forest dependence and participation in cpr management: empirical evidence from forest co-management...
TRANSCRIPT
E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
ava i l ab l e a t www.sc i enced i rec t . com
www.e l sev i e r. com/ l oca te /eco l econ
ANALYSIS
Forest dependence and participation in CPR management:Empirical evidence from forest co-management in Malawi
Charles B.L. Jumbea,b,⁎, Arild Angelsenb,c,1
aCentre for Agricultural Research and Development, Agricultural Policy Research Unit, University of Malawi, Bunda College,P.O. Box 219, Lilongwe, MalawibDepartment of Economics and Resource Management, Norwegian University of Life Sciences, PB 5003, N-1432, Aas, NorwaycCenter for International Forestry Research (CIFOR), Bogor, Indonesia
A R T I C L E I N F O
⁎ Corresponding author. Tel.: +265 1277 433; fE-mail address: [email protected]
1 Tel.: +41965700.
0921-8009/$ - see front matter © 2006 Elsevidoi:10.1016/j.ecolecon.2006.08.008
A B S T R A C T
Article history:Received 30 November 2005Accepted 11 August 2006Available online 9 October 2006
We develop an endogenous sample selection model to investigate how forest dependenceinfluences a household's decision to participate in forest co-management program. Usingdata from Chimaliro and Liwonde forest reserves in Malawi, we find that where forestsprimarily have a gap filling or safety net role in Chimaliro, high forest dependency induceshigher rates of participation. However, with more commercial forest uses and a moreheterogeneous social context as in Liwonde, high forest dependency reduces the incentivesfor participation. The findings point to the need to design parallel interventions alongsidethe forest co-management program in order to provide supplementary income sources toparticipants and increase the incentives for participation.
© 2006 Elsevier B.V. All rights reserved.
Keywords:Forest co-managementForest dependenceEndogenous sample selectionParticipationMalawi
1. Introduction
The key questions of this paper are: What makes people toparticipate in the forest co-management (FCM) program inMalawi? In particular, how does forest dependence (share offorest income) affect households' participation choice? Pro-viding answers to these questions is vital for assessingthe local response to devolution policies. This would give anindication of the appropriateness of devolution programs bothas a pro-poor and forest conserving strategy, and therebyyielding important insights into the design of future programs.
For many years, policies for managing common poolresources (CPRs) including forests had marginalized the localpeople by denying them access to these resources. Today,many developing countries have pursued policy reforms andimplemented devolution programs that allow for the greater
ax: +265 1277 286 / 364.(C.B.L. Jumbe).
er B.V. All rights reserved
involvement of local communities or user groups inmanagingthese resources (Meinzen-Dick et al., 1999). Although mostreforms have been promulgated by the failure of governmentsto implement effective strategies to curb overexploitation ofthe resources and the fiscal constraints faced by mostgovernments, it is widely argued that devolution of naturalresource management is the most viable option for ecologicaland economic sustainability of the natural resources (Conroyet al., 2002). In the forest sector, with the realization thatsubsistence forest use constitutes an integral part of rurallivelihood system, devolution of forest management is now atthe core of national forestry policies in many countries(Campbell and Luckert, 2002).
This paper assesses the conduciveness of the forestdevolution policies in Malawi by examining the link betweenforest dependence and participation. In theory, the finding of a
.
2 With the exception of land explicitly registered as private land,or gazetted as “government land”, all the remaining land fallingwithin the jurisdiction of a recognized Traditional Authoritygranted to a person or group and used exclusively for the benefitof a specific community is customary land (Malawi Government,2002).
662 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
positive correlation between forest dependence and participationmay indicate that the devolution policies foster cooperationamong rural households in forest conservation to sustain thefuture flow of benefits. Conversely, the finding of a negativecorrelation depicts an institutional failure (Dasgupta, 1996),which indicates that the program imposes costly constraintson forest use such that forest-dependent households stay outof the program. The finding of no association between forestdependence and participation may suggest that factors (e.g.,ideological, moral or ethical) motivate them to participate inthe program (Heyer et al., 2002).
Many studies have been conducted in the past to deter-mine factors or sets of conditions that induce participation orcooperation in CPR management (e.g., Molinas, 1998; Dayton-Johnson, 2000; Varughese and Ostrom, 2001). Although thesestudies have enriched the literature, and helped to shapepolicies for managing common pool resources worldwide,none of the studies has explicitly examined empirically howpeople's dependence on the resources influence their par-ticipation decisions.
In this paper, we first present a theoretical farm householdmodel to assess how households allocate their labor endow-ment to different productive activities including participationin the FCM program to maximize utility. We then develop anempirical endogenous sample selection model of participationas a system of simultaneous equations in which household'sparticipation decision is modeled as a function of forestdependence and forest use, a dummy variable indicating whetheran individual collects forest products from co-managed forestreserves. The specification of our model considers the fact thatquite a few households use co-managed forest reserves(illegally) but do not participate in the scheme.
Using two different household-level data sets from Chi-maliro and Liwonde forest reserves inMalawi, we estimate themodel in three steps to account for the contemporaneouscorrelation of unobserved factors that determine forest use,forest dependence and FCM participation. The first two stepsfollow the standard Heckman (1979) sample-selection correc-tion procedure to correct for selection bias in the estimates ofthe share of forest income or forest dependence (second step) byincluding the inverse Mills' ratios obtained from the first stage(forest use equation). The third step addresses endogeneity offorest dependence in assessing the impact of forest dependenceon participation by including predicted estimates of the shareof forest income from the second stage as one of the predictorsof participation. To compare the robustness of our results, weestimate the first and second stage equations (forest use andforest dependence) also using maximum likelihood.
This article contributes to the debate on whether devolu-tion of forest management is a universal solution to environ-mental degradation in different socioeconomic and ecologicalconditions by using unique data from two distinct sites. A keyelement in both theoretical and empirical models is howparticipation affects access to forest reserves in differentsocioeconomic, cultural and institutional settings. This is thefirst study to apply advanced econometric techniques toinvestigate the determinants of participation in CPR manage-ment from a developing country.
The rest of the paper is organized as follows: we present thebackground to Malawi's forest policies and co-management
program in Section 2. We develop and formalize our theoret-ical framework in Section 3. This is followed by the specifica-tion and estimation of our empirical model, and a descriptionof data used in the analysis in Section 4. The empirical resultsare discussed in Section 5, while some conclusions are given inSection 6.
2. Background
Malawi has a long history of involving local people to managelocal forests dating back to the 1920s. For many years, colonialadministration was preoccupied with controlling the use andconservation of natural resources, including trees and forests.By mid 1920s, most forests had been gazetted as protectedareas (Kayambazinthu, 2000). However, due to conflictsbetween the colonial government and the local communitiesover land, the colonial government established the CommunalForest Scheme managed by the central government (DistrictAdministration). Under the scheme, approximately 2.7millionha of forested area was allocated to communities for their useand management referred to as Village Forest Areas (VFAs)(Kayambazinthu, 2000). These weremanaged by Village ForestCommittees (VFCs) led by village heads. However, the schemeonly lasted one decade when the policy focus of the colonialadministration shifted from community forestry to forestestablishments for the export market.
After independence in 1964, all forest-related matters oncustomary land2 were handled by the local government(District Councils). In 1985, the management responsibilityreverted back to the central government (Forestry Depart-ment). By that time, the authority of village heads to controlthe VFAs was overpowered by the political influence whichdictated the composition and operations of the VFCs. Thenumber of active VFAs dropped from 5108 in 1963 to 1182 in1994 (Kayambazinthu and Locke, 2002).
The participatory-approach to natural resource manage-ment was revived in the 1990s, especially following the 1992United Nations Earth Summit in Rio de Janeiro during whichparticipatory development was accepted as an integral part ofthe development strategy. In 1996, the Malawi Governmentformulated the National Forestry Policy and the New ForestryAct was endorsed by parliament in 1997. The new legislationremoved a number of barriers to people's involvement in theconservation of trees, forests and protected forest areas, andempowered village heads to confiscate forest products ille-gally obtained from natural woodlands (Sakanda, 1996;Malawi Government, 1997).
With the support from the World Bank and UnitedKingdom (DfID), in 1996 the government launched the forestco-management (FCM) program in Chimaliro and Liwondeforest reserves located in the central/Northern and Southernregions of Malawi. The programwas designed to improve rural
3 Fukuyama (2000) defines social capital as an instantiatedinformal norm that promotes cooperation between two or moreindividuals. The norms (e.g., reciprocity, trust, networks, andrespect) constitute social capital which can range from a norm oreciprocity between two friends to wider social norms such ascollective action.
663E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
livelihoods by generating household income, contributing tofood security and providing environmental services whileenhancing the productivity of forests through sustainableforest management and utilization (Meyers et al., 2001).
Prior to the launch of the program, staff from the ForestryResearch Institute of Malawi (FRIM) who had been conductingresearch onwoodland ecology andmanagement in both forestreserves organized a series of consultation meetings tosensitize local leaders and communities on the impendinglaunch of the FCM program since 1994. A number of work-shops, meetings and training on silvicultural aspects of forestmanagement were conducted. In addition, local communitiesmade field trips to other parts of the country wheredeforestation was a serious problem.
These consultations culminated with the launch of theprogram in 1996. Approximately 210 ha and 1172 ha out of160000 ha and 274000 ha of Chimaliro and Liwonde forestreserves were respectively demarcated into three blocks. Thedemarcation process was participatory involving the localpeople, civil society, government and chiefs during whichancestral boundaries separating different clans were traced todetermine the customary boundaries (Jere et al., 1999). InChimaliro, the block sizes were 18, 118 and 74 ha, while inLiwonde they were 416, 288, and 468 ha. Both forest reservescomprise natural ‘miombo’ woodlands dominated by Brachys-tergia, Julbernadia and Isoberlinia species (Ngulube, 1999). Thereare no significant differences in the species composition,stocking densities and size classes across co-managed blocksin Liwonde (Makungwa and Kayambazinthu, 1999). In Chima-liro, species composition across blocks is generally the same,while stocking densities vary considerably due to differencesin soil characteristics (Chanyenga and Kayambazinthu, 1999).
The overall legal framework for the FCM program is guidedby a constitution (Marsland et al., 1999). The constitutionstipulates, inter alia, the rights and obligations of the commit-tees and government, conditions on the sharing of revenuebetween government and the community, and the types offorest products that can be legally collected from the forestreserves. The role of government is mainly to provide guid-ance, counseling and training of the local communities inforestmatters. Theprogramdoesnot provide long-termsecurerights to forests and their products and the co-managementstructuresdonothave the legalmandate to prosecute violatorsof forest regulations (Kayambazinthu, 2000).
The FCM program activities are implemented at the blocklevel. Within each block, a forest management committee(VFC) with representatives from the designated villagesprovides leadership in drawing up its own local bylaws andblock management plans. The FCM activities include bound-ary marking, firebreak maintenance, controlled early burning,fire fighting, and supervised harvesting. In general, theoperations of the program differ from block to block andbetween the two reserves due to the differences in theleadership and degree of tribal cohesion. Most co-manage-ment activities are undertaken during the dry season (July–October) when demand for agricultural labor is low and whenforest reserves become more susceptible to wild fires.
There are no strong restrictions regarding who shouldparticipate in the program. Participation is voluntary as longas the household lives within the designated villages, abide by
the local bylaws and participate in implementing forestmanagement plans besides attending FCM meetings andpatrolling to monitor illegal activities. In return, the schemelegitimizes participants' access and use of forest reserves tocollect various forest products. These include fuelwood,thatch grass, poles, fodder, mushrooms, wild fruits andother non-timber forest products (NTFPs) (Kayambazinthu,2000). These products, and especially fuelwood, are importantin people's daily livelihood. They also help to fill gaps in foodsupplies during the lean period of between November andMarch (rainy season) when most NTFPs especially mushroomand wild fruits become more abundant. Some householdsespecially in Liwonde survive by selling of fuelwood, curioproducts, cane baskets, mushrooms, honey, wild loquat(Uapaca kirkiana) and other fruits.
Institutional studies conducted in the research sites havedescribed the FCM program in Chimaliro as a model of asuccessful devolution program in Africa (e.g., Kayambazinthu,2000; Banda, 2001). This is in contrast to Liwonde where theFCM program has not been effective in halting excessiveexploitation of forest products for commercial purposesleading to a higher utilization pressure (Makungwa andKayambazinthu, 1999; Ngulube, 1999). Compared to Chimaliro,few institutional studies have been conducted in Liwonde.This study uses data from the two locations to understandfactors that influence participation decisions in order to tracesources of the unequal performance of the program.
3. Theoritical framework
The model developed in this section draws upon the economictheoryofagriculturalhouseholdbehavior (e.g., Singhetal., 1986)to analyze the question of how people decide to participate inthe FCM program in Malawi. The starting point is “the basiccost–benefit calculations of a set of users utilizing a resource”(Ostrom, 1999: 4). Each user compares the net benefits fromparticipation with the net benefits of non-participation. We putthis cost-benefit calculation into an agricultural householdmodeling framework,whichallowsus tounderstandbetterhowdifferent household characteristics and context specific factorsinfluence the participation decision, e.g., the degree of forestdependence (share of forest income) and social capital (peerpressure, past experience, tribal homogeneity)3.
Based on the insights from the fieldwork and analysis ofdata, we focus on three types of costs and benefits. First,participation affects access to the forest reserves as discussedbelow. Second, participation requires spending valuable timein the project (meetings, patrolling, etc.). Third, participationyields social benefits in terms of building the household'ssocial capital within the village. Themodel developed is static,i.e., it does not include the possible higher future benefitsparticipants can derive frombettermanagement of the forests.
f
664 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
We make a few analytical simplifications to make the modelmore tractable and enable us to focus on key aspects of theparticipation decision. The model assumes imperfections inthe labor market in that the household may rent out labor, butcannot hire labor4. The markets for the relevant outputs(forest and agricultural products in particular) are assumed tobe functioning perfectly5, which allows us to focus on totalincome and consumption rather than individual goods.
The household maximizes a twice differentiable quasi-concave utility function which depends on total consumption(C) and leisure (LH)6. The household also derives utility from a‘social good’ (S) as a reward for participating in the FCMprogram. Therefore, the household maximizes a utilityfunction of the following form:
Max U ¼ UðC; LH;S;HÞ ð1Þ
where H is a vector of household characteristics that affecthousehold preferences. The household faces the followingtechnological, time and budget constraints:
QF ¼ QFðLF;D;R;EÞ ð2Þ
QG ¼ QGðLG;M;RÞ ð3Þ
D ¼ DðP;H;V;RÞ ð4Þ
S ¼ SðP;H;VÞ; Sð1ÞNSð0Þ ¼ 0 ð5Þ
L ¼ LF þ LG þ LW þ LP þ LH ð6Þ
pFQF þ pGQG þwLw þ E ¼ C ð7Þ
LF; LG; LW ; LH;QF;QG;Cz0 ð8Þ
Eq. (2) is the production function for a (composite) forestcommodity, stating that output depends on family labordevoted to forest product collection LF and forest access (D). Atechnology parameter ℧ and a vector of forest resourcecharacteristics R also affect production, and are assumed to bebeyond household's control. Eq. (3) is a simple productionfunction for agricultural output, which is a function of familylabor (LG) andhousehold's land area (M) (historically given). Thisproduction is also conditioned on an exogenously productiontechnology R.
Eq. (4) is central in themodel, anddescribeshowaccess to theforest reserve (D) is affected by the household's participation inthe program (P). Participation is a binary variable taking thevalue 1 (participation) or 0 (non-participation). Access also de-pendsonhousehold andvillage characteristics (HandV), aswellas resource characteristicsR, e.g., distance to forest reserves andforest restrictions. In this paper, our broad definition of accessgoes beyond its legal definition. It includes how accessible the
4 This is broadly in line with rural households in Malawi wherevery few farmers rent in labor while some work off-farmparticularly during peak season on tobacco estates.5 There are reasonably well functioning markets for the major
agricultural markets in both sites and forest products particularlyin Liwonde.6 One may also think of C as a composite commodity, including
forest, agricultural and market purchased goods, with the priceset to unity.
forest reserves are both in terms of legal rights, but also thedegree of enforcement of regulations including punishment forviolating the rules. For example, non-participants may collectforest products during odd hours (at dawn or night), whileparticipants may collect openly during the day time.
According to the program, a household not participating inthe program should not have access to forest reserves to collectforest products (D(0)=0). However, the large number of non-participants using the forest reserves shows that this is notthe case. We therefore distinguish between two differentsituations. In the first case, the program functions reasonablywell (but not necessarily perfect) in excluding non-participants,thus,D(1)ND(0). In the second case, the program is ineffective inexcluding or restricting non-participants from using the forestreserves, while at the same time, restraining participants in thetermsof permissibleuses (e.g., frequency of collection).Where itis difficult for non-participants to violate the rules, participa-tion in the program can limit access, i.e., D(1)≤D(0).
Eq. (5) gives the “social good” as a function of participationin the program, or the “production of social capital fromparticipation”. We have normalized the non-participation tozero. A number of household and village variables affectparticipation, which in return yields social capital (respect,trust, solidarity).
Eq. (6) gives the household's total labor endowment or time(L), which is allocated to forest production (collection) (LF),agriculture (LG), off-farm wage labor (Lw), time spent on co-management activities (meetings, patrolling, etc.) (LP), andleisure (LH). LP is zero (0) if the household does not participatein the program, while it is fixed to L̄P when the householdparticipates, i.e., LP(P=1)= L̄PNL P(P=0)=0.
The LHS of Eq. (7) is total household income, whichincludes the value of produced forest and agriculturalcommodities (QF and QG), valued at market prices (pF andpG). The household earns wage income (wLw) if it participatesin the labor market. We also include exogenous income suchas remittances (E). Eq. (7) states that household consumption(C) cannot exceed total income. Eq. (8) represents the non-negativity constraints.
The choice variables are LF, LG, Lw, LH, QF, QG, C and P. SinceP is a discrete variable, the optimization strategy is first tooptimize labor allocation, for given P. We then compare theutility outcomes of the two values of P, and choose the Pwhich maximizes utility. We open up for corner solutions forboth forest production (LF=0) and off-farm labor (Lw=0), in linewith what we find in our data. Leaving out Eqs. (4) and (5), theLagrangian for this Kuhn–Tucker problem is given by:
S ¼ UðC; LH;S;HÞ þ k1½QF−QFðLF;D;R;EÞ�þ k2½QG−QGðLG;M;RÞ� þ k3½L−LF−LG−LW−LP−LH�þ k4½pFQF þ pGQG þwLw þ E−C� ð9Þ
The first order conditions can be summarized as follows(together with Eqs. (2), (3), (6), (7)):
pFQFLFV pGQG
LG ¼ ULHUC
¼ k3k4
zw ð10Þ
If the household is engaged in forest production, the firstinequality sign is replaced with an equality sign. Similarly,
participation (selling labor) in the labormarketmeans that the
665E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
second inequality sign is replacedwith an equality sign.Whenthe household participates in both activities, Eq. (10) is afamiliar optimality condition stating that the value ofmarginal labor productivity in agriculture and forestry shouldbe equal to the market wage rate, which again is equal to themarginal rate of substitution between leisure and consump-tion. In the case of no labor market participation, the house-hold's shadow wage rate is given by w=λ3 /λ4. The marketwage is below ω, and the household prefers working inagriculture, leisure and possibly also forestry.7
This paper focuses on the participation decision, and forthis problem we write the model in a semi-structural form(‘almost reduced form’, as P is an endogenous variable):
U ¼ U⁎ðP; pF;pG;w;E;E;R; L̄P;H;V;RÞ; P ¼ 0;1 ð11Þ
We further define the net gain from participation, B, as:
B ¼ U⁎ð1Þ−U⁎ð0Þ ¼ BðpF; pA;w;E;E;R; L̄p;H;V;RÞ ð12Þ
A household will participate in the program if thedifference in utility between participation and non-participa-tion (B) is non-negative, i.e.:
P ¼ 1 iff Bz0P ¼ 0 iff Bb0 ð13Þ
In this model, participation affects utility in three ways.First, participation influences access. In the first case, whenparticipation improves access to the forest reserve, D(1)ND(0),several factors will influence the value of better access. Higherprices of forest products will increase the benefits from betteraccess. Limited access to off-farm employment opportunitiesexpressed in terms of a low wage rate (w), has the same effect.In the case where the household is not participating in thelabor market (ωNw), factors such as small landholdings (M),low agricultural prices (pG) and poor technologies ðRÞ willincrease the value of B. In general, we can expect householdswith high dependence on forest products to be more inclinedto participate in forest co-management program (Baland andPlatteau, 1996, p 273). In the case where D(1)bD(0), all theseconclusions are reversed.
Second, there is a fixed labor cost by participating in theprogram, L̄P. Obviously, the higher this labor requirement,the lower is B, ceteris paribus.8 For households participating inthe labor market, the opportunity costs of time is given by themarket wage rate (w), and the participation cost increases asthe wage rate increases. For households not participating inthe labor market, we can expect poor households to have alower shadow wage rate, and therefore be more likely toparticipate, ceteris paribus.
7 An alternative assumption would be that the householdsparticipate in the labor market, but are quantity constrained, i.e.,they work L̄w and earn wL̄w. In this case the logic of the modelwould be as when the household does not participate in the labormarket, i.e., the relevant wage rate is the shadow wage rate of thehousehold (ω) rather than the market wage rate (w), cf. Angelsen(1999).8 In a more elaborate model both LP may be endogenous, and
also be an element of the S() function. Aggregate co-managementlabor inputs may also affect A(), e.g., more labor for policing canreduce access of non-participants.
Third, participation produces a social good or social capitalin forms of prestige, status and respect depending on a set ofcultural norms or values by which society or village commu-nity uses to reward, judge, approve or disapprove its citizens.In rural Malawi, participation in village affairs is crucial for thesocial acceptability of individuals and material or moralsupport during, for example, sickness, funerals, weddingsand rituals. Here, cultural norms act as standard for shapingthebehavior andactionsof villagemembers (Heyer et al., 2002).
In a homogeneous society with strong social norms andvalues, participants enjoy social benefits in the form of trust,respect, social acceptability and reputation which are ele-ments of social capital. Among the rich and elite, these benefitscan provide a strong incentive for participation in the program.Among the poor, povertymay also compel them to participatein the program to gain access to forest outputs from the forestreserves, but also for fear of being denied access to otherbenefits outside the FCM program such as benefiting fromsafety net programs,9 general vital information and othersocial benefits. Where communities are more heterogeneousor highly market integrated, the social benefits from partici-pating in the program are likely to be smaller. As such, socialpressure may have little or no influence on inducing greaterparticipation in the program.
4. Emperical model
From our theoretical framework, the decision to participate inthe FCM program depends, inter alia, on whether participatingin the scheme will facilitate access to the forest reserves, andthe importance given to this hinges on the households forestdependence.We take into account that some households haveaccess to forest outputs from co-managed forests withoutparticipating in the program. The key model is the probitparticipation equation which is a function of, inter alia, forestdependence. However, forest dependence is endogenous andis therefore estimated first. Since not all households use forestreserves, there is a (potential) selection bias.We correct this byusing two procedures, the Heckman two-step (Heckit) andmaximum likelihood estimation procedures.
Our model is thus specified as system of simultaneousequations to account for the interrelationships among forestuse, forest dependence and FCM participation as follows:
Ai ¼ Zigþ ei ð forest useÞ ð14Þ
yi ¼ xibþ ui ð forest dependenceÞ ð15Þ
Pi ¼ Wi1þ wyi þ ei ðparticipationÞ ð16Þ
where, Ai is forest use (or forest access) which is a dummyvariable indicatingwhether an individual derive income or notfrom co-managed forest reserves; yi denotes forest dependencedefined as the ratio of forest income (from the forest reserve)to the total household income; Pi is a dummy variable forparticipation; i=1,…, N denotes households; Zi,xi and Wi arevectors of exogenous variables that determine forest use, forest
9 In Malawi, traditional chiefs play a very important role indetermining who should be a beneficiary of such programs.
10 The membership status was crosschecked with the list ofparticipants within each village that was obtained throughParticipatory Rural Appraisals (PRA), and in some cases, suppliedby the committee.
666 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
dependence and participation, respectively; γ, β, ς and ψ areunknown parameters and ui, εi, and ei are error terms. Sinceour aim of this study is to examine the link between forestdependence and participation, we focus on the coefficient ψ inEq. (16) although many variables in the vector of thecoefficients ς are also important to gain insights into otherdeterminants of participation.
For our model, we consider forest income exclusively fromthe co-managed forest reserves. Consequently, yi is observed fora household i together with covariates xi and Zi if Ai=1. Wemake following distributional assumptions about the errorsterms ui,εi, and ei:ui∼N(0,σu2), εi∼N(0,1), ei∼N(0,σe2) and E(ui|Zi,εi)=E(ui|εi)=ρ, where, ρ is the correlation between ui and εi. σu2 and σe2
are the respective variances of ui and eiwhile the variance of theerror term in Eq. (14) is normalized to unity.
4.1. Model estimation
4.1.1. The three step estimationWe estimate our model in three systematic steps as follows:The first two steps are the standard Heckman (1979) two-stepsample selection correction procedure. The first step aims atobtaining the inverse Mills' ratios to correct selection bias inthe estimates of the share of forest income ( forest dependence).From Eq. (14) we specify the following reduced form of forestuse model:
Ai ¼ 1 if Zinþ xibþ viz00 otherwise
�ð17Þ
where, ξ=βγ, vi=βui+εi, and vi∼ (0,σv2) The associated log
likelihood function is
logLða;bÞ ¼XAi¼1
log UZinþ xib
rm
� �� �
þXAi¼0
log 1−UZinþ xib
rm
� �� �ð18Þ
where,Φ(.) is the cumulative function of the standard normaldistribution. By the normality assumption, we optimize thislog likelihood function by maximum likelihood to estimateparameters of the model. The dependent variable for forest useequation (Ai) was computed from the information given by arespondent if a household collects forest products from co-managed forest reserves coded as one (1) and zero (0) for the“yes” and “no” responses.
The second step aims at obtaining the predicted estimates ofthe share of forest income (forest dependence) corrected forsample selection bias. According to Maddala (1983), applyingtheordinary least squares (OLS) to Eq. (15)produces inconsistentestimates of the share of forest income since the expected valueof the error term conditional on forest use is non-zero. Thus, theconditional mean of the share of forest income in Eq. (15) is:
EðyijAi ¼ 1Þ ¼ xibþ EðuijZi; eiÞ ¼ xibþ EðuijeiÞ ð19Þsuch that E(ui|εi)≠0. The conditional expectation of the errorterms ui and εi is:
EðuijeiÞ ¼ EðuijeiV ZigÞ ¼ Eðru;qjeiÞ ¼ qru/ðZigÞUðZigÞ
; ð20Þ
where,ϕ(.) and Φ(.) are the standard normal density andcumulative distribution functions, respectively. We define
λi=ϕ(.) /Φ(.) as the inverse Mills' ratios, which is the covariancebetween residuals of the selection ( forest use) and the outcome(forest dependence) equations estimated from Eq. (18). ReplacingE(ui|εi) by the inverse Mills' ratios λi as a sample selection-biascorrection term in Eq. (15), we re-specify the forest dependenceequation as:
yi ¼ xibþ hki þ gi: ð21Þ
Whereηi is error termthat isassumed tohave theconditionalmeanzero (0) andvarianceση2,whileθ is anunknownparameter.The statistical significance of the coefficient for the inverseMills' ratio θ gives evidence of sample selection bias.
The dependent variable for Eq. (21) (forest dependence) wascomputed as the ratio of forest income to the total householdincome. Forest income includes cash income from sales ofdifferent products from the forest reserves including value ofdomestic forest use and income associated with participatingin FCM program. Total household income is calculated frominformation given by a respondent on cash and subsistenceincome from agriculture, fisheries, forests and livestock; laborincome from off-farm activities such as cottage businessesand employment; value of non-cash gifts, cash gifts andremittances received 12 months prior to the survey.
The third step addresses the problem of endogeneity inestimating the impact of forest dependence on participation(Eq. (16)). From Eq. (21), we derive the predicted estimates ofshare of forest income, we denote wyi. We then re-specify ourparticipation equation (Eq. 16) with predicted estimates of theshare of forest income included as one of the explanatoryvariables as:
Pi ¼1 if Wi1þ dwyi þ jiz00 otherwise
�ð22Þ
where κi∼N(0,σκ2), δ is our parameter of interest. We estimate
the model using maximum likelihood by optimizing thefollowing log likelihood function:
logLðs; dÞ ¼XTi¼1
log UWisþ dwyi
rj
� �� �
þXTi¼0
log 1−UWisþ dwyi
rj
� �� �ð23Þ
The dependent variable for participation in Eq. (23) isconstructed from the responses given by a respondent if anymember of the household participates in the program or not.10
We treat this as a discrete variable with a value of one (1) for“yes” responses and zero (0) for “no” responses.
4.1.2. Maximum likelihood estimationAlthough theHeckman (1979) two-stepmethod used to correctsample selection bias (the first two stages) has been widelyapplied in various studies (e.g., Wales and Woodland, 1980;Fernandez et al., 2001; Nawata, 2004), this estimation tech-nique is not efficient (Greene, 2000; Wooldridge, 2002). We
Table 1 – Comparison of characteristics betweenparticipants and non-participants
Chimaliro Liwonde
P NP P NP
Age of household head(years)
46.854 45.388 39.585 42.871⁎
Education (primary andabove=1)
0.921 0.888 0.667 0.783⁎⁎
Household type(female=1)
0.348⁎⁎⁎ 0.164 0.312⁎⁎⁎ 0.160
Household size 5.528 5.638 4.957 5.170Sex ratio (adult female/male ratio)
1.284⁎ 1.116 1.256⁎⁎ 1.047
Duration of foodinsecurity (months)
5.191 5.052 7.086 7.151
Share of forest income 0.018 0.021 0.136 0.312⁎⁎⁎Woodlot ownership(yes own=1)
0.708⁎⁎⁎ 0.483 0.269 0.226
Land holding size (ha) 5.676 5.613 2.490 2.212Livestock ownership(yes own=1)
0.382 0.448 0.311 0.283
Migration status(non-migrant=1)
0.526⁎⁎ 0.393 0.366 0.340
Duration of residence(years)
36.466⁎⁎ 30.652 27.151⁎⁎ 21.858
Tribal cohesion(belong to main tribe=1)
0.809 0.793 0.408⁎⁎ 0.255
Past group experience(yes=1)
0.618⁎⁎⁎ 0.026 0.107 0.094
Distance to forestproduct market (km)
6.903 10.178⁎⁎⁎ 4.242 5.246⁎
Distance to forestreserve (km)
1.168 1.210 0.361 0.545⁎⁎
Total annualincome (MK)
24443.89 31440.66⁎ 33610.32 27193.80
Total annual forestincome (MK)
449.45 962.11⁎⁎ 5313.63 9784.89⁎⁎⁎
Share of forest income 0.018 0.021 0.136 0.312⁎⁎⁎Forest business(participate=1)
0.562 0.491 0.797 0.887⁎⁎
Sub sample 89 116 93 106Total observations 205 199
⁎P=0.10, ⁎⁎P=0.05, ⁎⁎⁎P=0.001.Stars indicate that the means are statistically different betweenparticipants (P) and non-participants (NP).
667E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
therefore simultaneously estimate Eqs. (14) and (15) bymaximum likelihood to compare the results with those fromthe Heckmanmethod. Combining Eqs. (14) and (15), we derivethe following expression:
Ai ¼ Zigþ qððyi−xibÞ=ruÞ þ xi; xifNð0;1−q2Þ ð24Þ
where ωi=εi+βui. Since, yi, xi and Zi, are observed when Ai=1,it follows that
PrðAi ¼ 1jyi; xiÞ ¼ UZigþ qðyi−xibÞ=ruÞ
ð1−q2Þ1=2 !
; ð25Þ
and its corresponding log likelihood function is:
log Lðg; bÞ ¼XAi¼1
log UZigþ qððyi−xibÞ=ruÞ
ð1−q2Þ1=2 ! !
þXAi¼1
log /yi þ xib
ru
� �� �þXAi¼0
logðUð−ZgÞÞ ð26Þ
By the normality assumption, we estimate the model byoptimizing the log likelihood function directly by iterationalgorithm of a general non-linear optimization program(Greene, 2000). The impact of forest dependence on participa-tion is estimated by using the specification of Eq. (23) exceptthat the predicted estimates of the share of forest income(forest dependence variable) wyi are derived from Eq. (26). We useWhite-heteroscedasticity consistent estimator to obtain ro-bust standard errors to account for heteroscedasticity.
4.2. Data and variables
The main source of data in this study is the household surveyconducted in31villagesadjacent toChimaliroandLiwonde forestreserves between June andDecember in 2002. Prior to the survey,meetings, focus group discussions and key informant interviewswere held with leaders of the FCM committees, chiefs, govern-ment officials, local non-governmental organizations and inter-est groups (e.g., associations, forest traders and craftsmen). Theaims of these meetings were to get a general overview of theconduct, governance and performance of the program andcompile a list of participating villages from which we randomlyselected representative villages. We then held communitymeet-ings in each of the selected villages to learn about the operationsof the program at the village level, and compiled household liststo select randomly households for the survey.
A total of 404 households comprising participants and non-participants were selected for the survey: 205 households from20 villages in Chimaliro and 199 households from 11 villages inLiwonde. The survey questionnaire contained general ques-tions on demographic and socio-economic characteristics ofthe households, income, assets, and specific aspects of theprogram. Summary statistics for variables used in the analysisare in Appendix A.
5. Results and discussion
5.1. Descriptive statistics
Before we discuss the regression results, Table 1 describesfeatures that distinguish program participants and non-
participants in each reserve. The table shows that participantsin both Liwonde and Chimaliro are relatively more likely to befemale-headed households, have higher sex (female/male)ratio, and have lived longer in the same village. In addition,households in Chimaliro are older, more educated (90%), andhave more assets (land, livestock, and woodlots). Hence, theyseem to be relatively less exposed to the risk of food insecurity.Households in Chimaliro run out of own produced food(maize) for approximately 4 months before the next harvestcompared to 7 months in Liwonde. Nearly 28% of the house-holds in Chimaliro have previous group experience, comparedto 10% in Liwonde.
The table further shows that households in Chimaliro areless ethnically differentiated than in Liwonde. More than 80%of the households in Chimaliro belong to the main tribe(Tumbuka) compared to only 33% in Liwonde belonging to the
668 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
main tribe (Yao). We also note that the share of forest incomeis higher in Liwonde (23%) than in Chimaliro (2%). InChimaliro, participants are characterized by a higher propor-tion of households who possess private woodlots, beingpermanent residents and having past group experience. InLiwonde, a larger percentage of the participants belong to themain tribe. The non-participants in Liwonde are relativelyolder, more educated, have higher share of forest income,participate in forest businesses, and stay farther from theforest reserves and forest markets.
5.2. Determinants of participation
As discussed earlier, the first two stages of our estimationprocedureare onlynecessary toobtain selectionbias-correctedestimates of the predicted share of forest income to explorethe impact of forest dependence on participation. We thereforeskip the discussion of the results from the first two stagesand present results for the determinants on participation,except to note that we find evidence of sample selection inboth sites.
In Table 2, we present probit results of the determinants ofparticipation. The first column presents results without the
Table 2 – Determinants of participation in the FCM program in
Variables Chimaliro
No incomeshare
Heckit model
Individual variables Coef. S. E. Coef. S. E.
Age of house head 0.017 0.016 0.001 0.019 0Education (1=primary and above) 0.480 0.466 0.203⁎ 0.196 0Household type 0.518 0.387 1.157⁎⁎ 0.455 1Household size 0.127⁎ 0.070 0.129⁎ 0.076 0Sex ratio 0.472⁎⁎ 0.228 0.300 0.273 0Duration of food insecurity 0.168⁎⁎⁎ 0.046 0.144⁎⁎⁎ 0.068 0Woodlot ownership 0.655⁎ 0.388 0.149⁎⁎⁎ 0.049 0Land holding size 0.071⁎⁎ 0.029 0.237⁎⁎⁎ 0.057 0Livestock ownership 0.010 0.365 1.210⁎⁎ 0.604 1Migration status 0.485 0.507 0.921 0.579 0Group pressure 0.421⁎⁎⁎ 0.716 0.930⁎⁎⁎ 0.214 0Tribal cohesion 0.321 0.437 0.911⁎⁎ 0.452 0Years of residence 0.031⁎ 0.017 0.046⁎⁎ 0.019 0Past group experience 0.326⁎⁎⁎ 0.051 0.464⁎⁎⁎ 0.082 0Distance to forest reserve −0.058 0.106 0.061 0.110 0Distance to forest markets −0.021 0.050 −0.072 0.063 −Firewood price −0.009⁎⁎ 0.004 −0.017⁎⁎ 0.007 −Forest business −0.114 0.370 −0.001 0.386 −Block 1 dummy 0.154⁎⁎⁎ 0.048 0.158⁎⁎⁎ 0.055 0Block 2 dummy 0.834⁎ 0.456 0.893 0.544 1Forest dependence – – 0.193⁎⁎⁎ 0.047 0Interaction variable1 – – −0.187⁎⁎⁎ 0.057 −Constant −0.524⁎⁎⁎ 0.153 −0.830⁎⁎⁎ 0.243 −No. of observations 205 205Wald chi2(22) 72.23 63.40ProbNchi2 0.000 0.000Pseudo R2 0.711 0.756Log pseudo likelihood −40.521 −34.239
⁎P=0.10, ⁎⁎P=0.05, ⁎⁎⁎P=0.001.Coef. = Coefficient; S.E. = robust standard errors.1Interaction variable = group pressure ⁎ predicted forest income share.
forest income share (forest dependence) variable. The other twocolumns present results that include forest dependence variableestimated from the Heckman method (Heckit model) andmaximum likelihood (ML model). In both the Heckit and MLmodels, we also include a multiplicative interaction termbetween forest dependence and group pressure to investigatehow participation among forest dependent householdschanges with group pressure or alternatively how group pressureinfluences participation as people become more forestdependent.
We present estimation results from both the three-stepprocedure and the maximum likelihood for comparisonpurposes. More variables are statistically significant in themaximum likelihood (ML models) than the three-step proce-dure. Hence the discussion of the implications of our empiricalresults will focus on ML results. The validity of all models isconfirmed by the significance of the Wald Chi-square statis-tics indicating that the control variables in each model arejointly significant.
5.2.1. Forest dependence and participationOur primary focus in this study is to explore how forestdependence influences households' participation decisions.
Chimaliro and Liwonde
Liwonde
ML model No incomeshare
Heckit model ML model
Coef. S. E. Coef. S. E. Coef. S. E. Coef. S. E.
.010 0.018 −0.019⁎⁎ 0.010 −0.019⁎⁎ 0.010 −0.018⁎ 0.010
.244⁎ 0.135 −0.089 0.257 −0.053 0.269 −0.121⁎ 0.071
.665⁎⁎⁎ 0.619 0.289 0.254 0.209 0.264 0.244 0.249
.219⁎⁎ 0.091 −0.094⁎ 0.049 −0.094⁎ 0.049 −0.180⁎⁎⁎ 0.049
.018⁎ 0.010 0.233⁎ 0.133 0.222⁎ 0.133 0.186 0.135
.212⁎⁎⁎ 0.085 −0.005 0.037 0.012 0.041 0.029 0.040
.170⁎⁎⁎ 0.580 −0.184 0.271 −0.207 0.268 −0.035 0.277
.197⁎⁎⁎ 0.056 0.087 0.074 −0.059 0.079 −0.035 0.091
.529⁎⁎ 0.754 −0.191 0.254 −0.124 0.274 −0.021 0.275
.573 0.555 0.641 0.428 0.549⁎⁎ 0.240 0.660⁎ 0.340
.749⁎⁎⁎ 0.181 0.269 0.290 0.253 0.214 0.285 0.192
.636⁎⁎ 0.324 0.065 0.244 0.013 0.247 0.187 0.253
.050⁎⁎ 0.020 0.029⁎⁎⁎ 0.010 0.029⁎⁎⁎ 0.011 0.028⁎⁎⁎ 0.011
.386⁎⁎⁎ 0.062 −0.142 0.364 −0.063 0.388 −0.039 0.375
.042 0.114 −0.403⁎⁎ 0.184 −0.355⁎ 0.190 −0.355⁎ 0.1830.111 0.070 −0.148⁎⁎⁎ 0.036 −0.091⁎ 0.046 −0.103⁎ 0.0770.027⁎⁎⁎ 0.009 −0.004⁎⁎ 0.002 −0.004⁎⁎ 0.002 −0.006⁎⁎⁎ 0.0020.316 0.413 −0.145 0.290 −0.131 0.289 −0.110 0.291.196⁎⁎⁎ 0.056 0.131⁎⁎⁎ 0.041 0.580⁎ 0.302 0.130⁎ 0.079.242⁎⁎ 0.508 0.103⁎⁎⁎ 0.034 0.501 0.627 0.121⁎ 0.061.141⁎⁎⁎ 0.045 – – −0.196⁎⁎⁎ 0.051 −0.374⁎⁎ 0.1830.094⁎⁎ 0.038 – – 0.405 0.666 0.214 0.5660.722⁎⁎⁎ 0.209 −0.109 0.075 −0.181 1.133 0.547 1.247
205 199 199 19969.95 67.78 71.11 75.640.000 0.000 0.000 0.0000.742 0.245 0.250 0.267
−36.184 −106.849 −103.194 −100.840
669E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
Table 2 shows that forest dependence has a contrasting effect onparticipation between the two locations. The coefficient for for-est dependence is positive and statistically significant in Chima-liro, but negative and significant in Liwonde. This implies thathigh forest dependency is likely to induce participation in forestconservation amonghouseholds inChimaliro,while in Liwondeit reduces the incentives for participation. Thus, the two sitesseem to represent two contrasting cases discussed in the theorysection onhowparticipation affects access to the forest reserve.In the following sections,weelaborate possible explanations forthese contradicting results.
Firstly, unlikeChimalirowhich is located ina remotearea andwhere markets for primary forest products especially fuelwoodare almost non-existent, markets for forest products are wellestablished in Liwonde. This is largely due to scarcity of forestproducts in the Southern Region arising from high populationdensities. Liwonde is located close to urban cities of Blantyre andZomba, where demand for forest products especially firewood ishigh.Most forest products in Liwonde fetch higher prices than inChimaliro (Ngulube, 1999). Our estimates show that firewoodfetches MK199.00 per cubic meter in Liwonde (US$4.43/m3)compared to only MK66.00 per cubic meter (US$1.47/m3) inChimaliro. The profitability of forest products has inducedgreater commercialization of forest products in Liwonde. Largeamounts of various forest products harvested from the forestreserves are sold by the roadside to the traveling public andintermediate traders.
The finding of a strong negative effect of firewood price onparticipation in Liwonde suggests that market integration andincreased value of forest products reduce the incentivesamong households to participate in the program as it mayconstrain their access to the forest reserve. The programrestricts the frequency, type and quantity of forest productsthe participants can harvest from the forest reserve, while itseems very ineffective in excluding and regulate the use bynon-participants.
In addition, 70% of the revenue from commercial sales offorest products from co-managed forests by the FCM commit-tees is remitted to government while the community retainsonly 30% (Kayambazinthu, 2000). Although joint or bulkcommercial selling of forest products from the forest reservesis not common, it may have an impact of scaring awaypotential participants.
Overall, participants make considerable sacrifices to par-ticipate in the program. This is especially true for Liwondewhere the households are more forest-dependent as forestincome accounts for nearly a quarter of their total earnings(23%). As a result, most households who cannot afford the costof restrained forest use in the interest of conservation stay outof the program, and collect forest products from the forestreserve illegally.
Secondly, in Malawi local chiefs are the custodian ofdevolution policies and have the final verdict on mostdecisions made by the FCM committees (Kayambazinthu andLocke, 2002; Shackleton et al., 2002). Traditionally, theTumbukas (main ethnic tribe in Chimaliro) have deep respecttoward those in authority such as chiefs and politicians. Sincemost households in Chimaliro belong to the same tribe, localchiefs and leaders use their influence to foster cooperationamong individuals. The finding of positive impacts of ‘social
capital’ variables, namely group pressure, tribal cohesion andpast group experience in Chimaliro suggests that ‘socialcapital’ is vital for inducing greater participation. Shackletonand Campbell (2001) also attributed the success of the FCMprogram in Chimaliro to the respect people have toward localchiefs. This indicates that the FCM program is likely to besuccessful in isolated areas where ‘social capital’ exists withinthe community, consistent with the discussion in Section 3.
Finally, the coefficient for the interaction term betweenforest dependence and group pressure has also a contrastingeffect on participation between the two sites. In Chimaliro, thecoefficient for the interaction term is negative and statisticallysignificant, while in Liwonde, it is positive but not significant.The negative correlation between participation and theinteraction term in Chimaliro has implications for thesustainability of the FCM program. It implies that as peoplebecome increasingly dependent on forests as their mainsource of income beyond the current ‘safety net’ or ‘gapfilling’ consumption levels, group pressure or ‘social capital’ islikely to be ineffective in sustaining and fostering the existingnorms of cooperation in implementing the program.
In Liwonde, our data show that 65% of the households(N=199) are migrants belonging to different ethnic back-grounds. In addition, more than 80% of these migrants areengaged in forest-based enterprises as their main occupation.The combined effect of tribal differentiation and the prolifer-ation of forest-based enterprises (selling of forest products)due to the impact of market integration weaken the vitalitygroup pressure to influence greater participation in forestconservation or to control overexploitation of forest resourcesfor commercial purposes.
5.2.2. Other determinants of participationComing to other determinants of participation, we see that allparameter estimates for household variables in Chimalirohave the expected positive signs and are statistically signif-icant except for participation in forest business. None of thehousehold variables are significant in Liwonde except forhousehold size. Results indicate that both coefficients for ageand education of household head are negative and statisticallysignificant in Liwonde while in Chimaliro, both coefficientsare positive but only education is statistically significant.These results are consistent with our descriptive data where54% of participants in Liwonde (N=93) compared to 34% inChimaliro (N=89) are below 40 years old.
The finding of a strong positive effect of education onparticipation in Chimaliro supports our field observations thatpeople with formal education – especially retired publicservants and politicians – held key positions in blockcommittees in Chimaliro, but less so in Liwonde. Due totheir understanding of the importance of conserving forests,they are more likely to participate in the program themselvesand, therefore, motivate other villagers to participate as well.A possible explanation of the negative effect of education onparticipation in Liwonde is that wage employment opportu-nities are better than in Chimaliro. As such, educated peopleare more involved in off-farm and off-forestry activities andare, therefore, less interested in forestry issues.
An intriguing finding is that of a strong positive effect offood insecurity on participation in Chimaliro but not in
670 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
Liwonde, considering that the risk of exposure to foodinsecurity is more severe among households in Liwondethan in Chimaliro. Due to strong enforcement of rules inChimaliro, food insecure households participate in theprogram to gain access to the forest reserves. In contrast, theweak enforcement of rules does not compel food insecurehouseholds in Liwonde to participate in the program. This isanother indication of access being enhanced through partic-ipation in Chimaliro, while it has the opposite effect inLiwonde.
As mentioned earlier, forest products especially NTFPshelp to fill gaps in local food supplies during the rainy season.These findings indicate that the livelihood of food-insecurehouseholds, especially among female-headed households inChimaliro would have worsened if households did notparticipate in the program to gain access to forest outputsfrom the forest reserves. These findings are consistent withthe ‘gap filling’ or ‘safety net’ role of forests (Byron and Arnold,1999; Angelsen and Wunder, 2003).
In Liwonde, many people are engaged in the selling offuelwood and curio products by the roadside as they havesmall land holdings and unable to produce enough foodespecially maize. Thus, households are involved in suchbusinesses ‘not out of wish’, but as the only source of incometo support their families. Since participation in forest-relatedbusinesses conflict with the conservation objectives of theFCM program (and due to weak enforcement of rules) mosttraders refrain from participating in the program as implied bythe negative coefficient for the variable, participation in forestbusiness.
The coefficients for all asset variables have the expectednegative signs in Liwonde implying that asset holdingsreduce the incentives among households to participate inthe program. These results contrast those for Chimalirowhere household assets have strong positive effects onparticipation. Results from the first and second stage showa negative correlation between household assets and bothforest use and forest dependence, implying that asset-richhouseholds are less likely to exert pressure on forest reservesin Chimaliro. In the same way, one would have expected lessneed among asset-rich households to participate in thescheme. One possible explanation could be that most asset-rich households are influential and have strong ties with theruling elite. As such, they hold important positions thatcompel them to participate in the program to fulfill theirobligations.
Another explanation could be that asset-rich householdsparticipate in the scheme for social capital such as personalinterest, self-esteem, respect or personal sacrifices other thanthe short-term benefits in the form of better forest access. Thefinding of a positive effect of woodlot ownership on partici-pation in Chimaliro indicates that personal interest in forestsmotivates them to participate in the program. During projectinception, the government distributed free seedlings tointerested households to establish their own woodlots. Thissuggests that the distribution of seedlings has a motivatingeffect on a household's continued interest in forest activities,as further supported by the higher percentage of participants(28%) in Chimaliro who established private woodlots com-pared with Liwonde (10%).
We included binary variables for co-management blocks tocontrol for the differences in the condition of the forestresources and local policy environment (e.g., size of the blocks,species composition and density, access rules, benefit andcost-sharing arrangements and leadership) that conditionparticipation across blocks. The results indicate that the abovelocation-specific factors are important in influencing house-hold participation decisions.
6. Conclusions
Applying an endogenous sample selection model of participa-tion on household-level data from Chimaliro and Liwondeunder pilot co-management program in Malawi, we find thatforest dependence has contrasting effects on participationbetween the two sites. We find that the extent to which highforest dependency leads to more or less participation dependson the relative importance of forests on the local people'slivelihood, the degree ofmarket integration, existence of socialcapital within the local community, and the local economicenvironment.
Evidence for Chimaliro indicates that where forests haveprimarily a gap filling and safety net role, high forest depen-dency induces higher rates of participation, and that social(capital) variables namely past group experience, years ofresidency, tribal cohesion and group pressure are important ininducing greater participation amonghouseholds.However, thefinding of a strong negative effect of the interaction betweenforest dependence and group pressure on participation implies thatincreased forest dependence beyond the current ‘safety net’ or‘gap filling’ level is likely to jeopardize the effectiveness of grouppressure or ‘social capital’ in sustaining the program. As a result,the existing norms of cooperation thatmakes the FCMprogramin Chimaliro to be among the most successful devolutionprograms in Africa (Kayambazinthu, 2000) may dissipate aspeople increasingly become more forest dependent.
In Liwonde, with more commercial uses of forests and amore heterogeneous social context, high forest dependencyreduces the incentives among households to participate in theprogram, depicting an institutional failure (Dasgupta, 1996).The restrictions that the FCM program imposes on participat-ing households as discussed earlier, imply that people makeconsiderable sacrifices to participate in the scheme. Sinceforests are an integral part of rural livelihood in Liwonde,mosthouseholds especially among small collectors and traderscannot afford the cost of restrained forest use in the interest ofconservation. As such, they refrain from participating in thescheme. Besides, social heterogeneity, market integration andthe commercialization of forest products destroy the tradi-tional norms of cooperation in forest conservation such thatfree-riding behavior takes over and dominates.
Evidence presented in this study raises several importantissues that are critical to consider in the design of the FCMschemes. Overall, this study has revealed that the presentinstitutional arrangements for managing forest reserves inMalawi are not a panacea for preventing further degradationof forest resources in different socioeconomic, cultural andinstitutional settings. Our findings call for the need to besensitive to the short-term needs of the local people. In
671E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
particular, the distribution of benefits between the centralgovernment and villagers is one of the critical aspects thatcan enhance the benefits people derive from participation.Another aspect is the importance of linking participation inthe FCM program to other complementary livelihood inter-ventions in order to reduce pressure on the forest reservesespecially among forest-dependent communities. At thesame time, there is also the need to constrain free ridingbehavior by conceding greater autonomy and legal support tothe FCM structures. In this way, the local villagers will beempowered to effectively deal with the free-rider problem.
Since forest co-management inMalawi is a pilot program inits formative years, to gain more insights on how devolutionpolicies contribute to rural poverty reduction, further work isneeded to measure the impact of participating in the programand how the benefits (income) from the FCM program aredistributed among different groups of households. Thisanalysis is vital for designing appropriate interventions tomitigate the negative effects of future devolution programs bytargeting the most vulnerable households.
Acknowledgements
The authors thank the Norwegian Research Council for thefinancial support for a two-month stay at CIFOR in Indonesia.Thanks go to Bruce Campbell and David Kaimowitz (CIFOR),Olvar Bergland, Stein Holden, Baikuntha Aryal, Kassie Menale,Ronnie Babigumira, Mahari Tikabo, Eirik Romstad, Eric NævdalandDadi Kristofersson (UMB), Gustavo Sanchez of UniversidadDel CEMA (Argentina) for useful comments and suggestions.An earlier version of the paper was presented at the EuropeanAssociation for Ecological Economics (Portugal) and the UlvonConference on Environmental Economics (Sweden) in 2005.
Appendix A. Summary statistics for overall means
Variables
Chimaliro LiwondeALL
Std.Err.ALL
Std.Err.Individual characteristics
Age of household head(years)46.02⁎⁎⁎
1.017 41.120 1.054Formal education (yes=1)
0.902⁎⁎⁎ 0.021 0.729 0.032Household characteristics
Household type (female=1) 0.243 0.030 0.231 0.030 Household size 5.590 0.155 5.070 0.160 Gender ratio(adult female ratio)1.189
0.064 1.144 0.058Duration of food insecurity(months)
3.837
0.296 7.121⁎⁎⁎ 0.120Observed share offorest income
0.020
0.003 0.230⁎⁎ 0.022Household assets
Woodlot ownership(yes own=1)0.580⁎⁎⁎
0.035 0.246 0.031(continued)Appendix A (continued)
Variables
Chimaliro LiwondeALL
Std.Err.ALL
Std.Err.Household assets
Land holding size (ha) 5.641⁎⁎⁎ 0.324 2.342 0.125 Livestock ownership(yes own=1)0.420⁎⁎
0.035 0.296 0.032Migration status(non-migrant=1)
0.462⁎⁎⁎
0.035 0.352 0.034Duration of residence(years)
33.941⁎⁎⁎
1.276 24.331 1.189Group pressure (actual/potential participants)
0.653⁎⁎⁎
0.021 0.204 0.009Tribal cohesion(belong to main tribe=1)
0.800⁎⁎⁎
0.028 0.327 0.033Past group experience(yes=1)
0.283⁎⁎
0.032 0.101 0.021Spatial and market variables
Distance to forestproduct market (km)8.757⁎⁎⁎
0.317 4.711 0.309Firewood price (MK/m3)
65.65 2.042 201.10⁎⁎⁎ 3.762 Distance to forest reserve (km) 1.186⁎⁎⁎ 0.101 0.459 0.040 Forest business(Participate=1)0.522
0.035 0.844⁎⁎⁎ 0.026No. of observations
205 199⁎P=0.10, ⁎⁎P=0.05, ⁎⁎⁎P=0.01.Stars (⁎) indicates that the means are statistically differentbetween reserves.
R E F E R E N C E S
Angelsen, A., 1999. Agricultural expansion and deforestation:modeling the impact of population, market forces andproperty rights. Journal of Development Economics 58,185–218.
Angelsen, A., Wunder, S., 2003. Exploring the Forestry-PovertyLink: Key Concepts, Issues and Research Implications. CIFOROccasional Paper, vol. 40. CIFOR, Bogor.
Baland, J.M., Platteau, J.P., 1996. Halting Degradation of NaturalResources: Is There a Role for Rural Communities. OxfordUniversity Press, Oxford. 440 pp.
Banda, G.K.M., 2001. An evaluation of co-management of forests: acase study of Chimaliro forest reserve in Malawi. MSc. Thesis.University of Aberdeen.
Byron, N., Arnold, M., 1999. What futures for the people of thetropical forests? World Development 27/5, 789–805.
Campbell, B.M., Luckert, M.C. (Eds.), 2002. Uncovering the HiddenHarvest: Valuation Methods for Woodland Forest Resources.Earthscan. 262 pp.
Chanyenga, T.F., Kayambazinthu, D., 1999. Monitoring and evalu-ating productivity inmiombowoodlands under co-managementarrangements: the case of Liwonde forest reserve. In: Ngulube,M.R., Mwabumba, L., Chirwa, P. (Eds.), Community-BasedManagement of Miombo Woodlands: Proceedings of a NationalWorkshop. Forestry Research Institute of Malawi, pp. 131–159.September.
Conroy, C., Misha, A., Rai, A., 2002. Learning from self-initiatedcommunity forest management in Orissa, India. Forest Policyand Economics 4, 227–237.
Dasgupta, P., 1996. An Inquiry into Well-Being and destitution.Clarendon Press. 680 pp.
672 E C O L O G I C A L E C O N O M I C S 6 2 ( 2 0 0 7 ) 6 6 1 – 6 7 2
Dayton-Johnson, J., 2000. Determinants of collective action on thelocal commons: a model with evidence formMexico. Journal ofDevelopment Economics 62, 181–208.
Fernandez, A.I., Rodríguez-Póo, J.M., Stefan, S., 2001. A note on theparametric three-step estimator in structural labor supplymodels. Economic Letters 74, 31–41.
Fukuyama, F., 2000. Social Capital and Civil Society IMF WorkingPaper WP/00/74. April.
Greene, W.H., 2000. Econometric Analysis, fourth edition. PrenticeHall. 1004 pp.
Heckman, J., 1979. Sample selection bias as a specification error.Econometrica 47, 153–162.
Heyer, J., Stewart, F., Thorp, R., 2002. Group Behavior andDevelopment: Is the Market Destroying Cooperation. OxfordUniversity Press. 382 pp.
Jere, P., Valera, K., Voysey, B., 1999. Malawi national forestryprogram working group on co-management of forest goodsand services: synthesis study of initiatives in co-managementof natural resources in Malawi. Final Report.
Kayambazinthu, D., 2000. Empowering communities to managenatural resources: where does power lie? The case of Malawi.In: Shackleton, S.E., Campbell, B. (Editors), Empowering com-munities to manage natural resources: case studies fromsouthern Africa. SADC Wildlife Sector Natural ResourcesManagement Program, Lilongwe, Malawi, CSIR, Pretoria andWWF, Southern Africa, Zimbabwe, pp. 37–58.
Kayambazinthu, D., Locke, C., 2002. Forest Management andDiverse Livelihoods in Malawi. Ladder Working Paper, vol. II4.University of East Anglia. 32 pp.
Maddala, G.S., 1983. Limited dependent and qualitative variablesin econometrics. Econometrics Society Monographs. Cam-bridge University Press. 401 pp.
Makungwa, S., Kayambazinthu, D., 1999. Monitoring and evalu-ating productivity of miombo woodlands under co-manage-ment arrangements: the case of Liwonde forest reserve. FRIM,Community Based Management of Miombo Woodlands inMalawi. Zomba, pp. 115–130.
Malawi Government, 1997. The Forestry Act 1997. Ministry ofForestry, Fisheries and Environmental Affairs.
Malawi Government, 2002. Malawi National Land Policy. Ministryof Lands, Physical Planning and Surveys, Lilongwe.
Marsland, N., Henderson, S., Burn, B., 1999. On-going Evaluation ofFRP Project: “Sustainable Management of Miombo Woodlandsby Local Communities in Malawi” (R6709). Natural ResourcesInstitute and University of Reading.
Meinzen-Dick, R., Knox, A., Gregorio, M.D. (Eds.), 1999. CollectiveAction, Property Rights and Devolution of Natural ResourceManagement: Exchange of Knowledge and Implications forPolicy. Proceedings of the International Conference Held from21–25 June 1999 in Puerto Azul, the Philippines. DSE/ZEL,Germany. 294 pp.
Meyers, J., Ngalande, J., Bird, P., Sibale, B., 2001. Forestry Tactics:LessonsLearned fromMalawi'sNational ForestryProgram.PolicythatWorks for Forests and People, vol. 11. International Institutefor Environment and Development (IIED), London. 84 pp.
Molinas, J., 1998. The impact of inequality, gender, externalassistance and social capital on local-level cooperation. WorldDevelopment 26/3, 413–431.
Nawata, K., 2004. Estimation of the female labor supply models byHeckman's two-step estimator and the maximum likelihoodestimator. Mathematics and Computers in Simulation 64/3-4,385–392.
Ngulube, M.Z., 1999. The utilization of non-timber forest productsfrom the miombo woodlands in Malawi: a case study. FRIM,Community Based Management of Miombo Woodlands inMalawi. Zomba, pp. 44–68.
Ostrom, E., 1999. Self Governance and Forest Resources. CIFOROccasional Paper, vol. 20. CIFOR, Bogor.
Sakanda, G.P., 1996. National report on the forestry policy inMalawi. FAO, Forestry Policies of Selected Countries in Africa:Forestry Paper, vol. 132–341. FAO, Rome.
Shackleton, S., Campbell, B., 2001. Devolution in Natural ResourceManagement: Institutional Arrangements and Shifts—A Syn-thesis of Case Studies for Southern Africa. CIFOR, Boor. 79 pp.
Shackleton, S., Campbell, B., Wollenberg, E., Edmunds, D., 2002.Devolution of community-based natural resource manage-ment: creating space for local people to participate andbenefit? Natural Resource Perspectives 76, 1–6.
Singh, I., Squire, L., Strauss, J., 1986. Agricultural HouseholdModels: Extensions, Applications and Policy. Johns HopkinsUniversity Press, pp. 71–91.
Varughese, G., Ostrom, E., 2001. The contested role of heteroge-neity in collective action. Some evidence from communityforestry in Nepal. World Development 29/5, 747–765.
Wales, T.J., Woodland, A.D., 1980. Sample selectivity and theestimation of labor supply functions. International EconomicReview 21, 437–468.
Wooldridge, J.M., 2002. Econometric Analysis of Cross Section andPanel Data. MIT Press, Cambridge. 752 pp.