.-·~ -- . ;&~v~~-U--p'~ -) -(( -Qr, s:s·o _{':;f; [)'> . e}e-.. _ ··~ 93467 I BROADENING DEYE:t.OPMENT PROJECTS: FINAL REPORT ON . Easter11. Africa Department ...-. ·. -International Barik for Reconstruction and Development . - \ The views expressed here are those of the author and riot necessarily th~se of the. ~ank. This report may not be published n;,r may it be quoted as representing · the views of the Banl_c and it$ affiliated organizations. They do not ,accept responsibility for its accuracy_ or. · completeness.- I ! .',,',If II, B. H. Kinsey, Overseas _Developmeilt Group L-td. Consultant ! :·{':.) t{ni versity. of· ~a:st ~glia J\Ule '1975 _.- · '' . ~- · Norwich, E11gland \•, .:(.; . __ '· ' < r---- .! . .. 1 I I' i :.' l ~ i I I[ r I ,ACKNOWLEDGEMENTS ! Over the period of time from its conception to the writing of the final version of this report, ~he LilongWe Land. Development Program Follow-up Study has benefxted from . . . . the guidance,_ cooperation and assistance of a number of individuals to whom a very considerable debt of gratitude is owed~ For their hospitality and assistance in Lilongwe·, I shoui'd like to thank Andrew Mercer and Tony Standen, former and present managers of the Lilongwe project~ To Bob Reader, former evaluation officer at Lilongwe, go profound thanks for accomplishing in a very short time an enormous job of data·. ed.i ting without which the study could never have. b~en _done •.. · The usually thankless task of number .crunching was the responsibility of Ofelia Ela-Beguin 1 research'assistant in· the.Develqpment Economics Department, who .somehow managed che~Jfully_ and efficiently to get work done despite the many other claims on .her time. I am_very grateful to.uma Lele of the.E~stern Africa Department for her il)sp~ration.in the I I initial stages of the study CQ'ld for her helpful comments on. the f.inal draft. . .I also very much appreciate:~he thoughtftil . . and caref-q.l reading given to an earlier draftby Jim. Scott~ the present evaluat.ion_ officer at Lilon~e. Fina].ly, I stJ.ould . · like to express . my. th~ks to two colle,agues . . . . in the Overs~as Development. Group- Linden Vincent a,nd.Giltoy Coleman- · who s~ffered with ~e through· ~he ·.interpretation. of·. some· · . \' ., ... ' . . .·.·. ·. . . :. ' . .· .r .. dubious. statistical result.s, .and to the latter for sharing with.ine his wide knowledger o~ Malawi. . . -. I I ' B. H. Kin~ey .. · Overseas Development Group ·Ltd·. '-.': l ·i scho(?l ()f Development_.·Sttidies Norwich, Ef&glana·. · · · \ I June;: 1975 · . ' BROADENING THE EVALUA~ION OF RURAL DEVELOPMENT PROJECTS: FINAL REPORT ON THE LILONGWE.LAND ·DEVELOPMENT ·PROGRAM FOLLOW-uP·· STUDY (RP0-29,3) .. :..- SECTION I ·r IN~RODUCTION ,, 1. This report presents the findings and conclusions arisin·g from the analysis carried out in th~ Lilongwe Land · ',Development Program (LLDP) Follow-up·· Study· fRP0-293). · Certain . resu~ts from the study were distributed in preliminary summary· form.in May 19741 this report expands on the previously distributed summary findings and integrates theearlier work with.the results of analy,is carried out sinc::e May· 197·4. This ; - . - . . .. . final version has also benefitted from the coll'iments. provided on tbe .Preliminary report·::by the LLDP Evaluation. Officer' . Mr. P. J. Scott. 2 ~· Background--Familiar! ty ·is assumed with. the research project proposal in support of the LLDP follow...;up ·study and witll: .the Africa JRurai Dev~lopment study (ARDS) review of LLDP.: - J '· ' ' ' -: ' • ' . I Rural· Development in Maiawi: A Review of the Lilongwe Land · . I . Development Program (IBRD 'Studies ·in EIDPloyinent and Rural '\ . Development No.9; septeroqer 1974). The ARDS review provided a largely. ,quaiitative eV;~luatl·o~· of the effectiveness ·of th.e · .various components iJ;tcludedin LLDP as well as a general back,.;. grQ~d .and description of the project. The review. also generated - . - : - . a large number. of. hypotheses rei - a ted .to~· imPc>rtarit ' ~ aspects .of the· project which, it was thou.ght, were capable .of bein,g test~d with datacollectedb:Y.the .project's Evaluation .~ . I ' I I. L _ ... ,. ._ .. ,.- --. 2':__ -_ .. Unit and a small amount of supplemental data collected in the -f!eid.. 3. Selected data from the project were· shipped from-·,.-· Malawi to Washington, and the former headof the project•s. Evaluation Unit--Mr. R. A. Reader-;_was brought to the Bank to supervise editing and organ!zatiori of the data for analysis~ The supplem~ntal data were collected in Malawi iri September- October 1973. 4. · ·checking the data for consistency arid putting them into- a· form sui table for the computer processing envisaged .·-· - proyed .to.be enormously tedious and time-consuming~ At virtually every stage of data preparation, anomalies were detected in the data which had to be corrected before work could proceed-further, and it quickly beca:m~ evident that tbe phasing outlined for the study had been extremely optimistic. The study was launched two months later than planned, and ad(litional slippages began almost immediately. Analysis got underway only beginning in February 1974, ,wherea's the original· timetable called for the study' t6 have ~een comp1et~d by the· end of 1973 ~ 5. Object! ves-_-In the terms set out. in the c::>riginal proposal, the objectives of the·_ follow-up study were to assess the_ \lSefulness of a range of data collected by the Li:.OP .· .... •' Evaluation_ Unit. for the purpose of evaluating the .hnpact of the project. --. In particular, attention was to be focused on the abil;J.ty of the data to reveal what has happened at Lilongwe, ' ' - a .,..search-for by ·; ' project benefits which might be quantified in the data'- and to reveal why changes have occurre4, by atte~pting to assess the effectiveness of various' proj~ct components; - pa~ti~u,la~ly credit and the extension servic~. ·In additi<;>n, I th~. follow-up study . .. . was tc;> scrutirtize the ~at~ . to . dis~over; if . anything could be learned about the problem of declining ' grorindnut yields experienced in the project. area_, a matter of i I ~erious concern for project management . . Finally, it was hoped I that ~he- ~xperience. . gainE!cl thro~gh the foli~w--up ~tudy woul:d .' I, . . ' . . .. ' n. produce some guidelines for improving monitoring and evalua- tion. procedures both in the'LilonC}We project and-possibly in other Bani( projects as well.· 6. LLDP is among the most documertt~d of th~·Bank's rural development projects. .since 1968/69, the projee:t• s _Evaluation Unit ·:has collected an enormous range of data in an ambitious program of ~nnual and special-purpose surveys! The greatest part of the project's evaluation effort has been directed toward ascertaining the levels of crop yields obtained each season, since yields were the performance criterion given greatest stress in the first two appraisal reports. Even with t~is- strong focus, however, data were collecteu'i>oth through the yielci surveys and the other surveys which _app·eared to have- the potential of providing insights into projectimpact in areas not inUnedi.ately related to crop yields~ but the initial anal'ysis had not explored much of this peripheral-data in ~l'lY great gepth. Hence it was hoped to broaden the. evaluation of_ - LLI;>P through a re-exain:l..naticin of the data .. 7. _· · ·. Accordingly, in the analysis in ·the J:ollow-up study, pa.l;:ticu~ar attention was directed to '"three' areas of concern wh,i.ch had.]:)een largely neglected in the previous analytical work.,:. __ '- .' .. -- ··;, ·. (1) Determinants of the distributi,on pattern of:. 'credit~supplied inputs~ . . . . . . . . (2) Analysi~ of the rel(itionships·between apparent need -.for. credit, as defined by various measures 0! Wecdth.oand iriCOme 1, and credit.' distribUtiOn; and .. '• (-,3) . specific factors related to the.effectiveness of the extension sta~.f .in stimulating adoption ' of yield-increasing innovations~ ouri~g the analysis, the emphasis in (2) wa~ gradually broadened to ·take into acCount income -distrt'bution and the apparent·. determinants of income levels_. l:n addition, considerable ..,.~ ~fforts were made to analyze the principal' determinants of ·: ··' crop _yields in the Lilonc;Me area •. This· analysis sought to .det~imine why.yields were what they were r~therthanattempting to s~.eci£y more precisely the overall level·of yields obtained. Subs~'quently· this part of· the analysis was. restricted to those farin~rs who, at the time of data collection, were residing within. the boundaries of the proj~ct 1 sdeveloped area. 8. . . As the objectives of the evaluation carried out in ; ; the follow-up study differed considerably from those of the '' I. I . original ev'aluation effort; brief attention is given below ·t.o. the problems which arise from divergent evaluati()n goals. ,·;_: . . 9. Conflicts Among Objectives in Evaluation Work--In order.to relate the evaluation function to,project structure, . a .prpject should be conceptualized as a system c:)perating at< several different levels. At the first~-or project--lev~!,· goal~~ are spelled out in. broad terms, e •.g. raising family incomes by so much, doUbling yields of specified·crops, et~. I I Evaluation at this level involves finding measures of overall I I. I project effectiveness, measures which will relate "project ' ·effects and costs to project goals. For LLDP; the identifica- tion of .such measures was largely pre-empted by. the perhaps '. I ,. . • , , . , .. inappropriate· emphasis pl·aced upon yields· as the measure o'f project effectiveness. I, 10.: ·At the second--or section/division~~lev~:d . .. . .are spelled out the sub~goals which, ..when met iri a coordinated_ and phased. m~ne·r I enc;tble project go.lils. to be attained .. '. The sub~goalS I to a,laJ,=ge degree, ·define·the ftinctions of the various sections within - -. a project. Activities will be performed .by each SEfction i. · . and.·. the production . . . . . or other targets .. of .. . each activity . .. . ar~ ... . . . .. I .. specified, at least· in general terms, in· th~ a11nua1_· budgets i set dut ill· the appraisal reports •.. These sub~goals might i . I . ' iJ:l.c~~de, _~or exainple, the issuing of input~ by the.:credi.t sectio~ su.~fic;::ient to grow a specified nwnJ:?er·~of ~cres of a I given crop at a certain input intensity • . The relab:~d sub;..goal I ! 5 - - •· ;· •' ·'" for the extension section might be to ensure that the inputs are used correctly, while the section responsible_for marketing might be expected ·to make arrangements f~r moving· and stortng_the anticipated incremental output! Aily weak. link I in this chain of related sub-goals would ine~i tably have I . ;· . detrimental consequences for project goals, hence monitoring and . - evaluation . to detect.and . permit correctionof shortcomings at. this level are essential. 11. At the third level of operation, targets are set for . individual activities· carried out by the different section~. These targets will be expressed in terms· of.forecasts of inputs imd outputs· for each type of activity. Evaluation at this level, as well as at the section level, involves the identifi- cation of performance indicators--variables ~considered to }:)e of high priority in connection ·with speci~i,ed. goals-:--which ~re capable of measuring the effic:i.~ncy and level of. operations ·of the different sections ~nd activities within sections. . ' . . ' . . . . . 12. ~~aluation. . . units are placed in projects ~o provid~ .·. a service, but they can J?rovide this. service effectively. only when two conditions are met:· (1) the demand for information is .thoroughly and accurately specified~ and (2) they are ' . . . . . . . provided with the resources i;o permit thein ' . . to supply the ' infoJ:Iilatic;:m demanded. By and large, these two con.ditions appear not to have been met at Lilongwe. 13. The LLDP Evaluation Unit has at vario~s times had :.to · serve three different masters--the Bank, the Malawi Government and project management; each of which has imposed its own some- what divergent demands having diff~rent. implications for . eva! uation work. .(In fact, most of these· det;nanc;is were chanp.elled through project management, but· the:net effect was the same.) Unfortunat·ely, the most frequE!~t point of agreement . . . . I ' . among_ th~ three was that more or better information was needed . I. ·.On. yiel~S 1. thUS r~inforcing the StreSS On yieldS already I mentio~ed •.· As a result, the Evaluation un·it. concentrated ~n .. .. - ~.6 -~- . . providing data to permit an assessment of whether the proje·ct- level'_goal was being met and devoted little attention to identifying and examining performance .indicators at the low~r:- or. sub-goal--levels. Consequently, only when it .came to be . realized after several years that project~level goals did. not appear to be at predicted levels did the search begin for ·factors which might explain this phenomenon. A. inore management-.· ... oriented approach to evaluation could possibly have detected problems earlier than was the case and would certainly have provided more useful information for the diagnosis and correction of problems. The staff of the Evaluation Unit have not been unaware of the conflicts between the work they are i·. ' do~ngand the usefulness of work that 1 might be-done; they simply do not have the resources to do both. . . ' . 14. The process outlined above explains in significant. measure why the follow-up study, as will be shown in the I. secti·ons. ~hich follow, is able to say so little about why certain things appear to hav~·happened at LilongWe~d. why otner. t~ings, . . . which might be expected, ' have not happened. : Geneially speaking, these matters :were outside the scope o-f the evaluation effort at the -time the data were collected ~pon . . .· whicQ.· this study is based. Moreover, .on those. rare-occasions wherf . the Evaluation Unit was able to examine ·performi:mce at . other than the project level, the in~pection ·process is not reflected in . the data avail~ble to the follow-u~ ~.tudy. 15. F,i.nally, another important 'factor also deserves mention. For three seasons,· in addition to it_s normal arduous .work--load, ·the LLDP Evaluation ·unit also mounted .a farm· manag~mel)t'study of gargantuan proportions. it is estimated' I . ,. that-30-40 man.:..years and some $90,000 were invested in the I I' I st\}dy;, a . .drain on reso~rces which virtually ~rtstired that no ': f. other< evaluation work cou].ci be undertaken while. the study I I. c~ntihu~d. · lndeed, since the termination of the study; the • ? . ' .• ' . i . normal_. evaluation load alc:me has been Suff~cie~tly heavy to I I: . I prc;>h~bit the. analySiS' Of _the data COllected at SUCh COSt. - 7 - ··--:: .. ,- -16. A -footnote must be added to the reinarJ~s above to the. effe'ct that they apply most strongly to the peri.od 1968/69 to 19.71/72. Many improvements and· changes have since .been made. Sugge'stions for further changes· are contained in Se.ction VI-I. 17. Procedure--The data available in Washington included the project's yield samples for three seasons: 1969/70, 1970/ -71 arid 1971/72. In addition, data were· available. from the socio-econC)mic survey (SES) carried out in 1969/70, and from the supplementary survey carried o'ut ·in 1973 •. This iatter . . ' ' survey utiliz_ed a subsample of the raridom sample selected for the 1972/73 yield suryey and . was on a smaller . scale than any of the· other surveys. Sample siz.es and the number. of original and transformed variables included in each survey :are shown below: ··.! . I Number of variables ' Sample size .. Survey Origln.al Transformed. Original ·Study Difference :. 1969/70· 102 22 986 i. }.,005 +.2% .. 197otpl 12.2 64 873 895 +3% 1971/72 162 69 1,181 918 -22% SES 86 24 772 682 -12% IBRD/ARDS. 27 0 117 117 0 : .. :,,· 18. · With the exception of the .. IBR,D/ARDS survey, . . it . was ..: the _transformed variables which were emp~oyed'in the analysis. Each· ~f the three yield surveys was somewhat"different, arid a general tendency existed for. each. subsequent. ~urvey to be.come mor~ complex than its pr~decessor •. One· of the difficulties enqountered in preparing 'the data for· analysis was that of making data from the different ·surveys. comparable· in order that ' . infer~nces co.uld be made as to· ~changes over time. For the first· _.;two years • yield surveys, the samples lised in the follow- up. s:tudy ,were. larger than those. used in the or:iginal analysis. This ·was because the data received from Malawi· contained a . --·a: ...- . . I I. large· number of duplicate cases, not all of which could be ! identified.during editin9. ·It is thought, however, that arly - ' - I distortion caused by including the rel·atively small number c)'f duplicate cases will be slight. I 19. · With the 1971/72 yield su:x:-vey and the_ socio-economic survey, problems of a different nature were encountered~ · The data received from Malawi contained fairly large numbers' of invalid cases, and it proved impossible to identify some of· the .data received. Moreover, quite a large number of computer cards had been physically damaged in such a manner that the data ·On them could not-be rescued. The magnitude of the losses in these two .cases is clearly enough to influence the results obtained, but one redeeming factor is that the missing ' cases appear to occur in more-or-less random· fashion so that no consistent bias may be caused by the loss. 20. Because of problems with missing variables in the. . '. . . . . ' . data sets used, in only very few cases was t}le sample in _any · I given analysis the same as thosepreviotisly shown. Control cards were used to exclude. observations where variables were I ' ' missing and to defi~e criteria . for including·observations .. . in- any given analysis. Thus, even for·a given year, the-sample size- will' vary from ,analysis to analysis, but .almost ' . . .. consist.:. .. ' ently--with the exception-of ~he credit.subsamples--samples were·' large enough to give meB:ningfui re~ults. · 21. The fact that a second analysis was carried out on the d(lta collected by the LLDP Evaluation Unit is not meant in any way to imply criticism or deficiencies in the Unit's work. The Unit has had its problems and been beset with growing,-- , pains, but ·at all times it has made commend~ble progress. and i .. · has accomplished tasks which· seem·· beyond t}le. capability of the I . ' ' resour.ces allocated to the Unit. The al.'lalysis reported in _ _ this study is not intended as' a' crosscheck' on previous wo·rk of ' the. Unit; _in most cases,· in fact, the ·results reported here. I ·, - ' ' I ' '' • • , will.: not be directly comparable with the earlier work because I , , 9.·.:.1. : . . . .. • ·different procedures have been used on samples that themselves differ from the original . ' . .. . A quick comparison of results with those:obtained at Lilongwe indicates that :the.tw6.bodies of work are in broad agreement even if there might be some argu-. · ···m$nt .as to the precise level ·of· some· variables. 22 .· ·. . During the preparation for analysis and. actual \. I analysis two volumes of dpcumentationwere assembled. The firstdescribes.in minute detail the editing and compilation· ·procedures . .· ' and . the data format, while the second outlines the . ' hypotheses to be tested, :sets up the models to be used .. and . . defines the variables to l:,e employed .. Most of· this documentation · is· to,o:speciali_zed to be Of cjene~al. interest,: and it is not . reviewed here~ 23. ·Reporting Conventions--Ma:ny·of the results are i'. ; . reported in the form of data arrays o_r crosstabulation table.s : '. with classificatory va_riables as labels. In most cases~ the data are -reported ·along with either one or two. confidence · · interval.s. The first of these~-designated CI (x) :.-is· the i ·. co·n~idenqe interval around the fflean and the second--designated I I. i I CI (;n) .--is the confidence interval on the· observatio~s. The i. actual val~es of the confide~ce intervals are. reported as a 'I pair of numbers, the second in.parenthef:'es· following the first, _e ·.:9 .•.. ±3() (~2J. The. first riWnhe:i: represents· the Limits oi the int.e:r;;val a~ a 99 per c¢nt level of confidence, while the second is :the limits at a 95 per cent level of confidence. ; 24. :. The number of observations in a sample: or sub-sCln1ple , .. use~ 1n .the analysis .. is reported in one of two ways: . (1) . as . a :figure. in. parenthes_es after the category l(ibels, . as in "Maize .. (29>''~· in the first line of Table 1, or (2) · a:f:J a figure in p~re.rit}leSeS follOWing a reported. value for cl · varia})ie 1. as. in I "295 . ~ : (2)" in the fir~t line of Tabl~ 3~ ~- . 25. Four tests of significance have been.iused •.· These , . are a:s fqllows: '. :., . . ··..· ... -lb- ... l • :-!. '. z : Used to test the significance of ec.ch of a .~.:-: ., group of ~tratified sample means incornparisori with the population mean. · ·> t Used to· test the significance of .the difference between pairs of sample means from the same population • .F : Used to test the signific:ance ·of the difference between means where analysis of variance has.·· been used.· Only applied in cases where sums of squares were available with results-. · 2 x :. Used to test, the significance of the difference i between expected and . observed values of a I i variable. I . I ~ - I I I ~ . - .11 - SECTION II DETERMINANTS OF THE DISTRIBUTION OF. SEASONAL CREDIT . . 26. Previous analysis of the LLDP credit program (reported in the ARDS review of LLDP) had revealed that there appeared to be a distinct geographical pattern a~sociated with the use of s~asonal credit in the project area, and a nwnber of hypotheses relating to this pattern were generated for testing in the follow-up study. In addition, there are other. factors, s~ch as income and farm size, which one would expect to be related to seasonal credit and·for which hypotheses were tested. Because of problems with the data, certain of the hypotheses: identified could not be tested and these are discussed at the erid of this· section. 27. The principal determinant of the distribution·of seasonal credit was meant.to be the system of rating farmers• creditworthiness which has operated at LLDP in OI1e. form or .. another ';, . . since .the first se~sdn of ~redit operations. . - ·, ' .~.. . .. . .. . This. . :.. ' system operated somewhat differently in each of the two years examined in the study. In 1970/71, a lO..,...point scale .was used by extension staff ·to rate ' . creditworthiness while_,. in ,1971/72, . . a 5-point scale was used. Each . . scale was intended to assess . .manage:tial_abilityas reflected in the performance of a range of.sp~cific husbandry operations. The syst~mneve:r worked in. pr~c~ice .as it was meant to in principle, however; for.it. placed a very heavy, .. and· possibly inappropriate, burden. on extension staff, who would have had to visie·ev~ry·farm in the_ir a:r:ea .of responsibility~ often within a- p~rioCI of a f~w .· I days,. . ·' and to make some 10 such visits every season if the sy~t~m were to operate as designed. In. fact, the_ system o~ly . ; ever operated on ~n ad hoc basis and the majority of projec~ farmers were not. rated for the purpose of es.tablishing credit-· worthiness.'· (See further discussion at the end of this section.) 28. Credit Ratings and Crop Yields --If the credit rating system were··.object.ively and accurately appli~d, one . . . .. . . '. would expect to find· a fairly strong correlation between the I. i credit rating and crop yields since a higher rating implies better management and better management implies higher yields. This expectation was. exami_ned using data from the 1970/71 and 1971/-~2 seasons. 29. . Shortly before a yield sample is taken from a given plot,·Evaluation Section staff visually assess the condition of the plot and assign a subjective crop rating whi.ch indicates ·th~ir eXpectation of the yield to be obtained. Since such "eyeb~ll" yield estimates have in various . circumst.ances been proposed as. an alternative to expensive and tirne":"co_nsuming.: . yi'eid sampling,' it is consi,dered worthwhile. :to look .at the correlation between. the. ' crop . rating and : . the actual yield. Data :,for the same two seasons were ·also used for this ·purpose. 30~ Simple correlation coefficients were calculated between the appropriate pairs of ratings and crop yields for those farmers who had credit, and the correlation matrix is shown beiow: · - ... .. ~ ·~ 1970/71 yield of:: 1~71/72 yiel~ of: Maize Grouridnuts Maize Groundnuts Extension credit·raling o.i14 0.114 0.066 0.151 Eva.lU'at_~on crop rai;ing -0.113 0.248 -0.015 . o. 298 Maiz$'yieid 1.000 0.221 1.000' 0.105 Groundnut yield 0~221 1.000 O.lOS 1~000 All the coefficients relatirig·ratirigs to yields. are very.1ow, andonly the Evaluation-groundnut coefficierit exceeds 0.2. :Evilluation.Sectiori staff alsb ~eein consiste~tly. to have under-. ,---------------~- ------- rated maize yields. All in all, the ratings are weakly and inconclusively correlated with yields. In the case of credit ratings, this may be because the rating system was never applied as it was meant to be. The results for crop ratings suggest that visual assessment of yield$ would be a very poor substitute for actual yield sampling. The bottc.. m hm lines of the correlation matrix give the coefficients relating physical yields of maize to those of groundnuts; for both seasons the,re is a. positive hut weak association between yields of the two crops. 31. Holding Size in the Low-Credit Areas --As noted earli'er, there is a distinct geographical pattern associated with the use of seasonal credit. Those program units lying· adjacent to the city of Lilongwe receive less-than-average credit on a per-farmer basis while all other unit.s receive greater-than-average credit. One hypothesis advanced to explain this difference is that farmers in the peri-urban area have smaller holdings as a result of hiyher population I ,. densities caused by population spillover from the city. One criterion farmers must meet in order to receive credit is that-they must have at least one acre under crops. If,with normal distributions, the mean holding size is smaller for samp~e ~ than for sample ~, this suggests that a greater proportion of farmers in sample- ~ than in b will be inelig':"'"· ible ,fo~ credit on the acteage crit~rion. 32. The hypothesis was tested by stre~:tifyingthe entire project area into two groups: those units. lying. adjacent to Lilongwe (nos. 1, 2, 6, 9, 16) and all other units. Originally "holding" was defined in two ways: (1) as cropped area and (2) . as cropped area plus fallow; but as the size obt~ined using the second definition was on average only one-tenth of an acre larger than the cropped . this second measure . .was subsequently dropped area,. . . . from the analysis. Hereafter, unless "holding" is clearly_differen- tiated from "cropp-ed area", the term is used Cis a synonym .. .. · : ..... for cr6pped area. Data from the 1971/72 season were used in the testing 1. and the results of the stratification· are given: below: ·,···· Mean cropped_·· .. Coefficient. pf · area variation· . - - Peri~urban project units. (150) · - 4.30 acres 72% All 6ther units (768) .4.88 11 82% To.tal _ (9~8) II 81% CI (x)- · ±0.33(0.25) C:I (ri) .· ±10 . 02 ( 7 . 61 )' ' 33. Cropped . area in· the low-credit units . is 12 per cent . . smal'ler than in other units, and this difference is signifi.-. ·cant at five per cent (t = 1.67). The mean cropped area in the low-credit units near the city is 4.3. acresl while in · · other areas it- is 4. 9 acres. Both holding and c-ropped ar~a· 1 · however, ar~ somewhat more homogeno~s i~ the low-credit areas. Based· on confidence intervals for the ob~ervat.ions. in the. two sub-samples .and 'assuming a normal distribution of cropped ( . ·. . . . . . . . . .. . : . . . . ·. . area, some. 15 per cent of .farmers. near the town would appear to be ineligible for credit -on the one_-acre. rule. But for the. other areas and for the_pr6ject.area·as a whoie;-some 17 per ce~t ot_farmers appear tO be ineligible on this criterion. 3,1-~ .'i'htis ·farmers in the low-credit <;lreas near the city dQ .have.smaller-than-averacje'holdings,- but~- somewhat smaller- p~oportion of this group is ineligible: _for. credit ' tha,n in the high-credit areas away from the_city. The fact that. farmers· near the city do not take as much ·c.redi t. as other farmers _must ·be expiairt~d by ~ther factors~._ . 35. The findings here have important_implications _for the_design ·of cr~dlt programs which-hav~ as part of their objecti'V:~- reaching the small~r (and poor.:!r) farin~rs. ·The - 15 - ... results ~ugge~t that a st~ndardized one-acre crop ~redit package will be inapplicable to· some 17 per cer1t of growers. in the area .. and that breaking ~uch packages ~p into smaller uriits may be beneficial. For obvious reasons, ·however,· such an approach might result in.higher·administrative costs for the credit program. 36~ Incomes in the Low-:Creciit Areas --An alternative hypothesis which might explain why farmers in the peri~urban area receive les.s credit is that they have higher incomes than farmers in other areas and hence, for this group,_the use 6f credit is unn~6ess~ry~ 'This hypothesis ~as tested using 1969/70 data for total income (agricultural plus non--: agricultural income) arid.the same stratification scheme used above. The results are a's follows: Mean Total. Income ._ .... Peri~urban project units (·232) Kl21 All other units. (449) 218 Total (681) 185·. CI(x) ±22(17) · CI (n) · :!:576(438) 37. ·Mean total income in the peri-urban area is only 56·per bent of that obt~ined elsewheie in ~h~~projedt area and the difference between the mean incomes obtained in the · two· strata is ·significant at· the one· per cent· level (t = 2.68). What is ~tirprising, however, i.s that the difference in incomes i,s in the opposite.direction . . . of that expected, i.e. de:spite apparently greater opportunities to take wage employment in the city and thus augment their agricultural incomes, .there is no strong'evidencethat farmers living . . . near the.city in fact take advantage of these opportunities. :i:nCieed, the·total income of this group is c~mparable to ~he agricultural income of farinets who have had no wage employ-:- ··,, I.., -. ·._. 16 memt · (see discussion· of agricultural incomes and wage ·earnings in. section VI). 38~ ·These results are extremely difficult to interpre~ meaningfully. It has been shown that farmers living near the. ·city: have holdings somewhat smaller than those.· in other areas, b~t is the fairly sm~ll ditferenc~ enough to explain the .stri)Sing ,disparity iri.irtcomes? This explanation seems unlil<.ely~ The ARDS review of LLDP suggested that farmers in the peri- urba:Q area have only a weak commitment to agriculture as they.· are now primarily urban-oriented. The logical evidence.to support an urban orientation would be higher ' . . . total. - - incomes . . . arising from wage employment plus.t~e .sale of some agrfcult.,.. ·ural:produce, but such evideil~e isnot forthcoming. Perhaps the explanation for the relatively low tot~l c~sh incomes· 'near the city lies inthe.fact that each farm is supporting a .l.c:l:r<Jer ' number. of . people than in the more .rural areas so that there is less surplus produce to:market. But, if this. . ~- ·is the. case, it would be . thought that farmers here would have· . ~ 'St]:'Of!g incentive tO USe prQgrain Credit 1 . especially for fOOd · crOPi3.· 39.. ! In COI1Cltision, about the best that can be Said is that the reason for low iev~ls of credit use ,near the city r~~ains S()mething Of-an enigma. The reiative .levei of' credit us·~·: h~s b~en defined on a per~ farmer basis, I:J.owever; so the . population .:argurnemt above may Weli have COr:ISiqerable Validity • . ·.: . 40' •.. _Credit, Farm Size and Sdcioeconomic Status ~.;.If a cr~dit. p):'ogram .is to· have .~ signific~nt impact oq, income . . . distribution, it is imperative that .the p~ogram. ~.~ach those . who may:· need it most. By design the LLDP ·.credit program ·is . heitll~r· designed nor. intende~ to. achieve. incb~e .•~edistribu-· tion·, so one would not expect to find any bias in ·favor of. smarter and poorer·farmers •. It was. hypothesized, therefqre, tha~· within_ LLDP·. there is no significant d~~ference in farm size or.socioecbnomic' status between farme~s:teceiving crE:!dit - i7 -~ . . . . and those not re~eiving it, i.e. ~hat the credit progra~ is equitable in giving all farmers an equal chance.. The hypothesis ·was tested using data :fi:·om 1969)7o. , ·-1Jnfortunate1y 1969/70 is not a representative s.eason, for ~in this year dnly about four p~r cent of eligible farmer~ ~~ok credi~: this is the lowest proportion taking credit in any year since the program began, and as a result the sample of credit recipients is small. Moreover, 1969/70 was a seaso~ in. _which the creditworthiness criteria were rigidly-applied and the repayment rate as of the date when accounts became . delinque-nt was the high~st every. achieved (97. 2 ·per cen:t of· value loaned). ·.' .. ,. .. ' . 41~: Farm size was measured in terms of .cropped acres~ Socioeconomic ·status was measured b_oth by .·t:ota]. income; . ' i~cluding nonagricultural income, and by· a composite index I: '· intended to measure socia_l status. . (The components .. . .. of this· . ., index are discussed under agricultural income and·social status in Section VI. ) ·The· mean value of eac~ v~riable is shownbelow for those who-received ·credit and those who did not: Cropped· Stat·i.ls ·Total area index -income Credit recipients 6.29 acres (27) 5.69 (26) K324 • (26) Nonrecipients of credit 4.74 II (829) 3. 27 ( 656) 176(6'56) ·. ·; ·_ _T()tal ·4.79 il (856} 3.37 (682) .. ' . 181 (682) '. . CI (x) ±0. 28 (0.-22) +O. 2.6 (0. i 2) -· +44(34) - . -- ' ' ,:· .CI(n) ±8~18(6.21)' ±6.79(5.15) :tl '160 (882) 42. Each of the three . variables rneas1.ir'ed indi~ates t:;hat in' 1969/70 cred;it reciP,ients were a favored group and.'' were considerably better off than those whodid not receive credit. Credit recipients ·~ropped 33 per ce~t. more land; .·· had a ·status index 74 per ce~t higher and'had incoines84p'er ·- 18·- cent greater than those who did not re6eive credit~ ~he .d~fferences between the two groups have the. following leveis of· significance: cropped area· one per cent (t = 2.40); status index 0.5 per cent (t = 4.67); and total income five per cent (t = 1~65). 43. These results are not surprising. since the major · objective of assessing creditworthiness in 1969/70 was to identify those ~armers who were likely to make the'best · economic use of inputs and who were most likely to repay the ~oan. The evidence suggests, how.ever, that the farmers . who met these criteria were also those with the least need for· institutional credit. It is the dilemma of. almost .all institutional credit programs that the viability of the program in a commercial sense must be balanced against the objective .~ : . . of providing credit to those farmers.who need it but :Who are not necessarily in a' position to tak~ 11 ~est 11 · advantage of credit. In the 1969/70 s~as?n .the LLDP ~redit program placed great ~mphasis on the creditworthiness· crit.E3ria; commendable repayment results w~re.achieved though .. '· . number. the r .· . and. distribution of loans were limited.· In sub- c • '· sequent·., _, seasons the total. vol~me of credit, the number of . ., b()rr(>wers and the average loan size all increased substant~ ia~ly. The credit~orthiness criteria were relaxed, with sqmeadvers~ effects on r~payment rates,· but both e~tension arici' ~redi t staff sought to ensure that far~ers- -to whom imp:roved inputs were a complete novelty· received the i~struction necess.ary ·to use- the inputs effectively. The rel~t~on~hip betwee~ receipt of credit and~exten~ion visits in t_~e- two seasons following 1969/70 wiil be examined later. -- . . 44. Type of credit Package· itt Relation to Farm Size: and Labor- Force- --It was felt-that there rnight.be• importap.t associations between the; type.of-credit package taken and the s.i,ze of the farm holding and the· amount of._ labor available .in· relation to the area cultivated, so the hypothesis was foriil~lated and tested tha~t.' the~e exists no significi:mt - 19 - difference. with respect to these variables among farmers taking different credit packages. 45. The two variables of interest were defined as follows. First; holding size was measured as· the actual number of, acres planted to crops. Second, available labor force was measured in terms of the number of adult depend- . ents residing on th~ farm in relatibn to the area cropped, and this measure is referred to as the labor-land (LL) ratio. Data from the 1970/71 and 1971/72 seasons were used to test the hypothesis but, as the definition of a resident adult dependent differs somewhat for each year, comparisons of the LL ratio between the two seasons cannot be made. The sample used was obtained by collecting data from credit records on those farmers who, during the annual yield surveys, clai~ed to have received credit for the season in question. In fact the names of many of those who stated that they had received LLDP credit could not be located in program credit records. The sample, therefore, contains only those farmers who we1.·e confirmed as having received credit. The sample was stratified by type of credit package received and the . two variables were then crosstabulated against this other . stratification. The results are given below: Mean for 1970/71 Mean for 1971/72 Type_of Cropped· Cropped credit Eackase LL ratio area LL ratio area (1) Hyb_rid maize 0.39(25) 6.48 acres 0.27 (22) 4.36 acres (2)Synthetic maize 0.32 (52) 4 •. 33 II (3)Groundnuts 0.59(17) 4.48 II 0~ 40 (11) 4.01 II (4)F.ertilizer 0.26 (10) 5.98 " (5) 1 or 2 + 3 0.50(18) 5.46 II (6) 1 + 4 0.43(10) . 6. 39 II o. 30 (12) 7;.16 ·. II (7) 2 + 4 0.33 (28) 5.79 II n (8) Other 0.46 (8) 5.69 II 0.26 (40) 5.99 Total 0.47(78) 5.72 II 0.31(175) 5.21 II CI (x) ±0.08(0.06) ±0 • 6 5 (0. 4 9 ) ±0.04(0.03) ±0. 51 (Q. 39) CI(n) ±0.71(0.54) ±5. 74 (4.36-) ±0.53(0.40) +6.75(5.13) - 20 - 46. Unfortunately, the final sample sizes are too small to permit detailed statistical testing for all pack~ges and combinations of packages, but certain interesting patterns do emerge from the analysis. The first of these is that in both seasons farmers t~king gro~ndnut packages have LL ratios higher and cropped areas lower than both the mean values and ·the values for all other packages.· These differences in relation to the mean values have the following levels of significance: 1970/71- LL ratio z = 1.27, P = 0.102; cropped area-z = 3.89, P = 0.001; 1971/72- LL ratio z = 0.82, P = 0 . .206; cropped area z = 2.37, P = 0.009. The highly significant results for cropped area indicate that farmers taking groundnut packages cultivate on average an area some 22 per cent smaller than that cropped by the average credit recipient. Those taking groundnut packages also have LL ratios some 27 per cent above the mean, though the statistical significance of this re~ult is considerably less than that for ~rapped area. 47. A second pattern, associated with that alr~ady discussed, is that hybrid maize growers have LI. ratios which are lower than the means for both years and also lower than the LL ratios for groundnuts. · In the case of hybrid maize, however, the difference between the LL. ratio of those taking this package and the mean ratio is more significa11t ~han in the case of groundnut~ (1970/71: z = 2.4li P = 0.008; 1971/72: z = 1.23, P = 0.092). It is particularly interesting to note that groundnut growers have mean LL ratios on average 50 per cent greater than hybrid maize growers; the signifi- cance level for the difference between means is 2.5 per cent in 1970/71 (t = 2.22) and 10 per cent in 1971/72 (t = 1.41). For one of the years, . 1970/71, hybrid maize growers have . holdings larger than the total mean, but in the following year ._those growing hybrid maize have holdingS! smaller than the mean and a third smaller than in 1970/71. This reduction in holding.size is perhaps a reflection of the more lenient policies regarding creaitworthiness which were in force in 1 0., 1 ,., ., - 2'1 - 48. A third pattern that emerges from the data is to be expected, i.e. that those farmers taking more than one package have in general larger holdings than those taking only a single package. 49. The results of this part of the analysis suggest quite strongly that those farmers.who take only groundnut packages on credit differ significantly from those who take other packages or combinations of packages. The smaller cropped area for these groundnut growers is associated with higher LL ratios, which suggests that the smaller holdings may be supporting the same number of people as those with relatively much larger cropped areas. Does this fact imply that farmers growing groundnuts on credit are closer to a subsistence margin than other farmers? If so, are such "subsistence" farmers less capable managers, i.e. to what extent is their performance in terms of crop yields and credit repayment less satisfactory than the performance of other farmers? Answers to these questions are sought in later discussions bf other analyses. 50. The Relationship Between Credit and Extension Visits-- If farmers are to make efficient and effective use of new and possibly unfamiliar inputs, it is essential that they receive instruction in the correct use of these inputs. The most relevant instruction is that given on the farm where the farmer is.actually employing the inputs, and t~e most common delivery system for such instruction is by visits of extension staff. Consequently an effort was made to ascertain to what extent extension staff concentrated their efforts, in terms of numbers of visits, on farmers who in a given season had . received program credit. As an undue concentration of extension effort on credit farmers could have unfavorable results in terms of income distribution, attention was also directed to the extent to which noncredit farmers were system- atically excluded from the extension program. I·' ! I . ! 51. The proc~du~e followed was to formulate the null hypothesis, .i.e. the frequency of extension;visits is unrelated to whether'or not a farmer has·credit. This hypothesis was tested for two years·by comparing the mean code for the number of ~xtension visits (0 = non~, 1 = 1-5, ~ ~ 6-10, ·3 = 11-20, 4 =· 20+) for those with credit with the mean code for the I ! . entire sample, including those with credit. The codes were ' . then used to estimate the number of actual extension visits · received. The codes for the credit sample are a reliable reflection of the number of visits received, but the codes · for the entire sample are biased upward by the inclusion of the credit.recipients. Thus the codes for·the sample will, overstate the number ot extension visits the noncredit farmer. can expect to receive relative to the number the credit farmer .CC).n expect to receive. The code for extens.ton visits. was also crosstabulated by type of credit package to determine whether extension staff tend to concentrate visits on fax:mers.taki~g a particular package. The.crosstabulation· ·~ is given: below:· Mean for.l970/71 ·~ean for. 1971/72 Type ·Of Approx. no. Approx.no.· cre.dl t package Code ·of visits Code of visits (l) Hybrid inaize 2.88(25) ·14.4 2:82···(22)' 14.1 (2)S~ntheti~ m~ize ·-·- 2.52 (52) 10.8 ( 3) .Groundnut·s 3.24 (17) 19.4 . 2.00 ·.· (11) s ~o · <4 >Ferti·l'izer -- ·2 ~.6o ·(10) .13.0 (5) 1 or 2 + 3. 3.67(18) 10.6 (6)' 1 + 4 3.10(10) . 18.6 2 ~is Ci2> . 10. 8. (7) 2: + 4. 2. 64 .· ('28) i3~2 CS)Other 2. 88 .(8) 14.4 2. 50 (40) 12.0 .. · Total credit . 3~ 17 (78') 19.0. . 2~45(175) 11.8 . subsamp1e CI (x) ·±0.41(0.31) . . ±0.28(0.21) . 'rot.aJ. sample 2. u~· (333'> 10.5 1. 69 (390) ·. 6'.8· 52; The mbst striking thing about the fig~res above is the dramatic. decline in the mean number of·visits received ~ro~ 1970/71 to 1971/72. For ~he credit sub~an~le this decline was 38 per cent, while for the total sample it was 35 per cent. For both g:toups the decline is un-doubtedly related to the fact that, for the .average program unit, the number ofcredit·recipients virtually doubled between the two ye~rs, so that on this basis a.fixed number of extension staff were required to deal with twice as many cr~dit farm~rs~ '. I ~his task appears to have been accomplished-by reducing the average number of visits to all farmers. '•' .· 53. This interpretation, however, needs to be suppl-. emented with other data from the two seasons. For example, i' the proportion of the total samples above riot receiving any v;tsits from extension staff rose from 24 per cent in 1970/71 to 38 per cent in 1971/72. Also the proportion of growers_ t !. tec~iving 10 or more visits dropped from 63 to 57.per cent I over the two seasons. Thus the rapid ·expansion of the credit program is associated both witl;l an increase in the number of f'arx,ners,who receive' no extension visits and.witha decrease in.the a"erage number of visits received by farmers'who are ·visit:ed. 54.. Credit farmers received 81 and. 74 per cent' more . . ' Visits than . . the! total sample for the . . - . two seasons respectively. A~ noted earlier, though, these pe~centages understate the rel~tiv~ ~ositions of credit and nOncredit f~rmers becau~e credit farmers constitute 23 and 35 per cent of the total . sample in 1970/71 and 1971/7~ respectively. 55. Where credit packages for the two years arE! COf!lf>arable, it can be seenthat only hybrid maize growers hav~ not experienced an appreciable decliJ:le ir. the_ mean . . . . number of·visits received. At the other extreme, the'average number of extension visits made to groundnut growers dropped by well over half. ·Farmers taking packages other than'ttiese - 24 - two experienced intermediate declines in the number of visits. 56. In regard to declining groundnut yields in LLDP, it is particularly interesting to note the decreasing number of extension visits being made to groundnut growers and the fairly constant number of visits being made to hybrid maize growers. In 1970/71, the code for number of visits for both groups of growers is not significantly different from the mean code, and the difference between the two groups is signifi- cant only at the 25 per cent level (t = 0.79). In 1971/72, however, all the levels of significance rise; groundnuts (z = 1.36, P = 0.087) and hybrid maize (z = 1.16, P = 0.123) have roughly the same levels of significance respectively below and above the mean, while the level of significance of the ~ifference between the two groups rises to 10 ~er cent (t=l~56). 57. A concluding remark stems from the fact that, though for the two seasons examined some 30 percent of farmers received no visits from extension staff, 70 per cent did receive visits and this represents a very significant improvement over the performance in areas outside the project, where 80 per cent of farmers receive no visits. The fact that the. average project farmer receives some eight to nine visits a year is also commendable, though ~onsideration of the distribution of these visits across the farming population would temper the commendation somewhat. Nevertheless, there is no evidence to suggest that extension staff concentrate exclusively on credit growers~ although they do appear to :. devote a consider~ble proportiori of: their effort to this group. 58.. The Relationship Between Credit and Measures of Collateral --In an additional effort to examine the "~quity~ of credit distribution, it was postulated that receipt of credit in the current season is independent of the level of .:.. 25 - '. material well-being, as measured by a "collateral index", b'f. the credit recipient. :The hypothe~ls was tested in a some- whaf diffeient mann~r for each of two seasons~ 1969/70 and · . 1970/71, and the results .are reported separately below. 59. For this year the cbllat~ral index. Iri 1969/70: was 4efin~d as the sum of the following items own~d (with weightirig factors indicated in parentheses)~· work oxen, ' ' ' ag-ricultural· implements, lorry (xlO),: bicycle, watch/clock, radio (x2), paraffin lamp/stove, bed; table, chairs and sewing machine (x4). Two sources.of credit were disting- uished: government credit provided under the scheme that . . ' . operated i~ the area before LLDP began operatio!1s and. non-, government credit; including credit from LLDP. Th~ sample consisted of 682 farmers, the results were. as follows:· Mean collaterai Mean collateral index index No government 3.68(656) No nongovernment. 3.78(649) credit credit Government credit 9. 62 (26) .Nongovernment 6.45 (33) recipient: . credit·. recipient Totcil· 3.90(682) 3.· 90 (682) cr <x> · ±0 • 52 (0. 3 9 ) +0 ,; 52 . ( 0 • 3 9 ): ~ ..· . CI (n) ±13.34(10.13) ' ±13 • 3 4 ( 10 .; 13) 6o;· The collateral·. index ranged from 0 .to. 40, with a mean of 3.90 for the entire sample. Neither index· for those· farmers with no credit was significantly different fromth(;!' mean index. The index for farmers with government.credit was 9.62, .or some two.and a half times the mean and highly ·sig~ifi~antly different from the mean (z. = 3.46, p = 0~001). r;I'he index for farmers with nOiigovernmentcredi.t was 6.45, or.' ' 1 . .7 .. times the mean and somewhat less sign.ifi~antly different fro~ the mean (z = 1.85, . P = 0.032). . Thus the. data for . . . 1969/70 _refute-the. hypothesis and suggest that, while the .. . . ·'. ' . LLDP cr:edit:'prograrn jtends i 'bosconcentra•eetbn. f:helreiatfvelyh t ~-: more .wealthyF fa:r:mers;;.c:i tr:Jdoeteso1t0nae±esser LElegF~e £fiarie£:fie. government ··creditJ p1mgrain;<whicht~dperatedh±ni ·'tfie:·ai:-ea 1'Jpfl:or- to the· project.-. , ., ,. · . : · "' "' ·-·· ·-·· ,, 61. 1970/71: The collateral index was defined by summing the number of head of livestock a farmer owned, the number_ of pieces of farm equipment (carts, plmlS-, ~tc •• ) owned, and including in the sum the weighted va:lue of owner- ship of the following: lorry (xlO); grocery store (xlO), maize mill (x5),. and bar (x8). The mean collateral index . for those growers with credit in 1970/71 was compared with themean for the entire sample, ·including the credit growers. Thus, to the extent that credit growers have higher indexes, the sample index will be biased upward and there will be a_ tendency_ to understate the differ-ence between credit growers and noncredit growers. Comparison is also made of the collateral index for those project credit_farme:r.:s who repaid all and less than ali their -loans~ The mean indexes are given .. below: LLDP credit Mean collateral farmers by % of Mean ·collateral All farmers index lbans repaid ·index Prqjec-t \ l·:·.;: ·. ,". ·.farmers 3.21(370) 100 per cent. ' 3.55(60) . ; '!. 'j• 'J Nonproject 4.68(437) Less than,,-~C?P;-: ,, 1 ,-l 6.• 28 ·(! 8 ) ,-farmers per cent .. !· 1 ··Total' 4.00(807) Total·· ! ... · 4.i8(78) .-- ·... I <'1) .l !.t!J~(;UCiil CJ ~5\JrVt':'''t-' ·. CI (x) - +0.62(0.47)' . . ' CI(x) . c:r tn> ··;: · , .,. · , _ . · ·:_ ···±2.21(1.66) __ +r'~L's2 82> (r4. ·cr (n.). ±17.62 (13.39) ( ~ 1, i \ ~ ·: ~ ·t 1 1 ~ \. l" :. ; r l\. L: . -i) rt n<J · J'' , '/;:-,.,H),; , ;;.1 {•' \-{, .; ,; /··(~ \J'<·~.{ r.-·~~--~ :.:.· 62. The_ mean index for the sample was 4. oo with a range . . .. l~·~·· ... )'': (H~.. t,~.~:~•. ~:~:~;I_._,.L\.. _LIJ. t:_.._.:-~>.~-- . of 0 to 71; the mean index for the.cred1.t gro;wers was 4.18 . . ·. ~ ··:"!• ~ r ••••• : ~:·~- ·::··1. . ·; :.t ·i 1·-~c~t~.. -~c•·}_ 1_:-~:.l--:~•. ::·:::~~.~ • • • • :.-. :- with a range of 0 to- 5L. · These differences in the. mean index ·. ·.. .:~. :i\~·:l ·.:-·;~·c· . ~i<.·-. :':r:·~·: ..~.~~; ..L . ~·~~t~ ....: . . . . '·. · ... are. not. significant, thus tne. hypothesis i.s supported. Two J - ------------------------------ •. -.' 'c. 27 - oth~r interesting observations stem ftom the analysis for this ; ·hypothesis. . --. ·First, the collateral index outside the . projec~· . . ! i~ significantly higher than that~side theproject (with . . . '· i I project boundaries as of 1970/71). The mean r10'nprojeCt index !. ' is 46 per bent higher than the mean project inde~ (4.68 as compared with 3. 21). Such differences pale 1 however,· in view . of the fact that the maximum .inde~ is some. 18. times. the mean . index arid that 95 per cent of all farmers have·a mean index of less than 17.4. The second observation stemming from tbe analysis is that fa·rmers. repaying . . less. than 100 per cent of their loans in 1970/71 have a collateral index significantly greater than the mean 1 i • e • .the m~terially better-off farmcarS appear to. be the greater credit risks. .Note. s_hould be mad~ of the fact that the weights employed in· deriving the index would 'tendto give a higher index·score to those relatively. wealthy individuals whose commitment to agriculture is weakes.t .. 6 3. · As noted in the introduction to this secticYn, · . certairl of. the hypotheses... identified could. not be tested. . because of problems· .with the data ~equired to car:J:Y out th.e · . analysis ' ..... These hypotheses . .. ·arl.d. the nature . of the, problems ·encountered are discussed.briefly below. 64~.. The fir~t hypothesis . . not testable is. that the proportion . . of f~rmer$ applyirtg .for . br~dit i~ ide~t~cal i~, . . ~rea~ ~hi~h receive both greater-thari-average a~d less-than- average amounts_of credit on a per-farmer bc;isis. ·The data· were tp·be collected in the field in .1973 through~a survey of unit. center tecords. When the records were examined, however~ it 'lias found'that the unit centers tetain no information on those farmers who apply.for credit but who do not receive it. · It was t.herefore impossible .·to test this ._,. · hypothesis, .and it would likewise be impossible to. test a:ther. hypotheses which h.ing~·upon the distinction -between credit ap~iicants and ~~~d~t reci~ients~ · ··:' ·. ,,· ... ·I'. I ·"' . ·:• ..... ·:.: ......,: .... ~ 28 -. Mention was made in the ihtroductioh to .this s~ctiori. of the credit rating system. It was intended to examine the proposition ~h~t managerial ability is evenly distributed across the project population, i.e. there are aquivalent.· proportions of low, medium anCl high cr~<H t ·ratings in all- project units. Examin~tion of the data ~ugges~ed that the creditworthiriess rating system was ne~er applied as it was meant .to b~, for the ratings. were not ~ecorded by evaluation staff on the coding forms • . Inspection of unit. center records confirmed that the missing data were not a result of the failure of evaluation staff to collect the-information but rather a consequence of extension staff's n6~ h~ving produced ratings for most .farmers. In both 1970/71 and 1971/72 I some i 5Sper cent of project farmers included· in the samples had 1. not been, rated. In some units no. fa-rmers had beeri rated. The.dCita· reflect a patchy appllc~tion.of the cre4itworthiness rating system and differences in subj·ective asses.sments on thepe1rt of extension. staff. The small number of units rated ea,ch.' year. do~s. not permit' the initial proposition to be examined in any meaningful sense. ' .. 6:6 • .It was also proposed to test the hypothesis that· .· ' 0 . . receipt of credit in the current . season· is .independent of . t·he level of cash scil.es in th~ 'preceding seasC>n:~ This. I ,. • hypothesis could not be tested because a new sampling frame· is s~lected - ' . for each year's ·surveys at LLDP., .. so that the· • . . available .Ciata on cash.sales for one year's ~ample cannot. be related to what happens in the following season· .to the . same group. (This hypothesis . could be t.ested with the farm · management dat~:now at LL[)P and will be tested ifthe pro~osed ar1alysis of this d.ata is approved:·) ··.·: :. ,· 67. As a general procedural point, there are strong arg~ents for not selecting an_entirely new sample of farmers every year for the on..;going yiE!ld surveys. ·some · P.rop.ort~on of the most· recerit sample shotild selected for be ' ~ ' - 29 - re-surveying on an annual basi~ in order (1) to provide continuity and permit assessments of change over time within the same sample and (2) to provide a more realistic assess-. ment of the magnitude and importance of inter- and intra- ~Se~sonal yield variations in t;he area~ }\s farm management surveys tend to be expensive both in terms of time and money, ther~ would appear to be substJhfial e6onomies of scale to '·. be realized by collecting a l'imited amount of farm management data in the annual yield surveys as suggested in objective (1) above. It should, for example, be possible to_collect in the annual yield surveys a considerable amount of data. on'th~ use and effect of production credit. This is a point that is elaborated at several other places in this report. 68. A general comment is that perhaps too much attention has been given to the concern that the survey process will alter the sample in a significant way, and a totally new sample is still drawn for every season's surveys. Such ' concern .is no doubt valid where surveying is carried out in a larget~ static socioeconomic system, but surely in a dynamic situation such as that existing at Lilongwe, where fa,;rmers are subjected to barrages of change stimuli, the il'l:f luence of s.everal hours' interviewing a year could be dis.counted and more valuable inform'ation obtained. - 30 SECTION III CREDIT PERFORMANCE AND DEFAULTS 69. Having looked at the apparent determinants of the distribution of seasonal credit in Section II, this section wilL go on to attempt an evaluation of credit performance. The criterion for judging credit performance of farmers is whether or not they default on seasonal loans. There are other, arguably more valid criteria by which 'credit perform- a~ce might be assessed~ and looking at default rate~ rather tends to accentuate aspects related to the financial viability of the credit program. To a degree this emphasis has been avoided in the analysis by examining the factors which appear to influence defaulting, in an effort to get c~oser to a more meaningful interpretation of credit perform- ance. 70. . Defaults and Size of Holding-- For purposes of· the analysis a borrower becomes a defaulter if,. on August 31st of a given year, he has not repaid the total amount due on his loan from the preceding growing season. This date was. chbs'en because LLDP uses it to determine arrears and because penalty charges are meant to be imposed a~ter August 31st (though such charges are frequently not made in practice). Technically, then, the analysis is identifying farmers in arrears ia~her than defaulters~ though a proportion of the former will eventually become defaulters. For purposes of . deriving project policy, the analysis is ·most usefully regarded as an examination of factors associated with late .. 31 '. ,,. . . . repayment and nonrepayment of loans and how defaulters (tho~e in arrears) differ from .farmers who repay their lc;>ans on time. i : ! I · 71. . · The first. hypo~hesis tested in this section is. that I .defaults· are unaffected by size of holding. ·Size of holdi~g is measured here by cropped area, and defaults are measured by. the percentage of the total 'amount due ~epaid: 't,hrough Apgust 31st. The data used· are from the credit- subsamples o~ the 1970/71 and 19jl/72 surveys. The followih~ cross- tabulation table was gen~rated: Mean. cropped area in acres · P~rc~ntage r~paid 1970/71 1971/72. . . ~ 100.· 5.82(60) 5. 20 (116) Less than, 100 5. 37 (18) '5.25 (59) 51 99 5. 66 (10) 6.05(28) 0 50 .5~00 (8) 4.53(31) i, "' I . 'TOTAL 5.72(78) : . .5.21(175) '! I l . CI (x) '±0'. 65 (0. 4 9) . ±0. 51 (0. 39) ' CI (n) · '±5.91,(4.44)· ±6.71(5.10) :: I' I • F9r both 1976/71 and 1971/72 there is. no significarit . ' difference . {n.cropped area between those who have repaid ~11 their loan.· an¢1. those who have repaid less than the ent·irc loan, npr is t~~~e a·significant differe~ce betw~en the~eans for eac~.of · th~se grcmps. and the .total mean. ·Thus the hypothesis that · defaults are un~ffected by size of holdings is supported ~Y this. part of the analysis. .;'' 72. The only pattern that emerges from the data for .both · · seasons is that borrowers with the lowest repayment rates'".· I. (0-50 per cent). also have the smallest cropp~d areas. For 1970/71, the cropped area of 5.00 acres :for this group is q~it~ sigqifica~tly ~ifferent from th~ mean (z = 1~56, p = 0.059)_, while for 1971/72 the cropped area of 4. 53 acr.es. I I - 32,~ is highly significantly different from the mean (z = 2.19, P ~·0.014)~ Ther~fore, ~hile default per se is not~affected · by holding size, the proportion of the loan·not·repaid by defaulters is st~ongly, but inversely, related to holding size. Further comment wili be made on this:phenomenon later. 73. Default Rate in Relation to Type of Credit Package~- It was thought to be particularly important to test the hypothesis that there is no significant difference in the default rate among fa~mers growing maize, groundnuts or tobacco on credit. Such testing, it was felt, could be indicative as to whether certain types of credit packages are used more efficiently.than others and .whether the . returns to use of a given package are high enough to permit full . repayment of the loan while maintaining previous·Ieyels of. consumption. The v_ariables employed are.tlle.typE;!:of credit package taken--maize, ·groundnuts, some combination of maize and groundnuts, and other (principally ft;!rtilizer, thought . . to be applied largely to tobacco)--and the.percentage of the total amount due repaid on August 31st. The hypothes·is was tested for the two seasonsl970/71 and 1~71/72. 74. One component of the analysis was the .construction· of correlation matrices ·relating the perce'ntage repaid to a number of other variables, for each .season separat.ely and for both . seasons together. Iri all matrices, t,he percentage . ~ . ·. repaid is negatively but weakly correlated with the nwriber of'credit pac:kages taken,.with the value of the pac:kage, and withthe complexity of the packages. The last is measured • I • • • • by the number of "traits" inherent in the package taken: if five packages are taken and all are for the same kind of maize, this is considered a one-trait package;' ~f 'two or mor'e packages are taken, all of which are either.maize or ground- nuts, this is a two-trait package. Similarly I. e~ch additional . crop or technology represented in a. given combination of.· packages.is another distinctive.trait. The percentage of the i' amount due repaid is'also positively but weakly correlated with both cropped area and total holding size. Because of the. low absolute values of all the correlation coefficients~ .. the mAtrices aie not reproduced here. 75.. The relationship between' type of credit package and percentage of amount due repaid is shown in the following crosstabulation: Mean 12ercentag:e of loan value re12aid Type of credit :eackase 197oL71- 1971/72 (!)Maize 94 (35). 86(124) (2)Groundnuts 92(17) 62.(18) (3)Combiriation of 1 & 2 7 9 (21) 69 (23) l SUBTOTAL 89(73) 81(165) : . CI (x) +7 (5) ±7 (5) CI(n) :!;61(46) +84 (64) (4) Other 100 (5) 90 (10) . TOTAL* ·. 90'(78) . 82(175) 76~ Two patterns are revealed in the data· for both ,'II season_s: farmers·taking only maize packages r~pay'(l higher th~n,average percentage (by six per cent) of their. loans i while'fa:r;mers taking combinations of packages consistently repay a lower than average percentage (by13 per cent). _ These .,results have the following levels of significance: Ma!ze--1970/71: z ~· 1.47t· P = O.b71; 1971/72: z = 1.83, P = 0.034; Combinations-~1970/71: z = 1.46, P = 0~072; 19?~/72: z - 1.58, P = 0.057~ The improvement in 'signifi~ cance' levels between the two years probably stems both from the increase in sample size and from a seasonal effect~ In .. . . . the case of farmers taking groundnut package!;; the percent,;_ acje repaid.iri 1970/71 is slightly above the meart-~by three * ;.,Means. are weighted averages · .: .. . . ., . - 3A.· -· . per cent--but . this diffSrence is ndt significilnt. In the T971/72 season, ·however, the average farmer growing ground- nuts·~lone on credit repaid only 62 per c~rit of his lo~n-~ some •. 2 3 per cent less t,han ave rag~ and signLfic~ntly less than the mean (z = 2.16, P = 0.015). Finally, tho_ugh the . . . . . .. . significance of the difference cannot be measured with the results available, it is worth noting that farw.ers taking ·"other" packages--ma.inly. fertilizers applied to tobacco-~ ' hav~ the.highest.repaym~nt rates for both seasons. 77. A further breakdown of the 1971/72 data on maize packages reveals that farmers taking hybrid packages only (n = 34) repaid on average 81 per cent of the.l,.r loans, farmers taking synthetic packages only (n =. 8l).repaid 87 . . .per cent;. and farmers taking a combination of,hybrid and synthetic (n = 9) repaid on average 93 per.cent. Signifi- cance levels have not been calculated for this preakdown. 78. In summary~ . the results indicate that for.· each season considered separately repayment rat.es are ·highest· for maize and other--mainly fertilizer--packages and lowest for combinations of packages containing ma.ize and groundnuts... Additionally, on average fot the two seaso~s, the .repayment ' . ' ·' . rate.for groundnut packages is well below average. Further discussion of these results i~ reserved for.the concluding section. 79,~ . Defaults Related to Crop Yields and Productiori-- On .a priori grounds, . .it would be logical t~ assume·. that .agriCultural performance measured in terms of crop yields and total production "{OUld be the single most important determinant of credit repayment rates. Accordingly the nypothesis was formulated that default rates.are'independent of yields of maize and groundnuts, of total proc1uction of . each .of th.es~ . crops. and . of combined . product ibn . of both crops •.. '•. Productio~ of each crop was estimated.by m\,lltiplying the· yield recorded on a one-ninetieth of·a:n·acre subpl~t by the -'~ ;-~ - 35 - total area planted to the crop. While this method is.not necessarily completely reliable, it is the only way an estimate of production could be obtained. The hypothesis was tested using the credit subsamples for 1970/71 and 1971/72, and the results are given in Table 1. 80. The results obtained are generally disappointing for, although apparently important differences are detected' in the variables used, none of the differences is significant at even the 10 per cent level. Thus the hypothesis cannot be rejected, even though intuitively one would think that it should be. 81. Some broad patterns are revealed by the analysis, but in the absence of significance their interpretation is · · largely a matter of guesswork and thus all the more hazardous. For example, in both seasons farmers repaying 100 per cent of their loans have lower yields and prod~ction of both crops than farmers repaying less than 100 per cent. One possible reason for this phenomenon is that farmers·with higher yields . . . . and production make greater use of illicit marketing channels and thus escape repaying all or part of their loans, or that these farmers store surplus production to sell locally when prices rise after the post-harvest period and are thus late in repaying their loans. The fact that fairly large numbers of farmers repay their loans in cash, presumably obtained through nonofficial sales, tends to support the latter part of this conjecture. 82. Anothermteresting observation comes from the data in Table 1. Comparing the total samples for the two seasons, it will be noted that while the average maize yield rose by 15 per cent from the one year to the other, average total . production rose by le.ss than one per cent. This fact is consistent with two interpretations: first--farmers tend to reduce their maize acre~ge as yields rise, possibl~ to plant a more lucrative crop, i.e. tobacco; and second--the credit 36· -· .•) j Table 1: R~payment fu1tes Related to Credit Packag~s, Crop Yields and Production, 1970/71 and 1971/72* Percentage Type of Mean for m.:~ize Mean for c;roundnuts Nean total re,eaid . credit 12ackage Yield Production Yield Production production (lb/ac) (lb) (lb/ac) (lb) (lb} 1970/71: 100 Mai~e (29} 1,581 6,471 433 705 7,176 Groundnut (15) 1,065 3,152 389 427 3,579 Combination (11} 3,009 10,976 503 782 11,758 Subtotal (55} 1,726 6,467 435 675 7,112 CI (5() +407 ;!:1,919 :!: 95 .:!:176 .:!:1,927 Other (5) 845 2,428 334 462 2,890 Less than 100 Maize (6) 2,212 8,108 232 445 8,553 Groundnut (2) 1,435 3,731 377 151 3,881 Combination (10} 1,402 4,637 461 482 5,119 Subtotal (18)' 1,676 5,694 376 433 6,126 CI <x> :!;558 :!;2,387 ±158 :!:;214 .:!:2,439 Total (78) 1,658 6,029 415 584 CI <x> ±319 :!;1,473 ±76 ±137 -- 1971/72: 100 Maize (92) 1,974 5,753 326 470 6,222 Groundnut (8} 892 3,328 345 525 3,853 Combination . (9) 865 2,935 417 529 3,463 Subtotal (108) 1,811 5,361 335 478. 5,8:39 CI (Xl ±274 :!;952 :!63 :!;109 ;t982 Other (8} 1,940 7,120 403 614 7,733 Less than 100 Maize (32) 2,191 8,196 307 446 8,642 Groundnut (11) 1,605 ~.066 444 388 ' 5,45'4 Combination (14) 2',:090 7,128 478 779 7,907 Subtotal (57) 2,053 7, 330 375 516 7,846 CI (Xl ±433 :!;1,825 ;!:100 :!:203 ±1,896 Other (2) 2,107 4, 729. 114 69 4,798 Total (175) 1,899 6,075 349 492 CI <i<> :!;223 :!;854 !51 ±95 -- *The confidence intervals 9iven for the me~ns are those fer 95 per cent. - 37-- program in 1971/72 was extended to farmers with s~al1er holdings an~ smaller holdings.cultivated a smaller area of' maize.; The average yield of groundnuts, on the other hand, declined by 16 per cent between the two seasons and average total production fell by an identical amount.. This fact . ' ' . . tends to suggest that the average area :per holding.p1anted. to groundnuts remained.c6nstaht from 1970/71 tO 1~71/72 . . Finally, there is the curious fact that mean .total production for those repaying 100 per cent fell by 12 per cent from one year to the next, while total production for th.ose farmers repaying less than 100 per cent rose by 26 per cent (obtained· by taking weighted means of subtotals plus "others"). In I other words, full loan repayment is positively associated i with total production in.l970/71, as would be expected, but I i~ negatively associated with total production in 1971/72. 83. Defaults Related to Marketing Channel~ Utilized--_ 1: In a final· effort to isolate the determinant:s. of repa:Yment .. rates, it we1s hypothesized that defaults are independent of the marketing channels utilized. Data were available for the ! 1971/72 season which indicated the marketing channeis farm~~s intended-to use for their maize and grouridri.utc!:'ops. In the original data, each channel was given a code, ranging from . zerqfor no anticipated commercial dispOsal of crops to three a~d ~our, for malzeand gtouhdnuts respectively, if all sales were to be made to ADMARC, with intermediate codes represent- ing barter and other local and private sa].es. As the codes can l;le employed as a crude hierarchy with the higher.nuinbers signifying a greater orientation towa~d co~erc.ial and official marketing outlets, the percentage of creQ.it repaid was arrayed with: the mea~ code values, with values_ normalized ~o p~rmit comparison between t.he two d-ifferent ~~ales. The·.. results are shown below: - ;J8 .:.. . - . Mean code value ·Percentage of credit repaid Groundnuts Maize ~oo per cent (116) 2.51 2.45 Less than 100 per cent (59) 0.97 2.08 Total (175) 2.33 2.33 CI (x) ±0.37(0.28) ±0. 26 (0. 20) 84. The mean code value for those repaying all their loans is higher by 27 per cent for gro1.1ndnuts and 18 per cent fo.r inaize than the same value for farmers not repaying all .their loans, though only· the difference for grou11dnuts is significant at five per cent (t = 1 •. 77). In any event, the codes represent intentions rather than actuality So.that the .results can only suggest that farmers with better repayment records may have a somewhat stronger inclination to use comme~cial and/or official marketing outlets than farmers with pooJ::e.t.repayment performance . ., SECTION. IV EXTENSION .VISITS AND ADOPTION. .. 85. ' A subset of the hypotheses related to crop yields dealt with the interrelationships between extension visits and yields and adoption of new or· improved practices and in- .· ' ' . . . puts~ Because of strong interest· i:r;t the LLDP extension program, evidence as to its impact and effectiveness is presented sep~rately.in this .section. 86. '. Evidence of an Extension Impact on Yields-.,.In the annual yield surveys an item is included on the schedule for recording the number of extension visits tl'le .farmers remember · receiving. during the cropping season in question. .The number of visits is recorded as a range rather than.as a single. number, and the categories employed are as follows: o. (no v.i~;;its) ,, 1 .(1...;.5 ·visits), 2 (6-10 visits), 3 (11~.20 visits) ' ' . . . arid 4. (more than 20 visits.) • As a ·first: step. in this· part of the analysis, it was assumed th~t. there is a more~·or...;less l~~ear relationship between yields of maize and gr~undnuts an.d the number of extension visits received. To test this. assumption a multiple regression model was set up and.run with crop yield .as the. dependent. . variable and numbers of . visits as the independent. (dummy) variables. · ·Runs were made fqr ,two seasons, 1970/71 and 1971/72, with the total sample-- . il'lcluding both project and nonproj ec.t farme;rs.:..-and with a . · subsamp_le · qorisisting of only LLQP farmers~· . The . results are · · sqown· in Table 2. ; 87 •. ·.·~ ·Because the independent variables. are dummies, the· ex,peqted._yield for any given number ();f extension visits is. . . . - 4o ~ Table. 2 Relationship Between Number of Exte-nsion Visits···. and Yield~ of Maize and Groundnut~, 1970/71 ~nd 197t/72· C6efficiertts for no. of visits Eqtiation statistics constant 0 1_;5. 6-10 11-20 · .. R2 SE Maize 1970/71: 1,381 -538 -466 -21a -74 0.05 944 4.A7 LLDP (91) (146) (156) (221) (197) (4;278) Total 1,397 -424 -303 -286 -48 0.04 837 6.93 (7 2) (86). (106) (179) (157) (4;642) 1971/72: LLDP· 1,923 -951 -444 77. 160 0.09 1,218 7.15 (127) (211) (211) (231) . (210) (4;284) Total. 1~925 -926 -529 -106 103 0 •. 14 1,010 26~99 (105) (118) (146) (176) (172) (4 ;'641) 'i Grouildnuts 1970/71~ ss1 -364 -150 -234 -178· 0.28 255 23.87- LLDP (28) (38) (57) (7 9) (74) (4; 236> I I Total 5.61 ~209 -75 -178 ~148 p.ri6 ' 310 10.91 (39) (34) (46) (76) (73) (4; 604) 1971/72.: LLDJ?,· 394 -240 78 84 83 Q.19 293 21.31 (~l) (40) (53) (59). (57) (4;345) Total. . '401 -f66 ·91 85 76 0.12 ·. 281 27.88 (30) (32) . (43) .(52) (54) (4;753) .... NOTE.: Figures in parentheses under the coefficients are the ~-~andard errors while·those under the p-~alues are the appropriate degrees of . ·~· j. freedom. - 41 - simply the sum of the constant and the coefficient for that···· number of visits. Thus, for example, LLDP.farmers rece1v1ng no visits in 1971/72 would have had on average a ~aize yield of 972 lbs/acre. For maize, all the coefficients for 1-5 visits are significant at five per cent but none of the. coef:ficients for more than five visits is significant. even·· at ten per cent. The response in terms of· an increase in yie.ld to a higher level o.f extension exposure is greater half the time for the LLDP subsample and the other half for the . . total sample (or, by inference, for the nonLLDP farmers). The maize yield response to.extension visits is also very· much stronger and clearer· in 1971/72 than in the previous year. I: 88 •.· For groundnuts, only half the co_efficier1ts for 1-5 v:isits are significant at five per cent. None, of the 1971/7~ .. coefficients for more than five visits is significant at ten per cent, but all ·such co~fficients for 1970/71 are signifi- .c:;ant.at five per cent. : . . - . . . . ~ . . 89. . All equations are .significant beyo.nd the one per cent level even though the independent variables a$ defined 2 explain only a limited amount of the vari~nce~ ·The mean R ~or the f_our maize equations is just o. 08 ~hile ·that for the groundnut equations is 0.16. 90. · In. summary, · for both crops in both. sea~c;ms there is strong evidence of. a positive yield respohse.for.the first !ewvisits (1-5) a farmer_receives. 'Beyond_five visits the . resu,].ts are equivocaL For maize, the response. remains positive a~d strong (though not significant)· .in both years ali the .way up to 20 visits. For groundn':lts I the response .. I .to 6-10 visits·is negative in both seasons. (significant_ only ··in 1970/71); for more than 10 visits, the response is slightly. negative.inone season (1971/72) andsligbtly positive in,the other (1970/7i). - 42 - 91~. Essentially, however, the results above are at one ' . remove from the core of the issue, since in using extension visits as·the independent variables one is already using proxies for what is more immediately interesting, i.e. how extension visits and yields are linked thtough:±he adoption or nonadoption. of yield-increasing inputs and practices. This chain of causality will be considered explicitly below. 92. Adoption, Yields and Extension Visits--Tables 3 and 4 relate extension visits to groundnut yields through explicit inputs and practices, while Tables 5 and 6 do the same for maize. Each input/practice is considered separately by crop below. A. Groundnuts (1) Sulphur. Sulphur is recommended to control leafspqt disease in groundnuts. In 1970/71 five per cent of a sample of 759 used sulphur; nine per cent of those visited used it, while less than one per cent of those not visited used it. ~orne 95 per cent of users had visits, while only 50 per cent of nonusers had visits. In 1971/72 seven per cent of a sample of 725 used sulphur; 13 per cent of those visited used it, while less than one pei cent of those not visited used it. Some 92 per cent of users had visits, while only 43 per cent ·of nonusers had visits. There is nb significant difference between yields of sulphur-users and nonusers, though this fact may stern from either the relatively small size of the samples or from a level of application lower than that suggested by research result~. (2) Variety. Chalimbana is the.variety included in LLDP groundnut credit packages. The proportion of those growing each variety and receiving extension visits is as follows: 1970/71--Kalisere 50, Chalimbana 57 and mixtures 49 per c~nt;_ 1971/72-- ·Kalisere 27, Chalimbana 48 and mixtures 46 per cent. Kalisere, with fewer extension visits, yields consistently less than the average in both seasons. (P = 0.004 and 0.001). In 1970/71, neither Chali~- bana nor mixtures have yields significantly different from the mean; however, in 1971/72 Chalimbana yields more than the mean (P = 0.013) and mixtures yield less than the mean (P = 0.056). It is interesting to note that, on average for all varieties· in 1970/ 71, those farmers with no extension visits had .the highest yields. .... -.. ·-. -· ...... -----·-- - ========--==-~ Table 3: Interrelationships Among Extension Visits~ Agricultural Practices and Inputs, and Groundnut Yields, 1970/71* ·~·· r Confidenc-e. interval Number of Extension Visits 'None . 6-10 11-20 20+ Total on the mean: 99%(95%) (Yields in lb/ ac) ULPHUR Users 295 (2) 497 (2) 199 (1) 428 (S) 520 (28) 486 (38) _!128(96) Nonusers 520(364) 490.(140) . 362(32) 323(42) 502(143) 492(721) .!,33(25) Total 519(366) 490(142) 357(33) 334(47) 505(171) 492(759) +32(24). ARIETY- Ka1isere 434 (22) 190 (8) 434 (2) 425 (2) . 399 (10) 381 (41) .+112(84) Chalimban.;!. 534"(119) 609 (49) 420(15) 246(16) 452 (75) 502(274) +34(26) ~ W. Mixture 520(222) 447 (83) 288(16) 377 (28) 564 (86) 497(435) .:!,39.(30) I Total 519(366) 490(142) 357 (33) 334(47) 505(171) 492(759) +32(24) OBACCO Growers : 505(170) 538 (54) .. 387. (8) 414(15) 558 (60) 514(307) '!:_57 (44) ·. Nongrowers · · 313(332) . . 460 {88) 348(2'5) 297(32} 477(111) 367(588) '!:_3.5(27) Total 378(502) ·.490(142) 357(33) 334(47) 505(171) 417(895) - . +31(24) . Figur_es in parentheses following the. yield figures· are the number of observations in each group. Table 4: Interrelationships Among Extension Visits, Agricultural Practices and Inputs, and Groundnut Yields, 1971/72* Confidence interval Number of Extension Visits None 1-5 6-10 ll-20 20+ Total · on the mean: 99%(957,) (Yields in lb/ac) LPRUR Users 195 (2) 455 (13) 403 (8} 408 (6) 409 (8) 412 (37) +137 (103) Nonusers 360(347) 429 (86) 445 (47) 475(37) 381 (89) 387(606) +29 (22) Tol;a1 344(349) 432 (99) 439(55) 466(43) 383 (97) 391(643) +29 (22) \.RIETY .l:>o Kalisere 220 (33) 163 (4) 0 (1) 214 (4) 82 (3) 201 (45) +75 (56) oi:Oo Chalimbana 400(230) 435 (77) 399(46) 463(32) 358 (61) 405(446) +37 (28) Mixture 310(125) 423 (31) 503(18) 435 (14) 401 (43) 365(231) +44 (33) Total 355(390) 421(113) 421(65) 435(50) 367 (107) 379(725) +28 (21) BACCO Growers 271(204) 499 (42). 538(23) 495(17) 425 (49) 352 (335) +44 (33) N~Jngrowers 213 (333) 383 (57) 368(32) 447 (26) 340 (48) 266(501) +32 (25) - Total 235(542) 432 (99) 439(55) 466(43) 383 (97) 301(836) +2.6 (20) Figures ·in parentheses following the yield figures are the number of observations in each group. .. / ; - 45 - (3) Tobacco growing. The growing- of tobacco has been inciuded here because ot the belief that tobacco-gro~ers constitute a very special subset · ·- ·. of the LLDP population. - The evidence shown in Tables 3 and 4 strongly supports thi~ belief for,- although they do not receive a disproportionately large share of extension visits, toba~co-growers nonetheless obtain groundnut yields substantially greater than nongrowers. The difference in the ~ean yield obtained by growers and nongrowers is significant at a probability level of 0.005 (1~70/ 71: t =. 5.88; 1971/72: ·_t = 3.99). Not only is the niean yield of grourtdnuts .obtained by tobacco-. growers greater but also for every level of exten-_. sion visits for both years growers achieve higher yields than nongrowers. Bo Maize (1) Fertilizer. In 1970/71, 16 per cent of a sample of 856 used fertilizer; 21 per·cent of those visited used it, while 10 per cent of those not visited used it. Some 69 per cent of users were visited, but only SO per cent- of nonus'ers received visits. In 1~11/72, there was a cons{d- erable increase in fertilizer use and-there is !?tronger evidence of an extension irepa,ct. In this season, 26 per cent of a sample of 8R6 used fertilizer; 52 per cent of those: visited used it,_ while nine per cent of those not visifed used it. Among all fertilizer-users, 84 per cerit'were visited but only 34 per cent of nonusers.received visits. Using the chi-square tef:!t indicates that the relationship between fertilizer-use and . extension visits is significant at a probability . level of 0.001. for both seasons.. The maize yields achieved by fe~tilizer-users are significantly higher (P 'seasons. = 0.005)- than yiel(js of nqnusers. in both (2) Manure. The use of manure 'appears to_ be declining as the use of chemical fel:'tilizer is · increasing. Seven per cent of growers used manure· in 1970/71 and attained anaverage yield 56 per cent above that of nonusers (significant at a probability level of o.oo5). Only five per eent of growe~s used manure in 1971/72; 'they.achieved yields 21 per cent higher than_those.of nonusers but in this season_the differertc~ in yields i~ not particularly significant. The _P~ajority of >. · farmers using manure receive .no. or' less. than six: . visits, but on.averagt;;! 64 per cent of .. users were visited in the two seaso'ns~ (3) Variety-~ ·In 1970/71, .21 per cent of a sample of 856 planted improved varieties (IVs); Table 5: Interrelationships Among·Extension Visits, Agricultural P-ractices and Inputs, and Maize Yields, 1970/71* Confidence interval Number of Extension Visits None 1-5 6-10 11-20 20+ Total on the means: ·99%(95%) (Yields in 1b/ ac) ...;._.,;;;:..:_---.~ FERTILIZER Users 1,151 (42) -1,718 (19) 1,810 (5) 1,419(14) 1,695 (55) 1,505(135) +246(187) Nonusers 939(363) 878(151) 903(33)_ 1,045(41) 872 (133) 918 (721) ~ . +68 (51) Total 961(405) 972(170) 1,023(38) 1,140(55) 1,113(188) 1,011(856) +71 (54) MANURE Users 1,56,5 (24) 1,369 (13) 630 (3) 1,449 (3) 1, 705 (19) 1,516 '(62) +345(261) Nonusers 923(381) 939(157) 1,056(35) 1,122(52) 1,046(169) 971(794) +71 -(54) Total 961(405) 972(170} 1,023(38) 1,140(55) 1,113(188) 1,011(856) +71 (54) VARIETY Traditional 963(371) 873 (136). 961 (32) 1,079 (32) 842(107) 931(678) +70 (54) Synthetic 1,075 (14) 1,133. (24) 452 (2) 1,091 (9) 1,039 (39) 1,063 (88) +226(171)- H632 1,382 (6) 2,732 (5) 2, 4i2 (i). 1;491 (6) . 2,082. (32) 2,000 (50) +511 (384) Bingo 577 {13) 1~139 (5) 1,579 (3) 1,134 (7) 1,193· (10) 995 (38) +305(228) Total 961(405) . 972(i70)- 1,023 (38) 1,140(55). 1 '11-3 (188) 1,011 (856) +71 (54) TOBACCO· Growers i,064(161) 1,142 (57) 1~422 (9) 1,328 (17) 1,261 (63) 1,144(307) +128 (97) Nongrowers 770(283) 886(113) 899(29) 1,056(38) 1,038(125) 874(588) +83 (63) Total -877(444) 972 (170) 1,023(38) 1,140(55) 1,113(188) .. "967(895) . -- +83 ( 63) . *·Figures -in parentheses following the yield figures are .the number of observations in each group. ------;------.--·---.- - - - - - - - - - - - - - Table 6: Interrelationships Among Extension Visits, Agricultural Practices and Inputs, and Maize Yields, 1971/72* Confidence interval Number of Extension Visits :.:None 1,...5 6-10 11-20 20+ Total on the mean: 99% (95%) (Yields in lb/ac) .. . FERTILIZER Users. 1,677 (35) 1,645 (46) 7,292(30) 2,201(38) 2,016 (74) 1,955(223) +234(178) Nonusers 981(348) 1,160 (74) 1,349(30) 1,356(28) 1,296 (47) 1,075(529) +90 (69) Total . 1,045(383) 1,346(120) 1,821(60) . 1,843 (66) 1,736(121) 1,337 (750) +101 (77) MANURE. : I : Users 1,338 (12) 1,820 (11) 1,984 (4) 1,383. (2) 1,566 (7) 1,604 (36) +476(356) ~ Nonusers 1,035(371) 1,298 (109) 1~809(56). 1,857(64) 1, 747 (114) 1.323(714) +104 (79) ....... I Total 1,045 (383) 1,346(120) 1 ,821(60) 1,843(66) 1,736(121) t,337 (750) +101 (77). VARIETY Traditional 1,011(351) 1,164 (80) '1,460(32) 1,311 (29) 1,246 (56). 1,099(548) +89 (68) - Synthetic .. 1,389 (19) - 1 ;660 (28) 2;131(19) 2,073(26) 1,913 (47) 1,850(139) +277 (210) H632 -·. ·2,370 .. (3) 1 ~-964 '(9) : 2, 568 (7) 3:079 (9) 2,913 (l6) 2,661 (44) +712"(533) . . Bingo 1,192 (10). 1,412 . (3) 2,023 (2) 984 (2)- 1,892 (2) 1,366 (19) +549(401) Total . -~.045(383) 1,346(120) 1,821(60) 1,843(66) 1;736(121) . '1,337 (750) +101 (77) TOBACCO- Growers _1,164(188) 1,386 (55) 1,895(29) 1,577(26) . 11717 (59) 1,379(357) +257(195) Nongrowers · . 719(298) 933 (82}. 1,473(43) 1,453(55) 1,365, (82) 975(560) +101 .. (77) - Total·-~ 89-~ (486) 1 '115(137) 1,643(72) 1,4-93 (81) ;.1, 512 (141) 1 ~ 132 (917) .!_119 '(90) . * Figures in p~rentheses f6llowin:g the yield figures are the_ number of observations in each· group. · - '48 - 32 per cent of those visited used IVs, while only eight per cent.of those not visited used these varieties. Some 81 per cent of users received visits, but only 45 per cent of nonusers had vis~ts. As with fertilizer, there was an expansion in the use of IVs in 1971/72: 27 per cent of a sample pf 750 planted IVs; 46 per cent of those visi.ted used IVs, while orily eight per cent ofthose.not visited used IVs. In this season 84~p~r cent of user~ i were visited, but only 36 per cent of nonusers were visited. While the. yields of traditional maize are consistently significantly lower than the mean yield, the yield of only one improved variety-- H632--is consistently significantly greater than the mean yield (both at P = 0.01) ~. ·~ (4) Tobacco growing. As already noted, tobacco-· growers do not receive a disproportionately .·largie number of extension visits, i.e. the observed nwnber of visits i·s not significantly different from the number expected (x2 = 2.01 for 1970/7l.arid ·1. 53 for 1971/72) • Yet for maize, as with groundnuts, . . tobacco-growers attairi yields significantly grea,ter than those achieved by farmers who •do not grow .; · tobacco (signific~nce level for both seasons is I 0.005 per cent). On average, tobacco-growers have maize yields 36 per cent greater than nongrowers. ~oreover, not only is the yield of maize obtained by tobacco-growers greater overall, but . also. for· every level of extension visits :for bO.th years I.· growers achieve· higher.yields thari nongrowers .. 93. Credit, Yields and Extension Vi~its--Data are 1 available for 1970/71 which relate cropyie.lds and extens~on . . ' I visits to,wnether.or not a farmer had LLDPseasonal cred.:i.t. i I i'hese data are set out beiow. · · I. I No. of Credit reciEiEmts NonreciEiEmts . TOtal visits. Maize Groundnuts Maize Groundnuts Maize 'Groundnuts (lb per·.· ac) ·· 9 859 (24) . 196 ~97 (402) 406 .. 895 <426) 385· 1-5 1,208 (16) 444 928 (150) . 495 . 955 (166) 490 6.:..10 1,554 (6) 731 923 (32'). 290 . ' 11023 (38) . 357- 11-20 11484 · (15). 384 1,001 (39) 325 1 ,135. (54) 342 20+ ,, i,236 (71) 525 1,028 (110) 492 1,110 0.. 81) 505 Total. 1,209 (132) 421 . 930 (733) . 422 972 (S~5) 422. CI(x) ±231 (176) ±73 (55) ±72 (55) . ±35 (26) ±71 (541 .. ±32(24' r ,I ! - 49 - 94. As in previous results, figures in parentheses, except for those following the confidence interval, are the number of observations in each category. 95. The maize yields obtained by credit recipients ar~ significantly greater than those of nonrecipients at a , I probability level of 0.005 (t = 3.63), however the mean I groundnut yields obtained by the two groups are virtually I identical. Designating growers as cr~dit recipients, though, does not indicate what credit package they received. Enough maize packages were issued in 1970/71 to plant an area 10 times as large as could be planted with the groundnut packages ' issued, so it would be expected that the yield impact of the I credit program would be greater on maize than on groundnutsl. 96. Analysis using a chi-square test indicates, as would be expected, that credit recipients receive extension visit~ in significantly greater numbers than would be expected on a simple proportional basis. The expected and observed numbe~s of credit recipients in each extension category are shown below: Number of extension visits No. of credit.reciEients None 1-5 6-10 '11-20 20+ Observed 24 16 6 15 7l Expected 65 25 6 8 28 x2 = 1o1.26 (P< o.ool) What is most surprising about the observed number of visits is that 18 per cent of credit recipients received not even a· single extension visit. It would be extremely interesting to know to what extent, since 1970/71, the expansion of the credit program over a fixed extension base has resulted in greater and greater numbers of credit recipients not being visited by extension staff, but this question cannot be I ·- 50 ~xplored with the data available. 97. Extension Visits Related to FarmSize--As many extension programs have been criticized on the grounds·that staff tend to concentrate their efforts on the larger .farms, it was thought worthwhile to examine the extent to which a bias in favor of large farms exists within LLDP~ Farm size (measured as cropped area) was arrayed with the number of . · e:?Ctension visits received in the 1970/71 and 1971/72 seasons, and the results are shown below: Mean number of cropped acres No. of visits received 1970/71· 1971/72 None *3.83 (81) . *4.14(148) 1-5 *3.75 (65) ·*4.68 (60) 6-10 4.26 (27.) 5.68 (43) 11-20. 4.36 (34) . *5;73 (43) 20+ *5.05(126) 5~43 ( 96 )'. .Total· 4.37(333) 4.88(390) CI (x) ±0. 35 (0. 26) . +0. 39 (0. 30); ..· - . . . I . ! I . 98. Farm~rs receiving no extension visits had holdings· . I · 12 per cent smaller than ·the mean in 1970/71· (z =. 2. 46, 1 .. . . . .. . . . .! P = 0~007) ·and 15 per .cent smaller thi:m the mean in 1971/72! (z =4.33, p..( 0.001); these and.otherdifferehces signific:,ant at~P = 0.01 ..... . . or less hav~ been indicated with an a~terisk. . ·. . . . . . .. . . I ·. Fa:qrts receiving. more than 20 visits are on·. average 13. per c ent 1 largE!r than the total mean, but' only the difference for .197p/ 71.is significant. ()verall, for both seasons. together, thej . I weighted mean size of farms visited by extension .staff was 24 per cent greater.than the size of farms not visited.· .·, i 99. The evidence, theri is consist·ent with the inter- , pretation that extension ~taf:f favor the larger J;toldings in/ terms of number of visits. Though the absolute differences! i ·' - 51 - ~ . - in farm size are small, the relative differences ar~ fairly large and may be more significant than simple magnitudes alone would suggest. 100. To summarize the findings of this section, there is strong evidence of a positive extension impact on crop yields, though most of the available evidence indicates association rather than causation. The principal extension impact appears to derive from an interaction amung seasonal credit, adoption of practices and inputs, and extension staff visits. Extension staff concentrate their efforts most strongly on the relatively larger holdings and on holdings where the farmer is a credit recipient. Evidence is presented which suppotts the belief that tobacco-growers constitude a distinct subgroup of the.prograrn's target population, for they consistently outperform in yield terms other program farmers; yet they attain their superior performance levels without the benefit of an~ more assistance from extension staff than .that received by other farmers. r -"52 ~- I~ SECTION V DETERMINANTS OF CROP YIELDS lOL The Lilongwe project was originally conceived with objectives which differ considerably from those which sub-. seq~ently came to be espoused by project management and by Ban.k staff. The original· goal was essentially that of increasing production of three crops~-maize, groundnuts and tobacco; and increased production was to be achieved by raising yields, by expanding acre~ge and by relying upon market incentives to stimulate a positive supply response. 'The most important of these three factors was the project's intended impact on crop yields, and yields became the weather- vane that indicated the project's degree of success or failure. The bulk of the project's evaluation work is still directed toward asbertaining seasonal yields through ~ long- continuing series of annual yie_ld surveys; this emphasis on . .physical yields persists despite the embodiment of a broader coridept of rciral (rather than purely agricultural) develop- ment in the design of the second and third phases of the project • . 102. Since yields have·already received a great ~eal of attention in the project's annual yield reports, in the ARCS review Of·LLDP and in the Bank's performance audit of the first phase, there is little point in duplicating these earlier results here. The approach taken, therefore, is to try to broaden the perspective on yields by viewing the data in. a slightly d,ifferent context than that of the previous· work and by approachin,g the ana~ysls with a different set of questions. - 53 - hation of r~lative yield levels *nd mo~es on to assess the importance of a: range of factors in influencing crop yields; most ·.of the statistical results, are· reported in the Annex. (The discussion relating to yield~ in tile sections. on credit and.extension is·highly relevant to the treatment in this section.) 103. Comparison of Project and Nonproject Yields-~The first hypothesis tested was that there is no significant difference ip.mean yields between farmers inside the area administered by LLDP and those outside it. ·The hypothesis was testedusihg yielddata for the three seasons 1969/70· to 1971/72 and the results, together with statistical significance levels for the differences in the means, are given below: Mean Yield in lblac Significance level Year.· CroE LLDP NonLLDP .of difference 1969/70 Groundnuts 394 (229) 371(230) Not . Sl.g. at 0.10 Maize 1,178(184) 737 (167) o.oos 1970/71 Groundnuts 349(296) 451(415) o.oos .. Maize 1,105(343) 1,092(411) Not sig. at 0.10 1971/72 Groundnuts .313 (390) 289(446) Not sig~ at 0.10 Maize 1,494(390) 1 ·1 ,08 8 (446) o.oos 104. Maize yields are consistently higher within the· project than outside it, and the difference is significant two years out of three. For groundnut yields, only the higher yield outside the project in 1970/71 is signific~nt. The figures above should be interpreted bearing in mind the fact that the area covered by LLDP expanded in each of the three seasons shown, so that mean yields reflect differences in area productivity as well as seasonal differences. Unfortu- nately the method of sampling used by the LLDP evaluation 54 ,:;,. .; .· unit does not permit inferences to be made regarding possible improvements in crop yields which might result from ·farmers• - being taken into ·the program. Recommendations as to changes in, sampling procedures are contained :in the concluding· section. 105. · Growers' Opinions of Crop Yield_s--In each ~:~eason farmers ·were asked how their overall yield of maize ang groundnuts compared with previous seasons. If yields within LLDP are improving, it is important that they be improving _ quickly enough, or on a scale of magnitude great enough to be perceived by the farmers concerned.- Hence, as a matter of iriteres.t, growers' opinions of' their :crop ·yields were arrayed with the actual mean·physical.yields obtained. The· results at~ given· be!low:- Yield and 9:rowers• opinion of lield CrbE, _. area. arid season Bad- Average- ·Above_ averase -- (lb/acre) . MAIZE:·- LLDi? .1969/70 638 (50) 1,540(63) 2,089 (42) 1970/71 77 4 (130) 1,092(36) 1',523 (157) : 1971/72 1,091(125) 1,589(85) ' 2 ' 17 0 (13 9) NonLLDP '' 1969/70' 651 (16) 1,391(55) 1,722 (2l) 1970/71 862(140) 1,082(54) 1,363 (198) 1971/72 840(176) 1.,086 (87) 1,341(139) GROUNDNUTS: -t.LDP 1969/70 317 (12) 753 (76) 1,083 (27) 1970/71 250 (68) 519(30) 556(127) 1971/72 293(168) 489(54) 607 (77) ' NonLLDP 1969/70 258 (4) 684(62) -931 (45) 1970/71 405(124) 515(50) 633(176)· 1971/72 295(197) 454(62) 504 (85) - 55 - 106• Though differences in sample size make valid . comparisons. somewhat difficult, for LLDP growers the subjec- tive assessment as to what is a bad maize yield seems to indicate increasing expectations over time. In contrast, with the exception of the seasonal dip in 1970/71, there does not seem to be much change in the notion of what an average or above average yield is. The pattern is reversed in the case of groundnuts, for farmers' notions as to what is an average yield have declined every year, paralleling the decline in physical yields. Opinions as to what is a bad and an above~-. average groundnut yield also reflect a downwardtrend. 107. On average, for both maize and groundnuts, an average yield is some 68 to 82 per cent above a bad yield, and an a.bove-average yield is 24 to 32 per cent greater than an av~rage _yield. A bad yield is only 48 per .cent of-an abov·e_-average one. 108. Yields Related to Factors Beyond the ·proj"eet·• s Control--A number of economic and other factors cannot be influenced direct!~ by project activities and thus cannot be manipulated ' . so as to improve crop yields. . Among the most obvious factors which are beyond the control of the project are weather and soils. Analysis carried out for the ARDS rev_iew of LLDP. attempted to relate rainfall to crop yields . ,. through several functional forms but the results obtained were not significant. Attempts to use soil type as an. explanatory variable were thwarted by~a lack of disaggregated data. 109. Ultimately it proved possible to examine only two of the factors outside the project's control~-labour force size and cropped area. Physical yields of malze and ground<:' nuts were regressed on these two variables for the seasons 1969/70.and 1970/71. Results of these regressions, as well as the ~esults of all other regression equations in thi~ section, are set out in full in the Annex. As a rule, only the. major implications of the results will be discussed in - 56 - 110. The results are consistent in indicating that yields of both maize.and groundnuts are higher ori holdings with ·larger.cropped areas, though this finding is more significant in·the case of groundnuts'than for maize. As would be expected~ the two independent variables are quite strongly c6r~elated (coefficients range from 0.248 to 0.496), so that yields are also positively related to increments in the labor force (with the exception of groundnuts in 1969/70). 111. These results are important because they provide additional evidence relating to the problematical groundnut yields. An. inc::rease in per-unit yields with an increase in cropped area is in general the oppo.si te of what one would .. expect to find in most situations. This phenomenon is thought to.occur with groundnuts in the Lilongwe area because the. smaller holdings lack adequate land .on which to rotate ground- nuts with other crops or with a fallow. Short rotations or th~ complete absence of rotations is thought to contribute to a build-up·of .the pests and diseases which attack.ground- nuts. For comparasion purposes, a holding with five acres of cropped land can expect on average to achieve groundnut yields some 130 lb/acre greater than a holding with only two acres, assuming situations regarding labor availability remain fixed. 112. Yields Related to Specific Practices--Among the practices which the project could influence, and which would inyolve no direct financial outlay for farmers, are the time of planting and the per-acre planting density. Time of planting (with fiv.e different periods defined from "before the first rains" through "well after the main rains" and per- acre plant population (at time of harvest) were employed as . ind.ependent variables· to examine their :l:mpact on yields. Again, as would be expected," the two independent variables are highly correlated, though with a seasonal effect in· evidence; and the correlation coefficients range from -0.836 to +0_. 780, with no absolute value less than o. 500. Results ----~-------------------------- - ':j7 - for planting date are gener~lly riegative and not ~ignifican~; i.~. there is probably a negative but rioniinear relationship between yield and time of planting. The coefficients for plant population are, of course, all positive and also highly significant, with the coefficients for maize being larger than those for groundnuts. The equations for groundnuts explain on average about twice as much of the variation in yield as the comparable equations for maize. 113. Another factor which could be influenced by project extension staff with no direct financial outlay by farmers . is crop rotations. Crop rotations, and cropping practices, are·aspects of cultivation that have been little examined. by LLDP.- The follow-up study obtained results which indicate that rotations and intercropping have important consequences in terms ·of c.rop yields. The results are set out below: Yield of maize Yield of groundnuts 197oL71 1971L72 197oL71 1971L72 .. (lb ;eer ac·) Cro;es in ;elot before: (1) Fallow *210 . (54) *262 (4 7) *98(176) *37(217) (2) Maize *922(218) 1,229(204) *510(522) *399(590) (3) Groundnuts 1,116(319) 1,264(353) .*585 (5) 256 (14) ( 4) Tobacco ""·1,827 ( 95) *1,873(137) 229 (2) 324 (8) (5) Vegetables 1,361 (9) 704 (8) 351 (4) 18 (1) (6) Mixture, excl. maize 1;257 (46) 960 (28) Total 1,098 (7 54) 1 1 2 91 (7·8 2 ) 409 (711) 301(836) Cro;es in ;elot now: (1) None--crop failure 0 (39) 0 (31) 0(136) 0(193) (2) Maize 1,141(41.1) 1,366(665) -- ( 3) Groundnuts *534(192) *389(520) .i (4) 2 +3 519 (4) 1,545 (11) 504 (15) 424 (11) . ' (5) 2 +'other 1,199(297) *1,000 (66) , I ' I (6) 3 + other. *52 6 ( 2 67 ) * 3 9 7 ( 9 3 ) (7). 4 +other 270 (3) 1,355 (9) 3 73 (15) 38 4 (19) Total 1,098(754) 1,291(782) 410(625) 301(836) ~OTE: An asterisk indicates a yield which is significantly different - ,..,.., ,... ___ .&-_..__,_ ..:.--1,..,,:J:,..... Y't"i-lnn,... 1""~1"'\T'\C: ·. - 58 - 114. The most consistent result is that poor yields are obtained following·a fallow, regardless of crop. On average, groundnuts yieid better when they·· are planted f'ollowing a maize crop, as is the most common practice. Particularly striking are the very highmaize yields obtained when this crop follows tobacco; the total mean maize yield W<?uld be much lower if the yields of tobacco-gr~wers were excluded~ The high maize yields ~f tobacco-growers are thought to ste~ from residual effects ~f fertilizer applied t6 the t6bacco crop. And, 011be ~~ain,. tobacco~growers are· singl~d out as being di~ferent in an important respect from other project farmers. 115. With. regard to intercropping, groundnuts yield . significantly better either planted in pure stand or inter- planted with crops other than maize. In 1970/71, maize in pure stand or interplanted.with crops other than groundnuts _yiel~ed more than the mean (though not ~ignificantly)7 whil~· in 1971/72, maize interplanted with crops other than ground- nuts yielded significantly less than the mean, but ali' other mixtures and pure stand maize yielded more than the mean (not . significant). 116.. ··· As the regression results for time of planting were not particularly informative, it was decided to take another approach to see if better results might be obtained. Crop yields were crosstabulated with three times of plantinc:J: · (1) ·,early planting--befor~ the first rains, (2) medium planting--between the first and the main rains, and (3) late· planting--any time after the main rains. Results for tbis · sec:: and analysis are shown below: ·· . ,·.,··, . !· - 59.- . .. Tim~ of planting and yield . Crop'and :lear Earl:l Medium Late Total . (lb per acre) Maize: .1970/71 1,285(235) 1,117(437) *876 (43) 1,158(715) 1971/72 1,379 (99) 1,363(597) *974 (54) 1;337(750) Groundnuts: 1970/71 446 (5) 519(441) 462 (129) 505(575} 1971/72 500 (11) 406(534) *297 (98) 391(643} NOTE: An asterisk i·ndicates a yield which is significantly different from the mean at 0.01 per cent. 117. For the two seasons, the average yield for medium- planted maize is just seven per cent less than that for early- planted maize, although the difference between early and medium-planted maize is statistically significant only in 1970/71 (t = 2. 41, P = 0.01}. .The most dramatic decline in maize Yields occuts with late planting, which results in yield~ onaverage 26 per cent lower than with medium planting.· Maize planting in 1971/72 appears to be generally later than in the pr~vious year. 118. .Groundnuts are not generally planted early, as the . figures above clearly show. The average yield loss from late planting as compared with the normal medium planting is so~e . . 19 per cent, b~t only the figure for late planting in 1971/72 is statistically different from the mean. 119. Looking at the average coefficients of variation for yields for each of the three planting times reveals that· maize yields are most reliable with medium planting; both early and late planting result in somewhat more variable yields. For groundnuts the coefficient of variation for yields rises continuously from early through medium to late planting. The . average cioefficierits are: ii i! . ·- 60 - Coefficient of variation for time of Elantin'g: Early Medium Late Maize yiEdd 79% 75% 85% Groundnut yield 59% 62% 71% 120. As the coefficient of variation is defined as the standard <leviation divided by the mean (times 100), it can be· reduced either by decreasing the variance of the yields or by increasing the mean yield, or both. The figures above, then, suggest that when both the mean yield and its varianc~ are considered together maize should be medium-planted and ground- nuts early-planted. The small samples for groundnuts.are somewhat suspect, so that medium planting for groundnuts is pro~ably a safer recommendation. Medium~planting of ground- nuts, however, encounters an immediate bottleneck; which is caused by the planting of m~ize--the staple foodstuff-~in the same period.. As a consequence, the planting of groundnuts terids·~o be'delayed until even later~ 121. A note of caution should be injected here regarding the making of recommendations on the strength of results derived from analysis of yield data in isolation froin other types of data. It is now widely recognized that time of planting, .for example, is of crucial importance in determining yields in rainfed agricultural systems, and the results here corroborate this importance. It would be naive, however, .to recommend any particular time of planting without additional information on such factors as the existence of labour bottle- necks or the extent to which a trade-off exists for the. farmers between the level and the.reliability of yields. Where management decisions need to be made regarding the · adoption of cropping practices to be recommended for the project farming .system, this is the point where it becomes essential for the yield data to be supplemented by data based on farm management studies. Without this latter type of ~-- ·- 61- information, recorded yields simply indicate whether or not a given target has beert achieved and tell nothing as to why or how the· achievement .was accomplished. 122. Yields Related to Crop Variety--A variable on which the project has exert~d considerable inf1uende is the variety of crop--particularly maize--planted, though in this case the innovation is made only with some financial cost to the farmer~ The results of the regression models for crop variety.are given ~n the Annex, while the figures following below show the mean yields over three seasons-for maize and groundnuts. 123. Maize yields have been grouped by type .rather. than . by individual variety since in certain ins-tances one vartety of· a certain type has been phased out to be replaced by · · another var.iety •. The yields shown relate. only to the project area. Mean maize ~ield b~ t~Ee Local S~nthetic H~brid Total (lb/ac) 1969/70 1,336(105) 1,415 (26) 3,353 (5) ' 1,178(184)a/ 1970/71 1,006(203) 1,036 (59) 1,864(61) 1,173(323) 1971/72 1,228(168) 1,885(127) 2.357(54) 1,642(349) a/ Including 29 observations were zero yield where type was. unspecified and 19 observations where types were .mixed. 124. There is actually remarkably little one can conclude from results such as these. - It is possible to say that hybrids consistently outyield the other t~o types and that synthetics consistently outyield local maize~ but only the superiority of the hybrid is significant (at. P = 0.01) in every year, while the synthetics yield significantly more than local maize in· only one year (1971/72 at P =0.01) ~ Aside from these - 62 - ,. observations, one can conclude that maize yields partie~-· larly of the improved varieties., appear to be susceptible to · strong seasorial v~riations. 125. The variety of groundnut included in LLDP seasonal credit packages is the large, confectionery-type.Chalimbana. nut. Even without the credit program, however, the Chalim~ bana nut is the most common variety in the Lilongwe area . . Other varieties are grown in pure stands on a limited scale and. the interplanting of Chalimbana with other varieties is . quite common. The mean yields of Chalimbana compared with other varieties and mixtures of varieties are ·shown below for lg69/70 throtigh i971/72. Mean g:roundnut :Lield b:L variet:L ·chalimbana Others ·Mixtures Total (lb/ac) 1969/70 815 (92) 746(16) 752 (18) 383(45~)Y 19.70/71 .513(195) 423(38) 500(342) 505(575) 1971/72 424(391) 212(41) 363(211) 391(643) a/ Including 233 observations with zero yield where variety was unspecified.· 126. Because zero yields have been separated out from the 1969/70 variety data but included in the total for that year, it is possible to see what good yields at least some farmers.achieve. Due to the method of treatment, of course, all 1969/70 yields are significantly greater than the mean yield. The yields for 1969/70 should be contrasted with those for the other two years, .where zero yields are taken into account in the mean yield figures for varieties and·· mixtures. None of the individual results for 1970/71 is .significantly different from the mean yield, but in l971/72 the yield of Chalimbana is significantly greater than the L___ _ _ _ _ __:___ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ . _ _ _ ·- ·. - 63 - mean whLle ~hat of "others" is.significantly less (both at p == 0~001). 127. Treating the data in the manner used for 1969/70 results in an interesting hypothesis: LLDP is not experiencing a general decline in groundnut yields but' rather the devedop- ment of a bimodal pattern of yields in which substantial nuffibers of farmers receive excellent yields while large ·numbers of others suffer near or complete crop failures. For reasons explained in the introduction; this was another hypothesis where exploration of validity could be taken no further. If _the hypothesis is valid, however, it. should ma~e somewhat easier the identification of the factor(s) responsible for the poor-performance of groundnuts in the I • I project area. 128. As noted earlier, the statistical results for this . section are for the most part to be found in the Annex. Recommendations on the collection of yield-related data and how.these data might be more useful for· project ma11agement. purposes are-contained in Section VII. .---------------~- - - - - 64 - ·. SECTION VI DETERMINANTS OF INCOME DISTRIBUTION 12 9. As noted in the introductory paragraph to Section V, LLDP was conceived fundamentally as a crop productivity project. That Lilongwe is now regarded as a model for the Bank's "new style" projects is recognition of the far;_reaching evolutionary changes .that hav~ occured during some seven yea~s of implementation. It is interesting to note, however, that evolution has occurred at a faster rate in some areas of prbject operations than in others; despite changing priorities on the part of project management and the Bank, the evaluation effort does not yet reflect these changes in'direction and in general far more attention is directed toward questions relating to crop productivity than to issues relating to rural development in a much broader sense. Doubtless, though, the operations of the evaluation unit at LLDP will evolve as have the activities of other project sections (though the needed changes might take place faster and more efficiently with the assistance of external direction and manpower resources). What is of concern here is that the analysis of income distri- bution is further from the original purposes· of the data than any other subject examined in this report; as a consequence, f.ewer aspects of this subject can be ana.lyzed and the results of the. analysis need to be interpreted with more than the usual caution. 130. The data on income reported in this section derive from the socioeconomic survey carried out in 1969/70, a s~ason when LLDP operations covered a relatively small area. It is possible that over subsequent years LLDP's operations and .. - 65 - programs have resulted in substantial changes both ih absolute arid relative income levels, hence t~e resuits obtained here should perhaps best be. regarded as benchmarks.both to be verified and against which to assess the magnitude and direction of change in possible future survey work. 131. Agricultural Income and Social Status--Social. status could obviously be defined in many dimensions and using many different variables, but the variable used in the analysis was a composite one indicating {1) whether_or not a farmer held a position of traditional or nontraditional responsibility (with these positions ranked in a crude hierarchy and weighted) , {2) the highest level of formal education attained, and (3) the number of gardens for which ~n individual held cultivation rights. The mea~ of the status scores was 3.27 with a range of 15 {1-16). Two strata were defined: one for farmers with a status score of 10 or less and one for those with a score of more than 10. The results ofthis stratificatiori·are as follows: Status score Mean agricultural income 1 - 10 (668) K115 11+ (14) 219 Total (682). 117 CI (x) +30 (23) CI(n) ±780(592) _Range .0 - 3,130 The t value for the difference between the means of the two strata is significant at the 10 per cent level, which suggests that the definition of social status employed is reasonably useful in detecting differences in agricultural income. Only two per cent of the total sample falls within the upper stratum. The r_ange of individual agricultural -incomes is enormous, and the sample mean income ('at the-99. .-----~~~-~~--~ -- - 66 - ... per cent confidence level) lies between K87 and Kl47. 132 • The correlation coefficient between agricul t.ural income and the status score is only 0 •.196, much smaller than was expected; but it is interesting to go further arid look at the correlation coefficients. between total agricultural 'income _and the income received from various crops: Income received from: Maize Groundnuts Toba·cco Other crops Livestock Total ~gricult~~al 0.228 0~359 o. 934. 0.160 0.269 income·. i· I The. association b~tween income from tobacco and total . : .. . . . agricultural incom~ is striking and strongly s~ggests that tobacco sale$ are the major determinant of farm incomes in' . ·the area, ·with groundnuts second in importance and maize in fourth place, after livestock. The ARDS review of LLDP · shows that recorded cash receipts from maize and groundnuts as a proportion of total cash receipts have risen from some 22 per cent in 1969/70 to about 40 per cent in 1971/72, and this· can be taken as evidence that the dominant position of tobacco as the principal cash crop has been weakened. 133. Agricultural Incomes and-Migration--It was hypothesized that farmers with limited agricultural opport~ unities and low farm incomes would be more inclined to take advantage of the possibility of migrating to work temporarily outside Malawi, a practice common both in the LLDP area and elsewhere in the country. The hypothesis could not be ;: directly tested in the farm stated above wifh the available data, hence an ex post approach was taken instead. It is the pr~ctice for migrants to send part of their earhings · back.to their families who have remained in Malawi, therefore receipt of such remittances is the surest indication that a - 67 - .member of the·farnily is actually in paid.ernployrnent outside the country. Receipt of remittances then should be negatively asspciated with agricultural income if the hypothesis holds. A second, and also indirect, test was employed which consisted of comparing the current agricultural incomes of those Jarmers who had in the past worked outside Malawi with the incomes of farmers who had never worked outside.the country. Both sets of results are shown below: Mean agricu!'tural income Test I: Families not receiving · KllS ·.remittances (651) Families receiving remittarices(31) 165 Test II: Farmers who have never worked outside Malawi (346) .Kl07 Farmers· who have worked outside Malawi (336) 128 Total (682) K117 I ! 134. The confidence intervals and range for the total mean agricultural ~ncome of Kll7 are identical to those shown with.the results reported under social status; as these parameters are also identical for other .results in this section, · they will not. be reported again. 135. The results of both tests are the opposite of those hypothesized, i.e. former migrants and current migrants' families appear to have agricultural incomes well above those of nonmigrants or those who are not receiving remittances from a family member abroad. These differences have magni- tudes of 43 per cent (Test ·Il and 20 per cent (Test II) , but the. t values for the differences· between mean incomes· are significant onli at the 25 per cent level in both cases. · Th'e results, therefore, are not consistent with the inter- - 68 - pretation .that in the Lilongwe area it is farmers in the lower income groups who emigrate for employment. On the contrary, the findings suggest that there is some positive· link between emigration and agricultural income, possibly that income earned abroad is reinvested in the farming enterprise. The ana1y$is of this data carried out by LLDP concludes, however, tha.t .the most popular expenditure categories for repatriated earnings are,.in order of importance, consumer nondurables and durables, livestock, and gifts~ only five pet cent of recently returned migrants reported m~king on-farm investments. The nature of the link,. then, between emigration and higher agricultural incomes remains obscure~ it would, however, be of interest in future work to examine th~ hypothesis that re~urned migr~nts have higher innovation adoption rates, possible reflecting a sophistication acquired through the migration experience. 136. Agricultur·al In·comes and Wage· Earnings--A rough comparison between e~timated agricultural incomes in the Lilongwe rural areas and the earnings possible working at the statutory minimum wage level in the city of Lilongwe suggested. that on average gross incomes were higher in ·the rural areas. Estimated minimum cash earnings ·from urban employment were approximately K95 per annum while mean agricultural incomes, as shown earlier, were Kll7. It was hypothesized, therefore, that "farmers" who·have regular paid employment seek out such work because they have only limited opportunities in agric- ulture and that agricultural income would be negatively associated with wage income. This hypothesis was tested by again 'using a. simple two-way stratification of 'the sample: Mean agricultural income ·No wage employment during preceding ·year (625) Kl22 Regular· paid employment during preceding year (57) 70 - 69 - 137~ Full~time farmers with no outside employment had agricultural incomes 74 per cent greater than "farmers" with regular wage employment; the ·K70 agricultural income of the . latter is significantly lower than the total mean income (z ~ 2.27, P = 0.012). Such a result is not really surprising, however, as an individual in wage employment would have less. time to devote to farming. Indeed, what is surprising.is_that the agricultural income of wage-earners is as high as it is, though this fact is probably explained by the practice of wage earners' leaving their families behind to cultivate the family holding. More than eight per cent of the sample had been in regular wage employment during 1969/70; summing their agricultural income with their estimated wage earnings results in an estimate of total gross earnings of Kl65 per annum, 41 per cent above the total. mean agricultural income. Thus, rather than seeking wage employment because their farming opportunities are limited, wage-earners may simply be taking.advantage of an additional opportunity to increase :their total income. · 138. Agricultural Incomes and Nonagricultural Incomes-- It was expected that agricultural incomes would be strongly related to nonagricultural income (excluding wages) because of . the tendency for wealthier farmers to diversify by invest- ing in nonagriculturai enterprises. Data were avail~ble which indicated which farmers received income from maize grinding mills, grocery stores or small shops, brewing or distilling local alcohOlic beverages or oth~r sources. The sample was divided into two strata: those who had suc.h a nonagricultural income and those who did not. The. mean agricultural-income for each stratum is shown below: Mean agricultural income No nonagricultural investment ··income (418) Kl08 Income from mill,· shop, brewing, other (264) 131 ·. - 70 .:. 139. Though farmers with non.agricul tural investment income have agricultural incomes 21 per cent greater than those with no suc;h income, the former group's mean income is nof highly significantly greater than the total mean (z = 0.64, p = 0.~61). .. 140. This relatively low level of significance-probably stems from inclusion of those with brewing or distilling as a source of nonagricultural income. The making of "hoine brew" is a very widespread activity ·which can be expected to indrease wben maize yields and produetion.rise even further. This type of activity could be expected to exhibit a positive . but weak correlation with agricultural income since the principal participants will be farmers (or farmers' wives) who have a surplus of maize. It is also quite likely that brewing and distilling are among the first nonagricultural . . . . investments made by farmers since the techniques are known to all and the capital requirements.are small. Almost 40 per cent of the sample farmers have some source of nonagricultural income,· and it is thought that for most this source is brewing and/or distilling. 141. Agricultural. rn·c·ome·s and Gift·s· i"n Cash ·and in Kind-- The role of gifts in the Lilongwe area is an interesting one, for the. giving of gifts appears on. the one hand to be a manifestation of a built-in social welfare system and on. the other a device used to create a debt-bondage that is repaid by the recipients' supplying labour to the givers. The first hypoth~sis tested was that gifts of cash were in fact soci~l trans~er" payments, i.e. that such gifts were made by the relatively better-off farmers to those less well off. The 1969/70 SES sample was stratified according to whether or not farmers had received gifts of cash during the previous year. The mean agricultural income of·each stratum is shown below: - 71 - " ' ' ·-Mean a9:ricultural iricome Received no gifts in cash (612) Kl21 Rec~·ived gifts in cash (65) 87 Total (682) 117 142. The income of those receiving gifts is sig~ifi- cantly lower than the mean ·cz = 1.49, p· = 0.068) while the income of those not receiving gifts does not differ signifi- cantly from the mean (z = 0.25, P = 0.401). The evidence, therefore, supports the belief that it is the relatively less- well-off who receive gifts of cash but unfortunately provides little information as to the economic well-being of the givers. It would not be unreasonable to assume, however, even w~thout supporting data, that such gifts of cash come from donors with relatively larger cash incomes and therefore amount to a redistribution of cash income (provided no ' ' repayment is implied or understood) • .·~ 14 3. The term ''gift" is somewhat ambig~ous in the · Lilongwe context, as it is not known to what extent strings may be attached to what appears to the outsider to be a straightforward present. rt isknown, for example, that the relatively wealthy and those who are rapidly on the way to becoming relatively wealthy will make gifts in order to avoid recrimination and possibly retribution from relatives and others who do not enjoy equally high levels of prosperity. In this context "gift" should perhaps be read as "bribe"~ It is also known, though, that farmers frequently make gifts to members of their extended families, but even in these· cases the gifts may not be made entirely willingly. At any rate, there is almost certainly a reciprocity mechanism at work and, should there be a reversal in the relative fortunes of the better-off and the worse-off, there would likely be a reversal ·also in the roles of donor and recipient. 1~4. The discus~ion above relates primarily to one - 72 - . aspect of gift-giving: . . the more-or•less permanent . transfer ' of monetary wealth, for whatever motivation, from the relatively wealthy to the relatively poor. Work carried out as part of the ARDS generated a hypothesis regarding gift- giving less benign than that tested above. This hypothesis . is that the giving of gifts in the form of produce to distressed farmers perpetuates their poverty by obliging the· recipients to work for the giver at a time when their opportunity cost on their own holdings is highest. In order to obtain data relating to this hypothesis, a subsample of 117 farmers was drawn from the 1972/73 LLDP sample selected for yield studies, and interviews were carried out in October 1973. Information was collected on whether farmers had given or received gifts of maize or groundnuts, how much produce . was receiyed or given, and whether work--and how much and .what.kind--had·been or would·be done as aconsequence of the gift. 145. ·Thirteen per cent of the. sample had received gifts. in kind and 30 per cent had given them. Whether or not there ·is an obligation to work stemming from receipt of a gift in. kind varies somewhat depending upon whether one asks the donot or the recipient. Of those who received gifts, 5~ per cent stated that they had worked or w6uld work for the person who gave the produce to them. On the other hand, only 23_per cent of those giving away produce claimed that the recipient would work for them in return. While these different perspectives_on the obligations accruing from gifts may arise from an individual's,position as a giver or as a reci.~ient 1. they are alSO COnSiStent With the interpretatiOn. that the donors who expect work in return give produce to. more than one individual or family. The proportion of growers making gifts in the 1973 sample generally coincides with the results of a 1971/72 LLDP survey (using a sample. four times as l~tge), in which 23 per cent 6£ ·farmers made gifts of maiz.e and ·13 per cent made gifts of groundnuts. ~------------~---------------------------------------------------- - 73 - 146. Those who received gifts and worked put in an average of slightly more than 10 labor-days on behalf of those who·gave them·the gift; and most of this work was related to preparing land for planting, and thus came at the 'time of year when the opportunity cost of labor is at its highest. In Central Africa, it is not uncommon to find that maiz~ yields decline by up to 100 lbs/acre with each day'~ delay in planting. To the extent, then, that time spent preparing another's land for planting results in a compara- bl~ delay in preparing _and planting one's own .land, the average loss in yield could be as great as 1,000 lbs/acre.· Losses of this .magnitude would certainly tend to. weaken the economic position of· those supplying labor, and the labor supplied would strengthen the position of those for whom the work was done. Thus there would appear to be_ a self- perpetuating element in this system of debt-bondage. 147. What is uncertain is the extent to which debt- bond~ge amounts to a "system" or how pervasive it is._· Extrapolating the results discussed above suggests that some seven'per cent of project growers receive "gift~" in return for which tbey are "obliged" to work for the "donor". This percentage was equivalent to some 3,750 farmers in_ mid-:1973. If this extrapolation from a small s~mple is valid, does this seven per cent of farmer-s constitute the poorest stratum of Lilongwe rural society? If so, is this also the group which fails to receive visits from extension staff and which has not made use of LLDP-supplied credit? These_ are interesting questions, particularly in view of the Bank's increasing concern with income distribution, but it i~ beyond the bounds of the available data to seek further here for possible answers. This analysis does, however, indicate the problems which may exist and suggest directions for further research. 148 .. It was also postulated that gifts in kind on a large scale would result in a more equitable distribution ... 74 of project benefits but would reduce marketed surplus. The average "gift-wage" from the 1973 survey was 0.24 baskets of unshelled maize per labor-day, or a total of 2.5 baskets per person-working. The amount of maize disposed of in the form of such gifts may be in the neighborhood of 482 tons, or three per cent of total sales through the maize-shelling prog~am in 1973. Grourtdnrits are also g~ven as wage-gifts,. with one basket of unshelled groundnuts being given for each day's labor, though groundnuts are a less common form of gift than maize. Some 525 bags of shelled-equivalent groundnuts may have been disposed of in this manner in 1973. This.amount is an insignificant proportion of the tota~ quantity. of groundnut.s marketed in Lilongwe District in 1972. 149. The 1971/72 LLDP survey reported the average qift of maize as 1.67 bagst.Or 334 lbs. Using the figures reported above·of 23 per cent of growers making gifts· of maize, some 1,960 tons of maize--four times the 1973 estimate--may have been disposed ot as gifts in 1971/72. This level is equal to approximat·ely 25 per cent of maize sales to ADMARC and the LLDP shelling program in 1971/72; thus gifts of maize .do in certain seasons at least have the potential to reduce sign~ ficantly the marke~ed output of sUrplua production. 150. It was hoped to be able to test also the important hypothesis that incomes within the pr6ject area are positively related to active participation in LLDP-sponsored programs, but such testing proved in fact impossible to carry out. No difficulty was encountered in identifying a useful and· powerful definition of "active participation", for the annual list's of farmers receiving LLDP credit provide samples of participants who may be expected to benefit most in income terms from involvement in project activities. The problem was rather that no income data have been collected which can be related to participation in the credit program. Inferences about changes in income arising.from the use of credit could - 75 - possibly be drawn by making comparisons between crop yields or credit and noncredit farmers; the value of such an approach is diluted, however, because no data are available on tobacco yields and tobacco cultivation is strongly associated with cash income. (See related discussion. in Sections II and V.) ·. - 76 - SECTION VII CONCLUSIONS 151. The conclusions arising from this study are presented under two headings: (1) empirical findings, . · particularly where these reveal hitherto unknown or unapprec- iated relation·ships among important project variables, and (2) conclusions regarding evaluation procedures and operations. A final subsection as.sesses the possible utility of follow-up studies based upon data collected for other, often completely different purposes. 152 .. Principal Empirical Findings--Not all the hypotheses relating to the project's credit program could be tested, in part because the necessary records had not been maintained. It was found, for example, that no record was kept of farmers who had applied for credit but who had been denied it. · This . . . : gro..up would be well worth identifying.for a number of obvious reasons, one of which is that thos~ who are denied credit may be among the poorest of the poor. If this were the case,. it would be necessary to formulate an approach which could mean- ingfully ameliorate their position. The. analysis does suggest that some 17 per cent of project farmers are probably ineligible for credit on the grounds that they .crop insufficient land; this is a sizeable group to exclude from the projectis only program capable of dramatically affecting productivity and incomes. 153. Other findings relating to credit include the revelation that farmers taking groundnut credit packages have - 77 holding~·signif~cantly smaller than the mean for the area. These same farmers also have holdings with land-labor ratios smaller than average. .The implications of these findings are discussed in a later paragraph. 154·. There is evidence that the very ra-pid expansion of the. season.~tl credit program between 1970/71 and 1971/72 substantially curtailed the number of farmers receiving v:i,.sits from extension staff and also reduced the average number of visits made to farmers who were visited. The net result has been to increase the concentration of exte~sion visits among credit farmers, and this group can expect to receive 80-100_ per cent more visits than farmers.who do not have project credit. It_ should be emphasized, however, that ·a very large proportion of project farmers are visited by extension staff at least once a year: In fact, the proportion of farmers not visited within LLDP is similar to the proportion of farmers outside the project who· a·re visited. 155. Whether or not a farmer defaults on a loan. appears to·be unrelated to the size of his holding but, once a farmer has defaulted, then the proportion of the loan not repaid is inversely related to farm size. That is, among all defaulting farmers, those with smaller holdings pay back a· smaller proportion of the amount due. The analysis also confirms that the credit repayment rate is lower for farmers taking packages containing groundnuts than for those taking other credit packages. -The reason why the repayment rate is lower for groundnuts is not clear, for no obvious relationships emerge from the attempt to relate credit repayment to crop yields and production. Intuitively it would be thought that yields and production are significantly related to repayment of credit, but the small samples available for credit farmers make obtaining significant results difficult in several different parts of the analysis. - 78 - 156. The analysis reveals a posi~ive respons~ in terms of yield to the first few extension visits for both maize and groundnu~s and for maize a generally positive response to any number of visits. Beyond the first few visits made by extension staff, there is no conclusive evidence of an increase in groundnut yields; indeed, there is some indication of a negative g~oundnut yield response to more than five extension visits. This pattern is possibly not as illogical as it at first glance appears. Extension staff tend to concentrate their .visits on farmers who are growing improved varieties of maize;.hybrid maize growers probably receive more visits than any other group in the project area. If a· farmer who is growing hybrid maiz·e also grows groundnuts, the groundnuts are probably less important to him as a cash crop than they were before h~ adopted hybrid maize •.. Thus he will tend to neglect . the groundnuts somewhat and devote relatively more time and attention to his maize crop. In addition~ most of .the extension advice_rece;ved likely relates to the hybrid maize. 157. Evidence is also.found which indicates that extension staff concentrate a significantly greater proportion of their ef~ort on the larger farms, though to an unknown degree this observation corresponds with the fact that the smallest hold- ings are ineligible for project credit. A somewhat _surprising finding is that roughly one-fifth of all farmers using project credit'are never visited by. an extension field worker. 158. An examination of some of the determinants of c~op yields reveals that groundnut yields are significan~ly lower on. the smaller holdings in the project area. The exp~anation for· this phenomenon is thought to lie in the fact that the smaller holdings are subject to a land constraint and carinot therefore avoid replc3nting the same land to · groundnuts at. fairly short intervals, or even in successive seasons. This lack.of an adequate rotation, or rotation and fallow, results in the build-up in the soil of a number of pests and diseases which adversely affect groundnuts. - 79 - ·- 159.. There.is also evidence which suggests that the decline in mean groundnut yields in the project·area over the past several seasons may not be a decline in general·yields as previously.thought but rather the consequence of the develop- ment of a bimodal pattern in yields. Certainly many farmers continue to achieve· excellent yields on a par with the.best yield~ observed in the area eight or nine years ago._ But many other farmers manage to get only miserable yields if they get a yield at all; a disheartening aspect of the LLDP yield data on groundnuts is the frequency with which farmers abandon plots of groundnuts, earning no return for all the labor and seed invested. Overall some 22-23 per cent of groundn~t plots are complet.e ·failures,· while the comparable figure for maize is·only about four per cent. 160. Another interesting finding relating to yields is that, when either maize or groundnuts are planted in a plot which was under fallow the previoqs season, the yields will .. be dismally low. The explanation for such low yields is thought to.lie in poor preparation of these fields for planting. Fallow land is more difficult to prepare for planting using hand tools than is land which has been under cropping the: previous season. The result is that fields take longer to :till, so that crops are planted late, and often the fields are preP,ared badly so that they yield a luxuriant crop of :weeds but little else. 161. - Regarding cropping patterns and rotations, maize is found to yield significantly much more when it is planted on land which had been planted to tobacco the previous season. This pattern of response probably stems from the effects of residual fertilizer left in_the soil from the tobacco crop. The highest groundnut yields are obtained when this crop is plan~ed on land which has grown a maize crop in the previous season. Groundnuts also yield best either when grown in_pure stand or when grown interplanted with some crop other than ma~ze. The yield results for maize are at least consistent ... -- 80 - . ~ .,, -s.r- "Vl with the interpretation that higher yields~ might be obtained . ·- ··- -----p -···---··-· --· ... ·-··-· .... ~.rom interplanting with other crops, but the results are - ·--·-··------·"-----··-···---·--· -· ·-·- . . . .. inconclusive for the two seasons examined. With the_ currently very. high prices for nitrogenous ·fertilize.rs, a greater research effort might be directed toward establishing the agronomic advantages of intercropping, with reduced levels- of fertilizer application, versus the present recommended practice of planting maize in pure stands with fairly high levels of fertilizer use. 162 •. Incomes in the· LLDP area are found to be positively related to social status as measured on a scale which includes both-traditional and modern attributes of status. Further work could perhaps determine which set of attributes--tradi- . ticmal. qr modern--is the more strongly related to incomes. Somewhat surprising was the discovery that-migration to South Africa, Rhodesia or Zambia for wage employment is positively associated with agricultural incomes in the Lilongwe ~rea. Tpis finding tends to refute the argument sometimes put forward that it is only those with limited opportunities in agriculture who emigrate, however the causal link between. migr.ation and agricultural -incomes remains unclear. The link doe~ not appear to be the obvious one that migrants reinvest their repatriated earnings in agricultural enterprises, for most returning migrants spend their savings on a range·qf consumer durables and nondurables. 163. The phenomenon of gift-giving in the project area raises a nwnber of complex issues which are difficult to resolve, ··particularly since .there has never been a tho_rough anthropological study of the Chewa people. On the one hand, gift-giving·clearly serves as an income redistribution mechanism transferring cash from those relatively better off to those who are relatively worse off; how this mechanism is activated and .what demands may be made.on-both sides are . unknowns. On the other hand, gift-giving, particularly when the:~ gift is in the form of produce given to distressed families, . - 81 - -. appears in some cases to create a debt-bondage. In these .: • s instances the recipient is obliged to work on the donot. holding at a time when the opportUnity cost of the recipient's labor is greatest on his own.holding. It is easy to.see how a system like this could become self-perpetuating. l64. The recipients of.gifts are part of Lilongwe•s· rural poor. To a very considerable extent the institution of gift-giving helps them over times of hardship~ but there are at least two kinds of costs attached to this assistan,ce. The first, as already suggested, is that the result is a cyqle beginning.with need and going through gift, debt repay- ment with labor, shortfall in own crop production, and ending up .again with need. LLDP's standardized approach to increasing incomes through productivity gains achieved with credit..:. supplied inputs does not seem to be affecting this group of poor; indeed, it cannot affect them if they crop only very . vr«'.J. ',. small areas. By opting for a capital~intensive approach to 'lil \'11-''l v'-··).. II \IV"Vl ~· road-building and ·the construction of conservation works,· the ~:~:'.,\.,\'>"- · project may have missed an opportun·ity to help these people 1,t,v~""' out of a low-income trap through the device· of a rural· public 1 works program. The second kind of cost attached to the institution of gift-giving is that it can, in certain se~son~ at least, reduce substanti~lly the marketed output of food crops from the project area. In certain circumstances., such . . as·an adverse season, gift-giving may help to circumvent the need for famine relief or food charity programs; in more normal seasons, gift-giving where a labor. repayment· is· demanded may only ensure that those farmers near. the borderline of · sel.f-sufficd.ency do not rise above it.· 165. Evaluation Proce·dures and Data Requirements-- Discussion iri this section centers on proposals for making project evaluation both more useful in terms of improving project management.and more meaningful in terms of permitting valid assessment's of project impact. Discussion and proposals are divided into two sections, one dealing with sampling and ..;. 82 - . '· data collection and the other with data processing and evaluation procedures •. 166~ (1) Sampling and Data Collection. Where a project is to have its major. impact through changes in agricul t.ural productivity, there appears to be no alternative to sampling crop yields~ Analysis of the LLDP experience of having evaluation staff visually assess yields indicates that even well-trained staff.are incapable of providing acceptable· estimates of yields using this method. Having accepted the inevitability of yield sampling, the question then arises as to how this. samplil)g can. most usefully be done .. 16 7 · At :LLDP, the annual yield surveys have been based on a new sample every.year and none of the farmers selected for .one season '.s sample is included in that for the following season. The· reasons for drawing· .a completely new. sample every year are not altogether clear; they may be based on.a. desire to minimize possible resentment on the part of farmers at being subjected to a battery of interviews year after year, and·theymay have their origins in the belief.that continuous . . use.of the same sample alters the sample to such an extent 'that it is no longer representative of the population from. which it was drawn.· 168. This report recommends that a certain proportion of .the sample selected for yield studies be retained from one year to another. This part of the . sample would constitute in ef;ect a panel which would be resurveyed annually in-order :to be. able. to assess some of the dynamic elements of the project. With present sampling procedures, there is simply no way to relate what happens to. a farmer in one season with what 'happe~s in the next, though the two are obviously not. independent. It had been hoped, for example; to test the hypothesis that the receipt· of cre'dit by a farmer in one. sea13on ;f:.s unrelated to whether or not he had credit in the ~- ._ - 83 -· previous season, or to his · leve.l of performance in terms of yi~lds and. production the year before. Because each season's 'sample is independent of that of.the previous season, the testing could not be done. A time series of data on the same farmers would also permit one to begin to make inferences about rates of improvement in yields and changes in input use, etc. 169. The above recommendation, then, is leading toward a second, ·i.e. that a limited amount of farm management data be collected in the annual surveys. Farm management surveys are notoriously expensive in terms of manpower, funds and time,.and the results are often not available until long after the data were collected. tncluding in the annual yield surveys ~;i few simple but informative questions could yield useful results in a much shorter period of time. Evaluation staff might' consider recording, in addition to yields, the use o;f selected inputs, sources of seaso~al finance, farmers' opinions as to their major problems, etc. Changes in variables such as these could then be monitored for that part of the sample which ~s resurveyed every year, 170. The method of selecting samples for the annual sm:yeys appears to need reconsideration. The present method makes use of aerial photographs to identify villages from which the sample is randomly selected. The major criticism is.that while the population sample is random the sample of villages is.not, because the technique selects villages with a J;>robability proportional to the amount of land the village controls. Thus there is ~n inherent tendency to oversample certain types of viilages. The LLDP Evaluation unit has. been aware of this problem for some time and is engaged in trYing to develop a more satisfactory approach. 171. A £urther recommendation is that part of the J;>rOJ;>Ortion o~ the sample which is to be replaced every year be drawn on a nonrandom basis·. The objective here is to r-------------------~-- ·I ·. - 84 - sample' explicitly those farmers who are participating in project programs and who.~an therefore be expected'to benefit ·most. It is s~ggested,· for LLDP, that the most appropriate sampling.frame to use would be the annual lists of project farmers who receive seasonal credit. Actual selection of farmers could be random but the population, of course, . is deliberately selected. With the present sampling technique., it· is difficult to make any sort:. of systematic evaluation of the effectiveness of the credit program because so many features of interest are underrepresented in the samples. 172. Two other recommendations are specific to yield StUdieS·. firSt 1 Where pOSSible 1 thOSe VariableS Which Can }Je measured in a· continuous form---amount of fertilizer use~r, disease infe.station { pest damage, etc. --should be so measured rather than by dividing the responses up into discrete categories, or recording simply the presence or absence of. a factor. Only continuously measured variables can be used in .analysis· in such a way as to 9enerate recommendations on changes.in practices, levels of application, etc . . By way of illustration, the analysis had hoped to make use of data on· fertilizer use to ascertain what· sorts of responses Lilongwe farmers were getting in comparison to those achieved on the· research station whi·ch serves .the area~ the data on quanti ties of.fertilizerused·are, however,· known. to be inaccur~te so that the analysis couid only make use of a dummy variable indicating whether or not fertilizer was used. As would be expected, the ·variable employed indicates a positive .and· powerful response to the use of fertilizer, ·but· it is incapable .of. revealing whethe.r farmers wer~ ·using too much.· fertilizer, too little, or the right amount. R9ughly comparal:)le.re~ul:ts are ·achieved with . variables ·defining weed infestations ' ·····, and· ·attacks by pest~ and diseases .(see. the various regress~on results in the. Annex). Ideally,· one would ·also like to know· the precise date when a given plot.was planted sothat time of p~antirig could be systematically related to rainfall and possibly other climatic factors, but·this information cannot 85 .,.. be,: considered reliable when collected in e·x post surveys. 17 3.. The second· recommendation specifically· relating· to-yield studies is that where evaluation staff find there has been a complete crop failure on a sample plpt, additional information should be collected. A plot·· where there has been ·a crop failure should be considered just as important asone where there is a standing crop, though for somewhat different reasons. · Evaluation staff should make some ·attempt to ascertain why the crop failed or why the farmer abandoned, the plot. Staff should also try to find outhow much the farmer was relying for subsistence or cash income on the crop. that failed and whether the farmer has experienced crop .failures on other plots .as well. 174. An additional comment relating to yield studies is in. 'the forin of a question: are the sample sizes employed for yi~ld surveys, now ~nvolving more than 1,200 farmers annually, really essential in order to obtain acceptable levels of accuracy. It is known that the .LLDP yield data are highly variable, so that fairly large samples are required to provide a reasonably reliable estimate of mean yields_; nevertheless, large-sample theory and common sense suggest that, unless highly det~iled analysis is going to be done for small geog- . . raphical areas (.i.e . project units) , the sample sizes currently used are larger than they have to be. Statistical tests are available which can indicate how small a sample one can get away with to obtain given levels of confidenc~ in the results, and these could usefully be employed to ascertain whether and to what extent sample sizes in the yield surveys_ canbe reduced. If some of this extremely time-consuming type of sampling could be eliminated, then data. analysis could perhaps be speeded up and other less-demanding,· ·ad hoc types of surveys could be mounted. 175. As a final remark under this subject, it was noticed that LLD? evaluation staff allow interviews to become .-------~--~---- - 86 - social·occasions, with friends and neighbors--and possibly strangers--gathering around the farmer being interviewed and listening to all ·the questions and responses. There can be no objection to this at tim~~ and the atmosphere of a social. event may actually assist in removing any latent suspicions about the motives of evaluation staff. In certain cases, however,. such open interviewing is almost bound to distort · the·answers received. Examples would include a whole range of questions relating to income, sales, capital, cattle, future intentions, etc. In cases where information is sought on matters which are not normally publicly discussed or wil~ingly revealed, it would be extremely valuable if some way could be foUnd of preserving the confidentiality o( the answers. 176. (2) Data Processing and Evaluation Procedures·. An absolutely essential part of any evaluation effort is the documentation of work in progress. This means the·thorough recording 1 at every stage. of the process 1 of objectives 1 · definitions, methodology, criteria, etc.--vl.rtually ·every aspect of the evaluation effort, from recognition of the need for a survey. to printing of the final report. There are .· principally two justifications for the recording of the minutia of evaluation work. The ftrst of these is to·ensure tha.t a particular piece of work 1 once begun I can continue regardless of changes in personnel or other mishaps which sometimes befall data collection and processing exercises. The other justification is to permit the secondary use of the data collected should further analysis be deemed desirable at any time after the original purposes of survey work have been served. On no account. should survey. and evaluation work be allowed to develop so that successful· completion of a given project is contingent upon the involvement of a particular individual. To a certain extent the evaluation effort at LLDP has been hampered by the failure to observe this precaution. - 87 - 177. LLDP.suffered for some time from not having staff in the Evaluation Unit with the relevant statistical training for the work which was. being undertaken. This gap was filled, albeit not completely.satisfactorily, by consulting with ~taff of the National Statistical Office and personnel in. various ministries andin the university~ Particularly weak was the use of statistics in reporting the results of survey work, and for a good· number of years the only statisti.c used was the arithmetic mean. This selection was unfortunate because it t~nded to obscure the unreliability of the mean as a measure of many of the variables for which it was calcu- lated, e~g. crop yields, farm sizes, incomes, etc. Improve- ments have been made in this regard at LLDP, and it.is now common practice to indicate with one or more measures the inherent variability in the data employed. A useful amend.;... men~ would be to include with means, where relevant, the . . confidence intervals for given levels of probability. There is still, moreover, a need to make far greater use of modal and median values for certa.tn categories. of variables where use of the mean conceals the nature of the distribution·. Modes and medians, however, are not always available options in canned computer programs( indeed, they were not available in the program used for analysis in the follow-up ·study. The ideal program for the analysis carried out would have · .been the Statistical Package for the.social Scienc~s (SPSS), an extremely useful and e;fficient set of techniques .for dealing with a wide range of data-handling problems. 178. It would also be valuab,le if more results could be accompanied :by graphical representation, such as scatter dia,grams or bar charts,. d~pictinc;y expecially significant or informative relationships. 179. There appears to be a need as well for less simple reporting of results and more rigorous testing of hypotheses and· exploration of relationships. Much more use needs to be. made of tests of signtficance,·both of a parametric and - 88·- ·I ' nonparamet.tic nature. It should.be stressed here that one is not striving simply for more sophistication in the evaluation effort but rather for more meaningful ~valuation. The .common practice of reporting results·in the form of complex crosstabulation. tables with no indication. as to whether.the numerous relationships shown are significant or not. is extremely confusing, and if these results are used at all, they are q~ite likely to be misused. 180. For most projects, if evaluation is to j~stify· the very substantial investment of resources required to run a gqod program, there should be a well-defined link between evaluation output and management input. During the first few years of.a project's life, there will be a need for· a considerable amourit of positive analysis, particularly if little.basic information was available before the project was launched. As a project matures, however, a need is likely. to devel()p for relatively· more analysis capable of Yielding· n.ormati ve implications. To take an example from this analysis, it was discovered that yields are very'low when crops are planted in a field that had been allowed tq lie·fallow the previous season. This finding contradicts the experience in many other part's of· Africa, and signifi~ cant benefits could result from identifying the factors which result in the low yields. Identification of the f~ctors responsible would then permit management to design an extension program aimed at correcting the situation. Another example would be the· analysis of crop response· to fertilizer in actual farming situations as contrasted w~.th research conditions1 such analysis could be .used by rnanage- ment.·to modi£¥ the blanket recommendations for levels of fer1=-ilizer application often made by resea·rch stations •. 181. As a general point, there seems to be a need to include in evaluation work at LLDP more in the nature of performance checks.on various project components. Why, fo'r. example, do extension staff not visit one-fifth of ·. - 89 ... project farmers who have taken seasonal credit packages? To the extent that LLDP is regarded in the Bank as a proto- type for a certain approach to rural development, there should be rather more evaluation of the effectiveness of various project components. 182. Some of the recommendations made above will be fairly easy to implement, while others will be quite difficult. But one_ implication all share is that greater _rather than fewer resources need to be devoted to e.valuation if the necessary work is to be done and done in such a manner as to. be worthwhile. 183. ·Many aspects of evaluation are highly technical, and it is impossible to treat them in detail here. There remain also many open questions as to what are appropriate approaches to the on-going and ·ex po·st evaluation of rural development projects. Much experimentation needs to be done, for example, into the relative merits of alternative techniques. of surveying to acquire the different kinds of information required for various:evaluation purposes. Such experimentation could profitably focus as a starting point on such issues as costs, scale, accuracy and precision vs. speed, each of which might be expected to have a quite different importance from.one project·to another. This kind of experimentation, however, seems urilikely to take place in existing projects such as LLDP, where large investments have al:ready been made ;in designing systems and training staff.· to use them and where staff are alre_ady overburdened with the -existing evaluation load. It seems probable, therefore, that no experimentation .will take place and that there will be no consolidated effort to improve evaluation procedures and techniques unless external resources can be provided. 184. The· ut·i'li'ty of ·a· F'O'llow-up ·Study--How useful has it been to attempt to answer a wide range of questions on the basis of data collected often.for wholly different - 90 _; purposes? What kinds of data would be needed to answer · satisfactorily the questions posed? These questions and ' others which are at the core of the follow-up study are discussed below. 185. The essence.. of this type of evaluation effort is that one is examining a socioeconomic system that has bee~ exposed to planned intervention in the form of a project, and one wants· to know (1) what has changed, (2) how much of a change there has. been and (3) why and how change has occurred. The particular thrust of this effort has been to determine to what extent the existing data are useful in assessing the impact in specific areas of the Lilongwe pr~gram, i.e. in providing useful additional· knowledge of the type specified in the three points above. 186. Broadly speaking, attention has been focused on finding out what has happened at Lilongwe; and why,_ insofar as the data available might be expected to provide some indications as to where and in what magnitudes change has taken place. Unfortunately the data do not encapsulate many of the changes that are known to have occurre·d, so that on~ can say very little in this respect • . Even less can; be said with regard to explanatory· factors, the "why" of ch_ange., even where changes are reflected in the data. 187. LLDP was conceived with objectives which differ somewhat from those currently stressed in Bank thinking on rural development. As a consequence, the evaluation effort has always concentrated the bulk of its effort on finding out what has happened to crop yields, particularly those of . . maize and groundriuts. What consensus emerges from the efforts made by the LLDP Evaluation_Unit and the other researchers who have examined the Lilongwe yield data? The· first point on which all agree is that crop yields aresubject to very considerable seasonal variations due to a whole range of factors, many of. whiqh are not subject to project influence. l I . ,! [ ". - 91 - Even allowing for these seasonal variations, there· seems to be general agreement that groundnut yields experienced a . declining trend for several years, ·though more recent indica- tions are that this downward trend may have been reversed. The reasons for this decline have yet to be convincingly documented, and the explanations ventured include, with varying emphasis, agronomic factors, such.as unfayourable rainfall and pests and diseases, inadequate price incentives, social factors such as groundnuts• being a woman's crop, and the· relatively greater importance and attractiveness of maize. This study adds to existing knowledge about what has happened to groundnuts.by showing that the drop in yields has not been general_but rather that the most adverse effects have been experienced by a distinct, identifiable group of farmers. Additional information has also been forthcoming regarding cropping practices which appear to have·importanteffects on grqundnut yields~ 18.8. Somewhat more controversial is the matter of maize yields. In this case, there appears to be no discernable trend in average yields, though this point is not universally accepted and is open to varying interpretation where it is acknowledged. Perhaps the st.rongest point of agreement here is .that the technology and extension package appears to be the. right one for the area and the cropping system; every. I . study of LLDP concurs that the improved seed-fertilizer package. together with the practices promoted by extension staff' can and have ·raised maize productivity dramatically where I 1 .. they have. been used. The studies seem to agree also that the I I ' biggest proplem is maintaining a satisfactory rate of expan- sion in the use of improved inputs. Virtually every document rel~ting to developments at LLDP cites the need to review pricing policy for the project's principal crops. (Price ..policy has been outside the scope of this report but it sho~ld be stated, even if only parenthetically, that there is full agreement on the necessity to maintain a notional parity between crop prices and production costs and world market ·. - 92 - prices~.) This need has become all the more urgent because of sharp increases in the costs of seed and fertilizer . .189. The principal source of disagreement is the question of what has happened to maize yields as a result of the project. It is known that many farmers have been able to · increase their yields to very high levels, but the problem-. atical point is what has happened to overall yields in the areas developed by the project. The problem appears to stem first from the various methodologies-used to assess average yields and second from the interpretations given to the various sets of results. As yet, no two assessmen~of yield levels·appear to have employed consistently the same approach throughout..· Several fundamental points should be borne·in mind in any re-examination of the LLDP yield data. First, the Lilongwe Plain is far from being homogeneous, either in terms of soils, rainfall, population density, crop producti- vity or a number of other features: thus the project can };)e expected to have a relatively greater or lesser impact in different areas. 190~ ·second, the project has expanded annually by some 20-100,000 acres and has taken in·each year areas of often widely different productive potential. For the first few 'years, this expansion appears to have been into areas of relatively high potential; currently, expansion seems to be into areas of relatively lower potential. The consequence of this pattern of development is that ' . comparisons across ' . . ' time between _crop yields in "developed" and. "undeveloped" areas are invalid if they also attempt to attribute any observed difference in yields to. the project • s impact •. 191. Third, there have been several changes in bounda- · ries and composition of some of the higher-numbered evaluation areas used by LLDP staff as the basic land unit· for yield· assessment work. As a result, yields cannot be compared across seasons for all evaluation areas with the - 93 - same number; what is required is f~rst to disaggregate part of the data by project units and then to recombine them into new conglomerates which are·fully comparable with those for other seasons. 192. Fourth, and related to the point above, what is required for a valid assessment of the project's yield impact is physically comparabie areas for which good baseline data exist prior, to inclusion in the project and for which data are available for a series of seasons after inclusion. Since representative and reliable baseline data do not exist for the earliest evaluation areas taken into the project, it will be some time yet before this condition can be fulfilled. 193. Finally, there are obstacles to obtaining fully indicative results even from the above approach. The "undeveloped" areas outside the project do not exist in a static and pristine state. Among the least mentioned benefits of the project are two types of spread effects. The first of these is generated by what might be called a combination of extension and demonstration effects around the periphery of the developed area. LLDP extension staff not infrequently visit areas outside the project boundaries and give advice to farmers in these areas. There is, moreover, a constant to-and-fro stream of farmers visiting within the project who take home with them new knowledge based on what they have seen and heard. The second type of effect stems from the fact that the Regional Agricultural Office has extension staff worki~g in these undeveloped areas, though at a staffing intensity lower than that within LLDP. The regional staff liaise prote~sionally and socially with LLDP staff and many of the former acquire new knowledge, an increased interest in farming communities and a_greater sense of pride in their work from these interchanges. All of these factors impinge so a.s to be9in increasing yields before the project takes over a, new area. - 94 - 194~ . This study has·not attempted a full reassessment of actual yield levels in relation to the targets spelled out. in ·prbject appraisal reports. Data were available from ~nly .three seasons, which differed markedly in natural conditions, ·SO it was planned originally to concentrate on identifying factors whichappear to have an important influence on crop yields rather than devoting additional effort where so much work has already been done. It is felt, however, that an appropriate time to attempt an in-depth reassessment will be . when yield data are available covering the full period 1969/ 70 through 1974/75~ examination of these data should permit a fairly .gooci evaluation of yields during the years covered by ·the. first two project phases. · 195. The approach taken has yielded·useful results in ' . te:rms of technical factors influencing.yields, with. the principal factor identified being crop variety. In the LLDP area, crop variety--particularly maize--is a reasonably good proxy for the use of fertilizer, so for most purposes the one factor can be said to encompass the other. Some attention was also given to the use of sulphur on groundriuts, but the data available come from the early stages of the sulphur promotion campaign and include only small samples of users. Further useful work could be done on sulphur using .the data frqm seasons subsequent to 1971/72. 196. The influence of natural factors was also examined, bu~ preliminary analysis using a wide variety of functional fo~ms and models failed to give·m~aningful results for the , I , • most important of these factors--rainfall. These .disappointing results are .attributed to the fact that recorded rainfall data are little if at all related to the amount of rain which falls on·plots where yields are sampled. A complete lack of specific disaggregated data on soils prohibited any analysis of this factor. Attention was also directed to pests . and di·seases, .and significant results were obtained for groundnuts though no~ for maize. An unsatisfactory aspect of the data in this I '·· .. '- 95 - v, connection was that no distinction was made between pests on the one hand and·disease on the other,·and specific pests and diseases were not identified. ·197 •. A range of husbandry factors was subjected to scrutiny, and useful results were obtained for such factors . as weeding, time. of planting, planting densities - and . the - .use of farmyard .!manure. Much more valuable results could have . been obtained had certain factors, such as date of planting and·amount of fertilizer used, been quantified or quantified more. accurately. For example, LLDP' s former eval~ation offlcer.advised against attempting to use the data indicating amoWlt of fertilizer used because these were known to be extremely unreliable • . 198. A thoroughly comprehensive study of yields at LLDP should attempt to take account of a range of economic variables which were outside the terms of reference· of this study. Certain socioeconomic variables, such as farm size and labor force, were considered but other important variables, particularly crop·prices and input costs,·were not. Such a compFehensive study would be beyond the capacity ofthe LLDP Evaluation Unit with present maripower.and financial constraint~. 199. In summary, then, the. sections of this study relating to yields differ from previous work in concentrating on why particular yields are what they are rather than on ·defining precisely what average level of yields has been attained. · The i;ormer should be of more immediate value in terms of project management no matter what conclusions are ultimately reached regarding the latter. More useful work could also be done on the "why" of yields if the recommenda- tions made in the earlier part of this section are adopted. rn· its original conception, crop yields are the essence of success or failure at LLDP, yet the data available for analyzing the determinants of crop yields are, in one I . - 96 - i ~ J i importlimt. aspect, a~ong the least satisfactory used in the st~dY• ·The strongest criticism is that very little analysis is possible in which results can be translated into economic terms t~at have meaning for the farmers concerned. After re~alysis .of all the yield data, the end·result is still all too oft~n only a yield figure and a number of factors,· each of which can be said to bear a positive, negative or indeterminant relationship with yield. Without supplemental analysis of the farm management data collected at Lilongwe~ it . is impossible ·With the data: available to derive valid . no·rmati ve reco.mmendations sui ted to the resource endowments of .LilongWe farmers. 200. . The data which were available and used in the attempt to assess various aspects of ~he credit program were ' ' in themselves .reasonably satisfactory, particularly wheri supplemented . . with additional data on incomes and.. yields. . A ;fundamental problem, however, is that much useful data. relating to aspects of the credit program never gets rec;:orded in project records (e.g~ who applies for credit). The major · difficulty in this part of the study arose because the ·samples were so. small that, though the expected relationships oft:en . emerged from the analysis, the results not infr'equently fai.led to reach acceptable levels of significance. In large measure, the small sample sizes are a consequence of (1) the· sea:;:;ons . ,,· included '· in the study and (2) farmers• having or usf~g several names, one of which they give to credit .staff and. another of which they give to evaluation staff. In many instances it simply proved impossible to determine that t'wo or more names belonged to the same person. The suggestion al~eady made of using project credit lists to construct a sampling ;frame would largely overcome problems of this type. JU1 .alte;rnative possibility would be to issue all project farmers with an identity card bearing a unique record nuinber . and personalized with a thumbprint to help avoid misus~:. $uch an arrangement need not· be too burdensome to project sta'fi; since a census of farmers must be taken in any event,.· .• .,.. 97 ~ and it might well be administratively useful in many aspects unrelated to evaluation. 201. .While the Evaiuation Unit carried out a broadly socioeconomic survey early in the life of the project, LLDP has never hadthe benefit of a competent sociological or an.thropological study. The need for such a study assumes considerable. importance because so much basic in·forrnation on the Chewa, the·principal .ethnic group_in this part of Malawi, does not exist~ A number of well.,..qualified researchers have been keenly interested in carrying out field· studies in the project area, but such work was evidently accorded a low priority . by project management and the studies have ' . never . been done. Both the ARDS .review of LLDP and the follow-up st:udy.have been hampered by a lack of this kind of background information on the target population, particularly. when· at.tempts have been made ·to assess the land registration ' " . . ' . program or when analysis tackled areas relating·to poverty and: income distribution. Throughout the life of the_ project, ec.onomic and technical factors have received the dominant ·stress while social fact~rs have been relativeiy underplayed; there must be concern that lack of the latter type of know- ledge can 'lead to serious misinterpretation, to-underemphasis of; the significant and overemphasis of the trivial. It is not suggested that work of this nature be undertaken by staff of the Evaluation Unit, but ·rather that it could appropriately be undertaken by individuals or a team f~om the Universlty ' : 9f Malawi or, failing this, by staff of another suitable · university department or institute. · .202. The study also included a substantial number. of questions X"elati~g to incomes in the Lilongwe area. In ·large mea$ure, this focus was an afterthought and the 'result o't,the recent change of emphasis in official Bank thinking toward.income redistribution and assisting the poorer strata of rural soci·eties. The only data on which an examination ot income levels could be based were collected l~ng before . - 98 - the Bank's.change of emphasis and for the·purpose of ide.ntifying creditworthy farmers. To the extent that there wa~ any bias, it was in favor of. selecting a relati~ely higher income group in the belief that its members would be · more likely to repay a loan. The data were nevertheless useful as a first approximation in permiting an examination of ~ range of factor' associ•ted with both agricultural and nonagricultural incomes. The·original analysis done at LLDP •. ·had not stressed incomes nor had it sought to examine the correspondence between income from different sources and other. factors. A second al)alysis was therefore able to . ext·ract meaningful relationships from the .data which had . .·simply not been of high priority in the original analysis. 203. Perhaps the most important outcome of re-examining the income'data is that we are able to begin to identify the rur.al poor in the Lilongwe area and to learn some of th~ir · cha.racteristics. The greatest deficiency ·in this part of· the· analysis is not the unsuitability of the data but rather the.· fact that the data are so old. The survey work was done in . 1969/70; and LLDP has initiated many changes since. that time. So what the re-analysis gives,. then 1 is information on the determinants of income before the project could have '!. had much of.an impact~ What is most.needed now, or· in the. £u~ure~ is a replicatio~ of selected parts of the original su~:vey iri those areas that have been ·under project admini- . str.:ation f9r the longest time. 204. The analysis also produces evidence that certain . pro:grams inciuded in the project have operated in such a way that they cannot have hadthe beneficial impaCt·On the poorest part of the target populationthat is implicit in the . · Bank's new emphasis. It is shown, . for example 1 that a significant proportion ·of the target population is ineligible for credit in the form it has been made available in the past. Si~ilarly, extension staff are shown to have concentrated 'their efforts fairly heavily among farmers who have credit - 99 - ', and who, in the Lilongwe context,. farm relatively large· areas of land. Project staff, .however, are very responsive, , . ' . ' and '·. numerous changes have already.been effected in the credit I ' 1. program which have some potential at least to increase the equity of 6redit distiibqtidn. 205. In conclusion, the follow-up study could have ,had . . Potentially greater utility and have been completed much sooner had certain organizational problems been anticipated '! and avoided. As these problems have·implications for other ·similar studies, it is. worthw.hile indicating their nature •. The initial problem was that the time required to complete data. editing and analysis was severely underestimated. In large measure, this was because the team responsible for the anCJ.lysis had not also been. involved in .the collection of the . data and consequently had to devote much time to basic familiarization work. Additionally, the data were. in much .Poorer condition.than expected, so that more editing and checking of validity had to be done than was expected. The . . ti:me. taken in this basic groundwork then meant that .discon- tinuities inevitably resulted in phasing the work of the " ·. st.aff involved with the. analysis, a problem which had b_een paralleled in earlier experience at LLDP. Finally, the testing of the original set of hypotheses generated new · hypotheses, only a few of which could be explored iri_depth. because of the time c::onstraint. 206. The experience acqui~ed in carrying out this study suggests some guidelines for.similar efforts in the future. First,· it appears ·to be absolutely essential to have some participation on the part of the staff who were concerned with the original design and execution of the survey work and who have some experience with analyzing the data •. Secondly, in general no large~scale 'analysis of. data such as that reported here should be initiated unless it is' possible to ensure (1) that facilities for analysis, whether computers or otherwise, are adequate and (2) that there 100 .... ~i~l be no ~erious discontinuities in the p~rsonnel invdlved :with the .study. These guidelines are equally as applicable to relatively smet.ll-scale primary sttidies carried :out in .· projects ·or. reseet.rch units as to larger secondary studies performed at the·Bank or in other centralized facilities. There is a very powerful learning· curve asso.ciated with a $t.udy such. as this one; delays--whether avoidable or not, · inadequate pr~cessing facilities or access to facilities,and changes in Staff involved with the study weaken the learning .·curve and .dilute the value of. the final product. · It is . suggested,. therefore; that the best appr.oach to .a large-:-scale . · study is to assemble a team or task force which is to have ' ; as its principal responsibility'completing the study within a rea~onable . . .... period .of time,· and the estimate of time required should ~e on the generous side. In a recent major study '• of another Bank project in Malawi I an· interdisciplinary ' ' team approach has been employed, making use of both primary a11:d sepondary data, and the ·preliminary results 11appear to ~. corroborate the value of.. such an approach. 1/ Reported in '1 Economic and Sociological Issues in the Development of the ·Lower Shire Valley", by D. R. Colman and G. K. Garbett (University of Manchester) •. ANNEX :. •'' ~ l?~ge 1 r STATISTICAL RESULTS FOR REGRESSION MODELS This annex presents the statistical results for a ·number.of regression models, some of which are referred to in the main.body of the report. Results are reported by crop and season. ·The reporting format includes the constant for .the equation . and the coefficient for each of the indepe~dent variables; in parentheses below the value reported for each I of·these items is the standard error for the value. 111 the equation statistics, the R2 values reported are corrected for ' !. I degrees of freedom,·while SE represents the standard error of the .estimate. . ··In parentheses below the F value are its para- ' meters, i ~e. the number of·. degrees of freedom and .the relevant nulnber of observations. Because variable·definitions differ somewhat·in the data from the.different seasons, results are not strictly comparable across seasons. There. is, however, · greate.r comparability between results for 1970/71 and.l971/72 than · between either of these. seasons and 1969/70. Particular note should be made of the fact that the definition of labor force . ' in ,Model IV is si·gnificantly different for each of the two seasons for which results - are available. . The. results . - presented .refer only to theLLDP.developed area for each season •. As would·be expected, multicollinearity poses problems in certain of the results and is strong enough.in some instances to bias the outcome. Listed·below, by model, are the variables where the correlation coefficients exceed an absolute value of o. 5 (See the. notes to the individual models for further details): . w ANNEX Page 2 Correlation Model Variables and seasons coefficient II Maize: 1970/7.1 - Time x Population -0.55 1971/72 - Time x Population -0.55 Groundnuts: 1970/71 - Time x Popn -0.84 1971/72 - Time x Popn 0.78 V 1970/71 - Time x Population -0.55 Time x Fertilizer 0.77 1971/72 - Time x Population -0.55 Time x Fertilizer 0.58 VI 1970/71 - Time x Population -0.84 1971/72 - Time x Population 0.78 Summarized briefly, the implications of these correlation coefficients are as follows: For both seasons, the relationship between late planting of maize and plant population was negative and. equal. For groundnuts, late planting was negatively associated with plant population in 1970/71 but positively associated in 1971/72. The association between time of planting and plant population is stronger for groundnuts than for maize. There is a positive association between late planting of maize and the use of fertilizer, though the degree of association is weaker in 1971/72 than in 1970/71~ (: ""' ANNEX I"~, Page 3 ;.. STATISTICAL RESULTS' FOR REGRESSION MODELS MODEL I: CROP "YIELDS AND EXTENSION VISITS "The dependent variable is yield in pounds per acre and the independent variables arecthe number of extensi9n visits received during the season in question •. · Crop & . Independent variables Equation statistics Season __..;~;;=..;;;.;.;..;....._ Constant · · 0 · 1-5 · 6.;..10 11-20 R2 SE F .MAIZE: 1970/71 1,381 -538 -466 -218 . . -74 0.05 944 4.47 (91) (146) (156) (221) . (197) {4 i 278) 1971/72 1;923 -951 -444 77 160 0.09 1,218 7.75 (127) {211) (.211) c (2 ~1) . (210) (4,284) GROuNDNUTS: . 19..70/71 551 ... 364 -.150 -234 -178 0~ 28 . 255 23.87 (28) . ( 38) (57). (79) c( 74) (4,~236) 19ll/72. • . 394 -240 78 84 83 0.19 293 21.31 {31) (40) (53) (59) (57) . (4,345) .. MODEL II: CROP YIELDS AND AGRICULTURAL PRACTICES Two practices were considered: the timing .of planting (with five planting periods identified by dummy variables from one to five, with one being very early and.five very late planting7 other values are inter- mediate) and the density of planting (as measured by plant stands at the time of harvest) . Crop & Inde~endentvariables :Eg;uation statistics . 2 Season Const.ant Timing Plant ·;eo;eulation R SE F MAIZE: · 1970/71 135 . -160 0.1148 0.24 845 44.68 (166) (115) (0.0163) (2,280) 1971/72 ..;102 -47 . 0.1853 0.28 1,082 ',s6-. 49 (232) '(163) (0.0212) (2;286) GROPNDNUTS: 1970/71 . 216. ":"104 0.0230 . o. 56 19.9 154.59 (50) (26) (0. 0038) (2 ,238) 1971/72 25 10 0.0300 0.43 245 132.94 (25) (21) . (0.0030) (2,347) '1, ' 'c' ANNEX j- Page 4 .., " MODEL III: CROP YIELDS AND CROP VARIETY The depend~nt variable is yield in pounds per acre and the independent variables are crop varieties. For-maize, varieties are grouped by type of .variety· so that results may be comparable across s~a.sons. · For grollJ').dnuts, only one variety-~Chalimbana--is distinguished; all other named varieties are grouped together, but mixtures-of varieties are considered separa~ely. Crop·& Independent variables Equation~tatistics . 2 season Constant Local Synthetic Mixtures Hybrid ~ SE _F__ . MAIZE: 1969/70 0 1,522 1,803 3,587 1,579 0.52 744 38.08 (138). ( 16 3) ( 220). (360) ·. (273) (4,132) 1970/71 0 . 1,041 1,060 0 2,003 0.22 857 20.31 (202) (212) (236) (0) (2 38). (4 '278) 1971/72 b 1,295 2,036 0 2,651 0.23 1,120 22.69 (3/11) (326) (330) (0) ( 353) (4 ,284) GRQUNDNUTS: Chaliri\bana Other Mixtures 1969/70 0 814 889 741 0.70 .. 256 180.56 (24) (44) (85) (42) (3,225) 1970/71 66 . 424 314 422 "· 0.43 ~27 60.34 (26) (39) (229) (34) (3;237) 1971/72 20 -455 191 383 . 0.37 •25~ 68.10 (26) . ( 33) (260) (38) (3,346) MODEL IV: CROP YIELDS AND FACTORS BEYOND PROJECT 1 S CONTROL The dependent variable is crop yield in pounds per ·acre and the independent variables are labor .force (defined as available workers aged 15to 65) and farm size (defined as total cropped area). . crop· & Independent variables Equation statistics . 2 :. season · 'Con·s·tant Labor force Farm size R SE .F --'-~- MAIZE: 1969/70 655 118 13 0.02 998 4.56 (131) (52) (22) (2,285) 1970/71 764 3 49 0.04. 836 13.63 (78) (2) (13) (2,644) ...,, ANNEX Page 5 "' MODEL. IV, ·Continued. crop & IndeEendent variables · . Eg:uation statistics season Constant· Labor force . · Farm size R2 . SE F GROUNDNUTS: ·1969/70 199 -32 60 0.13 . 422. 34.07. . (41) (15) (7) (3,456) 1970/71 247 0.72 28 0.07 307. 25.55 (28) (0.64) (5)· (2,606) MODEL V: MULTIVARIATE ANALYSIS OF MAIZE YIELDS ·The dependent variable is maize yield in pounds per acre and the independent variables are: (1) plant population per acre--"Popn", (2) a.dummy variable (0 = no, 1 = yes) indicating whether or not f.ertilizer . was used--"Fert·", (3) a dummy variable (0 = none or light, 1 = heavy) indicating the degree of weed infestation--"Weeds", (4) a dummy variable (0 =none or iight, 1 = heavy) indicating the degree of infestation. bypests and disease•-"Pests", and (5) a dummy variable·. (0 =before the main rains, 1 = after the main rains) indicating; time of plan.ting;- "Time". IndeEendent variables · Eguation statistics Season Constant POE!L_ Fert Weeds Pests Time R2 SE F 1970/71 60 0.127 614 -358 -68 -862 0.39 754 37.37 (:].63) . (0.015) (95) (100) (152) (142) (5,277) 1971/72 -1. 0.169 503 -603 -138 -620 0.39. 997 37.44 (24 4) (0. 020) (104) (140) (237) (174) (5,283) MODEL VI: MULTIVARIATE ANALYSIS OF GROUNDNUT YIELDS The dependent variable is groundnut yield in pounds per acre and the independent variables are: (1) ·plant popu- lation· per acre..:-"Popn", (2) a dummy variable (0 =no, 1 = yes) indicating whether or not ,farmyard manure was applied to the crop--"FYM", (3) a dummy variable (O = none. or light, 1 = heavy) indicating the degree of weed infestation-"Weeds", (4) a dummy variable (0-"'7 none or light, 1 =heavy) indicating·the degree of infestation· by.pests and diseases--"Pests", and (5) a·dummy variable (0 = before the main rains, 1 = after the main rains)· indicating time of planting--"Time" .. l:. ' ' \ ' ' j. I ' ANNEX ~ Page 6 MODEL VI, Continued •. IndeEendent variables Eg:uation statistics Season Constant Pcipn FYM Weeds Pests Time R2 SE --- F· 1970/71 317 0.020 -12 -117 -277 -150 0.66 174 95.97 (45) (0. 003) (175) (45) (35) (24) (5,235) ' i___ .: ' 19,71/72 34 0.021 378 -33 -309 112 0.59' 207 103.46 (.21) (0. 003) {lOS) ( 38) (29) (20) (5,344) I. .·.' . l ~ - ,--~ -
Groupe de la Banque mondiale · Working Paper
Broadening the evaluation of rural development projects : final report on the Lilongwe land development program follow up study
Voir le document original
Le texte intégral est hébergé par l’organisation qui le publie. lawenc.com indexe les métadonnées et renvoie vers la source officielle.
Texte intégral
Informations clés
Organisation
Groupe de la Banque mondiale
Type de document
Working Paper
Pays
Malawi
Source
Banque mondiale