The David Kilcullen Collection (merged thread)

[marct]

Hi Jason,

(1) On the railway variable. I agree with the early start dates --- railroads played a huge role in the US Civil War, no question. This case isn't include in our dataset, however, since it was fought primarily (though not exclusively) along conventional lines. We tried to ensure that the conflicts we included were fought principally along guerrilla warfare lines. Of course, mixed cases exist, but I'd argue that the US Civil War was principally conventional, so it was dropped from the analysis.

Okay, I can understand that, but if that is so, why would you have included the Maoist revolution since it went conventional too? It strikes me that there is a very shady area between Phase 1 and Phase 3 wars that could loosely be termed as "insurgencies" or "revolts" that needs to be pulled apart.

(2) The first and second Riel Rebellions were considered but were excluded because they did not met the COW 1000 battle deaths threshold. The same holds for the Sandino Rebellion (1927-33), which I take as the Nicaragua example that was raised (136 Marines died, for example, and around 500-600 rebels). I'll have to recheck the Indian campaign dates again; thanks for pointing that out.

Totally agreed on the first Riel Rebellion - it was more a case of political theatre with only a few deaths associated. The second one, while it doesn't meet the 1000 cut off (directly), though, might be worth considering. I've always been leery of using an arbitrary cut off since I think it skews the data towards kinetic insurgencies. So, for example, insurgencies such as Estonia's Singing Revolution obviously wouldn't be included but, IMO, should be. The asymmetry of such conflicts is not on the battlefield per se, but is decisive in the political ends, and that can have some valuable lessons for other, more kinetically focused, insurgencies.

(3) I totally agree that the air power variable could be expanded. In fact, I'm just back from Afghanistan, where I've started a project on the impact of airstrikes on insurgent violence. So, totally agree here, and I think lots of interesting work could be done.

I look forward to reading it when you come out with it. You might think about parsing your airpower variable by function: movement, recon and strike as examples. If I remember correctly (and i could easily be wrong, it's been a while since I looked at it), in the Nuer war, the Brits had only 4 planes in operation and used them initially for recon. It was only at the end of the war when they started using them for strikes on the herds that they became decisive.

(4) Why not use the actual numbers of deployed mech? This presents three problems: (1) the data are surprisingly sketchy once the war has started, especially for earlier cases; (2) where you take the measurement (at the war's start? middle? end? an average?) is tricky; and (3) most important, the number of MECH deployed is endogenous to the war itself, meaning that your outcome measure and your measure of MECH become correlated, resulting in mistaken inferences given the selection effects present.

A good synopsis of the problems.

1. Agreed, but I suspect that much of it could be teased out. More importantly, the arsenal numbers probably do not reflect the same percentages as are actually being used in the field which throws a pretty serious question on the validity of your argument re: mechanization. It might be possible to take exemplar cases from the various time periods and use those to strengthen your argument.

2. Actually, I think this is a false problem: I would use a slope of initial, middle and end to reflect changes in the operations of the campaigns. You could, in effect, create a matrix of inter-related variables with, say, a "High" initial MECH value which drops during the campaign. This would allow you to have a much stronger argument about the relative value / cost of using MECH in various campaigns.

As an added note, I think you would also get a lot of additional information if you functionally split out the uses of MECH during the campaigns in a similar manner to what i was suggesting for airpower.

3. I think I would disagree with you on this point. I would argue that it is pretty strongly correlated at the start of a campaign, albeit modified by both doctrine and units organization/selection, but not necessarily so throughout the campaign. Selection pressures during the campaign would not be reflected in your model, which would result in mistaken inferences about the causal effects of MECH on the outcomes. In effect, the argument line you seem to be following appears to assume that outcomes can be predicted from starting conditions which ignores the actuality of the campaign process - the "friction" and "fog of war" as it were.

It's really the implications of this third point that concern me more than anything else. Once such a mechanistic (no pun intended ) model is accepted, then, I would argue, it only holds true while all of the sides involved accept the basic parameters of that model. But such a situation is what gives rise to variable asymmetries in the first place: a rejection of those parameters (aka "conventions"). I think that this is pretty clearly illustrated in both the primary AO of AQ (i.e. communications) and during the Millenium Challenge wargame when GEN Van Ripper showed how false those parameter assumptions were.

At any rate, that's moving beyond the scope of your initial paper, so I'll leave it there.

Cheers,

Marc