Fluctuations of maize and groundnut yields in the Lilongwe land development program

CONFIDENTIAL FLUCTUATIONS OF MAIZE AND GROUNDNUT YIELDS IN THE LILONGWE LAND DEVELOPMENT PROGRAM by Dennis Anderson Series: Studies in Employmenif and Rural Development No.28 Division: Employment and Rural Development Department: Development Economics Development Policy Staff International Bank for Reconstruction and Development The views expressed represent those of the author alone and not necessarily those of the Bank. This report may not be published nor may it be quoted as representing the views of the Bank and its affiliated organizations. Washington, D. C., June 1975 (Re-vised November 1975)  TABLE OF CONTEN;TS Lag e No SUMMAItY ACKNOWLEDGENENTS I. INT )l)UCTIO4 1 II. GENEIRAL MOVEM1-ENTS IN -LAIZE AND GROIUNDNUT YIELDS 2 III. I4AIZE-SPECIFIC FACTORS t4 (i) Spread Effects 4 (ii) The Gains from Husbandry 6 (iii) Fertiliser Shortage and Price Increases 8 (iv) M4aize Prices 10 IV. CGRUIMDI-UT-SPECIFIC FACTORS 10 (i) Improvements in Husbandry and Seeds 10 (ii) Groundnut Disease and the Decline in Yielcb 13 (i.ii) Groundnut Prices 15 V. GLIHI-AL FACTORS AFFECTIG YIELDS 15 (i) Rainfall 15 (ii) Mew Areas Coming into the Program 17 (iii) Migration 17 (iv) Samplin- Errors 17 (v) Baseline and Target Data 18 (vi) Variation of Yields Among Farms 19 VI. C0)NICLUSIONS LIST OF ANNEXES 1. SAMPLLE SIZJE 2. LZAIJIFALL, I)ISEASE AND GROU4DI'fUT YIELDS 1969-73 3. COMJhU'S 01.1 SlTATISiICAL ANlALYSIS 1X. I4(MONITORIMIG AWlD EVALUATIOIN AT L.L.D.P.  3U-141¶ARY (i) The report looks into the reasons for the mEasured fluctuations in maize and groundnut yields. It finds that groundnut yields were set back over a three year period by the growth of disease (rosette), which by 1972/73 afflicted over 50% of the crop. Now that the problem has been overcome, yields are recovering quicklly and should exceed expectations (though rust poses a new threat). (ii) Maize yields, after a good start, have recently been set back by high fertiliser prices without commensurate movements in crop prices. This is causing fairmers to use much less and inferior fertiliser, thus threatenin- good yields. Prizing policy for maize, and for crops in general, needs to be reviewed. (iii) Rainfalls and other feactors have also been a source of fluctuation, as one would expect. Rainfalls have not been good in recent ycars) and it is estimated that poor rains in 1972/73 cut yields by 20U as compared to the previous year. (iv) Even with these set backs, however, yields are higher than before the program, higher in the developed areas, and higher than what would have been achieved without the program. The reasons are that husbandry has improved remarkably - the improvements have also spread widely to the undeveloped areas - and, until recently, maize farmers were moving to the high yield seed-fertiliser packages. The response of the farmers to the program has been eniormous; hopefully the problems now occurring with fertiliser and crop prices will not be allowed to undermine the response. (v) The rcport is based on material produced over six-: years by the Lvaul'+io.l Unit at Liongvw . Comments on some of their difficulties ar c al so oA':.r.-d in thc annexr%es of this report. AC}IlO'TL qLGLI4,1ITS Many th'riks .ire due to TJ'ony,r Standen, Brian Wilsoin, Elwyn Mornda, and to n1i.ny L.L.D.P. staff for help and discussions.  ('') i'l port loof_l Tylln ,h re'risons for toL m.E:.sured fluctu:i GLons 11 in Inl:-e irti( ;rr,niindrnu.L y1olda. J L.finds that gruundrnit yields were set boc.. overs .i 'iLrce yerLr perioo by the growth of disease whichi by 1*Q'(/Y3 afflicted over 50% of the crop. Now (rosette), that the problem h1U-S be(en oV(!rCOm(e, yields are rr-coverring quickly and should exceed exp,c U Lions, (htlough rust pos8'S :, now -threat). (1i) -4a,k_ yields, after :: good start, have recently been set by high for'dliser prices without commensurate movements back in crop prices. This is c;tusing fr,.rmers to usa muchl less and inferior fertiliser, thus thr-eatening good yields. Pricing policy for ma-ize, and for crops in general, needs to be reviewed. (iii) Rainfalls and other factors have also been a source of fluctuation, as one would expect. Rainfalls have not been good in recent years, and it is estimated that poor rains in 1972/73 cut yields by 20/%' as copmpared to the previous ye .- (iv) Even with these se; backs, however, yields are higher than before the program, higher in the developed areas, and higher than what would have been achieved without the program. The reasons are that husbandry has improved remarkably - the improvements have tilso spreaid widely to the undevelopud areas - and, until recently, m:liz( farmers wore moving to the high yield seed-fertiliser packages. The response of the farmers to the program has been enormous; hopefully the problems now occurring with fartiliser and crop prices will not be allowed to undermine the response. (v) Thevc report is basi-d on mnterial produced over six years the I v,Tlur#tio LUnit at Lilongw, . Comments on some of by their difficulties ralso offered in the amnnexs of this report. !iP3 ACK11OWLEDGEMIENTS Miny thanks are due to Tony Standen, lIrian Wilson, Elwyn Mpondr., and to rl.an,y L.L.D.P. stalf' for help and discusslons.  FLUCTUATlONS OF MAIZE AND GROUNDNUT YIELDS IN THE 11LONCGWE LAND DEVELOPMENTT PROGRAM I. Introduction 1.1 The Lilongwe Land Declopmont Program (L.L.D.P.), which began in 1968, was the first of a series of comprehensive rural development schemes initiated by wiIe Malawi Government to modernise agriculture and raise the economic output and standard of living of the nationts smallholders. The program extends to a population of over half a million, and to an area of approximately 1.2 million acres, of which 900,000 are cultivated by over 100,000 smallholders. The first two phases of the program, the construction works for which are now complete, consisted of cutting crop extraction and feeder roads into the area; soil conservation works; establishing market and storage facilities at local centres; extension services and training courses for farm families in farm practice and household economy; the provision of seasonal credit for fertilisers, hybrid and synthetic seeds, and of medium term credit for farm implements and stallfeeders; land re-organisation and registration; and managerial training and research and evaluation services at the program's headquarters. The total program cost is estimated to be about $30 million (1974 prices). The first two phases, which covered roughly two thirds of the program area, concentrated mainly on the three traditional crops, maize, groundnuts and tobacco; construction for the third phase, which has now begun, is notable for adding health facilities to the infrastructure component, and for expanding the livestock, poultry and dairy aspects of the agricultural component. 1.2 The following report investigates the measured changes in the yields of the two principal crops in the program area, maize and groundnuts. When the program began, it was projected that maize yields would steadily increase from 1100 up to 2100 lbs per acre, and groundnuts from 500 to 650 lbs per acre, as a result of improved husbandry, seeds and l'ertilisers. However, while measured maize yields at first increased quiclcly, they IL.ve since declined quickly, while groundnut yields declined for several years in a row after the program began. This has pose?d somcthing of a paradox for many visitors to the program area (including several Bank missions), since it is obvious that husbandry has improved considerably in the area and new seeds and fertilisers are now much more commonly used. Consequently, opinions are still divided about the yield rneiasurements: it has been variously suggested that the me,asurtLmcnts uunreli.able, or that the baseline data were unrealistically high, or simply thn t analysis has been faulty.  -2- 1 .3 The report is based on material provided by the evaluation unit at Lilongwe during a visit by the author in June. One aim of the visit, besides looking into the above paradox, was to discuss with the unit various problems they have encountered in monitoring and evaluation; comments on these problems are to be found in the annexes. II. General Movements in Maize and Groumnt- t Yields 2.1 To begin, there are some encouraging reports that this year groundnut yields appear to have increased considerably. We have made some rough estimates based on the early samples that have come in from the field. It appears that this >war T s yields, from areas where the program is fullyestablished, may be something like 670 + 75 lbs. per acre; that is, 40% + 16% up on last yearfs yields and in excess of the projections made for 1979/80. I wish, of course, to make clear the tenxtative nature of these resulJ's; buit if they are roughly correct, there is at least nothing paradoxical about the measurements this year. 2.2 As regards the maize, and groundnut yields in previous years, the general concern seems to be about average yields in the program area, which were measured to be as follows: Table 1 Average Yields in Program Area, As Compared with Projections Shelled M4-tizeL Lbs/Acre Actu al Projected Act/ProJ. % 1969/70 1023 1263 81% 1970/71 1180 1332 89% 1971/72 1353 1384 98% 1972/73 1100 1449 76% 1973/74 1141 1517 75% 1979/80 - 1998 - Shelled Groundnuts, Lbs/Acre Actual* Projected Act/Proj i 1969/70 550 531 104% 1970/71 520 541 96% 1971/72 409 552 74% 1972/73 313 563 56% 1973/74 410 574 71% 1979/80 - 647 - * Not S.1lowing for plots that were abandoned; this matter is discussed separately below.  -3 It wolcd thus seem that groundnut yields have decreased, and that maize yiclds have remained static, with the exception of 1971/72 (whichi was a year of good rainfalls). 2.3 However, most of the evaluation unit's sampling has been outside !rc:,as so far developed (though in the program area), thus providing: - benchmark data for all areas developed or to be developed after 1969/70 (i.e. r'or about 90% of the program); - a basis for assessing the program's impact by reference to aLreas so far not developed; - data for analysis of the "spread" of benefits to outside areas. Up to 1974, the sample in the developed areas of the program varied betiween one sixth and one third of the total sample; hence average lfiguros are unrepresentative of the program's intended effects on yields. 2.4 The actual yields in the developed and undeveloped areas of the program were as follows (a fuller listing by area is provided on Table 3): Table 2 Average Yields in Developed and Undeveloped Areap Maize Lbs/Acre Developed Undeveloped Ratio 1969/70 1071 1010 106% 1970/71 1299 1152 113% 1971/72 1845 1115 166% 19'7/73 11143 1008 113% 1973/714 1236 988 125% Groundnuts, Lbs/Acre Developed Undeveloped 1969-/70 522 588 89- 1970/71 1490 540 91% 1('i7/72 1431 397 109% 1972/73 323 306 1 06 1973/74 1477 389 123'I Source: See Table 3.  The d-vcloped areas, it can be seen, have gained steadily on the undeveloe;d areas; it will be seen below that the gains would have been more marked had not the benefits of the extension services, new secd varieties and fertilisers spread significantly atre.ls. It is true, however, that (measured) to the um.'I.veloped yields have generally been less than hoped for - though maize, in 1971/72 exceeded forecasts by 33', and this year groundnuts may give good yields. It can also be seen (from Tabl s 2 and 3) that yields in all areas fluctuate considerably from one year to the next (thus .- ing it difficult surgesting that there may be some random, exogenous to discern trends), factors at work. 2.5 These points are now examined in more detail, at some maize and grouhidnut specilic factors, looking first and then at some general faectors. III. Maize Sp2ecific Factors (i) read Effects 3.1 A measure of the spread effects can be found of farm families who have gained access to new in the numter seed varieties and supporting fertilisers before the porgram has reached their areas. T'he proportion has declined in recent years, though it remains significant: % of Farms Samples in Undeveloped Year Areas Using New Seed Varieties* 1969/70 2Ly% 1970/71 19% 1971/72 7% 1972/73 n.a. 1973/714 10% See Table 5 3 In addition ,the evaluation unit's surveys slhow that husbandry improvements have also spread to undeveloped areas - in ridging and soil conservation, weeding, planting at the proper times and more densely, rotating crops, and a number of other factors.-i j Documented in the E.valuation Unit's Studies on Yield and M-Ianagerial Practices, 1970/71 and 1971/72.  Table 3 Annual Mlaize and Groundnut Yields (Lbs/Acre)/ - EIva 1:ation Area Dy 1969/70 1970/71 271/72 1972/73 1 973/74 474/75 Groundnu ts: 1 490 470 298 205 524 2 530 360 363 197 454. 3 62 0-,*i 520* 515 249 401 It 690**, 580* 455 417 416 690 5 550 4 480-* 367*9- 31 4* 404 6 61 O^-i- 478** 326-6 31 7* 7 480* *,* 309**^ IL22---r 341* 8 591 ,-, 191 *-* 389: 9 335* 262* 448** 644***- 10 259- - 193** 326i,r 11 3103-,. 370-'^' 12 349-** 310* - 13 287** 486** 64I Averages :- Deveiloaped Areas 522 490 1431 323 1427 ,laoped Und c-% Areas 5J8 540 397 306 389 Tot.l Lnarple 55,0 520 409 313 410 672 (+ 75) 1 1190 1330 1631 1308 2 970 1230 1628 1362 3 900--: 1110* 1845 1134 4 11 50'--* 1290*- 2038 1203 5 1o060*, 1l40O.:- 102G 1291 6 1100)H-- 12994**- 1197*$ 7 11 76*e*- 906Xo,- 8 1257-"-' 1081- * 9 876*6 1019489 10 9663'E' 803P,.,- 11 1 024'*'' 12 1112* 13 10032* Averages:- Developed Areas 1072 1299 1845 1143 1236 Unduveloped Areas 1010 1152 1115 1008 988 Tot!Ll Sinple 1023 1180 1353 1100 1141 Source: Evaluation Unit Rleports, Lilongwe j i.e. lbs. shelled maize and groundnuts * Units in transition from undevcloped to developed ** Undeveloped units v* Preliminary estimates based on early samples  -6- 3.2 It seems reasonable to suppose that, in the absence of the prour.m, nmi:'c yields would have been no more than is now obtained by farms using local seeds in the undeveloped areas; in fact, they might 1hiave been less, since husbandry practices would have been inferior. If hlis is So, then a proper comparison of yields " 7with and without" the project would be as follows: Table L4 Maize Yields (Lbs/Acre) in Devlopud Areas Compared With Yields in Undeveloped Areas, Assuming Local Seeds Only for the Latter*k Developed Areas Undeveloped Areas Gain, % 1969/70 1071 929 10 1970/71 1299 1086 20% 1971/72 1845 1074 72% 1972/73 1143 963 19% 1973/7h 1 1236 896 38% ' See Table 5 for the data Thus the gains have varied between 15 and 72% since the program began, amounting to about 35%, on average. (ii) The Gains from Husbandary and Technology 3.3 The gains in the developed areas have two sources - better husbandry, and a more extensive use of n-t seed varieties and fertiliser. Thus the data on maize yields, presented more fully in Table 5, show that, in all statistically significant cases since 1970, the yield on any seed variety is greater in the developed area (indicating better husbandry); and, of course, the new seeds are more commonly u5ed. The information for 1971/72 show this the most clearly (the differences were larger): Table 6 1971/72 Maize Yield Data (Lbs/Acre) Developed Area Undeveloped Area Seed E Farms Yield aFarms Yield % Gain Local 49 1351 93 1074 26% ) Weighted Average Synthetic 36 2169 4 1414 53% ) using seed mix Hybrid 12 2904 2 2324 25% ) of undeveloped area M1ixed 1727 1 _45% ) 27% 100 1845 100 1115 65%  -7- Table 5 S&urple Data onlize Yields (Lbs Acre) Developed Aruas Undevo dAeas Averages .)' d Varicty No. Fa,rms /O YiclrI No. Farms % Yield NIo. Farms % Yield 1969/70:- Local 197 84 1045 360 76 929 557 79 970 Synthetic 15 6 116)4 65 14 1208 80 11 1200 HIybrid 5 2 2289* 3 1 4 4154 8 1 2950* Iixed 18 8 978 41 9 1038 9 -9 1020 235 100 1071 469 100 1010 704 100 1023 1970/71: Locail 93 75 1110 454 81 1086 547 80 1090 yn Lhr tic 14 11 1260 68 12 1127 82 12 1150 Hybrid 11 9 2991 23 l4 2546 34 5 2690 Mixed 6 5 1190* 18 3 1177 24 3 1180 1214 100 1299 5o3 100 112 687 100 1170 1971/72:- LocEal 162 49 1351 632 93 1074 794 78 1130 Syntlletic 119 36 2169 28 4 1414 147 15 2025 Hybrid 40 12 2904 14 2 2324 54 5 2754 Mixed 9 3 1727 7 1 1 194* 16 2 1494 330 100 1845 681 100 1115 1011 100 1353 1972/73:- Local 936 963 yn thltic 1152 1323* Hybrid 2214 2358* Mixed 1521 1 539* 1143 1008 1973/714:- Local 202 64!. 925 180 90 896 382 74 911 Syn-thetic 45 14 1363 7 4 21 70* 52 10 1472 Hfybrid 60 19 2207 5 3 1843* 65 13 2179 Mixed 11 3 1127 5 1980* 16 3 1394 318 100 1236 197 100 988 515 100 1141 Sources: Various reports and records of the Evaluation Unit, Lilongwe Jj i.e. No. of farms in sample. * Note that these yield estimates are based on small samples (less than 10) and so are not necessarily statistically significant.  -8- Th}- wc-cihced aver-age gain, takcing the seed-mix to be as in the undoveloped areas, provides a meonoure of the gains from better husbandry (since thus we have held sced-mix and presumably fertiliser input, constant,, and random differenc( tend to cancel over large Li.am,les); the gain amounted to roughly 27%. The remainirg portion ot lt overall gain of 65;% came from improved seed and fertiliser mix. ln round numnbLers, there was a gain of about 25% from improved husbandry, and 40% from the new seeds and fertilisers. 3.t4 The major gains from improved husbandry have stemmed from highler planting densities. Yields can be doubled if the density is incre:,sed from 8,500 plants per acre (which was the average at' the Oegining of the program) to the ontimum of 14,500. This can be seen fIrom the results of the sample surveys (which check well with those of controlled experiments): Table 7 -aize Yields Lbs A-re as Function of Plant Densities Plants/Acre) 8000 10000- 12000 14000 Plant Dernsities: <8000 10000 12000 14000 16000 >16000 1971/72: Deerloped Areas 1062 1578 2077 2226 2734 2700 Undeveloped " 636 1541 1431 1520 1830 1550 1970/71: AvErf,qe 690 1120 1410 1790 1800 1400 Source: Yield Studies for 1970/71 and 1971/72 3.5 A l .rc,! effort has been made by program management and the extension wvorkcrC; to increase plant densities. In the developed areas, plant denrities increa'sed on average from 8,500 plants per acre in 1969/70 to something over 12,000 in 1973/74. In the undeveloped areas, aver,ge denisities increased to over 10,000 by 1973/74, possibly as a r; sult of the efforts of the extension workers (20% of the farms sampled in undlevelop,d areas in 1971/72 were regularly visited by extension workers). (iii) JFerl,ilise:r Shortages and Price Increases 3.6 While Lh, gains through improved husbandry have steadily incretsed e,nd can be expected to continue to do so, the gains from inproved se-dAs zand fertilisers have been seve-rely set back since 1971/72 by incresing shortages and prices of fertilisers; in conjunction with poor rainfalls vince then, this explains the observed setbacks in the lrrotrUh of maize yields.  -9- *W1t~-; l n f ertilisers make up most of the f'armers' cash costz. In 171/7, thie? e,penditure for a high yield seed-fertiliser p,ack(rlgo would cost t roundc K50 for a four acre lot; by 1974/75, it had risen to over K130, though mostly during 1974. The following data illstratc the price movements, which are typical for all of L L .L.D P)'s s t-d-fertiliser packages. 1971/72 1972/73 1973/74 1274/75 Cost/acre, K: 12 15 16-17 33 (Package: 221bs. S.R.52, Maize; 200 lbs. CAN; 110lbs. 20:20:0) Though the prices did not miove up so suddenly in 1973/74, the shortages were begininig to be apparent by then, and probably had some adverse effect on yields. 3.8 The response of the farmers has been to cut down on the use of feitiliser, or to use inferior types such as Sulphate of Ammonia, wihich is cheaper, but which acidifies the soil. The effect this year is most serious: Table 8 Fertiliser Inputs in Program Area (14etric Tos No. of Total rer Fertiliser (lMetric Tons) Unit 2 Unit Year CAN4 20 :20 :0 SA Urea Total Centres Centre 1970/71 559 500 - - 1059 5 212 1971/72 2014 1850 - 59 3923 14 280 1972/73 2543 1338 - 70 3953 17 233 1973/74 2771 2439 1492 100 6802 21 323(252) 1974/75 581 1295 2406 20 4302 27 159(70) j Calcium Anmoniumi N4itrate (CAN) and Sulphate of Ammonia (S/A) are alterna.tive top dressings. In 1970)Rinistry policy stipulated the use of CAN in theprogram area on account of the acidity of local soils. 2/ JlumnbEr of unit centres fully brought into the program- (i.e. centres in the developed areas). / L'ach unit centre is about 20,000 acres, so this coluinprovides a rough incication of intensity of fertiliser use. The figures in brackets exclude S/A from the data. The int-ensity of fertiliser use is down by over 50% on 1973/74, and by 70,; if one excludes S/A. (Note that in 1973/74, S/A was already being used extensively instead of CAN.) The use of fertilisers has also declined pari-pasu with the use of high yield seed varieties.  - 10 - 3.9 This is of course disappointing to progriim management :ndl Llie c ,:t nsiorn workers. A continual extension effort has been mnd.:r to xsu.ic the farmf:rs to use: CAN instead of S/A, and until 19'3/71'lthe eff'ort was .ippaLrcnfly successful. The above w a c(rious seLloack cnyields. While improved husbandry data point should mailLintain yields above pre-prog,ram levels (other things given), they will probably be down by over 30% on what otherwise could be comfortably achieved. (iv) iI ize Prices 3.10 One answLr to t'>, prolblem would be to raise prices restore farmerst incentives to op. fior high yield packages. so as to Maize prices, both. in private and public (ADIikC) markets,have recently risen, but not commrn nsurately with costs: 1971 1972 1273 1974 Private Market Prices, tAb 1.8 1.7 1.7 2.0 (yearly averages) AD14AHC Buying Prices, tAb 1.3 1.3 1.3 1.8 Private matrket prices have in fact risen recently, particularly in urban areas; and if yields fail to increase with rising demand, it is probable that prices will rise further, thus providing farmers the necessary incentives. 3.11 There remains, however, the problem of ADMARC's prices. There are, of course, considerable political difficulties in raising ADMARCIs prices by the necessary amount. Maize is a staple and it is argued that price increases would most affect the lowest income groups. But since many of the latter are smallholders growing) their own maize, they would be insulated from the price increases, and on the contrary stand to gain financially from the marketable surplus. There are cotton area,s and areas of low rainfall which are not self sufficient in maize; this widens the dilemma, since it raises obvious questions pricing policies for their crops. The situation brought about public about by the high fertiliser costs is thus most unwelcome; but it is hard to see how the small maize farmer can possibly respond to the situation without :idequate incentives. IV. Groundnut-Snecific Factors (i) Tnmrovements in Husbandry and. Seeds 4l1 Giroundnut yields declined for three years in a row dnveloped and uncleveloped areas, but have begun in both to recover strongly in the past two 1years. Repeating for convenience the yield measurements shotrm in 'abjle 2: Groundnut Yie-lds, Lbs/Acre 6/70 701 /72 72/73 7/74 /75 Developed Areas 522 )490 431 323 Undeveloped Areos 477 672+75* 88 54)0 397 306 389 :a. Ratio 89% 91% 109% 106% 123% n.a. * Tentative Note that the measured decline was stronger, and the recover, weaker, in the undeveloped areas.  L .2 Despite the initial declinie, there are Lwo reasons for thinking that bhe yields are sys-tematically and significantly larger than they would have been withiout the program: there is more emphasis now on better qui.lity nuts (Chalimuana nm'ts), which have been selectively bred to have higher yields; and husbandry has greatly improved in all areas. Cnnsider t.hese factors in turn. 4.3 The g-rowina emphasis on higher yield, better quality nuts been seen in the following data: can Table 9 l;: tent and Yields of Be-tter Quality (Chailimbana) Nuts-** 1969/70 1970/71 1971/72 1972/73 Chalimbana Nut Content in S,ample, %: Develcped Areas ) 32%* 66% 64% 81% Undeveloped Areas ) 58% 59% 74% Yields (lbs. kernels/acre) of Chalimbana HJuts: Developed Areas ) 580 ) 570 460 336 Undeveloped Areas ) ) 452 323 Yields of Other Varieties Developed Areas ) ) 397 245 Undeveloped Areas ) 557 ) 496 318 226 Source: Reports of the Evaluation Unit, Lilongwe * Probably low, since nuts weren't so well graded then. ** Not all Cnalimbana nuts arehigh quality; the distinction between Chalimbana and "low quality" isn't precise. (Notc: the higher yields in the developed areas after 1970.) M4uch of the extenision effort has been to encourage farmers to be more selective in the clhoice of seed, andto trade better seeds among themselves. ADMIRC, who operate in the undeveloped areas, have also co-operated with LLDP management in the effort; this explains, together with the visits of extension workers outside the developed areas, why the growth of better quality nuts has also spread to the undaveloped areas.  - 12 - The evidence on improved husbandry both inside the prugirmn is and outside rfilso convincing, (Table 10): Tablc 1 0 Sarple Data on Groundnut Husbandry 1969/o 1970/71 1971/72 1972/73 Average Plan t`n, Densities (Should be over 14,000 per acre37'" Developed Areas 8,900 11,857 13,000 13,407 Undeveloped Areas 9,200 12,245 13,000 13,917 Farms writh Proper Ridging (30"t - 42 If Ridge Widths): Developed Areas ) No data for '69 through '72. Undeveloped kreas 79% ) But contour ridging uncommon 76% befare proj ect. Farms with fleavy Weeds, %: Developed Areas n.a. 23w 15% 12 Undeveloped Areas n.a. 10% 10% 12% Planting before Main Rains*: Deve'lopod Are.as n.a. n.a. 85% 91% Undfeveloped Areas n.a. n.a. 84% 87% Ftarms writh Hea-vy Pest and Disrease Probl&;ms: Developed Areas ) UndevEloned Areas 36% 15% )negligible 15 27% 47% Farms Sulphur Dusting to Prf;vwnt, Leai Spot: Developedc Areas 0 0 13% 17% Undeveloped Areas 0 0 1% 2% Source: Yield Studi,es of the Evaluation Unit, Lilongwe. * Data not reliable due to faulty method of inferviewing' probably overstated. estimates 4* The optimum is about 25,000, but the major gains are obtained by raLising densities to 1l,000.  - 13 . 4,5 The prograrn's achievements in improving maize and groundnut husbandry are remarkably impressive (as many visitcrs have observed); the data in the above tables serve to confirm this. As regards groundnuts, the .icnievement of increasing plant densities from 9,000 plants/acre in 1969/70 to over 13,000 in 1972/73, with the associated improvements in ridf.inLr sulphur dusting, weeding, etc., and with the better seeds, should ultimrite)ly more than doublr' yields (the increased plant densities f1lo1n( could do Lhis, otler things given). One would expect, therefore, M iL ULthrc: would have been evidence of a transition to high yields during i.hig per.Iod. But measured yields d(clined in all ,ireas. Why? 4.6 S:tmpling errors are a possibility, of course, but the data, as itJ is hoped the analysis in this report will show, make too much sense to be rE-ject.ed lightly. There are problems (discussed later) with the samples, as w-as to be expected; but they do not seriously affect the results, nor do they explain the p-,radox. (ii) Groundnut Disease and the DB.cline in Yields 4.7 It is sug,gested that two groundnut-specific factors partly (?h-plain the decline: (;a) To the farmer, groundnuts are a low priority crop, and it is true that it is commorly regarded as the "old woman's crop", not often tended by the farmer himself. Maeize is the staple, to which on average he devotes 'four of his five acres. Groundnuts give a net income of $20 - 40 in a good year (say, 15% of the family income). (b) The incidence of pests and diseases increased considerably from 1969 to 1972/73. In addition, there seem to have been a number of general factors (such as population movements and rainfall) which have acted against good yrields. 4.8 The argument that groundnuts are a, low priority crop is difficult to reconcile with the substantial improvements in husbandry and the efforts of the extlension workers, reported above. It goes some way towrards explaining random fluctuations from year to year - it is the last crop to be planted, and presumably the last to be prepared and weeded - but it does nct seem reasonable to invoke this argwuent to explain the large decline between 1969/70 and 19712/73. 4.9 The incidence of heaLvy pests and diseases, on the other hand, -xplatins a lot. RJecalling thhe information in Table 10, the incidence was as follows:- 1'9)9/70 negli 6ible Yield 522 lb/acre 197u/71 15% farms sampled Yield 490 lb/acre 1971/72 36,A farms sampled Yield 431 lb/acre 1972/73 45% fTrms sampled Yield 323 lb/acre Thle vield studies of 1971/72 illustrate trie seriousness of the problem. In the developed areas, the incidenace of pests ard diseases on the farms sampled was as follows:  - 14 - Table 1 1 Condition No. Farms Yield, Lbs/Acre (i.e fgligible) noollb 63 698 ) Mean = 570 Lighl, Post ant;d Disealse 115 500 ) Heavy PCst tad Disease 10X 126 282 Mean 431 Hence with nugligible pest and disease problems, as was apparently the case in 1969•7,0, the yields miglht hIave been comfortably in excess of 600 lbs/acre- - a good increase, relative to the base year, and close to the targets set for 1979/80. Similar adjustments for 1972/73, when the pest and disease problem was even greater, show that yields would probably have been in the range of' 500-550 lbs/acre under normal conditions. 4.10 There has been of course a continous effort to overcome the pest -and disease problem. Sulphur dusting to control leaf-spot is being encouraged, and is gaining acceptance. But the major problem has been with rosette, for which the following actions are being encouraged: - burying the diseased plants as soon as they are identified; - early planting and correct (more dense) spacing to provide a cover, so as to prevent the aphids settling; more dense spacing also increases yields directly, as discussed abo-re; - planting more seeds as soon as gaps in the crop are identified. In alddition, a rosette resistant seed variety has been tried and released in some areas this year, and is being offer-ed in a special credit package. 4.11 Pest and disease problems are never over, however. It has recently been announced that minor outbreaks of groundnut rust have spread from Rhodesia, possibly (Ibut hopefully not) presenting yet another uphill struggle. Three measures of control are beiing initiated: cultural control, to eliminate the use of seeds and ot.her disease-keepers from the previous yearts crops; fungicidal control; and, a slightly longer term (2 to 3 year) but more robust measure, the breedincg of rust resistant varieties. 2 It isworth noting that in 1967/68, when low rainfalls occurred and pest and disea.se problems werc also prevalent, yields reportedly fell to 100 lbs/acre. / Incidence: llormal: 2e66 Light: 27,-, Heavy,: 47%  (1 1i) "frwui(frai 1. ri .ceos b.p JApf)L lr'-om sulphur diusLini;, whiih is not a major cost, ;,roun hnft.1 r (Jcu.jrP little cherLical cr fertiliser input, so that they have no L been seriously affected by the rising prices of chemicals and fertiliseirs. IHowever, the farmgate price and thus income incentives are not particularly strong, as ctan be seen from the following data:- Avera,e Groundnut Prices, Average Yearly Gross Shelled, Tambala/Lb Income, Kwacha / 1969/70 4.9 26 1970/71 4.8 23 1971/72 5.1 22 1972/73 5.7 18 1973/74 6.7 32 1974,/75 7 .4 50-k 1/ Assuming 1 acre of groundnut/farm, and the yields in the developed areas, reported earlier in Table 2. e rnaltive r'ie price incentives, if one allows for inflation, were particularly weak in 1972/73; so this, too, may add to the explanation of low yields (and -dlso the large number of abandoned groundnut plots) in that year. 4,13 The low prices, both farmgate and ADMARCts resale prices, are surprising in view of the very high and quickly growing demands overse_-3 for i4.;lavri1s high quality Chalimbana nuts; reportedly, it is very hard to meet the demand. Apparently, the prices are low on account of forward selling; if this is so, a gradual shift in pricing and marketing policy is worth contemplating; it would provide a valuable stimulus to thf., rowth in yields. V. General Factors Affectinz Yields ( i) Th'inTlf'-3l r el With gioundriuts and maize the rainfall needs to be good over the selson rand tobe well distributed: good rains early on, to develop thle roots; in the middle, to promote plant growth; and particularly in thlo latter hailf to promote fertilisation and ripening. The rainy season lsI3s,s roughly irom October to April, andi fertilisation occurs in February (which is thus the most critical month)* If we look at the rainfall and its distribution since the begining of the project, it is possible to see an additional e-.planation for yield fluctuations:  - 16 - Table 12 Rainal at LIlone Aiort Yieldsi / R-ainfall, Inches Remarks on Lbs/ e Year Total Oct/Nov DecJan Feb Rain QualitZ Nlii6 GInuts 1969/70 33.9 0O5.# 23.1 7.6 Quite Good 1072 522 1970/71 37.5 2.2 19.9 10.3 Good 1299 490 1971/72 29.3 6.8 10.5 6.0 Very Good 1845 431 1972/73 24$5 3 92 11.5 2.8 Unusually Poor 1143 323 1973/74 42.5 o.402 19.7 8.2 Moderate-late start 1263 477 30 Year M mn n 33.1 3.0 15.2 7.2 3] In Developed Areas. / A crucial difference between these late starts was that, in 1969/70, the rains began December 3; in 1973/74, the dry spell lasted until after mid December. Also, in 1973/74, the rains were very heavy once they started and, though effectively drowning groundnut rose'tte that year, caused tip rot and damage to the maize. J The rainfalls vary across the program area, often strongly; we have checked the data along with the fluctuations in yields in different areas; as far as we can judge, the above data give a reasonable indication of fluctua.tions from one year to the ne;t in the project area. In the tecinical annex, some suggestions are made for interpreting the data more rigorously on cross-section/time series ),-Si s. 5.2 A close look at the above data will show that yields tie in quite well %zith the observc-d fluctuations in rainfall (we have also looked at Iot records for some of the unit areas in the programn, and find the same result). rJft(. 1971/72 season was a very good year for rains, and this undoub!,edly h1npe(d maize yields; it is probable that groundnut yields wiould have risen too, w&.re it not for the severe attacks of rosette, djiscussld eP-rlier. The rain in the following season, 1972/73,was unusually poor, so it is not surprising that maize and groundnut yields Lell - groundriuts all the more so, bcc:-.use rosette was even worse. The rains recovered moderately in 1973/74, and so did yields; in this season it will br, recalled rosette was also leas of a problem, thus enabling groundnut yields to recover too. 5.3 'Quantitatively, we estimate that -the difference in yields betwevn good and moderate rain yaa is about 10%* this makes a difference  - 17 - of roughly 150-200 lbs/acre for maize, and 50-70 lbs/acre for groundnuts. The difference in yields between a good and a bad year is perhaps 20 or more, implying a difference of about 300-400 lbs/acre for maize and 100-140 lbs/acre for grouncinuts. (ii) New Areas Coming into the Program 5.4 Whilever the coverage of the program is expanding, average yield figures for the developed areas are held down by new areas (with low initial yields) coming into the program. An impression of the effect can be obtained from Table 2. (Note that in 1971/72, the opposite lhappened, in that average yields went up due to two good areas being added with high initial yields; but this was an ex-ception.) (iii) Migration 5.5 It is commonly known that there have been very substantial movements in population in the program area since the program began, though they have recently been curtailed by the Government. One of the important, unrestricted movements has been towards the city of Lilongwe, which has increased its population from over 20,000 to 80,000 in the past six years. This has undoubtedly affected local farming. One positive effect is that it has created additional demands for local food crops, poultry, dairy and meat; also, since the local prices have risen, it has provided additional incentives for farmers. It is also thought that it has contributed to the spread effects of the program since members of the (extended) families work in Lilongwe for higher cash incomes, and are known to -take farm inputs back to the rural areas, even over quite long distances. One negative effect is that some farmers have abandoned their groundnuts, and, though to a much lesser extent, their maize. In 1972/73, for example, 16% of the groundnut plots in the developed areas were abandoned (mostly near Lilongwe); significantly, a higher number (32%) were abandoned in the undeveloped areas. (iv) Sampling Errors 5.6 It has been suggested that the measurements of yields and &acreages are in error, either because sampling procedures are unsound and/or because the s ample is too small. j The estimates (in Annex 2) for groundnuts were based on a simple regression of the percent c'l-n'ge in yields Y%, from one year to the next, on the change in rainfall, X, and the change in the percent incidence of pests and disease, Z%. The results were Y% = 8X - 1 .01Z%, where X is 0 for no change in rainfall coniditions, 1 for change between poor and moderate, 2 for a chLnge from door to good, and -1 and -2 respcctively for clhanges in the oppj)osite direction. The co-efficient 1.01 denotes the percent clmn-u in crop yield for each 1% increase in the incidence of posts aLndl disease, and is perhaps a little on the high side. This regression is only a preliminary estimate to get a rouglh feel for orders of magnitude. A basis for a more comprehensive an--lysis is provided in Annex 3.  - 18 - 5.7 It is hard, however, to think of anything basically wrong wit- present, procedures (though one ca.n think of improvements which might reduce routine sampling errors; see Annexes). A large selection of farms is taken at random each year for the measurements of acreages and yields; and the only suggestion of bias is the observation that some farmers begin to eat their crops (e.g. green maize) before the harvest. Supervision of the ennumerators is a problem, but measurement errors from this should cancel over large samples. 5.8 If anything, the sample is far too large. The acreages of about 1590 farms nre measured each year, and the yields of 1/90th acre sub-plots taken as samples for the yield data. For the 1973/74 samples; we randomly threw out the data for half the farmers and recomputed the sample means -nd standard deviations; the differences in these parameters between the full and half samples were negligible, implying that the sample could at least be halved (see Annex 1). Cutting the sample would, if anything, increase accuracy in that more time could be spent on supervision, and the sample crops harvested before the farmers began to eat them! (v) Bslineanaret Da 5.9 Several people believe that the baseline yield figures of 500 lbs per acre for groundnuts, and 1100 lbs per acre for maize, were too high, and consequently that (a) declines relative to the baseline figures could be expected, and (b) target yields, which were based on percentage increases over the baseline figures, were too high. 5.10 With hindsight, the baseline data seems to about 10% on the high side for maize, and about right for groundnuts if one ignores pest and disease problems. In the undeveloped areas, maize yields on farms using local seeds have averaged about 1000lbs/acre since the program began (ce.f. baseline 1100lbs). Grolmdnut yields in the undeveloped areas, in disease free years, were respectively 588 and 540 lbs/acre in 1969/70 and 1970/71, when the rains were quite good, and 390 when the rains were moderate (but prices were poor) - an average of around 500 lbs/acre. 5.11 Provided there are no furtlher setbacks with pests and diseases, t,he target groundnut yields of 6,50lbs per acre (30% increase over the baseline) should be atlainable. Even when average yields are low, farms are able to obtain yields of 600 to 1000lbs per acre due to good husbandry. In 1973/74, for example, the average yield in the developed areas was only )47l7bs per acre; yet over one quarter of the f'arms obtained yields in excess of 600lbs/acre. As the extension effort spreads to the backward fa-.rms, and particularl, if it raises their planting densities, their yields should rise to 'i-gh levels too. 5.12 The inaize yield targets of 20OQLbs/acre constitute a much more Limbitious increase of nearly 100%. Good husbandry, in principle,could accomplish this, plantling densi{,ies being a critical factor. But realisation  - 19 - of the targets will in practice rely heavily on the high yield s,od-fertiliser packages; and, unless something is done about the current probelm of high fertiliser prices, it is difficult to sco how tho targets can be met. If the latter problem can be overcome, the trargets seem attainable. Hybrid seed-fertiliser packages, for example, have consistently given yields of 2000-3000lbs/acre since the progrnm began (see Table 5); and average yields over all seed varieties have exceeded 2000lbs/ acre when planting densities have been over 12,000 plants/acre (e.g. Table 7, for 1971/72). (vi) Variation of Yields Among Farms 5.13 To see what yields can be attained in the program, and thus to determine if targets are realistic, it is interesting to look at the frequency distribution of the yield measurements. These show that over one quarter of the farms are already obtaining yields above targets (in average rain and disease conditions). The following data illustrate this point. * Groundnuts, 1973/7L Mai z e-. 1972/7 22? Yilllsacre Farm Sam2led-1 Av.Yield, Lbs Ace%of Farms <100 3 1062 19 101-200 12 1578 24 201-300 18 2077 32 301-L400 22 2226 15 401-500 16 2734 8 501-600 11 2693 2 601-700 7 701-800 5 801-1000 3 1 000 2 100 / All Sample. ?I InlDeveloped Areas Only. ,k/ Taken from a classification according to plant densities, since classification according to yield range not given. Above table serves to show that over 25% of farms had maize yields in excess of 20001bs/acre. * It proved to be too time consuming during the mission to obtain the distribution data for each year.  - 20 - 5.14 Hence one of the main challenges now in is to lift front of LLDP the yields of fams that have remained somu progress is being r,vde in this direction backward. That is indicated by the Iatrguc number of farms that are improving their husbandry, as discussed earlier. VI. Conclusions (1i) Yield Fluctuations. The yields of both maize and groundnuts have fluctuated from year to year largely on account of viscissitudes not uncommon in agriculture: variable rains; pests and disease in the case of groundnuts; and (more recently) fertiliser and crop prices in the case of maize. (2) Husbandry. Husbandry practices for both maize and groundnuts have improved remarkably in the program areas, as reflected in contour ridging practices; weeding and crop care; and in particular in improved planting densities, which for groundnuts have increased on average from 9,000 to 13,000 plants per acre since the program began, and for maize from 8,500 to over 12,000 plants per acre. (3) Spread Effects. Improved husbandry and the fertiliser packages have spread to outside use of new seed the developed areas of the program. (4) Maize Yields. The yield estimates for maize in the developed areas arc compared with (a) projections, and (b) estimates of yields without the project, in the following table: Maize Yields, Lbs/Acre, in Developed Areas Yield Proj ected Actual Without 1/ 'Le OkI' Yield, P Gain: Remarks on Yield, A Proj ect W (A-P% AA.-W'2% liainfall I 90q/70?o 1263 1071 929 -15 15% Moderate 1970/71 1332 1299 1086 - 2;Z 20%' Good 1971/72 138h 1845 1074 197"/73 33% 72% Very good 141A9 11l43 963 -21% 1 973/71k 1517 19% Very poor 1236 8f)6 -23, 35% MIoderate j/ Based on yield data for farms using local seeds in undeveloped areas.  - 21 - (e) :11.ize Fcrtiliser Prices. Fertilizer prices have increased tlhr-ccoid in recent ycars, the largest increases occurring inl 197)4. This is causing farmers to cut back substantially on fTrtiliser use, and to turn to inferior fertilisers which acidify the soil. There is a corresponding cut-back in the use of high yield seed varieties. This is, of course, likely to seriously set back yields; pricing policy and farmer incentives in l4 alawi requires investigation. (6) Groundnut Yields. Yields declined for three years in a row due to an increasingly heavy incidence of disease (rosette), which in 1972/73 afflicted 50% of the crop. The problem has now passed however (though rust presents yet another threat) and yields appear to be recovering quickly. The following table summarises the yield data in the developed areas: Groundnut Yields, lbs of kernels/acre, in Developed Areas Yield!- Proj ec ted Actual Without Year Remarks Yield, P Yield, A (sease$ w (A-P)% (W-P)% Rains Pests 1969/70 351 522 522 - 2% - 2% Moderate None 1970/71 5)41 490 540 - 9% 0% Good Minor 1971/72 552 431 600 -22% 9% V, Good Bad 1972/73 563 323 525 -43% - 7% V., Poor V. Bad 1973/7)4 574 477 477 -17% -17% Moderate None 1974/75 585 672 + 75?I n.a. 15% 15% Moderate None j 13'Bsed on yield data for farms which escaped the problem. g/ Tentative (7) The Yield Para.dox. Most of the variations in yields, it can be seen, are explicable. Earlier reports on "disappointing yields" negloctod obvious factors as regards disease and rains, and also the overwhelming evidence on improved husbandry and on the use of new seeds and fertilisers. Earlier reports also took averages for the whole program, though over two thirds of the sample w.iere made (for control and benchmark purposes) outside the developed areas. Since yields in the developed areas have gained systemaitically on those in the undeveloped areas, averages gave a false impression.  - 22 - (8) Baseline and Target Yields. The baseline data for groundnut yields were about right, if one makes allowances for pests and disease, the target increase of 30% should be achievable (provided there are no further major setbacks with pests and disease). The baseline data for maize were about 10% too high. The maize yield targets, however, are quite ambitious and meeting them will depend heavily both on improved husbandry and on the widespread use of high yield seed-fertiliser packages; since the latter are being cut back on account of the high fertiliser prices, there is a likelihood that yields will be well below targets., (9) Distribution of Yields. Over 25% of the farms are able to attain yields comfortably in excess of targets; one of the main challenges being met by LLDP is to lift the yields of the farms that have remained backward. (10) Other Points. (a) Sampling procedures and the yield measurements make sense and stand up to scruitiny (some improvements can be thought of, however, as discussed in the Annexes); (b) sample size could be cut; (c) population movements in the Lilongwe area have recently been very substantial, and doubtless affectes yields indirectly; they also led to abandonment of crops in some areas; (d) groundnut prices are low relative to the export market clearing price; price increases would obviously offer useful incentives to improve yields.  ANNEX 1 Page 1 Sample Size (a) Yields and Acreages Should samples be larger or smaller than they now are? To .nswer this, we randomly excluded half the sample in the groundnut, maize and garden measurement survtys for 1973/74. The essential point is that, if the sample can be reduced by half, the sample means and standard deviations (or coefficients of variation) should be more or less the same. The following results are for groundnuts: Sample Means, Deviations and Standard Errors For Half and F'ull Samples Sample Standard size Mean Yield Coef. Variation Errors EVA Full Half2-/ Full Half Diff. Full Half Diff. Full Half 1 78 40 524 553 6% 39 33 -15% 23 28 2 34 18 454 458 1% 41 36 -12% 32 40 3 73 37 401 411 2% 43 46 7% 20 31 4 94 48 416 428 3% 50 51 2% 21 32 5 76 38 404 419 4a 45 51 13% 21 34 6 69 35 317 327 3% 56 55 - 2% 22 30 7 87 44 341 340 0% 50 49 - 1% 19 25 8 50 25 389 395 2% 46 50 9% 25 39 9 85 43 448 433 - 3% 43 4)4 2% 21 29 10 22 11 326 295 -10% 70 70 0% 49 62 11 45 21 370 304 -22,' 73 78 7% 46 51 12 7 4 310 368 10% 57 54 - 5% 67 98 13 57 486 525 8 25 6i g 1 36 63 767 393 410 415 1, 51 50 - 2% 8 10 Ever-y alternate case was taken, on the supposition that the cases were distributed randomly. The estimated means and standard deviations dontt differ very much be;tweein the two samples; from this it can be concluded that the samnle could be halved. (Th'ere is an exact statistical test, the It' test, which can be applied to establish this.) There are some discrepancies within the EVATs; the obvious thing to do to avoid sampling errors here is to set mLinimum levels for the samples in each EVA (also, it should be remembered that for most purposes the important classification is between Tdevelopedl and 'undeveloped' areas).  ANNEX 1 Page '2 Similar e-ercises were repeated for maize yields and acreages, witli the same conclusion: the s-ample could be cut, perhaps by half or iaore. Lven samples one quarter of the present size seem to give quite good results. (b) Explanatory Variables HIowever, there are a large number of explanatory variables to study, and the sample needs to provide a representative variation in each of these. One answer is that, provided there is good variation, and no serious multi-colinearity, the sample still need not be too large. The dr;gres of freedom are (N-n), where N, the sample size, is around two or thiree hundred, and n, the number of variables, is rarely more than twenty; so degrees of freedom should still be two or three hundred, which is quite good. In any case, a proper study of explanatory variables would pool the time series data for each year. Thus degrees of freedom rise by (N-n) each year, reflecting the accumulation of knowledge and ex<perience; cn these grounds, samples could be further reduced. Hlote The above calculations were repeated two or three times because of apparent aLrithnetical errors in either the full and/or half samples. In carlier caLlculations the differences were much less than quoted above. It is possible that some arithmetical errors still remain.  ANNEX 2 Pag e 1 RTh.infall, Disease and Groumdnut Yields 1969-73 a Simple R g.ression Study Rainfall at Lilongwe Airport, and yields and rosette in the developed areas of the program were as follows: Rain, Inches Rosette Yield, Year Total February % Crop Affected Lbs Acre Comment on Rainfall 69/70 33.9 7.6 5% 522 Moderate - late start 70/71 37.5 10.3 15% 490 Good 71/72 29.3 6.0 35,' 431 Good - well distributed 72/73 24.5 2.8 47% 323 Poor 73/74 42.5 8.2 2% 477 Moderate - late start Records for EVAs 3 and 4 were also available; respectively: 70/71 31 .0 4*5 28% 515 Moderate 71/72 2b.0 2.8 48% 219 Poor 70/71 34.0 7.7 34%0 455 Good 71 /72 27.0 4.0 , 41% 417 Moderate Since absolute yields change from one area to the next for many reasons, it is probably best, in pooling the above data, to correlate percentage changes in yields from one year to the next, with changes in rainfall and rosette. This gives: Change in Change in Change in Years Rains, X Rosette, Z Yields , Y Program: 69/70-70/71 M-G +1 10% - 6% 70/71-71/72 G-G = 0 20% -12% 71/72-72/73 G-P = -2 12% -25% 72/73-73/74 P-M = +1 -45% 48% iEVA 3: 70/71-71/72 14-P = -1 20% -52% EVA 4: 70/71-71/72 G-M = -1, 6%o - 8%  ANNEX 2 Page 2 wherc a constant-shift term is used for an estimate of the effects of changes in rains, that is Yi = aXi + bZ where Xi = 0 for no change between years, -1 for a change from good to moderate or moderate to poor, -2 for a change from good to poor; and 2 and 1 respectively for changes in the opposite direction. It would, with a larger sample, be preferable to estimate two constant- shift paraneters to allow for differences between changes from M to P and from G to M; the above sample was not big enough to do this. Tle regression is Y = 8.2X - 1.01Z!v8X - Z with a correlation coefficient of 0.88. This suggests that each 1% increase in the incidence of heavy rosetbe cuts yields, on average, by 1% (which makes sense) and that a change from good to moderate, or moderate to poor rains cuts yields by about 8% (slightly on the low side, one would think).  ANNEX3 Page 1 Comments on Statistical Analysis Mnny variables associated with improved husbandry nnd tuchnology move together; and there is even an association between t natural factors, such as rosette and rainfall. Regression, Variance and , Co-variance analysis (i.e. combined regression-variance analysis, discussed in Johnston's Econometrics) can bc used to determine independent effects. If Yirt is the yield on farm i in region (evaluation unit) r in year t, and if X irt are the corresponding explanatory variables, of which there are j - ... J, we could regress: Yirr -Larjirt + irt Note that the coefficients ma change between regions; equally, some maly not be statistically very different - one coefficient, e.g. on the percentage effects on yields of improved plant densities, may be estimated for all regions. There is an exact statistical test (the F test) to establish if coefficients are constant between regions, and if data can be pooled. Note that the Xts can be zero-one variables, as in variance analysis, to denote discrete effects such as changes between 'good' and 'bad' rainfalls from one year to the next. In much of covariance analysis, it is dften found that only one parameter shifts significantly from one region to the next, and this is the constant-shift parameter associated with j = 0 and Xorit is a zero-one dummy variable. There are, obviously, many correlations that can be tried; but the previous yield surveys seem to have fastened onto the important ones (except that rainfall was omitted - this could be studied by pooling cross-section and time series data). The best correlations would be to represent yields as a function of four groups of variables: ( ) Natura.l Factcrs - rain; - pests and disease; - soil condition; (2) Technical Factors - seeds; - f ertilisers; (3) .Husbandry - planting densities; - ridging; - weeding; - planting time; (4) Economic - prices.  ANNE;X 3 Page 2 -2- in some cases, it may not be worth distinguishing the independcent effects of, say, seeds and fertiliser. between best be represented by one variabl o, a dummy They might variable which changes with the seed-fertiliser package. crimilarly, one--might study planting densities independently, and reprtsent all other husbandry factors by one dummy variable, which shifts according to whether other aspects of husbandry are considered to be t7oodt, 'moderate' or 'poort.  ANNEX 4 Page 1 MONTL'LlNG AND E.VALUATION AT L.L.D.P. I nlro dues l. ion An inter-ministry and inter-program ,f .l;wlw working party in recently discussed in geii,. .1 terms ltrtsks of monitoring and evaluation the various purposes and in the four phases of project atnd program development; that is, in: - identification; - preparation; - appraisal; and 1/ - execution (on-going evaluation)- The following discussion is confined to phase as it is carried out at Lilongwe. the specifics of the fourth (very briefly) the work done to It 'begins by summarising date, and then offers some comments .ind suggestions. 14 Mnif-.oring and Lvaluetion to Da-te (a) Yields The major task of the Evaluation mu-sure, eacli year the yields, acreages Unit has been to survey and maize, igrounidnuts and tobacco. and farming practices regarding The surveys cover between 1000 and farmns each year, selected at random 1500 in the program area. Once the farms have been selected, the surveys begin with what are called the "garden surveys"'; these consist of measuring the farm size and acreages allocated to the various crops, and then of choosing at random and marking out the 1/190 lb, acre yield sub-plots which will subsequently be harvested for the yield measurements.-/ This is an arduous task which is in December e.ach year after the begun crops h-ave been planted, and which through until M4arch, not long before lasts the harvest. The second step is to harve.-` +he measurements, and interview the sub-plots for the yield farmr - c. obtain the "yield survey" data; these data include: crop speci`;-.r of' sl'fed; husbandry; type and use n, and type and sources of L 22 Lser; . natural factors, such as pe:sts, disease, waterlogring; e: -tr.nsion advice received. commenits of Uie farmer; and The last step is to collate and data onto punched cards, and begin put the analysis and reporting. ~/ Planning Unit, June 1975. File 14/12/1 3/3 v/ Instead of taking a sub-plot, it Blight be better to take samples from every . or 10 ridges. This would cut down random errors resulting from the sub-plot not being an average area for the farm.  ANI'NEX I Page 2 The garden measurement and yield surveys have been completed vacilyear as scheduled. Analysis and reporting has tended to lag, however, because of acute problems of staff turnover and the shortages of qu.nlified staff. So far, the data for 69/70, 70/71 and 71/72 have been fully collated and reported; but only the 72/73 groundnut data have been attended to since then. (b) Special Purposes Studies In addition to the work each year on yields, a good number of studies have been cunducted ad hoc into matters of particular interest and concern. Among those that have been completed are special crop surveys into the yields of new seed varieties; surveys of potential adopters of the new varieties; and surveys into maize damage, smallholder agriculture, the disposal of maize and groundnuts, stall-feeders, the extent of agricultural knowledge of small farmers, and various sociological matters. Most of these studies have been undertaken by people specially recruited for the work or by volunteers; this is to continue, and help from Bunda College has been discussed. There have also been several surveys which, though completed, remain unanalysed and on punched cards. The principle ones were household composition and behaviour; health; and farm labour utilisation. It should be added that the reasons why the information is unanalysed are the ones mentioned earlier - qualified people left, or could not be recruited. Also being contemplated, when resources permit, are several new studies, of which artificial insemination of livestock is a priority. In additii -, a person to carry out more extensive statistical analysis of existing data is being recruited. Comments and Suggestions All the above work is part of good evaluation, and the Unit has undoubtedly produced some good and useful material. The comments which arise, therefore, relate not to quality or relevance, but to the problems of backlogs, data compilation, and presentation. (i) Backlogs The immediate priorities would seem to be to remove the backlog on the 72/73 and 73/74 yield studies, and to process this year's data (particularly of maize). Artificial insemination is another priority of proaram management. Third in line, which could be done in parallel if ,esouaces permit, might be analysis of the information of the labour, heJilth and household composition surveys. (Given all the problems with recruitment experienced with thc special purpose surveys, it might be best in future to regard these as a windfall benefit, rather than as a basic task; the regualar staff could then devote more time to the basic work on yields.)  ANNEX 4 Page 3 (ii) Data Compilation Major factors in the backlog are the bottlenecks and difficulties witl comi,uterisation. The dataare put directly onto punched cards from tlhe questionnaires (schedules). The problems with this are that costs and delatys are high because of inaccurate punching; there are too many problems of debugging ; and it is very costly even if accurately done, since all the columns are thus compiled, when, for statistical analysis, rKlatively few columns are needc d in any one run. It would be better, I suggest, to first transcribe thy schedules onto ledgers, before any thought i.s given to computerisation, and to make preliminary tabulations by hand; this would make records more readily accessible, provide the unit with a better feel for the data, and provide more involvement and Lraining for the many junior level people involved with evaluation. (iii) Presentation and Reporting Oome consideration might be given to puting more of the Unit's findings eachlyear into the Annual Report of L.L.D.P. Things to report on would be: - the yield measurements in developed and undeveloped areas (and possibly by Evaluation Area), contrasted with measurements of previous years; - indicators of improvements in farm practices (seeds, fertilisers, pl.9nting densities and various aspects of husbandry) as compared with previous years; and - a short descriptive piece on the special factors affecting this yearts yields, zerhatps backed up with data (e.g. on fortiliser prices, rainfall, pests and diseases) to support the argumen ts. (Naot th.at such -information can be quickly obtained by hand, from ledgers At the moment the Annual Report contains considerable information on the progr:-nls inputs to rural areas - on conservation works, roads, boreholes and water supplies, buildings, finance and credit, training, extension, and various institutional and admrinistrative accomplishments. B3ut it contains little on the ultimate outputs of the program - acreages, crop yields, f.arm pricticos, l 7vestock and various economic and social changirs in the area. Yet there is a mine of infornation collected each year which remain in the Unit's files; survey schedules and technical reports. The communication benefits of brief annual reports produced before the main statistical surveys, and concentrating on movements in yields and explanatory variables, would be considerable; and more time could be taken on the more comprehensive analysis.