This post is co-authored with Michael O’Sullivan.
Effective property rights matter for development. And heck, they even got a couple of shout outs in the recently adopted Sustainable Development Goals. And we know from earlier work that weaker rights can lead to reduced agricultural productivity. So what happens when folks move to better property rights?
In a new paper with Kenneth Houngbedji, Florence Kondylis, and Harris Selod, we use the first randomized control trial of land registration to look at its effects on farming decisions in Benin. Before we get to the results, a bit about the program. Benin's Plans Fonciers Ruraux (or PFR) takes existing customary rights (which are identified during consultations as part of implementation) and formalizes them in land use certificates. As part of the process, parcels are clearly demarcated. This is all done in a mostly locally-managed and implemented process. At the end of the process, landholders have a certificate which conveys rights akin to ownership.
Using data from a survey that we supervised, we look at the early effects of this program (funded by the Millennium Challenge Corporation) that was randomized at the community level. The randomized design helps us deal with the identification challenges that bedevil much of the literature on property rights interventions. By the time our data was collected, everyone had had their fields demarcated within treated villages and about half of the folks had their fields demarcated at least 12 months prior to our survey. The issuance of land certificates had not yet occurred.
So what do we find? Despite the fact that it's early, we see an increase in investment. Compared to households in control communities, those in treated communities are 1.7 percentage points more likely to have planted trees, which is fairly large when compared against the control mean of 4 percent. And, when we look at the parcel level, these trees tend to be of the perennial, cash crop variety (e.g., oil palm and teak). Here the ITT estimate is 2.6 percentage points, which is a 39% increase relative to the control parcels.
When we take a look at gender differences, we find that female-headed households in treated villages were more likely to fallow their land – closing the control community fallowing gap they have relative to male-headed households. Given the earlier results of Goldstein and Udry in nearby Ghana that women fallow their land less (because they have weaker rights), this result gives us one dimension in which this program may help level the playing field between men and women.
When looking at average effects of this program on agricultural yield (i.e., value per hectare), we didn't see any significant results. However, when we look at female-headed households, we find a significant decline. Aside from increased fallowing (which isn't large enough to explain the size of the yield decline), we see no significant changes in labor, input use or crop choices for these women. So what's going on?
It turns out that this program was implemented systematically within communities – but all communities have a boundary. For parcels outside the village boundary, even if they were owned by folks within the treated village, land was not demarcated and registered. And when we look at female-managed parcels inside the village versus outside, we can see some explanations for this difference. Yields on women’s parcels inside the village go down, while they actually increase significantly for parcels outside the village. This seems to be driven by input shifts: women are using significantly less child labor and hired labor on within-village parcels and more hired labor on their fields outside of town. They are also using more fertilizer on the fields beyond the village boundaries. Why? As in the results from Ghana show, you have to use it, or you can lose it. So when women (who start with worse rights) get more secure rights on only some of their fields, they shift production to the less secure fields.
These results all come from a cross-sectional comparison across treatment and control. And that brings us to methodology. This paper taught us an important lesson on using national datasets. When you're evaluating a somewhat widespread program, there's often the tempting allure of a nationally representative dataset. When we first joined this project, it looked like we could use a national survey, so we were really excited. We looked at the national survey and thought that, while the questions clearly weren't as detailed as we wanted, we could use a handful of them for some baseline analysis. So we sampled the follow-up to line up with the national survey. And the problems began.
First, it's hard to tie into someone else's sample. Getting the sample listing, particularly if it wasn't set up for a panel in the first place, can be hard. And then finding the same people and the same places (again, if it wasn't set up and run by you) can be really hard. Finally, when you get the list of households, names and locations, and some simple covariates....you can find out that things look really different. In the end, we had to toss the idea of building a panel with these two really different datasets. We could, however, salvage some community level aggregates to see how well we were balanced across treatment and control (and we were).
We’re now diving into a newly collected round of data to examine the longer-run effects of the PFR program. Stay tuned for more results!
Effective property rights matter for development. And heck, they even got a couple of shout outs in the recently adopted Sustainable Development Goals. And we know from earlier work that weaker rights can lead to reduced agricultural productivity. So what happens when folks move to better property rights?
In a new paper with Kenneth Houngbedji, Florence Kondylis, and Harris Selod, we use the first randomized control trial of land registration to look at its effects on farming decisions in Benin. Before we get to the results, a bit about the program. Benin's Plans Fonciers Ruraux (or PFR) takes existing customary rights (which are identified during consultations as part of implementation) and formalizes them in land use certificates. As part of the process, parcels are clearly demarcated. This is all done in a mostly locally-managed and implemented process. At the end of the process, landholders have a certificate which conveys rights akin to ownership.
Using data from a survey that we supervised, we look at the early effects of this program (funded by the Millennium Challenge Corporation) that was randomized at the community level. The randomized design helps us deal with the identification challenges that bedevil much of the literature on property rights interventions. By the time our data was collected, everyone had had their fields demarcated within treated villages and about half of the folks had their fields demarcated at least 12 months prior to our survey. The issuance of land certificates had not yet occurred.
So what do we find? Despite the fact that it's early, we see an increase in investment. Compared to households in control communities, those in treated communities are 1.7 percentage points more likely to have planted trees, which is fairly large when compared against the control mean of 4 percent. And, when we look at the parcel level, these trees tend to be of the perennial, cash crop variety (e.g., oil palm and teak). Here the ITT estimate is 2.6 percentage points, which is a 39% increase relative to the control parcels.
When we take a look at gender differences, we find that female-headed households in treated villages were more likely to fallow their land – closing the control community fallowing gap they have relative to male-headed households. Given the earlier results of Goldstein and Udry in nearby Ghana that women fallow their land less (because they have weaker rights), this result gives us one dimension in which this program may help level the playing field between men and women.
When looking at average effects of this program on agricultural yield (i.e., value per hectare), we didn't see any significant results. However, when we look at female-headed households, we find a significant decline. Aside from increased fallowing (which isn't large enough to explain the size of the yield decline), we see no significant changes in labor, input use or crop choices for these women. So what's going on?
It turns out that this program was implemented systematically within communities – but all communities have a boundary. For parcels outside the village boundary, even if they were owned by folks within the treated village, land was not demarcated and registered. And when we look at female-managed parcels inside the village versus outside, we can see some explanations for this difference. Yields on women’s parcels inside the village go down, while they actually increase significantly for parcels outside the village. This seems to be driven by input shifts: women are using significantly less child labor and hired labor on within-village parcels and more hired labor on their fields outside of town. They are also using more fertilizer on the fields beyond the village boundaries. Why? As in the results from Ghana show, you have to use it, or you can lose it. So when women (who start with worse rights) get more secure rights on only some of their fields, they shift production to the less secure fields.
These results all come from a cross-sectional comparison across treatment and control. And that brings us to methodology. This paper taught us an important lesson on using national datasets. When you're evaluating a somewhat widespread program, there's often the tempting allure of a nationally representative dataset. When we first joined this project, it looked like we could use a national survey, so we were really excited. We looked at the national survey and thought that, while the questions clearly weren't as detailed as we wanted, we could use a handful of them for some baseline analysis. So we sampled the follow-up to line up with the national survey. And the problems began.
First, it's hard to tie into someone else's sample. Getting the sample listing, particularly if it wasn't set up for a panel in the first place, can be hard. And then finding the same people and the same places (again, if it wasn't set up and run by you) can be really hard. Finally, when you get the list of households, names and locations, and some simple covariates....you can find out that things look really different. In the end, we had to toss the idea of building a panel with these two really different datasets. We could, however, salvage some community level aggregates to see how well we were balanced across treatment and control (and we were).
We’re now diving into a newly collected round of data to examine the longer-run effects of the PFR program. Stay tuned for more results!
Join the Conversation