Effect of non-farm work on agricultural productivity

Size: px
Start display at page:

Download "Effect of non-farm work on agricultural productivity"

Transcription

1 WIDER Workng Paper 2017/38 Effect of non-farm work on agrcultural productvty Emprcal evdence from northern Ghana Benjamn Tetteh Anang* February 2017

2 Abstract: Ths paper nvestgates the factors nfluencng partcpaton n non-farm work and the effect of partcpaton on farmers productvty, usng survey data from 300 smallholder farm households n northern Ghana. The study employs an endogenous swtchng regresson model to address selecton nto non-farm work, and a treatment effects model to measure the effect of partcpaton on productvty. Factors determnng partcpaton n off-farm actvty nclude the head of household s gender and years of formal educaton, the locaton of the farm, ownershp of cattle, and the dependency rato. Factors affectng productvty nclude gender, years of formal educaton, farm sze, locaton of the farm, access to credt, herd ownershp, and degree of specalzaton n rce producton. Results from a treatment effects model ndcate a postvely sgnfcant effect of non-farm employment partcpaton on farm productvty. Income dversfcaton therefore remans an mportant lvelhood strategy among smallholders, and earnngs from off-farm work enable smallholders to mprove ther yelds. Keywords: non-farm work, northern Ghana, endogenous swtchng regresson, smallholder farmers, productvty JEL classfcaton: C21, D24, Q12 Acknowledgements: The author thanks Rachel M. Gsselqust for beng hs mentor durng hs PhD nternshp at UNU-WIDER n Helsnk. The author s also grateful to Mguel Nño-Zarazúa for hs comments and useful suggestons durng the preparaton of ths workng paper. * Unversty of Helsnk, Fnland, benjamn.anang@helsnk.f. Ths study has been prepared wthn the UNU-WIDER PhD nternshp programme. Copyrght The Author Informaton and requests: publcatons@wder.unu.edu ISSN ISBN Typescrpt prepared by Merl Storr. The Unted Natons Unversty World Insttute for Development Economcs Research provdes economc analyss and polcy advce wth the am of promotng sustanable and equtable development. The Insttute began operatons n 1985 n Helsnk, Fnland, as the frst research and tranng centre of the Unted Natons Unversty. Today t s a unque blend of thnk tank, research nsttute, and UN agency provdng a range of servces from polcy advce to governments as well as freely avalable orgnal research. The Insttute s funded through ncome from an endowment fund wth addtonal contrbutons to ts work programme from Denmark, Fnland, Sweden, and the Unted Kngdom. Katajanokanlatur 6 B, Helsnk, Fnland The vews expressed n ths paper are those of the author(s), and do not necessarly reflect the vews of the Insttute or the Unted Natons Unversty, nor the programme/project donors.

3 1 Introducton The farm economy lterature on developng countres n the 1950s and 1960s regarded farm households as dependent on ther own farmng to meet ther food securty needs. If households were unable to generate enough food to meet ther own requrements from the farm, they would engage n off-farm waged employment, or would send out mgrants to work n the ctes and send back remttances (Davs et al. 2009). However, new research fndngs show that rural non-farm employment currently comprses roughly 40 60% of rural ncomes n Latn Amerca, Asa, and Afrca. Farm households may now be decdng to work off-farm for varous reasons, ncludng lmted access to producton nputs, better off-farm wages, declnng farm wages, and the desre to mtgate producton rsks. The lkelhood of engagng n off-farm work ncreases wth the declnng proftablty of farmng and the emergence of rural non-farm employment opportuntes. In many developng countres, farm ncomes are declnng, and opportuntes for non-farm work are emergng, especally n the ctes. The declne n farm wages and emergng opportuntes for work outsde the farm sector can promote households partcpaton n rural non-farm work. As noted by Davs et al. (2009: 119), RNFE tends to be, n most countres, even much more mportant than mgraton and farm wage-labor ncome and often more mportant than cash croppng as a source of ncome and lqudty. The ncreasng mportance of non-farm work (and decreasng mportance of agrculture) to smallholder farm households s reported n some studes. For example, Mudhara (2010) reports that smallholder farm households n South Afrca derve only a small fracton of ther lvelhoods drectly from agrculture, wth remttances, off-farm work, and government transfers provdng the man source of lvelhood. The partcpaton of farm households n off-farm work s a major decson that has mplcatons for farm output and productvty n many developng countres. For example, nvestng the ncome from rural non-farm work n new producton technologes and farm nputs may enhance the household s producton effcency and agrcultural productvty. Income from off-farm work can also supplement on-farm ncome, thus contrbutng to household welfare and food securty. However, when the ncome from non-farm sources s spent on non-farm actvtes (e.g. acquston of household consumables, such as a new televson set), productvty growth may not be realzed. Thus the way farm households utlze ncome from rural non-farm employment s mportant for brngng about mproved farm effcency. Furthermore, f nvolvement n off-farm work places constrants on household labour for farm operatons, then ths wll adversely affect farm performance. Despte the mportant role of rural non-farm work n household welfare and food securty, not much s known about ts effect on farm performance, partcularly wth regard to the effcency and productvty of smallholder agrcultural producton n Ghana. Much of the research on off-farm work n developng countres, ncludng the lttle that relates to Ghana, has focused on the mpact of workng outsde the farm on household welfare and food securty (Owusu et al. 2011). What s unclear, partcularly n the Ghanaan context, s the effect of off-farm work on the productvty of smallholder farmers, who are the man food producers n Ghana. Ascertanng the effect of offfarm work on farmers productvty wll help the formulaton of polcy to address some of the challenges of rural poverty and low agrcultural productvty among smallholder farmers n Ghana and other developng countres. As ndcated earler, ncome from off-farm work may not necessarly support on-farm nvestment, even though ths s a common assumpton regardng partcpaton n off-farm actvty. Ths assumpton s based on the noton that smallholder farmers depend on agrculture as ther man source of employment and wll therefore nvest ther extra 1

4 ncome n farmng. Authors such as de Janvry et al. (2005), Ells and Freeman (2004), Anrquez and Dadone (2010), and Stampn and Davs (2009) have reported an ncrease n farm nvestment as a result of off-farm labour. However, ths vew contrasts wth Ahtuv and Kmh (2002) and Davs et al. (2009), who suggest that ncome from off-farm work does not necessarly support onfarm nvestment. The current study therefore emprcally nvestgates the effect of non-farm work on the productvty of smallholder rce producers n northern Ghana, usng household survey data collected n The rest the paper s structured as follows. Secton 2 provdes a bref lterature revew, followed n secton 3 by an outlne of the methodology used for the study. The methodology ncludes the conceptual framework, emprcal models, and data. The results of the study are presented n secton 4, wth a dscusson of the results n secton 5. The concluson and polcy recommendatons are presented n secton 6. 2 Lterature revew Among the challenges confrontng Ghana and many developng countres are food nsecurty and poverty (Owusu et al. 2011). Urban and rural poverty coexst n Ghana, but the ncdence of rural poverty seems to be exacerbated by declnng farm ncomes, low agrcultural productvty, and lmted employment opportuntes outsde the farm sector. Rural farm households have therefore devsed varous means to overcome the challenges of poverty and food nsecurty. Partcpatng n off-farm employment actvtes and sendng mgrants to waged employment n the ctes are some of the means adopted by rural households to obvate poverty and food nsecurty. There s a vew that ncreasng food producton as the means to tackle food nsecurty n Afrca may not be adequate; hence the need to promote rural non-farm employment to provde farmers wth addtonal ncome (Gladwn et al. 2001). In many rural areas of sub-saharan Afrca, non-farm employment opportuntes may be consdered a possble soluton to the vcous cycle of food nsecurty. In lght of the foregong, t seems logcal that attenton to the generaton of rural employment should be mportant n the fght aganst rural poverty and food nsecurty n Afrca. Avalable evdence ndcates that a hgh proporton of farm households and small-farm operators engage n non-farm waged employment n many developng countres as an addtonal source of household ncome. In addton, the proporton of ncome from off-farm work n the total household ncome portfolo n developng countres s reportedly hgh (Davs et al. 2009). For example, the percentage of rural ncome derved from off-farm work n Afrca, Asa, and Latn Amerca s reported as 42%, 32%, and 40% respectvely (Reardon et al. 1998). Escobal (2001) found rural non-farm employment to consttute 51% of rural household ncome n Peru. A study conducted n the Hube provnce of Chna by de Janvry et al. (2005) found an estmated 36% of rural household ncome to come from non-farm employment. The authors also found households that partcpated n non-farm employment to have hgher ncomes than households that dd not partcpate n non-farm employment. Furthermore, the authors found partcpaton n non-farm work to reduce ncome nequalty. In a study of Mexcan households, Pfeffer et al. (2009) found the proporton of household ncome from rural non-farm employment to be 33%. In addton, Klc et al. (2009) found the proporton of household ncome from rural non-farm employment to amount to 50% n Latn Amerca and 35% n Afrca. At the global level, the fgure s estmated at 58%. As noted by Chang and Wen (2011), partcpaton n off-farm work s a persstent phenomenon, and relance on off-farm work s expected to ncrease. Partcpaton n off-farm work s not a phenomenon restrcted to farmers n developng countres. Bag (1984), Fernandez- 2

5 Cornejo et al. (2007), and Kumbhakar et al. (1989) have reported partcpaton n off-farm actvty among farmers n developed countres such as the Unted States. A number of studes have sought to explan the factors leadng to the rse n off-farm actvty among farm households, partcularly n developng countres. For example, Lanjouw (1999) dentfed declnng farm ncomes and the need to mtgate producton rsks as factors leadng to the rse n off-farm actvty among farm households. Barrett et al. (2001) conceptualzed the relatonshp between the farm and non-farm sectors n terms of pull and push factors. The dea s that there are certan factors pullng or pushng farm households to engage n off-farm work. An example of ths s lmted landholdng, whch pushes farm households to work off-farm (van den Berg and Kumb 2006). Cunguara et al. (2011) also found that as a result of drought, households n Mozambque resorted to off-farm work as a copng strategy. Furthermore, a study by Mathenge and Tschrley (2015) ndcates that Kenyan farmers work off-farm n order to overcome antcpated rsks. Alasa et al. (2009) also vew partcpaton n off-farm work as a self-nsurance mechansm amed at ncreasng and stablzng the household s ncome. Partcpaton n off-farm work also enables farm households to reduce vulnerablty (Seng 2015) and stablze consumpton (Reardon et al. 1992). The poverty reducton effect of engagng n off-farm work s recognzed by Lanjouw and Lanjouw (2001) and Lanjouw and Sharff (2004). Off-farm work also enhances household consumpton expendture (Akaakohol and Aye 2014). Seng (2015) dentfed the head of household s age and educatonal attanment as well as landholdng as factors determnng partcpaton n non-farm work n rural Camboda. The study showed that older farmers were more lkely to work off-farm, and farmers wth larger landholdngs were less lkely to engage n off-farm work. Household heads wth better educatons also have a hgher propensty for workng off the farm (Akaakohol and Aye 2014; Lanjouw and Sharff 2004). Partcpaton n off-farm work s also expected to be motvated by shfts n producer ncentves and employment opportuntes outsde the farm. Farm households may vew workng off-farm as more desrable than workng on-farm f the ncentves for workng outsde the farm sector outwegh the benefts of workng on the farm. An ncrease n the opportuntes for waged employment outsde the farm sector may therefore motvate farm households to seek employment outsde farmng. On-farm and off-farm work play complementary roles wth potental benefts to the farm sector n many developng countres. For example, t s not uncommon for farm households to nvest ncome from off-farm work n farm operatons and vce versa. Income from workng off the farm can facltate the acquston of farm nputs or the adopton of new technologes, whle ncome from farmng can be nvested n commerce. Thus backward and forward lnkages exst between the two sectors n many developng countres. The exstng lterature alludes to two man potental drect effects of ncome from off-farm work (Babatunde 2015): the lqudty-relaxng effect, whch supposes a potental ncrease n farm expendture/nvestment, and the lost-labour effect, whch supposes a potental allocaton of labour away from the farm. Authors such as de Janvry et al. (2005), Ells and Freeman (2004), Ruben and van den Berg (2001), and Stampn and Davs (2009) pont to the lqudty-relaxng effect of ncome from off-farm work n dfferent studes. These studes hghlght the postve spllover effects of engagng n off-farm work on farm nput among producers. Ells and Freeman (2004) dentfed postve effects of off-farm ncome on land productvty, the hrng n of labour, and the acquston of farm nputs. Furthermore, Osen and Wnters (2009) observed a greater use of hred labour and norganc fertlzers among Ngeran farmers engaged n off-farm work. Smlarly, Anrquez and Dadone (2010) found off-farm work to enhance nvestment n farm nputs among farmers n rural Ghana, whle Maertens (2009) found off-farm employment to ncrease fertlzer use and cultvated areas n a study of Senegalese farmers. Pfeffer et al. (2009), on the other hand, found non-farm 3

6 employment to have a postve mpact on the demand for farm nputs but a negatve mpact on output and use of household labour n producton. An nverse relatonshp between the dversfcaton of household ncome through workng off-farm and farm nvestment s also reported by Ahtuv and Kmh (2002) and Davs et al. (2009). Kumbhakar et al. (1989) found the productvty of dary farms n the US to have a negatve relatonshp wth off-farm ncome. Yee et al. (2004) also reported a negatve relatonshp between off-farm work and productvty n the south-eastern regon of the Unted States. Smlarly, Nasr and Hunde (2014) found a negatve effect of non-farm employment on land productvty n southern Ethopa. However, Woldehanna and Oskam (2001), as well as Gebregzabher et al. (2012a), found off-farm ncome to ncrease agrcultural productvty n northern Ethopa. Ths was because off-farm ncome releved farm households credt constrants, thus enhancng expendture on farm nputs. Smlarly, Wang et al. (2011) found a postve effect of non-farm revenue on agrcultural productvty n rural Chna. As observed by Babatunde (2015), the emprcal lterature provdes mxed evdence about the effects of non-farm employment on agrcultural productvty and effcency. The am of ths paper s therefore to emprcally nvestgate the potental effects of partcpaton n off-farm work on agrcultural productvty. The current paper makes a sgnfcant contrbuton to the lterature on the effect of non-farm work on smallholder farmers agrcultural productvty n vew of the contrastng evdence provded so far by prevous studes on the subject. To the best of my knowledge, the current study s the frst to assess the effect of partcpaton n non-farm actvty on the productvty of Ghanaan smallholder rce farmers. The fndngs from the study wll be helpful n nformng polcymakers on ways to mprove agrcultural productvty and rural lvelhoods. 3 Materals and methods 3.1 Conceptual framework for non-farm employment partcpaton As reported by Jollffe (2004), roughly 74% of Ghanaan farm households are nvolved n some form of non-farm work. Owusu et al. (2011) dentfed agro-processng, commerce, charcoal producton, and seasonal mgraton as some of the non-farm ncome sources n rural Ghana. Other non-farm actvtes nclude gn brewng, basketry, and gatherng frewood for sale. Huffman (1991) provdes a useful framework for analysng farm household labour allocaton decsons. Owusu et al. (2011) used ths framework n ther study on off-farm work and food securty among farm households n northern Ghana, on whch the current study draws. Accordng to the model, households allocate ther tme to specfc actvtes that nclude non-farm work. The household s objectve s to maxmze ts total utlty, subject to certan constrants. The utlty functon that the household seeks to maxmze s expressed as U =U(Q,H), where Q s the household s consumpton of goods and H s lesure. Ths utlty-maxmzng behavour s subject to tme, budget, producton, and non-negatvty constrants (Owusu et al. 2011). The household s T L tme constrant s gven as 1 L2 H, where T represents the total household tme endowment, L1 s tme allotted to farm work, L2 s tme allotted to off-farm work, and H s lesure. The household faces a budget constrant on ts cash ncome gven by: 4

7 PQ p1 y1 w1 L1 w2 L2 R [1] where P represents the prce of goods bought by the household, w1 s returns from on-farm work, w2 s returns from off-farm work, y1 s farm output, p1 s the prce of the household s farm output, and R s non-labour ncome. As ndcated earler, households allot ther tme between farm work, off-farm work, and lesure. The frst-order condton for optmal tme allocaton for the three actvtes s as follows: U / L w U / Q U / L 0 [2] Rearrangng [2], we can derve the returns to labour from on-farm and off-farm work as follows: w ( U / L) / ( U / Q) [3] The labour supply functons for on-farm and off-farm work respectvely are gven by equatons [4] and [5]: L L ( w, w, p, p ; Z) L L ( w, w, p, p, R; Z) [4] [5] where Z represents ndependent varables affectng the household s reservaton and off-farm m r w wages. If we denote the potental market wage by and the reservaton wage 1 w by, then L 1 m r m r w f w L 0, and w f w. Both the reservaton and potental market wages are not observable, but we observe the decson whether or not to partcpate n off-farm work. Such a decson can be analysed usng a probt (or logt) model. 3.2 Emprcal models and data The study employs endogenous swtchng regresson (ESR) to analyse the effect of non-farm work on smallholders productvty. ESR s sutable for stuatons where we are nterested n the effect of beng n one of two dfferent postons or regmes (e.g. partcpaton versus non-partcpaton) on a desred outcome. In ths study, the two decson states or regmes are whether or not farm households engage n non-farm work, whle the outcome of nterest s agrcultural productvty (measured as rce yeld or rce output per area). Snce the decson to partcpate n off-farm work s voluntary, farm households may self-select nto non-farm wage actvty, resultng n a based sample and dffculty n determnng causaton. For example, partcpants n non-farm work may possess systematcally dfferent household attrbutes from non-partcpants as a result of selfselecton. The use of ESR controls for observable as well as unobservable factors (e.g. ablty) that 1 The reservaton wage refers to the mnmum wage at whch an ndvdual wll consent to work. For engagement n off-farm work, t equals the margnal value of a person s tme when t s all apportoned to farm and lesure (Owusu et al. 2011). 5

8 mght account for farmers propensty to engage n off-farm work as well as farm performance. The problem of selecton bas s therefore controlled usng ESR. The frst step n the applcaton of ESR s to estmate the determnants of non-farm actvty partcpaton usng a probt model (see Lokshn and Sajaa 2004), as follows: L Z * [6] L 1 0 f L * 1 otherwse [7] * L where s the latent dependent varable for partcpaton n non-farm actvty, whch s observed through the choce to partcpate n non-farm work. The observed dchotomous choce to work off-farm s gven by L, whch s equal to 1 for partcpants and 0 for non-partcpants. Z s a vector of farm and household characterstcs affectng partcpaton n off-farm work, α s a vector of unknown parameters, and μ s a random error term. The Z varables nclude the head of household s gender, age, and years of formal educaton, the locaton of the farm, farm sze, access to mcrocredt, herd ownershp, degree of specalzaton n rce producton, and dependency rato. The degree of specalzaton n rce producton s measured as the proporton of total landholdng allocated to rce producton. The second step n mplementng the ESR model s to derve separate productvty functons for the two farm groups. The productvty models are specfed as follows: Partcpants: Y1 1X1 1 f L 1 [8] Non-partcpants: Y2 2X2 2 f L 0 [9] Here, Y1 and Y2 are the dependent varables (log of rce yeld) n the contnuous productvty equaton for partcpants and non-partcpants respectvely; X1 and X2 represent vectors of weakly exogenous varables, whle β1 and β2 are vectors of parameters; and 1 and 2 are random dsturbance terms. In order to address the sample selectvty bas, the ESR technque reles on jont normalty of the error terms n the bnary partcpaton and contnuous productvty equatons. The u 1 2 error terms,, and are assumed to have a trvarate normal dstrbuton wth mean zero and covarance matrx, gven as: é ê cov(u,e 1,e 2 ) = W = êê 2 s u s 1u s 1u s 1 2 s 2u s 12 ù ú úú [10] ê ë s 2u s 12 s 2 2 ú û 2 where u represents the varance of the dsturbance term u n the probt partcpaton (selecton) 2 model n equaton [6]; 1 and s 2 2 represent varances of the dsturbance terms n the productvty 6

9 equatons; 1u s covarance of u and 1 ; s 2u s covarance of u and 2. It must be noted that the covarance between smultaneously. 1 and 2 s not defned, snce Y 1 and Y 2 cannot be observed Accordng to Lokshn and Sajaa (2004), an effcent way to estmate ESR models s by full nformaton maxmum lkelhood (FIML) estmaton. Ths procedure smultaneously estmates the probt selecton equaton and the productvty equatons to provde consstent standard errors. As noted by Lokshn and Sajaa (2004), the model s dentfed by constructon through non-lneartes. The FIML estmates of the parameters of the ESR model for ths study were obtaned usng the Stata command movestay (Lokshn and Sajaa 2004). To ensure that the model s properly dentfed, at least one ndependent varable n the frst-stage probt partcpaton regresson s not ncluded n the second-stage productvty regresson (Maddala 1983). A requrement for the selecton nstrument s that t must have a drect effect on the decson to partcpate n non-farm work but not on the outcome of nterest (.e. agrcultural productvty). The study used dependency rato as the selecton nstrument (or dentfcaton restrcton) because t drectly affects the decson to work off-farm but not agrcultural productvty. The study reled on data from a farm household survey coverng 300 smallholders n northern Ghana. Northern Ghana comprses the Northern, Upper East and Upper West Regons, whch together produced 66.7% of total rce output n Ghana n 2010 (Angelucc et al. 2013). Northern Ghana s consdered the country s breadbasket because of the vast amount of agrcultural land and the large volume of food produced n ths area. The regon s characterzed by savannah vegetaton and a short, unmodal ranfall regme. A multstage stratfed random samplng technque was used to select the respondents. In the frst stage, two of the regons comprsng northern Ghana were purposvely selected due to ther hgh nvolvement n rce producton. Next, the three major rrgaton schemes n the study area were selected: the Vea and Tono Irrgaton Schemes n the Upper East Regon, and the Botanga Irrgaton Scheme n the Northern Regon. Fve communtes were then selected at random from the vcnty of each rrgaton scheme. At the communty level, farm households were stratfed nto rrgators and non-rrgators. Equal numbers of rrgators and non-rrgators were then selected from each communty to gve a total sample of 300 rceproducng farm households. Table 1 provdes the defntons and summary statstcs for the sample. The mean productvty level ndcates that yelds are qute low among the respondents. The average farm sze s less than one hectare, and years of formal educaton are low. The average head of household s at an actve age for agrcultural producton. The sample contans fewer female heads of household. Roughly 45% of farmland s allocated to rce producton. The dependency rato s hgh at 85%. Table 1: Defntons and summary statstcs for the sample Varable Descrpton Mean Std dev. Productvty Output per area planted to rce Farm sze Farm sze n hectares Gender Dummy: equals 1 for male household head Age Age n years of the household head Formal educaton Years of formal educaton of household head Specalzaton Share of land allocated to rce Cattle ownershp Dummy: equals 1 for cattle ownershp Regon Dummy: equals 1for Northern Regon Dependency rato Percentage of dependent household members Access to credt Dummy: equals 1 for credt users Source: author s computaton. 7

10 Table 2 presents the descrptve statstcs of the respondents accordng to ther partcpaton status n off-farm work. Table 2: Descrptve statstcs of partcpants and non-partcpants n off-farm work Varable Partcpants (n = 128) Non-partcpants (n = 172) Mean dfference Mean Std dev. Mean Std dev. Productvty (kg/ha) Farm sze (ha) * Gender (1=male) ** Age (years) Formal educaton (years) ** Specalzaton (% rce land) Cattle ownershp (1=yes) * Regon (1=Northern) ** Dependency rato (%) Access to credt (1=access) *Sgnfcant at the 10% level. **Sgnfcant at the 1% level. Source: author s computaton. Partcpants n off-farm work reported hgher rce productvty and had more years of formal educaton as well as hgher dependency ratos. Non-partcpants n off-farm work were older and had larger farms but lower access to credt. Non-partcpants n off-farm work also had a hgher degree of specalzaton n rce producton. Ths shows that farmers wth a hgher degree of specalzaton n rce producton are less lkely to engage n off-farm work, other thngs beng equal. The proporton of non-partcpants ownng cattle was hgher than that of partcpants. In addton, a lower proporton of the partcpants n off-farm work were male farmers. Hence gender s lkely to affect partcpaton n off-farm work. In terms of regonal dstrbuton, 41% of the nonpartcpants came from the Northern Regon, as aganst 23% of the partcpants. Hence partcpaton n off-farm work s expected to be lower among respondents n the Northern Regon. 4 Results 4.1 Determnants of partcpaton n non-farm actvty Table 3 presents the results of the off-farm partcpaton model. The study showed that female farmers were more lkely to partcpate n non-farm ncome actvty n the study area. Smlarly, farm households wth a hgher dependency rato were more lkely to partcpate n off-farm work compared wth households wth a lower dependency rato. In addton, farmers wth more years of formal educaton had hgher partcpaton n non-farm employment compared wth farmers wth less formal educaton. Farmers located n the Upper East Regon were also more lkely to take part n off-farm work compared wth those n the Northern Regon, whle households wth cattle were less lkely to engage n non-farm actvty. The dependency rato varable, whch was ncluded n the adopton model but not n the productvty model n order to dentfy the model (Abdula and Huffman 2014), had a postvely sgnfcant effect on the decson to work off-farm. 8

11 Table 3: Determnants of smallholder farm households partcpaton n off-farm work Varable Coeffcent Standard error P-value Gender 0.500* Age Educaton 0.051** Farm sze Farm sze squared Regonal dummy 0.469* Access to credt Herd ownershp 0.392* Degree of specalzaton Dependency rato 0.125* Constant Observatons 300 *Sgnfcant at 5% level. **Sgnfcant at 1% level. Source: author s computaton. 4.2 Determnants of productvty The FIML estmates of the ESR of the determnants of farm productvty are presented n Table 4. The lkelhood rato test for jont ndependence of the three equatons s statstcally sgnfcant at the 1% level. The covarance term for the adopters ( ) s negatve and statstcally sgnfcant at the 1% level, whle the covarance term for the non-adopters s postve and statstcally sgnfcant at the 1% level. Table 4: Full nformaton maxmum lkelhood estmates of the endogenous swtchng regresson of the determnants of farm productvty Varables Partcpants (n = 128) Non-partcpants (n = 172) Coeffcent SE P-value Coeffcent SE P-value Gender 1.091*** * Age Educaton 0.077*** Farm sze 0.466* *** Farm sze squared *** Regonal dummy 0.996*** * Access to mcrocredt ** Herd ownershp 0.833*** Degree of specalzaton 0.022*** *** Constant 5.648*** *** ln 0.426*** u 1u 2u 0.951*** ln *** u LR test of ndep. eqns 13.05*** Log lkelhood *Sgnfcant at 10% level. **Sgnfcant at 5% level. ***Sgnfcant at 1% level. The dependent varable s the natural log of yeld, defned as output per area. The selecton equaton reported n Table 1 and the outcome equatons above were jontly estmated usng full nformaton maxmum lkelhood. Source: author s computaton. 1 u 9

12 Accordng to the results, the productvty of partcpants n off-farm work s affected by the head of household s gender and years of formal educaton, farm sze, farm locaton, herd ownershp, and the degree of specalzaton n rce producton. The results show that the productvty of partcpants ncreased when the household was headed by a male farmer, was located n the Northern Regon, possessed cattle, and specalzed more n rce producton, but decreased wth the number of years of formal educaton and farm sze. For the non-partcpants n off-farm work, productvty was affected by the gender of the household head, farm sze, locaton of the farm, and degree of specalzaton n rce producton. The productvty of non-partcpants n off-farm work ncreased when the household was headed by a male, had access to credt, was located n the Northern Regon, and specalzed more n rce producton, but decreased wth farm sze. The results reveal that access to credt affects the productvty of non-partcpants n off-farm work, but not the productvty of partcpants. The study also ndcates an nverse relatonshp between farm sze and productvty. Furthermore, the study shows that length of formal educaton s postvely assocated wth partcpaton n off-farm work but negatvely related to productvty. Fnally, herd ownershp enhances the productvty of partcpants, but has no effect on the productvty of non-partcpants n off-farm work. Factors that had the same productvty effects on partcpants and non-partcpants were the gender of the household head, farm sze, locaton of the farm, and degree of specalzaton n rce producton. 4.3 Estmaton of average treatment effect of non-farm work on productvty Table 5 presents the results of the average treatment effect of partcpaton n non-farm actvty on the farm households agrcultural productvty. To ensure robustness n the estmaton, dfferent estmaton procedures were used. The results ndcate that partcpaton n off-farm work ncreases rce productvty by an amount rangng between 175.4kg/ha and 218.5kg/ha. Table 5: Estmates of the average treatment effect of non-farm work on productvty Treatment effect estmaton method Coeffcent Robust SE P-value Nearest-neghbour matchng (NNMATCH) 218.5** Regresson adjustment (RA) 175.4* Inverse-probablty weghtng (IPW) 190.5** Inverse-probablty-weghted regresson adjustment (IPWRA) 175.4* * Sgnfcant at 10% level. ** Sgnfcant at 5% level. Source: author s computaton. 5 Dscusson 5.1 Partcpaton n non-farm work Women n many rural areas play crtcal roles n the home, augmentng the household ncome through petty tradng and other ncome actvtes. Due to ther entrepreneural abltes, women are more lkely to engage n off-farm actvtes, n consonance wth the fndngs of ths study. Ths result accords wth Man and Sadya (2009), who found male paddy farmers n Malaysa to be less lkely to engage n off-farm work. However, the result s at varance wth Matshe and Young (2004), 10

13 who found lower female partcpaton n the non-farm labour market n Zmbabwe. Corral and Reardon (2001) also reported hgher male partcpaton n non-farm waged employment n Ncaragua, although non-farm self-employment was found to be lower for males. Households wth hgher dependency ratos tend to have more famly oblgatons and hgher expendture, other thngs beng equal. Consequently, households wth more dependents are more lkely to partcpate n non-farm work to supplement ther ncome from farmng. Ths fndng of the study s consstent wth Man and Sadya (2009), who found that Malaysan paddy farmers lkelhood of engagng n off-farm work ncreased as the number of dependents ncreased. Rahman (2013) also found the dependency rato to postvely nfluence the decson to work n the servce sector n Bangladesh. On the other hand, Owusu et al. (2011) found that females partcpaton n non-farm employment n northern Ghana was nfluenced negatvely by the dependency rato. However, the authors found a non-sgnfcant postve effect of the dependency rato on males partcpaton n non-farm employment. Educaton enhances the value of human captal and opens up more opportuntes for rural people to engage n ncome-earnng actvtes. Educated farmers are therefore expected to have a hgher partcpaton n non-farm work, snce they are more lkely to fnd jobs outsde the farm sector compared wth uneducated farmers. Ths result accords wth Seng (2015), who found that the head of household s years of formal educaton enhanced partcpaton n non-farm work n rural Camboda. Abdula and CroleRees (2001) also found hgher partcpaton n off-farm work among households wth educated heads compared wth households wth llterate heads. Ths result of ths study also accords wth Ahtuv and Kmh (2002), Corral and Reardon (2001), Ferrera and Lanjouw (2001), Matshe and Young (2004), and Owusu et al. (2011). As ndcated by Ferrera and Lanjouw (2001), years of formal educaton make household members more employable. The lower partcpaton of herd owners n non-farm work mght ndcate a lower lkelhood of wealther households to engage n off-farm work, snce herd ownershp s an ndcaton of wealth n many rural communtes n Ghana. Ths result contrasts wth Owusu et al. (2011), who found cattle ownershp to have a postve and sgnfcant effect on males partcpaton n non-farm employment n northern Ghana but a non-sgnfcant postve effect on females partcpaton n non-farm employment. Beyene (2008), on the other hand, showed that herd sze postvely nfluenced women s partcpaton n off-farm work n Ethopa but dd not nfluence males partcpaton. The hgher partcpaton of farmers n the Upper East Regon n non-farm work, compared wth ther counterparts n the Northern Regon, suggests that regonal varatons account for farmers partcpaton n off-farm work. The Upper East Regon s further from the natonal captal and has a hgher populaton densty than the Northern Regon. Hence land for arable crop producton may be scarcer n the Upper East Regon compared wth the Northern Regon. These factors may account for the hgher ncome dversfcaton among farm households n the Upper East Regon. 5.2 Determnants of productvty The lkelhood rato test for jont ndependence of the three equatons ndcates that the equatons are dependent. The covarance terms ( 1 u and 2 u ) are non-zero, ndcatng that the model shows endogenous swtchng (Maddala 1986). Ths therefore justfes the use of the ESR model. The sgnfcance of the correlaton between ɛ1 and μ ( 1 u ) ndcates the presence of self-selecton n partcpaton n non-farm actvty. The result mples that partcpaton n off-farm work may not produce the same effect on non-partcpants, f they decde to partcpate (Lokshn and Sajaa 11

14 1 u 2004). Snce the coeffcent s negatve and sgnfcantly dfferent from zero, the model ndcates that partcpants n off-farm work have hgher productvty than a random household 1 u from the sample. The negatve sgn of can also be nterpreted to mean a postve selecton bas so that partcpants enjoy above-average productvty levels (yelds) once they take part n off-farm work. 2 On the other hand, the postve and sgnfcant value of r e2 u (correlaton between ɛ2 and μ) ndcates that non-partcpants enjoy above-average productvty levels when they do not take part n off-farm work. Thus farm households decsons to partcpate n off-farm work are based on the comparatve advantage (Fugle and Bosh 1995; Maddala 1983; Rao and Qam 2011). Smlar results have been obtaned by Abdula and Bnder (2006), Abdula and Huffman (2014), and Barrett et al. (2004). The fndngs pont to an nverse relatonshp between farm sze and productvty. Ths supports the almost stylzed fact of an nverse relatonshp between farm sze and productvty. Even though some researchers (e.g. Corna 1985; Yee et al. 2004) have found a postve relatonshp between farm sze and productvty, authors such as Larson et al. (2012), Sen (1962, 1966), and Thapa (2007) have found productvty to decrease wth farm sze. Snce smallholders are manly resource-poor and largely rely on ther own resources for producton, t s antcpated that they would be severely constraned n managng very large farms. It s expected that farmers manageral abltes are related to ther educaton and access to extenson servces and nformaton, whch are lackng n most rural communtes. Hence most smallholders are lkely to be less productve on larger farms, other thngs beng equal. Productvty decreased at an ncreasng rate for the non-partcpants, whle for the partcpants, the quadratc term of the farm sze varable had a non-sgnfcant postve effect on productvty. Ths result of the study accords wth Abdula and Huffman (2014), who reported a negatvely sgnfcant nfluence of farm sze on rce yeld among non-adopters of sol and water conservaton technology n Ghana but not among adopters. An mportant mplcaton of the current study s that educaton tends to draw labour away from the farm sector, thereby reducng productvty at the farm level. Ths s shown by the postve effect of educaton on partcpaton n off-farm actvty and the opposte effect on the productvty of partcpants. Hence there s a labour-loss effect assocated wth addtonal years of formal educaton, due to ts postve effect on partcpaton n off-farm actvty. On the other hand, educaton does not nfluence the productvty of non-partcpants. The educaton varable has a non-sgnfcant postve effect on the productvty of farm households that do not engage n offfarm work. Ths result of the study s at varance wth Abdula and Huffman (2014), who reported a postvely sgnfcant nfluence of educaton on rce yeld among both adopters and non-adopters of sol and water conservaton technology n four dstrcts n the Northern Regon of Ghana. The results of the study also revealed an mportant role of wealth status n agrcultural productvty among smallholders n the study area. In the current study, herd ownershp was used as a proxy for wealth status, because t s usually wealther households that own cattle n many rural communtes. As expected, partcpaton n off-farm work was lower for herd owners, suggestng that wealther households partcpated less n off-farm work. The postve and sgnfcant effect of cattle ownershp on the productvty of partcpants ndcates the mportant role of cattle as draught anmals n smallholder farmng. Cattle ownershp had no effect on the productvty of non- 2 The model satsfes the necessary condtons for consstency snce 2 u > 1 u, suggestng that partcpants acheve hgher yelds than would be the case f they dd not partcpate (Lokshn and Sajaa 2004; Trost 1981). 12

15 partcpants n off-farm work, suggestng that wealther non-partcpants may be usng more mproved technology, such as machnery, n producton. Another mportant mplcaton of ths study s that the mpact of credt on farm productvty seems to depend on the household s ncome dversfcaton. Access to credt s consdered an mportant determnant of productvty growth (Reyes et al. 2012). Anang et al. (2015) also noted that credt allows optmal nput use, resultng n a postve mpact on agrcultural productvty. Accordngly, there are several attempts by both governmental and non-governmental organzatons to ncrease smallholders access to agrcultural credt. The current study ndcates that access to credt enhances the productvty of non-partcpants n off-farm work but has a non-sgnfcant negatve effect on the productvty of partcpants. The result could mean that non-partcpants n off-farm work utlze credt more effcently n farmng. The result s consstent wth other studes, such as Abdula and Huffman (2014), who found a postvely sgnfcant nfluence of credt on rce yeld among both adopters and non-adopters of sol and water conservaton technology n the Northern Regon of Ghana. Furthermore, Reyes et al. (2012) dentfed access to credt as one of the factors nfluencng farm productvty and rural development n Chle. In lne wth a pror expectatons, the degree of specalzaton n rce producton had a postve effect on farmers levels of productvty. Ths result accords wth the economc theory that specalzaton leads to hgher productvty. The study also supports the fndngs of other researchers that show male farmers to be more productve n producton due to ther greater access to productve resources. The results further ndcate that productvty s nfluenced by the geographcal locaton of the farm, wth producers n the Northern Regon havng hgher productvty than ther counterparts n the Upper East Regon. As noted by Gebregzabher et al. (2012b), bophyscal and envronmental factors may nfluence farmers effcency and productvty. Fnally, the age of the household head had no sgnfcant effect on the productvty of farm households. Ths result accords wth Abdula and Huffman s (2014) study nvolvng rce-producng households n Northern Ghana. 5.3 Average treatment effect of non-farm work on productvty Ths study hghlghts ncome dversfcaton as an mportant lvelhood strategy among farm households, wth a postve effect on agrcultural productvty. Results from the ESR dagnostc coeffcents as well as the treatment effects model reveal a postve assocaton between off-farm work and agrcultural productvty. Hence ths study does not seem to ndcate a lost-labour effect from off-farm work. On the other hand, the study seems to support the noton of a lqudtyrelaxng effect of non-farm work on agrcultural productvty. In other words, partcpaton n offfarm work mght enable farmers to earn wages n order to fnance agrcultural producton, e.g. the purchase of seeds and hrng of labour. The results of the treatment effects model are robust and justfable from the dfferent methodologes employed n the estmaton. The fndngs accord wth Wang et al. (2011), who found a postve effect of non-farm revenue on agrcultural land productvty n rural Chna. Woldehanna and Oskam (2001) also found off-farm ncome to ncrease agrcultural productvty n northern Ethopa. 6 Concluson and polcy recommendatons The study examned the effect of partcpaton n non-farm actvty on agrcultural productvty among smallholder rce farmers n northern Ghana. Self-selecton nto non-farm work was addressed usng an endogenous swtchng regresson framework. Results of a probt analyss found 13

16 the head of household s gender and years of formal educaton, household herd ownershp, the dependency rato, and the geographcal locaton of the farm household to be sgnfcant factors nfluencng the decson to partcpate n off-farm work. Accordng to the ESR results, farm productvty for partcpants and non-partcpants n off-farm work alke was nfluenced by the gender of the head of household, farm sze, farm locaton, and degree of specalzaton n rce producton. These factors should therefore be taken nto account when strategzng how to enhance the productvty of smallholder farmers n northern Ghana. Herd ownershp and the number of years of formal educaton affected the productvty of partcpants n off-farm work, whle access to credt was sgnfcant n ts effect on the productvty of non-partcpants. Results from a treatment effects model showed that partcpaton n off-farm work had a postve effect on farmers levels of productvty. The study therefore concludes that ncome dversfcaton s an mportant lvelhood strategy among smallholders, and that earnng an ncome from workng offfarm enables smallholders to mprove ther yelds. The followng polcy recommendatons emerge from the study to mprove rce productvty n Ghana. Frst, the results ndcate that as farmers specalze n the producton of rce they become more productve. Thus productvty can be enhanced by encouragng farmers to specalze more n the cultvaton of rce. Second, rce productvty can be enhanced by makng agrcultural credt avalable to farmers. Ths s because credt s sgnfcant n ts effect on productvty. Even though the effect of credt on the partcpants productvty was not sgnfcant, ts effect on nonpartcpants, as well as results from other studes, ndcates that expandng access to credt s lkely to brng about mprovements n rce productvty. References Abdula, A., and C. Bnder (2006). Slash-and-Burn Cultvaton Practce and Agrcultural Input Demand and Output Supply. Envronment and Development Economcs, 11: Abdula, A., and A. CroleRees (2001). Determnants of Income Dversfcaton Amongst Rural Households n Southern Mal. Food Polcy, 26: Abdula, A., and W. Huffman (2014). The Adopton and Impact of Sol and Water Conservaton Technology: An Endogenous Swtchng Regresson Applcaton. Land Economcs, 90(1): Ahtuv, A., and A. Kmh (2002). Off-Farm Work and Captal Accumulaton Decsons of Farmers over the Lfe-Cycle: The Role of Heterogenety and State Dependence. Journal of Development Economcs, 68: Akaakohol, M.A., and G.C. Aye (2014). Dversfcaton and Farm Household Welfare n Makurd, Benue State, Ngera. Development Studes Research, 1(1): Alasa, A., A. Weersnk, R. Bollman, and J. Cranfeld (2009). Off-Farm Labour Decson of Canadan Farm Operators: Urbanzaton Effects and Rural Labour Market Lnkages. Journal of Rural Studes, 25(1): Anang, B.T., T. Splänen, S. Bäckman, and J. Kola (2015). Factors Influencng Smallholder Farmers Access to Agrcultural Mcrocredt n Northern Ghana. Afrcan Journal of Agrcultural Research, 10(24): Angelucc, F., A. Asante-Poku, and P. Anaadumba (2013). Analyss of Incentves and Dsncentves for Rce n Ghana. Techncal Notes Seres. Rome: FAO, MAFAP. 14

17 Anrquez, G., and S. Dadone (2010). Lnkages Between the Farm and Nonfarm Sectors at the Household Level n Rural Ghana: A Consstent Stochastc Dstance Functon Approach. Agrcultural Economcs, 41: Babatunde, R.O. (2015). On-Farm and Off-Farm Works: Complement or Substtute? Evdence from Ngera. Maastrcht School of Management Workng Paper 2015/02. Ilorn: Unversty of Ilorn, Department of Agrcultural Economcs and Farm Management. Bag, F.S. (1984). Stochastc Fronter Producton Functon and Farm-Level Techncal Effcency of Full-Tme and Part-Tme Farms n West Tennessee. North Central Journal of Agrcultural Economcs, 6: Barrett, C.B., C.M. Moser, O.V. McHugh, and J. Barson (2004). Better Technology, Better Plots or Better Farmers? Identfyng Changes n Productvty and Rsk Among Malagasy Rce Farmers. Amercan Journal of Agrcultural Economcs, 86(4): Barrett, C.B., T. Reardon, and P. Webb (2001). Nonfarm Income Dversfcaton and Household Lvelhood Strateges n Rural Afrca: Concepts, Dynamcs and Polcy Implcatons. Food Polcy, 26(4): Beyene, A.D. (2008). Determnants of Off-Farm Partcpaton Decson of Farm Households n Ethopa. Agrekon, 47(1): Chang, H.-H., and F.-I. Wen (2011). Off-Farm Work, Techncal Effcency, and Rce Producton Rsk n Tawan. Agrcultural Economcs, 42: Corna, G.A. (1985). Farm Sze, Land Yelds and the Agrcultural Producton Functon: An Analyss for Ffteen Developng Countres. World Development, 13: Corral, L., and T. Reardon (2001). Rural Nonfarm Incomes n Ncaragua. World Development, 29(3): Cunguara, B., A. Langyntuo, and I. Darnhofer (2011). The Role of Nonfarm Income n Copng wth the Effects of Drought n Southern Mozambque. Agrcultural Economcs, 42(6): Davs, B., P. Wnters, T. Reardon, and K. Stamouls (2009). Rural Nonfarm Employment and Farmng: Household-Level Lnkages. Agrcultural Economcs, 40: de Janvry, A., E. Sadoulet, and N. Zhu (2005). The Role of Non-Farm Incomes n Reducng Rural Poverty and Inequalty n Chna. CUDARE Workng Paper Berkeley: Unversty of Calforna, Department of Agrcultural and Resource Economcs. Ells, F., and H.A. Freeman (2004). Rural Lvelhoods and Poverty Reducton Strateges n Four Afrcan Countres. Journal of Development Studes, 40: Escobal, J. (2001). The Determnants of Nonfarm Income Dversfcaton n Rural Peru. World Development, 29: Fernandez-Cornejo, J., A.K. Mshra, R.F. Nehrung, C. Hencrcks, M. Southern, and A. Gregory (2007). Off-Farm Income, Technology Adopton and Farm Economc Performance. Unted States Department of Agrculture Economc Research Report, 36: Ferrera, F.H.G., and P. Lanjouw (2001). Rural Nonfarm Actvtes and Poverty n the Brazlan Northeast. World Development, 23(3): Fugle, K.O., and D.J. Bosch. (1995). Economc and Envronmental Implcatons of Sol Ntrogen Testng: A Swtchng-Regresson Analyss. Amercan Journal of Agrcultural Economcs, 77(4): Gebregzabher, K., E. Mathjs, M. Maertens, J. Deckers, H. Bauer, and K. Gebrehwot (2012a). Is Non-Farm Income Relaxng Farm Investment Lqudty Constrants for Margnal Farms? An 15