assessing the role of group heterogeneity in community ... · assessing the role of group...

Assessing the Role of Group Heterogeneity inCommunity Forest Concessions in Guatemala’s

Maya Biosphere ReserveLea Fortmann, Brent Sohngen, and Douglas Southgate

ABSTRACT. We use land use data from Guatemalato estimate the impact of a community forest conces-sion policy on avoided deforestation when accountingfor group heterogeneity. Our analysis includes com-munities with significant forestry experience, alongwith communities comprising recent settlers withprimarily agricultural backgrounds. Employing amatched difference-in-differences approach, we com-pare deforestation in community concessions relativeto a matched control area. We find the concessionpolicy reduced deforestation throughout the study re-gion, with the most significant effects in the recentlysettled concessions. While there is also evidence ofleakage in this group, overall, there is a net reductionin deforestation. (JEL Q57)

I. INTRODUCTION

The role for community management ofpublically owned forests, which comprise 80%of forestland throughout the world, has ex-panded in recent years, particularly in SouthAmerica and Southeast Asia (Food and Agri-culture Organization 2010). This move towarddecentralization largely results from increasedpressure on governments to acknowledge thatmany forest-based communities, especially in-digenous groups, depend on local resources fortheir livelihoods. In addition, shifting manage-ment responsibility to community groups canreduce the burden on governments (both finan-cially and administratively) of managing andprotecting forestland (Agrawal, Chhatre, andHardin 2008). Given the trend toward increas-ing areas under community forest management(CFM), along with the prominent role CFMplays in international climate negotiationsthrough the United Nations program on Re-

Land Economics • August 2017 • 93 (3): 503–526ISSN 0023-7639; E-ISSN 1543-8325� 2017 by the Board of Regents of theUniversity of Wisconsin System

ducing Emissions from Deforestation and For-est Degradation (REDD), it is critical to betterunderstand how community characteristics in-fluence deforestation rates (Agrawal, Chhatre,and Hardin 2008; Larson and Soto 2008;Bowler et al. 2010; Phelps, Webb, and Agrawal2010; Hayes and Persha 2010).

Typically, CFM involves granting someform of property rights (either formal or in-formal) to local communities as incentive formore sustainable resource management. Thisapproach is especially compelling where animbalance between expansive governmentalclaims on natural habitats and modest publicexpenditures on management and administra-tion creates de facto open access. While thelong-term, sustainable management of timberand nontimber forest product (NTFP) extrac-tion likely serves in the best interest of thecommunity members living in and around theforest, individuals have incentives to convertthe land for short-term, personal gains at theexpense of long-term sustainability. Throughthe assignment of shared property rights,CFM projects require communities to over-come these types of collective action prob-lems. Numerous studies have assessed factorsthat influence successful collective action inforest settings (Ostrom 1990; Poteete and Os-trom 2004; Gautam and Shivakoti 2005; Alix-Garcia 2007), as well as the outcomes of vari-ous CFM projects to date (Bowler et al. 2012;Porter-Bolland et al. 2012). However, few

The authors are, respectively, assistant professor,Economics Department, University of Puget Sound,Tacoma, Washington; professor, Department ofAgricultural, Environmental, and Development Eco-nomics, The Ohio State University, Columbus; andprofessor emeritus, Department of Agricultural, En-vironmental, and Development Economics, The OhioState University, Columbus.

August 2017Land Economics504

studies have analyzed factors affecting thesuccess of CFM, and those that have are ofteninconclusive or face limitations to their em-pirical estimation strategies, such as failing tocontrol for selection bias (see Bowler et al.2010 for further discussion).

Bray et al. (2008) observe that the case forCFM consists largely of “anecdotal evidenceand case studies,” many of which compareand contrast CFM and officially protectedparks and reserves. In a survey of various con-servation initiatives, Porter-Bolland et al.(2012) conclude that there is less deforesta-tion where local communities have some formof property rights than in parks and reserves.In contrast, Nepstad et al. (2006) identifiedmixed outcomes in the Brazilian Amazon,finding less encroachment on natural habitatsinside parks than in indigenous reserves.However, the parks they examined were alsounder less external pressure than the indige-nous holdings, which typically are located inthe vicinity of active agricultural frontiers.The same researchers found that, in the faceof similar outside pressures, parks and indige-nous lands were comparable in their capacityto restrict deforestation.

Additionally, most research on CFM todate has concentrated on long-term forestdwellers, including indigenous populations(see, e.g., Nelson, Harris, and Stone 2001;Gautam, Webb, and Eiumnoh 2002; Kumar2002; Somanathan et al. 2009; Nelson andChomitz 2011). One reason for this focus isthat few CFM projects involve recent mi-grants, who likely have significantly lessknowledge about forest management com-pared to long-established communities. Thishas resulted in a potential selection issue,where communities that are most likely to besuccessful in community-based forest man-agement initiatives are also more likely to bechosen to participate in such projects (Bowleret al. 2010). Furthermore, when consideringthe broader potential of CFM, agricultural mi-grants cannot be ignored, since a large portionof forest loss in Africa, Asia, and Latin Amer-ica is due to agricultural expansion (Barbier2004; Gibbs et al. 2010). If recent migrantsare unwilling or unable to engage in effectiveforest management, then CFM may ultimatelyprovide limited protection for forests.

Of the many tropical and subtropical set-tings where CFM has been undertaken, theMaya Biosphere Reserve (MBR) in northernGuatemala is among a handful where groupsof recent settlers have received land for forestmanagement under a policy that created 12community-based forest concessions. The le-gal framework for the policy does not varyacross the concessions, but there is substantialvariation among the participating communi-ties (Maas and Cabrera 2008; Bray et al. 2008;Radachowsky et al. 2012). Some of thesecommunities are made up of individualswhose families have been in the region forgenerations and who consequently are pre-pared to benefit from CFM. Other participat-ing groups comprise recent settlers from out-side the region with primarily agriculturalbackgrounds. These groups tend to have lim-ited forest experience, though CFM appealsto them insofar as it is a route to land own-ership. Additionally, some of the concessionsdedicated to CFM have residents, while othersdo not, typically because they are remote andlack road access. Under these circumstances,a rare opportunity exists to examine howmuch CFM is influenced by the heteroge-neous characteristics of the participating com-munities (Bray et al. 2008).

In this study we use land use data from theMBR to estimate the impact of the communityforest concession policy on avoided defores-tation for three categories of participatingcommunities: long-term inhabitants, recentsettlers, and nonresidents. We employ amatched difference-in-differences approach tocompare deforestation in each of the threetypes of community concessions with forestloss in matched control areas that are made upof nearby habitats that are inside the reservebut are unmanaged. Comparisons are madeboth before and after the concessions were es-tablished to estimate the amount of avoideddeforestation that can be attributed to the pol-icy. A further contribution of this study is ouranalysis of leakage, which might occur if re-duced deforestation inside community con-cessions can be linked to forest loss in adja-cent areas not under management. To ourknowledge, no existing CFM study addressesleakage. If CFM programs simply shift defor-estation to regions that are not under manage-

93(3) Fortmann, Sohngen, and Southgate: Community Forest Concessions in Guatemala 505

ment, either locally or through market prices,then the impact of CFM projects as a wholeis undermined because overall deforestationrates are not reduced.

Based on our analysis, the community for-est concession policy appeared to reduce de-forestation in all three types of concessions,with the largest impact in the recently settledconcessions. Leakage was insignificant in thenonresident and long-inhabited concessions;however, we find evidence of leakage in areasinhabited by more recent migrants. This is notsurprising given that the majority of the in-habitants come from agricultural backgroundsand had little experience with forestry prior tomoving to the MBR. These concessions haveexperienced significant challenges since theirinception, and the deforestation rates over theperiod of analysis are higher in these regionsrelative to other areas in the reserve; however,our results suggest that the concession policystill significantly reduced deforestation com-pared to what would have happened in the ab-sence of the policy. This result is meaningfulsince forests in areas where such migrantsnormally settle are also under greater threat ofland use change. While leakage was detectedin these concessions, the increased forest lossin surrounding areas that can be categorizedas leakage is not enough to negate the largerreductions in deforestation attributed to thepolicy. Overall, we conclude that the com-munity forest concessions were effective inreducing deforestation throughout the MBR,which suggests that this may be a promisingapproach to CFM for a variety of participantsand is not limited to indigenous groups andother well-established forest communities.

II. BACKGROUND

The MBR was established in the Petenlowlands of northern Guatemala in 1990 bythe National Council of Protected Areas(CONAP) in response to wasteful logging, anencroaching agricultural frontier, and mount-ing threats to historical and cultural resources.Encompassing 2.1 million ha, the reserve isdivided into three zones, each with its ownenvironmental objectives and managementstrategies (Figure 1). The core zone, which ac-counts for 36% of the reserve, is strictly pro-

tected and includes national parks, biotopes(areas of uniform habitat), and Mayan ruins—most prominently, El Tikal National Park,which the U.N. Educational, Scientific, andCultural Organization designated as a WorldHeritage Site in 1979. The remainder of theMBR is divided between a buffer zone (24%of the MBR), which separates the reservefrom the rest of Guatemala farther south, anda multiple-use zone (40% of the MBR), whichallows for sustainable resource extraction andis home to the community forest concessions(CONAP 1996; Nittler and Tschinkel 2005;Gomez and Mendez 2007).

While the MBR was established in 1990,the community concessions were introducedlater due to pressure from both conservationgroups that wanted to keep the forests out ofthe hands of private logging firms and localpopulations that wanted legal access to the re-serve and its resources. Left unchecked, theforests within the multiple-use zone presenteda collective action problem. Given the vast re-sources of the forests, local communitiescould manage the land collectively for long-term sustainability through the selective har-vesting of high-value timber (primarily cedarand mahogany) and nontimber forest products(NTFPs) such as chicle (a natural ingredientin chewing gum) and xate (used in floral ar-rangements). There is ample evidence thatforests in the MBR have long provided in-come for households living in the region(Schwartz 1990). However, new migrantswith different skills, or even individualswithin long-established communities, couldalso take advantage of insecure propertyrights and illegally clear the land for farmingor cattle ranching to maximize short-termrents.

To address the collective action problemand to satisfy the interests of both the conser-vation groups and resident communities, asystem was set up in which community as-sociations worked with NGOs to create a sus-tainable forest management plan and an en-vironmental impact evaluation, both of whichhad to be reviewed by CONAP (Nittler andTschinkel 2005; Gomez and Mendez 2007).The management plans were based on extract-ing timber, for commercial purposes, from aselected area of the concession one year and


FIGURE 1Zones and Concessions in the Maya Biosphere Reserve

then rotating to another section each subse-quent year, allowing time for regrowth in theinitial sections by the end of the contract pe-riod. Once the management plan and environ-mental evaluation were approved, CONAPgranted the community property rights to aforest concession area for a renewable periodof 25 years. Resource ownership was a keycomponent of the policy since many of themore recently settled communities inside theMBR were concerned that they would be re-located (Radachowsky et al. 2012). The as-signment of property rights to participatingcommunities secured the forest area for mem-ber use and provided legal grounds for thegroups to prevent outsiders from encroachingon the land (Nelson, Harris, and Stone 2001).In addition to providing property rights, theconcession model provided a framework forcommunities to collectively benefit from theresources provided by the forest.

In 1994, a pilot concession, San Miguel,was established which was followed by thecreation of 11 other concessions between

1997 and 2002 (Table 1). The staggered tim-ing in the formation of the concessions wasprimarily due to administrative and technicallimitations. It took several years to conductthe feasibility studies for the first concessions,which also had to coincide with assessmentsof the biological corridors within the reserve.Additionally, the availability of technical sup-port staff assigned to work with the commu-nity groups was limited, which prolonged theapproval process.

While the procedures for securing propertyrights to a forest concession area were uni-form, the participating communities and theirrespective land holdings differ substantially(Bray et al. 2008; Maas and Cabrera 2008;Radachowsky et al. 2012). With respect tospatial and environmental characteristics, sev-eral of the concessions are remote, withoutroad access, and are relatively farther awayfrom villages and larger population centers.As has been observed in various parts of theworld, remote forests are under less pressurefor land use change (Pfaff 1999; Cropper,


TABLE 1Overview of the Three Types of Community Forest Concessions

Concession Category Concession NameYear

FormedNumber ofMembers

ConcessionArea (ha)

Resident, long-inhabited Carmelita 1997 122 53,797Uaxactun 2000 244 83,558

Resident, recently settled San Miguel 1994 30 7,039La Pasadita 1997 110 18,817La Colorada 2001 39 27,067Cruce a la Colorada 2001 65 20,469

Nonresident Suchitecos (RıoChanchich)

1998 27 4,607

Laborantes (Chosquitan) 2000 78 19,390San Andres (AFISAP) 2000 174 51,940Arbol Verde (Las

Ventanas)2001 364 64,973

El Esfuerzo (Yaloch) 2002 39 25,386Custosel (La Union) 2002 96 21,176

Source: Data from Gomez and Mendez 2007.

FIGURE 2Deforestation in the Maya Biosphere Reserve: 1990–2008

Puri, and Griffiths 2001; Man, Wu, and Deng2013). In the MBR, areas lacking road accesshave had historically low deforestation rates(less than 1% per annum), thus even after con-cessions were established in such areas, thelikelihood of forest loss remained small (Maasand Cabrera 2008; Radachowsky et al. 2012).

In contrast, the central and western portionsof the reserve are more easily accessible andconsequently have experienced more settle-ment and economic activity, leading to greaterenvironmental disturbance. As indicated inFigure 2, which displays patterns of land usefrom 1986 through 2008, deforestation has


been particularly severe in the westernreaches of the MBR, which has absorbed alarge number of refugees during and since thecivil war that Guatemala endured from 1960to 1996, and more recently has become a routefor international drug trafficking (McSweeneyet al. 2014). General lawlessness and violencehave impeded CONAP operations, includingthe creation of concessions, which exist onlyin the central and eastern regions of the MBR(Gomez and Mendez 2007; Radachowsky etal. 2012).

The communities themselves also havesubstantial differences, which likely influencetheir respective incentives for managing theirconcession holdings sustainably. Other re-search supports this hypothesis. For example,Alix-Garcia (2007) examined the impact ofspatial and community characteristics on de-forestation rates in communal land holdingsin Mexico, commonly referred to as ejidos,and found that more educated communitieshad lower rates of deforestation. Other workby Alix-Garcia, Shapiro, and Sims (2012)found that Mexico’s forest conservation pro-gram was more effective in reducing defor-estation in areas with lower poverty levels. Wehypothesize that given the large disparities ineducation and income among the concessioncommunities that coincide with their respec-tive backgrounds in primarily agricultural orforest-related activities, the incentives of eachtype of community group will vary based onthese characteristics and thus need to be ac-counted for in the analysis. These differencesamong the concession communities arewidely recognized in the literature on theMBR (Bray et al. 2008; Maas and Cabrera2008; Radachowsky et al. 2012). In keepingwith previous work, we categorize the con-cessions into the following three groups: non-resident, long-inhabited, and recently settled(Table 1).

The nonresident category comprises sixconcessions made up of members with vari-ous backgrounds in farming and forestry. Thekey distinguishing features of this group arethat the members all live outside the conces-sion boundaries and, on average, report higherincome and education levels relative to theother two groups. Five of the six nonresidentconcessions are located in the more remote,

eastern half of the reserve, which is mostlyuninhabited. The members typically live inlarger towns along major roads in the southernportion of the MBR and travel seasonally toand from the concession areas to extract tim-ber, while often maintaining other jobs thatprovide their main source of income through-out the year. In a 2012 survey of 271 conces-sion members (Fortmann 2014), participantsof the nonresident concessions had the highestincome and education of the three groups, andover half of the members surveyed reportedhaving another job outside of farming or for-estry as their primary occupation (Table 2).The main benefit from joining these conces-sions was that it allowed members to comple-ment their primary earnings with income fromthe seasonal extraction of timber and otherforest products, and in some cases membersreceived significant dividend payments fromconcession profits. In addition to having thelargest income on average among the threegroups, these concession members also re-ported receiving the largest dividend pay-ments from concession profits, providing fur-ther incentives for these members tocollectively work together to keep their con-cessions in good standing.

We expect that leakage in these conces-sions would be minimal given the remotenessof the forest and the corresponding highercosts due to increased travel time, poor roadaccess, and greater distance to markets. Themembers that are engaged in farming orranching typically maintain landholdings out-side of the concession areas in the southernportion of the reserve. Thus, there is little in-centive for members of these nonresidentcommunities to clear land within the conces-sion boundaries.

One limitation in our analysis of the non-resident concessions is that they are both moreremote and managed by communities withhigher levels of education and income (rela-tive to the other concession groups). Thesecombined attributes predict that these conces-sions would experience the least forest distur-bance; however, we are not able to establishthe differential impacts of remoteness versesincome and education.

The second group, long-inhabited, consistsof two concessions, Uaxactun and Carmelita,


TABLE 2Member Summary Statistics Separated by Concession Group

Variable Nonresident Long-Inhabited Recently Settled

Age of head 49.6 47.8 51.0Education of head (years) 5.0** 4.1** 2.5***Born in Peten (percent) 52*** 76*** 29***Daily food expendituresa (dollars) 11.87** 8.77 7.41Annual incomea (dollars) 7,928* 3,867* 2,413**Own land (percent) 47*** 76.8** 82**Personal landholdings (ha) 13.8** 7.6** 48.8***Dividend payment (2011 dollars) 350*** 48*** 0***Farmer (percent) 34.2 37.1 86***Forest job (percent) 7.9** 25.8*** 1.7**Other job (percent) 52.6** 35.5** 10.5***Observations 152 62 57

Note: Concession membership ranged from 20 members to more than 300. Twenty to twenty-five percent ofmembers were randomly selected for interviews.

a Income and food expenditures are converted from quetzales to dollars based on October 2012 exchangerates (when the survey was conducted), Q1.00 = $0.1253 (source: www.freecurrencyrates.com/exchange-rate-history/GTQ-USD/2012). Significant difference from the other two groups based on independent sample t-tests.

*Significant at 10%; ** significant at 5%; *** significant at 1%.

with members coming from petenero com-munities that have been established in the re-gion since the early 1900s (Schwartz 1990).Given their history of harvesting timber andnontimber forest products (chicle in particu-lar), the communities are largely dependent onthe surrounding forest for their livelihoodsand most closely represent the typical com-munities that are the focus of much of the pre-vious research on CFM. While chicle collec-tion peaked in the 1940s and has sincedwindled (Schwartz 1990), a number of theresidents in these communities currently arexateros, collecting xate as a primary source ofhousehold income. Additionally, the commu-nities use the extensive tracts of forest that theconcession policy has provided to earn addi-tional income from ecotourism. All of theseincome-generating activities depend on theforest remaining intact, which provides addi-tional incentives for the concession membersnot only to refrain from illegal land clearing,but also to monitor and enforce the rulesamong the local populations of members andnonmembers alike. While these two commu-nities are located inside the concession bound-aries, they are more remote than the recentlysettled concessions to the south. The closeproximity in which they live, and isolationfrom other villages further enable enforce-ment of the concession policies. Uaxactun, in

particular, is protected by Tikal National Park,where there is only one road from which toaccess the village and it runs directly throughthe park, with guards monitoring all transpor-tation entering and exiting the park.

In a survey of members from long-inhab-ited concessions, 76% of the people reportedbeing born in the Peten (Fortmann 2014).Among the three concession types, this grouphas the largest percentage of members thatclaim a forest-related job as their primary oc-cupation (26%). A number of members alsoreported farming as their primary occupation(37%); however, this tends to be on land thatwas previously cleared and allocated to mem-bers for agricultural purposes when the con-cessions were first established. Given thatthese communities have existed within the Pe-ten for multiple generations and have greaterreliance on forest resources for their liveli-hood, they are more likely to successfully en-gage in collective action and thus have lowerdeforestation rates in their respective conces-sion areas compared to groups with less forestexperience (Baland and Platteau 2000, 255;Poteete and Ostrom 2004).

The four recently settled concessions, incontrast, are primarily comprised of individ-uals from agricultural backgrounds who wereborn in other parts of Guatemala and who con-tinue to rely heavily on farming for their live-


lihood. These concessions are located in thecentral region of the reserve, and the memberslive inside the concession boundaries. Com-pared to the other two concession groups, sur-veyed members of recently settled conces-sions reported the lowest income and level ofeducational attainment, and farming was re-ported as the primary occupation for 86% ofthis group (Fortmann 2014). Moreover, mem-bers of these concessions are not entitled todividend payments, since they were registeredas nonprofit associations at the time of for-mation and are eligible only for in-kind ben-efit distributions (e.g., life insurance, schol-arships). The nonresident and long-inhabitedconcessions, on the other hand, are allowed todistribute annual dividends to their membersbased on concession profits.

We suspect that the recently settled con-cessions are the most susceptible to leakagefor a number of reasons. Relative to the othergroups, these concessions are surrounded bya number of rural settlements populated pri-marily by migrants who are not participatingin CFM and come from agrarian backgrounds,placing these concessions under considerablepressure for land clearing. Residents of thesecommunities also have fewer outside optionsfor alternative work, given their greater dis-tance to towns and relatively low levels ofeducation. Part of the concession contract in-cludes forming vigilance committees that areresponsible for monitoring the concessionarea and keeping out encroachers; this addedenforcement may further result in shifting de-forestation from inside the boundaries to sur-rounding areas not under the purview of con-cession management. Survey data indicatethat members of the recently settled conces-sions have the largest private landholdings,with the members reporting 47.5 ha on aver-age compared to the other two concessiongroups, which averaged 13.8 and 7.6 ha forthe nonresident and long-inhabited members,respectively (Fortmann 2014), further sug-gesting that these communities have previ-ously engaged in land-clearing activities andthus making these concessions more prone toleakage.

The recently settled concessions are furtherdistinguished from the nonresident and long-inhabited groups in that the majority of these

concessions formed under pressure from thegovernment and NGOs. Many of the memberswere initially reluctant to join and did so onlyout of fear of being removed from the landthey were currently settled on after the estab-lishment of the reserve (Nittler and Tschinkel2005; Radachowsky et al. 2012). This pro-vides for a unique treatment group in whichthe members were not inclined toward forestmanagement and thus there is not the potentialfor selection bias that may be present in thelong-inhabited concession areas that wouldlikely have lower deforestation rates regard-less of concession status. Given the circum-stances, it is not surprising that the recentlysettled concessions have struggled more thanthe other two groups; two concessions (SanMiguel and La Colorada) were canceled as of2009, and another (Cruce a la Colorada) hasbeen under suspension (see Radachowsky etal. 2012 for more details).1 The reluctance toform groups and lack of initial interest in for-estry are a departure from many other CFMstudies that focus on long-term, forest-dwell-ing communities. The addition of migrantcommunities in this study provides new anal-ysis on how these types of groups will fair infuture CFM programs and whether they canbe effective in forest management and con-servation initiatives.

To account for the differences among theconcession communities in our assessment ofCFM, we separate the concessions into thethree categories described above, each com-prising a different treatment group that ismatched to a separate control group. The con-trol areas consist of surrounding nonconces-sion forestland that is part of the multiple-usezone within the MBR but is not under a man-agement plan. The core and buffer zones arenot included in this analysis since they are un-der different management strategies and thusdo not make a suitable control group for theconcessions. In addition to the features wehave highlighted, there are a number of otherdifferences among the concessions, including

1 In La Colorada, a number of the members illegally soldtheir land to cattle ranchers and fled the area. In 2008, alarge area of clear-cut forest was discovered and the con-cession was canceled shortly thereafter (Radachowsky et al.2012; Wildlife Conservation Society 2013).


size of membership and landholdings, profitdistribution (based on whether they are a for-profit or nonprofit association), and level ofengagement in NTFP collection. While manyof these community characteristics are of in-terest in determining the success of the con-cessions and the role of community hetero-geneity, we are unable to account for thesedifferences in the analysis, since the charac-teristics are not isolated to one type of con-cession group (Table 1).

III. DATA

Land use data for this study come from sat-ellite imagery of the MBR compiled by theCenter of Monitoring and Evaluation (CE-MEC) under CONAP.2 CEMEC classified theimages into various land uses at 30 m reso-lution. The data format is similar to that ofsurvival (or hazard) data. Each 30 × 30 mpixel is coded as “forest” if it remained for-ested from 1986 through 2008. If at somepoint it was determined that the parcel of landexperienced forest disturbance, based on sat-ellite images, that pixel is coded based on theyear the disturbance took place. Thus, refor-estation of a parcel in later periods is not ac-counted for in the data set. A further limitationis that we are not able to ascertain the densityor quality of the forest area, only whether aparcel is considered forested or not (see Saderet al. 2001 for more details). As a result, thedependent variable of interest is binary, equalto 1 if the pixel of land was deforested duringa given time period, 0 otherwise. Due to thelarge number of observations in the studyarea, we randomly selected 100,000 observa-tions from each type of concession area to cre-ate three separate treatment groups. The cor-responding control groups comprised 200,000randomly selected observations from the sur-rounding areas in the multiple-use zone thatwere not under concession management orpart of the core or buffer zones.3

2 All of the data used in this analysis originated from theNational Institute of Statistics and the Ministry of Agricul-ture and were complied by CEMEC. The data are availableupon request from the authors and CEMEC.

3 The total dataset including all observations in the con-cession and nonconcession areas of the multiple-use zonecontains 10,557,972 observations.

Environmental covariates used to matchobservations between the concession areasand the control areas include elevation, slope,soil acidity (pH), and aspect. The slope andelevation measures come from digital eleva-tion models (DEM) created by Guatemala’sMinistry of Agriculture, Livestock, and Foodand are analyzed using ArcGIS (ESRI 2011).Data on distances to roads and settlements (es-tablished prior to 1990 when the reserve wasformed) were obtained from the National In-stitute of Statistics. Spatial covariates used inthe analysis include distances to the nearestroad, pixel of land cleared prior to 1990, ar-cheological site, major town, and medium-sized village. Major towns are defined ashaving a population over 10,000 people.4 Me-dium-sized villages are defined as having apopulation greater than 200 people or existingwithin the reserve prior to its establishment in1990 (Grunberg and Hugo 1998).5 These vil-lages are included in the analysis separatelyfrom towns since populations in smaller set-tlements are more likely to be involved in sub-sistence agriculture and have different incen-tives for land clearing. Distance to the nearestmajor town serves as a proxy for transporta-tion costs and distance to markets and is morelikely to influence land use change on a com-mercial scale. To account for regional differ-ences in the reserve (discussed in the back-ground section), a dummy variable is includedin the matching models based on whether theobserved parcel is in the eastern or westernhalf of the reserve. For the four recently set-tled concessions, only observations in thewestern half of the multiple-use zone are in-cluded in the control group since all of theconcessions are located in this region.

Given that the primary unit of observationis a parcel of land and whether it was defor-ested or not, our analysis is driven by envi-

4 There are four towns in the Peten that meet this re-quirement: Flores, San Benito, Melchor de Mencos, andPoptun. However Poptun is located farther away from thereserve than the others, so no parcel of land included in theanalysis is closer to this town than compared to the othermajor towns.

5 Medium-sized villages within the reserve prior to 1990are based on a census report by Grunberg (1998) and cor-respondence with Dr. Norman Schwartz, an anthropologistand scholar of the Peten.


TABLE 3Comparison of Summary Statistics for Three Forest Concession Groups

Variable Nonresident Long-Inhabited Recently Settled

Road distance (km) 21.4 8.8 6.6Slope (degrees) 3.7 3.5 5.0Elevation (meters) 194 226 230Cleared distancea (kilometers) 9.4 5.7 1.6City distance (kilometers) 49.5 69.5 51.6pH 6.8 7.1 7.3Archeo distanceb (kilometers) 21.4 16.4 22.7Aspect (degrees) 175 159 185Village distance (kilometers) 25.3 13.1 5.1

Note: Data are from remote sensing satellite images. Observations are based on 30 m×30 m parcels of land.See Data section for more details. All of the variables for each concession group are significantly different fromthe other groups at the 99% level for all covariates.

a Cleared distance is the distance to the nearest pixel of land cleared prior to 1990.b Archeo distance is the distance to the nearest Mayan archeological site in the reserve.

ronmental and spatial covariates. The limita-tion of using spatial variables for matching asa proxy for differences in the concessiongroups is that each treatment parcel is influ-enced by the characteristics of its respectivetype of concession community. The controlareas, however, do not account for these dif-ferences among the communities, whichmight affect land use outside the concessionboundaries. While we cannot directly controlfor community variation in the control re-gions, we believe that many of the importantdifferences are captured by the spatial vari-ables (Table 3). For example, the distinctionbetween the three types of concessions is evi-dent when looking at the average distance tothe nearest road and village in the nonresidentconcessions (21 km and 25 km, respectively)compared to the long-inhabited (9 km and 13km) and recently settled concessions (7 kmand 5 km). The greater distance to roads andvillages in the nonresident concessions re-flects the remoteness of the land and highercosts of land clearing. These characteristicsare captured by this distance variable whenfinding a suitable matched parcel in the con-trol group. Additionally, the covariate for dis-tance to the nearest land clearing (prior to1990) further highlights how the spatial char-acteristics of the parcels reflect differencesamong the concession communities. In the re-cently settled concessions, the nearest clear-ing is 1.6 km away on average, compared to5.7 km and 9.7 km in the long-inhabited andnonresident concessions, respectively. Thus,

when observations in the recently settled con-cessions are matched to control areas, parcelsthat are closer to land clearings are also likelyto be closer in proximity to other, recently set-tled communities located in the MBR.6

The main limitation that we see in thematching process is the comparison betweenthe long-inhabited concessions and theirmatched control group, since the surroundingcontrol areas are mostly inhabited by recentsettlements and there are no similar forest-dwelling communities without concessions todraw comparisons. For these long-standingconcessions, we hypothesize that the pressurefor land conversion is reduced, since thesecommunities have been relying on forest re-sources for multiple generations. So, compar-ing these communities with control areas thatare mostly inhabited by recent migrants willoverestimate the effectiveness of the conces-sion policy, since treated areas are less likelyto experience deforestation compared to theirmatched control areas. While this is a concern,we believe that the main group of interest inour study is the recently settled concessions,since little research has been done on howthese types of communities would fare inCFM projects. The majority of settlements inthe control areas are made up of recent mi-grants who moved to the region in search ofland for farming and ranching and have many

6 The distance to the nearest clearing is based on clearedareas prior to 1990, before the reserve was established.


similarities to the recently settled concessioncommunities. Therefore, we expect that theunobservables due to community influenceare similar in both the treatment and controlgroups. Regarding the nonresident category,since these concession areas are remote andthere are very few settlements in the easternhalf of the reserve outside of the main town(Melchor de Mencos) where the majority ofthe population lives, the unobservables forboth the treatment and control areas are likelyto be similar after controlling for remotenessand location,

Finally, without baseline survey data, wecannot control for changes in the communitiesthat may have occurred due to the policy.However, we believe that it is unlikely thatthe defining characteristics of the three com-munity types, such as forest experience or mi-grant status, changed significantly betweenthe formation of the concessions and the timethe survey took place. Thus, we assume thatthe key differences among the concessioncommunities are time invariant over the pe-riod of analysis.

IV. METHODS

We use a matched difference-in-differences(DID) approach to assess the effectiveness ofthe three different concession communities onavoided deforestation in the MBR. The finaldata sets used in the postmatching regressionanalysis are based on nearest-neighbor match-ing with Mahalanobis distance metrics andcalipers to limit poor matches (Guo and Fraser2010, 147; Ho et al. 2007, 2011; Arriagada etal. 2012).

Approach to Matching

The first part of the analysis involvesmatching treatment group observations fromeach of the three forest concession areas withcontrol observations from the nonconcessionareas. The objective is to create a controlgroup of observations that resembles the treat-ment group to estimate how much deforesta-tion would have taken place in the absence ofthe concession policy. Matching methods arebased on the assumption that all of the factorsthat affect selection into the treatment group

are observable and controlled for in thematching process; this is often referred to asselection on observables or the conditional in-dependence assumption (Guo and Fraser2009). Failure to take into account differencesbetween the two groups may lead to biasedestimates of the treatment effect. The secondnecessary requirement for estimating causaleffects is referred to as the stable unit treat-ment value assumption (SUTVA), which re-quires that the treatment of one unit does notimpact the outcome of other units. To satisfythe SUTVA, the selection of forests into con-cession management would have to have noimpact on deforestation in surrounding, non-concession areas, which make up the controlunits in this case. This may not be true if leak-age is an issue, which occurs if increased man-agement in one area results in deforestationshifting to nonprotected areas. Given thatleakage is a concern, our sensitivity analysisincludes estimates of spillover (increased de-forestation) after the policy was implementedin nearby buffer areas surrounding the con-cession boundaries.

In the case of the MBR, we believe the re-sults would potentially be affected if we didnot account for differences among the threetypes of concessions (i.e., treat all the conces-sions the same), since the heterogeneous com-munity characteristics likely lead to variationin the effectiveness of the concession policyacross participating groups. To address thispossibility, we matched each of the threegroups separately under the assumption thatdifferences in the types of communities ap-plying for concessions and the spatial char-acteristics of the forest area will also influencethe ability of each group to manage the landand successfully engage in collective action.

Following Ho et al. (2007), we focus onnearest-neighbor matching, which allows forpostregression analysis with the matched datasets (Guo and Fraser 2009). The final samplesused in the DID analysis are matched basedon nearest-neighbor propensity score match-ing with Mahalanobis distance metrics. Pro-pensity scores are estimated with a logitmodel that predicts the probability of treat-ment (being selected into one of the three con-cession areas) based on a set of environmentaland spatial covariates (see Appendix Table A1


for logit results). The propensity score is usedto narrow down the range of possible matchesbetween the treatment and control observa-tions. For each treatment observation, onlycontrol observations that have propensityscores within a quarter of a standard deviationof the treatment observation’s score (cali-per = 0.25) are eligible for matching. After thepotential control observations are narroweddown, they are matched to the treatment ob-servation with the smallest distance metricbased on a vector of covariates (see Guo andFraser 2009 for further details).7 Balance isassessed by measuring the difference in meansbetween the treatment and control groups foreach covariate and the median difference inthe quantile-quantile (QQ) plots betweengroups, where zero indicates that the empiri-cal distributions are the same for each covar-iate (Ho et al. 2011) See Appendix Table A2for results on balance between matched treat-ment and control groups for each concessiontype.

Difference-in-Differences

After matching the data, we conduct DIDregression analysis with two cross-sections ofthe matched data to estimate the average treat-ment effect on the treated (ATT) observations,which is defined here as the expected differ-ence in deforestation in forest areas undercommunity concession management com-pared to the deforestation that would havetaken place had there been no concession pol-icy. The DID model compares differences inthe outcome variable between the treatmentand control groups before and after the policywas implemented. The regression equationcan be written as

y = β +β Treat +β Post +β Treatit 0 1 it 2 it 3 it

× Post +β X + ε ,it 4 it it

where y is a binary dependent variable equalto 1 if the parcel was deforested in a given

7 Mahalanobis distance metric is measured asD(i,j) = (u−v)

TC−1 (u−v), where i and j are treatment

and control observations, respectively, and u and v are vec-tors of covariates for the treatment and control observation.C is the sample covariance matrix (Guo and Fraser 2010).

period, i indexes the observed parcel of forestarea, and t indexes the time period, 0 beforethe concessions, or 1 after the concessionswere established. The coefficient of interest isb3, the DID estimate, which can be interpretedas the impact of the concession policy onavoided deforestation.

An advantage of using DID is that themethod controls for unobservable, time-invar-iant factors that might affect the outcome ofinterest and selection into the treatment group.However, this method rests on the paralleltrends assumption, which requires that thetrends in deforestation for treatment and con-trol group areas would have been similar ifthe policy had never been implemented(Wooldridge 2002). To test this assumptionwe reformat the data to represent a panel thatincludes three periods of observation for eachparcel before the concession policy was inplace and the concessions were actively man-aging the forest, from 1990 to 1997.8 We thencompare differences in deforestation in thecontrol group with treated concession areasbefore the policy was implemented using in-teraction variables ( )Control × Time_periodthat can be interpreted as the difference in de-forestation between the treatment and controlgroups prior to the concession policy. If therewere no differences in deforestation trends,we would expect the interaction variables tobe statistically insignificant, which is what wefind for all three types of concessions (see Ap-pendix Table A3 for results).

For the DID estimation, the prepolicy timeperiod used for this analysis is 1990 to 1997,after the reserve was created but before thecommunity groups were actively managingtheir concession areas. The first pilot conces-sion (San Miguel) formed in 1994, whichoverlaps with the prepolicy period; however,given the relatively smaller size of the con-cession area compared to the other recentlysettled concessions, we believe that the over-lap in time does not have a significant impact

8 Panel data format is based on periods where defores-tation was observed. There are three preconcession periods:1990–1993, 1993–1995, and 1995–1997. Each pixel wascoded as 1 if it experience deforestation during that period,and then remained 1 for all subsequent periods.


on the results.9 The postpolicy period variesslightly based on the concession group. Forthe nonresident concessions the postpolicy pe-riod is 2002 to 2008, so the analysis does notinclude the interim years when the conces-sions were being formed. For the long-inhab-ited concessions the postpolicy period is 2000to 2008, since Carmelita and Uaxactun wereboth established as of 2000. For the recentlysettled concessions the postpolicy period is2001 to 2008.10

For the main regression analyses, we use alinear probability model, since the dependentvariable is binary, equal to 1 if the parcel ofland was deforested in the period of analysis.The results include standard errors clusteredon regional management boundaries to con-trol for potential spatial correlation among ob-servations. We also conduct a number of otherrobustness tests discussed in the results sec-tion that follows.

V. RESULTS

The results indicate that the concessionpolicy reduced deforestation in all three typesof community concession areas relative towhat would have occurred had the policy notbeen implemented. The results are robust totests for unobserved bias using Rosenbaum’sbounds (Rosenbaum 2002). Given that we de-tect leakage in the recently settled conces-sions, to control for the impact of the conces-sion policy on nearby forest areas we reporttwo sets of results. Column (2) includes theentire matched control group and column (3)excludes control observations inside the 2 kmbuffer area surrounding the forest concessionboundaries.11 The results from both models

9 We believe the inclusion of a concession in the latteryears of the prepolicy period would result in lower defor-estation rates in the “pre-” period, which would bias theresults downward, underestimating the impact of the con-cession policy. In our analysis we also run a model withoutSan Miguel and the results do not change appreciably.

10 San Miguel was canceled in 2007; however, the de-forestation data is recorded in two-year increments to showdeforestation during the period of 2007 to 2008, so our anal-ysis extends to the end of 2008 for the recently settled con-cessions.

11 We exclude only the 2 km buffer and not the 5 or 10km buffer areas since those results are inconclusive regard-ing leakage.

are similar in magnitude, and the statisticalsignificance is improved when the 2 km bufferobservations are excluded.

Referring to the results from column (3),all three types of concessions experienced lessdeforestation compared to their control areas.The nonresident concessions had the smallesteffect, with a 4.3% reduction. This is not sur-prising given that these areas had the lowestdeforestation rates even prior to the conces-sions forming. The long-inhabited conces-sions reduced deforestation by about 6.4%compared to the matched control group, andthe recently settled concessions reduced de-forestation by approximately 7.7%. Table 4presents the DID results with clustered stan-dard errors, though some limitations on theclusters should be noted. Clustered standarderrors ideally control for correlation in the er-ror terms and reduce the reliance on the as-sumption of the errors being identically andindependently distributed. However, given thedata set, the best available variable on whichto cluster the errors is based on regional man-agement boundaries (concessions, biologicalcorridors, national parks, etc.; see Figure 2),which cover large tracts of land that have sig-nificant spatial variation and provide a limitednumber of clusters (25 when including allconcessions and control areas).

We use the DID results to estimate the areaof avoided deforestation that can be attributedto the concession policy for each type of con-cession. The long-inhabited concessionscover approximately 133,576 ha of forest, andthe resulting avoided deforestation is esti-mated to be 8,548 ha. In the nonresident con-cessions, avoided deforestation amounts toapproximately 8,348 ha (based on 194,157 to-tal ha), and 4,955 ha in the recently settledconcessions (out of 64,361 total ha).

As a further robustness check for the DIDresults, we also compare differences in meandeforestation rates between the matched treat-ment and control groups (ATT) before andafter the concession policy was establishedusing data matched with kernel and bias-cor-rected matching estimators (Abadie and Im-bens 2002).12 The results are comparable with

12 Smaller sample sizes (50,000 rather than 300,000) are


TABLE 4Results of Difference-in-Differences (DID) and Probit Marginal Effects (ME)

Unmatched(1)

Matcheda

(2)

Matched without 2 kmBuffer

(3)

Model Coeff.ClusteredStd. Err. Coeff.

ClusteredStd. Err. Coeff.

ClusteredStd. Err.

All GroupsDID −0.02 0.024 −0.042 0.038 −0.046 0.037Probit ME 0.022** 0.011 0.004 0.008 0.0098 0.009Observations 295,001 169,666 134,192R-squared 0.12 0.15 0.14

NonresidentDID −0.04** 0.02 −0.033 0.02 −0.043** 0.02Probit ME 0.003 0.02 −0.014** 0.007 −0.014* 0.007Observations 293,248 90,190 73,159R-squared 0.13 0.47 0.098

Long-inhabitedDID −0.043** 0.02 −0.050* 0.028 −0.064* 0.033Probit ME −0.018 0.013 −0.022** 0.011 −0.024** 0.012Observations 294,679 94,204 84,471R-squared 0.13 0.11 0.12

Recently settledDID 0.026 0.022 −0.078* 0.047 −0.077*** 0.019Probit ME −0.001 0.013 −0.040** 0.016 −0.015 0.014Observations 296,262 43,116 25,682R-squared 0.08 0.25 0.065

Note: Standard errors are clustered at the “concession” region level with 25 clusters for “all groups” and 17clusters in the subgroups.

a Matched samples are based on nearest-neighbor propensity score matching with Mahalanobis distance metricwith calipers set to 0.25 standard deviations taken from the sample with 300,000 observations. Results basedon a linear probability model where the dependent variable equals 1 if pixel is deforested, 0 otherwise.


the DID results (Table 5). Kernel matchingestimates indicate that deforestation was re-duced across all three types of concessions,with a 7.8% reduction in the recently settledconcessions (compared to a 7.7% reductionusing the DID estimates). Results from thebias-corrected matching show that the re-cently settled concessions reduced deforesta-tion by 8.4%, and the long-inhabited conces-sions reduced deforestation by 8.5% (resultsfor the nonresident concessions are not sig-nificant). These results are also robust to un-observed or “hidden” bias; see Appendix Ta-ble A4 for results and discussion onrobustness tests (Rosenbaum 2002; Beckerand Caliendo 2007).

As a final robustness check we transformthe data to a panel format, similar to the par-

used for these matching estimators due to the computationalcomplexity of the matching process.

allel trends analysis, and test DID including atime trend (see Appendix Table A5 for re-sults). The results do not change substantially,and in all cases, the estimated treatment ef-fects for the concessions are negative, furthersupporting the conclusion that the concessionpolicy reduced deforestation in all three cate-gories of concessions.

Leakage Analysis

We test for leakage that would occur if theconcession policy resulted in deforestationshifting from inside concessions to areas out-side their boundaries. If such negative spill-overs occur, then the estimates of avoided de-forestation, based on the results from Table 4,will be upward biased. Positive spilloverscould occur, as well, if the implementation ofthe concession policy and improved forestmanagement discourages deforestation out-


TABLE 5Average Treatment Effect on the Treated (ATT) Results for Kernel and Bias-Corrected Matching Estimators

Unmatched Kernel Bias-Corrected

Concession Group Time Period ATT Std. Err. ATT Std. Err. ATT Std. Err.

Nonresident Pre −0.036 0.002 −0.0028 0.004 −0.013 0.0006Post −0.070 0.003 −0.025 0.009 −0.015 0.004

Difference −0.034*** 0.004 −0.022** 0.010 −0.0015 0.004Long-inhabited Pre −0.036 0.002 −0.041 0.009 0.0080 0.004

Post −0.068 0.003 −0.144 0.011 −0.077 0.009Difference −0.032*** 0.004 −0.103** 0.014 −0.085*** 0.010

Recently settled Pre −0.019 0.002 0.013 0.02 0.038 0.002Post 0.068 0.004 −0.065 0.034 −0.046 0.020

Difference 0.087*** 0.004 −0.078** 0.039 −0.084*** 0.020

Note: Kernel matching performed in Stata Version 12 with “psmatch2” using clustered standard errors. Bias-corrected performed in Statausing the “nnmatch” command. The number of observations for each sample varies but is based on 50,000 total observations in the unmatchedsample, then split up into 25,000 each for pre and post samples. The standard errors for kernel results do not take into account estimatedpropensity scores, but we choose not to bootstrap given the potential for errors (see Abadie et al. 2004; Abadie and Imbens 2006), since theseare robustness checks. The bias-corrected ATT estimates do not include clustered standard errors.


side of the concession boundaries. Followingthe approach of Andam et al. (2008), we esti-mate leakage from the community concessionsby creating buffer areas around the outsideboundaries of the three concession areas (2 km,5 km, and 10 km) and then match randomlyselected observations from within the bufferareas to observations outside the buffers, far-ther away from the concession borders.13

Leakage is estimated based on the coeffi-cient for a buffer dummy variable, which in-dicates if an observation is within the desig-nated buffer area in the postconcession policyperiod. We estimate the model with ordinaryleast squares including the full set of environ-mental and spatial covariates (Table 3). A posi-tive buffer coefficient indicates that leakagemay have occurred, since observations in thebuffer zones had higher rates of deforestationcompared to matched observations fartheraway from the concessions. This approachcaptures leakage due to local residents in-volved in smaller-scale agriculture or ranch-ing. If the presence of the concessions dis-couraged large landowners and cattle ranchersfrom settling near concession boundaries, but

13 Andam et al. (2008) matched unprotected forest plotswithin a 2 km buffer surrounding the protected area (treat-ment) with unprotected plots beyond the 2 km radius (con-trol) and compared deforestation in the two areas to testwhether there were any spillover effects due to the desig-nation of protected area status in Costa Rica.

they moved elsewhere in the region and defor-ested outside the buffer areas, then this methodwould underestimate the total amount of leak-age taking place due to the concession policy.14

We find that leakage primarily occurrednear concessions managed by recent migrants.The results for these concessions indicate thatleakage varied across buffer distances, with ar-eas closer to the concession boundaries expe-riencing more deforestation than areas fartheraway. Deforestation inside the 2 km buffer sur-rounding the recently settled concessions was13% higher compared to matched control areasoutside the buffer region. In the 5 km and 10km buffer areas, the magnitude and sign (posi-tive or negative) of leakage varies, thus the ex-tent of the leakage outside of the recently set-tled concessions beyond the 2 km buffer is notclear (Table 6). Leakage estimates in the othertwo concession groups for all three buffer dis-tances are, for the most part, negative (indicat-ing positive spillovers) or not significant (Ap-pendix Table A6).

The presence of leakage in areas surround-ing the recently settled concessions is not sur-prising. We would expect to see leakage in

14 Our model estimates leakage only from activity shift-ing and not market-based leakage that would occur if thesize of the projects changes any prices (we assume pricesare exogenous). Other studies that estimate market leakageinclude those by Murray, McCarl, and Lee (2004) and Sohn-gen and Brown (2004).


TABLE 6Leakage Estimates for Buffer for Recently Settled Concessions

2 km Buffer 5 km Buffer 10 km Buffer

Model Unmatched Matched Unmatched Matched Unmatched Matched

OLS 0.089 0.145 0.078 −0.057 0.057 0.039Robust Std. Err. (0.004)*** (0.008)*** (0.003)*** (0.009)*** (0.004)*** (0.014)***Clustered Std. Err. (0.031)*** (0.003)*** (0.027)*** (0.005)*** (–0.054) (0.055)Kernel — 0.068*** — −0.13*** — −0.0004Std. Err. — (0.018) — (0.013) — (0.036)Observations 96,899 7,230 97,756 5,582 98,014 5,004R-squared 0.19 0.28 0.16 0.29 0.15 0.17

Note: Matched results are based on nearest-neighbor matching without replacement with Malahanobis distance metrics. Matching was donein R using “MatchIt” (Ho et al. 2011). Kernel results are from Stata Version 12 using “psmatch2.” Standard errors are in parentheses. Dependentvariable is binary, equal to 1 if observation is within the designated buffer zone. OLS, ordinary least squares.

*Significant at 10%; ** significant at 5%; *** significant at 1%. Significance designations based on clustered standard errors.

this area in particular for a number of reasons.The recently settled concessions are locatedin a more densely populated area of the re-serve, which includes a number of settlementsthat are not a part of the community forestconcessions. The majority of these settle-ments comprise more recent migrants thatcome from farming backgrounds, and as a re-sult, the main source of deforestation in thisarea is due to land clearing for subsistencefarming and cattle ranching. Furthermore, thelocal populations in this region have few al-ternatives to farming since they do not havebackgrounds in forestry and live inside the re-serve, farther away from towns and cities thatwould provide alternative employment oppor-tunities. Residents in these communities alsoare less educated, further limiting their op-tions for work outside of agriculture. In asurvey of both recently settled concessionmembers and neighboring nonmembers, par-ticipants from this region of the reserve re-ported the lowest educational achievementlevels (2.6 years of schooling on average forthe head of household) relative to survey par-ticipants in the nonresident and long-inhabitedcommunities (5.1 and 4.1 years, respectively).

Despite the negative spillovers, we stillfind that overall, the concession policy re-duced deforestation. Taking into considera-tion the increased amount of deforestation inthe 2 km buffer area that we attribute to leak-age, we calculate that avoided deforestationamounts to 3,233 ha in the recently settledconcessions (compared to 4,955 ha in gross).Other studies estimating leakage surrounding

protected areas or land enrolled in paymentfor ecosystem services programs found mini-mal leakage that was, if anything, positive.(Andam et al. 2008; Honey-Roses, Bayliss,and Ramırez 2011). The only study, to ourknowledge, that finds evidence of negativeleakage is by Alix-Garcia, Shapiro, and Sims(2012). The authors examined slippage due tofarmer enrollment in a payment for ecosystemservices program in Mexico that compensatesfarmers who conserve forestland. They foundthat in more remote areas with high programenrollment and low road density, deforesta-tion is higher in buffer areas surrounding en-rolled land compared to areas that are less re-mote and have better access to markets. Theseresults parallel our findings, where leakage isgreater in areas where the residents live far-ther away from markets and have few alter-natives to farming. Among other studies thatexamine community-managed forests, noneaddress leakage in their analyses (Nelson andChomitz 2011; Somanathan et al. 2009;Bowler et al. 2010)

VI. CONCLUSION

The majority of the literature on CFM hasbeen focused on indigenous or long-standingcommunities of forest dwellers. Given thatthese communities are more willing to engagein CFM projects, a selection bias problemarises, since they are also more likely to besuccessful given their background and depen-dence on forest resources (Bowler et al. 2010).The MBR presents a unique situation where


multiple types of communities have beengranted property rights to forest concessionareas, including long-time forest dwellers, aswell as more recent migrants with primarilyagricultural backgrounds. This allows us tocompare the two types of groups and theirrelative effectiveness in reducing deforesta-tion. The third group of nonresident conces-sion communities experienced the leastamount of deforestation, both before and afterthe concession policy was implemented. Thisfollows previous research, which mainly findsthat remote, uninhabited forests experiencelower deforestation rates compared to oneswith settlements (Nepstad et al. 2006; Joppa,Loarie, and Pimm 2008).

The members of the nonresident conces-sions come from mixed backgrounds, somewith prior forest experience, others without.They are also distinguished from the other twocommunity types by their relatively higherlevels of education and income (Fortmann2014). This further supports previous researchthat deforestation rates are lower in commu-nities with more educated and wealthier pop-ulations (Alix-Garcia 2007; Alix-Garcia, Sha-piro, and Sims 2012). Our findings indicatethat allowing nonresident community conces-sions access to remote forests for the purposeof harvesting timber and NTFPs does not leadto increased deforestation in uninhabited re-serves, but it does provide an alternativesource of income generation for local com-munities. However, this may be contingent onthese communities also having alternative em-ployment opportunities outside of agricultureand/or separate landholdings set aside forfarming, as the nonresident communities inthis study do, which reduces their incentivesto illegally harvest timber.

The main group of interest in our study isthe recently settled concessions. These com-munities primarily comprise migrants whowhere displaced from their previous landhold-ings due to the civil war in Guatemala, ormoved to the MBR in search of land for farm-ing (Schwartz 1990). We find that despitetheir lack of forestry experience, and in somecases, initial reluctance to join a communityforest group, the forest areas under concessionmanagement by these communities still faredbetter than the matched control areas in terms

of avoided deforestation, as predicted, thoughthese concessions also experienced leakage inareas adjacent to the concession boundaries(whereas the other two types of concessionshad minimal or positive spillover). The esti-mated leakage was not enough to negate theavoided deforestation attributed to the con-cession policy though, so overall, we find netavoided deforestation in the recently settledconcessions. These findings have broader pol-icy implications considering that a significantportion of tropical deforestation taking placeis happening at the agricultural frontier. Whileon the whole, these areas experienced higherrates of deforestation than other areas in thereserve, our results suggest that deforestationwould have been significantly higher in theabsence of the concession policy. The long-term viability of these concessions, however,is still in question. These communities haveexperienced more conflict and have ultimatelystruggled with concession management morethan the two other types of communitygroups, with two concessions being canceledand another under suspension. Given the lo-cation of these concessions, there was a hostof internal and external factors that likely con-tributed to this outcome, including corruptionwithin the organizations and external pressurefrom outside actors that resulted in land grab-bing and illegal land clearing for cattleranches (Nittler and Tschinkel 2005; WildlifeConservation Society 2013). Also, the his-torically weak governance of CONAP likelycontributed to the termination of these con-cessions, where illegal activities within theMBR have long gone unchecked and unpun-ished (see Radachowsky et al. 2012 for moredetails). On the whole, our evidence suggeststhat the agricultural migrant–based conces-sions did achieve some success with reducingdeforestation, though more research will needto be conducted to develop a longer-term, sus-tainable model for these groups. On the otherhand, in the two other types of concessions,we find that both groups were able to reducedeforestation with no signs of leakage, sug-gesting that the community concession modelcan be effective in similar settings of unin-habited reserves and the more traditional,long-term forest dwelling populations.


APPENDIX

Test for Unobserved Bias

TABLE A1Logit Results from Estimating Propensity Scores for Matching

Nonresident Long-Inhabited Recently Inhabited

VariableOddsRatio

ClusteredStd. Err.

OddsRatio

ClusteredStd. Err.

OddsRatio

ClusteredStd. Err.

Distance to road 1.41 0.209** 0.83 0.13 2.13 0.56***Distance to road2 1.00 0.005 0.99 0.005 0.97 0.008***Distance to major city 0.91 0.076 4.74 1.44*** 2.34 0.745***Distance to major city2 1.00 0.0006 0.99 0.002*** 1.00 0.004Elevation 1.00 0.0045 1.03 0.009*** 1.02 0.008***Slope (degrees) 1.05 0.02*** 1.04 0.016** 1.06 0.015***Distance to cleared parcel 2.08 0.28*** 1.62 0.314 ** 0.40 0.13***Distance to cleared parcel2 0.97 0.97*** 0.97 0.009*** 0.93 0.07Distance to nearest village 1.35 0.106*** 0.62 0.106*** 0.33 0.22*Distance to nearest village2 1.00 0.0016** 1.00 0.003 1.04 0.04Distance to archeo site 0.97 0.024 1.16 0.083** 0.47 0.04***pH 2.32 0.40*** 0.63 0.102*** 0.95 0.61Aspect 1.00 0.002 1.00 0.0002*** 1.00 0.0002West (dummy) 424.5 642*** 0.25 0.25 — —Constant 2.2e–6 5.2e–6*** 2.6e–23 2.7e–22*** 8.2e–7 8.9e–6Observations 293,248 294,208 232,984Pseudo R2 0.5828 0.7794 0.8384Log likelihood −78,344.1 −41,589.9 −25,643

Note: Logit results from Stata Version 12. The nonresident observations include 19 clusters, long-inhabited 15 clusters, and the recentlysettled 7 clusters (since only western concessions and controls are included).



TA

BL

EA

2C

ompa

riso

nof

Bal

ance

betw

een

Mat

ched

Dat

aSe

ts

Non

Res

iden

tL

ong-

Inha

bite

dR

ecen

tlyIn

habi

ted

Unm

atch

edM

atch

edU

nmat

ched

Mat

ched

Unm

atch

edM

atch

ed

Var

iabl

eM

ean

Dif

f.eQ

QM

ed.a

Mea

nD

iff.

eQQ

Med

.aM

ean

Dif

f.eQ

QM

ed.a

Mea

nD

iff.

eQQ

Med

.aM

ean

Dif

f.eQ

QM

ed.a

Mea

nD

iff.

eQQ

Med

.a

Dis

tanc

eto

road

6.86

6.75

1.81

1.72

−5.

716.

82−

0.03

0.67

−8.

938.

131.

061.

45D

ista

nce

toci

ty−

23.4

824

.92

−0.

966.

60−

3.48

23.1

65.

9619

.98

−14

.98

21.9

55.

206.

86D

ista

nce

tovi

llage

6.62

6.52

0.88

2.65

−5.

606.

60−

0.99

0.90

−9.

6610

.11

0.48

0.81

Dis

tanc

eto

villa

ge2

305.

921

7.7

32.0

671

.17

−23

1.9

262.

1−

40.1

021

.99

−24

0.3

200.

012

.31

5.74

Dis

tanc

eto

clea

ring

1.90

2.10

−0.

591.

36−

1.80

1.34

−0.

140.

56−

8.68

10.6

70.

860.

19D

ista

nce

tocl

eari

ng2

18.6

919

.14

−19

.24

27.8

2−

46.5

213

.78

−5.

323.

91−

152.

413

3.1

7.37

0.38

Dis

tanc

eto

arch

eosi

te−

0.86

5.22

−1.

892.

01−

5.85

3.27

−0.

972.

943.

645.

53−

3.19

2.87

Asp

ect

12.8

810

.01

1.01

4.40

−2.

606.

34−

0.36

1.30

20.2

714

.04

−4.

024.

07pH

0.34

0.36

0.10

0.05

0.55

0.50

−0.

090.

000.

510.

36−

0.13

0E

leva

tion

29.9

048

.25

12.2

211

.00

62.1

469

.00

−4.

2936

.00

29.9

335

.00

14.7

16Sl

ope

(deg

rees

)0.

570.

87−

0.36

0.00

0.38

0.46

−0.

110.

002.

201.

90−

1.86

1.97

Wes

t(l

ocat

ion

dum

my)

−0.

290.

290.

000.

00−

0.16

0.16

0.00

0.00

0.32

0.00

0.00

0.00

Obs

erva

tions

293,

248

90,1

9029

4,20

894

,204

296,

262

43,1

16

Not

e:M

atch

ing

met

hods

for

all

thre

eda

tase

tsar

ene

ares

tne

ighb

orw

ithou

trep

lace

men

t,w

ithM

ahal

anob

isdi

stan

cem

etri

cfo

rse

lect

vari

able

s.a

Med

ian

valu

esfo

rth

equ

antil

e-qu

antil

e(Q

Q)

plot

diff

eren

ces

are

used

for

all

ordi

nal

cova

riat

esan

dth

em

ean

isus

edfo

rca

tego

rica

lco

vari

ates

(wes

t).

Mat

chin

gis

cond

ucte

din

Rus

ing

“Mat

chIt

”(H

oet

al.2

011)

.


TABLE A3Panel Test for Parallel Trends in Deforestation Preconcession Policy

Recently Settled Long-Inhabited Nonresident

Variable Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Control 0.004 0.007 0.013 0.008 0.007 0.005Year 90_93 −0.003** 0.001 −0.0008* 0.0004 −0.00002** 0.00001Year 93_95 −0.002** 0.001 −0.0008* 0.0004 0.00000 0.00000Control*90_93 −0.004 0.005 −0.010 0.007 −0.005 0.003Control*93_95 −0.003 0.004 −0.003 0.002 −0.004 0.003Constant 0.007** 0.003 0.0008* 0.0004 0.00007** 0.00003Observations 129,099 282,537 270,564Groups 43,033 94,179 90,188

Note: Results based on a random effects linear probability model with panel data from three preconcessionperiods: 1990–1993, 1993–1995, and 1995–1997. Each pixel was coded as 1 if it experienced deforestationduring that period, and then remained 1 for all subsequent periods. Results based on a linear probability modelwhere the dependent variable equals 1 if pixel is deforested, 0 otherwise.


TABLE A4Test for Unobserved Bias Using the Mantel-Haenszel Test Statistic

Nonresident Long-Inhabited Recently Inhabited

Γ Pre Post Pre Post Pre Post

1 12.2 21.5 10.3 15.1 6.2 4.42 17.4 30.3 14.8 21.6 11.1 13.13 21.3 36.9 18.3 26.5 14.6 18.84 24.6 42.4 21.2 30.6 17.4 23.25 27.5 47.2 23.8 34.2 19.8 26.9

Note: Results were obtained using “mhbounds” in Stata Version 12 used for testing unobserved bias withbinary dependent variables (see Becker and Caliendo 2007). Reported values are for the Mantel-Haenszelstatistic Q+mh (assumption: overestimation of treatment effect). p-Values for all test statistics are less than0.001 and thus are not displayed in the table. Pre indicates the prepolicy period from 1990 to 1996, post indicatesthe postpolicy period from 2000 to 2008. Γ is a measure of the degree of hidden bias required to change resultsof the estimated treatment effect from the kernel matching results (Table 5).

The results for the Mantel-Haenszel test statistic(Table A4), which tests the null hypothesis that thetreatment effect is overestimated for the given levelsof Γ. We can reject the null hypothesis in all cases.The results are robust to unobserved bias for Γ up tofive (p < 0.001). Thus the significance of the treat-

ment effect would not change even in the presence ofa sizable unobserved, confounding variable. The up-per bounds on the significance levels for Γ are essen-tially zero (p < 0.001) for all levels of Γ, for allsubgroups (not shown in the table).


TABLE A6Leakage Estimates for “Buffer” Coefficient for Concession Groups

2 km Buffer 5 km Buffer 10 km Buffer

Model Unmatched Matched Unmatched Matched Unmatched Matched

Resident Indigenous

OLS −0.0411 −0.037 −0.027 −0.028 0.0008 −0.010Clustered Std. Err. (0.012)*** (0.014)*** (0.012)** (0.013)** (0.019) (0.015)Observations 96041 47246 95480 42136 94648 26932Pseudo R2 0.181 0.1474 0.1757 0.119 0.167 0.0896

Nonresident

OLS 0.0115 0.0056 0.015 −0.0014 0.015 −0.010Clustered Std. Err. (0.011) (0.009) (0.013) (0.008) (0.034) (0.032)Observations 95,007 38,176 94,723 29,674 94,602 24,568Pseudo R2 0.168 0.239 0.164 0.236 0.164 0.242

Note: Clustered standard errors in parentheses. OLS, ordinary least squares.*Significant at 10%; ** significant at 5%; *** significant at 1%.

Acknowledgments

We gratefully acknowledge Dr. Bayron Milian, ofthe Peten campus of the Universidad San Carlos, forhis assistance with the household survey. Addition-ally, the Centro de Monitoreo y Evaluacion (CEMEC)of the Consejo Nacional de Areas Protegidas

(CONAP) provided environmental data of critical im-portance for this study. We also thank Dr. Brian Roe,a professor in the Department of Agricultural, Envi-ronmental, and Development Economics at OhioState University, for his helpful comments on this pa-per and our reviewers for their thoughtful commentsand suggestions for revisions.

TABLE A5Panel Data Results with Time Trend

Recently Settled Long-Inhabited Nonresident

Variable(Dependentdeforest = 1)

With 2 kmBuffer

Without 2 kmBuffer

With 2 kmBuffer

Without 2 kmBuffer

With 2 kmBuffer

Without 2 kmBuffer

Coef. Coef. Coef. Coef. Coef. Coef.

Treat −0.003 −0.013*** −0.011* −0.014* −0.005 −0.008(0.006) (0.002) (0.006) (0.007) (0.004) (0.005)

Post 0.018 0.085*** 0.015 0.023 0.007 0.016(0.035) (0.024) (0.015) (0.019) (0.010) (0.013)

Treat × Post −0.081 −0.134*** −0.041** −0.052** −0.024 −0.034**(0.059) (0.013) (0.021) (0.023) (0.015) (0.017)

Time Trend 0.026*** 0.023*** 0.007* 0.008 0.004* 0.005(0.008) (0.006) (0.004) (0.005) (0.003) (0.003)

Constant −0.057*** −0.038*** −0.006 −0.005 −0.005 −0.003(0.016) (0.015) (0.008) (0.010) (0.005) (0.006)

Observations 344,264 204,832 753,432 675,576 721,504 585,256Groups 43,033 25,604 94,179 84,447 90,188 73,157

Note: Data are from remote sensing satellite images. Observations are based on 30 m × 30 m parcels of land. Panel format is based onperiods where deforestation was observed. There are eight periods from 1990 to 2009. Each pixel was coded as 1 if it experience deforestationduring that period and then remained 1 for all subsequent periods. Results based on a linear probability model where the dependent variableequals 1 if pixel is deforested, 0 otherwise.



References

Abadie, Alberto, David Drukker, Jane L. Herr, andGuido Imbens. 2004. “Implementing MatchingEstimators for Average Treatment Effects inStata.” Stata Journal 4 (3): 290–311.

Abadie, Alberto, and Guido Imbens. 2002. Simpleand Bias-Corrected Matching Estimators. Tech-nical report. Berkeley: Department of Economics,University of California.

———. 2006. “Large Sample Properties of MatchingEstimators for Average Treatment Effects.” Econ-ometrica 74 (1): 235–267.

Agrawal, Arun, Ashwini Chhatre, and Rebecca Har-din. 2008. “Changing Governance of the World’sForests.” Science 320 (5882): 1460–62.

Alix-Garcia, Jennifer. 2007. “A Spatial Analysis ofCommon Property Deforestation.” Journal of En-vironmental Economics and Management 53 (2):141–57.

Alix-Garcia, Jennifer, Elizabeth Shapiro, and Katha-rine Sims. 2012. “Forest Conservation and Slip-page: Evidence from Mexico’s National Paymentsfor Ecosystem Services Program.” Land Econom-ics 88 (4): 613–38.

Andam, Kwaw S., Paul J. Ferraro, Alexander Pfaff,Artur Sanchez-Azofeifa, and Juan A. Robalino.2008. “Measuring the Effectiveness of ProtectedArea Networks in Reducing Deforestation.” Pro-ceedings of the National Academy of Sciences US A 105: 16089–94.

Arriagada, Rodrigo, Paul Ferraro, Erin Sills, Su-bhrebdu Pattanayak, and Silvia Cordero-Sancho.2012. “Do Payments for Environmental ServicesAffect Forest Cover? A Farm-Level Evaluationfrom Costa Rica.” Land Economics 88 (2): 382–99.

Baland, Jean-Marie, and Jean-Philippe Platteau.2000. Halting Degradation of Natural Resources:Is There a Role for Rural Communities? Oxford:Oxford University Press. Oxford Scholarship On-line, 2005. doi: 10.1093/0198290616.001.0001.

Barbier, Edward. 2004. “Explaining AgriculturalLand Expansion and Deforestation in DevelopingCountries.” American Journal of AgriculturalEconomics 86 (5): 1347–53.

Becker, Sascha, and Marco Caliendo. 2007. “Sensi-tivity Analysis for Average Treatment E?ects.”Stata Journal 7 (1): 71–83.

Bowler, Diana, Lisette Buyung-Ali, John R. Healey,Julia P. G. Jones, Teri Knight, and Andrew S. Pul-lin. 2010. The Evidence Base for Community For-est Management as a Mechanism for SupplyingGlobal Environmental Benefits and Improving Lo-cal Welfare. A Scientific and Technical AdvisoryPanel advisory document. Rome: Global Environ-

ment Facility, FAO. Available at www.fao.org/family-farming/detail/en/c/284683/.

———. 2012. “Does Community Forest Manage-ment Provide Global Environmental Benefits andImprove Local Welfare?” Frontiers in Ecologyand the Environment 10 (1): 29–36. doi: 10.1890/110040.

Bray, David B., Elvira Duran, Victor Hugo Ramos,Jean-Francois Mas, Alejandro Velazquez, RoanMcNab, Deborah Barry, and Jeremy Rada-chowsky. 2008. “Tropical Deforestation, Com-munity Forests, and Protected Areas in the MayaForest.” Ecology and Society 13 (2): article 56.

CONAP (National Council for Protected Areas).1996. Plan maestro Reserva Biosfera Maya. Co-leccion Manejo Forestal en la Reserva de la Bios-fera Maya No. 2. Report sumbitted to the TropicalAgricultural Research and Higher Education Cen-ter (CATIE). Turrialba, Costa Rica: CONAP.

Cropper, Maureen, Jyotsna Puri, and Charles Grif-fiths. 2001. “Predicting the Location of Defores-tation: The Role of Roads and Protected Areas inNorth Thailand.” Land Economics 77 (2): 172–86.

ESRI. 2011. ArcGIS Desktop: Release 10. Redlands,CA: Environmental Systems Research Institute.

Food and Agriculture Organization. 2010. GlobalForest Resources Assessment 2010. FAO ForestryPaper 163.Rome: FAO.

Fortmann, Lea. 2014. “Assessing Factors that Con-tribute to Reduced Deforestation and SuccessfulCommunity Forest Management in Guatemala’sMaya Biosphere Reserve.” Ph.D. Dissertation,The Ohio State University, Columbus, Ohio.

Gautam, Ambika, and Ganesh P. Shivakoti. 2005.“Conditions for Successful Local Collective Ac-tion in Forestry: Some Evidence from the Hills ofNepal.” Society and Natural Resources 18 (2):153–71.

Gautam, Ambika, Edward L. Webb, and Apisit Ei-umnoh. 2002. “GIS Assessment of Land Use/LandCover Changes Associated with Community For-estry Implementation in the Middle Hills of Ne-pal.” Mountain Research and Development 22 (1):63–69.

Gibbs, H. K., A. S. Ruesch, F. Achard, M. K. Clayton,P. Holgrem, N. Ramankutty, and J. A. Foley. 2010.“Tropical Forests Were the Primary Sources ofNew Agricultural Land in the 1980s and 1990s.”Proceedings of the National Academy of SciencesU S A 107 (38): 16732–37.

Gomez, Ileana, and Ernesto Mendez. 2007. Associa-tion of Forest Communities of Peten, GuatemalaContext, Accomplishments and Challenges. Bo-gor, Indonesia: Center for International ForestryResearch (CIFOR).

Grunberg, Ramos, and Victor Hugo. 1998. Base deDatos sobre Poblcacion, Tierra y Medio Ambiente


en la Researva de la Biosfera Maya. Guatemala:CARE-Guatemala, Centro de Monitoreo y Eval-ucacion de CONAP, USAID, and CooperacionAustriaca para el Desarrollo.

Guo, Shenyang, and Mark W. Fraser. 2010. Propen-sity Score Analysis: Statistical Methods and Ap-plications. Thousand Oaks, CA: Sage Publica-tions.

Hayes, Tanya, and Lauren Persha. 2010. “Nesting Lo-cal Forestry Initiatives: Revisiting CommunityForest Management in a REDD+ World.” ForestPolicy and Economics 12 (8): 545–53.

Ho, Daniel, Kosuke Imai, Gary King, and ElizabethStuart. 2007. “Matching as Nonparametric Prepro-cessing for Reducing Model Dependence in Para-metric Causal Inference.” Political Analysis 15(3): 199–236.

———. 2011. “Matchit: Nonparametric Preprocess-ing for Parametric Causal Inference.” Journal ofStatistical Software 42 (8). doi: 10.18637/jss.v042.i08 react-text: 69.

Honey-Roses, Jordi, Kathy Bayliss, and Isabel Ra-mırez. 2011. “A Spatially Explicit Estimate ofAvoided Forest Loss.” Conservation Biology 25(5): 1032–43.

Joppa, Lucas, Scott Loarie, and Stuart Pimm. 2008.On the Protection of “Protected Areas.” Proceed-ings of the National Academy of Sciences U S A105 (18): 6673–78.

Kumar, Sanjay. 2002. “Does “Participation” in Com-mon Pool Resource Management Help the Poor?A Social Cost-Benefit Analysis of Joint ForestManagement in Jharkhand, India.” World Devel-opment 30 (5): 763–82.

Larson, Anne, and Fernanda Soto. 2008. “Decentral-ization of Natural Resource Governance Re-gimes.” Annual Review of Environmental Re-sources 33: 213–39.

Maas, Raul, and Claudio Cabrera. 2008. Forest Con-cessions Assessment in the Maya Biosphere Re-serve, Peten, Guatemala. Final report. Washing-ton, DC: U.S. Agency for InternationalDevelopment.

Man, Li, JunJie Wu, and Xiangzheng Deng. 2013.“Identifying Drivers of Land Use Change inChina: A Spatial Multinomial Logit Model Anal-ysis.” Land Economics 89 (4): 632–54.

McSweeney, Kendra, Erik Nielsen, Matthew Taylor,David Wrathall, Zoe Pearson, Ophelia Wang, andSpencer Plumb. 2014. “Drug Policy as Conser-vation Policy: Narco-Deforestation.” Science 343(6170): 489–90.

Murray, Brian C., Bruce McCarl, and Heng-Chi Lee.2004. “Estimating Leakage from Forest CarbonSequestration Programs.” Land Economics 80 (1):109–24.

Nelson, Andrew, and Kenneth M. Chomitz. 2011.“Effectiveness of Strict vs. Multiple Use ProtectedAreas in Reducing Tropical Forest Fires: A GlobalAnalysis Using Matching Methods.” PLoS ONE6 (8). doi.org/10.1371/journal.pone.0022722.

Nelson, Gerald, Virginia Harris, and Steven Stone.2001. “Deforestation, Land Use, and PropertyRights: Empirical Evidence from Darien, Pan-ama.” Land Economics 77 (2): 187–205.

Nepstad, D., S. Schwartzman, B. Bamberger, M. San-tilli, D. Ray, P. Schlesinger, P. Lefebvre, A. Alen-car, E. Prinz, Greg Fiske, and Alicia Rolla. 2006.“Inhibition of Amazon Deforestation and Fire byParks and Indigenous Lands.” Conservation Bi-ology 20 (1): 65–73.

Nittler, John, and Henry Tschinkel. 2005. CommunityForest Management in the Maya Biosphere Re-serve of Guatemala: Protection through Profits.Report for the U.S. Agency for International De-velopment, Washington, D.C., and SustainableAgriculture and Natural Resource ManagementCollaborative Research Support Program, Univer-sity of Georgia, Athens, Georgia.

Ostrom, Eleanor. 1990. Governing the Commons: TheEvolution of Institutions for Collective Action, Po-litical Economy of Institutions and Decisions.New York: Cambridge University Press.

Pfaff, Alexander. 1999. “What Drives-Deforestation in the Brazilian Amazon?: Evidencefrom Satellite and Socioeconomic Data.” Journalof Environmental Economics and Management 37(1): 26–43.

Phelps, Jacob, Edward Webb, and Arun Agrawal.2010. “Does REDD+ Threaten to RecentralizeForest Governance?” Science 328 (5976): 312–13.

Porter-Bolland, Luciana, Edward Ellis, Manuel Guar-iguata, Isabel Ruiz-Mallen, Simoneta Negrete-Yankelevich, and Victoria Reyes-Garcıa. 2012.“Community Managed Forests and Forest Pro-tected Areas: An Assessment of Their Conserva-tion Effectiveness across the Tropics.” ForestEcology and Management 268: 6–17.

Poteete, Amy R., and Eleanor Ostrom. 2004. “Het-erogeneity, Group Size and Collective Action: TheRole of Institutions in Forest Management.” De-velopment and Change 35 (3): 435–61.

Radachowsky, Jeremy, Victor H. Ramos, RoanMcNab, Erick H. Baur, and Nikolay Kazakov.2012. “Forest Concessions in the Maya BiosphereReserve, Guatemala: A Decade Later.” ForestEcology and Management 268: 18–28.

Rosenbaum, Paul R. 2002. Observational Studies.2nd ed. New York: Springer.

Sader, S. A., D. J. Hayes, J. A. Hepinstall, M. Coan,and C. Soza. 2001. “Forest Change Monitoring ofa Remote Biosphere Reserve.” International Jour-nal of Remote Sensing 22 (10): 1937−50.


Schwartz, Norman B. 1990. Forest Society: A SocialHistory of Peten. Pennsylvania Press.

Sohngen, Brent, and Sandra Brown. 2004. “Measur-ing Leakage from Carbon Projects in Open Econ-omies: A Stop Timber Harvesting Project as aCase Study.” Canadian Journal of Forest Re-search 34 (4): 829–39.

Somanathan, Eswaran, Rajagopal Prabhakar, andBhupendra S. Mehta. 2009. “Decentralization forCost-effective Conservation.” Proceedings of theNational Academy of Sciences U S A 106 (11):4143–47.

Wildlife Conservation Society. 2013. Maya BiosphereReserve Living Landscape (Guatemala). TheGlobal Conservation Program Achievements andLessons Learned from 10 Years of Support forThreats-Based Conservation at a Landscape andSeascape Scale. Report for U.S. Agency for Inter-national Development. Washington, DC: U.S.Agency for International Development. Availableat http://pdf.usaid.gov/pdf_docs/Pdacp725.pdf.

Wooldridge, Jeffrey M. 2002. Econometric Analysisof Cross Section and Panel Data. Cambridge, MA:MIT Press.

Copyright of Land Economics is the property of University of Wisconsin Press and its contentmay not be copied or emailed to multiple sites or posted to a listserv without the copyrightholder's express written permission. However, users may print, download, or email articles forindividual use.

assessing the role of group heterogeneity in community ... · assessing the role of group...

Documents