Chris Blattman

The economics of child soldiering

After years of plodding, revisions, additions and subtractions, I’ve finally finished what I hope is a final draft of my Industrial Organization of Rebellion paper (with Bernd Beber).

We finish none too soon. Daron Acemoglu just released a paper on the Economics of Labor Coercion with Alexander Wolitzky. (Am I the only one who wonders if Daron even sleeps?)

Our question: why do some groups recruit children while others do not? Why do some coerce while others reward?

To answer, we mine the literature on the industrial organization of firms. A key bit of our theory: high ability people have good outside options (and know it) and will run away if coerced. Only low ability people can be tricked or cajoled or threatened into fighting.

The thing is, if threatening and punishing people is just as costly as rewarding them, you’ll always prefer to pay the high ability people rather than coerce the low ability ones. It’s only when punishment is relatively cheap that it becomes optimal to coerce.

Here’s out output from a fancy pants principal-agent model of the return to a rebel leader (vertical axis) of people of varying ability (horizontal axis) for different costs of punishment (costly punishment, higher k):

(Coercive equilibria are dashed, while rewards-based recruitment is a solid line.)

When is punishment costly? Threats are always cheap. Mainly it’s when your civilian or foreign support isn’t contingent on respect for human rights. If, like Uganda’s LRA, you don’t even have any support, punishment is cheap indeed.

While the theory is important, mainly we use interviews and micro data from northern Uganda to build and substantiate our assumptions: for instance, that punishment is cheap and that children are lower ability guerrillas.

The key dynamic, however: children are more easily misled, indoctrinated and disoriented, and so they are much less likely to escape. Adults have better outside options, and are harder to mislead, and so it’s almost never easy to coerce them.

Here’s how many months it took you to escape the Lord’s Resistance Army depending on the age they kidnapped you:

The newest bit of the paper tests our predictions using “cross-rebel regressions”. We collected data on a random sample of rebel groups in Africa. If you hold your nose and believe the data, coercion and child soldiering have a pretty strong relationship, even if you ignore the LRA:

I’m not sure my data sausage factory is any worse than that for, say, GDP. But a sausage factory is a sausage factory. I think I’ll stick to survey data in future.

9 Responses

  1. Great stuff Chris. Also, I’m pretty sure Daron never sleeps. Someone should write a paper on that man’s productivity.

  2. Chris, this is really fascinating. I’m wondering whether the graph of months abducted in the LRA and age is based on “how many months it took to escape” or number of months abducted? What I mean is, does your data include only those who escaped or also those who were released by the LRA as well? Since it was a regular part of their MO to abduct adults to act as porters and then release them, I’m wondering how their inclusion/exclusion from your analysis would impact your observations. (Also, with all of your experience here in northern Uganda, I’m sure you know that to say the LRA doesn’t have ‘any’ support is not quite accurate and simplifies a very complex and fluid relationship with civilians and foreign support–although obviously their support doesn’t seem to have a great concern for human rights which would as you stated make coercion costly.)

  3. The content is interesting, but don’t you mean vertical and horizontal “axis” in paragraph 6? (As opposed to “access”?)

  4. Chris – this is fascinating, but I don’t like the last graph.
    The regression line looks like there is a continuous linear relationship – but forced recruitment is a dummy (or is it just almost a dummy? – that seems odd, too.) It looks weird and a little misleading in any case. Box and whiskers, maybe?

  5. Not to rain on the Acemoglu parade, but sentences like this (from the abstract) are why other people make fun of us: “Our model also predicts that the slave trade makes slaves worse off, conditional on enslavement…” But it is a fascinating sentence… have to read the paper… does slavery without the slave trade make the captured slave better off? If there is no resale market then your owner invests more in you? But suppose you could be bought by someone who would free you, would that not make you better off in expectations sense… or at least you could hope? But maybe your owner who would free you because becomes uncomfortable with owning slaves would now be able to get rid of you to a harsher master by selling you. But maybe by your work ethic you could affect your resale value. I wonder if there isn;t a philosophical parallel between people’s objections to certain things being bought and sold because it dehumanizes, to certain subjects being or not modeled, because dehumanizes….

  6. Incidentally, Chris, you are not the only one to have those doubts about Daron Acemoglu. I have long ago made up my mind about him: he is a robot who does not engage in any of the customary human physical functions with an army of research assistance to boot. Occam’s Razor pure and simple. Nothing else can even come close to explaining his monstrous productivity.

  7. Sorry for commenting without having read the paper, but my main concern from your description, to the extent that it’s an IO model, is that “high ability” as a rebel fighter would be quite different from “high ability” for the outside options otherwise available to a fighter.

Why We Fight - Book Cover
Subscribe to Blog