Thursday, November 16, 2006

Does troop size affect the number of deaths in Iraq?

This is an interesting analysis of the effects of troop size on the number of violent deaths in Iraq. The executive summary is: There's little evidence that putting more troops in will actually reduce the violence in Iraq. I'm not surprised. In fact I'm actually surprised that a rigorous analysis doesn't show violence increasing with troop size because the presence of troops acts as an irritant. I suspect that the explanation could be that there is an inflection point in the graph below which fewer troops leads to more violence because there aren't enough to do the job and above which more troops leads to more violence because they act as an irritant.

I do have a few bones to pick with BruceR's analysis.

1) First of all a link to the raw data would be nice, since it would clear up some questions about the interpretation of the data.

2) I would like to see at least one line graph rather than the bar graphs presented. I think that the line graphs will more clearly show a trend that I think I perceive in the data. That is that the violent deaths do have a tendency to vary inversely with the troop levels but this tends to lag behind the troop levels by about a month. There are a few factors which might contribute to this. One being that the graphs don't chop the data fine enough since troop levels may actually decrease on the 15th of a month but this would not be reflected until the 1st of the next month in the data. (I don't know if this is true since I have not seen the raw data). Another contributing factor could be that this really reflects reality. That increases in troops have a delayed impact because they have a tendency to break up terrorist rings and prevent future acts rather than having an immediate effect on attacks.

3) A related point is that his analysis does not discuss how violence levels are likely to directly impact force levels by causing the coalition to increase the forces during periods of increased violence. When you think about it, this could lead to that delayed inverse correlation that I think I see in the data. When things are bad, the generals start building up troops. This buildup continues until after the violence peaks. When violence starts going down, troop levels decrease until the next round of violence begins. According to this model, troop levels may have no causative effect on the violence, but there still would be a pretty good correlation.

4) As BruceR points out, there is an obvious and sensible correlation between increased violence and Ramadan (An event that comes 11 days earlier every year, so it has shifted more than a month during the time course of this graph). You can see the increases in violence around November or October of every year. But the seasonal increase in violence he points out in April is primarily due to one bad month in April 2004 when we fought to retake Falluja.

5) I think that if BruceR is going to bin the data by month and especially if he is going to compare it from one year to the next it would be much more informative to bin it by the Muslim calendar rather than the Gregorian. This would probably do a lot more to explain the seasonal variations in violence. Along these lines, I was a bit surprised to see violence against the coalition troops so clearly down in the months of February and March. Considering that this has been right around the time of Ashura and Arba'een and the period in between. This is a time of great conflict between Sunni and Shiite as these festivals highlight the events of a great Sunni victory over the Shia. Perhaps coalition deaths go down because the Iraqis are all together too focused on killing eachother at that time to bother with us.

These criticisms aside, this is a great analysis of the data. Surely there are folks at the Pentagon doing comething like this? Why don't we ever see that presented to the public?