impact of small-holders’ cattle fattening on household...

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

173

Impact of Small-Holders’ cattle Fattening on HouseholdIncome generation in Fadis District of Eastern HarargheZone, oromia, Ethiopia

Jafer Mume Ahmed 1* and Fikadu Tadesse Gute 2

Keywords: cattle fattening, Logit re-gression, propensity scorematching, household income

Received: 31 October 2017,Accepted: 21 April 2018 At the household level, livestock plays a critical economic

and social role in pastoralists and at the household level,livestock plays a critical economic and social role in pastoralistsand smallholder farm households. The objectives of this studywere to analyze factors affecting participation in cattle fatteningand its impacts on household income in Fadis district of EasternHararghe. Both primary and secondary data were used. Thedata were collected by means of a semi-structured questionnairefrom 124 samples during the period of April 20-May20/ 2017.Logit estimation revealed that participation in cattle fattening issignificantly influenced by five variables. Age of householdhead, labor force in family member, market information, accessto agricultural extension services and number of livestock aresignificant variables which affect the participation of thehousehold in cattle fattening practices. Propensity score matchingmethod was applied to analyze the impact of the cattle fatteningon the household income generation. In matching processes,kernel matching with 0.25 band width was resulted in relativelylow pseudo-R2 with best balancing test was found to be the bestmatching algorithm. This method was checked for standardizedbias, t-test, and joint significance level. Propensity scorematching results revealed that household participated in cattlefattening practice have got 14,071 more farm income and12,617 total household income in Ethiopian Birr (ETB) thanthose household that were not participated in fattening practices.This income difference shows how non-farm and off-farmincome compensated for income obtained from cattle fatteningactivities with farm income.

Abstract

International Journal of Agricultural Management and Development (IJAMAD)Available online on: www.ijamad.iaurasht.ac.irISSN: 2159-5852 (Print)ISSN:2159-5860 (Online)

1 Agricultural Economics, Oromia Agricultural Research Institute, Fadis Agricultural Research Center, Addis Ababa, Ethiopia2 Agronomist, Fadis Agricultural Research Center, Oromia Agricultural Research Institute Addis Ababa, Ethiopia* Corresponding author’s email: [email protected]

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

174

IntroDuctIonWorld meat production is anticipated to record

a modest expansion in 2015 to 318.7 milliontones, 1.3 percent, or 4 million tones, above2014. Even if, the cattle population in themajority of tropical country is higher, there is astrong unsatisfied demand, due to the incrementof population growth in the majority of tropicalcountries, for milk and meat (FAO, 2015).

Livestock is an integral part of Ethiopia’sagricultural sector and plays a vital role in thenational economy. At present, livestock con-tributes about 20% of the growth domesticproduct (GDP), supporting the livelihoods of70% of the population and the sub sector alsoaccount 11% of annual export earnings (SPS-LMM, 2010). Ethiopia is endowed with largestlivestock production, which ranks first in Africaand tenth in the world, it has much to gain fromthe growing global markets for livestock products.It is also known that Ethiopia is characterizedby a high livestock population with lowproduc-tivity of animal products, in terms of conventionalproducts such as meat and milk. Despite thelarge number of livestock, there has been adecline in national and per capita production oflivestock, livestock products, export earningsfrom livestock and per capita consumption offood from livestock (CSA, 2013). Meat pro-duction and consumption is important in theEthiopianeconomy and ruminants contributeover 3.2 million tons, representing over 72% ofthe total meat production (Belete et al., 2010).

Livestock production is of strategic economicimportance, not only for its number and diversity,but also for majority of the rural people uselivestock for various other activities like farmingand transportation of people and products(MoARD, 2006). In areas where mixed farming(crops and livestock production) undertaken,farmers use livestock for coping with adversesituations during crises of crop failure by sellinganimal products, as 72 percent of the householdsown cattle. With regard to direct food supplyand/or cash income generation, livestock playan increasingly significant role (MoARD, 2007).

At the household level, livestock plays acritical economic and social role in the lives of

pastoralists, agro-pastoralists, and smallholderfarm households. In the case of smallholdermixed farming systems, livestock provides nu-tritious food, additional emergency and cashincome, transportation, farm outputs and inputs,and fuels for cooking food. The governmentrecognizes the importance of livestock in povertyalleviation and has increased its emphasis onmodernizing and commercializing the livestocksub-sector in recent years. Eastern Hararghe iswell known for its best practices and indigenousknowledge in cattle fattening. Enhancing theproduction and productivity in the area withavailable indigenous technical knowledge willhelp the improvement of the sector in increasingthe sector contribution to national and agriculturalGDP. The subsectors contribute about 16.5% ofnational Growth Domestic Product (GDP) and35.6% of agricultural GDP (Metaferia et al. 2011).

The livestock production system in EastHararghe is market oriented. In the study area,there is little information available on determi-nants of cattle fattening and impacts of small-holder cattle fattening on households’ incomegeneration. Fattening is commonly practicedby some farmers in different places of the area.Farmers keep a small number of oxen whichare mainly purchased from market, fattened andsold for beef after a few month of work. There-fore, to plan and develop improved cattle fatteningand information sharing is very important toidentify the existing cattle fattening practices,determinants of cattle fattening and its impactson household income generation in selectedstudy area. So the specific objectives of thestudy were to identify factors affecting small-holders cattle fattening practice and analyzeimpacts of cattle fattening on household incomegeneration in the study area.

MAtErIALS AnD MEtHoDSDescription of the study area

The study was conducted in Fadis districts ofeastern Hararghe zone of Oromia region. It isfound in around 30 km distance from Harartown. The climate of the area is characterizedby warm and dry weather with relatively lowprecipitation. Agriculture is the major source of

Impact of Small-Holders’ Cattle Fattening ... / Mume Ahmed and Tadesse Gute

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

175

livelihood of the community. However, its pro-ductivity is dependent on the merit of rain-fedagriculture. The farming system is subsistencetype dominated by smallholder farmers. Sorghumand maize crops take the largest proportion ofcrop production. Similarly, chat and groundnutare also the main cash crops in the area. Eventhough livestock keeping constitutes an importantactivity, many households lost their livestockassets due to recurrent drought.

Fedis district is also found at latitude between8°22’ and 9°14’ north and longitude between42°02’ and42°19’ east, in middle and low landareas: altitude range is from 1200-1600 m.a.s.lmeters,witha prevalence of low lands. The areareceives average annual rain fall of 400 - 804mm. The minimum and maximum temperatureof the area is 20-25°C and 30-35°C, respectively(EHZARDO, 2015). The population’s livelihoodmainly consists of agriculture, husbandry andsmall-scale trade. The farm units are smallfamily holdings with an average agriculturalland area of less than one hectare. Agricultureis mainly rain fed. Similar to areas in the Hornof Africa, two rainy seasons characterize theFedis district’s climate: the first, named Belg,is the shortest one and takes place betweenMarch and May, while the second and most im-portant,named Meher, is between July and Oc-tober. The rainfall distribution during the yearis then bi-modal, with a dry spell period duringthe months of June and July, depending on itsduration, may affect crop growth. The Meher(Main) season is the most important one; whenthe intensity of farm practices and productionincrease.

Sampling technique and method of data collectionBoth primarily and secondary data sources

were used for this study. The data required forthis study were collected from sample respondentsusing a questionnaire. One day tutorial wasgiven to the enumerators about method of datacollection and the contents of the questionnaire.Secondary data that could supplement the primarydata were collected from published and unpub-lished documents obtained from Eastern Hararghezone. Total rural kebele in selected districts

were identified and arranged. The total ruralkebeles that are found in the Fadis district werecategorized. Total sample size for each kebelewas categorized as cattle fattening participantand non-participant for each sampled kebele.Toselect sample respondents from selected kebeles,first the household heads in the sampled kebeleswas identified and stratified in to two strata:cattle fattening participant and non-participant.Then the samples from each stratum wereselected randomly using simple random samplingtechnique. Since the number of household headsin the two groups was almost proportional,related number of sample was drawn from eachgroup, i.e. 70 participants and 54 non-participantswere selected. Then total of 124 respondentswere interviewed using questionnaire

Data analysis Based on the objectives of the study, both de-

scriptive statistics and econometric models wereemployed to analyze both qualitative and quan-titative data. From econometric model, logitmodel was applied to analyze factors affectingsmall-holder cattle fattening and propensityscore matching method (PSM) was also usedfor impact analysis.

Descriptive statisticsBy applying descriptive statistics, one can

compare and contrast different categories ofsample units with respect to the desired charac-teristics. It is used to explain the different so-cio-economic, institutional and other character-istics of the sample households. These includemean, percentage, standard deviation and fre-quency for fattening participants (treated group)and non-participants (non-treated group) farmers.

Households’ income measure Annual household income included both agri-

cultural (farming and non-farming) and non-agricultural off-farm incomes. The non-agricul-tural or income obtained from off-farm activitieswas considered because, income that could beobtained from cattle fattening activity can becompensated by non-agricultural or off-farmactivities. The contribution of cattle fattening


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

176

to household income might be exaggerated ifthe inclusion of non-agricultural or income ob-tained from off-farm activities is ignored. Itmeans that if the household income from non-agricultural or off-farm activities is omitted andonly agricultural income is considered the shareof income obtained from cattle fattening activitiesmight be higher than when income from bothagricultural and non-agricultural or off-farm ac-tivities are considered. Therefore, as much aspossible, it is plausible to include every sourcethat can generate income to household.

Econometrics analysisEconometric analysis for factors affectingparticipation in cattle fattening

The logit and probit are the two most commonlyused models for assessing the effects of variousfactors that affect the probability of cattlefattening of a given practice. These models canalso provide the predicted probability of cattlefattening practice. Both models usually yieldsimilar results. However, the logit model is sim-pler in estimation than probit model (Aldrich &Nelson, 1984). Hence, the logit model was usedin this study to analyze the determinants ofsmall-holders’ cattle fattening. Following Liao(1994), Gujarati (2003) and Aldrich and Nelson(1984) the logistic distribution function for thepractices of small scale cattle fattening:

(1)

where, Pi = is a probability of practicingsmall-scale cattle fattening for the ith farmerand it ranges from 0-1.

ezi = stands for the irrational number e to thepower of Zi.

Zi = a function of n-explanatory variableswhich is also expressed as:

Zi = B0+B1X1+B2X2+…+BnXn (2)

where, X1, X2… Xn are explanatory variables. B0- is the intercept, B1, B2 …; Bn are the logit

parameters (slopes) of the equation in the model.The slopes tell how the log-odds ratio in favor

of practicing small-holder cattle fattening changesas an independent variable changes. The unob-servable stimulus index Zi assumes any valuesand is actually a linear function of factors influ-encing decision of small-holder cattle fattening.It is easy to verify that Zi ranges from -∞ to ∞,Pi ranges between 0 and 1 and that Pi is non-linear related to the explanatory variables, thussatisfying two requirements:

As Xi increases Pi increases but never stepsoutside the 0 and 1 interval; and

The relationship between Pi and Xi is non-linear, i.e., one which approaches zero at slowerand slower rates as Xi gets small and approachesone at slower and slower rate as Xi gets verylarge. But it seems that in satisfying these re-quirements, an estimation problem has beencreated because Pi is not only non-linear in Xibut also in the B’s as well, as can be seenclearly below.

(3)

This means the familiar OLS procedure cannotbe used to estimate the parameters. But thisproblem is more apparent than real because thisequation is intrinsically linear. The interpretationof logistic regression coefficients (bi) is consideredby using odds ratio (Pi/(1-Pi ) and log of theodds ratio ln (Pi/(1-Pi) (Liao, 1994). The oddsvalue gives the expectedchange in the oddsratio of adopting the given farm activity versusnot adopting it per unit change in an explanatoryvariable, other things being equal. The same in-terpretation applies to both dummy and contin-uous variables (Liao, 1994). In this study, if Piis the probability of practicing a given small-holders’ cattle fattening then (1-Pi), the probabilityof not practicing, can be written as:

1-Pi=1/(1+eZi)

Therefore, the odds ratio can be written as:

Pi/(1-Pi) = (1+eZi)/(1+e-Zi) = eZi

Now Pi/(1-Pi)is simply the odds ratio in favor


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

177

of practicing small-holder cattle fattening. It isthe ratio of the probability that the farmer wouldpractice the cattle fattening to the probabilitythat he/she would not adopt it. Finally, takingthe natural log of equation 5, the log of oddsratio can be written as:

where, Li is log of the odds ratio in favor ofsmall-holder cattle fattening practices, which isnot only linear in Xi, but also linear in the param-eters. Thus, if the stochastic disturbance term,(Ui), is introduced, the logit model becomes:

Zi=B0+B1X1+B2X2+…+BnXn+Ui (7)

Impact evaluation methods using propensityscore matching (PSM) method

The first step in PSM method is to estimatethe propensity scores. A logistic model is usedto estimate propensity scores using a compositeof pre-participation characteristics of the sampledhouseholds (Rosenbaum & Robin, 1983) andmatching is then performed using propensityscores of each observation. The propensityscores themselves serve only as devices tobalance the observed distribution of covariatesbetween the treated and comparison groups.The success of propensity score estimation istherefore assessed by the resultant balance ratherthan by the fit of the models used to create theestimated propensity scores (Lee, 2006). Usingpredicted probabilities of participation in a givenfarm program (i.e. propensity score)match pairsare constructed using alternative methods ofmatching estimators. In this study, to analyzethe factors affecting households’ participationin cattle fattening practice, dependent variableis dichotomous in nature and represents the ob-served cattle fattening. It was represented in themodel as treated group (CatFat) =1 for a house-hold that participated in cattle fattening andnon-participated=0 for a household that do notpractice cattle fattening. In this study a VarianceInflation Factors (VIF (Xi) technique was em-ployed to detect the problem of multicollinearity

for all explanatory variables as (Gujarati, 2003). The impact of small-holder cattle fattening on

household income generation is the differencein households’ mean of farm income of the par-ticipant farmers and non-participant farmers incattle fattening. Thus, the fundamental problemof such an impact evaluation is a missing dataproblem. Hence, this study applies a propensityscore matching technique, which is a widely ap-plied impact evaluation instrument in the absenceof baseline survey data for impact evaluation.According to Caliendo and Kopeinig (2005),there are steps in implementing PSM. Theseare estimation of the propensity scores, choosinga matching algorism, checking on common sup-port condition and testing the matching quality.Imposing a common support condition ensuresthat any combination of characteristics observedin the treatment group can also be observedamong the control group (Bryson et al., 2002).The common support region is the area whichcontains the minimum and maximum propensityscores of treatment and control group households,respectively.

For any cattle fattening practicing household,there should be non-practicing household withclosest propensity score as the match. To ac-complish the match, the nearest neighbor (equalweights version) was tested. The nearest neighbormethod simply identifies for each householdthe closest twin in the opposite fattening group.Caliper matching which means that an individualfrom the comparison (non-treated) group wasalso tested as a matching partner for a treatedindividual that lies within a given caliper (propen-sity score range) and is closest in terms ofpropensity score and kernel matching estimatorswas also tested. However, for this specific studykernel matching was used to evaluate impact ofcattle fattening on households income generation.This is matching method whereby all treated unitsare matched with a weighted average of all controlswith weights which are inversely proportional tothe distance between the propensity scores oftreated and controls Becker and Ichino (2002)Venetoklis (2004). It then computes an estimateof the cattle fattening effect as the average dif-ference in households’ outcome variable between


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

178

each pair of matched households. The impactof cattle fattening for an individual i, noted δi,is defined as the difference between the potentialoutcome in case of cattle fattening and the po-tential outcome in absence of small-holder cattlefattening group using PSM.

δi= Y1i - Y0i (8)

In general, an evaluation seeks to estimate themean impact of the cattle fattening practice isobtained by averaging the impact across all theindividuals in the population. This parameter isknown as Average Treatment Effect or ATE:

ATE= E(δ) = E (Y1 −Y0) (9)

Where E(.) represents the average (or expectedvalue). Another quantity of interest is the AverageTreatment Effect on the Treated or ATT, whichmeasures the impact of the treatment on thoseindividuals who participated:

ATT = E (Y1 −Y0 | D =1) (10)

Finally, the Average Treatment Effect on theUntreated (ATU) measures the impact that thetreatment would have had on those who did notparticipate in cattle fattening practice:

ATU = E (Y1 −Y0 | D = 0) (11)

The problem is that, all of these parametersare not observable, since they depend on coun-terfactual outcomes. For instance, using the factthat the average of a difference is the differenceof the averages, the ATT can be rewritten as:

ATT = E (Y1|D =1)−E (Y0| D =1) (12)

The second term, E(Y0| D =1) is the averageoutcome that the treated individuals wouldhave obtained in absence of treatment, whichis not observed. However, we do observe theterm E (Y0| D=0) that is, the value of Y0for theuntreated individuals.

ATT = E (Y1|D =1)− E( Y0| D =0) (13)

rESuLtS AnD DIScuSIonDescriptive statistics results Households’ demographic and socio-economiccharacteristics

Table 1 shows descriptive statistics results ofsample household based on participation insmall scale cattle fattening practices. In thestudy area the average age of all sample re-spondents was 39.14. On average participanthousehold head have 37.3 years while that ofnon-participants of cattle fattening have 41.48years. There is a significant difference in theirage years. The survey results showed that meandifference between participants households incattle fattening and non-participants were foundto be significant at 5 % significant level basedon household head age in years. Similarly, theaverage year of formal schooling of participantis around grade 3 while that of non-participantin cattle fattening is around grade 2. The meandifference of the two groups is statistically sig-nificant at 5% of probability level. It showsthat, on average participant household havemore year of formal schooling compared to thatof non-participants in cattle fattening practice.

Farm size refers to the total area of farmlandthat a farm HH owned in hectares. In agriculture,land is one of the major factors of production.The average cultivated land of all sample re-spondents was 1 ha. On average participanthousehold have 1 ha while non-participantshave 0.91ha.There is a significant difference intheir cultivated land size. The survey resultsshowed that mean difference between participantand non-participant in cattle fattening was foundto be significant at 5% significant level basedon cultivated land.

Livestock is very important asset in farmhousehold. In this study, the average livestockholding of sampled household is 1.89 in TLU.On average participant household have 2.17while that of non-participant in cattle fatteningis 1.52 in TLU. Participant households havelarger livestock compared to non-participanthouseholds. The survey result revealed that, themean difference between participant householdin cattle fattening and non-participant householdwas significant at 1% level of significance based


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

179

on livestock holding in tropical livestock unit.Similarly, the cattle fattening participants havea larger number of labor force compared tonon-participants. The average number of laborforce of participants was 3 persons and that ofnon-participant is 2 persons. The result showedthat, the mean difference between numbers oflabor forces of participants and nun-participantswere also found to be significant at 1% signifi-cance level.

The descriptive results in Table 2 revealedthat, based on the source of market informationfor agricultural production, sample respondentsthat dot accessed market information in the areaaccount for about 72.7 % of the total non-par-ticipant of the cattle fattening respondents; while

other group of the respondents that dot accessedmarket information accounts for 26.3 % of par-ticipants in cattle fattening in the area Table 4.Similarly, it showed that, sample respondentsthat accessed market information from devel-opment agent account for about 69.2 % of thenon-participant and 30.8 % of participants.Other group of non-participant that obtain marketinformation by observing other market participantin the market accounts for 33.3 % while that ofparticipant in cattle marketing accounts for 66.7%. Brokers and local farmers themselves alsoservice as the source of market information forother farmers in the study area. The comparisonof the two groups depicted that a higher proportionof respondents that access market information


All VariablesAll samplesHH(N=124)

ParticipantsHH(N=70)

Non-participantsHH(N=54)

Mean difference

Mean SD Mean SD Mean SD Mean t-value

Age of HHEducation of HHMarket distance Family Size labor forceFarm size in ha Qty ProducedLivestock(TLU)Fod short month

39.142.216.975.753.191.038.741.895.15

10.3712.2572.5900.1951.3890.6145.8921.2561.482

37.332.616.995.943.501.138.902.175.03

10.6152.5612.6292.1261.4320.6316.5101.2171.372

41.481.696.945.502.800.918.531.535.31

9.6441.6692.5642.2301.2340.5715.0311.2221.612

4.15-0.93-0.04-0.44-0.70-0.23-0.37-0.640.29

2.246**

2.312**

0.081.13

2.88***

2.06**

0.352.92***

1.07

*** p<0.01 and ** p<0.05

Table 1Socioeconomic Characteristics of Sampled Respondents

Source of info.Household categories on fattening

TotalNon-participant Participant

Non

DA

Market

Broker

Other

local farmers

Total

Count% within source of information Count% within source of information Count% within source of information Count% within source of information Count% within source of information Count% within source of information Count% within source of information Chi2 = 35.58, p-value = 0.000, df =5

2873.70

969.20

1033.30

26.70

133.30

440.00

5443.50

1026.30

430.80

2066.70

2893.30

266.70

660.00

7056.50

38100.00

13100.00

30100.00

30100.00

3100.00

10100.00

124100.00

Table 2Source of Market Information for Agricultural Product in the Study Area

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

180

are participants of cattle fattening practice thanthat of not-participant of the fattening. This dif-ference is found to be statistically significantand the association between access to marketinformation for agricultural product and partic-ipation characteristics of the sample respondentswas found to be significant at 1 percent probabilitylevel for cross tabulation chi-square test.

In moisture stress area of Eastern Hararghezone, farmers use different sources of incomegenerating activities to diversify their source ofincome. The descriptive results presented inTable 3 revealed that, out of total non-participantof cattle fattening practice, sample respondentsthat use chat as main source of income accountfor 66.7 % while other group account for 25.9%, 1.9%, 1.9 % and 3.7 % from groundnut pro-duction, chat trading, livestock trading and othersource of income generating activity, respectively.On the other hand, out of total participant ofcattle fattening practice, participant respondentsthat use chat as main source of income accountfor 40 % while other group account for 21.4 %,2.9 %, 1.4 %, 27.1 % and 7.1 % from groundnutproduction, chat trading, livestock trading, cattlefattening and other source of income generatingactivity, respectively. The comparison of the

two groups depicted that a higher proportion ofrespondents that use non-cattle fattening as theirmain source income are non-participants ofcattle fattening practice than that of participantof the fattening. This difference is shown bycross tabulation chi-square test that found to bestatistically significant and the associationbetween main source of farm household incomeand participation characteristics of the samplerespondents was found to be statistically signif-icant at 1 percent probability level.

In the study area of Eastern Hararghe zone,farmers are facing different agricultural productionconstraints that challenge them in one or otherways. The descriptive results presented in table4 above revealed that, out of total non-participantof cattle fattening practice, sample respondentsthat replied oxen shortage as the main productionconstraints account for 16.7 % while other groupaccount for 11.1 %, 55.6 %, 7.4 % and 3.7 % aslabor shortage, disease, drought, weed and short-age of farm land as main constraints of agriculturalproduction, respectively. On the other hand, outof the total participant of cattle fattening practice,sample respondents that replied oxen shortageas the main production constraints account for15.7 % while other groups account for 19 %, 7.1


Source of income

Household categories onfattening Total

Non-participant Participant

Chat/coffee production

Groundnut production

Chat trading

Livestock trading

Cattle fattening

Other

Total

Count% within HH categories on fattening Count% within HH categories on fattening Count% within HH categories on fattening Count% within HH categories on fattening Count% within HH categories on fattening Count% within HH categories on fattening Count% within HH categories on fattening % of totalChi2 = 19.92, p-value = 0.001, df=5

3666.714

25.91

1.91

1.90

0.02

3.754

10043.5

284015

21.42

2.91

1.419

27.15

7.170

10056.5

7051.629

23.43

2.42

1.619

15.37

5.6124100100

Table 3Main Source of Household Income in the Study Area

Source: Own survey results

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

181

%, 54.3 %, 8.6 % and 4.3 % as labor shortage,disease, drought, weed and lack of pesticideand herbicide as main constraints of agriculturalproduction, respectively. The comparison of thetwo groups depicted that proportion of respon-dents that faced different agricultural productionconstraints to non-participants of cattle fatteningpractice and that of participant of the cattle fat-tening are almost equal. This difference is shownby cross tabulation chi-square test that is foundto be insignificant and the association betweenthe main agricultural production constraints andthe participation characteristics of the samplerespondents was found to be insignificant by

probability level. This implies that, sample re-spondents are facing similar agricultural pro-duction constraints even if the level of challengediffers between both groups.

Non-agricultural or income obtained from off-farm activities was considered because, incomethat could be obtained from cattle fattening ac-tivity can be compensated by non-agriculturalor off-farm activities. The contribution of cattlefattening to household income might be exag-gerated if the inclusion of non-agricultural orincome obtained from off-farm activities is ig-nored. Therefore, both off-farm income andnon-farm income that obtained from both ac-


Production Constraints



Oxen shortage

Labor shortage

Disease

Drought

Weeds

Lack of pest & herb side

Shortage of Land

Total

% within HH categories on fattening % of total% within HH categories on fattening % of total% within HH categories on fattening % of total% within HH categories on fattening % of total% within HH categories on fattening % of total% within HH categories on fattening % of Total% within HH categories on fattening % of TotalCount% within HH categories on fattening Chi2 = 6.27, p-value = 0.39, df=6

16.77.35.62.411.14.8

55.624.27.43.20.00.03.71.654

100

15.78.9

10.05.67.14.0

54.330.68.64.84.32.40.00.070

100

16.116.18.18.18.98.9

54.854.88.18.12.42.41.61.6124100

Table 4Agricultural Production Constraints for Sampled Respondents in the Area

Access to off-farm activity



Not-Access

Yes-Access

Total

Count% within off-farm activity% of totalCount% within off-farm activity% of totalCount% within off-farm activity% of total

4041.232.314

51.911.354

43.543.5

5758.846.013

48.110.570

56.556.5

9710078.227

10021.8124100100

Table 5Respondents Access to off-Farm Activity to Generate Income in the Study Area

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

182

tivities were gathered from sample respondentsand analyzed. Table 5 revealed that, out of thetotal non-participants of cattle fattening 85.2 %did not participate in non-farm activities whereas14.8 percent of them participated in non-farmactivities. On the other hand, 82.9 percent ofparticipants in cattle fattening did not participatein non-farm activities while 17.1 percent ofthem participated in non-farm activities to gen-erate addition income for household. This impliesthat non-participants of cattle fattening mostlycovered their family expenditure by non-farmincome that can be obtained from non-farm ac-tivities as described in Table 6 below. The sameis true for off-farm activities to generate off-farm income.

This study focused on the income that householdgenerate by participating in cattle fattening. Thebenefits which they gain from doing so and theconstraints they face in successful income gen-eration will help to draw out the potential roleof fattening in achieving beneficial income gen-eration and identify the kind of interventionsneeded to support this. Total household income

used in this analysis was the sum of total farmincome, non-farm and off-farm income generatedby farm household in the year.

The result presented in Table 7 above showssignificant difference in their farm and off-farmincome. The survey results showed that meandifference between participant households incattle fattening and non-participants were foundto be significant at 10 and 1 percent significantlevel based on the respondents’ farm and off-farm income, respectively.

Participation in cattle fattening and bestpractice in the study area

Information documentation and sharing experi-ences among the stakeholders would be an effectivestrategy. Farmers demonstrate the best techniqueto farmers in other areas. Farmers should shareinformation on this best practice. Farmers in thesame area with a given resource may practice dif-ferently in different areas because of the lack ofinformation and other technical support.

In the study area, cattle fattening was found to beone of the household income diversification strategies


Access to off-farm activity

Household categories oncattle fattening Total


Non-Access

yes-Access

Total

Count% within non-farm activity% of totalCount% within non-farm activity% of totalCount% within non-farm activity% of total

4644.237.1

840.06.5054

43.543.5

5855.846.812

60.09.770

56.556.5

10410083.920

10016.1124100100

Table 6Respondents Access to Non-Farm Activity to Generate Income in the Study Area

All Variables All samplesHH(N= 124)

Participants HH(N=70)

Non-participantsHH(N=54)

Mean difference

Non-Farm IncomeOff-Farm Income Farm Income

Mean1689892

26791

SD56693275

15945

Mean1205533

34258

SD29342054

16058

Mean2315135817111

SD791443619232

Mean1110825

17147

T-Value 1.081.39*7***

Table 7Description of Sampled Participants and Non-Participants’ Income

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

183

that was used to minimize the drought risk thatleads to farm income lose in the area. Farmers inthe area use different techniques to reduce fatteningduration and increase benefit that obtained fromsold fattened cattle. This practice varies fromfarmers to farmers depending on the capacity andskill of the farmers. Some farmers start fattening inthe rainy seasons when animal feed like grass andshrubs are adequate in the area. Similarly, othergroups of the farmers start fattening when theirplot maize starts to be rape as to increase body ofcattle then reduce fattening duration. Other groupsof the farmers use industrial by-products as mainfeed and mix other supplementary feed that acceleratefattening. These supplementary feeds are mixedfrom sugar, fenugreek seed, maize powder, sorghumpowder and other left of their own food after meal.Best practice on what they feed, how they feedand cattle management practice for fattening cattlewas observed by other farmers as best practice.

Regarding why farmers select cattle fattening,respondents replied that 71.4 % of cattle fatteningparticipants replied that they select fatteningdue to its higher profit while 25.7 % of participantshave chosen for its short term income generation.Similarly, around 2.8 percent of the participantfarmers have selected cattle fattening for itssimplicity of management. On the other hand,participant farmers were using different sourcesof cattle fattening information. Around 51.4%of the participants replied that they obtain mostly

cattle fattening information from other farmersas the sources of information while 34.3% usedtheir neighbors as the source of fattening infor-mation. Other farmers replied as they useddistrict information while 12.9 % of the partici-pants replied as they used extension workersinformation as main sources.

In the study area, the average cattle fatteningexperience of participant farmer was found to bearound 6 years which ranges from 1 year to 14years. Similarly, the average cattle fattening durationof participant farmer was found to be around 3.5months which ranges from 2 months to 7 months.

Cattle production and fattening can significantlybenefit the farmers of the Eastern HarargheZone in general and Fadis district in particular.Availability of the best fattening competitionamong the farmers, suitable fattening weatherand good indigenous knowledge of fattening,higher demand for their cattle or popularity offattened Harar bull in the country were an op-portunity for cattle fattening in the study area.Farmers in the area were also participating incattle fattening due to the presence of Somaliatraders in the area and road facility to transportcattle in all direction. Farmers using improvedcattle management and fattening practices inreducing fattening duration because they werebelieved that this improved cattle managementwill improve their efficiency and increased theirincome generating opportunity in the study area.


Variable Obser. Mean SD Min Max

Fattening ExperienceAverage fattening month

7070

6.373.51

2.543371.05971

12

147

Table 8 Description of Fattening Experience and Average Fattening Month

Fattening opportunity Frequency Percent Cumulative percent

Non-participants with no response Access to RoadHigher demand for cattlePresence of Somalia tradersAccess to MarketOther opportunityTotal

5421241717

124

43.516.919.413.70.85.6100

43.560.579.893.594.4100

Table 9Description of Cattle Fattening Opportunity in the Study Area

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

184

The descriptive results presented in Table 10above revealed that, out of total participant ofcattle fattening practice, sample respondents thatreplied lack of good market as main cattlefattening constraints account for 10%while othergroups account for 51.43%, 15.71% and 22.86%as low price for cattle, lack of market informationand broker problem as main constraints of cattlefattening in the study area, respectively.

Regarding when participants sell their cattle,the participants sell their cattle at different timesfor various reasons. The descriptive results pre-sented in Table 11 above revealed that, out ofthe total participant of cattle fattening practice,sample respondents that replied they sell theircattle at fixed month of Eidul-Arefa (when alarge number of cattle are slaughtered by Muslimcommunity both in the country and Somaliarea) account for 14.29% while other groupsaccount for 17.14%, 62.86 % and 5.71% thatsell when fatten observed, depend on price riseandsell as soon as money required, respectivelyfor time of selling cattle in the study area.

results of econometric analysis for factoraffecting participation in cattle fattening

Before proceeding to analyze factors affectingsmall-holder cattle fattening, Variance InflationFactor (VIF) was applied to test for the presenceof strong multicollinearity problem among theexplanatory variables. There was no explanatory

variable dropped from the estimated modelsince no serious problem of multicollinearitywas detected from the VIF results. Similarly,heteroscedasticity was tested by using Breusch-Pagen test. This test resulted in rejection of theexistence of heteroscedasticity hypothesis as(p= 0.346) using STATA 11. The pseudo- R2 in-dicates how well the regresses explain the par-ticipation probability. After matching thereshould be no systematic differences in the dis-tribution of covariates between both groups andtherefore, the pseudo- R2 should be fairly low(Caliendo & Kopeinig, 2005).

It was found that participation in cattle fatteningwas significantly influenced by five explanatoryvariables. Age of household head, labor forcein family member, size of livestock in tropicallivestock unit, market information and accessto agricultural extension service are significantvariables which affect the participation of thehousehold in cattle fattening practice. Age ofhousehold head shows negative relation withparticipation in small scale cattle fattening prac-tice. This implies that an increase in age ofhousehold head tends to decrease participationin cattle fattening practice. This is possible be-cause older farmers have not been capable tomanage cattle for fattening and resist to expensesfor cattle. They lack the use of best practice andbetter planning than the younger ones. As theage of household head increases the probability


Market constraints Frequency Percent Cumulative percent

Lack of good marketLow price for Cattle Lack of Market InformationBroker problemTotal

736111670

1051.4315.7122.86100

1061.4377.14100

Table 10Market Constraints for Cattle Fattening Participants in the Study Area

Time of cattle sale Frequency Percent Cumulative percent

At fixed month(Eid - Adeha)When fatten observedDepend on price riseAs soon as money requiredTotal

1012444

70

14.2917.1462.865.71100

14.2931.4394.29100

Table 11When Cattle Fattening Participants Sell Their Cattle

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

185

of household participation in cattle fatteningpractice decreases. The interpretation of theodds ratio also implies that if other factors areheld constant, the odds ratio in favor of partici-pation in cattle fattening practice decrease by afactor of 0.93 as the age of the household headincreases by one year (Table 12).

In Ethiopia, as in most of other developingcountries, labor is one of the most extensivelyused inputs of agricultural production. Theseare household members found between age of15 and 64. Furthermore; family is the majorand sole source of agricultural labor. Householdswith large number of economically active mem-bers have more number of agricultural laborsand hence, have more agricultural productionand more income provided that there is sufficient

land to employ the existing labor. Cattle fatteningrequires a large number of labor force in ruralareas. Households that have a larger number ofworking group members were more likely to beincluded in small-scale cattle fattening practicein the study area. As it is reveled from estimationof the logit regression analysis indicates that,participation in cattle fattening has a positiveand statistically significant association with theuse of higher labor, most likely due to the higherlevel of labor requirement during managementand feeding activities involved in the cattle fat-tening. The interpretation of the odds ratio alsoimplies that if other factors are held constant,the odds ratio in favor of participating in cattlefattening increases by factor of 2.08 as thenumber of working family member increases


Variable Coef. Odds Ratio Std. Err Z

Age of HHSex of HHEducation of HHMrketDistance Family SizeLabor ForceFarm SizeLivestock(TLU)Market InformationAccess ExtensionQtty ProducedFood Shortage Month_cons

-0.068-0.0940.125-0.026-0.0760.7310.4100.3951.5371.0890.002-0.058-1.593

0.930.911.130.970.932.081.511.484.652.971.000.94

0.0250.6260.1220.0920.1420.2550.4160.2060.5330.5090.0420.1591.712

-2.7***

-0.151.03-0.28-0.532.87***

0.991.92*

2.89***

2.14**

0.04-0.37-0.93

Number of obser = 124 Pseudo-R2 = 0.291Log likelihood = -60.2398

LR Ch2(12) = 49.35Prob> Ch2 = 000

Table 12Logistic Regression Results for Factor Affecting Participation in Cattle Fattening

***, ** and * means significant at the 1%, 5% and 10 % probability levels, respectively

Figure 1. Kernel density of propensity score distribution

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

186

by one person.Households who have a larger number of live-

stock in tropical livestock unit were more likelyto be included in the small scale-cattle fattening.This variable was found to influence participationof household in cattle fattening positively andsignificantly. The implication of the result wasthat livestock are an important source of cash inrural areas to allow purchase of important feed,medicine and other management that can beused to reduce the duration of cattle fattening.Farmers who have a large number of livestockmight consider their asset base as a mechanismof insuring any risk associated with cattlefattening practice. Given this potential contri-bution of livestock to sustainable householdfarm input supply and cash generation, they en-courage adoption of best practice in cattle fat-tening. The odds ratio of 1.48 implies that,other things kept constant, the odds ratio in

favor of participation in cattle fattening increasesby a factor of 1.48 for each increase in livestockin TLU (Table 12). This implies that livestockholding has an influence on the adoption ofbest fattening practice in different areas.

Market information is a dummy variable taking1 if the respondents had access to market infor-mation and zero otherwise. It is hypothesizedthat updated market information is positivelyrelated to participation in cattle fattening practice(Table 12). Access to market information wasfound to influence participation of householdin cattle fattening positively and significantlyat 1 percent probability level. Keeping otherthings constant, the odds ratio in favor of par-ticipation in cattle fattening increases by a factorof 4.65 as a household has access to market in-formation service in the study area.

Access extension service is a dummy inde-pendent variable taking the value 1 if a household


Figure 3. Kernel density of propensity scores of non-participant households

Figure 2. Kernel density of propensity scores of participant households

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

187

had access to extension services and 0 otherwise.It is expected that farm extension service widenshousehold knowledge with regard to the use ofthe best farm technology that enhances householdincome generation activity. Access to extensionservices on cattle fattening such as feedingsystem, cattle management and other best practicein cattle fattening received by households posi-tively and significantly affected participation incattle fattening at less than 5 percent probabilitylevel. Holding other things constant, the oddsratio in favor of participation in cattle fatteningincreases by a factor of 2.97 as a household hasaccess to extension service.

Impact Estimationresults of propensity scores matching

The logistic regression model was used to es-timate propensity score matching for participantand non-participants households in cattle fattening.The dependent variable in this model is a binaryvariable indicating whether the household was aparticipant in cattle fattening or not. The model

was estimated with STATA 11.2 computing soft-ware using the propensity scores matching al-gorithm developed by Leuven and Sianesi (2003).Results presented in Table 12 above shows theestimated model appears to perform well forthe intended matching exercise. The pseudo-R2

value is 0.291. A low pseudo-R2 value showsthat participant households do not have muchdistinct characteristics overall and as such findinga good match between participants and non-treated households becomes simple.

Figure 1 portrays the distribution of the house-hold with respect to the estimated propensityscores. In case of participant households, mostof them are found in the right starting from themiddle of the distributed propensity. On theother hand, most of the control or non-participantsof cattle fattening households are partly foundin the center and with the most part of distributionfound in the left side.

Matching participant and non-participanthouseholds


Variables SamplesMean % reduce t-test

Treated Control %bias /bias/ t p>/t/

_pscore

Age of HH

Sex of HH

Eductn of HH

MrketDistnce

Farm Size

Livestock(TLU)

MrketInformtn

Access Extentn

Foodshortage month

Family Size

Labor Force

UnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatchedUnmatchedMatched

0.71570.630237.32938.137

0.857140.823532.61432.33336.98577.05881.131

1.06022.17081.9693

0.857140.80392

0.80.74515.02865.07845.94295.8039

3.53.0392

0.368540.5862941.48137.992

0.777780.848651.68522.49466.94447.1971

0.905090.991541.52631.9121

0.50.765360.592590.683915.31484.9978

5.55.67092.79632.8991

145.718.4-40.91.4

20.5-6.543

-7.51.6-5.337.511.452.94.7828.9

45.913.5-19.15.4

20.36.1

52.610.5

87.4

96.5

68.3

82.6

-235.2

69.6

91.1

89.2

70.5

71.8

70

80.1

8.031.02-2.250.071.14-0.342.31-0.370.09-0.242.060.572.920.244.640.472.570.68-1.070.271.130.312.880.61

00.3110.0260.9430.2550.7350.0220.7130.93

0.8080.0420.5730.0040.811

00.640.0110.4990.2880.79

0.2620.76

0.0050.5

Table 14Balancing Test for Covariate for Matched and Unmatched Group

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

188

Three main tasks were accomplished beforematching. First, predicted values of treatmentparticipation (propensity scores) estimated forall participated households and non-participants.Second, a common support condition was imposedon the propensity score distributions of participanthousehold in cattle fattening and non-participanthousehold. Third, discard observations whosepredicted propensity scores fall outside the rangeof the common support region.

Imposing a common support condition ensuresthat any combination of characteristics observed inthe participant group can also be observed amongthe non-participant group (Bryson et al., 2002).The common support region is the area whichcontains the minimum and maximum propensityscores of participants or treated and controlhouseholds, respectively. It requires deleting ofall observations whose propensity scores issmaller than the minimum and larger than themaximum of participant and non-participantgroup, respectively (Caliendo&Kopeinig, 2005).For this study, the common support regionwould lie between0.0741972 and 0.8992712.In other words, households whose estimatedpropensity score is less than 0.0741972 andlarger than 0.8992712 are not considered forthe matching exercise. As a result of this re-striction, 22 households (19 participant and 3non-participant households) were discarded.

Balancing test is a test conducted to knowwhether there is statistically significant differencein mean value of the two groups of the respon-dents and preferred when there is no significantdifference after matched.

Accordingly, matching estimators were eval-uated via matching the participant and non-par-ticipant households in common support region.Therefore, a matching estimator having balancedor insignificant mean differences in all explanatoryvariables, bears a low pseudo- R2 value and alsothe one that results in large matched sample size is

preferred for matching exercise. In line with theabove indicators of matching quality, kernel match-ing with 0.25 band width is resulted in relativelylow pseudo-R2 with best balancing test (all ex-planatory variables insignificant) and large matchedsample size as compared to other alternativematching estimators indicated in Table 13. Then itwas selected as a best fit matching estimator.

testing the balance of propensity score andcovariates

After choosing the best performing matchingalgorithm the next step is to check the balancingof propensity score and covariate using differentprocedures by applying the selected matchingalgorithm (in our case kernel matching). As indi-cated earlier, the main purpose of the propensityscore estimation is not to obtain a precise predictionof selection into treatment or to participation,but rather to balance the distributions of relevantvariables in both groups. The mean standardizedbias before and after matching are shown in thefifth columns of Table 14, while column sixreports the total bias reduction obtained by thematching procedure. In the present matchingmodels, the standardized difference in covariatebefore matching is in the range of 1.6% and 82%in absolute value. After matching, the remainingstandardized difference of covariate for almostall covariates lies between 1.4% and 13.5%which is below the critical level of 20% suggestedby Rosenbaum and Rubin (1985). In all cases, itis evident that sample differences in the unmatcheddata significantly exceed those in the samples ofmatched cases. The process of matching thuscreates a high degree of covariate balance betweenthe participant and non-participant samples thatare ready to use in the estimation procedure.

Similarly, t-values in Tables 14 shows that,before matching more than half of the chosenvariables exhibited statistically significant differ-ences while after matching all of the covariates


Sample Pseudo-R2 LR chi2 p>chi2

Unmatched Matched

0.2910.021

49.482.94

00.996

Table 15Chi-square Test for the Joint Significance of Variables

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

189

were balanced and become statistically insignificant. The low pseudo-R2 and the insignificant likelihood

ratio tests support the hypothesis that both groupshave the same distribution in covariates after matching(see Table 15). These results clearly show that thematching procedure is able to balance the character-istics in the participant and the matched non-participant groups. We, therefore, used these resultsto evaluate the impact of cattle fattening on outcomevariables among groups of households having similarobserved characteristics. This allows comparing ob-served outcomes for participants with those of com-parison groups sharing a common support.

Sianesi (2004) suggests re-estimating thepropensity score on the matched sample, i.e.only on participants and matched non-participants,then comparing the pseudo-R2 before and aftermatching is important. The pseudo-R2 indicateshow well the regressors explain the participationprobability. After matching there should be nosystematic differences in the distribution of co-variates between both groups and therefore thepseudo-R2 should be fairly low. The low pseu-do-R2 (compared with other pseudo-R2 resultedusing different matching estimators) and the in-significant likelihood ratio tests (indicated bythe higher p-value after matching) support thehypothesis that both groups have the same dis-tribution in covariates after matching. All of theabove tests suggest that the matching algorithmsthat have chosen were relatively best with thedata we have at hand. Thus, we can proceed toestimate ATT for households.

Estimating treatment effect on treated (Att)How additional income gained from cattle fat-

tening is used depends on household priorities.For those earning small amounts of income only,meeting basic household needs for food and otherexpenses such as healthcare is usually the priority.

With a little more income, households often makeimprovements in their standard of living by up-grading their dwelling in quality or size (e.g. con-structing a new roof or adding a second buildingfor son marriage), buying more and better qualityfood. All of these investments are likely to improvethe overall health and welfare of the household.Households may also invest in livestock as a keyasset for further insurance and income-generation.Another priority is often children’s education.

In order to solve the second objective, the fol-lowing impact indicators of the treatment effecthave been performed using propensity scorematching model. In this section, the PSM resultsprovides evidence as to whether or not the cattlefattening practice has brought significant changeson households’ total farm income and totalhousehold income (farm, off-farm and non-farm income) of households in Ethiopian Birr.The estimation result presented in Table 16 pro-vides a supportive evidence of statistically sig-nificant effect of the cattle on household totalfarm income and Total household income (Farm,off-farm and non-farm income) in ETB.

After controlling for pre-participation differ-ences in demographic, location and asset en-dowment characteristics of the participants incattle fattening and non-participants in cattlefattening households it has been found that, onaverage, the participant households’ have in-creased total farm income by 14071 ETB thanthat of non-participant households in cattle fat-tening. Similarly, the participant households’have increased total household income(farm,non-farm and off-farm income) of participatinghouseholds by 12671.4 ETB than that of non-participant households in cattle fattening. Thisdifference between farm income and total house-hold income shows that non-farm and off-farmincome generating activities area not equality


Variable Sample Treated Controls Difference S.E. t-stat

Total Farm Income

Total Hh Income

UnmatchedATT

UnmatchedATT

34257.8632615.6935684.934061.6

17110.718544.720043.721444.2

17147.1214071.0115641.1512617.40

2449.162702.582587.22949.8

75.2***

6.054.28

Table 16Average Treatment Effect on Treated (ATT)

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

190

undertaken by treated group and control groupof the respondents. Main source of non-farmincome activities include petty trading in thevillage, salary, chat trading, groundnut trading,shopping and black smith were some the activitiesreplied by sampled respondents.

Sensitivity analysisRosenbaum (2002) proposes using Rosenbaum

bounding approach in order to check the sensi-tivity of the estimated average treatment effecton treated ATT. The basic question to be answeredhere is whether inference about treatment effectsmay be altered by unobserved factors. In orderto control for unobservable biases shows theresult of sensitivity of cattle fattening impactson different income as outcome variables. De-pending on Rosenbaum bounds it was calculatedfor cattle fattening impacts that are positive andsignificantly different from zero. The resultshows those outcome variables which bear sta-tistical difference between participant and non-participant households in this impact estimation.Results show that the inference for the effect ofthe fattening is not changing though the participantsand non participant households has been allowedto differ in their odds of being treated up to (e γ =3) in terms of unobserved covariates as shown bysignificant outcome variables with p-critical values(or the upper bound of Wilcoxon signify level -Sig+) at different critical value revealed. Thatmeans for all outcome variables estimated, atvarious level of critical value of e γ, the p- criticalvalues are significant which further indicate thatwe have considered important covariates that af-fected both participation and outcome variables.We couldn’t get the critical value e γ where theestimated ATT is questioned even if we have setlargely up to 3, which larger value compared tothe value is set in different literatures. Thus, wecan conclude that our impact estimates (ATT) areinsensitive to unobserved selection bias and are apure effect of cattle fattening in the study area.

concLuSIonS AnD rEcoMMEnDAtIonSTo expand improved cattle fattening and mar-

keting systems, it is very important to identifythe existing cattle fattening practices and mar-

keting systems in the study area. Based on theempirical findings reported in this paper, thefollowing conclusion and recommendations areforwarded: The main objectives of this studyare to analyze factors affecting participation incattle fattening and its impacts on householdincome generation in Fadis District of EasternHararghe Zone. Both primary and secondarydata were collected for the study. The data werecollected by means of a semi-structured ques-tionnaire from 124 sample respondents duringthe period of April 20-May 20/2017.The mainresearch question of the study was “what wouldhave been to the total household income andfarm income if cattle fattening was not in place?”Hence, this study applies a propensity scorematching technique, which is a widely appliedimpact evaluation instrument in the absence ofbaseline survey data for impact evaluation. An-swering this question requires observing outcomesof participant after and before participation forthe household. Beside PSM, logistic model wasused to analyze the factors affecting participationin cattle fattening in the study area. The studyimplemented binary logit regression model toanalyse factors affecting participation in cattlefattening. Binary logit regression estimationalso revealed that participation in cattle fatteningpractice is significantly influenced by five ex-planatory variables. Age of household head,labor force in family member, number of livestockin tropical livestock unit, market informationand access to agricultural extension service aresignificant variables which affect the participationof the household in cattle fattening practice.

For this study, the common support regionwould lie between 0.0741972 and 0.8992712.In other words, households whose estimatedpropensity score is less than 0.0741972 andlarger than 0.8992712 are not considered forthe matching exercise. As a result of this re-striction, 22 households (19 participant and 3non-participant households) were discarded. Indoing so, propensity score matching has resultedin 51 participant households to be matched with51 non-participant respondents after discardinghouseholds whose values were out of commonsupport region. In other words, matched com-


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

191

parisons of different outcome of interest wereperformed on these respondents who sharedsimilar pre-participation characteristics exceptthe treatment participation effect. The resultingmatches passed on many process of matchingquality tests such as t-test, reduction in standardbias and chi-square test.

Propensity score matching method was appliedto analyze the impact of the small-holders’ cattlefattening on total household income and farm in-come obtained from only farm activities. In match-ing processes, kernel matching with 0.25 bandwidth is resulted in relatively low pseudo-R2with best balancing test was found to be the bestmatching algorithm. This method was checkedfor covariate balancing with a standardized bias,t-test, and joint significance level tests. Propensityscore matching method results also revealed thathousehold participated in cattle fattening practicehave increased total farm income by 14071 ETBthan that of non-participant households in cattlefattening. Similarly, the participant households’have increased total household income(farm,non-farm and off-farm income) of participatinghouseholds by 12671.4 ETB than that of non-participant households in cattle fattening. Theaverage treatment effect on treated was found tobe significant at less than 1% of significantlevel. The impact estimation results then indicatethat there are significant differences in participantsin cattle fattening and comparison households,which could be attributable to the participationin cattle fattening.

The number of economically active membersin the family was found to be positive and sig-nificant at 1% significant level with participationin cattle fattening practice. The model result alsorevealed that all other things being kept constant,the odds ratio in favor participation in cattle fat-tening practice increases by a factor of 2.08 asthe number of economically active member ofthe farm family increase by one person. In thefarm community cattle fattening activity requiresadequate number of labor force in rural area.The results of logit models shows a positive andstatistically significant relationship between cattlefattening and use of higher labor, most likely dueto the higher level of labor requirement during

cattle fattening management activities involved.Households’ who have a larger number of

livestock in tropical livestock unit and numbersof oxen were more likely to participate in thecattle fattening. This variable was found to in-fluence the cattle fattening practice positivelyand significant at 10 percent significance level.The result show that all other things being keptconstant, the odds ratio in favor of participatingin cattle fattening practice increases by a factorof 1.48 as the number of livestock increase byone in tropical livestock unit. The implicationof the result was that livestock are an importantsource of cash in rural areas to allow purchaseall feed that required for cattle fattening and re-ducing fattening duration. Farmers who have alarge number of livestock might consider theirasset base as a mechanism of controlling anyrisk associated with cattle fattening and managing.Given this potential contribution of livestockand oxen to cattle fattening, it encourages foodsecurity and household income generation. There-fore, it is concluded that cattle fattening shouldbe facilitated by government and non-governmentorganizations. That means development partnershould focus on strengthening capacity of house-hold through providing credit facility in the di-rection of asset building like livestock purchasethought revolve funding system.

It is expected that farm extension servicewidens household knowledge with regard tothe use of the improved agricultural technology.Agricultural extension services are expectedto enhance households’ skills and knowledge,link households with technology and markets.Access to extension services such as information,training, field days, field visits and field toursreceived by households positively and signifi-cantly affected participation in cattle fattening.This implies farmers that have access to ex-tension service may analyze cattle price infor-mation and sell their cattle at appropriatemarket price.

AcknowLEDgEMEntI would like to thank Oromia Agricultural Re-

search Institute for financing data collectionactivity.


Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 8(

2), 1

73-1

92, J

une

2018

.

192

rEFErEncES Aldrich, J.H., & Nelson, F.D. (1984). Liner probability,

logit and probit model: Quantitative applications in the social science. Sera Miller McCun, Sage Pub. Inc, University of Minnesota and Iola, London.

Becker, So, Ichino, A. (2002). Estimation of average treatment effects based on propensity scores, The Stata Journal, 2, (4), 1-19.

Belete, A., Azage, T., Fekadu, B., & Berhanu, G. (2010). Cattle milk and meat production and marketing systems and opportunities for market -orientation in Fogera woreda, Amhara region, Ethiopia. IPMS K2 (Improving Productivity and Market Success) ofEthiopian Farmers Project Working Paper 19. ILRI (International Livestock Research Institute), Nairobi, Kenya, 2: 65.

Bryson, A., Dorsett, R., & Purdon, S. (2002). The Use of propensity score matching in the evaluation of labour market policies, Working Paper No. 4, Department for Work and Pensions.

CSA (Central Statistics Authority) (2013). Statistics report on farm management practices, livestock and farm implements. Part II, Addis Ababa, Ethiopia 1, 14-15.

Caliendo, M., & Kopeinig, S. (2005). Some practical guidance for the implementation of propensity score matching, Discussion Paper No. 1588, University of Cologne.

EHZARDO (East Hararghe Zone Agriculture and Rural Development Office) (2015). Annual report.Fadis District, East Hararghe Zone, Ethiopia.

FAO (2015). Food outlook, biannual report on global food markets.Food and Agricultural organization in the United Nation.PP 6.

Gujarati, DN. (2003). Basic econometrics (2ndEd.). McGraw Hill, Inc., New York.

Lee, WS. (2006).Propensity score matching and variations on the balancing test: Melbourne Institute of Applied Economic and Social

Research, the University of Melbourne.Leuven, E., & Sianesi, B. (2003). Psmatch2, stata

module to perform full propensity score matching. Retrieved from http://ideas.repec,org/ c/boc/ bocode/s432001.html.

Liao, T.F. (1994). Interpreting probability models: Logit, probit and other generalized models. Sage University paper series on Quantitative Applications in the Social Sciences, 07-101. Thousand Oaks, CA: Sage, California.

MoARD (Ministry of Agriculture and Rural Development), (2006). Poultry and poultry product development five years plan (1998-2003). Animal and Fishery Development Department, MoARD, Addis Ababa, Ethiopia.

MoARD (Ministry of Agriculture and Rural Development), (2007). Livestock development master plan study: Phase I Report – Data Collection and Analysis. Volume V-Poultry Production. Addis Ababa, Ethiopia.

Rosenbaum, PR., & Rubin, DB. (1983). Thecentral role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41–55.

Rosenbaum, PR. (2002).Observational studies (2nd Ed.),Springer-varlag, New York.

Sianesi, B. (2004). An evaluation of the active labor market programs in Sweden. The Review of Economics and Statistics, 186(1), 133-155.

SPS-LMM (2010).Trade Bulletin Issue I. Focus on Ethiopia’s meat and Live Animal Export Peters, K. J. and Thorpe, W. 1989. Trends in On-Farm Performance Testing of Cattle and Sheep in Sub-Saharan Africa. International Livestock Centre forAfrica, Addis Ababa, Ethiopia.

Venetoklis, T. (2004). An evaluation of wage subsidy programs to SMEs Utilising Propensity Score Matching: Government Institute of Economic Research, Helsinki.


How to cite this article:Mume Ahmed, J., & Tadesse Gute, F. (2018). Impact of small-holders’ cattle fattening on household in-come generation in fadis district of Eastern Hararghe Zone, oromia, Ethiopia. International Journal ofAgricultural Management and Development, 8(2), 173-192.urL: http://ijamad.iaurasht.ac.ir/article_540495_f42e57cd61867713987e013fe95e4341.pdf

impact of small-holders’ cattle fattening on household...

Documents