improvements in indigenous land rights and deforestation improvements in indigenous land rights and...

Improvements in Indigenous Land Rights and Deforestation:Evidence from the Brazilian Amazon

AidData and KfW

Geospatial Impact Evaluation

• Use spatial information on program activities• Merged with high-resolution geo-referenced outcomes

• Geo-referenced surveys• Remotely sensed (forest cover, nighttime lights)

• Causal attribution (identification) possible through matching, fixed effects, and discontinuity techniques

• Examples in growing number of fields/sectors• Land rights• Health• Governance• Post-conflict• Education

Does demarcating indigenous lands reduce deforestation?• Indigenous control, stewardship shown to be

associated with lower deforestation rates (Nelson et al. 2001, Nepstad et al 2006, Nelson and Chomitz2011, Nolte et al. 2013, Pfaff et al 2014, Vergara-Aseno and Potvin 2014)

• Most studies compare indigenous to other governance/rights Don’t consider time variation in protection status

• Given low rates of deforestation observed on indigenous lands, is demarcation likely to influence deforestation?

Does demarcating indigenous lands reduce deforestation?• Important to understand because alternative policies

show promise:• Monitoring technology and enforcement efforts (Hargrave and

Kis-Katos 2013, Arima et al, Assuncao et al 2014, Borner et al 2014, Borner et al 2015)

• Protected areas (e.g., Cropper et al 2001, Andam et al 2008, Borner et al 2010, Joppa and Pfaff 2011, Blackman et al 2011, Sims 2010)

• Payments for environmental services (e.g., Pfaff et al 2008, Robalino et al 2008, Honey-Roses et al 2011, Alix-Garcia et al 2012)

• Forest concessions (e.g., Mertens et al 2004) • Interventions in beef, soy supply chains (Nepstad et al 2014).

Project Description

• In 1988 constitution, Gov of Brazil committed to demarcating indigenous people’s territories

• Between 1995-2008, with funding and tech support from KfW and the World Bank, the PPTAL project identified, recognized, and studied 181 community lands.

• By 2008, 106 community lands demarcated, covering 38 million hectares (~35% of all indigenous lands in Amazon)

Project Description

• Demarcation: recognition by the Min of Justice

• Followed by regularization (entry into municipal, state and federal registries)

• Varied by community between 1995 and 2008• Median year is 2001

• Support for Boundary Enforcement

Data

• Treatment status• Boundaries of community lands

• Administrative data on demarcation dates

• Merged with satellite-based greenness measure• NASA Land Long Term Data Record (LTDR), 1982-2010

• Processed to Normalized Difference Vegetation Index (NDVI)

• Range is [0, 1] (0 = rocky, barren; 1 = dense forest)

• Annual NDVI max (and mean) measures

• Covariates• Climate (precip., temp.); topology (elevation, slope); distance

to rivers; gridded, interpolated population

Comparison Imagery from Manicoré Region, Brazil

Sample communities

Empirical Methodology

• Propensity Score Matching • Differences over time across matched treated/comparison

communities

• Match on baseline levels, pre-trends, & covariates

• Demarcated vs. not; “Early” (‘95-’01) vs “Late” (‘01-’08)

Δ𝑁𝐷𝑉𝐼𝑖𝑝= 𝛼 + 𝛽𝑇𝑖𝑝 + 𝜃𝑁𝐷𝑉𝐼𝑖𝑝1995 + 𝜃Δ𝑁𝐷𝑉𝐼𝑖𝑝 1982,1995 + Γ𝑋𝑖𝑝 + 𝐷𝑝 + 𝜖𝑖𝑝

Not Demarcated

Demarcated

Max

ND

VI

Year

NDVI Trends

Not Demarcated

Demarcated

Res

idua

l

NDVI Trends, Normalized by Year

NDVI Trends

Late Demarcation

Early Demarcation

Max

ND

VI

Year

NDVI Trends, Normalized by Year

Early Demarcation

Late Demarcation

Res

idua

l

Propensity Score Matching:1st Stage Results

Demarcation Year and NDVI Pre-Trends

Summary Statistics: Outcomes and Covariates

Cross-Section Results:Ever Demarcated

Differences-in-differences:

Demarcated vs. non-demarcated

Treatment = demarcated between ’95-’08

Outcome = Change in mean NDVI between ‘95 and ’10

Sample: 28 community pairs, matched on covariates

Cross-Section Results:Early Demarcation

Differences-in-differences:

“Early” vs. “Late”

Treatment = “Early” demarcation (‘95-’01)

Outcome = Change in max NDVI between ‘95 and ’01

Sample: 33 community pairs, matched on covariates

Cross-Section Results:Demarcation + Enforcement Support

Differences-in-Differences:

Treatment = Demarcation + Enforcement Support

Outcome = Change in max NDVI between ‘95 and ‘10

Sample: 44 community pairs, matched on covariates

Cell-year panel model

𝑁𝐷𝑉𝐼𝑖𝑐𝑡= 𝛼 + 𝛽1𝐷𝑒𝑚𝑎𝑟𝑐𝑎𝑡𝑒𝑑𝑖𝑐𝑡 + 𝛽2𝐸𝑛𝑓𝑜𝑟𝑐𝑒𝑚𝑒𝑛𝑡𝑖𝑐𝑡 + Γ𝐶𝑙𝑖𝑚𝑎𝑡𝑒𝑖𝑐𝑡 + 𝐷𝑐 + 𝐷𝑡 + 𝜖𝑖𝑐𝑡

• Treatment status at finer time intervals• Testing specific timing of effects (only after demarcation)

• Covariates available at finer spatial resolution – improved precision

• Fixed Effects• Control for time-invariant unobservables

Summary Statistics for Grid Cell Level Panel Dataset, Weighted by Community Size

Outcome = Level of max NDVI in year

Sample: 8,483 grid cells within demarcated communities with annual obs from 1982-2010

Standard errors clustered by community & year

Panel Results:Cell-Year Level

Panel Results: Cell Year Level

Post-2004: Satellite technology improves enforcement

Outcome = Level of max NDVI in year

Sample: 8,483 annual observations from 1982-2010 for grid cells within demarcated communities

Standard errors clustered by community & year

Robustness Checks

• Add community level trends as controls to cell-year panel model

• Limit propensity score estimation to 4 significant predictors, creating 37 pairs of communities; Use grid cells in this subsample of communities for cell-year panel model (n=404,405)

• Run cell-year panel model with full untrimmed, unmatched sample of cells (n=422,066)

• Run panel model using community-year data (n=1914)

Conclusions

• No clear, robust evidence of differences in deforestation attributable to the PPTAL project

• Project’s other aims included human rights protections

• Much lower rates of deforestation on indigenous lands in cross-section may not be related to land tenure status of these lands (or may be mediated through multiple, complex channels)

• Data deforestation rates in indigenous communities not necessarily available in early 90’s• Future programs may be able to target effectively

Next steps / future research

• Examining the impact on other outcomes – e.g. education outcomes using Brazil census data

• Identifying communities that have experienced conflicts or land disputes, where treatment effects may be larger

Extra Slides

Cell-year panel with community-level trends