ECONOMICS OF PAYMENTS FOR ENVIRONMENTAL SERVICES:
THREE ESSAYS ON KENYA, TANZANIA, AND MOZAMBIQUE
By
Rohit Jindal

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Community, Agriculture, Recreation, and Resource Studies
2011

ABSTRACT
ECONOMICS OF PAYMENTS FOR ENVIRONMENTAL SERVICES:
THREE ESSAYS ON KENYA, TANZANIA, AND MOZAMBIQUE
By
Rohit Jindal
The present study reviews important aspects of payments for environmental services
(PES). In particular, it focuses on three important questions around application of the PES
approach in developing countries: (i) how to assess the feasibility of PES projects on the ground,
(ii) how to estimate payments when markets for environmental services are missing, and (iii)
what kinds of local impacts can we realistically expect after PES projects have been functional in
the field? These questions are answered through field work on actual PES sites in Kenya,
Tanzania, and Mozambique by combining theory from environmental and resource economics,
information economics and microeconomics with econometric methods. Research methods
include field transects, exploratory interviews, focus groups, household survey, and field
experiments in the form of auctions to allocate tree planting contracts. The total sample size of
the study is more than 1,000 households in three different countries. The overall study consists of
three separate papers, each focusing on a specific question. Together, they present a
comprehensive review of PES in developing countries, including its entire project cycle.

ACKNOWLEDGEMENTS

Although I acknowledge individual contributions for each paper separately, I would like
to take this opportunity to express my gratitude to Dr. John Kerr who has been an inspiring and
extremely helpful guide over the last six years. He has been a constant source of motivation and I
have spent endless hours working with him, learning how to be patient and helpful, without any
qualms. I came to Michigan State mainly to work with him and I am glad I did.
I also thank other members of my PhD committee - Dr. Allan Schmid, Dr. Satish Joshi,
Dr. John Staatz, and Dr. Paul Ferraro for their excellent guidance. I did my first graduate course
at Michigan State with Dr. Schmid and to this date, his class on Institutional and Behavioral
Economics remains one of my favorite ones. There is much that I learnt from that course, but,
there is much more that remains to be learnt. Dr. Joshi came to my rescue when I had little hope
of raising funds for the year long fieldwork in Tanzania. Little did I know that his generous grant
would result in matching funds from other sources, which in the end helped me complete my
dissertation work. He has also been very interested in my work and has given helpful suggestions
on a number of occasions. I also thank Dr. Staatz, who introduced me to the fascinating world of
information economics, much of which forms the theoretical background for my second paper.
Dr. Ferraro is one of the most energetic and enthusiastic people I have ever met. He never says
no and has boundless energy to help anyone who comes in his contact. The result is my
embarking on a completely new topic and our successful collaboration in Africa and beyond.
Two other people who have played a very important role in my dissertation work are Dr.
Brent Swallow at University of Alberta and Dr. John Grace at University of Edinburgh. Dr.
Swallow first invited me to work in Kenya and as the collaboration with the World Agroforestry
iii

Centre (ICRAF) blossomed, he gave me generous funding to do action research under ICRAF’s
PES project in Tanzania. Dr. Grace first served as a supervisor for my master’s thesis. He was
instrumental in introducing me to climate change mitigation work and chose me as the lead
researcher to do the impact evaluation of the University of Edinburgh’s carbon forestry project in
Mozambique.
I appreciate and thank my wife Mamta Vardhan for being such a nice companion and for
always being there when needed. Little Kartik has brought happiness and joy to my life that I did
not know I missed. He suffered the hardships of an intense fieldwork without any complains and
kept us smiling all through. Finally, my parents have provided silently and smilingly. Through
the unspoken, I know what we have lost, but perhaps the gains and the hopes will help us all sail
towards a better tomorrow.
Thanks to you all!

iv

PREFACE

Payments for Environmental Services (PES) have become an important area for scholarly
work. Under PES, land stewards receive payments for adopting environment friendly practices
on their farms. After the failure of top down regulatory approaches in many parts of the world,
and the limited impact of both unconditional subsidies for conservation and integrated
conservation development project (ICDP) approaches, researchers, policy makers, and field
practitioners have started to focus their attention on PES as a way to secure and conserve
valuable landscapes. Hundreds of articles have appeared in prestigious journals and several new
projects are announced every month. PES has also become the main approach for climate change
mitigation and adaptation work on forested lands, with international financial commitments of
more than $4 billion to date to help reduce emissions from deforestation and degradation in the
tropics.
Amidst all this attention, however, literature on actual experience with PES on the ground
remains sketchy. While a good number of studies have started to appear on national level PES
programs in countries such as China, Mexico, and Costa Rica, a particular concern in this regard
is the relative absence of scholarly work on field experience with small community-based PES
projects in other developing countries. Important questions remain unanswered. What is the
actual scope of PES on the ground? What payments systems are in place or even how to estimate
payment levels that will result in effective provision of an environmental service? How to
address information asymmetry between service providers or sellers and people who want to buy
these services? Similarly, a lot has been written about the potential of PES approach in reducing
poverty among local communities. Again, detailed empirical studies on this subject remain
scarce.
v

The present study attempts to address some of these research gaps. In particular, it
focuses on three important questions around application of the PES approach in developing
countries: (i) how to assess the feasibility of PES projects on the ground, (ii) how to estimate
payments when markets for environmental services are missing, and (iii) what kinds of local
impacts can we realistically expect after PES projects have been functional in the field?
These questions are answered through field work on actual PES sites in Kenya, Tanzania,
and Mozambique by combining theory from environmental and resource economics, information
economics and microeconomics with econometric methods. Research methods included field
transects, exploratory interviews, focus groups, household survey, and field experiments in the
form of auctions to allocate tree planting contracts. In all, I surveyed more than 1,000 households
through three different surveys that I designed and carried out in three different countries. The
overall study consists of three separate papers, each focusing on a specific question. Together,
the three papers present a comprehensive review of PES in developing countries, including its
entire project cycle.
The first paper, “Exploring demand for forestry in Lake Victoria Basin, Kenya:
an econometric approach,” explores feasibility of the PES approach by estimating the local
demand for a tree planting program amongst rural households in western Kenya. It is based on a
field survey with 277 households, using a stratified random sampling approach. The study
follows an attribute-based method to elicit farmers’ preferences. Local demand for tree planting
is explored in terms of the number of trees that a household would like to plant under different
incentive levels, and the choice of tree species that it makes. The mean willingness to plant new
trees per household increases from 44 trees when farmers have to pay 10ksh/seedling, to 244
trees when farmers receive 10ksh/seedling, while a majority of local households prefer at least

vi

one timber species amongst the trees they propose to plant. The paper uses fixed effects, random
effects and random effects tobit, and random effects logit models to estimate relevant
parameters. In addition, the two demand equations are jointly estimated using the seemingly
unrelated regression technique. A unique contribution of the study is the combination of
conventional literature on determinants of agroforestry adoption in the tropics with PES literature
on potential impacts of conditional payments. The study finds that introduction of an incentive of
KSH 1 per tree seedling results in an increased demand for 18 seedlings per household. It
therefore shows that introduction of economic incentives can create demand for PES activities
among local communities in developing countries such as Kenya. It also identifies additional
socio-economic variables that affect this demand: availability of timber species (positive effect),
gender of the respondent (men likely to plant more trees than women), availability of agricultural
labor at the household (positive), and secure title to the land (positive) have a significant effect
on mean willingness to plant trees. Furthermore, farmers in the Yala River basin are likely to
plant more trees than those in the Nyando River basin with important lessons for the Western
Kenya Ecosystem Integrated Project being implemented in the area.
The second paper, “Estimating ‘payment’ in payments for environmental services:
Results from field auctions in the Uluguru Mountains, Tanzania,” looks at how to estimate
payment level in a PES project that can adequately compensate the opportunity cost of
participating farmers. It uses a set of field auctions to estimate an equilibrium price and allocate
conservation contracts under a PES project in Tanzania. 251 randomly selected local farmers in
the Uluguru Mountains submitted sealed bids for agroforestry contracts requiring them to plant
and protect 80 trees of selected species on a 0.5 acre plot in return for an upfront payment.
Winning bids were decided using the Vickrey Auction uniform pricing format, with the last

vii

rejected bid setting up the equilibrium payment. This ensures incentive compatibility for
participating farmers to reveal their estimated opportunity costs in the form of their bids. Results
show that bids across the two auction rounds were fairly similar. The mean bid for an
agroforestry contract was TSH 143,840 (USD 113.30), while the median bid is TSH 130,000
(USD 102.40). In all, 32 winners received a total of 2,560 tree seedlings to plant on their farms.
There is no evidence of collusion or cooperative bidding. Ordered bids can thus be used to
estimate a landscape level supply curve. Auction bids and analysis of socio-economic data
collected through a household survey show that many poor households are able to participate,
but not all.Regression of bids on household demographics revealed that male-headed households,
households located in either of the two main villages, smaller households with fewer members,
and households with less livestock assets faced higher opportunity costs. The model is also used
to predict budget requirements for alternate targeting of PES contracts in the area based on a mix
of socio-economic and environmental criteria. The paper thus demonstrates the feasibility of
using auctions in developing country setting as one of the ways to estimate payments and
allocate PES contracts. However, in contrast to some previous studies, it shows that alternate
targeting of contracts, though easy to conceptualize, is difficult to operationalize in the field.
The third paper, “Reducing poverty through carbon forestry? Exploring impacts of the
N’hambita Community Carbon Project in Mozambique,” uses a mix of household surveys and
focus groups to analyze environmental and livelihood impacts of the N’hambita carbon project in
Mozambique. The project makes payments for carbon sequestration and avoided deforestation
activities, generating 236,513 tCO2 of carbon offsets. Farmers receive US$433-808/ha over
seven years. The project has achieved an impressive diffusion rate with 852 households
representing 80% of all households in the area participating in the project. Econometric analysis
viii

shows that poor households are able to access the project. In addition, larger households,
residents who have been in the area longer, and households with off-farm income are also more
likely to participate in the project. A unique contribution of this study is the use of the differencein-difference method to differentiate the impact of carbon payments from the impact of other
development activities introduced by the project in the area. Although there has been a lot of
discussion in the environmental literature on potential poverty impacts of PES projects, the paper
shows that on their own, conservation payments are not enough to move households out of
poverty. Comparison between 2001 and 2008 shows that while carbon payments do supplement
incomes, the amount is small. In contrast, households employed in project-related
microenterprises are much better off. Further, permanence of carbon sequestration is an issue as
payments end after seven years.
By focusing on small community based projects in developing countries, the three
papers together add to the current discourse on PES. But they also show that field research is
messy and it is difficult to come up with conclusive answers. For instance, it is difficult to say
whether or not local farmers in Tanzania understood the auction process or its implications
completely, though absence of collusion and presence of normally distributed bids indicate that
they did understand the process to a large extent. At the same time, it is difficult to know to what
extent the auction findings can guide the design of follow-up PES projects. Similarly, measuring
the impacts of carbon payments in Mozambique required the use of several different analytical
techniques to try to distinguish between the effects of carbon payments from those of other
project activities and from wider scale economic progress in the area. The research neither says
that the carbon project is harmful, nor that the project has completely revitalized the local
economy. Instead, it offers more subtle and nuanced answers that help in linking PES theory

ix

with field application. While payments increase the likelihood of conservation, they alone are
unable to move people out of poverty. The study thus points out the limitation of the win-win
strategy that has been repeatedly highlighted in the environmental literature.
Such a lack of unambiguous, clear-cut answers is likely to be the norm in research on
PES, especially in developing countries, due to the complexity of PES and the methodological
challenges associated with the absence of environmental service markets and the confounding
effects of other economic activities.

x

TABLE OF CONTENTS

List of Tables

xii

List of Figures

xiv

Chapter 1: Exploring demand for forestry in Lake Victoria Basin, Kenya:
An Econometric approach
1.1 Introduction
1.2 Choice of Survey Methods
1.3 Brief description of the data
1.4 Econometric Analysis
1.5 Discussion: Significance of results
1.6 Conclusion: Limitation of the study
References

1
1
4
7
8
18
19
31

Chapter 2: Estimating ‘payment’ in payments for environmental services:
Results from field auctions in the Uluguru Mountains, Tanzania
2.1 Introduction
2.2 Information asymmetry and auctions
2.3 Data and Methods
2.4 Auction results
2.5 Alternate targeting of contracts
2.6 Discussion and extensions
References

33
33
37
44
50
62
69
90

Chapter 3: Reducing poverty through carbon forestry? Exploring impacts
of the N’hambita community carbon project in Mozambique
3.1 Introduction
3.2 The N’hambita Community Carbon Project
3.3 Data and Methods
3.4 Results: participation of the poor
3.5 Project impacts
3.6 Conclusion: Lessons from N’hambita
References

96
96
98
104
107
113
127
143

xi

LIST OF TABLES

Table 1.1 Descriptive statistics of variables used in the econometric models

23

Table 1.2 Descriptive statistics of dependent variables

24

Table 1.3 Determinants of demand for tree seedlings in Lake Victoria Basin

25

Table 1.4 Determinants of choice of tree species

26

Table 1.5 Joint estimation of the two demand equations by combining seemingly
unrelated regression with panel data model

27

Table 1.6 Trade-off between demand for number of trees and for timber trees
through single demand equation

28

Table 1.7 Results of the Hausman specification test

28

Table 2.1 Details of the study villages in Kinole catchment, Morogoro district

74

Table 2.2 Mean values for households that participated in the field auctions

75

Table 2.3 Characteristics and summary statistics of auction results

76

Table 2.4 ANOVA results for checking collusion among auction participants
seated together

77

Table 2.5 Determinants of auction bids

78

Table 2.6 Trade-offs from different targeting approaches

79

Table 2.7 Budgetary allocation under uniform pricing and alternate targeting

80

Table 3.1 Profile of project villages in Chicale Regulado

131

Table 3.2 Important components of the N’hambita project

131

Table 3.3 Carbon sequestration rates under the N’hambita project

132

Table 3.4 Determinants of participation in agroforestry activities

133

Table 3.5 Selected statistics for sampled households 2001 and 2008

134

Table 3.6 Cross-sectional analysis of the sampled households before project (2001)

135

xii

Table 3.7 Cross-sectional analysis of sampled household after project (2008)

136

Table 3.8 Impacts of the N’hambita project: Difference in difference from 2001 and
2008 status

137

Table 3.9 Comparison of the project area with the control area: Difference in difference
from 2001 and 2008 status
138

xiii

LIST OF FIGURES

Figure 1.1 Mean number of trees under different scenarios

29

Figure 1.2 Estimated demand schedule for tree seedlings

29

Figure 2.1 Checking for potential collusion among auction participants seated together

81

Figure 2.2 Box plots of auction bids received from groups of participants seated together 82
Figure 2.3 Estimated supply curve for enrolling land for tree planting

83

Figure 2.4 Relationship between the wealth status and the auction bids

84

Figure 2.5 Trade-offs in alternate targeting of carbon contracts (Round 1 bids)

85

Figure 2.6 Trade-offs in alternate targeting of carbon contracts (Round 2 bids)

86

Figure 2.7 Budgetary allocation under uniform pricing and alternate
PES targeting (Round 1 bids)

87

Figure 2.8 Budgetary allocation under uniform pricing and alternate
PES targeting (Round 2 bids)

88

Figure 3.1 Major components of the N’hambita community carbon project

139

Figure 3.2 Distribution of sampled households: Proportion of households in
different income categories (2001) and proportion of households with carbon contracts

140

Figure 3.3 Pattern of migration into Chicale Regulado, Mozambique

141

xiv

CHAPTER 1: EXPLORING DEMAND FOR FORESTRY IN LAKE VICTORIA BASIN,
KENYA: AN ECONOMETRIC APPROACH

1.1 INTRODUCTION
This paper estimates the impact of economic incentives and farmer characteristics on
demand for a forestry program in the Lake Victoria basin in Western Kenya. Demand is
quantified as the additional number of trees and the kinds of tree species (exotic or indigenous)
that a farmer is willing to plant on her farm, while economic incentives are explored in terms of a
hypothetical subsidy that farmers would receive for planting each additional tree. The results
presented in the paper are based on a survey with local households in the Yala and Nyando river
basins, which contribute a significant proportion of water flow into Lake Victoria.
It was conducted under the Western Kenya Integrated Ecosystem Management (WKIEM)
project, funded by the Global Environment Facility, which aims to conserve these river basins
through forestry activities with local farmers.
Sediment flow into Lake Victoria due to large scale soil erosion in its catchment is a
major concern for ecologists and environmentalists. Recent studies uniformly indicate the
occurrence of severe land degradation in important catchment areas such as the Nyando River
basin, which contributes to the growing silt inflow into Lake Victoria. Land degradation of this
magnitude has significant negative impacts on soil fertility and water quality in the surrounding
area, resulting in rapid colonization of the lake by water hyacinth and decreased fish and aquatic
plant diversity. The economic impact includes reduced fish catch from the lake and escalation in
maintenance costs for operating hydroelectric turbines in Uganda downstream from the lake
(ICRAF and KARI, 2004). Consequently, initiatives such as the WKIEM project focus on

1

reducing silt inflow into the lake by taking up afforestation and reforestation activities in the
upper catchment areas. The present study originated from the need to estimate the feasibility of
such a forestry program in the area. It had two main objectives: (i) to prepare a socio-economic
baseline of the area, and (ii) to assess the feasibility of a forestry program in the area by
estimating the effects of economic incentives and relevant demographic characteristics of
farmers on their willingness to plant new trees on their farms. It was conducted in collaboration
with the World Agroforestry Center (ICRAF), one of the implementing organizations for the
WKIEM project.
The study builds on existing research pertaining to adoption of agroforestry and farm
forestry by smallholders in developing countries. These research studies usually analyze whether
or not a household will adopt agroforestry practices and factors that determine this choice
(Mercer, 2004). For instance, Nkamleu and Manyong (2005) look at socio-economic factors that
affect adoption of agroforestry practices in Cameroon. They find that men are more likely to
adopt new agroforestry practices such as live fencing. Other significant factors include family
size (positive impact on adoption), security of land tenure (positive), and agroecological zone
(probability of adoption being low in forest margins). Similarly, Franzel (1999) point out a
strong association between wealth and adoption of agricultural fallows across Kenya and
Zambia. Other important factors included in this paper are gender of the farmer and significance
of off-farm income for the family. Marenya and Barrett (2007) use panel data to show that
resources available with a household (farm size, livestock, off-farm income, and family labor
supply) had a significant effect on likelihood of improved soil fertility management practices in
Kenya.

2

A valuable contribution to this literature is by Pattanayak et al. (2003), who reviewed 120
papers on adoption of agricultural and forestry technology by smallholders. They report five
categories of factors that are most significant – preferences, resource endowments, market
incentives, biophysical characteristics, and risk and uncertainty. The authors find that although
market incentives are important, only a handful of studies consider all of these factors.
Furthermore, only nine percent of the studies analyzed the explicit impact of prices on adoption
rates.
This gap in the literature on the potential role of economic incentives in forestry adoption
assumes significance amidst the recent emergence of payments for environmental services
(PES). PES pertains to a system of payments or other economic rewards to land stewards in
return for providing valuable environmental services such as carbon sequestration and watershed
conservation (Wunder, 2005). In Africa such payments have been used to encourage farmers to
invest in farm forestry and in protection of existing forests (Jindal et al., 2008). However, an
objective estimate of the impact of such payments on actual adoption rates in the field or on
demand from farmers for new forestry practices is still unknown in most cases. Therefore, a
study on the effect of direct economic incentives on farmers’ willingness to adopt new forestry
practices can tie these two different strands of research together. The present paper makes an
attempt in this direction by exploring the impact of prices of tree seedlings on farmers’
willingness to plant additional trees on their farms. Using this strategy, the paper is able to show
a direct relationship between environmental payments (in the form of a per seedling subsidy) and
the supply of conservation (in the form of additional trees that can be planted) in the Lake
Victoria basin. The paper also looks at the differential impacts of relevant socio-economic
characteristics. The methodology adopted is based on attribute-based survey methods.

3

1.2. CHOICE OF SURVEY METHODS
Since the objective of this study was to assess farmers’ preferences regarding ex ante
adoption of new agroforestry practices (say in terms of additional trees to be planted), it could
typically take two forms, i.e. either a contingent valuation type study or an attribute based study.
Contingent Valuation Method (CVM) is a survey-based methodology for eliciting values people
place on goods, services and amenities (Boyle, 2003). In a CVM survey, respondents are asked
to state their willingness to pay (WTP) for a good, or their willingness to accept (WTA)
compensation to voluntarily give up a good. Both of these are Hicksian consumer surplus
measures and the specific context determines which one is used – WTP is used if the respondent
does not have the property right over the good in the status quo while WTA is used if the
respondent has a legal entitlement over the good and is being asked to give up that entitlement
(Carson, 2000). For the present purpose, the study could ask farmers either their minimum WTA
to plant more trees on their farm (by giving up the option to other crops) or their maximum WTP
for an upstream afforestation project that reduces soil erosion downstream. However, such a
study would need to pre-determine the level of the desired environmental service, say the
specific number of trees to be planted, while farmers state their WTA (or WTP) for this service.
If, on the other hand, the objective is to assess the intensity of environmental service that
different farmers are willing to provide, then an attribute based method (ABM) is more
appropriate.
The objective of an ABM type study is to estimate respondents’ preferences for a
divisible set of attributes of an environmental good (Holmes and Adamowicz, 2003).
Respondents can be asked to choose between two versions of an environmental program that
differ by attribute levels. Including price as a variable can then help in estimating the economic

4

value of these attributes. For instance, potential participants of a forestry program can be asked
to choose between either planting additional timber trees or fruit trees on their farms, with the
program assistance varying depending on which tree species is selected. This will help in
determining different levels of environmental service that can be available as well as the total
program cost for each of these levels.
The present study therefore incorporated a variant of the ABM to assess the feasibility of
a forestry program in Lake Victoria basin by estimating demand for the number of tree seedlings
and the kinds of tree seedlings at different price schedules. Farmers were asked to elicit the kinds
of tree species and the number of additional trees they would be willing to plant under three
different scenarios, one where they would receive free seedlings, a second where they would
1

have to pay KSH 10 (Kenyan Shillings) per seedling, and a third where they would receive
KSH 10 per seedling. There were thus three observations per respondent, although in each case
the farmer would receive the seedling, and only the net price would vary.
One of the important requirements for a stated preference study is to make the scenario
realistic for the respondent by clearly specifying the good and reminding her that participation in
providing the good is voluntary (Carson et al., 2001). The present study incorporated this feature
by presenting realistic price schedules and by reminding respondents that they could decline to
plant more trees. Furthermore, respondents were told that payments would only be made six
months after the seedlings were planted and on the basis of the actual number of surviving
seedlings. Another requirement is to include relevant demographic characteristics of the
respondents (Carson, 2000). For the present study, the list of relevant socio-economic variables
was adopted from characteristics that have been found to be significant by previous agroforestry
1

The exchange rate at the time of the study was USD 1 = KSH 75.
5

studies: gender, age, and marital status of the respondent, total farm land available with the
household, ownership status (whether or not household has secure title), percent land area under
food crops, annual expenditure of the household in the previous year (as a proxy for the annual
2

income), total livestock owned by the household, kind of roof on the dwelling, total agricultural
3

labor available at the household, if the household had a member with a permanent job outside
the farm, and the geographical location of the farm (see table 1.1). The study estimates a set of
two demand equations:
1

Yi = f(P, Zi, Hi)
2

Yi = g(P, Zi, Hi)

------------------------------------- (1)
------------------------------------ (2)
1

2

Where the number of tree seedlings ‘Yi ’ and the kinds of tree seedlings ‘Yi ’ that a
household ‘i’ demands is a function of price of seedlings ‘P’, respective household
characteristics ‘Zi’ and farm characteristics ‘Hi’.
The two demand equations were estimated on with data from a survey of 277 households,
4

conducted from June to August 2005 in western Kenya . These households were selected with
stratified random sampling. The survey was conducted in the Nyando and Yala river basins
2

Since the primary purpose was to see if this variable was significant in explaining demand for
forestry, instead of calculating total livestock units, the study summed up the number of large
animals (cows, bulls, sheep, and goats) for each household.
3
Again, the study used a simple approach of summing up the total number of adults (>16 years)
at the household and applying the following weights for individual members: 0 = no involvement
on family farm, ½ = part-time involvement, and 1 = full-time involvement on family farm.
4
In all the survey covered 313 households. However, 36 observations had to be discarded due to
missing values and incorrect entries in the database. There is a slight probability that this
elimination could be biased against women headed households who could not respond to all the
questions. Alternatives such as assuming mean values for missing observations or analyzing an
unbalanced data set were not employed however.
6

where particular sub-locations were selected as the first level of stratification. The target
population for this survey therefore comprised all the inhabitants of the two river basins. For
each sub-location, we selected the farthest point from the main road that was accessible by car
and starting from this point, three researchers then went in opposing directions to interview the
first five households in each direction.
Respondents were usually the senior most male or female available in a house. Interviews
were conducted as per a survey instrument that was administered to all respondents. The survey
questionnaire was pre-tested and modified several times to make it realistic and culturally
appropriate for the local population. The data were analyzed using STATA software.
1.3. BRIEF DESCRIPTION OF THE DATA
Out of the 277 respondents included in this study, 44.8 percent were male, 69.7 percent
were married, while 28.2 percent were either widowed or separated. The average age of the
respondent was 46.4 years (see table 1.1). Most farmers were smallholders with 4.9 acres
average land ownership per household, 53.1 percent of which was under food crops in the year
before the survey. Only 38.9 percent of the households had at least one secure land title for
different pieces of land they farmed on. The average labor availability per household for farm
work was 3.6 units. The average annual expenditure per household was KSH 45,314 in the
previous year with 76.5 percent of the families living in dwellings with metal roofs. Only one
fourth (25.9 percent) of families had at least one member with a permanent job outside the
family farm. The respondents were about equally distributed across the two geographical strata
with 56.9 percent of respondents located in Nyando River basin.
Since each respondent was offered three price schedules, she decides about the specific
mix of tree species in each case and the number of seedlings she wanted to plant. Table 1.2

7

shows the mean response under each scenario: when the farmer buys seedlings at KSH
10/seedling, when the farmer is offered free seedlings, and when the farmer is paid KSH
10/seedling for planting and protecting additional trees.
If the farmers have to buy seedlings, they are willing to plant an average of 44 seedlings
per household. Demand per household increases to 203 seedlings if farmers receive free
seedlings and further to 245 seedlings per household if they receive a direct economic incentive
to plant additional trees (please see figure 1.1). 62.1 percent of the households prefer to plant at
least one timber tree species when they have to buy seedlings, increasing to 86.2 percent when
seedlings are available for free. Interestingly, the trend is non-monotonic as the willingness to
plant timber species reduces slightly to 82.3 percent when farmers receive economic incentives
to plant new trees along with free seedlings (table 1.2). It is also important to note that most
timber species listed by the respondents such as Eucalyptus, Casuarina equisetifolia, and
Gravellia pteridifolia are fast growing trees that are exotic to the area.
1.4. ECONOMETRIC ANALYSIS
We estimate the two demand equations (1) and (2) through three different procedures in
order to utilize the panel nature of our data (there are three observations per respondent for each
of the two demand equations). First, the two demand equations are estimated separately using
relevant panel data (fixed effects, random effects) methods. This is followed by joint estimation
of the two demand equations using the Seemingly Unrelated Regression (SUR) method. Finally,
we estimate the marginal effect of tree species on the number of trees that a household would
like to plant by combining the two equations into one. While the specifics of each method are
discussed separately, in general, each of them produces an identifiable model since our main
explanatory variable – price of tree seedlings – is introduced exogenously.

8

1.4a. Panel data model to estimate demand equations
In order to estimate the marginal effect of the price of the tree seedlings, the two demand
equations can be rewritten as:
Log (Yit ) = α1Pit + α2 Zi + α3 Hi + Ci + Eit
Wit = β1Pit + β2 Zi + β3 Hi + Di + Uit

----------------------- (3)
----------------------- (4

i

=

1,2,3,…..277

t

=

The three prices offered to each respondent, i.e. -10, 0, +10.

Yit

=

Number of trees farmer ‘i’ is willing to plant at price ‘t’

Wit

=

Choice of tree species farmer ‘i’ makes at price ‘t’

=

1 if farmer selects at least one timber tree species
0 if farmer doesn’t select any timber tree species

Pit

=

Price for individual ‘i’. It takes three values for each respondent: 0 (respondent
gets free seedlings), -10 (respondent needs to pay KSH 10/seedling), and +10
(respondent gets paid KSH10/seedling for planting trees)

α, β

=

Respective slope parameters for the two equations

Zi

=

Observable demographic characteristics for individual ‘i’.
These include age, gender, marital status of the respondent. Also included are
household characteristics such as farm land, labor availability, and annual
expenditure.

9

Hi

=

Observable farm level characteristics for respondent ‘i’, such as location of the
farm and if the household own the title to this farm.

Ci , Di =

Respective unobservable characteristics for individual ‘i’ in the two equations

Eit, Uit =

Respective error terms in the two equations

Both equations (3) and (4) represent panel data model with three observations per
individual farmer corresponding to three price schedules. In this model, the within farmer
variation is provided by price ‘t’, while between farmer variation is provided by demographic
characteristics ‘Zi’ and farm level characteristics ‘Hi’.
We estimate equation (3) by fixed effects (FE) and random effects (RE) panel models. As
discussed above, both fixed effects and random effects models will produce consistent estimates
since our key panel variable, ‘price’ is introduced exogenously (i.e. it is uncorrelated with the
unobserved heterogeneity Ci or Di). In section 4c below, when one of the panel variables is
likely to be correlated with the unobserved heterogeneity, we discuss the criteria for selecting
between the two models. Finally, to account for left censoring of the dependent variable in
equation 3 {log (Yit) or log (number of trees)} at zero (10 percent or 83 observations), we also
use random effects tobit (RE tobit) to estimate the relevant parameters. The results are reported
in table 1.3. Most variables are of the same sign and within the same range across the three
models, which indicates the robustness of these estimates.
Columns 2, 3, and 4 in table 1.3 indicate that apart from price, the other significant
explanatory variables include the gender of the respondent, the age of the respondent, labor
availability at the household level, and whether or not the household has a formal title to its farm.

10

Since the coefficients are virtually the same, we report estimates from the random effects model
(column 3) in the following discussion. As suggested by Pattanayak et al. (2003), the economic
incentive on seedlings has a strong positive effect on willingness to plant trees. Each KSH 1
incentive per seedling increases the mean willingness to plant trees by 11 percent (coeff. -0.11 on
price) i.e. by 18 trees (each percentage change is equal to 1.6 trees). This makes sense in a
developing country such as Kenya where access to seedlings or germplasm is an important
determinant of success of an agroforestry program and where farmers may not have the required
financial resources to bear the upfront costs of buying these seedlings (Roshetko et al., 2007a).
Similar to Marenya and Barrett (2007), gender has a strong effect on willingness to plant trees.
Ceteris paribus, each male is willing to plant almost 100 more trees more than a female (coeff.
0.66). Family labor supply also has a positive effect, with the presence of each additional
member with full time involvement in agriculture resulting in an average increase in 21 trees that
the household would like to plant (coeff. 0.13). Interestingly, older respondents are less likely to
plant trees than their younger counterparts, though the effect is small (coeff. -0.02). As suggested
by Roshetko et al. (2007b), we also find that secure tenure to farmland has a huge positive effect
and increases the mean willingness to plant trees by 30 percent or almost 50 trees (coeff. 0.30).
With a formal land title, farmers perhaps feel more secure in planting high value and long
gestation tree crops, although this result is in contrast to Nkamleu and Manyong (2005), who
found that farmers in Cameroon were more likely to invest in live fencing through trees as a way
to strengthen their claims over land for which they did not own a formal title. Further, the wealth
status of the household as reflected by the total expenditure in the previous year is only
significant at 88 percent. However, as expected, the sign is positive, which shows that better off
households are more likely to plant trees. However, many of the other variables such as access to

11

off-farm income (permanent employment outside the farm), farm area, and location of the farm
are insignificant at the usual confidence intervals.
In order to identify the marginal effects of various explanatory variables on the choice of
tree species (equation 4), we again use fixed effects and random effects models. Since the
dependent variable is binary (1 = respondent selects at least one timber species), we also use
5

random effects logit (RE logit) to estimate the demand parameters (table 1.4). As in the
previous case, the estimates from the three models (columns 2, 3, and 4) are similar in magnitude
and have the same signs. Since the same variables are significant across both random effects and
random effects logit models, for ease of interpretation we use results from the random effects
model (column 3, table 1.4) in the following analysis.
The three variables that have a significant effect on choice of tree species include price
along with the respondent’s age and gender. Surprisingly, none of the other household or farm
level variables is significant. As expected, price of tree seedlings influences choice of tree
species, though the effect is small; each incentive of KSH 10 per seedling increases the
probability of selecting at least one timber species by 0.1 percent (coeff. -0.01 on price). On the
other hand, older people are less likely to select timber trees, although the marginal effect is quite
small (coeff. -0.002). The strongest determinant of choice of tree species is the gender of the
respondent, with males much more likely to prefer timber trees than females (coeff. 0.09). We
think that this is due to existing customs amongst local communities in western Kenya, which
dictate that women are not allowed to plant trees, especially timber species on family farms
(Fortmann, 1985). Since women can only plant a restricted list of species, the effect shows up in
5

The RE logit model can be written as: log(Prit/1-Prit) = β1Pit + β2 Zi + β3 Hi + Di + Uit
where Probability ‘Pr’ = 1 denotes the probability that individual ‘i’ selects at least one timber
tree species at price ‘t’. The right hand side of the model is as before in equation (4).
12

their reduced preference for timber trees and as we observed above, even for the number of trees
that they would like to plant.
1.4b. Seemingly Unrelated Regression (SUR) to jointly estimate demand equations
An important aspect of the two demand equations (3) and (4) is that they refer to the
same individual ‘i’. Therefore, we expect that even after accounting for various explanatory
variables, the combined residual from equation (3) i.e. (^ci + ^eit,) will be correlated with the
same from equation (4), i.e. (^di + ^uit,). Indeed, we find that:
ρ{(^ci + ^eit,), (^di + ^uit,)} = 0.4107

----------------------------- (5)

Equation (5) indicates that there is medium to strong positive correlation between the two
combined residuals. This does not, however, affect the consistency of our previous results, since
as we noted earlier, the main explanatory variable, ‘price’ is introduced exogenously in both
equations. However, we can exploit this result to improve the efficiency of our model by jointly
estimating the two demand equations through Seemingly Unrelated Regression (SUR) (Zellner,
1962). Wan et al. (1992) follow this approach to estimate a set of production functions.
According to SUR, we can combine the two sets of demand equations (3) and (4), and rewrite
them as follows:
Ygit = βg1Pgit + βg2 Zgi + βg3 Hgi + µgit ----------------------- (6)
g

=

number of equations, i.e. g = 1, 2

=

1 denotes that dependent variable is log (number of trees)
2 denotes dependent variable is respondent’s preference for timber species

i

=

1,2,3,…..277, as before

t

=

The three prices offered to each respondent, i.e. -10, 0, +10, as before
13

Pit

=

Price for individual ‘i’ as before.

βg

=

Respective slope parameters for equation ‘g’

Zgi

=

Observable demographic characteristics for individual ‘i’ in equation ‘g’

Hgi

=

Observable farm level characteristics for respondent ‘i’ in equation ‘g’

Ci , Di =

Respective unobservable characteristics for individual ‘i’ in the two equations

µgit

Respective error terms in equation ‘g’ for each panel variable ‘t’

=

Equation (6) represents a set of six equations for each individual ‘i’: combination of two
demand equations (g = 1, 2), each of which in turn contains three panel equations (t = -10, 0,
+10). In order to estimate these six equations jointly, we use the method suggested by Biørn
(2004), which has been programmed into STATA command ‘xtsur’ by Nguyen (2008). The
estimation results are reported in table 1.5.
Comparison of results from table 1.5 with those from tables 1.3 and 1.4 show that
although most variables retain their coefficients as before, we gain on precision. However, there
are also some important variations. Table 1.5 shows that price, gender of the respondent,
household labor availability, economic well-being of the household (log of expenditure in the
previous year), size of land holding (square root of farmland), and possession of secure title to
farmland are all significant in explaining the demand for tree seedlings. As before incentive per
seedling has a positive effect on willingness to plant (coeff. -0.11 on price). Similarly, demand
for trees is much higher amongst males than females, and the coefficient value (0.71) is slightly
more than what we observed in table 1.3. Secure title has a positive effect on willingness to plant
as before (0.3), as is the effect of higher labor supply at the household level (0.14). There are two

14

additional variables that are now significant at 5%, log of annual expenditure (as a measure of
the household’s economic well being) has a positive effect to demand for trees. Increase in
expenditure by KSH 1 has an associated elasticity of 0.23 percent (coeff. 0.23). The size of
landholding also has a positive effect (coeff. 0.24), which is in line with the finding of Marenya
and Barrett (2007) in terms of factors that result in higher propensity of agroforestry adoption.
More farm size provides flexibility to the household to divert land from food crops to more
permanent tree crops.
Variables that affect choice of tree species now include price of seedlings, age and gender
of the respondent, socio-economic status of the household (as indicated by whether or not the
household dwelling has a metal roof), land holding, secure tenure to farmland, labor supply, and
location of the household’s farm (table 1.5). As before, incentive per seedling has a positive
effect on probability to select at least one timber species (coeff. -0.01 on price). Similarly, males
are more likely to pick at least one timber species, though the magnitude of the effect is now
larger (0.12). On the other hand, increasing age has a small positive effect on likelihood of
selecting timber trees (0.005) as compared to the earlier case when age had a small negative
effect on selecting timber trees (table 1.4). Also, the sign on age is now reversed between the
demand for trees (-0.002) and preference for timber trees (0.005). This indicates that older
farmers prefer to plant slightly fewer trees, but they have a higher preference for timber trees.
The same is also observed for location of the farm in either of the two watersheds
Nyando or Yala. Location of a farm in Nyando is associated with a strong preference for timber
trees (0.12), although it is also associated with a lower mean willingness to plant trees than in
Yala (coeff. of -0.11 in table 1.5 when the dependent variable is log of the number of trees that a
household demands). Field transects indicate that Nyando is drier and more erosion prone than

15

Yala. This implies that a higher proportion of farmland in Nyando is marginal and we believe
that farmers in Nyando therefore have less flexibility in terms of the area they can divert to trees.
However, they would still like to receive quick returns from this land by selecting timber trees
that mostly include fast growing exotics such as Eucalyptus. Higher labor availability (0.03),
more land holding (coeff on square root of land holding is 0.07), and secure tenure (0.07) all
have a positive effect on preference for at least one timber species. Similarly, the socio-economic
status of the household, as indicated by the presence of a metal roof on the main dwelling, also
results in positive demand for timber trees (0.14). These factors indicate that availability of more
resources for a household translates into higher demand for timber trees.
1.4c. Trade-off between demand for trees and timber species
In the previous discussion, we looked at the marginal effects of various explanatory
variables on demand for trees and on demand for timber species. In section (4a), we estimated
the two demand equations separately, while in section (4b), we estimated them jointly, though
still as a system of two independent equations. However, in order to estimate the marginal effect
of preference for timber species on demand for the number of trees, or to estimate the trade-off
between the two variables, we need to combine them into a single equation as follows:
Log (Yit) =

β1Wit + β2Pit + β3 Zi + β4Hi + Ci + µit

----------------------- (8)

i

=

1,2,3,…..277, as before

t

=

The three prices offered to each respondent, i.e. -10, 0, +10, as before

Yit

=

Number of trees farmer ‘i’ is willing to plant at price ‘t’, as before

Wit

=

Choice of tree species farmer ‘i’ makes at price ‘t’, as before

=

1 if farmer selects at least one timber tree species

16

0 if farmer doesn’t select any timber tree species
Pit

=

Price for individual ‘i’ as before.

β

=

Respective slope parameters

Zi

=

Observable demographic characteristics for individual ‘i’, as before

Hi

=

Observable farm level characteristics for respondent ‘i’, as before

Ci

=

Unobserved heterogeneity for individual ‘i’, as before

µit

=

Error term, as before

The main difference between equation (8) and the previous equations is that preference
for timber trees (Wit ) is now added on the right hand side as one of the explanatory variables.
Since both Log (Yit) and (Wit ) are jointly determined by the same household ‘i’, this may
perhaps lead to the notion that equation (8) suffers from simultaneity issues and is therefore
unidentified. However, as Wooldridge (2002b, page 529) explains, the two variables are not
simultaneously determined through a market equilibrium as is the case of typical demand and
supply equations. Although equation (8) can again be estimated by both fixed effects (FE) and
random effects (RE) methods, a crucial assumption for RE estimates to be valid is (Wooldridge,
2002a):
Covariance (Wit , Ci) = 0

------------------------------------ (9)

Table 1.6 reports the respective FE and RE parameters for equation (8). However, the
subsequent Hausman specification test (Hausman, 1978) is significant at 1% (table 1.7) which
means that the assumption in equation (9) is invalid and there exists correlation between choice

17

of tree species (Wit ) and the individual specific heterogeneity (Ci) in the right hand side of
equation (8). Hence, we can only consider the FE estimates as consistent:
Log(number of trees demanded) = 1.87 * - 0.09 price* + 2.2 species*
(0.108)
Overall R-sq = 0.3682

(0.005)

No. of observations = 822

-------- (10)

(0.133)

No. of groups = 274 * Significant at 1%.

Equation (10) indicates that selection of timber species is associated with more demand
for number of trees. Ceteris paribus, selection of at least one timber tree species results in twice
as much additional demand for number of trees as when compared to selection of all non-timber
species by a respondent. Further, the selection of timber species is equivalent to providing a
monetary incentive of KSH 25 per seedling to generate the same level of demand for trees in the
area.
1.5. DISCUSSION: SIGNIFICANCE OF RESULTS
The results presented in this paper confirm that direct economic incentives to land
stewards can indeed improve the provision of an environmental service; ceteris paribus, farmers
in Western Kenya are willing to plant about 18 more trees for every KSH of direct payment to
them. This result is robust to different econometric approaches. Admittedly, the price range
covered in this study is limited between +10KSH/seedling and -10KSH/seedling, but even within
this price range, the effect on demand for seedlings is significant. The partial effect is almost the
same as the slope coefficient on labor supply, which implies that provision of KSH1 of economic
incentive per seedling has the same effect as adding one adult member to a household who works
full time on the family farm. The estimated slope coefficients from the RE model (column 3 in
table 1.3) can also be used to construct a demand curve for a forestry program in the study area.
Demand is seen in terms of the number of additional trees that farmers are willing to plant, both
18

when farmers receive per tree economic incentive and when they have to pay for seedlings
themselves. Figure (1.2) shows that demand for tree seedlings amongst sampled households
(n=277) goes down from 125,000 seedlings at an economic incentive of KSH 25 /seedling to
almost zero when farmers have to pay KSH 25/seedling (this is the choke price at which a
forestry program is unlikely to generate much demand in the area).
These results have a direct significance for ICRAF and other implementing organizations
in the region. These organizations can use the results from this study to further explore demand
for forestry in the region. For instance, farmers in the Yala River Basin are much more likely to
plant trees than farmers in the Nyando River Basin. Therefore, a forestry initiative in the area
such as WKIEM is well advised to begin its activities in Yala rather than in Nyando. During this
phase, implementing organizations can try to identify factors that constrain adoption of forestry
in Nyando and as the program matures, use their experience to introduce appropriate activities in
Nyando.
The paper also raises concern for local and international NGOs like ICRAF that support
planting of indigenous tree species instead of exotics. The results presented here clearly show
that people are more likely to demand timber species such as Eucalyptus, Casuarina
equisetifolia, and Gravellia pteridifolia rather than fruit or slow growing indegenous trees.
Availability of such exotic timber species also raises people’s willingness to plant more trees on
their farms. Since exotics can sometimes be associated with long run ecological disaster,
especially on dry lands, research organizations will need to come up with suitable economic
incentives to promote indigenous tree species in the area. This may include provision of
differential economic incentives, for instance providing higher payment per tree when farmers
select indigenous tree species.

19

1.6. CONCLUSION: LIMITATIONS OF THE STUDY
The purpose of this study was to assess the feasibility of a forestry program in Lake
Victoria Basin by exploring farmers’ willingness to plant additional trees on their farms. The
study is able to confirm that there is significant potential for a forestry program in the region,
especially if farmers are offered direct economic incentives to take up plantations. Mean
willingness to plant trees increases almost six times from 44 trees per household to 244 trees per
household when farmers receive an economic incentive of KSH 10/seedling as compared to
when they have to pay KSH 10/seedling. At the usual planting density of 2.5m X 2.5m, this
translates to putting about 0.5 acres of farm land per household under tree plantations (10 percent
of the mean landholding per household in the area) when it receives an economic incentive of
10KSH/seedling along with free seedlings. The study is also able to confirm some results
reported by previous agroforestry studies. For instance, women are less likely to choose timber
species than men, while families with higher labor availability are more likely to plant new trees.
While these results are encouraging, the study also suffers from some important
limitations. The price schedule explored in the survey is rather limited and can only predict
demand within a narrow range. Since the purpose of the present study was to explore whether
economic incentives have any effect on provision of an environmental service, the answer is
unconditional yes. A subsequent study will however need to explore economic incentives in
greater depth by offering more price choices. Further, the study does not account for endowment
effects. For instance, it assumes a one to one correspondence between demand for additional tree
seedlings and the number of trees that a farmer is likely to plant. It does not deal with the
difference in farmers’ perceptions when they get free seedlings versus when they have to pay for
them.

20

Finally, while the study demonstrates the positive effect of payments that are conditional
on survival of the tree seedlings, it still does not address the question of impermanence. The
impact of contract duration in terms of the minimum number of years for which the farmers need
to protect their trees (beyond the first six months) on mean willingness to participate in the
forestry program remains unexplored. Similarly, the study does not include the effect of
provision of fines and penalties on farmers’ demand for tree seedlings. This is an important issue
because in many forestry-based PES projects, farmers receive upfront payments to take up new
plantations. However, once the payments end, they have little incentive to continue protecting
the trees, threatening the sustainability of the project. Therefore, more work is needed on how to
estimate demand for PES/forestry projects when the payments also include the provision of
continuity of protection in the long run.

21

Appendices

22

Table 1.1: Descriptive statistics of variables used in the econometric models
(n = 277)*

Variable
Price**
Gender
Age
Marital status
Land

Title
Percent
Farmland
Under
foodcrops
Livestock
Dwelling Roof

Labor
Availability
Access to
Permanent Job

Description
Net price per seedling for the
farmer (in KSH). Three price
schedules were offered.
Gender of the respondent.
1 = male, 0 = female
Age of the respondent in
years
Marital status of the
respondent.
0 = not married, 1 = married,
2 = separated/widowed
Land owned by the
household (in acres)
Possession of formal land
title.
1 = if the household has a
title to at least one piece of
land
0 = no formal title
Proportion of total land
under food crops (in percent)
Total number of large
livestock owned by the
household
Kind of roof on the dwelling.
1 = metal sheets,
0 = thatch/grass
Total agricultural labor
available at the household
(units) after accounting for
part-time and full-time
involvement
If a household member has a
permanent job
1 = at least one member has
a permanent job,
0 = no member has a
permanent job
23

Continuous
variable
Mean
Std.Dev.
0
8.17

Dummy
variable

1 = 44.8%
46.4

15.48
0 = 2.2%
1= 69.7%
2 = 28.2%

4.9

5.96

1 = 38.9%
53.1

25.98

7.1

8.58
1 = 76.5%

3.6

2.06

1 = 25.9%

Table 1.1 (cont’d)
Annual
Expenditure

Total annual expenditure of
the household during
previous year (in KSH)
Geographical location of the
farm
1 = Nyando river basin
0 = Yala river basin
If the household would like
to plant additional trees.
1 = yes, 0 = no

Block

Willingness to
Plant additional
trees

45,313.8 139,799.7
1 = 56.9%

1 = 99.6%

*Sample size is 277 for all variables except for Age (n = 275) and Dum_Block (n = 276).
** Price introduces panel effect in the model with three observations/respondent, n = 831

Table 1.2: Descriptive statistics of dependent variables
(n = 277)
Variable

Number
of
seedlings
demanded

Mean

Choice of
tree
species

Dummy variable (%)
1 = if the respondent
chose at least one exotic
timber species
0 = if no timber species
were selected

Buy Seedlings
(farmers pay
KSH10/seedling)
44

Std. Dev.

115.9

1 = 62.1%

24

Free
seedlings
(farmers
Get Paid
get free
(farmers get paid
seedlings) KSH10/seedling)
203
245
425.8
493.5

1 = 86.2%

1 = 82.3%

Table 1.3: Determinants of demand for tree seedlings in Lake Victoria Basin
Dependent Variable = Log (number of trees), i.e. Log (Yit)
Fixed Effects
Price per seedling (KSH)
Gender of respondent (1 = male,
0=female)
Age of respondent (years)
Secure land title (0/1)
Access to off-farm income (0/1)
Labor Supply per HH (number)
Presence of metal roof on dwelling
(0/1)
Log (annual expenditure in KSH)
Square root of land holding in acres
Location (1 = Nyando, 0 = Yala)
Constant
Number of observations
Number of groups

-0.11(0.006)*** -0.11(0.005)***
(dropped)
0.66(0.16)***

Random Effects
Tobit
-0.12(0.006)***
0.67(0.16)***

(dropped) -0.02(0.005)***
(dropped)
0.30(0.15)***
(dropped)
-0.19(0.18)
(dropped)
0.13(0.04)***
(dropped)
-0.13(0.18)

-0.02(0.005)***
0.33(0.16)***
-0.19(0.19)
0.14(0.04)***
-0.15(0.19)

(dropped)
(dropped)
(dropped)
3.58(0.0001)***
822
274
R-sq. = 0.2261

Random Effects

0.10(0.07)
0.09(0.08)
-0.07(0.17)
2.46(0.71)***
822
274
R-sq. = 0.3105

Prob. > F = 0.00 Prob. > F = 0.00

Figures in parentheses represent robust standard errors.
*** Significant at 1%
** Significant at 5%
* Significant at 10%

25

0.12(0.08)
0.10(0.09)
-0.05(0.19)
2.25(0.78)***
822
274
Log Likelihood =
-1508.77
Prob>chi sq = 0.00
83 left censored
observations
739 uncensored
observations

Table 1.4: Determinants of choice of tree species
Dependent variable: Choice of tree species, i.e. Wit
Fixed Effects
Price per seedling (KSH)
Gender of respondent (1 = male,
0=female)
Age of respondent (years)
Secure land title (0/1)
Access to off-farm income (0/1)
Labor Supply per HH (number)
Presence of metal roof on dwelling
(0/1)
Log (annual expenditure in KSH)
Square root of land holding in
acres
Location (1 = Nyando, 0 = Yala)
Constant
Number of observations
Number of groups

Random Effects

-0.01(0.002)***
(dropped)

-0.01 (0.002)***
0.09 (0.04)***

Random Effects
Logit
-0.1(0.02)***
0.89(0.40)***

(dropped)
(dropped)
(dropped)
(dropped)
(dropped)

-0.002(0.001)*
0.05 (0.04)
-0.04(0.05)
0.01(0.013)
0.01 (0.05)

-0.03(0.01)*
0.58(0.40)
-0.52(0.47)
0.09 (0.09)
0.15(0.48)

(dropped)
(dropped)

0.02(0.02)
0.009 (0.02)

0.18(0.21)
0.12(0.24)

(dropped)
0.78 (0.0001)***
822
274
R-sq. = 0.0376

0.47 (0.05)
0.59 (0.20)***
822
274
R-sq. = 0.0704

Prob >F = 0.00
Figures in parentheses represent robust standard errors.
*** Significant at 1%
** Significant at 5%
* Significant at 10%

26

0.55(0.45)
0.15(1.91)
822
274
Log likelihood =
-372.79
Prob >F = 0.00 Prob >chi sq. = 0.00

Table 1.5: Joint Estimation of the two demand equations by combining
Seemingly Unrelated Regression with Panel Data Model (n = 825)
Coefficient

Standard Error

Dependent Variable = Log (number of
trees), i.e. Log (Yit)
Log (annual expenditure in KSH)
Age of respondent (years)
Price per seedling (KSH)
Square root of land holding in acres
Secure land title (0/1)
Labor Supply per HH (number)
Gender of respondent (1 = male, 0=female)
Location (1 = Nyando, 0 = Yala)

0.23***
-0.002
-0.11***
0.24***
0.31**
0.14***
0.71***
-0.12

0.03
0.005
0.005
0.09
0.16
0.04
0.16
0.17

Dependent Variable =Choice of tree
species, i.e. Wit
Presence of metal roof on dwelling (0/1)
Age of respondent (years)
Price per seedling (KSH)
Square root of land holding in acres
Secure land title (0/1)
Labor Supply per HH (number)
Gender of respondent (1 = male, 0=female)
Location (1 = Nyando, 0 = Yala)

0.14***
0.005***
-0.01***
0.07***
0.07*
0.03***
0.12***
0.12***

0.05
0.001
0.001
0.02
0.05
0.01
0.04
0.05

*** Significant at 1%
** Significant at 5%
* Significant at 10%

27

Table 1.6: Trade-off between demand for number of trees
and for timber trees through single demand equation
Dependent Variable = Log (number of trees), i.e. Log (Yit)
Fixed Effects
-0.09 (0.005)***
2.24 (0.133)***
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
(dropped)
1.87 (0.108)***
822
274
R-sq. = 0.3682
Prob. > F = 0.00

Price per seedling (KSH)
Choice of timber species (0/1)
Gender of respondent (1 = male, 0=female)
Age of respondent (years)
Secure land title (0/1)
Access to off-farm income (0/1)
Labor Supply per HH (number)
Presence of metal roof on dwelling (0/1)
Willingness to plant trees (0/1)
Log (annual expenditure in KSH)
Square root of land holding in acres
Location (1 = Nyando, 0 = Yala)
Constant
Number of observations
Number of groups

Random Effects
-0.09 (0.005)***
1.96 (0.12)***
0.5 (0.15)***
-0.01 (0.005)**
0.19 (0.15)
-0.11 (0.18)
0.11 (0.04)***
-0.14 (0.19)
1.27 (1.21)
0.07 (0.08)
0.08 (0.09)
-0.15 (0.18)
0.08 (1.38)
822
274
R-sq. = 0.4242
Prob. > F = 0.00

Figures in parentheses are standard errors.
*** Significant at 1%
** Significant at 5%

Table 1.7: Results of the Hausman Specification Test

Price
Choice of tree species

Coefficients
FE (2)
RE (3)
-0.09
-0.09
2.24
1.96
Chi sq. = 18.66
Prob. > Chi sq. = 0.0001

28

Difference
(2) – (3)
0.00
0.28

Standard
Error
0.0002
0.065

Figure 1.1: Mean number of trees under different scenarios

Number of trees

300
244

250

202

200
150
100
44

50
0

Buy seedlings

Free seedlings

Get Paid

“For interpretation of the references to color in this and all other figures, the reader is referred to
the electronic version of this dissertation.”
Figure 1.2: Estimated demand schedule for tree seedlings

Price Per Tree Seedling (KSH)

30
20
10
0
0

25000

50000

75000

100000

125000

-10
-20
-30

Demand for Tree Seedlings (in number)

29

150000

References

30

REFERENCES

Biørn, Erik. 2004. Regression system for unbalanced panel data: a stepwise maximum likelihood
procedure. Journal of Econometrics. 122: 281-291.
Boyle, K.J. 2003. Contingent Valuation in Practice, in Champ, P.A., K.J. Boyle, and T.C. Brown
(eds). 2003. A Primer on Nonmarket Valuation. Kluwer Academic Publishers, Boston
Carson, R.T., N. E. Flores, and N.F. Meade. 2001. Contingent Valuation: Controversies and
Evidence. Environmental and Resource Economics. Vol.19. Pp. 173–210.
Carson, R.T. 2000. Contingent Valuation: A User’s Guide. Environment Science and
Technology. Vol. 34. Pp. 1413-1418
Franzel, S. 1999. Socioeconomic factors affecting the adoption potential of improved tree
fallows in Africa. Agroforestry Systems. Vol. 47. Pp. 305-321.
Fortmann, L. 1985. The tree tenure factor in agroforestry with particular reference to Africa.
Agroforestry Systems. Vol 2. Pp 229-251.
Hausman, J.A. 1978. Specification Tests in Econometrics. Econometrica. Vol.46. Pp.1251-1271
Holmes, T.P., and W.L. Adamowicz. 2003. Attribute-Based Methods, in Champ, P.A., K.J.
Boyle, and T.C. Brown (eds). 2003. A Primer on Nonmarket Valuation. Kluwer
Academic Publishers, Boston.
ICRAF and KARI. 2004. Western Kenya Integrated Ecosystem Management: Project Brief.
World Agroforestry Centre (ICRAF) and Kenya Agricultural Research Institute
(KARI), Kenya.
Jindal, R., Swallow, B., and Kerr, J. 2008. Forestry-based carbon sequestration projects in
Africa: Potential benefits and challenges. Natural Resources Forum. Vol. 32, Pp.116–
130.
Marenya, P.P., and C. Barrett. 2007. Household-level determinants of adoption of improved
natural resources management practices among smallholder farmers in western
Kenya. Food Policy. 32(2007) 515-536.
Mercer, D.E. 2004. Adoption of agroforestry innovations in the tropics: A review. Agroforestry
Systems. Pp. 311-328.
Nguyen, Minh Cong. (2008). XTSUR: STATA module to estimate seemingly unrelated
regression model on unbalanced panel data. Department of Economics, American
University, Washington, D.C.
31

Nkamleu, G.B., and V.M. Manyong. 2005. Factors affecting the Adoption of Agroforestry
Practices by Farmers in Cameroon. Small-scale Forest Economics, Management and
Policy. Vol. 4(2). Pp. 135-148.
Pattanayak, S., D.E. Mercer, E. Sills, and J. Yang. 2003. Taking stock of agroforestry adoption
studies. Agroforestry Systems. Vol. 57, Pp. 173-186.
Roshetko J.M., Lasco R.D., and Delos Angeles MD. 2007a. Smallholder agroforestry systems
for carbon storage. Mitigation and Adaptation Strategies for Global Change. 12:
219–242
Roshetko JM, Nugraha E, Tukan JCM, Manurung G, Fay C and Van Noordwijk M (2007b)
Agroforestry for livelihood enhancement and enterprise development. In
Djoeroemana S, Myers B, Russell-Smith J, Blyth M and Salean IET (eds) Integrated
Rural Development in East Nusa Tenggara, Indonesia. Proceedings of Workshop to
Identify Sustainable Rural Livelihoods, Kupang, Indonesia, ACIAR Proceedings No
126, 5–7, April 2006, p197
Wan, G.H., W.E. Griffiths, and J.R. Anderson. Using Panel Data to Estimate Risk Effects in
Seemingly Unrelated Production Functions. Empirical Economics. (1992) 17:35-49
Wooldridge, J.M. 2002a. Econometric Analysis of Cross Section and Panel Data. The MIT
Press, Cambridge.
Wooldridge, J.M. 2002b. Introductory Econometrics: A Modern Approach (2nd ed.). Thomson
South Western Publishers.
Wunder, S. 2005. Payments for environmental services: Some nuts and bolts. Occasional Paper
No. 42. Centre for International Forestry Research (CIFOR), Jakarta, Indonesia.
Zellner, A. 1962. An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests
for Aggregation Bias. Journal of the American Statistical Association. Vol. 57, No.
298 (June, 1962), pp. 348-368

32

CHAPTER 2: ESTIMATING ‘PAYMENT’ IN PAYMENTS FOR ENVIRONMENTAL
SERVICES: RESULTS FROM FIELD AUCTIONS IN THE ULUGURU MOUNTAINS,
TANZANIA

2.1. INTRODUCTION
This paper estimates the level of payment necessary to procure carbon sequestration
services through payments for environmental services in the Uluguru Mountains, Tanzania.
Payment for environmental services (PES) is a new conservation paradigm that focuses on
incentive payments to land stewards for investing in new land use practices that lead to
conservation or production of specific environmental services (i.e. positive externalities)
(Wunder, 2005). In general, conserving land-based environmental services with off-site benefits
is difficult when it is not in the private interest of the land stewards or the associated opportunity
cost is high (Pagiola and Platais, 2007). PES helps in aligning private interests of land stewards
with the value that wider society places in such services (Engel et al., 2008). Payments under
PES can either come from eventual users of environmental services or from conservation
agencies that are interested in environmental protection. The efficacy of the PES approach
depends on the level of economic incentive available to service providers; the incentive should
be direct and it should adequately compensate the service provider for the opportunity cost of
investing in a new land use practice (Ferraro and Kiss, 2002). PES thus mimics a market
transaction where the price determines how much of the good is produced.
In recent years there has been rapid growth in PES-based projects to secure valuable
environmental services across the globe (Huang et al., 2009; Southgate and Wunder, 2009). A
seminal paper on this work identified more than 250 PES schemes operational in different parts

33

of the world (Landell-Mills and Porras, 2002). PES examples vary widely and include such
projects as China’s Grain for Green program spread over more than 100 million hectares of
erosion prone land (Yin and Yin, 2010), the Nhambita Community Carbon Project in Africa that
aims to protect 11,000 hectares of forestland in Mozambique (Jindal, 2010), a local payments for
watershed conservation project in Heredia, Costa Rica that protects upstream watersheds for
downstream water users (Kenney, 2009), and a World Bank funded biodiversity conservation
project in Nicaragua (Pagiola et al., 2008).
As envisaged in the Kyoto Protocol’s Clean Development Mechanism and as observed
from several forestry projects across the globe, PES has also become the dominant approach in
securing forest based carbon sequestration services through absorption and storage of
atmospheric carbon by trees (Miles and Kapos, 2008). There are now numerous projects that pay
local land owners to sequester carbon by planting new forests or by protecting existing ones

6

(Hamilton et al., 2010; Jindal et al., 2008). Success of these projects is therefore directly linked
with the effectiveness of the PES approach, which in turn depends on identifying a price that
reflects the value of conservation while compensating land owners’ opportunity costs. If the
payment is too low, land owners will remain under-compensated implying that many potential
suppliers will opt out of the project. If on the other hand the payment is too high, service
producers will claim all the surplus from the transaction and the project will fail to deliver an
adequate level of environmental service for the buyers or what conservationists call ‘the biggest
bang for the buck’ (Pagiola and Platais, 2007; Jack et al., 2008). The challenge of achieving this
has brought a lot of skepticism to the PES approach (Kosoy and Corbera, 2010).
6

Technically, growing new trees in afforestation projects is different from conserving existing
forests that come under the purview of reduced emissions from deforestation and forest
degradation (REDD). There is a move now to combine the two under a proposed REDD+ regime
(Van Noordwijk et al., 2009).  
34

One of the constraints of the PES approach is that in the absence of competitive markets
for environmental services such as biodiversity and watershed conservation, it is hard to
determine the price or payment to offer to land stewards as suppliers. When markets do exist, as
in the case of carbon sequestration, they are so differentiated that there is no single price that can
be paid (Hamilton et al., 2010). Ex ante determination of price is also necessary because many
projects either include onetime contracts or are of long duration whereby renegotiation of the
contract is costly once it has begun. For instance, carbon sequestration projects need to ensure
the long duration or permanence of carbon stored through project activities (Haites, 2004).
Therefore, the terms of the project, including the payment level, have to be clearly laid out ex
ante in order to obtain a long term commitment from the suppliers. If these terms are changed in
the middle of the project, land stewards may discontinue their conservation efforts, jeopardizing
the entire carbon that has been stored historically. Moreover, it is difficult to directly transfer
cost estimates from one project to another since the cost of implementing a new land use practice
is often site (and farmer) specific. When measuring production costs is expensive, especially on
new project sites, providers may have little incentive in revealing their true costs. Estimating an
efficient payment level is therefore both methodologically and practically significant for PES
projects.
One potential method to estimate payments is through conservation auctions where PES
contracts are allocated to potential service providers through competitive bids (Ferraro, 2008).
Compared to conventional auctions, the roles of buyers and sellers (or service providers) are
reversed in these auctions and successful bids from potential service providers are decided on the
basis of how low they are rather than how high (Ferraro, 2008; Giampietro and Emiliani, 2007).
Although such conservation or reverse auctions have become popular in developed countries

35

such as the US, Australia and the UK, they have yet to be fully explored in developing country
contexts (Latacz-Lohmann and Schilizzi, 2005). This is an important gap considering that a high
proportion of PES projects including most carbon sequestration and REDD projects are proposed
for the developing countries. Many of these projects can gain from a method that can help in
calibrating an efficient level of payment to service providers. Moreover, most auctions are only
able to collect data on bids from potential environmental service providers but not on their socioeconomic profiles. As a result, researchers are unable to measure the extent to which poorer land
stewards participate in these auctions or whether or not they are actually awarded any PES
contracts. For a large number of developing country PES projects with a focus on poverty
alleviation, this is a serious limitation.
This paper addresses several of these concerns. It is based on field experiments in the
Uluguru Mountains of Tanzania where local farmers were invited to submit bids on the amount
of money they were willing to accept in return for providing carbon sequestration services
through adoption of agroforestry practices on their fields. It is organized as follows: the next
section presents a review of auction theory and empirical evidence to identify why an auction
can be an appropriate economic institution to estimate payments in PES projects. This is
followed in section three by a description of the context in which the study was undertaken,
including details on the specific auction format (second-price sealed bid, uniform pricing) that
was used in the field. Section four presents results from field trials and their implications for
project managers. Analysis of farmers’ bids and an econometric model of their household
characteristics help to answer several important questions that are discussed in detail: (i) what
level of payment is necessary to generate carbon sequestration services in the area, (ii) to what
extent are poorer households likely to be contracted, and (iii) how would different targeting

36

approaches affect the level of environmental service generated. This also includes a simulation of
different policy options such as giving higher weight to poor participants and how these options
affect payment levels, project costs and area covered.
2.2. INFORMATION ASYMMETRY AND AUCTIONS
Producing environmental services often requires a change in land use (Antle and
Stoorvogel, 2006). For instance, in order to generate carbon sequestration services, farmers will
need to plant additional trees on their farms. If such trees are not privately profitable this will
result in a decline in profits for the farmer, both from additional costs incurred in purchasing and
planting new seedlings, as well as from a change in labor inputs and a potential decline in crop
yields from the farm area now occupied by trees. This decline in profit or the opportunity cost of
adopting the new land use will inhibit the farmer from doing so. If a farmer were to be
compensated or paid for the loss in profit, she would be willing to adopt the new set of land use
practices. However, only the farmer knows a large proportion of this opportunity cost (e.g.
change in labor inputs) and this creates an information asymmetry between the farmer and the
project manager (Ferraro, 2008).
j

To formalize, let the production function for farmer ‘i’ be represented as yi where j = 0
denotes conventional set of practices and j = 1 denotes the new land use practices that includes
0

0

0

tree planting. Further, yi = f(hi xi ), where yi is the output per farmer from a conventional set of
0

practices, xi denotes the inputs used by the farmer (e.g. labor, seeds, fertilizers etc.), and hi is
the efficiency of with which these inputs are converted into outputs. Following Khanna et al.
(2002), the efficiency of input use can be construed as a function of the biophysical
characteristics of the farm li (soil quality, location of the farm) as well as the farmer specific

37

characteristics zi (age, gender, education of the farmer), i.e. hi = g(li, zi). The production function
0

f(•) has the usual properties with f’ > 0, and f’’ < 0. If c(xi ) is the cost function associated with
using the inputs and P the set of output prices, the gross profit for the farmer from the
conventional set of practices is found by solving:
0

0

0

πi = max {Pf(g(li, zi)xi )) – c(xi )}

--------------------- (1)
0

The farmer selects the optimal level of inputs xi such that:
0

0

Pf’(g(li, zi)xi )) – c’(xi ) = 0

-------------------- (2)

With the profit from the conventional set of practices being:
0

0

0

π*i = Pf(g(li, zi)x*i )) – c(x*i )

------------------- (3)

Similarly, under the new land use practice where the farmer plants additional trees on her farm (j
= 1), the maximum profit can be written as:
1

1

1

π*i = Pf(g(li, zi)x*i )) – c(x*i )

------------------- (4)

And, the change in profit or the opportunity cost of the farmer ‘i’ as:
0

Δ π*i = π*i – π*i

1

------------------- (5)

Usually, provision of carbon sequestration services requires not only planting of new tree
seedlings, but an ex ante assurance to protect the trees for a certain number of years (Haites,
2004). If under a new PES contract a farmer is required to protect her trees for say T number of
years, then her total opportunity cost can be written as (Miller and Tolley, 1989):

bi =

Δ π*i e

-rt

dt

------------------------- (6)

38

If there are N heterogeneous farmers who operate in a watershed, the farmers can be
ordered by their opportunity cost of providing carbon sequestration services (Paarsch and Hong,
2006):
b1:N ≤ b2:N ≤ . . . . . . . . ≤ bN:N

------------------------- (7)

Where b1:N is the opportunity cost of the lowest cost provider, and bN:N is that of the
highest cost provider. If the ordering as in (7) were known to the project manager, then she could
not only calibrate the level of payment that would induce the local farmers to adopt the new land
use practice but also estimate the supply of carbon sequestration services at each payment level.
However, there is an information asymmetry between the farmers and the project manager such
that only the farmers know their opportunity costs. Asking farmers to state their opportunity
costs (say in the form of a stated preference survey) may not result in correct ordering as farmers
lack incentive to reveal their true costs (Ferraro, 2008).
Some authors have suggested how a set of two different contracts, one for the low cost
providers and the other for high cost providers, can help address this information asymmetry and
encourage self-selection of potential service providers (Wu and Babcock, 1996; Gren, 2004).
However, if there are multiple cost types, it is not only tricky to formulate such screening
contracts but also extremely difficult to apply in practice unless one knows the distribution of
opportunity costs among the target population (Ferraro, 2008). A possible solution to this
problem is conservation auctions where potential service providers are invited to place bids on
what payment they are willing to accept in return for providing a specific level of the
environmental service, with the competition among bidders ensuring that they have an incentive
to reveal their true opportunity costs (Latacz-Lohmann and Schilizzi, 2005; Cason and
39

Gangadharan, 2004; Latacz-Lohmann and Hamsvoort, 1997). An ordering of bids can then be
used to estimate the supply curve for the provision of the environmental service across the entire
landscape (Platinga et al., 2001; Jack et al., 2008).
Conservation or reverse auctions have been tested in many developed countries both to
estimate the level of payment in PES projects and to allocate the actual conservation contracts.
Perhaps the best known example of such auctions is the US Conservation Reserve Program
(CRP), which pays farmers to set aside their land from production purposes to conservation use.
The program began in 1985 to protect ecologically vulnerable land from soil erosion and for
conserving other valuable natural resources (Rousseau and Moons, 2008). Farmers received an
annual payment per acre for removing land from crop production provided their bids were below
the maximum rental rate (bid cap) fixed by federal officials. Since 1990, bids from farmers have
been weighted on the basis of an environmental index that scores parcels of land on the basis of
the environmental benefits their inclusion in the program would provide to society. Parcels with
the highest score are enrolled first, followed by parcels with a lower score and so on until the
enrollment targets are met (Khanna and Ando, 2009). Nationwide, several million hectares of
land are enrolled under CRP through auctions with significant cost savings for federal agencies
(Classen et al., 2008). In Georgia, Cummings et al. (2004) used a series of auctions to inform
state policy makers on how best to buy back irrigation permits from local farmers in drought
years. Similarly, the BushTender program in Australia uses conservation auctions to promote
native vegetation and biodiversity protection on private lands (Latacz-Lohmann and Schilizzi,
2005). Stoneham et al. (2003) report how the allocation of contracts through these auctions
resulted in high biodiversity benefits at a reduced cost when compared to a fixed-price approach.
Despite several other examples of conservation auctions across the industrialized world, to the

40

best of our knowledge there are only two examples of conservation auctions in developing
countries – one pertaining to watershed management contracts in Indonesia (Jack et al., 2008)
and the other on tree planting contracts in Malawi (Jack, 2010). Both of these were experimental
auctions. Clearly, field application of conservation auctions in developing country settings
remains underexplored.
Perhaps one reason why conservation or reverse auctions have been used so rarely in
developing countries is their perceived complexity (Ferraro, 2008). In general, auctions need
thick markets or a large number of potential bidders to operate well (Klemperer, 2002a), which
may not be always feasible. In addition, there are many auction formats to choose from ranging
from the ascending English auctions to the descending Dutch auctions. Though in theory the
different auction formats produce the same outcome (Myerson, 1981), in practice the results may
vary due to diverse risk preferences and a divergence in information processing capability of
potential bidders (Athey et al., 2004). From a PES perspective, therefore, the specifics of auction
design affect the efficiency of contract allocation and the level of payment that service providers
receive.
Researchers differentiate between an independent values paradigm where bidders’ value
of a contract (or opportunity cost of a change in land use) is unrelated, and a common values
paradigm where bidders’ valuations are correlated (Milgrom, 1989). In a PES context, where the
cost of adopting a certain practice is farmer- and farm-dependent, and where service providers
cannot resell the conservation contracts they receive, an independent values paradigm is more
appropriate (Paarsch and Hong 2006). Auction outcomes also depend on whether the bids are
submitted orally or as sealed bids. While oral bids favor strong bidders, sealed bidding can

41

7

improve the chances of weak bidders to win the auction (Athey et al., 2004) . Further, sealed
bids help in estimating the supply curve for environmental service provision as the entire
spectrum of bids from potential service providers are observed by the auctioneer, as compared to
oral bidding where only a small sample of the bids are observed (Paarsch and Hong 2006).
For threshold benefits (e.g. a certain proportion of a watershed must be brought under
conservation for any discernable downstream benefit), or in order to produce a marketable level
of ES (e.g. a minimum number of carbon offsets that are needed to cover administrative costs of
a project), PES projects often allocate multiple contracts in the form of land parcels that are
required to follow the recommended land use. These multiple contracts can be auctioned
simultaneously and/or sequentially. If service providers can bid for multiple contracts, then the
marginal value to them of each additional contract decreases with the number of contracts they
have already obtained (Krishna, 2002). In a simpler design, each service provider can be asked to
bid for only a single contract, while the auctioneer can still allocate multiple contracts to all the
bidders whose bids were equal to or below the highest accepted bid. In this case, however, the
auctioneer still needs to decide between discriminative payments (where each service provider
receives a payment equal to her bid) and uniform payments (all winning bidders receiving the
same level of payment). In an auction experiment, Cason and Gangadharan, 2005) found that
even though bidders mark up their bids to earn profits in a discriminative price auction, it is still
more efficient than the uniform price auction. As expected, the conservation agency captures
most of the surplus (including producer surplus) in a discriminative price auction as compared to
the uniform case where most bidders receive much higher payments than their bids. However,
7

Strong bidders are the ones who are more likely to win an auction while weak bidders are less
likely. In a conventional auction, bidders with high valuation of the object being auctioned are
strong bidders while the ones with low valuation are the weak bidders.
42

for PES settings in developing countries, discriminative price auctions may be politically
infeasible or perceived as unfair by local landholders (Ferraro, 2008). Further, uniform price
auctions may provide a higher incentive for bidders to reveal their true opportunity costs (Cason
and Gangadharan, 2005).
In a seminal paper, Vickrey (1961) showed how second-price sealed bid auctions have a
dominant truth-revealing equilibrium strategy and produce efficient outcomes. In other words,
potential service providers can do no better than by revealing their true opportunity costs when
asked to bid in a second-price sealed bid or Vickrey auction. If they bid lower than their
opportunity cost, the value of the PES contract (i.e. the payment they receive) is less than their
opportunity cost and they may end up with a loss. If they bid higher than their opportunity cost
(in order to gain a profit), they may not get the contract at all. However, since the winners stand
to receive payment equal to the lowest rejected bid (which will be higher than their bid except
for the marginal bidder), their dominant strategy is to place a bid equal to their opportunity cost.
In spite of this incentive compatibility, however, Vickrey auctions are rarely used in practice, not
just in PES settings but also in sale of other objects that are routinely allocated through auctions.
Rothkopf et al. (1990) propose that the two most important factors that thwart the use of Vickrey
auctions are the fear of bidder collusion and resistance among bidders to reveal their true values
(or costs) to others. Klemperer (2002a) suggests that 1) the presence of thick markets where
many bidders compete and 2) using a sealed bid process should help address bidder collusion.
Further, resistance among bidders to conceal their true costs can be addressed by keeping the
winning bids secret (Rothkopf et al., 1990) and through the uniform payment system where only
the last rejected bid is announced by the auctioneer.

43

Econometrically, a second-price sealed bid (or Vickrey) auction model is identified
because of the dominant equilibrium strategy for all bidders to reveal their true costs (Paarsch
and Hong, 2006). Therefore, using a set of farm and farmer specific observables, one can
potentially estimate the marginal effect of each of these factors on the farmer’s bid as observed
in the auction, i.e. from eqs. 1 through 7:
bi = Ø(li, zi, xi)

-------------------------- (8)

where individual ‘i’ specific bid ‘bi’ is a function of farmer specific characteristics ‘zi’, farm
specific characteristics ‘li’, and set of inputs ‘xi’. Eq. 8 can be deterministically written in the
form of a typical econometric model where the log of observed bids is regressed on observed
characteristics:
log (bi) = α + li β1 + zi β2 + xi β3 + µi

-------------------------- (9)

to estimate the set of slope coefficients {β = (α, β1, β2, β3)} and the error term ‘µ’
accommodates for functional misspecification or any relevant unobservables. Lucking-Reiley et
al. (2007) use this approach to analyze the determinants of bids placed in online eBay auctions
on US one cent pennies, while Sun and Hsu (2007) specifically look at the effect of seller
reputation on bids in online auctions. Jack et al. (2008) attempt this approach within context of
conservation or procurement auctions for PES contracts but do not get a significant model.
2.3. DATA AND METHODS
The Ulugurus are part of the Eastern Arc Mountains, which extend from southern Kenya
to the southern highlands of Tanzania. Located in Morogoro district, Tanzania, the Uluguru
Mountains provide several valuable environmental services including biodiversity, carbon
sequestration, and watershed regulation. The mountains are an important center of floral and
44

faunal diversity and are home to at least 16 endemic vertebrate and 135 endemic plant taxa
(Polhill, 1968). This degree of endemism in the Ulugurus is exceptional in tropical Africa,
putting these mountains among the continent’s ten most important conservation sites (Burgess et
al, 2002). The Ulugurus are also the source of the River Ruvuu, which provides water to Dar-esSalaam, the biggest city in Tanzania and a major economic zone. However, many of these
environmental services are under threat due to rapid deforestation in the mountains. Recent
surveys report disappearance of some endemic faunal species while heavy flow of silt during
rains threatens the quality of water in the entire network of the River Ruvuu. Reversing this
deforestation and land degradation is a must if the valuable environmental services flowing out
of the area are to be preserved. As a result, many conservation projects have been initiated in the
Ulugurus, with many focusing on PES as a way to incentivize local farmers to adopt more
sustainable land use practices (Katoomba Group, 2007).
One potential way to revitalize the local ecosystem is by growing trees on agricultural
fields (Neufeldt et al., 2009; TAFORI, 2006). In addition to efforts to protect the remaining
forest, tree planting on agricultural fields could help revitalize the local ecosystem. With the
growth in international carbon markets, the trees could also generate saleable carbon offsets.
Replacing agricultural crops with trees would however reduce farmers’ incomes, requiring
sufficient compensation for them to voluntarily adopt such a practice. The objective of the
present study is to use conservation or reverse auctions in the field to estimate the level of
payment that would induce local farmers to adopt these carbon forestry practices on their farms.
The study also looks at ways in which PES projects could improve their targeting under different
policy objectives. The field work was undertaken in collaboration with the World Agroforestry
Centre (ICRAF) under its PRESA project (Pro-poor Rewards for Environmental Services in

45

Africa). The PRESA project is exploring the use of PES approach to conserve threatened
landscapes across different countries in Africa, and the Ulugurus in Tanzania are one of the core
project sites.
The field work for this study was undertaken in Kinole catchment of Morogoro district.
The catchment contains a sizeable chunk of the Uluguru South Reserve Forest and is the source
of one of the important tributaries of the River Ruvuu. The study area comprised ten villages in
this catchment, of which Tandai is the local government headquarters and the main market place
for all the other villages. The entire area is quite remote and there is only one fair weather road
that connects it to Morogoro and beyond. Three out of the ten study villages are located higher
up in the catchment, while five villages are not even connected to this fair weather road and
locals need to trek on steep slopes to access it (table 2.1). Agriculture is the main source of
livelihood in the area, with many households augmenting their income through casual labor or
small businesses, especially in Tandai. Maize and cassava are the main food crops while banana
and pineapple are the main cash crops. With the improvement in marketing infrastructure in
recent years, the production of banana and pineapple has grown many fold and each day several
lorries laden with bananas and pineapples can be seen leaving for the nearby towns of Morogoro,
Chalinze, and even Dar-es-Salaam.
Data for the study were collected over a twelve month period in 2008 and 2009, with a
set of field auctions conducted in March 2009. During several rounds of focus groups in the area,
local farmers expressed a high willingness to participate in a potential carbon project which gave
them incentives to plant trees on their farms. Many of them favored timber trees over fruit trees
because they felt that the local marketing infrastructure was inept to handle more horticultural
crops. These focus groups were followed by a survey with 400 randomly selected households in

46

the area to collect relevant demographic and agriculture related information. The survey was
administered through a written questionnaire, preferably with the head of the household or, in
his/her absence, with the next most senior member in the family. The questionnaire included
sections on household profile (gender of the household head, number and age of different
household members), agricultural profile (number of farms, major crops, farm expenditure in the
previous year), assets owned (type of house, ownership of livestock, and number of durable
assets), and questions on the household’s time preference with regard to any cash incentive. At
the end of the questionnaire, the household was given a written invitation (with a unique
identification number) to participate in the conservation auction in the area.
The auction was held at the main marketplace in Tandai village. Only the households that
had participated in the survey and thus had written invitations with them were invited to attend
the auction. Out of the 400 households covered in the survey, 268 attended the auction.
Comparison of household data showed no systematic differences between households that did
not attend the auction and those that did. After all the participants had assembled in the central
market place, several rounds of mock auction were conducted first where familiar objects such as
bananas and cash vouchers for cell phone minutes were auctioned to train farmers on how the
actual auction would operate. Any questions from the audience were duly answered and the
entire process was explained in detail several times. Once the participants said that they were
comfortable with the auction process, two separate auction rounds were conducted inviting
farmers to provide sealed bids for carbon or tree planting contracts. The entire exercise took five
or six hours with a break for snacks and refreshments in between.
Farmers were asked to bid for the minimum payment they would be willing to receive for
planting 80 trees over 0.5 acres (at a spacing of 5x5 m) and for protecting them for at least three

47

years. During these three years, farmers were responsible for looking after their trees, although
they were free to grow crops in between the trees. However, in order to reduce the transaction
cost of making repeated payments and in order to provide an upfront incentive to farmers, the
8

entire payment for the three year contract was payable up front . However, farmers were also
told that there would be external monitoring during these three years and if they looked after
their trees well, there was a good chance that the contract would be extended (at a renegotiated
price) after three years. The three year contract period was selected to give sufficient time to the
PRESA project to look for carbon buyers willing to purchase carbon offsets generated by the tree
plantations, while local farmers had enough time to look after the trees until they became well
established and needed less maintenance and protection. Farmers were also told that they were
free to decide how to use the trees if for some reason the contract was not extended after three
years.
There were two auction rounds, each consisting of a separate carbon contract focusing on
a different mix of tree species; Khaya anthoteca (African mahogany) and Tectona grandis
(Teak) in the first round, and Khaya anthoteca and Faidherbia albida (Winter thorn) in the
second one. These species were selected after consultations with the regional experts at Tanzania
Forestry Research Institute, taking into account the local ecology. Khaya anthoteca is an
indigenous tree in the area while both Tectona grandis and Faidherbia albida have done well in
experimental trials (Godziszewski, 2009; Okorio and Maghembe, 1994). In both rounds, farmers
were told that they would receive free tree seedlings procured from a reputed nursery in
Morogoro, and they were asked to bid for their cost of planting and maintaining the tree
seedlings. At an average price of 500 Tanzanian Shillings (TSH) per seedling, the total value of
8

This is similar to many PES projects where participants receive in-kind incentives that are
mostly provided up front and are semi-conditional at best.
48

9

tree seedlings per carbon contract was thus TSH 40,000 (USD 31.50) . In the first round,
farmers were invited to bid for a mix of 80 trees (40 each of Khaya anthoteca and Tectona
grandis) on an area of 0.5 acres, while in the second round, which was also the final round,
farmers were again invited to bid for planting a mix of 80 trees (this time a mix of 40 trees of
Khaya anthoteca plus 40 trees of Faidherbia albida) on 0.5 acres. Winners in both the auction
rounds were selected using the uniform second price rule (or Vickrey auction) with the last
rejected bid setting the equilibrium price. Price information was not shared between rounds and
winning bids were only announced after both the auction rounds had been completed. Each
participant could bid in both rounds but could only receive a single carbon contract. So in the
event of a participant winning in both rounds, she had to choose one of the two.
In all, 268 bids were received in each of the two rounds. However, 17 of these bids were
disallowed because they were either illegible or outrageously high. Subsequent discussions with
these farmers revealed that they had mistakenly added another zero in their bids. In the
limitations section, we discuss the implications of excluding these bids from our analysis. Table
2.2 presents details of the remaining 251 households whose bids were included in the auction. 69
percent of the participants were males, while almost 80 percent were born in the local area with
the rest migrating from outside. The average age of the participants was 43 years and they had
completed an average education of 4.4 school years. On average, each household consisted of 7
people (including children) and owned 5 farm plots. The average ownership of animals and
10

poultry birds was 0.16 livestock units . In the previous year, the local households had spent an
average of TSH 164,265 (USD 129.30) on agricultural expenses, out of which TSH 67,323 was
9

The exchange rate in March 2009 was USD 1 = TSH 1270.

10

Livestock units estimated according to ILCA (1990).
49

for hiring labor. Further, 30 percent of all households ran a small business or had a household
member with a regular job. Almost 68 percent of households had come in contact with the
Wildlife Conservation Society of Tanzania (another local NGO) to plant trees on their farms in
2002-04.
2.4. AUCTION RESULTS
One of the primary objectives of this study was to estimate the level of payment that
would elicit conservation investments from local farmers in a potential PES project. Required
payments should cover farmers’ opportunity costs, thus their bids from the two auction rounds
should represent their perceived opportunity cost of adopting the recommended land use change
(in this case planting a mix of 80 trees over 0.5 acre and protecting them for at least three years).
This assumption underlies the paper.
In terms of field results, the bids observed in the two auction rounds (round one: Khaya
anthoteca + Tectona grandis and round two: Khaya anthoteca + Faidherbia albida) were quite
similar (table 2.3). However, there was a marked heterogeneity across farmers as indicated by a
big difference in minimum and maximum bids. The minimum bid in round one for a three year
contract was TSH 1,400 while the maximum was TSH 450,000. In comparison, the minimum bid
in round two for a similar three year contract was TSH 2,000 while the maximum was again TSH
450,000. The distribution of the bids in both rounds was skewed to the right with the mean bid in
each round (round one: TSH 143,840 or USD 113.30 and round two: TSH 138,253 or USD
108.90)

11

being more than the respective median bid (round one: TSH 130,000 and round two:

11

As a point of reference, in 2008 the average per capita income in Tanzania was USD 440
(World Bank, 2010). Further, at an average wage rate of TSH 1,500 per day in the area, a mean
bid of TSH 143,840 represents 96 days of wage labor spread over three years, or about 32 days
each year.
50

TSH 126,000). The spread of the bids in the two rounds was also similar with a standard
deviation across bids in round one of TSH 96,105.5 and TSH 93,105.4 in round two.
Starting from the lowest bidders, farmers were contracted until the researchers’ small
conservation budget was exhausted. In all, the 32 lowest bidding farmers or households (15 in
round one and 17 in round two) received three year carbon contracts at the end of the auction.
Following the uniform pricing rule, each of the 15 winning bidders in round one received a
payment of TSH 30,000, while the 17 winning bidders in round two received TSH 20,000 each.
These payments were in addition to the free tree seedlings that were provided to each winning
bidder. In all, 2,560 trees were planted as a result of the carbon contracts allocated through the
auction.
2.4.1 Checking for Collusion
Collusion or cooperative bidding is an important concern in auctions (Klemperer, 2002a;
McAfee and McMillan, 1987). A tacit agreement among bidders to bump up the price can
seriously jeopardize the efficiency with which the conservation contracts are allocated. Repeated
interaction through multiple auction rounds facilitates communication among bidders, allowing
them to cooperate or collude (Klemperer, 2002b). Similarly, provision for revision of bids during
the auction process can help low cost providers inflate their bids, resulting in efficiency losses
(Cummings et al., 2004). In general, sealed bid auctions involving many bidders are less
susceptible to collusion (Klemperer, 2002a).
In the conservation auctions in Tanzania, though we used sealed bids and the participants
were discouraged from communicating during bidding, prior familiarity among participants
could have resulted in collusion. In addition, the seating pattern during the auction could

51

facilitate group bidding (where participants sitting together bid alike), a concern that has also
been raised by Jack (2010).
To check for potential collusion we followed a three-step process. During the auction, it
was observed that most participants sat together in small groups of five to seven people. When
the sealed bids were collected from these participants, we were able to identify the groups to
which these bids belonged. In all, 41 groups comprising a total of 232 participants were
identified. The rest of the participants were dispersed individually in the auction hall and their
bids were not included in this analysis as there was little risk of collusion from their side. To
check for collusion, we took the bids from round two, in which collusion was more likely
because round one may have already helped participants to become familiar with each other. As
the first step, we did an F-test for similarity of mean bids across these 41 groups (table 2.4). The
test was reported significant which implied that differences in bids across groups were more than
differences in bids within these groups.
To ascertain whether or not this indicated collusion between bidders, we plotted the mean
bids of different groups with respect to the overall mean of all the auction participants (figure
2.1). The graph shows that out of the 41 groups, mean bids from 39 groups were within one
standard deviation of the overall mean, which means that except for two groups, the average
bidding did not vary much across groups. However, high mean bids in these two groups (mean
bids TSH 292,917 and TSH 204,167) would indicate collusion only if there was low dispersion
or spread of the bids in these two groups with all group members bidding high. The box plots of
bids for each of the 41 groups (figure 2.2) show that in fact the bids in these two groups were
highly dispersed, implying that even though some group members bid high, others within these
two groups bid low (standard deviation TSH 180,419.07 and TSH 149,780.39 respectively).

52

Interestingly, the box plot for group number 37 did show joint bidding (mean TSH 140,000; SD
TSH 5,000). However, since this behavior was limited to only one group and the mean bid of the
group was close to the overall mean of the bids from all the auction participants, we can safely
infer that there were little or no collusion in the overall auction.
2.4.2 Landscape level supply curve
Many environmental services such as watershed management and carbon sequestration
have threshold effects whereby a minimum number of farmers (or acres) need to be contracted
across a landscape in order to produce a viable level of environmental service (Parkhurst and
Shogren, 2007). Platinga et al. (2001) use the equilibrium bids observed during the CRP auctions
to estimate regional supply curves for conservation lands across nine states in the US. Similarly,
Jack et al. (2008) use the auction data to estimate a hydrological supply curve for two microwatersheds in Indonesia. We follow a similar approach to estimate a supply curve for provision
of carbon sequestration services through tree planting on private lands in the Uluguru mountains.
The bids observed in the auction can therefore be ordered according to equation (7) to estimate
the marginal cost of environmental service provision from the Uluguru Mountains. There are two
important assumptions that we make in extrapolating auction results from a sample group to the
local population: first, that our auction participants are a good representative of the entire
population in the Kinole area, and second, that local residents would only enter into PES
contracts if the payment they receive is more than their opportunity cost. In other words if they
fail to comply with the contract requirements, they may face penalties which deter them from
accepting contracts that do not compensate them adequately.

12

12

While the first assumption is fairly reasonable in our case, the second is still to be tested in a
developing country context. Usually, non-compliance results in discontinuation of payments, but
we are unaware of cases where financial penalties have been enforced.
53

Figure 2.3 shows that the upward sloping marginal cost or supply curve with bids from
the two rounds mostly overlap with each other. The graph shows that at the mean payment of
TSH 148,000 (USD 116.60) per half acre, about 62 acres (out of the 125.5 acres that were
included in the auction) could be enrolled in a tree planting program for a minimum of three
years. Since our auction participants were a random draw from the area and each farmer was
eligible for only one tree planting or carbon contract over 0.5 acres, the curve can also be used to
estimate the provision of carbon services through tree planting for the entire Kinole catchment
representing 1,227 households (table 2.1, column 2). This fulfills another important objective of
identifying the level of payment necessary to enroll a given target of acreage (or households) or
to estimate the amount of area than can be brought under a PES project with a fixed conservation
budget. For instance, for a low enrollment target of 33% of the local households (or about 41
acres out of the 125.5 acres included in the auction), and using a uniform payment arrangement,
a PES project would need to pay TSH 100,000 per household (or per 0.5 acres). For the
catchment as a whole this would lead to enrollment of about 368 local households (or 184 acres)
at a total cost of TSH 368,000,000 (USD 289,763). Similarly, for a high enrollment target of
13

80%, the project would need to pay TSH 200,000 per household , leading to enrollment of 982
households (or 491 acres of private land) at a total cost of TSH 196,400,000.
Finally, although Figure 2.3 does give an idea of the loss of consumer surplus by using
uniform payments over discriminative pricing (the area between the horizontal dotted line and
the solid dots representing the supply curve), it should be used with caution since the bids from
potential service providers would have been different if they had been informed that the auction
process would involve discriminative pricing. In all likelihood, they would have inflated their
13

It is important to note that this excludes the cost of supplying tree seedlings and any other
project administrative costs.
54

bids (Cason and Gangadharan, 2005) leading to a different supply curve than the one we observe
now. However, in contrast to the BushTender trial auctions in Australia where Stoneham et al.
(2003) expect that landowners’ bids also included information rents, we believe that in the
present case the use of the uniform Vickrey pricing rule results in reduction (or even almost
disappearance) of rents as the dominant strategy of the bidders was to reveal their true expected
opportunity costs.
2.4.3 Participation of the poor
Participation of the poor is an important concern for PES projects in developing countries
since many projects are either specifically taken up to augment rural incomes through
conservation payments or are located in areas with widespread poverty (Gong et al., 2010;
Pagiola et al., 2008; and Uchida et al., 2007). There is contrasting evidence from the field with
some projects better able to demonstrate participation of poorer households (e.g. Jindal, 2010)
than others (e.g. Miranda et al., 2003). In general, participation of the poorer households is
linked to whether or not they are low cost providers. Under any payment system, PES projects
are more likely to contract low cost providers (who gain a higher surplus from the deal) than the
high cost ones. But even then, policy makers and project managers may be interested to know the
extent to which poor households are able to participate in PES projects in a given context.
For the PES work in the Uluguru Mountains, auction data supported by demographic data
from the household survey provide a reasonable estimate of the extent to which poor households
are low cost providers. While bids observed in the two auction rounds indicate the opportunity
cost of individual households to adopt agroforestry/carbon contracts, the value of their asset
ownership as collected during the household survey was used as an estimate of their wealth
status. The correlation between the bids (which represent the bidder’s expected opportunity

55

cost)

14

and the asset ownership (wealth status) is weak and negative at -0.07. This indicates that

many of the poor households may be high cost providers while many of the better-off households
may have a lower cost of providing carbon services. This is confirmed when the wealth status of
the households is plotted against their bids in the auction (figure 2.4). Many of the poor
households with asset ownership estimated to be less than TSH 50,000 did signal a low
opportunity cost in terms of their bids (and were indeed contracted in the auction), which were
lower than the mean bid of TSH 138,253 for the overall group. However, a significant proportion
of poor households also reported a high bid, below which they would evidently be disinclined to
enroll. On the other hand, many better-off households with asset ownership in excess of TSH
250,000 were also low cost providers and were in fact included in the list of 17 households that
were contracted under this study immediately after the auction. These results indicate that some
poor households but not all were able to participate in PES activities. For a project that would
specifically like to contract the poorer households first, there are thus efficiency and budgetary
implications to which we return in section five below.
There are many possible reasons for some poorer households’ high bids. They may own
only a small number of agricultural plots, so that diverting them from food and cash crops to
trees would have a high opportunity cost. Similarly, poor households may also face a labor
constraint, making labor-intensive investment in a new land use practice expensive, or they may
attach high risk to locking their land into a contract that requires maintaining tree cover for a
minimum of three years. We check for some of these variables presently by exploring more
carefully the determinants of bids received during the auction.
2.4.4 Determinants of bids
14

The bids in the two auction rounds were similar; we used the ones from round two for this
analysis.
56

An econometric analysis of auction bids with respect to bidder characteristics or
covariates is usually untenable for two important reasons: (1) many bids may remain unobserved
in auction formats that elicit oral bidding resulting in truncated data, and (2) when the auction is
not truth revealing, the auctioneer only observes the bid but not the underlying valuation of the
contract (Rezende, 2008). However, in the Uluguru auctions, use of sealed bidding with Vickrey
uniform second pricing helps to address both these constraints and makes the econometric model
identifiable (Paarsch and Hong, 2006). Sealed bidding ensures that all bids are observed and
uniform second pricing makes the auction incentive compatible for bidders to reveal their true
valuation in the form of their bids. Thus modifying equation (9) to account for two rounds of
auctions that we conducted, we can analyze the marginal effects of various characteristics on
observed bids as:
log bit = α +Dt + li β1 + zi β2 + xi β3 + µit-------------------------- (10)
where, as before, individual ‘i’ specific bid ‘bi’ is written as a function of farmer specific
characteristics ‘zi’, farm specific characteristics ‘li’, and set of inputs ‘xi’. The ‘t’ subscript
represents the two auction rounds (t = 1,2), with the dummy variable ‘Dt’ taking the value ‘0’
when t=1 (first auction round) and ‘1’ when t=2 (second auction round). β = (α, β1, β2, β3) are
the respective parameter estimates. Eq. 10 can be solved using the ordinary least squares for each
of the two auction rounds (columns 1 and 2 in table 2.5). However, by accounting for the fact
that we have two observations per individual in the form of separate bids in the two auction
rounds, we also use Pooled OLS (column 3 in table 2.5) and a random effects panel data model
(column 4) to get more precise estimates (Wooldridge, 2002). All four models report estimates
that are robust to system heteroskedasticity with standard errors clustered at the individual level.
57

As table 2.5 shows, the overall regression model is significant and most variables have the same
sign and magnitude across all the four models. For the ease of interpretation, unless mentioned
specifically, we therefore report the estimates from Pooled OLS (column 3).
As expected, the dummy on bid round (Dt) is insignificant since the bids across the two
auction rounds were quite similar (figure 2.3). Farmer-specific variables that returned significant
(zi) include gender and age of the bidder, livestock units owned by the household, and the wealth
status of the household as reflected by value of its asset ownership. On average, males tended to
bid higher than females. The slope coefficient for gender in the case of the Pooled OLS model
(column 3, table 2.5) is 0.21, which means that males bid 21 percent higher than females. Recall
that each percentage change in bid is equal to about TSH 1,400 (USD 1.10). This means that
ceteris paribus on average a bid from a male was higher by TSH 29,400 than a bid from a
female, perhaps representing the higher opportunity cost that males attached to their labor inputs
for the new land use practice. An increase in the age of the bidder by one year resulted in 0.6
percent increase in the bid. Similarly, increase in livestock ownership by 1 unit (which in turn is
equal to 10 goats or 100 chickens) reduces the bid by 29 percent or by TSH 40,600. Ownership
of livestock provides an alternate source of income to the household, thereby reducing the risk
attached with a new land use practice. However, as we observed earlier, asset ownership has a
small negative effect on the bid with increase in wealth by TSH 1,000 only reducing the bid by
0.01 percent. Although this is a small effect, it is significant across all the four models. We
already discussed the significance of this result in section 4.3 above on participation of the poor.
Since both labor constraint and number of plots are controlled in the model, it is more likely that
poor households bid higher as they faced a higher risk of locking their land in a new crop with

58

respect to the better-off households who have additional wealth/income to cushion them if the
new land use did not work out.
An interesting variable that has a significant impact on the bid is whether or not the
bidder also responded to the survey questionnaire administered a few weeks before the auction.
The questionnaire covered demographic characteristics as used in equation (10) above, along
with a few questions on the household’s willingness to participate in a carbon contract involving
agroforestry adoption. For this, the household was provided with some hypothetical incentives
that the project was considering and the household was asked to respond on whether or not it
would be willing to participate. The minimum incentive level was no payment and the maximum
hypothetical incentive was a payment of TSH 45,000 per annum. In most cases, it was the head
of the household who responded to the survey, while only in 8 percent of the cases (over a
sample of 400 households), the head was unavailable and so the questionnaire was administered
to the next adult member in the household. However, due to the economic significance of the
auction, we anticipate that it was only the head of the household who participated in bidding. The
dummy variable that reflects whether or not the head of the household responded to the survey is
significant and has a huge negative effect on the bids (43 percent). This probably implies that
farmers who were already aware of the nature of the carbon contract and had some idea about the
payment level bid much lower than others. Even though the variable is significant across all the
four models, it needs more investigation since there isn’t much variation in the variable itself
with only 8 percent of the bidders not contacted by the previous survey.
In contrast to some of the conservation auctions in Malawi (Jack, 2010), the farmer’s
time preference as reflected by whether or not she reported high preference for immediate returns
was insignificant across all the four models. This is perhaps due to the fact that the entire

59

payment for the three year contract was paid to farmers up front and so their time preference did
not affect their bid values.
In terms of inputs (xi), as expected the main variable that is returned significant is labor
inputs measured in terms of the number of household members or household size. On average,
each additional household member reduced the bid by 4 percent or by TSH 5,600 which reflects
the decrease in opportunity cost when a household has additional labor available to invest in a
new land use practice. In terms of farm-level characteristics (li), several variables are significant:
number of plots per bidder, average distance of the bidder’s plots to the nearest road, average
elevation of the plots, and location of the plots (whether or not the plots are located in either of
the two main villages Tandai and Kalundwa). Surprisingly, increase in ownership of farm plots
results in higher bids, with each additional plot resulting in an average increase by 4 percent or
TSH 5,600. This though is not a large increase. Managing diverse practices on a small number of
plots may thus have lower opportunity cost than managing them on more plots, especially if they
are located far away from each. (We were unable to get reliable data on inter-plot distances.)
During the survey (or even in the auction afterwards), we were unable to obtain precise
data on the location of each plot that a household owned. Instead, we demarcated the local area
into ten micro-catchments and estimated the average elevation of these micro-catchments and
their average distance from the nearest road, and whether or not they corresponded to one of the
two main villages of Tandai and Kalundwa. So, for each plot, we have data on whether or not it
lies in either of the two main villages, its average distance from the nearest road, and its average
elevation, based on which of the ten micro-catchments it is located in. Table 2.5 shows that
farmers with plots located in either of the two main villages bid higher by 28 percent which
indicates that land located in either of the two main villages is more valuable. On the other hand,
60

a greater distance from the nearest road resulted in higher bids, with a 100 m increase in distance
raising the bid by 2 percent on average. This is an unexpected result that needs deeper
investigation within the community, because we expected the opportunity cost of land to
decrease when it is located away from the road. Perhaps farmers find it difficult to access such
plots and so they attach a higher opportunity cost. But even then, putting such land under
permanent tree cover should actually be preferable to annual crops which may need more
frequent care. On the other hand, if it is more difficult to look after trees on plots that are too far
from the road, then it makes sense for farmers to associate higher cost for planting and protecting
trees on these plots. Finally, for every 10 m increase in elevation, the bids reduced by 3 percent,
which indicates the lower opportunity cost of land in the higher parts of the local watershed.
Although we do not have empirical data on average slope in each micro-catchment, transect
walks in the local area do show that upper parts of the watershed are more steeply sloped than
the lower ones, which is also confirmed by Yanda and Munishi (2007). So as the average
elevation increases, the increasing slope makes it more difficult to grow crops, thus reducing the
opportunity cost of land.
Finally, to the best of our knowledge this is one of the few econometric analyses of
auction bids in a developing country context. Even though the R square (the percentage of
explained variance) is low for all the four models, the analysis still provides useful insights to
policy makers and PES managers on the marginal effects of various socio-economic variables on
auction bids, which can help them in designing conservation contracts that are appropriate for a
given context.

61

2.5. ALTERNATE TARGETING OF CONTRACTS
Economists have suggested a range of targeting tools to capture the heterogeneity in
environmental benefits and opportunity cost of changing land use across land parcels that can
potentially be contracted in a PES project (Babcock et al., 1997; Ferraro, 2003). Using marginal
benefit and cost curves, these targeting tools can help in understanding trade-offs among
different contracting arrangements. They include (i) using a cost-only approach to target parcels
of land with the lowest opportunity cost first, which would maximize the acreage enrolled under
a PES project, (ii) using an environment-only targeting under which parcels with the highest
environmental benefits are enrolled first, and (iii) targeting parcels using the environmental
benefit-opportunity cost ratio, which would produce a cost-efficient outcome or what is called
‘the biggest conservation bang for the buck’ (Ferraro, 2003; Babcock et al., 1996).
In recent years, a lot of emphasis has also been placed on designing PES projects in a
way that enables poor households to participate as potential service providers (Pagiola et al.,
2008). Under a take-it-or-leave-it system where most PES projects offer a fixed level of
payment, poor households may find it difficult to participate if their opportunity cost of
providing the environmental service is more than the payment they are offered. However, under
a pro-poor approach, a fourth way to target PES contracts is that (iv) parcels of the poorest
households are enrolled first and the payments are scaled according to respective opportunity
cost of the household to adopt the recommended land use change (Gauvin et al., 2010; Jack et
al., 2008).
Often however, PES managers do not have reliable data on specific opportunity costs of
different land parcels and have to depend on average cost estimates to evaluate the outcomes
under different targeting arrangements. Data from field auctions such as in the case of the US

62

Conservation Reserve Program is handy in such cases since it provides a good measure of the
opportunity cost of changing land use over different parcels of land. The auction process in the
Uluguru Mountains is helpful in this regard since it not only provides a good estimate of the
opportunity cost of adopting carbon agroforestry activities, but it also generated information on
the wealth status of various households, which can be used for pro-poor targeting. Another
advantage of this approach is that PES managers and policy makers can ex ante estimate the
efficiency trade-offs from different kinds of targeting. The magnitude of the tradeoff will of
course vary from context to context depending on the extent to which poorer households occupy
environmentally sensitive lands and are also low cost providers (Babcock et al., 1997). In
general, the tradeoff is minimized when there is high positive correlation among the three
variables and is maximized under high negative correlation (Ferraro, 2003).
Following the usual formulation, we can represent ordering or ranking of land parcels as
per ranki = essi / bidi where ‘ranki’ the rank of an individual parcel ‘i’ is the ratio of the
environmental services that the parcel generates ‘essi’ to its opportunity cost or auction bid ‘bidi’
(Stoneham et al., 2003). Under a cost-only approach (i), parcels with the lowest bids ‘bidi’ are
selected first, while in the environment-only approach (ii), parcels with the highest
environmental benefits ‘essi’ are selected first. Recall, however, that in the case of the PES work
in the Uluguru Mountains, each carbon agroforestry contract requires maintaining a standard mix
of 80 trees on parcels of 0.5 acres, which would yield approximately the same amount of carbon
15

service or environmental benefit (essi) from each parcel of land . With no variation in carbon

15

Though the carbon sequestration rates vary with the tree species and the quality of land, such
effects can be ignored for a short duration contract as in the present case.
63

services ‘essi’, the cost-only approach (i) and benefit-cost ratio approach (iii) collapse into the
same thing (which we call here as efficient targeting) where parcels are ranked only using the
inverse of the opportunity cost or bid 1/ bidi . Finally, under the pro-poor approach (iv),
households with the lowest asset ownership ‘assti’, are selected first, and receive payments that
are commensurate to their respective opportunity costs.
Using data from the two auction rounds in the Uluguru Mountains, we explore the tradeoffs among these targeting approaches by constructing Lorenz curves as in figures 2.5 and 2.6.
The horizontal axis represents the cumulative acreage contracted under PES activities while the
vertical axis represents the cumulative budget necessary to pay for these contracts. In both the
figures, the solid line sloping upwards represents the efficient targeting approach, which enrolls
maximum acreage under a given budget. The pro-poor curve (dotted line) represents the acreage
when priority is given to the poorest households (which have the lowest ‘assti’ values). In order
to simulate outcomes when agroforestry adoption also produces a local environmental service
(e.g. slope stabilization on higher elevations) that varies depending on location of land parcels
(average elevation), we draw a third curve (hyphenated) that captures this heterogeneity in
environmental benefits. In the Uluguru Mountains, many government officials and forestry
experts have recommended extension of a line of trees on higher elevations in order to stabilize
mountain slopes (Yanda and Munishi, 2007). In this case, a PES project would select lands that
are located on higher elevation, followed by lands located on the lower reaches of the watershed.
Since we do not have elevation data for each individual parcel, we use the average elevation of
the micro-catchment in which a particular land parcel is located. An added advantage of this
approach is that it reduces the transaction costs associated with monitoring and supervision since
64

all available land in a micro-catchment (with the highest average elevation) is contracted first
before moving to the next micro-catchment (with a lower average elevation).
Figures 2.5 and 2.6 show the tradeoffs in terms of efficiency losses when efficient
targeting is replaced by pro-poor or environment targeting in auction rounds 1 and 2
respectively. In both cases, the curve for environment targeting lies in between the efficient cost
and the pro-poor cost curve, which indicates that enrolling poorer households first is associated
with a higher price compared to cost-only targeting, or even to targeting land parcels in priority
micro-watersheds (environment targeting). Table 2.6 estimates the magnitude of these tradeoffs
under different enrollment targets. Based on auction results from round 1 (where 100%
enrollment corresponds to 125.5 acres), enrolling 25% of the potential land (i.e. 31.375 acres)
would induce an additional cost of TSH 4,589,200 (USD 3,615 under environmental targeting,
and an additional cost of TSH 5,238,800 under pro-poor targeting when compared to the most
efficient cost-only targeting. The corresponding additional cost for 50% enrollment (as marked
in figure 2.5 by dotted lines) are TSH 4,835,900 under environment targeting and TSH
7,874,700 under pro-poor targeting. The corresponding additional cost for a 75% enrollment
target is still higher at TSH 5,551,900 for environment targeting and TSH 8,975,000 under propoor targeting.
In round 2, where a 100% enrollment target corresponds to 123.5 acres, the additional
cost for 25% enrollment targeting (or 30.875 acres) are TSH 4,544,000 under environment
targeting and TSH 4,870,000 under pro-poor targeting (table 2.6). Again as marked in figure 2.6
by dotted lines, for a 50% enrollment target, environment targeting will come at an additional
cost of TSH 4,380,600 and pro-poor at TSH 6,829,600. The corresponding numbers for 75%
enrollment target are TSH 4,714,600 under environment targeting and TSH 8,056,100 under pro-

65

poor targeting. Since farmers who participated in the two auction rounds were randomly selected
from the local area, these numbers can also be extrapolated to estimate the tradeoffs at the level
of the entire local landscape. For example, 251 bidding households in round 1 represent
approximately 20 percent of all households (1227) in the area (table 2.1). Therefore, enrolling
25% of all households across the entire landscape will induce additional cost of TSH 22,946,000
(USD 18,068) under environmental targeting and TSH 26,194,000 under pro-poor targeting.
These estimates explicitly state the magnitude of the tradeoffs involved in diverging from
most efficient targeting towards alternate targeting arrangements such as giving preference to
poorest households first. The objective of this analysis is not to take an ethical stand on whether
or not PES projects should target poorer households first, but rather to make it clear that policy
makers and buyers of environmental services should be prepared to bear additional cost for going
in for these alternate targeting approaches.
Again, while this analysis is useful in estimating the tradeoffs involved in following
alternate targeting approaches, it should be used with caution when estimating specific budgetary
allocations for these different approaches. There are two important reasons for this: (1) the
estimates are based on auction results where bidders were informed that contracts would be
awarded on the basis of the bids (or the opportunity cost of a specific land parcel) alone. If the
bidders knew that their bids would instead be ranked using alternate scaling criteria, such as a
poverty score, there is evidence from existing studies that they would change their bids to
maximize their gains from the scaling criteria (Kirwan et al., 2005). (2) Similarly, the bids were
selected using the uniform pricing rule (where each contracted household receives the lowest
rejected bid) while the trade-off analysis conducted above is based on discriminative payments,
where each household receives a payment equal to its opportunity cost. Again, studies suggest

66

that current bids would not constitute equilibrium bids (and hence could not be used to estimate
the additional budget needed for alternate targeting) if bidders knew that contracts would be
allocated under a discriminatory payment system (Cason and Gangadharan, 2005).
Therefore, in order to use the results of the current auction to estimate the specific budget
requirements for different targeting approaches, we need to proceed with the uniform pricing
arrangement under which all contracted households receive the same payment. For ease of
computation and inference, we consider only the efficient targeting under which land parcels are
contracted according to lowest opportunity cost first, and pro-poor targeting under which land
parcels belonging to the poorest households are contracted first. Figures 2.7 and 2.8 show the
cost of contracting under these targeting arrangements for the two sets of carbon contracts from
round 1 and 2 respectively, while table 2.7 presents the total budgetary requirements under these
two targeting arrangements for different enrollment targets.
The horizontal axis in the two figures (2.7 and 2.8) denotes the acreage enrolled as
before, but note that the vertical axis now represents the bid or the opportunity cost for a
particular land parcel and not the cumulative cost. Along the x-axis the bids of all auction
participants are presented in increasing order of the value of their assets, from the poorest to the
wealthiest. The efficient cost curve (hyphenated line) in both figure 2.7 and 2.8 represents an
ordering of bids with lowest bid first and so on, exactly the supply curve as in figure 2.3. The
pro-poor curve (solid line in both figures 2.7 and 2.8) on the other hand, is constructed by first
selecting the poorest households (with the least asset ownership or ‘assti’) and then plotting their
respective bids in the graph. The spikes in this curve indicate high bidders. As we noted in
section 4.3 above, many poor households are high cost providers and so their bids zig-zag up and
down without following a monotonic order.
67

Under efficient targeting and for a 25% enrollment target, the total budget required for
round 1 contracts is TSH 5,670,000 (USD 4,464.60). However, under a strict pro-poor targeting
where no household is excluded from enrollment, and for a 25% enrollment target (as shown by
the dotted line in figure 2.7), the total payments now rise to TSH 25,515,000. This represents an
additional budget of TSH 19,845,000 (USD 25,203). The reason for this massive increase in
budget is that one household in the bottom 25% of bidders reported a high bid or opportunity
cost of TSH 400,000 and as per the Vickrey rule, it would receive a payment equal to the next
rejected bid of TSH 405,000, but so would all the other contracted households in this group. It
would result in windfall gains for them and a significant increase in the overall budget. The same
trend is repeated for a 50% enrollment target % where the additional budget requirement is TSH
39,690,000 (USD 50,406). For 75% enrollment, it is TSH 48,786,000 (USD 61,958). Table 2.7
also reports respective budget increases for round 2 contracts under strict pro-poor targeting for
enrollment targets of 25% (TSH 19,530,000 or USD 24,803), 50% (TSH 39,060,000 or USD
49,606), and 75% (TSH 53,650,000 or USD 68,136).
This analysis presents a conundrum. On the one hand, using a Vickrey uniform pricing
rule auction format ensures that farmers’ bids represent their true estimates of their opportunity
costs. If, however, these bids are used for alternate targeting of PES contracts such as following a
strict pro-poor approach, and if poor households do not necessarily have low opportunity costs,
then the required PES budget can escalate substantially. In such a scenario, offering lower
payments to households may not be a good strategy because even though the households may
accept these contracts, they may not comply with the project requirements in the long run, thus
risking the sustainability of the project. If on the other hand, farmers are informed during the
bidding stage about the use of a poverty index in ranking their bids, it may present a potential

68

moral dilemma for them since in most developing country settings, rural incomes are selfreported by households. Although discriminatory price auctions can address some of these
concerns, they have yet to be tested in developing country contexts. Finally, one approach that
might work is to ease the strict pro-poor criterion to a more moderate pro-poor stance. Under
such an approach, the wealthiest households may be excluded from PES contracts as ineligible,
but thereafter all the remaining households are contracted as per the cost-efficiency approach.
Although not the first best strategy, this would help keep the project budget low while the
exclusion of the wealthiest households will ensure that at least some of the poorer households are
able to receive PES contracts.
2.6. DISCUSSION AND EXTENSIONS
This paper tests the feasibility of running field auctions in developing country settings to
both allocate PES contracts and estimate the payment that needs to be made to local land
stewards for adopting the recommended land use practices. In the case of the Uluguru Mountains
in Tanzania, we were able to attract bids from 268 local households for participating in
agroforestry activities to produce carbon services. Using the Vickrey uniform pricing format, we
contracted 32 farmers for carbon agroforestry contracts for a period of three years. Auction bids
also helped to estimate a landscape level supply curve which will assist local resource managers
to scale up PES activities across the entire landscape in the area. Using the auction results as well
as the socio-economic data we collected in a survey, we were also able to determine that many
but not all of the poor households were able to participate in the PES activity. Extending PES
contracts to all poor households would escalate overall project costs, making explicit the
tradeoffs involved in diverging from cost-only targeting to a more pro-poor targeting approach.

69

A critical concern underpinning the whole exercise, however, is the extent to which local
farmers were able to estimate their opportunity cost of adopting the new land use practice and
report it in their bids. Recall that out of 268 bids in each round, about six percent of the
observations (or 17 bids) were not included in the analysis because they were either illegible or
too extreme. Subsequent discussions with farmers who made these bids revealed that they had
mistakenly added extra zeroes or put the decimal in the wrong place. Even though we had hired
several educated people to help farmers in preparing and submitting their bids, clearly more
people were needed to help with the auction. Although this may raise the administrative
overheads, there would be several benefits including better quality data in the form of clearly
written auction bids. If, on the other hand, participating farmers underestimated their opportunity
costs, they would of course be unable to comply with the requirements of the contract. Initial
assessments from the field suggest that most contracted farmers did indeed comply with the
contract and duly planted tree seedlings on their farm plots.

16

In this regard, we think that

extensive training before the actual auction and spending enough time in the field to design PES
contracts that are appropriate for a given context are both very useful.
Validity of the auction results is also indicated by the auction bids which follow a normal
distribution and are consistent across the two rounds. If participants did not understand the
auction process well, we would expect the auction bids to be clustered together based on the
small group that a particular respondent was sitting with. However, as we observed in the
discussion on collusion and cooperative bidding, we find that even within small groups, the bids
are well distributed. Auction bids are also within the mean opportunity costs reported by some
other studies in the area with a focus on changing land use from seasonal cropping to more
16

More detailed field assessment of compliance rates are planned for December 2010.
70

permanent vegetation. Finally, participants knew well enough that the auction would be followed
by actual tree planting contracts. Since it was not a hypothetical exercise, they were more likely
to take it seriously and report a bid that was closer to the minimum payment they would need to
adopt the new land use practice.
An important limitation of this work is that due to constraints of time and resources, we
could not collect plot level data which reduced the precision of our estimates from the regression
analysis of auction bids on farm level and household level characteristics. A possible extension
of this work can therefore be a more comprehensive landscape level model that integrates socioeconomic data with more precise plot level biophysical data (Khanna and Ando, 2009; Lynch
and Lovell, 2003). Such a model would also be able to capture the potential threshold effects in a
landscape such that PES managers can quantify the minimum acreage of land that they would
need to enroll in order to produce a viable level of the environmental service. Another potential
benefit of such a model will be to reduce the threat of leakage where adoption of conservation
practices in one part of a landscape or watershed is offset by resource degradation in another
part. By explicitly working out the trade-offs from conservation and resource extraction in
different parts of a watershed, a landscape level model can help in estimating the level of
incentives necessary for local people to conserve the entire landscape. These incentives can be in
the form of an agglomeration bonus for individual landholders to pool their lands or as group
contracts where all individuals stand to gain by cooperating at the group level (Parkhurst and
Shogren, 2007).
Finally, the payment estimates presented here correspond to a very specific contract that
involves planting and protecting a standard mix of 80 trees over three years to produce carbon
sequestration services. It is hard to generalize these estimates for other environmental services or

71

even for additional species of trees that are not included in the present contract. On the other
hand, it would be quite infeasible to conduct repeated auctions in the same area for provision of
different environmental services. One potential solution is to use the auction data to design
screening contracts for provision of other environmental services, particularly the ones that are
strong complements of carbon sequestration from tree planting (e.g. slope stabilization through
tree planting). Screening contracts refer to a set of contracts with varying effort and payment
level such that when offered to potential service providers, they result in self-selection of the low
cost providers from the high cost ones. Although there is a voluminous literature on this subject,
screening contracts have rarely been applied in a field setting (Ferraro, 2008). One strong reason
they have not been used for conservation contracting is paucity of reliable data on distribution of
opportunity costs of different land parcels/service providers. Field auctions such as the one
conducted in the Uluguru Mountains in Tanzania fulfills this important gap and can potentially
be the first step in designing and testing screening contracts in actual field settings for provision
of various environmental services under PES projects. Clearly, there is ample scope for further
investigation on this subject.

72

Appendices

73

Table 2.1: Details of study villages in Kinole catchment, Morogoro district
Name of the
village
1. Tandai
2. Lukenge
3. Doga
4. Nyange
5. Chohola
6. Kalundwa
7. Tonya
8. Kisambwa
9. Jahimbwa
10. Lusegwa
Total/Average

Total
number of
households
375
72
69
123
90
138
96
160
34
70
1227

Number of
households
in study
sample

Distance
Distance
from nearest from nearest
road
market place
(meters)*
(meters)*
100
200
1850
1850
650
660
2300
2350
50
3800
100
3470
50
882
50
2080
300
2480
523
904
597.30
1867.60

Elevation
(meters)*
465
766
530
629
414
465
468
471
469
451
512.80

Note: * Pertains to average distance from the village center or the elevation of the village center

74

Table 2.2: Mean values for households that participated in the field auctions
(n = 251)
Variable
Mean
Male headed HH (0/1)
0.69
Age of the HH head (years)
43
HH Size (number of people)
7
Education of HH head (years
completed)
4
HH head born in the same village
(0/1)
0.79
Location of the HH in main
village (0/1)
0.47
HH reported high discount rate
0.4
(0/1)
Participation in WCST activities
(0/1)
0.7
Farm ownership (number of plots)
5
Farm ownership (area in acres)
8.9
Total Agricultural Expenditure
164,264.5
(TSH)
Expenditure of Hiring Labor
67,322.7
(TSH)
Animal Ownership (Livestock
0.16
Units)
House contains good toilet (0/1)
0.09
Value of Assets owned
(Thousand TSH)
250.60
Head of the HH reported to
survey (0/1)
0.9

Standard Deviation
0.46
14.85
3.1

0

10

0.40

0

1

0.50
0.49

0
0

1
1

0.47
2.6
13.1
292,624.5

0
0
0
0

1
17
177.3
2,426,000

138,937.5

0

1,500,000

0.33

0

2.58

0.28

0

1

1691.44

0

24,260

0.26

75

Maximum
1
90
17

3.4

Source: Author’s survey, 2008-09

Minimum
0
16
1

0

1

Table 2.3: Characteristics and summary statistics of auction results
Auction Details
Nature of contract
Auction
Format
Reservation price
Succeeding rounds
Bids
Number of bids
Minimum bid
Maximum bid
Mean bid
Median bid
Standard deviation
Salient rules
Payment criteria
Tie deciding rule

Round 1
Khaya anthoteca +
Tectona grandis

Round 2
Khaya anthoteca +
Faidherbia albida

Sealed bid second price
No

Sealed bid second price
No
Sequential

251
TSH 1,400
TSH 450,000
TSH 143,840
TSH 130,000
TSH 96,105.5

247
TSH 2,000
TSH 450,000
TSH 138,253
TSH 126,000
TSH 93,105.4

Uniform, lowest rejected bid
Random

Uniform, lowest rejected bid
Random

15
TSH 30,000
7.5 acres
1,200

17
TSH 20,000
8.5 acres
1,360

Auction outcomes
Number of winning bids
Payment per contract
Total Area contracted
Total number of trees
planted

Note: All bidders were eligible to bid in both rounds. Winning bids for each round was
announced only after the completion of both the rounds.

76

Table 2.4: ANOVA results for checking collusion among auction participants seated
together
Number of obs = 232 R-squared = 0.2875
Root MSE
= 85.2614 Adj R-squared = 0.1383
Source | Partial SS df
MS
F Prob > F
-----------+---------------------------------------------------Model | 560187.358 40 14004.6839
1.93 0.0019
|
group | 560187.358 40 14004.6839
1.93 0.0019
|
Residual | 1388476.65 191 7269.51124
-----------+---------------------------------------------------Total | 1948664.01 231 8435.77491

77

Table 2.5: Determinants of Auction Bids

Y = Log(auction bids)

OLS Round 1 Bids

OLS Round 2 Bids

Bid Round (Dummy = 1)
Gender of bidder (Dummy = 1)
0.16 (0.12)
0.26 (0.15)*
Age of bidder
0.009 (0.004)**
0.004 (0.004)
High Time preference (Dummy = 1)
0.12 (0.12)
0.08 (0.12)
Responded to survey (Dummy = 1)
-0.47 (0.19)**
-0.39 (0.18)**
HH Size
-0.05 (0.02)***
-0.04 (0.03)*
Job/Business (Dummy = 1)
0.09 (0.13)
- 0.02 (0.14)
Toilet in the house (Dummy = 1)
-0.18 (0.195)
-0.26(0.24)
Animal ownership (Units)
-0.43 (0.28)*
-0.15 (0.16)
Asset Value
-0.0001 (0.00002)*** -0.0001 (0.00003)***
Number of plots
0.04 (0.02)**
0.04 (0.02)*
Distance to nearest market
0.00 (0.00004)
0.00 (0.00006)
Distance to nearest road
0.004 (0.0002)*
0.00002 (0.0003)
Elevation
-0.004 (0.002)**
-0.0009 (0.002)
Located in main village (Dummy = 1)
0.26 (0.12)**
0.31 (0.15)**
Constant
6.4 (0.81)***
5.1 (1.04)***
N
250
246
Prob > F
0.0001
0.03
R sq.
0.1367
0.0913
Standard errors, robust to system heteroskedasticity in parentheses. ***Significant at 1%

78

Pooled OLS Both
Random Effects
Round Bids
Both Round Bids
0.07 (0.05)
0.06 (0.05)
0.21 (0.12)*
0.24 (0.11)**
0.006 (0.004)*
0.006 (0.004)*
0.1 (0.1)
0.07 (0.10)
-0.43 (0.16)***
-0.42 (0.23)*
-0.04 (0.02)**
-0.04 (0.02)**
0.04 (0.12)
0.02 (0.12)
-0.23 (0.19)
-0.23 (0.19)
-0.29 (0.20)*
-0.27 (0.16)*
-0.0001 (0.00002)*** -0.0001 (0.00004)***
0.04 (0.02)**
0.03 (0.02)
0.00 (0.00004)
0.0002 (0.00004)
0.0002 (0.0003)
0.0002 (0.0002)
-0.003 (0.002)
-0.002 (0.002)*
0.28 (0.12)***
0.27 (0.13)**
5.7 (0.85)***
5.7 (0.70)***
496
496
0.002
0.007
0.1005
0.099
**Significant at 5% *Significant at 10%

Table 2.6: Trade-offs from different targeting approaches

Auction
Round

Targeting
Approaches
Efficient

Cost of contracting (in Thousand TSH) under
different enrollment targets
25%
50%
75%
2802.7
9287.8
18351.8

Round 1*
Environment
Pro-poor
Efficient

7391.9
(4589.2)
8041.5
(5238.8)

14123.7
(4835.9)
17162.5
(7874.7)

23903.7
(5551.9)
27326.8
(8975.0)

2593.5

8963.5

17713.5

7137.5
(4544.0)
7463.5
(4870.0)

13344.1
(4380.6)
15793.1
(6829.6)

22428.1
(4714.6)
25769.6
(8056.1)

Round 2**
Environment
Pro-poor

Note: Figures in parentheses represent loss in efficiency with respect to cost of enrollment under
efficient targeting.
* In Round 1, there were a total of 251 valid bids, each corresponding to 0.5 acres. The total
acres that could potentially be contracted was 125.5 acres. Therefore, 25% enrollment target
corresponded to 31.375 acres, 50% to 62.75 acres, and 75% to 94.125 acres respectively.
** In Round 2, there were a total of 247 valid bids. So the total number of acres that could
potentially be contracted was 123.5 acres. Corresponding acreage for 25%, 50%, and 75% was
30.875 acres, 61.75 acres, and 92.675 acres respectively.

79

Table 2.7: Budgetary allocation under uniform pricing and alternate targeting

Auction
Round

Targeting
Approaches
Efficient

Cost of contracting (in Thousand TSH) under
different enrollment targets
25%
50%
75%
5670
17010
35814

Round 1*
Pro-poor

25515
(19845)

56700
(39690)

84600
(48786)

Efficient

5580

16740

29600

Pro-poor

25110
(19530)

55800
(39060)

83250
(53650)

Round 2**

Note: Figures in parentheses represent additional budget needed with respect to cost of
enrollment under efficient targeting.
* In Round 1, there were a total of 251 valid bids, each corresponding to 0.5 acres. The total
acres that could potentially be contracted was 125.5 acres. Therefore, 25% enrollment target
corresponded to 31.375 acres, 50% to 62.75 acres, and 75% to 94.125 acres respectively.
** In Round 2, there were a total of 247 valid bids. So the total number of acres that could
potentially be contracted was 123.5 acres. Corresponding acreage for 25%, 50%, and 75% was
30.875 acres, 61.75 acres, and 92.675 acres respectively.

80

Figure 2.1: Checking for potential collusion among auction participants seated together
350

Mean Bid (in thousand TSH)

300
250
200

Group Mean
Overall mean

150

Mean‐SD

100

Mean+SD
50
0
0

10

20

30

Groups of people seated together at the auction

81

40

Figure 2.2: Box Plots of auction bids received from groups of participants seated together

500

400

Bids

300

200

100

0
0

2

4

6

8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40

Group

82

Bids from local landstewards (in thousand
TSH)

Figure 2.3: Estimated supply curve for enrolling private land for tree planting

500
450
400
350
300
250

Round 1

200

Round 2

150
100
50
0
0

50

100

Private land enrolled under tree planting carbon contracts (in acres)

83

Figure 2.4: Relationship between the wealth status and the auction bids

Bids in Round 2 (in thousand TSH)

500

400

300

200

100

0
0

50

100

150

200

250

300

350

Asset Value per HH (in thousand of TSH) 

84

400

450

Cummulative Bids (Thousand TSH)

Figure 2.5: Trade-offs in alternate targeting of carbon contracts (Round 1 bids)
40000
35000
30000
25000

Efficient Targeting

20000
Pro-poor Targeting

15000

Environment
Trageting

10000
5000
0
0.5

25.5

50.5

75.5

100.5 125.5

Land Enrolled (Acres)

85

Cummulative Bids (Thousand TSH)

Figure 2.6: Trade-offs in alternate targeting of carbon contracts (Round 2 bids)
40000
35000
30000
25000
20000

Efficient Trageting

15000

Pro-poor Targeting
Environment Targeting

10000
5000
0
0.5

25.5

50.5

75.5

100.5

Land Enrolled (Acres)

86

Figure 2.7: Budgetary allocation under uniform pricing and alternate PES targeting
(Round 1 bids)

Cost of Enrollment (Thousand TSH)

500
450
400
350
300
Efficient
Targeting

250
200

Pro-poor
Targeting

150
100
50
0
0.5

25.5

50.5

75.5

Land Enrolled (in Acres)

87

100.5

125.5

Figure 2.8: Budgetary allocation under uniform pricing and alternate PES targeting
(Round 2 bids)

Cost of Enrollment (Thousand TSH)

500
450
400
350
300
250
Efficient
Targeting

200
150
100
50
0
0.5

25.5

50.5

75.5

Land Enrolled (in Acres)

88

100.5

References

89

REFERENCES

Antle, John M., and J.J. Stoorvogel. 2006. Predicting the supply of ecosystem services from
agriculture. American Journal of Agricultural Economics. 88: 1174-1180
Athey, S., J. Levin, and E. Seira. (2004). Comparing Open and Sealed Bid Auctions: Theory and
Evidence from Timber Auctions. Mimeo. Stanford University.
Babcock BA, Lakshminarayan PG, Wu J, and Zilberman D. 1997. Targeting tools for the
purchase of environmental amenities. Land Economics. 73:325–339
Babcock BA, Lakshminarayan PG, Wu J, Zilberman D. 1996. The economics of a public fund
for environmental amenities: a study of CRP contracts. American Journal of Agricultural
Economics. 78:961–971
Burgess, N., N. Doggart., J.C. Lovett. 2002. The Uluguru Mountains of eastern Tanzania: the
effect of forest loss on biodiversity. Oryx. Vol 36, No. 2. pp. 140-152.
Cason, T.N., and L. Gangadharan. 2005. A Laboratory Comparison of Uniform and
Discriminative Price Auctions for Reducing Non-point Source Pollution. Land Economics.
81 (1): 51–70
Cason, T.N., and L. Gangadharan. 2004. Auction Design for Voluntary Conservation Programs.
American Journal of Agricultural Economics 86 (Number 5, 2004): 1211–1217
Claassen, R., Cattaneo, A., and Johansson, R., 2008. Cost-effective design of agri-environmental
payment programs: US experience in theory and practice. Ecological Economics, 65 (4),
737–752.
Cummings, R.G., C.A. Holt, and S.K. Laury. 2004. Using laboratory experiments for policy
making: An example from the Georgia irrigation reduction project. Journal of Policy
Analysis and Management. Vol. 23(2), Pp. 341-363.
Engel, Stefanie, Stefano Pagiola, and Sven Wunder. 2008. Designing payments for
environmental services in theory and practice: An overview of the issues. Ecological
Economics. 65: 663-674
Ferraro, P. J. 2008. Asymmetric information and contract design for payments for environmental
services. Ecological Economics. 65:810– 821.
Ferraro, P. J. 2003. Conservation contracting in heterogeneous landscapes: an application to
watershed protection with threshold constraints. Agricultural and Resource Economics
Review. 32:53–64.

90

Ferraro, Paul J., and A. Kiss. 2002. Direct Payments to Conserve Biodiversity. Science. 29
November 2002. Pp 1718-1719.
Gauvin, C., E. Uchida, S. Rozelle, J. Xu., and J. Zhan. 2010. Cost-Effectiveness of Payments for
Ecosystem Services with Dual Goals of Environment and Poverty Alleviation.
Environmental Management. (2010) 45:488–501
Giampietro, C., and M. L. Emiliani. 2007. Coercion and reverse auctions. Supply Chain
Management: An International Journal. 12/2 (2007) 75-84
Gong, Y., G. Bull, and K. Baylis. 2010. Participation in the world's first clean development
mechanism forest project: the role of property rights, social capital and contractual rules.
Ecological Economics 69 (6), 1292–1302
Godziszewski, P. (2009). Estimating potential income from carbon sequestration payments
compared to current revenue from farming in context of planned reforestation in Uluguru
Mountains in Tanzania. Master’s Thesis. University of Aberdeen, UK.
Gren, Ing-Marie. 2004. Uniform or discriminating payments for environmental production on
arable land under asymmetric information. European Review of Agricultural Economics.
Vol. 31(1) Pp 61-76
Haites, E. 2004. Rewarding sinks projects under the CDM. Environmental Finance, March.
Hamilton, Katherine , Unna Chokkalingam, and Maria Bendana. 2010. State of the Forest
Carbon Market 2009: Taking Root and Branching Out. Forest Trends, Washington DC
Huang, M, S. Upadhyaya, R. Jindal, and J. Kerr. 2009. Payment for watershed services in Asia: a
review of current initiatives. Journal of Sustainable Forestry. Vol. 28 Pp 551-575
ILCA, 1990. Livestock systems research manual. Working Paper 1, Vol.1. International
Livestock Centre for Africa (ILCA), Addis Ababa.
Jack, Kelsey. (2010) Allocation in environmental markets: A field experiment with tree planting
contracts. Discussion Paper 2009-14, Cambridge, Mass.: Harvard Environmental
Economics Program, May, 2010.
Jack, B. K., Leimona, B. & Ferraro, P. J. 2008. A revealed preference approach to estimating
supply curves for ecosystem services: Use of auctions to set payments for soil erosion
control in Indonesia. Conservation Biology 23(2), 358 - 367
Jindal, R. 2010. Livelihood impacts of payments for forestry carbon services: Field evidence
from Mozambique in Tacconi L., Mahanty S., Suich H. (eds.) Livelihoods in the REDD?:
Payments for Environmental Services, Forest Conservation and Climate Change. Edward
Elgar, Cheltenham (forthcoming).

91

Jindal, R., B. Swallow, and J. Kerr. 2008. Forestry-based carbon sequestration projects in Africa:
Potential benefits and challenges. Natural Resources Forum. 32 pp 116–130.
Katoomba Group. 2007. East and Southern Africa Katoomba Group Newsletter,
http://www.katoombagroup.org/~katoomba/documents/newsletters/Vol1_No3.html#anchor
2, accessed on 27 August 2010.
Kenney, Alice. 2009. Water trading: Costa Ricans Keep it Simple. Ecosystem Marketplace.
http://www.ecosystemmarketplace.com/pages/dynamic/article.page.php?page_id=6663&se
ction=home. Accessed November 20, 2010.
Khanna, M., and A. Ando. 2009. Science, economics and the design of agricultural conservation
programmes in the US. Journal of Environmental Planning and Management.52:5,575-592
Khanna, M., M. Isik, and D. Zilberman. 2002. Cost-effectiveness of alternative green payment
policies for conservation technology adoption with heterogeneous land quality.
Agricultural Economics. 27 (2002) 157-174
Kirwan, B., R. N. Lubowski, and M.J. Roberts. 2005. How cost-effective are land retirement
auctions? Estimating the difference between payments and willingness to accept in the
Conservation Reserve Program. American Journal of Agricultural Economics. 87 (Number
5, 2005): 1239–1247
Klemperer, Paul D. 2002a. What Really Matters in Auction Design. The Journal of Economic
Perspectives, Vol. 16, No. 1. (Winter, 2002), pp. 169-189.
Klemperer, Paul D. 2002b. How (Not) to Run Auctions: The European 3G Telecom Auctions.
European Economic Review.
Kosoy, N., and E. Corbera. 2010. Payments for ecosystem services as commodity fetishism.
Ecological Economics. 69 (2010) 1228–1236
Krishna, V. 2002. Auction Theory. Academic Press, London.
Landell-Mills, N., and I.T. Porras, 2002. Silver bullet or fool’s gold? A global review of markets
for forest environmental services and their impact on the poor. International Institute for
Environment and Development (IIED), London, UK.
Latacz-Lohmann, U., and S. Schilizzi. 2005. Auctions for Conservation Contracts: A Review of
the Theoretical & Empirical Literature. Report to the Scottish Executive Environment and
Rural Affairs Department, UK.
Latacz-Lohmann, U., and Carel Van der Hamsvoort. 1997. Auctioning Conservation Contracts:
A Theoretical Analysis and an Application. American Journal of Agricultural Economics.
Vol. 79, No. 2, (May, 1997), pp. 407-418

92

Lucking-Reiley, D., D. Bryan, N. Prasad, and D. Reeves. 2007. Pennies from eBay: The
Determinants of Price in Online Auctions. The Journal of Industrial Economics. (June
2007) No. 2
Lynch, L., and S.J. Lovell. 2003. Combining Spatial and Survey Data to Explain Participation in
Agricultural Land Preservation Programs. Land Economics. 79 (2): 259–276
Milgrom, P. 1989. Auctions and Bidding: A Primer. The Journal of Economic Perspectives. Vol.
3 (3) (1989), pp. 3-22
Miles, L. and V. Kapos. 2008. Reducing greenhouse gas emissions from deforestation and forest
degradation: global land-use implications. Science 320:1454-1455.
Miller, T., and G. Tolley. 1989. Technology Adoption and Agricultural Price Policy. American
Journal of Agricultural Economics. Vol. 71, No. 4 (Nov., 1989), pp. 847-857
Miranda, M., I.T. Porras, and M.L. Moreno. 2003. The Social Impacts of Payments for
Environmental Services in Costa Rica: A quantitative field survey and analysis of the
Virilla Watershed. International Institute for Environment and Development (IIED), UK.
Myerson, R. B. 1981. Optimal Auction Design. Mathematics of Operations Research. 6, 58-73.
Neufeldt H, Wilkes A, Zomer RJ, Xu J, Nang’ole E, Munster C, Place F. 2009. Trees on farms:
Tackling the triple challenges of mitigation, adaptation and food security. World
Agroforestry Centre Policy Brief 07. World Agroforestry Centre, Nairobi, Kenya.
Okorio, J., and J.A. Maghembe. 1994. The growth and yield of Acacia albida intercropped with
maize (Zea mays) and beans (Phaseolus vulgaris) at Morogoro, Tanzania. Forest Ecology
and Management. 64 (1994) 183-190.
Paarsch H. J., Hong H. 2006. An Introduction to the Structural Econometrics of Auction Data.
MIT Press: Cambridge, MA.
Pagiola, S., A.R. Rios, and A. Arcenas. 2008. Can the poor participate in payments for
environmental services? Lessons from the Silvopastoral Project in Nicaragua. Environment
and Development Economics. Vol. 13 Pp 299–325
Pagiola, S., and Platais, G. 2007. Payments for Environmental Services: From Theory to
Practice. World Bank, Washington.
Parkhurst, G.M., & J. F. Shogren.2007. Spatial incentives to coordinate contiguous habitat.
Ecological Economics. Vol 64.
Platinga, A.J., R. Alig, and H. Cheng. 2001. The supply of land for conservation uses: Evidence
from the Conservation Reserve Program. Resources, Conservation and Recycling.31(2001)
Pp 199–215
93

Polhill, R.M. 1968. Tanzania. In Conservation of Vegetation in Africa South of Sahara. Eds. I.
Hedberg and O. Hedberg. Acta Phytogeographica Suecica, 54. pp. 166-78.
Rothkopf, Michael H., Thomas J. Teisberg, and Edward P. Kahn. 1990. Why Are Vickrey
Auctions Rare? The Journal of Political Economy. Vol. 98, No. 1 (Feb., 1990), pp. 94-109.
Rousseau, S., and E. Moons. 2008. The potential of auctioning contracts for conservation policy.
European Journal of Forest Resources. (2008) 127:183–194
Southgate, D., and S. Wunder. 2009. Paying for Watershed Services in Latin America: A Review
of Current Initiatives. Journal of Sustainable Forestry. Vol 28. Pp 497 – 524
Stoneham, G., Chaudhri, V., Ha, A. & Strappazzon, L. 2003. Auctions for conservation
contracts: An empirical examination of Victoria’s Bushtender trial. Australian Journal of
Agricultural and Resource Economics 47(4), 477-500.
Sun, Chia-Hung, and Ming-Fei Hsu. (2007). The Determinants of Price in Online Auctions:
More Evidence from Quantile Regression. WP 07-18. University of Wollongong
Economics Working Paper Series 2007.
Tanzania Forestry Research Institute (TAFORI). (2006). Popular Version of the National
Agroforestry Strategy. National Agroforestry Steering Committee. Morogoro, Tanzania.
Uchida, E., J. Xu, Z. Xu, Z., and S. Rozelle. 2007. Are the poor benefiting from China’s land
conservation program? Environment and Development Economics, Vol.12, Pg. 593–620.
Van Noordwijk M, Minang PA, Dewi S, Hall J, Rantala S. 2009. Reducing emissions from all
land uses (REALU): The case for a whole landscape approach. ASB Policy Brief 13. ASB
Partnership for the Tropical Forest Margins, Nairobi, Kenya.
Vickrey, W. 1961. Counter speculation, auctions, and competitive sealed tenders. Journal of
Finance. pp. 8.37.
Wooldridge, J.M. 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press,
Cambridge.
World Bank. 2010. Tanzania at a Glance. http://devdata.worldbank.org/AAG/tza_aag.pdf
(Accessed on November 22, 2010).
Wu, J.J., and B.A. Babcock. 1996. Contract Design for the Purchase of Environmental Goods
from Agriculture. American Journal of Agricultural Economics. Vol. 78, No.4 Pp. 935-945
Wunder, S., 2005. Payments for Environmental Services: Some Nuts and Bolts. CIFOR
Occasional Paper No. 42. Center for International Forestry Research, Bogor.

94

Yanda, P.Z., and P.K.T. Munishi. (2007). Hydrologic and land use/cover change analysis for the
Ruvu River (Uluguru) and Sigi River (East Usambara) Watersheds. Sokoine University of
Agriculture, Morogoro, Tanzania. Report for WWF/CARE. April 2007
Yin, Runsheng, and Guiping Yin. 2010. China’s Primary Programs of Terrestrial Ecosystem
Restoration: Initiation, Implementation, and Challenges. Environmental Management.
45:429–441

95

CHAPTER 3: REDUCING POVERTY THROUGH CARBON FORESTRY?
EXPLORING IMPACTS OF THE N’HAMBITA COMMUNITY CARBON PROJECT IN
MOZAMBIQUE

3.1

INTRODUCTION

For all the discussion of local benefits of community based carbon mitigation projects,
literature on actual measurement of such benefits is scarce (Engel et al., 2008; Uchida et al.,
2007). Most existing studies either present anecdotal evidence of project impacts or at best, an
estimate of potential benefits (e.g. Tipper, 2002; Aune et al., 2005). Even though some studies
have looked at impacts of large-scale national afforestation programs that are linked to carbon
mitigation activities, such as the Sloping Land Conservation Program (SLCP) and the Natural
Forest Protection Program (NFPP) in China (e.g. Liu et al., 2010; Uchida et al., 2007), empirical
evidence of welfare effects of small-scale carbon forestry projects remains sketchy.
This gap in the literature is disconcerting for several reasons. Carbon mitigation,
particularly through community forestry, follows the payments for environmental services (PES)
approach whereby service providers receive payments for their conservation efforts, in this case
through planting new forests and protecting existing ones (FAO, 2009; Jindal et al., 2007) Since
areas with potential to provide forest based carbon services also coincide with existence of largescale poverty, there is an expectation among policy circles that using PES mechanisms to support
carbon mitigation will lead to poverty alleviation (Perez et al., 2007; UNEP, 2002). As a result,
the number of forest conservation based carbon mitigation projects in developing countries has
grown steadily, with 19 reported from Africa alone (Hamilton et al., 2010; Jindal et al., 2008).
However, one cannot assume that these projects will definitely benefit the participating

96

households, or would even be accessible to the local poor. For instance, projects that involve
large-scale plantations can potentially have adverse outcomes for the poor, while in others the
local poor may be unable to participate due to insufficient resources (Eraker, 2000; Corbera et
al., 2007). Lack of information on poverty impacts of these carbon mitigation projects is thus a
legitimate concern.
Secondly, not all developing countries can afford large PES-like programs such as
China’s SLCP or South Africa’s Working for Water, which are funded from national budgets.
Instead, many poor countries view international carbon markets as a source of funds to support
their own forest conservation and poverty alleviation programs (Gutman, 2003). However,
without access to quality data on existing projects, national policy makers cannot design and
replicate appropriate projects in their own countries.
In addition, many PES projects either do not fulfill the strict notion of conditionality or
may have an add-on development component that is not linked directly to the provision of an
environmental service (Engel et al., 2008). While it is important to get a more realistic estimate
of the impact of the PES component by differentiating it from the developmental activities going
on in the area as part of a bigger project, most PES studies do not make this distinction. The
resultant estimates may thus be biased and will not reflect the true impact of the PES component.
Finally, concerns regarding impermanence and leakage of emission reduction from forestry have
been widely expressed in environmental literature (Alix-Garcia et al., 2010), but documentation
on actual experience in the field is limited. Field research on these issues can provide valuable
insights on how to address them in future projects.
This paper attempts to fill some of these gaps through a detailed investigation of the
N’hambita Community Carbon Project that pays local farmers in a remote part of Mozambique

97

for carbon mitigation services through on-farm agroforestry and avoided deforestation activities.
The paper addresses three gaps in the literature introduced above: (i) it estimates local impacts of
a small, community based project that makes conditional payments for carbon services, (ii) it
differentiates the impact of these payments from other development activities that are also part of
the project, and (iii) it assesses leakage and permanence of the carbon mitigation activities that
the project carries out.
The main finding from this review is that even though the project has been beneficial for
the community, carbon payments alone have been insufficient to move people out of poverty. On
the other hand, households employed in various microenterprises promoted by the project have
certainly become better off than other households, both those that participate in carbon
sequestration activities and those that do not. Upfront payments and flexibility regarding
adoption of agroforestry practices have been effective in creating a huge demand for these
carbon mitigation activities leading to a widespread diffusion of agroforestry in the area.
However, they also expose farmers to long-term contractual obligations and exacerbate the risk
of impermanence and leakage of emission reduction for which they are being paid. We review
these impacts in detail, starting with a description of the project.
3.2.

THE N’HAMBITA COMMUNITY CARBON PROJECT

The N’hambita Community Carbon Project is located along the periphery of the
Gorongosa National Park (GNP) in Sofala province of Mozambique. The local community was
relocated to this area after the creation of the park in 1948. However, the relative peace in the
area was disrupted when it became one of the focal points of armed struggle during
Mozambique’s long civil war from 1975-92. As a result, thousands of local families were
displaced to urban centers and lost access to their farmlands (Howell and Convery, 1997). With

98

the advent of peace in 1992, many families started returning to their villages. However, they had
few avenues for a decent livelihood and the poverty rates in the area remained among the highest
in the country with more than 88% of the population living below the poverty line (Simler et al.,
2004).
The N’hambita project was initiated to provide an alternate source of income to these
people through improved forest-based land use practices that could also produce carbon
reduction services for sale in international markets (UOE, 2002). The project began its field
activities in 2003, making it one of the earliest forest based carbon mitigation projects in
Mozambique. It was initially funded by the Eurpoean Union, but since the end of the pilot phase
in 2008 it has operated on the revenue from sale of carbon offsets (each offset equals one ton of
carbon dioxide or tCO2 sequestered by the project) to international buyers such as the MAN
17

group and the CarbonNeutral Company .
The project operates in all six villages of the Chicale regulado (a traditional
administration unit measuring about 20,000 hectares managed by a local chief or regulo). Five of
the six villages are located deep inside the forest, with the nearest paved road several kilometers
away (table 3.1). None of the villages is electrified and there is only one primary health centre
for the entire area, located in Pungue. N’hambita, Mbulawa and Pungue had small grocery shops
in the village, but the main market for the area is in Gorongosa, about 60 km away. Although all
villages except Bue Maria have a primary school in the village, the nearest secondary school is in
Gorongosa. According to Hegde and Bull (2008), there are a total of 1026 households in the
area, with Mbulawa being the most populated with 414 households.

17

Project details can also be found at www.miombo.org.uk.
99

The project carries out various activities, some of which produce carbon offsets while
others are more development oriented. For the purpose of this review, we have categorized the
main activities of the project into the PES component and the development component, based on
whether or not the benefits that local people receive are conditional on adoption of conservation
practices that yield carbon offsets (table 3.2).
(I)

PES Component
The PES component relates to activities that produce carbon offsets. The project offers

regular payments to the local community with the amount determined by the nature of the
activity and the number of carbon offsets its yields. This includes carbon sequestration through
adoption of agroforestry on private farms and reduced emissions from deforestation and
degradation (REDD) activities on community owned woodlands.
(i)

Carbon sequestration through agroforestry
The project invites local households to plant trees on their farms that result in

sequestration of atmospheric carbon into woody biomass. In return the households receive an
annual payment for seven years, depending on the amount of carbon sequestered on their farms.
The households can choose from a menu of agroforestry systems that the project offers: planting
mango (Mangifera indica) and cashew (Anacardium occidentale) orchards, setting up woodlots
with siris (Albizia lebbeck) and African mahogany (Khaya nyasica), intercropping with acacia
(Faidherbia albida), planting native hardwoods such as panga panga (Millettia stuhlmannii)
along the boundary of the m’shambas (farmers’ fields), or planting fruit trees such as tamarind
(Tamarindus indica) within the homestead. The specific carbon sequestration rate of different
systems depends on density of plantations and the choice of tree species (table 3.3) and ranges

100

from 10 to 181 tCO2/ha. These sequestration rates have been estimated on the basis of
maintaining trees for a period of 100 years (Tipper, 2008).
Each agroforestry system is designated as a separate contract and generally covers about
0.25-1.5 ha of a household’s field or m’shamba land. A household can enroll for multiple
contracts, either by adopting the same agroforestry activity on multiple plots or by taking up
different agroforestry activities on the same plot (e.g. combining boundary planting with fruit
orchards). In return, the N’hambita project monetizes carbon offsets generated by the new
agroforestry systems with an annual cash payment to the contracted household. It is important to
note that while carbon offsets are generated over 100 years, farmers are paid the entire value of
these offsets during the first seven years of the contract. For instance, intercropping with
Faidherbia albida generates a total of MTN 20,187/ha ($807.50) as carbon payments over the
first seven years.

18

An important concern regarding project’s sustainability is whether the newly

established trees will continue to be protected once cash payments end. We return to this issue in
more detail below.
(ii)

Reduced emissions from deforestation and degradation (REDD)
In addition to carbon sequestration, the local community also receives carbon payments

for REDD activities in communally owned miombo woodlands around the Gorongosa National
Park. These are open canopy dry deciduous forests dominated by trees such as Brachystegia,
Julbernardia and Pterocarpus. Major drivers of deforestation in these forests include clearance
for agriculture, tree felling for charcoal production, uncontrolled burning, and logging for timber
(Herd, 2007). Remote sensing images show that between 1999 and 2007, the average rate of

18

In September 2008, $1 = MTN 24.23
101

deforestation in the area was 2.4% per annum (Tipper, 2008). This would denude the entire
forest by 2040 unless conservation initiatives are effective in reversing the deforestation.
In order to reduce this deforestation, the N’hambita project has introduced two main
activities in the area: a total ban on tree felling, and formation of fire patrols that guard against
fire outbreak in the forests. These activities began in 2006 in a selected block of 5,000 ha and
since then have expanded to 11,071 ha. The project team estimates that over ten years, these
REDD activities will generate carbon offsets of 73.3 tCO2/ha, based on a net reduction in
biomass loss in the area (Tipper, 2008).
The N’hambita project combines these REDD offsets with carbon sequestration offsets
from agroforestry activities and sells them as one lot in international voluntary carbon markets.
After deducting overheads and commissions payable to brokers, the project has thus far paid
farmers an average of $4.50 per tCO2, which is much higher than the average price in most
voluntary carbon markets. For instance, in 2007, the average price of carbon in Chicago Climate
Exchange was only $3.13 per tCO2 and since then it has fallen below $1 per t CO2, while the
N’hambita project has continued to pay the much higher price of $4.50 per tCO2 (CCX, 2010;
Capoor and Ambrosi, 2009). This indicates the success with which the project is able to
negotiate deals with high profile buyers who are willing to pay this premium on carbon offsets.
However, this also results in significant transaction and administrative costs (such as high
brokerage fees) which account for almost 60% of the entire project cost. During the pilot phase
from 2003-08, most of these costs were supported by the European Union. Since then the project
meets them through the carbon revenue it generates. In order to scale up the number of offsets it
sells in international markets, it has increased the number of contracted farmers in the Chicale
102

regulado an expanded to newer regions (such as Zambézia and Quirimbas provinces in northern
Mozambique). In the last few years the project has sold 116,807 tCO2 worth more than
$900,000, an important achievement considering that these offsets have been mainly sold in the
voluntary market, which absorbs a smaller volume of carbon offsets than the Kyoto market.
Grace (2008) estimates that by 2007, the project had paid a total of $223,750 to the local
community for agroforestry and REDD activities, with more payments expected shortly.
However, the actual flow of money into the community varies by activity. In case of carbon
sequestration, most of the money is paid directly to individual contract holders, while a small
proportion goes into a community trust fund. REDD payments, on the other hand, are divided
into two; one-half is deposited into the community trust fund and the other paid as wages to
people who patrol the forest block against fire outbreak. Figure 1 shows the project’s
components.
(II)

Development component
The development component consists of various microenterprises that the project has

promoted in the area to provide alternate livelihoods. In addition, most of the project staff has
been hired from the local community and since their monthly salary contributes significant cash
inflows for the community, we also categorize this as part of the development component.
(i)

Promotion of microenterprises
To promote alternate livelihoods and improve local incomes, the project supports several

microenterprises (MEs) including nurseries, a community saw mill, a carpentry shop, beekeeping, and a vegetable garden. Employment in these MEs is not conditional on participation in
carbon mitigation activities though most people who are employed in MEs have also enrolled for
agroforestry contracts. While the MEs were initially supported from donor funding, the project
103

team is now working towards making them self-sustaining by linking them with the local market.
For instance, the carpentry shop received a large order to supply furniture to the tourist lodge in
Gorongosa National Park, while the nursery is trying to obtain a seedling contract from other
development agencies in the area.
(ii)

Project staff
A unique aspect of the project is that most of the staff is located on site and is drawn from

the local community. This includes agroforestry extension workers, administrative staff, drivers
and mechanics for project vehicles, and other casual staff. Like the ME staff, the project staff
receives a regular monthly wage that translates into additional cash inflow into the community.
In all, MEs and the project employ about 170 people with wages ranging from MTN 1,200 per
month ($49.50) for a forest nursery worker to MTN 15,000 per month ($619.10) for a senior
extension worker. In the following discussion of project impacts, we try to differentiate the
impact of these wages from that of the carbon payments that the local community receives.
3.3. DATA AND METHODS
PES projects differ from integrated conservation development projects in that payments
are directed only at those who can provide an environmental service, conditional on their
actually doing so (Wunder, 2005; Ferraro and Kiss, 2002). In this regard, Greig-Gran et al.
(2005) provide a useful framework that not only looks at impacts of PES on project participants
but also reviews the extent to which local poor people are indeed able to participate in such
projects. Further, Pagiola et al. (2005) suggest that impacts should not be limited to income alone
but should also include other non-income impacts such as changes in access to wage labor or
other employment opportunities in the area. We find both these approaches useful and
complementary. Therefore, we combine the two approaches to focus on five key themes with

104

respect to the N’hambita project: (1) the extent to which poor people participate in the project,
(2) impacts on project participants’ livelihoods, (3) impacts on non-participants, (4) wider
spillover effects in the community, and (5) environmental impacts. For livelihood impacts, we
considered indicators such as household income, asset ownership, livestock ownership,
education attainment, and access to wage labor. Relevant indicators to measure the
environmental impacts were the number of carbon offsets produced by the project as well as
leakage and threats to permanence of these offsets.
There were two main challenges in conducting an objective assessment of the project
impacts: isolating project impacts from wider changes in the region due to impressive growth in
the country’s overall economy, and differentiating the impact of the PES component (carbon
payments) from the development component (wages from employment in various
microenterprises). In order to address these challenges, our main estimation strategy was to
follow stratified random sampling whereby we distributed the local households into three
categories: households that participate in both agroforestry activities and MEs, households that
participate in only agroforestry activities, and households that participate neither in agroforestry
activities nor in MEs. Ideally, we would have liked to form a fourth category of households that
participate in only MEs but not agroforestry. However, we were unable to find such households
as everyone who was employed in any ME also possessed an agroforestry contract.
Continuing with the household as the unit of analysis, we compared the before-project
status of households in each of these categories with the after-project status. Assuming that
changes in the wider economy would have had similar effects on all households in the area, we
were thus able to differentiate the impacts of the project from those of macro changes as well as
impacts of the PES component from those of the development component of the project. Further,

105

to check for any systematic differences or potential selection bias across households that either
participated in the project (agroforestry contracts, employment in ME) or those that did not, we
conducted a before-project cross-sectional analysis of sampled households. Finally, in order to
estimate the spillover effects of the project on the entire community (impact of improvements in
educational infrastructure on community wide literacy rate) we compared averages across all
sampled households in the six project villages with randomly sampled households from six
neighboring villages outside the project area.
The main data for this review were collected through a household survey conducted in
May 2008. Based on the census compiled by Hegde and Bull (2008), we divided the local
population in the project villages into three strata and then randomly sampled about 25% of
households from each stratum: (i) households with both agroforestry contracts and with at least
one member employed in an ME (n=54), (ii) households with only agroforestry contracts
(n=170), and (iii) non-participating households that had neither agroforestry contracts nor
employment in MEs (n=46). We also conducted the same survey among 64 randomly selected
households in the control villages. The total sample size for the household survey is 334.
The survey focused on important demographic and socio-economic variables related to
project impact. There were two main sections in the survey questionnaire. Section one focused
on the status of a household in 2008 (or the after-project scenario), while section two collected
information from the same household for 2001 (before-project scenario) using the recall method.
Recall has been extensively used by researchers to study the impacts of an external intervention
(e.g. Uchida et al., 2007; Mullan et al., 2010). In order to minimize errors associated with recall,
we selected 2001 as the reference year since it was vivid in peoples’ memory as the year when
the local river had last flooded. Both section one (after-project) and section two (before-project)

106

covered the same set of variables, i.e. the number of people in the household, their educational
status, agricultural profile of the household including livestock ownership, extent of participation
in the N’hambita project, and assets ownership. It is also important to note that the year 2008 did
not imply the end of the project. Like most forest based carbon mitigation projects, the
N’hambita project is expected to run for many more years. Rather, it corresponded to completion
of the first five years of implementation and the conclusion of the donor-funded phase.
Finally, to triangulate survey data and add depth to our analysis, we also collected
qualitative data through semi-structured discussions with an additional set of respondents in the
project villages (Chung, 2000): (i) respondents employed in various MEs and also contracted
under agroforestry systems (group size 25), (ii) women, most of whom had only agroforestry
contracts (group size 25), (iii) respondents who had neither agroforestry contracts nor
employment in MEs (group size 14), (iv) new immigrants to the area, most of whom did not
have carbon contracts (group size 24), and (v) members of the community association (group
size 11). These discussions helped to understand people’s perceptions of the project and enabled
a richer interpretation of survey findings.
3.4 RESULTS: PARTICIPATION OF THE POOR
Although many PES projects have a pro-poor focus, there are concerns regarding the
extent to which local poor people and smallholders are actually able to participate in them (Gong
et al., 2010; Pagiola et al., 2008; and Uchida et al., 2007). In Costa Rica’s national PES program
for instance, Miranda et al. (2003) found that in their study area, most of the participants were
relatively well off landowners. In general, poor households may be unable to participate in PES
projects due to insecure tenure, insufficient land to set aside for PES activities, high transaction
costs, or high upfront investments needed to adopt new land use practices (Grieg-Gran et al.,

107

2005; Pagiola et al., 2005). Also relevant for PES projects are the factors that affect adoption of
new agricultural technology by smallholders: assured tenure, access to technical assistance, and
availability of savings to meet investment and maintenance costs of new land use practices
(Mercer, 2004).
The N’hambita project has tried to address many of these concerns. Benefits of REDD
activities, for instance, accrue to everyone in the local community irrespective of how much land
they own or even the extent to which they participate in other project activities. The REDD
payments go to a community trust fund managed by a democratically elected executive
committee and are used for the benefit of the entire community (more on this in the section on
spillover benefits). In contrast, participation in agroforestry activities is more selective and
depends on individual households for whether or not to enroll in the program. However, instead
of saying yes or no to one standard agroforestry contract, households have the flexibility to
choose their preferred land use from a menu of agroforestry systems that the project offers. Once
a household enrolls for a particular system, it receives free seedlings and technical assistance in
the form of training on how to plant and manage the new trees. Further, the project has
frontloaded the carbon payments, with 30% of the payment being offered in the first year, which
helps participants meet initial investment costs in the new land use system. As a result of these
flexible arrangements, there is high demand for agroforestry contracts, and as of 2008, 852 or
about 80% of all households in the area had enrolled in the project. Further, all land in Chicale
regulado is owned by the entire community as common property and after taking permission
from the local chief (the regulo), individual households can demarcate a piece of land to set up
their own farm (Jindal, 2004). In time, the household gets de facto ownership of this farmland. In

108

our field survey, we did not come across any landless household, and therefore the issue of
tenure insecurity does not apply to the N’hambita project.
Although this discussion indicates that a significantly high proportion of the local
community is participating in the project, nevertheless it is useful to investigate whether or not
the poorest households are indeed able to participate. We conduct this analysis through a twostep process. Following the approach of Pagiola et al. (2008), we first look at participation rates
of households in different income categories in our sample, followed by standard regression
analysis to identify any systematic barriers or factors that influence participation.
Figure 3.2 displays the distribution of sampled households in various income classes.
About 39% of the households in our sample had an annual per capita cash income of less than
MTN 100, while 20% of the households were between MTN 100 and 250 and so on. Only 4% of
the sampled households had a per capita cash income of more than MTN 2,000 per annum. Since
we did not use income as a variable in stratification of our sample, these percentages roughly
represent the income distribution for the whole community. When we look at the percentage of
sampled households within each income category that were enrolled in agroforestry contracts,
we find that the poorest households were slightly overrepresented while the wealthiest were
slightly underrepresented. For instance, 42% of participant households in our sample came from
the poorest income category, while only 3% of the participant households came from the
wealthiest group. This indicates that the poorest households were participating in agroforestry
activities in the N’hambita project.
For the regression analysis, we estimated two models. In the first model (column 2 in
table 3.4), we looked at factors that determine whether or not a household participates in an
agroforestry activity. Probability of participation can therefore be expressed as:

109

Pr(Pi=1) = Xiβ +Ui ……………………………………… (1)
Where probability Pi = 1 if household ‘i’ decides to enroll for an agroforestry contract
and 0 otherwise. X is the set of explanatory variables along with slope coefficientsβ, and Ui is the
error term. Equation (1) can be modified and solved using the standard Logit model where the
dependent variable Li is log of the odds ratio of participation:
Li = ln(Pi/1-Pi) = (Xiβ) +Ui ……………………………… (2)
In the second model (column 3 in table 3.4), we analyze the intensity of participation in
the N’hambita project. Since a household can simultaneously enroll in multiple agroforestry
contracts, we identify factors that determine the number of contracts that a household signs for.
Since the dependent variable is censored on left hand side (the minimum number of contracts
being 0) a Tobit model is appropriate in capturing the marginal effects of various household
characteristics on intensity of participation (Pagiola et al., 2008; Rajasekharan and Veeraputhran,
2002)

19

.

Denoting Yi as the observed number of agroforestry contracts for household i, and with X,

β, and U being defined as before, we can write the Tobit model:
Yi = (Xiβ) +Ui = Yi
if (Xiβ) +Ui > Yi

*

*

and

…………………….. (3)

19

An alternative is to use the two-stage Heckman correction. However, as Mercer (2004) points
out, if there is no prima facie reason to assume that variables that explain the dichotomous
decision of whether or not to participate and then the intensity of participation, the use of the
Tobit model is more appropriate.
110

Yi = 0, if (Xiβ) +Ui ≤ Yi

*

*

where Yi denotes the underlying latent variable, which in our case was only observed when a
household enrolled in a non-negative number of agroforestry contracts. We estimated both the
Logit (equation 2) and the Tobit model (equation 3) using the relevant commands in STATA.
The list of possible explanatory variables for the right hand side was drawn from relevant
literature on agroforestry adoption (Mercer, 2004; Nkamleu and Manyong, 2005; Franzel, 1999).
In a comprehensive review of such studies, Pattanayak et al. (2003) report five categories of
factors that were most important in explaining agroforestry adoption: preferences, resource
endowments, market incentives, biophysical characteristics, and risk and uncertainty. Based on
this work, we included the following variables in our two econometric models: (1) household
characteristics – gender of the household head, age of the household head, educational status of
the household, household size, and year of migration into the community, (2) resource
endowment – livestock ownership, number of m’shambas or fields, (3) off-farm income – any
permanent job or wage labor, and (4) location of the household. According to table 3.1, all
villages except one (Pungue) were located far away from the nearest paved road. Therefore,
taking Pungue as the base (dummy =0), we checked for the marginal effects of a household
being located in any of the other five villages viz. Bue Maria, Mbulawa, Munhanganha,
Mutiambamba, and N’hambita.
The overall results were similar across the two models. Although the gender of the
household head was insignificant, similar to Nkamleu and Manyong (2005) and Franzel (1999),
we found both household size, and off-farm income (income from a regular job or a business) to
be significant. Other factors that influenced participation were: whether the household migrated
into the community in the previous five years, total annual cash income of the household, and
111

location of the household represented by villages that were away from the paved road.
Household size had a strong positive influence on both the decision to participate and on the
number of agroforestry contracts a household enrolled for, once it decided to participate. A larger
household helped in supplying additional labor when a new land use practice was adopted.
Therefore, addition of each member increased the odds of participating by 1.34 (coefficient value
or log of odds ratio 0.29), while the predicted value of entering an extra contract increased by
0.14 However, if a household had migrated into the area within the last five years, it had a lower
likelihood of both participating in the project (coefficient value -1.32) as well as possessing more
agroforestry contracts (the predicted number of contracts fell by -0.53). This was an expected
result as recent migrants were still establishing themselves in the community and were probably
unaware of the project or how best to access it. Similar to the findings of Pagiola et al. (2008) in
Nicaragua, the econometric analysis confirmed that poorer households had a slightly higher
chance of both participating (-0.0008) in the project as well as in entering more agroforestry
contracts (-0.0003). Participation in agroforestry activities ensured a regular source of cash
income, which was important, especially for the poorer households. Interestingly, households
that already had a regular income source in the form of a job or a small business were also more
likely to participate in the project as well as to take up more agroforestry contracts. This could be
due to their ability to pay for the initial investment in adopting a new land use and to absorb any
associated risk. However, the extremely low magnitude of the respective coefficient values
(0.0009, and 0.0004) shows that households with lower off-farm income also did not find it
difficult to access the project. Finally, households that were located away from the paved road
had a much lower probability of participating in the project than households that were located in
the village Pungue, which was closest to the road (column 1). However, once the households

112

started participating in the project, their intensity of participation measured in terms of the
number of contracts they possessed, was largely unaffected by their location (column 2). This
showed that remotely located households had difficulty in first accessing the project, perhaps
because the information about the project activities was yet to reach them. But once they started
to participate they had an equal chance of entering additional contracts.
In terms of implications for the participation of the poor in the project, this discussion
indicates that the poorer households are indeed able to participate in the N’hambita project. Of
the variables that we found significant, only one – annual cash income – was related directly to
the economic status of the households, and even this showed that poorer households had a better
chance of participating in the project. The other factors such as location or size of the household
did not point towards any systematic barrier against the poor households.
3.5 PROJECT IMPACTS
According to our household survey, and based on the recall data reported to us for 2001,
before the N’hambita project started, an average household in the area consisted of 3.5 people, of
whom only one was literate (table 3.5). One third of all households were completely illiterate.
Households owned an average of 1.4 m’shambas or fields, while almost all households grew
food crops such as maize, cassava, pigeon pea, and sweet potato. 87.4% of households also
cultivated fruits and vegetables while a lower percentage (79.6%) cultivated a cash crop, mainly
sugarcane. 80% of households also purchased food from outside, spending an average of MTN
1973 for the entire year. Most households (91.9%) raised poultry birds while a smaller
percentage also owned livestock (63.3%). The average annual cash income was only MTN
975.40 while households possessed an average of 2.3 durable assets such as a radio, bicycle, or a
fishing rod etc.

113

By 2008, five years after the initiation of the N’hambita carbon project, many of these
statistics had changed significantly. The average household size had increased to 5.03, the
number of literates per household had increased to 1.6 while the proportion of completely
illiterate households had fallen to 22%. The number of m’shambas per household also increased
to 1.9, while the average livestock ownership fell. More than one-third of all households had at
least one member with a job or ran a small business. Both the annual cash income and the
average number of durables also increased to MTN 1740.60 and 2.44 respectively (table 3.5).
Clearly, there were marked changes in the area. In this context, there were two broad questions
that we wanted to explore: (i) the extent to which these changes could be directly ascribed to the
N’hambita project, and (ii) differences in project impacts on three groups of households in the
area – those that possessed both agroforestry contracts and jobs in MEs, those with only
agroforestry contracts, and those with neither.
We also wanted to check for potential bias in the selection of the project participants. For
instance, what if the participating households already owned more assets than the nonparticipants at the time the project first began? We compared this by conducting a before-project
(2001) and an after-project cross-sectional analysis of the three groups of households. Table 3.6
shows that before the project started, there were few differences across households in the three
groups except farm ownership (F-value from one way analysis of variance being 2.76), annual
20

expenditure on food (6.23), and access to wage labor within the village (4.11) . More
importantly, the F-values for differences in average annual cash income (0.21) and asset
ownership (0.56) were insignificant, which implied that there were no large systematic
20

Non-participating households were involved in more wage labor activities primarily in the
form of seasonal agricultural labor within the village. Perhaps, they received their wages in the
form of agricultural produce, which explains why they incurred relatively lower expenditure on
purchasing food from outside.
114

differences across households and they were mostly similar when the project activities first
began in the area.
However, after the initiation of project activities, the situation changed quite a lot and the
households in the three groups were no longer similar (table 3.7). A cross-sectional analysis of
the same households in 2008 revealed that there were significant differences in of the number of
literates per household (F-value 2.52), percentage of households with no literates (3.04), number
of farms per household (3.57), household’s annual expenditure on food (10.52), access to a
permanent job or a small business (19.42), household’s annual cash income (8.61), and the
number of assets it owned (3.82).
So, although the households in the three groups were fairly similar before the project
started, they underwent significant changes after the project was introduced. In order to assess
these changes and to answer the two questions raised above, we use the difference-in-difference
approach as follows (Uchida et al., 2007; Wooldridge, 2002, pg. 284):
DD = [E(T1|D=1) – E(T0|D=1)] – [E(T1|D=0) – E(T0|D=0)] ……………. (4)
Where DD is the difference-in-difference estimator, i.e. it compares before and after
project status for project participants with respect to a similar change for non-participants. E(.)
denotes the expected or the mean value for a particular group of households; D=1 represents
households that participated in the project, while D=0 denotes non-participants. T1 denotes the
after-project status for a household (year 2008 in our case), while T0 denotes the before-project
status (2001). So DD estimates the average impact on participating households between 2008
and 2001 compared to the average impact on the control group for the same time period. Since

115

we had three categories of households in the N’hambita project, we modified the approach
slightly:
DD0 = [E(T1|D=0) – E(T0|D=0)]
DD1 = [E(T1|D=1) – E(T0|D=1)]

…………………… (5)

DD2 = [E(T1|D=2) – E(T0|D=2)]
T1 and T0 have the same interpretation as before, i.e. they signify the after (2008) and
before project (2001) status respectively. DD0 measures the average change in non-participating
households (D=0) between 2008 and 2001; DD1 measures the average change over the same
time period for in households that possessed only agroforestry contracts (D=1), and DD2
measures the average change in the same period for households that had both contracts and jobs
in MEs (D=2). Once we estimated DD0, DD1, and DD2 for various impact variables, we
conducted F-tests to check whether or not the mean changes were significant across the three
groups of households. Our working hypothesis is that the project impact would be highest among
households that had both ME jobs and agroforestry contracts, followed by households with only
agroforestry contracts, and then by non-participating households that had neither of the two
(table 3.8). We present these results in the form of project impacts on participants, nonparticipants, and the spillover effects in the community.
3.5.1

Impacts on project participants
The N’hambita project provides two direct benefits to local people: carbon payments to

households that participate in agroforestry activities, and salaries to people who are employed in
various microenterprises. The schedule of carbon payments is as follows: 30% of the contract
116

value in the first year, 12% per year for the next five years, and then a final payment of 10% in
the seventh year. By the time of our field survey in 2008, most participating households had
received two or three rounds of payments; the average payment per household for the previous
year (2007–08) was MTN 1923 ($80), equivalent to about two to three months of wage labor. In
contrast, the average annual salary of an ME employee during this period was much higher at
21

MTN 12,484 ($519) . In our group discussions with project participants, many respondents said
they used their money to buy roofing material, food and clothes for the family, or books and
school stationery for their children. Some people also invested in agricultural seeds, while others
bought household durables such as a radio or a bicycle.
Table 3.8 shows that average annual cash incomes before (2001) and after (2008) the
project increased for households that possessed only agroforestry contracts (increase of MTN
566.30) as well as for households that had both agroforestry contracts and jobs in MEs (MTN
1865.11). The corresponding change for non-participating households, although positive (MTN
208.60), was relatively modest and statistically insignificant. Although an F-test for comparison
of the three mean increases was also significant (F-value 14.25), this was only because of the
huge increase in income for the group that possessed both agroforestry contracts and jobs in
MEs. Households that participated in only agroforestry activities had more income than before,
but this increase was statistically insignificant when compared to the mean change in income for
non-participating households. Thus the impact of wages earned from employment in MEs was
much stronger than the impact of carbon payments alone. The same was also verified when we
analyzed the change in average asset ownership. Again, households with employment in MEs
and with agroforestry contracts increased their ownership of durable assets by an average of
21

Some people such as the forestry extension staff receive a fixed monthly salary while others
such as carpenters get a piece rate.
117

0.41, which was much larger than the increase for households with only agroforestry contracts
(0.04) or for non-participating households (a decrease of 0.11 per household). Predictably, there
was also a huge increase in access to a regular job for households that received employment in
the local MEs (51.85%) but not for the other two groups.
These results substantiate our working hypothesis stated in the previous section as the
households with ME employment were much better off than others. However, there is one major
departure: households with only agroforestry contracts were slightly better off than before but
this change was not significant when compared to the change for non-participating households
with neither ME jobs nor agroforestry contracts. In other words, the development component of
the project had a significant livelihood impact on participating households, but the same cannot
be said for the PES component. However, it is also important to note that most households had
received only two to three rounds of carbon payments by the time this study was conducted.
With more payments to come in the near future, the impact on participating households may
increase. In addition, as the agroforestry systems mature, they will generate marketable products
such as fruits, with additional financial benefits for participants.
3.5.2

Impacts on non-participants
By design, payments under a PES project accrue to service providers (Ferraro and Kiss,

2002). As a result, participating households can expect to gain more than non-participating
households – an incentive mechanism to encourage land stewards to adopt more sustainable land
use practices. An important concern regarding non-participants, however, is whether the PES
project poses any risk to their livelihoods should they voluntarily decline to participate (Pagiola
et al., 2005).

118

Table 3.6 shows that before the project started, 60.9% of the non-participating

22

households earned wage labor in the village, mainly by providing seasonal agricultural labor to
other households. By 2008, the proportion of non-participating households earning wage labor
within the village dropped to 52.2% (table 3.7). This reduction of 8.7% (table 3.8) before and
after the project was much bigger when compared to the change for households with both
agroforestry contracts and ME jobs (-3.7%) and for households that had only agroforestry
contracts (2.9%). For those households now employed in various MEs, the reduced access to
wage labor is compensated. This is also true to a large extent for households that receive carbon
payments from agroforestry contracts. However, the same cannot be said for non-participating
households. There is insufficient data to establish whether or not the loss in wage labor income
for non-participating households has resulted in an overall decline in their economic status. Their
average asset ownership declined marginally (-0.11), but this change is statistically insignificant,
while their cash income showed a modest increase (MTN 208.60).
Non-participating households may also face increased hardship from a complete ban on
harvesting any resources from the large tract of miombo forest under REDD activities. However,
the project tries to ensure that benefits from REDD activities reach the entire community. For
instance, all REDD payments are transferred to a community fund that supports development of
community infrastructure such as construction and maintenance of school buildings. We explore
the spillover impacts of these activities presently.

22

Technically, there are no non-participants in the project area since even these households
participate in REDD activities. However, the attempt here is to explore the more direct impact of
participation or non-participation in agroforestry activities.
119

3.5.3

Spillover effects in the area
In recent years, there has been impressive growth in public infrastructure in the area,

mainly due to an overall improvement in Mozambique’s national economy rather than direct
project intervention. However, the project does play a role in creating demand for such
infrastructure. The $80 of carbon payments that each participating household receives on average
aggregates to about $70,000 per annum for the community. In addition, most people employed in
MEs belong to the local community and spend a considerable proportion of their monthly wage
within the area. This is a significant change for the local people, who until a few years ago had
very few sources of cash income.
Rising disposable incomes have also resulted in expansion of economic activity in the
area. With many households routinely buying household items such as soap and cooking oil,
many small provision stores and grocery shops have opened in all the six villages, which is a
considerable improvement given that most of these villages had almost no local shops till just a
few years back.
Another important spillover benefit is the community trust fund which receives half of all
REDD payments and a proportion of carbon payments from agroforestry activities. This fund is
managed by the local community association which consists of 24 members representing
different villages. In mid-2008, there were MTN 65,000 ($2,683) in the trust fund with more
REDD payments (about $22,942) expected shortly. The association has mainly used this trust
fund for community development activities such as construction and maintenance of school
buildings and a local health clinic, that benefit the entire area irrespective of the participation
status of individual households.

120

Apart from these qualitative impacts, there are two additional impacts that we explored
empirically: change in literacy rates and a decline in livestock ownership. During focus groups,
many participating households said they had used their carbon payments to pay for their
children’s school fees and school supplies. Both tables 3.5 and 3.7 show an impressive change
in overall literacy rates in the area between 2001 and 2008; the average number of literates per
household increased from 0.9 to 1.6 during this period, while the percentage of households with
no literates fell from 33% to 22%. This change was equally impressive across both participating
and non-participating households (table 3.8). In order to confirm whether or not the change in
literacy rates was a spillover effect of the project, we compared the change in project villages
23

with the change in control villages that were outside the project area . Using the difference-indifference approach as outlined before, but now comparing the mean change before-project
(T=0) with after project (T=1) for households in project villages (D=1) with the change over the
same period for households in the control villages (D=0):
DD = [E(T1|D=1) – E(T0|D=1)] – [E(T1|D=0) – E(T0|D=0)] ……………. (6)
Table 3.9 shows that even in the control villages, the average number of literates per
household increased from 0.8 in 2001 to 1.8 in 2008, while the proportion of household with no
literates fell from 40.6% to 23.4% over the same period. The difference between the mean
change in the number of literates per household in the project villages (0.7) is not statistically
significant from that in the control villages (1.0). The same result is found for the decline in
proportion of households without literates where the difference-in-difference or the DD estimate
(-6.1%) is close to zero. These results indicate that the increase in literacy rates is due to an

23

Six villages from the neighboring area of Cudzu were selected as control villages.
121

overall increase in educational attainment in the entire region rather than any direct or indirect
impact of the N’hambita project.
A ban on using a local forest or pastureland can also have an adverse impact on the
community, especially in terms of a drop in livestock ownership as the animals can no longer be
taken out for grazing (Jindal, 2000). Although, we do see a similar trend in the case of the
N’hambita project where the livestock ownership declined by an average of 2.8 animals per
24

household between 2001 and 2008 (table 3.8) , table 3.9 shows that even in control villages, the
average livestock ownership decreased from 12.6 animals per household in 2001 to 7.4 animals
per household in 2008. Comparison of the mean decline in the project villages (2.8) with the
same in control villages (5.2) shows that the two were not significantly different from each other,
i.e. DD was not statistically different from zero. This indicates that a fall in livestock numbers is
a wider trend in the area. We could not establish whether this decline was due to some livestock
disease or a result of a drought in the area or to changes in the structure of the economy, but as
we can infer, the decline in livestock ownership was not due to the project, which is also
confirmed by the decline in ownership of poultry birds in both the project and the control
villages.
3.5.4

Environmental impact of the N’hambita project
We explore the environmental impact of the N’hambita project in terms of carbon offsets

produced by agroforestry and REDD activities. The number of carbon sequestration offsets (in
tCO2) from agroforestry systems can be estimated as:
No. of carbon offsets = {(kind of agroforestry contract – baseline) – buffer} X area ………. (7)
24

Livestock here mainly refers to small ruminants such as goats and sheep. There were hardly
any large ruminants in the area such as cows or bulls. Therefore, we measured livestock
ownership in terms of number of animals owned per household.
122

The kind of agroforestry contract refers to different agroforestry systems offered by the
project, each with a specific carbon sequestration rate. The baseline is the carbon stock of
existing trees and shrubs on a site prior to new planting. To calculate the net number of carbon
offsets, the project team subtracts the baseline from carbon sequestered through project activities
(table 3.2). Since measuring actual carbon stock on each dispersed site is expensive, the project
uses sample plots to estimate the existing stock of carbon for the entire area (Williams et al.,
2008; Sambane, 2005 ).
By May 2008, the project had 1,234 agroforestry contracts in operation, covering about
1,000 ha of farmland. According to our survey, boundary planting was the most popular,
followed by fruit orchards, and homestead planting respectively. The project team estimates that
during the previous five years, local farmers had planted more than 500,000 trees under different
agroforestry systems, expected to generate a total of 82,056 tCO2 as carbon sequestration offsets.
Avoided deforestation or REDD activities consist of protection and management of
miombo woodlands around the Gorongosa National Park. On average, a well stocked area of
miombo woodland contains 95.42 tCO2 per ha, much of which is under threat of being emitted
into the atmosphere due to deforestation (Grace et al., 2007). To reduce deforestation, the
N’hambita project pays the local community to protect 11,071 ha of miombo woodlands outside
the GNP, while motivating community members to also conserve additional forest areas through
selective logging and reforestation. The project team estimates that these protection and
conservation activities have reduced emissions at the rate of 7.33 tCO2/ha per annum since 2006,
generating 154,457 tCO2 as REDD offsets. This implies that the project has generated a total of

123

236,513 tCO2 REDD and carbon sequestration offsets, placing it among the small to medium
sized forestry carbon projects in Africa (Jindal et al., 2008).
(i)

Leakage
Leakage refers to unplanned emissions of carbon arising from activities outside the

project boundary. For instance, project beneficiaries may plant trees at one site but cut trees in
another, resulting in net release of carbon to the atmosphere. The three main drivers of leakage in
the area are cutting of trees for charcoal production, uncontrolled burning of farmlands, and
clearing of forest for agriculture. Charcoal production is an important source of livelihood in
Mozambique (FAO, 2007). In Chicale regulado, Herd (2007) estimates that 35 ha of local
woodlands are lost every year to charcoal production. The N’hambita project has tried to address
this issue by promoting agroforestry as an alternative source of income and by educating
charcoal producers to use efficient kilns that reduce wood intake. The project also discourages
the local community from installing any new kilns in the area.
Burning of farmlands is an old cultural practice in the area. Farmlands or m’shambas are
burnt in preparation for cultivation, to clear undergrowth around settlements, for honey
collection, or to keep away dangerous animals. This increases the risk of carbon loss especially if
the fire escapes to nearby forest areas (Zolho, 2005). Therefore, the project strictly dissuades
contracted households from burning their m’shambas. However, in our survey, 16% of
agroforestry contract holders confirmed that they had burned their m’shamba in the previous
year. Although many of these respondents had also modified the burning to reduce the risk of
wildfire, clearly it is not easy to end this old cultural practice and the project needs to do much
more to reduce the chance of leakage.

124

The third and probably the most important driver of leakage in the area is the clearing of
forest for agriculture. During the Mozambique civil war from 1977 to 1992, a large proportion of
the population was displaced from rural to urban areas (Heltberg et al., 2003). Since the 1994
peace accord, many people have returned to rural areas in search of better livelihoods (figure 3).
For instance, 35% of our respondents migrated to the Chicale regulado from outside, As
expected, the biggest influx of people was immediately after the return of peace in 1994.
Relevant for the N’hambita project is the curve after 2003, which has lower peaks than earlier,
but the arrival of new households still continues. This translates into potential leakage for the
project as the incoming households clear forests to set up their farms. Dealing with this internal
migration, however, is a national policy issue and the N’hambita project has little control over it.
The project does try to reduce its impact though by encouraging migrants to enrol for carbon
contracts provided they agree to conserve existing forests by not clearing additional land for
m’shambas.
Interestingly, the practice of setting up new m’shambas is not limited to recent migrants
alone. According to our survey, the average number of farms in the area increased from 1.4 per
household in 2001 to 1.9 in 2008 (table 3.5). Previously settled residents may clear forests to set
up new farm plots as their old plots become less productive with time (Jindal, 2004). Another
reason could be that as more and more fields are put under agroforestry activities, farmers may
need to clear additional forestland to grow food crops for the household needs. The latter,
however, cannot be verified when we look at households in the control villages, where the
number of fields per household also increased from 1.3 in 2001 to 2.0 in 2008 (table 3.9). If at
all, this increase (0.7) is more than the increase in the number of fields per household in the
project villages (0.5), but this difference is statistically insignificant.

125

(ii)

Permanence
Permanence of carbon offsets is an important concern due to the temporary nature of

forestry carbon stocks – a forest can be cut at any time, eventually releasing most of the
sequestered carbon back into the atmosphere (Sedjo et al., 2001). For the N’hambita project the
most important threat to permanence is the extremely long contract period. The project estimates
its carbon offsets based on a 100 year contract period. Assuming such a contract is enforceable, it
produces high value long-term offsets but it subjects future generations to a rule they may not
agree with. An alternative for the project is to shorten the contract period, to say ten years.
However, that would produce temporary carbon offsets which carry a lower price in the market
(Haites, 2004). This would greatly reduce the carbon payments to farmers and perhaps make the
project financially unviable. So there is an inherent trade-off between contract duration and the
payment that local farmers will receive for carbon offsets.
A related issue is the timing of carbon payments. To help farmers cover establishment
costs of setting up new agroforestry systems, the project pays them the entire value of the carbon
offsets over the first seven years. Thereafter, the agroforestry systems are expected to provide
enough returns in the form of improved soil productivity, and timber and non-timber products for
farmers to continue managing them well. Many systems such as mango and cashew orchards will
also provide saleable products that can yield additional incomes. However, these benefits are yet
to accrue as most trees are still very young and in many cases may not compensate for loss of
carbon income. If some farmers do decide to cut their trees or stop caring for them after seven
years, the entire project may be jeopardized. The project tries to address this threat by retaining
15% of all carbon offsets as a risk buffer. However, future experience will determine if this risk
buffer is sufficient to address concerns regarding permanence of carbon offsets.

126

3.6 CONCLUSION: LESSONS FROM N’HAMBITA

A lot has been written about the potential poverty impacts of PES projects, especially
forestry based carbon mitigation projects. While this paper does not contradict these claims, the
analysis presented here offers a word of caution. The N’hambita community carbon project is a
well run project and within a short period of time it has penetrated deep within the community
with a participation rate of more than 80%. We also found that poor households are able to
access the project and many of them have multiple carbon contracts in the form of agroforestry
systems. The resultant carbon payments do supplement household incomes but not enough to
move the households out of poverty. Poverty alleviation is a long, complex process and it is not
realistic to ask PES projects, even with their community focus, to achieve it completely on their
own. In contrast, the development component of the project has had an immediate economic
impact through the employment opportunities it has created. However, the project cannot employ
everyone in the community and this may not be even replicable when most of the project finance
is raised from carbon sales in environmental markets.
There are also some useful lessons for the continuing negotiations on the role of forestry
carbon projects in carbon mitigation strategies. Combining carbon sequestration on individual
plots with REDD payments on community owned forests presents an interesting option. This
natural complementarity helps reduce transaction costs relative to overall project benefits.
Transaction costs for the N’hambita project are high but would have been much higher if the two
activities were not combined. It is important to design such combined projects in ways that
ensure that local communities retain flexibility to meet their timber and nontimber needs.
The menu of agroforestry systems offered by the N’hambita project also addresses the
issue of flexibility to a certain extent. In contrast to many carbon sequestration projects that

127

allow only one set of land use practices, this menu provides flexibility for individual households
to select systems that suit their specific needs. Mixing native trees with other multi-purpose
species also ensures that as the trees mature, farmers can fulfill many of their timber and nontimber requirements from their own farmlands, reducing the need to fell forest trees.
This flexibility comes at a price: escalating transaction costs related to monitoring and
supervising individual contracts. Even in N’hambita, where a large proportion of carbon offsets
comes from REDD activities, one third of all carbon revenue is used to meet local transaction
costs and another third is paid to international brokers and commission agents who help sell
carbon offsets. However, there are two interrelated issues here. One is that when the official
Kyoto market does not accept forestry offsets and the voluntary market is highly disaggregated,
hiring international brokers may be a necessary expense. In case of N’hambita, these brokers
ensure that the project continues to sell its offsets at a premium. This is not easy considering how
variable the carbon price is; even on a well established voluntary market such as the Chicago
Climate Exchange, the price had reduced from a high of $5 per tCO2 to less than $0.50 per tCO2
within a short span of time. In contrast, local farmers would need assurance of steady payments
in return for taking up carbon sequestration activities. How would community based projects
manage to establish these contracts if the carbon price in the voluntary market is so uncertain?
The second issue relates to high transaction costs when many small farmers are contracted,
instead of a few large ones (Kerr et al., 2006). When these costs are unavoidable, projects may
again need additional funds until they are able to raise sufficient carbon revenue from the market.
This raises serious concerns about the viability of community based carbon projects that are not
subsidized by donor funds, at least in the initial stages of project development.

128

Another concern about the N’hambita project is the duration of the contract and the
payment schedule. Monetizing carbon offsets over 100 years and disbursing payment over seven
years has the advantage of taking care of farmers’ upfront costs, but it locks them into very long
term contracts leaving little room for renegotiating the contract if market prices for carbon
offsets increase in the future. Also, after the last payment there is a real risk that farmers,
particularly future generations, may have little incentive to care for their trees. Addressing this
challenge poses a dilemma. Instead of monetizing carbon offsets over 100 years (long-term
carbon offsets), the project could pay farmers on an annual basis for the number of offsets they
produce in that year (temporary carbon offsets). Such a contract would be realistic but the value
of the carbon payments might be too low to interest farmers.
Finally, in terms of payment mechanisms for REDD, this project distributes payment
between wages for forest guards and a community fund. Judicious use of the fund is paramount
in giving individual households an incentive to conserve the forest. However, forest use is
dynamic and open to many conflicting claims. In the N’hambita project, migration into the area
and new migrants’ need to create farmlands places heavy pressure on forests, which cannot be
addressed only through REDD payments. This is a national phenomenon which requires a
country wide strategy.

129

Appendices

130

Table 3.1: Profile of project villages in Chicale regulado
Village

Distance
to tar road

Total number
of households*

Primary
School

1. Bue Maria
2. N’hambita
3. Munhanganha
4. Mutiambamba
5. Mbulawa
6. Pungue

20 km
12 km
12 km
8 km
6 km
1-4 km

42
64
65
56
414
385

2 km
In village
In village
In village
In village
In village

Location of nearest
Primary
Grocery
Health Center
shops
20 km
12 km
12 km
8 km
10 km
In village

27 km
In village
12 km
8 km
In village
In village

Source: Authors’ field work 2008. * Hegde and Bull (2008)

Table 3.2: Important components of the N’hambita project
I.

PES Component

II.

1. Payments for Carbon sequestration
•
•
•
•

1. Jobs in microenterprises (ME)

Agroforestry on HH farms
Payments to HH for 7 years (based on
rates for 100 years)
Avg price $4.50/tCO2
Payment: $400-$800/ha

•
•
•
•

Various microenterprises promoted by
the project
Carpentry shop
Tree nurseries
Salary $50 - $100 per month

2. Project and extension staff

2. Payments for REDD
•
•

Development Component

Protection of 11,000 ha forest block
Payment to community trust fund

•
•

131

From local community
Salary $100 - $600 per month

Table 3.3: Carbon sequestration rates under the N’hambita Project
(1)
Carbon
sequestered
(tC)
Boundary
planting
Interplanting
(Gliricidia)
Cashew
orchards
Mango orchards
Homestead
planting
Woodlots
Interplanting
(Faidherbia)

(2)
(3)
(4)
Baseline
Buffer
Net carbon
carbon stock carbon stock offsets (tC)
(tC)
(tC)
(1) – (2) – (3)

(5)
Net carbon
offsets
(tCO2)*

3.23/100m

0

0.48/100m

2.75/100m

10.03/100m

10.00/ha

0

1.50/ha

8.50/ha

31.16/ha

40.14/ha
34.00/ha

2.8/ha
2.8/ha

5.60/ha
4.68/ha

31.74/ha
26.52/ha

116.38/ha
97.24/ha

42.05/ha
61.30/ha

0
11.3/ha

6.30/ha
7.50/ha

35.75/ha
42.50/ha

131.08/ha
155.83/ha

58.20/ha

0

8.73/ha

49.47/ha

181.05/ha

* 1tC = 3.67 tCO2.
Estimates are based on projected tree growth under standard climatic and soil conditions, and
assume that all farmers will follow a standard set of silvicultural practices. These sequestration
rates may be affected by natural disasters such as prolonged drought or fire outbreaks.
Source: Tipper, 2008.

132

Table 3.4: Determinants of participation in agroforestry activities

Gender of household head (Male = 1)
Migration into the area within past 5
years (dummy)
Number of people in the household
Number of literates in the household
Age of the household head
Number of m’shambas per household
Number of livestock per household
Total annual cash income of the
household
Household Income from a regular job
and/or a business
Bue Maria (Dummy = 1)
Mbulawa (Dummy = 1)
Munhanganha (Dummy =1)
Mutiambamba (Dummy = 1)
Nhambita (Dummy = 1)
Constant

(1)
Dependent variable =
probability of participation
Logit Model
-0.17 (0.43)

(2)
Dependent variable =
number of contracts
Tobit Model
0.07 (0.17)

-1.32 (0.55)***
0.29 (0.15)**
0.12 (0.27)
-0.002 (0.015)
0.15 (0.31)
-0.03 (0.04)
-0.0008***
(0.0002)
0.0009***
(0.0004)
-17.85 (1.29)***
-19.45 (0.81)***
-19.36 (0.79)***
-18.64 (0.80)***
-19.76 (0.82)***
20.28

-0.53(0.27)**
0.14(0.06)**
0.07 (0.099)
-0.00 (0.006)
0.14 (0.13)
0.002 (0.127)
-0.0003**
(0.0001)
0.0004***
(0.0002)
-0.19 (0.416)
0.56 (0.32)**
0.07 (0.28)
-0.07 (0.32)
-0.46(0.35)
0.64 (0.36)*

32.68
0.0032
0.1532
-90.311

42.41
0.0001
0.1682
-334.1915

LR Chi sq
Prob > Chi sq
Pseudo R sq
Log likelihood

Notes: Figures in parentheses are standard errors
* Significant at 10% **Significant at 5% *** Significant at 1%

133

Table 3.5: Selected statistics for sampled households, 2001 and 2008 (n=270)
(1)
Before project in
year 2001
Household size
Number of literates per
household
Household with no
literates (%)
Number of m’shambas
per household
Household cultivating
food crops (%)
Household cultivating
horticultural crops
(%)
Household cultivating
cash crops (%)
Household bought food
from outside (%)
Number of months food
bought per
household
Household’s annual
expenditure on food
(MTN)
Household that own
livestock (%)
Number of livestock per
household
Household that own
poultry birds (%)
Number of poultry birds
per household
Household with at least
one permanent job or
a small business (%)
Household’s annual cash
income (MTN)
Asset ownership per
household (number)

(2)
After project in
year 2008

3.45 (1.77)

5.03 (2.19)

(3)
Difference in mean
(2) – (1) Paired tstatistics in parentheses
1.58 (15.9)***

0.9 (0.95)

1.6 (1.42)

0.7 (9.4)***

33 (47.2)

22 (41.7)

-11 (5.1)***

1.4 (0.67)

1.9 (0.88)

0.5 (8.3)***

95 (21.5)

99 (8.6)

4 (3.1) ***

87.4 (33.2)

93 (25.6)

5.6 (2.8)***

79.6 (40.3)

78 (41.7)

-1.6 (0.65)

80 (40)

82.6 (37.9)

2.6 (1.3)*

9.1 (3.9)

9.2 (3.8)

0.1 (0.7)

1972.80 (1952.8)

2452.30 (2159.5)

484.6 (3.97)***

63.3 (48.3)

53.3 (49.9)

-10 (3.01)***

5.1 (7.44)

2.3 (3.3)

-2.8 (6.1)***

91.9 (27.4)

91.5 (27.9)

-0.4 (0.18)

13.3 (13.81)

10.7 (11.61)

-2.6 (3.1)***

24.8 (43.3)

37 (48.4)

12.2 (3.9)***

975.40 (1417.15)

1740.60(2076.64)

765.10 (6.9)***

2.34 (1.23)

2.44 (1.29)

0.1 (1.3)*

Notes: Standard deviations are in parentheses for columns (1) and (2)
Column (3) reports absolute value of t-statistics. *** Significant at 1%. * Significant at 10%
134

Table 3.6: Cross-sectional analysis of sampled households before-project (2001)
Variables

Number of literates per
household
Households with no literates
(%)
Number of m’shambas per
household
Households that own
livestock (%)
Number of livestock per
household
Number of poultry birds per
household
Household’s annual
expenditure on
purchasing food (MTN)
Households with access to
wage labor in the village
(%)
Household with at least one
permanent job or a small
business (%)
Household’s annual cash
income (MTN)
Asset ownership per
household (number)
Number of sampled
households

(1)
Mean value
for nonparticipating
HH

(2)
Mean value
for HH with
only carbon
contracts

(3)
Mean value for
HH with both
carbon contracts
and ME
employment

(4)
F-value
from oneway analysis
of variance

0.85 (0.92)

0.99 (0.95)

1.09 (0.98)

0.82

41.3 (0.49)

34.1 (0.48)

24.1 (0.43)

1.73

1.44 (0.67)

1.38 (0.56)

1.64 (0.92)

2.76*

56.5 (0.50)

66.5 (0.47)

59.3 (0.49)

1.01

3.72 (4.85)

5.5 (7.95)

5.24 (7.80)

0.99

13.72 (15.14) 12.89 (13.13)
939.66
2205.19
(745.18)
(2108.59)

14.13 (14.63)
2053.59
(1841.04)

0.18
6.23***

60.9 (0.49)

37.6 (0.49)

40.7 (0.49)

4.11**

26.1 (0.44)
1017.87
(945.33)

26.5 (0.44)
998.98
(1488.54)

18.5 (0.39)
865.28
(1534.03)

0.71

2.17 (1.32)
46

2.39 (1.20)
170

2.37 (1.25)
54

0.56

Notes: HH Households.
Figures in parentheses are standard deviations.
* Significant at 10% ** Significant at 5%
***Significant at 1%

135

0.21

Table 3.7: Cross-sectional analysis of sampled households after-project (2008)
Impact variables

Number of literates per
household
Households with no literates
(%)
Number of m’shambas per
household
Households that own
livestock (%)
Number of livestock per
household
Number of poultry birds per
household
Household’s annual
expenditure on
purchasing food (MTN)
Households with access to
wage labor in the village
(%)
Household with at least one
permanent job or a small
business (%)
Household’s annual cash
income (MTN)
Asset ownership per
household (number)
Number of sampled
households

(1)
Mean value
for nonparticipating
HH

(2)
Mean value
for HH with
only carbon
contracts

(3)
Mean value for
HH with both
carbon contracts
and ME
employment

(4)
F-value
from oneway analysis
of variance

1.26 (1.29)

1.66 (1.39)

1.89 (1.54)

2.52*

34.8 (48.1)

21.2 (40.9)

14.8 (35.86)

3.04**

1.67 (0.83)

1.88 (0.81)

2.13(1.06)

3.57**

52.2 (50.51)

51.8 (50.1)

59.3 (49.59)

0.47

2.44 (3.15)

2.15 (3.37)

2.63(3.27)

0.45

8.13 (8.34) 10.82 (12.02)
1209.80
2885.10
(1401.81)
(2188.88)

12.67 (12.61)
2153.20
(2179.9)

1.92
10.52***

52.2 (50.51)

40.6 (49.3)

37.04 (48.74)

1.32

19.6 (40.11)
1226.50
(1430.11)

31.2 (46.46)
1565.3
(1844.45)

70.4 (46.09)
2730.40
(2824.38)

19.40***

2.07 (1.29)
46

2.43 (1.33)
170

2.78 (1.11)
54

3.82 **

Notes: HH Households.
Figures in parentheses are standard deviations.
* Significant at 10% ** Significant at 5%
***Significant at 1%

136

8.61 ***

Table 3.8: Impacts of the N’hambita project
Difference in difference from 2001 and 2008 status
Impact variables

Number of literates per
household
Households with no literates
(%)
Number of m’shambas per
household
Households that own
livestock (%)
Number of livestock per
household
Number of poultry birds per
household
Household’s annual
expenditure on
purchasing food (MTN)
Households with access to
wage labor in the village
(%)
Household with at least one
permanent job or a small
business (%)
Household’s annual cash
income (MTN)
Asset ownership per
household (number)
Number of sampled
households

(1)
Mean change
for nonparticipating
HH

(2)
Mean change
for HH with
only carbon
contracts

(3)
Mean change for
HH with both
carbon contracts
and ME
employment

(4)
F-value
from
one-way
analysis
of
variance

0.41 (0.13)***

0.67 (0.09)***

0.79 (0.16)***

1.46

-6.5 (0.037)*

-12.9 (0.03)***

-9.3 (0.04)**

0.67

0.35 (0.13)**

0.58 (0.059)***

0.70 (0.16)***

2.06*

-4.3 (0.069)

-14.7 (0.04)***

0.0 (0.07)

1.79

-1.28 (0.56)**

-3.3 (0.65)***

-2.61 (0.97)***

1.46

-5.6 (1.93)**
310.8*
(203.41)

-2.1 (1.08)**
550.97***
(140.86)

-1.5 (1.91)
191.57
(306.73)

1.49

-8.7 (0.04)***

2.9 (0.027)

-3.7 (0.065)

2.07*

-6.5 (0.057)
208.61
(208.35)

4.7 (0.03)
566.28***
(111.26)

-0.11 (0.18)
46

0.04 (0.08)
170

51.85 (0.08)*** 25.49***
1865.11***
(356.13) 14.25***
0.41 (0.17)***
54

Notes: HH Households.
Figures in parentheses are standard errors.
Columns 1, 2, and 3 represent intra-group differences between 2001 and 2008.
Column 4 represents F-values for difference-in-difference for each variable.
* Significant at 10% ** Significant at 5%
***Significant at 1%

137

0.89

2.94**

Table 3.9: Comparison of the Project area with the Control area
Difference in difference between 2001 and 2008 status

Household size
Number of
literates per
Household
Households with
no literates
(%)
Number of
m’shambas
per Household
Number of
livestock per
household
Number of
poultry birds
per Household
Sample Size

Project Villages
(1)
(2)
(3)
Mean
Mean Differenc
in 2001
in
e in
2008
Mean
(2) – (1)
3.4
5.0 1.6 (0.10)

(4)
Mean
in
2001

Control Villages
(5)
(6)
Mean Difference
in
in Mean
2008
(5) – (1)

(7)*
Difference
in
Difference
(6) – (3)

4.2

6.1

1.9 (0.23)

0.3 (0.25)

0.8

1.8

1.0 (0.24)

0.3 (0.25)

-11.1
(0.02)

40.6

23.4

-17.2
(0.05)

-6.1 (0.05)

1.4

1.9 0.5 (0.05)

1.3

2.0

0.7 (0.12)

0.2 (0.13)

5.1

2.3

-2.8
(0.46)

12.6

7.4

-5.2 (2.07)

-2.4 (2.11)

13.3

10.7

-2.6
(0.85)

17.5

17.5

0 (2.62)

2.6 (2.76)

0.9
33.3

1.6 0.7 (0.07)
22.2

270

64

Notes: Standard errors in parentheses
* None of the values in column 7 is different from zero at usual significance levels.

138

Figure 3.1: Major components of the N’hambita Community Carbon Project
Sale generates carbon revenue,
supports four project components
Sequestration offsets

1. Carbon
sequestration from
agroforestry
- taken up on
individual plots
- contracted
households receive
annual payments
- a small proportion
of the payment
goes to trust fund
managed by the
community
association.

REDD offsets

2. REDD activities
- taken up on
community owned
miombo
woodlands
- a proportion of the
payment made to
forest guards who
patrol the area
- the other half
deposited in trust
fund that supports
community
development
activities.

139

3. Micro-enterprises
- include activities
such as tree
nurseries,
carpentry shop,
and a saw mill
- workers drawn
from the local
community and
receive monthly
wages
- initially supported
through donor
money. Now
funded through a
proportion of
carbon revenue.

4. Project
administration
- performed
mainly by the
local staff
members of the
Envirotrade
Ltd.
- extension
activities,
project
administration
- funded as
administrative
overhead from
carbon
revenue.

Figure 3.2: Distribution of sampled households: Proportion of households in different
income categories (2001) and proportion of households with carbon contracts

Source: Authors’ survey (2008)

140

Figure 3.3: Pattern of migration into Chicale Regulado, Mozambique

Source: Author’s survey (2008).

141

References

142

REFERENCES

Alix-Garcia, J., E. Shapiro, and K. R.E. Sims. (2010) The environmental effectiveness of
payments for ecosystem services in Mexico: Preliminary lessons for REDD. Paper
presented at the World Congress of Environmental and Resource Economists, 2010,
Montreal, Canada.
Aune, J.B., A.T. Alemu, and K.P. Gautam. 2005. Carbon sequestration in rural communities: Is
it worth the effort? Journal of Sustainable Forestry, 21(1): 69–79.
Capoor, K., and Ambrosi, P. 2009. State and Trends of the Carbon Market 2009. The World
Bank, Washington DC.
Chicago Climate Exchange (CCX). (2010) http://www.chicagoclimatex.com/ Accessed on
August 25, 2010.
Chung, K. 2000. Using Qualitative Methods to Improve Data Collection and Analysis of the
Living Standards Measurement Surveys, in M. Grosh and P. Glewwe (eds) Designing
and Implementing the Living Standards Measurements surveys: Lessons from Ten Years
of Experience. The World Bank, Washington DC.
Corbera, E., N. Kosoy, and M. Martínez-Tuna. 2007. Equity implications of marketing
ecosystem services in protected areas and rural communities: Case studies from MesoAmerica. Global Environmental Change 17 (3–4), 365–380.
Engel, S., S. Pagiola, and S. Wunder.2008. Designing payments for environmental services in
theory and practice: an overview of the issues. Ecological Economics 65, 663–674.
Eraker, H. 2000. CO2lonialism in Uganda, NorWatch Newsletter, No.5, 2000.
http://www.norwatch.no/index.php
Ferraro, Paul J., and A. Kiss. 2002. Direct Payments to Conserve Biodiversity. Science. 29
November 2002. Pg 1718-1719.
Food and Agriculture Organization (FAO). 2009. State of the World’s Forests:2009. Rome.
Food and Agriculture Organization (FAO). 2007. State of the World’s Forests:2007. Rome.
Food and Agriculture Organization (FAO) 2003. Special Report: FAO/WFP Crop and Food
Supply Assessment Mission to Mozambique. FAO Global Information and Early Warning
System on Food and Agriculture, FAO and World Food Program (WFP), Rome.
Franzel, S. 1999. Socioeconomic factors affecting the adoption potential of improved tree
fallows in Africa. Agroforestry Systems, vol. 47, Pg. 305-321.
143

Gong, Y., G. Bull, and K. Baylis. 2010. Participation in the world's first clean development
mechanism forest project: the role of property rights, social capital and contractual rules.
Ecological Economics 69 (6), 1292–1302
Grace, J. (2008) Miombo Community Land Use & Carbon Management: N’hambita Pilot
Project, Annual Report 1 August 2006 – 30 November 2007. The University of
Edinburgh, UK. http://www.miombo.org.uk/Documents.html
Grace J, C. Ryan, and M. Williams (2007) An inventory of tree species and carbon stocks for the
N’hambita Pilot Project, Sofala Province, Mozambique. The University of Edinburgh,
UK http://www.miombo.org.uk/Documents.html
Grieg-Gran, M., I. Porras, and S.Wunder 2005. How Can Market Mechanisms for Forest
Environmental Services Help the Poor? Preliminary Lessons from Latin America. World
Development, vol. 33, no. 9, pp. 1511–1527.
Gutman, P. (ed.) 2003. From Goodwill to Payments for Environmental Services: A Survey of
Financing Options for Sustainable Natural Resource Management in Developing
Countries. World Wide Fund for Nature (WWF).
Haites, E. 2004. Rewarding sinks projects under the CDM. Environmental Finance, March,
2004.
Hamilton, K., U. Chokkalingam, and M. Bendana. 2010. State of the Forest Carbon Market
2009: Taking root and branching out. Ecosystem Marketplace, Forest Trends,
Washington DC.
Hegde, R. and G. Bull (2008) Economic Shocks and Miombo Woodland Resource Use: A
household level study in Mozambique. Department of Forest Resource Management,
University of British Columbia, Canada. Unpublished.
Heltberg, R., Simler, K., and Tarp, F. 2003. Public Spending and Poverty in Mozambique. Food
Consumption and Nutrition Division Discussion Paper no. 167, International Food Policy
Research Institute (IFPRI), Washington D.C.
Herd, A. (2007) Exploring the socio-economic role of charcoal production and the potential for
sustainable production in the Chicale Regulado Mozambique. MSc dissertation, School
of GeoSciences, University of Edinburgh, UK.
Howell, D., and I. Convery. (1997) Pilot Socio-Economic Study 1997: Nhambita Regulado,
Sofala Province. Forest and Wildlife Management Project (GERFFA), Mozambique.
Jindal, R., Swallow, B., and Kerr, J. 2008. Forestry-based carbon sequestration projects in
Africa: Potential benefits and challenges, Natural Resources Forum, vol. 32, Pg.116–130.

144

Jindal, R., J. Kerr, and S. Nagar 2007. Voluntary Carbon Trading: Potential for Community
Forestry Projects in India. Asia-Pacific Development Journal. Vol.14, No.2.
Jindal, R. (2004) Measuring the socio-economic impact of carbon sequestration on local
communities: An assessment study with specific reference to the N’hambita pilot project
in Mozambique. MSc dissertation, School of GeoSciences, University of Edinburgh, UK
Jindal, 2000. Barawa Village Case Study, in , Conroy, C. (ed.) 2000. Silvipasture Management
Case Studies. NRI Report No. 2540, DFID, Natural Resources Institute,UK, Seva Mandir,
BAIF Development Research Foundation, India.
Kerr, J., C. Foley, K. Chung, and R. Jindal 2006. Sustainable Development under Clean
Development Mechanism: Opportunities and Concerns. Journal of Sustainable Forestry.
Vol.23(1)
Mercer, D.E. 2004. Adoption of agroforestry innovations in the tropics: A review. Agroforestry
Systems. Pg. 311-328
Michaelowa, A., and F. Jotzo. 2005. Transaction costs, institutional rigidities and the size of the
clean development mechanism. EnergyPolicy, 33: 511–523.
Miranda, M., I.T. Porras, and M.L. Moreno. 2003. The Social Impacts of Payments for
Environmental Services in Costa Rica: A quantitative field survey and analysis of the
Virilla Watershed. International Institute for Environment and Development (IIED), UK.
Mullan, K., A. Kontoleon, T.M. Swanson, and S. Zhang. 2010. Evaluation of the Impact of the
Natural Forest Protection Program on Rural Household Livelihoods. Environmental
Management. 45: 513-525
Nkamleu, G.B., and V.M. Manyong. 2005. Factors affecting the Adoption of Agroforestry
Practices by Farmers in Cameroon. Small-scale Forest Economics, Management and
Policy. Vol. 4(2), Pg. 135-148.
Pagiola, S., A.R. Rios, and A. Arcenas. 2008. Can the poor participate in payments for
environmental services? Lessons from the Silvopastoral Project in Nicaragua.
Environment and Development Economics 13: 299–325.
Pagiola, S., A. Arcenas, and G. Platais. 2005. Can Payments for Environmental Services Help
Reduce Poverty? An Exploration of the Issues and the Evidence to Date from Latin
America. World Development, Vol .33 (2), Pg. 237-53.
Pattanayak, S., D.E. Mercer, E. Sills, and J.Yang, J. 2003. Taking stock of agroforestry adoption
studies. Agroforestry Systems, vol. 57, pp. 173-186.

145

Perez, C., C. Roncoli, C. Neely, and J.L.Steiner. 2007. Can carbon sequestration markets benefit
low-income producers in semi-arid Africa? Potentials and challenges. Agricultural
Systems, 94: 2–12.
Rajasekharan, P. and S. Veeraputhran .2002. Adoption of intercropping in rubber smallholdings
in Kerala, India: A tobit analysis. Agroforestry Systems 56: 1–11.
Sambane, E. (2005) Above ground biomass accumulation in fallow fields at the N’hambita
Community, Mozambique. MSc dissertation, University of Edinburgh, UK.
Sedjo, R., G. Marland, and K. Fruit (2001) Renting carbon offsets: the question of permanence.
Resources for the Future, Washington DC.
Simler, K. R., S. Mukherjee, G.L. Dava, and G. Datt .2004. Rebuilding after War: Microlevel
Determinants of Poverty Reduction in Mozambique. Research Report 132, International
Food Policy Research Institute (IFPRI), Washington D.C., USA.
Tipper, R.(2008) Template for Plan Vivo Technical Specifications on Avoided Deforestation
Conservation of miombo woodland in central Mozambique. Edinburgh Center for Carbon
Management (ECCM), UK http://www.miombo.org.uk/Documents.html
Tipper, R. 2002. Helping Indigenous Farmers to Participate in the International Market for
Carbon Services: The Case of Scolel Te, in Pagiola, S., J. Bishop, and N. Landell-Mills
(eds.) 2002. Selling Forest Environmental Services: Market-based Mechanisms for
Conservation and Development. Earthscan, London.
Uchida, E., J. Xu, Z. Xu, Z., and S. Rozelle. 2007. Are the poor benefiting from China’s land
conservation program?. Environment and Development Economics, Vol.12, Pg. 593–620.
United Nations Environment Program (UNEP) 2002. The Clean Development Mechanism.
UNEP Collaborating Centre on Energy and Environment, Risø National Laboratory,
Roskilde, Denmark.
University of Edinburgh (UOE) (2002). N’hambita Pilot Project: Project Proposal. Project
Proposal submitted by the University of Edinburgh (UOE), UK, to the European
Commission.
Williams M, C. M. Ryan, R.M. Rees, E. Sambane, J. Fernando, and J. Grace. 2008. Carbon
sequestration and biodiversity of re-growing miombo woodlands in Mozambique. Forest
Ecology and Management’ vol.254, Pg. 145–155.
Wooldridge, J.M. 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press,
Cambridge.
Wunder, S. 2005. Payments for Environmental Services: Some Nuts and Bolts. CIFOR
Occasional Paper No. 42. Center for International Forestry Research, Bogor, Indonesia.
146

Liu, C., J. Lu, and R. Yin. 2010. An estimation of the effects of China’s Priority Forestry
Programs on Farmers’ Income. Environmental Management. 45: 526-540.
Zolho, R. (2005) Effect of Fire Frequency on the Regeneration of Miombo Woodland in
N’hambita, Mozambique. MSc thesis, University of Edinburgh, UK.

147