Conflict and climate assertions

I always suspected El Niño had a violent past

So Nature has released a study by researchers from Columbia University purporting a link between the El Niño effect and violent conflict in a subset of countries (those that face large shocks in temperature when El Niño comes knocking).

There are tons of reasons to be skeptical of these results – one thing that made me pause was the complete lack of theory in the paper, and thus there’s no real attempt to discern why El Niño might increase the risk of violence. Even stranger – the results get stronger when the authors control for rainfall and temperature in the affected countries, which really doesn’t leave much of a theoretical channel for the resulting outcome. To make matters worse, the structure of the article makes it nearly unapproachable – which is not the fault of the authors, but the journal itself (why isn’t this appearing in a social science journal? I have a theory).

Edward Carr also stepped in and rightly choked-slammed the results. While I agree with him that the paper is a mess and we shouldn’t make too much of the results, I found some of his complaints to be off the mark:

This design makes sense only if you assume that the random back-and-forth shifting did not trigger adaptive livelihoods decisions that, over time, would have served to mitigate the impact of these state shifts (I am being generous here and assuming the authors do not think that changes in rainfall directly cause people to start attacking one another, though they never really make clear the mechanisms linking climate states and human behavior).  The only way to assume non-adaptive livelihoods is to know next to nothing about how people make livelihoods decisions.  Assuming that these livelihoods are somehow optimized for one state or the other such that a state change would create surprising new conditions that introduced new stresses is more or less to assume that the populations affected by these changes were somehow perpetually surprised by the state change (even though it happened fairly frequently).

Carr seems to be suggesting that, by using run-of-the-mill regression analysis, the authors are implicitly assuming that people in `treated’ countries don’t react in ways to offset future El Niño effects. It seems to me that this approach isn’t making that assumption at all, it’s attempting to measure the average impact on the incidence of conflict. We might expect that the impact to be higher under the assumption of non-adaptive livelihoods than under adaptive livelihoods (unless part of that adaptation is picking up an AK-47), but none of this feeds into the average effect.

Now – it is unclear what that average effect is measuring (and this might be what Carr is getting at), as the impact in year 0 is the impact of El Nino on the unadapted, where the impact in year 0 + T might be the impact on El Nino on the semi-adapted. While this might certainly be an issue for the external validity of the findings (if people have been adapting during the period observed, we’re not likely to see similar effects in the future), it doesn’t affect the researchers’ ability to say “here’s the aggregate impact of El Niño on X over this period.” It’s still a valid statement, if a less interesting one.

I should note this isn’t the first time I’ve taken issue with research on the impact of climate on civil conflict, especially with how this stuff gets reported in the news. The Guardian has already managed to mistakenly conflate the result of increased risk of conflict with direct culpability by claiming that the research “shows 50 of 250 conflicts between 1950 and 2004 were triggered by the El Niño cycle.” This is a common mistake, one I discuss in length here.

One thought on “Conflict and climate assertions

  1. Ed Carr

    August 31, 2011 at 3:37am

    You make an interesting point here – I may well have imputed too much of a conceptual link between environment and conflict to the piece (which you rightly note is totally absent). But I stand by my critique – not of the regression itself, but of its framing. The authors are treating the two ENSO states as two different worlds, looking at the divergence between them, and then arguing that an environmental stress (ENSO) is the (partial) cause of that divergence. But these are not, in fact, independent worlds – they are worlds in which people have long been adapting their livelihoods to these sorts of state changes, before and after the years of their data. This is a false parsing of the data, because the impact of the state changes is not felt directly (a bit more rain might make people more cranky, but is not going to start a conflict) – it is felt via its impact on people’s livelihoods and health . . . and if those livelihoods are already built to address these states, then parsing outcomes into one of two states doesn’t make much sense from an on-the-ground perspective. Statistically it may seem fine, but from the perspective of one who does qualitative work on livelihoods and adaptation in Africa, the parsing is nonsensical.

    To be honest, though, I can let that first point go and still level a really tough critique here. More important, perhaps, is their failure to control for livelihoods and institutions – you point out that when they control for rain and temp, the results get stronger . . . but of course rain is one of the big functions of ENSO, so that result seems counterintuitive. But even if we assume that they have controlled for the variable environmental impacts of ENSO (and they have not, because rainfall and temperature are only part of the issue – the interaction of rainfall with soils and local ecology, for example, is also critical and remains uncontrolled here), they have not controlled for livelihoods through which these impacts will be felt, and the institutions through which people can manage these impacts. Therefore, they have not constructed an “all things being equal” argument at all.

    In short, I am not against the idea of regression as a tool, or their mechanical use of the tool . . . but I think the framing of the concepts that informed the use of the tool was really problematic, with the result being a study that really doesn’t say much . . .

Comments are closed.