The Quality of Science in Participatory Research: A Case Study from Eastern Zambia

The Quality of Science in Participatory Research: A Case Study from Eastern Zambia

World Development Vol. 30, No. 4, pp. 523–543, 2002 Ó 2002 Published by Elsevier Science Ltd. Printed in Great Britain 0305-750X/02/$ - see front matt...

1MB Sizes 9 Downloads 40 Views

World Development Vol. 30, No. 4, pp. 523–543, 2002 Ó 2002 Published by Elsevier Science Ltd. Printed in Great Britain 0305-750X/02/$ - see front matter

www.elsevier.com/locate/worlddev

PII: S0305-750X(02)00002-5

The Quality of Science in Participatory Research: A Case Study from Eastern Zambia CHRISTINA H. GLADWIN University of Florida, Gainesville, USA JENNIFER S. PETERSON Africare, Niamey, Niger and ABIUD C. MWALE * Land Management and Conservation Farming, Kabwe, Zambia Summary. — Recent discourse in the development field has been directed to the question of how to maintain and enhance the quality of science in agricultural research using participatory methods. Discussion has also focused on the question of how to combine microlevel research/extension efforts using participatory methods with scientific methods employing rigorous and statistical testing techniques. Is there a tradeoff between researchers’ use of microlevel, gender-sensitive, ethnographic participatory methods and a commitment to ‘‘the scientific method,’’ with its conventional assumptions about sampling, data collection, hypothesis testing, and use of standard measures of statistical significance? If there is such a tradeoff, which of the two methods should be given the greater attention? Should scientific and rigorous testing methods take precedence in the agricultural science community over use of farmer-sensitive participatory methods? Should scientific rigor be sacrificed for ethnographic accuracy, or vice versa? Ó 2002 Published by Elsevier Science Ltd. Key words — participatory research, quality of science, agroforestry, improved fallows, decision models, small farmers, women farmers

1. INTRODUCTION Proponents of participatory rural appraisal (PRA) claim that farmer-sensitive participatory methods are necessary tools for the development analysts, theorists, and practitioners whose aim is to ‘‘catch the elusive ghost of development’’ (Hyden, 1998). 1 Such participatory methods have been used to shift the focus of agricultural research from commodityspecific activities toward a more farmer-centered approach, to expand the capacity of extension staff working for small-scale farmers, to allow them to discover the diversity within their clientele groups, and to empower farmers, communities, and extension staff in the process 523

of technology development and dissemination (Spring, Sullivan, & Litow, 2000). Increasingly, the aim has been to enhance collaboration among farmers, researchers, and extensionists, all working to develop appropriate strategies to address local problems and constraints. In a larger sense, the goal has been, as Hyden says, to ‘‘become more circumspect and more humble in our pursuit of development,’’ because . . .we have come to accept that development cannot be done for people, not even of or with them.

*

Final revision accepted: 20 November 2001.

524

WORLD DEVELOPMENT

Development is potentially effective only when it is done by people themselves. They are the primary ‘‘stakeholders’’—they must own the activity—for it to become reality. This also means that development is viewed more as the outcome of local micro action—by individuals, groups, or communities—than by formal institutions such as the state (Hyden, 1998, p. 1).

Historically, the PRA school evolved from several disciplines (anthropology, geography, rural development) that relied on fieldwork methods used by cultural anthropologists for over a century, such as participant observation (Spradley, 1980). PRA methods also built on the rapid rural appraisal (RRA) methods of the 1970s, e.g., the sondeos of Hildebrand (1981) and Norman (1975), used in development efforts that focused on the farmers’ role in deciding to adopt new practices (e.g., farming systems research/extension—FSR/E) (Chambers, 1981; Matlon, Cantrell, King, & BenoitCattin, 1984; McCracken, Pretty, & Conway, 1988; Rhoades, 1984). Since the late 1970s, FSR/E researchers realized that in order to design technologies that were potentially adoptable—or adaptable—by small-scale peasant farmers, the target clientele of most agricultural research centers, be they CGIAR centers or NARs or NGOs, it was necessary to involve the target clientele as active participants in the entire sequence of events in the technologygeneration process, from the initial design stage to the final adoption stage (Rolling, 1988). Chambers’ (1983) work to ‘‘put the last first’’ contributed to the evolution of development thought—and put the microlevel efforts of FSR practitioners into a broader macro-level development perspective—by showing how the popular development theories of the time failed miserably to alleviate poverty. Although billed as poverty-alleviation efforts, they usually bypassed the poor, whose livelihoods and living conditions were ‘‘local, complex, diverse, dynamic, and unpredictable (lcddu)’’ (Chambers, 1994a,b,c). (We have reservations only about the last adjective in the list—unpredictable—as the remainder of this paper will show.) Largely as a result of Chambers’ efforts, participatory methods have now become an accepted part of ‘‘bottom-up’’ poverty-alleviation efforts such as the sustainable livelihood (SL) approach that seeks to re-orient the development process away from top-down, technocratically planned interventions in favor

of finding ways for donors to enhance alreadyexisting modes of making a living and managing human and natural resources (Chambers & Conway, 1992). They are also used in diffusionof-technology efforts of all kinds, including the soil replenishment technologies described below. Currently, the distinctive feature of the PRA approach lies in the use by the farmers themselves of a variety of problem-solving exercises and visual tools (such as problem trees and Venn diagrams, stakeholder analysis, wealth and preference rankings, time and trend lines, food and labor calendars, gender activity analysis and gender resource mapping, pie diagrams, village role-playing) so that farmers would not only identify their own problems but also be empowered to take the necessary steps to reach a solution (Chambers, 1997, pp. 102– 140; Grandin, 1988; Nabasa, Rutwara, Walker, & Were, 1995). In spite of evidence long provided by noted anthropologists such as Hill (1956, 1963) that farmers rationally manage quite complex local agro-ecosystems in a dynamic, diverse way (the notion behind Chambers’ ‘‘lcddu’’), too many donors and institutions still propose answers that address the wrong question. For example, the USAID Soils CRSP (collaborative research support program) is now promoting and distributing a CD rom in Africa containing an ‘‘expert systems’’ software that will predict how much nutrient a farmer should put into his/her soil, given the desired yield. This is the wrong question. Clearly, African farmers desire high yields; but the right question to ask them is, ‘‘How much nutrient can you afford to put on, and how much yield will that get you, and how can you make up your household food gap?’’ Lack of attention to the right questions, and a focus on the wrong questions, as Chambers (1997, p. 17) points out, leads to a high proportion of failures in development projects, whose costs are borne not by the donor institutions but by the borrower countries. Yet there is no lack of new projects that propose grand innovations for farmers (e.g., conservation farming with ‘‘holey-holes’’ in Zambia, software ‘‘expert systems’’ for soil fertility replenishment) that will never be adopted because farmers have binding structural constraints to adoption (e.g., lack of land, labor, capital, education). Once funded, however, the projects become self-perpetuating because the institutions then feel the need to promote and distribute their technologies.

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

2. WHY PARTICIPATORY RESEARCH METHODS ARE NECESSARY BUT NOT SUFFICIENT The use of participatory methods is therefore necessary to enable researchers, extensionists, institutions, and donors to ask the right questions. As research tools, however, they are not sufficient. Why? The reasons are several. First, participatory methods have been criticized for being ‘‘quick and dirty’’ research methods. In some of our own previous work in Guatemala (Gladwin, 1983, p. 155), for example, we have noted that while innovative and useful for putting agronomists and farmers together before researcher-managed and on-farm trials are conducted, the length of time spent in the field ‘‘sounding out’’ farmers in a typical RRA or sondeo (Hildebrand, 1981) is far too short, taking a time period of one week instead of one month or one year, usually required by anthropological fieldwork. The same criticism applies to PRA teams who rely on RRAs and focus group discussions for their interviewing techniques, and ignore anthropological methods of in-depth ethnographic interviews (Spradley, 1979) and lengthy participant observation. What is the problem? The speed of the reconnaissance can result in superficial conclusions. Each member of the RRA team interviews only 10–15 farmers in three to four days, even though the whole team—of six to eight typically—may interview a total of 60 farmers in the region of interest. Each member of the team is therefore generalizing from too small a sample by any statistical standard. Arguments therefore ensue when the team meets to compare their findings, with remarks generated such as, ‘‘My three farmers said this or did that. . ..’’ Usually, observations of the team leader or the more assertive team members get written up in the report as the findings of the whole team. Any disagreements between team members are glossed over because there is not enough time to investigate further; the report must be written. Second, as a result the RRA report ignores individual variation in farmers’ decision rules and farming practices and concentrates on the similarities and shared behavior of farmers in the region. Statements in the report are typically of the form, ‘‘Farmers in this region are low-resource farmers who farm on family land. . .’’ rather than statements that observe diversity, e.g., ‘‘Of the 35 farmers interviewed, 15 identified vegetable production as their main enterprise’’ (Sullivan, Litow, & Spring, 2000).

525

Too often, the heterogeneous nature of farmer behavior is swept aside in the effort to generalize from the too-small sample to the population. Yet even in a region with homogeneous agroclimatic conditions, there are differences in farmer behavior, plans, and decision rules, as the decision models in this paper will show. Third and more important, there is no procedure to test the universality of the conclusions reached, or the importance of the constraints identified, in a typical RRA or PRA. The hypotheses and generalizations in the PRA or RRA report about farmer problems and constraints remain untested, mainly because most of the data gathered remain uncoded. The team expends a great deal of effort in collecting data from as representative a sample of farmers as is possible, over as wide a geographical area as is feasible in the short time usually provided them, e.g., Kanji and Gladwin’s (1999) report on gender and livelihood systems in Gorno Badakshan, Tajikistan, a former Soviet republic. They then draw their conclusions and make policy recommendations but leave the data uncoded and unanalyzed. This is a waste, as the PRA teams usually interview a good number of farmers and gather very useful information about a great deal of topics: farmers’ livelihood systems, wealth ranking and land tenure types, production activities, constraints to production, e.g., the Sullivan et al.’s (2000) report to FAO on St. Lucia in the Caribbean. But because the data are left uncoded, the major findings of the report are also left untested. Is this an accident? No, it is completely in line with Chambers’ (1997, pp. 33–53) sentiments that reject the scientific method along with quantification, reflected in statements such as: In the social sciences and policy, economics dominates, and gives primacy to mathematical analysis; what has been measured and counted becomes the reality (p. 33). . .. Many physical things are amenable to measurement. . .. The problem is that the idiosyncratic attributes of people are, in contrast, difficult to measure: their individual behavior is unpredictable (pp. 38–39). . .. Quantification and statistics can mislead, distract, be wasteful, simply not make sense, or conflict with common values. . .. Yet professionals, especially economists and consultants tight for time, have a strong felt need for statistics. . .. Numbers can also reassure by appearing to extend control, precision and knowledge beyond their real limits (p. 40). . .. Economists, like others, are then forced into a selfsustaining circularity: what is measured generates theory which leads to more measurements which maintain the theory. . .. Measurable is sustainable.

526

WORLD DEVELOPMENT

Irreverent imagery comes to mind: of a measurementtheory merry-go-round. . . (Chambers, 1997, p. 53).

What is wrong with this thinking? Clearly, there is something in what Chambers says. Numbers and statistics have too often been misused in the social sciences, and an obsession with precise measurement cannot substitute for the researcher’s doing her ethnographic homework and discovering the complex, diverse, dynamic realities of local livelihood systems first-hand by lengthy immersion in the culture complemented by in-depth ethnographic interviews with insiders to the culture. Professional status systems can trap the researcher whose aim is to understand poverty and elicit the livelihood strategies of the poor. The dangers of theorizing while safely ensconced in the ivory tower are not exaggerated. Yet by throwing out the methods of modern science along with quantification and statistics, Chambers is putting the researcher in more danger—the danger of being wrong with no way to show it. By contrast, the scientific method requires the researcher to model their interpretations of reality by generating a hypothesis about people’s behavior, then collect observational data to test the model, and then revise the model based on the test results (Lave & March, 1975). This hypothesis-testing sequence is the basis of science. Without it, researchers have no way to give themselves a reality check. Without a reality check, researchers have no way of sifting through all their ideas and ethnographic observations to cull the ones that are wrong; and unfortunately, untested ethnographic observations can give the researcher just as false a sense of security as do the numbers Chambers is criticizing. 2

3. COGNITIVE SCIENCE MODELS In our judgment, the solution to the problem that Chambers poses is not to throw out the hypothesis-testing sequence of the scientific method, but to build models that can predict individual behavior in local, complex, diverse, dynamic (lcdd) contexts. In this paper, we present a case study in which this was done. The model we present is predictive of individual behavior because it is a cognitive science model, made possible by the cognitive revolution of the 1960s and 1970s that swept through the social sciences and changed the fields of artificial intelligence, anthropology, linguistics, neuroscience, psychology, and philosophy (Gardner,

1985). Scientific assumptions about cognition were radically changed in 1956 with the publication of a seemingly-innocuous article entitled, ‘‘The Magic Number Seven, Plus or Minus Two,’’ by the psychologist George Miller (1956). Miller’s new idea was that the human computer is of limited capacity, unlike a real computer; its short-term memory is limited to roughly five to seven items at a time. Indeed, people seem to categorize or discretize variables, rather than deal with continuous quantitative variables as a real computer does. People use logic rather than perform complicated mathematical operations as a real computer does. The limited ‘‘rehearsal buffer’’ of humans affects the way people organize their language and thought processes, such that they are ‘‘hierarchically’’ organized (Miller, Galanter, & Pribraum, 1960). This revolutionary idea was taken up by the linguist Noam Chomsky, whose ‘‘trees’’ or transformational grammars spread to most known languages. In the field of cognitive science, it stimulated scientists to write computer programs which mimic the way humans think and do the clever things that people do using the same information processes that people use (Simon, 1972). Simon compared this field to that of artificial intelligence in which people programmed computers to do clever, humanoid things—but not necessarily to do them in a humanoid way. In psychology, this distinction generated new theories of problem solving (Newell & Simon, 1972) and decision making. Plans and scripts (Schank & Abelson, 1977) became the buzzwords of the 1970s; and decision-making models of expected utility were abandoned in favor of theories of ‘‘eliminationby-aspects’’ and ‘‘preference trees’’ (Kahneman & Tversky, 1972, 1982; Tversky, 1972; Tversky & Kahneman, 1981). In anthropology, the revolution resulted in the birth of cognitive anthropology which aimed to discover the meaning of ‘‘native’’ terms and domains of cultural meaning through new methods of frame analysis (Frake, 1964), taxonomic analysis (Berlin & Kay, 1969), componential analysis (Romney & D’Andrade, 1964), schematas or folk models (D’Andrade, 1981, 1987; H. Gladwin, 1974; Quinn, 1990), plans or scripts (Werner & Schoepfle, 1987), and ‘‘ethnographic decision tree modeling’’ (Gladwin, 1989). After the cognitive revolution, cognitive anthropologists and psychologists rejected the expected utility theory of choice (Von Neumann & Morgenstern, 1947), and searched for more cognitively-realistic models of the choice

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

process. They claimed that people do not make wholistic assignments of utility for each alternative and separately formulate subjective probabilities (Quinn, 1978) and then pick the alternative with the greatest expected utility, as does conventional microeconomic theory. In line with Miller’s results on the limitations of human computational capacities, it is posited that decisions are made in a decomposed fashion on a piecemeal basis, i.e., one dimension at a time (Schoemaker, 1982). This assumption fits in quite nicely with Chambers’ (1997, p. 41) call for rough comparisons rather than precise measurements. The resulting decision model is what Arrow (1951) calls a ‘‘choice function not built up from orderings’’ and Simon (1979) calls production rules, i.e., simply a set of if–then rules. Each path on the decision tree corresponds to one if–then decision rule and the tree as a whole is equivalent to a set of decision rules or a decision table. The resulting model is a hierarchical model or expert system that reflects the various expert rules elicited from the decision-makers themselves. The term hierarchical distinguishes decision trees from linear additive models such as linear regression analysis, probit analysis, or logit analysis. It refers to the fact that decision criteria in a tree are mentally processed such that alternatives are compared on each dimension or criterion separately, and all of them may not be processed by all individuals, which leads to the path structure of a tree. This simplifies the decision process considerably, and saves the individual cognitive energy. A linear-additive model, in contrast, assumes all the criteria or dimensions of each alternative are weighed by the decision maker, and each alternative is assigned a composite score, and the alternative with the highest score is chosen. Much debate about these two types of models of the searchfor-information process has occurred between psychologists (Rachlin, 1990, pp. 76–77). Ethnographic decision tree models differ from other ‘‘expert-systems’’ or rule-based models of the choice process, however, because the latter typically model the decision process of only one expert. With the ethnographic approach, a composite or group model is formed from individual tree models of individual decision processes, and is then tested against choice data collected from other individuals in the same population. The main aim is predictability of the choices made by individuals in a group rather than just one individual in that group (Breslin, Gladwin, Borsoi, & Cunning-

527

ham, 1999; David, 1992; Gladwin, 1976, 1979, 1991, 1992; H. Gladwin & Murtaugh, 1984; Gladwin & Zabawa, 1984; Mukhopadhyay, 1984; Swinkels & Franzel, 1997; Williams, 1997). Unfortunately, to date this method been used to predict choices of small samples of decision makers interviewed personally, usually by the model builders themselves, and thus has rarely been subjected to independent tests on large random samples of decision makers (Gladwin, Gladwin, & Peacock, 2001; Peterson, Tembo, Kawimbe, & Mwang’amba, 1999). The decision trees assume that decision makers go through two stages in the process of making a decision (Gladwin, 1980, 1989). During the first stage, decision makers pre-attentively eliminate all alternatives having an unwanted aspect (Gladwin & Murtaugh, 1979; Tversky, 1972). During the second stage, decision makers consciously choose among alternatives by ordering them on one aspect or feature or decision criterion, and then passing the alternative ordered first through another set of constraints. If that alternative passes all the constraints, it is chosen; if not, the second-best alternative gets a chance to pass its set of constraints. If no alternative passes, the decision maker searches further. The theory thus assumes that decision makers use a discrete form of the basic choice principle of microeconomics, ‘‘maximization subject to constraints;’’ but because the ordering is not connected, it is not necessarily transitive and the result is a ‘‘choice function not built up from orderings’’ (Arrow, 1951). The trees are relatively simple to design and test, as the model of the Zambian farmer’s decision to adopt improved fallow technologies (IFs) in Figures 1 and 2 shows, read from top to bottom. They have alternatives in set notation ðf gÞ at the top of the tree, decision outcomes in boxes ð½ Þ at the end of the paths of the tree, and decision criteria in diamonds ð< >Þ at the nodes of the tree. There are only two alternatives or decision outcomes in this set, [Plant an improved fallow] and [Don’t plant an improved fallo]. These are mutually exclusive choices, and so the model in Figures 1 and 2 is an example of a {Do it; don’t do it} decision, the simplest kind of decision to model. 4. PARTICIPATORY METHODS AND DECISION TREES IN EASTERN ZAMBIA The only trick to the trees is eliciting the decision criteria from the decision makers

528

WORLD DEVELOPMENT

Figure 1. Model 1 tested with ICRAF four-village data, 1998, 121 cases (49 FHHs, 32 MF, 40 MHHs).

themselves, who are the experts in making their decisions. They alone know how they make their choices, and so their decision criteria should be elicited from them in ethnographic interviews and by participant observation and other participatory methods (e.g., role playing). Given a particular sample of choice data collected from decision makers, a stratified sample of farmers in Eastern Zambia, one can test the tree easily by putting the data from each individual choice (as a separate, independent Bernoulli trial) down the tree and counting the errors in prediction on each path.

In Eastern Zambia, this methodology was used to study the decisions of small-scale farmers, including female headed households (FHHs), to adopt agroforestry innovations in the form of improved fallows, researched by ICRAF, the International Centre for Research on Agroforestry, and recently promoted and extended by World Vision International, and monitored by the University of Florida Soils CRSP (collaborative research support program). Previous research of the UF Soils CRSP had searched for a gender-neutral solution to soil fertility depletion, and showed that African

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

529

Figure 2. Constraints to IFs with Model 1 in ICRAF four villages.

farmers were all too well aware that their soils are depleted, and had increased their food insecurity problems. But how to solve the problem when structural adjustments reforms (SAPs) in the last two decades have made inorganic fertilizers on food crop unprofitable and unaffordable by most farmers? 3 Improved fallow technologies with various tree species (Sesbania sesban, Tephrosia vogelii, Gliricida sepium) were one possibility, and have been researched at Msekera Research Station in Eastern Zambia by ICRAF since 1988, and in

on-farm trials on farmers’ fields since 1992–93. Small improved fallow plots of under onefourth of a hectare are planted for two years with nitrogen-fixing tree species (Sesbania or Gliricidia seedlings or direct-seeded Tephrosia vogelii or Cajanus cajan (pigeon pea)), and followed by two or three years of maize. By far the most promising, although it may look like a ‘‘dinky little tree,’’ is Sesbania sesban, which is grown in a nursery three to six weeks before the rainy season. Results over the five-year cycle show improved fallows improve total maize

530

WORLD DEVELOPMENT

production 87% over unfertilized maize (even without any maize yield in year two of the fouryear cycle); Kwesiga, Franzel, Place, Phiri, and Simwanza (1997), for example, finds maize yields following two-year improved fallows approach those of fully fertilized fields (with 112 kg nitrogen per hectare). Moreover, with the rising prices of fertilizer in Zambia, fully fertilized maize is no longer an option, and even partially fertilized maize is not an option for many farmers who have neither the cash nor the access to credit to purchase fertilizer. Since 1997, therefore, the multi-year trials of improved fallow technologies (IFs) have been a unique African success story: over 3,000 farmers have participated, 49% of whom are women farmers (Franzel, Phiri, & Kwesiga, 1997). Since World Vision joined the effort to extend the IF technology to 50,000 people in the wider E. Zambia region in 1999, 5,000 households have tried planting IFs. Yet the question still unanswered is: why are improved fallows being adopted so readily in Eastern Zambia, especially by women and FHHs, and nowhere else? 4 Is their success due to the fact that Eastern Zambia is a region of lower population density than other African regions (e.g., Malawi, Western Kenya) so that women farmers have enough land to put some of it in fallow, or is it just a delayed reaction to structural adjustment policies that have raised the price of inorganic fertilizers to levels so high that women farmers have finally ‘‘adjusted’’ by deciding to ‘‘grow their own fertilizer’’ and adopt a substitute soil-fertility amendment? To answer this question, Jen Scheffee Peterson used participant observation and in-depth ethnographic interviews with women and men farmers who both were and were not testing and expanding their on-farm trials of IFs. Men and women adopters and non-adopters were interviewed in 1998, first by Jennifer Scheffee Peterson with three women in each of the four villages targeted by ICRAF with onfarm trials of improved fallows since 1992–93 (Peterson, 1999). After an initial composite model was built to represent the decision process of all 12 farmers interviewed, Gladwin and Peterson then conducted another set of personal interviews in June 1998, to jointly refine the model. A questionnaire was then jointly designed so that Peterson could test the revised composite decision model (Figures 1 and 2) during personal interviews with another test sample of 81 women farmers and 40 men farmers who also resided in the camps sur-

rounding ICRAF’s four target villages (Peterson, 1999). Women in both FHHs and MHHs were interviewed. The samples were chosen, after discussions with Steve Franzel and Donald Phiri of ICRAF, such that half the sample of each gender would be testers (who planted at least one improved fallow plot) and half nontesters (NT who did not plant even one improved fallow plot), and half of the sample of testers would be testers-expanders (TE who planted at least two improved fallow plots) and half testers-non-expanders (TNE who planted only one improved fallow plot). 5 Different versions of the adoption decision model were tested first by Peterson (1999) using an Excel spreadsheet, and then by Gladwin using simple SPSS syntax programs. The following model (Figures 1 and 2) is a close approximation to the first model elicited by Peterson and Gladwin, so that it has ‘‘descriptive adequacy,’’ meaning it matches informants’ statements about how they decided to plant an improved fallow. It differs from that model, however, in minor ways by the inclusion of other criteria, so that it is the ‘‘best-fit’’ model obtained with these data. (a) Motivations to plant trees The motivations to plant an improved fallow plot came from the very first interviews by Peterson, as nearly all women say they plant an improved fallow because their soils are tired (nthaka yosira/yoguga), fertilizer is too expensive (wodula ngako), and their maize harvest does not last all year until the next harvest. The model in Figure 1 says that any one of these reasons is enough for a farmer to consider planting an improved fallow; and thus sends them (i.e., their data) to the outcome, ‘‘Plant an improved fallow unless.’’ Note that in the Eastern Zambian sample, every farmer has at least one of these reasons to plant an improved fallow, and thus the whole sample passes on to the first set of ‘‘unless conditions’’—conditions or constraints which will block a farmer from planting an improved fallow, even though she or he has a good reason to. (b) Constraints to planting an improved fallow Figure 1 also lists the first set of constraints. It is a subroutine asking farmers if they are already satisfied with their soil fertility amendments so that they do not also need to plant an improved fallow. Farmers are sent to the

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

outcome, ‘‘Don’t plant an improved fallow,’’ if they can buy fertilizer, or barter for it, or get it on credit, and they are satisfied with the amount acquired; or they have used manure on field maize in the recent past, and they are satisfied with manure; or they rotate crops in the field, e.g., groundnuts with maize with cotton, and they are satisfied with their crop rotations; or they have land ready to come out of a natural fallow now. Results of testing this subroutine of the model show that most farmers can either buy or barter or get fertilizer on credit; but whereas women (especially female headed households) mostly barter for fertilizer, men mostly buy fertilizer. Almost no one gets credit for fertilizer in Eastern Zambia. Almost no one is satisfied with the amount of fertilizer that is acquired, as usually it is a big decrease from past use. In addition, almost no one uses manure on maize; it is saved for garden vegetables grown in the dimba in the dry season, not usually used on field maize. Finally, almost everyone rotates their crops as a soil fertility measure, but that does not satisfy their need for more soil fertility. Results further show there are a lot of errors with the fallow criterion, ‘‘Have land ready to come out of fallow now?;’’ but when we omit it (by running another version of the model without it) there are more errors in the model (29 vs. 21). We thus conclude that with these data, the fallow criterion clearly helps the prediction rate, so that it belongs in the model. 6 If the farmer is satisfied, he or she—really his or her data—is sent to the outcome [Don’t plant an IF now]. If the farmer is not satisfied, and also feels a need for the soil fertility amendment of IF trees, he or she is sent to the outcome, [Plant an IF plot unless. . .], meaning the farmer must pass another set of constraints in order to go to the outcome [Plant an IF]. The latter constraints in Figure 2 start with a benefits criterion (‘‘Have you ever seen the benefits of IFs in other people’s fields?’’). If yes, farmers are asked if they can wait two years to see the benefits. Because of the intense work of ICRAF in these four villages, most farmers have either seen the benefits of IF plots on their or their neighbors’ land, so most can wait the two years until the maize harvest after the improved fallow. Results show most (86) farmers in this sample proceed to the other constraints: lack of technical knowledge of how to plant the improved fallows (planting the nursery, transplanting the seedlings, or direct-seeding tephrosia), lack of time to plant an IF, lack of

531

strength and health, lack of access to seeds or seedlings, and lack of land. In addition, farmers were asked if their only access to land was borrowed land (so they would not plant an IF), or if villagers’ jealousy of early adopters of IF might be a problem. Results show only 54 of 86 farmers pass all these latter constraints and are predicted to adopt. The most important limiting factor (for 21 farmers) is lack of technical knowledge of how to plant an IF. Of the 86 farmers who make it down the tree to this constraint, lack of technical knowledge is a limiting factor for more married women (37%) than FHHs (24%) than men (17%). This gender difference is expected, based on previous literature, but it does affect adoption: this model predicts adoption for only 31% of the married women in MHHs compared to 47% of the FHHs and 52% of the men in MHHs. There are 22 total errors of the model, for an overall 82% success rate. Farmers thus say they plant an improved fallow because their soils are tired (nthaka yosira), fertilizer is too expensive (wodula ngako), and their maize harvest does not last all year until the next harvest. These farmers will adopt an improved fallow unless they can buy fertilizer, or barter for it, or get it on credit, and they are satisfied with the amount acquired; or they have used manure on field maize in the recent past, and they are satisfied with manure; or they rotate crops in the field, e.g., groundnuts with maize with cotton, and they are satisfied with their crop rotations; or they have land ready to come out of a natural fallow now. Results show while women mostly barter for fertilizer, men most buy fertilizer. Few farmers get credit for fertilizer in Eastern Zambia, because many have not repaid their fertilizer loans in the past, partly due to the 1991 drought. 7 Few farmers are satisfied with the amount of fertilizer that is acquired, as usually it is a big decrease from past use. In addition, few use manure on maize; almost everyone rotates their crops as a soil fertility measure, although rotation does adequately address the problem of soil fertility. Therefore, if farmers have seen the benefits of an improved fallow for themselves or can wait two years to see the benefits on their own plots, and they know how to plant the improved fallows (planting the nursery, transplanting the seedlings, or directseeding tephrosia), and have the time the strength and health to plant an IF, access to seeds or seedlings, and a small plot of land to experiment on, they will plant an IF.

532

WORLD DEVELOPMENT

(c) Alternative decision models Adoption decision trees presented here are built using participant observation and ethnographic interviews, and require an ethnographer to live in a village for a while in Africa—the longer the while, the better the tree model. 8 In the first stage of the research, the model-building stage, the ethnographer should know the informants so well that he/she knows when the informant is lying. In contrast, the second stage of the research is a testing stage that requires rigorous testing procedures. This stage can be done back in the office—in isolation as Chambers claims, as it means putting the cases down the tree a number of times and testing alternative models against each other to see which one is better. 9 This may be hard to

do by hand without fudging; to avoid the temptation, we used SPSS syntax programs to test the models for us. Several alternative models were also tested and are presented elsewhere (Peterson, 1999). The best of these, called model 3, is presented in Figures 3 and 4. It differs from model 1 in the following ways. First, the order of the criteria in Figure 3 is changed, so that the second criterion is now, ‘‘Ever seen the benefits of IFs in other people’s fields?’’ instead of ‘‘Are your soils tired?’’ This change gives primacy to a profitability or benefits criterion that used to appear much later in the decision process, and should make the economists happy. Now 82 farmers proceed to the next stage of the decision process because they have seen the benefits of IFs for themselves. The tired-soils criterion is

Figure 3. Model 3 tested with ICRAF four-village data, 121 cases (49 FHHs).

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

533

Figure 4. Constraints to planting an IF with Model 3.

now relegated to the third criterion, processed only if the farmer has not seen the benefits of an IF. If farmers think they have good soils, the model sends them (i.e., their data) to the outcome, [Don’t plant an IF]. Second, the fertilizer sub-decision is much simpler in this model, and consists of only three criteria on the ‘‘tired soils’’ path. Farmers will not plant an IF if they think fertilizer is not expensive (and therefore can afford it). Likewise, they will not plant an IF if they can buy at least one bag of fertilizer and are satisfied with the amount they can buy. But, the majority (113) of the 121 cases proceed to Figure 4 of the model because they either have bad

soils or think fertilizer is expensive or are not satisfied with the amount they can acquire. In Figure 4, the ‘‘able to wait two years to see the benefits of IFs’’ criterion leads the list of constraints farmers must pass in order to go to the outcome, [Plant an IF]. It is followed by constraints of: time in the busy planting season, knowledge of direct seeding of Tephrosia (which substitutes for a technical knowledge constraint), strength or health or labor to help transplant the IF seedlings, access to seeds or seedlings and owned land, and the jealousy constraint seen earlier in Figure 2. In spite of the greater simplicity of this model 3, it does not

534

WORLD DEVELOPMENT

predict as well as model 1: now 72% of the choice data are predicted instead of 82% with model 1. In yet another alternative model seen in Figure 5, we added a risk criterion we elicited from several farmers, ‘‘Do you fear animals’ (cattle, goats) grazing freely and eating the leaves of your IF trees during the dry season?’’ Unfortunately, the addition of this risk criterion added even more errors to model 3, so that we concluded the criterion did not ‘‘cut’’ (i.e., divide the sample into a set of adopters and non-adopters. This may be due to two factors.

Either farmers who did not even try an IF never got to the (mental) point where they considered the risks of animals in the fallows, or they did but also had a strategy to deal with it and thus reduce the risk. Such risk-reduction strategies include planting species of trees that resprout (e.g., Gliricidia sepium), planting the IF plot far away from the village, sending children to watch the field, informing the village leaders about the problem and getting new rules and regulations about animals in the post-harvest season, and using a barbed wire fence around the IF plot. If farmers have such a risk-reduc-

Figure 5. Adding a risk constraint to planting an IF with Model 3.

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

tion strategy, the model predicts they would then take the reduced risk of animals in their IF plot. We also elicited other sources of risks to the IF plots: children burning fields to catch mice during the hungry season, and damage to the trees due to beetles and termites. Unfortunately, we did not ask every farmer if they perceived these sources of risk or had one of these risk-reduction methods, but should do so in the future to properly test a ‘‘risk of the loss of the IF plot’’ subroutine. We therefore reject model 3 in favor of model 1; and discuss the policy implications of model 1, without a risk subroutine, in the discussion below. (d) More results: logit analysis of adoption of improved fallows In addition to decision tree models, we also tried quantitative methods as a further test of our hypotheses summarized in the decision tree models (Mwale, 2000) using data from Peterson (1999). Probit analysis and ordered logit analysis were also used to see if any of the variables in the decision criteria were significantly related to a dependent variable representing adoption of improved fallows. We proposed and tested two different dependent variables, representing two different definitions of adoption; and so used two different versions of multiple regression analysis: logit and ordered probit analyses, which like decision tree models are models of the choice process. Unlike decision trees, they are linear additive models amenable to multiple regression analysis which provides standard statistical tests of the significance of the different factors or dimensions which are purported to explain the choice. Logit analysis is used when the dependent variable takes on discrete, categorical ð0; 1Þ values rather than continuous numerical values. It is used here because adoption can be considered a discrete, categorical ð0; 1Þ variable equal to 1 if the farmer adopts or plants an improved fallow (IF), 0 otherwise. Ordered probit analysis is also used here with another dependent variable, TSYP or ‘‘tester type,’’ which equals ‘‘2’’ if the farmer is a tester-expander, ‘‘1’’ if the farmer is a testernon-expander, and ‘‘0’’ if the farmer is a nontester of IFs. It is appropriate to use here if there is an order of the different types of adopters, so that the tester-expanders (TSYP ¼ 2) are the real adopters of the technology, more so than the farmers who test IFs only one time (the tester-non-expanders, TSYP ¼ 1), and they

535

in turn are adopting more than the non-testers (TSYP ¼ 0). Ordered probit and logit analyses both measured the statistical significance of choice factors identified as important in the decision tree models. These included the independent variables listed and defined in Table 1: variables which measure the household size, available fallow land and labor for the household, the farmer’s age, wealth, educational level, club membership, gender and marital status, experience with hunger, and ability to buy fertilizer (see Adesina & Chianu, 2000 for a review of the adoption literature). Also included are dummy variables measuring whether or not farmer has seen the benefits of an improved fallow, and can wait two years to do so, and thinks his/her soils are tired. The results of the logit analysis (Table 2) indicate that only some of the independent variables, which were included as decision criteria in the decision trees in Figures 1 and 2, are significantly associated with adoption of improved fallows at the 5–10% level of significance. They are the farmer’s ability to wait two years to see the benefits of an improved fallow—significant at the 0.5% level, club membership at the 0.5% level, the availability of natural fallow land at the 2% level, gender and marital status at the 7% level, household size at the 4% level, and age at the 9% level. None of the other variables (e.g., farmer’s wealth or education or the availability of extra-household labor, the farmer’s experience with hunger or perception of tired soils) were found to be significantly associated with adoption of improved fallows, once the significant variables are held constant. The probit model predicts a high 75% of the farmers’ choices correctly, lower than the decision tree results, but still high. The log of likelihood function further indicates that the data fit the model well. The value of the result is further strengthened by the R-squared of 0.35 that is within the acceptable range of a regression equation with individual observation data. Of the variables that are significant in Table 2, some deserve more comment. First, the age of the farmer is significantly related to the likelihood of adoption at the 10% level of significance. Older farmers are more likely to adopt improved fallows. Second, the sign of whether the household has land in natural fallow is positive, meaning there is more probability of adoption of improved fallows if the household already has land in natural fallow. They are complements, not substitutes for each

536

WORLD DEVELOPMENT Table 1. Definition of variables in the logit and ordered probit analysis

Variable

Definition

USEIF TSYP FMPFAM

Use of improved fallow—1 if farmer has planted an improved fallow in the past, 0 otherwise Tester type—2 if farmer is a tester-expander, 1 if a tester, 0 if a nontester Gender and marital status of the farmer—0 for a male in a male headed household, 1 for female in a male headed household, and 2 for a female in a female headed household, i.e., female is the ‘‘marked’’ category Household size—the number of people considered to be in the household at the time of the study Household help—the number of relatives, friends available to carry out farm work, especially at peak times Household land in natural fallow—1 if household has land, 0 otherwise Farmer’s ability to buy chemical fertilizer—1 if able to buy, 0 otherwise Seen benefits oneself—1 if farmer has seen the benefits of an improved fallow, 0 otherwise Ability to wait two years—1 if farmer can wait two years to see the benefits of an IF for herself, 0 otherwise Hunger—1 if farmer stated the household had not experienced hunger at some time in the past year, 0 otherwise. Tired Soil—1 if farmer states his/her soils are tired or infertile, 0 if good soils Educational level of the farmer. Household wealth score measured by ICRAF (well off is 3, fairly well off is 2, poor is 1, very poor is 0) Club membership—1 if farmer is a club member, 0 otherwise Age of farmer Farmer’s ability to use oxen for land preparation (1,0) Farmer’s perception that fertilizer is expensive (1,0)

HHSZ HVHLP HVLND BFERTCH SNIFBFT ABWAIT HNGER TRDSOL EDLEV WLTHST CLBMEM HHAGE OXENUS FETEXP

Table 2. Logit estimation of factors affecting adoption of improved fallows, n ¼ 119 Variable Gender + marital status Household size Household help Ability to buy fertilizer Seen the benefits of IF Ability to wait two years Experienced hunger Natural fallow land Tired soils Education Wealth of household Club membership Age Intercept Percent correct predictions R-squared Log of likelihood function a

Estimate

Standard error

t-Statistic

P-valuea

0.482 0.150 )0.617 0.305 0.110 3.39 )0.66 1.58 )0.088 0.093 0.001 1.77 0.02 )7.73 74.8% 0.35 )56.78

0.322 0.084 0.648 0.541 0.543 1.14 0.531 0.827 0.690 0.093 0.362 0.556 0.022 2.15

1.490 1.79 )0.95 0.56 0.203 2.97 )1.24 1.91 )0.13 1.00 0.003 3.18 1.33 )3.60

0.07 0.04

0.005 0.02

0.005 0.09 0.005

One-tailed test.

other as suggested by the decision tree in Figure 1. Third, the significance of the gender and marital status variable means that women in female-headed households (FHHs) are more likely to adopt improved fallows than are married women or men in male-headed households (MHHs). This is completely in contrast with the WID literature (Due & Gladwin, 1991)

and our other findings in the UF Soils CRSP (D’Arcy, 1998; Uttaro, 1998), which tell us that FHHs are the last group to adopt, even though it is in accord with our participant observations in Eastern Zambia. We observed that FHHs in Eastern Zambia were very enthusiastic about improved fallow technologies (wrote songs about them, organized in clubs to adopt them,

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

taught each other the technology), and had pretty high adoption rates (47% FHHs adopt versus 31% of women in MHHs and 52% of men in MHHs). Regression results thus confirm ‘‘ethnographic observations,’’ because regression analysis allows the researcher to hold constant all other significant variables (e.g., household size, club membership, natural fallow land, ability to wait, and age) while focusing on one variable of interest. Because the gender variable is significant while holding these other variables constant, we know there is an additional independent positive impact on adoption from being an FHH. We return to this point in the conclusion. (e) Ordered probit analysis Will these results be supported by an ordered probit analysis? Ordered probit was used here because the dependent variable, tester type, is considered to be an ordered categorical variable: tester-expanders (TSYP ¼ 2) are the real adopters of the technology, more so than the farmers who test IFs only one time (the testernon-expanders, TSYP ¼ 1), and certainly more than the non-testers (TSYP ¼ 0). Results show the fit of the ordered probit equation is good (better than the logit analysis), as shown by the log of likelihood function and the significance of the coefficients. The results of the ordered probit analysis (Table 3) indicate that all the independent variables except educational level are significantly associated with whether or not the farmer adopts improved fallow technologies. The likely effect of the independent variables (all nominal variables except for education) in this case is that they significantly raise the probability of a farmer’s being an adopter from zero to a positive quantity, as all have a positive sign. Note that a farmer’s ability to wait two years to see the results of an

537

improved fallow has a more significant impact (at the 0.1% level) than either household size, the availability of natural-fallow land, and club membership at the 5% level. The least significant variable is gender and marital status (FMPFAM) at the 6% level. This means that men in male-headed households are less likely to adopt (i.e., become testers or tester-expanders) than are women in male headed households, and they both are less likely to adopt than women in female-headed households, holding the other variables constant, such as a farmer’s club membership and ability to wait two years to see results of an improved fallow, and household size (a labor proxy). This is a unique but not unwelcome finding for female-headed households. The wealth variable, as well as its proxy, ownership of oxen, was not included in this regression equation because it was not at all significant in the logit results in Table 2.

5. CONCLUSION In this paper, we have presented a method to predict farmer behavior, ethnographic decision trees, using a case study from Eastern Zambia in which we predicted 82% of farmers’ choices to adopt improved fallow technologies. This methodology, relying on data and observations gleaned from open-ended ethnographic interviews, agrees with Chambers’ (1983) philosophy of ‘‘putting the farmer first.’’ It goes a step further, however, and claims that participatory researchers who use decision-tree methodology in addition to participatory tools should be able to test and make predictions about peasant farmer choices, thereby eliminating the ‘‘u’’ in ‘‘lcddu’’ and refuting Chambers’ claim that peasant farmer behavior is unpredictable. Decision-tree methodology, however, requires the development analyst to use careful eliciting

Table 3. Ordered probit estimation of adoption of improved fallows, n ¼ 119 Variable

Estimate

Standard error

t-Statistic

P-value (1-sided)

Gender + marital status Natural fallow land Household size Club membership Ability to wait two years Education Intercept Log of likelihood function

0.233 0.764 0.068 0.444 1.85 0.027 )3.25 )78.9697

0.152 0.407 0.039 0.260 0.573 0.040 0.824

1.53 1.88 1.75 1.71 3.23 0.667 )3.95

0.06 0.05 0.05 0.05 0.001 0.001

538

WORLD DEVELOPMENT

techniques during in-depth ethnographic interviews to discover peasant farmers’ decision criteria. The sole use of PRA tools such as stakeholder analysis or Venn diagrams or preference rankings, although perfectly good tools for other purposes, are not going to substitute for the researcher’s hearing farmers verbalize their own decision criteria, the crucial step in the model-building phase of the research. Eliciting decision criteria from farmers takes time. Ethnographic interviews may take up to two hours apiece, followed by another hour or so when the researcher puts the information into a decision tree for that individual. The researcher should then verify that tree with another farmer in another ethnographic interview. Not uncommonly, that farmer may verbalize a completely different set of decision criteria, requiring the researcher to add a completely new path to the tree, that should then be verified with another farmer in the same culture or region. Because the researcher wants the tree to predict choice behavior of most of the farmers in a culture or region, this ‘‘test and revise’’ procedure used to build a decision tree can take months of careful interviewing of 25– 60 farmers, until the researcher is satisfied s/he has elicited all the criteria ‘‘shared’’ in that sample. Then the process of testing the tree begins, itself a lengthy process. Suffice it to say that ‘‘rapid’’ reconnaissance surveys are not usually long enough for researchers to build and test decision trees that predict. We conclude that both participatory methods and scientific rigor and testing procedures are both necessary and sufficient to guarantee that self-sustained adaptation in Africa takes place. Participatory methods are necessary to avoid the pitfalls of top-down approaches to development (Hyden, 1998), and to empower people so that they can voice their own solutions to their problems as they see them, so that the new technologies and interventions do solve local problems, e.g., soil fertility depletion in Eastern Zambia. But participatory research methods on their own are not sufficient to guarantee self-sustained adaptation of new technologies. Policy makers need to know when their programs are failing, and why, and so participant observers need to test their ethnographic observations and hypotheses. Both participatory and testing methods are necessary; and development analysts should in the future wear two interchangeable hats, one as participatory ethnographer and one as hy-

pothesis tester. To ask the right questions, the development analyst should do ethnography well and use all the methods available—cognitive, ethnographic, participatory—to formulate hypotheses about the people who own the development activity. As a follow up to use of these methods, the development analyst should also code the data that is now gathered but left uncoded; and test the hypotheses discovered via the use of participatory methods. We conclude that yes, now there is often a tradeoff between the quality of science and the use of participatory methodologies. Yet there does not have to be. There can be ‘‘a middle road’’ between ‘‘quick and dirty’’ RRA methods that rely on untested ethnographic observations, and on the other hand, long questionnaires that ask the wrong questions. As an example, we used a case study of adoption of improved fallow technologies by both men and women farmers in Eastern Zambia, to show how participatory methods, decision trees, and statistical tests can be used in complementary ways to answer important policy questions. 10 One example of such complementarities merits some discussion. The significance of the gender and marital status variable in the regression analyses means that, holding other variables constant, women in female-headed households are more likely to adopt improved fallows than are women or men in male headed households. This finding is consistent with the high adoption rates of improved fallows by FHHs one observes via participatory methods in the four ICRAF villages in Eastern Zambia, where 47% FHHs adopted versus 31% of women in MHHs and 52% of men in MHHs. It is confirmed by regression analysis, which holds constant other significant variables (e.g., household size, club membership, natural fallow land, ability to wait two years, and age) to focus on the one variable of interest. Here, the logit and ordered probit results clearly show FHHs are more likely to adopt improved fallows than men or women in MHHs, in contrast to the WID literature that suggests FHHs are likely to have less time to plant an improved fallow in the busy planting season due to the demands on their time from reproductive activities in the household. 11 Decision-tree results also agree, by showing that all FHHs passed the time constraint in the decision tree in Figure 2; and their most pressing constraint was land: four FHHs said they did not have enough land for a small improved fallow plot.

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH

Except for the land constraint, however, FHHs passed all the other constraints in Figure 2 almost as easily as did men in MHHs, and almost half of them adopted improved fallows. Why? We conclude that because they are poorer than the other two groups (married women and men in MHHs), FHHs realize they cannot afford inorganic fertilizers and therefore are now open to the idea of ‘‘growing’’ their own fertilizer. This shows the complementarity of the three methods used in this case study: participant observation, decision-tree models, and the logit and ordered probit analyses. This case also shows the value of hypothesis testing. Results from building and testing adoption-of-improved-fallow decision trees in Malawi (Uttaro, 1998) and Western Kenya (Williams, 1997), for brevity shown elsewhere (Gladwin et al., 2002), show the success of improved fallows in Eastern Zambia is partly due to the fact that Eastern Zambia is a region of lower population density than the other African regions. Farmers in Eastern Zambia have enough land to put some of it in a natural fallow, whereas farmers in Malawi and Western Kenya do not and so do not adopt improved fallows. Participant observation in these three locations suggests that natural fallows and improved fallows are complements, not substitutes for each other. Figure 1, however, posits they are substitutes, i.e., a farmer who has land ready to come out of natural fallow might be satisfied with his/her soil fertility measures and not need to plant an improved fallow. Yet results of testing this criterion show a high error rate: nine of the 18 cases sent to the outcome [Don’t Plant an IF] are errors of this model. In contrast, regression results are much more conclusive: in this sample, farmers with natural fallow land are more likely to adopt improved fallows, meaning they are complements.

539

Participatory tools, decision tree models, and regression analyses give different kinds of information to stakeholders and policy makers. Participatory tools allow the decision makers themselves to frame the decision context in terms they can understand. They also allow self-sustained adaptation of a new program or technology to occur, not merely top-down adoption, by encouraging individual and group experimentation with it. They do not, however, tell the policy maker why the people who are supposed to own the development activity are not doing so. Decision trees do that, by identifying all the decision criteria and limiting factors to adoption for the policy makers, the most idiosyncratic ones as well as the most important with the greatest number of nonadopters down their paths. Trees thus model a decision process in which people process a wealth of information and make many comparative judgments without precise numbers attached (Chambers, 1997, p. 41). As a result, decision trees are descriptively rich and give a wealth of information; and by doing so, they can predict individual behavior and thus take the ‘‘u’’ out of Chambers’ ‘‘lcddu.’’ They do not, however, provide the policy maker with a statistical test of significance that logit/probit and ordered logit models do. They do not allow the researcher to hold other significant variables constant while focusing on one variable of interest, to see if it is significant. Yet it is not unusual for only a few decision factors to be significant in a regression analysis, as the results in Tables 2 and 3 illustrate. Whereas trees contain all the criteria processed by individuals in a group, regression analysis highlights those that are significant in the behavior of that group. The three different kinds of methods therefore give different kinds of information to stakeholders and policy makers, and all are valuable.

NOTES 1. Hyden (1998) points out that ‘‘although we would like to believe that we are all engaged in a learning process and try to avoid the policy mistakes of the past, the new paradigms and approaches that we continue to develop have really failed to provide satisfactory answers. It seems as if we are caught more in a vicious circle than in a creative spiral when it comes to understanding what development is and how it can be promoted on an international level.’’

2. Ethnographic observations can be wrong. Unlike His Holiness the Pope when he speaks as head of the Catholic Church, ethnographers are not granted infallibility. 3. Previous UF Soil CRSP results in several African countries showed that men (and some women) farmers in male-headed households have either adopted or adapted various methods to replenish their depleted

540

WORLD DEVELOPMENT

soils, but that female-headed households (FHHs) have not (Gladwin et al., 1997). For example, small bags of fertilizer are usually bought for men’s fields and cash crops; FHHs do not risk getting credit, even microcredit, for fertilizer use on their food crops; vouchers and grants are not usually targeted at FHHs although they are among the poorest of the poor; and FHHs grow grain legumes only to eat, not turn under ‘‘green’’ to improve their soil fertility. Until we studied adoption of agroforestry innovations in the form of improved fallows, our results showed little chance of reversing soil fertility depletion and improving yields of food crops on women’s—especially FHHs’—fields, thus adding to the dismal reports and horror stories coming from Africa (prevalence of AIDs, corruption, soft governments). 4. In another paper (Gladwin et al., 2002), we present results from Western Kenya and Malawi which show that women do not adopt improved fallow technologies due to a lack of knowledge of the technology, a lack of enough land to put some of it in natural fallow, and a lack of labor. 5. At first it was planned to find 40 women who began testing improved fallows before 1995–96 in the four target camps. This was impossible, however, as only 28 women tested IFs before 1995–96, because most of the early testers were men. In many instances, however, farmers were so convinced of the success of the technology (especially after having visited farmers in other camps as part of field days or farmer-to-farmer visits) that they did not wait until they harvested their first IF before they planted another. Of the 81 women in the ICRAF sample, Peterson interviewed 40 NT, 23 TE, and 18 TNE; of the 40 men, she interviewed 15 NT, 16 TE, and nine TNE (Peterson, 1999, p. 4). 6. A more complicated subroutine may have to replace the simple criterion used here which assumes natural and improved fallows are substitutes for each other, e.g., add an additional constraint, ‘‘Do you have time and strength to clear this land?’’ 7. Alternatively, these results suggest adoption of improved fallows is a delayed reaction to structural adjustment policies that have raised the price of inorganic

fertilizers to levels so high that women farmers have finally ‘‘adjusted’’ by deciding to ‘‘grow their own fertilizer.’’ 8. One reviewer asked us to be more specific about the length of time a researcher needs to be immersed in a culture in order to elicit good decision criteria and build decision models that predict. Unfortunately, the answer to this question depends on several criteria, such as the extent of familiarity of the researcher with the culture and its language and customs, the previous fieldwork experience of the researcher, his/her ability to elicit decision criteria and recognize them when verbalized by informants, and whether cultural insiders feel threatened when asked to verbalize about a particular choice. Due to their experience and/or prior familiarity with the culture, some decision-tree modelers can elicit criteria and build decision models after only a few months in a culture; while some require years. Some decisions are of a more sensitive nature than others and thus require even more fieldwork time; for example, the decision to use illegal drugs—a choice we previously thought would be impossible to model with informants’ verbal reports—now has a decision-tree model (Johnson & Williams, 1993). 9. We are proposing an alternative model 3 (model 2 was discarded) in the belief that all models are simplifications of reality—just as model trains are simplifications of real trains—and the only test of a model is how well it predicts compared to another model (Lave & March, 1975). 10. Other researchers within the Sustainable Livelihood approach (Roe, 1998) have expressed the need for triangulation procedures with the use of participatory methodologies. Here, we are encouraging the use of rigorous testing procedures as one leg of the triangulation procedure. 11. When asked directly why FHHs adopted more than married women, women said lack of authority (malamuno) in the household was the reason married women did not adopt. But, this constraint could not be tested with these data because it was not asked of everyone.

REFERENCES Adesina, A., & Chianu, J. (2000). Understanding adoption processes of natural resource management technologies: adoption and modification of alley farming by farmers in Nigeria. Agroforestry Systems.

Arrow, K. J. (1951). The nature of preference and choice. In Social choice and individual values. New York: Wiley. Berlin, B., & Kay, P. (1969). Basic color terms. Berkeley: University of California Press.

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH Breslin, F. C., Gladwin, C. H., Borsoi, D., & Cunningham, J. A. (1999). Defacto client-treatment matching: how clinicians make referrals to outpatient treatments for substance use. Evaluation and Program Planning, 23, 281–291. Chambers, R. (1981). Rapid rural appraisal: rationale and repertoire. Public Administration and Development, 2(2), 95–106. Chambers, R. (1983). Rural development: Putting the last first. Harlow: Longman. Chambers, R. (1994a). The origins and practice of participatory rural appraisal. World Development, 22(7), 953–969. Chambers, R. (1994b). Participatory rural appraisal (PRA): analysis of experience. World Development, 22(9), 1253–1268. Chambers, R. (1994c). Participatory rural appraisal: challenges, potentials and paradigm. World Development, 22(10), 1437–1454. Chambers, R. (1997). Whose reality counts? Putting the first last. London: Intermediate Technology Publications. Chambers, R., & Conway, G. (1992). Sustainable rural livelihoods: practical concepts for the 21st century. Discussion Paper 296, Institute for Development Studies, University of Sussex. D’Andrade, R. G. (1981). The cultural part of cognition. Cognitive Science, 5, 179–195. D’Andrade, R. G. (1987). A folk model of the mind. In D. Holland, & N. Quinn (Eds.), Cultural models in language and thought (pp. 112–148). Cambridge: Cambridge University Press. D’Arcy, R. (1998). Gender and soil fertility project report: Dowa, Malawi. Report to the University of Florida Soils Collaborative Research Support Project, Gainesville, FL. David, S. (1992). Open the door and see all the people: intra-household processes and the adoption of hedgerow intercropping. Paper presented at the Rockefeller Foundation Social Science Fellows’ Meeting, CIMMYT, Mexico, November 9–13. Due, J., & Gladwin, C. H. (1991). Impacts of structural adjustment programs on African female-headed households. American Journal of Agricultural Economics, 73, 1431–1439. Frake, C. (1964). How to ask for a drink in Subanun. American Anthropologist, 66(2), 127–132. Franzel, S., Phiri, D., & Kwesiga, F. (1997). Participatory on-farm research on improved fallows in Eastern Province, Zambia. Zambia/ICRAF Agroforestry Research Project, Chipata, Zambia. Gardner, H. (1985). The mind’s new science. New York: McGraw-Hill. Gladwin, H. (1974). A study of the relationship between verbalization and deeper cognitive skills in learning a complex task. Final report, Project No. 2-0650, grant No. OECO-72-1879, US Department of Health, Education, and Welfare. Gladwin, C. H. (1976). A view of the Plan Puebla: an application of hierarchical decision models. American Journal of Agricultural Economics, 58(5), 881–887.

541

Gladwin, C. H. (1979). Cognitive strategies and adoption decisions: a case study of non-adoption of an agronomic recommendation. Economic Development and Cultural Change, 28(1), 155–173. Gladwin, C. H. (1980). A theory of real-life choice: applications to agricultural decisions. In Agricultural Decision Making: Anthropological Contributions to Rural Development (pp. 45–85). New York: Academic Press. Gladwin, C. H. (1983). Contributions of decision-tree methodology to a farming systems program. Human Organization, 42(2), 146–157. Gladwin, C. H. (1989). Ethnographic decision tree modeling. Newbury Park: Sage. Gladwin, C. H. (1991). Fertilizer subsidy removal programs and their potential impacts on women farmers in Malawi and Cameroon. In C. H. Gladwin (Ed.), Structural adjustment and African women farmers (pp. 191–216). Gainesville, FL: University of Florida Press. Gladwin, C. H. (1992). Gendered impacts of fertilizer subsidy removal programs in Malawi and Cameroon. Agricultural Economics, 7, 141–153. Gladwin, C. H., Buhr, K. L., Goldman, A., Hiebsch, C. K., Hildebrand, P. E., Kidder, G., Langham, M., Lee, D., Nkedi-Kizza, P., & Williams, D. (1997). Gender and soil fertility in Africa. In P. Sanchez, & R. Buresh (Eds.), Replenishing soil fertility in Africa (pp. 219–236). Madison, WI: SSA, Soil Science Society of America Special Publication no. 51. Gladwin, C. H., Gladwin, H., & Peacock, W. G. (2001). Modeling hurricane evacuation decisions using ethnographic methods. International Journal of Mass Emergencies and Disasters, 19(2), 117–143. Gladwin, H., & Murtaugh, M. (1979). The attentivepreattentive distinction in agricultural decision making. In P. Barlettt (Ed.), Agricultural decision making (pp. 115–134). New York: Academic Press. Gladwin, H., & Murtaugh, M. (1984). Test of a hierarchical model of auto choice on data from the national transportation survey. Human Organization, 43(3), 217–226. Gladwin, C. H., Peterson, J. S., Phiri, D., Uttaro, R., & Williams, D. (2002). Agroforestry adoption decisions, structural adjustment, and gender in Africa. In C. B. Barrett, F. M. Place, & A. A. Aboud (Eds.), Natural resource management in African agriculture: understanding and improving current practices. London: CABI Publishing. Gladwin, C. H., & Zabawa, R. (1984). Microdynamics of contraction decisions: a cognitive approach to structural change. American Journal of Agricultural Economics, 66(5), 829–835. Grandin, B. (1988). Wealth ranking. London: Intermediate Technology Publications. Hildebrand, P. E. (1981). Combining disciplines in rapid appraisal. The sondeo approach. Agricultural Administration, 8, 423–432. Hill, P. (1956). The Gold Coast cocoa farmer. London: Oxford University Press. Hill, P. (1963). The migrant cocoa-farmers of Southern Ghana: A study in rural capitalism. Cambridge: Cambridge University Press.

542

WORLD DEVELOPMENT

Hyden, G. (1998). Governance and sustainable livelihoods. Gainesville, FL: Center for African Studies. Johnson, J., & Williams, M. L. (1993). A preliminary ethnographic decision tree model of injection drug users’ (IDUs) needle sharing. International Journal of the Addictions, 28(10), 997–1014. Kahneman, D., & Tversky, A. (1972). Subjective probability: a judgement of representativeness. Cognitive Psychology, 3, 423–432. Kahneman, D., & Tversky, A. (1982). The psychology of preferences. Scientific American, 246, 160–173. Kanji, N., & Gladwin, C. H. (1999). Gender and livelihoods in Gorno Badakshan: a report to the Mountain Societies Development Support Program, Tajikistan. Aga Khan Foundation, Geneva, Switzerland. Kwesiga, F. R., Franzel, S., Place, F., Phiri, D., & Simwanza, C. P. (1997). Sesbania improved fallows in Eastern Zambia: their inception, development, and farmer enthusiasm. Zambia/ICRAF Agroforestry Research Project, Chipata, Zambia. Lave, C., & March, J. (1975). An Introduction to models in the social sciences. New York: Harper and Row. Matlon, P., Cantrell, R., King, D., & Benoit-Cattin, M. (1984). Coming full circle: Farmers’ participation in the development of technology. Ottawa: International Development Research Centre. McCracken, J. A., Pretty, J. N., & Conway, G. R. (1988). Introduction to RRA for agricultural development. London: International Institute for Environmental Development. Miller, G. (1956). The magic number seven, plus or minus two: some limits on our capacity for processing information. Psychological Review, 63, 81–97. Miller, G., Galanter, E., & Pribraum, K. (1960). Plans and the structure of behavior. New York: Holt, Rinehart, and Winston. Mukhopadhyay, C. (1984). The sexual division of labor among Californian families with working women. Human Organization, 43(3), 227–242. Mwale, A. C. (2000). Adoption of soil and land conservation technologies in Zambia: understanding farmers’ adoption decision criteria. M.S. Technical Paper, College of Natural Resources and the Environment, University of Florida, Gainesville, FL. Nabasa, J., Rutwara, G., Walker, F., & Were, C. (1995). Participatory rural appraisal: Practical experiences. London: Natural Resources Institute, Overseas Development Administration. Newell, A., & Simon, H. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Norman, D. W. (1975). Rationalizing mixed cropping under indigenous conditions: the example of Northern Nigeria. Samaru Research Bulletin 232, Institute for Agricultural Research, Samaru, Ahmadu Bello University, Zaria, Nigeria. Peterson, J. S. (1999). Kubweletza nthaka: ethnographic decision trees and improved fallows in the Eastern Province of Zambia. Report to the University of Florida’s ‘‘Gender and Soil Fertility in Africa’’ Soils Management Collaborative Research Support Program (CRSP) and the International Centre for Research on Agroforestry, May.

Peterson, J. S., Tembo, L., Kawimbe, C., & Mwang’amba, E. (1999). The Zambia Integrated Agroforestry Project baseline survey: lessons learned in Chadiza, Chipata, Katete and Mambwe Districts, Eastern Province, Zambia. Report to the World Vision International Zambia Integrated Agroforestry Project staff, the University of Florida’s ‘‘Gender and Soil Fertility in Africa’’ Soils Management Collaborative Research Support Program, the International Center for Research on Agroforestry, and the Zambian Ministry of Agriculture, Food, and Fisheries. Quinn, N. (1978). Do Mfantse fish sellers estimate probabilities in their heads? American Ethnologist, 2, 19–46. Quinn, N. (1990). A model of American marriage. Cambridge: Cambridge University Press. Rachlin, H. (1990). Judgment, decision, and choice. New York: W.H. Freeman. Rhoades, R. (1984). Breaking new ground: Agricultural anthropology. Lima, Peru: International Potato Center. Roe, Emory M. (1998). Policy analysis and formulation for SL. Available: Http://www.undp.org/sl/. Roling, N. G. (1988). Extension science: information systems in agricultural development. Wye studies in agriculture and rural development. Cambridge: Cambridge University Press. Romney, A. K., & D’Andrade, R. (1964). Cognitive aspects of English kin terms. American Anthropologist, 66, 146–170. Schank, R., & Abelson, R. (1977). Scripts, plans, goals, and understanding. New York: Wiley. Schoemaker, P. (1982). The expected utility model: its variants, purposes, evidence, and limitations. Journal of Economic Literature, 20, 529–563. Simon, H. (1972). On how to decide what to do. Bell Journal of Economics, 494–507. Simon, H. A. (1979). Information processing models of cognition. Annual Review of Psychology, 30, 363–396. Spradley, J. P. (1979). The ethnographic interview. Orlando, FL: Harcourt Brace Jovanovich. Spradley, J. P. (1980). Participant observation. New York: Holt, Rinehart, and Winston. Spring, A., Sullivan, A., & Litow, P. (2000). Participatory rural appraisal in St. Lucia: A report to the FAO. Gainesville, FL. Sullivan, A., Litow, P., & Spring, A. (2000). Region one, participatory rural appraisal report to the FAO. Gainesville, FL. Swinkels, R., & Franzel, S. (1997). Adoption potential of hedgerow intercropping systems in the Highlands of Western Kenya: II. Economic and farmers’ evaluation. Experimental Agriculture, 33, 211–223. Tversky, A. (1972). Elimination by aspects: a theory of choice. Psychological Review, 28, 1–39. Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453–458. Uttaro, R. P. (1998). Diminishing returns: soil fertility, fertilizer, and the strategies of farmers in Zomba RDP in Southern Malawi. Report to the University

THE QUALITY OF SCIENCE IN PARTICIPATORY RESEARCH of Florida Soils Collaborative Research Support Project, Gainesville, FL. Von Neumann, J., & Morgenstern, O. (1947). Theory of games economic behavior. Princeton: Princeton University Press.

543

Werner, O., & Schoepfle, M. (1987). Systematic fieldwork (Vols. 1 and 2). Newbury Park, CA: Sage. Williams, D. E. (1997). Gender and integrated resource management: the case of Western Kenya. Masters’ Thesis, University of Florida, Gainesville, FL.