Impacts of Land-titling on Conservation Behavior: Parcel-Level Results

4/23 Lab Presentation

Plan for Today

Introduction

Context: Benin

  • West African country, pop. 14 million
  • Roughly the size of Pennsylvania
  • Classified Lower-Middle Income by the World Bank
  • 3 climactic zones (coastal, forested, semi-arid desert)
  • Advanced in terms of efforts to consolidate land rights (Goldstein et al. 2018)

Context: Plan Foncier Rural (PFR)

  • First piloted in Benin in 1993, full-scale roll out in 2009 by MCC

  • Two-step formal land-titling program:

    Step 1: Program works with local communities to demarcate land parcels

    Step 2: Landholders issued certificates that formally recognize their land rights

(Goldstein et al. 2018)

Context: Plan Foncier Rural (PFR)

Rollout: 2009-2011

  1. Information campaign about PFR in villages across the country
  2. ~1,500 villages applied for program
  3. Lottery takes place to select villages for PFR
  4. 300 villages selected for PFR (283 completed), ~300 eligible villages assigned as controls
  5. ~70,000 parcels demarcated between 2009-2011 in treated villages

Contribution

  • All analysis has been at village-level

  • Limited study of variation across geography, HH characteristics

Methods

Data, Identification Strategy, Empirical Models

Data: Administrative PFR Data (Spatial Data)

  • Geographic parcel data (polygons/shapefiles)
  • Landholder age, sex, occupation
  • Mode of acquisition (Inheritance, Gift, Purchase, etc.)
  • N = ~62,000 treated parcels

Data: Household Survey Data (WB Data)

Source: World Bank

  • 2018 (baseline), 2021 (follow-up)

  • Detailed HH survey on second round of PFR

  • Only one department, ~50 villages

  • Treatment villages selected, then matched with control villages

  • N = ~3000 HHs

Identification Strategy 1: Spatial Data x Mode of Acquisition

RQ1: How do outcomes vary across parcels by mode of land acquisition?

  • Outcome: Share of tree cover loss (Hansen, 2000-2021) / Parcel Area
  • Controls: Region (South, Central, North)?
  • Model: ?

Identification Strategy 2: Survey Data

RQ2: What is the impact of the PFR on landholder behavior?

  • Outcome: Binary variable for engaging in conservation (soil improved, soil conservation, leaving land fallow)
  • Treatment: PFR
  • Method: Instrumental Variable
    • Instrument: 1/0 borders marked with sticks, trees, or stones

Identification Strategy 2: Survey Data

IV First Stage

\[ \text{borders}_i = \pi_0 + \pi_1\text{treatment}_i + \nu_{i} \]

IV Second Stage

\[ \text{conservation}_i = \beta_0 + \beta_1\text{borders}_i + \epsilon_{i} \]

Identification Strategy: Mode of Acquisition

Theoretical Model: TBD!

Identification Strategy: Mode of Acquisition

  • Same models, but with controls and interactions

Results

Spatial, Survey, Combined

Results: Spatial Data

Summary Statistics
Variable Mean SD 95% CI Lower 95% CI Upper
Parcel Area (m²) 17822.00 31137.00 17568.00 18076.00
Age 49.14 15.93 49.01 49.28
Loss Area - 2010 (m²) 160.00 4507.00 123.00 197.00
Loss Share - 2010 (%) 0.01 0.27 0.01 0.01
Loss Area - 2020 (m²) 3248.00 24721.00 3047.00 3450.00
Loss Share - 2020 (%) 0.30 2.20 0.28 0.32

Results - Spatial Data

Results - Spatial Data

RQ1: How do outcomes vary across parcels by mode of land acquisition?

  • Outcome: Tree cover loss (Hansen, 2000-2021) / Parcel Area
  • Controls: Region (South, Central, North)?
  • Model: ?

Results: WB Data

Summary Statistics by Treatment Group
Variable Control Mean Treated Mean Difference p-value
Land Area (ha) 438.751 532.180 93.429 0.84
Borders Marked 0.222 0.482 0.260 0.00
Any Conservation 0.116 0.158 0.042 0.00

Results: WB Data (IV Analysis)

Want to estimate:

\[ \text{Conservation}_i = \beta_0 + \beta_1\text{borders} + e_{i} \]

Instrument: PFR Treatment

  • Relevance: PFR treatment is correlated with whether the landowner reported that their parcel borders were marked

  • Exclusion Restriction: PFR treatment and control villages were matched on observable characteristics, and the only difference between the two groups is the PFR treatment. The PFR treatment should not affect conservation activities through any other channel.

Results: WB Data (IV Analysis)

Instrument: borders_marked - whether the landowner reported that their parcel borders were marked with sticks, trees, or stones (binary)

Border Marking Year by Treatment Status
treated_label 2016 2017 2018 2019 2020 2021 2022
Control 1 1 4 6 6 2 1
Treated 0 0 136 277 219 75 1
Total 1 1 140 283 225 77 2

Results: WB Data (IV Analysis)

Outcome Variables:

  • soil_conservation: Whether the landowner reported using soil conservation practices on their land (binary)

  • soil_improved: Whether the landowner reported that their soil was improved (binary)

  • planned_fallow: Whether the landowner reported that they planned to leave their parcel fallow (binary)

  • any_conservation: Whether the landowner reported soil conservation, soil improvement, or planning to leave land fallow (binary)

cor(ag_land2022.1$treated, ag_land2022.1$borders_marked,
    use = "pairwise.complete.obs")
[1] 0.2733252

Results: WB Data (IV Analysis)

First Stage Regression:

\[ \text{borders}_i = \pi_0 + \pi_1\text{treatment}_i + \nu_{i} \]

first_stage <- glm(borders_marked ~ treated, data = ag_land2022.1, family = binomial)
summary(first_stage)

Call:
glm(formula = borders_marked ~ treated, family = binomial, data = ag_land2022.1)

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept) -1.25366    0.05191  -24.15   <2e-16 ***
treated      1.18262    0.06941   17.04   <2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 5192.9  on 4034  degrees of freedom
Residual deviance: 4888.3  on 4033  degrees of freedom
AIC: 4892.3

Number of Fisher Scoring iterations: 4

Results: WB Data (IV Analysis)

Second Stage:

\[ \text{Conservation}_i = \beta_0 + \beta_1\text{borders}_i + \epsilon_{i} \]

library(AER)

iv.any_cons <- ivreg(any_conservation ~ borders_marked | treated, data = ag_land2022.1)
robust_se_any_cons <- vcovHC(iv.any_cons, type = "HC1")
coeftest(iv.any_cons, vcov = robust_se_any_cons)

t test of coefficients:

               Estimate Std. Error t value  Pr(>|t|)    
(Intercept)    0.080484   0.014539  5.5355 3.300e-08 ***
borders_marked 0.161677   0.041109  3.9329 8.533e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Interpretation: HHs that marked borders are 15.8% more likely to engage in conservation, accounting for endogeneity in border-marking.

Results: WB Data (IV Analysis)

Now with Mode of Acquisition:


t test of coefficients:

                                        Estimate Std. Error t value Pr(>|t|)   
(Intercept)                             0.096965   0.032220  3.0095 0.002633 **
borders_marked                          0.122702   0.078876  1.5556 0.119873   
mode_aqGift/Inheritance                -0.018223   0.039336 -0.4633 0.643190   
mode_aqPurchase                         0.056881   0.352985  0.1611 0.871988   
mode_aqOther/Unknown                   -0.020953   0.041253 -0.5079 0.611538   
borders_marked:mode_aqGift/Inheritance  0.077087   0.093859  0.8213 0.411522   
borders_marked:mode_aqPurchase          0.031144   0.566458  0.0550 0.956157   
borders_marked:mode_aqOther/Unknown    -0.074513   0.164890 -0.4519 0.651368   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Results: Combined

Mode of Acquisition Comparison (Borgou Only)

Results-Combined

Spatial Data - Borgou Only

Conclusion

Lots more analysis to do:

  • Spatial Data
    • Investigate regional variation
    • Look at other types of conservation activities
    • Machine learning techniques to compare loss in predicted control parcels
  • Survey Data
    • Explore other outcomes (e.g., land area, income, perceptions of tenure security)
    • Bring in non-agricultural data

References

Besley, Timothy, and Maitreesh Ghatak. 2010. “Chapter 68 - Property Rights and Economic Development*.” In Handbook of Development Economics, edited by Dani Rodrik and Mark Rosenzweig, 5:4525–95. Handbooks in Economics. Elsevier. https://doi.org/10.1016/B978-0-444-52944-2.00006-9.
Goldstein, Markus, Kenneth Houngbedji, Florence Kondylis, Michael O’Sullivan, and Harris Selod. 2018. “Formalization Without Certification? Experimental Evidence on Property Rights and Investment.” Journal of Development Economics 132 (May): 57–74. https://doi.org/10.1016/j.jdeveco.2017.12.008.