38468 Agricultural Extension Services In Indonesia: NEW APPROACHES AND EMERGING ISSUES May 2007 The World Bank East Asia and Pacific Region Rural Development, Natural Resources and Environment Sector Unit Sustainable Development Department ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES ACKNOWLEDGMENTS . This report is a product of the World Bank's Rural Development, l'\atural Resources and Environment Sector Unit ofthe East Asia and Pacific Region under the Raising Rural Productivity (P090503) umbrella AAA. This report draws on two core reports that were coordinated by EASRE: (i) Impact Evaluation of the Decentralized Agricultural and Forestry Extension Project - DAFEP (December 2006) carried out by the French Agricultural Research Centre for International Development - Centre de cooperation internationale en recherche agronomique pour Ie developpement (CIRAD) and (ii) Decentralization and Agricultural Service Delivery - Benchmarks, Transfers and Capacity Building in Intergovernmental Relations (June 2004) by the Social Monitoring and Economic Research Unit (SMERU), Jakarta. In addition, various World Bank reports have also been drawn on. Dely Gapasin was responsible for commissioning CIRAD to carry out the initial benchmark study in 2001. The DAFEP Impact Evaluation finally saw the light ofday thanks to financial support from the following sources: TF053716 (Community Driven Development Platform); TF054695 (Rural Investment Climate); TF051845 - Review of CDD Impacts in East Asia Region (TFESSD- multiple donors), and CIRAD.The main results of this report were presented at a workshop in \Vashington D.C during the Sustainable Development Network Week in l'\ovember 2006. 'Jhe report benefited from the feedback and discussions during this workshop. The team leader tor this report is Shobha Shetty (Sr. Economist, EASRE). The report was prepared under the guidance of Rahul Raturi, Sector Manager, EASRE and Stephen Mink, Lead Economist, EASRE. Peer reviewers were Gershon Feder (DECRG) and Riikka Rajalahti (ARD). Richard Chisholm (Sr. Agriculturalist, EASRE) provided useful comments. Dewi Sutisna (EACIF) provided valuable administrative assistance. The variolls project staff of AAHRD provided unstinting support during the preparation ofthis evaluation. PT Hutarnacipta Konsulindo and Surveymeter carried out the field survey at the benchmark and at the EoP stages respectively. Anne Gouyon from Ideforce was involved in the pilot methodology and contributed to improvements in many aspects of this work. The support of Scott Guggenheim (Lead Social Development Specialist, EASSO), Susan Wong (Sr. Social Development Specialist, EASSO),Jean-Guy Bertault (Regional Director CIRAD, South East Asia), and Robin BourgeOiS (Sr. Economist, CIRAD) is gratefully acknowledged. Last but by no means least, this work could not have been achieved without the understanding and cooperation of thousands of rural households throughout Indonesia. Our grateful thanks to all of them for giving so generously of their time and insights. r'.cknowledgments III AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES J ABBREVIATIONS AND ACRONYMS AAHRD Agency for Agricultural Human Resource Development ADB Asian Development Bank AIAT/BPTP Assessment Institute for Agricultural Technology ARMP Agricultural Research Management Project BIPP District extension center BPP Sub-district extension center ClRAD French Agricultural Research Centre for International Development DAFEP Decentralized Agriculture and Forestry Extension Project DEC District Extension Committee EoP End of Project FEATI Farmer Empowerment through Agricultural Technology and Information FO Farmers Organization FMA Farmer Managed Activity GoI Government of Indonesia lAARD Indonesian Agency for Agricultural Research and Development ICR Implementation Completion Report Logit Binary Logistic Regression Model NGO Non-governmental organization PIP Project Implementation Plan PPL Field extension worker PMU Project Management Unit RPO Rural Producers Organization RT/RW Administrative and geographic subdivisions at village level UPKG Village Extension Management Unit WE World Bank Cover photographs courtesy Robin Bourgeois, ClRAD IV Abbreviations and Acronyms ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES EXECUTIVE SUMMARY Indonesian agriculture is at a crossroads. Supporting the livelihood of millions of Indonesians, it needs to underpin renewed and robust growth of the economy; and be a key component of the Government's poverty alleviation strategy. The challenge for the future is to reinvigorate productivity gains among rural producers, and provide the foundation for long run sustainability of these productivity gains. Productivity gains are key to farmer income growth, and for this rebuilding the research and extension systems - that have seen a marked deterioration in recent years - will be critical. The experience of the Indonesian decentralization of its extension system has been mixed, with adverse impact on extension through sharp reductions in funding, and removal of central-level guidance. At the same time, a series of positive debates and experimentation in management have taken place from a shift on top-down to participatory approaches, input and technology dissemination to dissemination of market and upstream information and technology, from centrally managed extension services to decentralized services, and some movement toward privatization of extension. One such experimental approach was the World Bank-assisted DAFEP (Decentralized Agriculture and Forestry Extension Project 2000-2005) which provided an impetus for demand-driven extension and for institutional reforms at the local government level. The development objective of the Decentralized Agricultural and Forestry Extension Project (DAFEP) was to assist the Government ofIndonesia (GoI) in "... en.hancing farmers capacity to participate in extension activities, and in strengthening the capacity of the districl-Ievel integrated agricultural andforestry extension system which would promote economically viable, environmmtally sustainable, and socially acceptableforming practices and increasedfarmers income". In this context, an assessment of the agricultural extension services, as seen through the lens of the impact evaluation ofthe DAFEP project, was deemed to be timely and relevant. This report thus has the following objectives: (i) provide an overview of the institutional changes in agricultural extension in Indonesia; (ii) present the results of the impact evaluation of DAFEP; and (iii) discuss lessons learned and emerging issues in the new political and institutional context. The Impact evaluation comprised a Benchmark survey (2001) and an End-of-Project (EOP) survey for collection of data related to a set of indicators corresponding to the project expected outputs. SAlVIPLINI:; l\IETliODOLOGY The DAFEP Jample The choice of the appropriate sampling size was based on the decision to use income dispersion as the key variable whose variance will be used for setting the sample size so that it represents as much as possible the whole project population. With 30 000 households directly targeted by the project, and a 95% confidence degree, a sample size of 400 randomly selected farmers was obtained. Due to possible "losses" throughout the project duration, the sample was increased to 450. The field selection of villages was based on the identification of districts and related sub-districts were the DAFEP was planned to be Executive Summary V AGRICULTURAL EXTENSION SERVICES IN INDONESIA · NEW APPROACHES AND EMERGING ISSUES J implemented, and to select sub-districts that were the most representative of the district agro-ecological and socioeconomic conditions. 'Extreme' sub-district situations were first discarded based on meetings with local key respondents and consultation oflocal secondary data. Then, the sub-districts were randomly drawn. In each selected sub-district the process was repeated for the village selection, discarding 'extreme' villages. At village level, respondent households were randomly drawn using the list of registered DAFEP members. Replacement households were also drawn in case of the impossibility of finding any households in the first list. For the purpose of the "with and without" comparison, a reference household sample of similar size and characteristics was selected with the same method in non-DAFEP districts. The process for selecting the respondent households mirrored the process used for selecting DAFEP households. The main difference was that the choice of district, sub-district and villages was conditioned by their resemblance to the district, sub-district and villages in the DAFEP sample. Then, at village level, respondent households were randomly selected from the list of village households, excluding known households that were known to be completely non farm households (for instance, pure traders or civil servants without any farm activity, neither the head of household nor the spouse or dependents were disregarded in the drawing). This was intended to minimize differences between DAFEP and Reference households due to agro-ecological and socioeconomic conditions. Ibe Spi1!o'l'er sample In addition, a Spillover sample was added to estimate whether the project had also generated indirect effects in the area of implementation. The Spillover sample was built as a mix of households not participating in DAFEP activities but located in areas where extension workers were part of the DAFEP project and had been trained in and promoting DAFEP approaches. This sample included thus three cases of non DAFEP households according to whether they were located either: a) in DAFEP villages, b) in non DAFEP villages but still in DAFEP sub-districts, and c) in non DAFEP sub-districts but still in DAFEP districts. However, the results from this sample were inconclusive/or of marginal importance and hence, are not presented in the main report. Impacts Income and Jf!elfare The analysis of the income data indicate that DAFEP was successful in significantly increasing farmer net crop income per hectare as well as non-farm income. In fact, non-farm incomes showed a significant increase across all samples. However, due to the problem of a Significant number of cultivated but non harvested parcels, a better proxy of farmer welfare was obtained through the use of expenditure data. Three indicators were used for impact assessment on welfare: yearly expenses, assets value and share of food expenses in total consumption expenses. The difference between Before and After situations for these three welfare indicators is significant for all samples. DAFEP households experienced a significant improvement oftheir welfare situation based on yearly expenses and asset value per capita, a result that is associated with the significant improvement in nonfarm income. A welfare distribution analysis based on Gini coefficient, Skewness and Kurtosis indicates that inequality decreased in all samplesflryearly expenses but it increased significantly for the DAFEP sample with regards to asset distribution compared to the Reference sample. However, since the trend is common to DAFEP and Spillover samples it is difficult to relate it with an effect of the project. VI Executive Surrmary \.. AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Productivity and Technology. The project was expected to generate a 5% increase in productivity of the main farming systems by EoP. Using the gross value of all harvested crops per hectare and the yield of the main crops (rice, maize, soybean and coconut) indicated that. Tests results indicate that there is no significant change in the total value ofagriculturalproduction when comparing Before and After situation in DAFEP and Reference samples while the Spillover sample experienced a significant decrease. Rice yield has remained unchanged in all samples indicative of lack of up-take of improved technologies (including improved seeds) and lack of irrigation. Percentage of area devoted to ricefell sharply in all three samples. Only in the case of soybean yields a significant change can be observed in the DAFEP sample. DAFEP households achieved an 8% agricultural growth but significant attribution can be made to the project. Four proxy indicators were used as a measurement of technical changes: the total value ofinput per hectare, an input/output ratio measured in monetary value for the main crops, a diversification index, and the value of trade income per hectare. Results show that households in a1l samples have increased their use of inputs in constant value. Input use efficiency changes are significantfor rice in the DAFEP sample and for soybean in both samples. Diversification was high to begin with and did not change significantly. Comparison between DAFEP and Reference sample shows that the project cannot be associated in a significant way with an intensification process that could bear witness of a technological change. However, the magnitude of changes that took place in input efficiency between DAFEP and Reference tested significant for rice. In the DAFEP sample, households traded a larger amount ofproducts before and after the project compared to the Reference sample but the impact of the project on trade income per hectare is not tested significant. Extension. 1he cluster of qualitative extension indicators includes six ordinal indicators: availability of agriculture information (AAI), willingness to pay for agriculture information (WPAY), accessibility of extension workers (AEW), extension meeting frequency (EMF) and improvement in access to agricultural informat-ion (IAA1) , and general access to extension services (GAES), a composite indicator. Results indicate that households have better accessibility to agricultural information, but they don't show change in the willingness to pay for agricultural information. Accessibility to extension workers increased significantly in all samples. Improvement in the availability of agricultural information is rated as positive and non Significant for the DAFEP respondents and positive and significant for the Reference sample Links. The links refer to the relationships between farmers and the upstream and downstream agricultural environment. The cluster included six ordinal variables: accessibility of input markets (AIM), accessibility of agriculture input information (AAII) , joint purchase of agriculture input OPAI), accessibility of output markets (AOM), joint marketing of agricultural outputs (JMAO), and improvement in upstream and downstream links, a composite indicator (IUDI). All households have significantly better accessibility (fewer barriers) to input market in general. However, trends :'0 40 A ...J._........WWW.LIWJ.I.u.u..LU.E_ __ f~l Jil -I ~ N · .. . ~ ·:;0 .. ,... :-.0 ;40 tr;'l() ... ~ ~o · '0 o~ __~WW~WJ.I~UD~__~ o 20 40 ';0 80 100 20 40 'iO 80 100 Age Age Notes: For x (survey) l=Benchmark; 2~EoP Survey; For y (type sample) l=DAFE; 2~Spillover; 3=Reference 3.15 Education and Family Size. More than 60% of the respondents have an educational level equal or below the end of Elementary School. This proportion is similar and unchanged in the Before and After surveys and across the DAFEP, Reference and Spillover samples. The structure of education distribution shows no particular differences between samples. The average size of family members remained almost unchanged (4,68 before and 4,65 after). 14 The Decentralized Agr:ccltural and "ores try Ex~enslOn Project - - - - - - - - - - .. -.~.- - - . - - . - -..- - .. ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Figure 3. 4. Distribution of family size in the Benchmark survey. Survey 1 2 200 > ~ 1:.0 C · ij.loo Col · r.o ... I&- 0 200 Ii > ~ 150 C · ij.loo · ... ,,0 I&- 0 ~oo > 11 ~ 150 C · ij.loo ... · r.o ... I&- 0 I i i 0 o H) ~ 30 40 Household Household Notes: For xCsurvey) l=Benchmark; 2:EoP Survey; For y (type sample) l=DAFE ; 2:Spillover ; 3=Reference 3.16 Agricultural workers. The structure of the distribution of male workers in agriculture is similar in the three sa:nples in the Benchmark survey and similar also in the EoP survey. Most of the households (more than 60!1o) have two people working in agriculture. Changes in the Before/Mter situation are relatively small, with the exception of an increasing number of households with no workers in agriculture (from 1 to 10 %). lhis either indicates an increase in the income level that makes labouring as wageworker less needed or because other job opportunities have appeared. The Decentralized Agricu'tLral and Forestry Extension Project 15 ---_ _ - ... .... _ - - .... _ - - - - - - - - - .... _ - _ ... AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES ..J Figure 3.5. Evolution of male agricultural labour force (number of households members) DAFEP 100 Spillover Reference 300 Before After The structure of distribution of female agricultural workers has also evolved similarly across the three samples. The relative share of households with 0, 1 or 2 female agriculture workers, were respectively 17%, 56%, 18% in the Benchmark and 28%,54%, and 10% in the EoP survey. Figure 3.6. Evolution of Female agricultural labour force (number of households members) 300 7 DAFEP I o Spillover o Reference 16 The Deceni'allzed Agricultural and Forestry Extension Project "'~""','-------------------' ~ AGRICULTURAL EXTENSION SERVICES :N iNDONESIA. NEW APPROACHES AND EMERGING ISSUES 3.17 Crops. The total planted area did not significantly change across samples, but it did across time. It covered altogether almost 1,800 hectares in the Benchmark and only 1400 in the EoP survey. The reduction comes in tact from the rice cultivated area dropping from 1150 ha to 800. However, this is partly due to the reduction in the sample size (erosion of the sample). Importance of paddy for the respondent households is shown in Table 3.4 below. Table 3.4. Distribution oEland and cropped land in the samples BoP All Dafep Reference All Dafep Reference Nb of Households 1172 384 384 1067 329 330 iI'otal area 1729 560 597 1407 452 502 Land/Househ old 1,48 1,46 1,56 1,32 1,37 1,52 Paddy (Ha) 1109 359 362 797 234 281 64.1% 64.1% 60.6% 56.6% 51.8% 56.0% Maize (Ha) 143 38 54 129 48 49 8.3% 6.8% 9.0% 9.2% 10.6% 9.8% Soybeans (Ha) 59 18 22 54 13 21 3.4% 3.2% 3.7% 3.8% 2.9% 4.2% Others (Ha) 418 145 158 427 157 151 24.2% 25.9% 26.5% 30.3% 34.7% 30.1% The majoritv of households plant one or two crops per year and a few up to five different crops. There is no particular difference, except for a slight increase in the total number of households planting more than three crops in the DAFEP sample compared to the Reference sample in the EoP survey. Table 1\umber of crops grown and corresponding share of households Number of Household count Survey Total Crops DAFEP Spillover Reference 1 48% 51% 51% 575 2 40% 39% 37% 441 3 11% 9% 12% 121 Before 4 2% 1% 0% 9 5 0% 1% 1% 6 Total 384 384 384 1152 1 46% 53% 52% 488 2 41% 37% 35% 366 3 8% 7% 10% 81 i After 4 5% 2% 2% 28 5 1% 1% 0% 4 Total 329 308 330 967 Tre Decentra!ized AgrlcultLral and Forestry Extension PrOject 17 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES J In relation with income assessment and the measurement of the project's expected achievement to raise income by 5% over the project period, five indicators were used 10 : total net income per capita (TNIC), agricultural net income per capita (ANIC), agriculture-related income per capita (ARNIC), agribusiness net income per capita (ABNIC) and non farm income per capita (NFIC), the first one being the sum of ANIC, ARNIC and NFIC, while ABNIC is a subcomponent of ARNIC. As shown in the text box below, all these indicators are needed due to the diversity of income sources at individual household level. These indicators measure the net value of monetary inflows of the household during the surveyed period (one year starting from the date of interview backwards). ANIC refers exclusively to the total value of yearly agricultural production (crops, animals and forest products). ARNIC includes selling of labour force, renting out of production factors such as land or equipment, and processing of agricultural products (ABNIC). NFIC includes other jobs, trade, sources of income such as remittances, etc. Figure 3.7. Comparison of income indicators in 2001 constant value 2500000 Rp 2000000 1500000 1000000 500000 o Before After Before After Before After DAFEP SPILLOVER REFERENCE OANIC .ABNIC · NonABNIC .NFIC Note: NonABNIC= ARNIC-ABNIC The samples show a marked increase in non-farm incomes. Agricultural net income per capita in contrast has declined. The significance of these results will be analyzed in subsequent chapters. 10 All these indicators are only for farming households. In the EoP survey, 967 of 1067 households had cultivated lands. Other households are either landless or are not farming anymore. The income indicator for the whole sample is the yearly expenses per capita (see \Velfare indicators). 18 The Decentralized Agricultura i and Forestry ExtenSion PrOject DAFEP Impact: Analysis of Qyantitative and Qyalitative Indicators 4.1 1his chapter presents the detailed results of the data analysis starting with the quantitative indicators and followed by the qualitative indicators. Income and Wef(are Cluster 4.2 111e income data described in the earlier section have their limitations in that at the time of the EOP survey, about 20 percent of the plots were cultivated but not harvested. This affected the value of several indicators, in particular agricultural net income per capita, yields, total production value per ha, input use efficiency, trade income per capita. This situation influenced the results in particular when comparing the Bifore and After results, since at Benchmark data collection period tbose cases were almost non-existent. Some analytical adjustments were made (see Annex 1) in order to minimize the effect of the missing production data when comparing the results. The yearly expenditure data (which excludes input expenditures) was also analyzed separately as an indicator of farmer welfare. DAFEP Impact: Analysis of Quantitative and Quaiitative Indicators 19 ------------ ------ .. -.~ ..... ~. AGRICULTURAL EXTENSION SERVICES IN INDONESIA. ~ NEW APPROACHES AND EMERGING ISSUES Table 4.1. Before/After comparison of income indicators (farming households) Sample Condition TNIC ANIC CNIHA ARNIC ABNIC NFIC Before 1,375,117 545,359 7,296,341 I 215,599 80,054 614,160 DAFEP After 1,905,571 3~502,606 269,614 218,~ 1,293,366 530,454' (201, s 1,206,265 n5 54,015 138,689 I 679,206 Difference 11 38,6% (36,9%) 16% 25% 173,2% 110,6% Before 1,489,872 584,633 6,562,847 220,112 49,577 663,164 After 1,945,716 326,004 6,509,380 348,600 285,991 1,226,376 Reference 455,844' (258,629)' (53,468)n5 128,488 236,414' 563,212s Difference 30,6% (44,2%) (1%) 58,4% 476,9% 84,9% Notefor all quantitative tables: · indicates that the difference between Before and After is significant using either T -Test or Confidence Interval. *. indicates that it is not Interpretathm 4.3 The difference in TNIC between Before and After situation for the DAFEP and Reference samples is significant and quite high (more than +30%) in constant terms. In the case of ANIC the difference between Before and After situation for DAFEP and Reference samples is significantly negative and high (around -40%) in constant terms. However, for the reasons explained in the Methodology chapter, ARNIC cannot be used as an indicator of absolute performance because of the problem of not-yet-harvested parcels. For this reason, an alternative indicator was built - the crop net income per hectare (CNIHA) based on the results from the harvested parcels. This indicator, while it does not include all parcels in the sample, helps nevertheless to reduce the bias since changes are calculated per hectare and not per capita. The table shows that the difference between After and Before, based on this indicator, is not significant in both samples. Furthermore, what most matters, again, is not the absolute value of the changes but the significance between samples. As the problem similarly affected all samples in the EoP survey, the results of significance analysis presented are still valid. 4.4 Changes in ARNIC and in particular ABNIC are significant in both samples in constant terms, with a spectacular increase in ABNIC. The rate increase in the Reference sample is above that of the DAFEP sample. NFIC most contributes to the changes in TNIC. For this indicator, the Before and After situation is significantly different for the samples; income has been multiplied by t\vo. 1he analysis in the annex on multivariate analysis provides a better understanding of the observed trends in agricultural income and in non farm income. The graph below synthesises the above results, expressed in constant rupiah at 2001 value. '1 Significance of resulrsAlternative calculation taking into consideration the variation in the number of respondents between Before and Ajta sample is fully identical, therefore the simple difference, easier to understand is used here. This applies to al1 indicators using units per capita. 20 DAFEP Impact: Analys's Quantitative and Qualitative Irdlcators ·If' "-- AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Figure 4.1. Comparison of income indicators in 2001 constant value 2,500,000 2,000,000 1,500,000 1,000,000 500,000 o before afler after before arer DAFEP SPILLOVER REFERENCE Note: · AIC=Agi"ic, Production; AGBC, Non-AGBC=Agric.related; NFIC = Nonform. DAFEP Impact: AnalysIs of Quantitative a,cd Qualitative 21 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES J Table 4.3. Significance analysis of income indicators (DAFEP vs Reference) I Variable Condition Test Result Output TNIC Be ore Si ni cant REF CNIHA Before Sign ifican t DAFEP> REF After Sign ifican t DAFEP> REF Change Not Significant ARNIC Before " " t After Significant DAFEP REF After Not Significant Change Significant DAFEP REF Note: *signijicant at 10% with DAFElJ < REF) Interpretation 4.5 The table above presents the results of the logit analysis for each of income indicator, starting with the condition before the project (Bifore) and after the project (After). It shows also whether the changes in between Bifore and After conditions in the DAFEP sample are significantly different from the changes in the Reference sample and what is the nature of the change (> indicates higher; < indicates lower). 4.6 This table shows that the change in the situation of households in the DAFEP sample was significantly different from the Reference households for some indicators. DAFEP had no significant impact on TNIC. However, the following section on Welfare which uses another indicator as a proxy of household net income (YEC) for all households shows a positive effect attributable to DAFEP. Meanwhile, as far as farming households are concerned, the significance tests indicate that the project did not achieve the objective of raising households' income by 5 percent over five years, compared to a sample of reference farmers. However, given that TNIC is computed with the 22 DAFEP Impact: AnalYSIS of Quantitative and QcJalitative Indicators \.. AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES · incorporation ofARNIC and given that ARNIC cannot be satisfactorily used for the difference test, this conclusion must be taken with extreme caution. This does not mean either that DAFEP had no impact on households' income at aggregate level. In any case, the results show that there are other (more powerful) factors at work that have affected all samples in a similar way. 4.7 According to the testing method, change on households' agricultural income (ANIC) is related to the existence of DAFEP. However, for the reason indicated earlier (influence of 20% of not yet harvested parcels), no conclusion can be directly derived about the real significance of these changes. The results for CNIHA indicate that although DAFEP households had a significantly higher net income in both Before and After situations, there is no effect that can be directly related to DAFEP in the comparison of the change level according to the testing method. The DAFEP project cannot, thus, be associated with a significant change in farmers' agricultural incomes. 4.8 Reference households have performed better than DAFEP households with respect to ARNIC and ABNIC. Conversely, positive changes in NFIC are significantly higher for the DAFEP sample compared to the Reference sample. Further analysis will be needed to identifY which components of NFIC are related to this trend and whether they can be associated with some features of the project, and to clarifY the factors behind the observed significant results (see Annex 2). 4.9 It is problematic that even though the DAFEP sample showed significantly better income and welfare outcomes compared with the Reference sample, it is hard to explain why from the other evidence presented particularly since there is little change in the DAFEP sample on the agricultural practices/outcomes on which, presumably, DAFEP interventions would have had the largest impact. Most of the income growth was on the non-farm side of household activity. The difficulty in explaining why the DAFEP sample had these results opens the analysis to the critique that there was selection bias in the choice of DAFEP communities by the project managers in the districts e.g. more accessible, more open, more active, better organized, more social cohesion, which gave them a better foundation for improving non-farm income. However, the choice of the Reference sample was sufficiently rigorous to neutralize this critique of the results. One possible explanation could be that the farmers' benefited from the capacity building through DAFEP activities and this might have enabled farmers to become more confident and engage successfully non farm activities. However, this could not be tested. It is also theoretically possible of having positive externalities ofthe project that might have impacted on non-farm incomes, and thus that there might be an unexpected systemic effect affecting the DAFEP sample because of the DAFEP. ,im pact 4.10 TIlree indicators are used as proxies for DAFEP impact assessment on welfare: yearly expenses per capita (YEC) , assets value per capita (AVC) and share of food expenses per capita in total consumption expenses (SFC). As indicated earlier, the first one is equivalent to the total net income of all households, including households without agricultural income. It reflects the level of instant wealth in the household (amount of money circulating) while the second one reflects the state of more permanent wealth. Both are expressed in constant Rp/cap with 2006 values deflated by the increase in consumer price indexY The third one represents the relative importance of the budget needed by the household to fulfil the basic food needs of its members. It is calculated as a ratio between the amount of monetary expenses allocated for the purchase of food and the total budget 12 As the sample is not spatial and represent a population spread over nine provinces, a uniform deflator was applied based on the evolution of the national pI ice consumer index from July 2001 to July 2006 from the Indonesian Central Bureau of Statistics.1he deflator is 58,06%. DAFEP Impact: Ana:ysls of QuantltaCive and Qualitative Indicators 23 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES .J of routine consumption (including housing, clothing, education, etc.)13. The general trend observed everywhere in the world is that the less the share, the wealthier the household. Table 4.4. Before!Mter comparison of welfare indicators Sample Condition YEC AVC SFC Before 1,139,661 3,846,761 48.81% After 1,588,995 4,983,783 42.90% DAFEP 449,334" 1,801,296' -5.910/0" Difference 39,4% 46,8% 13,8% Before 1,229,715 4,347,160 50.12% After 1,587,028 5,232,200 43.57% Reference 357,313" 504,786ns -6.55°/0" Difference 29,0% 11,6% 12,8% Interpretation 4.11 The difference between Before and After for the three welfare indicator for all samples is significant. The result for YEC indicates that DAFEP households have seen their income increase by at least 10% more than the non DAFEP samples. They show also a 30% increase in assets per capita compared to the Reference sample. The changes in food expenses share to consumption expenses (-5%) are similar in all samples. 13In this calculation, rice produced and consumed by the household is not included since the objective is not to measure absolute but relative changes. It is assumed that the Dafep and the Reference samples do not differ significantly on rice self consumption. In the Benchmark survey, the average self sufficiency rate per household was 43 weeks for both Dafep and Reference samples. In the EoP this rate is 30 weeks also for both samples. Thus, the assumption is valid. 24 DAFEP Impact: AnalYSIS of Quantitative and Qualitalive Indicators \... AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES · Results of[t\FEPIReferencc comparison Table 4.5. Significance analysis of welfare indicators (DAFEP vs Reference) I Variable I Condition I Test Result I Output YEC Before Sign ifican t DAFEP REF AVC Before Not Significant After Not Significant Change Significant DAFEP> REF SFE Before Not Significant After Not Significant Change Not Significant Interpretation 4.12 The expenditures in the DAFEP sample were significantly lower before the project started, a result that is consistent with the observation ofDAFEP and Reference income situation before the beginning of the project (see Table 4.1). There are no significant changes in households' shares offood expenses between samples. Table 4.5 indicates that DAFEP households experienced a significantly higher change in the improvement of their welfare situation based on yearly expenses and asset value per ca::Jita. As YEC is also a proxy to household total net income, it indicates that a positive association can be made between the fact that the households belonged to the DAFEP project and an increase in their income and global wealth. This phenomenon however, is not independent from the results related to non farm income (NFIC). \Velfa... dj~ trilutioll analvsis 4.13 In addition to the analysis of changes related to individual household welfare level, the potential impact of the project on equity through the measurement of wealth distribution (consumption per capita and assets per capita) was also computed within each sample. The shape of data distribution (that is, to what extent the distribution differs from a normal-Gaussian- shape) is characterized with several indicators reflecting asymmetric distribution and/or inequalities such as Skewness, Kurtosis, and Gini coefficients. The changes in the distribution indicators are thus compared between samples so as to identifY whether the DAFEP sample significantly differs from the Reference sample. DAFEP Impact: AnalYSIS of Quantitative and Qualitative !ndicators 25 -"~.-"-"-"-"""-- """-- """- "--""--""--"""--"-------- AGRICULTURAL EXTENSION SERVICES IN INDONESIA: ...) NEW APPROACHES AND EMERGING ISSUES Table 4.6. Before/After comparison of welfare distribution Skewness Kurtosis Gini Coefficient Before After Before After Before After YEC DAFEP 1.68 0.95 8 2.83 0.10' 0.37 0.36ns Reference 1.48 1.00 s 2.13 0.19 s 0.36 0.36ns AVC DAFEP 2.39 6.23 s. 7.61 59.57 1 " 0.50 0.58 1 Reference 3.39 2.89'· 16.69 12.34 ns 0.60 053 ns Note: s· means significant at 10% Interpretation 4.14 In both samples, the Skewness and Kurtosis ofYEC decreased. This means that the shape of the curve representing the distribution of expenses tends to get closer to a Gaussian shape with a flatter form. It indicates an overall reduction of inequalities in expenses per capita. This result is confirmed by a slightly decreasing Gini coefficient in all samples. However, the test between DAFEP and Reference samples indicates that none of these changes is significant. 4.15 However, assets per capita shows a reverse trend for the DAFEP sample, the distribution becoming more unequal as both Skewness and Kurtosis increase, a fact confirmed by an important increase in the Gini coefficient. The Reference sample shows an opposite trend with an important drop in Skewness and Kurtosis as well as Gini coefficient. The tests indicate that this change is significant between DAFEP and Reference samples. Productivity and technology cluster 4.16 The project was expected to generate a 5% increase in productivity of the main farming systems by EoP in the villages. While not detailed, the concept of productivity seemed then to refer to land productivity, Le. the volume of agricultural output as measured in units per hectare. Participating farmers were also expected to adopt new/improved farming practices introduced by DAFEP extension workers. Productivity bnpact 4.17 Land productivity as described above could not be directly measured since it would require a compound of various crops and it is impossible to mix tons of paddy with number of coconuts. Therefore, two proxies were used to measure the impact of DAFEP on this issue. The first one is an aggregate indicator, the gross value of all harvested crops divided by the total area of cropped land (HCHA). The second one is a specific indicator (the yield) for each of the main crops that is then analysed at plot level {xYLD)14. However, this indicator cannot fully represent productivity changes since not all crops could be analysed for statistical reasons 15. 14 x is to be replaced by the name of the corresponding crop. 15Number of plots varies and the minimum number of required observations depends on the observed standard deviation of the Yield variable. The crops where the number ofplots meets statistical requirement are rice, maize, and soybean. 26 DAFEP Impact: Analysis cf Quantitative and Qualitative Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Results of Be forel After analysis Table 4.7. BeforelAfter comparison of productivity indicators Sample Condition HCHA RkeYLD MaizeYLD SoybeanYLD Before 10,547,283 3,587 2,247 1,074 After 11,416,023 3,565 2,400 1,831 DAFEP 868,740n· (22)"· 152 ns 7570 Difference 16 8% (1%) 6.8% 70% Before 9,007,712 3,680 2,095 785 After 9,052,828 3,684 2,127 813 Reference 45,l16ns 4ns 33 m 28 ns Difference 0% 1% 2% 4% Interpretation 4.18 There is no significant change in the total value of agricultural production when comparing Before and After situation in DAFEP and Reference samples. Rice yields have remained unchanged in all samples and changes in maize yields are not significant. Only in the case of soybean a significant change can be observed in the DAFEP sample. However, this result is based on a smaller number of observations and the sensitivity of the crop to local climatic conditions is high. I,;..ATte~;;ati~·~alccl;t;;n taking into consideration the variation in the number of respo~dents between Before and Mter sample have also been made. Significance of results is fully identical, therefore the simple difference, easier to understand is used here. This applies to all indicators using per capita unit. DAFEP Impact: Analysis of Quantitative and Qualitative Indicators 27 AGRICULTURAL EXTENSION SERVICES IN INDONESIA; ,...,) NEW APPROACHES AND EMERGING ISSUES Table 4.8. Significance analysis of productivity indicators (DAFEP vs Reference) Variable Condition Test Result Output HCHA Before Sign ifican t DAFEP> REF After Sign ifican t DAFEP> REF Change Not Significant RiceYLD Before Not significant After Not significant Change Not significant MaizeYLD Before Not significant After Sign ifican t DAFEP> REF Change Not significant SoybeanYLD Before Not significant After Sign ifican t DAFEP> REF Change Significant DAFEP> REF I nteqwetation 4.19 Households in the DAFEP sample show a significant difference for the total value of gross harvest in the Before and After conditions compared to the Reference sample, however the rate of changes observed between the two periods is not significantly different. DAFEP households achieved an 8% agricultural growth. The only significant difference in yield changes is observed for soybean where households in the DAFEP sample improved much more than in the Reference sample. Technology Impact 4.20 The measurement of the adoption rate of new technologies by farmers was not included in the methodology. The effect ofDAFEP on technological change and on change in production systems is also quite difficult to measure since new/improved farming practices introduced by extension workers are location-specific and thus vary Widely from one place to another. Four proxy indicators were identified whose combination can be reasonably considered an appropriate measurement of these changes. The first one is the total value ofinput per hectare (TVIHA). It is sought to reflect a higher intensity in cropping patterns and therefore through DAFEP/Reference comparison the impact of the extension services in terms of intensification. While this proxy could be disputable in developed countries where modern and already capital intensive agriculture is seeking to become more effective by reducing input use, it is still considered relevant for the case of Indonesia given the current low level of input use by households. The second indicator is an input/output ratio measured in monetary value (lOR) for rice, maize and soybeans. Thus, a lower ratio means higher input efficiency. 17 DAFEP targets were specified as a given percentage of farmers per village. Data collection would have required a representative sampling at village level for each project village resulting in a very high survey cost. 28 DAFEP Impact: AnalYSiS of Quantitative and Qualitative Indicators ... -~ ... --.-~ .....- -... - - - - ' - - " - - -_._-- "'-- AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES 4.21 Changes in production systems are estimated through a diversification index (Dl) and the calcula tion of trade income per hectare (TlHA). The former is simply obtained by counting the number of different crops grown by households in the sample, whatever the number of respondents growing the crops. The latter indicates an orientation towards more commercial agriculture and is therefore an indirect measurement of the capacity of the extension system to be more agri-business oriented 18 · When associated with a higher diversification index at sample level, it would be acceptable to consider that it helps measure an evolution of production systems. TlHA was calculated using the method indicated for computing CNlHA. Result!; of Before/After analysis Table 4.9. Before/After comparison of technology indicators Sample Condition TVIHA IORrice lt IORmz" IORsbn Ncrop TIHA Before 1,100,887 34.3% 33.5% 55.59% 20 8,034,8% After 1,419,359 28.5% 27.9% 31.26% 21 7,795,273 DAFEP 318,472 n& (5.9%)& (5.6%>,," (24.3%) 1 (239,622)n Difference 28.9% 17.2% 16.7% 43.7% (3.0% Before After 909,~ 1,146,2 32.48% 31.0% 14.73% 20.08% ~ 23.76% 19 20 6,954,288 7,227,059 Reference Difference 236,994'" (1.4%)n. (5.3%)'" (19.9%)" 1 272,771 26.1% (4.3%) (35.6%) (45.5%) (3.9%)m Note: TVIHA differencefor DAFEP and Reference is significant at 10% level Interpretat ion 4.22 TVIHA shows that households' inputs use increased in both samples in constant value. For lOR, changes are significant for rice in the DAFEP sample and for soybean in both samples. Diversification as measured by the total number of different crops grown does not change. The diminution in the value of trade income per hectare affects both DAFEP and Reference samples. As indiUlted in the Immediate Objectives in Appendix 2, DAFEP PIP Annexes. DAFEP Impact: AnalysIs of Quantitative and Qualitative Indicators 29 - - - - -.. -~---. ------ - - - - - - - - - - - - - - - - - - - - - _...._ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: .,,) NEW APPROACHES AND EMERGING ISSUES Table 4.10. Significance analysis of productivity indicators (DAFEP vs Reference) Variable Condition Test Result Output TVll-IA Before Significant DAFEP> REF After Not Significant Change Not Significant IORrice Before Not Significant After Not Significant Change cant DAFEP> REF IORmz Before Sign ifican t DAFEP> REF After Not Significant Change Not Significant IORsbn Before Significant DAFEP> REF After Not Significant Change Not Significant TIHA Before Sign ifican t DAFEP> REF After Not Significant Change Not Significant In tcrprctati on 4.23 Changes in the total value of input used per hectare are not significant between DAFEP and Reference. The project cannot be associated in a significant way, given the testing methods, with an intensification process that could bear witness of a technological change. However, the magnitude of changes that took place in input efficiency between DAFEP and Reference is tested significant for rice. In the DAFEP sample, households traded a larger amount of products before and after the project compared to the Reference sample but the impact of the project on trade income per hectare is not tested significant. (hmlitative assessment results 4.24 This section briefly summarizes the results of the assessment of qualitative indicators. As indicated earlier, these are grouped into four clusters so as to provide a more comprehensive approach. These clusters correspond respectively to extension, links, empowerment and awareness and are composed of nominal and ordinal variables. The following section presents the results of the extension and links cluster only. 30 DAFEP Impact: Analysis of Quantitative and Quantatlve hdlcators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Extenslon 4.25 This first cluster of qualitative indicators directly refers to this issue through a series of six variables that have been designed to assess the impact of DAFEP on access to information. These are five ordinal indicators: availability of agriculture information (AAI), willingness to pay for agriculture information (WPAY), accessibility to extension workers (AEW), extension meeting frequency (EMF) and improvement in access to agricultural information (IA<\I), and two nominal indicators: the best sources of agriculture information (SAl), and extension methods applied (EMA). The ordinal variables are combined into a single indicator reflecting general access to extension services (GAES). Comparing the Before and After situation ill the samples Table 4.11. BeforelMter comparison within each sample for ordinal extension indicators Sample Level AAI WPAY AEW EMF IAAI GAES 5% +5 +N5 +5 -5 +N5 +5 I DAFEP- Household 10% +5 +N5 +5 -S +NS +S 5% +NS +NS +5 -S -S +S Reference - Household 10% +5 +N5 +5 -5 -S +5 In terp ,·etatio 11 4.26 Households have better accessibility to agricultural information (AAI), but they don't show any change in the willingness to pay for agricultural information although there is positive trend in this attribute (WPA Y). Accessibility to extension worker increases significantly in all samples, but paradoxically there is a significant negative change in extension meeting frequency. Improvement in the availability of agricultural information is rated as positive and non significant for the DAFEP respondents and positive and significant for the Reference sample. Altogether changes are positive and significant for all household samples. However, chi-square tests applied to GAES indicates that the degree of these changes altogether is not significant. DAFEP Impact: Analysis 01 Quantitative ard Qualitative Indicators 31 AGRICULTURAL EXTENS!ON SERVICES IN INDONESIA. ..../ NEW APPROACHES AND EMERGING ISSUES 1 ;nation and Extension Methods Table 4.12. Before/After comparison of the best sources of information DAFEP- Household Reference- Household SAl Before After Sig. Before After - Sig. No know ledge 1.9% 1.4% -ns 3.8% 1.8% -s Other farmers 14.6% 27.0% +s 15.9% 30.5% +s Farmer trainer 10.50% 3.5% -ns 11.3% 3.9% -s Village leader 6.41% 9.9% +s 6.0% 8.0% +s Farmer groups 19.24% 13.7% -s 18.8% 16.4% -ns Extension workers 35.42% 29.4% -s 33.3% 25.7% oS Traditional market 1.60% 2.8% +ns 1.0% 1.3% -ns Sellers/distributors 0.44% 2.2% +s 0.9% 1.9% +ns Social or ganizations 0.87% 1.5% +ns 0.9% 1.0% +ns Traders 1.02% 1.39% +ns 1.21 % 1.13% -ns TV/radio 5.83% 3.55% -s 5.14% 4.66% -ns Newspapers 1.46% 1.08% -ns 0.45% 1.29% +ns Poster 0.58% 0.15% -ns 0.15% 0.16% +ns Other 0.15% 2.32% +s 0.15% 2.25% +s Interpretation 4.27 Households had been given the possibility to choose two best sources of information from a list of more than 10. 19 Significance tests highlight a shift in the sources of agricultural information from extension workers and farmer groups to other farmers. Here, "other farmers" refer to farmers with no specific position as "official" sources of information (as opposed to contact farmers, group leaders). Significance tests highlight a shift in the sources of agricultural information from extension workers and farmer groups to other farmers. This result is difficult to interpret. On one hand, one might say that extension workers were less important as best source of information, but on the other hand, it could be argued that extension workers were successful in linking farmers with other farmers, the latter becoming their best source of information. However, since the results are highly similar in all samples, it is hardly likely that the project had an effect on this trend. 19 This list was not given to the farmers so as not to influence them and make them choose the answers that they thought would please the enumerator. However, in the Benchmark survey enumerators had read the list to the household members at the time they discussed the topics, introducing thus a bias in the frequency of people unable to proVide an answer, which explain the difference berween Before and After on this point. This applies also for the next E.M.A. nominal indicator. 32 DAFEP Irnoact: AnalYSIS of Quantitative and Qualitative Indicators \.. AGRICULTURAL EXTENSION SERVICES IN INDONESIA NEW APPROACHES AND EMERGING ISSUES Table 4.13. Before/After comparison of extension methods implemented household samples DAFEP- household References- household EMA Before After Sig. Before After Sig. No knowledge 1.2% 15.7% +s 3.8% 21.1% +s PPL visits 21.5% 17.9% -5 21.5% 18.0% -5 Small meetings 23.5% 13.5% -5 18.2% 16.1% -5 Big meetings 16.0% 14.0% -5 17.2% 9.9% -5 Farmer teacher 3.7% 3.6% -ns 4.1% 2.5% -s Internship 1.5% 0.2% -s 0.9% 0.2% -ns Demplot 12.0% 21.5% +s 14.4% 19.8% +ns Evaluation meetings 1.2% 0.9% -ns 1.3% 0.8% -ns Radio Broadcast 0.0% 0.0% +ns 0.5% 0.4% -ns TV Broadcast 0.0% 0.5% +ns 0.3% 0.4% +ns Newspappers 0.0% 0.0% +ns 0.0% 0.0% +ns Field study 4.0% 4.2% +ns 2.8% 1.6% -s Meeting private indo 0.6% 0.2% -ns 0.3% 0.6% +ns Meeting researchers 1.0% 0.4% -ns 1.3% 1.0% -ns Courses/exercises 4.9% 5.1% +ns 2.4% 4.7% +ns Others 8.8% 2.4% -s 10.9% 2.7% -s Intcrpreta rion 4.28 The results show that in the household samples meetings with the PPL (field extension workers) were less frequent than in the Benchmark survey. Conversely, demonstration plots increased significantly in DAFEP and not in Reference. While the first results could be interpreted as the reduction of extension presence at village level, it could also be considered as a higher efficiency of extension activities, since less meetings and visits are accompanied with more hands-on field activities in the DAFEP sample. 'Ihe L,; oks c1u.;;ter 4.29 1his cluster relates to the expected effect of DAFEP on changes in farmers knowledge, attitudes and skills. The cluster here focuses on the links between farmers and their upstream and downstream agricultural environment. The cluster includes five ordinal variables: accessibility of input markets (AIM), accessibility of agriculture input information (AAII), joint purchase of agriculture input (TPAI), accessibility ofoutput markets (AOM), and joint marketing of agricultural outputs (TMAO). One nominal variable is also related to this cluster, sources of capital (SC). These variables are combined into a single indicator reflecting the improvement in upstream and downstream links (IUDI). DAFEP Impact: Analysis of Quantitative and Qualitative Indicators 33 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: .../ NEW APPROACHES AND EMERGING ISSUES Table 4.14 Before/After comparison within each sample for ordinal links indicators Sample Level AIM AAII JPAI AOM JMAO IUDI J 5% +5 +N5 - N5 +N5 -N5 +5 ! DAFEP- Household 10% +5 +N5 -5 +N5 - N5 +5 I 5% +5 +N5 -NS +N5 -5 +5 I Reference - Househol d 10% +S +N5 -5 +N5 -5 +5 I 4.30 All households have a significantly better accessibility (fewer barriers) to input market in general. Although there is no significant change in accessibility to input information, there is a positive trend. Joint purchasing ofinputs has not experienced a significant change, and the tendency is negative. There are no significant changes in the access to alternative output markets, although again the tendency is positive. Again, there is no significant change in joint marketing of agricultural commodities, and the negative tendency is similar to the joint purchased of inputs. The composite indicator (IUDI) resulting for the computation of the former ones shows a significant positive change in overall links upstream and downstream agriculture for all households samples. Cross comparison DAP'EP versus Reference 4.31 Chi-square tests indicate that for JPAI and JMAO there is a significant difference between DAFEP and Reference. For JPAI, the Reference sample shows a significantly higher positive change compared to DAFEP (that is more joint purchase of input). Conversely, DAFEP sample shows a significantly higher positive change in joint marketing of agricultural products. For all other indicators, including the composite indicator IUDI, there is no significant difference. Nominal variables ana~vsis: Source rifCapital Table 4.15. Before/Mter comparison of the sources of capital for the farmer sample DAFEP-household Reference-household SC Before After Sig. Before After Sig. Individual 49.7% 91.0% +s 49.0% 93.0% +5 Gose Relative 32.1% 4.6% -5 28.4% 4.6% -s Lenders 18.2% 4.3% -5 22.6% 2.3% -5 In tcrprctation 4.32 All samples follow a similar pattern with similar magnitudes showing a decrease in the share of external sources of capital in favour of own capital. This trend can be associated with the increase in total income at household level, in particular with non farm income. 34 DAFEP Impact: AnalysIs of Quanlltat:ve and QuaHative Indicators - - - - - - -....- -.... - -.... - -... - -... --~~-- ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Svnthe.;is ofthe results 4.33 The following tables provide a synthesis of the results. Highlighted in Bold case are indicators where changes occurred between Before and Mter situations, where there is a significant difference between DAFEP and Reference and where this difference favours the DAFEP sample. Indicators in Italic Bold case correspond to situations where the difference favours the Reference sample. When indicators are not highlighted, either there is no change in the Before and After situation, or that these changes are not significant when comparing DAFEP and Reference samples. 4.34 The table shows that most significant changes occurred in the Income and Welfare cluster of indicators, DAFEP households showing a significant improvement in their overall welfare situation compared to the Reference sample (positive changes in yearly expenditures that are a proxy of total net income for all households and in non farm income and assets per capita). The project thus succeeded in increasing households' income by 5% as targeted at the beginning of the project, (in constant terms). DAFEP households also performed better in agricultural income per capita, though the difference in the After and Before situation is negative for all samples. 4.35 However, the analysis of the technology and productivity cluster shows that significant changes in favour ofDAFEP households are rare. They relate to only three indicators: soybean yield and input me ratio in rice and soybean cultivation. Altogether the direct impact ofDAFEP on technology and productivity indicators has been limited. In particular, there are no significant changes in the yields of rice, the major and most common crop grown in the samples and in net income from trade. 4.36 In relation with qualitative indicators, this impact evaluation study shows that, based on the statistical approach used, sampling, conception of the questionnaire, data collection, data analysis and significance tests-, the project seemed to have not very much changed the target group as far as attitudes, skills and practices related to extension are concerned. Indeed, while positive changes have occurred in this field between the Benchmark survey and the EoP survey, they cannot be specifically considered as an impact of the project. DAFfY impact: Analysis of Quantitative and Qualitative Indicators 35 AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES J dts fo.- DAFEP performance indicators (at 5% ,Hors Significance I Before! After'" DAFEP vs Reference"''' 20 Total Net Income per capita (Rp) farming households Yes+ Yes DAFEp Agricultural Net Income per Capita (Rp) Yes- Yes DAFEP Crop Net Income per Hectare (Rp!Ha) No No Agriculture Related Net Income per Capita (Rp) Yes + Yes Reference Agribusiness Net Income per Capita (Rp) Yes + Yes Reference Non Farm Net Income per Capita (Rp) Yes+ YesDAFEP . Total Yearly Expenses per capita (Rp) all households Yes + Yes DAFEP I Total Asset Value per capita (Rp) Yes+ YesDAFEP Share of food expenses (%) Yes + No Notes: * 1be + or - sign indicates the sense ifthe change (better or worse) ** 1be name ofthe sample (DAFEP or Riference) indicates which one shows abetter peiformance **.Nominal indicator, no test ifsignificance, but changes in all samplesfollow a similar trend. 'I'able 4.17. Presentation of significam'e results for DAFEP performance indicators: Cluster ofTechnology and productivity .-elated indicators Significance I Before! After'" DAFEP vs Reference"" Gross crop value per hectare (Rp/ha) No No Rice Yield (t/ha) No No Maize Yield (t/ha) No No Soybean Yield (tlha) Yes" YesDAFEP Coconut Yield (t/ha) No No Total value of input per hectare (Rp/ha) No No Input/output ratio rice (%) No Yes DAFEP Input/output ratio maize (%) No No Input/output ratio soybean (%) Yes+ YesDAFEP Diversification index (number) No No Trade income per hectare (Rp/ha) No No N?tes: · 1be + or - sign indicates the sense ifthe change (better or worse) ·· 7be name of the sample (DAFEP or Reference) indicates which one shows a better peiformance ... Nominal indicator, no test ofsignificance, but changes in all samplesfollow a similar trend. At 10% level 36 DAFEP Impact: Analysis of Quan:i:ative and Qualitative Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Table 4.18. Presentation of signiflcance resnlts " j Cluster of Extension related indicators Significance Before/ After" DAFEP vs Reference"" Extension methods implemented Yes No...... Sources of agriculture information Yes No""" Access to agriculture information Yes + No Willingness to pay for agriculture information No No Access to extension workers Yes + No Extension meeting frequ ency Yes - No Improvement in access to agricultural information Yes+ No . General access to extension services Yes + No Notes: * The + or sign indicates the sense ~fthe change (better or worse) ·· 7he name ofthe sample (DAFEP or Reflrmce) indicates which one shows abetterpeljOrmance ..* Nominal indicator, no test ofsigniftcance, but changes in all samples follow a similar trend. 'Elble -4.19 Pnsentation of Significance results for DJ\ FEP perfonnance indicators Clustc of ;~in·(s Related Indicators Significance I Before/After'" DAFEP vs Reference*'" Accessibility of input markets Yes + No Accessibility of agriculture input information No No Joint purchase of agriculture input No Yes Reference Accessibility of output markets No YesDAFEP Joint marketing of agricultural outputs Yes - (Reference) No SourcE~s of capital Yes No....'" Improvement in upstream and downstream links Yes + No Notes: · The + or - sign indicates the sense I!lthe change (better or worse) .. The n,/me qlthe sample (DAFEP or Reflrence) indicates which one shows abetter peljOrmance ... Sommal indicator, no test ofsigniftcance, but changes in all samples follow a similar trend. ']]IC result!- from the multivariate analysis (sec Annex 2) hugely endorse the above findings. 4.37 Overall, the results indicate that DAFEP extension agents were judged as "not good" to "good enough" on important key factors such as being experienced, having suitable knowledge on new technologies, having knowledge/understanding of local conditions, etc. Furthermore the study showed that the main qualities that were expected from extension workers by farmers are "experience" and "technical knowledge". This still corresponds to the traditional approach of extension services as DAFEP Impact: AnalYSIS of Quantitative and QLalitalive Indicators 37 ----- - - - - - - - - -....- - -....- -... ~ ....- -...- . - -...- - - - - AGRICULTURAL EXTENSION SERVICES IN INDONESIA: ~ NEW APPROACHES AND EMERGING ISSUES the means for transfer of technology. Finally, according to the farmers, the benefit ofjoining farmer groups were mainly exchange on farming/cultivating techniques (30.3%) and exchange of village development (23.0%). The benefit of exchange on farming/cultivating techniques shows a significant decrease meanwhile other benefits have increased significantly. Unfortunately detailed data on the other benefits cannot be processed in a similar way. The conclusion is that there was no significant change of routines done by extension workers, the DAFEP activities did not result in something more ground-breaking such as joint marketing in groups or joint purchasing. 4.38 Weaknesses of the Impact Evaluation: Given the fact that the Farmer-Managed Activities (FMAs) - Component 1 was a major institutional innovation piloted by the project, the design of the EOP survey did not specifically identity the beneficiaries of the FMAs and of its impacts. The households selected in the DAFEP sample are households who participated in DAFEP activities, but since not all households received grants for FMA, there is no perfect overlap between DAFEP membership and FMA. The DAFEP sample was a sample of households participating in DAFEP activities. Many of the households in the benchmark survey (carried out before the implementation ofthe FMA component in 2002) did not participate in the FMA activities. This limits the usefulness of the evaluation to a certain extent. The analysis may have been more useful had it targeted FMA grant recipients versus the rest - DAFEP ( FMA recipients) Spillover ( same village not FMA recipients) and the Reference ( non-DAFEP, non-FMA similar villages). The funding of grants for addressing market failures in information was at the core of the project. The analysis unfortunately does not reveal whether this was a necessary part of the project or a desirable feature of future projects. Finally, the number of indicators (39) while attempting to be comprehensive perhaps could have been simplified to focus on key indicators that would have sufficed to demonstrate the impact of the project. Data quality and reliability are in general satisfactory (see Annex 1 for more details). 38 DAFEP Imoact: klaiYSls of Quantitative aCid Qua:itat've ICidicato's · $.;;1 "'-. AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Assessment of F armer Managed Activities and Institutional Reforms 21 5.1 A major weakness of the impact evaluation was that many of the households in the DAFEP sample did not participate in the farmer-managed activities (FMA) of component 1, the core institutional innovation of the project. Further, the impact ofthe institutional reforms at the district level were also not addressed. This section draws from the data and analysis of the F AO team that was responsible for the preparation of the ICR. 5.2 Development of FMAs and action planning to village commuOltles to assist in change from reactive to proactive farming was facilitated by participatory rural appraisal (PRA) techniques whereby farmers having common interests were involved in assessment and ranking of problems, potentials and alternative solutions. Various PRA techniques were evaluated in early phases and these were reduced to five to seven assessed by farmers to be the most useful. A supplementary PRA approach was specifically developed to cater for agro-forestry to focus on the community and its needs and identifY potential benefits of trees. 5.3 1he first cycle ofFMAs was started in most districts in September 2002 with sub-district extension workers playing a leading role in their preparation. Understanding of extension staff, farmers and other stakeholders ofDAFEP principles and concepts was poor. Farmer participation rates were low, planning processes were poorly understood and FMAs often had little relationship to better farm management or improved profitability of farm enterprises. Nevertheless, many of these early FMAs proved highly successful (Annex 4) thereby demonstrating to farmers what could be achieved. The second cycle ofFMAs was again delayed (until late in 2003), as was the third cycle, which commenced This chapter draws from DAFEP supervision reports and the DAFEP Implementation Completion Report, Report No. 33582, 21 November 2005. Assessment 01 Farmer·Managed Acltvlties and Institutional Reforrr·s 39 AGRICULTURAL EXTENSION SERVICES IN INDONESIA, ...) NEW APPROACHES AND EMERGING ISSUES in November 2004. I'\evertheless, by 2005 most participating villages had enthusiastically adopted the DAFEP model of farmer-led learning. There had been a marked improvement in the quality of FMA submissions, associated family agro-business plans and village action plans. ;Most participating villages visited by the ICR team reported examples of high adoption rates following training and numerous examples of high returns to the training investment (Annex 3). Isolated examples were noted where elite groups appeared to have captured the F;vtA process for personal gain with limited flow-on to other village members. Participation ofwomen has increased dramatically under DAFEP at all levels from farm budgeting, planning, income generating activities to village decision making through UPKGs and representation on DECs. 5.4 From the Impact evaluation however, there were also complaints that the Farmer ;Managed Activities (FMA) activities did not allow for the acquisition of physical items directly related to agricultural production such as inputs, tools and equipment, credit/revolving funds. Building or rehabilitation of farm infrastructure was prohibited too. The philosophy behind the FMAs was that farmers participation in extension activities would be funded. Purchasing of inputs and equipment was authorised provided these would be used exclusively for extension purpose. Other types of activities included the funding of studies by farmers, exhibitions, visits to other areas, farmer field schools, demonstration plots, etc. Many participants expressed regret that the project did not supply them directly with credit, equipment or inputs for agricultural production. Institutional Reforms: 5.5 The establishment of the Rural Extension and Information Centers at district (Balai Informasi dan Penyuluhan Pertanian BIPP) and sub-district (Balai Penyuluhan Pertanian BPP) levels were meant to consolidate extension programs and staff of isolated district agricultural service units for provision of sub-sector advice (food crops, estate crops, livestock, fisheries, and forestry). DAFEP aimed to strengthen this process by facilitating team-building within and between sectors and encouraging staff to develop an integrated extension program using farming systems approaches including agro forestry. 5.6 At district level, the institutionalization of integrated extension services remains tenuous. Some districts, recognizing potential benefits from farmer-led approaches to extension using the DAFEP model, have supported integrated extension centers through guaranteed on-going finances and Bupati decree. In districts where decentralized, farmer-led extension is working well, and benefits accruing from application of DAFEP principles are recognized, there has been strong support from the head of the district (Bupati) and district parliaments in institutionalization of BIPP's (e.g. Magelang district Central Java and Maros District in South Sulawesi). Where there was strong support for DAFEP, funding has also been provided for expansion ofDAFEP concepts to other districts (e.g. in Maros district Rp 800 m for expansion to an additional 20 villages during 2005). Elsewhere support is weaker and in some cases BIPPs have either been abolished (Banyumas, Kotabaru, Tanah Laut, Timor Tengah Selatan Districts), or their echelon level downgraded in effect restricting the career path and status of extension staff. In many cases BIPPs/KlPPs are seen simply as cost centers and their potential contribution to poverty alleviation is not recognized. 5.7 District Extension Committees (DECs) were established by Bupati decree in the participating districts to provide district level policy support for farmer-led extension, facilitate flows of information to meet farmer demands, assist in building linkages between farmer groups with NGOs, Universities and Industry. As such, the DEC was envisaged as potentially playing a pivotal role in achievement ofDAFEP objectives. While their objectives rely on committee members representing 40 Assessment of Farmer-Managed ActiVities and hstitutlOnal Reforms - - - - - - - - - - -....- - -.... - - - - - - - - - - - - - - - - - - - - - - - - "-- AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES all influential sub-sectors (Government, NGO, Research Institutes, Farmers, Agro-business, etc.), most DECs remained dominated by Government through an over-representation (usually 50 % or more) or perceived authority of Government members, and as yet have largely failed to realize their potential. DECs have not received information from BIPP/KIPPs on actual benefits accrued from implementation ofDAFEP at farmer level except for verbal reports and, where DEC is active (e.g. Magelang district), members have assessed impacts of FMAs for themselves through farm visits. This information failure significantly reduced the effectiveness ofDECs in fulfilling their roles such as establishing priorities for extension effort. Further external support and clearer definition of roles and responsibilities for both BIPPs and DECs is essential to consolidate progress so far. 5.8 Farmers and NGOs however were enthusiastic about the DECs: NGOs because of the informa tion forum it provides to focus and refine their district programs and farmers because, often for the first time, they have direct access to decision-makers. On the other hand, DEC meetings have often dealt with inconsequential issues and failed to engage decision-makers who are often represented by staff members with no end-of-line authority. There are examples of how DECs have facilitated engagement of research institutions in partnerships with farmers to solve issue-specific topics however, it will take time before these committees mature and begin to tackle long-term, industry wide or whole farming systems approaches to improving productivity. Partnerships with industry initiated through DECs have also begun to show benefit. For example, farmers have been able to negotiate a direct marketing agreement for rice produced throughout the Gowa district whereby they now receive 20% more for their product. But overall, institutional development impact has been modest at best. 5.9 DAFEP's principles of integrated, farmer-led extension based on establishing public-private partnerships have been enthusiastically accepted by many front-line extension workers. Such change requires extension workers trained in new extension methodologies of facilitation, participative processes, media etc. These extension workers require a whole systems approach to agriculture including business management and marketing. They are not technical specialists but generalists, and their role is one of facilitating technology transfer rather than providing this themselves. Under the DAFEP model, technical expertise in specialist fields is provided by agro-business, research institutions and relevant Dinas, facilitated by DEC. Management at the district level has yet to understand and put these principles into practice. Outside ofDAFEP, training for extension workers continues to focus on acquisition of technical skills and not new extension methodologies. Assessment ~, Farmer-Managed Activities Instltu'lonal Refon'1s 41 --- ----------------- --- ----------- Concluding Discussion and Emerging Issues 6.1 Overall, the report card of DAFEP is mixed. While on the one hand, the project succeeded in improving farmer welfare, the impact on productivity and technological change was limited. Nonetheless, the lessons learned from DAFEP have important implications for the future of agricultural service delivery in Indonesia in the context of the broader structural transformation that is already underway. This chapter discusses some of the emerging issues. Diversification and Rural Productivity 6.2 Perhaps the most striking result from the DAFEP impact evaluation is the secular increase in non farm incomes across the board. This is a clear indication that diversified farming will be the solution for farmers whose scale of operations or land quality does not enable them to support a family from rice farming income alone. Between 2001 and 2006, there was further fragmentation in the landholdings with the average area/household declining from 1.48 hectares to 1.32 hectares. Rice areas also shrunk from 64 percent of total cropped area to 56 percent. There was a dramatic increase in the cultivation of chillies across all samples. This also confirms broader trends in the economy. The most rapidly growing type of farming is horticulture, with horticulture farmers nearly doubling between 1993 and 2003 to 38 percent of all farm households (2003 Agriculture Census). Rapid urbanization and income growth is fueling changes in food consumption which increased by 8 percent (per capita, real) over 1996-2002. (Susenas, 2005). This consisted almost entirely of growth in high-value food consumption while per capita consumption oflow-value grains and tubers actually declined (Susenas, 2005). Even rural households remaining primarily engaged in agriculture can be expected to continue to rely on diversified income from other labor and business sources (because of small land holdings); these already account for half ofrural household incomes nationally (PATENAS). Concluding Discussion and Emergmg Issues 43 .j AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES ...J 6.3 Diversification means switching to higher-valued crops, livestock and fish production in response to new types of consumer demand. Small farmers will need technical assistance from either the public or private sector if they are to respond successfully to these new market opportunities. However, with the recent steep increase in rice prices 22 , the Government is taking important steps in several areas to boost productivity in the agricultural area notably the promotion of hybrid seeds, irrigation investments etc. However, current policies still fail to address the structural problems which inhibit productivity growth in the sector. In particular, public expenditures are still currently biased towards subsidies (on fertilizer, seed and credit) despite evidence suggesting that these subsidies have very limited impact on production or productivity. 6.4 Farmer associations/rural producers organizations (RPOs) in Indonesia are still considerably weak. During the 32 years of "New Order" era (1966-98), Government policies called for a total control of the State over all the functions that could have been performed by RPOs which inhibited the rise ofgenuine and independent RPOs. Official organizations such as HKTI (Association ofIndonesian Farmers Groups and Fishermen Groups), the KrNA (National representative of the farmer, the "best farmers" are the members), and the village cooperatives (KUDs) were the only ones permitted to operate under the aegis of the state. While there was some success from these state-sponsored groups in supporting the rice intensification package in some irrigated areas, the numerous cases of malfeasance and negative environmental impacts (decreasing soil fertility, increasing pests) resulted in a general distrust oHarmers towards all forms of State intervention in organizational development. Yet experience from other countries indicate that strong RPOs can be a key asset for agricultural development. However, due to the unpleasant recent history, the development of genuine farmers' organizations is still a slow process and RPOs in Indonesia are still on need of a firm basis on which to establish their growth (Roesch et aI, 2002). 6.5 With support from DAFEP, the Village Extension Management Unit (Unit Penge10la Kegiatan Gabungan UPKG) received an annual grant (FMA grant) to finance extension activities identified by a group of farmers and approved by the District Extension Committee. The UPKG scheme is an embryo of a village farmer organization/association and through this farmer groups developed project proposals to receive training to start up or improve their farming activities (for example fish growing, duck rising, biological coffee, corn, vanilla, bee-keeping etc.). There is a need however to clearly separate the role of UPKG which is a farmer village committee (UPKG selects project proposals) from the role of producer groups who prepare and implement projects. The UPKG/FMA scheme can be improved and systematized as an effective way to facilitate small farmers' access to technology and markets. UPKG leaders should be accountable to the farmers to ensure that UPKG is inclusive and not serving a few large or influential farmers as some of the evidence seems to suggest. The village-level producer groups need to be encouraged to federate at district and provincial levels. These federations could then engage in partnership activities with private enterprises and/ or extension and/or research institutions with appropriate technical assistance that keeps the government at arm's length. In addition to technical support, farmer associations/federations also need to receive organizational support to help them resolve inevitable organizational management 22 Rice prices have increased by 110% since January 2004 - far faster than the general rate of inflation (32% over the same period). 111is large increase in the price was the main reason for the increase in poverty between 2005 and 2006 the 37% increase in the rice price over the least year alone makes it likely that poverty will increase again in 2007. The main reason for the dramatic increase in prices is government policy Although the fuel price rises in 2005 triggered high inflation, the increase in rice prices has been far than the increase in input costs for rice production. 1\1oreover, world rice prices have been quite stable - Indonesian prices are now 73% above world prices. The main reason for the increase in prices is the rice import ban in place since January 2004 periodic but increasing shortages in supply as imports can not fill the gap. The effect has been particularly severe this year because of a moderate El Nino which has delayed planting and will delay the harvest. (\Vodd Bank, 2007, forthcoming) 44 "'-. AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES difficulties which will arise when implementing their activities (issue offree riders, internal conflicts, accountability mechanisms to members and member organizations, transparency, participation etc.). While the approach has its danger in that farmer groups remain active only as long as a project provided subsidies or access to resources (Cary, Zijp, Byerlee, et al. 2002), there are other examples (Senegal, Colombia, Mali, Equador) where RPOs are increasingly partnering with research and extension agencies. With Indonesia moving towards a more pluralistic institutional arrangements for extension, building the capacity of RPOs and other service providers will be essential to empower users and expand the pool of qualified service providers. Research-Extension Linkages 6.6 The institutional and management reforms (supported by the \-\Torld Ban, \VB and the Asian Development Bank, ADB) has helped IAARD to transform its organizational structure and institutional culture towards a demand-driven, "farmer first" strategy. The establishments of the 1423 Assessment Institutes for Agricultural Technology (AIAT) to serve the regionalization ofagricultural R&D has arguably been the most significant change in the National Agricultural Research System Ci'-IARS). While some progress has been made over the past 10 years, capacity remains generally weak and is substantially failing to deliver improved technologies and practices to farmers and agribusiness. Indonesia's agricultural research expenditure has declined dramatically since the early 1990s compared with its neighbors. Real expenditure on public agricultural research in 2000 was no greater in real terms than a decade earlier, and presently ranks Indonesia near the bottom compared with other Asian countries in terms of agricultural research spending relative to agricultural CDP. However, the separation of the research and extension functions within the organization of the l\10A (between IAARD and AAHRD) has also militated against both ensuring focus on farmer's problems while setting the research agenda, and effective dissemination of research results which have till now, relied heavily on the use unidirectional 'technology transfer' approaches, supported by field extension methods such as demplots, as the primary farmer extension method over the past decade. lhe province-level BPTPs are the major adaptive R&D providers at this level, and they are being required to continue an evolution from being research organizations to technology assessment and knowledge transfer units. 6.7 Development of a more multidisciplinarylfarming systems approach as piloted by DAFEP is needed to tackle identified real-world production issues that often span several segments of a value chain, rather than the present situation where researchers usually work in disciplinary isolation, often producing research outcomes that address only part of the original problem and therefore have limited immediate application. The stagnant level of rice yields points to the need for research on how to increase yields without increasing costs i.e., how to improve rice profitability- in the interests of both farmers incomes and stability of food prices. Nevertheless the fact that rice based indicators were not significantly different in DAFEP and reference groups is not surprising since the choice of enterprises for focus by farmers was open and incentives for rice have been poor. Donors (including the Bank) have also pursued separate projects under the research and extension agencies that has also contributed to the disconnect. Building on the lessons of DAFEP, the Bank's new Farmer Empowerment through Agricultural Technology and Information (FEATI) project, will for the first time combine research and extension in a more comprehensive approach) which responds to the above set of issues. 13 of" hich are fully functional today :.vlaluku having been burned down during sectarian violence in 1999, Concluding Discussion and Emerging Issues 45 ---,-_._---...._ . __ _ - _ _ - - - - - - - - - - ... ... ....- - - - . ------------- AGRICULTURAL EXTENSION SERVICES IN INDONESIA: ",.,) NEW APPROACHES AND EMERGING ISSUES 6.8 DAFEP was predicated on the idea that lack of information was a binding constraint in farm incomes. The data on diversification and decline in rice areas shows that other factors were probably at least as important - for example agricultural prices, which would have been declining for rice during the implementation period. Secondly, the constraint of credit to apply any new ideas, particularly in Eastern Indonesia but also elsewhere owing to various access constraints in the banking system was a critical factor the FlVIA grants did not increase access to implementation credit. 6.9 Evidence from the survey shows that extension workers are not the most important source of information and extension. This confirms other findings elsewhere (Faure and Kleene, 2004) stating that farmers learn more from others farmers than from external advisors. Projects like DAFEP intend to promote a paradigm change in the profile of effective extension workers. Technical skills become less important than communication skills, ability for facilitation, and a capacity for human relations. Commitment is also a key issue that is facilitated if the extension worker comes from the area, has rural roots, speak the local language. As a key issue is whether the farmers trust them and confide in their capacity, we should also make sure that the beneficiaries are involved in "difining criteriafor selection and quality standards (an expert is only an expert ifslhe is recognised)" (Christoplos and Kidd 2000). Contemporary thinking on new extension roles emphasizes that the new extensionist will iften need to be one-third management specialist, one-third communications specialist, and one-third technical specialist. (Gary, Zijp, Byerlee, et al. 2002:26). 6.10 "While DAFEP envisaged that information of interest to farmers would be disseminated through printed, visual and audio media, there is little evidence to indicate that this was successful. It was intended that a Farmers' Information and Technology Promotion Service (FITPS) be established at the district extension centers to improve farmers' access to information on technology, credit, markets and farmer network activities. However, the ICR notes that it was difficult to ascertain how demand for the material was identified and farmer groups have not rated it highly. Similarly, there has been little or no recognition of other services ofFITPS. 6.11 ICTs as tools are still underutilized in extension services (Alex et aI, 2004). The low capacity of rural information and communication technologies (ICT) development is widely acknowledged to be a barrier to increasing rural incomes and productivity in Indonesia. Rural communities need up-to-date information on sources, availability and costs of inputs for production, and also on the potential of different techniques and technologies used for production, processing and marketing. The information that is often most relevant to improving livelihoods is non-technical, including the role and responsibilities of different institutions in the provision of key services, such as agricultural extension, credit, health and education, and where to go and who to ask for more specific information. ICT offers an opportunity to make rapid progress. It is important that information is available in an appropriate format and language, and that rural communities have the capacity to access, analyze and act on it. With decentralization and the new political and institutional environment in Indonesia, there is an opportunity to use ICT to support the agricultural development agenda and to improve the delivery of agricultural services in innovative ways. Future of Extension Services in a Decentralized Context 6.12 Extension faces a major challenge in Indonesia, similar to public sector extension systems in many countries that are seeking to advance structural, financial and managerial strategies to reform extension (see for example Rivera, 1991; Rivera and Gustafson, 1991; Rivera and Zijp, 2002). The 46 Concluding Jiscuss:on ard Emer gl1g "--. AGRICULTURAL EXTENSION SERVICES IN INDONESIA NEW APPROACHES AND EMERGING ISSUES literature indicates that there is no one-size fits all even in the same country. However, the public sector continues to play an important role in coordination and poverty reduction objectives. Various extension approaches need to be considered and have to be flexible to accommodate changing policies, technologies and the needs of farmers. While there is a move towards a more pluralistic system of extension funders and service providers, the public sector still continues to play an important role. Private extension services will assume greater importance in the dry-land cash cropping sub-sector in eastern Indonesia. This is because exportable commodity production is being increasingly supported by the private sector. However, quality enhancement for service providers - public or private continue to be a big gap in Indonesia and elsewhere (Alex et aI, 2004). 6.13 National development priorities include improving extension, and an acknowledgment that new approaches will be required in the context of the changed institutional context of decentralized extension service delivery. However, evidence from DAFEP indicates that the commitment oflocal governments is key to improving extension service delivery. Local governments need to be assisted to take up applicable models as a part of: (i) moving from top-down to participatory approaches to extension subject matter prioritization and delivery; (ii) shifting the balance from input and technology dissemination to dissemination of market and upstream information and technology; and (iii) moving from centrally-managed extension services to decentralized services, and increase space for private delivery of extension services. 6.14 PartiCipatory extension approaches such as those pioneered by DAFEP face their own challenges (Alex, et al 2004). These approaches require changes in the roles for extension workers - from messengers to facilitators - and changes in the way messages are transferred to farmers, organizational srructures, facilities offered to local communities. While DAFEP was reasonably successful in some respects, the results of the qualitative analysis shows that it is quite a challenge to change the mental processes and working routines that extension workers have had for over two decades. In these routines, technology and production are the key words and the processes are invariably linear and top-down. Similarly, changing farmers' routine and expectations with regard to extension services i~ also challenging. Farmers too, for decades have lived with the custom of grouping in order to get fimds and doing what the government tells them to do. Indonesia, as elsewhere, much learning is still needed on how to develop democratic procedures, inelusiveness, and linkages that integrate r~lral communities. The dramatic increase in the importance of non-farm incomes underscores the premise that extension is beingjorced to embrace a broadened mandate [ ..} shiftingfrom an "agricultural" /.1 a "rural"focus in programs, recognizing that agricultural productivity may not always be the best way /.? improve peoples' livelihoods (Alex, et a1. 2002). The promise of participatory extension is that local people who have a sense of ownership in projects and activities learn to be independent through appropriate technical support. This independence and self-reliance is the ultimate purpose of promoting participation in extension development Concluding Discussion and Emerging Issues 47 ----------------- -------~--------~-------------- AGRICULTURAL EXTENSION SERVICES IN INDONESIA. "",,) NEW APPROACHES AND EMERGING ISSUES 48 References !l '6 "-- AGRiCULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES ANNEX 1. Methodology and Indicators 1.1 A list of 39 indicators was established in consultation with DAFEP PMU and World Bank staff. These indicators reflected the concern of the project designer to assess various (economic, social and technical) impacts the project was expected to have on the target group. As a result of the multidimensional function ofthe project, this list included quantitative and qualitative indicators divided into six clusters. The final list of indicators includes 15 quantitative indicators measuring income, welfare, productivity and technology changes and 30 qualitative indicators related to changes in extension, links, empowerment and awareness, as presented below. Clu.fter ofIncome and Weifare indicatorJ Total Net Income per capita (Rp) · Agricultural Net Income per Capita eRp) Crop Net Income per Hectare (Rp/Ha) · Agriculture Related Net Income per Capita eRp) · Agribusiness Net Income per Capita (Rp) NOl~ Farm Net Income per Capita eRp) Clu.fter ofUTeifare indicators (all indicators are in Rp/capita) Total Yearly Expenses per capita eRp) · Total Asset Value per capita (Rp) Share of food expenses in the household consumption budget (%) Cluster ofProductivity and Technology indicatorJ Gross crop value per hectare (Rp/ha) · Yield (t/ha) for rice, maize, soybean, and coconut · Total value of input per hectare (Rp/ha) Input/output ratio (%) Di\-ersification index (number) Trade income per hectare (Rp/ha) Cluster ofExtension related indicatorl4 · Availability of agriculture information (perception, ordinal) · VVillingness to pay for agriculture information (perception, ordinal) · Accessibility of extension workers (perception, ordinal) Extension meeting frequency (factual, ordinal) Im:Jrovement in access to agricultural information (perception, ordinal) Sources of agriculture information (factual, nominal) Extension methods applied (factual, nominal) General access to extension services (composite, ordinal). 24 'Ordhal'refers to the possibility to rank the factors so that results can be measured with scores. 'Nominal' indicates that there are several options and results are expressed in a distributive form (share). 'Factual' relates to what the respondents do, or to what happened. 'Perception' relates to what the respondents think awj !nd,cators 49 AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES ..J Cluster ofLinks related indicators · Accessibility of input markets (perception, ordinal) · Accessibility of agriculture input information (perception, ordinal) · Joint purchase agriculture input (factual, ordinal) · Accessibility of output markets (perception, ordinal) · Joint marketing of agricultural outputs (factual, ordinal) Sources of capital (factual, nominal) · Improvement in upstream and downstream links (composite, ordinal) Cluster of Empowerment related indicators · Participation in implementing development projects (factual, ordinal) · Desire to participate in decision making (perception, ordinal) Group meeting frequencies (factual, ordinal) · Interest in group activities (perception, ordinal) · Participation in village joint activities (factual, nominal) · Participation in village decision making (factual, nominal) Knowledge on decision making process at a village level (perception, nominal) · Benefits from joining group activities (perception, nominal) Confidence to join decision making processes (composite, ordinal) Cluster qfAwareness related indicators · Poor people's participation in extension activities (perception, ordinal) · Willingness to contribute for poor people's participation in extension (perception, ordinal) · Awareness of poor people (perception, nominal) · Women's participation in extension activities (perception, ordinal) · Willingness to contribute for women participation in extension (perception, ordinal) · Women's role in agriculture (perception, nominal) Table A.l summarizes the sampling approach used for the DAFEP performance evaluation process and the number of respondent households in each sample and for the two surveys. Sampling Methodology The DAFEP sample The choice of the appropriate sampling size was based on the decision to use income dispersion as the key variable whose variance was used for setting the sample size so that it is representative of the project population. With 30,000 households directly targeted by the project, and a 95% confidence, gave rise to a sample size of 370 randomly selected farmers. Due to possible "losses" throughout the project duration, the sample was increased to 450. However, a full random selection of households among project participants was likely to yield a widely dispersed, and therefore cost wise unmanageable, sample. Thus, a two-step process was used without loosing the degree ofconfidence (Deaton, 1997). This process consisted in randomly drawing first a number of villages, and then similarly, a number of households in these villages. Based on calculations (Salant and Dillman, 1994) a draw of 30 villages and 15 households per village satisfied at the same time confidence level requirements, equal distribution of village (three villages per participating province x 10 provinces) and a 450 households sampling. The field selection of villages was based on the identification of districts and related sub-districts were the DAFEP was planned to be implemented, and to select sub-districts 50 ANNEX 1. Methodology and Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES that were the most representative of the district agro-ecological and socioeconomic conditions. 'Extreme' sub-district situations were first discarded based on meetings with local key respondents and consultation of local secondary data. Then, the sub-districts were randomly drawn. In each selected sub-district the process was repeated for the village selection, discarding 'extreme' villages. At village level, respondent households were randomly drawn using the list of registered DAFEP members. Replacement households were also drawn in case of impossibility to find the corresponding households in the first list. 7he Reference sample For the purpose of the "with and without" comparison, a Reference household sample of similar size and characteristics was selected with the same method in non-DAFEP districts. The process for selecting these households mirrored the process used for selecting DAFEP households. The main difference was that the choice of districts, sub-districts and villages was conditioned by their similarities to the districts, sub districts and villages in the DAFEP sample. Then, at village level, respondent households were randomly selected from the list of village households, excluding households that were known to be completely non farm households (for instance, pure traders or civil servants without any farm activity, including activities of the spouse or dependents). With this process, differences between DAFEP and Reference households were minimized as external heterogeneity due to agro-ecological and socioeconomic conditions was sought to be reduced. 7he Spillover sample In addition, CIRAD proposed the addition of a Spillover sample, in order to estimate whether the project had alm generated indirect effects in implementation areas. The Spillover sample was built as a mix of households not participating in DAFEP activities but located in areas where extension workers had been trained to, and expected to apply, the DAFEP approach. This sample included three different situations of non DAFEP households according to where they were located: a) in DAFEP villages, b) in non DAFEP villages but still in DAFEP sub-districts, and c) in non DAFEP sub-districts but still in DAFEP districts. ANNEX 1. Methodology and Indicators 51 AGRICULTURAL EXTtNSION SERVICES rN INDONESIA · NEW APPROACHES AND EMERGING ISSUES J llentalion Implementation IDistrict Sub- district Village Household Sample Noof name Households TheolJ Benchmark EoP DAFEP DAFEP 30xl5 450 480 360 DAFEP Non lOxl5 ! DAFEP DAFEP DAFEP Non Non Spill-ove! lOxl5 450 528 350 DAFEP DAFEP Non Non Non lOxl5 DAFEP DAFEP DAFEP Non Non Non Non Reference 30xl5 450 480 357 DAFEP DAFEP DAFEP DAFEP Collecting data Data was collected at household level with the same interface forms used for data entry in the Access database where the Benchmark data was stored. This option was selected so as to limit the problem of data coding, a frequent source of errors. Supervisors in charge of each province were trained by Cirad staff and in turn they trained filed enumerators in each province. These five-to-six day training events enabled the participants to familiar with the aim of the project, with the purpose of the survey, with the sampling method, with the interview techniques, and with the various forms. In addition, a special training was given to the staff in charge of operating computers for data entry and to the supervisors for quality control. All supervisors and operators did a pre-testing exercise under the supervision of the trainers to get used to the forms. They did also a pre-test with the enumerators before starting the field work. During the training of supervisors and enumerators the need to interview household members together and not only focusing on one respondent (usually the male active head of household) was emphasized. Therefore each survey unit consisted of a household and not a farmer. Data analysis The objective of data collection was to provide reliable material for DAFEP impact assessment in relation with specific indicators. The data analysis component was aimed at processing this data in order to perform this impact assessment. Since what matters is whether DAFEP had an impact, the key indicator in fact was the significance of a difference observed between the DAFEP and the Reference sample, difference being expressed for each indicator presented above, as the variation between the changes that occurred in the DAFEP sample and the changes that occurred in the Reference sample. This can be represented with formula (1) below as: where: I x is the impact of the project in relation with the indicator X (I x can be true or false) 52 Af\JNEX 1 Methodoogy and Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Sig. is the result of the significance test is the difference in the terms expressed in the parentheses is the mean or median of the indicator X for the Dafep Sample in the EoP survey (and according ~( ) to the indices, for Reference sample; the Benchmark survey is referred to as After) Statistical treatment of data thus differed according to the type of indicator and whether the samples were considered as independent (for comparison of means and averages and dispersion, and for Before and After comparison) or as dependent (for comparison of individual longitudinal data and for regression analysis). The process of analysis included two steps. The first step consisted in conducting statistical work on each indicator in order to assess the impact of the project. This step had two subcomponents: i) analysis of quantitative indicators (see Figure 1), and ii) analysis of ordered (ordinal) and non-ordered (nominal) qualitative indicators (see Figure 2). The output of this step was the identification of the indicators for which it could be concluded that the project had a significant impact according to formula (1). 1be second step consisted of multivariate analyses such as principal component and factorial analysis, applied to these indicators in order to provide, whenever pOSSible, a more complete understanding of how this impact took place and to what extend it was due to the project. Statistl cal t rea/ment ofindividual indicators Quantitative data (income, assets, yields, costs) In the f'reparation of the data sets corresponding to each indicator, outlier25 identification was conducted to eliminate unreliable data. Two methods were used. First, unreliable data was eliminated (for instance yields that were impossible to obtain, or a share that is more than 100%). Alternatively percentile limits set at 95% or 99% were used to identifY and eliminate extreme data that could influence or bias the results. SAS software was used to conduct two different tests on the data sets. AT-test was used (with adjustment to the error degree of freedom using Satterthwaite's formula) within each sample to assess whether the situation After was significantly different from the situation BifOre. This test was selected since the After sample~; are not independent form the BifOre samples due to the selected method of conducting the second survey with the same households. Then, Binary Logistic Regression Model (logit) was applied to test the significance of the means' difference between DAFEP and Reference. For each model, the dependent variable was the membership of the respondent household to the different samples and the independent variable was the indicator. Each indicator was analysed separately. 1bis method uses the significance of the regression coefficient, estimated with the maximum likelihood method, as a significance test of the differences observed between samples in the inifal situation, in the final situation and ofthe difference as indicated in formula (1). Significant changes in the latter indicate that the observed difference was related to the existence of the project; in other words 25 According to Iglewicz and Hoaglin (1993) outliers can be defined as observations that represent a discrepancy with the rest ofthe collected data according to the investigator. Several methods can be used for identifYing and eliminating outliers in normally distributed data sets (Fallon and Spada).lhe Box plot method (Tukey 1977) is a graphic method that does not include the extreme outliners in calculating the dispersion because it is based on median and inter-quartile range calculation from which error limits are set at 5% and 95% confidence intervals. vVe had for instance the case of one household whose total non farm income represented 26% of the DAFEP sample total non farm income. It turned out that the respondent household had borrowed and invested a huge amount of money for developing a cocoa trading business that year. . .NNEX I. \;lethocio:ogy and :,~d(ijtor5 A 53 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: .....) NEW APPROACHES AND EMERGING ISSUES Analysis of Quantitative Data (Mean and Median) the project had an impact27 · Before After Difference In addition, tests were also - carried out on the distribution DAFEP r of the quantitative variables Reference using classic indicators such as skewness, kurtosis, and Gini Spillover coefficients. As there is no ~ , · formal test to check whether the difference between Before and U U U 5 · i After conditions are significant, Tests: . . ......· .. ·.........!: the Bootstrap method was used. 28 . : : The steps of the simulation were: 1. Log it Test maximum likelihood: l:etween samples (with and without) I: 1. Re-sampling the data for : each sample (Monte Carlo 1_2_. _-v_e_s_t_Wl_·thin_·_s_am_l_e_s_{l:e_fo_Ie_an_d_after} _ _ _ _ _ _ _ _ _ _---'~...i .... T p ___ method) 2. Compute the skewness and kurtosis of the variable at before and after condition for each sample 3. Compare the skewness and the kurtosis of before and after condition 4. Repeat 10.000 times steps 1- 3 5. Count the situation where 'after'is less than 'before', and calculate the percentage. If the percentage is less than 0.05 or greater than 0.95 we conclude that they are significantly different. Figure 1. Statistical approachfor quantitative data 27 Note that although the figures in the tables are sometimes presented with a number of significant figures, this is not a reflection of the precision of the estimates, merely the nature of the output of the software. 2' See for example Efron and Tibshirani 1993. 54 ANNEX J. Methodology ana Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Coping with missing production data At the time of the EoP field survey, more than 400 parcels, a non negligible share of the total number of parcels, were cultivated but not yet harvested. The majority of these parcels are mainly cropped with rice and to a lesser extent with maize and soybeans. As a result, cost data were collected but not production data. This affected the value of several indicators, in particular agricultural net income per capita, yields, total production value per ha, input use efficiency, trade income per capita. This situation influenced the results in particular when comparing the Before and After results, since at Benchmark data collection period those cases were almost non-existent. The following adjustments were made in order to minimize the effect of the missing production data when comparing the results. According to the type ofindicator, the adjustment process took different forms. Act/ustmentsfor yield calculation. Three crops were concerned: rice, maize and soybeans. For coconut, there was no missing data. The method for adjustment consisted in calculating a proxy for the expected yields with the following procedure: 1. Split DAFEP and Reference samples into two sub-samples, one without the missing parcels (SSI) and one with only the missing parcels (SS2). 1. Calculate the new average yield in SSI for each of the three crops 2. Calculate the input cost per ha (ICl) in SSI and input cost per ha (IC2) in SS2. 3. If leI differs from IC2 by more than 10%, calculate the average yield for a subset of parcels in SSI centred on the value ofIC2. This calculation was made using a 10010 range from the centre value ofIC2. 29 4. Use this yield to calculate the harvest for each parcel multiplying it by the cultivated area (this makes a new SS2 sub-sample) 5. Use the new data to complete the sample by merging SSI and the new SS2 and recalculate the new proxy yield. 6. Do the impact analysis with the recalculated yields. Adjustment for agricultural income calculation. The agricultural income per capita cannot be recalculated because the period of analysis is a one year period. Including an expected but virtual income would have altered the consistency of the whole data set. Thus, this indicator is not used to assess the project impact on this specific point. An alternative indicator, the crop net income per hectare, was thus established as a proxy of project impact on income. The crop net income per hectare is calculated from crops for SSI, followed by running comparisons of Before / After, and DAFEP/Reference. The same procedure was applied to recalculate the trade net income per hectare. Another consequence was that the total net income per capita was also to some extent affected by this situation. However, given that both Reference and DAFEP samples were similarly affected, the difference in total net income indicator per capita was not expected to be significantly affected. Furthermore, another indicator, the net yearly expenses per capita is also a proxy of the total income and therefore analysis on income and welfare could be performed. Adjustment for input use efficiency indicator. The input/output ratio (lOR) was calculated for rice maize and soybeans after proceeding as for yield calculation adjustment. In addition the value of production using the average price in the sample was calculated without missing data. For rice this approach is particularly acceptable, since farm gate price of rice is determined by GoI floor price. 3o 2. This method was based on the hypothesis of a positive correlation between input use and output volume. The coefficient of correlation between input costiha in rupiah and the yield in Kg/ha for the EoP survey, without the parcels not yet harvested is 0,89 for rice, 0,76 for maize and 0,40 for soybeans. This indicates that the hypothesis was verified for rice and maize. lhe H PP Inpres No. 131200.5 sets the floor price of paddy (Gabah Kering Panen) at Rp 1,730/kg.lhe median price of paddy in the three EoP samples is Rp1.500/kg. Similarly, the price for maize was Rp1,200/kg and for soybean Rp2,700/kg. ANNEX 1 r'/etilodoiogy and Irldli~atcrc 55 - - - - - _ . _... ---~--------- ... ~--- ...- - -....- - -... -~-- ...- - -.... - - - - - - - AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES J Qualitative data Data is sorted into two categories, ordinal (numbered) and nominal data, For each ordinal indicator, the difference in the value Before and After was processed for each respondent. Ihis difference is classified into three categories: improvement (positive), stable (neutral) and worsening (negative), The testing of ordinal indicators consisted of measuring the difference between Before and After for each sample using a related sample sign test. Then, the difference in the change rate between samples was measured based on the different values of the indicators over the time period and using a chi-square homogenised proportion test, For nominal indicators, a representation of the variation in the distribution of respondents' answers was used. The estimates between Before and After conditions were based on two-samples proportion test. For some clusters a new synthetiC indicator (composite indicator) was also used with a weighted combination of ordinal indicators. Then, change and significance tests were conducted as indicated above for ordinal variables. Figure 2. Statistical approach for qualitative data Analysis of Qualitative Data (Ordinal and Nominal) Before After Difference ,-- DAFEP(+ Y&W) - Reference (+ Y &W) Spillover (+ Y&W) .. " · U U U i · : Tesls: · ··ed : ·..................._! : ! U::hi square test between samples for ordinal variables (with and without) :.! 2. Sample sign test within samples or two samples proportion test (before and after) j 56 A.NNEX 1. Methodology and Indicators ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES · At the start of the project, DAFEP was planned to be implemented in 10 provinces. On this basis, the sample was design to cover and represent the target group of 30000 households. However, activities were cancelled in one province because the extension services were "re-structured" and disappeared as such. This reduced also the size of the sample. There was also a "natural" erosion due to the impossibility in the EoP survey to find all the households interviewed in the Benchmark survey. This case had been foreseen in the project survey design and an erosion rate of 20% had been anticipated. The real erosion rate is more or less 10%, mainly due to households that have moved and/or that the enumerators were unable to locate. Thanks to the overestimation of the erosion rate, the size of each sample in the EoP survey was large enough to carry out the statistical analysis with the expected confidence level. Data quality and reliability: How reliable is the data ? There are clear limitations of the data. The first limit is inherent to the type of work and to the collection of socioeconomic and agronomic data from farm households. Collected data is not similar to data used in laboratory experiences or in field trials. 1V108t information comes from the recounting by respondents of facts that have occurred in a more or less recent past, without the possibility to check its accuracy. In addition, many indicators proposed for the assessment of the project were qualitative and some of them based on perceptions of fact, not even factual ones with no way to measure their accuracy. Thus, accuracy and data inconsistency were the main problems in this survey. Several methods were used to counter these. 4.36 A~ income was a central issue in the evaluation (the sampling rate was based on household income distribution), particular attention was paid to income data accuracy and inconsistency. For this purpose a household budget approach was used with on one hand the income flow and on the other hand the expenses flow. Data was collected independently for the two flows and a ± 10% match considered as acceptable over a one year period. The requested presence of all household members at the time of the interview was also a means to increase the reliability of data and in many cases proved to be necessary. Enumerators were requested to check consistency in income flows at the end of the interview and built-in queries were added in the database so as to automatically detect these. 4.37 1hus, it is considered that data is reliable with a ± 10% range and this although was true for the calculation ofsignificant differences. In fact, since what mattered in the evaluation was the significance of the differences and not the absolute values taken by each indicator, reliability of data was more a function of using similar methods and tools in the DAFEP and Reference sample than which method and which tools were used. As enumerators were assigned to perform both DAFEP and Reference sample survey, possible biases due to different teams assigned to different samples were eliminated. The second limit was linked to two classic problems: outliers and missing data. For outliers, a median based aDproach was used rather than a mean based approach. It is acknowledged that alternative methods and approaches may be used the aim is to provide transparency on the methods used so that it can be replicated and checked and alternative methods can be tested. The third source of uncertainty is the respect of data collection protocol by the teams of enumerators and supervisors. Feedback from the EoP survey team related to how data collection was performed during the Benchmark indicates that there were some possible breaches in the data collection protocol. Cases were mentioned where apparently respondents were not interviewed in their house and in per households, but regrouped in one place. The selection of respondents in some place was not fully random and respondents belonged to one farmer group (reference sample). M'lNLX 1 fviethodoIG,s' anel 57 "-. AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Annex 2: Multivariate Analysis This annex presents multivariate analysis applied to the indicators that have been identified as significant in the main report. The objective is to provide further arguments to discuss the significance of the changes earlier observed. It intends also to find out whether specific features of the DAFEP sample are related with these changes in a causal way, in which case the impact of the project would be considered as even more likely. This work is firstly conducted with factorial analysis for the whole set of significant performance indicators, distinguishing qualitative and quantitative indicators. Then speCific correlation analysis is conducted between significant indicators. It is not the purpose of this report to perform systematic multivariate analysis of changes that have occurred between the Before and After situations if the latter are not significant in the comparison across the DAFEP and Reference samplesY Section A2.1 presents the results of the analysis conducted on the whole set of significant indicators and section A2.2 has the results of the analysis for the specific indicators for which significance tests were positive. A2.1. hlCtorial Analysis Quantitative indicators Methodology In this part of multivariate analysis, principal component analysis (PCA) and Biplot are applied to samples from which multi-dimension outliers have been eliminated. Outliers were detected and removed using the Mahalanobis distance method. For independent variables this is a measure of how much a case values differ from the average of all cases. A large distance points out a case having extreme values for one or more of the independent variables. The numbers of outliers detected with this method are displayed below: Table A2.1. Outliers identified with Mahalanobis distance method for multivariate analysis Before After Type of Outlier DAFEP Spillover Reference DAFEP Spillover Reference Extreme Outlier 18 21 20 28 22 32 ''Usual'' Outlier 14 14 15 17 10 23 After detecting the outliers, they were classified into three categories: not an outlier, usual outlier, and extreme outlier. These categories were used as dummy variables in a regression analysis to see how much they affected the variables used in the multivariate analyses. A low R-square value indicated that the outlier's effect was not significant. The results pointed out that the extreme outliers have a more significant effect (R-sq = 34% for Before and R-sq 40% for Mter) compared with the usual outliers (R-sq = 27% for Before and R-sq = 35% for After). Then, dummy variables were regressed excluding the extreme outliers. The results are R-sq 6% for Before and R-sq = 15% for After, showing that after excluding the extreme outliers, usual outliers presence was non-significant. Among the extreme outliers, seven were common to the Benchmark and EoP surveys. 31However, the authors in collaboration with the \Norld Bank plan to conduct further analysis on this data set and to present the results in other media. 59 - - _ . _..... __ _ - - - - - - - - - - - - - - - - - - - - - - - -ANNC:X - Mutl'Janate- - ... -- --- 2. Analysis AGRICULTURAL EXTENSION SERVICES IN INDONESIA. ,,) NEW APPROACHES AND EMERGING ISSUES After elimination of extreme outliers, the data set was ready for a principal component analysis. This analysis is used to understand the covariance structure in the original variables and/or to create a smaller number of variables using this structure. Results of the PCA are displayed with Biplot, a way to present information on the characteristics of many objects in a 2D graph. With Biplot, it is possible to display the similarity of characteristics among objects (here each sample in each survey, so that there are six objects), the variance of variables (see the list of variable below), relations/ correlations among variables/ characteristics and the main characteristics of each object. Two scaling methods were combined the correlation method and the distance method, in on columns centered and standardized correlation biplot representation. Figure 1 below synthesises this process. Figure 1. Process of multi-dimension outlier identification and removal Boxplot Extreme Outlier Outlier Not an Outlier Regression with dummy variables If R2 is small IfR2 is big Keep Outliers Delete Outliers Results The results of PCA applied to a selected list of quantitative variables and indicators in each sample are presented in the Biplot figures 2 and 2 bis below. Variables are Type of sample (l=DAFEP; 2=Spillover; 3=Reference); Age of the household head; Education level of the household head; Family size; Land area in ownership; Cultivated area; :Number of crop grown; YEC; AVC; SFE; NFIC; AB:NIC; AR:NIC; TIHA; TVIHA; CNIHA; ARNIC ; RiceYld; MaizeYld; SoybeanYld; IORrice; IORmz; IORsbn. The objects are the different types of sample while the vectors represent the variables. 60 ANNEX Multivariate AnalYSIS ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Figure 2. Correlation biplot representation (columns centered and standardized) for selected quantitative variables 0.8 0.6 TVIHA -0.8 -0.6 0.8 Land Owned Cultivated Area ·0.4 Education -0.6 Figure 2bis. Extracted from Figure2. 0.4 · RS SFE · DB 0.8 A.;mEX 2. MLiriva' ate A.naiysls 61 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES J Ibe characteristics of a correlation Biplot are as follows: ¢ Correlation between variables is represented by the cosinus angle of two factors. If the correlation is 0 positive, the two vectors form an angle that is <90°, i f negative it is >90 and if the correlation is 0 the 0 angle is _90 ¢ The variance is equivalent to the length of the vector; the longer the length, the higher the variance. ¢ Object Characteristics: two objects with similar characteristics will be shown as tvvo dots close to each other; if an object and variable vector are in the same space (close to each other) the closest vectors indicates the characteristic of the object. ¢ The distance between objects (samples) is a proxy of their Mahalanobis distance in the multidimen sional space. Most correlations are pOSItive and some are very strong (the more acute the angle the stronger the correlation). Most income and welfare variables are strongly correlated except ANIC with ABNIC and NFIC. 1he proximity ofYEC and TNIC confirms that YEC is a good proxy for the total net household income per capita. As the length of the line that links a variable point to the graph's origin represents its contribution in this space, we can see that most of the characteristics of the sample are explained by welfare and income indicators. However, the projection of the objects coordinates to these vectors indicates a rather large homogeneity among the samples in the Benchmark survey and among samples in the EoP survey. This result confirms the trend in the statistical analysis. Samples were very close in their characteristics before the project and remain so after. It confirms also the importance of the change (positive) in relation with NFIC for both DAFEP and Reference samples between Before and Mter and the relative decline of ANIC and other variables that are related to agriculture (number of crops harvested, TVIHA, CNIHA, TIHA, Rice, Soybean and Maize Yield, and cultivated area) have relatively decreased. Qualitative indicators Methodology Factor analysis is used to identifY underlying variables, or factors, that explain the pattern of correlations within the set of observed variables. As indicated in table A2.2 eight components explain more than 75% of the total variance. A rotated component matrix highlights which indicators mostly contribute to each of these components as displayed in table A2.3 where, for example, factor 1 represents ind-19 (Accessibility to extension workers), factor 2 represents ind-14 (Interest in group activities) 62 ANNEX 2. Multlv2r1ate Ana:ysis ""- AGRICU~TURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Table A2.2. Total Variance Explained Component Initial Eigenvalues Total % of Variance Cumulative % 1 2.449 18.840 18.840 2 1.407 10.822 29.662 3 1.184 9.107 38.769 4 1.098 8.446 47.215 5 1.046 8.046 55.261 6 .966 7.430 62.690 7 .928 7.139 69.830 8 .866 6.665 76.494 9 .842 6.474 82.969 10 .780 6.003 88.971 11 .738 5.677 94.648 12 .380 2.919 97.567 13 .316 2.433 100.000 Extraction Afethod' Principal Component Analysis. Table J\2.3. Rotated Component Matrix (a) Component 1 2 3 4 5 6 7 8 ind-19 .942 -.067 -.030 -.097 -.018 .009 -.077 -.042 ind-14 -.071 .926 .087 .012 .082 .033 .039 .056 ind-8 -.029 .076 .985 .000 .070 .073 .031 .028 Ind-23 -.086 .010 .000 .991 -.027 -.044 .028 .028 ind-16 -.018 .071 .070 -.027 .984 .035 .040 .070 ind-11 .006 .029 .071 -.044 .035 .991 -.050 .036 Ind-27 -.066 .031 .030 .028 .039 -.050 .993 .041 ind-18 -.038 .047 .028 .028 .068 .036 .042 .990 Ind-24 -.031 .02~ .024 .017 .065 .009 .014 .068 Ind-26 -.008 .004 .008 -.003 -.025 .010 -.017 .002 1 ind-41 -.010 .009 .007 -.024 .030 .022 -.012 .000 ind-13 -.065 .351 : .095 .001 .070 .050 -.001 .042 ind-20 -.329 .090 i .084 .096 i .069 .054 .011 .034 The variables selected for further correspondence analysis are thus: ind-19 : Accessibility to extension workers ind-14 : Interest in group activities ind-8 : Participation in Village Decision Making ind-23 : Barriers of access to input markets ind-16 : Accessibility of agriculture information ind-ll : People's desire to participate in decision making process at a village level Ind-27 : Joint marketing of outputs/agriculture commodities ind-18 : Willingness to pay for agriculture information ANNEX 2. Mdtlvanale AnalySIS 63 AGRICULTURAL EXTENSION SERVCES IN INDONESIA. ....) NEW APPROACHES AND EMERGING ISSUES 1he next step was to conduct a multiple correspondence analysis and represent the results on a graph. In Figure 4, the square dots represent the values taken by the selected indicators and triangular dots represent the various samples (12 samples in this case after separating household respondents from Y&W respondents). lnterpretation When samples and indicators values are close it means that these values are associated with the sample. Based on this, in the multiple correspondence analysis graph displayed in figure 3, we can differentiate two main clusters: the Benchmark cluster and the EoP cluster. This confirms that there is generally a Significant difference for many of the key qualitative indicators between the Before and After situation. The Benchmark (Before) cluster is characterized by the following respondents' perception: having easily access to extension workers, often having barriers to input market, participating in group activities for between 1 and 5 years, and having no desire to join in village decision making. The EoP (After)cluster is characterized by the following respondents' perception: having no barriers to input market, having sufficient access to agriculture information, being a member of group activity for more than five (a logical result since this survey was conducted five years later) and having the desire to join in village decision making. As indicated by the arrows, the direction of the evolution (the After cluster has moved to a position closer to improved values for many variables) is rather positive with the exception of access to extension workers but it is compensated by improved access to agriculture information. However, these changes occur again between the Before and After surveys, not across sample (Reference versus DAFEP). 64 ANNEX 2. Multivariate Analysis "-. AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Figure 3. Representation of multiple correspondence analysis results I I I I I I 22 I I ···· ···· · I ...... 1 I ..... I ····· I ·· f! &F I "t~' · ... R-F ~'i". ~\)o·t··· I · .( +-' ...2i..A .'--."· I: · I· 11 16 C Q) \. II> I I · c /9\ : 8. 4 E o ---------~-~~~- · 1 - · · --~--------------~R_F---------- I. o 5-.1,.5 .. () · '.. · t 7 -: ~\ I 1 · :;!~5 19\1 · 1 .. i ... A-&F 6 ~-D-Y&W · \: I 10 -1 · -1 o 1 ComjDnent1 Note: B-D F j, BefOl'e DA FEP farmer (household) B-D-Y&\V is Before DAFEP youth and women sample B-R-F is Before Reference farmer (household) B- R -Y&\V is Before Reference youth and women sample B-S-F is Befort' Spillover farmer (household) A-S-Y&VV is Before Spillover youth and women sample A-D-F ii After DAFEP farmer (household) A-D-Y&\Vis After DAFEP youth and women sample A-R-F if Mter Reference farmer (household) A-R-Y&VV is After Reference youth and women sample A-S- F is After Spillover farmer (household) A-S-Y&Vvis After Spillover youth and women sample 1 ;:-.!O barriers to input market 2 Sometimes barriers to input market 3 Often have barriers to input market 4 No jont marketing 5 Joint marketing (0-2 years) 6 Joint marketing (3-5 years) 7 Joint marketing (>5 years) 8 No access to agriculture information 9 Lack of access to agriculture information 10 Sufficient access to agriculture information 11 Not willing to pay for agriculture information 12 Willing to pay for agriculture information 13 Easy access to extension worker 14 Accessibility to extension worker is average 15 Hard to access to extension worker 16 No respond to accessibility to extension worker 17 No participation in village decision making 18 Not to Active in village decision making 19 Active in village decision making 20 No desire to join in village decision making 21 Desire to join in village decision making 22 Not a member of group activity 23 Member of group activity (0-2 years) 24 Member of group activity years) 25 Member of group activity (>5 years) M,i\IEX iVlu!tlv3nate Analys.s 65 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES ...J !Catol'S The results of the significance tests applied to the performance indicators are summarized below cluster by cluster, and corresponding hypothesis and multivariate analysis method are indicated. The significance analysis shows that Reference households have performed better than DAFEP households with respects to agriculture related net income per capita (ARNIC) and within this category agribusiness net income per capita (ABNIC). Conversely, positive changes in non farm net income per capita (NFIC), yearly expenses per capita (YEC) and asset value per capita (AVe) are significantly higher for the DAFEP sample compared to the Reference sample. We firstly examined whether a change in the number of household members could explain the differences observed, since all these indicators are calculated per capita. The results are presented in table A2A. Table A2A. Value and significance of correlations between family size and selected per capita indicators Sample YEC AVC NFIC ARNIC Dafep Before -0.3* -0.2* -0.2* -0.1 Dafep After -0.2* -0.3* -0.2* 0.0 Reference Before -0.3* -0.3* -0.1 * -0.1 * Reference After -0.3* -0.3* -0.2* -0.1 · Spillover Before -0.3* -0.3* -0.2* -0.1 * I . Spillover After -0.3* -0.2* -0.2* -0.1 * 'Significant (2-tailed) Interpretation There is a correlation between the size of the family and TEC, AVC and NFIC for all samples in both Benchmark and EoP surveys. For ARNIC the correlations are not always significant. However, since the values of the correlation coefficient are all very low «0,3), it is likely that other factors explain the observations. Decomposition ofsign!ficant income and welfare indicators Agriculture-related net income per capita (ARNIe) In table A2.5 is displayed the share of households earning an income from agriculture related activities and then the share of its three components, respectively agro processing (ABNIC), renting of production factors (RPFC) and selling labour force (LF). Table A2.5. Decomposition of ARNIC and correlations with its constituents ABNIC RPFC LF Decomposition Before After Before After Before After I DAFEP 27% 45% 24% 21% 49% 34% . Spillover 19% 38% 26% 23% 55% 39% Reference 18% 41% 29% 25% 53% 34% ! ABNIC RPFC LF Correlation Before After Before After Before After DAFEP 0.6* 0.9* 0.3* 0.3* 0.6* 0.4* Spillover 0.8* 0.9* 0.4* 0.3* 0.4* 0.1 Reference 0.7* 0.9* 0.5 0.3* 0.3* 0.1 'Significant (2-tailed) 66 ANNEX 2. Multivariate AnalYSiS '".1-'$ "Ii" .J hud !t1 . · n \... AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Interpt'etati on The results indicate that in all samples the percentage of households with income from agriculture related activities has increased. This increase is characterised by a drop in the relative weight of selling labour force as source of income and the surge in agro processing. This is particularly true for the Reference sample where the earlier analysis had highlighted the weight of this source ofincome in agriculture related income. The correlations between ARNIC and its components confirms that ABNIC is the main factor explaining the observed trend and that the two other factors get less correlated with ARNIC in the EoP sample. Analysis ofNon Farm Net Income per Capita In table A2.6 is displayed the share of households with a non farm income and then its four different sources, respectively non farm jobs (NF]C), sale of assets (SOAC), financial transactions (FTC) and remittances (Rl\IIC). Table A2.6. Decomposition ofNFIC and correlations with its constituents NFIC NFJC SOAC FTC RMC Sample Before After Before After Before After Before After Before After DAFEP 77% 94% 61% 49% 11% 12% 23% 21% 4% 18% Spillover 74% 95% 61% 52% 9% 9% 23% 20% 7% 19% Reference 75% 94% 59% 48% 9% 11% 22% 22% 8% 18% NFJC SOAC FTC RMC Sample Before After Before After Before After Before After DAFEP 0.6* 0.1* 0.5* 0.1 0.6* 0.98* 0.1 0.04 Spillover 0.5* 0.8* 0.6* 0.2* 0.6* 0.5* 0.1 0.2* Reterence 0.8* 0.7* 0.1* 0.4* 0.8* 0.6* 0.3* 0.2* 'Significant I)-tailed) Interpret;Jtion In all samples the percentage of households with non farm income has drastically increased reaching 95% (NFIC column represent the share of households with NFIC income). The structure of non farm income remains stable as far as cash flow aspects (little change in SOAC) and financial transactions within the households (little changes in FTC) are concerned. However, the relative share of remittances and other subsi&es have gained importance. These results indicate that these households become more dependent from external sources of income while the share of purely non-farm household activities has decreased. This trend is similar for all sample but more accentuated in the case of the DAFEP sample since the Before situation of RMC was lower. The correlations are more difficult to explain. Non farm jobs income is strongly correlated with all samples before but very loosely with DAFEP after, while there is a stronger correlation with financial transactions in the EoP samples. Remittances are not strongly correlated with NFIC. Ana{y.5is o.fAsset Value per Capita In table A2.7 the structure and evolution of assets using five main categories whose share is represented through the estimates of their monetary value with regards to total asset value is presented. These are respectively animal assets (AA), AgriEquipment (AE), House&Furniture (HF), Transportation Means (TM), and Capital Savings (CS). ANNt:X Multivariate Analysis 67 AGRICULTURAL EXTENSION SERVICES IN INDONESIA. ",,) NEW APPROACHES AND EMERGING ISSUES Table A2.7. Decomposition ofAVC and correlations with its constituents I Decomposition AA AE HF TM CS I Before After Before After Before After Before After Before After DAFEP 18% 9% 5% 4% 70% 76% 7% 10% 0,0% 0,1% J Spillover 13% 10% 6% 5% 74% 76% 7% 9% 0,0% 0,3% Rererence 16% 10% 5% 5% 72% 75% 8% 9% 0,2% 0,2% I Correlation AA AE HF TM CS Before After Before After Before After Before After Before After· DAFEP 0.37'" 0.23* 0.42* 0.61* 0.9}* 0.92* 0.43* 0.66* -0.04 0.13 Spillover 0.23* 0.12* 0.09 0.07 0.99" 0.99" 0.60* 0.17'" -0.03 -0 Rererence 0.92" 0.24" 0.09 0.17* 0.84" 0.96* 0.88" 0.41* -0.02 0.04 *Significant (2-tailed) In all samples house and furniture represent more than 70% of the assets both Before and After conditions. The weight of animal assets has rather strongly dropped in the DAFEP and Reference. This may be interpreted as a change in the consumption patterns of the household and in their accumulation strategy. With higher income the needs for savings through animals has decreased and households acquire more consumption goods. The correlations confirm this trend with a decline in the association between assets and animals in the After samples while the correlation get stronger with house and furniture. Analysis ~fYearly Expenses per Capita (VEC) In table A2.8 we present the structure and evolution of yearly expenses per capita using six main categories whose share is represented through the estimates of their monetary value with regards with total expenses value. These are respectively, food consumption (FC), non food routine consumption (NFR), education costs (EC), social expenses and transfers, asset acquisition (AA) and financial transactions (FT). Table A2.8. Decomposition ofYEC and correlations with its constituents FC NFR EC SC AA FT I Decomposition Before After Before After Before After Before After Before After Before After DAFEP 36% 31% 31% 34% 5% 6% 7% 8% 11% 12% 10% 9% I ! Spillover 36% 32% 33% 35% 5% 6% 7% 8% 9% 11% 10% 9% I Rererence 36% 32% 31% 33% 5% 6% 7% 8% 11% 11% 10% 10% I ! FC NFR EC SC AA FT Correlation : Before After Before After Before After Before After Before After Before After ! DAFEP 0.49* 0.18* 0.59* 0.22" 0.29* 0.03" 0.53* 0.11" 0.80* 0.37" 0.55" 0.94" Spillover 0.46* 0.48" 0.64" 0.58* 0.27* 0.41" 0.37* 0.42" 0.78" 0.73" 0.59" 0.67" Reference 0.53" 0.63* 0.51* 0.66" 0.18" 0.26" 0.30" 0.41" 0.77'" 0.72" 0.68" 0.57'" Interpretation In all samples the trend is that monetary non food routine expenses are progressively taking over monetary food expenses. This indicates some improvement in general welfare level; however, as education expenses 68 ANNEX MullivJllale An3iYs,s ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES and assets acquisition remain stable, this improvement does not translate into an accumulation process or a long term strategy. Transfers from the households are limited and confirm that the net flow of transfer is positive towards the rural households. Deepening the analysis ofwelfare distribution As indicated earlier, while AVC in the DAFEP sample is significantly higher than in the reference sample, AVC distribution in the EoP survey is significantly different than in the reference sample, showing a higher level of inequality. In order to better understand this situation, we split the DAFEP sample into two sub,amples of the same size using the median value ofA VC in the EoP sample. A correlation analysis was run for the variables displayed in Table A2.9. Figure 4 plots the variations in asset value per capita between the Benchmark sample and the EoP sample against the variation in non farm net income per capita £)r the two subsamples (called "Upper Stata" and "Lower Strata"). Table A2.9. Correlation coefficients between AVC and NFIC for DAFEP upper and lower strata UPPER AVC-l AVC-2 Dif AVC NFIC-l NFIC-2 DifNFIC AVC-l 1,00 AVC-2 0,14 1,00 DifAVC -0,21 0,94 1,00 NFIC-l 0,29 0,07 -0,03 1,00 NFIC-2 0,20 0,23 0,16 0,28 1,00 I- DifNFIC -0,05 0,16 0,17 -0,52 0,68 1,00 LOWER AVC-l AVC-2 Dif AVC NFIC-l NFIC-2 DifNFIC AVC-l 1 I AVC-2 0,19 1,00 DifAVC -0,95 0,14 1,00 NFIC-l 0,18 -0,06 -0,21 1; NFIC-2 0,16 0,17 -0,10 0,16 1,00 DifNFIC 0,03 0,19 0,03 -0,45 0,81 1,00 Note: 1 and 2 refer respectively to Benchmark and EoP surveys. The results show that there is no significant correlation between the variation in asset value per capita and the variation in non farm income (0,17 and 0,03 respectively for Upper and Lower Strata). The difference in AVe is strongly positively correlated (0,95) with the level of asset in the 2006 for the Upper strata while it is strongly negatively correlated (-0,94) with the level of assets in 2001 in asset value per capita for the lower strata. Figure 11 shows that in the higher strata, an increase in non farm in come is more often linked with an increase in asset value, while in the lower strata the two elements are almost completely de-linked. Furthermore, the trend in the lower strata is a de-capitalization of assets in value (deflated) while there is a capiting and Study tour Citrus Technology transfer; and training 31% 3.1% 6.8 33.1 cultivation organized in collaboration with pest management school Apprenticeship program for technology 9.2% 4.1% 15.0 1.4 I Pruldy Threshers transfer to make indigenous wooden thrasher i Partnership External expertise provided training on 100% 3.1% NA NA building partnership activities I c~~ning growth and weight gain-enabling two Training for technology transfer-faster 27% 4.7% 2.2 NA cycles in 2 years I ! Impro, cd paddy Rainfed paddy field school 8% 3.1% I 2.2 NA technol~ I Bamboo Demonstration and supervision by 15% NA 15.7 handcraft for invited artisan. NA youth I Impro,ed cattle Technology transfer and training 100% NA 2.8 1.9 by breedingl ration fonnulation I I I Local Chicken Technology transfer for better rearing More 4% 9.0 9.0 increased stocking and production than I 50% I Proces;;ing of Technology transfer for spices Less than 14.5% 2.1 NA Spices for value I processing in 2002 and field meeting 1% addition for spice crops cultivation in 2003 Rambutan Technology transfer and training 75% 13% NA NA selection, budding and production I 11Annualized investment cost is estimated based on 10% interest rate and 3 to 5 years of repayment period wherever appropriate; NA for Not available/applicable ANNEX 3. Farmer Managed Activity Interventions 75 ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA NEW APPROACHES AND EMERGING ISSUES Annex 4. Statistical treatment of univariate outliers Statistical treatment done for the outliers in this report for the univariate indicators are shown in the figure below. The first step is to build a database of respondents who are in the before and after survey and eliminate those who don't match. Mter getting a paired data set we start detecting outliers for each individual indicator. For indicators that can be judged as rational or irrational value, such as yield of paddy it is unlikely to get a harvest of 10 tons from 1 Ha or for lOR it is unlikely to have a value above 1, we eliminate the irrational values. The second method is used when we have indicators which might have a long range of values, such as income, consumption, assets, etc it varies from 0 to 1.000.000.000 for assets, and then we use a 5% or 10% distribution method. This method depends on the distribution of each indicator, if the indicator has a wide distribution then we use a 10% limit and when we have a narrow distribution we use a 5% limit. Graph of univariate outliers treatment algorithm Eliminate Outlier Diagnostics Method I: Method II : Subjective Method Outlier treatment based on distribution I · Set up a reasonable cut · Data has a wide off point distribution, eliminate · Eliminate the data that 10% of the is above/under the cut highest/lowest values. off point · Data has a narrow Example: For lOR and distribution, eliminate yieldIHa 5% of the highest/lowest values. Example: assets, income, consumption, expenses, etc ANNEX 4, Statistical treatment of u'ilvanale outliers 77 \.. AGRICULTURAL EXTENSION SERVICES IN INDONESIA. NEW APPROACHES AND EMERGING ISSUES Annex 5. Analysis of Soybean Yield in DAFEP Sample 1. Correlation o Correlations between Yield and Cultivated area Correlations Cultivated Cu ltiwted Area Pearson Correlation 1 -.074 Sig. (2-tailed) .585 N 57 57 Yie IdIH a Pearson Correlation -.074 1 Sig. (2-tailed) .585 N 57 57 The correlation reported in the table is negative, although it is not significantly different from 0 because the p-value of 0.585 is greater than 0.05. o Correlations between Yield and Total input Correlations YieldlHa Total input/ha YieldlHa Pearson Correlation 1 .221 Sig. (2-tailed) .098 N 57 57 Total inputfha Pearson Correlation .221 1 Sig. (2-tailed) .098 N 57 57 The correlation is positive and the p-value of 0.098. It indicates that the correlation is not significantly different from 0 at level 5% but it is significantly significant at level 10%. 2. Association o Relationship between Yield and Location OIi-Square Tests Value df ~ymp.Sig. (2-sided) Pearson Chi-Square 12.937a 15 .607 Likelihood Ratio 12.655 15 .629 N of V'dlid Cases 57 a. 20 cells (83.3%) have expected count less than 5. The minimum expected count is .21. The significant value of the Chi-square statistic is greater than 0.05. It means that there is no relationship between yield and location. The chi-square test is useful for determining whether there is a relationship. ANr-;EX 5. A.nalyss of Soybean Yield ie, DAFEP Sarlple 79 -------- AGRIC:JLTURAL EXTENSION SERVICES IN INDONESIA: ,...,) NEW APPROACHES AND EMERGING ISSUES o Relationship between Yield and Type ofvariety OIi-Square Tests Value df ~ymp.Sig. (2-sided) Pearson Chi-Square 31.412 a 21 .067 Likelihood Ratio 31.983 21 .059 N of Valid Cases 57 a. 28 cells (87.5%) have expected count less than 5. The mi nimu m expected count is .21. The significant value of the Chi-square statistic is 0.067. It means that there is no relationship between yield and location at level 5%. At level 10%, there is relationship between Yield and Type of variety. The Chi-square test doesn't tell you the strength of the relationship. Symmetric measures attempt to quantifY this strength. These measures are based on the chi-square statistic. Symmetric Measures Value ""prox. Sig. Nominal by Phi .742 .067 Nominal Cramer's V .429 .067 ContingencyCoefficient .596 .067 N of Valid Cases 57 a. Not as sum ing the null hypothesis. b. Using the asymptotic standard error assuming the null hypothesis. The significance values of a1l three measures are 0.067, indicating a statistically significant relationship at level 10%. The values ofall three measures are around 0.5; it subjectively indicates the association yield and type of variety is not strong enough. o Relationship between Yield and Type ofland OIl-Square Tests Value df ~ymp.Sig. (2-sided) Pearson Chi-Square 25.051 a 21 .245 Likelihood Ratio 25.247 21 .237 N of Valid Cases 57 a. 28 cells (87.5%) have expected count less than 5. The minimum expected count is .21. The significant value of the Chi-square statistic is greater than 0.05. It means that there is no relationship between yield and type of land. 80 ANNEX 5. Aralysl$ of Soybean Yield DAFEP Sample ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES o Relationship between Yield and Tenancy OIi-$quare Tests Value df /J.s ymp. Sig. (2-sided) Pearson Chi-Square 4.893 a 6 .558 Likelihood Ratio 5.935 6 .430 N of Valid Cases 57 a. 7 cells (58.3%) have expected count less than 5. The minimum expected count is .63. The significant value of the Chi-square statistic is greater than 0.05. It means that there is no relationship between yield and tenancy. - - - _... __ ___ ... .. ._.. _-_ __ _ _ - - - - - - - - .. .. .. ~ AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES Annex 6. Analysis ofInput Efficiency for Paddy in DAFEP Sample 1. Correlation between lOR and variables: Correlations lOR Total InpuVHa Yield/ha lOR Pearson Correlation 1 -.108' -.070' Sig. (2-tailed) .000 .022 N 1060 1060 1060 Total Input/Ha Pearson Correlation -.108' 1 .697* Sig. (2-tailed) .000 .000 N 1060 1060 1060 Yield/ha Pearson Correlation -.070* .697*' 1 Sig. (2 -tail ed) .022 .000 N 1060 1060 1060 '*. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed). The correlation reported in the table for Total input/Ha and Yield/Ha towards lOR paddy is negative, and it is significantly different from 0 because the p-value is smaller than 0.05. 2. Asso ciatlon o Relationship between lOR and Type ofVariety. Chi-Square Tests Value df Asymp. Sig. (2-sided) Pearson Chi-Square 20.358a 12 .061 Likelihood Ratio 21.163 12 .048 N of Valid Cases 1064 a. 4 cells (20.0%) have expected count less than 5. The minimum expected count is .17. The significant value of the Chi-square statistic is greater than 0.05. It means that there is no relationship between lOR paddy and type of variety at a 5% significance level, but the relationship is significant at 10%. Symmetric Measures Value Approx. Sig. Nominal by Phi .138 .061 Nominal Cramer's V .080 .061 Contingency Coefficient .137 .061 N of Valid Cases 1064 a. Not assuming the null hypothesis. b. Using the asymptotic standard error assuming the null hypothesis. ANNEX 6, AnalYSIS of InPJt EffiCiency for Paddy In DAFEP Sample 83 ---.~-.- ..· ----- -- AGRICULTURAL EXTENSION SERVICES IN INDONESIA · NEW APPROACHES AND EMERGING ISSUES J The significance values of all three measures are 0.061, indicating a statistically significant relationship at level 10%. The values of all three measures are around 0.1; it subjectively indicates the association lOR and type of variety is not strong enough. o Relationship between lOR and Type of Land: Chi-Square Tests Value df Asymp. Sig. (2-sided) Pearson Chi-Square 31.447 a 28 .298 Likelihood Ratio 25.202 28 .617 N of Valid Cases 1064 a. 16 cells (40.0%) have expected count less than 5. The minimum expected count is .02. The significant value of the Chi-square statistic is greater than 0.05. It means that there is no relationship between lOR paddy and type ofland is not significant at 5% and 10%. o Relationship between lOR and JPAI: Symmetric Measures Asymp. a b Value Std. Error Approx. T Approx. Sig. Ordinal by Kendall's tau-b -.018 .027 -.668 .504 Ordinal Kendall's tau-c -.014 .021 -.668 .504 Gamma -.031 .046 -.668 .504 I N of Valid Cases 1064 I · a. Not assuming the null hypothesis. b. Using the asymptotic standard error assuming the null hypothesis. The significant value of the all three measures of ordinal association is greater than 0.05. It means that there is no relationship between lOR paddy and JPAI is at 5% and 100/0. o Relationship between lOR andJMAO: Symmetric Measures Asymp. a b Value Std. Error Approx. T Approx. Sig. Ordinal by Kendall's tau-b -.018 .027 -.668 .504 Ordinal Kendall's tau-c -.014 .021 -.668 .504 Gamma -.031 .046 -.668 I .504 N of Valid Cases 1064 I a. Not assuming the null hypothesis. b. Using the asymptotic standard error assuming the null hypothesis. The significant value of the all three measures of ordinal association is greater than 0.05. It means that there is no relationship between lOR paddy and JMAO at 5% and 10%. 84 DAfEP Sample "-. AGRICULTURAL EXTENSION SERVICES IN INDONESIA: NEW APPROACHES AND EMERGING ISSUES REFERENCES Alex, G, Byerlee, D, Collion, M, and Rivera, W. 2004 Extension and Rural Development: Converging Views en Institutional Approaches. ARD discussion Paper 4, April. The World Bank. Alex, G., W. Zijp, D. Byerlee, et al. 2002b. Rural Extension andAdvisory Services: New Directions. Rural Development Strategy Background Paper #9. Washington, D.C.: Agriculture & Rural Development Deparment, World Bank. Anderson, J, Feder, G, and S. Ganguly. 2006. Ihe Rise and Fall if Training and Visit Extension, World Bank Policy Research Working Paper no. 3928. Bourgeois R. and F. Jesus (2002). Technical Assistance to the Benchmark Study, Final Report. CIRAD, Jakana. CIRADIIDE-FORCE (2000). DAFEP, Guidelines/or the Benchmark Survey. CIRAD,Jakarta. -----------------------------------. DAFEP, Report if the Pilot Benchmark Study. CIRAD, Jakarta. Christoplos 1., and A. Kidd (2000). Guide for Monitoring, Evaluation and Joint Analyses 0/ Pluralistic Extension Support. Neuchatel Initiative, LBL, Lindau. Deaton, A. (1997). Ihe Analysis if Household Surveys: a Microeconometric Approach to Development Policy. Washington D.C. and Baltimore: ~lhe World Bank and Johns Hopkins University Press. Efron B. ans R.J. Tibshirani (1993). An introduction to the bootstrap. Chapman & Hall. Fallon A. and C. Spada (1997). Environmental Sampling and M=onitoring Primer: Detection and Accommodation ~f Outliers in Normally Distributed Data Sets. Virginia Tech. http://ewr.cee.vt.edul environrnentallteachlsmprimerloutlierloutlier.html. Faure, G. and P. Kleene (2004). Lessons from New Experiences in Extension in West Africa: Management Advicefor J;amily .Farms and Farmers' Organizations. JAEE, vol 10, No 1, pp 37-49 Iglewicz. B. and D. C Hoaglin (1993). How to Detect and Handle Outliers. American Society for Qyality Control, Milwaukee, WI National Centre for Agricultural Extension (1999). DA.FEP Project Implementation Plan Annexes. Ministry ofAgriculture, Jakarta. PT. Hutamacipta Konsulindo (2002). Benchmark Study Final Report. Jakarta. Rivera, W. 1996. Agricultural Extension in Transition Worldwide. Public Administration and Development, Vol. 16, pp. 151-161. Rivera, VvT and Gustafson, D. 1991 eds. Agricultural Extension: Worldwide Institutional Evolution and .Forces for Change. Elsevier, Amsterdam 85 AGRICULTURAL EXTENSION SERVICES IN INDONESIA: .,) NEW APPROACHES AND EMERGING ISSUES Rivera, Wand Zijp, W (eds). 2002. Contractingfor Agricultural Extension: International Case Studies and Emerging Practices (CABI:Wallingford, UK) Salant P. and D. A. Dillman (1994). How to conduct your own survey. New York: John Wiley and Sons. SMERU Research Institute. 2004. Decentralization andAgricultural Service Delivery- Benchmarks, Transfers and Capacity Building in Intergovernmental Relations T ukey J.\V. (1977). Exploratory Data Analysis. Addison- Wesley, Reading, Mass. World Bank, 2000. Full Supervision Report (Office Memorandum), September 6,2000, Jakarta. World Bank, 2005. Iriformation and Communication Technologies for Rural Development. Washington D.C World Bank, 2006. Making the New Indonesia Workfor the Poor, Washington, D.C World Bank2007a Spendingfor Development: Indonesia Public Expenditure Review, \Vashington, D.C World Bank, 2007b Issues in Indonesian Rice Policy, forthcoming 86 References