One of the more exciting sessions I went to at the recent Centre for the Study of African Economies Conference was on the relationship between agricultural plot size and productivity. I walked out of the session not sure of the shape of the relationship, but I was sure of the fact that there is a lot of measurement error going on. And this is measurement error that matters a lot.
So for those who don’t follow this, there has long been a puzzle in development economics where, per unit of land, small farms produce more than larger farms. This goes back to the 1920s (see Leah Bevis’ mini-history here). And there is a legion of explanations. In a 2012 post, Mark Bellemare lays out three of them: market failures (e.g. extra workers hanging around family farms), omitted variables (e.g. soil quality), and/or measurement error. Bellemare talks about a paper (gated) he has with co-authors that concludes it’s just a bit of soil quality, and a lot of market failures.
Recently, there has been a surge in the measurement error work. First up, we have a paper by Gero Carletto, Sara Savastano and Alberto Zezza (ungated version here) which looked at how accurate farmers are in estimating the size of their fields. The answer is: not so much, and there is a method to the madness. Comparing GPS-measured field size and farmer reports, Carletto and co. find that farmers with smaller fields overstate their field size, while those with larger fields understate the size. Their conclusion: this measurement error goes in the opposite direction – making the inverse size-productivity relationship stronger.
Now there are some new kids on the block. A recent paper by Leah Bevis and Chris Barrett brings a behavioral lens to this relationship. They have good data on soil quality, which they can use to rule out this as an explanation for the inverse size-productivity relationship – as well as a host of other potential explanations. And they are also using GPS measured plot size, so we are starting from a Carletto and co. world of better size measurement. Their story to explain the inverse size-productivity relationship: smaller fields have more edge-area, and edges are more productive.
It turns out there is a literature on the more productive boundary areas of plots. And this has produced a set of biophysical explanations: edges may get more sunlight and/or plants on the edge might get better pollination, less pests or have less competition for nutrients or water. But Bevis and Barrett favor a behavioral explanation: crops at the edge get more attention and problems are more visible. They show that fields with a higher edge to interior (i.e. perimeter to area) ratio get more labor per hectare. They also lay out an interesting second behavioral explanation, which builds, in a way, on Carletto and co. Bevis and Barrett show that the more farmers overestimate the size of their field (relative to GPS measured size), the more inputs (labor) they seem to use, even when controlling for the perimeter to area ratio. Conversely, when they underestimate the size of their field, they use less. Now, Bevis and Barrett can’t rule out the possibility that farmers are also consistently over or under estimating their yields, but this is an intriguing other avenue.
And walking down that avenue comes two papers: one by Sam Desiere and Dean Jolliffe and another by Sydney Gourlay, Talip Kilic and David Lobell. Let’s start with the one by Desiere and Jolliffe. First off, they spend a fair amount of time discussing the farm-size vs plot-size productivity relationship (I am focusing on the latter here, but see his paper for a review of the history of this discussion). Desiere and Jolliffe are after better yield measurement, and so they are using crop cuts. For those of you who aren’t familiar with this, this involves harvesting a small part of the field (usually 4x4 meters) and then weighing that harvest.
Using the ESS/LSMS-ISA from Ethiopia, Desiere and Jolliffe show that, when you use farmers’ self-reports of yields, there is, in fact, an inverse plot size-productivity relationship. But when you use crop cuts, this relationship goes away and may actually switch to a positive one (there are some issues around what kind of crop measurement you use). Desiere and Jolliffe show that this is likely to be due to systematic over and under reporting – by large margins. As they put it “Our best estimates indicate that production needs to be over reported by 40 percent to 100 percent on plots of 100m² and underreported by 10 percent to 35 percent on plots of 3000m² to explain the IR [inverse relationship].”
In their paper, Gourlay and co. give a somewhat less extreme version of the overreporting story. Gourlay and co. take us over to Uganda, where in another LSMS-ISA supported dataset, they have a host of variables. They control for objective measures of soil quality, genetic heterogeneity (!) of maize, and edge effects, among a host of other characteristics. They, too, are using crop cutting (with two rounds of data) but they also complement this with remote sensing data (potentially neat – but a subject for a future post). When they use the farmers’ estimates of yield: presto – a 1 percent increase in size leads to a 0.35 to 0.9 decrease in yields (i.e. output per unit of land). And when they use crop cut or remote sensing yields? No relationship. Indeed, Gourlay and co. show that the farmers’ overestimates of yields (relative to the more objective measures) are quite high in the lowest part of the distribution and then decline.
So it looks like we need to rethink the assumption that smaller fields produce more per hectare than larger ones. But we’re not quite there on why. Many of these papers are early working papers/drafts and it will be interesting to watch this new crop mature. We’re headed to a new world of measurement – and a better understanding of agricultural productivity.
So for those who don’t follow this, there has long been a puzzle in development economics where, per unit of land, small farms produce more than larger farms. This goes back to the 1920s (see Leah Bevis’ mini-history here). And there is a legion of explanations. In a 2012 post, Mark Bellemare lays out three of them: market failures (e.g. extra workers hanging around family farms), omitted variables (e.g. soil quality), and/or measurement error. Bellemare talks about a paper (gated) he has with co-authors that concludes it’s just a bit of soil quality, and a lot of market failures.
Recently, there has been a surge in the measurement error work. First up, we have a paper by Gero Carletto, Sara Savastano and Alberto Zezza (ungated version here) which looked at how accurate farmers are in estimating the size of their fields. The answer is: not so much, and there is a method to the madness. Comparing GPS-measured field size and farmer reports, Carletto and co. find that farmers with smaller fields overstate their field size, while those with larger fields understate the size. Their conclusion: this measurement error goes in the opposite direction – making the inverse size-productivity relationship stronger.
Now there are some new kids on the block. A recent paper by Leah Bevis and Chris Barrett brings a behavioral lens to this relationship. They have good data on soil quality, which they can use to rule out this as an explanation for the inverse size-productivity relationship – as well as a host of other potential explanations. And they are also using GPS measured plot size, so we are starting from a Carletto and co. world of better size measurement. Their story to explain the inverse size-productivity relationship: smaller fields have more edge-area, and edges are more productive.
It turns out there is a literature on the more productive boundary areas of plots. And this has produced a set of biophysical explanations: edges may get more sunlight and/or plants on the edge might get better pollination, less pests or have less competition for nutrients or water. But Bevis and Barrett favor a behavioral explanation: crops at the edge get more attention and problems are more visible. They show that fields with a higher edge to interior (i.e. perimeter to area) ratio get more labor per hectare. They also lay out an interesting second behavioral explanation, which builds, in a way, on Carletto and co. Bevis and Barrett show that the more farmers overestimate the size of their field (relative to GPS measured size), the more inputs (labor) they seem to use, even when controlling for the perimeter to area ratio. Conversely, when they underestimate the size of their field, they use less. Now, Bevis and Barrett can’t rule out the possibility that farmers are also consistently over or under estimating their yields, but this is an intriguing other avenue.
And walking down that avenue comes two papers: one by Sam Desiere and Dean Jolliffe and another by Sydney Gourlay, Talip Kilic and David Lobell. Let’s start with the one by Desiere and Jolliffe. First off, they spend a fair amount of time discussing the farm-size vs plot-size productivity relationship (I am focusing on the latter here, but see his paper for a review of the history of this discussion). Desiere and Jolliffe are after better yield measurement, and so they are using crop cuts. For those of you who aren’t familiar with this, this involves harvesting a small part of the field (usually 4x4 meters) and then weighing that harvest.
Using the ESS/LSMS-ISA from Ethiopia, Desiere and Jolliffe show that, when you use farmers’ self-reports of yields, there is, in fact, an inverse plot size-productivity relationship. But when you use crop cuts, this relationship goes away and may actually switch to a positive one (there are some issues around what kind of crop measurement you use). Desiere and Jolliffe show that this is likely to be due to systematic over and under reporting – by large margins. As they put it “Our best estimates indicate that production needs to be over reported by 40 percent to 100 percent on plots of 100m² and underreported by 10 percent to 35 percent on plots of 3000m² to explain the IR [inverse relationship].”
In their paper, Gourlay and co. give a somewhat less extreme version of the overreporting story. Gourlay and co. take us over to Uganda, where in another LSMS-ISA supported dataset, they have a host of variables. They control for objective measures of soil quality, genetic heterogeneity (!) of maize, and edge effects, among a host of other characteristics. They, too, are using crop cutting (with two rounds of data) but they also complement this with remote sensing data (potentially neat – but a subject for a future post). When they use the farmers’ estimates of yield: presto – a 1 percent increase in size leads to a 0.35 to 0.9 decrease in yields (i.e. output per unit of land). And when they use crop cut or remote sensing yields? No relationship. Indeed, Gourlay and co. show that the farmers’ overestimates of yields (relative to the more objective measures) are quite high in the lowest part of the distribution and then decline.
So it looks like we need to rethink the assumption that smaller fields produce more per hectare than larger ones. But we’re not quite there on why. Many of these papers are early working papers/drafts and it will be interesting to watch this new crop mature. We’re headed to a new world of measurement – and a better understanding of agricultural productivity.
Join the Conversation