Understanding, Quantification, and Modelling
in Farming Systems Research:
Results of a Simulation Study in Northern Nigeria
Eric W. Crawford
Ag. Econ. Staff Paper No. 80-19
Department of Agricultural Economics
Michigan State University
March 13, 1980
MSU IS AN AFFIRMATIVE ACTION/EQUAL OPPORTUNITY INSTITUTION
Understanding, Quantification, and Modelling
in Farming Systems Research:
Results of a Simulation Study in Northern Nigeria
Eric W. Crawford
The paper discusses understanding versus quantification in farming systems
research. The strengths and weaknesses of a large-scale and rapid exploratory
surveys are compared. The value of a comprehensive rather than partial approach
is illustrated with results from a recent simulation study of farm households in
Eric W. Crawford is an Assistant Professor of Agricultural Economics at
Michigan State University. He holds a B.A. from Pomona College (1967), and
an M.P.A.from Princeton University (1969). He received his Ph.D. from
Cornell University in 1980. His field is international agriculture and farm
Understanding, Quantification, and Modelling in Farming Systems Research:
Results of a Simulation Study in Northern Nigeria*
There has been a growing tendency to recommend that research on small farm
development in the Third World should be undertaken within a farming systems con-
text (Norman, 1978, 1980; CGIAR/TAC). One of the key elements of the farming
systems research (FSR) approach is its holistic focus. FSR examines the full
range of household activities--including livestock production, off-farm enterprises,
and domestic tasks--and the interactions among them, rather than studying only
cropping activities within the farm operation (Gilbert, Norman, and Winch).
It is clear that subsistence or semi-commercialized farm households are complex
systems whose activities, goals, and decision-making processes must be thoroughly
understood if policy interventions or technology development are to benefit them
(Vincent). However, to what extent comprehensive data collection and analysis is
needed in order to achieve this understanding is a matter of current debate. Do
we need to analyze whole farm models constructed from detailed field data, or can
we achieve sufficient understanding by focussing more narrowly on key constraints
and activities identified through rapid exploratory surveys carried out directly
with farmers?1 Collinson stresses the need for "understanding rather than
quantifying the farming system" (p. 4), and asks whether whole farm modelling
and the quantification it requires can do more than marginally improve under-
*The author would like to acknowledge the helpful comments of Carl K. Eicher on
1Methods developed at ICTA (Hildebrand) and CIMMYT (Collinson; CIMMYT) are
two examples of the latter approach.
This paper presents some evidence on the strengths and weaknesses of modelling
and large-scale farm management surveys with respect to understanding farm house-
hold behavior, based on a recent simulation study of farm households in northern
Nigeria (Crawford, 1980a,b). Without citing the full results of the study, it is
possible to suggest circumstances in which quantitative modelling has a useful
role to play in conjunction with other FSR activities taking place on the experiment
station or in farmers' fields.
The paper first describes the objectives and setting of the study and the
main components of the model, in part to illustrate how the farming systems approach
was incorporated. Second, the analysis undertaken with the model is outlined and
selected results discussed to illustrate the value of a comprehensive rather than
partial approach. Third, implications of the study for rapid versus detailed survey
methodologies are considered. Conclusions are then presented.
Objectives and Setting of the Study
The objective of the study was to examine the effect of variable crop yields,
physical resource limitations, and family size on the growth of income and con-
sumption among traditional small farm households in the dryland savanna zone
of northern Nigeria. Particular emphasis was given to seasonal cash and subsistence
food requirements, and to the interaction of these constraints with weather-induced
variability in crop yields and returns over time.
The study was based on data obtained from a comprehensive farm survey conducted
during 1974/75, as described in Matlon. Households in the area produce both for
home consumption and for the market, using traditional hoe technology. Principal
crops (grown in mixtures) are millet, sorghum, groundnuts, cowpeas, onions, sugar
cane, and vegetables. Off-farm activities are an important source of income.
The study develops a simulation model which covers a series of farm house-
hold situations common to the area. The principal components of the model are
shown in Figure 1:
1. A stochastic production/marketing model which maximizes net farm income
subject to resource availability and basic consumption needs, solved in two stages
each year using multiperiod linear programming (MLP).
2. Two user-written simulation routines: (a) SIMI, which generates stochastic
crop yields, investment returns, special consumption expenditures, and derived crop
prices and harvest-y4eUt requirements; and (b) SIM2, which allocates surplus income
among consumption, savings, and investment, and uses the final solution for each
year to update the prices, investment activity levels, and resource availabilities
incorporated in the production/marketing model for the following year.
As shown in Figure 1, the two stages of the MLP component are referred to as
"Model.A" and "Model.B." Model.A is divided into three time periods covering the
main phases of the farm operation: land preparation, planting, and weeding (Period
1); initial harvest and marketing (Period 2); and final harvest, marketing, and
dry season activities (Period 3). The basic version of Model.B is simply a subset
of the rows and columns of Model.A. Following the ex ante farm plan generated by
Model.A, the average yield, price, and labor coefficients in Model.B are replaced by
randomly determined "actual" values. Given the cropping plan and resources available
at the end of Period 1 of Model.A, and the "actual" coefficients, Model.B is then
solved to give a revised output and income position. This serves as the input for
the year-end consumption, savings, and investment decisions. The results of these
decisions are used to revise the prices and resource levels in Model.A for the
following year. Model.A is then re-solved, and the process continues.
A major emphasis of the study--the behavior of the farm household over time--
provides the principal justification for employing the simulation approach. The
FLOW CHART OF MODEL COMPONENTS AND DECISION POINTS.
Model. A: Make
Decision for Year t
L Period I Period 2 -Period 3
SIM I : (a) Calculate Simulated
Yields, Prices,and Labor; Random
(b) Revise Model.B to Reflect Process
Simulated Actual Conditions.
Model. B: Revise Ihe Harvest and
Marketing Plan for Periods 2
Period 2 Period 3
. .._ __-
SIM 2: (c) Revise Year-end
(d) Revise Prices and Resources
for Model.A in Next Year.
SIM 2" (a)Compute Year-end
Income and Inventory;
(b) Make Additional
Consumption, Savings and
aim was to go beyond a static, one-year analysis of the farm system and its constraints.
In a sense, the nature of the farming system and its environment demanded a systems
simulation methodology. Farm and household activities interact with off-farm occupa-
tions. Crop yields, investment returns, and major consumption expenditures fluctu-
ate widely from year to year. Under these circumstances, simple linear projections
over time, based on alternate initial assumptions, would not provide the depth or
accuracy of analysis that a stochastic, recursive model allows.
The flexibility of form which the simulation approach allows makes it possible
to incorporate many of the complexities of real farm household systems (Johnson
and Rausser), such as a sequential decision process involving multiple goals, as
well as the interaction between family size and resource characteristics (Vincent).
A theoretical model of this degree of complexity would be impossible to evaluate
analytically, especially in a dynamic context (Naylor).
Analysis and Results
The analysis carried out during the study includes: (1) calculation of in-
put-output relationships for the common intercropped mixtures used as production
enterprises in Model.A; (2) calculation of one-year farm plans for different
household types, using only Model.A; (3) a set of deterministic runs of the full
simulation model (Model.A, Model.B, SIM1, and SIM2 linked together), designed
to identify key constraints and to investigate the effect of specified experi-
mental factors such as household type, and predetermined sequences of crop
yields, investment returns, and major family consumption expenditures on house-
hold income and consumption; (4) a set of 20 replications of the full stochastic
model for each of two household types, with the values of stochastic variables
randomly obtained from their respective probability distributions. Experiments
conducted under (3) and (4) are referred to as deterministicc" and "stochastic,"
Several findings of the study have implications for the issue of under-
standing versus quantification, and the relative merits of detailed surveys
and modelling versus more rapid surveys and partial techniques. First,
Matlon's survey was very valuable in documenting the great diversity which
characterized his area despite its ecological homogeneity and traditional level
of technology. It was therefore possible as well as desirable to incorporate
several household types and a range of crop mixture types and production tech-
niques in the model, rather than utilizing average categories. The use of
average relationships would have given misleading results, a point noted by
Palmer-Jones, and Upton and Casey. For example, the average labor profile
for fields growing a given mixture will be smoother than the labor profile of
a particular field. By underestimating peak labor demands, a model using
average coefficients will not represent the labor constraint realistically.
A second, related, result suggests that models based on average yields
and prices may overestimate the potential for income growth, at least for
the poorer household types. It was found that average levels of accumulated net
capital and total consumption per consumer, over the six-year time period of
the experiments, were lower in the stochastic runs than in base runs where all
stochastic yield and returns variables were fixed at average levels. The gap
between stochastic and base run values was statistically significant at the
.01 and .05 levels, depending on household type.
Two aspects of the model--both of which reflect observed complexities--explain
1. There is what might be called an efficiency loss which occurs in Model.B
as a result of the sequential decision process. The cropping pattern which is
formulated ex ante in Model.A cannot be changed in Model.B, although labor use,
marketing, borrowing, and resource transfer plans can be altered. A cropping
plan which is optimal for the average yields and prices of Model.A is unlikely to
be optimal given the particular set of simulated "actual" yields and prices in-
corporated in Model.B. The resulting ex post sub-optimality can reduce net farm
2. In the model, households can use cash reserves to invest in soil fer-
tility improvements and non-agricultural occupations off the farm, e.g., trading.
Threshold levels of working capital must be accumulated before these activities
can be undertaken. Studies by Matlon and Hill indicate that the more profitable
off-farm activities also have higher working capital requirements. Accordingly,
if working capital can be accumulated steadily, the household will be able to
engage in activities of increasing profitability, accelerating the growth of income
over time. On the other hand, if surplus capital is reduced or wiped out by
stochastic crop failures, investment losses, or special consumption obligations,
the household must operate in the following year with fewer if any growth-inducing
activities. Although crop yields were assumed to be symmetrically distributed,
their effect on income growth was not. In a good year, not all additional income
is devoted to on-farm or off-farm investment, since larger amounts are consumed
or saved before investments are made.
A third result worth noting in this connection is that the importance of the
cash constraint--emphasized by Matlon and Hill--was clearly revealed by the full
multi-year simulation model, but not when Model.A was used alone. Although there
was scope in the MLP model (Model.A) for the cash reserve to influence incomes
and cropping patterns, it was not nearly as powerful a constraint as land and
labor. Since cash inputs for crop production were low, cash reserves did not
21In reality, farmers may employ a variety of methods for adjusting their
farm plans during the season as additional information about growing conditions
is gained. Nonetheless, some such losses are inevitable as events happen upon
the farmer too quickly for effective adjustment to be made.
constrain the choice of production enterprises. In the full model, however,
cash reserves are the crucial factor determining the extent to which the house-
hold can undertake on-farm improvements and off-farm investments which boost
income over time. This helps illustrate the value of specifying a full range
of household activities, and of conducting the analysis within a dynamic framework.
A fourth result was that the household with the poorest resource endowment
(i.e., the lowest land/man ratio) not only experienced slower income growth and
incurred emergency debts more often, but also experienced greater variability
of income and consumption levels over time, by comparison to the better endowed
household. This occurred because the poorly endowed household chose labor-inten-
sive crop mixtures emphasizing grain rather than groundnuts, an optimal choice
given relative prices and the household's abundant family labor supply and cor-
respondingly large subsistence grain consumption needs. However, grain in the
model had a higher yield variance than groundnuts, which predominated in the
better endowed household's cropping plan. This adds weight to other findings
(Sektheera) that differences in household types, expressed partly in terms of
the land/man ratio, should be incorporated in the analysis.
Implications for Survey Methodology
Matlon's survey of three villages covered 35 households with twice-weekly
interviews, and another 105 households in monthly interviews. The model de-
scribed in this paper was constructed almost entirely from the 35-household
sample, which provided an excellent picture of the range of farming conditions.
The multiperiod linear programming model of Sektheera was based on a sample of
30 households in one village, which was judged adequate in that case. Rapid
exploratory surveys can also reveal the diversity in an area, but would be un-
likely to provide the detail needed for modelling of this kind. However, it
does not appear necessary to mount detailed surveys covering hundreds of house-
holds, as done by World Bank projects in northern Nigeria. If the aim is to
identify only the principal farm enterprise and the critical resource constraint,
the rapid survey approach and partial budgeting can be utilized. On the other
hand, a thorough evaluation of the impact of new technology or policies is
likely to require some quantification of the full range of household activities.
As Collison (p. 11) suggests, the trick is to identify key areas of the farm
system where quantification is warranted.
An advantage of devoting survey resources to open-ended interviews by re-
searchers rather than to administration of detailed questionnaires by enumerators,
is that interviews can provide information on the pattern of events over time
(costly to obtain via large-scale surveys), and on the crucial but hard-to-
quantify sociocultural factors and of their impact on the farm system. Such informa-
tion provides an understanding that is crucial to any analysis, including the
design and use of models. For example, are there intangible barriers to invest-
ment in profitable off-farm enterprises? The financial ones may be obvious, but
social and political status are also probably important. Or, what sociocultural
factors affect how farm households allocate surplus income to consumption, savings,
and investment? These are important questions when evaluating the survivability
or income growth prospects of different household types. However, it is difficult
to deduce the answers by relying only on numerical survey results.
In addition to missing important sociocultural information, large-scale
surveys do not provide a sufficiently accurate picture of sensitive variables
such as income and assets which are important in economic analysis (Palmer-Jones).
Moreover, even the most comprehensive farm management surveys may lack information on
field-level soil, rainfall, and pest conditions, or on plant density. Physical
and agronomic information of this kind is essential for the analysis of production
relationships. Such information was not collected by Matlon, largely because of
the inherent difficulty of obtaining such measurements without the participation of
technical scientists. Accordingly, it proved very difficult to fully analyze
the sources of observed variability--in terms of labor inputs and crop output
per hectare--among the different fields growing a given crop mixture type.
This paper has attempted to illustrate some of the benefits of quantification
and modelling in understanding farm household behavior. Unless the objectives of
farming systems research are limited to understanding the principal farming activi-
ties and resource constraints--as exemplified by CIMMYT's work in East Africa--it
may be desirable to expand the scope of the study to encompass several sub-sectors
of the rural household and the range of variables which influence them. The ex-
tent to which information gained in rapid exploratory surveys can generate the
range of data involved in more complex systems is something which further ex-
perience may answer.
It is clear, however, that large-scale farm management surveys cannot be
expected to provide sufficient data for complete quantification of the farm
system, even if this were desired. Where time and resources are limited,
it may be desirable to collect less detailed economic information and broaden the
survey to generate information on sociocultural and agroclimatic variables.
Some of this information is best collected in open-ended interviews, where multi-
disciplinary teams employing rapid exploratory survey methods have the compara-
tive advantage. This approach may provide a more complete, profound under-
standing of the farm household than can be obtained from assembling the myriad
pieces of data from a large-scale survey.
There are circumstances in which simulation modelling may be a worthwhile
adjunct to experiment station or field-based FSR, despite the recognized data,
expertise, and computer requirements of the modelling approach. For example,
testing prototype technology via on-farm trials is an essential process, but a
slow, sometimes expensive, and expertise-demanding method itself (Collinson, p. 11).
Where the manpower and computational resources are available (e.g., at an inter-
national research center, or middle-income or developed country institution),
simulation studies might usefully proceed side by side with on-farm trials, facili-
tating a timely assessment of the sensitivity of new technology under different
assumptions regarding household type and environmental variability. The analysis
can easily be extended over a five to ten year period, guarding against the
possibility that the modified farm system might be subject to unacceptable cumula-
tive, seasonal, or bad-year losses. Such losses are unlikely to be picked up in
a single-year analysis which relies on average returns to evaluate profitability.
The improved evaluation of the dependability of returns which the simulation ap-
proach can afford would be particularly helpful where there are many marginal
farmers, and/or where weather conditions are highly variable.
CGIAR/TAC (Consultative Group on International Agricultural Research, Technical
Advisory Committee). "Farming Systems Research at the International
Agricultural Research Centers." Mimeo. September, 1978.
CIMMYT Economics Program. Planning Technologies Appropriate to Farmers: Concepts
and Procedures. Mexico: Centro Internacional de Majoramiento de Maiz y
Collinson, Michael. "Micro-Level Accomplishment and Challenges for the Less Developed
World." Paper presented at the XVII Conference of the International Association
of Agricultural Economics, Banff, Canada, September 3-12, 1979.
Crawford, Eric W. "A Programming-Simulation Study of Constraints Affecting the Long-
Run Income-Earning Ability of Traditional Dryland Farming Systems in Northern
Nigeria." Ph.D. thesis, Cornell University, 1980a.
___ "A Stochastic Farm System Simulation Model, with Emphasis on Experimentation."
Mimeo. Department of Agricultural Economics, Michigan State University, January,
Gilbert, E. H., D. W. Norman, and Fred Winch. "Farming Systems Research for
Agricultural Development: A Review of the State of the Arts in Low Income
Countries." Forthcoming Michigan State University Rural Development Paper.
Hildebrand, Peter E. "Generating Small Farm Technology: An Integrated Multi-
disciplinary System." Paper given at the 12th West Indian Agricultural
Economics Conference, Antigua, April 24-30, 1977.
Hill, Polly. Rural Hausa--A Village and a Setting. London: Cambridge University
Johnson, S. R., and Gordon C. Rausser. "Systems Analysis and Simulation: A
Survey of Applications in Agricultural and Resource Economics." A Survey
of Agricultural Economics Literature, Vol. 2, ed. G. G. Judge et al.,
pp. 157-304. Minneapolis: University of Minnesota Press, 1977.
Matlon, Peter J. "The Size Distribution, Structure, and Determinants of Personal
Income among Farmers in the North of Nigeria." Ph.D. thesis, Cornell University,
Naylor, Thomas H. Computer Simulation Experiments with Models of Economic Systems.
New York: John Wiley and Sons, 1971.
Norman, D. W. "Farming Systems Research to Improve the Livelihood of Small
Farmers." American Journal of Agricultural Economics. 60(1978): 813-818.
_ "The Farming Systems Approach: Relevancy for the Small Farmer." Michigan
State University Rural Development Paper No. 5, 1980.
Palmer-Jones, R. W. "Field Research Methods Used in Farm Management Studies."
Mimeo. Department of Agricultural Economics and Rural Sociology, Institute
for Agricultural Research, Zaria, Nigeria, 1977.
Sektheera, Rapeepun. "The Allocation of Family Resources to Farm and Non-Farm
Activities in a Village in Northern Thailand." Ph.D. thesis, Michigan
State University, 1979.
Upton, M., and H. Casey. "Risk and Some Pitfalls in the Use of Averages in Farm
Planning." Journal of Agricultural Economics. 25(1974): 147-152.
Vincent, W. H. "Farmer's Decision-Making Behavior with Regard to Cropping Systems
Research." Proceedings, Symposium on Cropping Systems Research and Development
for the Asian Rice Farmer, pp. 127-137. International Rice Research
Institute, Philippines, 1977.