1) A study analyzed the impact of a castor outgrower scheme in Ethiopia on farm household decisions and food security.
2) Preliminary results from an endogenous switching regression model found that participation in castor production was positively associated with owned land size and a pre-program asset indicator, and negatively associated with the proportion of the household's labor force.
3) For participating households, higher crop income was associated with more land, extension visits, and media access, while non-participating households saw higher income with a larger labor force and non-agricultural income sources.
3. LICOS
Biofuel development : controversial
Disadvantages:
price increase & volatility –worsens food security (IFPRI
2008; Mitchel 2008);
- 10% biofuel expansion in EU and NAFETA – a reduction in GDP
by 1% for most poor African countries. (FAO, 2008)
weak land governance institutions may favor investors–
risk to vulnerable hhs (Cotula et al 2010)
3
4. Advantages:
- biofuels boost growth - (Arndt et al 2011)
- using partial equilibrium analysis (Lashitew, 2011)
reported ‘food and fuel’ – complement eachother in
Ethiopia
- clean & cheaper energy source to remote rural areas (IIDA,
2008; FAO, 2008)
5. Evidence in current literature: no consensus
-largely focused on developed economies
-based on aggregate economic wide simulations or
qualitative studies
-actual impact analysis on smallholder farmers -
limited
6. LICOS
- what explains farm household's biofuel
crop adoption decisions?
- how participation decision affects food
security?
7. price hikes
Classification of liquid biofuels: &volatility have
– ethanol been attributed?
- biodiesel
Said to be less
Feedstock sources for liquid biofuel: food security
threatening?
– edible crops e.g. corn
- non-edible e.g. castor bean
Table: Inventory of biodiesel feedstock projects in Ethiopia (active in 2010)
Type of business No of Production Type of Total area (ha)
model project location feedstock
specialized
Total Under
allotted cultivation
(‘000 ha) (‘000 ha)
Plantations 4 SNNPR, Jatropha, 66.7 3.1
Oromia, Pongamia,
Beneshangul Castor
Outgrowers 1 SNNPR Castor _ _
PPP 1 Tigray Jatropha, 15 7
Candlenut,
Croton, Castor
8. Studied: castor outgrower scheme in Ethiopia
Castor
45% oil bearing seed
grows best in arid zones
1100~1600 m.a.s.l
poisnous – non food, for
chemical & biofuel industry
4-5 months maturity
(shifting is possible based on
market conditions)
good for soil fertility but bad to
biodiversity (invasive species)
9. foreign company contracting farmers to grow castor
common form of ‘input loan’ for ‘pay in output’ arrangement
allocate a maximum of ¼ but keep traditional crops on the
side – food security reasons
Castor has no other use in the area - default is minimal once
farmers make decision to grow
the remaining default is often – redirecting inputs to other
crops
thus contract farmers may in general record higher
productivity
11. LICOS
- 4 districts –that represent castor growing zones in
southern region
- all villages in altitude range of 1100– 2000 m.a.s.l.
covered by the program – included in our
sampling frame
- 24 villages randomly selected
- 18-21 households per village
- total of 478 household
- 30% participants (who received seeds &other
inputs)
- participation – allocated piece of land for castor &
entered contractual agreement w/t company
12. Source: FEWS, 2010
Policy – may direct biofuel projects to dry & arid
zones – to ease resource competition w/t food
14. Overall observation
- dissemination of the castor crop into inaccessible
& remote places
- widespread adoption rate (20-33%) in three years
of promotion -unlike low rate of other technology
adoptions in developing countries
- vast diversification (7 crops types on 0.81ha (3.2
timad) - adoption may interact with performance
of other crops
15. Descriptive (explanatory variables)
Participants Non-participants |t/chi-stat|
Household wealth variables
Owned land size (in ha) 0.93 0.72 3.54***
Own land per capita 0.15 0.13 1.00
Farm tools count (Number) 4.20 3.84 1.48
Proportion of active labour 0.49 0.51 0.99
Access related variables
Formal Media (TV/radio/NP) main info. source (1=yes) 0.27 0.18 1.73***
Fertilizer use(kg/ha) 33 24 9.0***
Borrowed cash money during the year (1=yes) 0.42 0.36 1.14
Distance from extension center (Minutes) 27.53 27.80 0.10
Contact with govt. extension agent (Number of visits) 12.63 11.08 0.98
Household characteristics
Gender of the HH head (1=female) 0.06 0.14 2.95***
HH head attended school (1=yes) 0.60 0.50 1.67*
Family size 6.87 6.10 2.98***
* p<.1; ** p<.05; *** p<.01
16. Descriptive – welfare indicator variables
Outcome Variables Participants Non-participants Diff
Crop income ('000 Birr) 5.141 4.491 769 **
Per capita crop income 824 770 54*
Food gap (months) in 2010 1.02 1.58 - 0.56***
Food consumption per capita (birr) 534 458 75***
Total expenditure (‘000 Birr) 7.144 6.292 852*
Per capita expenditure 1130 1062 67*
* p<.1; ** p<.05; *** p<.01
Definition:
Food gap months - hh short of own stock & cash to buy food (lean seasons)
- easily memorable esp. for long periods of scarcity
- a decrease in value – improvement in food security
17. 3. Method
We analyzed using
- Endogenous Switching Regression (ESR) &
- Two Step Heckman selection (TEM)
Selection (di) – to participate or not participate in castor
Potential correlation (ui & ℇji)
Selection: (1)
Participants: (2)
Non-particips: (3)
18. Identification assumption to estimate using Heckman & ESR
the error terms in (1) , (2) and (3) are jointly normally
distributed
(assumption specification varies slightly b/n the two models)
Adding exclusion restriction –makes estimates more robust
Excluded variables
village level past adoption rate X eligibility criteria
past asset indicator (livestock holding in TLU)
pcorr significant for participation but not to income
19. Table: Estimation of treatment effects under ESR model
Using the info. contained in the distribution of the error terms
The model allow us to get predictions of the counterfactuals
Selection decision Treatment
Participation Non-participation effect
Participant (a) E(𝑦1𝑖 𝑑 𝑖 , 𝑥 𝑖 = 1 (c) E(𝑦1𝑖 𝑑 𝑖 , 𝑥 𝑖 = 0
= =
households 𝛿 𝜀1𝑢 𝜙 𝑧𝑖 (a)-(c) = TT
𝛽1 𝑋1𝑖 + 𝛿2 Φ 𝑧𝑖
𝑢 𝛿 𝜀2𝑢 𝜙 𝑧𝑖
𝛽2 𝑋1𝑖 +
𝛿2
𝑢 Φ 𝑧𝑖
Non-participant (b) E(𝑦2𝑖 𝑑 𝑖 , 𝑥 𝑖 = 1 (d) E(𝑦2𝑖 𝑑 𝑖 , 𝑥 𝑖 = 0 (b)-(d) = TU
= =
households
𝛿 𝜀1𝑢 𝜙 𝑧𝑖 𝛿 𝜀2𝑢 𝜙 𝑧𝑖
𝛽1 𝑋2𝑖 − 𝛽2 𝑋2𝑖 −
𝛿2
𝑢 1−Φ 𝑧 𝑖 𝛿2
𝑢 1−Φ 𝑧 𝑖
Source: Verbeek, 2009
where Φ = is the standard normal cumulative function of the selection equation distribution
and ϕ = - standard normal probability density function of the distribution
20. 4. Preliminary results
4.1A. Crop income determinants – endogenous switching regression
Table 1– Switching regression estimation (Joint participation selection & crop income determinants)
Selection Non-
Variable (Jointly estimated Probit) Participants participants
Per capita owned land size (ha) 6.91** 7.76*** 4.60***
Per capita owned land size squared -10.37* -7.63* -2.78**
Pr of maize before planting made (in birr) -0.42** 0.31 0.01
Media (1= main info source) 0.31** 0.24 0.07
Family member with non agri inc source (1=yes) -0.14 0.13 0.17*
Log of number of govt. extension visits 0.02 0.12** 0.14***
Log of number of social contact and freinds -0.17** 0.06 0.00
Log of distance from extension center -0.05 0.10 0.10
Proportion of labour force -0.53 -0.10 0.38**
Gender of the head (1=Female) -0.44* 0.15 -0.4
Household head attended school (1=yes) 0.25 0.02 0.05
Log of number of enset trees 0.02 -0.03 -0.13
Age of the head (years) 0.04 -0.02 -0.00
Age squared 0.00 0.00 0.00
EligabilityXintensity indicator 0.12
Pre program asset indicator 0.67**
District dummies yes yes yes
_cons -0.74 3.75*** 4.20***
prob >F/ chi2 0.000
Pseudo R2 0.11
ρ 0.13* -0.50***
LR test of independent equations (prob>chi2) 0.05
N 467
* p<.1; ** p<.05; *** p<.01
21. Table 4.1B : Average expected crop income for castor adopters and non-adopters
Decisions stage Treatment
Effect
(TT/TU)
Sub-sample To participate Not to participate
Log per capita annual crop income (birr)
Households who participated (a) 6.37 (c) 5.93 (treated) 0.44***
Households who did not participate (b) 6.06 (d) 6.16 (untreated) -0.10***
• Average crop income gain of participants (TT)is 44%
• Non-participants would have lost 10% had they enter into contract
• Suggests households selected into where they be better off
22. 4.1C. Crop income determinants – standard Heckman two steps
Heckman two step
Probability to (treatement effect
Dependent var. log per capita crop income participation model)
Participation 0.57***
Per capita owned land size (ha) 8.12*** 4.61***
Per capita owned land size squared -12.66** -2.65
Pr of maize before planting made (in birr) -0.32* 0.06
Media (1= main info source) 0.27* 0.03
Family member with non agri inc source (1=yes) -0.22 0.19**
Log of number of govt. extension visits -0.08 0.08*
Log of number of social contact and freinds -0.18** 0.05
Log of distance from extension center 0.00 0.09**
Proportion of labour force -0.4 0.21
Gender of the head (1=Female) -0.42* 0.03
Household head attended school (1=yes) 0.24 -0.11
Log of number of enset trees 0.02 -0.02
Age of the head (years) 0.03 -0.01
Age squared 0.00 0.00
EligabilityXpast part.rate indicator 0.09**
Pre program asset indicator 0.54
_cons -0.22 4.59***
District dummies yes yes
ρ -0.61***
23. 4.2 Food gap months
Table 2A– Switching regression estimation (dependent=Food gap months in last 12 months)
Variable Participants Non-participants
Log of asset value per capita -0.47** -0.55**
Owned land size per capita (ha) -4.88** -4.67**
Polygamy family (1=yes) 0.43* 1.12*
HH head attended school (1=yes) -0.46 -1.46***
Family member with non agri inc source (1=yes) -0.78 -0.80*
Log of number of social contact and friends 0.52 0.67**
District dummy yes yes
_cons 1.82 2.04**
Table 2B: Average expected food gap months for castor adopters and non-adopters
Decisions stage Treatment
Effect
Not to (TT/TU)
Sub-sample To participate participate
Log per capita annual food gap (no. of months)
Households who participated (a) 1.84 (c) 2.42 (treated) -0.58***
(-16 days)
Households who did not participate (b) 3.05 (d) 2.31 (untreated) 0.24***
(22 days)
24. 4.3 Consumption and expenditure effects
Table 4.3A: Average expected consumption for castor adopters and non-adopters
Decisions stage Treatment
Effect
Sub-sample To participate Not to participate (TT/TU)
Log per capita annual food consumption (birr)
Households who participated (a) 6.06 (c) 5.46 (treated) 0.39***
Households who did not participate (b) 5.42 (d) 5.84 (untreated) -0.35***
Table 4.3B: Average expected expenditure for castor adopters and non-adopters
Decisions stage Treatment
Effect
Sub-sample To participate Not to participate (TT/TU)
Log per capita annual exp. (birr)
Households who participated (a) 6.60 (c) 6.58 (treated) 0.02
Households who did not participate (b) 6.30 (d) 6.52 (untreated) -0.22***
25. 4.5. Factors that explain participation
Table 4.5. : Selection into participation
ESR Probit Tobit
Jointly estimated (Liklihood to Probit (Liklihood of
Selection equation adopt castor) dy/dx allocating an
Variable (Probit) extra ha of land)
Per capita owned land size (ha) 6.91** 5.25*** 1.60*** 6.31***
Per capita owned land size squared -10.37* -7.40** -2.26** -8.94*
Pr of maize before planting made (in birr) -0.42** -0.39** -0.12** -0.40**
Gender of the head (1=Female) -0.44* -0.45* -0.14* -0.48*
Household head attended school (1=yes) 0.25 0.25 0.08 0.27*
Log of number of social contact and freinds -0.17** -0.17** -0.05** -0.18**
Media (1= main info source) 0.31** 0.32** 0.10** 0.31*
Pre program asset indicator (livestock in TLU) 0.67** 0.71** 0.09** 0.40*
Farmers choice indicator 0.12 0.15* 0.05* 0.13*
Log of distance from extension center -0.05 -0.06 -0.02 -0.07
Log of number of extension visits 0.02 0.02 0.01 0.02
Log of number of enset trees 0.02 0.03 0.01 0.03
Family member with non agri inc source (1=yes) -0.14 -0.15 -0.05 -0.15
Age of the head (years) 0.04 0.04 0.01 0.05
Age squared 0.00 0.00 0.00 0.00
Proportion of labour force -0.53 -0.46 -0.14 -0.5
District dummies yes yes yes
_cons -0.74 -0.1 -0.13
prob >F/ chi2 0.000 0.000
Pseudo R2 0.10 0.11
N 467
* p<.1; ** p<.05; *** p<.01
26. Factors positively associated with adoption
Assets – land & livestock
Formal media (+10%)
Negatively associated with adoption
Land squared
Price of maize (main food crop) (-12%)
Female
More social contact
Non-associated
Government extension visit
Distance from village center
27. Effect of participation:
Impact is heterogeneous - implying presence of rational sorting
Castor growers gain from participating which they would not otherwise
Policy implication : grant farmers more choice
: as farmers with comparative adv. will engage in biofuel supply chain
Determinant of adoption:
HH assets are key factors for adoption
Adoption of biofuel declines with price of food crop
Physical accessibility showed no significance unlike most studies
Policy implication: privately organized techn. transfer –may efficiently surpass physical barriers
-Arndt etal--Biofuelcancreat ‘growthlinkages’ – i.e. demandforother sectors, thuswage and incomeincrease, thusspending …spillovers throughconsumerdemand….Biofuelinvestement increases significantly domestic product (GDP) in Mozambique and Tanzania by between 0.2 and 0.4 percentage points - Lahitew,didsimilarpolicysimulationexcerciseforEthiopia …rise in oilprices …biofuel is predicted to increase in itsproduction…&alongwithreduction in processing cost –biofuel has lesstendency to crowd out agriculturalproduction in Ethiopiagiven –priceincreasewillcreat more demandforotheragriculturalproducts – and thatenhancesproductivity –sincethere is largepotentialforproductivitygrowth in Ethiopia-
The outgrowerscheme – we looked is similarwithcommonform of input loanfor output payarrengement. Farmers are contracted to allocate part of their land &labour, the companylendsthemfertilizer and supervision
- The domesticchain is short – farmers delivarunsheldcastor to village stores, companytransportsit the Zone towncollection center, companyusessimplemill to remove the sheld, and exports the seeds to china.
Family size, land size, and fertilizerapplication are significantlyhigherfor participant groupsParticipants are mostly men, & educatednodifference in land size per capita and suggestingthatparticipants are notassetrichon a per capita bases- Proportion of participantsowning mobile phones are lessthan NP althoughnot significant - -distance to anextenstion center are not different among the twogroups-participants shows smalldifference in farm tools, extenstion contact
- Cropincome is significant portion of totalincome. The remainingincome types consitutessmalldividedshareeachbut are a source of substantialvariation in the totalincome (close to a double standarddevation). - To assessbiofuelcropinduceddifference we choose to start lookingcropincomedifference, sincecastor is alsoannualcrop and cropincomeonlyconstitutesannualcrops…analysingcropincomegains/lose is important.-For thatreasonincome is best approximatedbyexpenditurewhich is less variant.-food gap
Dummy participationpositionvariablecontinousoutcomevariable – log of percapitacropincome, foodconsumption and expenditurefood gap is excluded in thisanalsysisbecause the outcomevariable is notcontinous –but is censerodvariable at 0, manyreported 0 food gap soi’llnot talk aboutit, estimationapproach is different forlimiteddependent variablesIn such a setting, ATE and TT are the common effect we wouldlike to quantifyATE-measure the effect of the program on a random personfrom the population, while TT quntifies the effect on a random personfrom the treatedpopulationDependingwhat effect we wouldlike to quantify and assumption we wouldlike to bearthen, we have different availableoptions of methods
Dummy participationpositionvariablecontinousoutcomevariable – log of percapitacropincome, foodconsumption and expenditurefood gap is excluded in thisanalsysisbecause the outcomevariable is notcontinous –but is censerodvariable at 0, manyreported 0 food gap soi’llnot talk aboutit, estimationapproach is different forlimiteddependent variablesIn such a setting, ATE and TT are the common effect we wouldlike to quantifyATE-measure the effect of the program on a random personfrom the population, while TT quntifies the effect on a random personfrom the treatedpopulationDependingwhat effect we wouldlike to quantify and assumption we wouldlike to bearthen, we have different availableoptions of methods