He then shows a break-down of how this hypothetical “organizational structure” variable, which claims to represent the coherence and size of the various subgroups, has shifted in Iraq. Starting with a relatively unified opposition to the US’s invasion, his model shows that by 2002/3 the number of splinter groups within the insurgency had spiked up to what his finding show is an ideal level of ‘structure’. Following the ’06 elections, his model than shows a further splintering of groups for the next year, until the start of the surge, at which point α drops back down, only to rise back up. (Chart is at 6:48 in the video).
His theory is that when α is high or low, we are better able to best insurgencies, either by bringing unified groups to the negotiating table or crushing small groups, and that the ideal organization of insurgencies, revealed in the patterns of attacks, will be consistent in all conflicts. While intriguing, there are some quiet serious questions I have about this theory.
First of all, the idea that violence as a broad concept follows power-law distribution is not only not new (as noted by Drew Conway regarding violence more generally, and this study that looks at terrorist attacks only, both worth a glance in their own right), but also not terribly surprising. Smaller incidents generally cost less than larger ones. In a shot out, killing more people needs you need more men, in the case of an IED more casualties require a higher level of technical proficiency and more materials. Man, experience and materials are all expensive so can only be used a finite number of times.
That of course doesn’t mean that the work couldn’t still be useful to the discipline, particularly the idea of a mathematically ideal type of organizational structure; however, I think it is this ability to intuit the basic finding that there are likely to be more small attacks and fewer large attacks that Ricks is responding to in his pieces, when he asks why this research should be seen as useful. Gourley responds to these critiques by claiming:
With this new approach… We can understand the underlying structure of an insurgency i.e. how an insurgency 'decides' to distribute its forces (weapons, people, money etc). Further, we can explain why this kind of insurgent structure emerges in multiple different conflict zones around the world. We can estimate the number of autonomous insurgent groups operating within a theatre of war. We can monitor and track a conflict through time to see how either side’s strategies are affecting the state of the war. Finally we can compare the mathematical patterns of current ongoing wars with past wars to estimate how close they are to ending.
Well all this sounds great, but does it actually work?
The first problem is the source of the data. Not only does media usually not report ‘failed’ attacks, potentially lowering the reported number of incidences, they often miscode violence in confusing ways. Kalyvas (article gated but abstract here, or in book form) has done a lot of work highlighting the extent to which micro level killings are coded by elites (including the media) to read as 'insurgent' or 'ethnic' violence, instead of homicide or gang violence. Expecting that bias, the problem in the data is not just that failures are excluded, but also that successes are over reported.
The second issue is partially caused by a lack of clarity in the data presented. Because the data tables aren’t available on the lecture slides and I couldn’t find a paper citing this data, I’m actually unclear on what exactly he means by number and size of incidents. Given the difficulty he had in obtaining date from the Pentagon, he has to be looking at more than just coalition casualties*. Assuming that they include other security personnel, insurgent dead, or even civilians, how are Gourley and co. coding the number killed? This isn’t just nitpicking, the study means deferent things when placed in actual context depending on what variables he’s using.
Is the question about the number of casualties on the side of the Counter insurgents (which in and of itself is going to be a contentious definition when coding), as inflicted by the insurgents? Given that the causal story that is told about the relationship is one about the organization of the insurgency, that coding would make some sense (and seems to be the reading that Drew is taking). However, he’s applied the same model to conflicts with 2, 3, or more independent and active parties involved in the war, with no attempt to distinguish what’s going on. Given that there are often multiple factions that are just as often fighting each other as the COIN force (which isn’t necessarily unified itself) assuming a simple binary conflict is intensely problematic, particularly in cases like Iraq where there is quite a lot of insurgent on insurgent violence. To me, that either indicates that the models is being applied without really thinking about how the case fits into the parameter, or that in fact what he is interested in is the total number of deaths, on both sides and including civilians that occur during a conflict.
Leaving this problem aside for a moment, Drew Conway (who is much better at modeling then me) points out the gap between the observed distribution of attacks, and the causal mechanism that Gourley suggests:
Power-law distributions can provide a categorical approximation of a network's underlying structure because in these cases the distribution in question refers to the frequency of edge counts among nodes, a structural measurement. Even for networks, however, the actual underlying structures of networks following a power-law can vary wildly. Attack frequencies, on the other hand, have nothing to do with structure. In what way, then, is this metric valid for measuring the structure or distribution of insurgent forces?
In many ways this in not only a modeling problem, it goes back to my point about insurgencies not being as simple a structure as this model assumes. Gourley’s understanding of insurgent structure is far too simplistic; when the reality is that the social networks that support these groups vary widely in sophistication and resources without necessarily varying in size. Because there is no effort to compare α to any other indicators of insurgent structure one could look at (ex. Number of groups claiming credit for attacks, existence of sections of the organization devoted to nonviolent efforts like social services, or qualitative accounts from intelligence about the nature of the structure) there is little effort made to see if the narrative α presents holds up. To take the example of Iraq, the fragmentation of the insurgency following the 06 elections that is seen in Gourley’s findings doesn’t match up terribly well to the narrative that has been told by observers. Without some qualitative work to back up the narrative, its hard to take such sweeping findings too seriously.
*At least I assume such based on colleagues’ attempts to gather data of a similar nature
No comments:
Post a Comment