Impact of land fragmentation and resource ownership on productivity and efficiency: The case of rice producers in Bangladesh

Size: px
Start display at page:

Download "Impact of land fragmentation and resource ownership on productivity and efficiency: The case of rice producers in Bangladesh"

Transcription

1 Unversty of Plymouth PEARL Faculty of Scence and Engneerng School of Geography, Earth and Envronmental Scences Impact of land fragmentaton and resource ownershp on productvty and effcency: The case of rce producers n Bangladesh Rahman, S /j.landusepol Land Use Polcy All content n PEARL s protected by copyrght law. Author manuscrpts are made avalable n accordance wth publsher polces. Please cte only the publshed verson usng the detals provded on the tem record or document. In the absence of an open lcence (e.g. Creatve Commons), permssons for further reuse of content should be sought from the publsher or author.

2 IMPACT OF LAND FRAGMENTATION AND RESOURCE OWNERSHIP ON PRODUCTIVITY AND EFFICIENCY: THE CASE OF RICE PRODUCERS IN BANGLADESH SANZIDUR RAHMAN AND MIZANUR RAHMAN Abstract The paper analyzes the mpact of land fragmentaton and ownershp of resources on productvty and techncal effcency n rce producton n Bangladesh usng farm level survey data. Results reveal that land fragmentaton has a sgnfcant detrmental effect on productvty and effcency as expected. The elastcty estmates of land fragmentaton reveal that a one percent ncrease n land fragmentaton reduces rce output by 0.05 percent and effcency by 0.03 percent. On the other hand, ownershp of key resources (land, famly labour, and draft anmals) sgnfcantly ncreases effcency. The mean elastcty estmates reveal that a one percent ncrease n famly labour and owned draft anmal mprove techncal effcency by 0.04 and 0.03 percent, respectvely. Also, a one percent ncrease n the adopton of modern technology mproves effcency by 0.04 percent. The mean techncal effcency n rce producton s estmated at 0.91 ndcatng lttle scope to mprove rce producton per se usng exstng varetes. Polcy mplcatons nclude addressng structural causes of land fragmentaton (e.g., law of nhertance and poltcal economy of agraran structure), buldng of physcal captal (e.g., land and lvestock resources), mprovements n extenson servces and adopton of modern rce technology. JEL Classfcaton: O33, Q18, and C21. Keywords: Stochastc producton fronter, effcency elastcty, land fragmentaton, resource ownershp, rce, Bangladesh 1

3 1. Introducton Land s the major source of wealth and lvelhood n rural Bangladesh, as n other South Asan countres, although the land person rato s one of the lowest n the world estmated at 0.12 ha (FAO, 2001). The agrcultural sector n Bangladesh, domnated by rce producton, s already operatng at ts land fronter and has very lttle or no scope to ncrease the supply of land to meet the growng demand for food for ts ncreasng populaton (Rahman, 2003). The expanson n crop area, whch was the major source of producton growth untl the 1980s, has been exhausted and the area under rce started to declne thereafter (Husan et al., 2001). The converson potental from local to modern varetes of rce has stagnated at 69 percent of total rce area (BBS, 2000), mplyng that the prncpal soluton to ncrease food producton les n rasng the productvty of land gven the exstng varetal mx. Ths s further complcated by the shrnkng avalablty of land per farm holdng, as the long establshed debate of nverse sze- productvty relatonshp has now been weakened (Ram et al., 1999) n favour of postve sze-productvty relatonshps (Wattanutcharya and Jtsanguan, 1992). In Bangladesh, the szeproductvty relatonshp vares across regons dependng on the level of technologcal development and envronmental opportuntes. The relatonshp s postve n technologcally advanced regons, whereas the classc nverse relatonshp stll exsts n backward areas (Toufque, 2001). In addton to shrnkng avalablty of land for farmng, land fragmentaton 1 s on the rse n Bangladesh. Table 1 presents farm dynamcs and the extent of land fragmentaton based on three censuses of agrculture and lvestock over the past three decades. The number of farm holdngs ntally ncreased rapdly but then slowed down and there has been a major shft n the composton of farm sze groups. Unlke the experence n East Asan countres, e.g., Japan and 1 Land fragmentaton here refers to farmng of a number of non-contguous owned or leased plots (or parcels) of land as a sngle producton unt (McPherson, 1982). 2

4 Korea, where farm szes are gettng larger as the number of operatonal holdngs are gong down (Nroula and Thapa, 2005), Bangladesh s experencng rapd declne n farm szes coupled wth an ncrease n the number of operatonal holdngs. The average farm sze shrank to a level (0.68 ha) at whch t s unlkely to sustan lvelhoods 2. The number of small farms ncreased dramatcally at the expense of a reducton n the number of large and medum szed farms. The stuaton deterorates further when one consders the fragmentaton of total land holdngs nto parcels. Overall, the number of fragments per holdng as well as average sze of fragments declned n Bangladesh. Nevertheless, the average sze of fragments ncreased for the large farm sze categores, mplyng that some consoldaton s takng place for ths sze group, perhaps through purchase or smple appropraton from margnal or landless farmers through an explotatve tenural system. Table 1 further reveals that not only the avalablty of land n Bangladesh s shrnkng, but also another key farm resource endowment, the ownershp of a draft anmal, whch s used exclusvely to substtute for farm power requrements n land preparaton and the transportaton of agrcultural products, s also declnng rapdly. Furthermore, although landlessness s on the rse, the number of agrcultural labour households s decreasng, whch potentally has mplcatons for the sze and operaton of the hred labour market. [Insert Table 1 here] A host of supply sde and demand sde arguments exst to explan the persstence of land fragmentaton. The supply sde arguments treat land fragmentaton as an exogenous mposton on farmers as a result of nhertance laws, populaton pressure and scarcty of land (McPherson, 1982, Bentley, 1987). The demand sde explanatons vew land fragmentaton as a postve choce by farmers n order to reduce rsk from natural dsasters (such as floods, droughts), 2 Small farmers wth less than 1 ha of landholdngs cannot fulfl ther subsstence requrements through agrculture (Nroula and Thapa, 2005). 3

5 promote crop dversfcaton, as well as to ease allocaton of labour over croppng seasons (Fenoaltea, 1976; Ilbery, 1984, Tan, 2005). Studes of the constrants mposed by land fragmentaton on productvty and effcency n agrculture are mxed and nconclusve. For example, Blake and Sadeque (2000) hghlght that land fragmentaton s becomng a crtcal constrant n ncreasng productvty n Nepal, Inda and other nearby regons. In contrast, farmers n the hghly land fragmented regons of Malaysa and Phlppnes do not consder t as a problem n paddy farmng (Hoo, 1978; Wong and Geronmo, 1983; cted n Nroula and Thapa, 2005). In case of Chna, Wu et al., (2005) conclude that land fragmentaton does not have any sgnfcant mpact on productvty, whereas Wan and Cheng (2001) conclude that land fragmentaton reduces productvty. Smlar contrastng arguments exst on the effects of land fragmentaton on effcency. For example, Schultz (1953) vews land fragmentaton as the msallocaton of the exstng stock of agrcultural land, mplyng t as a source of neffcency. Dovrng and Dovrng (1960) dentfy dstance between parcels as the man source of neffcency created by land fragmentaton. Recent studes, Sherlund et al., (2002) and Tan (2005) conclude that the ncrease n the number of plots has a postve relaton wth techncal effcency n rce producton n Cote d Ivore and Chna, whereas Parkh and Shah (1994) and Wadud and Whte (2000) report that land fragmentaton reduces effcency n rce producton n Pakstan and Bangladesh, respectvely. Apart from land fragmentaton, ownershp of key producton resources such as, land, draft anmal power and famly labour can also potentally mpact effcency. The man argument n favour of resource ownershp s the tmelness of operaton as well as control over the qualty of the resource. For example, owner operators are lkely to be more effcent than tenants or sharecroppers as they could hold on to the best qualty of land whle rentng or leasng out relatvely poorer qualty land. And snce control over land s hgh, the owner operator s able to conduct all requred farmng operatons n a tmely manner. There s a belef that farmers n 4

6 developng countres overuse famly labour and hence lkely to be neffcent n producton. However, a large pool of famly labour may enable farmers to use labour on tme partcularly durng the peak season when hred labour becomes relatvely scarce (Dhungana et al., 2004). Smlar arguments can be put forward for draft anmal power ownershp. Frst, the rental market for draft power s relatvely smaller than the hred labour market and can result n acute shortages durng the peak season for ploughng, partcularly n Bangladesh. Second, ownershp of draft power enables the farmer to plough and prepare the land at the rght tme. However, no sngle study has examned the nfluence of all these key resources jontly on effcency. Only a few studes used any one of these resources, e.g., ether tenural status or famly labour but none used ownershp of draft anmal power. Even then, the results are mxed. For example, Helfand and Levne (2004) concluded that tenants are more effcent than the sharecroppers and owners n Brazl, whereas Rahman (2003) concluded that owner operators are more effcent than tenants n Bangladesh. Tzouvelekas et al., (2001) note that famly operated farms are relatvely more neffcent than farms usng hred labour n olve farmng n Greece, whereas Dhungana et al., (2004) conclude that use of famly labour s postvely related wth effcency n rce farmng n Nepal. On the other hand, Battese et al., (1996) conclude that both hred and famly labour are equally effcent n wheat producton n Inda. Gven ths backdrop, the present study sets out to analyze explctly the mpact of land fragmentaton on productvty as well as on techncal effcency n rce farmng, usng farm level survey data n Bangladesh. In addton, the study also analyzes the jont mpact of the ownershp of three key resources (land, famly labour and draft anmal power) on techncal effcency. The paper proceeds as follows: secton 2 descrbes the analytcal framework, the study area and the data; secton 3 presents the results and dscusses polcy mplcatons; and secton 4 draws some conclusons. 2. Research Methodology 5

7 2.1 Analytcal framework Applcaton of the stochastc producton fronter framework s approprate to analyze the mpact of land fragmentaton and resource ownershp on productvty and effcency. Three basc hypotheses are tested: (a) whether land fragmentaton affects productvty; (b) whether land fragmentaton affects producton effcency; and (c) whether resource ownershp affects producton effcency. The mpact of land fragmentaton on productvty s captured by specfyng number of plots farmed 3 as an ndependent varable n the stochastc producton fronter functon. The mpact of land fragmentaton on effcency s examned by placng the same n the neffcency effects model n addton to resource ownershp varables and ndcators representng other farm characterstcs to explan the underlyng causes of devaton from the fronter. In the stochastc producton fronter framework, the output (rce producton) s treated as a stochastc producton process and s defned as (Agner et al., 1977): Q = f X ; A).exp( ε ) ( (1) where X s the (NxJ) matrx of the nputs, Q s the (Nx1) vector of output, f(.) s the best practce producton fronter, A s the technology parameter vector, and the subscrpts ndvdual farm households, respectvely. The error term ε s composed of two components: ε = v u ( 1a) 3 The potental ndcators to measure land fragmentaton are: the number of plots, average plot sze, average dstance of plots to dwellngs, and the Smpson ndex (Tan, 2005). Wadud and Whte (2000) used average plot sze as the ndcator. We have used number of plots as the ndcator of land fragmentaton for two reasons: (a) to avod collnearty between average plot sze and total land under cultvaton n the specfed producton functon; and (b) to provde an explct measure of the mpact of an ncrease n the number of plots on productvty as well as on effcency. 6

8 where the component v s are assumed to be dentcally and ndependently dstrbuted 2 N(0, σ ) two sded random errors, ndependent of the u s, representng random shocks, such v as exogenous factors, measurement errors, omtted explanatory varables, and statstcal nose. The u s are non-negatve random varables, assocated wth neffcency n producton, whch are assumed to be ndependently dstrbuted as truncatons at 0 of the normal dstrbuton wth mean, 2 2 µ = δ 0 + δ dwd and varance σ u ( N( µ, σ u ) ), where W d s the d dth explanatory varable assocated wth neffcences of farm and δ 0 and δ d are the unknown parameters. The producton effcency of the farm s defned as: EFF = E[exp( u ) ε ] (2) where E s the expectaton operator. Ths s acheved by obtanng the expressons for the condtonal expectaton u upon the observed value of ε. The method of maxmum lkelhood s used to estmate the unknown parameters, wth the stochastc producton fronter and the neffcency effects functons estmated smultaneously. The lkelhood functon s expressed n terms of the varance parameters, σ = σ v + σ u and γ = σ 2 /σ 2 u (Battese and Coell 1995). 2.2 The study area and sample of farmers Prmary data for the study pertans to a farm survey of rce producers conducted durng early 2000 n the Barsal dstrct located n the southern part of Bangladesh, whch n turn s located 162 km away from the captal cty, Dhaka. The dstrct s composed of 10 thanas (subdstrcts), 86 unons and 1,069 vllages. Samples were collected from four vllages n two subdstrcts, Hzla Thana and Mulad Thana, respectvely. A total of 298 farm households were selected followng a multstage stratfed random samplng procedure. Factors consdered n stratfcaton nclude degree of uncertanty (.e., rsk of floodng) at the subdstrct level, type and dstance from the local market, transport and road facltes of the 7

9 vllages at the unon level, and farm holdng sze and tenural classes of the farmers at the vllage level, respectvely (for detals, see Rahman, 2004). Detaled nput and output data were collected for the Aman (monsoon) season rce of the crop year 1999 because rce produced n ths season provdes the bulk of the foodgran supples n Bangladesh, and farmers largely produce local varetes of rce whch s dependent on monsoon ran. In Barsal dstrct, 69.5 percent of the total cultvated area s devoted to Aman rce producton (BBS, 2000). 2.3 The emprcal model The producton structure of rce farmers n Bangladesh s specfed usng a sngle output multple nput stochastc producton fronter. The general form of the extended flexble translog stochastc producton fronter for the th farm s defned as: lny and = α j= 1 + φ ln L p α ln X p j j p= 1 j= j= 1 k= 1 ϕ ln L pj β ln X pj jk jk + v u ln X jk + τ D (3) m m u = δ 8 * 0 + δ d Z d + ψ q Lq + ζ (3a) d= 1 where the dependent varable Y s the aggregate of rce output produced (kg per farm) n the Aman season; X s are the nputs, L s the varable representng land fragmentaton; and D s the dummy varable used to account for the zero values of nput use 4 ; v s the two sded random error and u s the one sded half normal error n eq. (3); and ln s the natural logarthm; Zs n eq. (3a) are the varables representng resource ownershp as well as farm 4 Inputs contanng zero values for some observatons are specfed as ln {max (X j, 1 D j )} followng Battese and Coell, (1995). However, the nteracton effects of these dummy varables n a translog framework are avoded n order to preserve degrees of freedom. 8

10 specfc characterstcs to explan neffcency; L s the land fragmentaton varable, ζ s the truncated random varable; α k, β 0, β j, τ m, ϕ p, ψ q, δ 0, and δ d are the parameters to be estmated. A total of sx producton nputs (X) are used n the stochastc producton fronter model and eght varables representng resource ownershp and other soco-economc characterstcs 5 of the farm (Z) are ncluded n the neffcency effects model as predctors of techncal neffcency. Table 2 presents the defntons, unts of measurement, and summary statstcs for all the varables. [Insert Table 2 here] 3. Results From the nformaton provded n Table 2, we see that the average farm sze s small (0.78 ha) and the average level of land fragmentaton s 4.4 wth a range from a sngle plot farm to a maxmum of a 21 plot farm 6. Only 27 percent of the farmers are owner operators, the number of workng members n the famly s 1.9 persons and the number of workng anmals s only 1.2. Seventy three percent of the farmers have some educaton 7 and 34 percent had extenson contacts. Level of modern technology adopton s low, because only 33 percent of the farmers produced modern varetes of rce n addton to tradtonal varetes. The Maxmum Lkelhood Estmaton (MLE) procedure s used to estmate the parameters of the stochastc producton fronter and neffcency effect models jontly n a 5 Choce of these varables (e.g., experence, educaton, non-agrcultural work, extenson contact, etc.) s based on exstng lterature (e.g., Coell et al., 2002; Rahman, 2003; Wadud and Whte, 2000; Tzouvelekas, et al., 2001; Sherlund et al., 2002). 6 The fgures are slghtly lower than the natonal averages presented n Table 1 but exactly match the data for the Barsal dstrct as a whole (not shown), thereby, provdng confdence n the representatveness of the selected households for ths study. 7 Barsal dstrct as a whole s regarded as a relatvely hghly lterate part of the country based on lteracy rate 9

11 sngle stage 8 usng STATA Verson 8 (Stata Corp, 2003). Two versons of the model were estmated. In Model 1, the land fragmentaton varable s ncluded n the producton functon, and n Model 2 the same s ncluded n the neffcency effects functon. 3.1 Productvty effects of land fragmentaton The second column of Table 3 presents MLE estmates of the extended translog stochastc producton fronter whch ncorporates the land fragmentaton varable ncludng full nteractons wth producton nputs n order to account for ts total effect on productvty. A test of hypothess on the choce of functonal form (Cobb-Douglas vs. translog) confrms that the choce of translog producton functon s a better representaton of the producton structure for both models (Table 4). All basc resource nputs except seed sgnfcantly nfluence rce producton. The pestcde varable recorded some zero observatons, and was therefore, corrected wth dummy varables as mentoned n footnote 4. Contrary to expectaton, labour seems to be the domnant factor followed by draft power servces and fertlzers. Output elastcty of labour s estmated at 0.31 (0.33 n Model 2) ndcatng that a one percent ncrease n labour use wll ncrease output by 0.31 percent 9 (Tables 3 or 5). Land fragmentaton sgnfcantly reduces rce nformaton. 8 The sngle-stage approach s consdered superor to the conventonally used two-stage approach wheren the frst stage nvolves estmaton of the stochastc producton fronter and the predcton of neffcency effects under the assumpton that these neffcency effects are dentcally dstrbuted wth one-sded error terms. The second stage nvolves the specfcaton of a regresson model for predcted neffcency effects, whch contradcts the assumpton of an dentcally dstrbuted one-sded error term n the stochastc fronter (Battese and Coell, 1995). 9 All the resource nput varables (ncludng land fragmentaton) were mean-dfferenced pror to estmaton. Therefore, the coeffcents on the frst order term can be read drectly as elastctes. Nevertheless, the fgures are reproduced n Table 5 for ease of exposton. 10

12 output as expected 10. The test of null hypothess that the effect of land fragmentaton on productvty (ncludng ts nput nteractons) s jontly zero s strongly rejected at 1 percent level of sgnfcance (Table 4). The output elastcty of land fragmentaton wth respect to productvty s estmated at -0.05, mplyng that for a one percent ncrease n the number of plots, output s reduced by 0.05 percent. Increasng returns to scale prevals n rce producton n Bangladesh (Tables 3 and 5). The null hypothess of constant returns to scale s rejected n favour of ncreasng returns to scale for both models (Table 4). [Insert Tables 3, 4 and 5 here] 3.2 Effcency effects of land fragmentaton and ownershp of resources Gven the robust detrmental effects of land fragmentaton on rce productvty, we next nvestgate ts nfluence on techncal effcency. We also nvestgate the jont nfluence of the ownershp of key resources (land, famly labour and draft anmal power) on techncal effcency. Pror to the dscusson of these effects, we brefly hghlght the farm specfc effcency scores presented n Table 6. The mean effcency level s estmated at 91 percent (92 percent n Model 2) ndcatng that rce producton can be ncreased by 9 [(100-91)/91] percent by mprovng techncal effcency alone wth no addtonal use of resources. The mnmum effcency level s 62 percent (63 percent n Model 2) whle the maxmum s 99 percent. The estmates are slghtly hgher than those reported by Rahman (2003), Coell et al., (2002), and Wadud and Whte (2000) on Bangladesh rce producton. [Insert Table 6 here] Among the nne varables selected to explan techncal neffcency, the coeffcents on the seven of them were sgnfcantly dfferent from zero at 1 percent level wth consstent expected sgns (lower panel of Table 3, column 4). The null hypotheses wth regard to the exstence of neffcency and valdty of the specfed predctors of neffcency were tested 10 We have allowed full nteracton of the land fragmentaton varable wth all the producton nputs. 11

13 and rejected at the 5 percent level of sgnfcance for both specfcatons (Table 4). The detrmental mpact of land fragmentaton on techncal effcency n rce farmng s hgh as expected. The elastcty 11 estmate reveals that a one percent ncrease n the number of parcels reduces techncal effcency by 0.03 percent (Table 5). Ownershp of key resources,.e., land, famly labour and draft anmal power, seem to have a sgnfcant nfluence n ncreasng effcency. The null hypothess of no nfluence of resource ownershp on effcency was tested and strongly rejected at the 5 percent level of sgnfcance at least for both specfcatons. Owner operators, n other words, land owners perform sgnfcantly better than tenants or part tenants. The elastcty estmate reveals that a one percent ncrease n the proporton of owner operators wll ncrease effcency by 0.01 percent (Table 5). The reason may le wth the qualty of land. In general, tenants receve less than an deal type of land from the landlords to farm, whch may lead to lower effcency. Farm households wth hgher numbers of famly labour operate at a hgher level of effcency. The elastcty estmate reveals that a one percent ncrease n the number of famly labourers wll ncrease effcency by 0.05 percent (Table 5). The mplcaton s that the substtuton of famly farm workers wth hred labour does affect rce producton effcency. Ths may be due to the unavalablty of hred labour servces at peak perods, partcularly n rce growng regons. A smlar effect s evdent wth respect to the ownershp of draft anmals. The elastcty estmate reveals that a one percent ncrease n the number of owned draft anmal power wll ncrease effcency by 0.03 percent (Table 5). Ths may agan be due to unavalablty of hred anmal power 11 The coeffcents n the neffcency effects model show only the drecton of nfluence but do not provde nformaton on the magntude of nfluence. We computed techncal effcency elastctes for these predctors by adoptng the framework of Frame and Coell (2001). As a result, we are able to provde a specfc measure of responsveness of each predctor on techncal effcency, whch s not commonly seen n the exstng lterature (for detals see Appendx A). 12

14 servces at the rght tme, partcularly durng peak plantng season. Among the farmers soco-economc characterstcs, educaton does not seem to have a sgnfcant role n mprovng effcency. The nfluence of extenson contact s sgnfcant, as expected n a country lke Bangladesh where extenson servce s nascent. Opportuntes for off farm work, and hence access to non-agrcultural ncome reduces techncal effcency, as expected. Adopton of modern rce technology sgnfcantly mproves effcency. The elastcty estmate reveals that a one percent ncrease n the proporton of modern rce adopton wll ncrease effcency by 0.04 percent (Table 5). 3.3 Polcy Implcatons The results of ths study clearly reveal that productvty and effcency are adversely affected by land fragmentaton n Bangladesh, a key nsttutonal factor that has the potental to be redressed through approprate polcy nstruments. In fact, ts detrmental mpact on productvty and effcency s hgher than compared to other constrants that the farmers face. The access to extenson servces s also a sgnfcant constrant, yet another mportant nsttutonal factor equally amenable to polcy adjustments. Most mportantly, ownershp of key resource endowments (.e., land, famly labour and draft anmal power) s another major factor. The elastcty estmate reveals that the combned effect of a one percent ncrease n the ownershp of these resources can mprove techncal effcency by 0.08 percent, and hence deserve proper attenton. Land fragmentaton s not only acceleratng the pace of degradaton and constranng agrcultural development, but also dscourages farmers from adoptng agrcultural nnovatons (Nroula and Thapa, 2005). Although the Green Revoluton has been vgorously promoted n Bangladesh over the past four decades, there are sgnfcant regonal varatons n adopton levels. Only a thrd of the sample farmers cultvated modern rce n addton to tradtonal rce varetes, although we have demonstrated the postve mpact of modern technology adopton 13

15 on effcency (Table 3). Land fragmentaton may partly be responsble for the slow and uneven dffuson of modern technology n Bangladesh. Khan (2004) rghtly ponts out that ncreasng fragmentaton of land n Bangladesh s a cause of worry rather than an ndcaton of a well-functonng land market as the World Bank clams. The general mplcaton of a lberalsed land market s that t could enable landowners to consoldate ther plots by sellng land further away from home and purchasng land closer to home and/or exstng plots. In ths way, the farmer could mtgate the constrants mposed by a wde scatter of plots to some extent. Land consoldaton measures amed at preventng land fragmentaton have been largely unsuccessful n South Asa for several reasons, ncludng demographc, economc and cultural factors (Nroula and Thapa, 2005). For example, the consstently declnng land person rato over tme shows the mportance of demographc pressure n Bangladesh (Table 1). Presence of ths demographc pressure together wth nhertance laws, whch dvde land equally amongst all brothers and half of brothers share to ssters (occasonally), provdes a powerful tendency towards ncreasng land fragmentaton (Khan, 2004). Hstorcally, land s seen as the ultmate source of wealth n rural Bangladesh. Farmers tend to hold onto even a tny parcel of land, whch may stll provde subsstence support for few crucal months n a year. The redstrbutve land reform polces undertaken n Bangladesh, such as settng a celng on land ownershp of a maxmum of 11 ha per farmer, has been largely unsuccessful. Ths s because dvdng the land between members of a household crcumvents the problem easly, and enables the household to retan the total land holdng whch usually exceeds the maxmum lmt of 11 ha. Also, redstrbuton of land (ncludng those reclamed from rsng Char lands n coastal areas) to a landless populaton had lttle mpact. Ths s because of the msmatch between the numbers of elgble members versus the total area reclamed, thereby, resultng n the redstrbuton of small parcels of land to a small number of households. 14

16 Furthermore, such programmes only add to vested poltcal popularty rather than address the ssue of landlessness per se. On the other hand, rural development programmes amed at land consoldaton va formng cooperatves also faled largely due to the power weldng of the landed eltes. These eltes often turn out to be the landlords, vllage leaders, as well as key players n the management of these cooperatves, thereby, resultng n poor partcpaton of small holder farmers. Therefore, the man polcy thrust should be amed at addressng the structural causes underlyng the process of land fragmentaton. These nclude among others, the law of nhertance and the poltcal economy of the agraran sector n Bangladesh. The latter conventonally favours accumulaton of land by vested ndvduals and groups wth factonal connectons up to the top end of the natonal poltcal herarchy. Wth respect to the law of nhertance, modfcatons are requred to mplement measures that would dscourage splttng land nto tny parcels amongst hers. Our reservaton about the success of the radcally redstrbutve land reform suggested by Grffn et al., (2002) s largely based on two consderatons plus a revew of past performance dscussed above. Frst, s the techncal and economc lmtaton, and second, s the poltcal economy of the agraran structure n Bangladesh. Grffn et al., (2002) made a strong theoretcal case for truly radcal land reform amed at transferrng land from large landowners to small owner operated holdngs to mprove productvty and effcency. Ther argument s based on the classc premse of nverse sze-productvty relatonshp, whch has now been weakened to some extent (Toufque, 2001; Ram et al., 1999; Wattanutcharya and Jtsanguan, 1992) However, even such deal compulsory redstrbuton (probably mpossble to mplement) would leave each landless household wth only 0.21 ha of land, whch s unvable as a lvelhood resource. Khan (2004) concludes that strateges of nsttutonal (land) reform that focus only on technocratc ssues (as above) wll not work unless the poltcal 15

17 nature of the problem n specfc countres (e.g., Bangladesh) s addressed n some way. He further stresses that wthout addressng the poltcal economc forces at work, no amount of loans from the World Bank to carry out nsttutonal reforms would have any dscernble effect on the bg pcture, wth whch we clearly agree. Hence, t s mperatve that an array of wde rangng cross-sectoral polces s devsed to address ths complex ssue of land fragmentaton nstead of concentratng only on narrowly defned land reform measures. The other sensble approach would be to undertake massve rural nfrastructural development amed at promotng non-farm employment and ncome generatng opportuntes. Ths would dvert people away from the already overcrowded agrcultural sector, thereby, releasng pressure on land, and hence, the process of land fragmentaton. Tan (2005) provdes evdence that the presence of land rental markets and off farm employment reduces land fragmentaton by 2 and 15 percent respectvely n Chna. Wu et al., (2005) conclude that farm productvty under a comprehensve agrcultural development 12 (CAD) programme added 1.5 percent to household productvty n Chna. The argument n favour of enhancng agrcultural extenson servces s straght forward. Injecton of resources s requred to mprove the physcal and nfrastructural facltes of the agrcultural extenson system as a whole. Not only nvestment s necessary, but also the core of the dscplne requres all round mprovement to make t attractve, remuneratve and effectve as compared to ts peer workforce, e.g., members of Thana cvl admnstraton or health and other servce sectors. The present level of coverage by each block supervsor (the lowest admnstratve level for agrcultural extenson) s n the regon of 650 farmers to 1 wth a spread of at least sq km, whch clearly ndcates the dffculty 12 CAD, launched n 1988, s a land development programme amed at nducng nvestment by farm households, cooperatves and the state to mprove the nfrastructure of farmng, partcularly, qualty of land (Wu et al., 2005). 16

18 of renderng effectve support to all elgble farmers 13. Few large non-governmental organzatons (NGOs), such as BRAC, PROSHIKA, UBINIG, etc., provde supplementary agrcultural extenson support, whch s largely confned to vegetable producton and ktchen gardenng, targeted exclusvely at ther clentele of women from landless households, and hence lmted n scope and content. The other mportant area of nterventon s n the lvestock sector whch needs revtalzaton as t has a drect mpact on farmng effcency. The draft anmal s the most mportant source of farm power n Bangladesh, although the lvestock sector s n total neglect from a polcy perspectve. The market for draft power transacton s hghly fragmented as well. Landlords and tenants share draft anmal power costs nstead of the commonly practced sharng of fertlzer and rrgaton costs n areas of draft power shortage because of ts hgh rental rates (Rahman, 1998). We also see that the use of famly labour mproves techncal effcency sgnfcantly. The mplcaton s that households wth large pool of famly labour are perhaps able to use labour at the rght tme, partcularly durng peak perods. Reducton of agrcultural labour households between the census perods (Table 1) ndcates tghtenng of the hred labour market, at least durng peak plantng and harvestng tmes. Such a stuaton ponts towards a combnaton of polces amed at promotng labour savng technologes as well as the smooth functonng of the hred labour market. 4. Conclusons The present study analyzes the mpact of land fragmentaton on productvty and effcency n rce producton n Bangladesh. The study also examnes the nfluence of the ownershp of key 13 The Department of Agrcultural Extenson has a total of 23,954 employees ncludng 460 Upazla Agrcultural Offcers, 963 Agrcultural Extenson Offcers, and 18,338 Class III employees who are predomnantly the block supervsors (DAE, undated). 17

19 resource endowments on techncal effcency, whch s not commonly explored n the lterature. Results demonstrate that land fragmentaton s an nfluental predctor of techncal neffcency and loss of productvty. Ownershp of resource endowments (land, famly farm labour and draft anmal) sgnfcantly ncreases techncal effcency, ndcatng that the substtuton of famly labour and owned draft anmal wth hred labour and anmal power servces has a detrmental effect on effcency. Access to extenson servces as well as adopton of modern rce technology sgnfcantly mproves effcency, as expected. Off farm work, on the other hand, reduces effcency, as expected. The polcy mplcatons are clear. Frst, polces geared towards addressng the structural causes of land fragmentaton are vtal. These nclude, modfcaton of the law of nhertance, regulatons to prevent land fragmentaton, rural nfrastructural development and the promoton of non-farm ncome and employment opportuntes n order to release pressure on the land, and hence retard the process of land fragmentaton. Second, polces that postvely encourage buldng up of physcal resources, e.g., the draft anmal, by developng the lvestock sector as a whole. Thrd, the mprovement of extenson servces whch has been consstently hghlghted n the lterature as well. The key s to have an effectve mechansm n place to reap the benefts of extenson servces whch remans elusve n many developng countres. And fourth, s to ncrease the adopton of modern rce technology. However, ths would requre concerted effort not only to develop new varetes suted for vared and/or ranfed condtons, but also to effectvely dssemnate them to farmers. Although a total of 38 rce varetes have been produced by the Bangladesh Rce Research Insttute durng , only a few are wdely avalable at farm level. As Ahmed (2001: 70) ponts out, mere avalablty at research statons does not guarantee that farmers wll be able to make use of t. The delvery of ths technology to farmers s the crux of the problem n ncreasng rce producton, whch wll reman a formdable challenge for polcy makers. 18

20 References Ahmed, R Retrospects and Prospects of the Rce Economy of Bangladesh. The Unversty Press Lmted, Dhaka, Bangladesh. Agner, D.J., Lovell, C.A.K., Schmdt, P Formulaton and estmaton of stochastc fronter producton functon models. Journal of Econometrcs, 6: Battese, G.E., Coell, T.J A model for techncal neffcency effects n a stochastc fronter producton functon for panel data. Emprcal Economcs, 20, Battese, G.E., Malk, S.J., Gll, M.A An nvestgaton of techncal neffcences of producton of wheat farmers n four dstrcts of Pakstan. Journal of Agrcultural Economcs, 47: BBS, Statstcal Yearbook of Bangladesh Bangladesh Bureau of Statstcs, Dhaka. Bangladesh. BBS, The Bangladesh Census of Agrculture and Lvestock: , Volume 1. Bangladesh Bureau of Statstcs, Dhaka. Bangladesh. BBS, Census of Agrculture, 1996: Structure of Agrcultural Holdngs and Lvestock Populaton, Natonal Seres, Volume 1. Bangladesh Bureau of Statstcs, Dhaka. Bangladesh. BBS, 1999a. Census of Agrculture, 1996: Agrcultural Sample Survey, Natonal Seres, Volume 2. Bangladesh Bureau of Statstcs, Dhaka. Bangladesh. BBS, Statstcal Yearbook of Bangladesh Bangladesh Bureau of Statstcs, Dhaka. Bangladesh. Bentley, J Economc and ecologcal approaches to land fragmentaton: n defence of a much-algned phenomenon. Annual Revew of Anthropology, 16: Blake, P.M., Sadeque, S.Z., Polcy n the Hgh Hmalayas: Envronment and Development n the Hmalayan regon. ICIMOD, Kathmandu. 19

21 Coell, T., Rahman, S., Thrtle, C., Techncal, allocatve, cost and scale effcences n Bangladesh rce cultvaton: a non-parametrc approach. Journal of Agrcultural Economcs, 53: DAE, undated. Department of Agrcultural Extenson, Mnstry of Agrculture, Dhaka, Bangladesh. Dhungana, B.R., Nuthall, P.L., Nartea, G.V Measurng the economc effcences of Nepalese rce farms usng Data Envelopment Analyses. The Australan Journal of Agrcultural and Resource Economcs, 48: Dovrng, F., Dovrng, K Land and labour n Europe n The Hague: Martnus Nyhoff. FAO, Supplements to the Report on the 1990 World Census of Agrculture. Statstcal Department Seres (9a). Food and Agrculture Organsaton, Rome. Fenoaltea, S Rsk, transacton cost and the orgn of medeval agrculture. Exploraton n Economc Hstory, 13: Frame, W.S., Coell, T.J U.S. Fnancal servces consoldaton: the case of corporate credt unons. Revew of Industral Organsaton, 18: Grffn, K., Khan, A.R., Ickowtz, A Poverty and the dstrbuton of land. Journal of Agraran Change, 2: Husan, A.M.M., Hossan, M., Janaah, A Hybrd rce adopton n Bangladesh: socoeconomc assessment of farmers experences. BRAC Research Monograph Seres No. 18. BRAC, Bangladesh, Dhaka. Ilbery, B.W Farm fragmentaton n the Vale of Evesham. Area, 16: Khan, M.H Power, property rghts and the ssue of land reform: a general case llustrated wth reference to Bangladesh. Journal of Agraran Change, 4:

22 Kodde, D., Palm, F., Wald crtera for jontly testng equalty and nequalty restrctons. Econometrca 54: McPherson, M.F Land fragmentaton: a selected lterature revew. Development Dscusson Paper No Harvard Insttute for Internatonal Development, Harvard Unversty. Nroula, G.S., Thapa, G.B Impacts and causes of land fragmentaton, and lessons learned from land consoldaton n South Asa. Land Use Polcy, 22: Parkh, A., Shah, K Measurement of techncal effcency n the North-West fronter provnce of Pakstan. Journal of Agrcultural Economcs, 45: Rahman, M The applcaton of fronter approaches to model the effcency of rce producers n Bangladesh. Unpublshed Ph.D. thess. Imperal College London at Wye, Kent, UK. Rahman, S Soco-economc and envronmental mpacts of technologcal change n Bangladesh agrculture. Unpublshed Ph.D. thess. Asan Insttute of Technology, Bangkok, Thaland. Rahman, S Proft effcency among Bangladesh rce farmers. Food Polcy, 28: Ram, K.A., Tsunekawa, A., Sahad, D.K., Myazak, T Subdvson and fragmentaton of land holdngs and ther mplcaton n desertfcaton n the Thar Desert, Inda. Journal of Ard Envronments, 41: Schultz, T. W The Economc Organzaton of Agrculture. New York: McGraw Hll. Sherlund, S.M., Barrett, C.B., Adesna, A.A Smallholder techncal effcency controllng for envronmental producton condtons. Journal of Development Economcs, 69: STATA Corp, STATA Verson 8. Stata Press Publcatons, College Staton, Texas, 21

23 USA. Tan, S Land fragmentaton and rce producton: a case study of small farms n Jangx Provnce, Chna. Ph.D. thess. Wagenngen Unversty. Toufque, K.A Structural constrant to agrcultural growth: an explanaton of the farm sze productvty relatonshp n Bangladesh. Workng Paper. Bangladesh Insttute of Development Studes, Dhaka. Tzouvelekas, V., Pantzos, C.J., Fotopoulos, C Techncal effcency of alternatve farmng systems: the case of Greek organc and conventonal olve growng farms. Food Polcy, 26: Wadud, M. Whte, B Farm household effcency n Bangladesh: a comparson of stochastc fronter and DEA methods. Appled Economcs, 32: Wan, G., Cheng, E Effects of land fragmentaton and returns to scale n the Chnese farmng sector. Appled Economcs, 33: Wattanutcharya, S. Jtsanguan, T Increasng the scale of small-farm operatons n Thaland. Food & Fertlzer Technology Centre Wu, Z., Lu, M., Davs, J Land consoldaton and productvty n Chnese household crop producton. Chna Economc Revew, 16: Helfand, S.M., Levne, E.S., Farm sze and determnants of productve effcency n Brazlan Centre-West. Agrcultural Economcs, 31:

24 Appendx A Dervaton of margnal effects and elastcty of techncal effcency 14 The predcted techncal effcency usng the condtonal expectaton for the th frm s: TE = E[exp( u ) ε ) = e 1 Φ( σ µ * / σ *) = exp( µ + 1 * 1 Φ( µ 2 * / σ *) = ( B / C) A = AD * 2 σ * ) where µ = (1 γ ) zδ γe, σ * 2 = γ (1 γ ) σ, 2 * s A B {exp( µ = * σ * )} = { 1 Φ( σ * µ * / σ *)} C = { 1 Φ( µ / *)} * σ D 1 Φ( σ * µ * / σ *) = 1 Φ( µ * / σ *) we wsh to obtan the partal dervatve of the techncal effcency measure wth respect to the j th element of the z vector. Use of chan rule we have: 15 TE TE * = * µ z µ z j * j ( a1) Furthermore, we have: 14 The dervaton strategy essentally follows from Frame and Coell (2001). However, the detals of formula used n the dervaton are slghtly dfferent due to the defnton of techncal effcency used n the STATA software. 15 The th subscrpt s dropped from ths pont forward for the ease of exposton. 23

25 µ * z j = (1 γ ) δ j ( a2) A µ * = A= A' B µ * C µ * 1 = {1 φ ( σ * µ * / σ *)} = σ * 1 = {1 φ ( µ * / σ *)} = σ * C' B' and D µ * CB' BC' = = D' 2 C Usng these results, we obtan: TE µ * = AD' + DA' = A( D' D) CB' BC' B = A 2 C C A = ( CB' BC' CB) 2 C Thus, usng ths result and equatons (a1) and (a2) we obtan the margnal effect of techncal effcency of the th frm wth respect to j th z vector as: TE z j A = ( CB' BC' CB)(1 γ ) δ ( a3) 2 j C and the elastcty of techncal effcency of the th frm wth respect to j th z vector as: η TE j = TE z j z j TE (a4) 24

26 Table 1. Farm dynamcs and land fragmentaton n Bangladesh. Varables 1977 Census Census 1996 Census Farm holdngs, operaton sze, draft anmal ownershp and labour Number of farm holdngs ( 000) 6, , , Percentage of small farms ( ha) Percentage of medum farms ( ha) Percentage of large farms (above 3.03 ha) Absolute landless households as percent of total holdngs (farm + non-farm) Average farm sze (ha) Per capta net cultvated area (ha) Draft anmal per holdng (number) Draft anmal per capta (number) Agrcultural labour households (% of farm holdngs) Number of persons engaged n agrcultural work (as self-employed and famly helper) per holdng (farm + non-farm) All holdngs (number) Owner holdngs (number) Owner-cum-tenant holdngs (number) Tenant holdngs (number) Changes between nter-census perods (%) Number of farm-holdngs Number of small farms ( ha) Number of medum farms ( ha) Number of large farms (above 3.03 ha) Number of absolute landless households Average farm sze Per capta net cultvated area Draft anmal per holdng Draft anmal per capta Agrcultural labour households Land fragmentaton 25

27 Average fragments per farm holdng (number) Small farms Medum farms Large farms Average sze of fragments (ha) Small farms Medum farms Large farms Source: Computed from BBS, (1981, 1986, 1999, and 1999a). 26

28 Table 2. Defnton, measurement and summary statstcs of varables Varables Measure Mean Standard devaton Mnmum Maxmum Inputs and output Rce output Kg per farm Land cultvated Hectare Fertlzers Kg of actve nutrents (N, P, and K) per farm Labour Persons-days per farm Draft power Anmal par-days per farm Seed Kg per farm Pestcdes Mg/ml of actve ngredents per farm Cow dung Kg per farm Cow dung users Dummy (1 = Used cow dung, 0 = No) Pestcde users Dummy (1 = Used weedcdes, 0 = No) Techncal neffcency predctors Age experence rato Number Educaton of the farmer Dummy (1 = have some educaton, 0 = No) Number of workng members n the household Number of workng anmal n the households Number Number Off-farm workng hours Hours per month Extenson contact Dummy (1 = f had contact n the past year, 0 = No) Tenural status Dummy (1 = f owner operator, 0 = No) Cultvated modern varety Dummy (1 = f also produced modern rce varetes, 0 = No) Land fragmentaton Number of plots farmed Total number of observatons

29 Table 3. Average producton functon and stochastc producton fronter estmates of rce producton. Varables Stochastc fronter model 1 Stochastc fronter model 2 Coeffcent t-rato Coeffcent t-rato Producton functon Constant *** *** ln Land *** ** ln Labour *** *** ln Fertlzer nutrents *** *** ln Anmal power *** *** ln Seed ln Pestcdes *** *** ln (Land) * ln (Labour) ln (Fertlzer) ln (Anmal power) ln (Seed) ln (Pestcdes) ** ln Land x ln Labour ln Land x ln Fertlzer ln Land x ln Anmal power ln Land x ln Seed ln Land x ln Pestcdes ln Labour x ln Fertlzer ln Labour x ln Anmal power ln Labour x ln Seed ln Labour x ln Pestcdes ln Fertlzer x ln Anmal power * ln Fertlzer x ln Seed ln Fertlzer x ln Pestcdes ln Anmal power x ln Seed ln Anmal power x ln Pestcdes *** *** 28

30 ln Seed x ln Pestcdes Pestcde users *** *** Land fragmentaton and ts nteractons ln Land fragmentaton *** ln (Land fragmentaton) ln Land fragmentaton x ln Land ln Land fragmentaton x ln Labour ** ln Land fragmentaton x ln Fertlzer ** ln Land fragmentaton x ln Anmal power ln Land fragmentaton x ln Seed ln Land fragmentaton x ln Pestcdes Model dagnostcs σ 2 = σu 2 + σv *** *** γ = σu 2 /(σu 2 + σv 2 ) *** *** Log lkelhood Techncal neffcency predctors Constant *** *** Age-experence rato of the farmer Educaton of the farmer Number of workng members n the HH ** ** Number of workng anmal n the HH ** *** Off-farm workng hours * * Extenson contact * Tenural status * * Cultvated modern varety *** *** Land fragmentaton *** Total number of observatons Note: All the resource nput varables were mean-dfferenced pror to estmaton and therefore the coeffcents on the frst order term can be read drectly as elastctes (see also footnote # 9). *** sgnfcant at 1 percent level (p<0.01) ** sgnfcant at 5 percent level (p<0.05) * sgnfcant at 10 percent level (p<0.10). 29

31 Table 4. Hypothess tests Null Hypothess Stochastc fronter model 1 Stochastc fronter model 2 Choce of functonal form Cobb-Douglas vs. Translog Model (H0: βjk = 0 for all jk) Lkelhood Rato test statstc (χ 2 ) Degrees of freedom p-value (Prob > χ 2 ) Decson Reject Reject Producton structure exhbts constant returns to scale (H0: Σαj = 1 for all j) Lkelhood Rato test statstc (χ 2 ) Degrees of freedom 1 1 p-value (Prob > χ 2 ) Decson Reject Reject No effect of land fragmentaton on productvty (H0: φpj = 0 for all pj) Lkelhood Rato test statstc (χ 2 ) Degrees of freedom 8 -- p-value (Prob > χ 2 ) Decson Reject -- No neffcences present n the model a (H0: µ = γ = 0) Lkelhood Rato test statstc (χ 2 ) Degrees of freedom 3 3 p-value (Prob > χ 2 ) Decson Reject Reject No neffcency effects (H0: δd = 0 for all d) Lkelhood Rato test statstc (χ 2 ) Degrees of freedom 8 9 p-value (Prob > χ 2 ) Decson Reject Reject Effcency effects of all owned resources are jontly 30