Content uploaded by Julian Park
Author content
All content in this area was uploaded by Julian Park on Jul 27, 2019
Content may be subject to copyright.
Learning selection: an evolutionary model for
understanding, implementing and evaluating
participatory technology development
B. Douthwaite*, J.D.H. Keatinge
1
, J.R. Park
Department of Agriculture, University of Reading, Earley Gate, PO Box 236, Reading RG6 6AT, UK
Received 16 July 1999; received in revised form 25 January 2001; accepted 2 April 2001
Abstract
This paper develops a model of the early adoption process that takes into account modi-
fications made by users. The model is based on data from 13 attempts to introduce six post-
harvest technologies into the Philippines and Vietnam. It is built on an analogy between
technology change and Darwinian evolution. At the core of the model is the interactive
experiential learning process — learning selection (LS) — that is analogous to natural selec-
tion in the living world. In learning selection stakeholders engage with a new technology,
individually playing the evolutionary roles of novelty generation and selection, and in their
interactions creating recombinations of ideas and experiences and the promulgation of bene-
ficial novelties. Peoples’ motivations to engage in learning selection, and its outcomes, are
influenced by the interaction between their lifeworlds and their environments. The model has
implications for management of agricultural technology change. It suggests the need for a
nurturing of new technology during its early adaptation and adoption, until the point where
the beneficiary stakeholders (manufacturers and users) are sufficiently numerous and have
adequate knowledge to play the evolutionary roles themselves. The LS model, while devel-
oped with data from agro-mechanical technologies, could provide a theoretical underpinning
for participatory technology development. #2002 Elsevier Science Ltd. All rights reserved.
Keywords: Participatory technology development; Social construction of technology; Model of adoption
process; Postharvest technology; Philippines; Vietnam
0308-521X/02/$ - see front matter #2002 Elsevier Science Ltd. All rights reserved.
PII: S0308-521X(01)00071-3
Agricultural Systems 72 (2002) 109–131
www.elsevier.com/locate/agsy
* Corresponding author. Present address: International Institute of Tropical Agriculture (IITA), Iba-
dan, Nigeria.
E-mail address: b.douthwaite@cgiar.org (B. Douthwaite).
1
Present address: International Institute of Tropical Agriculture (IITA), Ibadan, Nigeria
1. Introduction
The Food and Agriculture Organization estimates that international assistance to
agricultural development fell from $16 billion in 1988 to $10 billion in 1994 (ODI,
1997). In 1996 the Organisation for Economic Co-operation and Development
(OECD) agreed international development targets that included a halving of the
proportion of people living in extreme poverty by 2015 (OECD, 1996). Poverty
reduction has since been adopted as a goal by many donor organisations. The
implication of these two developments is clear. If international agricultural
research wants to increase its funding levels then it must demonstrate that it is
having an impact in poor farmers’ fields to a greater degree than it has done in the
past.
Unfortunately though, Public-sector agricultural research is sometimes seen as
having more impact in richer farmers’ fields than in poorer ones. One commonly
cited reason is that researchers implicitly assume a relatively simple mental map
of the research, development and transfer process that better matches the rela-
tively simple farming systems found in favourable production environments
(where farmers tend to be richer) than those found in less favourable production
environments. This simple mental map has been called the Transfer of Tech-
nology (TOT) model or Central Model (Chambers and Jiggins, 1986; Biggs,
1989). It is based on the positivist paradigm that underpins conventional science
and considers that reality is objective, independent and based on natural laws
that science can uncover (Ro
¨ling, 1996). According to the TOT view, scientific
method can be used to understand reality and design technologies. Hence agri-
cultural scientists believe that they can and should be able to deliver technolo-
gies that work in farmers’ fields. Local knowledge might be important for fine-
tuning, but this can be captured during on-farm testing prior to release. The
technology should not be released before it has been ‘perfected’ by which time
the researchers have finished their job. It is then up to the extensionists to
deliver the package to the farmers who either do or do not adopt, but are not
expected to make innovative changes. This model has worked well in generating
and delivering simple technologies — high yielding crop varieties — into rela-
tively favourable production systems, and in the process spawned the Green
Revolution.
In 1995 the International Rice Research Institute (IRRI) and the University of
Reading began a study to examine the extent to which the TOT view fitted the
innovation processes of relatively complex new agricultural technology. This paper
presents an evolutionary model of the early stage of the adoption process which may
fit reality much better than the TOT model. While developed from agricultural
engineering data it is suggested here that the model is also relevant to planning,
implementing and evaluating participatory technology development in other dis-
ciplines. Participatory approaches are increasingly being recognised as being more
effective at achieving adoption and impact in poor farmers’ fields than the technol-
ogy generation and transfer approaches traditionally used by much agricultural
research (World Bank, 1996).
110 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
2. Methodology
One of the challenges facing an ex-post study of innovation is that there is no
evidence left of the critical steps in the innovation process, and that people have
forgotten about them. This study considers agricultural machinery because it has a
physical manifestation that is relatively easy to modify (cut and weld) and it then, in
effect, leaves a ‘fossil’ record of adaptation because the changes are difficult to
destroy.
A case study methodology is used because technology adoption is a complex
process (Tidd et al., 1997) and case study methodology is, ‘‘a method for learning
about a complex instance, based on a comprehensive understanding of that instance
obtained by extensive description and analysis of that instance taken as a whole and
in its context.’’ GAO (1987, p. 9).
The case study technologies chosen are all the rice harvesting and rice drying
technologies introduced to the Philippines and Vietnam after 1975. The definition of
‘introduced’ used in the paper was that at least 100 machines had been used in either
country. Two types of harvester qualified. Both are relatively cheap and light
mechanical harvesters, and achieve this by being controlled by an operator who
walks behind the machine rather than riding on it. Four types of dryer were also
eligible. They ranged in capacity and cost from the locally-made SRR dryer — SRR
is an acronym for ‘‘very low cost’’ in Vietnamese — which can be bought for $100
and dries 1 t of rice in 2–4 days, to recirculating dryers imported from Taiwan which
cost 150 times more but can dry 6 t in 8–10 h. The case study technologies are shown
in Tables 1 and 2.
The Philippines and Vietnam were chosen to take advantage of a natural experi-
ment (Freedman, 1991) created by the Postharvest Technologies for Rice in the
Humid Tropics project.
2
This project gave National Agricultural Research and
Extension Systems (NARES) in the Philippines and Vietnam similar resources to
pilot test two technologies — the stripper gatherer (SG) harvester, and the low-
temperature in-bin drying system (LT-IBDS), from which the SRR dryer was
developed. The inputs into the pilot testing, modification and initial dissemination
programs set up by the NARES in both countries were the same although the out-
comes were quite different. This natural experiment allowed an analysis of the effect
of people (their lifeworlds)
3
and the environment on the innovation process. Also,
the GTZ Project generated a great deal of process documentation which is necessary
for developing a detailed understanding of the adoption process. Furthermore
the first author was a complete participant (project leader and chief designer) in the
2
This project was co-ordinated by IRRI and funded by the German government, through GTZ. In the
implementation phase of the project NARS chose postharvest technologies which they were then funded
to evaluate and initiate commercialisation. The implementation phase ran from January 1993 to Decem-
ber 1997.
3
Lifeworlds are the realities that people adaptively construct for themselves. They are the sum total of
the mental maps and models that people have built to allow them to cope in their environments and as
such are made up of past experience and personal and shared understanding (Long, 1992).
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 111
Table 1
Description of case study technologies
Technology Description Adoption status Lab. prod
a
.ha
1
Cost $
Stripper Gatherer (SG) harvester Walk-behind harvester 140 units sold in 5 years (Philippines) 7.5 2000
Mechanical reaper Walk-behind harvester 1071 units sold in 8 years (Philippines) 15 3000
SRR dryer Low temperature dryer 700 units sold in 3 years (Vietnam) 6.4 100
Flatbed dryer Heated air dryer with manual mixing
a
1000 units sold in 17 years (Vietnam) 4.8 2000
Flash dryer High temperature dryer 2000 units donated in 4 years (Philippines) 3
b
3500
Recirculating dryer Heated air dryer with mechanical mixing 1500 units sold in 6 years (Philippines) 1.5 15,000
a
Drying to 18% m.c. (wet base) not 14% m.c. as the other dryers.
b
Lab. prod.=labour productivity measured in person hours per tonne of paddy rice (rough rice).
112 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
development of the SG harvester and a participant observer
4
(team member) in
the development of the SRR dryer. Participant observation is a way of gathering
data on the motives, meanings and experiences of stakeholders in a social process
like innovation (Burgess, 1984). Hence the SG harvester and SRR dryer case studies
were constructed with much more detail and data than the four others.
Although there is debate about whether the technology generation and adoption
process is linear there are nevertheless discernible stages in the life of an innovation.
One categorisation, after Yin (pers. comm. with S. Sechrest, 1996), is shown in
Fig. 1. According to the TOT approach researchers should hand-over their tech-
nology to extension workers at the beginning of the adaptation phase, hence this
phase and the start-up phase before it, are the two of interest in this study to test the
validity of the TOT view of innovation.
Case studies may be criticised for lacking both internal and external validity
(GAO, 1987; Yin, 1989; Sechrest et al., 1996). The main threat to external validity is
that generalisation of the findings to other situations is not legitimate. The four
‘minor’ case studies provided a check to the relevance of the findings from the two
‘main’ ones. Life histories of all 13 innovation attempts were constructed describing
the time-ordered sequence of events, the stakeholders who were responsible for or
influenced the events, and other contextual influences (Sechrest et al., 1996). As well
as analysing the individual life histories, cross case comparisons were also made.
Inaccuracy, bias and lack of objectivity are the main threats to internal validity
(Yin, 1989). The use of life histories is an important guard against these. In addition,
the large amount of process documentation for each of the main case studies, and the
use of multiple data sources for all case studies, helped maintain internal validity.
Table 2
Introduction of case study technologies in the Philippines and Vietnam
Technology Source of innovation Philippines Vietnam
Harvesting
SG Public @(2) @
Reaper Public @@
Private @
Drying
SRR Public @
Flatbed Public @(2) @
Flash Public @(2)
Recirculating Private @
Bold indicates main case study technology.
4
Complete participant and participant observer are two ideal field roles described by Gold (quoted by
Burgess, 1984). For more details see Douthwaite (1999).
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 113
Multiple data sources included:
1. surveys of key stakeholders (manufacturers/suppliers and users);
2. participant observation (for the main case studies only);
3. process documentation and other secondary literature;
4. key informant interviews (often recorded and transcripts made); and
5. documentation of physical evidence by photograph and video.
The whole population of manufacturers was interviewed for the main case study
technologies, and half the population of adopters, or 40, whichever was smaller. For
the minor technologies an attempt was made to interview two manufacturers/sup-
pliers and six adopters. Another methodological safeguard was that key people were
asked to read the main case studies as a further guard against inaccuracy, and as a
check against subjectivity and bias.
5
One challenge to constructing effective and persuasive case studies is to avoid
becoming swamped in data and detail. Sechrest et al. (1996) recommend construc-
tion of a theory of the case to act as a guide to data gathering and analysis. Our
theory of the case was based on an analogy that has been made between technology
change and Darwinian evolution (Nelson and Winter, 1982; Mokyr, 1990). If this
analogy is valid then technology change must be driven by a process analogous to
natural selection. Hence our theory of the case is simply that there is an analogue,
which we call learning selection, shown in Fig. 2. Learning is central to innovation
(Nelson and Winter, 1982; Mokyr, 1991; Clark, 1995; Leonard, 1995) which is why
it is the basis of the learning selection model. Kolb’s (1984) experiential learning
Fig. 1. Stages in the innovation process.
5
Caesar J.M. Tado, Senior Research Specialist currently working on stripper harvesting at the Phi-
lippine Rice Research Institute commented on the SG harvester case studies, and Martin Gummert, for-
mer GTZ project manager and drying expert commented on the SRR dryer case study.
114 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
model was chosen in particular because the two types of learning that characterise
the innovation process — ‘learning by using’ and ‘learning by doing’ (Rosenberg,
1982) — are both types of experiential learning. Rosenberg defines ‘learning by
using’ as learning during manufacturing that leads to improvements in the manu-
facturing process while ‘learning by doing’ is learning during the use of the tech-
nology that make it work better.
The analogy suggested here between natural selection and learning selection is not
perfect. Rather, it is an ‘analogy as a heuristic’; an analogy that suggests ways of
thinking about innovation processes from the much better understood evolutionary
process (Ruse, 1984). One obvious difference between natural and learning selection
is that natural selection is ‘mindless’ while learning selection is not — genetic muta-
tions occur at random but farmers make changes to their machines for a reason.
This difference means that to make learning selection into a useful tool for under-
standing and predicting the outcomes of the early stages of innovation processes, an
understanding of the motivations of people involved in learning processes is
required. For instance, what motivates people to want to interact with new tech-
nology in the first place? ... what is likely to influence the outcomes of the learning
selection iterations they go through? Some factors will help the technology evolve.
Fig. 2. The learning selection algorithm, analogous to natural selection.
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 115
Long’s (1992) actor-oriented approach is designed to understand the ways in which
groups interact with an intervention such as a new technology, and how these
interactions are affected by cultural, social, political and power dynamics.
Douthwaite et al. (2000) examine how the actor-oriented approach and the learning
selection approach can be integrated.
Fig. 2 is a model of how learning selection works. It illustrates two of the poten-
tially many participants involved in their own learning cycles while at the same time
interacting with others. Individually participant iand jare carrying out two of the
three roles necessary within an evolutionary system:
1. Novelty generation that creates differences between individual members of the
species (Nelson, 1987), e.g. individual differences between machines of the same
type, or the way they are used.
2. Selection of beneficial novelties (Nelson, 1987).
They do this during an experiential learning process based on work by Kolb
(1984) and Hunt (1987) and consisting of four stages shown in Fig. 2.
Experience — suppose a farmer finds that the rice miller pays her a low price for
the grain dried in her dryer because some of it is not properly dried.
Making Sense — she reflects on this experience from different points of view to
give it meaning. For example, she realises that uneven drying is losing her money
and that it might be sensible to try and improve the dryer’s performance.
Drawing conclusions — she then develops personal explanations of what hap-
pened from her own or others previous experience or theories. For example, she
hypotheses that if she reduces the amount of paddy she loads into the dryer then
drying will be more uniform.
Action — she then decides to test her hypothesis, and in so doing generates a novelty.
Testing the novelty begins another learning cycle. Her selection decision to adopt
or reject the novelty will depend on whether the rice miller pays her more for her
product. The miller will make this price decision after going through his own learn-
ing cycle when he tests a sample of her rice for milling quality. If the farmer is par-
ticipant iin Fig. 2 then the miller represents participant j.
The third component required of an evolutionary system is a promulgation and
diffusion mechanism. In our example promulgation of the novelty occurs when the
farmer tells people in her social network, represented in Fig. 2 by the ‘other partici-
pants’ box, about the benefits of her novelty and they select to adopt it. Moreover,
many of these people may be going through their own learning cycles creating the
conditions for the recombination of differing observations and experiences that can
lead to the generation of novelties that have ‘hybrid vigour’. In the process the
technology evolves and with it the participants’ opinions and knowledge of it and
the way they organize themselves to use and promote the technology. These pro-
cesses are all involved in learning selection, and the implicit assumption is that they
will lead to improvements in fitness of the technology, where fitness is taken in the
biological sense to mean improvements in the likelihood that the technology will be
adopted and promulgated. This concept of fitness, or adoptability, is similar to
116 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
Lyotard’s (1996) concept of performativity, which he defines as the best possible
input/output ratio. Lyotard argues that performativity itself is the main form of
legitmisation of knowledge.
Part of the purpose of developing the learning selection model is to guide case
study data collection, as already discussed. The model does this by suggesting that
life histories describe the degree to which learning selection took place. Modifica-
tions made to a technology are evidence of learning selection having been carried
out. The life histories should therefore identify modifications and evaluate their
effect on the performance of the technology. This is an investigation of the novelty
generation function. Data should also be collected on the selection and promulgation
functions, namely, which modifications were selected and who promoted the mod-
ifications and the technology itself, and how. Contextual data should be gathered on
participant characteristics and their environment to explain the outcomes of the
learning selection iterations. This data will also help identify factors that affected
peoples’ motivation to participate.
3. Case study findings
The six technologies shown in Table 1 were major inventions when they were first
introduced into Vietnam and the Philippines. The 13 subsequent innovation attempts
provide data to support Mokyr’s (1990) conclusion that major inventions are nearly
always followed by a learning process which improved the ‘fitness’ of the technology.
The case studies of the technologies with a public source of innovation (public tech-
nologies) clearly show that manufacturers, users and sometimes researchers made
many modifications and selection decisions after commercialisation. While the overall
effect of these changes was to generally reduce production and operation costs
(improve input/output ratio), some changes, particularly in the early adoption stages,
proved detrimental to the technology’s fitness and its evolutionary prospects. Exam-
ples of equipment evolution after commercialisation are given for four technologies,
followed by an examination of the degree to which learning selection occurred. The
full details of the case studies are presented in Douthwaite (1999).
3.1. Examples of equipment evolution
3.1.1. The flatbed dryer in Vietnam
A university lecturer installed the first flatbed dryer in 1983 in Soc Trang Province
in the Mekong Delta. The dryer was introduced into an area where the move from
single rice cropping to double rice cropping, made possible by the development by
IRRI of shorter growth duration rice varieties, was being seriously constrained
by problems drying the first crop in the middle of the wet season. By 1985 local
entrepreneurs began copying the design but made mistakes in replicating the blowers
due to a lack of technical knowledge. As a result rice traders paid 5% less for flat-
bed-dried rice compared with sun-dried rice. Nevertheless farmers used the dryers
because they were often the only way of drying their produce. Over a 10-year period
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 117
a whole set of improvements to the dryer and the way it was used meant that the
price paid for mechanically dried paddy was by 1994 5% higher than the price paid
for sun-dried paddy. These improvements were made by manufacturers and users as
part of ‘learning by doing’ and ‘learning by using’, respectively. In this period the
cost to farmers of using flatbed dryers fell from as much as 12% of the value of
the yield to 5%. A reduction in fuel costs was largely responsible for this fall,
brought about by village craftspeople realising that their indigenous rice hull cook-
stove could be used as the basis of a furnace design that would burn rice hull instead
of more expensive wood. As the flatbed dryer became cheaper to use, it spread
without public sector intervention, beyond its cradle in Soc Trang Province to other
areas of the Mekong Delta. There are now an estimated 1000 dryers in the Mekong
Delta drying about half a million tonnes of paddy per year, and in so doing saving
Vietnam millions of dollars.
3.1.2. The SG harvester in the Philippines
The SG harvester was first commercialised in 1993. The first farmers to adopt in
the Philippines, and the first larger-scale manufacturer to start building, did so after
hearing about the technology in newspaper and other media stories that resulted
from an IRRI press-release. The initial units sold had problems with the reliability
of the ground-drive transmission, due in part to the substitution by manufacturers of
cheaper and lower quality components. In 1995 IRRI released the Mark II SG
harvester design which had a heavier duty transmission, as well as other strength-
ened components. This reduced the reliability problem, but increased the weight by
31%, making the machine harder to manoeuvre and more liable to bog down in soft
field conditions. In 1996 a manufacturer — Morallo Metal Industries — developed
its own version of the Mark II design that was almost as light as the first version but
far more reliable. Another manufacturer developed an idea for an improved wheel
design that was further developed by IRRI and proved to give much better mobility
in muddy field conditions. In 1997 IRRI released drawings of the Mark III SG
harvester based on the Morallo SG harvester and incorporating the novel wheel.
3.1.3. The SRR (‘‘very low cost’’) dryer in Vietnam
The first SRR dryer was sold to a farmer in October 1995, built by an R&D team
at the University of Agriculture and Forestry (UAF). The first units sold were fitted
with a 1 kW electric cooker element to heat the air blown into the paddy being dried.
Again, the first farmers to buy heard about the dryer from the media and came to
UAF to find out more about the technology. Initially sales were limited to areas
near Ho Chi Minh City where farmers had a good electricity supply. In 1996 the
R&D team added a coal stove as an optional replacement for the electric cooker
element. This meant that the SRR dryer could be used in areas with a poor elec-
tricity supply (low voltage). Another manufacturer started copying the SRR dryer in
1995 and began supplying dryers with a 110 V motor rather than 220 V one to
operate in areas with poor electricity supply. He also reduced the power demand and
capacity of the blower. These modifications increased the potential market for the
SRR dryer still further but the 110 V motors proved more liable to damage from
118 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
voltage fluctuations and the reduced blower capacity reduced drying efficiency. The
net effect of these manufacturers’ modifications on the fitness of the technology was
negative.
Owners of the SRR dryer made important modifications to the operation instruc-
tions provided by UAF with the machine. UAF’s instructions to turn the heater off
during the day were to minimise fuel costs, but led to long drying times. Most farmers,
however, had more than one batch (1 t) to dry at a time and so reducing drying time
was their main priority. Hence most ignored the UAF recommendation and kept the
heater on all, or nearly all, of the time. Even if they did turn the heater off, hardly any
used the UAF-recommended drying strategy, choosing instead to devise versions of
their own, matched to their electrical supply, the initial moisture content of paddy to
be dried, and personal preference. By adopting such strategies owners were able
to reduce the drying time by 39%, or by 42 h for a 22% increase in energy costs if coal
was used, or 37% increase if an electric heater was used.
3.1.4. The mechanical reaper in the Philippines
The Kubota mechanical reaper sold in the Philippines is one of the two technol-
ogies that had a private sector source of innovation (private technology). The same
machine is still being sold and has not been modified at all. This is in contrast to the
public technologies which were all modified, some a great deal, after first commer-
cialisation. Very limited numbers of changes have been made to the other private
technology in the survey, the recirculating dryer. The private technology hardware
evolved less than the public technologies because both had been developed, com-
mercialised and ‘perfected’ in other countries before introduction into the Phi-
lippines, and because the designs were being built by only one manufacturer. Public
technologies were often built by many manufacturers and so had multiple sources of
novelty generation and selection. For example 12 manufacturers have built the SG
harvester. The advantage, therefore, of private technology is that it can be expected
to work better when first released onto the market. The disadvantage is that there is
just one source of hardware novelty generation that is often geographically distant
from the market. As a result private technology can be expected to be less responsive
than public technology to changing local needs.
While the private technology hardware changed little, owners and operators made
important changes to the way they used the technologies, as they did with the public
technologies. For example, teams of harvest labourers now hire Kubota mechanical
reapers when prior to 1983 all harvesting was done by hand. When the Kubota
reaper was first introduced owners competed directly with manual harvest teams for
work. Manual harvesters responded in some cases by sabotaging the machines by
placing iron rods in unharvested crops to destroy the reaper cutterbar.
6
Another
strategy for discouraging the use of the machine was for labourers to refuse to hand
harvest crops for reaper owners when the reaper did not work, for example in badly
lodged crops and deep mud. The organisational innovation that helped solve this
6
The sabotage technique was also used when mechanical harvesting was first introduced into the UK
(Farming Today on BBC Radio 4, 7 November, 1998).
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 119
problem is one that allows the labourers to decide if and when to use the reaper, so
they no longer see it as competition. Rather they see it as a useful tool that allows
them to harvest a greater area and earn more. Increasing labour shortage due to
alternative off-farm employment has meant that harvesting teams have not come
into conflict with each other. Harvest teams now use the machine in good crop
conditions, while in poor crop and field conditions they hand harvest but charge
farmers more. This willingness to compensate for the technical shortcomings of the
machine has made them less of a constraint to usage and adoption. The SG har-
vester case study showed that similar technical limitations have reduced the utilisa-
tion rate of the technology.
3.2. Learning selection carried out by stakeholders
3.2.1. Researchers
The researchers’ main role was to develop a promising prototype of a new technol-
ogy. This ‘plausible promise’ began the learning selection process by convincing at
least some manufacturers and farmers that it could be of benefit to them, and hence to
adopt. However, rather than handing the technology over to an extension organisa-
tion at this point, as the TOT model dictates, in many cases the R&D team continued
to stay involved. In eight out of the 13 case studies the R&D team played a clear
‘product champion’ role in working after initial adoption ‘‘to push the new technology
through the innovation process and overcome obstacles’’ (Peters and Waterman,
1982, p. 64). In terms of the learning selection model the researchers championed their
technologies by selecting beneficial modifications made by manufacturers and users,
and then promulgating them. Researchers often facilitated learning selection by iden-
tifying ‘knowledge gaps’ that were causing the key stakeholders problems, and then
working to fill them, while at the same time recognising and filling the gaps in their
own learning. The net effect of the R&D teams efforts was often to build a ‘common
property regime’ in which manufacturers, who would otherwise be competing with
each other, contributed their ideas towards the common goal of developing a better
technology. Engineer Lawrence Morallo of Morallo Metal Industries explained why
he let IRRI take his improved design (see Fig. 3) and then give the drawings out for
free. ‘‘Further improvements can still be made on the design that we gave to IRRI.
Improvements can always be made to the stripper. Even other manufacturers can still
make changes. At least we can also get ideas from them. We can adapt their ideas.’’
One of the main impediments to researcher learning selection were government
programs that began widespread diffusion of technologies too early because they
assumed the machinery was sufficiently perfected. Eight of the 11 public sector
innovations were promoted in nation-wide programs which began very early — on
average just 2.3 years after research started. According to Collinson and Tollens
(1994) it can take 10 years to produce a useful technology if beginning with basic
research. Nevertheless R&D teams and their donors agreed to the inclusion of their
technologies in these programs because they saw this as an indicator of success
because it meant large numbers were built and sent to the field. However, this
effectively set the design in stone (it is hard to tell a manufacturer you made a
120 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
mistake and to recall hundreds of machines) and made the R&D team defensive to
subsequent criticism. Furthermore, even if the R&D team remained receptive
to suggestions for improvement the SG harvester case study showed that once the
government promotion program began most of the R&D team’s time was taken up
with training regional technicians.
3.2.2. Extension workers
The degree to which extension workers became involved in learning selection
depended on how the extension program was organised. If the program assumed the
TOT model, as was the case with the promotion of the SG harvester and flash dryer,
then they were not expected to make modifications and were not given the resources
or responsibility to do so. Worse still, they felt that their recommendations and
suggestions were ignored. They had very little incentive to be pro-active in solving or
reporting problems. In contrast, when extension workers were able to modify and
promote the SRR dryer on their own initiative, they became a large driving force
behind the dryer’s refinement and rapid adoption.
3.2.3. Manufacturers
Manufacturers modified the technology hardware a great deal, making, for
example, an average of 23 changes to the basic design of the SG harvester they were
copying. These changes came in four categories:
1. changes to the design to make it cheaper or easier to build;
2. changes to the design to improve the performance of the machine;
3. continuing to use a feature of an older design that they had been building prior to
adopting the new design, and which they did not think was worth changing; and
4. mistakes or oversights.
Fig. 3. Net effect of manufacturers on the design of the Mark II SG harvester.
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 121
Manufacturers were behaving as novelty generators when making the first two
types of change. In the third category they were behaving as selectors in deciding not
to adopt certain aspects of the design. The fourth category of modification did
not immediately involve learning because it was a mistake or oversight. Once the
mistake was made, however, feedback sometimes led to a changed perception,
learning and modification. Promulgation occurred when manufacturers copied
changes made by other manufacturers, or detailed in the periodically updated
drawings circulated by the R&D group.
Manufacturers made some very important improvements to the technology. For
example, Morallo Metal Industries reduced the weight of the SG harvester by 25%
making it cheaper and easier to use. In the medium-term manufacturers improved
the fitness of the technology, but when they first started building machines there was
a tendency for them to make more detrimental changes than improvements. Fig. 3,
which shows the net effect of the modifications made by nine SG harvester manu-
factures, indicates that only Morallo Metal Industries would have had a net positive
effect on the design without some ‘industrial extension’ by the research team.
7
An
example of one manufacturer modification that was particularly disastrous was the
reduction of the rotor and forward speed on the 14 units supplied to regional
demonstration centres in the Philippines. The change meant the machines harvested
with high loss and as a result much damage was done to the reputation of the tech-
nology amongst extensionists and co-operative members who attended the demon-
strations. The manufacturer did change back to the original speeds a few months
later but did not recall the demonstration units or even mention the problem. The
R&D team who perhaps should have noticed the problem were too busy training
regional technicians to thoroughly test the units before they went out to the regional
demonstration sites.
Most of the modifications made by manufacturers were to the machine hardware,
i.e. the embodied knowledge. Nevertheless they also made some important innova-
tions to the software knowledge, that is the knowledge not embodied in the machine
itself, but necessary to build the machine cheaply and well, or to use it properly. For
example, some manufacturers developed jigs and fixtures to make fabrication
quicker and easier. An example of an innovation made in the way labour was
organised in building the machine is the adoption of the pakyaw system of hiring
contract labour to avoid employer obligations under Philippine labour laws. The
workers were paid a piece rate according to their output but as a result tended to
rush the work. Without strict quality control the pakyaw system had a large detri-
mental effect on quality. Even though some manufacturers and customers knew this,
adequate quality control was rarely in place.
Government machinery programs were found to be a disincentive to learning
selection, partly because of the tendering process. Manufacturers wishing to build
machines for a government program had to copy a standard design and then submit
a prototype to be checked for compliance with the standard. This accreditation
7
Each modification made by a manufacturer and identified during the survey was rated on a +5 to –5
scale. The sum of these valuations is shown in Fig. 3. See Douthwaite (1999) for methodological details.
122 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
process, plus the fact that manufacturers were not selling directly to the intended
users, hampered learning selection by reducing the manufacturers’ scope to make
changes and by reducing feedback about potential shortcomings, respectively.
3.2.4. Users
In contrast to the manufacturers, users made most of their changes to the tech-
nology software. The organisational innovation already discussed above surround-
ing the use of the reaper is one example. Another was that some SG harvester
owners paid their operators according to area harvested, rather than a daily wage,
and enjoyed a significantly higher seasonal usage rate as a result (see Table 3).
Although owners were making less than one tenth of the number of modifications
made by manufacturers they nevertheless represented an important source of design
improvement through their recommendations for modifications. Over half of the 24
non-trivial recommendations recorded by owners in the survey were incorporated in
later designs by manufacturers, or in the drawings produced by IRRI.
Users also played an important promulgation role. This was particularly clear in
the case of the SRR dryer where on average 68 people visited each unit in the
survey sample. UAF built their extension strategy for the SRR dryer around users
in key villages who would teach and promote the technology to others. In the case
of the recirculating dryer, adopters said that recommendations made by neighbours
and associates were important in persuading them to buy the machine. The SG
harvester and flash dryer case studies showed how adopters who have a negative
experience with a technology could dissuade others from buying or using the
machine.
Government programs also tended to limit user learning selection because by
giving equipment to users at a highly subsidised rate they reduced the incentive to
sort problems out when they occurred. As one manufacturer said, ‘‘farmers don’t
appreciate the machine if it is a dole-out (given for free).’’ (pers. comm. with A.
Atienza, 1997). Co-operatives who acquired the SG harvester virtually free under a
government program used the machine less than private individuals who had paid
the full price for the machine.
Table 3
Level of harvester usage by incentive for SG operators
Harvester usage rate
(ha per season)
Incentive for SG operators
Piece rate Daily wage None Totals
>3 5 1 0 6
1–3 1 4 3 8
<11045
Totals 7 5 7 19
P=0.011 [significant at the 5% confidence limit level according to the Fisher exact test (Everitt, 1992)].
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 123
4. Discussion
4.1. Developing a conceptual model of the innovation process
The case studies show that a very large amount of innovation took place after
release of the technology. The conventional TOT model did not fit reality — scien-
tists and engineers were not able to produce useful technologies, only prototypes
that promised to be useful.
Fig. 4 shows a schematic representation of what really happened, and as such is a
component of a new conceptual model of the innovation process. It shows a suc-
cessful case-study technology beginning its life history as a bright idea that is then
developed during the development phase. At this stage the R&D team are driving
the process and the farmers and manufacturers — the key stakeholders — partici-
pate, if at all, as consultants. This laboratory phase ends when the researchers
develop their ‘best bet’ — a prototype embodiment of what they believe will benefit
the key stakeholders — and take it to the field. During the start-up phase the R&D
team demonstrates the machine or loans it, and then seeks feedback from potential
customers and manufacturers. On the basis of this feedback the R&D team modify
their ‘best bet’. Key stakeholders become increasingly interested in the technology to
the point where one or two manufacturers and a few farmers believe that the
machine makes a ‘plausible promise’ of being of benefit to them. The essential dif-
ference between a ‘best bet’ and a ‘plausible promise’ is, therefore, that the former is
Fig. 4. Evolution of knowledge during the innovation process and how stakeholder participation changed.
124 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
defined by the R&D team and the latter is decided by the key stakeholders. Another
difference is that the ‘plausible promise’ embodies more knowledge, some of which
comes from the key stakeholders. This is shown in Fig. 4 as an increase in the area of
the gear-wheel depiction of the technology, and a change in its shading.
The adaptation phase begins when the key stakeholders show they believe in the
‘plausible promise’ by adopting, and in so doing invest materially in the technology.
This research found that the first farmers to adopt closely resembled Rogers (1995)
definition of an innovator, the first in five categories of adopter types he identified.
The others are early adopters, early majority, late majority and laggards. Rogers
described innovators as venturesome, enjoying the technical challenges posed by
new technologies and actively seeking them out.
It has been suggested that the R&D team needs to remain involved during the
adaptation phase to nurture learning selection by filling knowledge gaps amongst
the key stakeholders, selecting beneficial modifications and promulgating them, and
in carrying out their own learning selection iterations. It is exactly this phase,
though, that the conventional mental maps of the innovation process ignore, or
regard as a black box — somewhere inside which one cannot look and therefore
understand or use as the basis for predictions.
Fig. 5 shows how the learning selection model can explain the workings of the
black box. It shows that the ‘plausible promise’, which embodies some key stake-
holder knowledge but is largely a creation of the R&D team, gains knowledge and
evolves into a fitter ‘widely adoptable technology’ through many learning selection
iterations. The widely adopable technology is depicted as a set of interlocking gear-
wheels to represent the ‘meshing in’ that takes place during the adaptation phase.
Fig. 5. The learning selection model of how a ‘plausible promise’ evolves into a ‘widely adoptable tech-
nology’ during the adaptation phase.
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 125
This ‘meshing in’ process, in which a new technology is shaped by the community in
which it is adopted, and that community is in turn shaped by the technology has
been called ‘the social construction of the technology’ by Bijker (1994). The labour
arrangement innovation negotiated between the owners of the mechanical reaper
and the manual rice labourers is a good example of part of the social construction of a
new technology. As already discussed Long’s (1992) actor-oriented approach can help
understand how learning selection and the social construction process is affected by
cultural, social, political and power dynamics between groups in a community.
5. Implications of the learning selection (LS) model to managing R&D — the LS
approach
The conceptual model of the innovation process based on learning selection (i.e.
the LS model) is a better match to reality than the TOT conceptual model for the
case study innovations presented here because it explicitly recognises key stake-
holder innovation during the early adoption phase, something that the TOT model
does not. The LS model, together with the experience derived from the case studies,
has some clear implications for the design and management of public sector agri-
cultural engineering projects, which is referred to as the Learning Selection
approach. Below are seven principles for setting up and managing a successful co-
development effort.
5.1. Start with a ‘plausible promise’
The key to developing a successful innovation is a successful partnership between
the stakeholder, who holds the technical knowledge about the new technology,
and the key stakeholders, the people who are going to build and use it. This part-
nership needs to be motivated by the belief amongst at least some of the key stake-
holders that the prototype technology makes a plausible promise to benefit them. A
measure of whether the R&D team’s ‘best bet’ makes a ‘plausible promise’ is
whether some key stakeholders adopt it and hence invest time, money and effort.
5.2. Keep the ‘plausible promise’ simple
A plausible promise should be simple, flexible enough to allow revision by the key
stakeholders, and robust enough to work well even when not perfectly optimised. It
should match the key stakeholders’ needs and knowledge levels rather than the
R&D team’s inclination to technical sophistication and elegance.
5.3. Find a product champion
During the early adoption of a new technology many things can go wrong, any
one of which could potentially kill off the technology. It is, therefore, important to
identify an individual or a small group who have sufficient interest and knowledge
126 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
to nurture the innovation. Nurturing means building a common property regime
and promoting learning selection — working with the key stakeholders to identify
and promulgate beneficial modifications, weeding out detrimental ones, plugging
knowledge gaps while at the same time engaging in experiential learning oneself.
Nurturing also means being a good facilitator of adult learning, for which the lit-
erature on people-centred learning can provide a guide (e.g. MacKeracher, 1994). In
practice the product champion is likely to come from the R&D team who developed
the ‘plausible promise’ because they have both the technical expertise and the moti-
vation to do the job.
5.4. Work in a pilot site or sites where the need for the innovation is great
The key stakeholders will be influenced by their environment. Their motivation
levels will be sustained for longer if they live or operate in an environment where the
new technology promises to provide great benefits. In addition, they are more likely
to receive encouraging feedback from members of their community.
5.5. Work with innovative and motivated partners
The outcome of learning is a function of the interaction between the learner and
his or her environment (Lewin, 1951; MacKeracher, 1994). Having chosen the right
environment the next step is to choose innovative-adopters who possess the ability
to make improvements and are drawn to the challenge of doing this. The SG har-
vester and SRR dryer case studies showed that media coverage was very effective at
prompting innovative-adopters to seek out the technology and then buy it, hence
effectively selecting themselves. Enquirers should be charged the market value for
the machine to ensure they are adopting because they believe in the ‘plausible
promise’ and not to get something cheap or for free. Also, people generally value
something more highly if they have paid for it and they will be more committed to
sort out the problems when they emerge. On the other hand the first adopters will
need to know that they are working with the product champion as co-developers,
and as such will not be left taking all the risk.
5.6. Don’t release the innovation too widely too soon
For the innovation to evolve satisfactorily, the changes the stakeholders make to it
need to be beneficial. As those generating the novelties will have gaps in their
knowledge, product champions should restrict the number of co-developers so that
they can work with them effectively. When people show enthusiasm for a prototype
it is very tempting to release it as widely as possible but this entails jumping from the
start-up to the expansion phase and missing out the adaptation one. This should be
resisted. However promising the technology might appear, there are many things
that can and will go wrong. First adopters need to be aware of this and have ready
access to the product champion. Otherwise, their enthusiasm will quickly turn to
frustration and the product champion will end up defending the technology against
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 127
their criticisms when the problems appear. Once the product champion becomes
defensive, he or she will be far less useful at sorting out problems.
5.7. Know when to let go
Product champions need to become personally involved and emotionally attached
to their projects to do their jobs properly. This makes it easy for them to continue
championing a technology when it has become clear to everyone else that the tech-
nology is not going to succeed. Equally, project champions can continue trying to
nurture their innovations long after they entered the expansion phase and proper
market selection has begun. Market selection begins when enough of the key stake-
holders know enough, and are sufficiently motivated, to ensure the technology con-
tinues to evolve and diffuse through their own novelty generation, selection and
promulgation efforts alone.
6. Application of the LS model beyond agricultural engineering
The LS model describes a process by which key stakeholders, helped by research-
ers, experiment with technology and make it their own through adaptation. The
researchers are learning in the process and making their own innovations. This is the
essence of participating technology development (PTD) which van Veldhuizen et al.
(1997) describe as a process by which outside facilitators and rural people interact so
that the target groups have a greater capacity to adapt new technology to their
conditions and the facilitators have a better understanding of traits and character-
istics of local farming systems. Therefore, the LS model may help understand PTD
by focusing attention on the fundamental process by which rural technology change
occurs: interactive experiential learning, or learning selection as it is referred to here.
Loevinsohn (1998) has gone further than this in saying that evolutionary theory
could provide the needed theoretical underpinning to assist understanding and
design of participatory research in general.
As well as helping to provide a fundamental understanding of PTD, learning
selection should also be able to provide a guide to planning and managing a PTD
approach through implementing the seven steps of the LS approach. Furthermore,
learning selection may be able to help in the monitoring and evaluation of PTD
projects by measuring increases in the capacity of the target groups to interact with
technology through identifying and explaining the novelties generated, selection
decisions made and promulgation mechanisms. Combining the LS approach with
Long’s actor-oriented approach can give a full picture of how cultural, social, poli-
tical and power dynamics affected the process.
The LS approach may be relevant beyond agriculture. Von Hippel (1988) in his
influential book the Sources of Innovation found that people working in the USA
industry employed a mental map of the innovation process that is similar to the
TOT model. He writes: ‘‘It has long been assumed that product innovations are
typically developed by product manufacturers. Because this assumption deals with
128 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
the basic matter of who the innovator is, it has inevitably had a major impact on
innovation-related research, on firms’ management of research and development,
and on government innovation policy. However, it now appears that this basic
assumption is often wrong.’’ (von Hippel, 1988, p. 3). Douthwaite (2001) finds evi-
dence of LS-type innovation processes in areas as far apart as the development of
the Linux computer operating system and the Danish wind turbine industry. He
concludes that the LS approach is relevant beyond agriculture.
Further evidence that the LS model and approach may be valid beyond agricultural
engineering is that others, from different backgrounds and disciplines, have developed
similar models. Eric Raymonds (1997) in his paper The Cathedral and the Bazaar
describes what he calls a bazaar innovation model to develop software. In the bazaar
approach a product champion uses a plausible promise to build a co-developer com-
munity to write code, fix ‘bugs’ and build the program into something better. Ray-
mond grounded his bazaar model on the innovation history of the very successful
Linux operating system (Douthwaite, 2001). The LS approach is also similar to the
Center for International Forestry Research’s (CIFOR) adaptive co-management (pers.
comm. with C. Colfer, 2000; Belcher et al., 2000) developed for forestry.
7. Conclusions
In agricultural research it has often been assumed that researchers develop new
technology and farmers either adopt it or not, without significantly adapting it
themselves. A similar assumption in industry, that manufacturers are the sole source
of innovation for new products, has been shown to be incorrect with some types of
technology. This paper has shown that at least with agricultural equipment, user,
manufacturer and researcher innovations are present and essential in machinery that
is widely adopted. Failure by public sector researchers to realise farmers and man-
ufacturers must first adapt new technology to local conditions before widespread
adoption will occur has led to the too-early promotion of some equipment and
wastage of public money. The learning selection (LS) cognitive model of the early
adoption process can help research planners and managers see innovation as an
evolutionary process, and manage it as such.
The LS model sees technology evolving during adoption as a result of stake-
holders making modifications (novelty generation), and then selecting and pro-
mulgating some of these. The model explicitly recognises that during the initial
adoption of a publicly-developed technology, the key stakeholders (those with
most to gain) may not know enough about the new technology for this learning
selection to improve the fitness of the new technology. Also, the technology may not
work well enough for them to be motivated to persevere when unforeseen problems
arise. Researchers need to be active participants in the early adoption process to
nurture new technology until market selection begins to work. In this way early
release of new machinery can provide a mechanism to create a valuable synthesis
between local and research knowledge that should lead to more appropriate tech-
nology, and increase key stakeholder capacity to interact with new technology.
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 129
The LS model is based on the view that new technology can be seen as an increment
in the knowledge of the system into which it is introduced. Although developed based
on data from equipment technology this generic basis makes the LS model applic-
able to any technology where ‘learning by using’ and ‘learning by doing’ are likely to
significantly improve its fitness and adoption prospects.
Acknowledgements
The research presented in this paper was carried out while the first author was
employed at the International Rice Research Institute (IRRI), PO Box 3127, Makati
Central Post Office, 1271 Makati City, Philippines, in collaboration with the Philippine
Rice Research Institute (PhilRice), Mun
˜oz, Nueva Ecija, Philippines and the University
of Agriculture and Forestry (UAF), Thu Duc, Ho Chi Minh City, Vietnam.
The research was funded by IRRI, the Impact Assessment and Evaluation Group
(IAEG) set up by the Consultative Group on International Agricultural Research
(CGIAR) and by the German Government through Deutsche Gesellschaft fu
¨r
Zusammenarbeit GmbH (GTZ). The authors wish to thank the anonymous reviewer
for helping to make this a better paper.
References
Belcher, B., Colfer C., MacDicken, K.G., 2000. Toward INRM: three paths through the forest. Paper
presented at the workshop INRM in the CGIAR: Approaches and Lessons, ICLARM, Penang,
Malaysia, August 21–25.
Biggs, S.D., 1989. A multiple source model of innovation of agricultural research and technology pro-
motion. ODI Agricultural Administration Network Paper: 6. Overseas Development Institute, London.
Bijker, W., 1994. Of Bicycles, Bakelites and Bulbs. Steps Towards a Theory of Sociotechnical Change.
MIT Press, Cambridge, MA.
Burgess, R.G., 1984. In the Field: An Introduction to Field Research. Contemporary Social Research: 8.
Routledge, London.
Chambers, R., Jiggins, J., 1986. Agricultural research for resource poor farmers: A parsimonious para-
digm. Discussion Paper 220. Institute of Development Studies, University of Sussex, Brighton.
Clark, N., 1995. Interactive nature of knowledge systems: some implications for the Third World. Science
and Policy 22, 249–258.
Collinson, M.P., Tollens, E., 1994. The impact of the international research centers: measurement, quan-
tification and interpretation. Issues in Agriculture: 6. Consultative Group on International Agricultural
Research (GIAR) Secretariat, Washington.
Douthwaite, B., 1999. Equipment evolution: case studies of changes in rice postharvest technologies in the
Philippines and Vietnam. PhD thesis, University of Reading, Reading, England. Available at:
www.douthwaite.net.
Douthwaite, B., 2001. Enabling Innovation: A Practical Guide to Understanding and Catalysing Tech-
nological Change. Zed Books Ltd, London.
Douthwaite, B.N. de Haan, N.C., Manyong, V.M., Keatinge, J.D.H., 2000. Blending ‘hard’ and ‘soft’
science: the ‘Follow the Technology’ approach to catalysing and evaluating technology change. Paper
presented at Integreated Natural Resource Management in the CGIAR: approaches and lessons. 21–25
August 2000, Penang, Malaysia. Available at: http://www.inrm.cgiar.org/Workshop2000/abstract/
Douthwaite/fulldouthwaite.htm.
130 B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131
Everitt, B.S., 1992. The Analysis of Contingency Tables. Monographs on Statistics and Applied Prob-
ability 45, 2nd Edition. Chapman and Hall, London.
Freedman, D.A., 1991. Statistical models and shoe leather. In: Marsden, P. (Ed.), Sociological Method-
ology 21. Basil Blackwell Ltd, Oxford.
GAO (General Accounts Office), 1987. Case Study Evaluation. Program Evaluation and Methodology
Division, Transfer Paper 9. GAO, Washington DC.
von Hippel, E., 1988. The Sources of Innovation. Oxford University Press, New York.
Hunt, D., 1987. Beginning with Ourselves: In Practice, Theory and Human Affairs. OISE Press, Toronto.
Kolb, D.A., 1984. Experiential Learning: Experience as the Source of Learning and Development. Pre-
ntice-Hall, Englewood Cliffs NJ.
Lewin, K., 1951. Field Theory in Social Science. Harper and Row, New York.
Loevinhsohn, M.E., 1998. Practising evolution: theory for understanding and evaluating participatory
research. Discussion Paper No. 98-14. International Service for National Agricultural Research
(ISNAR), The Hague.
Leonard, D., 1995. Wellsprings of Knowledge: Building and Sustaining the Sources of Innovation. Har-
vard Business School Press, Cambridge, MA.
Long, N., 1992. From paradigm lost to paradigm regained. The case of actor-oriented sociology of devel-
opment. Chapter 2. In: Long, N., Long, A. (Eds.), Battlefields of Knowledge: The Interlocking of Theory
and Practice in Social Research and Development. Routledge, London and New York, pp. 16–43.
Lyotard, J.F., 1996. The post-modern condition: a report on knowledge. In: Cahoone, L. (Ed.), From
Modernism to Post-Modernism: An Antholog. Blackwell Publishers, Oxford.
MacKeracher, D., 1994. Making Sense of Adult Learning. Culture Concepts Inc, Toronto.
Mokyr, J., 1990. The Lever of Riches: Technological Creativity and Economic Progress. Oxford Uni-
versity Press, Oxford.
Nelson, R.R., 1987. Understanding Technical Change as an Evolutionary Process 8. Professor Dr. F. de
Vries Lectures in Economics, North Holland, Amsterdam.
Nelson, R.R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. The Belknap Press of
Harvard University Press, Cambridge, MA.
ODI (Overseas Development Institute), 1997. Global hunger and food security after the world food
summit. Briefing paper 1997 (1) February. Published on the internet at http://www.oneworld.org/odi/
briefing/1_97.html.
OECD (Organisation for Economic Cooperation and Development), 1996. Shaping the 21st Century: The
contribution of development co-operation. Development Assistance Committee. OECD, Paris.
Peters, T.J., Waterman, R.H., 1982. In Search of excellence: Lessons from America’s best run companies.
Harper & Row, New York.
Raymond, E.S., 1997. The cathedral and the bazaar. Downloaded from http://www.openresouces.com/
documents/cathedral-bazaar/index.htm.
Rogers, E.M., 1995. Diffusion of Innovations. 3rd Edition. Rev. Edition of: Communication of innova-
tions. The Free Press, New York.
Ro
¨ling, N., 1996. Towards an interactive agricultural science. Journal of Agricultural Education and
Extension 4, 35–48. Available at: http://www.agralin.nl/ejae/v2n4-5.html.
Rosenberg, N., 1982. Inside the Black Box: Technology and Economics. Cambridge University Press,
Cambridge.
Ruse, M., 1986. Taking Darwin Seriously. Basil Blackwell, Oxford.
Sechrest, L., Stewart, M., Stickle, T.R., Sidani, S., 1996. Effective and Persuasive Case Studies. Jaguar
Graphics, Tucson, Arizona.
Tidd, J., Bessant, J., Pavitt, K., 1997. Managing Innovation: Integrating Technological, Market and
Organizational Change. Wiley, Chichester, UK.
Van Veldhuizen, L., Waters-Bayer, A., de Zeeuw, H., 1997. Developing Technology with Farmers: A
Trainer’s Guide for Participatory Learning. Zed Books Ltd, London.
World Bank, 1996. The World Bank Participation Sourcebook. Published on the Internet at http://
www.worldbank.org/html/edi/sourcebook/sbhome.htm.
Yin, R.K., 1989. Case Study Research: Design and Methods. Sage Publications Inc, California.
B. Douthwaite et al. / Agricultural Systems 72 (2002) 109–131 131
