More strictly protected areas are not necessarily more protective: evidence from Bolivia,
Costa Rica, Indonesia, and Thailand.
Paul J. Ferraro1*, Merlin M. Hanauer2, Daniehela A. Miteva3, Gustavo Javier CanavireBacarreza4, Subhrendu K. Pattanayak3, Katharine R. E. Sims5
1
Department of Economics, Andrew Young School of Policy Studies, Georgia State University,
Atlanta, Georgia, USA.
2
Department of Economics, Sonoma State University
3
Duke University, Durham, North Carolina, USA.
4
Centro de Investigaciones Económicas y Financieras – CIEF, Escuela de Economía y Finanzas,
Universidad EAFIT, Medellín, Colombia
5
Department of Economics and Environmental Studies Program, Amherst College, PO Box
5000, Amherst, MA 01002, USA
*Corresponding author:
Tel: 404-413-0201
[email protected]
Short Title: More strictly protected areas are not necessarily more protective
National parks and other protected areas are at the forefront of global efforts to protect
biodiversity and ecosystem services. However, not all protection is equal. Some areas are
assigned strict legal protection that permits few extractive human uses. Other protected area
designations permit a wider range of uses. Whether strictly protected areas are more effective in
achieving environmental objectives is an empirical question: although strictly protected areas
legally permit less anthropogenic disturbance, the social conflicts associated with assigning strict
protection may lead politicians to assign strict protection to less threatened areas and may lead
citizens or enforcement agents to ignore the strict legal restrictions. We contrast the impacts of
strictly and less strictly protected areas in four countries using IUCN designations to measure de
jure strictness, data on deforestation to measure outcomes, and a quasi-experimental design to
estimate impacts. On average, stricter protection reduced deforestation rates more than less strict
protection, but the additional impact was not always large and sometimes arose because
of where stricter protection was assigned rather than regulatory strictness per se. We also show
that, in protected area studies contrasting y management regimes, there are y2 policy-relevant
impacts, rather than only y, as earlier studies imply.
1
Introduction
National parks and other protected areas are at the forefront of global efforts to protect
biodiversity and ecosystem services. Understanding how these reserves affect environmental and
social outcomes is thus crucial to building the evidence base for conservation policy. Recently,
scholars have made advances in attempting to isolate the causal effects of protected areas
separately from confounding factors that jointly affect where protected areas are placed and the
outcomes of interest (e.g., Andam et al., 2008; Andam et al., 2010; Sims, 2010; Ferraro &
Hanauer, 2011; Ferraro et al., 2011; Joppa and Pfaff, 2011; Nelson & Chomitz, 2011).
Although these analyses have shed light on protected area impacts, most treat
“protection” is if it were homogenous. In practice, however, protected areas are heterogeneous in
their management objectives and the human uses that they permit. Protected areas that severely
restrict human uses have been a focal point of controversy in the “fortress conservation” debates
(Brockington, 2002). Thus understanding how their environmental and social effects compare to
less strictly, and often less controversial, protected areas is important. This understanding would
also be a useful input into recent debates about “degazetting” and “downgrading” protected areas
to less strictly protected or unprotected status (Mascia and Pailler, 2010).
As noted by Andam et al. (in press), theory is not clear about the way in which
environmental impacts vary with the strictness of protection. Holding all other attributes equal,
stricter protection implies a lower probability of anthropogenic disturbance and a higher
probability of restoration. However, the social conflicts that strictly protected areas can engender
may lead to lower overall enforcement or compliance with protected area regulations. Moreover,
establishing strictly protected areas is politically more difficult on accessible, productive lands
and may often be guided by aesthetics or other criteria (Dudley and Stolten 2012) that favor
2
more remote, less productive lands. For example, Joppa and Pfaff (2009) show that the stricter
the management category of tropical forest protected areas, the more likely the area is found on
“higher and steeper lands further away from roads and urban centers.” Thus the lands to which
strict protection is assigned may be, on average, less likely to be disturbed and more likely to
regenerate in the absence of protection compared to less strictly protected areas. This
phenomenon is called the “residual reserve problem” in the conservation planning literature
(Pressey and Bottrill, 2008). If less strict protection is more likely to be assigned to lands facing
higher human pressures, less strictly protected areas may result in greater avoided losses even
though they legally permit more disturbances than strictly protected areas.1
Without theory to provide guidance, the effect of different management categories is
ultimately an empirical question. One study (Nelson & Chomitz, 2011) estimates the average
effect of protected areas on fire density in the tropical forest biome conditional on the strictness
of protection as measured by International Union for Conservation of Nature (IUCN) protected
area categories from the World Database on Protected Areas (WDPA). Their results suggest that,
for Latin America and Asia, the average effect was larger in multiple-use protected areas than in
strictly protected areas (in Africa, the estimates are imprecise). However, the estimates may be
sensitive to hidden bias: most of the protected areas lack true baseline (pre-protection) measures
of the outcomes and the study uses measures of seven confounding variables to control for the
assignment of all kinds of protection across all countries.
1
The same theoretical ambiguity is present in the context of social impacts. Strictly protected areas may preclude
more productive uses than less strictly protected areas, but they also may be established on lower opportunity cost
lands and potentially induce more tourism business opportunities, more infrastructure and more ecosystem services.
Thus the net social effect of the management rules is difficult to predict based on theory alone.
3
Country-level studies that consider the particular aspects of the country’s protected area
assignment process may provide more accurate estimates of impacts. Three recent studies
examine the relative effectiveness of strictly protected areas, multi-use protected areas and
indigenous reserves in the Amazon (see also Nepstad et al. 2006). Pfaff et al. (2012) claim that
sustainable use areas and indigenous reserves generated more avoided deforestation, on average,
than strictly protected areas. Soares-Filho et al. (2010), using a different sample and (unusual)
empirical design, claim that strictly protected areas were more effective than sustainable use
areas (and indigenous reserves were the most effective). Nolte et al. (in press) generate
ambiguous results. In Thailand, Sims (2010) claims that wildlife sanctuaries and national parks
were more effective than less strictly protected forest reserves in generating additional forest
cover. In Costa Rica, Andam et al. (in press) estimate that less strictly protected areas induced
more regrowth on cleared forests than strictly protected areas, but this difference is not
statistically different from zero.
We estimate the effects of differing levels of protection on avoided deforestation in four
countries that are important for biodiversity conservation: Bolivia, Costa Rica, Indonesia
(Sumatra) and Thailand. We improve upon the aforementioned literature in four key ways. First,
with the exception of Andam et al. (in press), previously published studies do not clearly define
the estimated treatment effects (in particular, the counterfactual status is often vague): yet when
comparing across two types of governance regimes, there are at least four policy-relevant
treatment effects. Second, no study has used estimates of these effects to gain more insights into
the mechanisms through which legal restrictions operate. Third, previous studies do not test the
statistical significance of the differences in treatment effects across protection types (i.e., how
confident can one be that the differences in the point estimates are not simply the result of
4
sampling variability?). Fourth, by using consistent empirical designs and definitions of treatment
effects, our study facilitates cross-national comparisons.
Material and Methods
Data
Table 1 summarizes the data from the four countries. The Bolivian and Costa Rican data
span the entire country (Costa Rica data come from Andam et al., 2008). The Thai data cover the
north and northeastern part of the country (see Ferraro, Hanauer & Sims, 2011). The Indonesian
data cover the island of Sumatra. All pixels are forested at baseline and thus the relevant
outcome is whether the pixel is forested or deforested at the end of the study period. The
Supplementary Information (SI) file describes the data sources, definitions of “forested” and
“deforested,” and other data characteristics (SI Tables 1-2).
Attribute
Pixel Size
Full Sample
Strict Definition
Strict Pixels
Less-Strict Definition
Less-Strict Pixels
Forest Cover Period
Protection Period
Matching Weighting
Bolivia
2
100m
20,000
IUCN II
2,069
IUCN VI
(Integrated Mgmt)
1,721
1991-2000
1992-2000
Inverse Covariance
Table 1. Design Summary by Country
Country
Costa Rica
Indonesia (Sumatra)
2
3ha
20,000
IUCN Ia, II and IV
1,414
1km
26,154
IUCN Ia, II and IV
2,271
IUCN VI
IUCN VI
1,395
1960-1997
1973-1980
Mahalanobis
441
2000-2006
prior to 2000
Mahalanobis
Thailand
30m2
20,000
IUCN I and II
2,796
IUCN VI
(Forest Reserves)
8,958
1973-2000
prior to 1985
Inverse Covariance
Defining Strictness of Protection
We cannot observe de facto enforcement of regulations. Instead, we define the strictness
of protection based on de jure criteria; i.e., what the law says the regulations are. As noted above,
de facto enforcement may differ across protected areas. For example, according to one report
5
(ICEM, 2003), more than 500,000 people may have been living inside wildlife sanctuaries and
national parks in Thailand, contrary to the regulations for these reserves.
Following previous studies (e.g., Soares-Filho et al., 2010; Nelson & Chomitz, 2011;
Joppa and Pfaff, 2011; Pfaff et al., 2012), we define the “strictness” of protection in the Bolivia,
Costa Rica and Sumatra analyses by using the six IUCN management categories (World
Database on Protected Areas; WDPA). These categories use standardized definitions and imply
decreasing strictness of regulations as the category number increases (IUCN, 1994). We define
protected areas as “strictly protected” if they fall into categories I – IV, and as “less strictly
protected” if they fall into categories V-VI. The former category includes strict nature reserves,
wildlife refuges and national parks, and the latter includes multiple-use protected areas and
integrated management areas. These categories match closely the de jure protected area rules in
the three countries (SERNAP, 2007; Evans, 1999; Jepson 2001). In Thailand, we define wildlife
sanctuaries and national parks as “strictly protected” (IUCN I-II) and forest reserves as “less
strictly protected.” Thailand’s forest reserve area statutes do not match exactly with official
IUCN categories, but they are, in practice, multiple-use reserves, i.e., IUCN category VI (ICEM
2003, Emphandhu and Chettamart 2003, Fujita 2003, FAO 2009). Thus the analyses in all four
sites contrast protected areas that strongly restrict human use to protected areas that allow
extractive uses.
Estimands
To precisely describe the relevant treatment effects, we introduce some notation. Let
Ds=1 if the forest parcel is strictly protected, Dls=1 if it is less strictly protected, and Ds= Dls=
0 if it is unprotected. Let Y = 1 if the forest is deforested during the study period, and Y = 0 if it
remains forest. Thus, for every parcel, there are three potential outcomes: (1) Ys, the potential
6
deforestation under strict protection, (2) Yls, the potential deforestation under less strict
protection, and (3) Y0, the potential deforestation under no protection. Given three treatment
states and three potential outcomes, each country analysis has three observable outcomes and six
counterfactual outcomes.
For each country, we estimate four average treatment effects on the treated (ATT)2:
(1) ATTs,0=E(Ys- Y0| Ds=1); the expected difference between deforestation on strictly protected
forests and deforestation on strictly protected forests had they instead not been protected at
all; (2) ATTls,0=E(Yls- Y0| Dls=1); the expected difference between deforestation on less
strictly protected forests and deforestation on less strictly protected forests had they instead not
been protected at all; (3) ATTs,ls=E(Ys-Yls| Ds=1); the expected difference between
deforestation on strictly protected forests and deforestation on strictly protected forests had they
instead been less strictly protected; and (4) ATTls,s=E(Yls-Ys| Dls=1); the expected difference
between deforestation on less strictly protected forests and deforestation on less strictly
protected forests had they instead been strictly protected. All ATTs are measures of avoided
deforestation.
With the exception of Andam et al. (forthcoming), the studies cited in the Introduction
are not clear on what treatment effects they are estimating and how they test for differences
across effects. Their text suggests they estimate ATTs,0 and ATTls,0 and then informally contrast
their magnitudes and p-values. The studies do not attempt to differentiate among three reasons
for any observed differences: sampling error, the strictness of protection, or the underlying
2
As noted by Andam et al. (2008), the average treatment effect (ATE), which is the mean effect for a randomly
chosen parcel from the population of forested parcels, is not policy-relevant given that protection is not assigned to
randomly chosen parcels. The more policy-relevant treatment effect is the average effect of protection on the parcels
actually assigned protection (ATT).
7
characteristics of the land protected (i.e., strictly and less-strictly protected lands may differ in
characteristics that affect potential deforestation in the presence and absence of a given level of
protection; for example, holding protection strictness constant, protection placed on lessthreatened forests would generate less avoided deforestation).
The other two treatment effects, ATTs,ls and ATTls,s, also address policy-relevant
questions: how much different would deforestation on strictly protected forests have been had
these forests instead been less strictly protected, and vice-versa? Estimating the two additional
treatment effects allows one to compare strict and less strict protected areas with similar
covariate distributions. Thus by combining estimates of ATTs,ls and ATTls,s with other data and
statistical tests, one can also elucidate the relative contributions to ATTs,0 and ATTls,0 from the
strictness of protection and the location of protection.
Empirical Design
To estimate the counterfactual deforestation rates E(Y0| Ds=1) and E(Y0| Dls=1), we use a
quasi-experimental matching design like that used in Andam et al. (2008) and others. The
matching algorithms reweight the unprotected forest parcels to create a control group that, on
average, is observably similar to the protected parcels in the distributions of important baseline
covariates known to jointly affect the placement of protected areas and deforestation in the
absence of protection (i.e., factors known to affect land use). The baseline characteristics
selected for each country are described in the SI.
Under the assumption that there are no systematic unobservable differences among the
matched protected and unprotected parcels in characteristics that affect deforestation in the
absence of protection, the expected deforestation of the matched unprotected parcels represents
8
the expected counterfactual deforestation of the protected parcels had they not been protected.
Thus the difference between deforestation on the protected and matched unprotected parcels is
an unbiased estimator of ATTs,0 and ATTls,0 (for more conceptual detail, see Ferraro 2009). The
same approach is used to estimate ATTs,ls and ATTls,s (the matched control parcels are from the
other category).
For each country analysis, we select the matching algorithm that achieves the best
covariate balance after matching. The final algorithms chosen use (single) nearest-neighbor
covariate matching with replacement, but each analysis uses a different weighting approach (see
final row of Table 1). Multiple measures of the differences in the covariate distributions between
protected and unprotected parcels are used to measure covariate balance (SI Tables S3-S18). We
further control for bias that can remain after matching in finite samples by using a post-matching
bias-correction procedure (Abadie & Imbens, 2006). A final source of potential bias comes from
spillovers, both positive and negative, whereby protection affects forest cover in matched control
units (Andam et al. 2008). We find no evidence of spillovers, on average, in our study sites (SI
Table S19).
To characterize the precision of our estimates, we calculate heteroskedasticity robust
standard errors (Imbens & Wooldridge, 2009; Abadie & Imbens, 2006; Abadie et al., 2004).
These standard errors allow for heteroskedasticity both within and across treatment arms by
calculating conditional variances via a secondary matching algorithm that matches units within
treatment arms (treated units to treated units, etc.). To test the null hypothesis that ATTs,0 =
ATTls,0, we use Welch’s t-test.
9
Results
Table 2 presents the estimates of the four treatment effects for each of the four countries
(see Table S20 for breakdown by treated and control units). If the estimated ATT is less than
zero, the treatment reduced deforestation compared to the counterfactual condition; i.e.,
generated avoided deforestation. If ATTls,0 - ATTs,0 is greater than zero (first row, third column
where t-statistic is presented in brackets), more strictly protected areas reduced deforestation by a
greater amount than less strictly protected areas did (i.e., generated more avoided deforestation).
To illustrate the importance of estimating all four estimands and the statistical
significance of the differences between ATTls,0 and ATTs,0 , consider the Bolivia results. Strict
protection reduced deforestation on strictly protected forests by, on average, an additional 2.3
percentage points (p<0.01) compared to their counterfactual deforestation with no protection
(ATTs,0): in other words, 2.3% of the strictly protected forests would have been deforested had
they not been strictly protected. Less strict protection reduced deforestation on the less strictly
protected forests by an additional estimated 1.3 percentage points on average (p>0.10) (ATTls,0).
The difference between these two estimates is small (one percentage point) and not statistically
different from zero (p>0.10). However, were strictly protected areas instead assigned less strict
protection (ATTs,ls), the deforestation observed would have been, on average, 2 percentage points
higher (p<0.05). In contrast, assigning stricter protection to less strictly protected areas (ATTls,s)
would have had little impact on average (two-tenths of a percentage point; p>0.10).
Which areas, on average, experienced more avoided deforestation (ATTs,0 vs. ATTls,0)?
Comparing avoided deforestation on strictly protected areas (ATTs,0) to the avoided
deforestation on less strictly protected areas (ATTls,0), our estimates imply that more strictly
10
protected areas reduced deforestation by a greater amount in three countries (see first row of
each country panel). Costa Rica, Sumatra and Thailand’s strictly protected areas experienced an
estimated 10-13 percentage points less deforestation than less strictly protected areas. In Bolivia,
the difference between the estimated treatment effects is small and statistically insignificant.
How different, on average, would avoided deforestation have been on strictly protected areas
had they instead been less strictly protected (ATTs,ls)?
The estimates imply that assigning forests to stricter protection, instead of less strict
protection, reduced deforestation in three of the four sites: Bolivia, Sumatra and Thailand (see
second row of each country panel). In Costa Rica, the estimated effect is similar in sign and
magnitude to the estimate in Bolivia, but we cannot reject the null hypothesis of zero ATT.
How different, on average, would avoided deforestation have been on less strictly protected
areas had they instead been more strictly protected?
The estimates imply that assigning a forested area to less strict protection, instead of more strict
protection, increased deforestation in three of the four countries: Costa Rica, Sumatra and
Thailand (see third row of each country panel). In Bolivia, the estimated effect is near zero and
we cannot reject the null hypothesis of zero ATT.
Selection or Restriction?
Although we only observe the de jure strictness of the regulations, we can combine the
results from Table 2 with information on the characteristics of the protected and unprotected
forested lands (Tables S3-S17) to clarify whether the results in Table 2 are driven by the
strictness of protection or by selection (i.e., where protected areas are located). Recall there are
11
Table 2. Heterogeneous Treatment Effects
Effect on Deforestation by Level of Protection
Strictness of Protection
Strict
Less Strict
(ATTs,0 )
(ATTls,0 )
Difference
Counterfactual:
No Protection
ATTs,ls
Bolivia
-0.023***
(0.005)
-0.013
(0.015)
0.010
[0.645]
NA
NA
-0.020**
(0.009)
NA
NA
0.002
(0.003)
ATTls,s
Counterfactual:
No Protection
ATTs,ls
Costa Rica
-0.167***
-0.036**
(0.033)
(0.028)
NA
NA
-0.032
(0.028)
NA
NA
0.046*
(0.027)
ATTls,s
Counterfactual:
No Protection
ATTs,ls
Indonesia (Sumatra)
-0.104***
-0.003
(0.053)
(0.022)
ATTs,ls
0.101***
NA
NA
-0.209***
(0.054)
NA
NA
0.190***
(0.028)
-0.008
(0.012)
0.131***
[4.882]
NA
-0.165***
(0.021)
ATTls,s
Counterfactual:
No Protection
0.131***
[3.027]
Thailand
-0.139***
(0.024)
NA
NA
NA
0.200***
ATTls,s
(0.013)
ATTx,y = Average Treatment Effect on the Treated for x category of
protection compared to the counterfactual y, where s=strictly protected,
ls=less strictly protected, and 0 = unprotected
(Abadie-Imbens heteroskedacticity robust standard errors)
[t statistic based on Welch's t -test; H0: ATTs,0 = ATTls,0 ]
***, **, * Significant at the 1%, 5% and 10% levels respectively
12
reasons for the estimated difference between ATTs,0 and ATTls,0: sampling error, differences in
the levels of strictness (the treatment), or differences in the underlying productivity and
accessibility of the land assigned to strict and less strict protection. Uncertainty about sampling
error is captured in the t-test statistic in Table 2. The underlying productivity and accessibility of
the land is held constant for the ATTs,ls and ATTls,s estimands, while only strictness is changed.
Combing this information yields insights into the relative contributions of legal
restrictions and selection to protected area effects on deforestation. For example, holding site
selection in Sumatra constant, changing strictly protected areas to less strictly protected status
would reduce the protected area impacts substantially (forgo 21 percentage points of avoided
deforestation), whereas moving less strictly protected areas to stricter protection status would
increase impacts substantially (19 percentage points increase in avoided deforestation).
Furthermore, strictly protected areas are, on average, located on no better lands in terms of
productivity and accessibility (i.e., no more threatened). These observations imply that the
strictness of protection is very important in explaining the differences between ATTs,0 and
ATTls,0. Following the same logic, we observe that, like in Sumatra, the strictness of protection is
the main driver of differences between ATTs,0 and ATTls,0 in Bolivia and Thailand. In contrast, a
combination of selection and, to a lesser extent, strictness drives differences in Costa Rica.
Discussion
Our results imply that the effects of regulatory strictness are heterogeneous within and
across countries: although greater strictness of a protected area’s IUCN management category is
often associated with greater levels of avoided deforestation, the differences in the effects across
management categories are not always large and can arise from differences in the way in which
13
the strictness of protection is spatially assigned (i.e., site selection), as well as the strictness per
se. Stricter is not necessarily better and more evidence is needed to guide policymakers in the
choice of protected area management categories.
To help build the evidence base, our study can be extended in seven ways. First, although
our results are sufficient to show differential effects conditional on the management category,
they are not sufficient to form a global picture of such heterogeneity. Our sample was one of
convenience rather than representation. Our study should be replicated in other nations and
extended to include other environmental outcomes (e.g., degradation, regrowth; Andam et al. in
press), biomes and time periods. Second, to understand the full effects of protected areas, the
effects of management categories on social outcomes should also be investigated (Sims, 2010).
One can explore then explore way in which the environmental and social impacts of different
management categories vary conditional on observable characteristics of the landscape and the
people (Ferraro et al., 2011).
Third, our study design should be extended to contrasts of community-managed,
government-managed and co-managed ecosystems (Somanathan et al. 2009). Fourth, there is a
glaring lack of cost data in the literature on environmental impact evaluations. Even if more
strictly protected areas generate greater avoided deforestation, they may be less cost-effective.
Fifth, the way in which ex post impact studies like ours can inform ex ante conservation planning
exercises to site new protected areas needs further exploration (Pressey et al. 2007). Sixth, our
study only looks at de jure regulations rather than de facto enforcement, yet theory points to the
strong effects that enforcement can have on outcomes (e.g., Albers, 2010; Robinson et al., 2011).
Future studies could improve on our design through collaborations among scholars and
practitioners that help identify de facto enforcement levels. Without these finer resolution data,
14
decomposing any observed heterogeneous effects into their constituent mechanisms will be
difficult. Yet probing the heterogeneity of conservation actions and the mechanisms through
which they operate is a crucial step in building the evidence base in environmental policy, an
effort that Miteva et al. (2012) call Conservation Evaluation 2.0.
References
Abadie, A, D Drukker, J Herr, & G Imbens. 2004. Implementing matching estimators for
average treatment effects in Stata. Stata Journal 4: 290–311.
Abadie, A, & G Imbens. 2006 Large sample properties of matching estimators for average
treatment effects. Econometrica. 74: 235–267.
Albers, HJ. 2010. Spatial modeling of extraction and enforcement in developing country
protected areas. Resource and Energy Economics 32:165-179.
Andam, K, PJ Ferraro, & M Hanauer. In Press. The Effects of Protected Area Systems on
Ecosystem Restoration: a quasi-experimental design to estimate the impact of Costa Rica’s
protected area system on forest regrowth. Conservation Letters.
Andam, KS, PJ Ferraro, A Pfaff, GA Sanchez-Azofeifa, & J Robalino. 2008. Measuring the
Effectiveness of Protected Area Networks in Reducing Deforestation. Proceedings of the
National Academy of Sciences 105: 16089-16094.
Andam, K, PJ Ferraro, KE Sims, A Healy, & MB Holland. 2010. Protected Areas Reduced
Poverty in Costa Rica and Thailand. Proceedings of the National Academy of Sciences 107:
9996-10001.
Brockington, D. 2002. Fortress conservation: the Preservation of the Mkomazi Game Reserve,
Tanzania (Indiana U Press, Bloomington, IN).
15
Canavire-Bacarreza, G & MM Hanauer. 2013. Estimating the impacts of Bolivia’s protected
areas on poverty. World Development 41: 265-285.
Dudley, N & S Stolton. 2012. Protected landscapes and wild biodiversity, Volume 3 in the
Values of Protected Landscapes and Seascapes Series, Gland, Switzerland: IUCN.
Emphandhu, D & S Chettamart. 2003. 2003. Thailand’s Experience in Protected Area
Management. Faculty of Forestry, Kasetsart University, Bangkok, Thailand.
Evans, S. 1999. The Green Republic: A conservation history of Costa Rica. University of Texas
Press: Austin, TX.
Food and Agriculture Organization (FAO) Regional Office for Asia and the Pacific. 2009.
Thailand Forestry Outlook Study II. Working Paper No. APFSOSII, Bangkok, Thailand.
Ferraro, PJ. 2009. Counterfactual thinking and impact evaluation in environmental policy. In
Special Issue on Environmental Program and Policy Evaluation, M. Birnbaum & P. Mickwitz
(Eds.). New Directions for Evaluation 122: 75–84.
Ferraro, PJ & M Hanauer. 2011. Protecting Ecosystems and Alleviating Poverty with Parks and
Reserves: ‘win-win’ or tradeoffs? Environmental and Resource Economics 48:269–28.
Ferraro, PJ, M Hanauer & KE Sims. 2011. Conditions Associated with Protected Area Success in
Conservation and Poverty Reduction. Proceedings of the National Academy of Sciences 108:
13913-13918.
Fujita, W. 2003. Dealing with Contradictions: Examining National Forest Reserves in Thailand.”
Southeast Asia Studies 41: 206-238.
Imbens, G, & J Wooldridge. 2009. Recent developments in the econometrics of program
evaluation. Journal of Economic Literature 47: 5–86.
16
International Center for Environmental Management (ICEM). 2003. Thailand National Report
on Protected Areas and Development. Review of Protected Areas and Development in the Lower
Mekong River Region (Indooroopilly, Queensland, Australia).
International Union for the Conservation of Nature (IUCN). 1994. Guidelines for Protected
Areas Management Categories (IUCN, Gland, Switzerland).
Jepson, P. 2001. Biodiversity and Protected Area Policy: Why Is It Failing in Indonesia?
Dissertation Thesis, University of Oxford.
Joppa, LN & A Pfaff. 2009. High and far: biases in the location of protected areas. PLoS ONE 4:
e8273
Joppa, LN & A Pfaff. 2011. Global protected area impacts. Proceedings of the Royal Society B
278:1633–8.
Mascia, MB & S. Pailler. 2011, Protected area downgrading, downsizing, and degazettement
(PADDD) and its conservation implications. Conservation Letters 4: 9–20.
Miteva, D, SK Pattanayak, & PJ Ferraro. 2012. Evaluation of Biodiversity Policy Instruments:
What works and what doesn’t? Oxford Review of Economic Policy 28: 69-92.
Nelson A, & KM Chomitz. 2011. Effectiveness of strict vs. multiple use protected areas in
reducing tropical forest fires: a global analysis using matching methods. PLoS ONE 6: e22722.
Nepstad D, S Schwartzman, B Bamberger, M Santilli, D Ray, P Schlesinger, P Lefebvre, A
Alencar, E Prinz, G Fiske & A Rolla. 2006. Inhibition of Amazon Deforestation and Fire by
Parks and Indigenous Lands. Conservation Biology 20: 65–73.
Pfaff A, J Robalino, LD Herrera, E Lima, & C Sandoval. 2012. Estimating Avoided
Deforestation by Protection Type: can governance plus location equal forest plus livelihoods?
Duke Working Paper (Duke University, Durham, NC).
17
Pressey, RL & M Bottrill. 2008. Opportunism, threats, and the evolution of systematic
conservation planning. Conservation Biology 22: 1340-1345.
Pressey, RL, M Cabeza, ME Watts, RM Cowling, & KA Wilson, 2007. Conservation planning in
a changing world. Trends in Ecology and Evolution 22: 583-592.
Robinson, EJZ, & RB Lokina. 2011. A spatial-temporal analysis of the impact of access
restrictions on forest landscapes and household welfare in Tanzania. Forest Policy and
Economics 13: 79-85.
Servicio Nacional de Áreas Protegidas (SERNAP). 2007. Informe sobre el sistema de áreas
protegidas. SERNAP: La Paz.
Sims, KRE. 2010. Conservation and development: evidence from Thai protected areas. Journal
of Environmental Economics and Management 60:94-114.
Soares-Filho B, P Moutinho, D Nepstad, A Anderson, H Rodrigues, R Garcia, L Dietzch, F
Merry, M Bowman, L Hissa, R Silvestrini & C Maretti. 2010. Role of Brazilian Amazon
protected areas in climate change mitigation. Proceedings of the National Academy of Sciences
107:10821–10826.
Somanathan, E, R Prabhakar, & SM Bhupendra. 2009. Decentralization for cost-effective
conservation. Proceedings of the National Academy of Sciences 106: 4143-4147.
18
Supporting Information for Ferraro et al. “More strictly protected areas are not necessarily more
protective: evidence from Bolivia, Costa Rica, Indonesia, and Thailand”
Data
Sources
Bolivia- Spatial data for forest cover, protected areas, roads, cities and digital elevation models
were obtained from Conservation International, Bolivia (see Canavire-Bacarreza and Hanauer 2013 for
details).
Costa Rica- Spatial data for forest cover comes from Andam et al. (2008). Data on protected
areas (source: National System of Conservation Area Office, Ministry of Environment and Energy, 2006),
and the locations of major cities were provided by the Earth Observation Systems Laboratory, University
of Alberta, Canada. Other GIS layers are land use capacity (source: Ministry of Agriculture) and roads
digitized from hard copy maps for 1969 (source: Instituto Geográfico Nacional, Ministerio Obras Publicas
y Transporte).
Indonesia- Spatial data for forest cover comes from the TREES 1998-2000 dataset, generated by
the EU Institute for Environment and Sustainability, and GlobCover 2006, generated by the ESA/ESA
GlobCover Project. Spatial data for major cities, ports, slope, and soil properties were generated from
publically available geospatial datasets. Socioeconomic data come from the PODES census datasets.
Thailand- Spatial data for forest cover are based on Landsat MSS images interpreted by the
Tropical Rain Forest Information Center (Michigan State University) and Landsat TM images interpreted
by the Thai Royal Forestry Department (courtesy of Marc Souris, See Andam et al. 2010 and Sims 2010
for full details). Data on forest reserves comes from the Thailand Environment Research Institute’s
19
“Thailand on a Disc” (1996). Spatial data for protected areas, slope, elevation, roads, cities and rivers
were generated from publically available geospatial data sets (see Ferraro et al. 2011 for full details).
Unit of Analysis
In all analyses the relevant unit is a land parcel, or pixel. Following Andam et al. 2008, the baseline forest
layer in each country is broken into pixels (100m2 pixels for Bolivia, 3ha pixels for Costa Rica, 30m2
Thailand,3 and 1km2 pixels for Indonesia, see Table S1). To mitigate the potential for spatial correlations
we take a random sample of pixels from the initial forest layers (20,000 pixel sample in Bolivia, Costa
Rica and Thailand; proportional samples from protected and unprotected forest in Indonesia result in a
final sample of 26,154 pixels). The resulting sample contains forested pixels that subsequently were
either protected or left unprotected.
Protection and Outcome
Using ArcMap 10.x we overlay the relevant protected areas from each country to determine: (1) if the
forested pixels from our respective samples are considered protected, and (2) for those pixels that fall
within a protected area, the relative level of protective strictness to which those pixels were exposed.
Due to the timing of the establishment of protected areas, and availability of data, the relevant time
period for the evaluation of the impacts of protected areas differs for each country. Therefore, baseline
years and the periods of time considered differ across countries.
For each country the outcome of interest (Y) is change in forest cover between baseline and
end-period. In other words, each pixel receives two binary indicators: (1) 0 or 1 to indicate absence or
presence of forest at baseline, and (2) 0 or 1 to indicate absence or presence of forest at end-period.
3
2
2
The precision with which forest cover was measured in Thailand differed between 1973 (60m ) and 2000 (30m ).
Because our designation of deforestation depends on the precision of the 2000 data, we use the 30m 2 resolution
in the description of our data set.
20
Therefore, due to the fact that, by construction, all parcels are forested at baseline, Y=0 indicates that
the pixel has remained forested, whereas Y=1 indicates that the pixel has been deforested.
Bolivia- We measure change in forest cover 1991-2000, during which time forest is defined by
>30% canopy cover. 3,790 forested pixels were protected between 1992 and 2000. We define “strict”
protected areas as national parks (2,069 pixels) and “less-strict” protected areas as integrated
management areas (1,721 pixels)
Costa Rica- Using the same sample as Andam et al. 2008, we measure change in forest cover
between 1960 and 1997, during which time forest is defined as >80% canopy cover. The final sample
comprises 13,100 forested pixels (see Andam et al. 2008 for exclusion restrictions), 2,809 of which were
considered protected between 1973 and 1980 (see Andam et al. 2008 for choice of time period). We
define “strict” protected areas as International Union for the Conservation of Nature (IUCN) categories
Ia, I, II and IV (1,414 pixels) and “less-strict” protected areas as IUCN category VI (1,395 pixels).
Indonesia- We measure change in forest cover between 2000 and 2006, during which time the
definition of forest ranges from >15% to >70% depending on the category of canopy. Our final sample
comprises forested 26,154 pixels (pixels within protected areas established 2000-2006, and pixels in
villages adjacent to protected areas are excluded from the analyses), 2,712 of which were considered
protected prior to 2000. We define “strict” protected areas as IUCN categories Ia, II and IV (2,271 pixels)
and “less-strict” protected areas as IUCN category VI (441 pixels).
Thailand- Using the same sample as Ferraro et al. 2011, we measure change in forest cover
between 1973 and 2000, during which time a hybrid method of visual analysis and digital data are used
to define forest cover (see Skole, Salas and Silapathong (1998) for further details). Our final sample
comprises 16,417 forested pixels (pixels within protected areas established after 1985 are excluded
from the analyses), 11,754 of which were considered protected prior to 1985. We define “strict”
21
protected areas as wildlife sanctuaries and national parks (IUCN I and II, 2,796 pixels) and “less-strict”
protected areas as forest reserves (IUCN IV, 8,958 pixels).
Covariates
The effectiveness of a protected area in preventing deforestation depends on the amount of
deforestation that would have taken place had the protected area not been established. This
counterfactual depends on the deforestation pressure facing a particular land parcel. For instance, a
parcel that lies adjacent to a road and 1km from a major city faces greater pressure (probability) of
deforestation, on average, than a similar parcel that is far from both a road and major city. Therefore, a
protected area (regardless of strictness) composed of parcels with high deforestation pressure will be
more effective than a protected area composed of parcels with low deforestation pressure. Thus,
although the specific covariates differ across countries, all covariates aim to capture factors that jointly
affect deforestation pressure and the selection process through which protected areas were established
(see Andam et al 2008 for theory underlying the choice of covariates). See Table S1 for a list of
covariates used for each country-specific analysis and Table S2 for a description of unique covariates
used in the Indonesia (Sumatra) analysis (covariates that are not described in Andam et al. 2008 or
Ferraro et al. 2011).
Heterogeneous Treatment Analyses
All countries in our analyses contain protected areas with different levels of land-use restrictions.
Because protected area conservation definitions and implementation vary by country, we attempt to
define the relative strictness of a protected area in terms of the amount of human intervention allowed
within the protected areas. In each country, a protected area is classified as “strict” if little or no
extractive activities are allowed. Conversely, a protected area is classified as “less-strict” if some degree
of managed extraction is permitted.
22
Recall from the main text there are four estimands in which we are interested: (1) ATTs,0=E(YsY0| Ds=1); (2) ATTls,0=E(Yls- Y0| Dls=1) ; (3) ATTs,ls=E(Ys-Yls| Ds=1) ; and (4) ATTls,s=E(Yls-Ys| Dls=1) .
Matching followed by a calculation of the mean differences in outcomes of the treated and matched
control units are used to estimate the four ATT estimands. The choice of matching algorithm is based on
the algorithm that achieves greatest post-match balance across the covariates of interest. If matching is
effective, the covariate balance measures in Tables S3-S17 should move dramatically toward zero after
matching. All matching is followed by post-match regression bias-adjustment (Abadie and Imbens 2006).
To estimate (1) and (2), we run two separate matching analyses using the full set of unprotected parcels
as potential matches in each analysis for each country. In each analysis we exclude the complimentary
set of protected units in order to prevent matching across protected units of disparate strictness
designations. The precision of each analysis is calculated using heteroskedasticity robust standard errors
(Abadie and Imbens 2006). These standard errors allow us to test the statistical proximity of the point
estimates (see Table 2) using a Welch’s t-test.
In the second analysis (for (3) and (4) from above) we match strict protected pixels to less-strict
protected pixels, and then match less-strict protected pixels to strictly protected pixels. Although the
first analysis described in the previous paragraph controls for differences in observable characteristics
between protected and unprotected pixels, the covariates that determine deforestation pressure (and
therefore the effectiveness of protected areas) may differ between strict and less-strict protected areas.
If this is the case (see balance Tables S3-S18) then the relative effectiveness of protected areas inferred
from the first analysis may be driven by the underlying land characteristics rather than the strictness of
protection. For instance, if high pressure land falls disproportionately under strict protection, then we
might observe differentially higher avoided deforestation under strict protection using the first analysis.
However, without controlling for observable differences between the two types of protection (if
differences exist) then differences in avoided deforestation estimates might be falsely attributed to
23
protection strictness. Therefore, this analysis addresses the question, “what would deforestation in
“strict” protected areas have been had they been protected but with “less-strict” land-use restrictions?”
or, similarly, “what would deforestation in “less-strict” protected areas have been had they been
protected but with “strict” land-use restrictions?”
Spillover
If the establishment of protected areas simply pushes deforestation, that would have occurred in the
absence of protection, to surrounding forest (i.e., leakage) then the measurement of protected areas’
impacts could be confounded. Andam et al. (2008) found no evidence of deforestation spillover around
protected areas in Costa Rica. For Bolivia, Indonesia and Thailand we run a similar analysis to Andam et
al. (2008) to test for local spillover effects. Using GIS we select all unprotected pixels that lie within a
5km buffer of any protected area and designate them as “treated.” In a new matching analysis, we
match these “treated” buffer pixels to observably similar unprotected pixels that lie outside of the
buffer. A statistically significant and positive ATT estimate would provide evidence that protected areas
are simply pushing deforestation to their borders.
The results from the spillover analyses can be seen in Table S19. We find no evidence of local
deforestation spillovers in Bolivia, Indonesia or Thailand. These findings are concordant with the findings
of Andam et al. (2008) from Costa Rica. We acknowledge the inability of our local spillover analyses to
capture potentially more complex spillover scenarios. However, more complex, distant scenarios would
be exceedingly difficult to model and likely impossible to measure.
References
Abadie, A., & Imbens, G. (2006). Large sample properties of matching estimators for average treatment
effects. Econometrica, 74(1), 235–267.
24
Andam, K. S., Ferraro, P. J., Pfaff, A., Sanchez-Azofeifa, G., & Robalino, J. A. (2008). Measuring the
effectiveness of protected area networks in reducing deforestation. Proceedings of the National
Academy of Sciences, 105(42), 16089–16094.
Andam, K., Ferraro, P., Sims, K., Healy, A., & Holland, M. (2010). Protected areas reduced poverty in
Costa Rica and Thailand. Proceedings of the National Academy of Sciences, 107(22), 9996.
Ferraro, P. J., Hanauer, M. M., & Sims, K. R. (2011). Conditions associated with protected area success in
conservation and poverty reduction. Proceedings of the National Academy of Sciences., 108(34), 13913–
13918.
Skole, D. L., W. A. Salas, and C. Silapathong. 1998. Interannual Variation in theTerrestrial Carbon Cycle:
Significance of Asian Tropical Forest Conversion to Imbalances in the Global Carbon Budget.Pp. 162-186
in Galloway, J. N. and J. M. Melillo, Eds, Asian Change in the Context of Global Change. Cambridge:
Cambridge University Press.
Sims, K. R.E. (2010). Conservation and development: Evidence from Thai protected areas. Journal of
Environmental Economics and Management, 60(2), 94–114.
25
Item
Bolivia
Pixel Size
Full Sample
Protected Pixels
Strict Definition
Strict Pixels
Less-Strict Definition
Less-Strict Pixels
Forest Cover Period
Protection Period
Matcing Metric
Covariates
Table S1. Full Design Summary by Country
Country
Costa Rica
Indonesia
2
2
Thailand
100m
20,000
3,790
National Parks
2,069
Integrated Man.
1,721
1991-2000
1992-2000
Inverse Covariance
3ha
20,000
2,809
IUCN Ia, II and IV
1,414
IUCN VI
1,395
1960-1997
1973-1980
Mahalanobis
1km
26,154
2,712
IUCN Ia, II and IV
2,271
IUCN VI
441
2000-2006
prior to 2000
Mahalanobis
30m2
20,000
11,754
IUCN I and II
2,796
IUCN IV
8,958
1973-2000
prior to 1985
Inverse Covariance
Euclidian Distance to
Road (km)
Euclidian Distance to
Major City (km)
Euclidian Distance to
Forest Edge (km)
Slope (degrees)
Euclidian Distance to
Road (km)
Euclidian Distance to
Major City (km)
Euclidian Distance to
Forest Edge (km)
High Land Use Capacity
Euclidian Distance to
Major River
Euclidian Distance to
Major City (km)
Euclidian Distance to
Nearest Port (km)
Aridity Index
Euclidian Distance to
Road (km)
Euclidian Distance to
Major City (km)
Euclidian Distance to
Forest Edge (km)
Elevation (m)
Elevation (m)
Medium-High Land Use
Capacity
Medium-Low Land Use
Capacity
Slope (degrees)
Slope (degrees)
Elevation (m)
Euclidian Distance to
Major River
Socioeconomic
Measures (see Table S2)
Soil Measures (see Table
S2)
* Thai analysis also includes exact matching on district.
Covariate
Clay
Conductivity
Organic Carbon
Average Slope
Baseline Forest
Poverty
Population Density
Table S2. Indonesia Soil and Socioeconomic Covariates
Description
Weight% of particles < 0.002mm (clay) in fine earth fraction in the topsoil, pixel-level
Measures the electrical conductivity, which proxies for salinity of the topsoil, pixel-level
Weight% of total organic carbon (Walkley-Black method) in the topsoil, pixel-level
Average slope (in degrees), village-level
Baseline forest as fraction of village area, village-level
Baseline poverty in a village, village-level
Baseline population density in a village, village-level
26
Covariate
Status
Distance to Forest Unmatched
Edge
Matched
Distance to Major Unmatched
City
Matched
Distance to Road Unmatched
Matched
Elevation Unmatched
Matched
Slope Unmatched
Matched
Table S3. Bolivia Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
1.570
0.926
0.644
0.195
1.570
1.512
0.058
0.016
243.277
216.943
26.334
0.096
243.277
228.956
14.321
0.046
32.153
42.726
-10.574
0.125
32.153
30.232
1.921
0.039
702.599
436.376
266.223
0.204
702.599
687.189
15.410
0.011
8.712
3.724
4.989
0.285
8.712
8.520
0.193
0.009
Mean eQQ
Difference
0.647
0.061
34.914
15.638
13.807
2.099
267.733
17.161
4.980
0.236
% Improve
Mean Diff.
0.00%
91.02%
0.00%
45.62%
0.00%
81.83%
0.00%
94.21%
0.00%
96.14%
Covariate
Status
Distance to Forest Unmatched
Edge
Matched
Distance to Major Unmatched
City
Matched
Distance to Road Unmatched
Matched
Elevation Unmatched
Matched
Slope Unmatched
Matched
Table S4. Bolivia Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.848
0.926
-0.078
0.026
0.848
0.818
0.029
0.012
320.804
216.943
103.861
0.357
320.804
292.542
28.262
0.080
60.777
42.726
18.051
0.199
60.777
56.546
4.232
0.054
557.650
436.376
121.273
0.096
557.650
539.948
17.702
0.013
6.956
3.724
3.232
0.190
6.956
6.737
0.219
0.012
Mean eQQ
Difference
0.084
0.037
103.837
29.093
21.041
4.304
127.336
22.673
3.226
0.283
% Improve
Mean Diff.
Covariate
Status
Distance to Forest Unmatched
Edge
Matched
Distance to Major Unmatched
City
Matched
Distance to Road Unmatched
Matched
Elevation Unmatched
Matched
Slope Unmatched
Matched
Table S5. Bolivia Strict to Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
1.569
0.848
0.721
0.373
1.569
1.449
0.120
0.055
243.220
320.800
77.580
0.343
243.220
248.430
5.210
0.025
32.145
60.777
28.632
0.648
32.145
33.034
0.889
0.029
702.600
557.650
144.950
0.163
702.600
705.670
3.070
0.003
8.708
6.956
1.752
0.126
8.708
6.737
1.971
0.007
Mean eQQ
Difference
0.719
0.129
77.702
18.892
29.147
1.563
144.250
34.529
3.226
0.283
% Improve
Mean Diff.
62.64%
72.79%
76.56%
85.40%
93.22%
83.36%
93.28%
96.90%
97.88%
-12.51%
27
Table S6. Bolivia Less-Strict to Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.848
1.569
0.721
0.373
0.848
0.837
0.011
0.007
320.800
243.220
77.580
0.343
320.800
308.860
11.940
0.051
60.777
32.145
28.632
0.648
60.777
57.671
3.106
0.059
557.650
702.600
144.950
0.163
557.650
562.010
4.360
0.006
6.956
8.708
1.752
0.126
6.956
6.712
0.244
0.019
Mean eQQ
Difference
0.719
0.039
77.707
19.821
29.147
4.946
144.250
18.189
3.226
0.283
% Improve
Mean Diff.
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Table S7. Costa Rica Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.014
0.205
-0.191
0.296
0.014
0.014
0.000
0.000
0.050
0.198
-0.148
0.222
0.050
0.050
0.000
0.000
0.071
0.507
-0.436
0.580
0.071
0.071
0.000
0.000
3.314
2.045
1.269
0.248
3.314
3.097
0.218
0.044
20.352
15.336
5.015
0.190
20.352
19.234
1.118
0.042
80.322
80.515
-0.193
0.002
80.322
81.448
-1.126
0.010
Mean eQQ
Difference
0.192
0.000
0.149
0.000
0.436
0.000
1.307
0.235
5.023
1.637
11.060
2.716
% Improve
Mean Diff.
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Table S8. Costa Rica Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.002
0.205
-0.203
0.319
0.002
0.002
0.000
0.000
0.007
0.198
-0.191
0.300
0.007
0.007
0.000
0.000
0.090
0.507
-0.417
0.547
0.089
0.089
0.000
0.000
2.393
2.045
0.349
0.072
2.391
2.332
0.059
0.015
14.315
15.336
-1.022
0.041
14.317
14.127
0.190
0.009
73.591
80.515
-6.923
0.071
73.638
74.383
-0.745
0.007
Mean eQQ
Difference
0.204
0.000
0.191
0.000
0.417
0.000
0.597
0.092
1.358
0.516
20.872
2.349
% Improve
Mean Diff.
Covariate
Status
Distance to Forest Unmatched
Edge
Matched
Distance to Major Unmatched
City
Matched
Distance to Road Unmatched
Matched
Elevation Unmatched
Matched
Slope Unmatched
Matched
Covariate
High Land Use
Capacity
Medium Land Use
Capacity
Medium-Low Land
Use Capacity
Distance to Forest
Distance to Road
in 1969
Distance to a
Major City
Covariate
High Land Use
Capacity
Medium Land Use
Capacity
Medium-Low Land
Use Capacity
Distance to Forest
Distance to Road
in 1969
Distance to a
Major City
98.47%
84.61%
89.15%
96.99%
86.07%
100.0%
100.0%
100.0%
82.9%
77.7%
-484.3%
100.0%
100.0%
100.0%
83.1%
81.4%
89.2%
28
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Table S9. Costa Rica Strict to Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.014
0.002
0.012
0.095
0.014
0.014
0.000
0.000
0.050
0.007
0.043
0.184
0.050
0.050
0.000
0.000
0.071
0.089
0.018
0.049
0.071
0.071
0.000
0.000
3.314
2.393
0.921
0.296
3.314
3.077
0.237
0.072
20.352
14.315
6.037
0.362
20.352
18.142
2.210
0.125
80.322
73.591
6.731
0.096
80.322
78.149
2.173
0.032
Mean eQQ
Difference
0.011
0.000
0.043
0.000
0.019
0.000
0.917
0.276
6.017
2.328
11.496
6.793
% Improve
Mean Diff.
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Table S10. Costa Rica Less-Strict to Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
0.002
0.014
0.012
0.095
0.002
0.002
0.000
0.000
0.007
0.050
0.043
0.184
0.007
0.007
0.000
0.000
0.089
0.071
0.018
0.049
0.089
0.089
0.000
0.000
2.393
3.077
0.684
0.296
2.393
2.393
0.000
0.018
14.315
20.352
6.037
0.362
14.315
14.243
0.072
0.005
73.591
80.322
6.731
0.096
73.591
71.773
1.818
0.025
Mean eQQ
Difference
0.011
0.000
0.043
0.000
0.019
0.000
0.917
0.276
6.017
0.662
11.496
4.384
% Improve
Mean Diff.
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Distance to River Unmatched
Matched
Elevation Unmatched
Matched
Table S11. Thailand Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
109.900
114.345
-4.445
0.040
109.900
111.404
-1.504
0.016
31.240
16.537
14.703
0.414
31.240
29.217
2.022
0.044
3.634
2.097
1.537
0.259
3.634
3.209
0.425
0.069
7.969
4.893
3.075
0.252
7.969
7.499
0.470
0.037
36.108
26.731
9.377
0.230
36.108
34.934
1.174
0.027
698.909
465.983
232.925
0.344
698.909
629.914
68.995
0.105
Mean eQQ
Difference
9.414
4.645
14.713
2.614
1.538
0.432
3.080
0.596
9.384
2.319
233.058
69.013
% Improve
Mean Diff.
Covariate
High Land Use
Capacity
Medium Land Use
Capacity
Medium-Low Land
Use Capacity
Distance to Forest
Distance to Road
in 1969
Distance to a
Major City
Covariate
High Land Use
Capacity
Medium Land Use
Capacity
Medium-Low Land
Use Capacity
Distance to Forest
Distance to Road
in 1969
Distance to a
Major City
Covariate
Distance to Major
City
Distance to Road
in 1962
Distance to Forest
Edge
Slope
100.00%
100.00%
100.00%
74.27%
63.39%
67.72%
100.00%
100.00%
100.00%
100.00%
98.81%
72.99%
66.2%
86.2%
72.3%
84.7%
87.5%
70.4%
29
Covariate
Distance to Major
City
Distance to Road
in 1962
Distance to Forest
Edge
Slope
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Distance to River Unmatched
Matched
Elevation Unmatched
Matched
Covariate
Distance to Major
City
Distance to Road
in 1962
Distance to Forest
Edge
Slope
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Distance to River Unmatched
Matched
Elevation Unmatched
Matched
Covariate
Distance to Major
City
Distance to Road
in 1962
Distance to Forest
Edge
Slope
Status
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Unmatched
Matched
Distance to River Unmatched
Matched
Elevation Unmatched
Matched
Table S12. Thailand Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
113.580
114.339
-0.759
0.007
113.907
114.016
-0.109
0.001
17.516
16.543
0.973
0.030
17.422
16.773
0.650
0.021
2.366
2.061
0.305
0.056
2.375
2.122
0.253
0.053
5.105
4.831
0.273
0.024
5.115
4.791
0.324
0.032
27.536
26.750
0.786
0.020
27.530
26.887
0.642
0.017
498.545
462.621
35.924
0.053
499.892
479.045
20.847
0.030
Mean eQQ
Difference
3.308
2.166
1.161
0.845
0.427
0.253
0.425
0.324
1.522
0.743
36.257
21.108
% Improve
Mean Diff.
Table S13. Thailand Strict to Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
109.900
114.240
4.340
0.073
109.900
108.850
1.050
0.021
31.240
17.401
13.839
0.556
31.240
28.031
3.209
0.113
3.634
2.348
1.286
0.332
3.643
3.188
0.455
0.107
7.969
5.027
2.942
0.376
7.969
7.405
0.564
0.069
36.108
27.524
8.584
0.319
36.108
34.912
1.196
0.045
698.910
497.340
201.570
0.498
698.910
625.160
73.750
0.194
Mean eQQ
Difference
8.361
6.103
13.843
3.357
1.286
0.447
2.944
0.705
8.585
2.763
201.570
73.801
% Improve
Mean Diff.
Table S14. Thailand Less-Strict to Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
114.240
109.900
4.340
0.073
110.930
110.400
0.530
0.009
17.401
31.240
13.839
0.556
17.547
18.626
1.079
0.058
2.238
3.634
1.396
0.332
2.467
2.960
0.493
0.139
5.026
7.967
2.941
0.376
5.126
5.903
0.777
0.121
27.524
36.108
8.584
0.319
26.906
28.275
1.369
0.061
497.340
698.910
201.570
0.498
501.300
565.980
64.680
0.159
Mean eQQ
Difference
8.360
6.350
13.843
2.149
1.286
0.502
2.944
0.941
8.585
2.713
201.570
67.211
% Improve
Mean Diff.
85.7%
33.3%
17.0%
-18.6%
18.3%
42.0%
75.81%
76.81%
64.62%
80.83%
86.07%
63.41%
87.79%
92.20%
64.68%
73.58%
84.05%
67.91%
30
Table S15. Indonesia Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
cted Plots
rol Plots
Mean
Difference
4058.80
8579.10
-4520.30
-0.23
4058.80
4085.60
-26.80
-0.01
110000.00
87047.00
22953.00
0.31
110000.00
100000.00
10000.00
0.15
44212.00
39504.00
4708.00
0.12
44212.00
41445.00
2767.00
0.08
746.28
413.89
332.39
0.48
746.28
734.52
11.76
0.02
12.84
13.46
-0.62
-0.05
12.84
13.17
-0.32
-0.03
0.10
0.14
-0.03
-0.12
0.10
0.11
-0.01
-0.03
1.64
2.16
-0.52
-0.23
1.64
1.58
0.06
0.04
11.53
5.39
6.14
0.71
11.53
11.07
0.45
0.05
84.75
66.88
17.88
0.54
84.75
83.08
1.67
0.06
0.45
0.50
-0.05
-0.14
0.45
0.46
-0.01
-0.03
29.95
52.77
-22.82
-0.10
29.95
43.44
-13.49
-0.11
Mean eQQ
Difference
4554.08
436.49
23139.09
6726.67
4715.81
4404.35
332.06
25.41
1.85
0.69
0.52
0.12
0.04
0.02
6.14
0.78
0.06
0.02
28.37
15.22
17.89
2.09
% Improve
Mean Diff.
Table S16. Indonesia Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
Status
cted Plots
rol Plots
Mean
Difference
Unmatched
3772.00
8579.10
-4807.10
-0.24
Matched
3772.00
3927.80
-155.80
-0.03
Unmatched 54776.00
87047.00
-32271.00
-0.55
Matched 54776.00
54897.00
-121.00
0.00
Unmatched 34273.00
39504.00
-5231.00
-0.19
Matched 34273.00
34464.00
-191.00
-0.01
Unmatched
1275.30
413.89
861.41
1.18
Matched
1275.30
1261.50
13.80
0.02
Unmatched
13.66
13.46
0.21
0.02
Matched
13.66
13.59
0.07
0.01
Unmatched
0.10
0.14
-0.04
-0.18
Matched
0.10
0.10
0.00
0.02
Unmatched
3.67
2.16
1.52
0.19
Matched
3.67
3.68
-0.01
0.00
Unmatched
14.56
5.39
9.17
1.28
Matched
14.56
14.53
0.02
0.00
Unmatched
81.61
66.88
14.74
0.41
Matched
81.61
81.88
-0.26
-0.01
Unmatched
0.46
0.50
-0.04
-0.11
Matched
0.46
0.46
0.00
0.00
Unmatched
31.96
52.77
-20.82
-0.09
Matched
31.96
31.83
0.12
0.00
Mean eQQ
Difference
5197.74
158.55
37000.95
384.19
8855.17
336.35
860.09
17.46
3.64
0.07
2.05
0.01
0.04
0.00
9.18
0.11
0.04
0.00
57.93
0.73
14.81
0.37
% Improve
Mean Diff.
Covariate
Status
Distance to River Unmatched
Matched
Distance to Port Unmatched
Matched
Distance to City Unmatched
Matched
Average Elevation Unmatched
Matched
Soil: Clay Unmatched
Matched
Soil: Conductivity Unmatched
Matched
Soil: Organic Unmatched
Content
Matched
Average Slope Unmatched
Matched
Baeline Forest Unmatched
Cover
Matched
Baseline Poverty Unmatched
Matched
Population Unmatched
Density
Matched
Covariate
Distance to River
Distance to Port
Distance to City
Average Elevation
Soil: Clay
Soil: Conductivity
Soil: Organic
Content
Average Slope
Baeline Forest
Cover
Baseline Poverty
Population
Density
99.40%
72%
41.20%
96.50%
47.30%
77.50%
89.20%
92.60%
90.60%
77.50%
40.90%
96.80%
99.60%
96.40%
98.40%
64.80%
98.30%
99.60%
99.70%
98.20%
97.60%
99.40%
31
Covariate
Distance to River
Distance to Port
Distance to City
Average Elevation
Soil: Clay
Soil: Conductivity
Soil: Organic
Content
Average Slope
Baseline Poverty
Population
Density
Baeline Forest
Cover
Covariate
Distance to River
Distance to Port
Distance to City
Average Elevation
Average Slope
Baeline Forest
Cover
Baseline Poverty
Population
Density
Table S17. Indonesia Strict to Less-Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
Status
cted Plots
rol Plots
Mean
Difference
Unmatched
4058.83
3772.04
286.78
0.04
Matched
3772.04
3927.80
-155.76
0.02
Unmatched 107695.20
54776.43
52918.79
0.74
Matched 54776.43
54896.60
-120.17
0.00
Unmatched 44211.67
34273.49
9938.18
0.19
Matched 34273.49
34463.76
-190.27
0.00
Unmatched
746.28
1275.28
-529.00
0.38
Matched
1275.28
1261.45
13.83
0.01
Unmatched
12.84
13.66
-0.82
0.04
Matched
13.66
13.59
0.07
0.00
Unmatched
1.64
3.67
-2.03
0.12
Matched
3.67
3.68
-0.01
0.00
Unmatched
0.10
0.10
0.01
0.03
Matched
0.10
0.10
0.00
0.01
Unmatched
11.53
14.56
-3.03
0.24
Matched
14.56
14.53
0.02
0.00
Unmatched
0.45
0.46
-0.01
0.01
Matched
0.46
0.46
0.00
0.00
Unmatched
29.95
31.95
-2.01
0.01
Matched
31.95
31.83
0.12
0.00
Unmatched
84.75
81.61
3.14
0.05
Matched
81.61
81.88
-0.26
0.00
Mean eQQ
Difference
353.60
158.55
53344.93
384.19
12562.44
336.35
531.97
17.46
2.83
0.07
2.17
0.01
0.03
0.00
3.13
0.11
0.05
0.00
12.08
0.73
4.43
0.37
% Improve
Mean Diff.
Table S18. Indonesia Less-Strict to Strict Matching
Mean Prote- Mean Cont- Difference in Normalized
Status
cted Plots
rol Plots
Mean
Difference
Unmatched
3772.00
4058.80
-286.80
-0.06
Matched
3772.00
3779.20
-7.20
0.00
Unmatched 54776.00
110000.00
-55224.00
-1.04
Matched 54776.00
57236.00
-2460.00
-0.08
Unmatched 34273.00
44212.00
-9939.00
-0.31
Matched 34273.00
32691.00
1582.00
0.08
Unmatched
1275.30
746.28
529.02
0.66
Matched
1275.30
1158.00
117.30
0.15
Unmatched
14.56
11.53
3.03
0.40
Matched
14.56
14.22
0.34
0.06
Unmatched
81.61
84.75
-3.14
-0.10
Matched
81.61
79.25
2.36
0.07
Unmatched
0.46
0.45
0.01
0.02
Matched
0.46
0.43
0.03
0.08
Unmatched
31.96
29.95
2.01
0.02
Matched
31.96
32.23
-0.28
0.00
Mean eQQ
Difference
353.60
158.55
53344.93
384.19
12562.44
336.35
531.97
17.46
3.13
0.11
4.43
0.37
0.05
0.00
12.08
0.73
% Improve
Mean Diff.
45.69%
99.77%
98.09%
97.39%
91.17%
99.73%
88.85%
99.23%
86.03%
93.95%
91.58%
97.50%
95.40%
84.10%
77.80%
88.80%
24.70%
-308.40%
86.20%
32
Table S19. 5K Spillover Analysis
Bolivia
Indonesia
Thailand
ATT
-0.0096
0.005
0.0088
SE
0.0114
0.0232
0.0128
N
772
1807
3386
Controls
15448
7177
10223
Abadie-Imbens heteroskedacticity robust standard errors
Table S20. Observed and Counterfactual Deforestation for all ATT Estimates
Country
Matching
Strict to Unprotected
Less-Strict to Unprotected
Strict to Less-Strict
Less-Strict to Strict
Y(T=1)
0.0097
0.0087
0.0097
0.0087
Ŷ(T=0)
0.0327
0.0217
0.0297
0.0067
ATT
-0.023
-0.013
-0.02
0.002
Costa Rica
Strict to Unprotected
Less-Strict to Unprotected
Strict to Less-Strict
Less-Strict to Strict
0.089
0.137
0.089
0.137
0.255
0.173
0.1206
0.091
-0.166
-0.036
-0.0316
0.046
Indonesia
(Sumatra)
Strict to Unprotected
Less-Strict to Unprotected
Strict to Less-Strict
Less-Strict to Strict
0.146
0.213
0.146
0.213
0.287
0.242
0.355
0.023
-0.141
-0.029
-0.209
0.19
Thailand
Strict to Unprotected
Less-Strict to Unprotected
Strict to Less-Strict
Less-Strict to Strict
0.082
0.398
0.082
0.398
0.221
0.406
0.247
0.198
-0.139
0.008
-0.165
0.2
Bolivia
Y(T=1) is the observed level of deforestation for the treated units in each analysis
Ŷ(T=0) is the counterfactual level of deforestation estimated from the matched controls
in each analysis
33