The Adoption Of Agricultural Innovations

TECHNOLOGICAL

FORECASTING

AND

SOCIAL

CHANGE

43, 215-239

(1993)

The Adoption of Agricultural Innovations A Review GERSHON

FEDER

and DINA

L. UMALI

ABSTRACT

This paper reviews the theoretical and empirical literature on the adoption of agricultural innovations during the last decade and the impact of policy interventions promoting technology adoption. The analysis of the final stage of the Green Revolution technology diffusion cycle reveals that the agroclimatic environment is the most significant determinant of locational differences in adoption rates. The linkage between micro-adoption and the aggregate diffusion process needs to be more firmly established to achieve a clearer understanding of diffusion patterns. Several studies showed that the impact of policy interventions to promote technology adoption depends on the type of technology, market structure, and the nature and duration of the policy intervention.

Introduction The analysis of diffusion processes has been given significant attention by social scientists in the past few decades, because such processes are an important determinant of economic growth. Agricultural innovations have been of special interest in this context, because the decision makers in the agricultural case are typically individual households, rather than firms. Thus, household behavioral issues are hypothesized to be a key to understanding the observed processes. The Green Revolution from the 1960s to the early 1980s motivated numerous studies to explain the determinants of adoption during the early stages of the diffusion process. A clear distinction between recent works analyzing adoption and works in the 1970s and early 1980s relates to the evolution of the Green Revolution. By the early 199Os, most of the “first-wave” Green Revolution technologies had long attained or were close to the final stage of the diffusion cycle. Correspondingly, many recent studies focused on refining the underlying theory of agricultural technology adoption based on the completed diffusion cycle, from introduction to the adoption of the second wave of agricultural technologies, and the empirical testing of these theories, so as to draw new lessons and new directions for future development strategies. Feder et al. [l] conducted a comprehensive survey of the theoretical and empirical literature on the adoption of agricultural innovations; the majority of these studies focused on the initial stages of diffusion of Green Revolution technologies. This paper complements this work and surveys the various studies which have now mainly graduated to the analysis GERSHON FEDER and DINA L. UMALI are, respectively, Chief of and a consultant to the Agricultural Policies Division of the Agriculture and Rural Development Department of the World Bank. The views expressed are those of the authors and do not necessarily represent policies or views of the World Bank. Address reprint requests to Dr. Gershon Feder, Agricultural Policies Division, Agriculture and Rural Development Department, The World Bank, 1818 H Street, N.W., Washington, DC 20433. 0

1993 Elsevier Science Publishing

Co., Inc.

0040-1625/93/$6.00

216

G. FEDER

AND D.L. UMALI

of agricultural innovations that have reached maturity. Because the volume of such published research is overwhelming, we will attempt simply to review representative works rather than present an exhaustive discussion of all work to date. For the purposes of the present discussion, an innovation is defined as a technological factor that changes the production function and regarding which there exists some uncertainty, whether perceived or objective (or both) The uncertainty diminishes over time through the acquisition of experience and information, and the production function itself may change as adopters become more efficient in the application of the technology. This definition facilitates the analysis of innovation diffusion as a dynamic process. Technology adoption may also be viewed from two perspectives. At the micro level, each decision unit must choose whether to adopt the innovation and its intensity of use if adopted. Many adoption studies, therefore, examine the factors influencing the firm’s or household’s adoption decision and may be viewed from a static or dynamic (if learning and experience are incorporated in the decision model) perspective. At the macro level, the adoption pattern of the whole firm or household population is examined over time to identify the specific trends in the diffusion cycle. “Diffusion studies do not consider the innovation process, but begin at a point in time when the innovation is already in use. . . . Thus, relative to adoption, diffusion may be viewed as a dynamic, aggregative process over continuous time” [2, p. 781. The earliest adopters may be referred to as innovators, and the diffusion process is essentially the spread of the new technology to the other members of the population. The organization of the paper is as follows. The next section surveys the theoretical literature on adoption patterns at the farmer/household and sector levels. The third section reviews the empirical studies that have attempted to validate various aspects of the adoption process in light of the theoretical literature, while the fourth section discusses methodological issues. The fifth section examines the rationale and implication of government interventions to promote technology adoption, drawing from the results of theoretical studies of the issue. The final section draws conclusions from the review. A Survey of Adoption REVIEW

OF MODELS

and Diffusion OF ADOPTION

Models BEHAVIOR

OF INDIVIDUAL

HOUSEHOLDS

Most agricultural technologies introduced in the last three decades, and the highyielding varieties (HYVs) in particular, are in fact a package of interrelated technologies (for example, fertilizer, herbicides, and chemicals). Accordingly, one major focus in the literature in recent years has been the investigation of the decision-making process characterizing choice of the optimal combinations of the components of a technological package over time. The impact of factors such as credit, information availability, risk, and farm size on farmer adoption behavior, which has been a common preoccupation of adoption studies, has been investigated in many recent works. One of the first models dealing with a technological package was developed by Feder [3], incorporating technological complementarity and adoption under uncertainty.l The study examines a case where farmers face the choice between a traditional technology (for example, a traditional crop variety) and two innovations-a divisible, scale-neutral technology (for example, a modern variety or MV) and a lumpy technology subject to decreasing costs with respect to farm size (for example, a tubewell). The two innovations are interrelated because I Complementarity is defined as a situation wherein augmenting one innovation (for example, a tubewell) by another innovation (for example, an HYV) increases profits (corrected by a risk factor) by more than is obtained when one innovation alone is adopted [3].

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

217

potential output is higher if both technologies are adopted than when only one is adopted. Furthermore, the adoption of the lumpy technology influences the perceived risks associated with the divisible technology. Given a perceived output risk associated with the divisible technology, and farmer risk aversion, farmers maximize their expected utility through the dichotomous choice of whether or not to adopt the lumpy technology and the choice of the proportion of land to be allocated to, as well as the intensity of use of, the divisible technology. The model shows that the optimal level/intensity of divisible input use is not affected by uncertainty and risk aversion, is higher when the lumpy technology is adopted, and is negatively related to the input/output price ratio. The proportion of land allocated to the divisible technology decreases if (a) output variability (that is, uncertainty) increases, (b) the fixed cost associated with the lumpy technology increases (assuming absolute risk aversion declines with income), (c) the farm size decreases (assuming relative risk aversion increases with income), and (d) the cost of the divisible input increases. Assuming a logarithmic utility function,2 it is shown that adoption of both the lumpy and divisible technologies is worthwhile only beyond a critical minimum farm size. Moreover, the farm size threshold declines as uncertainty regarding perceived output variability decreases. Perceived output variance associated with new technology may decline over time as a result of learning by doing or better dissemination of information. Thus, the model implies a pattern whereby initially larger farms adopt both technologies, while smaller farms adopt the divisible technologies to a limited extent. Over time, with the reduction in perceived uncertainty, smaller farmers adopt the divisible technology more intensively, and eventually, many of them adopt the lumpy technology as well. The imposition of a credit constraint does not affect the intensity of use of the divisible technology; rather, its impact is manifested in the adjustment in the proportion of land allocated to the divisible technology. Leathers and Smale [5] attempt to explain sequential adoption even when farmers are risk neutral and unconstrained in their expenditures using a dynamic Bayesian model. Farmers maintain their subjective expected utility of income which is a function of a decision set of technologies whose returns are conditional on the validity of information received (extension report). The model depicts a farmer who is uncertain about the validity of the information received from an extension agent regarding the technologies’ (MV seed and fertilizer) productive performance in terms of the revenues generated. Because uncertainty is reduced through experience, the model demonstrates that in order to learn more about the innovation, the farmer may choose to adopt a component of the package rather than the complete package. Moreover, while early adopters may adopt only parts of a package, later adopters, whose confidence has been raised by the positive experience of their neighbors, may adopt the whole package. Pitt and Sumodiningrat [6] model the simultaneous decisions involved in the adoption of new technologies under risk and uncertainty using the metaprofit function approach.’ They note that if the cultivator’s seed technology choices are the result of profitmaximizing behavior, then varietal choice will depend on the determinant of profits, that is, divisible factor prices, variety specific output prices, and the level of fixed factors including the agroclimatic environment. Each variety will be characterized by a different * The logarithmic

utility function,

as used by Feder and O’Mara [4], is selected simply for ease of manipula-

tion.

’ The metaprofit function is dual to the metaproduction function introduced by Hayami and Ruttan [7] and has all the usual properties of profit functions. By Hotelling’s lemma, the demand for the variable input is equivalent to the negative of the first derivative of the metaprofit function with respect to the input price [6].

218

G. FEDER

AND D.L. UMALI

profit function to reflect the differences in its biological characteristics. If seed variety choice is itself a component in the farmer’s profit maximization decision problem, then there exists a single profit function-a metaprofit function, which treats variety choice as a divisible input and from which all variety-specific profit functions can be derived. By Le Chatelier’s principle, treating seed variety as fixed, as most studies do, leads to the underestimation of the absolute values of input demand and output supply elasticities. The introduction of constraints, such as the fixing of seed variety choice in this case, cannot increase the opportunity to substitute other inputs. As a result, the response to price and fixed factors is greater for movements along the meta surface relative to the individual variety-specific profit surface. The farmer’s dichotomous variety choice decision (also referred to as the switching equation or metaprofit function),“ as formulated by Pitt and Sumodiningrat, is approximated by a linear function of the difference between a given pair of variety-specific profit functions (HYVs vs. traditional variety) and a vector of other variables influencing variety choice. It is further assumed that only one seed variety is chosen per plot. These varietyspecific profit functions are represented by transcendental logarithmic flexible functional forms and are a function of the ratio of input/output prices and a vector of fixed factors. Risk is incorporated in the variety-specific profit function in the form of price variability such that prices are a function of a fixed and a stochastic element. Variety-specific profit share equations having additive errors are derived using Hotelling’s lemma from the profit functions. This model is applied to farm-level data from Indonesia and joint estimation of the complete model (the metaprofit function, the variety-specific profit functions, and the input demand equations) is performed using the method of maximum likelihood. The model does not incorporate features that entail a dynamic diffusion process (for example, learning). The study found that the higher profitability of a seed variety is positively associated with a higher probability of its adoption. The adoption of HYVs is positively associated with the likelihood of flooding, quality of irrigation (conditional on its effect on relative profit), and the availability of credit, and negatively associated with the likelihood of drought and land owned. Education was not found to be a significant determinant of variety choice. Traxler and Byerlee [8] explore complementarity from the product output perspective to explain the differential variety diffusion at the farm level in South Asia. They draw attention to the fact that some innovations produce by-products that have economic value, a feature which has been neglected by previous models. They note that the straw produced from MV wheat is valuable as animal fodder in many developing countries and should thus be included in the farmer’s objective function. To illustrate, they assume a simple linear profit function subject to a production function with two products (grain and straw) and one input (nitrogen) and show that the choice of variety is determined by both a technical constraint (the grain/straw harvest index) and the relative input/ output and grain/straw prices. For example, given a fixed level of nitrogen and grain/ straw ratios for a traditional variety (TV) and an MV, they show that the MV will be revenue superior to the TV only if the ratio of the reduction in straw yield to the increase in grain yield is less than the existing grain/straw price ratio. Rauniyar and Goode [9] provide some evidence that farmers do adopt technologies in clusters. They investigated the interrelationships among technological practices adopted

4 The decision choice variable takes the value of 1 if the HYV seed is adopted and zero otherwise.

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

219

by maize-growing farmers in Swaziland. The seven practices included per hectare application of MV maize seeds, basal and topdress fertilizer, insecticides, tractor plowing, planting density, and average planting date. Using factor analysis, they found that farmers adopted the seven technologies in three independent packages: (a) MV maize seed, basal fertilizer, and tractor plowing; (b) topdress fertilizer and chemicals; and (c) planting date and density. Interfactor correlations do not support the sequential adoption hypothesis. The study, however, did not investigate the process by which each component in a set is adopted. Byerlee and Polanco [lo] present empirical evidence on the sequential adoption of technologies by plotting the time pattern of the diffusion of three technologiesMV maize, fertilizer, and herbicides- in the Mexican Altiplano. Fitting logistic diffusion curves of the cumulative adoption levels based on the farmers’ recall of adoption years, they show that farmers in the wet zones followed the following sequences of adoption: MVs, herbicides, and fertilizer sequentially; the sequence of adoption for farmers in the dry zone was MVs, fertilizer, and herbicides. Two major underlying assumptions of most farm household technology adoption models are that markets are perfectly competitive and the production and consumption decisions are separable. However, the economic environment of rural households in developing countries is often characterized by imperfect or missing markets, resulting in nonseparability of the household production and consumption decisions. Nonseparability also occurs when there is imperfect substitutability of family and hired labor, differences in purchase and sales prices of inputs and outputs, and in the presence of interlinked transactions [l l-141. Pradhan and Quilkey [15] apply the household model to capture farmer technology adoption behavior while taking into account the existence of imperfect factor and commodity substitution, as is found in many developing countries. The household production function undergoes a change as a result of the introduction of a new technology (for example, MV rice) which may be used concurrently with the existing technology. Given these technologies and assuming that they produce an identical commodity and compete for the fixed amount of land, the two production functions are combined to derive the household’s total output equation. The farm output equation is thus a function of the labor supply and hired labor demand, the cash inputs, the proportion of land allocated to the new technology, and other relevant exogenous variables. In this model, farmers maximize their utility from the consumption of the produced commodity, leisure, and a residual cash endowment to purchase other consumption commodities subject to (a) a production constraint which is a function of family and hired labor, other cash inputs, the area allocated to the new and traditional technology, and other exogenous variables; (b) a time endowment constraint; (c) a cash endowment constraint; and (d) a market balance constraint. The solution to the constrained optimization yields the family and off-farm labor supply, the hired labor demand, and the technology adoption equations. In specifying the empirical model, Pradhan and Quilkey rely on intuition and experience to augment insights from the theoretical model. Using data from a study of farm households in Orissa, India, and applying three-stage least squares, their results confirm that the farmer’s decision about the degree of land allocation to HYVs of rice is made simultaneously with that of the input levels of family labor and hired labor, and expenditures on purchased inputs such as fertilizer and chemicals. Other factors affecting the land allocation decision include the percentage of irrigated area, experience with growing the MV rice, and education. Several studies have examined the role of prices and learning in technology adoption. Rosenberg [16, 171 stressed the role of learning by using and expectations of future prices of the innovation on the takeoff and shape of the diffusion of new technology. Other

220

G. FEDER

AND D.L. UMALI

studies that investigate the impact of price expectations and learning by doing on the process of diffusion of an innovation using a firm-level framework include Kislev and Shchori-Bachrach [ 181, Feder [ 191, and Just and Zilberman [20]. Producers are assumed to choose the level of use of a new technology according to a static criterion, but at the industry level, the aggregate adoption pattern changes (with more firms joining over time) as elements such as perceived risks change. Stoneman and Ireland [21] investigated the role of price expectations, while Feder and O’Mara [22] demonstrate the plausibility of Bayesian learning as the underlying dynamic process. Tsur et al. [23] study the effects of risk and learning on intrafirm diffusion of new technology, assuming the decision makers to be risk averse and to vary according to their learning ability and firm size. Unlike most previous models, they explicitly model the farmer’s adoption decision in a multiperiod dynamic framework under risk and uncertainty. The diffusion process in the Tsur et al. model is derived as a result of a dynamic optimization procedure under the assumption that decision makers have perfect foresight regarding the effects of present decisions on future events. Since conditions change over time, entrepreneurs update decisions and a diffusion process occurs. For example, the adoption of a new technology today will be considered worthwhile if, as a result of learning, it would improve future performance, even though it involves losses at the present time. Likewise, adoption will be accelerated or postponed if the price of the innovation is expected to change. The dynamic model developed by Tsur et al. accounts for risk aversion by assuming that the firms maximize expected utility, defined over the present value of a stream of future profits. The returns from the traditional technology are assumed to be stochastic; the returns from the new technology are also stochastic, but with a structure that is affected by learning. Two types of learning are assumed: the unconscious process of learning from own use and from the use of others. Total returns at time tare a function of returns from the new technology given an associated information set, the returns from the traditional technology, and the adoption cost. A utility assessment mechanism is assumed so that the decision maker first calculates the present value of the profit stream associated with a given bounded adoption path (the rate at which area is transferred to [or returned to] the new technology [the old technology]) and then assigns a utility level to this value. A constant absolution risk aversion utility function is further assumed. The adoption decision problem therefore involves the selection of the adoption path that maximizes the expected utility from the profit stream subject to a bounded rate of adoption. Solving the maximization problem, Tsur et al. investigate the effects on adoption decisions of changes in the discount rate, firm size, and the level of risk aversion. While most models of technology adoption under uncertainty predict that risk aversion delays adoption, the results of Tsur et al. [23] indicate that risk aversion positively affects adoption. This seemingly counterintuitive result is explained by the fact that although mean profit at time t is negative, compensation occurs in future periods as a result of the decline in future risks which results from the learning process. Therefore, the higher the risk aversion, the greater is the appreciation of these future declines in risk. If the current mean profit is positive, risk aversion leads the decision maker to diversify in order to reduce undesired income risk. In the absence of uncertainty or risk aversion and in the absence of economies of scale with respect to the area allocated to the new technology, there will be no incentive to diversify. Another finding is that if the discount rate is greater than the rate of increase in the variability of income, the intrafirm diffusion process eventually settles in the long term at a certainty world solution, that is, full-scale adoption or no adoption at all. If conditions were reversed, risk considerations become increasingly important and dominate at later stages of the adoption process.

THE ADOPTION

OF AGRICULTURAL

221

INNOVATIONS

The models discussed above conform to a portfolio selection model under uncertainty. In the case of the HYVs, the land allocation choice in technology adoption becomes a portfolio decision, in which farmers maximize the expected utility of income by choosing a specific combination of varieties given their risk aversion levels, the stochastic interrelationship between the technologies, and the effects of other socioeconomic factors such as wealth, age, and education [24]. The other models of farmer decision making under certainty which have been employed to explain technology adoption include the safety-first algorithms [25-271, farmer experimentation and the learning process [lo, 28-301, and models that contain specific constraints on the variability of some decision variables (for example, land and credit) [31, 321. However, Smale [24] observes that the ex ante decision to choose one model as a basis for empirical analysis of technology adoption can result in estimation errors, such as the omission of relevant variables and biased estimators. To overcome this limitation, Smale proposes a general model that contains all four special specifications. Applying the general model to the case of farmers in Malawi, the farmer is assumed to maximize the expected utility of net income and information over a multiperiod horizon subject to an area constraint and a cash/credit constraint on production expenditures. Net income is assumed to be equal to the net returns from production less a penalty associated with falling below local subsistence requirements. The penalty function assigns penalties to combinations of choice and output which are disastrous. Yields and penalties are stochastic, and output and input prices are known with certainty. In any production period, the farmer gains utility from current net income and the discounted future flow of information that results from current adoption decisions. The solutions to the problem in any period can be generally characterized by employing different assumptions about farmer behavior and technology. For example, a farmer who is risk neutral will only sow the variety with the highest returns when any or all constraints are binding. In this case, introducing a convex penalty function or a concave information function generates an interior solution. Increasing competition for often finite water resources, the environmental costs associated with the buildup of salinity and waterlogging, and the high investment costs associated with agricultural drainage have provided significant impetus for the development of water-conserving technologies, such as drip and sprinkler irrigation systems. Because of the important role these technologies can play in environmental conservation, the factors influencing their adoption have received increasing attention in recent studies. Caswell and Zilberman [33] developed one of the first models examining the adoption of irrigation technologies, taking account of the land-augmenting and engineering properties of the technology and land quality of the farm.5 Their study assumes that a single crop is produced with a constant returns-to-scale technology. The output of a particular plot with a given land quality and well depth is a function of the amount of water applied and the irrigation effectiveness of the technology. Irrigation effectiveness (measured by the ratio of the amount of water utilized by the crop over the total quantity of water applied to the field) is also influenced by land quality. Irrigation cost for a given land quality and well depth consists of two components: a fixed irrigation cost per acre and a variable component accounting for the energy cost of pumping and pressurization. In choosing between traditional gravity irrigation (for example, flood, level border, or furrow) and modern irrigation (drip and sprinkler) systems, the farmer follows a two-stage 5 Land quality is measured by the plot’s water-holding texture, slope, salinity, and depth [33].

capacity,

which depends

on factors

such as soil

222

G. FEDER

AND D.L. UMALI

procedure.6The farmer must first choose the optimal amount of water for each technology and then choose the irrigation technology yielding the highest operational profit. Thus, at each location, farmers will adopt the technology with the highest quasi rent per unit area, given that it is greater than the land rent. Caswell and Zilberman found that land quality and well depth determine the threshold levels at which the adoption of the modern technology will occur. Modern irrigation technologies will be adopted at well depths greater than the threshold level ry, which is a function of land quality. Furthermore, when the elasticity of the marginal productivity of effective water (EMP, which measures how responsive the crop is to further irrigation) is high, the modern technology becomes relatively more profitable as well depth increases. At low EMP, there is a range of well depths, YYto y”, where the modern technology becomes relatively more profitable as well depth increases and the switch from the traditional to modern technology occurs. However, at well depths greater than the critical level y”, an increase in well depth will improve the profitability of the traditional relative to the modern technology, so that a switch back to the traditional technology occurs.’ The study also showed that at high EMP and well depths greater than ry, the modern technology becomes relatively more profitable as land quality declines. Moreover, there exists a critical land-quality threshold level below (above) which the modern (traditional) technology will be adopted. For a range of high land qualities, the operational profits differential between the modern and traditional technologies increases as land quality declines. This result suggests that at low EMP, there may exist either one or two switching land-quality thresholds: the upper threshold separates a tier of high-quality land utilizing the traditional technology from a tier of medium-quality land utilizing modern technologies and a lower land-quality switching threshold separating the tier of medium-quality land from a tier of low-quality land which utilizes the traditional technology.8 Caswell and Zilberman also found that if the new technology increases output supply and output demand is not infinitely elastic, consumer welfare will improve because of the reduction in prices, the welfare of owners of low-quality lands (or deep wells) may improve because the decline in price may be offset by the increased output, but the welfare of owners of prime lands may decline. Several empirical studies have partially confirmed these theoretical results.’ Negri and Brooks [35], in a study of farmer adoption of sprinkler irrigation and tail water recovery pits in the United States, found that land quality (slope, productivity constraints, and soil texture) is the most important determinant of adoption. Other factors, such as 6Under the traditional system, large quantities of water from a distribution pipe at the head of the plot are applied in a short period of time and spread across the plot by gravity, which often results in a nonuniform application of water. With modern irrigation technologies, small quantities of water are applied over a continuous period of time, and both equipment (tubing, valves, filters, emitters) and pressure are used to distribute water uniformly throughout the field. ’ The second switching at well depths greater than y” may not occur in reality because pumping water from such depths may not be profitable under either technology [33]. 8 Caswell and Zilberman [33] also found that (a) the adoption of the modern technology will reduce water use per unit area under high EMP, but may increase water use under low EMP; (b) where the modern technology increases yield, its adoption will save both water and energy if the EMP is high, save water but increase energy use if EMP is low and well depth is between yy and y” (the reduction in energy due to the decreased water demand may be more than offset by the energy required for pressurization), and increase both water and energy use at low EMP and well depths greater than y”; and (c) regions with a relatively higher proportion of low qualityhigh well depth lands are likely to be major adopters of new technologies, while regions with low well depths and high-quality lands will continue to use traditional technologies. 9 Caswell [34] provides a detailed survey of the empirical literature on irrigation technology adoption.

THE ADOPTION

OF AGRICULTURAL

223

INNOVATIONS

small farm size and high water or labor costs, also increase the likelihood of adopting sprinkler irrigation. Lichtenberg [36] examined the impact of land quality on cropping patterns of field crops (maize, sorghum, soybeans, and small grains) and the adoption of the center-pivot irrigation technology in Nebraska using county-level data. He found that land quality-augmenting technologies like center-pivot irrigation tend to be adopted especially rapidly on lower qualities of land and are shown to have induced significant changes in cropping patterns. Nieswiadomy [37] studied the irrigation practices of cotton and sorghum growers in Texas using farm-level data. The study found that the likelihood of adoption of water-saving technologies increases as water and output prices increase and the quality of land decreases. Dinar and Yaron [38] studied the determinants of adoption of sprinkler and drip irrigation technologies in citrus groves in Israel and Gaza. They found that modern irrigation technologies tend to be adopted sooner and to a larger extent on groves located on relatively low-quality land, in regions with higher evaporation rates, on groves planted with more sensitive rootstock, and on groves grown under conditions of restricted water allotments and higher water prices. In summary, while many models of individual adoption behavior in the past followed a static framework, more recent models have sought to incorporate the dynamic features of the adoption decision process. Some models, while describing adoption behavior at the farm level according to static criteria, nevertheless, account for dynamic features, such as the learning effect or the reduction of uncertainty, thereby allowing the prediction of adoption patterns over time at the aggregate/sector levels. The following section reviews representative works theoretically describing aggregate adoption patterns. It presents the new directions being pursued in the study of the diffusion cycle, particularly those of agricultural innovations. REVIEW

OF MODELS

OF THE AGGREGATE

DIFFUSION

PROCESS

Aggregate models of technology diffusion in the past were generally founded upon the epidemic or logistic model, wherein the diffusion process is viewed to be formally akin to the spread of an infectious disease. The analogy is that contact with other adopters and exposure to information on the innovation leads to adoption [39]. The demonstration effects and learning from the experience of others underlie the logistic model. It assumes a homogeneous population whose members have an equal probability of coming into contact with each other. The logistic model follows the general form:

an, QN-

at=

n,)

(1)

where n, is the number of individuals who have adopted the innovation at time t, N is the fixed population of potential adopters, and S is a parameter reflecting the rate of adoption (probability of adopting upon learning about the innovation). Given a fixed value of l3, the number of additional adopters at any time t is a function of the total number of adopters, so that the basic characteristic of the process is imitative behavior [2]. The value of l3 will depend upon such factors as the nature of the specific innovation, economic factors, the social system in which it is introduced, and the channel and change agent used to diffuse it. However, the absolute increase in the number of adopters at any point in time is the product of opposing forces; as the proportion of the number of adopters increases (MN), the number of potential adopters (N - n,) declines. This results in a bell-shaped frequency distribution of adopters over time and a sigmoid cumulative

224

G. FEDER

AND D.L. UMALI

density function of the logistic frequency distribution. loThe logistic model, thus, imposes a symmetric S-shaped diffusion trend which attains a maximum diffusion rate when 50% of the potential cumulative adopters have adopted the innovation [40]. The seminal works of Griliches [41] on hybrid corn adoption in the United States and Mansfield [42] on interfirm diffusion of 12 innovations in four US industries are founded on this S-shaped diffusion path. Recent studies have further sought to extend the basic logistic model. Doessel and Strong [43], in investigating the diffusion of a new pharmaceutical drug, modify the logistic model to incorporate uncertainty in the potential user population and population variability during the diffusion period. Knudson [44] relaxes the assumption of a fixed adoption ceiling and allows for the possibility of disadoption and changes in the technology. She then applies the modified logistic model to data on semidwarf wheat varieties in the United States; the results suggest a better fit to the data as compared to the standard logistic model. Gore and Lavaraj [45] use the logistic model to describe diffusion between two spatially separated groups. Diffusion in the first group (within the town) follows the logistic model, while diffusion in the second group (a village outside of town) is a function of information received from adopters within the village and the town. The model is applied to data on the diffusion of crossbred goats in villages in Pune, western India, and the results yield only a marginal improvement over the logistic model. As an alternative to logistic growth patterns, the symmetry of which does not always fit observed patterns, a family of exponential growth models, which Gregg et al. [46] refer to as the “modified exponentials,” have been developed to explain the diffusion process. These include the Gompertz, the log-normal, and the flexible logistic or FLOG models.” The Gompertz curve imposes an asymmetric (positively skewed) trend; it attains its point of inflection when diffusion has reached approximately 37% of the upper bound. Dixon [48] employed the Gompertz model when reworking Griliches’s study [41] and found that 21 of the 31 states studied had a positively skewed diffusion curve so that the Gompertz curve was more appropriate than the logistic curve. The skewness arose because the data subsequently showed that adoption had exceeded the ceiling assumed by Griliches. Griliches [49] in a reply noted that the diffusion curve was not asymmetric once it is recognized that the adoption ceiling should be specified as a function of economic variables that change over time. Metcalfe [50] continued the debate and argued that instead of a single diffusion curve, there exists an envelope of successive diffusion curves, each associated with a given set of innovations and environmental characteristics, adoption ceiling (N) and coefficient of adoption @). The cumulative log-normal is another member of the exponential growth curves which can reproduce a whole family of asymmetric S-shaped curves, because the inflection point is variable. Maddala [Sl] argues that since many economic variables cannot take negative values and do not have symmetric distributions as the normal, the log-normal distribution may be more appropriate in some economic applications. Bewley and Fiebig [52] criticized the logistic and Gompertz models because of their rigidity. They developed the FLOG model whose point of inflection and degree of symmetry are determined by I0 Solving for n, of the logistic differential

where a is the constant of integration. I’ For a detailed survey of the exponential

equation

generates the result:

growth models, see Meade [47] and Thirtle and Ruttan 121.

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

225

the data set rather than imposed. The model is used to analyze the diffusion of telecommunication technology. The underlying assumptions of exponential growth models, namely, a homogeneous population, complete intermixing of members of the population, and a singular source of information have been questioned in several studies. Lekvall and Wahlbin [53] point out that internal (learning from other adopters) and external (learning from sources other than the adopters, that is, the mass media) sources both shape the diffusion process. The effects of heterogeneity of the population are explored separately by Coleman [54] and Davies [55]. In response to these criticisms, the specifications of endogenous and exogenous sources of information and heterogeneity of the population have been combined in a “new product growth model” generally attributed to Bass [56] and further developed by Mahajan and Schoeman [57]. Essentially, the Bass model classifies the population into two groups: the innovators, who adopt the new technology independently of others in the system, arriving at their decision based on exogenous information, and the imitators, who are influenced by those who have adopted. It has the general form:

2 =flow-n,) (innovator)

+ h(N-n,) (imitator)

where /30is the “coefficient of innovation,” or the rate of adoption of the proportion of the population whose adoption decision is influenced by exogenous information, and B, is the “coefficient of imitation,” or the rate of adoption of the population whose adoption decision is based on personal interaction [58]. When PO = 0, the Bass model reduces to the logistic model. Akinola [59] applied Bass’s innovator-imitator model to analyze the diffusion of cocoa-spraying chemicals among Nigerian cocoa farmers. His results show that although the Nigerian data fit the model well, statistical evaluation using various criteria (that is, R*, the correlation between actual and predicted values, and long-term forecasting efficacy) shows little improvement over the standard logistic model. Recent extensions of the Bass model relax some of the original assumptions. Sharif and Ramanathan [60] show that the assumption of a constant equilibrium number of potential adopters does not always hold because if the demand for the innovation is a function of the demand for the output (as is the case of a process innovation), then if the demand for the output increases over time (for example, as a result of population growth), this will tend to increase the demand for the innovation as well. Akinola [61] models such as a case wherein the equilibrium number of potential adopters and the adoption parameters (SO and l3]) vary over time. Empirical results indicate that the data on the diffusion of cocoa-spraying chemicals among Nigerian cocoa farmers fit the model fairly well [61]. Karshenas and Stoneman [62] similarly allow POand 8, to change over time and disaggregate the innovators further into two groups: those who continue to be instrumental in the learning process and those who no longer contribute to the learning process. The influence of economic factors (disposable income, the price of the product, and credit conditions) on the speed of diffusion are incorporated through their influence on PO and PI. The model was applied to the diffusion of color television in the United Kingdom and was found to outperform the logistic and Gompertz models. It was found that the exogenous factors dominate the diffusion process. Tanny and Derzko [63] developed a model similar to that of Gore and Lavaraj [45] described earlier, which they call the two-compartment model: the first compartment is composed of innovators while the second compartment is composed of imitators. In the second compartment, adoption

226

G. FEDER

AND D.L. UMALI

results from the interaction with other imitators who have already adopted, from interaction with innovators from the first compartment, or from exogenous sources of information. The model is applied to data on new product sales of selected consumer goods. They found that “the two compartment model fails to provide superior structure [relative to the Bass model] for explaining observed data on new product diffusion” [63, p. 2311. Most economic literature has concentrated on modeling the diffusion cycle of commodities. Little work has been done, however, on examining the rate and time at which a technology will be abandoned. Dinar and Yaron [64] developed one of the first models describing the diffusion and abandonment process of agricultural technologies. They define the technology cycle as the time period between the adoption of a particular technology by the decision maker (firm or household) and its abandonment or replacement by another technology.” The model also assumes that each technology is associated with a given life span (cycle) which does not change over time as a result of market changes. Using data on the adoption of seven irrigation technologies for citrus production from Israel and Gaza,u they first estimate the technology cycle for each technology. Economic variables (output price, water price, government subsidies) are used to estimate the logistic diffusion equation. Using the estimated technology cycle, they subsequently estimate the process of diffusion and abandonment of technologies and of technologies already in the process of renouncement using simulation programs. Their results suggest that the technology cycle is dependent only on the technology and not on physical conditions (for example, weather or soil types) and is similar in length for all regions. The model also predicts the year of full abandonment of each technology. The use of the estimated equations for policy purposes suggests that water pric_e and irrigation equipment subsidies can be used to control the speed and ceiling of irrigation technologies. In summary, the various models of diffusion are founded on the spread of information either through interpersonal interactions between adopters and nonadopters or through exogenous sources. The type of potential adopters may follow from the source of information; that is, imitators are influenced by endogenous information, while the innovators are influenced by exogenous information. Diffusion has been modeled to account for changing equilibrium populations, changing technologies, changing rates of adoption, spatial differences, and the rate of abandonment. However, it is apparent that no general model perfectly fits all situations and that in some cases [59, 611 different diffusion models can describe a single event effectively. Empirical

Studies of Adoption

DETERMINANTS

OF TECHNOLOGY

ADOPTION

As environmental conservation issues have come to the forefront, increasing focus is being directed to the study of the determinants of environmental conservation technology diffusion. Another area that has received considerable focus is that of soil conservation. Forster and Stem [66], Baron [67], Ervin [68], and Norris and Batie [69] found that older farmers are less likely to use soil conservation practices because of their shorter planning horizons and the less than perfect capitalization of yield changes in land prices. Moreover, younger farmers may be more educated and more involved with more innovative farming. I2 This concept has been used broadly in models of equipment replacement that suggest replacement patterns for equipment used by identical producers 1651. I1 The technologies include traditional irrigation such as border and furrow, hand-moved sprinklers (aluminum pipes), solid-set sprinklers above canopy, drag-line sprinklers under canopy (plastic pipes), solid-set sprinklers under canopy (plastic pipes), low-volume microsprinklers and microjets, and drip irrigation.

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

221

Forster and Stem [66], Baron [67], Ervin and Ervin [70], and Norris and Batie [69] found that education has a positive impact on soil conservation technology adoption. The recognition of the erosion problem has been found to positively influence conservation behavior [69-721, while farm size has been found to have a positive influence on the adoption of conservation practices [67, 69, 72, 731. Income has also been found to have a positive influence on adoption of erosion control practices [69, 731. It is generally held that renters of farmland are less likely to invest in conservation practices; however, Lee and Stewart [74] found that renters were more likely to use conservation tillage than full owners. A recent study by Sureshwaran et al. [75] examined the determinants of adoption of sloping agricultural land technology (SALT) in 14 villages in Leyte, Philippines.i4 Contrary to expectations, they found that income and education did not have a significant impact on adoption intensity. Studies of the early phases of diffusion processes show that farm size, tenure status, education, access to extension services, and credit were major determinants of the speed of adoption by various users [l, 761. Recent studies examining these factors at a phase when the technology has reached the final stage of the diffusion process indicate that the impact of many of these factors is no longer significant. Ramasamy et al. [77] found that tenure, education, and farm size were not significant determinants of MV rice adoption in Tamil Nadu, India. David and Otsuka [78] and Otsuka and Gascon [79] found that farm size and tenure were not significant determinants of MV rice adoption in the Philippines. Upadhyaya et al. [80] found similar results for MV rice adoption in Nepal. Otsuka and Gascon [79] also found that age and education were not significant determinants of MV rice adoption. Alauddin and Tisdell [81] found that although large farms were early adopters of MV rice in Bangladesh, smaller farms rapidly caught up such that farm size is not a significant factor in technology adoption in the later phases of the adoption process. They also found that the intensity of adoption of MVs and related technologies was generally higher for small farms than large farms at any point in time. CLIMATIC

AND

INFRASTRUCTURE

IMPACT

ON DIFFUSION

RATES

The long-run upper limit or adoption ceiling is hypothesized to be determined by the economic characteristics of the new technology and by the state of the economy. Completion of the diffusion cycle may be observed with some farmers remaining as nonadopters. Several empirical studies attempt to explain this phenomenon. Several recent studies initiated by the International Rice Research Institute (IRRI) examined the adoption patterns for MV rice in several Asian countries. The studies focused on areas of differing natural and socioeconomic endowments. Generally, it was found that the production environment, and in particular, water availability and control which depend on irrigation and natural conditions such as topography and rainfall patterns, was the most important factor explaining differential adoption patterns [77, 78, 80, 811. In addition, David and Otsuka [78] and Ramasamy et al. [77] found that relative factor prices (fertilizer/paddy price ratio) were important determinants of adoption of labor-saving technologies. Jansen et al. [82] posit that complete diffusion may not be achieved until the introduction of second-generation technologies. For example, under the idealized case of a homogeneous region, the first released groups of MVs are uniformly economically superior to local varieties. Consequently, the long-run 100% level of adoption can be quickly I4SALT involves planting contours across the slopes in the uplands with hedgerows of fast-growing legumes like the giant Ipil-ipil (Leucaena kucocephala). The hedgerows reduce soil erosion, and the Leucaena, while building soil fertility, can also be used as timber, pulpwood, and fodder for livestock.

G. FEDER

228

AND D.L. UMALI

attained. In contrast, in a heterogeneous region, the first-generation MVs will be more profitable than the local varieties in only some selected areas, most likely in the more favorable, higher potential areas, so that the adoption rate falls below 100%. As Griliches [49] points out, the slow upper tail may be associated with the lack of well-adapted MVs, thereby affecting the estimated speed of acceptance of the innovation in the later stages of the adoption cycle. However, if the introduction of a second-generation MV is successful in addressing the location-specific problems of earlier MVs, full adoption may still occur, but actually on a secondary diffusion path.‘j Jansen et al. [82] use a modified version of Griliches’ [41] two-stage estimation approach to examine adoption behavior. The first stage involves fitting a logistic curve to historical diffusion data (secondary district data on coarse cereals, that is, sorghum, millet, and maize in India) to estimate the diffusion speed and the adoption ceiling. The second stage involves the examination of the determinants of cross-district variations in adoption. For sorghum and millet, the estimated adoption ceilings were statistically significant at the 5% level in 71% and 58% of the districts respectively. In the case of maize, only 36% of the districts displayed statistically significant adoption ceilings. They attribute this poor performance to the release and partial adoption of “second-generation” shorter duration MVs in some districts of north India. Their results also show that infrastructural variables (use of irrigation, access to fertilizer, markets and roads, population per unit area) explained a large share of the variation in MV adoption, particularly for pearl millet and maize, but agroclimatic variables, particularly excess moisture, had a higher explanatory power. These results are consistent with those obtained in the IRRI studies on MV rice adoption under different production environments. TECHNOLOGY

ADOPTION

AND

EQUITY

CONSIDERATIONS

The welfare implications of technology adoption are a continuing concern to many social scientists. The equity impact of modern rice varieties, in particular, has received considerable attention. Most of the MVs that have been developed have been more suited to irrigated and favorable rain-fed conditions with adequate water control [83]. Consequently, the less favorable production environments where farmers are generally poor have often been bypassed [76, 841. MVs require greater liquidity to finance the higher demand for fertilizer and other purchased inputs and may, therefore, favor larger farmers who have better access to credit markets [SS, 861. It has been observed that the spread of MVs precedes a wider adoption of labor-saving technologies, for example, tractors, threshers, and direct seeding [87]. Thus, although the adoption of MVs results in increased labor use per unit area by increasing the labor requirements of crop management and harvesting [88], if MV adoption induces the use of labor-saving technologies, then the welfare of the landless households will be adversely affected [78]. These conclusions were drawn from analysis conducted when the diffusion process had not reached its equilibrium. The studies conducted during the final stages of the diffusion process, however, showed mixed results. Hazel1 and Ramasamy [89] found that while MV rice used 5% to 10% more labor per hectare than locally improved varieties, the green revolution did little to increase labor employment. The increased mechanization of irrigation pumping and paddy threshing were not adequately offset by the substitution of paddy for groundnuts in the cropping pattern. I6 On the other hand, David and Otsuka [78] and Ramasamy et al. [77] investigated the same question, using farm household data from the Philippines Is Jansen et al. (821 also note that expanding and deepening rural infrastructure in the adoption ceiling. I6 Groundnuts use about half to two-thirds as much labor per hectare [SS].

could lead to an increase

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

229

and Tamil Nadu, India, and found that MVs do not directly induce the adoption of labor-saving technologies such as tractors and threshers. Rather, relative factor prices are the major determinants of the adoption of these technologies. They point out, however, that MVs indirectly promoted tractorization by raising the profitability of irrigation investments, which is a significant factor inducing higher cropping intensity. Otsuka and Gascon [79] also found no clear association between MV rice adoption in Central Luzon, Philippines, and the adoption of tractors. Methodological Issues In empirically testing for the determinants of the intensity of use of an innovation, studies in the past typically used some form of the linear, log-linear, or semilogarithmic regression equation, and the parameters were estimated using ordinary least squares (OLS) [l]. For agricultural innovations, adoption was usually expressed as the percentage of area cultivated with the new technology over total cultivated area. Nonadopters were frequently excluded from the sample, thus resulting in sample selection bias and consequent biases in the estimated coefficients. Even in cases where nonadopters are included, the parameter estimates will still be biased and inconsistent, since the clustering of observations violate the OLS assumption of a continuous variable and there is a high probability that the predicted value of the dependent variable is negative, which is nonsensical [90]. OLS estimation of equations with a dichotomous or otherwise limited dependent variable is not appropriate either, because the error structure is heteroscedastic and the resulting parameter estimates will be inefficient. The classical hypothesis testing cannot be applied either because the error terms for a limited dependent variable will not be normally distributed [91, 921. In the case of dichotomous adoption decisions, one can use logit or probit models. The adoption variable is expressed in binary form (1 if the farmer adopts, 0 otherwise); if the error terms are assumed to follow a normal distribution, the result is the probit model, and if they follow a logistic cumulative distribution, the result is the logit model [92]. Application of these approaches can be found in Jamison and Lau [93], who analyzed the adoption of chemical inputs by Thai farmers; Rahm and Huffman [94], who studied the adoption of reduced tillage in Iowa; Duraisamy [95], who investigated the determinants of adoption of MV rice under single and multicrop production in India; Harper et al. [96], who examined the adoption of the sweep net/treatment threshold and rice stinkbug spraying technologies in Texas; and Strauss et al. [97], who studied the role of education and extension on the adoption of new technologies (soil analysis, MV seeds, fertilizer, planting method, disease control) in the production of rice and soybeans in central-west Brazil. The analysis of the binary adoption decision is not appropriate when adoption also involves the simultaneous decision regarding the intensity of utilization if adoption occurs. Consequently, the tobit model, also known as the censored normal regression model, has been increasingly used in adoption studies in recent years. It assumes that variables have a lower (upper) limit and take on this limiting value for a substantial number of respondents. For the remaining respondents, the variables take on a wide range of values above (below) the limit [98]. The explanatory variable is thus expected to influence both the probability of the limit responses and the size of the nonlimit responses. Because observations beyond some limit are omitted, the tobit model accounts for the fact that the error terms will follow a truncated normal distribution. In the case of adoption analysis, the tobit model takes account of both the truncated adoption distribution and the intensity of adoption. This approach was applied by Akinola and Young [99] in the analysis of the adoption of cocoa-spraying chemicals in Nigeria; Norris and Batie [69]

230

G. FEDER AND D.L.

UMALI

and Sureshwaran et al. [75] for farmer adoption of soil conservation technologies in Virginia and the Philippines respectively; and Upadhyaya et al. [80], Otsuka and Gascon [79], and Lin [lOO-1021 for farmer adoption of MV rice in Nepal, the Philippines, and China, respectively. Policy Interventions and Technology Adoption New technologies offer opportunities for increasing productivity and incomes and for improving product quality. However, what determines the actual improvements in productivity and product quality, thereby enhancing economic welfare, is not the rate of development of new technologies, but the speed and extent of their application in commercial operations [103]. In a general sense, the faster a superior technology is diffused, the larger the improvement of social welfare, as higher income (or larger consumption) can be enjoyed earlier. Experience has shown, however, that several factors can constrain technology adoption: lack of credit, limited access to information and inputs, and inadequate infrastructure. Furthermore, the nature and intensity of the impact of these constraints may vary according to the type of technology, for example, a divisible versus a lumpy technology. Some of these problems arise as a result of market failures, leading to a loss of potential welfare gains. To overcome these constraints, governments have generally pursued two general strategies: information provision (for example, extension programs) and the provision of subsidies and support programs (output, input, and credit subsidies, the provision of complementary infrastructure, and risk-reducing programs). Whether these policies are effective in fostering technology adoption and, if effective, what their optimal levels are, have been issues that have drawn considerable attention. Feder and Slade [104, p. 4231 postulate less than socially optimal levels of adoption by farmers to “the divergence between the true (objective) distribution of net benefits This divergence can be traced to limited and the perceived (subjective) distribution.” information, which affects farmers’ perception of the risks (for example, yield, input and output price) involved in adopting the new technology. The impact of risk and uncertainty on technology adoption has been extensively studied, and Feder et al. [l] provide a comprehensive survey. The rationale for public sector intervention, therefore, rests on the need to eliminate or at least to minimize this divergence. Feder and Slade [ 1041 explore various policy options in the case of a farmer whose decision behavior is described by a mean-variance utility function with risk aversion. The farmer has to decide how much land to allocate to the new technology, whose yield is subject to variability. Because the farmer’s perception of yield risks is an overestimation of the true risks (by a factor 6), utility maximization results in less than socially optimal levels of adoption. It is shown that several policy options can be used to reduce 6 and thus achieve optimal adoption: an output price subsidy, a subsidy on the cost per acre, and per unit area subsidy. Feder and Slade [30] also highlight the issue of externalities associated with the dissemination of information. Early adopters provide information and the learning that affects subsequent adopters. However, the effects on other future adopters are not taken into account by the first adopters. Thus, there is a positive externality which can be corrected by, for example, subsidies to the early adopters. Feder [3] investigates the impact of policy intervention in the case of the adoption of complementary technologies (a divisible and a lumpy technology) under uncertainty. He finds that an input subsidy on the divisible technology results in higher utilization of the divisible input (fertilizer) and the area allocated to it, but its impact on the adoption of the lumpy technology (tubewell) is conditional on its effect on the farm size threshold which determines the lumpy technology

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

231

adoption. A price subsidy affects the divisible input adoption in the same direction as the input subsidy, although its effect on the lumpy technology adoption is not necessarily positive. Under conditions of a binding credit constraint, a output price subsidy will reduce the volume used per unit area of the divisible input, but will increase the share of the area grown with the divisible input (whether the lumpy technology is adopted or not). A divisible input subsidy will increase the farm size threshold for the adoption of the lumpy technology, thereby inhibiting its adoption. The impact of government price stabilization activities is further investigated by Kim et al. [lo51 in the case of intrafirm diffusion of a yield-increasing vs. a cost-reducing technology.” Assuming a single farm with fixed landholdings, no adjustment costs, and output price as a random variable, the farmer must allocate his landholdings between the two innovations.r8 The farmer maximizes the expected utility from his or her income from the new technologies. They show that as price variability decreases, ceterisparibus, the proportion of land allocated to the yield-increasing technology increases. Conversely, the land allocated to the cost-reducing technology decreases. Modeling the long-run equilibrium choices of the firm, they assume an adjustment cost associated with the yieldincreasing technology, which is a function of the size of the landholding. They find that lower price variability will increase the speed of diffusion of the yield-increasing technology and will reduce that of cost-reducing technologies. A common approach to overcoming farmer credit constraints in the past has been to provide directed subsidized credit. These programs typically have not led to a significant increase in the adoption of technologies because they have not adequately addressed the risk-bearing aspect of technology adoption and, in fact, have at times compounded the riskiness of adoption by adding financial risks [106]. Several studies [107, 1081 suggest that opportunities to spread risks to other parties are insufficient, thus limiting farmer use of credit and the adoption of new technologies. Krause et al. [ 1061 therefore investigate the efficacy of three risk-sharing alternatives that the government can undertake instead: a risk-bearing credit system with the repayment program adjusted against the amount of rainfall, a scheme of risk sharing by hired laborers wherein part of the laborers’ wages paid at harvest are proportional to the revenues of the cropping year, and a risk-sharing scheme with fertilizer suppliers wherein payment for the inputs is proportional to the revenues from the fertilizer-using technology. I9These schemes are applied in the context of the adoption of new cultivation practices for cowpea and millet in Nigeria. The results of the study indicate that offering low-interest credit is a relatively ineffective strategy in encouraging the adoption of agricultural technologies: substantial changes in the interest rate result in only modest changes in the rate of adoption. However, all three proposed schemes increased technology adoption. When at least the principal was repaid, regardless I7 A technological innovation is “yield increasing” if it increases yield per unit area and does not reduce optimal variable costs per unit area (for example, hybrid livestock) and “cost reducing” if it reduces optimal variable costs per unit area but does not increase yield per unit area (for example, improved feed formulation). These two types of technologies have similar effects on profits: both reduce average total costs, but while yieldincreasing technologies reduce average ftxed costs, cost-reducing technologies reduce the average variable cost [105]. I8 The traditional technology will not be preferred under conditions of zero adjustment costs. I9 Krause et al. [106] utilize discrete stochastic programming (DSP) to obtain a simultaneous focus on both technology adoption and credit. The DSP model incorporates choices of crop mix, production practices (the use of fertilizers, planting densities, and intercropping) and the level of credit from traditional and government sources. A biological simulation model is used to estimate yields (millet and cowpea production). The farm planning model maximizes expected utility, which is a function of wealth generated by each state of nature, the probability of each state of nature, and the risk aversion coefficient.

232

G. FEDER

AND D.L. UMALI

of weather under the risk-sharing credit program, it achieved the same level of adoption as a lower interest rate under the traditional credit system. Risk sharing by laborers produced substantial or modest increases in technology adoption depending on the premium offered to the laborers to participate in the program. The most effective system was risk sharing with fertilizer suppliers. The system produced significant increases in the area planted with reasonable increases in government borrowing. Lunn [109] investigates the merits of tying agreements (for example, leasing contract) as a strategy for reducing risks. He shows that by leasing the good at a relatively low rate and tying the sale of a complementary good at above the market price to the lease, the lessor reduces the risk to the lessee and more rapid diffusion of the innovation occurs. The effectiveness of policy intervention in influencing the extent and speed of adoption is also determined by the structure of the market. Stoneman and David [103] investigate whether and/or under what market conditions subsidies and information provision will affect the speed and extent of diffusion and whether and/or under what conditions the impacts of these policies are socially desirable. Their model assumes two periods, a new process that can be acquired by the purchase of one new capital good, a fixed population of potential adopters, a logistic diffusion curve, rational profit-maximizing users who have perfect foresight of output prices, and suppliers who have perfect knowledge of the technology demand environment. Under a competitive environment, the price of the technology will equal the marginal cost of producing the technology in each period. In the presence of a monopolist supplier and in the absence of learning about the effect of the first-period sales on the second-period demand, the monopolist output and consequently the number of adopters will be less than the competitive case in both periods.20 If learning occurs, the first-period monopoly output may be greater than, less than, or equal to the competitive case. In the absence of any subsidy and under conditions of competitive supply, the benefits obtained by the marginal adopters in both periods equal the opportunity cost of their provision. Under monopolized supply, the benefits exceed the cost of provision in period 2, while in period 1, if the learning effect is strong (weak), the marginal adopter receives benefits that are less (greater) than the opportunity costs of provision. Subsidization policies have two effects: the incentives cause movement down the benefit distribution, thus extending usage, and the increased usage generates information that stimulates further use. Stoneman and David [103] show that under competitive conditions, subsidy policies increase technology adoption but may not be welfare increasing because such policies encourage use by firms for whom the benefits are exceeded by the cost of their provision.” Under monopolistic conditions, a subsidy will likewise increase usage. However, the welfare effect depends on the learning effect between the two periods. Alternatively, a policy of information provision may be pursued. Increased awareness is incorporated in the model by means of either increasing the proportion of the population who know about the technology in the first period or increasing learning 2oWith a monopolist supplier, the monopolist will choose an output that equals the marginal cost in the second period. In the first period, the monopolist chooses an output level taking into account the effect that the first-period sales will have on the second-period demand through learning. In the absence of learning, monopolist output would be determined such that first-period marginal revenue will equal the opportunity cost of supply both periods. ‘I The net effect on welfare of subsidies under competitive conditions cannot be signed because movements down the benefit distribution during the two periods result in welfare losses, but the increased intensity of use in the second period implies a welfare gain. Thus, the policy would achieve the aim of increased usage but would not necessarily increase welfare [ 1031.

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

233

effects between periods. Their results indicate that under competitive conditions, the extra information provided in periods 1 and/or 2 increases the intensity of use of the new technology within unchanged margins. The marginal adopters in both periods have the same benefits, but a greater proportion of intramarginal adopters make use of the technology as information increases. Stoneman and David therefore conclude that information provision under competitive conditions increases both usage and welfare. Under monopolistic conditions, the monopolist supplier reacts to government intervention, which could counteract the intent of the policy and thus definitive welfare implications could not be inferred. Increased learning in period 2 increases the number of intramarginal adopters and under monopolistic conditions represents an increase in welfare. Multiperiod approaches to similar issues can be found in David and Olsen [llO] and Stoneman and Ireland [21]. Stoneman and Ireland [21] also examine the effect of market structure on the speed of diffusion. They find that the optimal diffusion path is generated by either a competitive supply industry with perfect foresight or by a monopoly supplying an industry with buyers who are myopic. Myopia with a competitive supply industry, however, yields diffusion that is too fast, while perfect foresight with monopoly supply yields diffusion that is too slow. Policies designed to correct market failure which constrain technology adoption may themselves lead to distortions and thus resource misallocation. Miller and Tolley [ill] address this tradeoff between adoption benefits and resource misallocation. Using a fourpart welfare expression based on the profitability of production using the old and new technology, before and after price intervention, they derive the optimizing conditions for the degree of price intervention (price support and fertilizer subsidy) to maximize net welfare gains, the length of time for the price intervention program for a new technology, and the costs to the government. Their results suggest that output price and fertilizer subsidies can affect the rate of adoption, but still have only a minor effect on social welfare because the movement away from an old technology may result in a welfare gain that is partially offset by the inefficiency in resource use. Mesak and Coleman [ 1121 also developed a model to estimate the optimal price intervention policy. They assume that the optimal subsidizing policy and the optimal rate of government spending are a function of the extent of market adoption of the new technology; they apply the model to data on the adoption of photovoltaic systems in Kuwait. Policy interventions can also be used to reduce the use of technologies that may exhibit some negative externalities or to promote resource conservation. The increasing depletion of water resources in addition to the problems of salinity, waterlogging, and water contamination associated with irrigation have highlighted the need for irrigation technologies with greater water use efficiency. Caswell et al. [113] extend the model developed by Caswell and Zilberman [33] for analyzing the adoption of input-conserving technologies (drip and sprinkler irrigation systems) to include the spillover effects of water use and the role of environmental policies in conserving resources and reducing pollution. The study introduces a pollution coefficient, which is a function of land quality, to the farmer’s profit-maximizing equation and examines the impact of a pollution tax per unit of drainage effluent on technology adoption. The study showed that an increase in the pollution tax tends to increase the threshold, as, for the land quality at which a switch from traditional to modern technology occurs, thus encouraging owners of high-quality land to switch to the modern technology. On the other hand, a higher pollution tax will increase the land quality threshold at which production ceases because operational profit is zero. The pollution tax may, therefore, encourage the adoption of modern technologies in locations where it has not yet been adopted before, but may reduce the extent of

234

G. FEDER

AND D.L. UMALI

irrigated agriculture. The pollution tax may also increase overall output, because of the higher yield associated with the switch to the modern technology. Overall, the implementation of a pollution tax is likely to reduce water use and pollution on farms using traditional technology, encourage the adoption of the modern, less polluting technology, and encourage the retirement of low-quality lands. Caswell et al. simulated the effects of a pollution tax on technology choices (furrow, shortened furrow, sprinkler and drip irrigation), crop yields, water use, and drainage levels of cotton producers in the San Joaquin Valley. The stimulation results showed that increases in output prices affect technology choices mostly by extending the range of water prices and drainage (pollution) charges under which operations are profitable. However, although the greater efficiency of drip and sprinkler irrigation results in the highest levels of water and drainage fee savings, these savings are insufficient to offset the increased fixed costs associated with their adoption. Consequently, the output price must be relatively high before these more efficient technologies will be adopted. They demonstrate that an effective way of achieving greater reduction in water use and drainage is to reduce the fixed costs incurred by farmers in adopting the modern technology. An annual subsidy increases the profitability of the modern technology even for very low output and water prices. The combination of a subsidy on the fixed costs and a drainage (pollution) charge will further decrease the requisite subsidy. Dinar and Yaron [38] examined the use of other intervention measures to promote the adoption of sprinkler and drip irrigation technologies in citrus groves in Israel and Gaza. The study found that water price and water quota rates can be substituted to achieve similar adoption levels: increasing both the water price and the water quota can result in the same rate of conversion of regional citrus area to modern technologies. In summary, the literature suggests that different government policy interventions clearly serve as effective devices for promoting the adoption of new technologies or discouraging the use of traditional technologies. However, speeding the process of diffusion is not always welfare increasing. Conclusions and Implications for Further Research Much of the research on the adoption of agricultural technologies has paralleled the progress made in the diffusion cycle of the Green Revolution technologies first introduced almost three decades ago. Often, the factors that were empirically found to be critical determinants in the initial phases of adoption (for example, farm size, credit, tenure, and education) have faded into insignificance in the later stages of the diffusion cycle. A clearer, though still imperfect understanding of the complex adoption cycle is emerging. What the true shape of the diffusion cycle is and whether a singular pattern holds true in every adoption situation remains a serious topic of debate. It was also shown that policy interventions to promote technology adoption are not always welfare increasing. Moreover, the type of technology and its interactions with related technologies, the structure of the market, and the nature and length of the policy intervention are major considerations. From the survey of the technology adoption literature, several issues requiring greater research efforts come to the forefront. These issues are discussed below. Significant effort has been invested by scholars in analyzing the determinants of the individual agent’s adoption behavior and in identifying the role of various constraints and infrastructure, environmental, and climatic factors in the adoption process. In parallel, a large body of literature has developed on aggregate diffusion patterns, and empirical specifications have been developed, which have the capacity to fit observed data well.

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

235

While the parameters of these specifications are often given plausible economic interpretations, the literature contains few models that link in an explicit and rigorous fashion the microbehavioral model with the aggregate diffusion model. Such a linkage will enable a more precise interpretation of parameters in the aggregate diffusion model (for example, how the adoption rate ceiling is determined) and will facilitate the analysis of policy instruments. Market failures constraining technology adoption may be corrected through appropriately designed public policy interventions. However, the determination of appropriate policy options and their optimal intensity and duration require greater study, if resource misallocation is to be avoided. Empirical evidence on policy effectiveness is limited and merits more attention as well. Whether technologies are adopted in packages, individually, or in combination following a sequence also requires further research, because empirical evidence has provided mixed results. A fuller understanding of this process is essential, because it will contribute to the more cost-effective design of government policy interventions. The identification and timing of appropriate “packages of policy interventions,” conforming with the farmer’s technological adoption chain, can substantially increase the efficacy of such interventions. References 1. Feder, G., Just, R. E., and Zilberman, D., Adoption of Agricultural Innovations in Developing Countries: A Survey, Economic Development and Cultural Change 33, 255-298 (1985). 2. Thirtle, C. G., and Ruttan, V. W., The Role of Demand and Supply in the Generation and Dt#usion of Technical Change, Harwood Academic Publishers, London, 1987. and the Impact of Risk, 3. Feder, G., Adoption of Interrelated Agricultural Innovations: Complementarity Scale and Credit, American Journal of Agricultural Economies 64, 94-101 (1982). 4. Feder, G., and O’Mara, G., Farm Size and the Adoption of Green Revolution Technology, Economic Development and Cultural Change 30, 59-76 (1981). 5. Leathers, H. D., and Smale, M., A Bayesian Approach to Explaining Sequential Adoption of Components of a Technological Package, American Journal of Agricultural Economics 68, 519-527 (1991). G., Risk, Schooling and the Choice of Seed Technology in Developing 6. Pitt, M. M., and Sumodiningrat, Countries: A Meta-Profit Function Approach, International Economic Review 32, 457-473 (1991). 7. Hayami, Y., and Ruttan, V., Agricultural Development: An International Perspective, Johns Hopkins University Press, Baltimore, MD, 1985. 8. Traxler, G., and Byerlee, D., A Joint Product Perspective on the Evolution and Adoption of Modern Cereal Varieties in Developing Countries, Paper presented during the annual meeting of the American Agricultural Economics Association, Baltimore, MD, 9-12 August, 1992. 9. Rauniyar, G. P., and Goode, F. M., Technology Adoption on Small Farms, World Development 20,275282 (1992). 10. Byerlee, D., and de Polanco, E. H., Farmers’ Stepwise Adoption of Technological Packages: Evidence from the Mexican Altiplano, American Journal of Agricultural Economics 68, 519-527 (1986). 11. Iqbal, F., The Demand and Supply of Funds Among Agricultural Households in India, in Agricultural Household Models, I. Singh, L. Squire, and J. Strauss, eds., Johns Hopkins University Press, Baltimore, MD, 1986, pp. 183-205. 12. Lopez, R. E., Structural Models of the Farm Household That Allow for Interdependent Utility and Profit Maximization Decisions, in Agricultural Household Models, 1. Singh, L. Squire, and J. Strauss, eds., Johns Hopkins University Press, Baltimore, MD, 1986, pp. 306-325. 13. Pitt, M. M., and Rosenzweig, M. R., Agricultural Prices, Food Consumption, and the Health and Productivity of Indonesian Farmers, in Agricultural Household Models, I. Singh, L. Squire, and J. Strauss, eds., Johns Hopkins University Press, Baltimore, MD, 1986, pp. 153-182. 14. Singh, I., Squire L., and Strauss, J., Methodological Issues, in Agricultural Household Models, I. Singh, L. Squire, and J. Strauss, eds., Johns Hopkins University Press, Baltimore, MD, 1986, pp. 48-70. 15. Pradhan, J., and Quilkey, J., Modelling the Adoption of HYV Technology in Developing Economies: Theory and Empirics, Paper presented at the conference on Mechanisms of Socio-Economic Change in Rural Areas, 25-27 November, 1991, Australian National University, Canberra, Australia. 16. Rosenberg, N., On Technological Expectations, Economic Journal 86, 523-535 (1976).

236

G. FEDER

AND D.L. UMALI

17. Rosenberg, N., Inside the Black Box: Technology ondEconomics, Cambridge University Press, Cambridge, 1982. 18. Kislev, Y., and Shchori-Bachrach, N., The Process of an Innovation Cycle, American Journal ofAgricultural Economics

55, 28-37 (1973).

19. Feder, G., Farm Size, Risk Aversion Economic

Papers 32, 273-283

and the Adoption

of New Technology

Oxford

Under Uncertainty,

(1980).

20

Just, R. E., and Zilberman, D. J., Stochastic Structure, Farm Size, and Technology Adoption in Developing Agriculture, Oxford Economic Papers 35, 307-328 (1983). 21. Stoneman, P. L., and Ireland, N., Technological Diffusion, Expectations and Welfare, Oxford Economic Papers 38, 283-304

(1986).

22 Feder, G., and O’Mara, Cl., On Information Journal

of Agricultural

23. Tsur, Y., Sternberg,

Economics

and Innovation

64, 145-147

M., and Hochman,

Diffusion: A Bayesian Approach,

American

(1982).

E., Dynamic Modelling

of Innovation

Process Adoption

with

Risk Aversion and Learning, Oxford Economic Papers 42, 336-355 (1990). 24 Smale, M., Competing Explanations for Land Allocation in HYV Adoption Models, Methodological Issues, Paper presented during the annual meeting of the American Agricultural Economics Association, Baltimore, MD, 9-12 August 1992. 25. Bell, C., The Acquisition of Agricultural Technology: Its Determinants and Effects, Journalof Development Studies 9, 123-159

26. Moscardi,

(1972).

E., and de Janvry,

American

Journal

A., Attitudes

of Agricultural

Economics

21. Parton, K. A., Household Goals presented during the Development lia, 25-27 November 1991. 28. Lindner, R. K., Fisher, A. J., and 29. O’Mara, G., The Microeconomics

Towards Risk Among Peasants: 59, 710-716 (1977).

An Econometric

Approach,

and Technology Adoption of Smallholders in Eastern Kenya, Paper Studies Conference, Australian National University, Canberra, Austra-

Pardey, P., The Time to Adoption, Economic Letters 2, 187-190 (1983). of Technique Adoption by Smallholding Mexican Farmers, in The Book of Chat: Programming Studiesfor Mexican Agriculture, R. D. Norton and L. Solis, eds., Johns Hopkins University Press and the World Bank, Baltimore, MD, 1983. 30. Feder, G., and Slade, R., The Acquisition of Information and the Adoption of New Technology, American Journal

of Agricultural

Economics

66, 312-320

(1984).

31. Shumway, C. R., Pope, R. D., and Nash, E. E., Allocatable Production: Implications for Economic Modelling, American

Fixed Inputs and Jointness in Agricultural Journal

of Agricultural

Economics

66, 72-

78 (1984).

32. Leathers,

H. D., Allocable

American

Journal

Fixed Inputs as a Cause of Joint Production:

of Agricultural

Economics

73, 1083-1090

A Cost Function

Approach,

(1991).

33. Caswell, M. F., and Zilberman, D., The Effects of Well Depth and Land Quality on the Choice of Irrigation Technology, American Journal of Agricultural Economics 68, 798-811 (1986). 34. Caswell, M. F., Irrigation Management

Technology

Adoption

Decisions: Empirical Evidence, in The Economics and A. Dinar and D. Zilberman, eds., Kluwer Academic

of Water and Drainage in Agriculture,

Press, Boston, 1991. 35. Negri, D. H., and Brooks, Agricultural

D. H., Determinants of Irrigation Technology Choice, Western Journal of (1990). E., Land Quality, Irrigation Development, and Cropping Patterns in the Northern High

Economics

15, 213-223

36. Lichtenberg, Plains, American Journal of Agricultural Economics 71, 187-194 (1989). 31. Nieswiadomy, M. L., Input Substitution in Irrigated Agriculture in the High Plains of Texas, 1970-80, Western Journal

of Agricultural

Economics

13, 63-70 (1988).

38. Dinar, A., and Yaron, D., Influence of Quality and Scarcity of Inputs on the Adoption of Modern Irrigation Technologies, Western Journal of Agricultural Economics 15, 224-233 (1990). 39. Arrow, K., Classificatory Notes on the Production and Transmission of Technical Knowledge, American Economic Review 81, 29-35 (1968). 40. Mahajan, V., and Peterson, R. A., Models for Innovation, Sage Publications, Beverly Hills, CA, 1985. 41. Griliches, Z., Hybrid Corn: An Exploration in the Economics of Technological Change, Econometrica 25, 501-522

(1957).

42. Mansfield, E., Technical Change and the Rate of Imitation, Econometrica 29, 741-765 (1961). 43. Doessel, D. P., and Strong, S. M., A Neglected Problem in the Analysis of the Diffusion Process, Applied Economics

44. Knudson,

23, 1335-1340

(1991).

M. K., Incorporating

tural Economics

73, 724-733

Technological (1991).

Change in Diffusion Models, American Journal of Agricul-

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

231

45. Gore, A. P., and Lavaraj, U. A., Innovation Diffusion in a Heterogeneous Population, Technological Forecasting and Social Change 32, 163-168 (1987). 46. Gregg, J. V., Hassell, C. H., and Richardson, J. T., Mathematical Trend Curves: An Aid to Forecasting, Oliver and Boyd, Edinburgh, 1964. 47. Meade, N., The Use of Growth Curves in Forecasting Marked Development-A Review and Appraisal, Journal of Forecasting 3, 429-45 1 (1984). Dixon, R., Hybrid Corn Revisited, Econometrica 48, 1451-1461 (1980). Griliches, Z., Hybrid Corn Revisited: A Reply, Econometrica 48, 1463-1465 (1980). Metcalfe, J. S., Impulse and Diffusion in the Study of Technical Change, Futures 13, 347-359 (1981). Maddala, G. S., Econometrics, McGraw-Hill, New York, 1977. Bewley, R., and Fiebig, D. G., A Flexible Logistic Growth Model with Applications in Telecommunications, International Journal of Forecasting 4, 177-192 (1988). 53. Lekvall, P., and Wahlbin, C., A Study of Some Assumptions Underlying Innovation Diffusion Function,

48. 49. 50. 51. 52.

Swedish Journal of Economics 75, 362-377 (1973). 54. Coleman, K. S., Introduction to Mathematical Sociology, Collier Macmillan, London, 1964. 55. Davies, S., The Dtffusion of Innovations, Cambridge University Press, Cambridge, 1979. 56. Bass, F. M., A New Product Growth Model for Consumer Durables, Management Science 15, 215-217 (1969). 57. Mahajan, V., and Schoeman, M. E. F., Generalized Model for the Time Pattern of the Diffusion Process, ZEEE Transactions of Engineering Management 24, 12-18 (1977). 58. Mahajan, V., and Peterson, R. A., First Purchase Diffusion Models of New-Product Acceptance, Technological Forecasting and Social Change 15, 121-146 (1979). 59. Akinola, A. A., An Application of Bass’s Model in the Analysis of Diffusion of Cocoa-Spraying Chemicals Among Nigerian Cocoa Farmers, Journal of Agricultural Economics 37, 395-404 (1986). 60. Sharif, M. N., and Ramanathan, K., Binomial Innovation Diffusion Models with Dynamic Potential Adopter Population, Technological Forecasting and Social Change 20, 63-87 (1981). 61. Akinola, A. A., Dynamic Innovator-Imitator (IN-IM) Diffusion Model, Canadian Journal of Agricultural Economics 34, 113-124 (1986). 62. Karshenas, M., and Stoneman, P., The Role of Exogenous and Endogenous Learning and Economic Factors in the Diffusion of New Technology: An Epidemic Based Study of the Spread of Color Television Ownership in the UK, Warwick Economic Research Papers No. 356, Department of Economics, University of Warwick, Coventry, England, 1990. 63. Tanny, S. M., and Ruttan, V. W., The Role of Demand and Supply in the Generation and Dtj&sion of Technical Change, Harwood Academic Publishers, London, 1987. 64. Dinar, A., and Yaron, D., Adoption and Abandonment of Irrigation Technologies, AgricuNuralEconomics 6, 315-332 (1992). 65. Rifas, B. E., Replacement Models, in Introduction to Operations Research, C. W. Churchman, R. L. Ackroff, and E. L. Arnoff, eds., Wiley, New York, 1957. 66. Forster, D. L., and Stem, G. L., Adoption of Reduced Tillage and Other Conservation Practices in the Lake Erie Basin, US Army Corp of Engineers, Buffalo District, 1979. 67. Baron, D., Landownership Characteristic and Investment in Soil Conservation, United States Department of Agriculture ERS Staff Report No. AGES810911, 1981. 68. Ervin, D. E., Soil Erosion on Owned and Rented Cropland: Economic Models and Evidence, Paper presented at the annual meeting of the Southern Agricultural Economics Association, Atlanta, GA, 1981. 69. Norris, P. E., and Batie, P. E., Virginia Farmers’ Soil Conservation Decisions: An Application of Tobit Analysis, Southern Journal of Agricultural Economics 19, 79-90 (1987). 70. Erwin, C. A., and Ervin, D. E., Factors Affecting the Use of Soil Conservation Practices: Hypothesis, Evidence, and Policy Implications, Land Economics 58, 277-292 (1982). 71. Earle, T. R., Rose, C. W., and Brownlea, A. A., Socio-economic Predictors of Intention Towards Soil Conservation and Their Implication in Environmental Management, Journal of Environmental Management 9, 225-236 (1979). 72. Lasley, P., and Nolan, M., Landowner Attitudes Toward Soil and Water Conservation in the GrindstoneLost Muddy Creek Project, Unpublished manuscript, Department of Rural Sociology, University of Missouri, 198 1. 73. Carlson, J. E., Dillman, D. A., and Lassey, W. R., The Farmer and Erosion: Factors Influencing the Use Of Control Practices, Agricultural Experiment Station Bulletin No. 601, University of Idaho. 74. Lee, L. K., and Stewart, W. H., Land Ownership and Adoption of Minimum Tillage, American Journal of Agricultural Economics 65, 256-264 (1983).

238

G. FEDER

AND D.L. UMALI

75. Sureshwaran, S., Londhe, S. R., Frazier, P., and Pascual, N. P., A Tobit Analysis of Factors Affecting the Intensity of Adoption of a Soil Conversion Technology, Paper presented during the annual meeting of the American Agricultural Economics Association, Baltimore, MD, 9-12 August 1992. 76. Ruttan, F. W., The Green Revolution: Seven Generalizations, International Development Review 19, 1623 (1977). 77. Ramasamy,

Labour

C., Paramisivam, P., and Otsuka, K., Modern Seed-Fertilizer Technology and Adoption of Saving Technologies in Rice Production: The Tamil Nadu Case, Indian Journal of Agricultural

Economics

47, 35-47 (1992).

78. David, C. C., and Otsuka,

K., The Modern Seed-Fertilizer Technology and Adoption of Labour-Saving Technologies: The Philippine Case, Australian Journal of Agricultural Economics 34, 132-146 (1990). 79. Otsuka, K., and Gascon, F., Two Decades of Green Revolution in Central Luzon: A Study of Technology Adoption and Productivity Changes, Paper presented at the IRRI Rice Research Seminar, Social Science Division Paper No. 90-15, International Rice Research Institute, Los Banos, Laguna, Philippines, 1990. 80. Upadhyaya, H. K., David, C. C., Otsuka, K., and Thapa, G., Adoption and Productivity Impacts of New Rice Technology Across Production Environments in Nepal, Social Science Division Paper No. 90-12, International Rice Research Institute, Los Banos, Laguna, Philippines, 1990. 81. Alauddin, M., and Tisdell, C. A., Dynamics of Adoption and Diffusion of HYV Technology: New Evidence of Inter-farm Differences in Bangladesh, Occasional Paper No. 155, Department of Economics, The University of Newcastle, N.S.W., Australia, 1988. 82. Jansen, H. G. P., Walker, T. S., and Barker, R., Adoption Ceilings and Modern Coarse Cereal Cultivars in India, American Journal of Agricultural Economics 72, 653-663 (1990). 83. Barker,

R., and Herdt, R. W., The Rice Economy of Asia, Resources for the Future, Washington, DC, 1985. 84. Lipton, M. P., with Longhurst, R., New Seeds and Poor People, Unwin Hyman, London, 1989. 85. Frankel, F. R., India’s Green Revolution: Economic Gains and Political Costs, Princeton University Press, Princeton, NJ, 1971. 86. Bhalla, S. S., Farm and Technical Changes in Indian Agriculture, in Agrarian Structure and Productivity in Developing Countries, R. Berry and W. Cline, eds., Johns Hopkins University Press, Baltimore, MD, 1979.

87. Jayasuriya, S. K., and Shand, R. T., Technical Change and Labor Absorption in Asian Agriculture: Some Emerging Trends, World Development 14, 415-428 (1986). 88. Barker, R., and Cordova, V. G., Labor Utilization in Rice Production, in Economic Consequences of New Rice Technology, International Rice Research Institute, Los Banos, Philippines, 1978. 89. Hazell, P. B. R., and Ramasamy, C., The Green Revolution Reconsidered, Johns Hopkins University Press, Baltimore, MD, 1991. 90. Goldberger, A. S., Econometric Theory, John Wiley and Sons, New York, 1964. 91. Amemiya, T., Qualitative Response Models: A Survey, Journal of Economic Literature 29, 1483-1536 (1981). 92. Maddala,

G. S., Limited Dependent and Qualitative Variables in Econometrics, Cambridge University Press, London, 1983. 93. Jamison, D. T., and Lau, L. J., Farmer Education and Farm Eficiency, Johns Hopkins University Press, Baltimore, MD, 1982. 94. Rahm, M. R., and Huffman, W. E., The Adoption of Reduced Tillage: The Role of Human Capital and Other Variables, American Journal of Agricultural Economics 66, 405-413 (1984). 95. Duraisamy, P., Human Capital and Adoption of Innovations in Agricultural Production: Indian Evidence, Discussion Paper No. 577, Economic Growth Center, Yale University, New Haven, CT, 1989. 96. Harper, J. K., Rister, M. E., Mjelde, J. W., Drees, B. M., and Way, M. O., Factors Influencing the Adoption of Insect Management Technology, American Journal of Agricultural Economics 72,997-1005. 97. Strauss, J., Barbosa, M., Teixeira, S., Thomas, D., and Junior, R. G., Role of Education and Extension in the Adoption of Technology: A Study of Upland Rice and Soybean Farmers in Central-West Brazil, Agricultural

Economics

5, 341-359

(1991).

98. Tobin,

J., Estimation of Relationships of Limited Dependent Variables, Econometrica 26, 24-36 (1958). 99. Akinola, A. A., and Young, T., An Application of the Tobit Model in the Analysis of Agricultural Innovation Adoption Process: A Study of Cocoa Spraying Chemicals by Nigerian Cocoa Farmers, Manchester Discussion Papers in Development Studies No. 8501, International Development Center, Faculty of EcOnomic and Social Studies, University of Manchester, Manchester, England, 1985. 100. Lin, J. Y., Education and Innovation Adoption in Agriculture: Evidence from Hybrid Rice in China, American

Journal

of Agricultural

Economics

73, 713-723

(1991).

THE ADOPTION

OF AGRICULTURAL

INNOVATIONS

239

101. Lin, J. Y., The Household Responsibility System Reform and the Adoption of Hybrid Rice in China, Journal of Development Economics 36, 353-372 (1991). 102. Lin, J. Y., Hybrid Rice Innovation in China: A Study of Market-Demand Induced Technological Innovation in a Centrally-Planned Economy, Review of Economics and Statistics 74, 14-20 (1992). 103. Stoneman, P. L., and David, P. A., Adoption Subsidies Vs Information Provision as Instruments of Technology Policy, The Economic Journal (Supplement) 96, 142-150 (1986). 104. Feder, G., and Slade, R., The Role of Public Policy in the Diffusion of Improved Technology, American Journal of Agricultural Economics 67, 423-428 (1985). 105. Kim, T. K., Jayes, D. J., and Hallam, A., Technology Adoption Under Price Uncertainty, Journal of Development Economics 38: 245-253 (1992). 106. Krause, M. A., Deuson, R. R., Baker, T. G., Preckel, P. V., Lowenberg-Deboer, J., Reddy, K. C., and Maliki, K., Risk Sharing Versus Low-Cost Credit Systems for International Development, American Journal of Agricultural Economics 72, 911-922 (1990). 107. Binswanger, H. P., and Sillers, D. A., Risk Aversion and Credit Constraints in Farmers’ Decision-Making: A Reinterpretation, Journal of Development Studies 20, 5-21 (1983). 108. Bromley, D. W., and Chavas, J. P., On Risk, Transactions, and Economic Development in the Semi-arid Tropics, Economic Development and Cultural Change 38, 719-736 (1989). 109. Lunn, J., Tie-in Sales and the Diffusion of New Technology, Journal of Institutional and Theoretical Economics 146, 249-260 (1990). 110. David, P. A., and Olsen, T., Anticipated Automation: A Rational Expectations Model of Technological Diffusion, Center for Economic Policy Research Publication No. 24 (CEPR, Technology Innovation Program Working Paper No. 2), Stanford University, Stanford, CA, 1984. 111. Miller, T., and Tolley, Cl., Technology Adoption and Agricultural Price Policy, American Journal of Agricultural Economics 71, 847-857 (1989). 112. Mesak, H., and Coleman, R. W., Modeling the Effect of Subsidized Pricing Policy on New Product Diffusion, OMEGA International Journal of Management Science 20, 303-312 (1992). 113. Caswell, M. F., Lichtenberg, E., and Zilberman, D., The Effects of Pricing Policies on Water Conservation and Drainage, American Journal of Agricultural Economics 72, 883-890 (1990). Received IO September

1992; revised 15 October 1992