Christina H. Gladwin, “Production Functions and Decision Models: Complementary Models,” American Ethnologist 6, no. 4 (Nov. 1979), 653-674.

 

Production functions good for examining outcomes of adoption decisions, not for answering why adoption (didn’t) happen(ed)

---

Notes

• 653 – “Recent studies in economic anthropology use the production function to answer empirical questions about production in an indigenous industry.”
• “In my judgmenet, the use of the production function is appropriate in both of the above-mentioned studies because both studies ask empirical quesitons that the production function was designed to answer. What, for instance, is the maximum output attainable from a given set of inputs, or factors of production? Are entrepreneurs in an indigenous industry, such as farming or peddling, allocating their variable resources efficiently? Are they using ‘too much’ or ‘too little’ of a variable input? What should be the percentage share accruing to sharecroppers who provide some of the inputs to production, based on the contribution of the inputs to output? Do all productive inputs contribute equally to output? In order to answer such questions, the production function is the appropriate model to propose and specify.”
• “However, there are empirical questions regarding production in an indigenous industry that cannot be answered through production functions or the ‘allocative efficiency’ conditions derived from them.”

 

• “The purpose of this paper is to show how decision models and production functions can be used in a complementary fashion to answer different empirical questions raised about an important production process — adoption of new agricultural technology.”

• “Economists use the maximization assumption to explain changes in behavior, such as the adoption of new farming practices in Puebla, Mexico. But this tests the maximization assumption itself and does not test models of farmers’ actual decisions. Better explanation results from using natural decision models of the adoption process. Economic models, such as the production function, may be used to show the efficiency of practices after adoption decisions have been made.”

 

• 653n1 – “This paper is based on the author’s Ph.D. dissertation, “A Model of Farmers’ Decisions to Adopt the Recommendations of Plan Puebla,’ Food Research Institute, Stanford University, 1977, and is dedicated to the memory of Benjamin Coyotl, for whom the Plan Puebla was a hope.”
• 654 – “The particular new technology studied, during fieldwork in one village in Puebla, Mexico, in 1973-1974, was the Plan Puebla, an agricultural development project started in 1967 by the Centro Internacional de Mejoramiento de Maiz y Trigo (the International Center for the Improvement of Maize and Wheat) and now run by Mexico’s Graduate College of Agriculture at Chapingo, Mexico. Its aim was and is to increase yields of maize on rain-fed farms. Initially, the project’s breeding program attempted to find improved maize varieties or hybrids that performed appreciably better than the local variety (criollo). When the program found none, the project focused on deriving recommendations about fertilizer use and timing and plant population for the local variety.”

 

• “Given the considerable amount of publicity and high expectations generated by the Plan Puebla, the results or measures of farmers’ adoption have been disappointing: less the 20 percent of Poblana farmers were on Plan credit lists in 1973 (CIMMYT 1973:79-80). As a result, many ‘visiting firemen’ descended on Puebla to answer the ‘unasked question,’ ‘why didn’t the farmers adopt?’”

 

• “The methodology used in the study follows from the basic assumption that adoption or nonadoption is a decision made by individual farmers, and not just an S-shaped function over time or a spherical wave spreading out from a geographical ‘center’ (Griliches 1957; Huke and Duncan n.d.).”
• “The way to explain farmers’ adoption decisions, then, is to propose and directly test a decision model. If a particular decision model predicts actual adoption choices made by individual farmers, then one may conclude that the model explains adoption. If the predictive decision model can also pinpoint or identify the factors or reasons leading to the farmer’s choice not to adopt, then one has explained why farmers have not adopted. A decision model is therefore the appropriate model to answer the question, ‘why didn’t the farmers adopt, or even try the new technology?’”
• “However, a production function can appropriately be used to answer the questions, ‘what happened to the corn yields of farmers who did try the new technology? Did the recommendations improve the corn yields of the farmers who tried them, holding quantities of other inputs constant?’”

 

• 654-655 – “The present analysis differs from a more traditional economic treatment of adoption only by replacing the economist’s behavioral assumption about the farmer’s decision process with a model of that adoption-decision process. This is done because the farmer’s adoption-decision rules observed during fieldwork were too complex to fit into one mathematical behavioral assumption. Most behavioral assumptions used by economists are simple: farmers maximize profits, they minimize risk subject to a profit constraint, or they maximize expected profits or income subject to risk and credit constraints, and so forth. Upon observation and elicitation of farmers’ actual decision rules and strategies or plans, however, even the last behavioral assumption was considered to be, for some of the adoption decisions, too simple a reflection of the actual decision process.”

• Does this constitute a ‘next step’ in social science studies of the green revolution in mexico — a step away from algorithmic quantification of farmers’ values and principles made by the assumption of their analogousness to capitalistic estadounidense farmers’ values — made possible by talking with actual mexican farmers? Is there evidence that any of the (non-anthropological) economists (using behavioral assumptions and algorithms) I’m studying actually interviewed any Mexican farmers?

 

• 655 – “However, simplified behavioral assumptions have characterized other economic studies of the Plan Puebla: Villa Issa assumes farmers maximize profits; Moscardi also includes the ‘safety-first’ risk constraint; Martin del Campo and Winkelmann assume that farmers maximize income subject to a risk constraint. These studies thus ignore other possible constraints on farmers’ adoption, such as lack of knowledge of the new technology and lack of capital or credit. Moreover, the authors did not interview farmers directly, to see if other unexpected constraints could be holding back adoption.”
  • What does the failure of the social scientists she mentions to include alternate assumptions about farmers’ decisions say about the social construction of the social sciences, about their reductiveness? About the inappropriateness of reduction to the complex ecology of farmers’ attitudes and knowledges?
• “‘But,’ counters the economist bored with the ethnographer’s perennial objection to unrealistic assumptions, ‘the only proper test of a model is results.’ Of course, but what kind of results are offered by an indirect test of a behavioral assumption, as compared to a direct test of a decision model? Consider the results gathered by other studies of the Plan Puebla. Moscardi and deJanvry’s results imply that risk-aversion explains nonadoption of high levels of fertilizer use, whereas Martin del Campo’s and Winkelmann’s results show that risk-aversion is not important. All of these results are clouded by the fact that the researcher’s aim is not a direct test of the relationship that is actually estimated or fitted by multiple regression analysis. Instead, the results are used to test indirectly the underlying behavioral assumption about how farmers make adoption decisions.”
  • THE MOST IMPORTANT QUOTE OF THE ARTICLE — puts her decision model in direct conflict with behavior assumption based regression analyses. Claims that econometric approaches don’t test Mexican farmers, they test their own assumptions.
• 655n3 – “One such unexpected constraint holding back farmers’ adoption of the Plan’s recommended level of fertilizer use in the village was that the quantity of fertilizer recommended was too low. This was partly due to the banks’ dragging their feet about the quantity of fertilizer given on credit. Even though a poster in the Plan office stated that farmers in the village should use sixteen (fifty-kilogram) bags of fertilizer per hectare, Banco Agricola in 1973 was only giving credit for ten bags per hectare to village farmers. Credit was increased to twelve bags per hectare in 1974, but the average farmer in my sample applied sixteen bags per hectare. After many complaints and the appointment of a new representante in the village, with a subsequent improvement in communication and village participation in the Plan, credit was again increased to sixteen bags per hectare in 1975. However, even sixteen bags per hectare was too low for the best farmers or managers in the village, who were applying eighteen bags per hectare and more. This may have been due to the fact that the Plan’s recommendation was based on the worst soil type in the village.”
• 655n4 – “Economists concerned with estimation of demand for inputs, such as fertilizer, do not have to ask this question, because they are using the (simplified) behavioral assumption only to test a derived model or demand relationship. In this case, the model that is tested is the derived or demand model itself. It is tested directly, by fitting data to the derived relationship. Hence, one can look at the results to decide if the demand model is adequate.”
• “Of all the results gathered from Puebla, however, Villa Issa’s are the most pertinent to this paper because he uses (or misuses) the production function and ‘allocative efficiency’ conditions to explain the nonadoption decision.”
• “His hypothesis is a reasonable one: farmers do not adopt Plan Puebla recommendations due to their involvement in the off-farm labor market in the Puebla-Cholula-Mexico City area. To test this hypothesis, he does not build a decision model with availability of labor or management as one of the constraint (as in Benito 1976). Instead, he assumes that farmers maximize profits. Further, he assumes there are two sets of farmers, the adopters of Plan Puebla (defined as those with ‘high’ use of fertilizer) and the nonadopters, or those with ‘low’ use of fertilizer. . . . He then estimates two production functions, one for adopters and one for nonadopters, and uses the allocative efficiency conditions to show that the ‘average’ farmer in both groups is allocating his labor time efficiently. Thus, if the non-adopters (or those with low fertilizer use) were to put any more labor into their farm operations, as required by the Plan Puebla, they would not be maximizing profit.”
• “The allocative efficiency conditions are thus used to test indirectly the underlying assumption that farmers maximize profits. They can be tested for each farmer or adopter by calculation of the ration of the marginal value product (MVP) to the marginal cost (MC) for each factor of production. But even though the calculations can be done for each farm-firm separately, the ‘normal’ procedure is to examine the marginal returns and opportunity costs for the ‘average’ farmer. This procedure can be justified on the grounds that it is normally used to see if farmers as a group are using too much of one input and not enough of another, or allocating their resources inefficiently. The procedure cannot be justified, however, when one wants to test a hypothesis about how individuals make decisions.”
  • Why is it important that Gladwin emphasizes that the generalizing and homogenizing tendencies of the econometric approach don’t apply here?
• 656 – “Moreover, Villa Issa’s use of the production function and allocative efficiency conditions does not explain why farmers with little or no off-farm labor also do not adopt Plan recommendations. This is partly due to the fact that he does not know that farmers with little or no off-farm labor also do not adopt, because he defines adoption as ‘high’ fertilizer use and does not do his ethnographic homework and measure adoption precisely.”
• “However, a view from the field in 1973-1974 revealed taht there were four Plan recommendations: to provide credit for fertilizer, increase plant population, increase the number and change the timing of fertilizer applications, and use a recommended level of fertilizer per hectare. Because credit was not tied to mandatory adoption of the agronomic recommendations, farmers could and did adopt one recommendation and not another. Further, the recommendations were not blanket ones, that is, there was a specific recommended level of fertilizer use and plant population for each of the sixteen homogeneous producing sub-regions in Puebla. To employ fertilizer use as a measure of adoption, one should therefore match the farmer’s actual level of use with the recommended level of use in his region, because farmers could use more than the recommended amount of fertilizer.”
• “CIMMYT also neglected to measure individual adoption of each of the Plan’s four recommendations in each of the sixteen producing systems. . . . Rather than accept the too-low percentage of farmers on Plan credit lists as a measure of adoption, CIMMYT then implied that increased fertilizer use, which was considerably higher for Puebla farmers with credit than for those without, was a measure of adoption of the Plan’s technological package.”
• “The results of this study, however, show that high use of fertilizer did not imply either use of the Plan’s recommended quantity of fertilizer or adoption of the other agronomic recommendations of the Plan (for example, to increase plant population and to fertilize twice, at planting and the second weeding). Village farmers in my sample in 1973-1974 did not use the Plan’s recommended level of fertilizer because it was too low; but 53 percent were on Plan-sponsored credit lists; only seven percent adopted the plant population recommendation, and no one adopted fertilizer at planting and the second weeding.” Because adoption rates vary widely across Plan recommendations, one should treat adoption of each recommendation as a separate decision.”
  • What does it say about indigenous innovation that farmers found the Plan recommendations for fertilizer to be too low?
• 657 – “The decision criteria can either be simple orderings of alternatives on some aspect or dimension or factor of the alternatives or they can be constraints that must be passed or satisfied. In either case, the criteria or constraints are discrete rather than continuous; that is, the alternative ‘fertilizer’ either passes the criteria or constraints or it does not. A decision tree is thus a sequence or series of discrete decision criteria, all of which have to be passed along a path to a particular outcome or choice.”
  • Does the fact that the criteria must be discretized saying anything methodologically about this process that might make it less suitable to studying Mexican farmers than some other method? Is it just another way to anatomize and deconstruct problems a la mechanistic modes of thought?
• 658 – “The second reason why tree models were used is that they predict. . . . The predictability of these ‘hierarchical’ models has been high: the models usually predict 85 to 95 percent of the actual choices made by farmers . . ..”
  • Why is this predictability matter — how do these anthropologists intend to utilize this predictability?

 

• “Given a form of decision model, the crucial question becomes, ‘how are the decision nodes selected?’ After all, there are a multitude of possible adoption factors. A neoclassical economist would look at factors such as profitability, riskiness, costliness of the innovation, and farmers’ access to information, inputs, and credit (Winkelmann 1976). A Marxist or dependency-theorist, on the other hand, would look at the relations of production, distribution and the political control of land, credit, and other inputs (deJanvry 1979).”

 

• “The answer to this question is currently under much debate in the anthropological decision-making literature. Some anthropologists follow the ‘statistical aggregate approach,’ that is, they ‘observe behavior, record outcomes, and then analyze the patterns in the outcomes to construct a statistical profile of the people who choose different options’ (DeWalt 1975).’ They rely on observation and factual data because, in some instances, decision rules may not be elicitable; or people may not be able to tell what goes on in their heads.”

 

• “Cognitive anthropologists, on the other hand, assume that ‘people’s own knowledge of their world is the best starting point for model-building.”

• “Because decision making is a cognitive process (that is, it occurs within the human ‘black box’), decision models that ignore cognition are ‘machines with their insides missing.’”

• What does it say that this ethnographic model building is akin to using a machine (in their own words) to understand cultures? Is the anthropological industry just an extension of the epistemic colonization enterprise along with the Green Revolution itself?

 

• “The decision criteria selected in this study had to satisfy two requirements. First, they had to be formulated in terms understandable to the decision makers . . . Second, they had to ‘cut’ the sample of farmers into subsets of adopters and nonadopters. Thus the criteria had to be elicited or based on farmers’ reports and had to be observed to predict some farmers’ behavioral outcomes.”
• 658-9 – “Due to recent advances in linguistics and cognitive science, anthropologists now have a way to model individual behavior observed in the field and test that model against actual historical choices. The way in which decision criteria are selected is unimportant, as a test of the model will show incorrectly specified criteria.”
• 659 – “Given the set of decision criteria, one can identify the main factor limiting adoption by simply counting the number of nonadopters on each path leading to the outcome, ‘don’t adopt.’ One only looks at nonadopters because, in a tree model, an adopter has to pass all constraints. With the nonadopters , one can pick out the reason farmers do not adopt.”