Duncan Green, touting an IMF report on the effect of international food price fluctuations, feels like he’s had the last word on this argument:
I usually prefer ‘man bites dog’ research that comes up with unexpected answers, but sometimes it’s helpful to have the opposite – number crunchers who back up what you always suspected, thereby increasing your certainty and confidence.
So people riot when food prices go up. And the effect is bigger in poor countries, where food can constitute 80% of a household’s expenditure. Well duh. But still, now we know that it’s true (because the IMF says so). And that Bob Marley was right (and Chris Blattman wrong).
Not so fast Duncan! There’s a big difference between statistical significance (we can say with some certainty that these two things are connected) and the size of the predicted causal effect. This is a major problem in most econometric literature – we’re so focused on statistical relationships that we forget to discuss how important those relationships really are. From the cited paper:
The estimates in columns (2) and (3) imply that on average a one standard deviation increase in the food price index increased the number of anti-government demonstrations and riots by about 0.01 standard deviations.
So an absolutely massive increase in the food price index leads to an infinitesimally small increased in anti-government demonstrations and riots. Let’s put this into perspective: the effect of an increase in one standard deviation in the food price index raises the number of riots in a country in a given year by 0.0143 riots and 0.0175 demonstrations (I’m using the information from the descriptive statistics for low income countries to obtain the standard deviations for riots and demonstrations).
These are not large increases, and so while you might be able to say there’s a connection between food prices and demonstrations/riots, you really can’t say that it has much of an effect, and probably isn’t what is driving the story here.
The paper suffers a bit for it – it shies away from discussing the size of the effects across all indicators and focuses on statistical significance, even eschewing the kind of statistics that would give us an idea how useful the whole model is. Some comparisons with other exogenous impacts might be useful, but it’s just not there.