THEWORLD BANK Discussion Paper EDUCATION AND TRAINING SERIES Report No. EDT49 Farm'er-Education & Farm Efficiency in Peru: The Role of Schooling, Extension Services & Migration Daniel Cotlear December 1986 Education and Training Department Operations Policy Staff The views presented here are those of the author(s), and they should not be interpreted as reflecting those ot the World Bank. Discussion Paper Education and Training Series Report No. EDT49 FARMER EDUCATION AND FARM EFFICIENCY IN PERU: THE ROLE OF SCHOOLING, EXTENSION SERVICES AND MIGRATION Daniel Cotlear (consultant) Education Policy Division Education and Training Department December 1986 The World Bank does not accept responsibility for the views expressed herein, which are those of the author(s) and should not be attributed to the World Bank or its affiliated organizations. The findings, interpretations, and conclusions are the results of research or analysis supported by the Bank; they do not necessarily represent official policy of the Bank. Copyright © 1986 The International Bank for Reconstruction and Development/ The World Bank ABSTRACT The purpose of the study is to explore the relation between the formal, nonformal, and informal education that farmers in Peru have received and their subsequent efficiency as farm operators. Education has an important role, but this role depends crucially on its technological and economic context. The effects of schooling are stronger in the more modern regions. A minimum level of schooling is necessary for the positive effects to appear, and this level increases with the modernity of the environment. Agricultural extension has a direct effect on contacted farmers, but the effectiveness of extension depends on the appropriateness of the message to the stage of agricultural development of the farmers. An indirect effect of extension through imitation is also of importance. Migration experience has powerful effects as a form of informal education. Table of Contents Page No. EXECUTIVE SUMMARY ......... i - ii 1. INTRODUCTION .................................. 1 - 2 2. BACKGROUND .................................. 2 - 15 3. THE SURVEY .................................. 15 - 19 4. THE DATA ............... . ........ . 19 - 26 5. METHODS FOR THE ESTIMATION OF THE DETERMINANTS OF PRODUCTIVITY .................................... 26 - 29 6. PRODUCTIVITY: BASIC RESULTS .......................... 29 - 49 A. PHYSICAL INPUTS ................................ 32 -33 - B. FORMAL EDUCATION ................... 33 - 38 C. NON-FORMAL EDUCATION: DIRECT EFFECTS ........ 38 - 41 D. NON-FORMAL EDUCATION: INDIRECT EFFECTS ....... 42 - 43 E. INFORMAL EDUCATION ................. 43 - 45 F. TECHNOLOGY AND CREDIT . .............. 45 - 46 G. QUALITY OF FORMAL EDUCATION .................... 46 - 49 7. METHODS FOR THE ESTIMATION OF THE DETERMINANTS OF ADOPTION BEHAVIOUR ............................. . 49 - 52 8. ADOPTION: BASIC RESULTS ............................ 53 - 61 9. SUMMARY AND CONCLUSIONS .......................... . 61 - 65 REFERENCES .. ........................................ 66 - 68 ACKNOWLEDGMENTS The research reported here was prepared during consultancy: for World Bank Research Project 673-26 "Education and Employment in the Informal Sector." Adolfo Figueroa, John Knight and Joy de Beyer provided useful advice and assistance on various aspects of this work. Dennis de Tray and Peter Moock made comments to a first draft of the paper. I wish to acknowledge my indebtedness to all of them. The data for the study comes from a survey conducted by the author in the context of an ECIEL study on rural productivity financed by the Inter-American Development Bank. The results reported here appear in a chapter of my D. Phil. Dissertation "Technological and Institutional Change Among the Peruvian Peasantry." EXECUTIVE SUMMARY The purpose of the study is to explore the relation between the formal, nonformal, and informal education received by farmers in Peru and their efficiency as farm operators. The effects of schooling, agricultural extension services, and migration experience (as a form of informal education) are analyzed. Specific indicators of greater farm efficiency are the achievement of higher agricultural productivity and the adoption of modern mechanical, biological, and chemical inputs. The geographical focus of the study is the Peruvian sierra. In simple terms, Peru is divided into three ecological zones running in parallel strips from north to south. The westernmost zone is a mostly desertic coast, the easternmost zone is constituted by the jungle; separating the two strips is the sierra (the Andean range). Approximately three-fourths of the rural population of the country live in the sierra; and a similar proportion of the cultivable land is located in this area. It is here that the poorest segment of the population lives, working on small household farms. The study's data come from a survey of rural households in three regions of the Peruvian highlands. The regions were selected because of agro-ecological similarity (in all three the agricultural land was located at a height of 3,500 to 4,000 meters above sea level and agriculture is unirrigated), and to show a gradient of technological modernity: a modern, an intermediate, and a traditional region were chosen. Differing levels of modernity were obtained by selecting ecologically similar regions located at different distances from urban markets. The total sample consists of 555 observations obtained in nine villages in the modern region, five villages in the intermediate region, and four villages in the traditional region. The questionnaire was designed for a single visit of approximately two hours. A revisit was made when the completed questionnaire was found to be incomplete or inconsistent. The survey took place in August and September 1983 and the information refers to the 1982-1983 agricultural year. The analysis of the effects of education on productivity is approached by means of production functions. Differences in productivity imply that different outputs will be obtained from a given bundle of physical inputs. As we were interested in examining whether the educational characteristics of the household have an effect on productivity, we specified the production function to include, on the right hand side, in addition to measured physical inputs, measures of the educational characteristics of the household. Adoption of modern inputs is examined as a dichotomous dependent variable, the results being obtained with the use of logit analysis. Education (of some type) was found to be an important determinant of adoption and productivity. With respect to adoption, the. role of education was found to be greater in the early stages of the diffusion process. Later on, imitation of the successful early adopters becomes generalized. Three patterns were found in relation to formal education. First, the effects of schooling are stronger in the modern than in the traditional regions. This suggests that to be successful, educational policies must be accompanied by policies which dynamize the technology in use, and that this dynamization of agriculture is easier to achieve if it is accompanied by the supply of formal and nonformal education. Second, the results show the existence of a threshold effect, by which formal education begins to influence output only after a certain number of years of schooling has been obtained. This threshold seems to be higher where the complexity of the technology and the economic choices available are greater. This implies that while basic levels of education may be effective in speeding traditional regions into the first stages of technological modernization, further technological development will require higher levels of formal education. Third, the quality of schooling can strongly influence productivity. Urban schooling was found to have stronger effects on agricultural productivity than rural schooling. When the direct effects of extension were examined a significdnt impact was found only in the intermediate region. The main explanation for this seems to be related to the adequacy of the message transmitted by the extension service. A similar message is transmitted in the three regions, but it seems appropriate only for the intermediate region, being too far advanced for the traditional region and failing to include any important innovations for the technological levels already achieved in the modern region. The existence of indirect effects of extension by imitation - was noted. Recent extension was found to affect productivity. However, farmers who had been contacted more than three years prior to the survey no longer distinguished themselves from farmers who had never been in contact with the extensionists. Further, there was some indication that in those villages which in the past had been more intensely visited by extensionists, the average level of productivity was higher than elsewhere. The importance of these indirect effects is something that must be taken into account for the assessment of extension services designed to reach only a selected group of farmers. Finally, the effects of two forms of informal education on output were examined, that of age (as a proxy for experience) and that of migration experience. The expectation with respect to age was of a positive effect on output. Instead, a negative effect was found in each of the three regions, the effect being more pronounced the more modern the area. These findings were interpreted as reflecting a "cohort effect," according to which older peasants tend to stick to older, less productive technologies. Migration experience was found to influence adoptive behavior in the traditional region and productivity in the intermediate region. It was noted that migration to urban areas was of greater value than migration to other rural research - it was found that when migration experience was not included in the analysis, the effects of formal education were being overestimated. FARMER EDUCATION AND FARM EFFICIENCY IN PERU. THE ROLE OF SCHOOLING, EXTENSION SERVICES AND MIGRATION 1. Introduction The purpose of the study is to explore the relation between the formal, non-formal and informal education farmers have received and their subsequent efficiency as farm operators in Peru. A monograph by Jamison and Lau (1982) reviews the literature on this area- through 1978 and develops a thorough theoretical and methodological framework for examining the role of education in production. Later, in an article designed to ascertain further the relation betweeen education and farmer efficiency, Jamison and Moock (1984) provided an excelent exposition of the issues involved, and developed a useful framework for the examination and presentation of results. The paper is written following that tradition. The data for testing aspects of the causal relationships between education and efficiency were obtained in the Peruvian Andes as part of a larger study into technological change in peasant agriculture and its determinants.I Our data set allows testing some of the major hypotheses in the literature, providing further results which can be compared with those previously obtained from a different agricultural environment. The results are also of interest because they include previously overlooked variables that can be used to extend our understanding of the relationship between education and efficiency in production. Possibly the most important contribution of this study arises from 1 See Cotlear(1986). 2 the way in which the sample was designed: It includes households from regions at different degrees of development which are otherwise comparable, hence, we will be able to relate to specific findings concerning the effects of education to specific modernity contexts. This paper is divided in the following way. The next section presents a general discussion on the determinants of agricultural productivity and adoption of technological innovations, with special focus on the role of education. The discussion is summarized in the form of a set of testable empirical hypotheses. Section 3 describes the data collection methodology. Section 4 presents the data that will be used to test the hypotheses. A test concerning the determinants of agricultural productivity is presented first with the use of production function analysis: the method is described in section 5 and the results are presented in section 6. The subsequent two sections describe the method used to test the determinants of adoption behaviour, and the results obtained. Section 9 compLetes the paper with a,general summary and conclusions. 2. Background Three different types of education are often distinguished in the literature: "Formal", which consists mainly of schooling; *non- formal", which includes different kinds of extension and organized apprenticeships; and "informal", which refers to a wide definition of learning-by-doing including not only direct experience in a particular job but the multi-dimensional processes of learning that arise from being exposed to different circumstances.3 In the Andean context a major source of informal education affecting attitudes and providing information has been shown in many studies to be migration.4 3 Coombs and Ahmed(1974), Figueroa(1985). 4 See for instance Adams(1959), Arguedas (1964), Oegregori et.al.(1973), Laite(1981). 3 Education can have "cognitive" and "non-cognitive" effects. The cognitive effects consist of the development of general reasoning skills and the transmission of specific knowledge. The non-cognitive effects modify attitudes and beliefs. The relative importance of these effects is much debated but still poorly understood.5 In the cognitive area there exist strong interactions between developing a generalized capacity for thinking and learning on the one hand, and the specific subjects learned on the other. The different types of education are better suited for the provision of some of the cognitive effects. For instance, it has been argued that the greater structure, longer duration and specific age group of school attendance makes formal education best suited for the "formation of competences", while the greater flexibility of non-formal services which allow them to deliver a message closer to the work place makes this type of education best suited for the "transmission of information".6 Informal education can provide either aspect of the cognitive effects depending on the specific type of experience: for example, a migration experience as an urban street seller may improve the numeracy capabilities of a peasant, facilitating future calculations of costs and returns on the farm, while his experience as a farm wage labourer can put him in touch with specific details related to the use of new technologies which he can then apply to his own farm. Many of the non-cognitive effects of education -receptivity to new ideas, competitiveness, and willingness to accept discipline- are directly relevant to productive economic activity. Others - tolerance, self-confidence, social and civic responsibility- are more personal or political in nature, but may also affect economic performance in the farm.7 Formal and informal education are likely to be the most important processes for the change of attitudes and beliefs. 5 See for instance the debate between the "anthropological" and the "economic" views in Wharton(1969). 6 Bowman(1976), quoted by Jamison and Moock(1984) p.69. 7 World Bank Development Report 1980, p.47. 4 What are the mechanisms through which education can have an effect on output and incomes? Education may have productive value because it enables the farmer to produce larger output quantities from the same measured quantities of inputs and because it helps the farmer to allocate resources in a cost-efficient manner, choosing which output to produce, how much of each output to produce, and in what proportions to use inputs in the production of any output.8 Welch has labelled these effects of education as the "worker" effect and the "allocative" effect.9 The former is related to the enhanced ability of production with a given set of inputs, and the latter has to do with the farmer's ability to acquire and decode information about costs and productive characteristics of other inputs. A central aspect of the allocative effect is the capacity to evaluate and adopt profitable new technologies. What is the role of education in the process of adoption of innovations? Rogers defines the adoption process as "the mental process an individual passes from first hearing about an innovation to final adoption".10 This definition stresses the fact that adoption is not a single action referred to the moment of introduction of a new technology into the farmer's production process, but rather it is a process which takes time. A major role of education can be to reduce the time lag involved in adoption. The required time lag to final adoption can be conceptually divided in two stages: the discovery stage leading from the availability of the new technology to awareness of its technical characteristics and the evaluation stage which leads from awareness to use.11 Education can have a role in reducing the time required to complete both stages. The time lag involved in the discovery stage will depend on the peasant's "information field", and this is likely to be increased by all three types of education discussed above. Formal education is likely to facilitate the capacity to search for information, and to 8 Jamison and Moock (1984) p.68. 9 Welch(1970) 10 Rogers (1962) p.17. 11 Lindner et. al.(1979). The two stages are not fully independent one from the other, and there may exist some overlap; in this sense the distinction is more conceptual than empirical in nature. 5 order and systematize this information. Agricultural extension programmes are specifically designed to take this type of information to the farmer, explaining the technical details and the likely consequences of the use of the new technologies. Some forms of informal education are equally likely to put the farmer in contact with the effects of the use of new technology. The evaluation stage usually includes the assessment of two aspects. One is whether the new technology will be appropriate to the farmer's technical and economic conditions of production, e.g. is it appropriate to his own soils, his use of bullock traction, his availability of land and labour, or his access to finance or markets?. The second assessment is of whether the technology will be profitable under those conditions. This assessment is often rendered particularly difficult by the existence of uncertainty in production, e.g. net incomes were higher last year with the new technology, but will that be the cas.e under different weather conditions? Perhaps average incomes are higher over the years but, are the probabilities of total failure in any year larger with the new technology? Moreover, and especially important in the early stages of modernization, innovation will imply further involvement in the markets for inputs and outputs and this will add uncertainty from market fluctuations to the already uncertain natural environment of the peasant. Further still, innovation will often imply not only a change of inputs, but an increase in the total expenditure required for production. Under these conditions, even if the average profitability is increased and the probabilities of failure reduced, the harm caused by failure will be greater. Many variables must be taken into consideration when assessing the new technologies. Imposing order on the existing evidence and understanding the results is a difficult process, and here education may be expected to play an important role. Schooling can facilitate the process in several ways. Increased numerical skills are likely to be of importance. A greater capacity of abstraction will make it easier for an educated farmer to uncover causal relations between technology and outputs which -because of long lags between 6 application and results and weather related randomness influencing the results- may remain obscure to less educated farmers. Well designed extension programmes are also likely to help the farmer through this process by demonstrating the technologies under conditions which are similar to the farmer's own, by pointing out the causality between the use of the new inputs and specific results and by facilitating the calculations of profitability. Also, crucial in this phase are the non-cognitive roles of education which can make the farmer more receptive to new ideas, and more self-confident and consequently more willing to innovate. With many innovations, the time required by individual farmers to go through the two stages of adoption described will be reduced once adoption by other farmers has occurred. Adoption by the early starters will facilitate the process for the followers. The discovery stage will be shortened because, as the new technologies are employed in neighbouring fields, it will be much easier to identify the relevant technical characteristics of the technology. The evaluation stage will also be shortened because the followers will have the opportunity to observe the use of the technology under conditions similar to their own. Further, for many followers the main criterion utilized to establish the superiority of the new technologies over the traditional ones will be the continuous use of these techniques by the early adopters. In other words, the early adopters can act as "technological leaders": they experiment with the new techniques, select those which are superior to the traditional technology and adapt them to local conditions. At this point the decision-making process of the followers is made much simpler. An implication of this process could be that, where imitation is important, the role of education will be especially important in the initial stages, when an innovation is being introduced into an area. After its initial introduction, a main channel of further adoption by individual farmers will be imitation, and the educational requirements at this stage are likely to be lowered. Most extension programmes are based in the assumption that after the earlier stages, 7 diffusion occurs largely by imitation of technological leaders.12 These methods involve concentrating the attention of the extensionists on a few farmers, expecting to reach the non-contacted farmers by indirect means.13 In the discussion above, we have referred to adoption as a process largely consistent on the identification of "superior technologies". In this situation, the main role of education was described as one of aiding in this process of identification and hence shortening the time lag between first hearing about an innovation and final adoption. This leaves aside an important possibility. Education itself can be a crucial complementary input in the new technological package. The superiority of the new technology over the traditionally used ones may require the presence of high levels of education. The productivity levels obtained with the new technology may crucially depend on the farmer's education. The use of some modern technologies may involve a large number of alternative procedures, and the choice of a particular procedure may depend on the conditions of the natural environment or the market. When this is the case, recourse to memory may be insufficient and personalized transmission of information inefficient. In this situation, literacy may be needed to facilitate the storage of the large amounts of information involved, and ease its depersonalized transmission. When chemical inputs are introduced, numeracy may be required to calculate the correct proportions in the use of the inputs. All this suggests that there may exist high costs 12 For instance, this is the case of the "Training and Visit" system used by the Peruvian Extension Service (INIPA) at the time of our survey. A description of this system is provided by Benor and Harrison(1977). 13 Clearly a necessary characteristic that the early starters must have to become "technological leaders" is that they must produce under technical and economic conditions similar to those of the followers. If early adoption occurs mostly because these farmers have, say, more access to irrigated lands, or because these farmers are wealthier and have greater possibilities of financing the investment in new inputs, they are unlikely to produce the "leadership effect". The other farmers will not regard their selection of innovations as a valid proof of the superiority of the new techniques under their own conditions, and little imitation will occur. 8 to learn.ing the use of the new techniques, and these costs are likely to be smaller for the more educated farmers. The level of education required for the efficient use of inputs could depend on the sophistication of the new technologies. For the simpler technologies no formal education may be necessary, or basic literacy might be enough. For more complex technologies, higher levels of education will be needed. An implication of this is that when modern technologies are in use, education is likely to have a positive effect on productivity. Thus, education may favour adoption not only because of the role it can play in the faster discovery and assessment of the new technologies, but because it acts as a complementary input for the appropriate use of the new technologies. Hence, adoption will be more profitable for the more educated peasants. In other words, a major reason for the early adoption of the more educated farmers may be that the gain in productivity that can be obtained from the new technologies is larger for them. The main hypothesis which derives from the preceding discussion is straightforward. Education helps people to obtain and evaluate information about improved techniques and new economic opportunities, to keep track of past events, and estimate the returns of potential innovations. Education also helps people to learn the adequate use of the new techniques, with a lower learning-cost. In consequence, we expect to observe education to be associated with higher productivity. Several studies have attempted to test this hypothesis. Lockheed, Jamison and Lau reviewed studies from 37 regions which tested the effects of farmers' schooling on agricultural productivity. Most of these studies used production function analysis to test the hypothesis: output was regressed against physical inputs a.-d education indicators. The survey summarized the results and found that in 15 cases education had a significantly positive coefficient at the .35 level, in 6 cases the sign was negative but not significant, and in 16 cases the sign was positive 9 but not significant.14 Jameson and Phillips commenting on the review noted that for the Latin American countries only four of thirteen coefficients were positive and significant. They then estimate the effects of education in four Latin American countries, finding a positive and sigificant coefficient for only one country. These ambiguous results demonstrate the need for a more precise specification of the hypotheses to be tested. Schultz has argued that the value of education is likely to be greater in a modern environment. In traditional environments, technology and markets change very slowly. Here, discovery of optimal economic behaviour in the use of technology and the the allocation of resources has occurred by a long process of trial and error and the results of this process can be replicated over the years and even from generation to generation. By contrast, a modern environment is characterized by constant changes in technology and market situations. Facing a dynamic environment of this sort, there exists the constant need to adjust to the new opportunities by taking decisions concerning new options which did not exist before. Thus, changes in the technological environment increase the value of a farmer's "ability to deal with disequilibrium", and it is here that education may play a major role.15 Evidence favourable to this hypothesis has been found in previous studies. When the results summarized by Lockheed, Jamison and Lau were classified according to whether the regions are modern or traditional, they found that the effects of education are much more likely to be positive in modernizing agricultural environments than in traditional ones. By way of summarizing the above discussion, a number of hypotheses concerning the effects of education on adoption and productivity can be presented: 14 Lockheed, Jamison and Lau(1980). A survey describing the effects of schooling on adoption is provided in Feder,Just and Zilberman(1985). 15 The original argument was presented in Schultz(1964); it has been further refined and developed in Schultz (1975). The quotation is from the latter article. 10 - Formal,non-formal and informal education have a positive effect on adoption and productivity. - The role of education is likely to be stronger in modern, dynamic regions than in traditional regions. - The required level of education for the efficient use of inputs is likely to depend on the degree of sophistication of the new technologies. - The effects of schooling will depend not only on the number of years of exposure to the schooling system, but also on the quality of the education received. -Since agricultural extension is more concerned with "the transmission of information" than with "the formation of competences" and given that the value of having received specific information about the use of new agricultural practices is likely to diminish with time as newer technologies are introduced, the value of extension contacts will diminish with time. - The role of education on adoption is likely to be of special importance for the more recent innovations; in later stages diffusion occurs by imitation, and the education of individual farmers loses some of its initial importance. Recent studies have argued that the differential propensity to adopt innovations by different farmers can also be related to the existence of constraints such as inadequate farm size, or lack of credit.16 An often-mentioned impediment to the adoption of new technologies by smaller farms is related to costs of implementation. Large fixed costs reduce the tendency to adopt and slow down the rate of adoption by smaller farms.17 It is important to note, however, 16 Inadequate tenancy arrangements are also often mentioned in this regard. We shall not deal with this because most land in our regions is owned by the farmers and usual tenancy arrangements often regarded as potential inhibitors of adoption such as share-cropping are of little importance. 17 These hypotheses have been supported by much empirical evidence. It was found true for example for ox cultivation in Africa(Weil 1970) and tractor machinery in South Asia (Binswanger 1978). LL that the relative lumpiness of some technology can be mitigated by a larger variety of desiz-s and by the emergence of markets for hired services. Also, some inputs such as chemical fertilizer and HYVs are divisible. Even in these cases, some authors have found evidence that the incidence of adoption of these technologies is positively related to farm size. 18 Other studies suggest that the smaller farms sometimes lag behind the larger ones in adopting HYVs, but eventually catch up.19 Several factors can explain this behaviour. One possibility is that the larger farms may face a less binding capital constraint, either because of greater savings, or because of greater access to the credit market. Larger farmers may be able to "afford being less risk-averse". Even seemingly neutral technologies such as HYV may entail significant set-up costs in terms of learning the use of a technology and locating markets for inputs and outputs. These factors can be considered as fixed expenses and can be quite high. Returns to this investment will be roughly proportional to the amount of land under cultivation, and this could explain why the larger farms are the first to adopt. Farm size and education tend to be correlated. In consequence, it will be important to include farm size variables when we analyze the education coefficients to reduce the chances that any effects that education may seem to have on adoption are really only a reflection of larger farm size. The need for credit to facilitate adoption has been the subject of heated debate. Capital, in the form of either accumulated savings or access to capital markets is required to finance many new agricultural technologies. Thus, differential access to capital may be a factor explaining differential rates of adoption. This is the case with indivisible technology, in particular, such as tractors or other machinery that require a large initial investment. There is some debate, however, concerning the need for credit in the case of 18 Feder et. al.(1985) p272 claims this to be the case. 19 Ruttan and Binswanger(1978) _2 divisible innovations such as fertilizers and new seed varieties. What seems clear is that lack of credit is likely to be a constraint on adoption even in the case of divisible inputs if there exist complementary inputs which are indivisible (such as tube wells) and indispensable for adoption. Some authors have argued that lack of credit alone does not inhibit adoption of innovations that are scale- neutral, and that, specifically in the case of HYV, the profitability of adoption will induce even small farms to mobilize (from whatever sources to which they have access) the relatively small cash requirements for necessary inputs.20 A number of studies, however, .have found that lack of credit does significantly limit adoption of HYV technology even where fixed pecuniary costs are not large.21 The inputs which embody the modern technology in the areas of our study are all divisible -including the use of tractor services, which can be rented on a per hour basis. We shall present an empirical test of the importance of credit as a determinant of adoption and productivity. Where the market for credit is rationed, education can play a role in facilitating access to credit. The more educated farmers are likely to be able to complete application forms, and deal with bank officials with less difficulty than uneducated ones. It is therefore possible that an observed correlation between education and productivity is mediated by the role of credit. If this was the only way in which education leads to increases in productivity, the policies which would be called for would not be to expand education, but to expand the funds available and reduce the rationing. Hence, it is of great interest to establish the indirect effects of education (i.e. those which are mediated by other factors such as credit) and the direct effects i.e. those which persist when the mediating factors are controlled for. It should be noticed that the interpretation we are favouring of the potential correlation btcween education and productivity is 20 Several authors who agree on this point are quoted by Schujter and Van der Veen(1977) 21 Opinion surveys of farmers leading to this conclusion are presented by Bhalla(L979) and Khan(1975). 13 that education functions to enhance the productive capacity of workers and the allocative capacity of managers. An alternative explanation suggests that education does not make workers more productive but merely identifies those who were more productive to begin with. This argument, usually identified as the "screening hypothesis" has been originally utilized to explain the correlation between education and earnings among employees. In this hypothesis, employers find it difficult to identify the ability of employees and so they need a screening mechanism to differentiate the candidates. Workers provide themselves with education as a mechanism to signal their ability to potential employers.22 Since in self-employment there are no employers and employees, there is no need for signals concerning the employee's ability. The screening hypothesis cannot be totally ruled out by the finding of correlation between education and earnings among the self-employed, but it is weakened by it.23 A defendant of the screening hypothesis may argue that since the self-employed may believe that education enhances their productive capacity (even if it does not) they may demand education in the same way as workers do in the case of wage employment. The argument would be that workers in all sectors believe that education enhances their productive capacity. Even if education does nothing to productive capacity, in the wage sector it signals ability, and ability-blind employers are forced to use education as a signal for ability, and pay according to education. This behaviour reinforces the belief that education adds to productive capacity. The same cycle of belief leading to action reinforcing the belief may occur in self-employment. Since it is easier for the more able people to obtain more education, as in the wage sector, these two variables will be strongly correlated, and it is impossible to discern which of these variables affects earnings. The argument is 22 This description of the posible interpretations of the correlation is taken from Jamison and Moock(1984) p.68. 23 A strong version of the screening hypothesis asserts that education merely identifies students with particular attributes, acquired either at birth or by virtue of family background but does not itself produce or in any way improve those attributes. Since there is little point in self-screening, this version implies that education has no effect on earnings when it comes to the self- employed. i4 weakened by the fact that in this context there are no penalties against those with high ability and low education, and no bonuses for those with low ability and (exceptionally obtained) high education levels. In wage employment, according to the hypothesis, ability- blind employers are forced to pay roughly according to education. It is this that leads to the strong association of education and earnings. Since the same thing will not happen in self employment, the relationship will be weaker and will tend to break down as people realize that education has no economic returns.24 It remains true that care must be taken when interpreting empirical results since there may be an upward bias on the education coefficients due to the positive correlation which is likely to exist between ability (IQ, drive, etc.) and education. On the other hand, if we measure the effects of education by the size of the difference in output or wages obtained by individuals of different educational levels, there is a likely tendency to underestimate the benefits of education caused by the existence of positive externalities which arise in the form of the less educated imitating the more educated. This is likely to be common in a peasant environment where agricultural activities are done outdoors and in small visible plots.25 The screening arugment also has implications for the interpretation of other variables such as credit and extension. The agents in charge of these services may select the more able farmers. In consequence part of the impact measured by the variables for credit and extension can possibly correspond to ability. 24 Jamison and Moock(1984) did a study specifically designed to provide a test between these two hypothesis among small farmers. They controlled for "ability" with the use of variables measuring family background. Their findings support the claim that under conditions of self-employment like the ones described, if a correlation is found, it is likely to be due to the productive effect of education. 25 It may be noticed that if the benefit of education is measured in the way indicated, then if imitation reduces the gap by X, it underestimates the true benefit obtained by at least 2X. 15 In this paper we will present results which enrich those obtained in previous studies. We use data specifically collected with the above issues in mind, and hence more precise; in several respects than the data used in some of the other studies. Also, we include in the analysis new variables previously overlooked in the literature. However, possibly the most important improvement of this study over the previous ones is that our sample was designed to include regions of different degrees of development which are otherwise comparable. This means that the sample design will allow us to relate our findings to specific contexts. 3. The survey26 The data comes from a survey of rural households in three regions of the Peruvian Highlands. The regions are the valley of Yanamarca in the Department of Junin in the Central Highlands and the plateau of Chinchero and the pampa of Sangarara in the Department of Cusco in the Southern highlands. The total sample consists of 555 observations. The survey was conducted by the author while he was Assistant Professor in the Department of Economics of the Catholic University of Lima. It was done in the framework of an ECIEL study and financed by the Interamerican Development Bank. In simple terms, Peru is divided into three ecological zones running in parallel strips from north to south. The westernmost zone is the coast, this is a mostly desertic area crossed by narrow valleys formed by rivers which ran from the highlands to the Pacific Ocean. All the agriculture is done by irrigation and the most modern agriculture of the country is concentrated in these valleys. Most of the largest cities of the country (including Lima) are also located in these valleys. The easternmost area is a wide strip of jungle including a very small proportion of the total population and an 26 For a fuller description of the methodology utilized see Cotlear (1986) chapter III and appendix Al. 16 insignificant proportion of the agricultural production. The middle strip consists of the sierra (the Andean Range). Three fourths of the rural population of the country live here and a similar proportion of the cultivable land is located in this area. The highland area is of particular developmental interest because most of the population classified as "poor in absolute terms" by international criteria are small farmers from the most traditional areas of the sierra.27 Agriculture in this region is mostly unirrigated. The areas of the survey were specifically selected to comply with two main conditions required by the study. Firstly, the objectives of the study required the comparison of production characteristics from regions with a varying degree of technological dynamism. Hence we chose a modern region where modern inputs and practices are widely in use and in continuous change: Rl (the valley of Yanamarca), a traditional region where they are scarcely used: R3 (the pampa of Sangarara) and an intermediate region: R2 (the plateau of Chinchero). The second condition was ecological homogeneity. Given the small size of the sample, the results concerning the effects of technology on productivity were likely to be darkened by agro-ecological factors if we allowed the whole range of ecological variation found in the sierra to be present in the sample. Hence, the sample had to be restricted to areas from a narrow range of ecological variation. The highlands show a wide variation of agro- ecological conditions changing mostly with altitude. Six different subregions are often distinguished, two of them are above 4000 meters above sea level and do not contain agriculture. Our sample is taken from the "suni" region which is the highest region containing agriculture (3500 to 4000 meters above sea level) and is characterized by being agriculture "above the line of maize". This sub-region includes approximately a fifth of the rural population of the sierra and approximately the same proportion of the agricultural land. 27 Thomas(1978) 17 Two additional considerations further restricted the choice of regions for the survey. Several parts of the highlands had been affected by drought in the agricultural year of our study and we had to choose among the non-drought areas. Finally, the search for areas for the study was restricted to regions where terrorist activity was not intense. The regions chosen were the first we found to satisfy all the required conditions. Our sample is a stratified random sample of households in nine villages in Yanamarca, five villages in Chinchero and four villages in Sangarara. The universe was stratified according to whether the households had obtained agricultural credit from formal sources or not during the 1982-83 agricultural year. Lists of all household heads were obtained for the eighteen villages. In most cases the lists had not been updated and were often incomplete and had to be completed and checked. This was done in a long process which included the help of authorities and other residents of each village. The revision of the lists included a classification of households which separated households headed by women and households which did not engage in agriculture, these were not considered for the survey. The male headed households which engage in agriculture were then classified according to whether they had obtained credit in the previous agricultural year. This was done with the help of lists obtained from the financial institutions which operate in the regions of the study. It was estimated that the available budget allowed the completion of approximately 550 interviews. The sample size was chosen to be of 150 observations in each of Chinchero and Sangarara and 250 observations in Yanamarca (a larger sample size was chosen for the modern region considering that the analysis of data independently for that region could be of particular interest). The households to be interviewed where chosen in a public lottery and so were the households selected for replacement of households lost through sample attrition. The lottery was held by the village authorities who had to choose a set number of households 13 from the credit-holder and non credit-holder lists. In each village the number of observation was chosen to be proportionate with the weight of that particular vilLage in the total region. These procedures led to a slight departure from the originally planned sample size, the samples were increased in Yanamarca by six observations and in Chinchero by one observation. Before the survey began in each village, a campaign of public relations was developed to prepare a good reception for the surveyors. Several visits were made to each village, and in these visits first the authorities of the village, and later the villagers at large in public assemblies were informed of our presence in the region, the objectives of our study and the procedures of the survey and contents of the questionnaire. In most villages discussions were held to motivate the participation of the villagers. We -often obtained a gesture of official approval for the study. This occurred sometimes in the form of a "vote of welcome" from the communal assembly, others by the offer of some form of material support for our work such as the appointment of an aide for the duration of the field-work or lodgings in a communal building. The lottery to select the households to be interviewed was often held in a public meeting. These proceedings were useful to obtain understanding and goodwill from the respondents and to secure a good reception for the surveyors (there were only 14 cases of rejection to answer the questionnaire among all the households selected). Field workers for the interviewing were recruited in Cusco for Chinchero and Sangarara, and in Lima for Yanamarca. The interviewers were university students in the Yanamarca team and university graduates in the Cusco team. In the latter team most were sons and daughters of farmers, they all spoke Spanish (the language in which the questionnaires were written) and Quechua (the language of the respondents in Cusco). The interviews were done by individual surveyors who worked in sub-teams in charge of a supervisor. One of the responsibilities of the supervisor was to daily revise the completed questionnaires for arithmetic consistency and completeness. 19 The instrument was designed for a single visit of approximately two hours. A revisit was made when the completed questionnaire was found to be inconsistent or incomplete by the field-supervisor. The survey took place in August and September 1983. These months were chosen to accommodate to the agricultural calendar. In our regions the last harvest ends in July, hence the respondents already knew the amount of output harvested in the year, and the data was still fresh in their minds. Also the new agricultural year had not started and so the chances of confusing inputs used in 1982 with those of 1983 were reduced. Finally, these dates were the most appropriate because the survey was done during the seasonal lull in activity which made the respondents more willing to collaborate with their time. The survey instrument included- questions on the stocks and flows for the agricultural production of the household, given the large fragmentation of land we found it necessary to facilitate the respondent's task by asking the questions on agricultural production at the plot level. Other questions were referred to non-agricultural farm production, off-farm income producing activities, credit, household member characteristics on education, extension and seasonal and long-term migration. In this paper we will restrict our attention to agricultural production and more specifically to the impact of household members characteristics on productive efficiency and adoptive behaviour. The agricultural production data covers the 1982-83 agricultural year and includes data for all crops with particular attention to potato. Potato is the most important and technically most dynamic crop grown and it is on this crop that we will concentrate the analysis. 4. The Data Table 1 presents the definitions of the variables used in the regression analysis. The input and output variables refer to the Table _1 Variable Definitions Variables Definition ._. ... . ... .. .. ._.W._...... ......... . . .. ..... l. output Kg. of potatoes harvested. 2. Area cultivated Rrea cultivated with potatoes in metres. 3. Labour input Person-da7s utilized for potato production; iale and female, hired and family labour days are weighted equally. 4. Animial and tractor input Measured in equivalent animal days utilized for potato production; the inputs are weighted in terms of the time required to bring an area of land into the sane state of readiness for planting. S. village duamy Dummy variable for houeholds from Acolla in region 1. 6. School attainment Of head of household, in years. 7. 1-5 year of schooling Dummy variable; I if schooling of head of household in this range, 0 otherwise. B. 4-5 yr. of schooling Dummy variable; I if schooling of head of household in this range, 0 otherwise. 9. 6+ yr. of schooling Dumar variable; I if schooling of head of household in this rtange, 0 otherwise. 10. 4-5 yr. of rural schoeling Dummy variable; 1 if schooling of head of household in this rtange in rural schools, 0 otherwise. 11. 6+ yr. of rural schooling Dusmy veriable; I if schooling of head of household in this range in rurall.schools,0 otherwise. 12. 4-5 yr. of urban schooling Dumny variable; I f schooling of head of houshold in this range in urban schools, 0 otherwise. 13. 6+ yr. of urban schooling Dummy variable; I if schooling of head of household in this range in urbafn schools, 0 otherwise. 14. Recent contact with extension Agent Dusmy variable; I if there was any contact in the 3 years previous to the survey, 0 othervise. 15. Old contact with extension agent Dummy variable; I if there was any contact 3 years before the survey, 0 othervise. 16. Migration experience Number of years the head of the household has spent away fron the village. 17. Migration experience included Dumny variable; 1 if the head of the household has any agricultural activity migration experience which includes agricultural vork away from the village. 18. Age Of head of household, in years. 19. Households in village with Proportion of households in the village where the household with recent extension contact lives, who have received extension contacts in the last three years. 20. Households in village with Proportion of households in the village where the household old ext. contact lives who have received extension contacts three years before the survey. 21. Households in village with 85S of Proportion of households in the village potato land vith HYVs 22. Uses HYU Dunmy variable; I if some of the area cultivated is planted vith high yielding varieties, 0 othervfse, 23. Received formal credit Dummy variable; I if the household received credit from a formal institution for the agricultural campaign in the year previous to the survey, 0 othervise. 221 production of potatoes. The education characteristics refer to the head of the household. The means and standard deviations of the variables in each of the three regions of the study are presented in table 2. Here we review them very briefly.28 Output, the dependent variable, is measured in physical terms. This is also the case for two of the other inputs, land and labour. Traction power for cultivation can be provided by animals and/or tractors. Tractor inputs have been converted into equivalent animal- days and added to the animal traction by applying technical coefficients which reflect the time required to complete each task with the two- alternatives. The use of modern inputs is represented by a dummy variable for the use of high yielding varieties of seed. Table 2 shows that the mean potato output in Rl is almost four times as large as' the mean output in R2 and seven times larger than in R3. The ratio of average output to average land or to average labour input shows the "gradient of modernity" which characterizes the comparison of the three regions: productivity of land and labour are highest in RI and lowest in R3. This same gradient is found for the use of HYVs: most households use them in Rl, about a third use them in R2 and only 3% use them in R3. While we made special efforts to obtain samples from homogeneous regions, we could not avoid including in RI the village of Acolla, the capital of the district of all the villages sampled in Rl. Acolla's agriculture is in the same valley as all the other villages and its ecological conditions are basically similar to those of the other villages. However, it is at a slightly lower altitude than the other villages and was this year slightly drier than the other villages. Further, there are important differences in that it is a place where non-agricultural activities are of great importance, notably commerce and music playing -it hosts a large regional music school- and as a consequence agriculture often receives less attention than elsewhere. For these reasons, the households in 28 For a fuller description of the economic anatomy of the households in the three regions see Cotlear(1986) chapter 4. 22 Table 2 Means and Standard Deviations of Variables ___RI___ R2 ---R3 Mean s. d. Mon s. d. Mean sd 1. Output (Kg.):* 12183 21317 3072 2792 1708 1520 2. Area cultivated 12,618 172031 7019 4865 5199 4452 (metres)** 3. Labour input 159 215 86 62 99 73 (person days)** 4. Aniul I tracter 16 48 12 7 1.0 5.8 (leqiv. animal days) 5. Village (Dumy 0.22 0.42 - - - for Acalla) 6. School attainment 6.2 2.7 4.32 2.81 3.7 3.0 (yeats)t 7. 1-5 yr. of schoolt 0.33 0.47 0.36 0.48 0.41 0,49 8. 4-5 yr. of schools 0.19 0.39 0.25 0.43 0.15 0.36 9. .6+ yr. of schools 0.65 0.48 0.43 0O50 0.33 0.47 10. 4-5 yr. of rural 0.08 0.26 0.09 0.28 0,15 0.36 schoolingt 11. 6t yr. of rural 0.29 0.45 0,08 0.27 0,21 0.41 schoolings 12, 4-5 yr. of urban 0.11 0.31 0.16 0.37 0 0 schooling: 13. 6+ yr. of urban 0.37 0.49 0.35 0.48 0.12 0.33 schoolilgS 14. Recent contact 0. 10 0,30 0.29 0.45 0.07 0.25 with ext. Ag, (O,1): 15 Old contact vith 0.24 0.43 0.30 0.46 0.16 0.37 ext. Pg. (0,1): 16. Migration experience 2.90 4382 2.13 4.12 1,91 3,41 (years): 17. Migration experience 0.13 0.33 0.16 0.37 0.10 0.30 inc. agriculture (O,l)t 18. Aget 43.8 13.5 43,4 14.9 47,3 12.9 19. Households in village 0.10 0.12 0.28 0.14 0,06 0.04 with recent ext. contact (proportion) 20. Households in 0.24 0.18 0.37 0.05 0.14 0.04 village vith old ext. contact (proportion) 21. Households in village 0,70 0.17 0.09 0,06 0.01 0.01 vith 85X of potato land under cultivation vith HYvs 22. Uses HYV (0,1)*: 0.92 0.27 0.36 0.48 0.03 0.i9 23. Received formal 0,24 0.43 0.50 0.50 0.22 0.42 credit (0,1) Humber of observations 254 151 150 t Referred to household head; 00 Referred to potato production. 23 Acolla may have a lower agricultural productivity than those of the other villages. We include a dummy variable for Acolla, which includes 22 per cent of our sample of Rl. The educational indicators of main interest for policy purposes are formal educational attainment and exposure to non-formal education (extension and practical course-training). However, informal education indicators will also be included in the analysis because they can sometimes be influenced indirectly by government policy, and because it is possible that their exclusion from the analysis may produce misleading results concerning the effects of the other types of-education. School attainment shows the same gradient as productivity in the sample, being largest 'in Rl and lowest in R3. The heads of households studied on average 6.2 years in RI, 4.3 in R2 and 3.7 in R3. One of the hypotheses discussed above states that a minimum level of education is required for certain purposes. This effect is easier to *test if schooling is measured by discrete rather than by continuous variables. Hence, school attainment will be divided into three categories and handled with the use of two dummy variables.29 The three categories of classification chosen are: (i) Never went to school or studied for up to three years (the omitted variable),(ii) studied more than three but less than six years, and (iii) studied six years or more. This breakdown was chosen to reflect the institutional structure of the educational system in Peru. In rural villages where a school is available, the minimal educational supply usually found extends to the fourth grade: we have chosen completion of this basic cycle as the first dividing line. Complete primary education in Peru consists of six years of schooling: this has been taken as the second dividing line. These dividing lines where also found satisfactory to examine descriptive data on labour- and land- productivity: in at least one of our regions, we found the level of these variables to remain in narrow ranges within particular 29 In the initial stage of the analysis, we also experimented with the use of a continuous variable for schooling. The t coefficients for this variable were found to be generally lower than for the dummy variables. 24 educational categories, and to change noticeably between categories. Below we discuss the results obtained when the omitted category is redefined to include only household heads with no schooling. The percentage of farmers who had not completed the basic cycle of education was 16, 32 and 52 in RI R2 and R3 respectively. At the other extreme of the education cycle, the proportions who completed primary school where 65, 43 and 33 percent. We argued above that if non-tormal education is concerned with the transmission of specific information about technologies, and market structures, the value of this type of education is likely to diminish with time as the specific technologies or market situations grow obsolete. Also, the effects of non-formal education as measured by the differential between households will tend to diminish as other households learn the new techniques by imitating the innovators. The above suggests that recent contacts should be more important determinants of productivity differentials than old contacts, as the households who receive the more recent bits of relevant information take advantage of them by increasing their productivity. We use two dummy variables to measure the impact of non-formal education and test this hypothesis. The two variables show whether there has been direct contact in the last three years, only before that period, or not at all (variables 10 and 11).30 There are large differences in the proportion of households with extension contact in the samples of the three regions. Ten percent of the households sampled in RI had had recent contact. This proportion is 29 percent in R2 and 7 per cent in R3. The respective proportion for old contacts are 24, 30 and 16. We will also test the existence of indirect effects of extension through imitation. This is done with the use of special variables which assign to each household the village-wide proportion of households who have been contacted by extensionists. Two such variables are considered: one measures the indirect effect of recent 30 Notice that these variables do not overlap. In the case of households with recent and old extension contacts, only the former has been considered. 25 extension (less than three years) and the other measures indirect effects of previous extension.31 A similar variable is used to measure the impact of imitation in the use of HYVs by "technological leaders" who are defined as farmers who cultivate most of their potato land (85%) with HYVs. - Informal education occurs in many ways. In this analysis we will include two. The first is the farmer's work experience. The indicator for this will be the age of the head of household, and the expectation is that greater age, indicating more work experience, will increase productive efficiency. Table 2 shows that the average age of farmers is similar in the three regions, namely 43 years i-n R1 and R2 and 47 in R3. The second form of informal education that seems relevant in the Peruvian context is migration experience. Large numbers of people leave the villages for several years to enter external labour markets and often to study in contexts different from that of the village. Experience of this sort is likely to affect attitudes and develop certain skills. Some of these may be general skills which may increase productive efficiency in agriculture e.g. any urban job will give the farmer the training in the Spanish language which will facilitate buying inputs, obtaining credits or bargaining at the market place; a former street vendor will have improved his numeracy by dealing with money and this may facilitate his calculations on the use and mix of chemical inputs. Other skills may be more specific such as the acquisition of technical information about the use of new agricultural inputs obtained by doing agricultural tasks away from the village. Two variables will be used to measure the effects of migration. One is the number of years spent away from the village by the head of the household. The averages for this variable (including the heads who had never migrated) show that this experience is slightly more common in RI (2.9 years) than in R2 (2.1 years), and that it is Lowest in R3 (1.9 years). The second variable is a dummy with a value of I for households where the farmer had migration experience which included work in agriculture. 13, 16 31 Similar variables were used by Jamison and Moock(1984). 26 and 10 percent of households had this experience in Rl,R2 and R3 respectively. Finally, the other variable to be included in the equation is an indicator of whether the household received credit from a formal institution in the agricultural year of the survey. In our sample, the proportion of households with credit is shown in table 2 to be 24, 57 and 22 percent. As was explained before, the high proportion of households which received credit in R2 is due to the presence of an integrated rural development project in that region. 5. Methods for the Estimation of the Determinants of Productivity A useful way of approaching the analysis of productivity is by the use of production functions. The amount of output obtained by a farm from a given bundle of physical inputs depends on the "technical efficiency" and the "technological level" of the farm. These concepts can be defined with the use of a one-input one-output production function illustrated in figure 1. The production function is of the form Y=F(X), where Y is the level of output, X the level cf input, and F describes the technology used to transform inputs into outputs. We assume diminishing returns to the single input factor which cause the incremental gain in Y for a fixed increase in X to decrease as X grows larger. - ~ ~~ _ ' e F(Xe) x FIGURE I PRODUCTION FUNCTIONS 27 If the farm is technically efficient,it will produce at a point on the production function like a, b or c. The points above the production function like e cannot be obtained Qith the technological level depicted by F. Points below the production function like d are obtainable, but are inefficient because with the technology described by F the farm could obtain more output for the same amount of input. Efficient farmers operate in the production function which is for them their production frontier. Inefficient ones operate within the farm's production frontier located below the production function (point d in figure 1 is a point on an inefficient farm's production frontier); The technical efficiency of a farm has been defined in terms of the range of options opened to it (shown by F above). Some farms will however be better able to produce because they have access to higher technological levels. These are shown in figure 1 where H represents a technology capable of producing a larger amount of output for the same level of input compared to F, i.e. H is said to be a production function with a higher technological level than F. In the absence of information on the precise shape of the production functions such as would theoretically be obtained from agronomic research, it is difficult to disentangle "technical efficiency" from "technological level" since the observed effect of a higher output for the same level of input could be due to either. The combination of these two effects will be termed "productive efficiency". Presumably a farm that has higher productive efficiency than another possesses more information about the production process. For example, a more efficient farmer possesses better information on the spacing of plants, the handling of chemical pesticides, the mixing of fertilizers, the timing of all inputs -on all of the techniques of production not reflected in the variables used to quantify physical inputs. 28 Differences in productivity imply that different outputs will be obtained from a given bundle of physical inputs. If we are interested in examining whether certain characteristics of the household, like its level of education, have an effect on productivity, we can do this by means of production function analysis. We can specify the production function so that it includes on the right hand side, in addition to measured physical inputs, measures of the characteristics which we presume to be linked with the use of more productive techniques. Thus the production function reads Y=f(X,E) where Y is the quantity of output, X is a vector of physical inputs, and E is a vector of variables that characterize a particular farm. The parameters of f apply to all farms in the sample. The regression equation is an estimate of the "average" production frontier for the sample. This procedure provides us with a test to explore whether certain characteristics which may be of particular interest are related to productivity. It will be used to assess whether formal, non-formal and informal education allow peasants to obtain a larger product from a given bundle of physical inputs. The comparison of our regions will give some indication as to whether the results can be presumed to be general, or if they pertain to particular contexts. The functional form for the production functions estimated is a modified version of the Cobb-Douglas production function. There is no theoretical basis to prefer this or any other form of estimation. The choice is arbitrary and we have preferred this form for ease of computational manipulation and because it produces coefficients which are easy to interpret. An additional advantage is that most previous studies have used similar forms. Our choice would thus allow an interested reader to make direct comparisons of our results with those of previous studies (for instance with those reported in Jamison and Lau(1982)). in linear form the equation estimated can be represented as: ln Y = ln a. + Bi ln Xi + I yi Ei 29 Let Y represent level of output, X a vector of physical inputs, and E a vector of variables that characterize a particular household. a is an efficiency parameter, and Si the elasticity of output with respect to physical input i. The interpretation of the parameter yi depends on whether Ei is a continuous variable or a dummy variable. In the 'first case Yi, can be interpreted as indicating an approximaticn to the proportionate change in output that results when characteristic Ei (say migration experience) increases by one unit (a year). In the case Ei is a 0-1 indicator variable, then Yi indicates the approximate proportionate difference between the output produced by a household demonstrating this particular characteristic (say, having had recent contact with the extension services) and one produced by a household not demonstrating this characteristic. 32 6. Productivity: Basic Results Table 3 shows the least-square regression results. For each region the variables are entered in four steps, adding at each additional step new variables to the equation. The physical inputs, schooling, extension and experience variables are entered first. The migration variables are added in a second step to allow us to examine whether the common neglect of these variables is likely to affect the coefficients of formal and non-formal education. At a third step, the variables indicating use of credit and "technological level" are 32 This is only an approximation. The proportionate change in output that results when Ei increases by one unit from Ei to Ei +1 is actually the following, where Yo is output at Eo and Y1 output at EO+l. (YL -YO)/YO = (Yj/YO) - 1 = (eY'(EI + l)/eyiEi) -1 = eyi This value is approximated by yi when Yi is a small fraction. (The larger the absglute value of Yi, the poorer it is as an approximation of eYl 1. This interpretation of the coefficient was developed by Jamison and Moock(1984) p.73. Table 3 Production Function Estimates _ __ Rl____ __ ___RR2 R3 1 2 3 4 1 2 3 4 1 2 3 4 Constant .26 .25 .26 .24 .74 1.00 1. 22t 2.76 .88 .84 .77 .32 (0.56) (0.53) (0.50) (0.44) (1.35) (1.69) (2.34) (1.03) (1.63) (1.55) (1.37) (0,56) 2. Area cultivated .77s8 .778t .7588 .768t .56tt .52*t .49*8 .488t .58*s .59*u .60*8 .60s (in mettes) (10.88) (10.78) (10.32) (10.3) (5.22) (5.02) (4.83) (4.80) (7.95) (7.99) (7.95) (8.04) 3. Labaur input .35tt .35tt .32*t 32*8 .49*t .52u .49*8 .51*8 .34t* .34t* .34t* .32t* (in persoo dais) (5.27) (5.22) (4.79) (4.86) (4.37) (4.69) (4.57) (4.72) (4.85) (4.65) (4.66) (4.35) 4. Animal and tractor .02 .02 .02t .02 .01 .01 .01 .02 .01 .01 .01 .01 (is equivalent animal days). (1.83) (1.85) (2.00) (1.91) (0.28) (0.21) (0.34) (0,49) (0.75) (0.86) (0.85) (0.90) 5. Uillage (Acolla=l, -.21* -.21s -.19s -.06t etherwO) (-2.23) (-2.23) (-1.99) (-0.54) - - - - - - 8. 4-5 yr. of schooling (0,1) .14 .15 .15 .11 .13 .12 .13 .17 .13 .14 .14 .15 (1.27) (1.28) (1.27) (0.97) (1.35) (1.23) (1.62) (1.86) (1.15) (1.21) (1.22) (1.32) 9. 6 or more yr. of .37tt .36*8 .358* .31tt .24* .16 .16 .18 .05 .07 .08 .08 schoolirtg (0,1) (3.52) (3.45) (3.28) (2.87) (2.48) (1.68) (1.65) (1.89) (0.56) (0.70) (0.80) (0.72) 18. Age (years) -0,18 -.Olt -0.1*t -O.Itt -.004 -.01* -.004 -.003 -.00 -.00 -.00 -.00 (-2.81) (-2.81) (-2.74) (-2.90) (-1.66) (-2.17) (-1.60) (-1.17) (-0.16) (-0.09) (-0.02) (-0.16) 14. Recent contact with .09 .10 .08 .07 .29t* .33tt .27*t .27t* .11 .13 .13 .15 pxt. Agent (01I) (0.84) 10.85) (0.75) (0.58) (3.42) (3.93) (3.25) (3.14) (0.66) (0.80) (0.76) (0.89) 15. Old contact with -.05 -.05 -.05 -.10 -.00 .01 -.01 .01 .04 .04 .04 .04 ext. Agent (0,1) (0.60) (-0.62) (-0.54) (-1.15) (-0.05) (0.15) -(0,18) (0.16) (0.37) (0.3)3) (0.40) (0.39) 16. figration experience .00 .00 .00 .03*t .034s .03*t -.01 -.01 -.01 . (years) (0.37) (0.53) (0.62) .(3.36) (3.06) (3.11) (-0.58) (-0.47) -(0.50) 17. fig. exp. included agric. -.02 -.02 -.01 -.22t -.21* -.240 .13 .11 .14 activity (0,1) (-0.18) (-0.19) (-0.05) (-2.21) (2.14) (-2.35) (1.00) (0.86) (1.05) (1M (Table 3: Continuation) -_ __ __ Rl _ __ ____ R2 R3 1 2 3 4 1 2 3 4 1 2 3 4 23. Received foreal credit 11 .09 .08 .11 -.06 -.06 (0,1) (1.23) (0.99) (1,20) (1.51) (-0.58) (-0.54) 22. USES HYUs (0,1) .26t .288 .20*t .18t .07 .07 (1.99) (2.16) (2.68) (2.44) (0.32) (0.30) 19. Households in village .30 -1.00 3.19* with recent extension (0.86) (-0.59) (1.98) contact (proportion) 20. Households in village .71t -3.65 2.89*t with old extension (2.36) (-0.73) (2.64) contact (pruportios) 21. Households in villoge -.32 .17 2.33 with 85SX of potato land (-1.19) (0.10) (0.21) under cultivation with HYirs R squared 0.853 0.853 0.857 0.860 0.782 0.800 0.813 0.821 0.717 0.719 0.719 0.736 Adjusted R square 0.848 0.847 0.849 0.851 0.769 0.786 0.797 0.801 0.701 0.699 0.695 0.707 Note: Numbers in parentheses are t values. *t coefficient significant at 0.01 level in two tail test. * coefficient significant at 0.05 level in two tail test. 32 introduced, in order to discuss some of the direct and indirect effects of education on productive efficiency. We can thus investigate whether the main effects of the different types of education on productivity occur indirectly, through the greater access to credit and the greater propensity to adopt, or whether education has direct effects on productivity when use of credit and technological level are controlled for. Education influences the adoption of modern technology and facilitates access to credit, hence we can expect that the inclusion of these variables in the third step will rob education of some of its indirect effects on output. This implies that equation two can be regarded as a reduced form equation, combining direct and indirect effects of education, and the coefficients estimated at that stage may be preferred for some effects. Separating the direct and indirect effects, however, can give us important insights into the processes through which education and the supply of new inputs and of credit interact and lead to output growth. Finally, in the fourth step we include in the equation the variables measuring indirect effects of extension and learning the use of new technologies by imitation. a.Physical inputs We look first at the coefficients for the physical inputs. The coefficients for land and labour are positive and highly significant in all regions. The coefficient for land is very stable across the different specifications of the production function in all regions, the greatest fluctuations being in R2 where the range is .56 to .48. The coefficient for labour is also stable in all regions. The coefficient for animal and tractor inputs is positive and stable in all three regions, but it is significantly different from zero at the .05 level only in Rl. The elasticities for the physical inputs (from equation 3), add up to 1.09 in Rl, 0.99 in R2 and 0.95 in R3, implying that 33 essentially constant returns to scale exist in the production of potatoes for the three regions. In Rl, the village dummy for Acolla is, as was expected, negative. The coefficient is significant in equations 1-3, but loses significance and falls in absolute value, when the village-wide variables 15, 16 and 17 are introduced. b.Formal Education We turn now to the educational variables, which constitute the central interest of this paper. The estimated effects of these variables are summarized in table 4. We look first at the impact of formal education (variables 8 and 9) on output. Both variables are positive for all regions. However the estimated effects of. schooling are seen to be larger in the more modern region, and to weaken as the region becomes more traditional. It is only in RI where one finds a schooling variable significantly different from zero at the 0.01 level. For this region the regression results suggest that the completion of primary school has a strong positive effect on agricultural productivity, increasing potato output by approximately one third. In this region, incomplete primary education seems to have no effect on output. In R2, equation I suggests that complete primary education is important, but this variable loses significance when migration is included in the production function (more about this below). Equations 3 and 4 show that complete and incomplete primary schooling have t values which closely approach levels conventionally considered satisfactory for statistical significance.33 In this region the effects of complete or incomplete primary education on output are 33 Both coefficients are different from zero at the 0.06 level in a two tail test, or at the 0.03 level in a (possibly more adequate) one tail test. Table 4 Summary_of_Production Function Regression Results _____RI ____R2 ________R3____ ..... ~~~~~~~~... ..... . . . ._. .. _.... 1 2 3 4 1 2 3 4 1 2 3 4 Frio Table 3 4-5 years of ochooliog 0 0 0 0 0 0 + + 0 0 0 0 6+ years of schooling +++ +.+ .+. +tt ++ + + + 0 0 0 0 Receat extensios 0 0 0 0 .++ .+t +4+ +t+ 0 0 0 0 Old extession 0 0 0 0 0 0 0 0 0 0 0 0 Age - -- - 0 0 0 0 0 Migration eXpetieoce 0 0 0 .+t+ +.. ++ 0 0 0 Migration to agricultutre 0 0 0 -- -- -- 0 0 0 Indirect recert exteisioo 0 0 ++ Indirect old extension ++ 0 ++ Received formal credit 0 0 0 0 0 0 Uses HYU ++ 4+ .++ ++ 0 0 From Table 5 1-5 years of schooling 0 0 0 0 ++ ++ ++ +4 6+ years of schooling 0 0 0 0 + + + + From Table 6 4-5 years of rural 0 0 0 0 0 0 schooling 6+ Yeatrs of rire + + 0 0 0 0 schooling 4-5 years of arbas + + .+. +++ a a schoeling 6+ yeats of urban .++ +++ . +.. ++. 0 0 schoolirg Migration esperienice 0 +.+ 0 Netation: + significant at .10 level; ++ significant at .05 level; ++ sitgnificant at .01 level. a. there are no observations in this category. 35 similar, and they imply an increase of about 15 percent in output as compared to farmers with less than 4 years of education. Table 3 shows that in R3 there is no significant difference between different education levels. However, since we have seen the threshold of "impacting schooling" fall as we move from a modern into a more traditional region, could it be the case that as we move into a still more traditional region the threshold would fall even further? To test this hypothesis we ran the same regressions shown in table 3 replacing the variable used to measure incomplete primary education. In the original equations "incomplete primary" begins with 4 years of schooling. We redefined the variable to start with one year of schooling.34 Unfortunately, by doing this we were left with scarcely any observations in the base group in RI, and it was not possible to run the new regressions for this region. The results for regions 2 and 3 are shown in table 5 and the estimated effects of the education variables are summarized in the second block of -table 4. Table 5 shows that in R3 schooling does have an effect, but only when one compares peasants with schooling with those with no schooling at all. The effect of completing primary education does not add to that of having "some education". Notice in contrast that in R2, with the new specification there is not only no effect of incomplete primary education, but that the effect of complete primary is much weaker than that found with the previous specifiqation. This suggests that in R2 "a couple" of years of schooling does not help and may even harm output.35 34 It should be noticed that by doing this we redefine the "base group" with respect to whom the impact of education is being calculated, the original base group includes all farmers with less than 4 years of schooling, the new base group includes only those with no schooling at all. 35 A similar pattern was found in a study of Kenyan farmers by Moock(1981) who argues that "...an a posteriori justification would need to show that those who begin school but drop out before earning even the first credential are dull from the start, or that they become demoralized or, conversely, that they develop a self- importance that blinds them to their technical incompetence."(p.732). 36 Table 5 Production Function Coefficients for Formal Education using Farmers_with__zero_years of schooling as the base Group (t) _ ._ _ R _ _ - - - _ I 2 3 4 1 2 3 4 1-5 Jr. of -0,3 -0.3 -.00 00 .24t .24* 25* .24s schooling (0,1) (-0.29) (-0.28) (-0.04) (0.91) (2.39) (2.37) (2.45) (2.40) 6+ y1. of .15 .07 .08 .09 .20 .21 .24 .2t schooling (0,1) (1.31) (0.69) (0.71) (0.81) (1.66) (1.72) (1.87) (1.72) R squa?e 0.78 0.80 0.81 0.82 0.73 0.72 0.73 0.74 Adjusted R square 0.77 0.78 0.79 0.80 0.71 0.71 0.70 0.72 Note: Numbers in pa?enthebis are t values. Equations 1-4 include all vdriables included in equations 1-4 respectively in table 3.tt: coefficient significant at 0.01 level in two tail test; t: coefficient significant at 0.05 leYel in two tail test. 1. We cotuld not run the regressions for RI because we have only four observations vhere the head of household had no education. 37 In summary, the analysis seems to bear out two patterns with respect to formal education. The first is that there seems to be an effect associated with formal education in the production of potatoes for all regions in our sample, but the larger and stronger effect occurs in the modern region. This finding is fully consistent with what was suggested by Schultz. Schooling has a greater impact where conditions are dynamic and there is a greater need for economic adjustment. In traditional regions, where the economic and technological environment changes slowly, there are less tasks for education to fulfill, and schooling has less effects on productivity. None of our regions is completely stagnant, however, and that is probably what accounts for the positive effect found in all regions. The second pattern is a related one. There seems to exist in the three regions a level of schooling which represents a threshold at which formal education begins to have an effect. However, the level of this threshold differs across regions, increasing with the complexity-of the technology found in each region. It is lowest in R3 where "some" schooling has an effect, increases in R2, where 4 years are required to obtain an effect, and is largest in Rl, where only complete primary shows an effect. Adoption of the simplest modern inputs is only starting in R3, and it is in Rl where one finds more modern inputs and practices being used and in more complex combinations. There is good reason to expect the educational threshold to be different where the tasks imposed upon education are different.36 The pattern found in the Peruvian highlands suggests that there is a threshold regarding the amount of schooling needed to have an impact on productivity, and that the level of this threshold depends on the degree of modernity, i.e. the complexity of the technological and market problems which need to be solved. The 36 The existence of these thresholds has been noted before in the literature, e.g. Chaudhri(1979) found a threshold for farmer's education. The existence of an educational threshold to obtain preferential treatment for skilled jobs in non-agricultural rural labour markets has also been documented in India by Heyer (1985)p.13. However, I have not found any attempt to explain why a particular number of years should have that effect in the specific context under study. 38 greater their complexity, the larger the minimum level of education required to obtain an impact on productivity. This suggests that while the generalization of literacy in the traditional areas may act as a catalyst speeding traditional regions into the first stages of technical change, further technological development will require higher levels of formal education. c. Non-formal education: Direct effects We argued above that if extension services transmit specific information about technologies or market structures, the impact of extension contacts -as measured by a productivity differential- is likely to diminish with the passing of time as the specific technologies become obsolete, or the differential is lost by other households imitating those who had the direct contact. If this is true we may expect recent contacts to show a greater effect than old contacts. It can be seen in table 3 that that is what the coefficients for the extension variables suggest. The table shows, however, that the direct effects of extension are only significantly different from zero in R2. In this region, other things being equal, output is increased by a quarter to a third in those farms which received extension contacts. Many factors may help explain why non-formal education has an impact in some regions and not in others. No doubt an important one concerns the quality of the extension service (as an institution), and the quality and commitment of the individual extensionists. It is likely that PRODERM, the institution giving most of the extension in R2, being a non-state, experimental, better paying institution would get higher marks on these two counts. Also, the sort of additional services provide( (e.g. provision of inputs, commercialization facilities) may be crucial, and PRODERM provided a wider range of these than the other institutions in the regions of the study. Finally, it is posible that in RI there exist many sources 39 of information about the modern inputs, so that the impact of extension becomes very small. Farmers may still be willing to receive contacts to confirm information obtained by other means, or may prefer it to other means because it is cheaper, but it may have ceased to be the only source to obtain information. It should be noted that if this was effectively the case, it would not imply that extension is irrelevant for modern regions. The implication would be that to be relevant it will have to be accompanied by a constant flow of new technologies. The crucial difference between regions may reside in an even more essential question. The main point of extension is the transmission of specific information. The impact of extension will thus depend on the potential value of the information which is being transmitted in the context in which extension is occurring. It is possible that the message transmitted by the extensionists in R2 is the only one which is adequate for the technological stage each region is going through. We have hypothesized elsewhere that there are four differentiated stages of technological development in the Andes.37 The first stage is traditional agriculture. The second consists of the introduction of chemical innovations, the third is the addition of biological innovations to the technological package, and the fourth corresponds to fine-tuning of the techniques in use. The introduction of chemical fertilizers at the second stage initially has its main impact in allowing an increase in the cultivation intensity (i.e. the amount of land left fallow is reduced). Productivity of the total farm-land (including fallow) is increased, but yields of the cultivated area do not change in an important way. Entering this stage is likely to be slow for two reasons. The first is that this is the first step away from traditional agriculture, hence everything is totally new and uncertain. An important part of what has to be learned is how to learn about changing. If traditional agriculture is merely 37 Cotlear (1986), chapter Vt. 4 0 repetitive, this will be the first time that people are faced with experimentation and having to learn from its results.38 Further, the benefits to be gained from fertilizing accrue through several channels (e.g. a reduction in fallowing, an increase in the yields of fertilized products and an increase in the yields of products which absorb the fertilizer residual) and often with long lags between application of the input and benefiting from the output. This makes it much harder to assess the benefits of the input and establish causation between inputs and outputs than in the case of inputs which have a strong short-term impact on yields. The second reason for the slow pace in the introduction of the second stage is that in many places there exist institutional constraints on the change in the intensity of cultivation, such as the communal restriction on cultivating certain areas in some years.39 There is little tnat extension can do to cope with these problems. The third stage of technological adoption is the introduction of HYVs. The main advantage of HYVs is their higher average yields. They are usually regarded as more risky than traditional varieties, and they are regarded by peasants as gastronomically inferior to the traditional varieties. The new varieties of seed are fertilizer- intensive. Higher yields will be obtained but only when fertilizer is applied. Once fertilizer is widely in use in a particular region the introduction of HYVs appropriate to the weather and soil characteristics of the region is likely to have a strong, and relatively quick effect on yields per unit of cultivated land. An extension organization guiding the peasants into the choice and techniques of utilization of the new varieties is likely to increase the rhythm of adoption of the new varieties, and to improve upon the techniques applied to their use, hence allowing a more marked 38 Some anthropologists believe that even in "traditional agriculture" there is experimentation (e.g.Rhoades 1978). This is held for things like regular selection of new native seed from other areas to repla-e local seed which after a while may risk genetic degeneration. This sort of "experimentation" is also likely to be repetitive in the sense that it will consist of "set tests", and set forms of assessment. Experience on this is unlikely to be of great help to model new forms of experimentation, but still a lot of new things will have to be learned. 39 This is described in detail in Cotlear(1986), Chapter II. 41 increase in productivity than would otherwise be the case. New varieties require new techniques of production in many ways, but their introduction does not represent a "new concept' in the way fertilizer did in the second stage, and so the process of learning their use is not as difficult. Also, expectations about their potential benefits tend to be higher, and more certain than in the fertilizer stage because (a) farmers have previous experience with innovations and (b) the causality between HYVs and higher yields is easier to prove in short experiments than longer term relations such as that between fertilizer-use, intensity of cultivation and long- term soil erosion. The phases sketched above can be identified with the regions of -the sample. R3 is entering phase 2, R2 is entering phase 3, and Rl is in phase 4. It is easier for extension to have an effect on phase 3, and it is only in the region entering that phase (R2) where one finds that extension is statisticalLy associated with a positive effect on output. The weak effect of extension in R3 type regions also seems to be due to the message transmitted by the extension services. Possibly impressed by the welcome HYVs receive in regions of R2 type, they tend to push for the introduction of the whole package in the traditional regions. This seems to be too complicated and is thus greeted with very slow results. The problem in Rl regions seems to be the non-existence of "superior technologies" in the message the extension services have to offer. There are no firm results of research about new superior inputs for regions in the stage of RI. New productivity increases arise from a more efficient use of the modern inputs already adopted, the extension organizations do not have clearly superior recommendations on this direction, the impressions from the field work were rat,er that the extensionists are in the process of learning from the most efficient farmers, and transmitting this to the Less efficient farmers. 42 d. Non-formal education: Indirect effects In RI, recent extension shows no indirect effects, and only the coefficient for the indirect effects of old extension is significantly different from zero. In the case of Rl,the regression coefficient for indirect effects of old extension is 0.71. The interpretation of this variable is that a 10 percent increase in "old" extension coverage in the farmer's village is associated with a 7 per cent increase in the farmers potato output, other factors being the same, including the farmers own direct extension exposure. The implication here would seem to be that a farmer can acquire technical information relevant to the production of potatoes indirectly from other farmers who have been in direct contact with extension agents, but that this indirect transmission occurs with a certain lag. In R2 both variables measuring indirect effects of extension are not significantly different from zero. In R3 they are both positive and significant. The value of the coefficients there seems to suggest that 10 percent increases in extension coverage could have indirect effects of increasing output by about 30 percent, both with and without a lag. This seems to me to be too large a value to be reflecting the real impact of extension, especially since the effect of extension on productivity should occur through a wider use of modern technologies. That is what extensionists in those regions push for and, that does not seem to be explaining a significant part of the variance in productivity in that region. We must conclude that at least for this region this variable is measuring something different from what we originally tried to measure. Possibly this is due to the fact that there are only 4 villages in R3 (5 in R2 and 9 in RI) and hence these variables could be correlated with many other characteristics of the villages. Also, it is possible that extensionists choose to wo.k in villages which already have high productivity levels. For this reason we will exclude variables 15 and 16 from further analysis. Variable 17 (proportion of households with more than 85 per cent of their potato land planted with HYVs), is 43 insignificant in all regions and suffers from the defect discussed above and so will also be excluded from all further analysis. e. Informal education Three variables are in use to indicate the impact of informal education on output. The first is a conventionally used measure of on-the job experience approximated by age. The other two variables measure the impact of migration experience on output. Age, being a proxy for experience, is expected to have a positive impact on output. Table 3, however, shows that the coefficient has a negative sign in all equations for the three regions. One can also observe that the coefficient is insignificant in R3, it increases in absolute value in R2 (and is significant in this region in equation 2) and is most negative and significant in Rl, i.e. the negative effect of age on productivity increases with "modernity". This is probably reflecting a "cohort effect", by which the old tend to stick to the older less productive techniques they started off with. "Age" is then acting here as a proxy for the "vintage" of the techniques in use. Since "old" techniques are obsolete in Rl where technology has been going through a stage of change, but have not been much improved upon in R3 where agricultural technology has been comparatively less active, we find "age" to be significant in the modern region but not in the traditional region. The reluctance to change among the old is likely to reflect their greater risk-averseness, true even when formal education is taken into account. An alternative explanation would be to suggest that age has a negative effect because farmers become weaker for physical work as they become older. This effect is unlikely to be of importance, because the farmer's own labour constitutes only a fraction of the total labour involved, and this fraction is likely to become smaller as the syblins grow up, and later as more wage-workers are hired. 4 4 We suggested above that a migration experience could provide the peasant with general skills which may influence productivity. if migration had included work in agricultural activities this could have developed specific skills. There is an an even greater expectation that this sort of experience will have a positive effect on productivity. Table 3 shows that in R2 both variables are significant. The coefficient for the number of years of migration is 0.03, suggesting that, other things being equal, output will increase by about 3 per cent for each year the peasant has spent away from the village. The mean number of years spent away by returnees is 4.8, hence the coefficient suggests that a returnee with average experience will have an output almost 15 per cent larger than non-returnees. Surprisingly, the dummy for migration experience in agriculture has a negative sign. The coefficient for this variable shows the additional effect that an agricultural experience has on top of the migration experience variable. The relative values of the coefficients of variables 12 and 13 suggests that for long periods of migration there will be a slightly positive effect, but for short periods there may be a net negative effect. The negative coefficient is probably not due to a negative effect of learning on productivity (I found no evidence that people were unsuccessfully trying to apply inappropriate agricultural techniques learned away from the village). Instead, it probably reflects a process of self-selection: Wages obtainable by migrating to agricultural activities are lower than those obtainable in the more selective urban labour markets, and so migration to external agricultural labour markets constitutes a process which may self-select peasants who have lower productivity to start with. This finding is also a reminder that the positive coefficient of migration must be taken with some care because it may partly be reflecting some above average abilities which led to migration in the first place. 45 Notice that in R2, when the variables for migration experience are introduced, the coefficients of the schooling variables are reduced. Particularly, the coefficient pointing to the effects of complete primary loses a third of its value and the associated t coefficient also loses its value. The same effect occurs in Rl but in a much slighter way. Many of the peasants studied while in migration, and the most educated are more prone to migration, hence there is a correlation between schooling and migration experience. However, in R2 when both variables are introduced in the equation, only migration experience remains significant. This suggests that it is the migration experience, and not schooling, that is relevant for the differences in output. One may also hypothesize that both variables are simply acting as a proxy for a variable of urban schooling which, having higher quality, has a greater effect than rural schooling. This hypothesis will be explored below. It is important to notice that the exclusion of relevant categories of informal education (as is usually done in the literature) may, as in the case of our sample, produce estimates on the effects of formal education which overestimate its true effect. f.Technoloqy and credit The effects of education on output are obtained partly by increasing the efficiency in the use of a particular technology, but they also occur indirectly by elevating the technological level utilized in the farm. Two indirect ways in which this may occur are through the selection of more productive technology and through allowing greater access to credit which in turn allows the adoption of modern inputs. In this section we discuss the effects on the regression coefficients of explicitly including variables which control for adoption of mode.n technology and use of credit. The use of high yielding varieties (variable 17) is significant in Rl and R2. In both regions it shows that the shift to this new 46 technology can have substantial effects on output. The very few households who are already using HYVs in R3 seem to be still in the phase of learning the use of this new technology since in this regicn the variable for HYVs is positive but not significant. Use of formal credit does not have a coefficient different from zero in any of the regions. The main role of credit is to give access to inputs, and we would expect it to be a major determinant, say, of the adoption of HYVs. The effects of credit use on adoption will be seen below to be positive and important. Once the inputs are utilized in production, however, it is these and not credit which play a role in increasing productivity, and in this sense the results obtained are plausible. When these variables are introduced, controlling for some of the "indirect effects" of education, we notice that in RI the coefficient on formal education falls by one point, and that in R2 the coefficient on non formal education (variable 10) falls by six points. This is consistent with the argument that part of the impact of education on productivity can be traced to the indirect effects it has through gaining better access to the credit markets, and through adoption of modern technology. The fact that the coefficients of education remain positive and significant once these effects are controlled for, suggests that the direct effects are themselves important. g. Quality of formal education It was shown above that the amount of formal education received by peasants has an effect on their productivity. Visits to the schools in the area where the regions are located suggested the existence of wide difference in the quality of educatio-. These differences seemed to be especially marked between urban and rural schools, where the gap in quality of teachers, equipment and material available for students and the quality of infrastructure seems to be 47 so wide that one is forced to ask whether the effects of the same amount of such different inputs can be the same.40 In this section we explore the effects of formal education in rural and urban schools on productivity. The effect of studying in the two settings can differ for two reasons:(a) quality is different in the two settings and (b) the urban experience in itself provides skills independent of schooling, which have an effect on productivity. In the analysis below we will try to separate these two effects. The hypothesis is that quality does matter. We test it by running again regressions 1 and 2 shown before in table 3, replacing the two formal education dummies by four dummies: two for urban schooling and two for rural schooling. - If quality matters, then the impact of the same number of years of schooling will be larger for urban than for rural schooling. The results are shown in table 6. In RI of the four formal education variables shown in regressions 5 and 6, only complete urban primary is significantly different from zero. The coefficient for complete rural primary is lower than the coefficient for incomplete urban primary in both magnitude and significance. The coefficient for incomplete rural primary has a negative sign. In this region, migration experience has no impact, and its introduction does not have an effect on the schooling coefficients. In R2 the effect is even more clear-cut. The coefficients of both rural schooling variables have negative signs, but are not significant. Both coefficients for urban schooling are positive and highly significant. As was found before when discussing threshold 40 Arriagada(1983), using data from an ECIEL study by Rivera(l979),examined the effects of school quality characteristics in urban and rural areas in reading and science achievement. She showed that achievement in both of these areas was higher in urban schools. However, when one controls for quality indicators of the schools involved, the urban/rural location of the schools loses significance in regression analysis. This suggests it is the quality and not the location of the schools (or other variables different from quality but correlated with location)that affects achievement. Table 6 Production Function Estimates of the Effects of Urban and Rural School_n ______1 __ __RI R2 ____ R3 1 2 1 2 1 2 4-5 yr. of ruritl schooling{01 -t -.03 -l-10.13 IS (-0.20) -0. 1i) (-1.64) (-1.43) (1.18) (1 24) 6 or ote It, if toral schoolitg (0,1) 23 .23 -.02 -.06 .01 .02 (I .1) 1.90) (-0.17) (-0.47) (0.07) (0. 13) 4-5 yr. of urban schooling (0,1) .25 .25 32*4 .Bta (1.93) (1.93) (3.13) (2.85) 6 or Note Jr. of urbau schooling (0,1) .424 42$1 .32$* 25t8 15 .25 (3.95) (3.89) (3,33) (tt55) (1.15) (1.58) ttigratioa experience (yeats) 00 - 02*; - -.02 (0.05) (2.89) (-1.2) Note: Numbets la pareethesis are t values. Equntion I lacludes a total of 9 variables (10 it RI including the viilaqv dusty), equatioo 2 includes a total of 11 Atriables t12 it Ri). a. there are no observations in this category. *t# coefficieot sigoificant at 0.01 level in tho tail test; * cteffictent significant at 0.05 level in t'o tall test. 49 effects, the completion of 4-5 years of schooling has an effect in this region, and further years of education do not add to this effect. We notice now, however, that while the pattern remains, it is valid only for urban schooling. In R2 migration experience was previously shown to be of significance. Is urban education simply substituting as a proxy for migration experience? Or was migration experience just a substitute for urban schooling? Equation 6 in table 6 shows that neither is true. When both urban schooling and migration experience are introduced independently in the equation, both effects remain significant. In R3 we find that neither urban nor rural education makes a difference with respect to the results found above: neither the completion of the basic cycle of primary, nor the completion of primary, have an effect on productivity in the production of potatoes. 7. Methods for the Estimation of the Deteriinants of Adoption Behaviour Many studies have attempted to explain the decision to adopt or not to adopt with the use of ordinary regression methods. For example, it has been a common practice to explain adoption empirically by an ordinary least squares regression of a zero-one adoption variable on explanatory variables such as farm-size, tenure, or location.41 However, the assumption of normality of disturbances is inappropriate for such regressions, and thus the estimated standard errors and t-ratios produced by an OLS regression are not appropriate for investigating hypotheses about the role and importance of various factors in the adoption process. Additionally, ordinary linear regression estimates produce predictions other than 41 e.g. Colmenares(1976) 50 zero or one for the dependent variable; if these predictions are considered as probabilities, then predictions less than zero or greater than one are nonsensical. There now exist appropriate estimation methodologies for the investigation of the effects of explanatory variables on dichotomous dependent variables.42 The most commonly used qualitative response models are the probit model and the logit model. These models specify a functional relation between the probability of adoption and various explanatory variables. Several of the more recent studies on adoption, of modern technology have utilized these new methodologies. Gerhart used probit analysis to explain adoption rates for hybrid maize in Kenya, Jamison and Lau applied logit analysis to investigate factors affecting the adoption of chemical inputs among Thai farmers, Nerlove and Press used logit analysis to study adoption of several new inputs in Philippine agriculture, Jamison and Moock used logit analysis to investigate factors leading to adoption of chemical fertilizer and wheat cultivation among Nepalese farmers 43 Our results will be obtained with the use of logit analysis. The logit model assumes there is an index which is a linear combination of the independent variables. The dependent variable (adoption/non-adoption) is expected to equal zero or one depending on whether this index is greater than or smaller than some threshold value. These threshold values are assumed to be logistically distributed over the population of potential adopters. The coefficients obtained in the analysis can be used to calculate changes in the probability of the occurrence of adoption as a function of changes in the values of the independent variables.44 42 See e.g. Amemiya(1973), Maddala(1985) 43 Gerhart(1975) Jamison and Lau(1982), Nerlove and Press(1976), Jamison and Moock(1984). A review of the methods utilized in the literature for the analysis of adoption behaviour can be found in Feder et.al.(1985). 44 The description of the logit model was taken from-Jamison and Moock(1984) p.81. The estimations where done in an ICL 2988 computer in Oxford University. The software utilized was SPSS-X version 2.1. 51 The adoption of four modern inputs will be examined: pesticides, high yielding varieties of potato seed, chemical fertilizers and tractors. As explained above, the three regions are in different stages in the adoption of the chemical-biological technology. R3 is in the initial stage where the use of pesticides and chemical fertilizers is relatively new, while these inputs are used almost universally in the other regions. The use of HYVs -is almost non-existent in R3, it is of recent introduction in R2, and is in very common use in RI. The more recent innovation to be considered is the use of high levels of chemical fertilizers per hectare as recommended by the extension service. A small proportion of farmers use this high density fertilization i-n RI, and an even smaller proportion of farmers in the other regions. In consequence, this choice of variable will allow us to examine adoption behaviour for "old" and "new" modern inputs in each region -except R3 where there are no "old" modern inputs. The adoption decisions are treated as discrete and dichotomous. During the year prior to the survey, each of the households did or did not use pesticides, use tractors, use a density of fertilization at least as high as the recommended dose, and use HYVs for more than 20% of the area cultivated with potatoes (this, rather than a yes/no variable was chosen to exclude from the group of adopters those farmers who are still in the experimentation phase). Table 7 -describes the variables used on the analysis of adoptive behaviour. The independent variables include measures of the schooling of the household head, his age and his migration experience, and the presence of extension contacts in the three years prior to the survey. Also included in the analysis are the total farm-size, and the use of credit in the year prior to the survey. As with the production function regressions, the independent variables are entered in a series of steps. The education characteristics are entered first, followed by the age of the farmer. In the third step the variables for farm-size and credit-use are entered in the equation. Table 7 Variables for the Analysis of Adoption Behaviour: Descripton_and Means RI R2 R3 Pesticides (I if used, 0 if not) .97 .99 .54 HYUs (1 if more thai 20$ of potato seed is of HYU, 0 if not) .92 .33 .03 High density of fertilizatiso (1 if uses recommended deosity or nore, 0 if not) .67 .43 01 Tractor (1 if esed, 0 if not) .38 .19 .01 Incoeplete primary (4-5 years of schooling; 0,1) .19 .25 .15 Complete primary (6 or more years of schooling; 0,1) 65 . 43 .33 Recent contact with extension agent (0,1) 10 .29 .07 Higratiom experience (0,1) .59 .45 .40 Agt (years) 44 43 47 Total fare-size (metres) 40,760 35,127 27,162 Used credit in the last agricultural year (0,1) .24 .57 .22 ul 53 8. Adoption: Basic Results The results are presented in tables 8a to 8d and are summarized in table 9. We see that schooling has an effect on adoption behaviour but only for some inputs and in some regions. In the modern region it affects the choice of high density fertilization, and the use of tractors. In the former case, farmers with complete or incompLete primary education show-a higher propensity to adopt than farmers with less than four years of schooling. In the latter case, only complete primary education makes a difference. It is worth noting that in these regions, the coefficient for schooling shows a statistically significant effect even when one controls for the farmer's wealth and use of credit. In the intermediate region the coefficient for schooling is significant for the use of tractors and for the adoption of HYVs. However, the coefficients are unstable. In the case of tractor use, the schooling coefficient loses significance when the variable for land-holdings is introduced, suggesting that the crucial characteristic for the adoption of mechanized practices in this region is farm size rather than education. In the case of HYVs, too, the education coefficient is unstable. In this instance, when the farmer's age is introduced into the equation, schooling loses statistical significance, but age itself does not appear to be statistically significant. This suggests that neither effect is strong enough on its own, but that both characteristics affect the adoption decisions of HYVs. In R3 schooling is not related to the adoption of any of the inputs considered. Turning to the effect of extension programmes, we find statistically significant coefficients for the adoption of pesticides in R3, HYVs in R2 and high-density of fertilization in RI. In all Table 8a Results of it Regressions for Adoption of Pesticides - -_ .__ ____ __ Regions _ - RI ____ R2 R3 1 2 3 1 2 3 1 2 3 Incomp. priscro -.12 -.27 -.51 2.42 -.31 2.40 -.19 2.50 -.41 (-0.26) (-0.54) (-0.97) (0.40) (0.63) (0.47) (-0.78) (0.41) (-1.39) Complete primary .74 .35 .25 2.31 .22 2.24 .09 2.54 -.28 (1.42) (0.66) (0.46) (0.50) (0.45) (0.55) (0.46) (0.55) (-1.03) Recent ext. 3.21 2.74 1.98 1.83 1.61 1.55 .91 1.81 .57 (0.34) (0.30) (0.32) (0.39) (0.41) (0.37) (1.65) (0.39) (0.98) Higratiom exp. -.09 -.01 .18 1.75 -.05 1.61 .42 1.74 .32 (-0.24) (-0.01) (0.41) (0.43) (-0.13) (0.46) (2.23) (0.43) (1.51) Age -.04 -.03 -,04 .01 0.1 -. 01 (-2.26) (-1.95) (-2.36) (0.30) (0.30) (-0.76) Total fare size .00002 .00 .00003 (1.53) (-0.06) (3.37) Credit use 2.15 1.76 .66 (0.51) (0.52) (2.39) Intercept 6.47 8.64 7.72 6.6 8.75 5.77 4.86 5.94 4.56 (15.9) (7.66) (6.44) (12.8) (8.71) (2.17) (39.9) (2.71) (9.11) Note: Numbers in parenthesis are asymptotic t values. L Ji Table 8b RESULTS OF LOGIT REGRESSIONS FOR ADOPTION OF HIGH YIELDING VARIETIES Regions RI ___ R2 ___ _ R3 1 2 3 1 2 3 1 2 3 Intercept 6.17 7.15 6.45 4.22 4.67 4.44 2.98 3.76 2.48 (19.7) (12.1) (10.36) (21.1) (L1.15) (10,1) (6.83) (3.11) (0.64) Incomp. primary -.32 -.37 -.52 .19 .12 .05 .52 .36 .37 (-0. 86) (-0.98) (-1.31) (0.76) (0.46) (0. 18) (0. 96) (0.61) (0.29) Comp. prisary -.03 -.24 -.34 .50 .35 .25 -. 53 -.72 -2.16 (-0.09) (-0.67) (-0.91) (2.27) (1.38) (0.94) (0.74) (0.93) (-1.63) Recest ext. 3.92 3.76 3.28 .38 .37 .31 .68 .76 .68 (0.40) (0.38) (0.39) (I.95) (1.93) (1.58) (0.99) (1.07) (0.52) higration eip. .12 .16 .25 .07 .08 .03 .52 .51 1.64 (0.53) (0.69) (1.01) (0.39) (0.43) (0.14) (1.05) (1.02) (1.53) Age -.02 -.02 -.01 -.01 -.02 -.01 (-2.06) (-1.80) -1. 19) (-1.22) (-0.67) (-0.29) Total farm size .00002 .00 -.00013 (2.61) (0.72) (-1.71) Credit-use .11 .42 4.83 n (0.28) (2.14) 1.78)tn Note: Numbers io parenthesis are asymptotic t-values. Table 8c Results of Logit Regressions for Adoption of High Density of Fertilization _ _ __ ____ _ __ Regi1ons _ ____ _ _ Rl RI R2 _ __. R3 1 2 3 1 2 3 1 2 3 Intercept 4.85 5.97 5.71 4.70 5.32 5.27 s s s (28.3) (17.5) (15.4) (28.0) (13.8) (13.0) locomp. primary .53 .50 .42 .08 -.02 .11 I s I (2,33) (2.17) (1.73) (0.36) (-0.06) (0,44) Coop. primary .64 .42 .34 .13 -.08 -.001 s s (3.43) (2.10) (1.65) (0.65) (-0.36) (-.004) Recent ext. .56 .44 .39 .10 .10 .05 s (1.94) (1.49) (1.29) (0.56) (0.53) (0.24) Migration exp. -.05 -.002 -.01 .13 .14 .13 s a s (-0.38) (-0.01) (-. 09) (0.75) (0.82) (0.75) Age -.02 -.02 -.01 -.01 (-3.79) (-3.42) (-1.79) (-1.21) Total fara size .00000 -.00001 (1.23) (-1.89) Credit use . BO .40 (3,08) (2.17) Note: Numbers in parenthesis are asymptotic t-values: s:coefficient and asymptotic t-values are extremely small. Table 8d Results of Logit Rressions for Adoption of Tractors RI R2 R3 1 2 3 1 2 3 1 2 3 Intercept 4.50 4.06 3.59 3.84 3.69 3.38 -1.06 -0.67 -5.32 (24.3) (12.15) (9.7) (14.7) (7.22) (5.92) (-0.23) (-0.14) (-1.02) Incoep. primary .13 .16 .01 .33 .35 -.01 .09 .01 2.14 (0.56) (0.65) (.05) (1.04) (1,09) (-0.04) (0.01) (0.00) (0.36) Coop. primary .35 .46 .37 .50 .56 .22 2.56 2.45 4.02 (1.81) (2.21) (1.75) (1.80) (1.75) (0.64) (0.63) (0.60) (1.42) Recent ext. .10 .16 .10 -.12 -.12 -.22 -3.04 -2.98 -3.66 (0.48) (0.72) 10.43) (-0.49) (-0.47) (-O.81) (-0.25) (-0.25) (-0.40) Mligration exp. -.05 -.07 -.11 .27 .26 .22 2.27 2.29 2.52 (0.40) (-0.53) (-0.74) (1.24) (1.23) (0.93) (0.63) (0.64) (1.10) Age .01 .01 .003 -,01 -.008 -.00 (1.60) (2.14) (0.34) (-0.52) (-0.24) (-0.00) Total farm size .00001 .00002 .00004 (2.62) (3.49) (1.521 Credit use .57 .11 1.48 (3.21) (0.46) (0.71) Ln Note: Numbers in brackets are asymptotic t-values. - Table 9 Summary.. of Results of Logit Regressions for Adoption of modern Inputs Dependenit Variables Independent Variables Region EjuationnIncomplete Complete Extension Migrationl Age Firm Used ,.Primary Primary Eup. Size Credit Uses tractor RI I 0 + 0 0 2 0 i++ Q 0 + ... .......... .................. 3.. . ..... 0. ....0 . ++ +4+ ... R2 I + 0 0 2 0 + 0 0 0 3 0 0 __0 0 -0 ... 0 R3 10 0 0 0 2 0 0 0 0 0 3 0 00 0 0 +3 Uses pesticides RI I 0 0 0 0 2 0 0 o 0 - 3 0 0 0 0 0 R2 1 0 0 0 0 2 0 0 0 ~ 0 0 .... ........ . ..3 0 0 .. 0 .0 0 0 0 R3 10 0 + +4 2 0 0 + +4 a 3 0 0 0 + a + ++ Uses HYU RI I 0 0 0 0 2 0 0 0 0 - 3 0 3 3 3 +4+ 0 R2 10 ++ +4 2 0 0 +00 30 0 (+ 0 0 0 R3 I 0 0 0 0 2 0 0 0 0 0 3 0 ()0 ()0 -+ Uses High DensitT of Fertilization RI I +4 +4+ *+ 0 2 4+ + N+ 0 -- . ......... . ... 3 + + 0 0 --- 0. . R2 I 0 0 0 0 2 0 0 ~~0 0 - 3 0 0 0 0 0 -+ R3 I 0 0 0 0 2 0 0 0 0 0 3 0 0 0 0 0 a a Notation: +4:Positiye and significant at .01 level: +4: Positive and significant at .05 level; +: Positive and signi-ficant at .10 level: (+): t coefficienit larger thon 1.5; minus signs are used for negative coefficients at the same levels of significance. 59 three cases, however, we find that the link appears to be mediated by the use of credit. Migration experience appears to have a significant effect for the adoption of pesticides in R3.45 In the other regions we find no effect of adoption associated with migration experience. This finding contrasts with our previous finding that migration experience affects productivity in R2. The coefficient for age usually shows a negative sign for the adoption of biological and chemical inputs in all regions, suggesting that older farmers are more conservative in their outlook. In RI the coefficients for this variable are larger and more significant than in the other regions. It is interesting to notice that in this region "age" not only distinguishes the early adopters from the rest (as is the case with the adoption of high density fertilization), it is also a factor distinguishing those few laggards who have not yet adopted the use of pesticides (3%) or HYVs (8%). The only exception concerning the sign of the age variable is the case of tractor use in RI: here the effect of age in adoption is positive. This could be due to less family labour being available in the households of the older farmers, after the children are gone. Farm size is seen to be an important determinant of adoption, especially for the case of tractor use, where the coefficient is significant for the three regions: the larger farms are the adopters of mechanical practices. With respect to the other inputs, too, we find that farm size can be of importance: the larger farms are the first to adopt pesticides in R3, and a small farm size is a characteristic which distinguishes the non-adopters of HYVs in Rl. Large farm size is not always an incentive for adoption, however, especially for land-substituting inputs, as can be seen by the negative sign of farm size for the adoption of high density fertili.ation in R2 and HYVs in R3. 45 When the equation was run for the adoption of chemical fertilizer, this variable also showed a strong effect. 60 Access to credit influences adoption of modern inputs in all regions. In R3, it increases the propensity to adopt pesticides and HYVs; in R2 it increases the likelihood of adopting HYVs and high density fertilization, and in Rl its effect is found positive for the probability of adoption of high density fertilization and of mechanized practices. Several general patterns arise from the above discussion. We find that the different types of education are associated with a higher probability of adoption but only in the early stages of the diffusion process. We find significant coefficients for the education variables for pesticides in R3, HYVs in R2 and high density fertilization in RI. In the case of "old innovations" such as pesticides in R2 or HYVs in Rl, the education variables are not significant in explaining the difference between adopters and non- adopters. If we divide the farmers into early adopters, followers and laggards, according to their behaviour in the diffusion process, then this result suggests that more education distinguishes the early adopters from the rest. Our results for RI suggest that the laggards are distinguished more by their old age and atypically small farm size than by their educational characteristics. This is consistent with the view that, once diffusion is under way, it occurs mostly by imitation, with the individual educational characteristics of the farmer having only a secondary role. The effects of credit, too, are particularly noticeable for recent innovations. The coefficients for credit use were seen to be significant for pesticide and HYVs adoption in R3, HYVs and high density of fertilizers in R2 and high density of fertilizer and tractor use in RI, and it was not significant for the adoption of the older modern inputs. This could be due to a pattern by which availability of total finance increases with modernization: capital is initially a constraint on adoption, but the increased net incomes which follow adoption give access to a greater working capital, and allow the farmers to become independent on the credit institutions. Additionally this finding is also likely to be due to the fact that 61 loans are often given in the form of specific inputs.46 For recent innovations, it may be difficult to obtain these inputs from other providers. Also, the subsidy implicit in *the low interest rate charged may constitute an additional incentive to convince still undecided farmers to experiment with a new input. Our results suggest that farm-size can be an important factor influencing technical change, but they do not indicate an absolute bias in favour of the larger farms.47 The results are consistent with the view that farm-size can influence the direction of technical change: the larger farms have a higher propensity to mechanize some of their field activities, while the smaller farms have a somewhat higher propensity to adopt land-substituting inputs. 9. Sum-ary and Conclusions In this section we summarize the main results discussed in this paper with respect to the effects of general education on productivity in potato production, and in the adoption of modern inputs. Education (of some type) was seen to be important for adoption and productivity. For adoption the role of education was found to be greater in the early stages of the diffusion process. The education characteristics seem to separate early adopters from followers. At the other end of the diffusion process, these characteristics do not seem to matter as much. Rather than very low levels of schooling or lack of extension contacts, we found old age and very small farm size to be the main characteristics of the laggards. 46 This was discussed in section IV.8. 47 This cannot be generalized, however, as our sample was taken from regions where practically all farms had less than 10 hectares, and a larger size in absolute terms could be an advantage. Also, we did find some indication that very small farm size can be a disadvantage -forcing peasants actively into off-farm activities- and seems to be a characteristic of the laggards. 62 Schooling and extension contacts are correlated with farm-size. When they are introduced into an equation without controlling for the wealth of the farmer, they sometimes act as a proxy for this characteristic. There is little doubt that a farmer's wealth can be an important determinant of adoption behaviour, but even if the inclusion of variables measuring farm size weakens the coefficients of education, in several equations they remain stable and statistically significant determinants of adoption behaviour. With respect to the effects of formal education on productivity, two interesting patterns were found. First, the effects of schooling are stronger in the modern region than in the more traditional ones. While R2 and R3 cannot be described as "technologically stagnant", the complexities of technology (and their rate of change) and the market variations are greater in RI. The results are thus consistent with the Schultz hypothesis that the value of education consists in enhancing the peasant's "ability to deal with disequilibrium". Second, the results show the existence of a threshold-effect, by which formal education begins to have an effect on output only after a certain number of years of schooling have been obtained. Far from being a general threshold, the number of years necessary to obtain an effect on output seems to increase with the complexity of the technologies involved. This suggests that while basic levels of education may be effective in speeding traditional regions into the first stages of technological modernization, further technological development will require higher levels of formal education. Quality of formal education varies in different schools, and this difference is particularly marked between urban and rural schools. When these are distinguished, striking differences are found. The effects of urban schooling on productivity are much stronger than those of rural schooling (even if we control for "urban socialization"). In RI, we find that the effect of completing primary education in urban areas is much larger that that of completing 63 primary education in rural schools. In this region, the estimated coefficient on complete rural primary is lower than that of 4-5 years of urban schooling. In R2, when the location of schools attended is considered, the effect is even more striking since more than 4 years of urban schooling are seen to have strongly significant effects, but the coefficients for the effects of rural schooling are all negative (but statistically not significant). With respect to non-formal education, we examined the direct and indirect effects of extension, i.e. the effects on those directly contacted by extension services, and the effects on those who imitate them. The first hypothesis with respect to the direct effects of non- formal education on productivity was simply that there is a positive effect. The results show that this positive effect existed only in R2, while in the other regions the coefficients are positive, but not significantly different from zero. We suggested several reasons for this finding. Two factors which are likely to be important are (a) the quality and commitment of the people working in PRODERM (the agency which gave most of the extension in R2) is higher than that found in the other regions and (b) PRODERM gives more additional services than the state agencies, in particular access to inputs and especially to HYV seeds. Apart from these factors we stressed the adequacy of the message transmitted by extensionists. We argued that the message transmitted is similar in the three regions, but seems to be appropriate only in R2: it is too advanced for R3 and does not include any important innovations for the technological levels already achieved in RI. We argued that if non-formal education transmits specific information on new technologies, its effects should become obsolete as new technologies appear. Also, the differential between those who directly receive the information and those who do not should fade away as imitation occurs. The hypothesis is that the effects of recent contacts on output should be larger than the effects of old extension. Implied in this hypothesis is the related one that 64 imitatian takes time. Our empirical results are consistent with the first hypothesis, in that in the three regions the value of the coefficient for recent extension is larger than the coefficient of old extension. However, it is only in R2 that any of the coefficients is statistically significant, but in this region only the coefficient for recent extension is significant. The second hypothesis is sustained by the data in RI. In this region imitation does appear to have an effect, but only with a lag, since it is only the indirect effects of old extension that appear to affect output. In R2, we find no positive effect from imitation. in R3, the coefficients of the variables used to measure the indirect effects of old and new contacts are statistically greater than zero, but the fact that their value is very high in a region where we find no direct evidence of positive effects of extension on productivity via the use of modern technoloqies casts some doubt on the variables in use. The variables used consist of assigning to each household of a village the village-wide proportion of households being contacted. Hence all households of a village are given the same value in these variables. Obviously these variables are likely to be correlated with other village characteristics which may affect output (especially since the number of villages in each region is small i.e. 4 in R3, 5 in R2 and 9 in RI), and so these results must be interpreted with great care. Finally, the effects of two forms of informal education on output were examined, that of age (as a proxy for experience) and that of migration experience. With respect to age the expectation was of a positive effect on output, but negative coefficients were found in the three regions, the negative effect being more pronounced the more modern the area. These findings were interpreted as reflecting a "cohort effect", according,to which older peasants tend to stick to older, less productive technologies. This could be due to the existence of greater risk-aversion or conservatism among the old even when formal education is taken into account. 65 Migration experience appears to have no effect in RI, but shows important effects for adoptive behaviour in R3 and for productivity in R2. Several points were noted with respect to migration. Firstly, it is the urban experience that is of greater value; return from rural areas has a smaller effect on output. Secondly, the significance of the coefficients remains even when the (urban/rural) location of schooling is considered. This suggests that the migration experience variables are not simply a proxy for quality of education. Finally it was shown that when migration experience was not included in the regression, the effects of formal education were overestimated. Credit was seen to be important as a determinant of adoption, giving capital-constrained farmers access to the modern inputs. As was expected, once these new inputs are included in the analy.sis, we find that credit does not have a role in determining productivity levels. The role of credit on adoption appears to be significant only for recent innovations. This is consistent with the view that modernization leads to capitalization of the farm. Lack of capital is a constraint for initial adoption, but the adoption of successful technologies leads to an increase in net incomes which then permits the farmer to achieve an increase in his working capital to the higher levels required for investment in the use of the new inputs. In other words, credit seems important to start the farmers off in the use of the new inputs. In a second stage, the new technology produces higher incomes which allow the farmer to become independent of the credit-giving institutions. 66 REFERENCES Adams, R.N. 1959. A Community in the Andes: Problems and Progress in Muguiyauyo. Seattle: University of Washigton Press. Amemiya, T. 1973. "Regression Analysis when the Dependent Variable is Truncated Normal" Econometrica 41. pp.997-1016. Arguedas, J.M. 1964. Todas las Sangres Buenos Aires: Editorial Losada. Arriagada, A.M. 1983. "Determinants of Sixth Grade Achievement in Peru". Mimeo, Education Department, The World Bank. Benor, D. and J.Q. Harrison. 1977. Agricultural Extension: The Training and Visit System. Washington D.C.: The World Bank. Berry, R.A. and W. CLine. 1979. Agrarian Structures and Productivity in Developing Countries, Baltimore and London: Johns Hopkins University Press. Bhalla, S.S. 1979. "Farm and Technical Change in Indian Agriculture" in Berry and Cline (L979) Binswanger,H;P. And V.W. Ruttan (eds.) 1978. Induced Innovation: Technology, Institutions and Development, Baltimore and London: The Johns Hopkins University Press. Binswanger, Hans. 1978. The Economics of Tractors in South Asia: An Analytical Review, New York: Agricultural Development Council and The International Crops Research Institute for the Semi- Arid Tropics. Bowman, M.J. 1976. "Through Education to Earnings?" Proceedings of the National Academy of Education Vol.3, pp.221-92. Chaudry, D.P. 1979. Education, Innovations and Agricultural Development ILO, Vikas. Colmenares, H.J. 1976. Adoption of Hybrid Seeds and Fertilizers amonq Colombian Corn Growers, Mexico City: Centro Internacional de Mejoramiento de Maiz y Trigo. Coombs, P. and Ahmed, Manzoor. 1974. Attacking Rural Poverty: How Non-Formal Education can Help, Baltimore: The Johns Hopkins University Press. Cotlear, D. 1986. 'Technological and Institutional Change Among the Peruvian Peasantry". D. Phil. Dissertation, Oxford University. Degregori C.I., Golte, J. 1973. Dependencia y Desintegracion Estructural en la Comunidad de Pacaraos, Proyecto de Estudios Etnologicos del Valle de Chancay, Monografia No. 3. Lima: IEP. Feder, G., R.E. Just, and D. Zilberman. 1985."Adoption of Agricultural Innovations in Developing Countries: A Survey". Economic Development and Cultural Change, Vol. 33, pp.255-298. Figueroa, A. 1985. "Productividad y Aprendizaje en el Medio Rural: Informe Comparativo", Research Report presented to ECIEL, Rio de Janeiro. Gerhart,J. 1975. The diffusion of Hybrid Maize in West Kenya, Mexico City: Centro Internacional de Mejoramiento de Maiz y Trigo. 67 Heyer,J. 1985. "Landless Agricultural Workers n Coimbatore Villages". Oxford: Draft. Jamison, D.T. and L.J. Lau. 1982. Farmer Education and Farm Efficiency, Baltimore: The Johns Hopkins University Press. Jamison, D.T. and Moock, P.R. 1984. "Farmer Education and Farm Efficiency in Nepal: The role of Schooling, Extension Services, and Cognitive Skills". World Development Vol.12, pp.67-86. Khan, M.H. 1975. The Economics of the Green Revolution in Pakistan. New York: Frederick A Praeger. Laite, A.J. 1981. Industrial Development and Migrant Labour Manchester: Manchester University Press. Lindner, R.K., A.J. Fischer and P.Pardey. 1979 "The Time to Adoption'. Economic Letters Vol.2, pp.187-90. Lockheed, M.E., D.T. Jamison, and L.J. Lau. 1980. "Farmer Education and Farm Efficiency: A Survey". Economic Development and Cultural Change Vol.29, pp. 37-76. Moock,P.R. 1981. "Education and Technical Efficiency in Small Farm Production". Economic Development and Cultural Change, Vol. 29, pp.723-39. Rivera, I. 1979. "Los Determinantes de la Calidad de la Educacion en el Peru". Universidad Catolica, Documento de Trabajo CISEPA No. 44, Lima. Rogers, E. 1962. Diffusion of Innovations, New York: Free Press of Glencoe. Ruttan, V. and H.P. Binswanger. 1978."Induced Innovation and the Green Revolution" in Binswanger and Ruttan (1978). Schultz, T. W. 1964. Transforming Traditional Agriculture, New Haven: Yale University Press. -------------. 1975. "The Value of the Ability to Deal with Disequilibria". Journal of Economic Literature Vol.13, pp.872- 76. Schujter, W. and M. Van der Veen. 1977. "Economic Constrains on Agricultrual Technology Adoption in Developing Countries" U.S. Agency for International Development, Occassional Paper no.5.Washington D.C. Thomas, V. 1978. "The Measurement of Spatial Differences in Poverty: The Case of Peru" Staff Working Paper no. 273, The World Bank. Weil, P.M. 1970. "The Introduction of the Ox Plow in Central Gambia." In Mc. Laughlin, P. (ed.) African Food Production Systems: Cases and Theory. Baltimore: The Johns Hopkins University Press. 68 Welch, F. 1970. "Education in Production". Journal of Political Economy. Wharton, C.R. (ed.). 1969. Subsistence Agriculture and Economic Development. Chicago: Aldine.