Impact of Seed Voucher System on Rice Yield, Income Inequality and Poverty Reduction in Rural Nigeria: A Randomized Control Trial Approach

Size: px
Start display at page:

Download "Impact of Seed Voucher System on Rice Yield, Income Inequality and Poverty Reduction in Rural Nigeria: A Randomized Control Trial Approach"

Transcription

1 Impact of Seed Voucher System o Rce Yeld, Icome Iequalty ad Poverty Reducto Rural Ngera: A Radomzed Cotrol Tral Approach AWOTIDE, B.A. * 1, A. DIAGNE 2, T. T. AWOYEMI 3 ad V.E.T. Ojehomo 3 A Cotrbuted Paper Prepared for Presetato at the Iteratoal Coferece of the Cetre for the Studes of Afrca Ecoomes (CSAE), St Cathere s College Oxford, U.K March, Abstract Emergecy Rce Itatve ( ) was formulated by AfrcaRce respose to the global food crss of 2008, to mtgate the adverse effects o poor rce farmers. It made use of Seed Voucher System (SVS) to grat farmers access to Certfed Improved Rce Seed (CIRS) at a subsdzed rate. The SVS was to geerate creased yeld ad a reducto poverty ad come equalty. However, the extet to whch the SVS has creased rce yeld, reduced poverty ad come equalty has ot bee fully documeted. Therefore, ths study examed the mpact of SVS o yeld, poverty ad come equalty reducto amog the beefcares. The study used structured questoare to collect both basele (2008) ad post-voucher (2010) data, usg multstage samplg procedure. I all, 600 rce farmers were selected based o probablty proportoate to the sze of rce farmers the vllages, out of whch 160 farmers receved the seed voucher ad the others dd ot. The LATE estmato techques were used to assess the mpact. The results revealed that the seed voucher system mpacted yeld sgfcatly. Poverty ad come equalty also reduced sgfcatly after the terveto ad betwee the treated farmers. Therefore, the use of seed voucher to grat farmers assess to producto puts ca actually be a way out of poverty ad ca be used to redstrbute come rural Ngera. Keywords: Impact, rce, voucher, poverty. Icome equalty, LATE, Ngera *1 Correspodg author Ph.D Studet, Departmet of Agrcultural Ecoomcs, Uversty of Ibada, Ngera E-mal:bawotde@yahoo.com Tel.(Moble) :

2 1.0. Itroducto Globally, crease agrcultural productvty has bee vew as a way out of the scourge of poverty ad come equalty that has bee the bae of ecoomc growth ad developmet, partcularly developg coutres. Poverty ad come equalty are closely related ad t has bee argued that come equalty s a mafestato as well as a strog cause of poverty (UNU/WIDER, 2000). Poverty most developg coutes lke Ngera has bee descrbed as a rural pheomeo. For stace, I Ngera poverty cdece the rural area was 46.0 %, 69.3%, 63.3% 1992, 1996 ad 2004 respectvely. Whle the correspodg fgures for the urba area was 37.5%, 58.2% ad 43.2% for the same perods (Natoal Bureau of Statstcs (NBS), 2005). Of equal cocer also, s the hgh come equalty Ngera as show by G-Coeffcet. The G-Coeffcet was for the atoal, ad for urba ad rural areas respectvely (NBS, 2005). Furthermore, whle populato s growg at the rate of 3.5% per aum, the food growth rate s stll as low as 2.5% per aum. Thus geeratg a wde gap demad ad supply of basc food ad threateg the food securty of the ato. Cosequetly, crease food producto, reducto of poverty ad come equalty s the most dffcult challege facg Ngera ad the greatest obstacle to the pursut of sustaable ecoomc growth (Natoal Bureau of Statstcs (NBS), 2005). The rural dwellers Ngera are bascally farmers, producg a wde rage of staple food crops. However, rce s the most mportat staple food crop cosumed ad has thus become a atoal commodty. Rce accouts for 12% of the total output of cereals produced Ngera (CBN, 2009). Majorty of the populato lve o rce, ad ther prmary food securty s etrely depedet o the volume of rce produced. Ideed, rce avalablty ad food securty have log bee syoymous Ngera. Especally the producg areas, t provdes employmet for more tha 80 per cet of the habtats as a result of the actvtes that take place alog the dstrbuto chas from cultvato to cosumpto (Ogudele ad Okoruwa, 2006). Whle the aual domestc cosumpto rose from 5kg/perso 1970 (Obecha ad Ott, 1985), to 25kg/perso 2004 (Husse, 2004) rce producto has bee expadg at the rate of 6% per aum, meag that the demad for rce s growg faster tha producto. The self-suffcecy rato has also decled from 0.87 ad 0.93, the 80s ad 90s to 0.64 betwee 2001 ad 2005 (AfrcaRce, 2005), thus, makg the coutry depedet o mported rce to meet the hgh demad to the extet that Ngera was raked frst Afrca ad secod the world wth mportato of 1.6 mllo tos 2006 (AfrcaRce, 2008). I 2007, mport rose up to 1.7 mllo tos, supplyg about oe-thrd of total demad of 5 mllo tos (The Puch, 2008). Ths costtuted a massve dra of the ato s exteral reserves ad had compouded the poverty stuato of the populace; ths has made producg rce locally ad reducg ts mportato a atoal cocer. Area expaso ad rrgato have already become a mmal source of output growth at a world scale. Hece, agrcultural growth wll deped more ad more o yeld-ehacg techologcal chage (Datt ad Ravallo, 1996; Hossa, 1989). It s beleved the adopto of ew agrcultural techology, such as hgh yeldg varetes that kck-started the Gree Revoluto Asa, could lead to sgfcat creases 2

3 agrcultural productvty Afrca ad stmulate the trasto from low productvty subsstece agrculture to a hgh productvty agro-dustral ecoomy (World Bak, 2008). Cosequetly, several mproved rce varetes have bee developed by the atoal ad teratoal research sttutes ad dssemated to the farmers through dfferet programs. The qualty of the mprove rce varetes were further ehaced through certfcato by the approprate certfyg ageces ad passed through the exteso agets, research sttutes ad regstered seed compaes to the rural farmers. Frequetly, farmers fd t dffcult to obta good qualty seeds that are sutable to ther local codtos. Most of the recorded crop falure s attrbuted to late avalablty of seed ad the poor adaptablty of some of these avalable varetes to local agro-ecologcal ad soco-ecoomc codtos (Awotde et al., 2010). Also, recurret droughts some areas have resulted to local seed stocks beg exhausted due to the coverso of seeds to food ad the stocks are ot beg repleshed year year out due to crop falure. I addto, commercal producers of certfed seed are ot avalable remote dstrcts ad local busess people are reluctat to stock seed because demad s ucerta. Hece, access to certfy seeds have bee oe of the problems mltatg agast crease food producto Ngera. Oe opto out of ths predcamet accordg to Bramel, Remgto ad McNel (2004), s smart seed subsdes that target partcularly eedy farmers. The targeted approach of put subsdes, whch s supported by a rage of doors ad govermets, advocates the use of vouchers that ca be exchaged at agro-dealer shops across rural areas. The fertlzer ad seed subsdy adopted Malaw has helped doubled ts agrcultural productvty, turg the coutry to a et food exporter after decades of fame as a pereal food mporter (Uted Natos, 2008). I addto, the bumper harvests resultg from the program are helpg the poor farmers to ear more come, thus reducg ther propesty to fall to poverty. I vew of ths, the Emergecy Rce Itatve (ERI) to boost rce producto Sub-Sahara Afrca was lauched 2008 by the Afrca Rce Cetre (Ex-WARDA), the Iteratoal Fertlzer Developmet Cetre (IFDC), ad Catholc Relef Servces (CRS), wth the framework of the Food ad Agrcultural Orgazato s (FAO) Itatve o Soarg Food Prces (ISFP) (AfrcaRce, 2009). The ERI adopted the seed voucher system to grat farmers access to certfed mproved rce seed at a subsdzed rate. Specfcally, the tatve was: to mprove farmer s access to rce seed ad meral fertlzer, expad ad ehace farmer s kowledge o best-bet rce techologes, reduce post-harvest losses ad mprove processg techques, mprove farmer s access to market, mprove food securty, establsh or stregth puts dealers etwork. The emergecy rce tatve adopted the seed voucher system. Some radomly selected rce farmers from four coutres (Mal, Seegal, Ngera ad Ghaa) were gve voucher to procure certfed mproved rce seed at a subsdes rate ad some farmers were ot gve, for comparso sake. Based o the declarato by Chapma et al.,(2003) that agrcultural formato teracts wth ad flueces agrcultural productvty a varety of ways, ot oly ca t help form decsos regardg lad, labour, captal ad maagemet, t ca also lead to mprovemet agrcultural productvty. Therefore, several meas of commucato were used to dssemate formato o the best-bet techologes that wll ehace crease rce output to the radomly selected rce farmers. I addto, exteso servces s wdely beleved 3

4 to possess the ablty to reach llterate farmers ad provde them wth formato relatg to all aspects of agrcultural producto a laguage they uderstad, hece exteso servce was complemeted wth rural rado broadcast ad vdeo shows for effectve results. O the overall, the terveto was to mprove the farmers access to certfed mproved seeds whch s expected to have drect poverty ad come equalty reducg effect through the crease yeld ad come. Idrect medum ad log-term poverty reducto effects were also expected as a result of mproved access to educato ad health servces brought about by crease come. However, the extet to whch the seed voucher system has mpacted rce yeld, reduce come equalty ad poverty has ot bee assessed. Therefore, ths study emprcally vestgated the mpact of seed voucher system o rce yeld, come equalty ad poverty reducto amog the rce farmg households Ngera. The rest of the paper s orgazed as follows: secto two preseted the theoretcal framework. The aalytcal framework s preseted secto 3. Secto cotas the data ad descrptve statstcs. The results of the aalyses are preseted secto 5. Fally, secto 6 cotas the summary, cocluso ad a bref polcy recommedato Theoretcal Framework The basc am of mpact assessmet s to assess a program s performace agast a explct couterfactual, such as the stuato the absece of the program. It focus maly o estmatg the drect effects of the program ad also o uderstadg what the program effects would be f the program offer were exteded to other dvduals ot curretly partcpatg. Ths study adopted the potetal outcome framework. Uder the potetal outcome framework developed by Rub (1974), the rce farmg household has ex-ate two potetal outcomes: a outcome f a farmer receved the seed voucher ad a outcome f a farmers dd ot receve the seed voucher. For stace, the terest of ths study s o the effect of the recept of seed voucher (treatmet) whch s represeted by the bary varable T, o outcomes of terest Y, If T = 1 for farmer who receved seed voucher ad T=0 for farmer who dd ot. Assocated wth each state s a potetal outcome that a dvdual would atta wth ad wthout beg exposed to the treatmet (Rub, 1974, 1977). If Y T deotes the potetal outcome the treated state ad Y C the potetal outcome the utreated state, each perso has a (Y C, Y T ) par that represets the outcome that would be realzed the two states. The treatmet effect of the recept of seed voucher o the outcome of terest s the defed as Y T - Y C. However, there s a detfcato problem due to the fact that t s mpossble to observe the two outcomes Y C ad Y T for a dvdual at the same tme; oe perso ca oly be oe state at a tme. What ca oly be observed s Y= Y T.T + Y C. (1-T). Due to the mssg of oe of the potetal outcome, t s therefore mpossble to compute the treatmet effect, Y T - Y C for a dvdual (Abade, 2001). I ths study, oe caot observe what would happe to the outcomes (yeld, per capta rce come ad household expedture) of the farmers that 4

5 receved the seed voucher f they had ot receved t. The fudametal evaluato problem arses because we ca ever observe both states for the same dvdual at the same tme,.e. oe of the states s couterfactual. That s, there s a problem of mssg data. The mpact of the program caot be evaluated wthout a soluto to the mssg data. Hece, dfferet methods have bee proposed the lterature to solve the problem of mssg data mpact assessmet these are: No-expermetal ad expermetal methods. No-expermetal estmator of program mpacts typcally use two types of data to compute the mssg couterfactual outcomes for programs: data o partcpats at a pot tme pror to eterg the program (Basele data) ad data o o-partcpats. Oe major problem assocated wth o-expermetal methods s the problem of selecto bas. Selecto bas s a statstcal bas whch there s a error choosg the dvdual or groups to take part a scetfc study. Selecto bas arses because, programs are placed specfc areas, dvduals are screeed for partcpato ad the decso to partcpate a program s ofte volutary, creatg self-selecto. Hece those who are ot exposed to the programs are ofte poor comparso group for those who are, because ay dfferece betwee the groups ca be attrbuted to both the mpact of the programme or pre-exstg dffereces (the selecto bas). Wthout a relable way to estmate the sze of ths selecto bas, oe caot decompose the overall dfferece to a treatmet effect ad a bas term (Duflo ad Kremer, 2007). For stace, suppose the mea effect of the program o poverty a populato s E (Y T - Y C ). Such a populato parameter s called the Average Treatmet Effect (ATE) the lterature (Imbes ad Wooldrdge, 2009).ATE measures the effect of a treatmet o a perso radomly selected the populato (Wooldrdge, 2002). It s also possble to estmate the mea effect of seed voucher o the sub-populato of the farmers that receved the seed voucher E (y t -y c t=1) whch s called the average treatmet effect o the treated ad s usually deoted by ATT. The Average treatmet Effect o the utreated E (y t- y c t=0), deoted by ATU s aother parameter that ca be estmated. I ths study some farmers receved the seed voucher ad others do ot. Oe smple approach s to take the average of both groups ad exame the dfferece betwee yeld, come equalty ad poverty amog the farmers wth seed voucher ad those wthout. I a large sample, ths wll coverge to: T=E [Y T farmer receved voucher] - E [Y C Farmer dd ot receve voucher] = E[Y T T=1]-E[ Y C T=0] Subtractg ad addg E[Y C T=1], that s the expected outcome for a farmer the treatmet group had the farmer ot bee treated ( a quatty that caot be observed but s logcally well defed) we obta: D=E[Y T T=1]-E[Y C T=1]-E[Y C T=0] + E[Y C T=1] =E[ Y T -Y c T=1] +E[Y C T=1]-E[Y C T=0]. (1) D=ATE1+B 5

6 The frst term [Y T -Y C T=1] equato 1, s the treatmet effect, whch s called the Average Effect o the Treated ATE1 or ATT. It s the mea mpact o all the outcome of terest amog those who actually receved the seed voucher. It also aswers the questo: O average, amog the treated farmers, what dfferece dd the seed voucher make? The secod term B or E[Y C T=1] E[Y C T=0], s the selecto bas caused by edogeous selecto to treatmet. It captures the dfferece potetal outcome betwee the farmers that receved the seed voucher ad those that dd ot. Farmers that receve seed voucher may have had dfferet yeld, come or per capta expedture o average eve f they had ot bee treated. Sce E [Y C T=1] s ot observed, t s therefore mpossble to assess the magtude (or eve the sg) of the selecto bas ad therefore the extet to whch selecto bas explas the dfferece outcomes betwee the treatmet ad the comparso group. The basc objectve of a soud mpact assessmet s therefore to fd a way to get rd of the selecto bas (B=0) or fd a way to accout for t. The expermetal method also refers to as radomzato or Radomzed Cotrol Tral (RCT) s geerally cosdered to be the most robust of the evaluato methodologes. Expermetal desgs are coducted by radomly allocatg the terveto amog elgble beefcares such that the assgmet process wll create comparable groups: the treatmet ad cotrol groups (Duflo ad Kremer, 2007). Oe of the hallmarks of RCT s ts effectveess dealg wth the problem of selecto bas program mpact evaluato. Therefore, for ths study, samples of 600 rce farmers were selected from the populato of rce farmers Ngera. Ths expermetal sample s the dvded radomly to two groups: the treated group (T=1) ad the comparso group (T=0). The farmers the treated group receved the seed voucher. The cotrol group s the expermetal group of farmers who would have receved the seed voucher, but were radomly deed access to the seed voucher.if the Average Treatmet Effect o the treated s estmated as: E(Y 1 T=1)-E(Y 0 T=1) 2 I other to estmate equato (2), t s extremely very mportat to kow what would have bee the outcomes of the farmers had they ot receved the seed voucher. Ths mples that a estmate of E(Y 0 T=1) s requred, whch s uobserved by the researcher. I geeral, the observed outcomes of the farmers that dd ot receved the seed voucher (E (Y 0 T=0)) caot be used as proxy sce those farmers who dd ot receved the seed voucher mght be dfferet from those that receved t may ways. The goal of RCT s to geerate a expermetal group of farmers who would have partcpated the program but were radomly deed access to the seed voucher. Hece, radomly selected group were used as the cotrol group ad ther resposes were also used as the desred couterfactuals, E (Y 0 T=1). If T*=1 for the farmers who receved the seed voucher the presece of radom assgmet ad T*=0 otherwse. Let r=1 doate the group of farmers who were radomly assged to receve the seed voucher ad r=0 for those farmers who dd ot receve the seed voucher (Cotrol group). By troducg T* ad r equato 2 ca be rewrtte as: E (Y 1 T*=1, r=1) - E (Y 0 T*=1, r=1) 3 6

7 Where T s replaced by T*=1, r=1 thus t s expected that: E (Y 0 T*=1, r=1)= E (Y 0 T*=1, r=0) 4 E (Y 0 T*=1, r=1) s the average outcome that would have bee obtaed had the farmer ot bee treated. The radomzed expermet made t possble to observe E (Y 0 T*=1, r=0) whch ca be used as a replacemet for E (Y 0 T*=1,r=1). Sce the treatmet has bee radomly assged, dvduals assged to the treatmet ad cotrol dffers expectato oly through ther exposure to the treatmet. Had they ot bee treated, they would have bee expected to gve the same outcome. Ths suggests that the selecto bas, E[Y C T=1]-E[Y C T=0], s equal to zero. Therefore, by usg equato (3) the Average Treatmet Effect o the treated ca be estmated as follow: E(Y 1 -Y 0 T*=1)= E(Y 1 -Y 0 T*=1,r=1) =E(Y 1 T*=1,r=1)- E(Y 0 T*=1,r=1) = E(Y 1 T*=1,r=1)- E(Y 0 T*=1,r=0) Thus radomzato serves as a strumetal varable by creatg varatos amog farmers who receved the seed voucher, because some of them dd receve the seed voucher radomly whle some of them dd ot receve the seed voucher. Ths eables a accurate estmato of the ATE o the treated. The ma beeft of ths techque s the smplcty terpretg results the program mpact ca be measured by the dfferece betwee the meas of the samples of the treatmet group ad the cotrol group o yeld, per capta rce come ad come equalty. However, accordg to Imbes ad Agrst (1994), soo after radomzato some farmers assged to the treatmet group wll ot take the seed voucher, but re-assged themselves to the cotrol group. Whereas, some farmers assged to the cotrol group wll maage to get the seed voucher at all cost. For stace, f T (r),s a dcator for the recept of seed voucher (1 for treated ad 0 for cotrol). Gve the assgmet r:t =T (r ) be the actual treatmet receved. The two dcators T (1) ad T (0) descrbed the complace behavour ad are used to partto the populato uts to four dstct types : The complers, always takers, ever takers ad defers. However, the mpact o the outcome of terest ca oly be observed o the sub-populato of complers, thus ecesstatg the use of other mpact evaluato methods that cosstetly gve the mea mpact o the outcomes of terest the sub-populato of the complers Aalytcal Framework 3.1. Measuremet of Poverty Idces The stadard Foster-Greer-Thorbecke (FGT) (1984) was employed to geerate the poverty profle of the respodets before ad after the project for the two groups (the treatmet ad the cotrol group). 7

8 FGT takes the form; P 1 Z Yp q 1 Z 5 Where Z = the poverty le q= umber of dvdual below the poverty le = umber of dvduals the referece populato Y p = per capta come of the th household = FGT dex whch takes values 0, 1, 2. Z-Y = poverty gap of the th household Z Y Z = poverty gap rato Ths class of poverty measure s flexble two ways. Oe, α s a polcy parameter that ca be vared to approxmately reflect poverty averso ad two, the P α class of poverty dces s sub-group decomposable. Whe α = 0 equato 5 P o = 1/ (q) = q/ = H 6 The head cout s the umber of people a populato who are poor, whle the headcout rato (H) s the fracto of the populato who are poor. The poverty gap measures the total amout of come ecessary to rase everyoe who s below the poverty le up to that le, Whe α = 1, the poverty measure becomes the povertygap dex (PG) P α - 1 = PG = 1 Z Yp q 1 Z = HI 7 1 Z Yp Where I = q = HI 8 q 1 Z s the come gap rato. I s the mea of the poverty gaps expressed as a porto of the poverty le. Ths measure s sestve to come dstrbuto amog the poor. Whe α = 2, the squared poverty gap dex (SPG) s geerated gve by, P α-2 = SPG = 1 Z Yp q 1 Z 2 8 9

9 P α-2 measure s creasgly used as a stadard poverty measure by the World Bak, the regoal developmet baks, most UN ageces ad has bee used most emprcal work o poverty because of ts sestvty to the depth ad severty of poverty. The cdece s measured by the umber of people the total populato lvg below the poverty le whle the poverty testy reflected the extet to whch the comes of the poor fall below the poverty le. Aother advatage of the P α measure s that t s decomposable by populato subgroups. That s : m P α = K P 10 j j Where: j 1 j =1,2,3.m, k j s the populato share of each group, cotrbuto of each group C j to overall poverty ca be calculated as follows: C j = K j j 11 P P P j s the poverty measure of group j. The Ths property of the dex mples that whe ay group becomes poor, aggregate poverty wll crease. Hece poverty ca be dsaggregated by subgroup such as geder ad rego Determats of poverty amog the Rce Farmg Households Ths study adopted the logstc regresso to assess the factors that determe the farmers poverty status. The respodets were classfed to poor ad o-poor usg the poverty le. The relatve poverty le of two-thrd of mea per capta come was used. Farmers that have per capta come below the poverty le were classfed as poor ad o-poor otherwse. The respose varable was bary, takg values of oe f the farmer s poor ad zero otherwse. However, the depedet varables were both cotuous ad dscrete. Although, several methods have bee adopted the lterature to aalyse data volvg bary outcomes, for ths partcular study; logt model was selected over dscrmat ad lear probablty models. The justfcato for usg logt s ts smplcty of calculato ad because ts probablty les betwee 0 ad 1. Moreover, ts probablty approaches zero at a slower rate as the value of explaatory varable gets smaller ad smaller, ad the probablty approaches 1 at a slower ad slower rate as the value of the explaatory varable gets larger ad larger (Gujarat, 1995). The probablty that a farmer wll fall below the poverty le was postulated as a fucto of some socoecoomc/demographc characterstcs ad sttutoal factors. Therefore, the cumulatve logstc probablty model s ecoometrcally specfed as follows: P F( Z ) F( X 1 ) 1 e 12 Z 9

10 Where P s the probablty that a farmers wll fall below the poverty le or ot gve X; e deotes the base of atural logarthms, whch s approxmately equal to 2.718; X represets the th explaatory varables; ad ad are parameters to be estmated. The logt model could be wrtte terms of the odds ad log of odds, whch eables oe to uderstad the terpretato of the coeffcets (Hosmer ad Lemeshew, 1989). The odds rato mples the rato of the probablty (P ) that a farmer s poor to the probablty (1-P ) that the farmer s ot poor. 1 ( 1 P ) 13 Z 1 e Therefore P 1 P 1 e 1 e z z e The atural log of equato (14), wll gve: Z P 1 P z l 1 X 1 2 X 2... m X m 15 If the dsturbace term ( U ) s take to accout, the logt model becomes: Z m X U 16 1 Equato (14) was estmated by maxmum lkelhood method. Ths procedure does ot requre assumptos of ormalty or homoskedastcty of errors predctor varables. Ths aalyss was carred usg STATA verso11.0. Note: The defto of the varables cluded the logstc regresso s preseted table Measuremet of Icome Iequalty Icome equalty ca be measure by usg the G-coeffcet. Followg Morduch ad Scular (2002), where come are ordered so that y 1 y 2 y 3 y 4 y. The G-coeffcet s computed as: 14 I G (Y) = ( 1 ) y 2 17 Where : =umber of observato = mea of dstrbuto y = come of the th household. 10

11 3.4. Ecoometrc Impact of Seed Voucher System o Productvty ad Household Icome I the treatmet effect lterature, bases that ca arse whe estmatg causal effects are of two types: overt bas ad hdde bas (Rosebaum, 2001; Lee, 2005). Overt bas s the dfferece the observed outcomes y ot caused by the seed voucher but whch s due to dffereces observed characterstcs of the farmers. Hdde bas s the dfferece the observed outcome y ot caused by the seed voucher but whch s due to dffereces as a result of uobservable characterstcs of the farmers. A thrd problem s the problem of o-complace also called the edogeous treatmet varable problem ecoometrcs (Imbes ad Rub, 1997; Imbes ad Agrst, 1994; Heckma ad Vytlacl, 2005). The o-complace problem arses because the subjects of treatmets are people who may or may ot stck to ther assged treatmets eve f the treatmet was assged radomly. Cosequetly, the dfferece a dvdual perso s potetal outcomes may ot be due to the seed voucher but rather to the uobserved factors that cause that farmer ot to stck to hs or her assged treatmet. As a result, the average treatmet effect for the etre populato s dfferet from the mea treatmet effect that would obta whe the seed voucher was radomly assged ad every farmer the populato compled wth ther assgmet (Imbes ad Rub, 1997; Imbes ad Agrst, 1994). Cosequetly, addto to the RCT approach, other methodologes that ca remove both hdde bas ad o-complace was adopted order to gve a cosstet estmate of the mpact of seed voucher o all the outcomes of terest Estmatg Treatmet Effects uder Radomzed Expermet Uder a assumpto of codtoal exogeety of program placemet, a ubased estmate of program ca be estmated. Followg Ravallo (2008), deotg for smplcty Y (1) as Y T ad Y (0) as Y C. The equato ca be appled to sub-sample of farmers that receved seed voucher ad those that dd ot receve the seed voucher as follows: Y T T T T X 18 f T =1,=1... Y C C C C X 19 f T =0, =1 For ths study equato 18 ad 19 was estmated as a sgle regresso by poolg the data for both the treated ad cotrol farmers as follows: Y C T C C T C ) T X X ( ) ( T 20 T C Where: T ( ) The treatmet Effect from equato 20 ca be wrtte as C t C T C ATE1= E( Y T 1, X ) E X ( )

12 ATE1 s the treatmet effect o the treated. From equato 20, the cosstet estmate of the program effect was T C calculated usg OLS uder the assumpto that: E( X T t) E( X T t) 0, t (0,1 ).That s there was o selecto bas because of radomzato. I practce a commo mpact model that s usually adopted T assumes C T C. The ATE1 s the smply. Apart from RCT approach that removes selecto bas program mpact evaluato, other methods desged to remove bases ca be classfed uder two broad categores based o the types of assumptos they requre to arrve at cosstet estmators of causal effects (see Imbes 2004). Those methods desged to remove overt bas oly are based o the gorablty or codtoal depedece assumpto (Rub, 1974; Rosebaum ad Rub, 1983) that postulates the exstece of a set of observed covarates x, whch, whe cotrolled for, reders the treatmet status d depedet of the two potetal outcomes y T ad y C. O the other hads, the pure parametrc regresso-based methods adopted the codtoal depedece assumpto whch the covarates are possbly teracted wth treatmet status varable to accout for heterogeeous resposes, or they are based o a two-stage estmato procedure where the codtoal probablty of treatmet P(t = 1 x) P(x) (called the propesty score), s estmated the frst stage ad ATE, ATE1 ad ATE0 are estmated the secod stage by parametrc regresso-based methods or by o-parametrc methods; the latter clude varous matchg method estmators such as those used by Medola (2006) Iverse Propesty Score Weghtg Techque Ths study adopted the codtoal depedece-based estmators of ATE, ATE1 ad ATE0 whch s the socalled verse propesty score weghg estmators (IPSW) ad are gve by the followg formulae (see Imbes, 2004; Lee 2005, Dage ad Demot, 2007; Dotsop-Nguezet et al., 2011; Awotde et al, 2011): t pˆ x y x 1 px ATEˆ pˆ ˆ t ˆ x px p y ATEˆ 1 ˆ pˆ pˆ x x t y ATEˆ Where s the sample sze, propesty score evaluated at x. t 1 1 s the umber of treated ad p ˆ( x ) s a cosstet estmate of the ATE= s the mea mpact of the seed voucher the populato ATE1=s the mpact of the seed voucher o the subpopulato of the farmers the treated group. ATE0= s the mpact o the subpopulato of the farmers the cotrol group. Ths s equally of terest case the program s to be exteded to those farmers who curretly dd ot receve the seed voucher. 12

13 A probt specfcato was employed to estmate the propesty score. However, the result of the ATE caot be terpreted as the mpact of the terveto. Due to the fact that the ATE estmates do ot correct for hdde bas (selecto o uobservables) whch s due the fact that farmers decso to receve the seed voucher could be based o some atcpated beefts ad problem of o-complace or edogeety whch may arse as a result of the fact that the farmer ca decde to receve the seed voucher or ot. Hece t s ecessary to use other methods that ca elmate these problems Istrumetal varable Estmato Methods Local Average Treatmet Effect Estmato Techque The strumetal varable methods are desged to remove both overt ad hdde bases ad deal wth the problem of edogeous treatmet. The strumetal varable (IV)-based methods was used by Heckma ad Vytlacl (2005, 2007a, 2007b); Heckma et al,(1997); Card,2001; Imbes (2004); Abade (2003); Imbes ad Agrst (1994); Dage ad Demot (2007) ad Dotsop-Nguezet et al.,(2011) to deal wth overt ad hdde bases ad also deal wth the problem of edogeous treatmet. The method volves fdg a varable (strumet) that s hghly correlated wth program partcpato but s ot correlated wth uobservable characterstcs affectg outcomes (Khadker et al., 2010). I other words, the IV-based methods assume the exstece of at least oe varable z called strumet that explas treatmet status but s redudat explag the outcomes y T ad y C, oce the effects of the covarates x are cotrolled for(rub, 1974; Rosebaum ad Rub,1983, Dage ad Demot, 2007; Dotsop-Nguezet, 2011). The methods rely o fdg a varable excluded from the outcome equato but whch s also a determat of programme partcpato. It s ofte the case socal expermet that some of those radomly selected for the programme do ot wat to partcpate. Hece, beg radomly assged to receve the seed voucher oly affects outcome va actual recept of seed voucher. Therefore, to estmate the causal effect of the treatmet whe the complace s ot perfect, the radom assgmet s used as a atural choce of strumetal varable (Katz et al., 2001, Galasso et al., 2004 ad Ravallo, 2005) Imbes ad Agrst (1994) solve the problem of o-complace the populato after RCT by dvdg the populato to four groups based o complace status: complers (those who adhere to ther assged treatmet), always takers (those who maage to always take the treatmet regardless of ther assgmet), ever takers (those who ever take the treatmet regardless of ther assgmet) ad defers (those who do the opposte of what ther assgmet asked them to do). The mportat pot made by Imbes ad Agrst (1994) s that oly the mea treatmet effect for the subpopulato of complers ca be gve a causal terpretato ad they called such a populato parameter the local average treatmet effect deoted by LATE. Thus, LATE estmate provdes the mpact of seed voucher o all the outcomes wth a causal terpretato. I other for IV estmate to be terpreted as the causal effect of a treatmet o the complers both mootocty ad the depedece assumpto must hold (Imbes ad Agrst, 2004). The depedece assumpto requres that potetal 13

14 outcomes of ay treatmet state ( y, y ) are depedet of the strumet z..e. y y, T (1), T (0) s T C T, C depedet of Z. The mootocty assumpto requres that the strumet makes every perso ether weakly more or less lkely to actually partcpate the treatmet (o defers),.e. T (1) T (0) for all. The mootocty assumpto s trvally satsfed the seed voucher case because oe caot receve the seed voucher wthout beg radomly selected to receve t. Ths effectvely rules out the cases of defers ad always takers. Thus, for assessg the mpact of the seed voucher o ay farmer s outcome, the populato was parttoed to oly two dstct groups: the group of complers, whch s the group of potetal recevers (those who wll receve the seed voucher whe they are radomly selected to receve t), ad the group of ever takers, whch s the group of farmers that wll ever receve t eve whe they are assged to receve t. Hece, the LATE estmate of the mea mpact of seed voucher o all the outcomes of terest has a causal terpretato, apples oly to the sub-populato of potetal recevers of the seed voucher. Specfcally, the Local Average Treatmet Effect (LATE) estmates the treatmet effect oly for those who decde to partcpate because of a chage Z(Agrst 1994). Ths study adopted the smple o-parametrc Wald estmator proposed by Imbes ad Agrst (1994) ad whch requres oly the observed outcome varable y, the treatmet status varable t, ad a strumet z. To gve the expressos of the Imbes ad Agrst (1994) LATE estmator ad that of Abade (2003), we oted that the radom assgmet s a atural strumet for recept of seed voucher e (whch s the treatmet varable here). Ideed, frstly oe caot receve the seed voucher wthout beg radomly selected to receve t. Secod, t s atural to assume that beg radomly selected to receve the seed voucher actually affect the overall poverty dcator through the recept of the seed voucher. That s beg radomly selected have o mpact o ay of the outcomes. Yeld, come equalty ad Poverty of the farmers are actually affected oly whe the farmers receved the seed voucher. Hece the two vtal requremet of the radom assgmet to be a vald strumet are met. The mea mpact of the seed voucher o poverty outcome of the sub-populato of Complers (.e. the LATE) s as gve by Imbes ad Agrst, 1994; Imbes ad Rub 1997, Lee, 2005: IV late = Ey yc t1 1 y z 1 Ey z 0 z 1 Et z 0 E T 25 E t The deomator equato (25) s the dfferece the probablty of partcpato the program (probablty of T=1) uder the dfferet values of the strumet. The rght had sde of (25) ca be estmated by ts sample aalogue: 14

15 1 1 y z z 1 y 1 1 z 1 z 1 t 1 z z t z 1 z 1 26 Ths s the well-kow Wald estmator. The Wald estmate gves the effect of the treatmet o those whose treatmet status wll be affected by the strumet, whch s kow as the Local Average Treatmet Effect (LATE) (Agrst ad Imbes, 1994). These are those who the absece of the radomly assged strumet, would ot have bee treated but are duced to receve treatmet by the assgmet. They are ofte referred to as the complers. Because the recept of the seed voucher s ot radom the populato due to the fact that farmer the cotrol group may oe or the other obtaed the seed voucher thus affectg ther outcomes. Also, farmers who were radomly selected to receve the seed voucher may evetually ot receve t. I addto, the receved of the seed voucher s also ot radomly dstrbuted the populato. It was targeted at rural based rce farmers ad also, oly farmers the three otable rce producg ecologes were targeted for terveto. Hece, the study adopted the Abade s estmato of LATE usg the LARF, whch requres the codtoal depedece assumpto stead of the radomess assumpto Local Average Respose Fucto Abade s (2003) geeralzato of the LATE estmator of Imbes ad Agrst (1994) to cases where the strumet z s ot totally depedet of the potetal outcomes 15 y T ad y C, but wll become so codtoal o some vector of covarates x that determes the observed outcome y. Wth these assumptos, the followg results ca be show to hold for the codtoal mea outcome respose fucto for potetal complers f(x,t) E(y x, t; t 1 = 1) ad ay fucto g of (y, x, t) ( Abade, 2003; Lee 2005): f (x,1) f (x,0) = ( y T - y C x, t 1 = 1) 27 1 E gy, t, xt1 1 Ek gy, t, x 28 P t 1 Where 1 z k p t z 1x Equato (29) s a weghted fucto that takes the value 1 for a potetal compler ad a egatve value otherwse. The fucto f(x, t) s called a Local Average Respose Fucto (LARF) by Abade (2003). Estmato proceeds by a parameterzato of the LARF ; x, t Ey x, t; t1 1 f 30 The, usg equato 23 wth gy, t, x y f ; x, t 2, the parameter s estmated by a weghted least squares scheme that mmzes the sample aalogue of E{κ (y f (θ ; x,t))2}. The codtoal probablty P(z=1 x) appearg the weght κ s estmated by a probt model a frst stage. Abade (2003) proves that the

16 resultg estmator of θ s cosstet ad asymptotcally ormal. Oce, θ s estmated, equato (28) s used to recover the codtoal mea treatmet effect Ey y x, t 1 1 obtaed by averagg across x usg equato (28) For example, wth a smple lear fucto f, t, x t x, the Ey y x t 1 Where:, 0, T C, 1. T C as a fucto of x. The LATE s the 0 I ths case, there s o eed for averagg to obta the LATE, whch s here equaled to α. Hece, a smple lear fuctoal form for the Local Average Respose Fucto (LARF) wth o teracto betwee t ad x mples a costat treatmet effect across the sub-populato of potetal complers. I ths study, we postulated a expoetal codtoal mea respose fucto wth ad wthout teracto to guaraty both the postvty of predcted outcome (Poverty) ad heterogeety of the treatmet effect across the sub-populato of potetal recevers( Those who wll receve the seed voucher whe radomly selected to receve ). Because bee radomly selected to receve the seed voucher s a ecessary codto for the recept of the seed voucher, t ca be show that the LATE for the subpopulato of potetal recevers (.e. those wth t1=1) s the same as the LATE for the subpopulato of actual recevers (.e. those wth t=zt1=1) Data ad Descrptve Statstcs Ngera wth a populato of 140 mllo (NBS, 2006), o al lad area of 924,000 square klometres s purely a agrara ecoomy. The rural ecoomy remas largely agrcultural based a tropcal clmate wth a varety of vegetato belts,ragg from the forest the south to the Sahel savaah the orth. Rce s grow all agro ecologcal zoes of Ngera uder three major producto systems amely; rrgated, ra-fed uplad ad lowlad whch accout for 16%, 30% ad 47% respectvely to the total lad area devoted to rce, ad they jotly cotrbute 97 per cet to the atoal rce output (Daramola, 2005).The study used both basele data (2008) ad Post-voucher data (2010) collected by AfrcaRce/NCRI through multstage samplg techque. Kao, Osu ad Nger states were purposvely selected to represet the three major rce producg systems respectvely. From each of the three states, fve rce producg Local Govermet Areas (LGAs) were selected ad three vllages were selected from each of the LGAs to geerate a total of 45 vllages. I all, 600 rce farmers were selected based o probablty proportoate to the sze of rce farmers the vllages, out of whch 160 farmers receved the seed voucher (Treated Farmers) ad the others dd ot (Cotrol Farmers). Data o socoecoomc/demographc characterstcs, treatmet status, expedture, come, ad sttutoal varables were collected usg structured questoare. Data were aalyzed usg descrptve statstcs, Foster-Greer-Thorbecke (FGT) poverty measure ad Iverse Propesty Score Weghtg Techque ad Local Average Treatmet Effect (LATE). After data cleag, 563 were used for the aalyss. As show Table 2, agrculture was the ma occupato of the respodets as 90.0% of the respodets had agrculture as ther ma occupato. Because of the tedousess assocated wth farmg, t s ot a surprse that 16

17 majorty of the respodets (80.6%) were males, whle oly 19.4% were females. I terms of age dstrbuto, a hgher percetage ( 44.8%)of the respodets were wth the age group of years, whle a eglgble proporto (0.9%) were above 70.0 years of age ad a total of 76.2% were betwee years of age. Ths shows that majorty of the respodets were ther actve ad productve age ad ths could have a postve fluece o rce productvty. The household sze was relatvely hgher the study area. Majorty of the respodets (76.2 per cet) were wth the household sze group of 1-10 people per household. About 87.0% of the respodets were atve of ther respectve vllages ad 52.0% have spet betwee years the study area. The educatoal backgroud of the household s head revealed that majorty of the respodets (32.0 per cet) lacked formal educato. Whle 15.0% had at least prmary educato, 10.0% had secodary educato ad 40.0% had Islamc educato. Oly 5 of the respodets represetg 0.9% had uversty educato Results ad Dscusso 5.1. Descrptve aalyss of the Impact of Seed Voucher The descrptve statstcs was adopted to assess f there was ay sgfcat dfferece the mea yeld, per capta rce come ad per capta cosumpto expedture before ad after the terveto ad also betwee the treated ad the cotrol farmers. The sgfcace of ay observed dfferece was also tested usg the t-test. The results revealed that there was a dfferece yeld of 760kg/ha sgfcat at 1% after the terveto. I the same ve, a sgfcat dfferece of N ad N 7104 was recorded for the per capta come ad per capta cosumpto expedture respectvely after the terveto. The aalyss was exteded further order to also exame f there were sgfcat dfferece the yeld, per capta come ad expedture betwee the treated ad the cotrol farmers. The result showed that the treated farmers had sgfcat hgher yeld, per capta come ad cosumpto expedture of 453kg/ha, N 1272 ad N respectvely. Cosequetly, oe ca coclude that the seed voucher system has sgfcat mproved the welfare of the treated farmg households. However, these observed creases caot be gve ay causal terpretato, because o bas was removed from the aalyss, ths could overestmate or uder estmate the true mpact of the terveto Poverty Profle of all Treated Farmers, before ad After the Iterveto A poverty profle descrbed the patter of poverty but, t s ot prcpally cocered wth explag the causes of poverty, although t s also a approprate startg pot for aalyzg the determats of poverty. The am of the poverty profle s to brg to lmelght the ma facts o poverty amog the respodets. Poverty was measured by three poverty measures, (1) the headcout dex, (2) the poverty defct dex or poverty depth, ad (3) the poverty severty dex (Froster et al., 1984). The relatve poverty le of 2/3 of the mea per capta come was used. The result showed that poverty cdece, depth ad severty reduced by 23.41%, 26.28% ad 17

18 34.48%amog the treated farmers after the terveto. Dsaggregato by geder revealed that the female headed households had a sgfcat reducto all the poverty measures tha the male headed household Determats of Poverty amog the Respodets The logstc regresso result of the determats of poverty amog the rce farmers s preseted table 6. A addtoal sght was also provded by aalysg the margal effects, whch was calculated as the partal dervatves of the o-lear probablty fucto, evaluated at each varable sample mea (Greee, 1990). The log-lkelhood of , the Pseudo R 2 of 0.21 ad the LR (ch2) of (sgfcat at 1% level), mples that the overall model s ftted ad the explaatory varables used the model were collectvely able to expla the correlates of poverty amog the rce farmg households Ngera. The aalyss showed that the recept of seed voucher (-0.63), educatoal level of household head ( ), Vocatoal trag (-0.09), come from other crops (-0.54) sgfcatly reduced the probablty of the farmers fallg below the poverty le. Whle household sze (0.22) sgfcatly creased the probablty of a farmg household fallg below the poverty le. Large household sze dcates that a farmer would have more mouth to feed; cosequetly the per capta expedture would reduce. The age of household head although egatve t was ot sgfcat. Ths mples that the probablty of fallg to poverty creases as age decreases ad poverty s more prevalet amog the youg farmers tha the old oes. However, ths probablty s reduced at old age ad very youg age of the household head as show by the postve sg of the age squared. I the same ve, the coeffcet (-0.66) of geder of household head was also egatve, but ot sgfcat. Ths suggests that poverty s more prevalet amog the female headed households the study area. The result of the margal effects revealed that a addtoal year of educato wll reduce the probablty of a farmg household fallg below the poverty le by 1.6 per cet, whle a addtoal perso to the household creases the probablty of fallg below the poverty le by 0.5%. Furthermore, a crease the umber of years of experece uplad rce farmg wll geerate a sgfcat (4.1%) reducto the probablty of fallg below the poverty le. The practce of multple-croppg system ca also have a sgfcat effect o poverty reducto. Ths was supported the large percetage (12%) reducto the probablty of fallg below the poverty le as a result of a addtoal come from other crop producto Ecoometrc Impact Evaluato of Seed Voucher System Impact of Seed Voucher System o Rce Yeld Ths study adopted varous estmato techques to assess the mpact of seed voucher system o dfferet poverty dcators, such as yeld, per capta come from rce producto ad cosumpto expedture. The aalyss of the mpact of the seed voucher system o rce yeld s show Table 7. From the result of the mea dfferece, t was observed that there was a sgfcat postve dfferece of kg/ha yeld betwee the 18

19 farmers that receved the seed voucher ad those farmers that dd ot receve. The result of the Iverse Propesty Score Weghtg (IPSW) estmato techque showed that the mea Average Treatmet Effect (ATE) o the populato was kg/ha, sgfcat at 1%. The ATE1 was kg/ha sgfcat at 1%. Also, the ATE0 was kg/ha. However, the ATE estmates of the mpact of seed voucher o yeld, does ot have a causal terpretato due to the problem of o-complace. Due to the problem of edogeety the program, cosequetly, other to assess the true mpact of the terveto we adopted the LATE both by WALD estmator ad the LARF. The LATE estmates by WALD estmator ad LARF revealed that the seed voucher system sgfcatly creased the yeld of rce by ad kg/ha respectvely. The results represeted the average chage yeld due to the recept of the seed voucher. The dsaggregato of the mpact seed voucher o yeld by socoecoomc characterstcs of the farmers shows that the mpact was hgher amog the male farmers (793kg/ha) tha the female farmers (15.49kg/ha). Also, comparso across the major rce producto ecologes Ngera revealed that the terveto had a sgfcat postve mpact across all the major rce producg ecologes Ngera. However, there was varato the mpact across the ecologes. The uplad rce producg ecology have the hghest mpact of kg/ha, followed by lowlad (639.72kg/ha) ad rrgated rce ecology wth a mpact of kg/ha. I terms of the mpact o the poverty status of the respodets, the terveto has a sgfcatly postve mpact o both the poor ad o-poor. However, the mpact o the o-poor (872.15kg/ha) was hgher tha the poor (445.13kg/ha). Ths mples that although the terveto had postve mpact, the degree ad the level of mpact was also determed by some socoecoomc characterstcs of the beefcares. The determats of rce yeld as gve by the Local Average Respose Fucto (LARF) are preseted table 8. The result showed that there were coeffcets of o-teracted terms (depedet varables of yeld) ad teracted terms whch are teracto betwee depedet varable ad yeld. The o-teracted terms showed that some other soco-ecoomc/demographc characterstcs of the farmers apart from the seed voucher sgfcatly explaed varato rce yeld. These soco-ecoomc characterstcs cluded household sze, trag, secodary occupato; umber of years of experece uplad ad lowlad rce producto. From the aalyss, t was dscovered that female headed households ted to have hgher yeld tha the male headed households. Ths could be due to the fact that the male headed households most of the tme have other secodary occupatos apart from farmg ad devotes less tme ad efforts to farmg those that have secodary occupato also have a hgher yeld. Probably, secodary occupato meas a addtoal source of come. Also as the years of experece uplad rce farmg creases, yeld also creases. Educatoal backgroud s also postvely related to rce yeld, those wth formal educato havg a hgher yeld tha those wthout. Furthermore the result dcated that teracto betwee the seed voucher ad the covarates were statstcally sgfcat (prob >F=0.000), Suggestg that the teracto had effect o the yeld, thus cofrmg the presece of heterogeety the mpact of the recept of seed voucher o rce yeld. I addto, the postve sgfcat of the teracted terms of trag, farmg ad years of experece lowlad rce producto 19

20 mples that the mpact of seed voucher s hgher amog those farmers that had atteded trag before, had agrculture as a ma occupato ad also had a hgher umber of years of experece lowlad rce producto, whle the egatve sgfcace of the teracto of secodary occupato mples that the mpact of seed voucher wll be low for farmers wth secodary occupato. Although ot sgfcat, the postve coeffcet of the teracto term of educato meas that the mpact o yeld wll be hgher for those farmers that were educated tha the o-educated farmers Impact o Per Capta Rce Icome Several methods were also adopted to assess the mpact of seed voucher o per capta come of the rce farmg households Ngera. The result s preseted table 9. The mea dfferece was used as a tal aalyss sce the study adopted the RCT approach. The result showed that there was a postvely sgfcat observed dfferece of 17, per capta come betwee the treated ad the cotrol groups. For the estmato of the ATEs, the parametrc fuctoal form ad the Iverse propesty weghtg Estmates was adopted. The result of the parametrc estmato was postvely sgfcat. Usg The IPSW estmato techque, the ATE, ATE1 ad ATE0 were 3240, , ad respectvely. However, due to the problem of ocomplace, the ATE estmatos of the mpact of seed voucher o per capta come do ot have a causal meag. Hece, other to assess the true mpact of seed voucher o the rce farmers per capta come, two dstct approaches were adopted to calculate the LATE. Frst LATE was calculated usg the WALD estmator, followed by the Local Average Respose Fucto (LARF). The result of the WALD estmates reveals that the seed voucher has creased the farmg household s per capta come from rce producto by However, the result of the LARF showed a very low mpact of N The mpact across socoecoomcs characterstcs also vares. The mpact of the seed voucher s sgfcatly hgher for the female headed households ( ) tha the male headed households ( ). I addto, there was also a sgfcat dfferece the mpact across the otable rce producg ecologes Ngera. For stace, t has a hghest mpact rrgated ( ) followed by the lowlad ecology ( ) ad the uplad rce ecology ( ). Most mportatly, the seed voucher system ca also be sad to be pro-poor ature, sce t had a hgher mpact o the per capta come of the poor ( ) tha the o-poor ( ). Ths showed that t ca actually be employed as a tool for poverty eradcato, partcularly amog the farmg households where poverty s reportedly predomat. The determats of per capta rce come as revealed by the LARF estmate s show table 9. The aalyss dsaggregated the explaatory varables to two dfferet groups: 9 o- teracted terms ad 8 teracted terms. The o-teracted terms are the depedet varables that expla varato per capta come from rce producto.the result showed that apart from the recept of the seed voucher there were other socoecoomc characterstcs of the farmers that had sgfcat effects o the per capta come from rce producto. These varables cluded geder of household head, secodary occupato ad farm sze. 20