Saweda Onipede. L. Liverpool

Transcription

1 Povey saus and he impac of social newoks on smallholde echnology adopion in ual Ehiopia Saweda Onipede. L. Livepool Univesiy of Illinois, Ubana-Champaign Alex Wine-Nelson Univesiy of Illinois, Ubana-Champaign Seleced Pape pepaed fo pesenaion a he Agiculual & Applied Economics Associaion s 2009 AAEA & ACCI Join Annual Meeing, Milwaukee, WI, July 26-28, 2009 Copyigh 2009 by Livepool, S.O and Wine Nelson, A. All ighs eseved. Reades may make vebaim copies of his documen fo non-commecial puposes by any means, povided his copyigh noice appeas on all such copies.

2 Absac: Despie ecen aces of economic gowh, Ehiopia emains one of he pooes counies in he wold. Though abou 80% of is populaion is engaged in agiculue, agiculual poduciviy emains low and exemely vulneable o climaic condiions. The adopion and use of moden echnologies is geneally acceped as a poenial vehicle ou of povey fo many bu adopion aes in he couny emain low wih he naue of he adopion pocess lagely unsudied (Spielman e al, 2007). This pape sudies he impac of social newoks in he echnology adopion pocess in ual Ehiopia. In addiion o geogaphic newoks, i consides he ole played by ohe newoks wih moe puposeful ineacions such as a household s fiends. We also exploe he diffeenial impacs of social newoks by newok ype, echnology and he asse povey saus of households. Key wod: social newoks, povey, echnology adopion, Ehiopia JEL classificaions: O31 O33 Q12, Q13 Inoducion The longsanding effo o undesand he pesisence of povey has exposed he complexiy of is undelying sucue and dynamics. While hee is now a geneal consensus on he failue of canonical gowh models o saisfacoily explain pesisen chonic povey, moe complee models emain in developmen. Recen effos have idenified he ole of vaious exclusionay mechanisms which peven some households fom escaping povey and explain he divegen povey oucomes obained by diffeen goups. The suggle o undesand why ceain households o goups ae excluded fom economic gowh emains an acive aea of eseach. The ole of social newoks in shaping consumpion, poducion and exchange behavio is one aea of cuen debae (Bae, 2005).

3 Relaionships ae impoan o he adopion and disseminaion of moden echnologies in agaian economies, and paiculaly in ual Ehiopia. Despie he geneal view ha adopion and use of moden echnologies (like iigaion fo poducing high value cash cops) could seve as a vehicle ou of povey fo many, adopion aes in Ehiopia emain low (Spielman e al, 2007). Fame adopion of moden echniques and innovaion may be inhibied by lack of sufficien cedi o acquie inpus and make necessay invesmens o due o limied access o inpu and oupu makes. Ye anohe poenial deeen o he adopion of new echniques is inadequae infomaion on hei pacice. A common soluion o his consain has been he expansion of exension sevices, whose fom and efficacy can depend on he naue of social leaning as well fomal insucion. While he effec of social newoks in educing he infomaion consain has been shown o exis, i is only ecenly ha eseaches have begun o sudy moe how his effec occus and o disinguish beween effecs afe adopion and hose causing adopion. (Munshi, 2004, Bandiea and Rasul, 2006). Fuhemoe, vey lile eseach has been done o see if and how hese newok effecs diffe acoss households chaaceized by diffeen foms and levels of povey. In his ligh, his pape conibues o his eseach wofold: fis by invesigaing he oles ha social newoks in household echnology adopion in ual Ehiopia, and second by exploing he diffeenial effec of newoks acoss households a diffeen levels of povey and hus hei poenial o help households gow ou of povey. This sudy aemps o answe he following fou quesions: Fis, do newoks conibue o echnology adopion? Second, if newoks affec echnology adopion, wha kinds of newoks mae? Thid, can we find evidence of social leaning in newok effecs and fouh, ae hese newok effecs he same acoss households in diffeen foms of povey.

4 Social Leaning and Technology Adopion The heoy of social leaning in echnology adopion looks a how infomaion inenionally o uninenionally made available o a fame as a esul of decisions of ohe fames affecs echnology use. The fac ha social newoks affec echnology diffusion has been sudied widely. Howeve only ecenly have eseaches begun o sudy moe how his effec occus and o disinguish beween effecs afe adopion and hose causing adopion. Besley and Case (1994) pesen an ealy model of infomaion exenaliies in he adopion and diffusion of impoved coon culivas in he semi aid opics. Fose and Rosenzweig (1995) fuhe develop his model in a sudy of high yielding vaieies (HYVs) of whea and ice in India duing he Geen Revoluion. Munshi (2004) exends he analysis of Fose and Rosenzweig o show how social leaning diffes acoss heeogeneous populaions. These sudies disinguish beween he effecs of leaning by doing and leaning fom ohes. They show how infomaion consains limi echnology adopion and how own and neighbo expeiences educe his consain. This eseach follows moe ecen sudies o exploe newok effecs and social leaning pio o adopion. While seveal easons fo a posiive elaionship beween an individual s newok and hei pobabiliy of adoping a new echnology exis, social leaning heoies indicae ha he diecion of his elaionship beween newok size and adopion is ambiguous. A lage newok migh indicae access o moe infomaion abou a echnology fom he newok and hus encouage adopion. Howeve, infomaion fom pesonal expeience may be cosly o acquie and he expeience of ohes can subsiue i. Hence, a lage newok could encouage households o delay adopion and fee ide on he expeience

5 of membes of hei newok. (Badhan and Udy, 1999 ;Bandiea and Rasul,2006) 1. Fuhemoe, he effecs of social newoks could be heeogeneous depending on he kind of newok as well as on he chaaceisics of fames such as how infomed hey ae geneally and wih espec o he echnology. Mos pevious sudies focus on geogaphic poximiy as he causal explanaion fo coelaed adopion choices wihin social newoks. The geogaphic explanaion assumes all fames have unhindeed access o he necessay infomaion on he use of he new echnology when i is used in hei aea. Thus, eihe neighbos willingly shae infomaion o fames coslessly obseve each ohe s inpu use and oupu. Bief discussions wih ual fames acoss Ehiopia eveal ha his assumpion is no necessaily ue. Wih land allocaed by govenmen and passed on fom geneaion o geneaion, fames have lile choice as o who hei neighbos ae and ae no always on he bes ems wih hem. Fuhemoe vaious pocedues associaed wih a new echnology such as quaniy and applicaion ime of vaious inpus as well as iming of vaious managemen aciviies may no be casually obsevable bu necessiae moe puposeful ineacion. Social newoks based on chaaceisics ohe han physical poximiy migh be woh exploing in he bid o undesand how infomaion consains could be educed in ual Ehiopia. As evidence ha adopion newoks need no be based on physical poximiy, sudies like Sliche von Bah (1963) eveal ha duing he English agiculual evoluion, i was no uncommon o see fields being culivaed wih vey old adiional echniques shaing boundaies wih lands culivaed by newly inoduced cop oaion. Moe ecenly, Bandiea 1 This occus because expeced pofi is inceasing in boh he infomaion eceived fom own ials as well as fom ials fom ohes in he newok. Fuhemoe, he addiional infomaion gained fom pesonal ials declines, he lage he newok available fo he fame o lean fom. This will be discussed fuhe in he heoeicafamewok.

6 and Rasul (2006) find ha fame adopion decisions wee coelaed o he decisions of fiends and family as well as hose of he same eligion bu no fo hose in diffeen eligions. Similaly, in hei sudy on echnology adopion in Ghana, Conley and Udy (2001) find ha fames end o have a limied numbe of incomplee echnology infomaion souces no necessaily based on geogaphic poximiy. As fa as we ae awae, no such sudy has been conduced in ual Ehiopia. Given he impoance of infomaion in echnology adopion and he numeous effos o esucue and impove exension sevices in Ehiopia, i is impoan o undesand he naue and qualiy of social leaning among ual households. Beyond he infomaion exenaliy offeed by newoks, hee ae ohe possible easons why adopion choices could be elaed wihin vaious goups. As menioned, bu no fully exploed, by Bandiea and Rasul (2006), decisions wihin goups could be coelaed if hee ae ohe shaed goals fo, o consains o he adopion decision, such as economies of scale in commecializaion of a commodiy. Fo example, if hee is isk shaing wihin newoks o if he echnology in quesion is oo expensive fo an individual fame o buy and opeae, one migh expec a high degee of coelaion of adopion among goup membes. Similaly, goup effecs and dynamics could educe he willingness of individual fames o engage in new aciviies. Thus his pape sudies he effec of a household s newok of neighbos and fiends on hei echnology adopion decisions wih a view o disinguishing beween social leaning and ohe pee effecs. Finding an invese u elaionship beween he pobabiliy of adopion and he numbe of adopes in a households-infomaion newok as in Bandiea and Rasul (2006) will eveal social leaning and hus he poenial fo using ceain goups as a

7 vehicle o disseminae infomaion of new echnologies. While finding a sicly linea o u elaionship could indicae he pesence of social leaning, such a esul could also be explained by ohe newok effecs. Fo example, a u shape migh indicae a heshold effec whee a smalle newok wih shaed isks educes incenive o adop bu as his isk shaing goup ges lage, he high cos of failue is miigaed, hus encouaging adopion. A linea elaion shop could sugges benefis of pooling esouces o educe uni coss. Thus, findings may sugges whehe newoks have an impac and whehe ha impac is hough social leaning. Fuhemoe, idenificaion of diffeenial newok effecs acoss povey classes will also infom he planning and design of exension as well as ohe povey educion saegies in ual Ehiopia. Even if social newoks encouage echnology adopion, i is impoan o undesand if and how hei effecs diffe acoss households chaaceized by diffeen povey foms and dynamics. Pevious wok has indicaed ha educing fomal cedi consains ends o have no effec on he use of moden echnology fo he pesisenly asse poo hough he use of ceain echnologies, like feilize, assiss in hei abiliy o accumulae asses ove ime (Livepool and Wine-Nelson, 2009). Thus exploing whehe educing infomaion consains hough social leaning has a posiive effec on he use of hese echnologies by pesisenly poo could indicae is ole as a poenial vehicle ou of pesisen povey. Theoeical Famewok Social Leaning Social leaning is ofen measued using a age inpu model o a pofiabiliy model. The age inpu model lays emphasis on he fame s poblem of decipheing he opimal

8 managemen of a new echnology. This appoach conass wih ohe models of social leaning like Besley and Case (1993;1994) and Ellison and Fudenbeg (1993) which focus on he poblem of deemining he ue pofiabiliy of a new echnology fom pesonal and newok expeience. This sudy adops he age inpu model o focus on he ole ha newoks play in leaning when new cops o echnologies ae inoduced and he evidence of leaning abou he bes use of inpus fom ohes (Fose and Rosenzweig,1995). Secondly, i can be shown ha unlike in he case of unceain bu exogenous pofis, he pofiabiliy of any new echnology gows ove ime as knowledge accumulaes. Thus, as poined ou by Fose and Rosenzweig (1995), we can es fo leaning exenaliies diecly by looking a poduciviy. Inceasing pofiabiliy wih inceased knowledge accumulaion implies ha echnology adopion is an absobing sae. The above assumpion appeas moe appopiae in ou conex han he assumpion of complee leaning abou he echnology needed fo idenificaion as made by Besley and Case (1994, page 17). Also, ulimaely, he educed foms ha emege fom boh heoeic models capue fames leaning by doing and leaning fom ohes. The age inpu model 2 developed hee follows ha of Badhan and Udy (1999) as well as Bandiea and Rasul(2006). I assumes ha fames use Bayesian updaing o lean abou he paamees of a new echnology. While fames ae awae of he undelying poducion echnology, hey ae unawae of one paamee, i.e he age inpu level. The age inpu model assumes ha fames oupu in ime ;, declines in he squae of he disance beween he inpu used k, and he unknown inpu age, i qi c i 2 The age inpu model is a longsanding model which has been developed by Pesco(1972), Wilson (1975), Jovanovic and Nyako(1994) and applied wih egads o leaning in agiculue by Fose and Rosenzwig(1995).

9 q i = k i 2 1 ( c i ) (1.1) Though he age inpu level c i is no known a ime, afe he fame has seleced his inpu level k i and sees his yield, he updaes his belief abou wha he age inpu is. Each ime he fame makes a selecion of k i and ges a paicula yield is a ial afe which he is povided moe infomaion abou he disibuion of c i. Thus fames lean by doing. Because of fame and ime specific effecs, he opimal age fo fame i flucuaes aound c * is defined as: c i = c * + e i (1.2) Whee e i efes o hese ansioy fame specific shocks o he opimal age inpu c *. The eo is assumed o be independenly and idenically disibued nomal wih E( e ) =0 and V( e ) = s 2 u. i i A any ime, fame believes * ~ ( * 2 2 i c N c i, s c i ). The model assumes ha s u is known and also ha he inpu is cosless so ha fame s pofi is jus pice (nomalized o 1) q i muliplied by. Since E ( e ) =0, o maximize his expeced pofi, fame i uses his expeced opimal i age level as his new level of inpus. Thus, =E( c )= c * and expeced oupu is k i i E ( q ) = 1 - E [ -E ( c )] 2 = 1 - s i - s 2 2 k i i c i u (1.3) showing ha oupu inceases wih lowe levels of unceainy abou age inpu c. i Wih egad o leaning by doing, in each peiod, fame i engages in a ial wih a ceain level of age inpu k i, sees he oupu afe which he modifies his belief abou he

10 age inpu level. A ime, he vaiance of fame i s belief abou c * is s 2 ci. Afe obseving he age inpu fo he pevious peiod, c i 1, he fame updaes his belief abou he vaiance of c * applying Bayes s ule and as shown by Badhan and Udy (1999), his poseio belief becomes: s 2 c+ 1 = s 1 2 ci s 2 u (1.4) 1 If we define he pecision of infomaion geneaed by a fames own ial as 2 = s u o and io = s 1 2 ci0 as he pecision of fame i s iniial belief abou he vaiance of c *, we can show by subsiuion ha s 2 ci+ 1 = 1 + io I o (1.5) Whee I is he numbe of ials fame i has had wih he new echnology on his own fam beween peiods 0 and. Subsiuing (1.5) ino (1.3) we can expess cuen expeced pofis 3 as: E ( ) = 1- q i 1 + io I 1 o - s 2 u (1.6) Fom equaion (1.6), we can see ha oupu inceases wih he numbe of ials, i.e. E( q I i 1 ) o ) = 1-2 ( + I io 1 o ) >0 (1.7) 3 1 We acually have E( q i + 1 ) = 1 - io + I o - s 2 u which when pu in cuen ems gives us (1.6)

11 Now, conside he case whee a fame can impove his esimae of he age inpu by leaning fom he ials of ohe fames. If we define he newok of fames who shae infomaion as n( i) and assume ha fames in his newok coslessly shae infomaion, hen afe each peiod, fame i updaes his belief abou he age inpu wih no only infomaion fom his pevious ials, bu also fom hose of ohe newok membes j i. This means ha a ime whee fame i has had I 1 ials and he newok n i) ( 1 ials, his poseio belief abou he vaiance of c * will be s 2 ci = io + I 1 1 o + n( i) 1 o (1.8) wih expeced oupu now being E [ qi, n( i) 1 ] = 1- io + I 1 1 o + n( i) 1 o - s 2 u (1.9) wih oupu also inceasing wih he numbe of ials of he newok. 4 E[ qi, n( i) n( i) 1 1 ] o ) = 1-2 ( + I n( i) io 1 o + 1 o ) >0 (1.10) The echnology adopion decision Given he exisence of some available adiional echnology (adiional cop o vaiey), wih a known eun of q T a fame is faced wih he decision o adop a new echnology o no. Le he adopion of new echnology by fame i in ime be a a i dichoomous vaiable, such ha =1 if adopion occus and 0 ohewise. If leaning akes a i place as suggesed in he pevious secion, fame i s adopion depends on he adopion 4 I can be infeed unde his assumpions ha he lage he newok size, he lage he numbe of ials available fo fame i fom he newok.

12 decision of ohes in his newok. The value of fuue seams of pofis o fame i fom peiod o T is: V I, n( i) ] [ 1 = max 0,1} a i 1 = max 0,1} a i T s E d { 1 ais ) qt + aisqs[ I s 1, n( i) s 1]} ] (2.1) { s= ( 1 ai ) qt + ai Eq[ I, n( i) 1 ] + dv + 1[ I, n( i) ] { 1 ] (2.2) Whee s 1 0 is is he oal numbe of ials ha fame i has conduced up o and I = s = a n( i) including peiod s. and s 1 efes o he numbe of ials ha fame i s newok has had ove he same peiod. d is he discoun ae. Fom equaions 2.1 and 2.2, we can see ha echnology adopion by fame i depends on his expecaion of cuen pofis as well as he fuue expeced pofiabiliy of adopion. Expeced pofis ae inceasing in he numbe of ials of he new echnology. Thus, he numbe of ials posiively affecs expeced pofi which deemines echnology adopion. Fuhemoe, he fac ha expeced pofis incease wih he numbe of ials indicaes ha echnology adopion is an absobing sae. While seveal sudies have evealed examples of innovaions ha ae aemped and abandoned once poven less pofiable han alenaive echnologies, his would end o occu in places whee new echnology has no been esed fo conexual appopiaeness befoe inoducion. Based on infomaion in he suvey aeas, we feel safe o assume ha he new cops and vaieies exploed in his sudy have geneally poven o be supeio o he adiional cops on aveage once appopiae complemenay inpus and pacices ae also adoped; hese being he unknown in his pocess. Issues of elaive pofiabiliy will be capued by conols fo possible heeogeneiy among fames.

13 Equaions 2.1 and 2.2 allow he possibiliy ha adopion in ime migh occu even if he echnology is less pofiable han he adiional pacice in ha paicula peiod, as long as he benefi in he fuue fom an addiional peiod of pesonal ial and/o he ials of ohes in ime is sufficienly lage. If he loss in cuen expeced pofis is less han he discouned gain in fuue pofiabiliy fom he addiional ial of he new echnology, hen he echnology will be adoped in ime even if he cuen pofiabiliy is less han he adiional vaiey. This esul obains if he igh hand side of he following equaion: q T Eq(, n( i) ) d[ V+ 1 ( + 1), n( i) V+ 1( ), n( i) ] (2.3) is geae han zeo; [ V T s + ( + 1), n( i) V+ 1( ), n( i) ] = d ( q( s) q( s s= )) T = d s 1 + ( s 1) + n( i) + s 1 + n( i) s= + 1 io 0 0 i0 0 0 > 0 (2.4) Whee [ V+ 1( + 1) V+ 1( )] efes o he diffeence in value funcions if adopion occued in ime +1 when fame i has had s ials and value funcion esimaed in ime +1 whee fame has had s-1 (one less) ials. The igh hand side of equaion 2.4 is posiive, eflecing he incease in he expeced pofis due o he infomaion goen by he fame fom expeimening in ime. Howeve, while he enie value on he RHS is posiive, i is deceasing in n(i). If infomaion fom pesonal ials and he ials of ohes ae subsiues, hen as moe ohe fames use he new echnology, less addiion infomaion is gained by he individual fame s own expeimening. Thus if many of fame i s neighbos o associaes have chaaceisics ha

14 would lead hem o adop a new echnology ealy, i migh be in s bes inees o efain fom expeimening unil she has seen how ohes have done wih he new echnology. (Udy and Badhan, 1999). These opposing effecs can be seen by aking he deivaive of 2.5 which eflecs he necessay condiion fo a fame o adop he new echnology(cop) in ime, based on 2.3. ] ) ( ), ( [ ]) ) ( 1, ( [ ] ) (, [ 1 1 (2.5) which is ) ( ] ) ( ), [( ] ) ( 1, [ { ) ( ] ) (, [ 1 1 (2.6) = 2 1 ) ) ( ( ] ) ( [ ] ) ( 1) ( [ (2.7) The fis em in equaion 2.7 indicaes he posiive effec of leaning fom he newok hough i is deceasing in ) (. Howeve, given ha infomaion fom pesonal ials and one s newok ae subsiues, he lage a fames newok of adopes, he lowe he value of addiional infomaion fom pesonal ial is, ceaing an incenive fo he fame o saegically delay adopion. This can be seen by he negaive sign of he second pa of equaion (2.7). The sign of he oveall equaion depends on which value dominaes and consequenly he ne gains fom adopion in ime can be an inceasing o deceasing funcion of he numbe of adopes in a fame s newok. Bandiea and Rasul explain he sign of his elaionship as an indicaion of myopia amongs fames wih he moe myopic i T i n V q i n V i n q E i n i n V i n V i n i n q E o io o i n T s i io s i n s i n s i n = d d d d d

15 fames being less likely o delay saegically (hence posiive sign of ne gains) and vice vesa fo he less myopic fames. I is also impoan o noe fom equaion 2.7 ha he ne gains o adopion in ime is also a deceasing funcion of he accuacy of fame i s iniial infomaion abou he echnology. The moe accuae his own pesonal infomaion, he less impoan he addiional infomaion fom he newok will be, and he less sensiive he is likely o be o he numbe of adopes in his newok. Also, given ha he effec of newok size n(i) on adopion is posiive, bu deceasing in he size of he newok, ess fo a non-linea elaionship beween adopion and newok size can povide evidence of social leaning. Empiical Technology Adopion Decision Model We have shown via he social leaning model ha fames lean fom hei expeience as well as he expeience of ohes in hei newoks. This infomaion inceases he pofiabiliy of he echnology. The moe ials (and consequenly infomaion) fame i has had access o (fom he numbe of adopes in his newok) a ime, he moe pofiable echnology i is o him. We have also shown ha adopion of a new echnology fo any fame is a funcion of he value of cuen and fuue seams of pofis (given adopion) fo ha fame. Given ha pofiabiliy is an inceasing funcion of infomaion and ha infomaion fom one s own ial and he ials of his newok ae subsiues, we have shown ha while he diecion of he associaion is no ceain, he decision o adop a echnology a any poin in ime will depend o some exen on he size of his newok. The elaionships can be measued empiically hough esimaion of:

16 (TA ikv ) x = (a A ) x + ( l A Z iv ) x +f [n(i )] x + ( y AV i ) x + (E1 iv ) x i=(1 N); whee TA ikv x is he adopion of echnology k by household i whee household i belongs o povey saus class x. Z iv efes o a veco of exogenous vaiables capuing household i s demogaphic chaaceisics, as well as ohe facos ha affec a household s decision o adop a paicula echnology. This includes, household size, sex of household head, age of household head, highes yeas of educaion in he household, disance o he neaes make, size of land culivaed by household(in hecaes),value of household implemens, numbe of household membes engaged in full ime agiculual aciviies. unobseved vaiaions acoss villages ha could affec a household s echnology use decision. n( ) capues he social newok effecs and E iv is he eo em capuing unobseved individual and newok chaaceisics which affec household paicipaion. In he model, he social newok vaiable is measued as he epoed numbe of adopes among he fame s social newok, a he ime of adopion. The newoks exploed ae fiends and neighbos. Vaious specificaion fo f[n( )] ae exploed. The main ones ae a quadaic funcion o es fo he diecion of he elaionship beween newok size and pobabiliy of adopion. The ohe wo appoaches ae he use of splines o exploe possible heshold effecs in he size of vaious newoks and non paameic esimaions o confim he esuls of ohe specificaions. As in he adiional laen vaiable analysis, TA iv * epesens he household s pesen value of ne gains fom paicipaing in agiculual innovaion a ime. TA * iv = A [Z iv,v i f[n(i )] E iv ] While we canno see he ne pesen value ascibed by each household, we do obseve hei dichoomous decision o use a moden echnology o no. V i is a dummy o accoun fo

17 TA ikv =1 if TA ikv * >0 and TA ikv =0 ohewise we assume ha pob (TA ikv =1) = pob (E 1iv >-{f[n(i )] + Z iv,+ V i }) = F( -{f[n(i )] + Z iv,+ V i }) assuming symmey of he funcion descibing F() aound zeo and whee we exploe vaious specificaions fo f[ ]. As discussed ealie, social newoks ae expeced o have an effec on echnology adopion, bu he naue of his elaionship is no immediaely obvious. Thus in he fis model specificaion, we include he squaed newok effecs as anohe vaiable o es fo a quadaic polynomial fi and he u o invese u shape. P [TA ikv = 1] x = F[(a A) x + ( λaz iv ) x +? 1[n( i )] x +? 2[n( i ) 2 ] x + ( ψav i ) x + (E1 iv ) x ] Then we also exploe possible heshold effec in newok size esing he diffeenial effec of having a newok size of (1-4),(5-8) and 8+ membes in he newok engaged in a paicula echnology a he ime of adopion elaive o having no one in he newok using he echnology. P [TA ikv = 1] x = (a A ) x + ( l A Z iv ) x + ß 0 [0]+ ß 1 [1] + ß 2 [2] +ß 3 [3] + ( y AV i ) x + (E1 iv ) x Whee [0], [1], [2] and [3] eflec diffeen splines Daa This sudy uses a subse of he Ehiopia Rual Household Suvey (ERHS) daase and addiional daa colleced fom wo ERHS villages. The ERHS daase conains deailed infomaion on consumpion expendiue, asses and agiculual aciviies of ual Ehiopian households and is he poduc of a longsanding daa collecion effo by Oxfod Univesiy, he Univesiy of Addis Ababa, and he Inenaional Food Policy Reseach Insiue (IFPRI).

18 I saed in 1989, when a suvey eam visied seven peasan associaions in Cenal and Souhen Ehiopia. In 1994, he suvey expanded o 15 peasan associaions (PAs) acoss fou egions, yielding a sample of 1477 households. Addiional ounds wee conduced in lae 1994, 1995, 1997, 1999 and Duing 2007, supplemenay communiy level suveys wee adminiseed in all 15 PAs o idenify ecen changes in he villages, paiculaly beween 2004 and Supplemenal household suveys wee also adminiseed in 2 ou of he 15 PAs. These 2 villages, Haesawe and Koodegaga, wee seleced based on local infomaion and communiy level suveys ha indicaed ha innovaive echnologies had been inoduced in hese egions. Daa on adopion of impoved echnologies including impoved vaieies of vaious ceeals and iigaed vegeables wee colleced fom 186 households in hese wo PAs. Demogaphic infomaion as well as infomaion on hei asses, access o vaious insiuions, social newoks and he pevalence of echnology adopion wihin hese newoks wee also colleced. Ohe daa ae based on he peviously colleced daa fom he 6 ounds of he suvey conduced beween 1994 and The 2 main echnologies exploed hee ae impoved ceeals and iigaed vegeables and pulses. While hese echnologies ae no band new, discussions wih fames and developmen agens indicaed ha hee has been a ecen emphasis on he poducion of high value cops such as fuis and vegeables as well as ohe makeable cops like pulses and impoved ceeals. As a conol fo new echnologies, we exploe he social newok effecs in a elaively old echnology (chemical feilize) on ecen adopes. Social leaning is no expeced o affec adopion of his olde echnology. In Haesawe, he fames and he developmen agen cied 2004/2005 as he peiod of majo shif in he village in ems of

19 inceased focus on field pea poducion and 2004 as a yea fo inceased vegeable poducion. Fo iigaion of pulses, fuis and vegeables, we esiced he analysis o households who had adoped iigaion since Niney wo pecen (92%) of hese households wee engaged in iigaing fuis, vegeables, pulses o oil seed in This sudy explicily disinguishes among households by hei asse povey saus o discen diffeenial effecs of social newoks on he pobabiliy of fames adoping echnology. This analysis begins by using he complee ERHS daase o classify households by hei asse povey saus. The asse based appoach o povey measuemen classifies as asse poo asse hose households whose asses ae inadequae o geneae an income seam suppoing consumpion above he expendiues povey line (Cae and Bae, 2006). An asse povey line is defined as he asse value ha exacly suppos consumpion a he expendiues povey line. In his applicaion an asse index is esablished as a funcion of he household s land, livesock, fam implemens, ohe physical asses, and educaion. The weighs on each componen of he asse index ae based on an esimae of he elaionship beween asses and consumpion as descibed in appendix A. Households whose asse index was below he asse povey line in each suvey yea ae classified as always asse poo. Neve asse poo households ae hose whose asse index was above he povey line each yea, and households whose saus changed beween yeas ae classified as Tansioy asse poo. Because vey few households in he wo villages wee in he neve asse poo caegoy, he analysis disinguishes only beween households who wee pesisenly asse poo (consideed o be in a povey ap) and hose who wee no. 5 Though some aspecs of iigaion such as pesonal digging of wells, seing up wae havesing ponds and seing up of small scale dip iigaion have been ecenly inoduced on a wide scale.

20 Tables 1(a) o 1(c) descibe newok sizes fo adopes and no adopes acoss newoks and acoss povey classes fo he echnologies consideed. The esuls show ha on aveage, adopes of impoved seed had moe fiends who had peviously adoped han had no adoped. Howeve, he mean numbe of neighbos who had adoped ealie is highe fo non adopes han fo adopes. Wih egad o povey saus, adopes in povey aps had fewe adopes in hei newok han pesisenly poo non adopes. In conas hose adopes who wee no in a povey ap had moe adopes among hei fiends, bu fewe among hei neighbos han non-poo non-adopes. Compaed o hose in povey, he non-poo have moe adopes in hei newoks of fiends bu no in hei newoks of neighbos. Fo neighbos, we find lage numbe of adopes in he newok of non adopes compaed o adopes acoss boh povey caegoies, hough he diffeence in means is much highe amongs hose no in a povey ap. This suggess ha leaning fom newoks may no be defined by space bu ahe by ohe ineess. Wih egads o iigaed cop we find a highe mean of adopes in boh he fiend and neighbo newok among adopes ahe han non-adopes hough he diffeence in mean adopes is highe wihin he fiend newok han in he neighbo newok. While he mean adopes in he newok of non-adopes is simila acoss povey saus, adopes of iigaed cops amongs households in a povey ap ae significanly highe han hose no in a povey ap; 8 vs 5. Feilize, like iigaed cops eveals moe adopes in he newok of adopes compaed o non-adopes. Thus ables 1a 1c appea o indicae he pesence of some so of newok effecs, possibly diffeen acoss newok ypes and povey saus. The descipive saisics in able 2 eveal he elaively poo naue of ou sample. Households end o culivae abou 2 hecaes of land, be headed by middle aged men of

21 abou 50 yeas old and have on aveage someone wih a maximum of abou 5 yeas of educaion. Thei asses end o compise of 1 o 2 head of livesock valued a abou 400EB 6 wih households in povey ap ending o have lowe asses, moe people engaged in full ime faming and less accessibiliy o makes. Esimaion Resuls: Impoved Ceeals Given he binay naue of he adopion vaiable, we exploe pobi, logi and linea pobabiliy models. In he case of ecenly adoped impoved vaieies of ceeals, we find evidence of social leaning ha vaies by newok ype. As can be seen fom able 3 below, he pobabiliy of adoping impoved seeds exhibis he invese u elaionship wih espec o he numbe of fiends who had peviously adoped impoved seed use. While he maginal effecs on he level em ae posiive and significan, he maginal effecs on he squaed em ae negaive and significan. On he ohe hand he neighbo newok ends o have an insignifican effec on he odds of adopion. We also find ha younge households and households culivaing lage landholdings ae moe likely o adop he impoved seed. The close he household is o a paved oad, he moe likely hey ae o adop. We diffeeniae beween he closes make used (usually he local peasan associaion make) and access o ohe makes which is an indicao of moe commecializaion oppouniies. Nex we es o see if his behavio diffes by povey saus. Resuls (shown in able 4) indicae ha evidence of social leaning sill exiss bu vaies acoss boh newok and povey saus. While he effec of fiends coninues o exhibi he invese u elaionship fo 6 One US dolla is equivalen o abou 11 Ehiopian Bi (EB). The PPP convesion faco is appoximaely 0.25.

22 he households no in he povey ap, he effec is no saisically diffeen fom zeo fo hose households in a povey ap. This indicaes ha hee ae diffeenial social leaning effecs no only acoss newok ypes bu also acoss povey levels. I appeas ha while he level vaiable fo he numbe of neighbos who have adoped has no effec on he odds of adopion fo all households, he squaed em is posiive fo he households in a povey ap, hough only significan beyond 10% in he logi esimaion. This diffeence acoss povey levels may eflec diffeen kinds of newoks o efficacy of newoks by povey class. Whee newok membes ae less knowledgeable o infomaion ansfe is less efficien, a lage numbe of infomans is needed fo adequae infomaion o igge adopion and he pomise of gaining infomaion fom he newok is less likely o dee own expeimenaion. Poo households may be moe likely o be in such newoks. Alenaively, he diffeen esuls by povey class could also indicae ohe newok effecs such as economies of scale if indeed hese ae cops fo commecializaion. Since he neighbo effec is no saisically significanly diffeen acoss povey saus, such an explanaion migh be feasible fo households likely o commecialize Iigaed fuis, vegeables and pulses Nex we exploe he same model fo iigaion of pulses and vegeables. Consideing he adopion pocedue fo ecen adopes (fom 2004), again we find ha newok effecs fo fiends who had peviously adoped exhibis he invese u elaionship bu his is no eviden fo neighbos who have aleady adoped. Consideing he sepaae effecs of he vaious newoks only he effec of fiends is significan, sill exhibiing he invese u elaionship indicaing social leaning. Coefficien values sugges inceasing pobabiliy of

23 adopion up o abou 10 fiends, and hen evesal in he effec. To validae hese esuls, we aemp non-paameic esimaion of he effecs of hese newoks. Figues 1 and 2 show ha while he fiends newok seems o exhibi he invese u elaionship, he neighbo newok effec appeas o be moe of an inceasing funcion of he newok size. While hee ae some newok effecs of neighbos, social leaning is moe eviden in newoks whee hee is moe inenional ineacion. Ohe facos ha appea o affec adopion of iigaed cops ae access o commecializaion oppouniies, households wealh capued by non poducive asses including jewely and ohe household iems. Again we es fo a diffeenial effec of he fiend s newok acoss povey class. Fom able 6, we find ha households wih moe non poducive asses, bee access o exenal makes and lage culivaed land size ae moe likely o adop iigaed cops. We find ha hough a newok effec of fiends is pesen, i does no clealy eveal he invese u elaionship and does no appea o be saisically significanly diffeen acoss povey classes 7. Howeve, he small sample size may be esponsible fo low levels of saisical significance. Thus we fuhe exploe any diffeence acoss povey saus by a non paameic esimaion. The non paameic esimaes in figues 3 and 4 confim ha hee ae simila effecs of a household s newok of fiends who had peviously adoped iigaed cops on hei pobabiliy of adopion acoss households in diffeen povey caegoies. The esuls eveal an invese u elaionship beween newok size and pobabiliy of adopion, ha may have been obscued in he paameic esimaion due o he small sample size. 7 Given ou small sample size Jus o confim hese esuls, we exploe moe pasimonious specificaions such as dopping he non poducive asse measue given ha his migh be coelaed wih povey saus. This does no change he esuls Fuhemoe he coelaion coefficien beween vaious vaiables indicaes ha we do no have a poblem wih mulicollineaiy.

24 Given he weakness of he esuls in able 6, we es fo a heshold effec on he fiends newok by inoducing splines. Table 7 shows he maginal effecs of vaious facos on he adopion of iigaed cops 8. The esuls eveal ha compaed o households who have no fiends who have adoped iigaed cops, households wih beween 1 and 4 fiends who had adoped have a 40% highe pobabiliy of adoping iigaed cops. Fo hose wih beween 5 and 8 fiends ae abou 67% moe likely o adop he echnology and while having moe han 8 fiends using a echnology ae less likely han hose wih 5 and 8 o adop, hey ae 62% moe likely han hose who have no fiends using a echnology. This fuhe indicaes ha he elaionship beween he size of he fiend newok and pobabiliy of adopion is shaped as an invese u. A es on he equaliy of coefficien eveals ha he coefficien on 1-4 fiend adopes is saisically significanly diffeen fom having 5-8 membes a 1%, and while we fail o ejec ha having 1-4 is saisically significanly diffeen fom having moe han 8 membes we also fail o ejec ha having 5-8 is saisically significanly diffeen fom having moe han 8 membes. Feilize Nex we exploe he esuls fo feilize. Since feilize is an old echnology, we do no expec o find song evidence of social leaning, bu newok effecs could sill exis. We exploe he facos likely o deemine feilize adopion among households who have adoped feilize since We find ha households who had ecenly begun o gow pulses and oil seed, impoved seeds and vegeables wee moe likely o be ecen adopes of feilize use. Wealhie households and households wih moe educaed membes wee also moe likely o be ecen feilize adopes. When he size of diffeen newoks was 8 As expeced, he logi and pobi esuls ae consisenly he same acoss his analysis. Howeve, we decided o use he maginal effecs hee o ease he inepeaion of he newok effec of he spline esuls.

25 consideed sepaaely, we find evidence of newok effecs among fiends only. The esuls in columns 5-7 in able 8 show ha hee appeas o be a sicly inceasing elaionship beween numbe of adopes in a households goup of fiends who had peviously been using feilize and hei likelihood of adoping feilize use fo ecen feilize adopes. This seems o indicae some newok effecs exis, bu no necessaily social leaning. Inceasing euns o he numbe of adopes of feilize use migh be indicae economies of scale effecs of a household s newok. Given ha he use of feilize is songly associaed wih commecial cop poducion, i makes sense ha scale effecs in coodinaing inpu pocuemen and oupu sale migh be pesen. Finally we exploe he diffeenial effec acoss povey classes. The main conclusions fom able 9 is ha social newok effecs hough weake sill appea o exhibi inceasing euns o scale, bu only fo hose households who ae no in a povey ap. Fo hose in a povey ap, i appeas ha social newok effecs ae no saisically significanly diffeen fom zeo. We find evidence of some newok effec exclusion fo he pesisenly poo households hough we canno idenify he mechanism hough which his exclusion occus. This is woisome if hose households in a povey ap ae no able o ake advanage of newok effecs such as coodinaion in he pocuemen of inpus o makeing of oupus necessay fo hei adopion of yield enhancing echnologies. Howeve, his highlighs possible diffeences beween he ype of newoks ha households in a povey ap use compaed o hose no in povey ap as well as hei easons fo he use of feilize. If hese poo households ae using feilize o incease poducion o impove poduciviy and no necessaily fo commecializaion, i migh make sense ha newok effecs suppoing commecializaion oppouniies migh no be necessay. Howeve, if hee ae inpu

26 pocuemen benefis o newok membes, his lack of significance implies ha he pooes households ae excluded fom such oppouniies wih moe sevee consequences. Conclusions: This pape exploed he ole of social newoks on adopion of new echnologies in ual Ehiopia. I found evidence of social leaning, hough his diffes acoss newok ypes and wih povey saus. Fo moden echnologies mos ecenly inoduced ino he villages sudied (impoved ceeals), we find evidence of social leaning, ha opeaes hough he newok of fiends among households no in a povey ap. We find ha newoks of neighbos who have adoped is only significan fo households in a povey ap and even hee, i is only a vey lage newok sizes ha he neighbo newok is posiively associaed wih adopion of impoved seed Fo ecenly adoped iigaed vegeables, pulses and oil seeds, we sill find evidence of social leaning and encouagingly his is no saisically significanly diffeen acoss povey class. Howeve, again we find evidence ha social leaning occus acoss newoks fo which hee is moe puposeful ineacion ahe han ha povided by geogaphical poximiy. Fo a well known echnology, feilize, we find evidence of newok effecs bu no social leaning. Again we find ha newok effecs diffe by povey saus, wih newoks unfounaely no affecing feilize adopion fo he pooes households. The finding ha social leaning effecs ae available in ual Ehiopia is and ha hey emege hough puposeful ineacion ahe han poximiy is significan fo exension planning. Technology diffusion is likely o be enhanced if exension can each moe newoks of inees. This implies a need o age inenional goups of ual people ahe han spaial cluses. Idenifying such goups pesens a challenge o exension sevices.

27 One mechanism fo ageing goups migh be local iddis. Iddis ae adiional communiy based insuance schemes o which households peiodically conibue a pedeemined amoun of money o seve as insuance in he even of deah of a membe of he family o ohe shock like healh elaed advesiies. Though Iddi aangemens ae infomal, hey ae well coodinaed and oganized wih long life spans and elaively high levels of us amongs membes. This oganizaion aleady plays muliple oles in he ual envionmen, and migh be a condui fo social leaning ha exension sevices could employ. Ou peliminay esuls indicae he pesence of social leaning effecs in Iddis bu due o mulicollineaiy beween he Iddi newok adopes vaiable and fiends newok, we focus on he fiends in his pape

28 Tables and Figues: Table 1a. Mean numbe of newok membes aleady using impoved seed by newok ype fo ecen adopes of impoved seed Newok ype Complee Sample Povey Tap No in a Povey Tap Fiends Toal Adopes Non Adopes Adopes Non Adopes Adopes Non Adopes (10.30) (11.02) (9.47) (7.25) (13.95) (14.45) (4.74) Neighbos (12.56) (4.63) (7.24) (4.99) (14.44) (3.48) (10.64) Obsevaions (Sandad deviaion in paenhesis) Table 1b. Mean numbe of newok membes aleady pacicing iigaion by newok ype fo ecen adopes of iigaed cops Newok ype Complee Sample Povey Tap NonPovey Tap Fiends Neighbos Toal Adopes Non Adopes Adopes Non Adopes Adopes Non Adopes (7.747) (9.505) (3.290 (11.121) (2.629) (7.559) (3.761) (3.975) (4.455) (2.925) (4.893) (2.262) (4.028) (3.346) Obsevaions (Sandad deviaion in paenhesis)

29 Table 1c. Mean numbe of newok membes aleady using feilize by newok ype fo ecen adopes of feilize Newok ype Complee Sample Povey Tap NonPovey Tap Fiends Toal Recen Adopes Non Adopes Recen Adopes Non Adopes Recen Adopes Non Adopes (7.182) (9.789) (5.403) (11.364) (4.689) (8.563) (5.344) Neighbos (8.177) (9.143) (6.754) (10.877) (6.052) (3.528) 6.059) Obsevaions (Sandad deviaion in paenhesis) Table 2. Descipive saisics Descipive Saisics Vaiable Complee Sample Povey Tap Non Povey Tap MaleHead(1/0) (0.49) (0.50) (0.50) Household Livesock(Ehiopian Bi) (461.86) (302.17) (491.03) Household Non Poducive asses (Ehiopian bi) (526.07) (419.27) (559.99) Agehead (yeas) (13.99) (12.45) (17.07) Disance o closes make(km) (3.66) (7.96) (8.04) Disance o paved oad(km) (8.06) (4.23) (3.05) Household Land Culivaed (hecaes) (2.74) (2.05) (3.15) Fullime Fam Labo (Numbe) (1.14) (1.15) (1.10) Mos Educaion(yeas) (3.19) (3.29) (3.09) Numbe of obsevaions (Sandad deviaion in paenhesis)

30 Table 3: Social Newok effec on he adopion of impoved ceeals Logi esimaion esuls Pobi esimaion esuls Impoved seed Odds Raio P>z Coefficien Maginal Effec P>z MaleHead(1/0) Household Livesock Household NonPoducive asses Agehead (yeas) 0.967* * Disance o closes make(km) Disance o paved oad(km) Household Land Culivaed 1.099* Fullime Fam Labo (Numbe) Mos Educaion(yeas) Fiend Newok size ** ** Fiend Newok size Squaed ** ** Neighbo newok size Neighbo newok size squaed Haesawe *** *** Numbe of Obsevaions Pob > chi Pseudo R *=significan a 10% **=significan a 5% ***significan a1%

31 Table 4. Social Newok effecs esimaion by povey saus Logi esimaion Resuls Pobi esimaion esuls Odds Maginal NewCeeal Raio P>z Coefficien effecs P>z MaleHead(1/0) Household Livesock Household NonPoducive asses Agehead (yeas) * * Disance o closes make(km) Disance o paved oad(km) Household Land Culivaed Fullime Fam Labo (Numbe) Mos Educaion(yeas) No in a povey ap Fiends*PovTap Fiends*PoveyTapSquaed 1.015* * Fiends*NonPoveyTap 1.538* * Fiends*NonPoveyTapSquaed 0.985* * Nighbos*PoveyTap Neighbos*PoveyTapSquaed 1.000** Neighbos*Non PoveyTap Neighbos*NonPoveyTapSquaed Haesawe 0.050*** *** Numbe of Obsevaions Pob > chi Pseudo R *=significan a 10% **=significan a 5% ***significan a1%

32 Table 5: Social Newok effec on he adopion of iigaed cops Iigaed cop Logi esimaion Resuls Odds Raio P>z Coefficien Pobi esimaion esuls Maginal effecs MaleHead(1/0) Household Livesock Household NonPoducive asses 1.002*** *** Agehead (yeas) Disance o closes make(km) Disance o paved oad(km) 0.929*** *** Household Land Culivaed 1.461*** *** Fullime Fam Labo (Numbe) Mos Educaion(yeas) Fiend Newok size 2.718*** *** Fiend Newok size Squaed 0.924** *** Neighbo newok size Neighbo newok size squaed P *** *** Numbe of Obsevaions Pob > chi Pseudo R *=significan a 10% **=significan a 5% ***significan a1% P>z

33 Table 6: Social newok effecs on he adopion of iigaed cops Iigaed cop Logi esimaion Resuls Odds Raio P>z Coefficien Pobi esimaion esuls Maginal effecs MaleHead(1/0) Household Livesock Household NonPoducive asses *** *** Agehead (yeas) Disance o closes make(km) Disance o paved oad(km) *** *** Household Land Culivaed *** *** Fullime Fam Labo (Numbe) Mos Educaion(yeas) No in a povey ap Fiends*PovTap ** *** Fiends*PoveyTapSquaed Fiends*NonPoveyTap Nighbos*PoveyTap P *** *** Numbe of Obsevaions Pob > chi Pseudo R *=significan a 10% **=significan a 5% ***significan a1% P>z

34 Table 7: Social newok effecs by splines on he adopion of iigaed cops Iigaed Cop Maginal effecs Robus sandad eo P>z MaleHead(1/0) Household Livesock Household NonPoducive asses Agehead (yeas) Disance o closes make(km) Disance o paved oad(km) Household Land Culivaed Fullime Fam Labo (Numbe) Mos Educaion(yeas) Having 1-4 adopes ** Having 5-8 adopes *** Having moe han 8 adopes ** Haesawe *** Numbe of Obsevaions 85 Pob > chi Pseudo R *=significan a 10% **=significan a 5% ***significan a1%

35 Table 8: Social Newok effecs on he adopion of feilize use Fiends and Newoks Fiends only Neighbos only Odds Raio P>z Coef. P>z Odds Raio P>z Coef. P>z Odds Raio P>z Coef. P>z Feilize Male Head (1/0) HH Livesock New Pulse New Ceeal NewVeg Ohe Asses Agehead (yeas) Disance o closes make (Km) Disance o paved oad(km) HH Land Cul Fullime Fam Labo (Numbe) Mos Educ. (yeas) Fiend Newok size ** Fiend Newok size Squaed ** ** Neighbo newok size Neighbo newok size squaed