lable at ScienceDirect

Journal of Environmental Management 130 (2013) 48e54

Contents lists avai

Journal of Environmental Management

journal homepage: www.elsevier .com/locate/ jenvman

The impact of a shade coffee certification program on forestconservation: A case study from a wild coffee forest in Ethiopia

Ryo Takahashi*, Yasuyuki TodoDepartment of International Studies, University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8563 Japan

a r t i c l e i n f o

Article history:Received 6 December 2012Received in revised form30 July 2013Accepted 10 August 2013Available online 21 September 2013

Keywords:Shade coffeeImpact evaluationPropensity score matching methodRemote sensingAfrica

* Corresponding author. Tel.: þ81 4 7136 4852; faxE-mail address: ggt000@hotmail.com (R. Takahash

0301-4797/$ e see front matter � 2013 Elsevier Ltd.http://dx.doi.org/10.1016/j.jenvman.2013.08.025

a b s t r a c t

In recent years, shade coffee certification programs have attracted increasing attention from conservationand development organizations. Certification programs offer an opportunity to link environmental andeconomic goals by providing a premium price to producers and thereby contributing to forest conser-vation. However, the significance of the conservation efforts of certification programs remains unclearbecause of a lack of empirical evidence. The purpose of this study was to examine the impact of a shadecoffee certification program on forest conservation. The study was conducted in the Belete-Gera RegionalForest Priority Area in Ethiopia, and remote sensing data of 2005 and 2010 were used to gauge thechange of forest area. Using propensity score matching estimation, we found that forests under the coffeecertification program were less likely to be deforested than forests without forest coffee. By contrast, thedifference in the degree of deforestation between forests with forest coffee but not under the certifi-cation program and forests with no forest coffee is statistically insignificant. These results suggest thatthe certification program has had a large effect on forest protection, decreasing the probability ofdeforestation by 1.7 percentage points.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Deforestation in less developed countries, especially in sub-Saharan Africa and Latin America, is a widespread problem.Tilman et al. (2001) estimated that 1 billion hectares of forestlandwill be converted to agricultural land by 2050. Because deforesta-tion is proceeding the most rapidly in areas that are rich in biodi-versity (Balmford, 1994), a loss of forest area directly leads to a lossof biodiversity.

Many of the studies on forest management have noted theimportance of traditional coffee production for forest conservationand biodiversity protection. Coffee is traditionally grown in theunderstory of shade trees, and the agroecosystems of shade-growncoffee preserve the forest and provide important refuges forbiodiversity (Greenberg et al., 1997; Mas and Dietsch, 2004;Mogueland Toledo, 1999; Perfecto et al., 1996; Perfecto and Snelling, 1995;Wunderle Jr. and Latta, 1996). In addition, Toledo and Moguel(2012) showed the potential value of traditional shaded coffeesystem in terms of the sustainability of livelihoods and landscape.However, forest areas that currently operate under the shadedcoffee system are rapidly being converted into plantations, whichoffer few or no shade trees, for modern industrial coffee production

: þ81 4 7136 4853.i).

All rights reserved.

(Moguel and Toledo, 1999; Perfecto et al., 1996). One of the majorreasons for the rapid transformation is the low yield of the shadedcoffee system (Rappole et al., 2003b). Although the coffee yield hasbeen improved by the modern coffee system, the modern systemalso has a higher environmental cost than in the past. The moderncoffee system triggers forest reduction, increased erosion, chemicalrunoff, and consolidation, the combined effects of which threatenthe long-term sustainability of the ecosystem (Perfecto et al., 1996;Rappole et al., 2003b; Staver et al., 2001). Hence, conserving forestarea under the shaded coffee system and preventing themovementto the modern industrial system are essential from an environ-mental perspective.

Particularly in recent years, shade coffee certification programshave attracted increasing attention from conservation and devel-opment organizations interested in reducing producer incentives toconvert shaded coffee forest areas into plantations for industrialproduction (Fleischer and Varangis, 2002; Perfecto et al., 2005;Philpott and Dietsch, 2003; Taylor, 2005). Such certification pro-grams offer an opportunity to link environmental and economicgoals by providing a premium price to producers who maintainshade trees and thereby contribute to the protection of forest coverand biodiversity.

The relationship between the shaded coffee system and biodi-versity (i.e., an environment rich in orchids, birds, ants, butterflies,and hymenopterans) has been the subject of many studies (Moguel

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e54 49

and Toledo, 1999; Nestel and Dickschen, 1990; Nestel et al., 1993;Nir, 1988; Perfecto et al., 2003). Many other studies have focusedon the impact of certification systems on the livelihoods of pro-ducers (Arnould et al., 2009; Barham et al., 2011; Barham andWeber, 2011; Bolwig et al., 2009; Ruben and Fort, 2012). Howev-er, few empirical studies have examined the significance of theconservation efforts of certification programs. The first ecologicalevaluation was conducted by Mas and Dietsch (2004) in Mexico,who attempted to evaluate the effect of coffee certification onbiodiversity conservation. Unfortunately, the authors selected anarea in which no farms had obtained any certification, although thearea had a high probability of being granted certification as aresearch site. Although the area was likely to meet the environ-mental criteria used by major certification programs, the resultsmay not ensure the conservation effects of the certificationprogram.

A study in Mexico by Philpott et al. (2007) evaluated theecological benefits of coffee certification programs (fair-trade la-beling organizations and Certimex producer lists) by examining thearea’s vegetation and the richness of the area’s species, such as antsand birds. Their results revealed that there were no significantdifferences between certified and uncertified areas in terms ofshade management or in the diversity of ants and birds. However,the coffee certification programs in their study issued only organicand/or fair trade certifications, and shade coffee certification wasnot part of their research, simply because no farms in the areaunder study had obtained shade coffee certification. Althoughother certification programs, such as fair trade certification andorganic certification, include environmental criteria, the main goalsof the different certification programs differ (Ponte, 2004). Forexample, the purpose of fair trade certification is to guarantee aprice floor for marginal producers in less developed countries (Basuand Hicks, 2008). Hence, it is less clear whether other certificationprograms can be expected to participate in forest conservation ef-forts (Philpott et al., 2007).

These previous studies have used regression procedures toexamine the differences in the mean outcomes of the target areasand the control areas. The results of this method are most likelybiased by the endogeneity of the criteria that are used to select

Fig. 1. The Belete-Gera regional forest priority area. The areas shown in dark gray representwithout forest coffee. The areas marked with a black color are the study areas for this inve

certificated areas (Blackman and Rivera, 2011; Imbens andWooldridge, 2009). Therefore, although a large number of theo-retical studies have focused on shade coffee certification, whether ashade coffee certification program provides a disincentive for shadecoffee producers to convert to the modern system remains unclear,and there is still a need for further empirical studies using detailedanalyses to evaluate the impact of shade coffee certification onforest conservation.

Therefore, this study aimed to evaluate the impact of the shadecoffee certification program on forest conservation. More precisely,we evaluated the impact of the certification program on theprobability of deforestation using a propensity score matching(PSM) method to control for possible biases, such as selection bias.This study provides initial insight into the impact of the shadecoffee certification on forest conservation based on detailed anal-ysis. We chose to analyze the certification program of the RainforestAlliance, a major international non-governmental organization(NGO) based in the United States, andwe selected Ethiopia as a casestudy because there is a lack of empirical evidence of this impact inthe African context (Donald, 2004).

2. Description of the study area

2.1. Description of the Belete-Gera RFPA and the project

We selected the Belete-Gera Regional Forest Priority Area (RFPA)as the study area. The Belete-Gera RFPA is 150,000 ha in size and islocated in the Gera and Shabe Sombo Districts in the Oromiya Re-gion of Ethiopia (Fig. 1). This region is part of the highland rain-forest, and the natural vegetation in this area is subject to 1500mmof annual precipitation and an annual average air temperature ofapproximately 20� Celsius. The topography of the Belete-Gera RFPAis complex, consisting of undulating hills of 1200 to 2900 m abovesea level (masl) in height, with steepmountainous terrain in certainlocations. The forest cover in the RFPA has decreased significantlydespite the government’s prohibition of wood extraction in theforest area. Between 1985 and 2010, 40 percent of the forest areawas cleared (Todo and Takahashi, 2011).

the sub-villages that produce forest coffee, and the light gray areas are the sub-villagesstigation.

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e5450

In 2003, the Japan International Cooperation Agency (JICA), aJapanese foreign aid agency, and the Oromia Forest and WildlifeEnterprise, a public institution that is responsible for forest con-servation in the Oromia Region, launched the Participatory ForestManagement Project (hereafter referred to as “the project”) toestablish participatory forest management associations that wouldbe responsible for forest protection in the RFPA. Three forest as-sociations were initially established in 2007, followed by 32 and 58associations in 2008 and 2009, respectively (Table 1). In 2010, theremaining 32 forest associations were established, and the entireforest area in the Belete-Gera RFPA was covered by 125 forestassociations.

Before the implementation of the project, the issues ofextracting wood from the forest and expanding farmland or grazingland into the forest area were left to the judgment of the individualvillagers. However, each time that a forest association was estab-lished, the border between the homestead/farmland and the forestwas clearly identified by markers painted by the associationmembers. The use of the registered forest area for purposes such asexpanding farmland, extracting wood, and planting trees wasstrictly prohibited after the establishment of a forest association,except for the necessary thinning. The members of the associationsregularly monitored and evaluated the condition of the forest. Eachtime they found open space in the registered forest area, the forestassociation held a meeting to plan for the care of the forest andprevent further degradation. In addition, forest associations plan-ted trees along the edges of the forest area to mark its boundariesand to distinguish the forest area from agroforestry. In anotherstudy, Takahashi and Todo (2012) examined the impact of forestassociations on the conversion of forest area into farmland andfound that the establishment of forest associations resulted in a 1.5percent increase in forest cover.

2.2. Wild coffee production and coffee certification

Coffee (Coffea arabica) is a native species that grows wild in theBelete-Gera RFPA. Because coffee production is not economicallypractical at high elevations (above 2300 masl), wild coffee is typi-cally found in the forest at an altitude of approximately 2000 masl(shown in the dark gray and black areas in Fig. 1). Each wild coffeearea is managed by an individual producer, and the right to harvestwild coffee is given to the producer according to traditionalagreements among villagers. Producers commonly dry the wildcoffee after harvesting and sell it as sun-dried, shade-grown coffeeto local markets, but the selling price has traditionally been fairlylow (i.e., approximately 1 US dollar per kg in 2007 and 2008).

To provide a channel of income for association members, theproject supported the forest associations in their efforts to obtainshade coffee certification (hereafter, “forest coffee certification”)from the Rainforest Alliance. Although the Rainforest Allianceoriginally worked primarily with producers of larger plantations(Méndez et al., 2010), the Rainforest Alliance also oversaw a certi-fication program that excludes modern industrial coffee producers

Table 1The number of forest associations and coffee-producing households participating inthe coffee certification program.

2007e2008 2008e2009 2009e2010 2010e2011

The number of forestassociations

3 35 93 125

The number of participatingforest associations

3 21 48 58

The number of participatinghouseholds

555 1701 2808 3050

in an effort to encourage the shaded coffee system to move towardgreater sustainability (Mas and Dietsch, 2004). The criteria used inthe program include shade criteria for tree species richness andcomposition, tree height, tree density, the number of strata in thecanopy, and the percentage of canopy cover (Philpott et al., 2007).

In the area under study, certification was extended to the forestcoffee areas within the registered forest and the areas under thenatural forest coffee system. Most of the producers also producedthe coffee using improved seeds outside of the registered forestarea, but under the modern coffee system, such coffee is strictlyexcluded from certification. Moreover, because certification wasgranted at the association level, all producers, including poor pro-ducers, had an equal opportunity to participate in the certificationprogram. In 2007, a group of 555 coffee-producing households wascertified by the NGO, and they obtained a price that was 15e20percent higher than the regular price in that year (Table 1). In 2010,58 of the 125 forest associations participated in the certificationprogram. Thus, the total number of participating coffee producerswas 3050. Approximately 60 tons of certified coffee were collectedfrom the participating associations and were sold to the interna-tional market in 2010.

In another study, Philpott and Dietsch (2003) noted theimportance of regulation by certifiers in prohibiting the extractionof wood and the conversion of primary forest areas: certifiersshould suspend certification if a regulation has been violated. In thearea under study, an auditor from the NGO visited and surveyed thecondition of the certified area and the surrounding forest area onceper year. If an expansion of the forest coffee area or a degradation ofthe forest (e.g., the logging of shade trees) was observed in thecertified area, then the certification would be withdrawn. Such anauditing system may encourage certified producers to abide by theregulations and maintain the forest areas.

3. Data

3.1. Remote sensing data

In this study, we considered only the forest-covered area in2005 and sought to determine whether that area was deforested in2010. For our analysis, we used Landsat 7 ETMþ (path/row 170/55)satellite images, with a resolution of 30 m, from January 2005 and2010.

We classified the satellite images into two groups: forest areasand non-forest areas. Forest areas are defined as areas that functionas forests either physically or socially for local communities(Southworth and Tucker, 2001). Non-forest areas include agricul-tural lands, young fallow, rangelands, cleared areas, bare soil areas,and urban areas. To distinguish the forest areas from the non-forestareas, we used the normalized difference vegetation index (NDVI),the values of which were calculated using the raster images of band3 and 4 from the satellite images. The NDVI is a measure of vege-tation that is commonly used in remote sensing studies (e.g., byTucker et al. (1985), Davenport and Nicholson (1993), and Tuckeret al. (2001)). The NDVI provides a value ranging from �1.0 to 1.0that increases with the degree of vegetation biomass (Jensen,1996).Following Southworth et al. (2004), we determined a thresholdvalue of the NDVI for each year of the satellite images to distinguishbetween the forest areas and non-forest areas. In other words, wedefined the areas with NDVI values that were greater than thethreshold value as forest areas and the areas with NDVI valuesbelow the threshold value as non-forest areas. Therefore, we simplymerged many types of forest areas, such as natural forests andcoffee forests, into one group (i.e., a forest area).

To determine the threshold value, we conducted ground-truthlocational data collection at 17 points along the boundaries

Table 3Characteristics of the households in the sample.

Certified in2007 (1)

Certified in2009 (2)

Total (3)

Number of observations 122 83 205Age of the household head 42.3 41.6 42.0

(13.0) (12.8) (12.9)Proportion of female household heads 4.1% 3.6% 3.9%Years of formal education of

the household head2.6 3.2 2.9(2.7) (2.7) (2.7)

Proportion of Muslim household heads 93.4% 98.8% 95.6%Number of household members 5.8 7.0** 6.3

(1.7) (3.6) (2.7)Proportion of male adult members

(age 15e60) (%)30.0 33.0 31.2(12.3) (22.2) (17.0)

Proportion of female adult members(age 15e60) (%)

28.2 27.1 27.7(12.2) (19.7) (15.6)

Total area of agricultural land (a)a 71.0 85.2 76.7(67.0) (69.2) (68.1)

Total area of modern coffee (a) 13.7 33.0** 21.5(15.1) (30.3) (24.4)

Note: standard deviations are in parentheses; ** in column 2 indicates that themeans in two groups (column 1 and 2) are significantly different at the 1% level.

a total area of agricultural land excludes the certified forest coffee area.

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e54 51

between the forest and non-forest areas that existed during theperiod of our study. Those ground points were established on thebasis of interviews with several local residents and forest experts.Using geographic information system (GIS) techniques, we chosethe area with the highest NDVI value for each year as the thresholdvalue for the forest areas. Although thismethodology has been usedin previous studies (e.g., Southworth et al. (2004), White andNemani (2006), and Takahashi and Todo (2012)), the NDVIthreshold value may be misestimated, which can lead to errors inestimating the probability of deforestation (Huang et al., 2008;Southworth et al., 2004). However, such errors would not causeproblems in our study for two reasons. First, because the same errorwould affect any locational unit within the same year, the foresttransition in a forest area with and without certification would beoverestimated or underestimated to the same extent. Therefore,possible errors in the estimation of forest area from the satelliteimages would not lead to bias in our estimation. Second, as Huanget al. (2008) stated, threshold-based classification can be applied ifthe threshold values are established based on expert knowledge oflocal land cover conditions. Because the values in our study weredetermined on the basis of interviews with several local residentsand local forest experts who have extensive experience in theBelete-Gera, we assume this process provide a better accuracy indelineation.

Table 4Characteristics of the observation grids.

Forest area certified in 2007: Forest area certifiedin 2009:

Forest coffeearea (1)

Forest areawithoutforest

Forestcoffeearea (3)

Forest areawithoutforest

3.2. Certified forest coffee area and observation grids

The producers who were certified by the NGO typically had therights to manage and harvest one or two forest coffee areas in theregistered forest area. To obtain the locational data of a forest coffeearea, we used global positioning system (GPS) devices to map all ofthe forest coffee areas managed by the certified producer. Weselected four forest associations (i.e., areas marked with a dark graycolor in Fig. 1) as the studied areas: two associations obtainedcertification in 2007, and the other two obtained certification in2009. We studied all of the certified producers from the associa-tions, including a total of 205 producers, and obtained the loca-tional information of 240 forest coffee areas. Of these forest coffeeareas, 148 were certified in 2007, and the remainder were certifiedin 2009 (Table 2). In this study, we defined these areas as certifiedforest coffee areas, whereas the remainder of the areas in theregistered forest area were defined as forest areas without forestcoffee.

The general characteristics of the certified forest coffee areas inthe sample are summarized in Table 2. Columns 1 and 2 provideinformation on the forest coffee areas that were certified in 2007and 2009, respectively. The average size of the forest coffee areasthat were certified in 2007 (56.4 Are) is greater than that of the

Table 2Description of certified forest coffee areas in the sample.

Certified in2007 (1)

Certified in2009 (2)

Total (3)

Number of plots 148 92 240Size of forest coffee area (a) 56.4 40.3 50.2

(107.7) (75.9) (96.9)Distance to the village (m) 203.3 86.6** 158.6

(334.1) (136.9) (281.1)Average elevation (m) 1971.0 1913.0** 1948.8

(124.8) (68.2) (110.2)Average slope (%) 13.8 12.3* 13.2

(5.9) (5.6) (5.8)

Note: standard deviations are in parentheses; ** and * column 2 indicates that themeans in the two groups (columns 1 and 2) are significantly different at the 1 and 5percent level, respectively.

areas certified in 2009 (40.3 Are), but the difference was not sta-tistically significant. Other environmental characteristics of thecertified forest coffee areas are significantly different between thegroups. Table 3 summarizes the general socioeconomic character-istics of the certified producers, who managed 240 plots. Only twocharacteristics were significantly different between the producerswho obtained certification in 2007 and those who obtained certi-fication in 2009: the number of household members and the totalarea of modern coffee.

The target forest areas were divided into square-shaped cells(30 m by 30 m). In this study, we used these grids as observationunits for the analysis. The total number of observation grids was85,323, and they were divided into four categories: forest areascertified by the program in 2007, forest areas certified by the pro-gram in 2009, forest areas without forest coffee for which a forestassociation was established in 2007, and forest areas for which anassociation was established in 2009 (Table 4). All of the

coffee (2) coffee (4)

Number of observations 1362 20,962 646 62,353Distance to the project

office (km)10.2 11.7** 7.6 11.8**(3.5) (2.7) (1.6) (2.7)

Distance tothe village (km)

0.4 1.5** 0.2 4.1**(0.4) (1.1) (0.2) (2.8)

Distance to the mainroad (km)

1.2 1.6** 2.0 4.7**(1.2) (1.1) (1.2) (3.4)

Average elevation (m) 1920.5 1955.2** 1890.4 2033.5**(126.7) (159.1) (96.0) (175.5)

Average slope (%) 11.9 15.0** 11.9 13.3**(6.0) (7.7) (4.9) (7.6)

Proportion ofAcrisol overthe observations

2.9% 18.1%** 1.9% 12.9%**

Note: standard deviations are in parentheses; ** indicates that the means of thecertified forest coffee area in 2007 and 2009 (columns 1 and 3, respectively) aresignificantly different from the means of the forest area managed by the association(columns 2 and 4, respectively) at the 1 percent level.

Table 5Results from the probit estimation.

Benchmarkestimation (1)

Additionalestimation (2)

Distance to the closest project office (km) 0.39** �0.333**(16.82) (�12.385)

Distance to the village (km) �0.87** �2.251**(�18.23) (�21.916)

Distance to the main road (km) �0.38** �0.083(�9.60) (�1.858)

Average elevation (m) �0.00 �0.000(�0.46) (�0.674)

Average slope (%) �0.03** �0.035**(�10.09) (�7.639)

Acrisol dummy �0.30** 0.146(�2.91) (1.020)

Village dummy �3.71** 1.025**(�19.39) (7.510)

Constant �1.48** 2.143**(�2.95) (4.238)

Observations 22,324 62,999Pseudo R2 0.24 0.42

Note: z-statistics are in parentheses; ** indicates statistical significance at the 1percent level.

Table 6Balancing tests.

Benchmark estimation Additional estimation

Differencebeforematching (1)

Differenceaftermatching (2)

Differencebeforematching (3)

Differenceaftermatching (4)

Distance tothe closestproject office (km)

�1.51** �0.11 �4.19** 0.01(�19.82) (�0.75) (�39.29) (0.09)

Distance tothe village (km)

�1.13** �0.02 �3.86** �0.01(�36.77) (�1.30) (�35.60) (�1.11)

Distance to the mainroad (km)

�0.47** 0.00 �2.69** �0.06(�14.64) (0.01) (�19.97) (�0.80)

Average elevation(m)

�34.70** �3.30 �143.10** 1.20(�7.87) (�0.63) (�20.70) (0.24)

Average slope (%) �3.15** 0.27 �1.41** 0.26(�14.77) (1.03) (�4.70) (0.92)

Acrisol dummy �0.15** 0.01 �0.11** 0.00(�14.49) (1.91) (�8.35) (0.23)

Village dummy �0.29** �0.01 �0.39** �0.02(�29.31) (�0.44) (�37.99) (�0.79)

Pseudo R2 0.24 0.00 0.42 0.00

Note: t-values are in parentheses; ** indicates statistical significance at the 1 percentlevel.

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e5452

characteristics of the certified forest coffee areas and the forestareas without forest coffee were significantly different (p < 0.01).

4. Methods

In this study, we evaluated the impact of forest coffee certifi-cation on forest conservation using a PSM method to reducepossible bias, such as selection bias. The PSM method is commonlyused in the study of impact evaluation. For example, Ruben and Fort(2012) used the PSM method to evaluate the impact of fair tradecertification on production and livelihood in Peru. The currentpaper specifically examines the average effect of treatment on thetreated (ATT), developed by Rosenbaum and Rubin (1983). Oneadvantage of using remote sensing data is that we can obtain paneldata for the analysis. If panel data are available, a difference-in-differences (DID) PSM estimation of the ATT can be employed, asproposed by Heckman et al. (1997, 1998). Thus, we can eliminatetime-invariant effects on the outcome variable.

In this study, to quantify the impact of the forest coffee certifi-cation program, we performed two types of estimation. First, in thebenchmark estimation, we compared the probability of deforesta-tion between the forest coffee areas certified in 2007 and the forestareawithout forest coffee, which served as the control area. Second,we compared the forest coffee area certified in 2009 and the forestarea without forest coffee in an additional estimation. Becausecertifications were issued in November 2009 and the remotesensing data that were used in this study were collected throughJanuary 2010, we defined the forest coffee area certified in 2009 asthe forest coffee area without certification. By comparing the re-sults of the two estimations, we can estimate the effect of thecertification program.

To obtain the PSM estimator of the effect of the treatment, wefirst used a probit model to examine how a target area for theprocurement of forest coffee certification is selected. We primarilyfollowed the method used in the study by Takahashi and Todo(2012) to select the variables that were used as covariates. Inaddition, because previous studies have shown that road con-struction triggers deforestation (Chomitz and Gray, 1996; Cropperet al., 1999), we added a variable for the distance to the mainroad. Therefore, the following variables were used as covariates inthe probit estimation in this study: the distance to the closestproject office (the project has two offices located in the Belete-GeraRFPA), the distance to the village, the distance to the main road, theaverage elevation, the average slope, a dummy variable for acrisol,and a dummy variable for the village. Acrisol is a soil with sub-surface accumulation of low-activity clay and low base saturation;in other words, acrisol is infertile. In this study, we included onlyenvironmental factors and excluded the socioeconomic character-istics of the producers in the estimation because there is no infor-mation on the socioeconomic characteristics of producers in theforest areas without forest coffee.

5. Results

5.1. Matching procedure

The results of the probit estimations are presented in Table 5.We found that nearly all of the variables had significant effects. Thegoodness of fit can bemeasured by the pseudo R-squared value, andboth probit estimations showed fairly large pseudo R-squaredvalues: 0.24 and 0.42.

Based on the propensity score from the probit estimation, wecreated a new control observation group to ensure that the treat-ment group and the new control group would have similar char-acteristics. In this study, we employed one-to-one matching with a

caliper size of 0.001. A common support condition must beimplemented to satisfy the overlap assumption. In other words, inthe treatment group, we omitted observations whose propensityscores were higher than the maximum score or lower than theminimum score of the observations in the control group. Eachtreatment observationwas compared with the weighted average ofall of the control observations in the common support region.

To check the characteristics of the treatment group and thecontrol group after the matching procedure, we conducted twotypes of balancing tests. First, a t-test was used to compare themean of each covariate between the treatment group and thecontrol group after thematching procedure. The results of the t-testfor the PSM estimation are presented in Table 6. The first column foreach estimation (i.e., columns 1 and 3) shows the mean differencebetween the treatment group and the control group for each co-variate before matching, and the second column (i.e., columns 2and 4) shows the mean difference after the matching, with t-valuesin parentheses. The results show that the differences in the cova-riates before matching were insignificant after the matching

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e54 53

procedure, indicating that the characteristics of the control groupwere sufficiently similar after matching.

Subsequently, we performed the probit estimation using thesample after the matching and compared the pseudo R-squaredvalue with that obtained from the probit estimation using thesample before the matching procedure. If the matching was suc-cessful, then the pseudo R-squared after the matching would have alower value than that before the matching, which would indicatethat the after-matching probit has no explanatory power. The re-sults that are shown in the lower rows of Table 6 indicate that thepseudo R-squared values of both estimations drastically decreasedto 0.00. The second balancing test confirms that there is no sys-tematic difference between the treatment and after-matchingcontrol groups.

5.2. Impact of the forest coffee certification program

After thematching procedure, we computed the PSM estimationbased on the treatment and control groups. In most studies, thestandard error is obtained by bootstrapping (Caliendo andKopeinig, 2008). In this study, we also used the bootstrappingstandard error based on 100 replications, following Smith and Todd(2005).

The results of the benchmark PSM estimation, which are givenin the first column of Table 7, show that the probability of defor-estation in the certified forest area (2.8 percent) is significantlylower than that of the forest area without forest coffee(4.5 percent), i.e., the difference is 1.7 percent. By contrast, the re-sults of the additional PSM estimation, which are given in thesecond column of Table 7, indicate that the mean difference be-tween the two groups is fairly small (i.e., 0.6 percent) and, mostimportantly, that the difference is statistically insignificant at anyreasonable level of significance. These results lead to the conclusionthat the forest coffee certification program reduced the probabilityof deforestation by 1.7 percent.

6. Discussion and conclusions

We conducted two PSM estimations to evaluate the impact ofthe certification program. We found that the probability of defor-estation in the certified forest coffee area was 1.7 percent smallerthan that of the forest area without forest coffee, but there was nosignificant difference in the probability of deforestation betweenthe non-certified forest coffee area and the forest area withoutforest coffee. Therefore, our empirical results show that the forestcoffee certification program has had a positive effect on forestconservation, which is consistent with the findings of previous

Table 7Average effects of the forest coffee certification program on the probability ofdeforestation.

Benchmarkestimation (1)

Additionalestimation (2)

Treatment group Certified forestcoffee area

Forest coffee area

Control group Forest areawithout forestcoffee

Forest area withoutforest coffee

Mean of treatmentgroup 0.028 0.118Mean of matched control group 0.045 0.124Difference: average treatment effect �0.017 �0.006Standard error 0.009 0.022t-value �2.02* �0.29Number of observations 2546 1274

Note: * indicates statistical significance at the 5 percent level.

studies on certification systems (Fleischer and Varangis, 2002;Perfecto et al., 2005; Philpott and Dietsch, 2003; Taylor, 2005).

These results could provide useful information to the fields offorest conservation and forest management. Although we selectedthe forest area with forest coffee in Ethiopia as the case study area,we believe that our findings can be applied to other forest areasoutside of Ethiopia. The results of our analysis show that we canexpect a conservation effect by integrating the certification systeminto the agroforestry system. In addition to forest coffee, forestshave other natural resources that have the potential to be certified,such as honey and herbs. Thus, we can promote forest conservationin other regions or counties by focusing on the natural resourceswithin forests and by granting certification to those resources toincrease owner incentives to engage in proper managementpractices.

In addition, these results provide empirical evidence of therelationship between different property regimes and forest con-servation. It is well recognized that establishing property rights isan important step toward achieving sustainable forest manage-ment (Arnot et al., 2011; Owubah et al., 2001; Tucker, 1999).However, the debate as to whether private or common ownershipleads to more sustainable forest management has not yet beenresolved (Glück, 2002). In the additional PSM estimation in thisstudy, we compared non-certified forest coffee area, which is al-ways privately owned and inwhich the ownership of coffee is clear,with forest areawithout forest coffee, which is commonlymanagedby the community, and we found no significant difference. Thisresult suggests that the property regime alone may not be a goodindicator of forest conservation, which is consistent with thefindings of Tucker (1999).

However, this study examined only the impact on forest con-servation (i.e., quantity), not the impact on forest degradation (i.e.,quality). As Rappole et al. (2003b) noted, one potential problemwith certification programs is that they can create incentives forproducers to convert an existing primary forest area into an areathat produces shade coffee. Although a shaded coffee system rep-resents an important refuge for biodiversity (Estrada et al., 1993), itcannot provide an ecosystem that is comparable with that of anative forest (Perfecto et al., 2003). Therefore, a certification pro-gram may instead encourage forest degradation. However, Philpottand Dietsch (2003) dispute the claims of Rappole et al. (2003b) andargue that such degradation can be prevented. However, the debatebetween Philpott and Dietsch (2003) and Rappole et al. (2003a) hasnot yet reached a consensus. Although our analysis showed that thecertification program alleviated forest clearing or transformationinto agricultural land, we could not assess forest degradation in ourcurrent analysis. Therefore, further study is needed to identify theimpact of the certification program on forest degradation.

Acknowledgments

This workwas supported by JSPS KAKENHI Grant-in-Aid for JSPSFellows Number 25:3204. This study was conducted as part of theresearch project “Impact Evaluation of Aid Projects of the JapanInternational Cooperation Agency” in the JICA Research Institute.The authors would like to thank Yasuyuki Sawada (the projectleader) and Keiichi Tsunekawa (the former director of the JICAResearch Institute) for providing us the opportunity to perform thisresearch. The authors would also like to thank Naoki Ando, MegumiMasui, Koji Noda, Shinji Ogawa, Yasuo Takahashi, Toru Inada, andparticularly Tsutomu Nishimura and Fumiaki Saso for their greatassistance in the data collection. The opinions expressed and thearguments employed in this paper are solely those of the authorsand do not necessarily reflect those of the JICA, the JICA ResearchInstitute, or any institution with which the authors are associated.

R. Takahashi, Y. Todo / Journal of Environmental Management 130 (2013) 48e5454

References

Arnot, C.D., Luckert, M.K., Boxall, P.C., 2011. What is tenure security? Conceptualimplications for empirical analysis. Land Econ. 87, 297e311.

Arnould, E.J., Plastina, A., Ball, D., 2009. Does fair trade deliver on its core valueproposition? Effects on income, educational attainment, and health in threecountries. J. Public Policy Mark 28, 186e201.

Balmford, A., 1994. Avian endemism and forest loss. Nature 372, 623e624.Barham, B.L., Callenes, M., Gitter, S., Lewis, J., Weber, J., 2011. Fair trade/organic

coffee, rural livelihoods, and the “agrarian question”: southern Mexican coffeefamilies in transition. World Devel. 39, 134e145.

Barham, B.L., Weber, J.G., 2011. The economic sustainability of certified coffee:recent evidence from Mexico and Peru. World Devel. 40, 1269e1279.

Basu, A.K., Hicks, R.L., 2008. Label performance and the willingness to pay for fairtrade coffee: a cross-national perspective. Int. J. Consum. Stud. 32, 470e478.

Blackman, A., Rivera, J., 2011. Producer-level benefits of sustainability certification.Conserv. Biol. 25, 1176e1185.

Bolwig, S., Gibbon, P., Jones, S., 2009. The economics of smallholder organic contractfarming in tropical Africa. World Devel. 37, 1094e1104.

Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation ofpropensity score matching. J. Econ. Surv. 22, 31e72.

Chomitz, K.M., Gray, D.A., 1996. Roads, land use, and deforestation: a spatial modelapplied to Belize. World Bank Econ. Rev. 10, 487e512.

Cropper, M.L., Griffiths, C., Mani, M., 1999. Roads, population pressures and defor-estation in Thailand, 1976e1989. Land Econ. 75, 58e73.

Davenport, M., Nicholson, S., 1993. On the relation between rainfall and thenormalized difference vegetation index for diverse vegetation types in EastAfrica. Int. J. Remote Sens. 14, 2369e2389.

Donald, P.F., 2004. Biodiversity impacts of some agricultural commodity productionsystems. Conserv. Biol. 18, 17e38.

Estrada, A., Coates-Estrada, R., Meritt Jr., D., 1993. Bat species richness and abun-dance in tropical rain forest fragments and in agricultural habitats at LosTuxtlas, Mexico. Ecography 16, 309e318.

Fleischer, G., Varangis, P., 2002. Toward more sustainable coffee: consumers fueldemand for more sustainable agriculture. Agr. Technol. Notes 23.

Glück, P., 2002. Property rights and multipurpose mountain forest management.For. Policy Econ. 4, 125e134.

Greenberg, R., Bichier, P., Angon, A.C., Reitsma, R., 1997. Bird populations in shadeand sun coffee plantations in central Guatemala. Conserv. Biol. 11, 448e459.

Heckman, J.J., Ichimura, H., Todd, P., 1998. Matching as an econometric evaluationestimator. Rev. Econ. Stud. 65, 261e294.

Heckman, J.J., Ichimura, H., Todd, P.E., 1997. Matching as an econometric evaluationestimator: evidence from evaluating a job training programme. Rev. Econ. Stud.64, 605e654.

Huang, C., Song, K., Kim, S., Townshend, J.R., Davis, P., Masek, J.G., Goward, S.N.,2008. Use of a dark object concept and support vector machines to automateforest cover change analysis. Remote Sens. Environ. 112, 970e985.

Imbens, G.M., Wooldridge, J.M., 2009. Recent developments in the econometrics ofprogram evaluation. J. Econ. Lit. 47, 5e86.

Jensen, J.R., 1996. Introductory Digital Image Processing: a Remote SensingPerspective. Prentice-Hall Inc., Englewood Cliffs, NJ, USA.

Méndez, V.E., Bacon, C.M., Olson, M., Morris, K.S., Shattuck, A., 2010. Agro-biodiversity and shade coffee smallholder livelihoods: a review and synthesis often years of research in Central America. Prof. Geogr. 62, 357e376.

Mas, A.H., Dietsch, T.V., 2004. Linking shade coffee certification to biodiversityconservation: butterflies and birds in Chiapas, Mexico. Ecol. Appl. 14, 642e654.

Moguel, P., Toledo, V.M., 1999. Biodiversity conservation in traditional coffee sys-tems of Mexico. Conserv. Biol. 13, 11e21.

Nestel, D., Dickschen, F., 1990. The foraging kinetics of ground ant communities indifferent Mexican coffee agroecosystems. Oecologia 84, 58e63.

Nestel, D., Dickschen, F., Altieri, M.A., 1993. Diversity patterns of soil macro-Coleoptera in Mexican shaded and unshaded coffee agroecosystems: an indi-cation of habitat perturbation. Biodivers. Conserv. 2, 70e78.

Nir, M., 1988. The survivors: orchids on a Puerto Rican coffee finca. Amer. OrchidSoc. Bull. 57.

Owubah, C.E., Le Master, D.C., Bowker, J.M., Lee, J.G., 2001. Forest tenure systems andsustainable forestmanagement: the case ofGhana. For. Ecol.Manag.149, 253e264.

Perfecto, I., Mas, A., Dietsch, T., Vandermeer, J., 2003. Conservation of biodiversity incoffee agroecosystems: a tri-taxa comparison in southern Mexico. Biodivers.Conserv. 12, 1239e1252.

Perfecto, I., Rice, R.A., Greenberg, R., Van der Voort, M.E., 1996. Shade coffee: adisappearing refuge for biodiversity. BioScience 46, 598e608.

Perfecto, I., Snelling, R., 1995. Biodiversity and the transformation of a tropicalagroecosystem: ants in coffee plantations. Ecol. Appl. 5, 1084e1097.

Perfecto, I., Vandermeer, J., Mas, A., Pinto, L.S., 2005. Biodiversity, yield, and shadecoffee certification. Ecolog. Econ. 54, 435e446.

Philpott, S.M., Bichier, P., Rice, R., Greenberg, R., 2007. Field-testing ecological andeconomic benefits of coffee certification programs. Conserv. Biol. 21, 975e985.

Philpott, S.M., Dietsch, T., 2003. Coffee and conservation: a global context and thevalue of farmer involvement. Conserv. Biol. 17, 1844e1846.

Ponte, S., 2004. Standards and Sustainability in the Coffee Sector: a Global ValueChain Approach. International Institute for Sustainable Development, Winni-peg, Manitoba, Canada.

Rappole, J.H., King, D.I., Rivera, J.H., 2003a. Coffee and conservation III: reply toPhilpott and Dietsch. Conserv. Biol. 17, 1847e1849.

Rappole, J.H., King, D.I., Rivera, J.H.V., 2003b. Coffee and conservation. Conserv. Biol.17, 334e336.

Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score inobservational studies for causal effects. Biometrika 70, 41e55.

Ruben, R., Fort, R., 2012. The impact of fair trade certification for coffee farmers inPeru. World Devel. 40, 570e582.

Smith, J.A., Todd, P.E., 2005. Does matching overcome LaLonde’s critique ofnonexperimental estimators? J. Econometrics 125, 305e353.

Southworth, J., Munroe, D., Nagendra, H., 2004. Land cover change and landscapefragmentationdcomparing the utility of continuous and discrete analyses for awestern Honduras region. Agric. Ecosyst. Environ. 101, 185e205.

Southworth, J., Tucker, C., 2001. Forest cover change in western Honduras: the roleof socio-economic and biophysical factors, local institutions, and land tenure.Mt. Res. Dev. 21, 276e283.

Staver, C., Guharay, F., Monterroso, D., Muschler, R., 2001. Designing pest-suppressive multistrata perennial crop systems: shade-grown coffee in Cen-tral America. Agroforest. Syst. 53, 151e170.

Takahashi, R., Todo, Y., 2012. Impact of community-based forest management onforest protection: evidence from an aid-funded project in Ethiopia. Environ.Manage. 50, 396e404.

Taylor, P.L., 2005. A fair trade approach to community forest certification? Aframework for discussion. J. Rural Stud. 21, 433e447.

Tilman, D., Fargione, J., Wolff, B., D’Antonio, C., Dobson, A., Howarth, R., Schindler, D.,Schlesinger,W.H., Simberloff, D., Swackhamer, D., 2001. Forecasting agriculturallydriven global environmental change. Science 292, 281e284.

Todo, Y., Takahashi, R., 2011. Impact of farmer field schools on agricultural incomeand skills: evidence from an aid-funded project in rural Ethiopia. J. Int. Devel.25, 362e381.

Toledo, V.M., Moguel, P., 2012. Coffee and sustainability: the multiple values oftraditional shaded coffee. J. Sustain. Agr. 36, 353e377.

Tucker, C.J., Goff, T., Townshend, J., 1985. African land-cover classification usingsatellite data. Science 227, 369e375.

Tucker, C.J., Slayback, D.A., Pinzon, J.E., Los, S.O., Myneni, R.B., Taylor, M.G., 2001.Higher northern latitude normalized difference vegetation index and growingseason trends from 1982 to 1999. Int. J. Biometeorol. 45, 184e190.

Tucker, C.M., 1999. Private versus common property forests: forest conditions andtenure in a Honduran community. Hum. Ecol. 27, 201e230.

White, M.A., Nemani, R.R., 2006. Real-time monitoring and short-term forecastingof land surface phenology. Remote Sens. Environ. 104, 43e49.

Wunderle Jr., J.M., Latta, S.C., 1996. Avian abundance in sun and shade coffeeplantations and remnant pine forest in the Cordillera Central, Dominican Re-public. Ornitol. Neotrop. 7, 19e34.