Conserving Forests: Mandates, Management or Money?

Conserving Forests: Mandates, Management or Money?

Kathy Baylis1, Jordi Honey-Rosés2, M. Isabel Ramírez3

(Very) Preliminary Draft, March 2013

1

Assistant Professor, Department of Agriculture and Consumer Economics, University of Illinois, UrbanaChampaign, 1301 W. Gregory Dr., Urbana. IL 61801, (217) 244 6653, baylis@illinois.edu 2 Postdoctoral Researcher, Catalan Institute for Water Research (ICRA) Edifici H20, Parc Científic i Tecnològic de la Universitat de Girona, Girona, Spain 3 Research Professor, Centro de Investigaciones en Geografía Ambiental (CIGA), Universidad Nacional Autónoma de México (UNAM) The authors would like to thank the University of Illinois and CONACYT for financial support.

1.

Abstract Decision-makers are keen to learn which policy instruments are most effective at preserving forest cover. Using data from a patchwork of programs designed to preserve the overwintering forest habitat of the Monarch butterfly in central Mexico, we compare the effectiveness of three instruments to limit deforestation and forest degradation: logging bans, payment for ecosystem services (PES), and forest management. Using a matched sample of one hectare parcels and a spatial lag model of deforestation, we find that for preserving forest, PES is the most effective, particularly in those communities that previously had a forest management plan in place. PES and logging bans on their own are of limited use in limiting forest degradation, while in those communities with pre-existing management plans, PES appears to protect dense forest. We also observe positive conservation effects of management on both forest and dense forest cover and this effect increases the longer the plan is in place.

Keywords: Payment for Environmental Services; Mexico; deforestation; logging regulation; protected area status; community forest management; spatial econometrics.

JEL Codes: Q23, Q28, Q56, R14

2.

Introduction As pressure builds to protect valuable carbon stocks and other forest ecosystem services, forest conservation policies are attracting tremendous attention from policy makers (Pagiola et al.

2002; Wunder 2007; Harvey et al. 2010). The possibility that forest conservation policies will be funded by carbon sequestration projects will soon oblige decision makers to decide how to invest conservation dollars. To make wise choices, policy makers will need to know which conservation instruments are more likely to achieve desired outcomes. The United Nations Program on Reduced Emissions from Deforestation and Forest Degradation (REDD)+ offers national governments flexibility in choosing which forest protection measures to use (Harvey et al. 2010), and most agree that national governments will adopt a combination of instruments, including outright logging bans (protected areas), improved forest management, integrated community development programs, and performance payments programs such as payment for environmental services (PES) schemes. Views on which conservation measure is preferable has largely broken down along ideological lines (Berkes 2003, Neumann 2005), with little empirical work comparing the effectiveness of these measures in meeting their conservation goals (Ferraro & Pattanayak 2006).

In this paper we compare conservation outcomes of three forest conservation policies: (1) legal protection through the declaration of protected areas, such as establishing a logging ban, (2) economic incentives through a payment for environmental services (PES) program and (3) community forest management (CFM). We explore both deforestation and forest degradation outcomes from a patchwork of these programs in place at the overwintering site of the monarch butterflies (Danaus Plexippus) in central Mexico.

Establishing logging bans within protected areas is a well-established approach for protecting forests. Empirical research has shown that protected area designations have been effective (Bruner et al. 2001, Nelson and Chomitz 2011) although somewhat less effective than what naĂŻve comparisons might lead one to believe (Andam et al 2008, Joppa and Pfaff 2010). Since protected areas are not designated randomly, to determine their effect one needs to estimate what would have happened in absence of protection. Only recently have empirical studies begun to assess the effectiveness of protected areas using empirically-generated counterfactuals (eg Andam et al 2008, Nelson and Chomitz 2011, Gaveau et al. 2012).

Protected areas have been criticized for not attending to the needs of local communities and excluding them from decision-making (Berkes 2004; Neumann 2005). The “fortress model� of 3.

protection, whereby local people are removed or excluded from protected areas, has given way to community-based measures that seek to engage with land owners or indigenous peoples through community forestry, co-management or development projects (Klooster 1999, Bowler et al. 2012). To avoid some of the resentments and monitoring problems associated with protected areas, policy makers have developed community based strategies that are less restrictive and more participatory (Berkes 2004).

Forest management is another approach for maintaining forest habitat, with particularly strong support in Mexico (Bray et al. 2003; Antinori & Bray 2005). Establishing a forest management plans have the potential to help improve forest governance through community education and involvement. This approach advocates for engaging forest owners in the management of their natural resources to generate both conservation and development benefits (Bray et al. 2008). A number of governments, including Mexico, facilitate and subsidize the development of forest management plans in the hopes that community forestry can both provide a long-run source of income and protect forests (Klooster 2002; Antinori & Rausser 2007). Because the plans allow resource extraction and their implementation is highly varied, establishing their contribution to forest conservation is challenging (Porter-Bolland et al. 2012).

Most recently, conservationists have begun to pay land owners to meet conservation objectives. This payment approach has been proposed as an alternative to protected areas, forest management or integrated community development programs (ICDPs) (Ferraro and Kiss 2006). In the last decade, Payment for Environmental Services (PES) programs have been rolled out throughout the tropics in countries such as Mexico, Ecuador, and Costa Rica (Pagiola et al.

2002, Wunder 2007). By creating financial incentives for forest protection, a PES program can improve monitoring and enforcement outcomes. To be successful, however, PES programs must be executed in sites with good governance and tenure security (Engel & Palmer 2007; Engel et al. 2008). Further, because most participation in PES programs is voluntary, concern has been raised that PES funders may be paying to protect forest that would have remained unlogged regardless (Sánchez-Azofeifa et al. 2007). As with protected areas, the evidence of PES program success has shown mixed results. Pfaff, Robalino and Sánchez-Azofeifa (2008) found that PES contracts signed between 1997 and 2005 prevented deforestation in less than 1% of the land under contract. Yet when looking at the PES program on a regional scale, Sarapiquí region of Costa Rica, showed that participation in PES contracts increased farm forest cover by about 11% to 17% over eight years (Arriagada et al. 2012).

Mexico’s National Payment for Hydrological Services Program was one of the first in Latin America (Muñoz-Piña et al. 2007). The program paid land owners between $300 and $400 MX Pesos ($26-$37 USD) per hectare of conserved forest between 2003 and 2009, covering 2.27 4.

million hectares of forest in Mexico. A recent evaluation of this program shows that enrolled lands with PES contracts probably would have remained forest anyway. The avoided deforestation that can be attributed to the program is modest at around 1.17% (Alix-Garcia, Shapiro and Sims 2011). The authors also found evidence of leakage or “slippage”, in which property owners shifted logging activities away from forest lands enrolled in the PES program to other areas where no restrictions were applied. Alix Garcia and colleagues estimate that leakage was more pronounced in poorer communities, and slippage effects reduced overall estimates of avoided deforestation by 4%.

Only a handful of authors have compared the effectiveness of various conservation instruments. Theoretical modeling of forest management policies suggest that command and control measures - such as protected areas - are least likely to produce efficient outcomes or optimize the supply of ecosystem services when compared to other policies (Riera et al. 2007).

To understand the extent to which community forest management is effective, and to understand the conditions that improve the probability of their success, several researchers have conducted meta-analysis on community forestry projects (Agrawal & Chhatre 2006; Pagdee et al. 2006; Lund et al. 2010). These studies conclude that successful forest conservation is associated with a handful of characteristics including secure land tenure, the effective enforcement of community rules, congruence between the biophysical and socioeconomic boundaries of the resource, strong leadership, and the capability to mobilize local resources.

A few case studies compare the effectiveness of at least two conservation policies. A comparison of community managed forest and protected areas in the Maya Forest of Guatemala and Mexico found that deforestation rates were higher in protected areas than in community forests, but the differences were not statistically significant (Bray et al 2008). The authors conclude that governance is key to conservation outcomes as poorly governed forests were not effective at protecting forest cover regardless if the forests were managed by central governments or communities.

Nelson and Chomitz (2011) compared the effectiveness of different types of protected areas on a global scale. They compared “strict” protection with “multiple-use” areas that permit various forms of resource extraction. In Latin America they included a comparison of indigenous protected areas. They used a binary measurement of deforestation, with forest fires as a proxy for deforestation. Across each continent in the tropics, multi-use protected areas were found to be more effective than strict protected areas, and in Latin America, indigenous protected areas were almost twice as effective as any other form of protection (Nelson and Chomitz 5.

2011). In general, they concluded that some mixed-use areas are as effective or more effective as strict legal protection.

Researchers have also used meta-analysis to compare the effectiveness of protected areas and community forest management. Two comprehensive meta-analyses have recently sought to compare legal protection and CFM in the tropics (Porter-Bolland et al. 2012; Bowler et al.

2012). Porter-Bollund et al. (2012) searched the literature for studies that reported deforestation rates in protected areas or in community managed forests, and found 40 and 33 studies respectively. To compare the effectiveness of these conservation approaches, they calculated the mean deforestation rate for protected areas (1.47) and community managed forests (0.24). Given that community managed forests, on average, reported less forest change, they conclude that community managed forests are at least as effective, if not more effective, at protecting forest cover than protected areas (Porter-Bolland et al. 2012). Porter-Bolland et al. recognize that this meta-analysis suffers from a selection bias. The outcomes of CFM are likely to overstate conservation outcomes, because successful CFM projects are more likely to be studied. In addition, the outcomes of protected areas may overstate deforestation, if particular attention is placed on problematic protected areas where there is known deforestation.

Bowler et al (2012) also used a meta-analysis to compare the effectiveness of PA and CFM however they were more selective in the studies chosen. They limited their meta-analysis to studies that explicitly compared the environmental and livelihood outcomes of community forest management with a comparison site. Only 42 studies fit this criterion. Nevertheless, most of these studies still had basic methodological issues. For example, the majority did not have baseline information prior to the implementation of CMF, few investigated potential confounding variables, and many lacked quantitative data on ecosystem conditions. Bowler et al. (2012) find weak evidence that community forest management provided better environmental outcomes than protected areas. They conclude that the evidence base for community forest management is weak because of the lack of rigorously designed studies.

In this paper, we seek to compare the effectiveness of a protected area, a PES program and forest management in the overwintering habitat of the Monarch butterfly. To our knowledge, the effectiveness of PES programs has not been compared to the effectiveness of other conservation policies such as legal protection or CFM. Nor has a study compared the effectiveness of all three conservation measures in the same study area or through a metaanalysis. In earlier work, we demonstrate the combined protected area status and PES appears to preserve these forests (Honey-RosĂŠs et al 2011). Our previous research does not disentangle the different effects of these instruments and does not consider the effect of forest management in the region. Here, we use detailed remote sensing imagery between 1986 and 6.

2012 to evaluate whether these individual instruments and/or their combination has changed community behavior to conserve forest. Unlike much previous work, we have the advantage of having data from before all three programs were implemented, allowing us to use the timing of implementation to identify the programs’ effects.

Since the decision to deforest one parcel is not independent of the decision to deforest its neighbors, we take parcel location into account (Robalino and Pfaff 2012). To analyze the effect of these instruments, we use a novel spatial matching technique developed in Honey-RosĂŠs et al (2011) that matches over the characteristics of neighboring parcels along with time-invariant characteristics of the parcel itself. We use this method to create a dataset and then use a spatial lag panel data model to estimate the effect of these programs on deforestation and forest degradation taking neighboring outcomes into account. Further, like our earlier work, we are able to distinguish between the effect of these programs on deforestation and forest degradation, where deforestation is a loss in forest, defined as having a minimum 40% forest cover, while degradation is a loss in dense forest cover, defined as having 70% canopy cover.

Since environmental benefits, such as providing overwintering habitat for the Monarch Butterfly, relies upon not only forests, but dense forests, we feel it is important to consider the both outcomes. We find that the expansion of the protected area had little effect on limiting illegal logging, but that payments help reduce deforestation but not maintain the dense forest cover. Forest management appears successful in limiting forest degradation and therefore preserving dense forest cover.

Data and Methods Our study area covers 342,774 hectares in central Mexico, and includes four Mexican states and 24 municipalities. Lower elevations have been converted to agricultural fields of corn, beans or horticulture, while forests of oak (Quercus sp.), pine (Pinus sp.) or fir (Abies sp.) thrive in higher elevations. Most land is held communally as Ejidos or Indigenous Communities. Our study area is shown in figure 1.

Our study area includes the overwintering habitat for Monarch butterflies east of the Rocky mountains. Each fall, billions of butterflies migrate to a handful of mountain tops in central Mexico where they find the unique climatic conditions necessary for winter survival (Calvert & Brower 1986; Brower et al. 2008).

7.

Figure 1. Study area surrounding the Monarch Butterfly Biosphere Reserve in Mexico (source: Honey-Roses et al 2011)

Protected Area: The Monarch Butterfly Biosphere Reserve To protect this natural wonder, The Monarch Butterfly Biosphere Reserve (MBBR) was originally created by the federal government set aside 4,514 ha of forest as a no-cut zone within a protected area. Yet by the late 1990s, butterfly conservationists deemed this protection insufficient. Legal restrictions on timber harvest were met with considerable local resistance, and viewed as an unwelcome imposition (Chapela and Barkin 1995). The continued threat to the monarch led an international team of biologists to map out the minimum habitat necessary for the butterflies’ survival in Mexico, and soon momentum gathered behind their proposal to enlarge the protected area (Missrie & Nelson 2007). In 2000, the Mexican federal government enlarged the boundaries of the protected area to include 56,259 hectares, of which 13,551 8.

were declared as no-cut zones. The boundary of the no-cut zone was established using transparent geographic criteria such as elevation, slope and aspect to capture the prime habitat for the monarch butterfly. Unlike most protected areas which often fall in remote and uninhabited areas, the MBBR is surrounded by inhabited areas, and the sites being protected are under considerable logging pressure.

PES: The Monarch Butterfly Conservation Fund The Monarch Butterfly Conservation Fund (MBCF) was established in 2000 as part of a negotiated agreement between the Mexican Government, local communities, conservation groups and foundations. The MBCF makes two types of payments to participating landowners.

In June of each year, timber payments are made to communities who lost logging rights as a result of the new protected area boundaries. The second payment is made on a per hectare basis to all communities in the core zone of the protected area. Of the 40 properties in the core zone of the MBBR, the federal and state properties are ineligible for participation in the fund program. Of the 38 eligible property owners, 31 agreed to participate in the MBCF. An interesting attribute of the fund was that it did not cover all regions that saw a change in regulation. Within the protected area, lands are owned by state of Michoacรกn and the federal government. Similarly, areas in the earlier reserve only saw the introduction of a payment, but no change in legal protection status. Last, some areas had forest management plans prior to the legal restrictions while others did not. This mosaic of affected communities and community parcels allows us to explore the differential effect of these various conservation instruments.

Forest Management To legally harvest trees, a property owner needs to establish a forest management plan.

Management plans are developed by certified foresters and then approved by the forestry department. Rotations are on a ten year cycle. We collected data on management plans by community in the States of Mexico and Michoacan, including the number of permits, when those permits began and how long the permits last. Anecdotally a number of communities rushed to establish logging permits when it became evident that a PES program might be instituted to pay out based on existing logging rights. Therefore we test whether those PESeligible parcels that saw their first forest management permits established in 1999 and 2000, right before the PES payment came into place, have a different effect than those management plans initiated at other years.

9.

Figure 2. Forest communities with forest management plans are in darker grey. The boundaries of core zone the Monarch Butterfly Biosphere Reserve is in black.

The study area was divided into a uniform grid of cells 1 ha each (100 m x 100 m) for a total of 342,774 cells. We used ESRI ArcMap 9.3 for spatial transformations and analysis. We linked each cell with basic biophysical information: mean elevation, slope, aspect, distance to roads, and presence or absence of monarch colonies. Each cell also included political information such as State Government, Municipality and local community. Most rural communities in Mexico are community owned and managed as either Ejidos or Indigenous Communities. In areas where the management structure was unknown, we assumed the area to be a private property for a total of 1143 property units. We combined various sources of ownership data in order to generate the most complete dataset. These sources included data from World Wildlife Fund and the Mexican Federal Government, including the formal boundaries established under the PROCEDE (Programa de Certificaci贸n de Derechos Ejidales) program, which certified land tenure rights for common property. Using data from Landsat imagery in 1993, 2000, 2003, 2006, 2009 and 2012, we calculated the total hectares of conserved forest and total forest per cell (0.0-1) and the dominant tree species. We compare the forest cover in these cells by treatment group over time, in a difference-in-difference approach.

10.

Data Analysis We estimate the effect of protected area designation and payments for environmental services by comparing forest outcomes in those 1 hectare parcels with and without the various ‘treatments’. One concern is that allocation of the ‘treatments’ was not random, and therefore their assignment needs to be taken into consideration when developing an appropriate counterfactual (Ferraro 2009). When the no-logging areas were redefined in 2000, the areas set aside for no-logging were determined based on observable geographic and biological features such as slope, aspect, elevation, and presence of monarch colonies. Our matching procedure finds similar parcels based on these characteristics, as well as additional features such as distance to road, property tenure type, and state, to ensure that difference in these characteristics do not bias our estimate of treatment. However, communities chose to initiate management plans based on unobservable characteristics, including governance ability.

Governance ability also clearly affects forest outcomes. We address this concern in three ways.

First, in our main specification we control for parcel-level fixed effects, so that if unobservable governance characteristics do not change over time, these effects will be captured by the parcel fixed effects. We identify the effect of management by considering those parcels who fall under a management regime for the first time during our study period. Second, we estimate the effect of management explicitly controlling for other governance measures in the spatial lag regression to derive a lower bound of the effect of forest management. The intuition behind this approach is that the governance measures should control for the direct effect of governance on forest outcomes leaving the coefficient on management to pick up the effect only coming from forest management. Note that this approach assumes that instituting a forest management plan has no effect on improving community governance, which is likely untrue. Therefore, we treat this second estimate as a lower bound estimate on the effect of management.

We begin by considering only those 1 hectare parcels that are at least part forest in 1986. We use a spatial matching technique described in Honey-Roses et al (2011) to generate a set of control observations that share similar values of covariates with our treatment observations.

Matching allows researchers to circumvent the obstacles created by a non-random distribution of protected areas or program participation (Joppa and Pfaff 2010). Estimates of avoided deforestation using matching estimates usually reduce more naïve comparisons of insideoutside. Matching estimates are increasingly used to evaluate conservation programs. When investigating the impact of PES programs in both Costa Rica and Mexico, the use of matching estimators were found to reduce the estimates of avoided deforestation (Andam et al 2008, Alix-Garcia, Shapiro and Sims 2012).

11.

We use spatial matching to take the spatial nature of deforestation into account, where deforestation in one observation makes deforestation in the neighboring observation more likely (Pfaff 2009, Robalino and Pfaff 2012). The matching routine chooses control observations based on the method developed by Abadie and Imbens (2006) but matches over characteristics of the parcel itself and the characteristics of its neighbours. We define a ‘neighbour’ using a contiguity-based weights matrix because we observe that deforestation spreads contiguously, presumably due to the construction of logging roads and the movement of machinery makes it easier to deforest contiguous plots. A second aspect of the spatial match is that we exclude those control observations that are contiguous to a treatment group, as implementation of the treatment may directly affect the forest outcome in the neighbouring control observation, essentially contaminating that observation as a clean control. We specifically match over physical characteristics of the parcel, such as slope, elevation, distance to road, dominant tree species, size of the ecological patch and governance characteristics such as parcel ownership and state.

Once we generate our sets of comparable control observations, we use a linear spatial regression to estimate the percent of forest and the percent of dense forest in an observation.

We determine whether deforestation follows a spatial lag and/or error process and explicitly control for this spatial effect using panel data methods.

We use a difference-in-difference approach to identify the effect of treatment as shown in equation (1). We estimate the number of hectares offorest and dense forest cover(Fit) as a function of a spatial lag (WFit), the variables used for matching (Zit), the year (Yt), an indicator of whether the parcel receives a specific treatment program (Pi) and the interaction between the year the treatment is introduced and the dummy for those parcels receiving treatment (Yt Pi).

In the case of payment, we estimate the regression both with a dummy variable for payment.

The coefficient on last interaction term (δ) is the measure of treatment. In our case we will have three terms, one to capture the effect of being in the protected area, one to capture the effect of payment and one to capture whether the community has a forest management plan in place.

(1) Our treatment group for the protected area is defined as those parcels that moved into the protected area in 2000. Thus, as part of our control, we consider those parcels that were already part of the buffer zone of the protected area in 1986, but within the core area where logging was banned entirely. Treatment parcels for the PES program included all those joining the PES program in 2000. We drop communities from our sample who were eligible for the PES, but chose to not participate. In later work we hope to both estimate an intent to treat effect and instrument for this participation decision. We also drop the landcover data from 12.

2000 since while the satellite imagery came several months before the implementation of the PES program and the logging ban, anecdotal evidence indicates there was substantial anticipatory logging in 2000 which would bias our treatment estimates. Last, we use data on community forest management plans to generate a variable that designates when the plan is in place. All parcels in a community when such a plan is in place are designated as being ‘treated’ by the management plan. Select summary statistics are given in table 1 for each of our treatment and control groups. We include the summary statistics for the total study area, the treatment groups and their matched controls for comparison.

13.

Table 1: Summary statistics for treatment and control groups

Study Area Treatment PES Matched PES Control Treatment New Ban Matched Controls for New Ban Treatment CFM Matched Control for CFM

% dense forest in 1986 0.54

Fir in 1986

Pine in 1986

slope

Elev.

Dist. to road

Indig.

MichPriv.

Comm. oacan

N

0.22

0.64 31.45

2635 5234

0.18 0.29

0.66 126522

0.81

0.68

0.31 38.23

2983 6365

0.30 0.07

0.68

9038

0.83

0.68

0.32 38.18

2978 6403

0.30 0.06

0.68

9038

0.78

0.64

0.35 36.78

2965 6110

0.30 0.04

0.71

6359

0.79

0.64

0.35 36.76

2963 6155

0.30 0.04

0.71

6359

0.60

0.33

0.59 33.59

2707 5614

0.18 0.01

0.63

39000

0.60

0.33

0.58 33.40

2723 5175

0.18 0.01

0.63

39001

14.

As can be observed, before matching, the average characteristics of our control observations are very different from our treatments. On average, parcels in our study area have lower dense forest cover in 1986, are at a lower elevation and are less sloped than our treatment parcels, particularly those parcels in the core zone. Given the high degree of deforestation in the region, it makes sense that the parcels targeted by preservation are less accessible than other lands. Because we have multiple treatments, we first find control observations for each treatment (payment, regulation and management) and generate a final dataset that includes all treatment and all matched control observations. Effectively, this approach means we exclude those untreated observations that are not good controls for any of our treatment observations.

Our final dataset has 404685 observations.

We use a panel fixed effects regression to control for unobservable time-invariant site characteristics. One drawback of this approach is that we are unable to use our measures of community governance since they do not change over time. To observe the effects of these governance measures, we also run the regression using random effects, recognizing the results may suffer from endogeneity.

Instead of random sampling (Alix-Garcia et al 2010; Joppa and Pfaff 2010), we use all treatment observations. Because our observations are small and their outcomes are likely highly spatially correlated, we need to control for spatial spillover effects in our data. We do this using a spatial panel spatial lag regression, where the spatial lag is the average outcome of the neighbouring units. Because the outcome of the observation will also affect the outcomes of its neighbours, we instrument for the spatial lag using the second order spatial lag of the outcome, and the temporal lag of the spatially lagged outcome.

Results We regress both forest and conserved forest on the pairwise treatment and control categories using fixed effects and a spatial lag. Regression results are presented in table 2a for forest, defined as forest with greater than 40 percent canopy cover, and in table 2b, we present results for dense forest, defined as greater than 70 percent canopy cover. In model 1, we consider the effect of establishing a payment for environmental service (PES) by comparing those parcels that received a payment after 2000 against their matched controls. In model 2, we compare results for those parcels that were included in the logging ban (Ban) and therefore were subject to a logging ban, against their matched controls.

15.

Table 2a Spatial lag fixed effects regression on percent forest using treatment-specific matched controls Forest Wy PES

1 0.957*** (0.00237) 0.00458*** (0.000957)

Ban

2 0.949*** (0.00274)

3 0.956*** (0.00193)

0.00349*** (0.00114)

Permit parcel fixed effects year fixed effects Constant

yes yes 0.0359*** (0.00212)

yes yes 0.0445*** (0.00240)

0.000240 (0.000408) yes yes 0.0512*** (0.00146)

Observations Number of parcels

97,490 19,498

70,780 14,156

381,475 76,295

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 2b Spatial lag fixed effects regression on percent dense forest using treatment-specific matched controls Dense Forest Wy PES

1 0.939*** (0.00229) -0.00161 (0.00122)

PA

2 0.952*** (0.00278)

3 0.914*** (0.00152)

-0.00467*** (0.00148)

Permit parcel fixed effects year fixed effects Constant

yes yes 0.0438*** (0.00178)

yes yes 0.0369*** (0.00211)

0.00122** (0.000482) yes yes 0.0528*** (0.000905)

Observations Number of parcels

97,490 19,498

70,780 14,156

381,475 76,295

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

16.

From table 2a, we observe that both the payment and the logging ban considered separately help preserve forest cover. We see no change in forest cover after a community obtains a forest permit. As we can observe from table 2b, the policies are less promising for preserving dense forest cover. While establishing a payment scheme does not improve dense forest cover, the logging ban appears to decrease the amount of dense forest cover. Thus, establishing a protected area appears to result in forest thinning, likely by encouraging selective logging. On the other hand, establishing a forest management plan and obtaining a forestry permit appears to increase the amount of dense forest. In all cases, the spatial lag is highly significant, implying a strong spatial correlation in outcomes across parcels.

One concern with these results is that regions with management plans include parcels that have very different characteristics than those in the core zone, thus, the set of controls differs greatly from regression to regression. To explore the effect of these policies on the same sample, we repeat the above regressions with only those parcels located in the core and buffer zones and their matched controls, with the results presented in table 3.

Table 3a Spatial lag fixed effects regression on percent forest using limited matched sample Forest Wy PES

1

2

3

4

0.962*** (0.00186) 0.00435*** (0.000883)

0.962*** (0.00186)

0.962*** (0.00186)

0.962*** (0.00186) 0.00322*** (0.00119) 0.00199 (0.00131) -0.00286*** (0.000685) yes yes 0.0343*** (0.00165) 120,340 24,068

PA

0.00433*** (0.000971)

Permit parcel fixed effects year fixed effects Constant

yes yes 0.0334*** (0.00164)

yes yes 0.0340*** (0.00162)

-0.00279*** (0.000685) yes yes 0.0356*** (0.00162)

Observations Number of parcels

120,340 24,068

120,340 24,068

120,340 24,068

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3b Spatial lag fixed effects regression on percent dense forest using full matched sample Dense Forest

1

2

17.

3

4

Wy PES Ban Permit parcel fixed effects year fixed effects Constant

Observations Number of id2

0.938*** (0.00205) -0.00188 (0.00114)

0.938*** (0.00206)

0.938*** (0.00205)

0.938*** (0.00206) -0.00129 (0.00154) -0.00189 -0.000930 (0.00126) (0.00170) -0.000542 -0.000509 (0.000887) (0.000887) yes yes yes yes yes yes yes yes 0.0458*** 0.0455*** 0.0456*** 0.0462*** (0.00157) (0.00154) (0.00157) (0.00161) 120,340 24,068

120,340 24,068

120,340 24,068

120,340 24,068

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

As can be seen in table 3, the policy results for the payment and the ban more closely reflect the earlier results when we compare each treatment group to its separate set of controls.

However, the having a forest management plan in this setting has little effect. We also regress the effect of the program on the intersection of results and find qualitatively similar results (available from the authors).

Having experience with community forest management plan may also improve a communityâ&#x20AC;&#x2122;s ability to enforce compliance with the payment for environmental service. To explore this effect, we interact whether the community had a management plan in place before the introduction of the payment and the logging ban in 2000. Results are presented in table 4.

Table 4: Spatial lag fixed effects regression on forest and dense forest preservation when considering pre-existing management plans

Wy (spatial lag) PES Ban Management Plan in Place

Forest

Dense Forest

0.961*** (0.00188) 0.00169 (0.00133) 0.00157 (0.00150) -0.00243** (0.000968)

0.937*** (0.00207) -0.00385** (0.00173) -0.00123 (0.00195) 0.00284** (0.00125) 18.

PES x Management Ban x Management parcel fixed effects year fixed effects Constant

Observations Number of parcels

0.00554** (0.00269) 0.000234 (0.00298) yes yes 0.0374*** (0.00182)

0.0111*** (0.00348) -0.00313 (0.00385) yes yes 0.0543*** (0.00180)

120,340 24,068

120,340 24,068

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

As can be seen from the results in table 4, management appears to affect the success of the other policies, in particular increasing conservation created by PES. This effect is particularly larger for conserved forest, where having a PES on without pre-existing management actually decreases dense forest cover.

Next, we explore the effect of management on the larger sample. In tables 5a and 5b, we present various regression results on the effect of new forest management on percent forest and percent dense forest cover. We first consider the effect of new forest management compared to parcels without forest management in place. Second, one might be concerned that some forest management plans were introduced in anticipation of the PES payments, so we exclude those management plans initiated in 1999 and 2000. Last, we use the full set of management controls and consider the effect of new management plans controlling for existing management plans.

Table 5a Spatial lag fixed effects regression on percent forest resulting from forest management forest Wy First permit

1 Control group has no permit

2 Without anticipatory permitting

3 Includes control for existing permit

0.937*** (0.00250) 0.000835** (0.000423)

0.937*** (0.00250) 0.00106** (0.000434)

yes yes 0.000676 (0.000448)

yes yes 0.00584*** (0.000431)

0.955*** (0.00193) -0.00426*** (0.000545) 0.00730*** (0.000588) yes yes -0.00162*** (0.000448)

Permit in place parcel fixed effects year fixed effects y2009

19.

Constant

Observations Number of id2 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

0.0656*** (0.00189)

0.0605*** (0.00198)

0.0496*** (0.00146)

358,930 71,786

358,930 71,786

381,475 76,295

20.

Table 5b Spatial lag fixed effects regression on percent dense forest resulting from forest management dense forest

1 Control group has no permit

2 Without anticipatory permitting

3 Includes control for existing permit

0.908*** (0.00163) 0.00111** (0.000486)

0.908*** (0.00163) 0.00119** (0.000499)

yes yes 0.0571*** (0.000973)

yes yes 0.0652*** (0.00103)

0.914*** (0.00152) -0.00161** (0.000644) 0.00459*** (0.000694) yes yes 0.0516*** (0.000920)

358,930 71,786

358,930 71,786

381,475 76,295

Wy First permit Permit in place parcel fixed effects year fixed effects Constant

Observations Number of id2 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

In column 1 of tables 5a and 5b we present the results using those parcels without management as our control group. By contrast, the results in table 2 column 3 use parcels that see no change in management as controls. Comparing these two results, we can observe that the effect of new management plans is larger when compared against those parcels with no management in place. One concern about our measure of management is that we have anecdotal evidence that a number of communities established management plans in the years leading up to the PES purely to generate larger payments for foregone logging rights. In the data, we see a substantial increase in new permits in 1999 and 2000 which appears to support the anecdotal evidence. Since these parcels are likely to be well conserved as the community never truly intended to log these areas, one might worry that we would bias our estimate of the effect of management upward by including these anticipatory permits. In column 2 we restrict our measure of treatment to exclude those management plans established in 1999 and 2000.

We see that the measured effect remains essentially unchanged. In column 3 we explicitly control for those parcels with management in place to explore the dynamics effect of management. We see that the conservation effect of management appears to increase over time, in that the net effect of a first permit is positive but significantly smaller than the conservation effect of having a permit in place. Note that for all of these regressions, the effect of management is larger on dense forest than overall forest cover.

In table 6, we present the results using a random effects regression to explore the effect of time-invariant characteristics on forest preservation outcomes. However, these results should 21.

be treated with caution since they may be biased. First, we note that the estimated policy effects are similar to those from the fixed effect regression above. When the full matched sample is considered, as in the regressions presented in table 3, the introduction of payments and protected area status decreases both forest cover and dense forest, while forest management improves dense forest. Conversely, when only those parcels in the core and buffer region and their matches are considered, payment and regulations both improve forest cover outcomes.

22.

Table 6: Random effects spatial lag regression on forest and dense forest preservation Full Matched Sample

Wy PES PES x after 2000 PA PA x after 2000 PES x PA PES x PA x after 2000 Initiates first permit during period Permit in place % dense forest in 1986 rprocede1

ejidatars slope

elevation

distance Pine Fir Contiguous area in 1986

Core & Buffer Matches dense forest dense forest forest forest 0.917*** 0.854*** 0.946*** 0.910*** (0.00170) (0.00136) (0.00180) (0.00193) -0.0392*** -0.0591*** -0.0150*** -0.0328*** (0.00596) (0.00516) (0.00439) (0.00488) -0.00675*** -0.00913*** 0.00387*** -0.00162 (0.00125) (0.00153) (0.00127) (0.00166) 0.0210*** 0.0485*** 0.00355 0.0293*** (0.00517) (0.00445) (0.00399) (0.00441) -0.00129 0.00456 0.00623** 0.00214 (0.00335) (0.00403) (0.00314) (0.00405) 0.0176*** 0.0177*** 0.0185*** 0.0209*** (0.00400) (0.00351) (0.00282) (0.00316) 0.00364 -0.00252 -0.00445 -0.00145 (0.00368) (0.00443) (0.00344) (0.00445) -0.00658*** 0.00123 -0.00583*** -0.0100*** (0.00149) (0.00128) (0.00200) (0.00221) 0.000911** 0.00120** -0.00287*** -0.00128 (0.000399) (0.000488) (0.000683) (0.000892) 0.151*** 0.337*** 0.138*** 0.299*** (0.00181) (0.00168) (0.00267) (0.00301) 0.0195*** 0.0170*** 0.0180*** 0.0164*** (0.00188) (0.00161) (0.00226) (0.00248) -1.62e-6.55e-06 3.06e-06 05*** 1.76e-07 (4.48e-06) (3.83e-06) (5.04e-06) (5.54e-06) 0.000924*** 0.000379*** 0.000289*** -4.12e-05 (4.17e-05) (3.56e-05) (5.00e-05) (5.49e-05) -5.31e-3.41e4.46e05*** 05*** 2.68e-05*** 05*** (2.97e-06) (2.55e-06) (4.81e-06) (5.29e-06) -5.32e-1.95e-1.98e2.87e-08 07*** 06*** 06*** (2.36e-07) (2.02e-07) (4.30e-07) (4.73e-07) -0.242*** -0.0291*** -0.131*** -0.0253 (0.00639) (0.00540) (0.0185) (0.0202) -0.245*** -0.0212*** -0.132*** -0.0208 (0.00807) (0.00685) (0.0187) (0.0205) -1.20e-3.60e-3.59e1.16e-06** 23.

Michoacan spatially weighted time-invariant controls panel fixed effect year fixed effect Constant

Observations Number of id2 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

06***

07***

06***

(7.14e-08) -0.0236*** (0.00154)

(6.10e-08) -0.0286*** (0.00132)

(4.23e-07) -0.00392* (0.00215)

(4.65e-07) -0.0133*** (0.00237)

yes no yes 0.0917*** (0.00801)

yes no yes -0.000492 (0.00680)

yes no yes -0.0785*** (0.0181)

yes no yes -0.109*** (0.0198)

404,865 80,973

404,865 80,973

120,340 24,068

120,340 24,068

As can be seen, several governance variables appear to affect forest outcomes. A larger number of decision makers in the communities increases both deforestation and forest degradation in the core and buffer zone. We define the number of decision makers as the number of ejidatarios for the ejidos, the population for the indigenous communities and assumed to be an individual for private property. Second, those communities that tenured their property under the Procede program also had lower rates of deforestation and forest degradation. Note that the ability to register a property with the federal government under Procede is related to forest outcomes but is also related to general governance of the community and is therefore likely endogenous.

Other characteristics affect deforestation and degradation as expected. Slope, elevation and distance from road all indicate that those parcels that are less accessible having more forest and dense forest cover. Being in the state of Michoacan, which is known for having more lax enforcement than its neighbour, edo. de Mexico, is associated with higher levels of deforestation and degradation. All of the spatially weighted variables including the spatial lag are also highly significant.

Discussion and Conclusions In this paper we provide empirical evidence that alternate policy instruments generate different conservation outcomes. We observe little evidence that protected area status generates benefits on its own in our study region. We find that the PES helped increase forest conservation, but not dense forest cover. Thus, we see indications that communities may have 24.

received payments for conserving forest and then engaged in some selective logging, reducing dense forest cover. One possibility is that while clear cutting is known to be highly visible, selective logging may be less easy to observe by community or state monitors, leading to substitution between harvest methods. We also find that these results are very sensitive to the set of control parcels used. When we use a broader range of control observations that less precisely match the characteristics of those parcels receiving a PES, the effect switches signs.

Perhaps most promising, we see evidence that management helped preserve conserved forests, and that those parcels with pre-existing management plans also conserved more forest under the PES and logging bans.

There are a number of limitations of this paper. First, we ignore the potential endogeneity associated with forest management planning except to the degree that this endogeneity is generated by time-invariant community characteristics. Future work will use measures of community governance to instrument for forest management plans and for participation in the PES. Second, the PES program explored in our paper is never totally unbundled from regulation, so we do not observe the variation in PES participation that would occur with a truly voluntary program such as Mexicoâ&#x20AC;&#x2122;s payment for hydrological services.

That said, we believe this paper makes several contributions. First, unlike most literature, we not only see the coincidental move from no regulation to regulation + payment, we observe the initial imposition of regulation, then the expansion of the regulated area + payment. We also observe some regions that had regulatory changes but were ineligible for payment, and other regions that had no regulatory changes but did have payments. Second, we are one of the few papers to empirically estimate the effect of forest management. Third, unlike most literature, we are able to observe forest disturbance such as might occur with selective logging, not only complete deforestation. These data are particularly important for policy since the illegal logging often occurs as selective logging and the move to full deforestation is often much harder to reverse than when communities have only thinned forest. Last, we are fortunate to have data before and after the program, within the â&#x20AC;&#x2DC;treatedâ&#x20AC;&#x2122; region as well as clearly outside the region. These data better allow us to construct counterfactuals for our various treatments.

25.

References Alix-Garcia, J., E.N. Shapiro, K.R.E. Sims. 2010. Impacts of payments for ecosystem services on deforestation in Mexico: Preliminary lessons for REDD, Land Tenure Briefs. Alix-Garcia, J., E.N. Shapiro, K.R.E. Sims. 2011. Forest Conservation and Slippage: Evidence from Mexico’s National Payment for Ecosystem Services Program. Working Paper.

Alix-Garcia, J. A spatial analysis of common property deforestation. Journal of Environmental Economics and Management 53:141-157.

Andam, K., P. J. Ferraro, A. Pfaff, A. G. Sanchez-Azofeifa, and J. A. Robalino. 2008. Measuring the effectiveness of protected area networks in reducing deforestation. Proceedings on the National Academy of Sciences 105:16089-16084.

Arriagada, R.A., P.J. Ferraro, E. Sills, S.K. Pattanayak, S. Cordero-Sancho. 2012. Do payment for environmental services reduce deforestation? A farm-level evaluation from Costa Rica.

Land Economics 88(2): 382-399 Berkes, F. 2004. Rethinking Community-Based Conservation. Conservation Biology. (18)3:621630.

Bruner, A.G., R.E. Gullison, R.E. Rice, G.A.B da Fonsea. 2001. Effectiveness of Parks in Protecting Tropical Biodiversity. Science 291:126-128.

Bray, D.B., L. Merino-Pérez, P. Negreros-Castillo, G. Segura-Warnholtz, Juan Manuel TorresRojo, H.F. M. Vester 2003. Mexico' s Community-Managed Forests as a Global Model for Sustainable Landscapes. Conservation Biology 17(3): 672-677.

Bray, D. B., Merino-Pérez, L., & Barry, D. (2005). The community forests of Mexico: Managing for sustainable landscapes. Mountain Research and Development (Vol. 27, p. 372).

University of Texas Press.

Bray, D. B., E. Duran, V. H. Ramos, J.-F. Mas, A. Velazquez, R. B. McNab, D. Barry, and J. Radachowsky. 2008. Tropical deforestation, community forests, and protected areas in the Maya Forest. Ecology and Society 13(2): 56. [online] URL: http://www.ecologyandsociety.org/vol13/iss2/art56/ Bowler, D.E., Buyung-Ali, L.M., Healey, J.R., Jones, J.P.G., Knight, T.M.,and Pullin, A.S. 2012.

Frontiers in Ecology and the Environment 10(1): 29–36, doi:10.1890/110040 Chapela, G and D. Barkin 1995. Monarcas y campesinos: estrategia de desarrollo sustentable en el oriente de Michoacán. Centro de Ecología y Desarrollo. Mexico D.F., Mexico.

Chhatre, A., and A. Agrawal 2008. Forest commons and local enforcement. Proceedings of the National Academy of Sciences 105 36:13286-91.

Engel, Stefanie, Stefano Pagiola, and Sven Wunder. 2008. Designing payments for environmental services in theory and practice: An overview of the issues. Ecological Economics 65(4): 663-74.

26.

Ferraro, P. J. 2009. Counterfactual thinking and impact evaluation in environmental policy. In M. Birnbaum & P. Mickwitz (Eds.), Environmental program and policy evaluation: Addressing methodological challenges. New Directions for Evaluation 122: 75–84.

Ferraro , P.J., Kiss, A., 2002. Direct payments to conserve biodiversity. Science 298:1718-1719.

Ferraro , P.J., Simpson, R.D., 2002. The cost-effectiveness of conservation payments. Land Economics 78:339-353.

Ferraro, P.J. S. K. Pattanayak. 2006. Money for Nothing? A Call for Empirical Evaluation of Biodiversity Conservation Instruments. PLoS Biology 4:4. 482-488.

Ferraro P.J., Lawlor, K., Mullan, K.L., Pattanayak, S.K., 2012. Forest Figures: Ecosystem Services Valuation and Policy Evaluation in Developing Countries. Review of Environmental Economics and Policy. 6(1): 1-26. [Double check year] Gaveau, D. L. A., L. M. Curran, G. D. Paoli, K. M. Carlson, P. Wells, A. Besse-Rimba, D. Ratnasari, & N. Leader-Williams 2012. Examining protected area effectiveness in Sumatra: importance of regulations governing unprotected lands. Conservation Letters 5: 142– 148 Harvey, C. A., B. Dickson, and C. Kormos. 2010. Opportunities for achieving biodiversity conservation through REDD. Conservation Letters 3:56-61.

Honey-Rosés, J., K. Baylis and I. Ramirez. 2011. A spatially explicit estimate of avoided forest loss Conservation Biology 25(5):1032-1043.

Joppa, L, Pfaff, A 2010. Reassessing the forest impacts of protection: The challenge of nonrandom location and a corrective method. Annals of the New York Academy of Sciences 1185: 135–149.

Klooster, D. 1999. Community-based forestry in Mexico: Can it reverse processes of degradation? Land Degradation Development 10 (4): 365-381.

Klooster, D. 2003. Campesinos and Mexican Forest Policy during the twentieth century. Latin American Research Review 38(2): 94-126.

Nelson, A., K. Chomitz 2011. Effectiveness of Strict vs. Multiple Use Protected Areas in Reducing Tropical Forest Fires: A Global Analysis Using Matching Methods. PLoS ONE 6(8):e22722 doi:10.1371/journal.pone.0022722 Neumann, Roderick P. 2002. Imposing wilderness: struggles over livelihood and nature preservation in Africa. California Studies in Critical Human Geography. Vol. California.

University of California Press. http://www.amazon.co.uk/dp/0520234685.

———. 2005. Making Political Ecology. Ed. Alexander B Murphy. Human geography in the making. Hodder Arnold. http://www.amazon.com/dp/0340809396.

Millennium Ecosystem Assessment. 2005. Ecosystems and human well-being: synthesis.

Washington D.C.: Island Press.

27.

Muñoz-Piña, C., A. Guevara, J. M. Torres, and J. Braña. 2008. Paying for the hydrological services of Mexico's forests: Analysis, negotiations and results. Ecological Economics, 65, (4) (5/1): 725-36.

Pattanayak, S.K., S. Wunder, P.J. Ferraro 2010. Show me the money: Do payments supply ecosystem services in developing countries? Review of Environmental Economics and Policy 4(2): 254-274.

Pagiola, S., J. Bishop, and N. Landell-Mills, editors. 2002. Selling forest environmental services market-based mechanisms for conservation and development. Earthscan Publications, London, UK. Pfaff, A. 1999. What drives deforestation in the Brazilian Amazon. Journal of Environmental Economics and Management 37, 26-43.

Pfaff, A., Robalino, J.A., Sanchez-Azofeifa, G.A., 2008. Payments for Environmental Services: Empirical analysis for Costa Rica. Working Paper Series SAN08-05. Durham NC: Terry Sanford Institute of Public Policy, Duke University.

Pfaff, A., E.O. Sills, G.S. Amacher, M.J. Coren, K.Lawlor, C.Streck 2010. Policy Impacts on Deforestation: Lessons from Past Experiences to Inform New Initiatives. Nicholas Institute for Environmental Policy Solutions & Packard Foundation.

Porter-Bolland, L. E.A. Ellis, M.R. Guariguata, I. Ruiz-Mallen, S. Negrete-Yankelevich, V. ReyesGarcia. 2012. Community managed forests and forest protected areas: An assessment of their conservation effectiveness across the tropics. Forest Ecology and Management 268:6-17.

Riera, P., L. Aranda, R. Mavsar. 2007. Efficiency and equity of forest policies: A graphic analysis using the partial equilibrium framework. Forest Policy and Economics 9:852-861.

Robalino, J.A., A. Pfaff 2012. Contagious development: Neighbor interactions in deforestation.

Journal of Development Economics 97:427-436.

Sánchez-Azofeifa, A. G., A. Pfaff, J. A. Robalino, and J. P. Boomhower. 2007. Costa Rica's payment for environmental services program: Intention, implementation, and impact. Conservation Biology 21:1165-1173.

Sills, E., Arriagada, R., Pattanayak, S., Ferraro, P., Carrasco, L., Ortiz, E., Cordero, S., 2006. Impact of the PSA Program on land use. Paper presented at the Workshop on Costa Rica's Experience with Payments for Environmental Services. San José, 25–26 September 2006.

Weiss, G. 2000. Evaluation of policy instruments for protective forest management in Austria.

Forest Policy and Economics 1:243-255.

Wunder, S. 2007. The efficiency of payments for environmental services in tropical conservation. Conservation Biology 21:48-58.

28.

Data Appendix I. Census Data We obtained social and demographic data from the 2010 Mexican national census (Censo de Poblacion y Vivienda) collected by the National Institute of Statistics and Geography (INEGI â&#x20AC;&#x201C; acronym in Spanish).

The census data were organized by localities that had a specific point location which could be projected onto our study area. Each forest community could be comprised of several localities. A locality could be comprised of a large urban center, a small town, a neighborhood or a collection of rural households.

The census included data on the demographic characteristics of the locality (total population, gender, age,); fertility (percent women under and over 12, average births per municipality, marital status); mortality (child mortality rates); migration (percent born in other municipalities by age and gender); indigenous language (by age and gender); education (literacy and schooling by age and gender); employment (by age and gender), religion, marital status, and material wealth and conditions in the household (access to water, electricity, number of telephones, televisions, construction material of home, number of rooms, etc..).

While the census data by locality provided abundant information on the individuals living in each locality, the census data did not report how localities were related to each ejido or indigenous community. In cases where the locality (town, neighborhood, or small group of homes) could be spatially located within a known ejido or indigenous community, the relationship could be spatially determined. Using the spatial join tool in ArcMap (ESRI 9.3), we liked the social attributes of communities in the census (locality) data with the biophysical and forestry features of each ejido or indigenous community.

Several localities were located in sites where we did not have ownership data (ejido or ingigenous community) and we assumed that there areas were private properties. These properties were located largely in areas on control areas located on the fringes of our study area. In cases where the locality information was not tied to a community but yet was located within 100 meters of a known community, we connected the locality data with the known community. In the remaining cases, we attributed the locality data to site where it was spatially located, which consisted of a generic private property within a municipality where no other community data was available.

29.