Quote Originally Posted by Rex Brynen View Post
It is also all about how you count insurgency.

If you do COIN well, you nip an insurgency in the bud very early, before it reaches the stage of full scale civil war. None of these small proto-insurgencies (say, Uruguay vs the Tupamaros, Egypt vs Islamic Jihad, even Germany vs the Red Army Faction or Canada vs the FLQ) are counted in the "win" category in the RAND data, however.

Of course, you can choose not to "count" insurgencies until they reach a threshold of casualties--Lyall and Wilson use "a minimum 1,000 battle death inclusion rule, with at least 100 casualties suffered on each side." However, this is rather like not including boxing matches that feature a first round KO. (It is also never clear to me why insurgency should be measured by an absolute threshold, rather proportional to population or some other indicator of relative intensity and threat).

Second, there seem to be lots of cases missing from the RAND study. Omani suppression of the Dhofari insurgency? The consolidation of the Islamic Republic in Iran against various internal challengers, 1979-80? Syrian suppression of the Hama uprising (20,000 casualties in 1982, so not a minor case)? Iraqi suppression of the southern and Kurdish uprisings in 1991? The civil war (and collapse) of PDR Yemen? Those are just the ME examples.

Large n studies are very useful. They do (as you note) tend to have a lot of devils hiding in the details of how they are coded, however.
As a grad student working full time, I had neither the time nor resources to create my own dataset. RAND's was the least bad. Lyall and Wilson have yet to release their dataset, which only increases my ire and skepticism about their results. I even wrote IO about why they were published in a peer reviewed journal without a dataset released for replication, but apparently IO doesn't require it. Repeated emails to both LTC Wilson and Lyall resulted in refusal to release the data. My K-State professor even wrote Lyall asking for the data. Therefore, I am skeptical of the dataset. When combined with the spuriousness of the whole argument, it's just a bad and misleading paper. There isn't a Correlates of War equivilant for insurgency.

One also has to deal with definitional issues about Civil War vs. Insurgency vs. Terrorist Groups, each of which is like a Venn diagram but have different characteristics.

If I had dedicated time to persue a PhD perhaps I could create a useful dataset.

Another problem is accurate data on the insurgents. For example, if you are coding insurgency "size", do you include fighters, auxilliaries, and sympathizers? How do you get an accurate count of insurgents? They don't usually have a G1. Inclusion and exclusion of these groups changes the data and results.

RAND also has a terrorist data set that includes 640 groups. However, that data has even more issues since it includes groups as small as 10 individuals, which makes it useless for an insurgency study. You are right that the threshold of what constitutes an insurgency in the IR academia sense is in flux.

It's not an easy subject to do large-N work on.

Niel