A Case Study on Afghan Statistics: Does Poverty Reduce Conflict
Edward Kenney Afghanistan Study Group
In Washington DC there is a strong bias towards empirical research over either comparative or historical studies, but I have increasingly become convinced that most people cannot identify the characteristics of a good empirical study over a bad one. And no; shouting about correlation and causation does not signify an understanding of regression analysis. Last month, Jay Ulfelder contributed his thoughts to the subject. Here is his critical paragraph shortened for brevity.
But what does that association tell us about the causes of civil violence, really? …For starters, we have grievance-based theories, which see the roots of violence in poor people’s anger and frustration over their meager living conditions… More recently, some economists have argued that poverty breeds civil violence by lowering the opportunity costs involved with participation in armed conflict… Finally, still other scholars interpret poverty in models of civil violence as a feature of the state rather than its citizens.
Ulfelder is on the right track, but his argument is missing some key points and is confusing in some areas. For the layperson, there are three key questions a researcher must answer to establish causality that are highlighted by this analysis of poverty and civil conflict:
1. Theoretical Framework. For the poverty and civil violence example Ulfelder provides two avenues of causality. 1. Poverty leads to political grievance which in turn leads to violence and 2. Poverty leads to economic opportunism and violence. In Afghanistan there is a third potential “causal story” which Kandahar researcher Felix Kuehn described last month. Poorer Afghans are more likely to be educated and radicalized in the free-of-tuition madrasahs (call it indirect economic opportunism, if you like), which in turn has been linked to the insurgency. The fact that there are at least three mechanisms through which poverty leads to violence actually strengthens the causal claims linking poverty to violence.
2. Omitted Variables Bias (OVB). A variable such as state capacity affects both poverty and civil violence. Capable states generally have both less poverty and less violence. If we fail to include this variable in our study, our analysis almost certainly will be incorrect: After all, how can you separate the effects that are really due to poverty, and those that are due to poor state capacity “masquerading” as poverty. Looking at Afghanistan there are a whole host of other potentially omitted variables. As Felix Kuehn pointed out, the Ghilzai Pashtun tribes (from which the Taliban draw support) tend to be poorer, so our poverty explanation may be capturing a tribal dynamic. Even static variables like geography (which can be both favorable to an insurgency and destructive to economic development) must be included in an analysis to avoid OVB.
3. Reverse Causation. We believe we have a reasonable explanation for how poverty causes civil violence, but is the opposite also true? Can’t civil violence and the associated instability cause increased poverty? As with the previous example, “reverse causation” makes it impossible to judge the extent to which poverty causes civil violence. This is a really big problem and requires a more complex statistical process to resolve.
4. Making Policy Recommendations: Finally, Ulfelder is correct that caution is needed when moving from a descriptive model to a prescriptive one: Just because we may be able to prove that poverty causes civil violence, does not mean that economic aid reduces civil violence. Trying to model prescriptive policies in a conflict environment is incredibly difficult. The key players—whether they be the government, the insurgency, or local civilians—act strategically and take their opponents’ strategies into account. If the civilians in a town believe that the U.S. is likely to respond to an increase in violence by providing aid money, they may choose to increase their support for the insurgency, even if this equilibrium is suboptimal (see Nash, John). Furthermore, policy variables almost always fail the reverse causality test: In Afghanistan, aid is frequently sent to districts with higher than average violence.
Empirical work gets far too much credit in this town, often at the expense of better non-empirical work. I urge analysts to read statistical work critically, don’t be cowed by the fancy lingo, keep these basic points in mind, and don’t be afraid to critique ruthlessly.
