JUL -5 L ,',
April June 1984
CIMMYT EASTERN AFRICA ECONOMICS PROGRAMME
INTERNATIONAL MAIZE AND WHEAT IMPROVEMENT CENTRE (CIMMYT)
P 0 Box 25171, Nairobi, Kenya, Telephone 592054/592206
THE NEWSLETTER ARTICLE
ON-FARM EXPERIMENTS IN FARMING SYSTEMS RESEARCH
This paper argues that the conventional on-farm experimental methods employed
by many Farming Systems Research programmes are inappropriate to FSR because
they lack its essential components. It is suggested that on-farm experimental
methods which do have the unique characteristics of FSR should be developed.
To this end, an illustration is provided by an experimental method and by
some analytical techniques. The argument is developed in two parts. The
first part describes conventional on-farm experimentation and the unique
characteristics of Farming Systems Research. This is followed by a demonstra-
tion, with the aid of examples from FSR programmes in Southern Africa, of how
conventional on-farm experimentation does not satisfy the special requirements
of FSR. The second part describes an experimental method based on the super-
imposition of treatments on to existing crops. A range of analytical techniques
is also described. These descriptions show how on-farm experimental methods
can incorporate the unique characteristics of FSR.
On-farm experiments have been a part of national and international research
programmes for many years. These conventional on-farm experiments were
primarily developed to examine 'site effects' on largely known biological
parameters. A typical example would be the FAO's fertilizer programme. Thus
those experiments were largely miniaturized research station experiments.
They were randomized block designs of two or more replicates with four or
more treatments, which were sometimes even factorially structured. Adequate
precision in the data was guaranteed by enormous research input into the
management and implementation of each experiment. This forced researchers
with limited resources into conducting a few complex experiments on a handful
of farms. .The farmers' involvement in these experiments would usually
extend only to the lending of land. Being good biologists, the researchers
only made those measurements necessary to test their hypotheses, so grain
yield per hectare was the principle criterion of evaluation. This method of
on-farm experimentation was successful insofar as it produced information
on the effects of site on biological parameters. Farming Systems Research
however, goes far beyond the determination of site effects on biological
parameters. Not only must this be understood but also be placed in a
perspective with economic and social parameters, more especially as FSR is
immediately concerned with adoption rather than the description of biological
response curves. Herein lies a major difference between FSR and other types
of agricultural research.
Farming Systems Research was first specified by Norman in 1976 at a seminar
in Mali on 'Improved Agricultural Production Systems' (Institut d'Economie
Rurale, 1976) and later more fully developed by Norman (1980) and Collinson
(1982). This development naturally borrowed ideas from current research and
so more research projects that resembled FSR were not called FSR and
conversely some research was called FSR but obviously, by its methods, was not.
In an attempt to reduce the confusion of what was and what was not Farming
Systems Research, Norman outlined seven unique characteristics to FSR which
he called 'common elements'
* Consultant: 80 Avenue Rosemere, Ottawa, Ontario. '"I 1 r-
- 13 -
The intention was to maintain the unique identity of FSR even though research
programmes may have different names. Of these seven 'common elements', three
are pertinent to the processes of on-farm experiments. They are multidisci-
plinary involvement, farmer participation and evaluation relevant to the
So far, the major developments in Farming Systems Research have been in the
fields of socio-economic investigations. This is hardly surprising as FSR
was developed largely by economists. Development in the technical component
has limited itself to merely borrowing wholesale from the existing methods
of conventional on-farm experimentation. The following set of examples shows
how these conventional on-farm experimental methods neglect the unique
characteristics or 'common elements' of Farming Systems Research.
Although the examples are taken from arable enterprises of FSR programmes
in Zambia and Zimbabwe, the experiences are not uncommon to those of other
programmes (Lightfoot,1982 and 1982a). The complexity of field layout and
treatment structure in conventional on-farm experiments makes these experiments
hard for farmers to understand, mainly because they are so foreign to their
experience. Not only does such complexity inhibit farmers' understanding
but also farmers' participation in implementation. Thus researchers are bound
to plant the crops and carry out the treatments. Here, the logistical
problems of communication and transport between the centre and experimental
sites virtually ensures that experiments are rarely planted at the optimum
time. Thus in Zambia and Zimbabwe, it was not uncommon to see farmer planted
crops in better health than the research crops, purely because they were
planted at the right time. Such poor growth of research crops can easily
confound the results of conventional experiments. For example, nitrogen
top-dressing of maize, normally providing a large yield response, had no
effect when applied to an unhealthy crop. As a result, the analysis across
all sites proved inconclusive. The inflexibility of these experiments can
reduce experimental precision such that conclusive results are difficult
to obtain. To make matters worse, farmers are unlikely and certainly would
be recommended not to top-dress a poor maize crop. Another example of
confounding is offered in weeding experiments, where the weed burden is
negligible at some sites and therefore the benefits of weeding become unclear.
The likelihood of inconclusive results is further increased by the considerable
risk of experimental failure and the unavoidably small number of farms used.
Experiments can fail for a variety of reasons, one of the more common being
crop damage by livestock. On one occasion in Zimbabwe, the farmer was responsible
for the failure of a tillage experiment. This farmer decided to weed the plots
at crop emergence thereby losing information on the all important tillage
weeding interaction. The difficulty farmers have in understanding complex
experiments is further illustrated in the following example. Another farmer
running the same tillage trial found that there was little difference in growth
of single and double ploughed crops. To the researcher, this meant that the
return to draught power had been increased but the farmer did not appear to
appreciate this. Such misunderstanding could have arisen if draught power was
not a limiting resource but it was more likely that the farmer did not
understand the experimental objective. Either way, the researcher would have
benefitted from the involvement of the farmer.
Those examples show that, not only are conventional experiments difficult
to implement but also that they offer little chance for farmers' participation,
which is one of the important features of FSR. Obviously, using farmers' land
- 14 -
involves a level of farmer participation and it is likely that some dialogue
between technologist and farmer occurs during the implementation of these
experiments; however, this dialogue is limited by the training of the
technologist. In order to attain a useful depth of dialogue, some input by
appropriately trained social scientists is necessary.
The importance of co-operation between the disciplines is further illustrated
by the inadequacies of conventional evaluation of experimental results.
Conventional evaluation almost exclusively employs mean grain yield per
hectare as the sole criterion for the assessment of technologies. There are
several deficiencies associated with slavish use of this measure. Firstly,
farmers in variable environments are more likely to be interested in the
range and stability of performance. In addition, from the statistical
viewpoint, means of variable data are of dubious value. Secondly, cash or
nutritional value may be more relevant returns than grain yield. Finally,
farmers may seek to maximize returns to more limiting resources such as
labour and capital, than to the land. However, selecting a criterion or a
combination of criteria most appropriate to the farmer requires knowledge
outside the scope of the technologist. Hence the importance of co-operation
between technologists and social scientists, whose job it is to provide that
In conclusion, conventional on-farm experimental methods and analyses are
difficult to implement and the results difficult to interpret. They-also
offer little scope for meaningful farmers' participation and co-operation
between the disciplines involved. This paper proposes that new experimental
methods must be developed to overcome the difficulties described and
incorporate the unique features of FSR. The following on-farm experimental
method and analyses are presented to illustrate more appropriate methodologies
for Farming Systems Research.
The on-farm experimental method entails the super-imposition of treatments
or technologies on to the appropriate existing crop or soil condition. For
example, to test the benefit of nitrogen top-dressing of maize, an area of
healthy maize (i.e. a crop that farmers would be advised to top-dress) is
bisected and one half receives nitrogen. Objectively in selecting which half
gets the nitrogen must be ensured by random allocation of treatments. Another
example of a technology that lends itself to super-imposition is weeding.
At the time of first weeding, an area with the appropriate weed burden
(especially important if parasitic weeds are involved) is selected and divided
into control and treatment plots. More complex treatment structures than
presence or absence can be used only if the farmer understands them.
Replication is achieved between farms and not within farms (for statistical
reasons never by a mixture of the two) which will require an increase in the
number of participating farms. For reasons of statistical precision (adequate
error degrees of freedom for testing variance ratios) at least twelve farms
are required in two treatment experiments but seven in three treatment
experiments. Although it is hard to imagine, difficulty in locating enough
farms with the right crop conditions would present problems here. The
simplicity of this experimental method permits farmer implementation. Ideally,
the inputs of seed, fertilizer or whatever are given to the farmers and they
then implement the treatments. The researchers' involvement extends to the
selection of the area, instruction on the method of treatment implementation
and some checking on the accuracy with which the experiments are conducted.
Of course, all measurement and data collection are the responsibility of the
- 15 -
The shift in level of input from the researcher to the farmer in the
super-imposed method increases the scope for farmer participation and
gives time for researchers to contact many more farms. This can only
result in more rigorous and more precise testing of technologies. Farmer
implementation in the appropriate conditions on many farms can improve
the timeliness of operations, reduce the degree of variation and increase
the number of data points for analysis. Farmers' participation, an
important feature of FSR, could be deepened by soliciting the farmers'
reactions to the technologies. The depth of such questioning is likely
to be greater when conducted by researchers trained specifically in this
area, again a point of contact between the disciplines.
The importance of co-operation between technical and social science
disciplines is further emphasized in the analysis of experimental results.
In the variable environmental conditions that prevail throughout the semi-
arid tropics, farmers are more likely to be interested in the stability
of technologies rather than their average performance. While, stability
can be defined and measured in many ways, only three methods are presented
here. Firstly, the likely range pf outcomes can be estimated by inter-
quartile ranges and confidence intervals. In other words, the situation
in which technologies perform poorly and those where they perform well are
measured and defined. In this way, crop failures that were usually put
to one side become useful results. The variation in the performance of
technologies across different environments, (i.e. on different farms) can
be estimated by the regression of outcomes against an appropriate index of
the environment. Lastly, the risk of crop failure to farmers who adopt new
technologies can be estimated by calculating the probability of failing to
achieve specified disaster outcomes. The relevance of such estimations
to farmers can only be ascertained by consultation between social scientists
and the farmer. The same applies to the selection of returns, (grain, cash,
calories, etc.) and to resources, (land, labour, capital, etc.). Such
consultation increases the chance of technology assessment being relevant
to the farmer. Thus, the remaining important feature or common element of
FSR, technology assessment relevant to the farmer, is addressed by the
analytical techniques described.
In conclusion, the super-imposed experimental method and the analytical
techniques described incorporate the unique features of Farming Systems
Research. Participation of the farmer is ensured by farmers' implementation
of the treatments and by consultation for the evaluation of technologies.
In such consultation, the evaluation of technologies is made more relevant
to the farmer. Lastly, co-operation between technical and social science
disciplines is engendered by their interdependence for specific kinds of
information. In addition, this experimental method makes the implementation
of on-farm experiments easier and can increase the precision and rigor with
which technologies are tested.
Obviously, the experimental method described here has limited application
since not all technologies can be tested by super-imposition. Indeed, if
this method is restricted to simple presence or absence studies, there is a
distinct danger of slipping into purely demonstration work. For such studies
to retain their title of research, the results must add to the existing body
of knowledge. Here, the facility of conducting many experiments could greatly
increase the supply of technology options for extension.
- 16 -
More experiments and thus more participating farmers will also mean more
direct beneficiaries of the research programme. The analytical techniques
described make greater use of the data produced, but again their applications
are limited. The details of analysis are always peculiar to each experiment.
The intention here is to show that on-farm experimental methods which include
the unique features of FSR can be developed. Technologies should not consider
themselves, or behave as if they were, straight jacketed by conventional
on-farm experimental methods. FSR does have unique characteristics, thus
new experimental methods are required to reflect this. By paying greater
attention to farmer participation and interdisciplinary co-operation,
researchers will stand a greater chance of developing on-farm experimental
methods that are true to the principles and spirit of Farming Systems Research.
Collinson, M.P. 1982. Farming Systems Research in Eastern Africa: The
experience of CIMMYT and Some National Agricultural Research Services,
1978-81. MSU International Development Paper No. 3. East Lansing, Michigan:
Michigan State University.
Institute d'Economie Rurale. 1976. Rapport de synthese sur les systems de
culture et d'elevage dans le context de Mali. Bamako: Institut d'Economie
Lightfoot, C.W.F. 1982a. Some observations for the On-Farm Experiments.
Report to the Dept. of Specialist Services, Govt. of Zimbabwe, Harare, Zimbabwe.
1982. Some Guidelines for the Adaptive Research Planning
Team On-Farm Trials. Report to the Ministry of Agriculture and Water
Development, Govt. of Zambia, Lusaka, Zambia.
Norman, D.W. 1980. The Farming Systems Approach: Relevancy for the Small
Farmer. MSU Rural Development Paper No. 5. East Lansing, Michigan: Michigan