateral damage as measured in this analysis. Hardly seems meaningful.
[I will crosspost this to Drezner's comments]
Your posting here short-changes the most substantial problems with using this data, and even to some extent trying to analyze this issue.
1) The missing wars problem is about far more than civil wars. As I noted in my comments on Drezner's blog, Downes leaves out several major 19th century interstate wars and numerous extrasystemic wars (colonial conquest and the like). You can't derive any meaningful comparison about different time periods when the data exhibit severe missing data problems, especially when those are concentrated in certain time periods.
2) The civilian casualty counts that Downes provides for earlier wars, particularly 19th century ones, are most likely to be drastically low, even his "high" values. That's because his data are based on reported civilian casualties for the various conflicts included. While that might provide accurate counts today, it is unlikely that anyone today would be able to pull together anything close to an accurate count of civilian casualties for a 19th century war because the documentation won't be there.
Even conflicts earlier in the 20th century will exhibit the latter problem, even if Downes managed to include all of them in his data set.
If you do want to look at civilian victimization over time, you'll need to control for relative intensity of war and increasing population along with all of the other problems noted. Good luck!