Illlll'llllllllllllIlllllllllllilllll'llll

3 1293 01025 3981

This is to certify that the
dissertation entitled

Agroecosystem Analysis in a Semi-arid, Smallscale
Farming System of Zimbabwe: A Systems Approach

presented by

Timothy J. Lynam

has been accepted towards fulfillment
of the requirements for

 

Dnrrnr' 0. degree in Ebj Igsgpb¥

M a jdr professor

%J.M4/z

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

 

 

 

 

 

 

 

 

 

- LIBRARY
Michigan State
University

 

 

PLACE IN RETURN BOX to roman this checkout from your record.
TO AVOID FINES rotum on or before date duo.

 

DATE DUE

DATE DUE DATE DUE

 

$53 c*£§,f‘t.-g';;;:r '

 

 

[Rat u l. . ..
f

 

 

mg;
M3

5;)

 

 

NT. 2 3 2008

 

 

 

 

 

 

 

 

 

 

 

 

MSU ION! ”“

AGROECOSYSTEM ANALYSIS IN A SEMI-ARID, SMALLSCALE
FARMING SYSTEM OF ZIMBABWE: A SYSTEMS APPROACH

BY

Timothy Jen Philip Lynam

A DISSERTATION

Submitted to
Michigan State University
In partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Crop and Soil Sciences

1 993

ABSTRACT

Effective agroecosystem analysis is made difficult by the lack of suitable
philosophical and methodological approaches and the difficulty of applying
performance criteria, such as sustainability, to agroecosystem behavior. A research
methodology for the analysis of agroecosystem behavior in data scarce
environments was described. The approach was used to examine an
agroecosystem in the mid-Zambezi Valley.

A questionnaire survey and RRA techniques were used to develop a general
understanding of the needs, resources and constraints households faced in
satisfying their needs. Thereafter an extended analysis was conducted with the
objective of developing a computer simulation model of the agroecosystem. Five
male villagers were elected and three women volunteered to act as village
representatives (VRs) in the analysis. Data collection methods were developed and
then used with the VHS to identify. and weight, by relative importance, major
household needs and the production enterprises used to satisfy those needs as
well as all inputs to and outputs from each production enterprise.

Household needs and yield probability density functions derived from the
VHS were used to develop a computer simulation model of the agroecosystem. Up

to 300 households were randomly placed on a raster based GIS image of the

Masoka agroecosystem. Each cell of the landscape represented one acre.
Households were allocated cells (fields) around their house sites and on soils
adjacent to the Angwa River. The model simulated the productive activities of up
to 300 households, updating variables in each cell and for each household once
a season. Households could be allocated to one of four production strategies
which determined the crops as well as the proportion of household land that was
planted to each crop each year. Households could employ labor from other
households to make up for deficits failing which their yields were reduced. A partial
budget format was used to estimate returns to land, labor and initial investment.
Soil erosion was modeled using the SLEMSA erosion model.

The effects of changing important model inputs and parameters on model
response variables were examined. Factors that had notable effects on the
proportion of deficit households, yields and returns to land. labor and initial
investment were rainfall, the land area available and the cash needs required by

households.

Copyright by
Timothy Jan Philip Lynam

1994

To Limited, who never got to see a rhino
and

to the spirit of CAMPFIFiE.

ACKNOWLEDGEMENTS

In a study spanning two continents and of considerable breadth l have had
to rely on the goodwill and assistance of many people and organizations. I am
especially grateful to the WWF Multispecies Project, for providing me with the
institutional and financial support for this research and to the Project Leader. Dr.
D.H.M Cumming, whose enthusiasm, mentorship and unfailing support were
crucial to every aspect of my research. I would like to thank the Rotary Foundation
for the generous scholarship which made it possible for me to attend MSU. I would
like to thank Dr. J.T. Ritchie, my advisor, who provided the impetus that brought
me to MSU, the institutional support while I was at MSU and for leaving me to go
my own way. My committee members Dr. Campbell, Dr. Crawford, Dr. Manetsch,
Dr. Robertson, Dr. Paustian proved invaluable as teachers, advisors and friends.
I am very grateful to them all. Like so many Zimbabwean colleagues I am indebted
to Peter Frost. His remarkable stream of ideas, criticisms, support and friendship
have enriched my work and my abilities. Working with Gift Chisunga, George
Kanoderuka, Mai Kajengo, Mai Tekenyeka, Mai Kowe, Albert Fero. Saraoga Mbisvi
and lnocent Fakero was both a delight and a pleasure. Without their enthusiasm

and dedication I could not have achieved a fraction of what we did. I am grateful

to Henry Elwell for his invaluable help with erosion modeling. for doing the
erodibility tests for me and for the many ideas and thoughts he has so generously
provided. All errors and omissions in the erosion model used here are however,
my responsibility! Sarah Funkhouser made a great difference to my life and to my
work, I am grateful. Bill Verboem and Katherine Verbeek were invaluable in helping
with soil mapping and classifications. I am particularly grateful to Stanford Mushiri
of the Chemistry and Soil Research Institute for the tremendous assistance he
gave in having my soil samples analyzed. Bill Derman, Stuart Gage. Sharlene
Rhines, Karen Potter-Witter, Brian Baer and Ivan Bond provided invaluable support
and help over the years. I would like to thank the staff at WWF MAPS Project
Office Harare, for all they have done and continue to do. To all those I have

inadvertently missed, thank you.

TABLE OF CONTENTS

LIST OF TABLES ........................................... vii
LIST OF FIGURES ............................................ x
ABBREVIATIONS ........................................... xiv
Chapter 1. ................................................ 1
Background and justification ................................. 1
INTRODUCTION ...................................... 1
BIBLIOGRAPHY ...................................... 10
Chapter 2. ................................................ 12
Background data on the Masoka agroecosystem ................. 12
INTRODUCTION ...................................... 12

Study site description .............................. 13

Summary ....................................... 26
BIBLIOGRAPHY ........................................... 28
Chapter 3. ................................................ 30
Towards a general understanding of the Masoka agroecosystem ..... 30
INTRODUCTION ...................................... 30
METHODS ........................................... 35
RESULTS AND DISCUSSION ............................ 40
Household needs for food and cash crops .............. 41

Production ...................................... 43

Income generating potential ......................... 47

Cropland areas required by households to meet needs ..... 50

Implications ..................................... 52
CONCLUSIONS ....................................... 54
BIBLIOGRAPHY ...................................... 57
Chapter 4. ................................................ 61

iv

Needs analysis, system Identification and problem formulation In the

Masoka agroecosystem ................................ 61
INTRODUCTION ...................................... 61
METHODS ........................................... 68
Selection of community representatives ................. 68
General approach to recording local knowledge .......... 70
Identification of key individuals and groups within the
community ................................. 72
Identification of household needs ..................... 73
Inputs to, outputs from and measures of performance of, the
Masoka agroecosystem ....................... 74
Methodological validity and data accuracy ............... 80
RESULTS ........................................... 85
Key actors and household classes .................... 85
Household needs ................................. 88
System identification ............................... 94
Maize production ................................. 95
Cotton production ................................ 99
Sorghum production .............................. 102
Summary of inputs to and losses from crop production sub-
systems ................................... 102
Rainfall analysis .................................. 105
Labor requirements for each crop enterprise ............. 112
Soils and soil fertility ............................... 118
Crop yields ..................................... 122
Performance measures ............................. 130
Problem formulation ............................... 143
SUMMARY AND CONCLUSIONS ......................... 150
BIBLIOGRAPHY ...................................... 153
Chapter 5. ................................................ 160
Simulation model: development, structure and testing ............. 160
INTRODUCTION ...................................... 160
MODEL DESCRIPTION ................................. 165
Overall approach ................................. 165
Model overview .................................. 167
Rainfall generating model (RAINGEN) .................. 169
Initializing model (INITIAL) ..................................... 171
Household size generation and household field
allocation ............................. 171
Cropping history ............................ 174
Vegetation cover ............................ 174

Outputs from INITIAL ......................... 176

Agroecosystem model (MASOKA1) .................... 178

The STORAGE sub-system ..................... 178

The ALLOCATIVE sub-system ................... 182

The LABOR sub—system ....................... 186

The CROP subosystem ........................ 191

The EROSION sub-system ..................... 200

The ECONOMIC sub-system ................... 205

The OUTPUT sub-system ...................... 207

METHODS ........................................... 213
Verification, validation, usefulness ..................... 213
Sensitivity analysis ................................ 215
VERIFICATION AND VALIDATION ......................... 222
Rainfall and seasons ............................... 222

Initial conditions .................................. 222

Maize yields ..................................... 227
Cotton yields .................................... 229
Sorghum yields .................................. 230
Erosion ........................................ 232
RESULTS ........................................... 233
Sensitivity analysis ................................ 233
Household needs satisfaction ................... 234

Crop yields ................................ 243

Economic response variables ................... 246

Erosion, erodibility and "fertility" .................. 253

SUMMARY AND CONCLUSIONS ......................... 257
BIBLIOGRAPHY ...................................... 262
Chapter 6. ................................................ 267
Overview and key Issues for further research .................... 267
INTRODUCTION ...................................... 267
OVERVIEW .......................................... 267
KEY ISSUES FOR FUTURE RESEARCH .................... 272
BIBLIOGRAPHY ...................................... 276

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 3.1

Table 3.2

Table 3.3

Table 3.4

Table 3.5

Table 4.1

Table 4.2

LIST OF TABLES

Major soil types of Kanyurira Ward, mapped at 1:250 000 (Anderson,
1987). .......................................... 15

Total revenues generated by wildlife in Masoka in each year since 1989,
dividends paid per household and total revenues paid to households In

Zimbabwe dollars (ZS). ............................... 23

Census population estimates of large herbivores in Kanyurira Ward, 1988
to 1992. .................................................. 24

Land types within the Masoka fenced area with Zimbabwean and USDA soil
classifications. ..................................... 25

Household needs for maize. cotton and sorghum, by household size
class. Data are mean _-I_-_ SEM, with n in parentheses. ............. 43

Maize, cotton and sorghum yields obtained by respondents for the
1990/91 season on three soil types. Data are mean .t SEM, with n in
parentheses. ...................................... 45

Anticipated yields for hand cultivated maize, cotton and sorghum on three soils
In good and bad rainfall seasons. Data are mean‘ i SEM, with n In
parentheses. ...................................... 46

Perceived yields for tractor cultivated maize. cotton and sorghum on
three soils in good and bad rainfall seasons. Data are mean 3; SEM, with
n in parentheses. ................................... 47

Cropland areas (ha) required to satisfy household needs. Data are total area for
household size class with the area per adult equivalent for each size class In
parentheses. ...................................... 50

Subjective probabilities indicating VR perceptions of the probability of the number
of rainfall events, of each rainfall type. occurring in any two week period of the
rainy season ....................................... 107

Classification of ten rainfall seasons ending 1991/92 as good (G).

average (A) or poor (P) by village representatives (VRs), a local informant
and questionnaire survey (1988/89 to 1990/91 only). ............. 111

vii

Table 4.3

Table 4.4

Table 4.5

Table 4.6

Table 4.7

Table 4.8

Table 4.9

Table 4.10

Table 4.11

Table 4.12

Table 4.13

Table 4.14

Table 4.15

Table 5.1

Table 5.2

Labor requirements (days ha“), for each activity of maize production in each
month ofthegrowingseason. ........................... 113

Labor requirements (days ha") for each activity of cotton production for each
month of the growing season. ........................... 114

Labor requirements (days he") for each activity of sorghum production, in each
month of the growing season. ........................... 114

Relative importance weights (5 highest to 1 lowest) of tree and herb or grass
species after a jecha field Is left fallow. ..................... 119

Mean (standard deviation in parentheses) yield estimates (kg ha' 1) for
maize, cotton and sorghum, derived from household field maps and

household respondent information. ........................ 123

Maize, cotton and sorghum yields reported for the mid-Zambezi Valley. . 123

Expected values for maize, cotton and sorghum yields (kg ha' 1) on three
soils and for good, average and poor rainfall seasons. Source. possibility
diagrams. ........................................ 126

Mean (standard deviation in parentheses) number of adults, number of
children and years residence in Masoka for classes of household well-

being. .......................................... 1 35

Mean (standard deviation in parentheses) for characteristics of
households in each of four employment classes. ............... 136

Partial budgets for one hectare maize production enterprises with different input
and pricing scenarios. 1991/92 season prices. ................. 138

Partial budgets for one hectare cotton production enterprises with different Input
scenarios. 1991/92 season prices. ........................ 139

Partial budgets for one hectare sorghum production enterprises with
different input and price scenarios. 1991/92 season price data. ...... 140

Expected returns (23 ha") to land, to labor and to Initial investment for
maize, cotton and sorghum on each soil type. ................. 141

Erodibility values for cultivated and uncultivated jecha and bepe soils in Masoka.
Source Elwell, personal communication. ..................... 202

Summary of output variables written to household output file
(HOUSEHLDDUT). .................................. 209

viii

Table 5.3

Table 5.4

Table 5.5

Table 5.6

Table 5.7

Table 5.8

Table 5.9

Table 5.10

Table 5.11

Table 5.12

Table 5.13

Summary of output variables written to SUMMARY.OUT and to
MONTECAR.OUT. ................................... 212

Factorsandfactor levelsusedin sensitivity analysisofMASOKA1. . . .. 216

Goodness of fit test statistics for the distribution of numbers of children
per number of adults. ................................ 23

Simulated maize yields (kg ha") for each soil and in each season. Data
are means with standard deviations in parentheses. medians and the

sample size. ...................................... 228

Simulated cotton yields (kg ha") for each soil and In each season. Data
are means with standard deviations In parentheses, medians and the

sample size. ...................................... 29

Simulated sorghum yields (kg ha") on jecha in each season. Data are
means, with standard deviations in parentheses, medians and the sample

size. ........................................... 231

Sensitivity of the mean of the 96 of households that were cash (ACDF),
maize (AMDF) and sorghum (ASDF) deficit, each year, to changes in
factor levels. ...................................... 236

Sensitivity of mean annual yields of maize (MZYLD), cotton (CTYLD) and
sorghum (SRYLD) to changes in factor levels. ................. 241

The effects of factor level changes on the means of total retums
(MZREI), returns to land (MZHA). labor (MZLAB) and to initial investment

(MZINV) for maize. .................................. 247

The effects of factor level changes on the means of total returns (CTRET),
returns to land (CTHA), labor (CTLAB) and to initial investment (CTINV)
for cotton ......................................... 248

The effects of factor level changes on the means of total returns (SRRET),
returns to land (SRHA), labor (SRLAB) and to initial investment (SRINV)
to sorghum. ...................................... 249

ix

Figure 2.1

Figure 2.2

Figure 2.3

Figure 2.4

Figure 3.1

Figure 3.2

Figure 3.3

Figure 3.4

Figure 4.1

Figure 4.2

Figure 4.3

Figure 4.4

LIST OF FIGURES
Map of Zimbabwe showing the location of the study site ..... 14

Map showing the base of the Zambezi escarpment (600m). major
road routes and important towns or settlements near Masoka . 16

Precipitation (mm). from Angwa Bridge (n=15) and evaporation
(mm) from Muzarabani (n=6). Data are monthly means with bars
showing SEM. ................................... 18

Map showing Masoka land types and game fence ......... 21

Estimated numbers of households in Masoka, 1985 to 1992
and predicted numbers of households through 2010 based
on a linear and an exponential projection model. .......... 42

Household needs (ZS) and earning potential for large
households over the period 1978 to 1992 ................ 49

Household needs (ZS) and earning potential for small
households over the period 1978 to 1992 ................ 50

Arable land required and remaining under the population
growth scenarios presented in Figure 3.1 and the land
requirements presented in Table 3.5.

Major actors influencing agricultural production in Masoka
and their relative importance weights. Female VR perspective . 86

Major actors influencing agricultural production in Masoka
and their relative importance weights. Male VR perspective . . . 86

Household classification and the relative importance weights
of each household class. Male VR perspective. ............ 87

Household classification and the relative importance weights
of each class. Female VR perspective. .................. 87

Figure 4.5

Figure 4.6

Figure 4.7

Figure 4.8

Figure 4.9

Figure 4.10

Figure 4.11

Figure 4.12

Figure 4.13

Figure 4.14

Figure 4.15

Figure 4.16

Figure 4.17

Household needs. with their relative Importance weights (RIW).
as well as the major inputs to those needs with their RIW.
Female VR perspective. ............................. 90

Household needs, with their relative importance weights (RIVV),
for a large household as well as the major inputs to those
needs with their RIW. Male VR perspective. .............. 91

Household needs, with their relative importance weights (RM),
for a medium household as well as major inputs to those
needs with their RIW. Male VR perspective. .............. 92

Household needs, with their relative importance weights (RIVV),
for small households as well as major inputs to those
needs with their RIW. Male VR perspective. .............. 93

Female VR perspective of what is required for, and what detracts
from, the achievement of good maize yields. .............. 94

Male VR perspective on what is required for, and what detracts
from, the achievement of good maize yields. .............. 95

Female VR perspective on what is required for, and what detracts
from, the achievement of good cotton yields. ............ 100

Male VR perspective on what is required for, and what detracts
from, the achievement of good cotton yields. ............ 101

Female VR perspective on what is required for, and what detracts
from, the achievement of good sorghum yields. .......... 103

Male VR perspective on what is required for, and what detracts
from, the achievement of good sorghum yields. .......... 104

Probability of a rainfall event of each of four rainfall types
occurring in each week of the rainy season. ............. 108

Probability of a rainfall event lasting at least a certain time
for each of the four types of rainfall. ................... 110

Increase in total tree biomass with time after field abandonment

as a proportion of the maximum achievable in 15 years for major
tree species growing on jecha. ....................... 120

xi

Figure 4.18

Figure 5.1

Figure 5.2
Figure 5.3
Figure 5.4

Figure 5.5

Figure 5.6

Figure 5.7

Figure 5.8

Figure 5.9

Figure 5.10

Figure 5.11

Figure 5.12

Figure 5.13

Figure 5.14

VR derived performance measures to assess the liklihood of a

household satliying its basic needs. ................... 131
Major components of the Zambezi Valley Agroecosystem

Model (ZVAM) ................................... 166
Input and output files for the initializing model (INITIAL) ..... 175

Input and output files for the agroecosystem model (MASOKA1J77

Sequence of major components of MASOKA1 ............ 178
Flow chart illustrating sequence of events in STORAGE

sub-system ..................................... 179
Flowchart of activities in the ALLOCATIVE sub-system ...... 181

The cumulative probability of leaving a jecha field fallow
for households using income maximizing (MAXINC), variance
minimizing (MINVAR), needs satisfying (SATIS) and environ-

mentally protective (GREEN) production strategies ........ 184
Flowchart of major activities in the EMPLOY component of the
LABOR sub-system. ............................... 185
Flowchart of major activities in yield correction component

of the LABOR sub-system. .......................... 187
Flowchart of major activities in CROP sub-system. ......... 190

The proportion of respondents allocating each growing season
from 1978/79 to 1991/92 to good, average or poor seasons
plotted against the rainfall for that season measured at Angwa
Bridge. Rainfall data from the Department of Meteorological
Services, Harare. ................................. 192

Membership functions for good, average and poor seasons used
In MASOKA1. .................................... 193

Example yield generation function from MASOKA1. Dependent
variable is yield (kg ha ) and independent variable is the
random variate p. ................................. 197

Flowchart of major activities in EROSION sub-system. ...... 204

xii

Figure 5.15

Figure 5.16

Figure 5.17

Figure 5.18

Figure 5.19

Figure 5.20

Flowchart of major activities in ECONOMIC sub-system. . . . . 206

The cumulative proportion of a) actual (n=70) and b) simulated
(n=1000) Masoka households cultivating a given area of non-
riverland. ....................................... 223

The cumulative proportion of a) actual (n=70) and b) simulated
(n=1000) Masoka households cultivating a given area of
riverland. ....................................... 224

Causal loop diagram of the effects of increasing household cash
needs on household maize and cash balances. .......... 237

Good (G), average (A) and poor (P) season membership functions
used for +1 and -1 levels in sensitivity analysis. .......... 243

Total number of years each cell had been cultivated, summed
over the landscape for +1 fallow practice factor level (min=0,
max=10 years) and -1 factor level (min=5, max=7 years). Result
of single simulation. ............................... 254

xiii

ABBREVIATIONS

AGRITEX ...... Department of Agricultural Technical and Extension Services
CA ............................................ Communal Areas
DMS .......................... Department of Meteorological Services
FSRU .............................. Farming Systems Research Unit
GOZ ................................... Government of Zimbabwe
MAPS .................. Multispecies Animal Production Systems Project
MLARR ............. Ministry of Lands, Agriculture and Rural Resettlement
MZVRDP ................. Mid-Zambezi Valley Rural Development Project
VR ........................................ Village Representative
WWF ................................. World Wlde Fund for Nature

xiv

Chapter 1.

Background and justification

INTRODUCTION

Two problems need to be overcome in analyzing the behavior of
agroecosystems‘. The first of these Is the lack of a suitable and philosophically
and methodologically coherent approach to the analysis of agroecosystems. The
second is the problem of meaningfully applying performance criteria such as
sustainabilty or stability in real world decision making.

Generally speaking science is ill-prepared, both methodologically and
philosophically, to solve real world problems of agroecosystems where the
performance criteria are sustainability, stability or similar measures requiring a
predictive understanding of agroecosystem behavior. MacRae et al. (1989) review
several scientific barriers to achieving sustainable food production. They discuss

the failings of the reductionist approach to scientific investigation, question

 

Agroecosystems are defined as ecosystems in which basic biological processes are managed
to obtain food and fibre as well as other goods and services. They Include social. managerial or
decision making and economic as well as biophysical components. In Zimbabwe Communal
Areas and at the scale of this analysis they incorporate a number of households. They include
therefore, a political component as well.

2

scientific objectivity, particularly in fields of research such as agriculture, and the
common belief that quantifying Is an essential precursor to rational evaluation of
facts.

There can be little doubt that reductionist science has been a powerful
problem solving paradigm. It is limited, however, in its ability to deal with problems
associated with systems and especially where multiple performance criteria are to
be used (Bawden, 1991; MacRae et al., 1989). The perception of science as an
objective and value free search for the truth has not been substantiated by
research (MacRae et al., 1989; Busch and Lacey, 1983; Mahoney, 1979). The
perception of some agricultural scientists that it is possible to remain objectively
detached from social and economic processes Is a dubious assumption at best.
Values influence what problems are addressed and which solutions are considered
acceptable (or considered at all). These decisions have political, social and
economic consequences that should be recognized. Failure to explicitly define
values in agricultural research could result in the introduction of sources of bias
not accounted for in subsequent analyses.

Agricultural scientists rely on quantitative data to support or refute their
assertions and are more likely to believe quantitative arguments (MacRae et al.,
1989; Mahoney, 1976). Quantifying states and outputs in agroecosystems, with
their tightly coupled bio-physical, social and economic sub-systems, is difficult.
Reliance on quantified relationships could lead to emphases that would not be

supported by giving equal weight to those system attributes that are difficult to

3
quantify and for which there might only be qualitative relationships. By quantifying

relationships and states analysts also stand the risk of expressing a precision that
is not a true reflection of their knowledge of the world.

Philosophically, much of contemporary science uses the hypothetica-
deductive approach in establishing whether or not assertions are false. For the
most part scientists are concerned with the probability of making a Type I error
(rejecting a true null hypothesis) but rarely examine the likelihood of making a
Type II error (accepting a false null hypothesis). Yet, in the world of decisions,
costs and benefits, these errors may have equally expensive consequences
(MacRae et al., 1989; Officer and Dillon, 1968). Aversion to making Type I errors
is a strong incentive to reduce problems to single variables to ensure that
observed effects are due to factor changes. The assumptions imposed on
observation, in order to use classical inferential statistics, constrain our ability to
react to "scientific complexity" (Box and Tlao, 1992). The hypothetico-deductive
methodology, as formulated by Popper (1968), also makes the assumption that,
a priori, we know nothing whatsoever about the subject matter of interest (Stove,
1982). In basing decisions on this view a tremendous body of knowledge and
previous experience are discarded. Given the complexity of the performance
criteria used in comparing agroecosystems or agroecosystem outputs as well as
the sometimes high costs of making either Type I or Type II errors, this seems a
wasteful practice; in essence the tools used are dictating the nature of the

problems that are addressed.

4

The systems approach is one that offers promise for dealing with the
complex requirements of design and analysis in agricultural systems (Bawden,
1991). The soft systems approach of Checkland (Checkland, 1981; Checkland and
Scholes. 1990) was specifically developed to deal with problems where neither the
objectives nor the system boundaries are clearly defined, or perhaps even clearly
definable. This may often be the case with agricultural systems. The soft systems
approach does not stress the value of computer modeling as is the case in
conventional or "hard" systems approaches. The systems approach is not without
its critics however; Belinski (1976) is scathing in his attacks on the misuse of logic,
mathematics and statistics in systems science and particularly when these
methods have been applied to social and biological systems. Hoos (1972) is
critical of the ability of the systems approach to adequately deal with public policy
issues. She is even more damming of attempts to predict the future of social
systems (Hoos, 1974). These criticisms are often valid and should not be ignored.
Criticism is a necessary but, not sufficient, condition for change and improvement.
Change also requires trying new techniques or taking what is deficient and
improving it. So, whilst the systems approach is clearly not perfect it is a useful
approach with which to start and from which to develop new approaches to
agroecosystem analysis. The well developed and tested approaches to analysis,
including the large body of knowledge on many aspects of modeling dynamic
systems, offers a rich tool box with which to undertake the analysis of

agroecosystem behavior.

5
Perhaps the most widely advocated approach to the analysis of agricultural

systems in the developing world is the agroecosystem analysis approach of
Conway (Conway, 1987, 1986. 1985). Agroecosystem analysis (after Conway) uses
four system properties to describe system behavior. These are productivity,
stability, sustainability and equitability. Grimm et al. (1992) provide a checklist of
factors that should be explicitly defined when discussing ecosystem stability
concepts; a checklist that appears equally appropriate for use with sustainability,
productivity and equitability as it does with stability. These factors are: a) the level
of description (whether individual. population or ecosystem for example); b) the
variable of interest; c) the referential behavior of the variable of interest; d) the
nature of the disturbance; e) the spatial scale; and f) the temporal scale.

In the view of many systems’ theorists, systems exhibit what are called
emergent properties; these are properties that are not deducible from observations
of the individual parts of the system (Bawden and lson, 1992). If we accept these
assertions then the selection of the level of analysis as well as the spatial and
temporal scales of analysis are made very much more complex. Changes in
factors at one level of analysis could lead to unexpected changes at others or an
agroecosystem could be stable at one scale but not at another.

Data suitable for a quantitative description of both the current state of the
agroecosystem as well as those describing the behavior of the agroecosystem
over space and in time are seldom likely to be available and particularly so in

deveIOping countries.

6

These problems are compounded in practical decision making by issues
of whose values need be considered and the large number of variables and
performance criteria that need to be considered simultaneously. Policy makers and
individual heads of household are unlikely to share the same set of values.
Theoretically at least, the policy maker could be expected to act in the best
interests of society whilst the household head could be expected to act in the best
interests of the household. As analysts we are faced with the dilemma of selecting
which set of values are to be used in the analysis. There do not seem to be any
clear cut criteria for selecting one view of the world as opposed to another.

In Conway’s approach (Conway, 1985. 1987) these issues are not
addressed. The approach is orientated toward the analyst’s world view.
Techniques that include local farmer participation are used for data collection and
the farmers in the target agroecosystem appear to participate only as passive
sources of data rather than active participants in an client orientated analysis.
Conway (1985) admits that the properties he defines are more easily defined than
measured and that "satisfactory methods of measuring sustainability still need to
be found." In more recent applications of the methodology (Conway. 1987) a team
of experts is used in a workshop format, to subjectively select patterns of space.
time, flow and decisions, that are likely to reveal the key functional relationships
that determine system properties.

The agroecosystem analysis approach of Conway (1986, 1985) does not

advocate predictive modeling. Evaluations of sustainability, stability and equitability

7
appear to be made subjectively by the team of experts. With scarcely any data at

all, let alone time series data, this may be the only approach that is possible, yet
it can lend itself to biased and erroneous interpretations. Under Zimbabwean
communal area conditions, farmers and analysts are often from different cultures
and backgrounds and share very different futures. Few of the analysts are likely
to have spent more than a few weeks, let alone months, in a communal area
farming system. Fewer still of the analysts will have futures that are more than
cursorily linked to the outcome of the analysis. Given that the analysts and
communal area farmers have neither values nor objectives in common and the
analysts are not accountable to households for their decisions it appears unlikely
that a team of professionals can reliably predict the sustainability, stability and
equitability performances of an agroecosystem.

It Is undeniable, that there are a number of philosophical and
methodological tools available, which could be gainfully used to develop insights
appropriate to the analysis of complex agricultural systems, particularly those in
the developing world. There does not. however, appear to be a philosophically and
methodologically coherent approach to agroecosystem analysis for use with
performance criteria such as sustainability, stability, resilience or equitability, and
in particular where there are no data except local or indigenous knowledge.

The World Wildlife Fund Multispecies Animal Production Project (WWF
MAPS) was established in 1988 with one of its major objectives being to critically

examine the ecological and economic implications of multispecies and single

8
species animal production systems in Zimbabwe (Cumming. 1991). A number of

the hypotheses to be tested by the project used sustainability, stability and
resilience as performance criteria. As a research fellow working on the WWF MAPS
project, I developed a research program designed to examine the sustainability of
Masoka, an agroecosystem located in the mid-Zambezi Valley.

The following chapters describe the research I conducted as part of the
WWF MAPS Project. My original objective was to examine the sustainability of
Masoka. The result is a first, and admittedly rather crude, methodology for
conducting agroecosystem analyses where the local people were seen as equal,
if not major participants in the analysis.

I have not been able to address any of the philosophical problems
associated with research of this kind. I have also failed to address any of the
difficulties of statistically analyzing the data collected in the field as well as the
simulated results from the model. Despite the obvious importance of resolving
these philosphical and analytical issues I have not had the time. in this study, to
do more than recognize their existence.

In Chapter 2 I briefly present background data on the Masoka
agroecosystem. In Chapter 3, material that was used to develop a general
understanding of the agroecosystem and the constraints household face in
satisfying needs are presented. Chapter 4 is a description and evaluation of the
methods used for field data collection; summaries of these data are included. In

Chapter 5, development of the Masoka agroecosystem model is described and

9

the results of preliminary sensitivity analyses of the model are presented. In the
final chapter, major features of the research are reviewed and I make a few

suggestions about key issues for future research.

BIBLIOGRAPHY

Bawden, R.J. and BL lson. 1992. The purposes of field-crop ecosystems: Social
and economic aspects. In: Pearson, C.J. (Ed.), Ecosystems of the world. Field
crop ecosystems. Elsevier, Amsterdam, Holland. 11-35.

Bawden, R.J. 1991. Systems thinking and practice in agriculture. Journal of Dairy
Science, 74, 2362-2373.

Belinski, D.J. 1976. On systems analysis: an essay concerning the limitations of
some mathematical methods in the social, political, and biological sciences. The
MIT Press, Cambridge, MA, USA.

Box, G.E.P. and 6.0. 1130. 1992. Bayesian inference in statistical analysis. John
Wiley and Sons, New York, USA.

Busch, L and W.B. Lacy. 1983. Science, agriculture, and the politics of research.
Westview Press, Boulder. CO, USA.

Checkland, P. and J. Scholes. 1990. Soft systems methodology in action. John
Wiley & Sons, New York, USA.

Checkland, P. 1981. Systems thinking, systems practice. John Vlfiley & Sons,
Chichester, UK

Conway, GR. 1985. Agroecosystem Analysis. Agricultural Administration, 20.
31 -55.

Conway, GR. 1986. Agroecosystem analysis for research and development.
Winrock International Institute. Bangkok, Thailand.

Conway, GR. 1987. The Properties of Agroecosystems. Agricultural Systems, 24,
95-117.

Grimm, V.E. Schmidt and C. \Mssel. 1992. On the application of stability concepts
in ecology. Ecological Modelling, 63, 143- 161.

10

11

Hoos, LR. 1974. Criteria for ”good" futures research. Technological Forecasting
and Social Change, 113-132.

Hoos, LR. 1972. Systems analysis in public policy. University of California Press,
Berkeley, CA, USA.

MacRae, R.J., S.B. Hill, J. Henning and GR. Mehuys. 1989. Agricultural science
and sustainable agriculture: a review of the existing scientific barriers to sustainable
food production and potential solutions. Biological Agriculture and Horticulture.
6, 173-219.

Mahoney, M.J. 1979. Psychology of the scientist: An evaluative review. Social
Studies of Science, 9, 349-375.

Mahoney, M.J. 1976. Scientist as subject: the psychological imperative. Ballinger,
Cambridge, MA, USA.

Marten. G.G. 1988. Productivity, stability, sustainability, equitability and autonomy
as properties for agroecosystem analysis. Agricultural Systems, 26, 291 -31 6.

Officer, RR. and J.L Dillon. 1968. Probability and statistics in agricultural research
and extension: A pro-Bayesian view of modern developments. The Journal of the
Australian Institute of Agricultural Science, September, 121-129.

Popper, Sir KR. 1968. The logic of scientific discovery. Harper and Row, New
York, USA.

Stove. DC 1982. Popper and after. Four modern irrationalists. Pergamon Press,
Oxford, UK

Chapter 2.

Background data on the Masoka agroecosystem

INTRODUCTION

In this chapter I present background data on Kanyurira ward. Kanyurira
ward was selected as a study site because: a) it was one of the first CAMPFIRE
projects in the Zambezi Valley; b) it has a small and relatively discrete human
population; c) studies by the Centre for Applied Social Sciences (CASS) at the
University of Zimbabwe (Cutshall, 1989) as well as those by the WWF Multispecies
Animal Production Systems project (T aylor, 1993) provide useful background
resources; d) the community has been living in the area for, at least, several
generations and thus could be expected to exhibit traditional. as well as the more
modern, technologies that are spreading through the mid-Zambezi valley area; 9)
the absence of cattle considerably simplified analysis of the agroecosystem; f)
Masoka is still relatively undeveloped. The community has therefore. a large set of

development options that it could follow.

12

13
Study site description

Administratively, Zimbabwe is divided into provinces, districts, wards and
villages. Kanyurira Ward (Masokaz) is situated in Guruve District, Mashonaland
Central Province, Zmbabwe, between 30° 10’ E and 16° 15’ S (Figure 2.1). Nearly
170 km2 (40%), of the 400 kmz, lie above the Zambezi escarpment (Taylor, In
preparation). The altitude of the ward ranges from 1120 meters at the top of the
escarpment to about 400 meters where the Angwa river leaves the ward in the
north east (Figure 2.2). The topography of the ward ranges from steeply dissected
slopes on the escarpment to gently sloping alluvial terraces along the major rivers.

The geological formations of the area are largely recent to Pleistocene
alluvial sands and gravel, Triassic sandstone (upper Karoo group) and micaceous
sandstone, the upper portion of the latter alternating with mudstone (Broderick,
1989; Oesterlen, 1989). There is little likelihood of useable shallow aquifers in the
ward (Owen. personal communication, 1991; Owen, 1989).

In general, the river systems of the ward drain north and south into the
Angwa River which rises about 120km to the south. above the escarpment (Figure
2.1). All rivers flow seasonally. The Angwa River is dry for most of the year (May
to October), floods after the onset of the rains (November or December) and has
peak flows in January and February.

Masoka has a long dry season from April to November and a wet season

 

The name Kanyurira is derived from the sub-Chief of the Ward, Headman C. Kanyurira. Masoka
Is the name of the royal spirit, (Mhondoro) of the area and Is the name In common usage. It Is not
used for the whole ward but is used to refer to the area of present habitation. I shall use Masoka
when describing the area of current habitation and Kanyurira to refer to the Ward.

     

96:25

 

 

/
Z\

ugm<m§N

14

80.6:

So can m“— «new.
.27. Sam 9: com
8:82 9: 932m oEBEQ So no: 3 .mcm

15

from December to March. Mean annual rainfall (725mm, n=15, CV=34%) was
recorded at Angwa Bridge, some 15km north east of Masoka. Mean annual
evaporation (2050mm, n=6. CV=11%) was measured at Muzarabani,
approximately 90km south east of Masoka (Figure 2.2). Mean monthly rainfall
values and mean monthly pan evaporation are shown in Figure 2.3. Mean monthly
precipitation seldom exceeds mean monthly evaporation. The Zimbabwe
Department of Meteorological Services, (Department of Meteorological Services,
1981) report increasing evaporation northward from the Zambezi escarpment.
Evaporation at Masoka ls therefore, likely to be somewhat higher than that
measured at Muzarabani. The rainfall of the area is highly erosive with annual
energy values estimated by Stocking and Elwell (1976) to be between 11000 and
13000 J mm" m'zh'1.

Anderson (1987) identified three major soil types at a 1:250 000 scale (T able

2.1). The soils were described and classified following Thompson and Purves

Table 2.1 Major soil types of Kanyurira Ward, mapped at 1:250 000 (Anderson, 1987).

 

 

Zimbabwe soil name Classification
Zimbabwe USDA FAO
Sialitic 4S Typic ustorthent Calcaric regosol
Strongly sodic 8N Typic naturastalf Orthic solonetz
Alluvial vertisols SS Typic pellustert Pelllc vertisol

 

(1978) and Thompson, (1965). The soils of this part of the Zambezi Valley are

1(3

 

 

 

<
2

95.50
Estabofl 39:5

8... 589;.

      

 

coo coo .2 28m

.3082 too: $5553
.5 8:3 Ectoaé nco «2:8 88 Lace Apnoea
$8983 _NBEoN on. 2o 38 9: 9:65 no: Nu .m:

 

 

 

 

ragga
o; .2 ice.

<_m2<N

 

 

17

considered to be moderately erodible; low vegetation cover and high soil
erodibility being the major factors contributing to this erodibility (Madhiri and
Manyanza, 1989).

The vegetation of Kanyurira has been described at scales of 1:250 000
(Anderson, 1987) and 1:50 000 (T aylor, 1993). Taylor describes 14 different
vegetation types, broadly grouped into i) riverine and alluvial vegetation; ii) dry
deciduous forest; iii) Colophosperrnum mopane communities; iv) miombo
communities; v) Termina/ia communities; and vi) mixed mopane-miombo
communities. Taylor suggested a close association between these vegetation
types and the soils of the area.

Most residents of Masoka claim to have lived at Marla Angwa, 10km south
west of the present centre of habitation (Figure 2.2), prior to 1965 and to have
lived in the general area of Masoka for several generations. In 1979. all
households3 were moved into a "protected village" at Angwa Bridge as part of the
Rhodesian government’s attempts to isolate nationalist guerrillas from the
population. The households moved back to their current centre of habitation,
Masoka, in 1980. By the end of 1992 there were about 143 households in the
ward, an almost 140% increase since 1988, when Cutshall (1989) surveyed the
community and counted 60 households. Part of this rapid growth can be attributed

to the arrival of between 30 and 35 households of VaDema people who settled on

 

The definition of what constitutes a household Is somewhat difficult (Hammel, 1984; Guyer, 1981).
Throughout this study I use the definition supplied by key informants. A household is defined by
one of the following: a married couple; a widow or widower; or a couple who have lived together
all their lives.

ITLCLL
S: .
“a

H:

 

 

18

300 I I I I I I I I I I I I

l- T T
275 Precipitation: Mean 725, SEM 63.8 mm Y!“
250 ' Evaporation: Mean 2052, SEM 78 mm ‘

225 - v _
200 - v/ i
175 - .
150 - BIZ V/ .
125 » \§—§\v/ «

100 -

i -
50L \ l
25' §\ i

 

Precipitation and Evaporation (mm)
IGI

 

 

 

 

O 1 I I I H . Q l l I
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month
v Evaporation O Precipitation

Figure 2.3 Precipitation (mm), from Angwa Bridge (n=15) and evaporation
(mm) from Muzarabani (n=6). Data are monthly means with bars
showing SEM.

19
the northern banks of the Angwa River sometime in 1990/1991. Excluding this

immigration the growth rate in the number of households in the ward is still about
10% per year. Ethnically, the community is dominated by Korekore (Cutshall, 1989)
but has a large population of VaDema, as well as a few Chikunda and Malawian
households. In late 1992, ten households from Masvingo Province were allowed
to settle in the ward.

Politically, Kanyurira Ward is divided into three Village Development
Committees (VIDCOS). A ward councillor, who represents ward interests in
meetings of the district council, is elected by popular vote of all households in the
Ward. These political institutions were initiated after Independence in 1980.
Traditional leadership is vested in the Chief (in this case sub-chief Kanyurira), the
spirit medium and village headmen. In Kanyurira these traditional leaders still play
a major role in directing community affairs. Important organizations that influence
community planning and decision making are the Department of Agricultural,
Technical and Extension Services, (AGRITEX) which has a representative at Angwa
Bridge, the Department of National Parks and Wild Life Management (DNPWLM)
which has a large station at Mkanga Bridge, 11km north west of Masoka, and the
Tsetse Control Branch of the Department of Veterinary Services which has a base
at Mana Angwa.

In 1988, Guruve was one of the first districts in Zimbabwe to begin
exercising authority over its wildlife resources. This was formalized in 1991 when

Appropriate Authority status was granted by the central government in terms of the

20
1975 Parks and Wild Life Act. With Appropriate Authority status, district councils

have control over the use of their wildlife resources, with any income generated by
wildlife utilization projects being available to the district council (Cumming, 1991;
Peterson, 1991). During 1989 the Kanyurira community developed a wildlife land
use plan (Kanyurira Wildlife Committee, 1989). Based on this plan, the community
chose to fence slightly more than 14km2 along the south bank of the Angwa River,
and to limit settlement and cultivation to within this area (Figure 2.4). All of the
households except those of the VaDema and one of the original Korekore
households live and carry out most of their cultivation within this fenced area. A
number of households in the community (n=42), however, have fields on the
recent alluvium outside of the fenced area.

Vlfildlife in the ward generate revenues in the form of trophy and
accommodation fees paid to the professional safari hunter who leases the ward
hunting concession from the district council‘. The professional hunters pay the
district council trophy fees and a lease or concession fee for use of the ward. The
district council deducts a district council levy, a budget allocation for resource
management and a levy for the Campfire Association; the remainder of wildlife
revenues belong to the ward. Theoretically, each ward should decide what to do
with these revenues but in practice the Guruve district council makes these

decisions (Peterson, 1991). Each household in Masoka received direct payments

 

For the first three years the District Council managed the safari operation. More recently the
wildlife enterprise is managed by a professional safari operator with contracts being awarded, by
the District Council, on consideration of submitted tenders.

21

 

 

EELS _m;::¢ E
880: a

atom a
Press I
/—\oc:e:a a
3:: a

z .3

 

 

 

........

08mm“— 2on
96 m8: 9.2 308: 938.0. no: am 8%:

      

 

 

 

 

 

In

rll.

ar

SO
is)
rIVi
rivi

rec

22

(dividends) from ward wildlife revenues three times since 1989 (Table 2.2). These
dividends are equivalent to between 9 and 2800% of the annual income that
Cutshall (1989) has reported households derive from cotton production. The wide
range of areas that households reported planting to cotton is likely to have
resulted in the variability in total returns to cotton reported by Cutshall.

lMld animal populations in Masoka have been estimated regularly since
1989 (Mackie, 1993; Taylor. 1991; Taylor and Cumming, 1989). The populations
of the major herbivores show considerable inter-annual variation (T able 2.3),

indicating that these populations are using resources that expand over an area
greater than that of the ward.

The 14km2 fenced area is of primary concern in this study. The soils of this
are mostly alluvial deposits of varying ages and were mapped with 1:9750 air
photography (Figure 2.4). Four major land classes, comprising four soil types,
were identified (Table 2.4). Two additional land types were identified (hilly and
riverine) but were not investigated as they comprise a small proportion of Masoka
and are not important agriculturally. The mopane woodland area comprises deep,
medium textured sandy Ioams with small, localized patches of fine textured sodic
soils. The alluvial or river terraces slope away from the original river bank. Soil
' texture on these terraces grades from medium textured sands near the historical
river bank to fine textured sandy clays on the down-slope edge of the terrace. The
river terraces have been the focus of traditional agriculture in Masoka. More

recently however, settlers have been moving into areas of elevated old alluvium

23

Table 2.2 Total revenues generated by wildlife In Masoka In each year since 1989,
dividends paid per household and total revenues paid to households In
Zimbabwe dollars (ZS).

 

 

YEAR Total Wildlife Dividend per Total Revenues
Revenues Household paid to
Households
(Z$) (23) (25)
1989 47,310 200 17,200
1990 78,170 NIL NIL
1991 89,293 NIL NIL
1992 276,746 400 56,000
TOTAL 491,519 600 73,200

 

and even onto some sodic areas.

Traditionally, households satisfy their food and cash needs largely
from rainfed crop production and illegal huntings. Crop production is based on
grain crops, primarily maize (Zea mays L.) and sorghum (Sorghum bicolor L.)
inter-cropped with cucubits such as watermellon, pumpkin and cucumber. Less
commonly grown are small plots of groundnuts (Arachis hypogaea) and sweet
potatoes (lpomoea batatas). Mean total areas cultivated for small, medium and
large households are about 1.3, 1.8 and 2.2 ha (this study, Chapter 3). Dry season
production is limited to vegetable crops grown in small gardens along the banks
of the Angwa River that are irrigated from shallow wells dug into the sand of the

river bed.

 

5

Vlfildlife in Zimbabwe belongs to the state. Hunting Is strictly controlled by the DNPWM. Poachers
in the Zambezi Valley face jail sentences if caught In the war against rhino and elephant
poaching, poachers are often shot.

24

Table 2.3 Census population estimates of large herbivores in Kanyurira Ward, 1988 to

 

 

 

 

1992‘.

{galloelynflgor 1988 1989 1990 1992
Numbers of animals

Elephant 209 412
Buffalo 1824 28 149
Rhino 24
Impala 459 380 82
Sable 74 15
Kudu 12 29
Escarpment
(1 82ka)
Elephant NOT SURVEYED 70 1 16
Buffalo NOT SURVEYED 23
Rhino NOT SURVEYED 23
Impala NOT SURVEYED 14
Sable lNOT SURVEYED 42
Kudu NOT SURVEYED 8

 

1. Source: Mackie, 1993; Taylor, 1991; Taylor and Cumming, 1989

 

 

0L

Ar

inc

inf

25

Table 2.4 Land types within the Masoka fenced area with Zimbabwean and USDA soil
classifications.

 

Land type Soil classification
Local soil description
Zimbabwe USDA

 

River terraces
Jecha Calcimorphlc 4U Typic rhodustalf
Bepe 3U Fluventic ustochrept
Elevated old alluvium
Jecha Calcimorphlc 4U Typic haplustalf
Mopane woodland Calcimorphlc
4U Typic haplustalf
Natric order Typic natrustalf
8NU
Sodic soils
Shapi Natric order 8NU Typic natrustalf

 

Cotton (Gossypium hirsutum) was introduced into Masoka in the mid-1970s
and it is now a major cash crop with 73% of the households growing it In 1988
(Cutshall, 1989) but only 57% growing it In 1991 (this study, Chapter 3). Fewer
than 7% of the households reported using inorganic fertilizers in the 1990/1991
growing season. Soil fertility in the older alluvial terraces is restored with a tree
fallow dominated by Acacia tortilis (subsp. heterocantha). In the recent alluvium
outside the fenced area, field fertility is restored by the periodic flooding of the
Angwa River.

Obtaining reliable estimates of the contribution of wildlife meat to household
incomes is difficult owing to the technical illegality of hunting in Masoka. Village

informants suggest that wild meat was always in plentiful supply. This source of

26

food has dropped to much smaller amounts since Masoka was fenced in 1989.

Masoka is not on a major transport or bus route. The nearest bus route is
on the road from Mkanga Bridge to Angwa Bridge. These buses stop on the north
bank of the Angwa but the river is not crossable during periods of heavy rainfall.
Most farm and household inputs are purchased In Guruve, about 140km south
east of Masoka (Figure 2.2), and then transported by bus to Angwa Bridge or to
the north bank crossing point. There is no public transport that goes directly to
Masoka. Cotton harvests are taken by tractor to the Cotton Marketing Board
(CMB) depot at Mushumbi Pools. Although a clinic is scheduled to be built using
revenues derived from wildlife, Masoka does not yet have either a clinic or a bank
(although the Agricultural Finance Corporation has a local agent who helps
villagers complete applications for credit). There is no electricity or piped water in
the community. Households obtain all drinking water from the Angwa river or from
shallow wells dug into the river bed during the dry season. The AGRITEX

agricultural extension agent visits the community less than once per season.

Summary

Despite having constrained household production activities to an area of
14km2, the Masoka community is growing very rapidly, both within and outside of
the fenced area. Household needs are satisfied from a diversity of rainfed crop
production activities, from cash earned from local agricultural or external (i.e. not

within the ward) wage labor and from the revenues generated by the ward wildlife

27

project. The rainfall is however, low and variable with evaporation exceeding
precipitation over most of the season. Crop yields are consequently poor.

Transporting inputs to Masoka and harvests to markets is difficult and expensive.

BIBLIOGRAPHY

Anderson, I. 1987. Communal land physical resource inventory, Guruve District.
Chemistry and Soil Research Institute, Harare, Zimbabwe.

Broderick, T.J. 1989. Annals of the Zimbabwe geological survey.

Cutshall, CR. 1989. A socio-economic baseline survey of community households.
Masoka / Kanyurira Ward. Centre for Applied Social Sciences, University of
ambabwe, Harare, Zmbabwe.

Department of Meteorological Services. 1981. Climate handbook of Zimbabwe.
Government of Zimbabwe, Harare, Zimbabwe.

Guyer, J.l. 1981. Household and community in African studies. African Studies
Review, 24, 87-137.

Hammel, EA. 1984. On the *** of studying household form and function. In:
Netting, R.McC., R.R. Wilk and E.J. Arnould (Eds.), Households. Comparative and
historical studies of the domestic group. University of California Press, Berkley,
California, USA.

Kanyurira Wildlife Committee. 1989. Wildlife land use plan Kanyurira Ward, Dande
Communal Area. WWF Multispecies Project, Harare, Zimbabwe.

Mackie, C. 1993. Aerial census of large herbivores in Western Dande Communal
Land (Kanyurira Ward). WWF Multispecies Project, Harare, Zimbabwe. MAPS
Project Paper In preparation.

Madhiri, L.W. and RC. Manyanza. 1989. Erosion hazard mapping of the SADCC
Region. Part I. Zimbabwe. Southern African Development Coordination Conference
(SADCC), Soil and water conservation and land utilization sector, Maseru, Lesotho.
Report No. 18.

Oesterlen, PM. 1989. The geology of the Dande west area (western Cabora Bassa
basin) - a preliminary report. Annals of the Zimbabwe Geological Survey, 12-20.

Owen, R.J.S. 1989. The use of shallow alluvial aquifers for small scale irrigation.

Overseas Development Agency (ODA), Harare, Zimbabwe. Consultant's report to
ODA project R4239.

28

29

Peterson, J.H. Jr. 1991. CAMPFIRE: A Zimbabwean approach to sustainable
development and community empowerment through wildlife utilization. Centre for
Applied Social Sciences, Harare, Zimbabwe.

Stocking, MA and HA Elwell. 1976. Rainfall erosivity over Rhodesia. Transactions
of the Institute of British Geographers, 1, 231-246.

Taylor, RD. 1993. Resource survey report on Masoka / Kanyurira. WWF
Multispecies Project, Harare, Zimbabwe. MAPS Project Paper in preparation.

Taylor, RD. 1991. Aerial census of large herbivores in pilot project areas. WWF
Multispecies Project, Harare, Zimbabwe. MAPS Project Paper No. 11.

Taylor, RD. and D.H.M. Cumming. 1989. Aerial census of large herbivores in pilot

project areas. WWF Multispecies Project, Harare, Zimbabwe. MAPS Project Paper
No. 4.

Thompson, JG. 1965. The soils of Rhodesia and their classification. Department
of Research and Specialist Services, Harare, Zimbabwe.

Thompson, J.G. and W.D. Purves. 1978. A guide to the soils of Rhodesia.
Rhodesian Agricultural Journal, Technical Handbook No. 3.

Chapter 3.

Towards a general understanding of the Masoka agroecosystem

INTRODUCTION

The ability to predict the behavior or future state of an agroecosystem, such
as Masoka, is constrained by the difficulty of dealing with complexity. Masoka is
socially, politically, economically and bio-physically complex. To predict the future
behavior or state of this agroecosystem the major components and behavioral
trends need to be identified from among the plethora of system components and
behaviors. This will: a) enable one to establish whether or not to proceed with the
analysis and b) facilitate a more focused and therefore, efficient analysis.
Preliminary investigations of the agroecosystem had four main objectives. These
were firstly, to obtain a general understanding of the agroecosystem. The second
objective was to identify the needs the agroecosystem is to satisfy. Identification
of major trends and patterns in the structure and functioning of the agroecosystem

was the third objective and the fourth was to identify important environmental6

 

The term 'environment' is used here in the systems theoretic sense, where environment
implies everything external to the system of interest

30

variables and their trends. It is important, at this general level of analysis, to keep
the investigation as broad as possible.

The purpose of this chapter is to describe the methods and results of the
coarse level analysis of the Masoka agroecosystem and to discuss the results with
regard to the satisfaction of household needs. The analysis is also aimed at
identifying issues that will focus the intermediate level analysis.

Viewed from outside the agroecosystem Masoka can be seen as one
element in a biophysical, social, economic and political matrix. The smallest unit
of resolution is the household whilst the extent of the analysis depends on the
criteria being evaluated; economically and politically it is the national system,
socially it is the area of Chief Chisunga and bio-physically it is a part of the Angwa
and Zambezi River catchments.

For the remainder of this introductory section I review pertinent literature to
provide an overview of the environment in which Masoka exists and changes.
Thereafter, I present the methods used in this level of analysis; describe and
discuss the results of the analysis; and end by discussing the implications of these
findings to the necessity for further analysis and to the ability of households in
Masoka to continue to satisfy their needs.

Zimbabwean communal area (CA) farmers manage their agroecosystems

to satisfy their food and cash needs7 (Shumba, 1989; Reh, 1986; FSRU, 1985).

 

The considerable debate on what constitutes 'basic needs' and the measurement thereof,
Is beyond the scope of this study. Streeten (1986) provides a useful discussion of the
Issues. Hopkins and Van Der Hoeven (1983) discuss the main concepts.

31

 

 

Soc
(30v
the:
and

deve

nee:
1989
“vest
1990
wfidfl
sYste
ennui
livesn
Cons;
inCIUC

mOrsi;

WOOdl
Vaiiey

othOU

32

Societal needs are reflected in the development objectives of the Zimbabwe
Government’s First Five-Year National Development Plan (GOZ, 1986); "...raising
the standards of living of the population, enlargement of employment opportunities
and manpower development and maintenance of a correct balance between
development and the environment."

The major components that households manage or use to satisfy these
needs are: 1) cropping systems (Stack and Chopak, 1990; Campbell and Swift.
1989; Jackson and Collier, 1988; Murindagomo, 1988; Stilz and Weyl, 1986); 2)
livestock systems (Cumming and Bond, 1991; Stack and Chopak, 1990; Ndlovu,
1990; Murindagomo, 1988; Scoones and Wilson, 1988; Stilz and Weyl, 1986); 3)
wildlife systems (Cumming, 1991; Cumming and Bond, 1991); 4) woodland
systems (Lynam et al., 1993; Swift et al., 1989) and 5) systems providing paid
employment (Jackson and Collier, 1988; Stilz and Weyl, 1986). Development of
livestock production systems in most of the mid-Zambezi Valley area has been
constrained by the disease trypanosomiasis that is fatal for many livestock species,
including cattle (Jordan, 1986). The vector of this disease, the tsetse fly Glossina
morsitans, is present throughout much of this area. The satisfaction of needs in the
Zambezi Valley therefore, largely depends on rainfed crop production, wildlife and
woodland use and the acceptance of paid employment off-farm. The European
Economic Community (EEC) initiative to eradicate tsetse fly from the mid-Zambezi
Valley area could, however, result in livestock becoming an important component

of household and community production systems.

 

 

sails
stud
area
(199
189C
yeah

throL

OffiCi

are c

33
There are few data on the extent to which Zambezi Valley households

satisfy their needs. A number of indicators from national statistics as well as from
studies in other areas of the country indicate, that for a large number of communal
area farmers, household needs for cash and food are not being satisfied. lllife
(1990) found famine to have been a regular phenomenon in Zimbabwe between
1890 and 1960. Rukuni and Wyekoff (1991) report that even in normal rainfall
years, about 40% of rural households do not produce sufficient food to last them
through the dry season. During periods of drought the government feeds as much
as 8.5% of the population (Rukuni et al., 1990). The Zimbabwe Central Statistics
Office (CSO, 1989) reports that thirty percent of Zimbabwe’s children under five
are chronically malnourished. Malnutrition is the highest cause of mortality in
children between the ages of one and four (Mason, 1990). Households in semi-arid
areas, such as Masoka, are among the most vulnerable groups (Jayne et al.,
1990; CSO, 1989).

Factors that are both internal and external to the agroecosystem can
increase the vulnerability of households to failure to meet their needs. Internal
factors are often related to high population density and subsequent inefficient or
poor use of resources (Campbell and Swift, 1989; Whitlow and Campbell, 1989;
Stilz and Weyl, 1986). External factors that Impact the ability of households to
satisfy needs include soil type, rainfall, technology, markets and social or political
stability. Most communal area soils are infertile sands (Grant, 1981) as is the case

with Kanyurira Ward where shallow, infertile sands cover most of the area

 

 

(An:

lfigh

advil
term
beca
deve

1991;

34
(Anderson, 1987). Rainfall in much of the mid-Zambezi Valley area is low and

highly variable (Anderson, 1987; Vincent and Thomas, 1965).

Drinkwater (1991) discusses how variable and inconsistent the extension
advice to communal area (CA) farmers has been since the early 19005. The long-
term political environment has also shown itself to be variable and uncertain
because, within 15 years, CA farmers have faced a violent war, Marxist
development policies and then market-oriented development policies (Drinkwater,
1991)

Many of these variables show increasingly unfavorable trends when viewed
from the perspective of the CA farmer. Land has become a scarce commodity in
Zimbabwe to the extent that "...it must be acknowledged that it is no longer
possible to honor every citizen’s claim to land rights." (World Bank, 1991). Elwell
(1992) suggests a downward trend in mean annual rainfall across the country.
Producer prices for maize and cotton have declined by 15 to 20% in the period
1985 to 1990 (Jansen and Rukovo, 1992). Government expenditure on agriculture
has declined from an average 9.8% of gross domestic product (GDP) in the early
seventies to 6.4% of GDP in the period 1986 to 1990 (Jansen and Rukovo, 1992).
Government expenditure on agricultural research and extension has also declined
since 1984 (Rukovo et al., 1991).

These facts create an image of an uncertain and variable bio-physical,
social, economic and political environment in which CA householders have had

and will continue to have difficulty satisfying their basic needs for cash and food.

35
This situation is likely to be exacerbated If declining government expenditure on

agricultural research limits the development and dissemination of technologies

appropriate to the conditions and needs of CA farmers in the Zambezi Valley.

METHODS

Secondary data of relevance to the study area were reviewed. Thereafter,
academics and professionals from several university departments, non-
governmental organizations (NGOs) and government departments, were brought
together for a one day workshop with the objective of developing a conceptual
model of a Zambezi Valley farming system. After a few paper presentations,
participants split up into small groups to develop conceptual models of
components of a Zambezi Valley farming system. In a final session, the whole
group attempted to bring these components together into a farming system model.

In a 1991 reconnaissance trip to Masoka, a local informant, Mr. Gift
Chisunga, arranged meetings with community leaders and individuals. I used semi-
structured interviews to identify components of the agroecosystem, constraints or
problems faced in attempting to satisfy household needs, and key Individuals or
social groups in the community. To obtain a general understanding of the layout
of the community, the distribution of soils, cropping practices, and land ownership
I mapped these factors along four transects through the area; from the Angwa

river to the southern most edge of the cultivated area. Transects were placed to

 

 

rn

36

cut through the major areas of habitation and crop production as well as through
areas that were more recently being opened for cultivation. Transect starting points
and routings were identified from air photographs. These routings were followed
as far as was convenient and appropriate. If farmers were encountered during this
exercise they were asked questions about their fields, their period of residency in
the area, how they had been allocated land, their yields, and their production
practices.

During this reconnaissance field trip it seemed that a number of households,
and especially the poorer households, could not produce enough to satisfy their
household needs from their crop production system. It seemed that access to
good quality soils, access to labor, and the mix of crops planted by households,
were key constraints to households being able to satisfy needs. A more detailed
survey was planned to examine these Ideas as well as to provide information that
would be used to focus the next level of the analysis.

A questionnaire was developed, and administered in September 1991, in
which respondents were asked to state which crops they planted and on what soil
type; what acreage they planted; and the yields they harvested for the 1990/91
growing season (i.e. the season for which harvesting had just ended). These latter
data provided estimates of actual yields. To obtain estimates of perceived yield (i.e.
the yields farmers believed could be obtained), respondents were presented with
scenarios of crop, soil type, hand or tractor tillage and good or poor rainfall

season, and were asked what yields they would expect to get under these

 

 

amdi
and h
needs
16 yea
had bi
dassm
(see Tl
state h
using 1
theycc
Drevlou
or pOOr
T
English,
e”Uf'l'ier;
daY- Tr;
qUeSUOn
finally, SL

(”:80 hr

Pr.

N

Thi
SLIP
bar

37

conditions. Respondents were also asked how many bags of maize and sorghum
and how many bales of cotton their household required to meet basic household
needs for one year. Data were collected on the number of adults and juveniles (<
16 years old) in the household and the number of years that the household head
had been resident in the community. Soils were classified according to local soil
classification practices as shapi (sandy loam), bepe (sandy clay) and jecha (sand)
(see Table 2.1, for FAO and USDA classifications). Respondents were asked to
state how many acres of the different soil types their household could manage
using traditional hand cultivation and how many acres of the different soil types
they could manage using a tractor. Finally respondents were asked to class the
previous three seasons (1988/89, 1989/90, 1990/91) as being either good, average
or poor rainfall seasons.

The questionnaire was translated into Shona and then retranslated into
English, by Independent translators, to ensure translation accuracy. Three
enumerators were employed from within the community and were trained for one
day. Training included questionnaire translation, detailed discussions of each
question, group and individual practice with administering the questionnaire and
finally, supervised field administration. I attempted to survey the entire community
(n=80 householdss) but sampled only 74 households.

Prior to analysis, data that did not conform to a normal distribution were

 

This is the number of households reputed to be in the community at the time of the
survey. It excludes a group 30 to 40 households of VaDoma who settled on the north
bank of the Angwa river at about this time.

 

transfc
transfo
analysi:

rules:

Selectlo
assistan‘
A
Configure
needs of
WhEIher (
and food.
LP model
iderltified l
clasSeS bl
in any One
The
h0u$ehold
and 80’9h
manage. A

38

transformed using natural logarithms (x=ln(x+1)) and, in one case, the square root
transformation. All data were converted to SI units. For the purposes of data
analysis, households were classified as small, medium or large using the following

rules:

if # adults + 0.5 * # children <- 4 then household - small
it # adults + 0.5 * # children > 4 and <= 7 then household = medium

If # adults + 0.5 * # children > 7 then household 8 large

Selection of these cutoff values was based on discussions with the three field
assistants.

A simple linear programming model was used to establish the optimal
configuration of crops grown on each of the three soil types that would meet the
needs of each household size class. The model was expected to roughly identify
whether or not households, in each size class, were likely to satisfy needs for cash
and food. If it were possible to satisfy needs with the available resources then the
LP model should find solutions that could achieve the objective. The solutions
identified by the LP model were not expected to be ideal solutions for households
classes but only indicators as to whether or not households could satisfy needs
in any one year.

The objective function of this model was the maximization of gross
household income subject to the household satisfying its needs for maize, cotton
and sorghum. Households were constrained in the number of hectares they could
manage. All data used in the model were derived from the questionnaire survey.

Two fifths of the cropping area available to each household was allocated as

 

shapi
thougl
croppl
in the
progra
years).
consid
(mean-
CrOp p
Zlmbat

maize.

39
shapi, two fifths as jecha and one fifth as bepe soils. These proportions were

thought to represent the probable distribution of the different soils currently
cropped in the community. To evaluate the ability of households to satisfy needs
in the long run, the optimal crop and soil mixes derived from the linear
programming model were used for each season from 1977/78 to 1991/92 (is. 15
years). Rainfall data from Angwa Bridge were used and the season was
considered good if rainfall exceeded the 15 year mean less 20% (Le. IF rainfall >
(mean-20%) THEN season = "good"). Otherwise, the season was considered bad.
Crop production that was surplus or deficit to household needs was converted to
Zimbabwe dollars (23) using 1990/91 season producer prices; 23264.40 ton’1 for
maize, 251660 ton'1 for cotton and Z$213.35 ton'1 for sorghum.

The potential of households to generate income was calculated using crop
producer prices for 1978 to 1991, converted to 1991 dollars. The 1978 to 1991
Angwa Bridge rainfall data were used to determine whether a season was good
or bad as described above. Yields were set as functions of soil type and season.
The land allocations derived from the linear programming models were used for
the optimal scenario. For the average scenario no cotton was planted. Of the
average areas that different households in each size class claimed they were able
to manage, one quarter was planted to sorghum and the rest to maize. Sorghum
was always planted on shapi soils.

The analyzed results of the survey were presented to the community using

graphic techniques: colored cardboard cutouts were made of three sizes of

hou:

CUIO

sorg

sque

tract

BVGI’E

resul

 

type
the b

Svaej

fairming
a cone
appropl
ambitioL
aCtUally
illustrate,

Mos; hotter

4o

household to represent large, medium and small household classes. A number of
cutouts were also made to represent bales of cotton, bags of maize and tins9 of
sorghum. Other cutouts of a tractor, a man and woman hoeing, different colored
squares and different sizes of cloud were used to represent the two forms of
traction discussed, different soil types and whether a rainfall season was good,
average or poor. These cards or icons were stuck on a large board to illustrate the
results of the survey. For example, yields on each soil type in each rainfall season
type could be discussed by pinning maize bag or cotton bale shaped cards on
the board to represent the average number of bags or bales reported in the

survey.

RESULTS AND DISCUSSION

The attempt during the workshop to develop a conceptual model of a
farming system was inconclusive, in the sense that participants failed to agree on
a conceptual model of the agroecosystem and were unable to agree on the
appropriate scales of analysis. Much of this uncertainty could be attributed to the
ambitiousness of the objective as well as lack of clarity in defining what was
actually needed. In another sense however, this result was conclusive in that it
illustrated that, as a group. the workshop participants did not have a sufficiently

clear understanding of a typical farming system to derive a model or to derive

 

9

Most households reported sorghum yields In twenty kilogram tins or buckets.

 

 

appro

1992 v
increas
model
(populi
numbe
I
In the 1
second,
Present.
village a
gross in.
CTOplanc
SeCtl'On.

needs ar

Th.

41

appropriate scales of analysis.

Estimated numbers of households in the community between 1985 and
1992 were provided by the village informant. These data show the average annual
increase in household numbers to be ten percent (Figure 3.1). A linear regression
model (population=-16291.4+8.23*year; df=6, r2=.94) and an exponential model
(population(t +1) = populationt*exp(°1)) were used to extrapolate household
numbers to the year 2010 (Figure 3.1).

Questionnaire survey results are presented and discussed in six sections.
In the first, household needs for food and cash crops are described. In the
second, the areas of land that households claimed to be able to manage are
presented. Production and production potential of the different soil types in the
village area are presented in the third section. The fourth section deals with the
gross income generating potential of households and in the fifth section, areas of
cropland that households require to satisfy their needs are presented. In the last
section, the implications of these results for satisfying household and societal

needs are discussed.

Household needs for food and cash crops

The requirements for maize (n=72; r2 = 0.915; p<.001), cotton (n=56; r2
= 0.87; p<.001) and sorghum (n=13; r2 = 0.907; p<.001) increased with family
size (Table 3.1). These values were derived by linear regression of kilograms of

cotton, maize or sorghum (dependent variable) against the total number of adult

LT
PC
ICU. _r\.£

cc

L aw F~EsJ7§

 

Figure

42

 

700

  
   
  

pop(I)=pop(l-1 )*exp(0-J-)—>

Population = 220

eooeeeomuomomoe

N O) -I> 0'1 0\
O O O O O
O O O O O

i i l l i

   

Number of households

\

pop=-16291.4+8.23‘yeor

Observe
r d“

 

 

 

0 I r T i i
1985 1990 1995 2000 2005 2010

Year

Figure 3.1 Estimated numbers of households in Masoka, 1985 to 1992 and
predicted numbers of households through 2010 based on a linear
and an exponential projection model.

 

Tame

Ic' : cnlzr

equw
Maze
vvere

ZS35€

EQUNa

43

Table 3.1 Household needs for maize, cotton and sorghum, by household size class. Data
are mean i SEM, with n in parentheses.

 

 

NEED MAIZE COTTON SORGHUM
“008900” Size 09) (1‘9) 0‘9)
Small 720 1: s (19) 1620 :32 (13) 170 .t 14 (4)
Medium 1270 1: 9 (24) 2840 i 36 (21) 230 1 40(3)
Large 1610 i. a (29) 3020 -_i_ 34 (22) 470 .t 42 (6)

 

equivalents (two children = one adult) in each household (independent variable).
Maize requirements were about 180 kg per adult equivalent, cash requirements
were about 23400 per adult equivalent for small and medium households and
Z$350 per adult equivalent for large households. Sorghum requirements per adult

equivalent were 45, 35 and 50 kg for small, medium and large households.

Acreage that households can manage

Respondents indicated that they would be able to manage greater areas it
they had tractors for ploughing than they could manage by hand, suggesting that
labor for land preparation is an important constraint. The mean areas that small
households claimed to be able to manage were 1.34 ha by hand and 2 he with a
tractor (t=6.26, df=18, p<.001). For medium households these means were 1.8
ha by hand and 3 he with a tractor (t=7.29, df=22, p<.001) and for large
households the means were 2.2 he by hand and 3.4 ha with a tractor (t=6.48,

df=24, p<.001).

 

Produc
5
one. The
Yields ii
season 2
for any
consider
and sorg
same se
Har‘lzi us
YIEIdS (m.
yields (m,-
et. a], (19'
and Com:
in Table 5
Fal
these that.
higher We
make defi,
7 (10%) o:

Production

Seventy two percent of farmers believed the 1990/91 season was a bad
one. The very low rainfall measured at Angwa Bridge (480mm) supports this belief.
Yields for maize, cotton and sorghum that were obtained by farmers for this
season are shown in Table 3.2. Yields were low but did not differ among soil types
for any of the three crops investigated. The cotton and sorghum yields were
considerably lower than the poor season (1983/84) yields for cotton (950kg ha’1)
and sorghum (450kg ha") but considerably higher than the maize yields of the
same season (180kg ha") reported by Harizi (1985) for the Muzarabani area.
Harizi used Agritex crop yield data to estimate these yields as well as average
yields (maize, 720kg ha", cotton, 1200kg ha", and sorghum 900kg ha") and best
yields (maize, 1800kg ha", cotton 1600kg ha'1 and sorghum, 1600kg ha"). Brunt
et. al. (1986) report yields for the mid-Zambezi Valley area of 700kg ha‘1 for maize
and cotton and 600kg ha‘1 for sorghum which are similar to the yields reported
in Table 3.2.

Farmers who ploughed with a tractor (n=14) had yields no greater than
those that used hand cultivation, nor did those who used fertilizer (n=5) achieve
higher yields than those who did not. The sample size is, however, too small to
make definitive conclusions about the effects of tractors or fertilizer on yields. Only
7 (10%) of the respondents had used credit facilities in the last growing season
whilst 13 (18%) intended doing so in the forthcoming growing season.

In general the yields that farmers stated they were likely to get on a given

 

In

LII

hc

45

Table 3.2 Maize, cotton and sorghum yields obtained by respondents for the 1990/91
season on three soil types. Data are mean _-I; SEM, with n in parentheses.

 

 

MAIZE COTTON SORGHUM
SOIL <--------—I(g ha"--—-----—->
SHAPI 670 _-'I_-_ .18 (28) 673 i .19 (12) No data
BEPE 540 .t .34 (11) 476 i .29 (4) 490(1)
JECHA 482 i .19 (24) 600 i .12 (26) 198 :_I-_ .64 (6)
Mean 570 i .12 (63) 600 i .1 (41) 225 i .55 (7)

 

soil with a bad season (T able 3.3) were slightly higher than the yields they actually
achieved (Table 3.2). Acceptance of the anticipated yield figures presented in
Table 3.3 assumes; a) that farmers have had some experience in growing the crop
and b) that they can predict the yields they are likely to obtain under a given set
of conditions. To satisfy the first assumption respondents were asked only to
respond to creps they had grown. As for their ability to predict yields under given
conditions, a fair assumption to make is that farmers, whose families depend on
their ability to produce sufficient food and cash, would be able to estimate
potential yields and plant sufficient areas to meet their needs. One would in fact
expect these estimates to be slightly conservative because mistakes could have
serious implications for the well being of the farmer’s household. Comparisons of
the yields in Tables 3.2 and 3.3 suggest that farmers tended to overestimate their
poor season maize and cotton yields, on bepe soils in particular, and
underestimate their sorghum yields. The anticipated yields presented in Table 3.3,

however, compare favorably with the yield figures reported by Harizi (1985). The

46

 

 

 

 

 

Table 3.3 Anticipated yields for hand cultivated maize, cotton and sorghum on three soils
in good and bad rainfall seasons. Data are mean 1 SEM, with n in parentheses.
SOIL SHAPI BEPE JECHA AVERAGE
CROP _1
MAIZE
Good 1650.b 1 20 23306“ + 18 1610“ + 24 1870 +12
(52) (56) (39) (147)
Poor 760," + 22 1080,11 22 760.cf1 24 870 1 13
(52) (56) (39) (147)
COTTON
Good 1780h 1 42 2510ij 1 62 1970h 1 50 2050 1 31
(37) (28) (16) (81)
Poor 900 147 1150156 6105162 910132
(33) (27) (16) (81)
SORGHUM
Good 1260m1148 8101113 470 158 860111
(8) (6) (5) (19)
Poor 5401128 160n143 170158 300156
(8) (5) (5) (13)
1. Means followed by the same letter are significantly different at a = 0.05.

major differences are that Masoka farmers anticipated much higher poor season
maize and good season cotton yields whereas Harizi reported much higher good
season sorghum yields. In general, the anticipated yields appear to be adequate
estimates for use in the current analysis.

Respondents clearly perceived that yields, for all crops, on tractor cultivated
fields would be higher than those on hand cultivated fields (Table 3.4; Cotton,
n=324, df=323, p<.001; maize, n=585, df=585, p<.001; sorghum, n=72, df=71,
p<.001). Since few households had actually used tractors in production of these

crops, these yield data are to be treated with some caution.

47

 

 

 

 

Table 3.4 Perceived yields for tractor cultivated maize, cotton and sorghum on three soils
in good and bad rainfall seasons. Data are mean1 1 SEM, with n In parentheses.
SOIL SHAPI BEPE JECHA AVERAGE
once .1
Season < "9 ha >
MAIZE
Good 3090_b122 4620cw119 25403125 3430113
(52) (55) (39) (147)
Poor 1390“ 123 2320f 121 1330”_+_ 23 1690114
(51) (55) (33) (144)
COTTON
Good 2930,155 44309,,164 3230132 3452137
(37) (23) (15) (81)
Poor 1380111 1 51 1860g167 1330174 1521135
(37) (28) (15) (81)
SORGHUM
Good 1130,,1124 17601133 920,1 55 12361 60
(3) (5) (5) (19)
Poor 10201164 9701133 470k 1 36 3331 74
(7) (5) (4) (16)

 

1. Means followed by the same letter are significantly different at a = 0.05.

Income generating potential

The gross annual incomes that households could earn from the production
of rainfed, hand cultivated maize, cotton and sorghum were 25430 (CV = 46%)
under the average scenario and 231400 (CV = 52%) under the optimal scenario
for large households. For small households these values were Z$260 (CV = 46%)
under the average and 231015 (CV = 52%) under the optimal scenario. The mean
values for each year 1978 to 1991 are shown for large households (Figure 3.2)
and for small households (Figure 3.3). These data highlight a) how far from being

able to satisfy their needs most households are likely to be; and b) the potential

 

impel
needs
2.2) ii
prodL

Some
Cr0pl

were c
yields

QVOWn

indicate
none 0-
their be

”Ct at a
DGeds.‘
e”°U9h
manage

Medium

 

48

importance of wildlife revenues (Table 2.2) in assisting households to satisfy their
needs. Average annual household income from wildlife dividends (23150 pa, Table
2.2) could be equivalent to between 15 and 88% of incomes derived from crop
production for small households and between 11 to 35% for large households.

Some implications of these earning potentials are discussed later in this chapter.

Cropland areas required by households to meet needs

The areas required by each household class to satisfy their perceived needs
were calculated by dividing the household needs for food and cash crops by the
yields achieved for that crop on any of the three soil types where the crop was
grown (T able 3.5).

Linear programming analyses based on the acreage that households
indicated that they could manage and the yield figures from Table 3.3, indicate that
none of the household size classes could produce anywhere near enough to meet
their basic needs in bad rainfall seasons. To do so, households of all sizes would
have to increase their yields or acreage by well over 100% (Table 3.5).

These figures are daunting to say the least. Even in good rainfall years it is
not at all clear that production from crops is sufficient to meet basic household
needs. Using the yield figures of Table 3.3 only small households could produce
enough to meet their basic needs given the land areas they claimed they could
manage by hand (and the allocation of soil types assumed in this analysis).

Medium and large households would need to increase the areas cultivated in good

vii

if”

I
V

Figure

49

 

 

 

 

A 5000
8 4500 -
4000‘

3 .

c 3500

E 3000 - f

E 2500 - \

U 2000 ' - ‘

g 1500 - ,2. / /

3 1000 A . .. a

3 500 z : 2 fl. '-' 1‘ .: a i a

Z 0 .1: = i i = 1 i i
1978 1982 1986 1990

1980 1984 1988 1992
Year

+Needs fi-nverage stratequ -+- Optical strategu

*Dptteal + midlife +flveraqe + uxldltie

Figure 3.2 Household needs (23) and earning potential for large households
over the period 1978 to 1992.

50

 

 

 

 

 

A 2500
GB
'0’
2000 r /
In
0‘
C
H 1500 1 '
E f‘
8
U 1000; / '1’
C “I
(U “in"
g 500 ‘W I
3
Z 0 c . . . . . .
1978 1982 1986 1990
1980 1984 1988 1992
Year
+ Needs -8- Average strategu +0ptinal strateqtj

-+- 0pt1eal+u11d11ie -)(— fiveraqe+u11dltfe

Figure 3.3 Household needs (23) and earning potential for small households
over the period 1978 to 1992.

 

Tablt

years I
sensni
areas

housel

 

rainfall

 

the ma
Size Cla
with no.
rairlfall l

DroducE

hOUSth

 

51

Table 3.5 Cropland areas (ha) required to satisfy household needs‘. Data are total area for
household size class with the area per adult equivalent for each size class in

 

 

parentheses.
HOUSEHOLD SMALL MEDIUM LARGE
SIZE
Manageable (ha) 1.3 (.33) 1.3 (.26) 22 (.24)
Good season (ha) 1.3 (.33) 2.3 (.33) 2.9 (.32)
Change 0% +30% +30%
Poor season (ha) 3.0 (.75) 5.2 (.74) 6.2 (.70)
Change +125% +190% «113093
Weighted 2.0 (.5) 3.5 (.5) 3.8 (.4)
Total (ha)

 

1. The total areas required during good and bad seasons

are obtained by weighing the areas required In those

seasons by the proportion of good (0.6) and bad (0.4)

seasons at Angwa Bridge.
years by up to 30% (Table 3.5). Solutions from the linear programming model were
sensitive to reductions in areas of soil types and the price of maize. Using the
areas that households reported as being manageable with tractor tillage, all
household classes could produce enough to meet their basic needs in good
rainfall seasons, but still not In poor rainfall seasons.

If we accept that the yields of Table 3.3 are conservative and add 40% to
the maize and sorghum yields and 50% to the cotton yields, then all household
size classes are theoretically capable of meeting their needs in good rainfall years,
with notable cotton surpluses. If, however, we take the long-term view and use the
rainfall data from Angwa Bridge for the past 15 years, and the 1990/91 season

producer prices for C grade crops, a different picture emerges. Large and medium

households could show gross deficits of between 234000 and 2$6000 over the 15

52
year period whilst small households could show gross surpluses of 2510000.

These figures assume everything else being held constant and optimal allocation
of land to crops. The surpluses of the small household are attributable to their

lower grain needs and therefore, the relatively larger areas they are able to plant

tO COttOI'I.

Implications

When presented to members of the community there was general
agreement with the results of this analysis. In seeking an answer to the question,
'Why do households not grow more cotton and use the higher returns to purchase
their maize and sorghum needs?" community members responded that they
wished to ensure their maize supplies - their staple food crop - and could not rely
on their ability to purchase and get transport for maize supplies from Guruve. In
the 1992/93 season these farmers were proved correct; the drought resulted in
national maize shortages and households were unable to purchase sufficient
amounts to meet their needs.

A number of patterns are evident from the coarse level analysis. First, large
and medium sized households are unlikely to be able to satisfy their basic needs
for food and cash over the long term. They are constrained by their need to
produce sufficient grain to meet household food needs. Second, households that
concentrate their productive activities on cotton production appear more likely to

be able to satisfy household food and cash needs than those who attempt to

 

Iev

53
satisfy food needs from local grain production. Third, the ability of households to

satisfy needs is partially attributable to the areas of each soil type to which they
have access. Fourth, households would need to double the average annual wildlife
dividends they received between 1989 and 1992 (is. 23400 pa) to offset the
deficits the large and medium sized households could be accruing.

One way for households to improve their overall yields in bad seasons is
to increase the areas they cultivate so that they always plant sufficient area to
satisfy household needs even in a bad season. The recent calls for a tractor are
clear indicators that, in the eyes of many households, this is perhaps a more viable
option than trying to improve yields by 130% or more. If this is a route that the
community chooses, the choice has important implications for the carrying
capacity of the arable lands in Masoka and needs to be carefully evaluated. The
fine soil filth and larger field sizes that result from tractor cultivation practices could
increase the erosion risks on the erodible soils of Masoka. The jump from hand
cultivated technologies to tractor technologies is not always advisable (Pingali et
al., 1987). The poor fence management performance of the Masoka community
is illustrative of the need for caution in introducing relatively sophisticated
technologies.

l have not dealt with the need to allow fields to go to fallow. Discussions
with farmers indicate that fertility declines significantly after three to five years of
continuous cropping. With the high risks of crop failure due to low rainfall, low

levels of fertilizer seems the rational choice. This means however, that the

54

community needs to plan for twice as much land for each household (assuming
five years of continuous cropping and five years of fallow).

Based on the hand cultivation figures households claimed to be able to
manage and assuming (i) technology remains constant at current levels; (ii) the
proportions of large, medium and small households within the community remain
constant; (iii) two fifths of the land area is in each of shapi and jecha soil types and
one fifth is bepe and (iv) all land within the fenced area is arable, then the number
of households supportable within the fenced area is about 20. Given current
household population growth rates the community could reach this number of
households within six years (Figures 3.1, 3.4).

Rapid population growth is a rational response where large families provide
labor in a labor constrained environment. Population increases, the need to
expand cropped areas to satisfy needs and the tendency of labor constrained
households to use existing fields rather than to open new fields (as well as
looming land shortages) imply the biophysical resources of Masoka will come
under increasing pressure. Wildlife revenues, by buffering households from the
variability in needs satisfaction that is a result of the climate, could also undermine
the resilience of the agroecosystem by capitalizing the expansion of cropland

areas without improving the underlying technology.

 

 

 

 

 

 

 

4500 , 1500
A 4000-1 1000
‘0 A
5 3500- 500 r0
5
E 3000 - - --------0 O
E, c
g 2500‘ "500 E
a: ' H
L m
3 2000‘ “1000 E
U 8
0: 1500‘ -1500
L "U
U 1000- --2000 g
C _.l
'0
—I 500- ;-2500
0 r i 1 ' I l T -3000
1985 1990 1995 2000 2005 20 10
’ Year
+ Linear need -8- Linear remain + Exp need -+- Exp remair

Figure 3.4 Arable land required and remaining under the population growth

scenarios presented in Figure 3.1 and the land requirements
presented in Table 3.5.

56
CONCLUSIONS

The majority of households in the community are unlikely to be able to
produce enough to meet their basic needs from their crop production system,
using current technology. Low and unreliable rainfall, a shortage of labor or
traction, and inadequate information on optimal cropping patterns and technology
appear major constraints to this agroecosystem being able to satisfy household
needs. Smaller households and households with greater access to resources have
better chances of satisfying their needs. The use of a tractor to improve tillage and
the areas households can manage is one possible solution, but one that needs to
be evaluated with caution. The calls from the community for a tractor have
important implications both for assisting households to stabilize their needs
satisfaction and also for increasing the erosion hazard in Masoka. These
implications need to be more fully investigated. Average wildlife dividend payments
would have to be more than doubled to ensure continuous needs satisfaction, for
large and medium households, under the conditions presented in this analysis.
Declining national (and global) cotton prices as well as the failure of the GMB to
provide adequate food in drought years suggest the common practice of
managing a portfolio of productive activities in Masoka is well advised.

The fact that households have been living in the area for several generations
suggests that the agroecosystem is more resilient than this coarse level analysis
would have us believe. We need therefore, to investigate more completely the

productive practices of households in different resource access classes.

BIBLIOGRAPHY

Brunt, MA, V.J. Clarke and C. de Greling. 1986. Tsetse area development and
land use planning in the Zambezi Valley, Zimbabwe. Food and Agriculture
Organization of the United Nations, Rome.

Campbell, 3M. and M.J. Swift. 1989. Links between the environment and basic
needs satisfaction. In: Campbell, B.M., R.F. Du toit and CAM. Attwell (Eds), The
Save study. Relationships between the environment and basic needs satisfaction
in the Save catchment, Zimbabwe. University of Zimbabwe, Harare, Embabwe. 70-
83.

Central Statistics Office. 1989. fimbabwe demographic and health survey 1988.
Ministry of Finance, Economic Planning and Development, Harare, Zimbabwe.

Cumming, D.H.M. 1991. Multispecies systems and rural development in southern
Africa: opportunities, constraints and challenges. World Wide Fund for Nature,
Multispecies Animal Production Systems Project, Harare, Zimbabwe. Project Paper
No. 19.

Cumming, D.H.M. and l. Bond. 1991. Animal production in southern Africa: Present
practice and opportunities for peasant farmers in arid lands. World Vlfide Fund for
Nature, Multispecies Animal Production Systems Project, Harare, Embabwe.
Project Paper No. 22.

Department of Meteorological Services. 1981. Climate handbook of Zmbabwe.
Government of Zimbabwe, Harare, Zimbabwe.

Drinkwater, M. 1991. The state and agrarian change in Zimbabwe’s communal
areas. St. Martin's Press, New York.

Elwell, HA 1992. Cropland management options for the future. Proceedings of the
third annual scientific conference of the SADCC-Land & Water Management
Research Programme. Harare, Zimbabwe, 57 October, 1992. In press.

FSRU. 1985. Annual Report of the Farming Systems Research Unit, Department
of Research and Specialist Services. Government Printers, Harare, Zimbabwe.

Government of Zmbabwe. 1986. First five-year national development plan
1986-1990. Volume 1. Government of Zimbabwe, Harare, Zimbabwe.

57

58

Grant, PM. 1981. The fertilization of sandy soils in peasant agriculture. Zimbabwe
Agricultural Journal, 78, 169-175.

Harizi, KEI. 1985. Zimbabwe. Mid-Zambezi Valley agricultural development project.
Agricultural production. FAO, Rome. FAO/ADB Working paper No.2.

Hopkins, M. and R. VanDerHoeven. 1983. Basic needs In development planning.
Gower Publishing Company Limited, Aldershot, England.

llifie, J. 1990. Famine in Zimbabwe 1890-1960. Mambo Press, Gweru, Zimbabwe.

Jackson, J.C., and P. Collier. 1988. Incomes, poverty and food security in the
communal lands of Zimbabwe. RUP Occasional Paper, Department of Rural and
Urban Planning, University of Zimbabwe, Harare, Zimbabwe.

Jansen, DJ, and A. Rukovo. 1992. Agriculture and the policy environment: Zambia
and Zimbabwe. Political dreams and policy nightmares. Study prepared for OECD
Development Centre.

Jayne, T.S., M. Chisvo, B. Hedden-Dunkhorst, and S. Chigume. 1990. Unravelling
Zimbabwe’s food insecurity paradox: Implications for grain marketing reform.
Proceedings of the First National Consultative Workshop on Integrating Food,
Nutrition and Agricultural Policy. Juliasdale, Zimbabwe, 15-18 July, 1990.

Jordan, AM. 1986. Trypanosomiasis control and African rural development.
Longman, London, England.

Lynam, T.J.P., S.J. Venneulen and BM. Campbell. 1993. Contingent valuation of
multipurpose tree resources in the smallholder farming sector, Zimbabwe.
Accepted for publication in the Journal for Farming Systems Research-Extension.

Mascarenhas, J.P., Shah, S. Joseph, R. Jayakaran, J. Devararam, V.
Ramachandran, A. Fernandez, R. Chambers, and J. Pretty. 1991. Participatory rural
appraisal. Proceedings of the February 1991 Bangalore PRA Trainers Workshop.

RRA Notes, No. 13. International Institute for Environment and Development (IIED),
London, England.

Mason, E. 1990. The nutrition situation. Current strategies and plans.
Consequences of malnutrition. Proceedings of the First National Consultative
Workshop on Integrating Food, Nutrition and Agricultural Policy. Juliasdale,
Zimbabwe, 15-18 July, 1990.

Murindagomo, F. 1988. Preliminary investigations into wildlife utilization and land
use in Angwa, Mid-Zambezi Valley, Zimbabwe. University of Zimbabwe, Harare,

 

an

Ndb
sysié

lnSi
Rn.
fie

Pres

Reh,
Syste

 

Zhnb
Ruku
SAIM
Secu
RUkc
agnc

Pfimd
900k
CONN

Shur
Farnl
Stag

Fhsn
“syn.
30b:
135991
Dow

Zimbabwe.

Ndlovu, L 1990. The role of ruminants in promoting food security in farming
systems in the SADCC region. Proceedings of the conference on Food Security
in Southern Africa. Harare, Zimbabwe, 16-18 October, 1989.

Pingali, P.L, Y. Bigot and HP. Binswanger. 1987. Agricultural mechanization and
the evolution of farming systems in sub-Saharan Africa. John Hopkins University
Press, Baltimore, U.S.A.

Reh, l. 1986. Farming systems research program and progress. Mimeo, Farming
Systems Research Unit, Department of Research and Specialist Services, Ministry
of Agriculture, Hmbabwe Government, Harare, Zimbabwe.

Rukuni, M., T.S. Jayne, J. Tagwireyi, and L Rugube. 1990. Alleviating hunger in
Zimbabwe: Towards a national food security strategy. World Bank, Washington
DC, USA. Issues paper for the World Bank Agriculture Sector study for
Zimbabwe.

Rukuni, M. and J.B. Wyekoff, (Eds.). 1991. Market reforms, research policies and
SADCC food security. University of Zimbabwe/Michigan State University Food
Security Research in Southern Africa Programme.

Rukovo, A.T. Takavarasha, R. Thiele and M. Wlebelt. 1991. The profile of
agricultural protection in Zmbabwe. Kiel Working paper No. 457.

Scoones, l., and K. Vlfilson. 1988. Socio-economic dimensions of livestock
production in the communal lands of Zimbabwe. Households, lineage groups and
ecological dynamics: Issues for livestock research and development in Zimbabwe’s
communal lands. Harare.

Shumba, E. 1989. Maize technology research in Mangwende, a high potential
communal area environment in Zimbabwe. Part 1: Developing a research agenda.
Farming Systems Newsletter, 34, 12-34.

Stack, J.L, and C.J. Chopak. 1990. Household income patterns in Zimbabwe’s
communal areas: Empirical evidence from five survey areas. Proceedings of the
First National Consultative Workshop on Integrating Food, Nutrition and
Agricultural Policy. Juliasdale, Zimbabwe, 15-18 July, 1990.

Stilz, D., and U. Weyl. 1986. A model of an ecology-oriented agricultural
development programme and its transformation into a coordinated attack on rural
poverty. University of Osnabruck, Osnabruck, Germany.

60

Streeten, P. 1986. Basic needs: Some unsettled questions. The challenge of
employment and basic needs in Africa. Oxford University Press, Nairobi, Kenya.

Swift, M.J., P.G.H. Frost, B.M. Campbell, J.C. Hatton, and KB. Wilson. 1989.
Nitrogen cycling in farming systems derived from savanna: Perspectives and
challenges. In: M. Clarholm and L Bergstrom (Eds.), Ecology of arable land.
Kluwer Academic Publishers, Netherlands.

Whitlow, R. and BM. Campbell. 1989. Factors influencing erosion in Zimbabwe:
a statistical analysis. Journal of Environmental Management, 29, 17-29.

World Bank. 1991. Zimbabwe agriculture sector memorandum. World Bank,
Washington DC, USA. No. 9429-Zim.

 

To SL
SYSIel
IS th
Bawd

part c
mOtiva
c0W3:

exam p

 

 

CUTTSn
minim);
plans C

failed S

 

Chapter 4.

Needs analysis, system Identification and problem

formulation In the Masoka agroecosystem

INTRODUCTION
To successfully improve a system’s performance we need to first identify the
system and second define what constitutes an improvement. What is not as clear
is whose perspective we are to adopt in making these decisions (Bawden, 1992;
Bawden and lson, 1992). A large number of actors (individuals and organizations),
with an equally large set of motivations, have varying degrees of influence on a
part or, the whole of, the Masoka agroecosystem. These perspectives and
motivations carry with them value systems that the actors use to weight the inputs,
components or outputs of the agroecosystem: the National Government, for
example, might emphasize cotton production activities to maximize its foreign
currency earnings. Whereas individuals might place greater weight on activities that
minimize the risk of their household going without food that year. Development
plans or projects that use values other than those of the target community have

failed so often the credibility of this “top-down" development model has been

61

 

ur
fin
re:
50!
efft
inc

bee

ana:
DBTS
knov
expe
Work
Chang
diner,
ahern,
persp,
Zimba
Camp
1939; E
1985. :

ThOnRD

 

62
undermined (Bawden and lson, 1992; Chambers 1983; Redclift, 1987). These

failures often arise from inadequate understanding of household needs, household
resources, household objectives and the constraints that households face and
suggest that, for the purposes of this analysis, we stand a greater chance of being
effective if we adapt the perspective of Masoka households. The value of
incorporating local knowledge and perspectives into development planning has
been receiving increasing attention (Brokensha et al., 1980; Warren, 1991).
Whilst using local knowledge may be a necessary condition for successful
analysis of the Masoka agro-ecosystem, assuming that adopting a household
perspective would be a sufficient condition for a successful analysis is naive. Local
knowledge is often restricted to what farmers can see and what is within their
experience (Richards, 1980) and may be unevenly distributed in a community (IDS
Workshop, 1983). If either the agroecosystem or its environment are likely to
change faster than household adaptive capabilities, or in ways that are radically
different from local experience or perception, then we are compelled to incorporate
alternative perspectives into our analysis. The obvious source for these alternative
perspectives is the large body of empirical knowledge that has been built up in
Zimbabwe on many aspects of household resource use (Campbell et al., 1989;
Campbell and Du Toit, 1988), crop production practices (FSRU, 1985; Reh et al.,
1989; Shumba, 1984a, 1985, 1986a,b), crop yields (Akwabi-Ameyaw, 1990; FSRU,
1985; Shumba, 1984b), soils (Anderson, 1986; Grant, 1981; Thompson, 1965;

Thompson and Purves, 1978), household labor use (Reynolds, 1991), household

63
economics (FSRU, 1985; Shumba, 1985; Stanning, 1987a,b; Chopak, 1991;

Mudimu et al., 1988) and soil erosion (Elwell, 1980, 1985; Elwell and Stocking,
1984, 1988). Much of this information Is, unfortunately, inadequate to develop the
predictive capabilities we require In our analysis. Few of these authors provide the
kind of data (statistics of population distributions) needed for stochastic simulation.
With the exception of the detailed study on child labor in the Zambezi valley by
Reynolds (1991) and Elwell’s erosion model (Elwell, 1980), little of this information
is directly relevant to the mid-Zambezi Valley.

A further problem with using indigenous knowledge as the basis for an
extensive analysis is the difficulty in evaluating the degree to which the information
presented by local informants is an unbiased representation of the populations of
interest. With few exceptions (Farrington and Martin, 1987; Gil, 1991; Swift, 1979)
this issue has received little attention in the literature on indigenous knowledge or
RRA. The literature on potential biases in human responses to questions is large
and has been examined from several perspectives: Aler and Settle (1985)
discuss sources of bias from a survey perspective and Tversky and Kahneman
(1974, 1981) discuss the subject from a psychological perspective. Adelman, in
describing the problems of biases in expert knowledge suggests, “At the very
least, knowledge engineers and evaluators should be aware of their existence and
look out for instances where elicited judgements may reflect a bias," (Adleman,
1992) which is hardly a rigorous strategy for establishing the accuracy and

precision of a set of knowledgel Of particular relevance to this study is the

64

extensive treatment of bias and measurement validity by Mitchell and Carson
(1989).

There are two issues of concern here. The first is knowing whether one has
measured that which was set out to be measured. Mitchell and Carson, (1989.
p190) call this the validity of a measure and identify three indicators of a valid
measure - content, criterion related, and construct. Content validity concerns
establishing whether the method used to measure the parameter of interest
constrains or biases the elicited response. A measurement is valid if it does not
bias or constrain the response. Criterion validity concerns the degree to which
other measures of the population of interest can be used as criteria to establish
the validity of the elicited measure. The ideal measure here would be actual
measures of the population under consideration. For example, measures of actual
maize yields would be an ideal criterion to establish the validity of the yields
farmers say they obtain. Direct measures are unlikely to be available (othenivise we
would not be using an indirect method) but other measures that are “unequivocally
closer to the theoretical construct than the measure whose validity is being
assessed" (Mitchell and Carson, 1989, p192) are necessary to assess this type of
validity. Construct validity concerns the degree to which other measures of the
same population either converge towards the same result or are consistent with
what we would predict from theory.

A second issue of concern is the accuracy and precision of the

measurements one makes. In evaluating sources of bias in respondent responses,

65
Mitchell and Carson (1989) note four major sources of systematic bias. The first

is the use of scenarios that contain strong incentives for respondents to
misrepresent their true beliefs either because they think they can influence some
result of the study or because they wish to comply with the interviewer’s beliefs to
please either the interviewer or the study sponsor. The second source of bias
arises when respondents, unsure of their responses, rely on the information
relayed to them through study questions to formulate their responses. The third
and fourth of these sources of systematic bias involves the incorrect specification
of the scenario about which the interviewer seeks information. This may occur
either because the scenario is incorrectly specified from a theoretical perspective
or the scenario is correctly specified but communicated in such a way that the
respondent does not perceive it the way the researcher intended.

To my knowledge no predictive analyses of community level production with
differentiated households has been undertaken in Zimbabwe. There can be no
doubt that analyses at the scale of the household are difficult: Households are
structurally heterogenous (Cutshall, 1989); they have different labor, land and
capital resource bases (Coudere and Marijse, 1991; Reynolds, 1991; Shumba,
1984b, 1985); they exhibit different production objectives (Stanning, 1987) and they
engage in a diverse set of productive activities (Jackson and Collier, 1991), not all
of which are agriculturally based (Helmsing, 1991). The importance of inter-
household sharing, hire and employment (Shumba, 1985; Stanning, 1987; Znyama

et al., 1987) as well as the major role of women and children in household

66
production (Adams, 1991; Mehretu and Mutambirwa, 1992; Reynolds, 1991)

exacerbate the difficulty of analyzing agroecosystem performance from a
household perspective.

From this daunting complexity we must construct a model that we believe
incorporates the essential elements (properties) of the agroecosystem and Is
capable of accurately simulating agroecosystem functions and of predicting future
states and outputs of the agroecosystem. We quite obviously have to simplify the
system considerably. In this simplification process we are faced with two major
methodological problems. First, how do we identify major actors, their needs and
key agroecosystem components? Second, how do we collect the data we require
to develop and test our predictive model? These problems are particularly difficult
firstly because, Masoka household decision makers are, for the most part, illiterate
and inumerate and therefore keep no records of production inputs, areas or
outputs. Second, time and money constraints allow us only a single cropping
season for data collection.

There is little in the scientific literature to guide our search for appropriate
methods. Most household analyses rely on household questionnaire surveys
(Campbell et al., 1989; Jackson and Collier, 1991; MLARR, 1989; Stanning, 1987),
which are questionably useful for the rapid and flexible learning recording required
when no a priori model exists to guide data collection. The long term immersion
approach of anthropologists often lacks the quantitative information required to

predict changes and to compare alternative development or production strategies

67
over long time periods (Hasler, 1992; Reynolds, 1991). The more informal open-

ended interview techniques commonly used by social anthropologists (Derman,
1990, 1992) yield qualitative information on many aspects of a production system
but lack temporal breadth and are not likely to yield the general functions required
for predictive simulation analyses. Recent developments in techniques classed as
rapid rural appraisal (RRA) or participatory rural appraisal (PRA) show great
promise in terms of their flexibility and the range of both quantitative as well as
qualitative data they are able to record (Khon Kaen University, 1987; Mascarenhas
et al., 1991). The depth and breadth of data that are required for predictive
simulation analyses require, however, far more than the single short visit with a
multi-disciplinary team of investigators, that is standard practice in RRA. Where
RRA techniques have been used in Zimbabwe (Carter of al., 1993), there has been
little attempt on the part of the researchers to test the validity of the methods used
or the accuracy and precision of the information collected.

The nature of the analysis required to develop a predictive model of the
Masoka agroecosystem. from a household perspective, demands some hybrid of
the in-depth, qualitative analysis of anthropologists, with the rapid and multi-
disciplinary approach of RRA as well as the more formal methods of agricultural
research. For the most part appropriate data collection methods have not been
developed.

In this chapter, the field data collection methods developed for the analysis

of the Masoka agroecosystem are described. The results of applying these

 

to:

agi
ecc
to e
is 0

asa

anal
nun
reqm

apprc

0f vill'

I

Ih80(

Selec

Wore

i”form

 

68

methods to the Masoka agroecosystem are described and discussed with regard
to: 3) identifying major actors and client groups influencing agroecosystem
behavior; b) establishing the needs and objectives client groups expect the
agroecosystem to satisfy; 0) Identifying the major components of the agro-
ecosystem and their relationships and; d) Identifying the criteria client groups use
to evaluate the performance of the agroecosystem. The focus of the data collection
is on developing the ability to predict the state or behavior of the agroecosystem

as a function of controls that Masoka households can impose on their resources.

METHODS
Having a multi-disciplinary team carry out the Masoka agro-ecosystem
analysis for the period required by this analysis was not possible. Expecting large
numbers of Masoka residents to give as much of their time as this analysis
required was also not feasible. I decided therefore, that the most appropriate
approach to knowledge collection would be to have the community elect a group
of villagers to represent the community perspective. This group provided most of

the community level knowledge presented in this chapter.

Selection of community representatives
At an open meeting held on February 2, 1992, Masoka community members
were informed of the objectives of my research. Community members were

informed that I would need to work intensively with a small group of people from

69

the community. No reward or bench fees for the members of the group with whom
I would work were mentioned. The community members were explicitly asked if
they wanted me to proceed with the analysis. Upon receiving the community’s
approval, people at the meeting were asked to identify classes of household in
Masoka and to proportionately weight these classes according to the number of
households in each class. Community members at the meeting were asked to
elect representatives from each of these classes to form the team of village or
community representatives (VRs) who would work with me. The community was
specifically requested to include women in this elected group but this request was
refused on the grounds that the men in the community did not wish their wives or
daughters to interact with unknown men. Six of the men present at the meeting
were elected. All except one were young and either recently married or unmarried.

At the first meeting with the VHS, n early March 1992, I was informed that
the eldest VR had withdrawn. The village informant who assisted me during earlier
survey work (described In Chapter 3) was elected by the remaining VHS to replace
the missing man. The ward councillor was approached to have women included
in the VR team. He subsequently sent three women who had agreed to join the VR
team. One of these women was the councillor’s wife, one a widow and one the
wife of a locally respected farmer. The women VRs were aged between 38 and 58,
had between three and eight children each, and none of them had been to school.
The male VRs were younger, between 22 and 26 years old and had between nine

and 11 years of schooling each. Two of the men had one child each, the others

70

had none. Each of the men spoke at least some English but none of the women
did. All except two of the VHS were born in Masoka and had lived there for most
of their lives.

In discussing the work that we were to do together, I broadly explained my
objectives and the methods I would attempt to use. I volunteered to pay the VRs
and to provide food for the group while we worked. A daily wage and a meal
schedule were agreed upon. In this, as in all such discussions, my approach was
to develop a sense of equal participation in a team effort. Each member of the
team was encouraged to express his or her views and decisions were reached by
consensus. In this visit I had brought with me a field assistant and translator, Mr.
Limited Mukusanya, a permanent staff member of the WWF Multispecies Project.
For this first visit, the VHS were split into two groups, one comprising the men and
the other the women. In all other visits the VRs worked as mixed male and female
groups.

Altogether eight visits were made to Masoka over the period March to
December 1992. Each visit lasted from seven to 14 days. After March 1992
translation during interviews and all VR activities was carried out by two of the VRs;

Mr. Gift Chisunga and Mr. George Kanoderuka.

General approach to recording local knowledge
Field trips were generally structured as follows: I arrived in the early

afternoon and used the remainder of the day to send messages to the VRs that

71
I wanted to meet the following morning. One of the VRs would find someone from

the community to prepare food for the team for the duration of my visit. A morning
and midday .meal were prepared for all VRs by the hired cook. By tradition men
and women ate separately. I shared all meals with the VRs. The) issue of food and
meals requires some clarification. Zimbabwe, at this time, was in the middle of the
worst drought in memory. Many of the people in Masoka were living on drought-
relief food supplied by the government. Expecting hungry people to work the long
and intense hours that the VRs worked was neither realistic nor ethical. I therefore
undertook to feed the VR team whilst we were working together. The sharing of
meals is also an important social process in the rural areas of Zimbabwe, and
facilitated an open and trusting relationship among the VRs and me.

At the first meeting the VHS were asked to describe events and issues that
had occurred or were being debated in the community since my previous visit.
Thereafter, the objectives of the visit were outlined and work sChedules were
proposed, discussed and arranged. All work times and arrangements were
arranged to suit VR needs.

New concepts and techniques were illustrated with simple, and locally
relevant examples. VRs would then carry out a task using the new technique or
concept. Once a task was complete the entire group would collect to discuss and
seek consensus on the outcome of the task.

When the basic data collection procedures and methods had been used for

several field trips, VRs were asked to prepare projects that they would like to see

72
developed in Masoka. The VRs were asked to do this in my absence with the

objective of evaluating whether the techniques that had been developed were likely

to be useful outside of the framework of my own research.

Identification of key Individuals and groups within the community
Perhaps the most important data collection method developed during this
analysis was what is called a spidergram: a graph with a central node and arms
or branches leading off this node. Each branch may end in a sub-node which may
also have branches leading from it to further sub-sub-nodes and so on (Figure
4.1). Three points are allocated for each branch emanating from a node. These
points are allocated to each branch of a node to weight the relative importance
(RIW) of that branch. The central node is designated as the main concept or
question about which we are seeking information. For example, Figure 4.1 shows
the spidergram developed to answer the question 'Which individuals or groups in
Masoka have most influence on agricultural production?" Each branch emanating
from the core node represents an individual or a group. The numeric values are
the weights attached to these individuals or groups in terms of their degree of
influence on agricultural production. RIWs were normalized by dividing each
branch score by the total number of points allocated to that node (i.e., three times
the number of branches).
VRs drew these spidergrams in the sand and used stones or sticks to

indicate the RIW of each branch. VRs were encouraged to use appropriate

73
symbols to represent each node so that even those VRs who could not read could

understand the representation. When a spidergram was complete the group would
gather around it and the VHS who had drawn the diagram would carefully explain
what each node and branch represented and then have to defend, to the entire
group, the branches they had included or left out and the RiWs they had allocated
to each branch. After consensus was reached that all relevant branches were
present and that the Rle were correct, the spidergram was copied into a field
notebook. Exceptions to the consensus rule were the actors and needs
spidergrams developed separately by male and female VRs. Consensus was

reached within these groups but not among the entire VR team.

Identification of household needs

VRs were asked to develop spidergrams to express the needs that
households required to live a basic, but adequate life. During a subsequent field
trip the identification of these needs was developed in greater detail using needs
calendars or matrices. VRs were asked to draw columns in the soil to represent
each month of the year. Then, for each of their major needs they made a row and
filled in the amount of that need required for each month. Separate matrices were
developed for the needs of an average man, an average woman and an average
child. in a thirteenth column the VFls were asked to allocate a number of points
between one and five to indicate the degree to which they could not live without

that need (five points) or if it was more of a luxury than a necessity (one point). A

74
row was added to the matrix and once again the VHS were asked to allocate

points between one and five, but this time to indicate the degree to which they
experienced needs shortages (five being none and zero being extreme) in each
month of the year. When a group of VRs had completed a matrix the whole team
gathered around it and the authors were asked to explain the matrix and defend
their inclusion or exclusion of items, the amounts they claimed were required in
each month and the relative weights they gave to each need and to each month.
Once consensus was reached by the VHS on the diagram and its content, it was

copied into a field notebook.

Inputs to, outputs from and measures of performance of, the Masoka
agroecosystem

Identification of the major inputs and outputs from the agroecosystem built
upon the spidergrams developed for needs identification. Once the needs nodes
were identified. they became central nodes for more detailed input/output
spidergrams. The question posed in the central node then became, for example,
"What is needed in order to produce a good yield of maize?" Branches from this
node indicated the major inputs to, and losses from. the production of maize.
These branches were weighted and sub-branches were added where necessary.
All spidergrams were discussed and defended before final acceptance.

A third major tool for recording local knowledge was the possibility diagram.

The objective of these diagrams was to enable VHS or other village informants to

75
express the probabilities that they believed described the relationships under

consideration. A possibility diagram for maize yields is described for illustrative
purposes. Specific maize production scenarios were identified from the production
spidergrams and from discussion with VRs. Soil type and rainfall, for example,
might be identified as key determinants of maize yields. For each soil type and

rainfall season combination a possibility diagram was developed. A single stick or

‘, was placed at the base

stone. representing a maize yield of one 91 kg bag acre'
of a column drawn in the sand. VRs were then asked to indicate the possibility of
obtaining at least that yield, under the specified conditions, by placing between
zero and 10 sticks or stones in the column above the symbol representing the
single bag maize yield. A score of 10 indicated that households could definitely
obtain at least that yield whilst a score of zero indicated they would definitely not
be able to achieve that yield. Scores between 10 and zero indicated the probability
that a household could achieve at least the yield specified at the base of the
column, under the given circumstances. A column was then added with stones
representing the number of bags yielded incremented by one. The scoring
process was repeated as before. Columns were added (and possible yields
steadily increased) until a score of zero was given to a particular yield to indicate
that it could definitely not be achieved.
Possibility diagrams were developed for maize, cotton and sorghum yields

on each of the soil types on which they were cultivated (bepe, jecha and river

lands) and for a good, an average and a poor season. Possibility diagrams were

76

also developed for maize, cotton and sorghum in each season and grown on
virgin jecha (i.e., never cultivated) and jecha that had been continuously cultivated
for five years. The need for this differentiation was realized following discussions
with VHS and other community members as yields on the jecha soils decline with
continuous cultivation. Possibility scores were normalized to lie in the range of zero
to one. These diagrams represent the cumulative probability function of achieving
at least a given yield. The probability of achieving a given yield can be calculated
by subtracting the normalized probability of achieving a particular yield from the

probability of achieving the next highest yield. The mean of a yield distribution

(E[x]) was calculated as:

Eixl ‘3 2 PDQ"?

The variance of a yield distribution (V[x]) was calculated as:

Vin = 2: (mix?) - 5x12
Where:

P(xi) = the probability of achieving yield level xi
xi = the ith yield level.

Where weighted means and variances were used they were calculated according

to the following:

Weighted an = 23 Wt, (E[x])

Weighted M = )3 (Wt, Mm + wr, (Eixlizll - am:
Where:
Wti = the probability of a particular event occurring, such as the probability

of a particular season or soil type.

Possibility scores were also used to analyze the rainfall of Masoka. VRs
were asked to identify the types of rainfall that occurred in Masoka. For each of
these rainfall types possibility scores were developed to indicate a) the probability
of a rainfall event of each type occurring in any week of the rainy season; b) the
probability of a given number of these rainfall events occurring in any two week
period of the rainy season; and c) the probability of a rainfall type lasting for a
given period of time.

VRs were asked to map the soils of Masoka, map the placement of
households in Masoka, as well as the areas that were unsuitable for crop
production due to soil, slope or cultural reasons. These maps were drawn in the
sand or on canvas, using locally available materials for differentiation. The maps
were described and defended by their authors and then copied onto paper. A
paper map was also drawn by two of the VHS to indicate the historical distribution

and movement of households within the Ward.

To elicit information on cropping decisions, yields and farmer perceptions

of soil condition, a combination of field mapping and semi-structured interviews

78
was used. When used with VRs as respondents these were combined with what

were called field interviews. A hypothetical example is described for illustrative
purposes. The respondent was asked to point to where he or she lived and was
asked to draw a circle on the grbund to represent their household. Respondents
were asked to stand within that circle and were orientated towards the Angwa
river, a feature all could point to from any place within Masoka. Then for each year
since their arrival in Masoka, respondents were asked to draw the fields they had
opened or used. Respondents were asked to indicate the size (acres) and soil
type(s) of these fields as well as the crops they had grown, the yields they had
achieved and any problems they had encountered. When this exercise was carried
out with VRs as respondents the VR was asked to stand in particular fields and to
pretend that they were the field. They were then interviewed by me and other VRs
as if they were the field. in some instances, a second VR was asked to pretend
that he or she was the farmer so that the field interview went through the "farmer"
to the "field". Questions related to weed populations, yields, fertility and general soil
condition were asked.

To obtain a clearer picture of the politically sensitive issue of land allocation
to new settlers or sons of local residents VFls were asked to adopt the roles of the
various actors involved in land allocation and to present a short play of the
proceedings. Once a play was completed the actors were switched around so that
different perspectives and emphases could be observed. Plays were observed for

a would be new settler and a long-term resident seeking land. The dialogue in

79

these plays was translated for my benefit and discussed after each performance,
which lasted only a few minutes.

Spidergrams were used to identify the attributes that VRs considered useful
in determining a household's ability to continuously satisfy needs. The central node
asked the question 'What do you look for to tell you that a household is able to
satisfy its needs?" Major factors, and sub-factors that could be used to measure
these major factors were identified, added to the branched nodes and weighted
in the usual way.

Well-being ranking (based on wealth ranking methods, Grandin, 1988) of
all households in Masoka was used to identify classes of household based on their
ability to satisfy needs. VRs were asked to define a household and to define
classes of well-being in Masoka. The name of each household head was then
written on a card and the cards shuffled together. The VRs, working as a group,
selected each card and discussed the attributes of the household before placing
it into one of the well-being classes. Decisions were reached by consensus.
Thereafter, VRs ranked the cards in each well-being class according to the
household’s ability to satisfy needs. The name of each household was read out to
the group of VRs. VRs discussed whether or not that household was better or less
able to satisfy needs than each card that had already been discussed. The card
under discussion would be moved through the stack, from bottom to top. until the
household was no longer considered to be better able to satisfy household needs

than the household represented by the next card on the stack. VRs worked in

80
smaller groups to identify the number of children and adults in each household as

well as the number of acres of river-land that each household possessed and
whether or not the household was an employer or an employee of other
households, had a family member in other forms of employment or received
remittances.

Using an open-ended interview (format, VRs were asked to recall the
expenses they had incurred and the prices they had received in producing each
crop (maize, cotton and sorghum) in the previous season (1990/91). VFts also
developed labor requirement calendars (matrices) for each crop that indicated the
number of labor days (per acre) required for each task, in each month of the
growing season. Possibility diagrams were developed to indicate the possibility of
one adult completing each activity on one acre in at least a given number of days.
These data (expenses, prices and labor budgets) were used to develop cash and

labor budgets for each crop.

Methodological validity and data accuracy

The issue of the validity of the measures made in this study as well as their
accuracy and precision is of obvious importance and is perhaps the most difficult
issue to logically and statistically resolve. As discussed previously there are two
broad areas of concern. The first of these is the degree to which the methods
employed bias the results obtained and the second is the accuracy and precision

of the measures made.

81

All interactions with VHS and other community informants required some
measure of translation; translation of the question from English to ChiSezuru
(Shona) and then the translation of the responses from Shona to English. To avoid
gathering biased information then we need to be confident that: a) the translation
of English symbolism into Shona symbolism was accurate; b) how the respondent
perceived the translated symbols was consistent with how the analyst perceived
these symbols; c) the respondent was accurately portraying their knowledge of the
phenomena under consideration; d) the translation of Shona symbolism into
English symbolism was accurate; and e) how the analyst perceived these
translated symbols was consistent with how the respondent perceived these
symbols. This process alone, had enormous potential for introducing error. The
approach to dealing with these problems in this study had been to use, as far as
possible, graphic representations of questions, situations and responses. The
assumption being made was that graphic representations provide a common set
of symbols that minimize the translations required as well as forcing both analyst
and respondent to simplify all communications. Wherever possible specific, real
world symbols were used to illustrate important phenomena; an old maize cob
might be used to represent maize in a spidergram, for example.

Content validity (Mitchell and Carson, 1989) was assumed to be minimized
by the following circumstances. if there was a distortion in the translation process
we might expect answers to either be inconsistent with expectations or for

divergent answers to be expressed. All data were discussed almost as soon as

82

they were generated. There was therefore, opportunity for immediate discussion
if the data did not conform to expectations or appeared inconsistent. Consensus
was sought on all data developed by the VRs and in most cases groups of VRs
independently developed data representations and then had to defend their
representations to the entire VR team. This worked much like the scientific peer
review system. In most situations, the VHS themselves defined the specific
information they presented. The analyst posed general questions or concepts but
allowed VHS to define the specifics of each issue. in some instances, the use of
different elicitation methods enabled us to identify whether or not there was
divergence in the recorded information.

The second test of the validity of the methods used was the extent to which
they generated results that were consistent with other measures of the same
population. For most data collected in this study, the requirement that these
alternative criteria be "unequivocally closer to the theoretical construct than the
measure whose validity is being tested“ (Mitchell and Carson, 1989) was
impossible to satisfy. There just are no reliable data for most of the factors of
interest in Masoka or in similar situations. The best that can be done is to use
published results that are most likely to have populations, of the factors of interest,
that are similar to those under scrutiny in Masoka.

The final test of the validity of the methods used was the degree to which
they produce results that were either consistent with what might be predicted by

theory or were correlated to measures from the same theoretical population. For

83
example, the Department of Meteorological Services (DMS) literature on rainfall in

Zimbabwe (DMS, 1975) provides a distribution of storm duration that we can use
to test the information provided by VRs. We might also expect different correlations
between yield and rainfall on different soil types - expectations that we can test.

The issue of data accuracy and precision was very difficult to completely
resolve. To some extent the tests of validity discussed above will enable us to, at
least qualitatively, establish the accuracy of the results we observed. if the
methods are shown to be valid, then if the same populations are measured using
different methods and each method yields the same result then we may feel
confident in the accuracy of these results. in addition, if the sample statistics we
observe are close to those published in the literature then we may feel confident
in the accuracy and precision of these results. Problems obviously arise when
different methods produce different results or the results we obtain are inconsistent
with the literature or there are no comparable results. The approach adopted in
these cases was firstly, to examine the validity of the measures using the three
types of validity tests discussed by Mitchell and Carson (1989). We recognize
criterion validity as being the strongest test and content validity as being the
weakest. if the measure is shown to be valid and we find no reason to believe that
the responses are Systematically biased, then we will accept the information. Quite
clearly, the degree of confidence we hold in particular information will be related
to the strengths of the tests that information has withstood.

The theoretical model (logical hypothesis) which underpins our data

34

collection is as follows: Masoka residents have a long history of depending on the
biophysical environment to provide most of their food and other material needs.
Household well-being has been, and is, dependent on their knowledge of the
structure and patterns of their environment as well as their ability to manipulate
their environment and the resources at their disposal to satisfy their needs. Two
sets of predictions are deducible from this model. The first of these is that Masoka
residents (as represented by the V83) can accurately describe the structure and
patterns of key environmental inputs to household needs satisfaction. The second
of these is that VRs can correctly describe the results they are likely to achieve
using commonly used combinations of controllable inputs with different resources
and under commonly experienced environmental situations. The grounds for
accepting or rejecting a prediction in this study are difficult, because in most
cases, there are insufficient data or their form is such that statistical analyses are
impractical. The decision to accept or reject a prediction will therefore, be based
on careful evaluation of the validity of the measure and comparison of the measure
to comparable results published in the literature. These decisions will be made as

transparent as possible so readers are able to reach their own conclusions.

85
RESULTS

Key actors and household classes

in response to the question 'Which individuals or groups have greatest
influence on agriculture in Masoka?" both male and female VFI groups identified
the spirit medium, headman and ward councillor as key actors (Figures 4.1 and
4.2). These data also illustrate the power split between traditional leaders
(Headman and Kraalheads) and the more recent elected leaders (ward councillor
and Village development committee (VIDCO) chairmen). The importance of spirit
mediums in community decision making within Zambezi Valley communities has
been noted by Lan (1985) and by more recent anthropological research (Hasler,
1992; Derman, 1992). The split of power between traditional leadership and the
post 1980 elected leadership appears common through much of Zimbabwe
(Drinkwater, 1991). The male and female VRs differ most in their perceptions of the
importance of VIDCO Chairmen, with the women perceiving them to be the third
most important group whilst the men considered them least important. ‘

Both male and female VRs used size as a primary factor for classifying
households within the community (Figures 4.3 and 4.4). Male VRs extended their
classification to include wealth as a second factor to differentiate household
classes (Figure 4.3). Male VRs perceived most households to be in the large and
wealthy category. Using the household size classification rules of Chapter 3 to

allocate female VR classes to large medium or small, their analysis indicates that

86

Spirit
medium
.28
Chairmen\-05\. '
Major
Actors
11 .17\C ll
- ounce or
/
ZANU PF .17
Branch
Chairman
Households

Figure 4.1. Major actors influencing agricultural production in
Masoka and their relative importance weights. Female VB
perspective.

Spirit
Medium

Headman
Households

\03 ' .17/

ZANU PF \ v / VIDCO
Branch Chairman \ 07 17/ Chairmen
° \9 Major /'

NJ
o

 

School .———-—-'

Chairman / \ Councellor
.07
.07
.07
Health / / \
Workers Wildlife Committee
Kraal Chairman
Heads

Figure 4.2. Major actors influencing agricultural production in
Masoka and their relative importance weights. Male VFl

perspective.

87
Wealth Poor
y‘.67 .33’

 

Wealthy W l
\ ea thy
.83 .22) /
b/
.17 .33
/ \
Poor Poor
ND = No data

Figure 4.3. Household classification and the relative importance weights of
each household class. Male VB perspective.

MFG-4*

M3F0-17 -
\20 l / MFS e

     
    

Household
classes

M2F4-14 MF7-12

* M-> number of male adults
F-> number of female adults
0-4->number of children

Figure 4.4. Household classification and the relative importance
weights of each class. Female VR perspective.

88

medium sized households dominate (Figure 4.4). For all household size classes
male VBs indicated wealthy households dominate. This latter result is in
contradiction to the result presented by the community members present at the
meeting when the VBs were elected. At this meeting households were classed as
poor (BlW=.8) and not so poor (BIW=.2) indicating a predominance of poorer
households. This contradiction suggests that the male VB spidergram did not
adequately representthe question of interest - 'What types of household occur in
Masoka and which are the most common?" it appears that the male VBs were
perhaps answering the question 'What types of household occur in Masoka and
which have the greatest influence?" The male VB spidergram does however,

provide useful information.

Household needs

In answering the question 'What do households need in order to live a
satisfactory life in Masoka?" female and male VB groups developed the
Spidergrams shown in Figures 4.5 through 4.8. Both female and male VBs
identified food and cash as most important household needs as has been
reported elsewhere in Zimbabwe (Shumba, 1988). The female VBs also indicated
the importance of trees and water, land and a clinic (Figure 4.5). From their
perspective, major inputs to satisfying household food needs were maize. cash,
Chickens, pumpkins and finger millet. Major inputs to household cash needs

satisfaction were maize, cotton, chickens and brewing beer.

89

in the male VB perspective. household needs and their BIVV were indicated

for each household class (Figures 4.6 to 4.8). For most household classes male
VBs identified maize, chickens, sorghum, pumpkins and vegetables as major
inputs to household food needs satisfaction. Male VBs also noted the greater
relative importance of food to poor households compared to wealthy households.
They perceived cotton and maize as major inputs to household cash needs
satisfaction for wealthy households whilst cotton, craft and sales of fish or honey
were identified as major inputs to household cash needs satisfaction for poorer
households (Figures 4.6 to 4.8). These data indicate a decrease in the number of
production activities with decreasing household size and to a lesser extent with
wealth. Similarly, Jackson and Collier (1991) found an increase in the per capita
incomes of households with increasing number of productive activities.

Also notable from Figures 4.6 to 4.8 is the shift from relying almost totally
on crop production, for wealthy households, to increased reliance on indigenous
resources by the poor households of each size class. This latter observation is in
accord with the coping strategies of households in deficit conditions described by
Zi nyama et al., (1987) in which households make increasing use of indigenous
(Wild) resources as food scarcity intensifies. These authors also suggest that it is
the women who tend to guide the household through these difficult times. The
sensitivity of women, as expressed by the female VHS, to household needs for
indigenous resources (Figure 4.5) supports this proposition. Women are also more

directly concerned with the daily collection of water and fuelwood and are

90

828828 m> 295.... .>>_m :9: 5? atom: 9.05 2 £35
BEE 9: 3 =95 mm 92$ £299; mocmtoag $26.9 :9: 53, «com: 20:33: 2m V 9:9“.

.353”. EapoE
/ =25 5.8 cases:

2228 a. N z \

/mN. I ‘2 5N. 5N.
we 6.8-5.8....
52m \8. 412/5.535
, // v\:.\ 2.x 085

5330588 852 «new...

@3521.“
1:. 38
Extra. «”6 t. «poo: : @aiillit. ...l.ll..l .o
20:09.0: .8. IlllcoamZomcoo
sumo \h d/mM.
/ 50¢“;
\.NN\

bogkm
c.3822 V/NN/

9.....iil :80
8. r .mfiI'OCOv—nucc
EoE>o_aEm\ \8.\\1.‘\.\@/.fl/ .
t. FN/NEZ
at 5.

=80\ // 3. B. .25:
0 Bl I
$33oao> \8 .580 B .35...

mo 8 mo \ o==2
_ a G @5390 />E=On_ 5:298 / tawn—
ficE

  

 

91

mm 228:0; moi m be

23.530
830 I or.
:r t

mcidEad ISTIV

\ B
n . ‘
Eacmaow\ p &\t

8.
\ \

>25: u__>>\ &\ 9.

628828 ¢> 9m

2 .>>_m :9: 5? name: $0.: 2 235 SEE 9: mm :95
555 2:995 wocmtoaé 9:29 :9... 5:5 .mp8: 20:832.. .3. 9:9“.

«mos
Eageom/ .80

\

             

 

m: QE:
29.0.50 2. 9. \ .3 d
. or.
/t. IQSwScucaew
cums. lth /flo¢omo.lmo.nflomo>
. /
8. 8/mco__oE-._2m>>
B. 8. _ mzogaoao
i=2
cozoo
ON. NEE
K 9.\o
\

sKa/

_ 5:00
520

T .
5. .1 @d/ 91600
wo_fi.omo> \ 8.\t \ /1/ or.
. 5. /.
mco=mE¢2m>>\ \8 9. m 9.020

/. Manuaow
830

9:: u: 20

 

628898 m> was. .BE .65 55, mp8: 805 o. 985 BEE 9: 8 =95
8 20:83: 6:68. a .2 55.5 £5995 8:2..an 8:39 .65 £3, .88: 20582.. .5. 9:9“.

           

$388; E3590 goofing
mco__oE.._2a>>/ m’o. w/mw. 3\
m: QE: 2.15 t 89.020 /2. .b:o=oE.._2£$
.2 .. lwflfi gel... 9 a
\ 8.
EEQMMono\t. “a / mm. /mc_xaE:d
. . one: 8. B.

«2.22

92

29.820:

\2. 8
\éc/ . 6%

:90

B. 8 \ d/ .
\ 9. / . 8
.68: was \ 8 R 8:00 2 3 / 5.80
not _ / \ t. /

28 >30 “no”

 

cums.
>801

628828 m> was. .>>_m .65 53> mp8: 805 9 98:. SEE 05 mm
:95 mm 8.88:0; =me .2 .925 9:995 85:08:: 8:39 :8. 2:; .88: 29.8.8... .m.v 9:9“.

 

menses;
89.0.8
2 29.35 \
mcflaEsd/ B. / or
.7. .
£1.49
2095030 \
ON. 9.
mo_§omm>\ \ /
one: 8.
«Sc 9265
205830:
% \el :95 ollé/
\BF. 8.
9» /6/
«N\ a 3 / a / B
\ A 8:8 N. / :88
_.P.
.. _ /

94

therefore, more likely to be aware of the importance of these resources than are
men. The greater reliance of poorer households than wealthy households, on craft
production for the generation of cash incomes (Figures 4.6 to 4.8) suggests that
these households could be important initiators of informal seetor production
enterprises. Boserup’s (1965) model of agricultural development implies these
groups play an important role in rural development.

What is surprising from these analyses, is that only the female VRs indicated
the importance of land to household needs satisfaction. (Figure 4.5). This is
particularly noteworthy in that women do not have the same rights to land as do
men. Their rights of use are normally derived from their husbands or fathers. and
sometimes even their children (Reynolds, 1991). Perhaps because of this relative

landlessness, women have a greater sensitivity to land scarcity than do'men.

System identification

Spidergrams were developed to illustrate the inputs to. and losses from,
each component of household needs. Only the spidergrams for maize, cotton and
sorghum are developed and discussed in this chapter. The importance of these
crops as inputs to both household food and cash needs (Figures 4.5 to 4.8 and

Tables 4.16 to 4.18) suggests their selection is appropriate.

95
Malze production

In answering the question 'What is needed to produce a good yield of
maize?" both the male and female VRs identified rainfall, soil, crop variety and labor
as key inputs (Figures 4.9 and 4.10). Female VRs noted the importance of land but
male VRs did not. Major losses from the maize production sub-system were
identified as being caused by wild animal damage (particularly from elephant.
buffalo and wild pigs) by both male and female groups of VRs. Insect pests were
not identified as major loss causing agents although the male VRs identified
termites as causing loss.

Female VRs developed a more detailed breakdown of labor inputs to each
task than did male VRs. Gender differentiation of tasks is evident as is the
importance of the labor of children, findings that are in accord with those of
Reynolds (1991) for the Tonga people of the Zambezi Valley. Female VRs also
indicated a greater appreciation of natural resources than did male VRs through
their inclusion of trees and grass as inputs to maize production; These inputs are
important in restoring field fertility.

Male and female VR perceptions of the relative importance of different soil
types for maize production differed considerably. The female VFi group stressed
the importance of katondo and bepe with shapi and jecha being similarly
weighted. Very few households in Masoka have access to katondo soils. These
soils are red soils with a high clay content. Discussion with the VFls indicated that

they were identifying these soils with the red soils of the major maize growing

96

.mp8; mums. boom do
EoEo>oEom oz. .80: $09.me .23 can g2 8.559 m. 8:; so 95328 m> p.95“. .m.v 9:9”.

moot.
5965 896
.28; mo. .o5___tmu_
/ 8. mo. \

      

03:5

mo
mo.

82 . 2. /
an . m . . mw. /»
Euro \ \3 \ax $2. 5320 \ e 9 ma 25>

SE; _mcoaushtmmué

x1 >
m&m\ 8 8

95K 5%. Ag
a.) must
\ \. .

8.1 9:235

 

=£Emm

 

.833 came boom do EoEo>mEom
on. .80: $09.8 .9? can ..2 .8569 w_ “an; :o 9:83on m> was. .94. 9:9“.

383 80: roam—BF 9.0mm

 

98

areas of Zimbabwe. Thus the women were being true to the question in identifying
these soils as being best for achieving good maize yields. Of the soils that are
generally available to households, both male and female VR groups agreed on the
relative importance of bepe over jecha. There appears to have been some
confusion as to the identification of shapi soils. Only one household in the
community was observed to grow crops on shapi, suggesting that few households
had any experience with this soil type. it is also important to recognize that at this
early stage of the analysis the VR groups were not distinguishing between the
riverine sandy soils (river jecha) and the alluvial terrace sandy soils (jecha). These
distinctions only became apparent later in the analysis when dealing with the yield
potential of different soils.

Female VRs identified maize variety R201 as being the best which agrees
with the varietal recommendations made by Whingwiri and Harahwa (1985) for NR
Ill and with those of Shumba (1984a). The selection by male VRs, of variety R215,
a long season variety. in preference to R201 and R200, both short season varieties
is more difficult to understand. Whingwiri and Harahwa (1985) suggest that R215
is a poor performer in a low yielding environment but can respond significantly to
an improvement in the environment. The interaction of soil and variety was not

investigated in this study but could be a fruitful subject for future research.

Cotton production

In contrast to the maize production sub-system, female and male VR group
perceptions of the relative importance of inputs to cotton production were quite
different. In answering the question 'What is needed to achieve good cotton
yields?", both groups indicated the importance of soil and chemicals (Figures 4.11
and 4.12). Female VRs considered rainfall and crop variety to be less important
than labor, whereas male VRs perceived rainfall to be the most important input and
both rainfall and crop variety were perceived as being more important than labor.
The Rle ascribed to women’s labor by both male and female VRs (Figures 4.11
and 4.12) suggests why female VRs perceived labor to be of such great
importance. Women are the dominant source of labor for all of the major activities
except land clearing. Cotton production adds appreciably to the demands made
on household labor and particularly that of women and children.

Male and female VRs differed also in what they perceived to be the major
losses from the cotton production sub-system. Male VRs identified insect pests as
being less important than monkeys and baboons whilst the female VRs identified
red bollworm as the major loss and did not note losses due to wild animals.

Both male and female VR groups identified jecha as the most important soil

type for cotton production but differed greatly on the relative importance of the

other soil types.

100

.mu_o_> cozoo boom .0 EoEm>mEom
9: .80: Eugen 8...? new .8. 09309 m_ .23 :0 93898 «S 295“. .P 3. 9:9“.

5.4.. a»...
8
x

832

\.

          

Ezoz,__oo
.x 528:3

Euros B.
\ 9.an

58; a o. .
583 0:0 o

cozoo

  

9.520

23 E82509 \m.

com 8 /mc_>maw

8.
m_mo_Emno\ \ P «P. /bo:m>
was. n. 2.8%

8mm 8 Macon
Barman

 

8:23.

101

.333 8:8 boom
so 59:99:08 on. .80: «88:8 .853 new :8 8.509 m_ .23 :0 026828 m> 28.2 d; 959“.

29:25 53882

 

  
 

.25...
SEQEm ofisc 28:50.98
cops; .4 /ON. \agw_o 5.0
/8 :. 822328 B. 8. mean 2.8””. 532.3 580$
éf beam 8.1 \
c238\8..1 - _ £9.85. Ken
2 a 8.88 _ Q,
. mm.
insists? 9. /3 x 8\ x {2‘2
Siol:.t.\ c2823.. . .om.Emc 8/
C0300 no.0 4.8 ofl/ 9.280 .80..
\ mp.
82 53;. bo:m>\ x3928;
ma/cmcaom
:Emm
w_mo_EonO\
8. @Mgofl.

_amnw

oo/om

1 02
Sorghum production

In answering the question 'What is needed to achieve good sorghum
yields?" both male and female VFi groups identified rainfall, labor and seed variety
as major inputs (Figures 4.13 and 4.14). The most notable differences in the female
and male VR perceptions were the inclusion of land by the female VHS and the
much higher importance of soil type in the male VR analysis than in the female VR
analysis.

As with the maize and cotton production spidergrams, the female VRs
developed more detailed labor components to their spidergrams than did the male
VRs. The importance of male labor in so many of the activities of sorghum
production (Figure 4.13), the identification of a male householder as the local
expert on sorghum production, as well as the greater detail in the variety
component of the male VFi spidergram (Figure 4.14) suggested that sorghum is
less of a women’s crop than is generally believed.

The relative importance of local seed sources compared to commercially
available hybrids and the relatively high weights given to bepe soils by both male

and female VRs are consistent with Reynolds (1991) findings elsewhere in the

Zambezi valley.

Summary of Inputs to and losses from crop production sub-systems
For maize, cotton and sorghum rainfall, soil type, seed variety and labor

were identified as the major determinants of good yields by both male and female

103

 

.mp6; 53:98 noon be .5592ch
on. .89. £858 “on; new c2 8.53. m_ “an; :0 93828 «S macho“. .96 2:9".

magneto

3%: t. 8.8 88 .535

     

cozoaooa
Eacmoom

/. /® Tamilaom
m. w...

.323”. . /
. 52m. /

02.93.
9.3...

 

104

.333 £398 boom so
298538 we .89. 38:8 an; new c2 8562 m_ “as; :0 920898 m> 292 .36 059....

Magnum
8.

K
madcap no.0
\

mp.

 

8.
/...E_m

oozes—boa
Eacmoow

om. _Esam
§mfinm\¢. 8. W
goals. .. I
8836\8. V8. om/.
28.. 80m

 

105

VR groups. Wild animals were held responsible for a large proportion of the
perceived losses from maize, cotton and sorghum production whilst insect and
mite pests were important loss causing agents only in cotton. Female VRs
consistently expressed their perception of the importance of land and, in the case
of maize, their perception of the importance of trees and grass for fertility
restoration. Women appeared to be the dominant source of labor in cotton and
perhaps in maize although the trend was not as clear as for cotton. Men seemed
to have a greater role in sorghum production than is generally recognized.
Children contributed importantly to the labor requirements of the household in
most major crop production activities. Commercially available seed varieties
(hybrids) are most important for maize and cotton but traditional (open pollinated)
varieties are most important in sorghum production.

More detailed analyses of the rainfall, soil and labor inputs to crop
production were conducted and the results are presented below. Despite the
stated importance of crop variety in achievement of household needs satisfaction,

this factor was not investigated further.

Rainfall analysis

The VRs identified four distinct rainfall types: The first, named Mhepo (called
thunder in subsequent discussions), occurs throughout the season and is
characterized by wind and thunder with only a light rainfall over short periods of

time. The second rainfall type, Mhunurukwa (called heavy rain in subsequent

106

discussions) is characterized by heavy rainfall with storms lasting longer than
several hours. This rainfall type occurs mostly in January and February. The third
rainfall type Mvuramabwe (called hail in subsequent discussions), is characterized
by hail, occurs infrequeme and usually lasts for less than 30 minutes. The fourth
rainfall type, Pfunambuya (called light rain in subsequent discussions), is a light
rain that continues for several hours and sometimes even a whole day.

The VR data indicated a high probability of the rainfall season beginning in
the first two weeks of November and certainly beginning in the last two weeks of
November (Table 4.1). The Department of Meteorological Services (DMS) indicated
November 22"d as the mean date of the first rainy pentad for the eastern mid-
Zambezi valley and the end of the rainy season occurring in the first two weeks of
March (DMS, 1981 ). The VR data suggested the last certain rains occur in the first
two weeks of March (Table 4.1) but recognized a high probability of light rain
occurring through until the first two weeks in April. The DMS also indicated the
mean number of raindays of 1mm or more as being between 50 and 60 days for
that region of the Zambezi Valley (DMS. 1981). if we take the sum of events with
possibility scores greater than zero as an upper bound on the number of rain
events and those with a possibility score of 0.5 or greater as a lower bound (Table
4.1) then the VHS indicated a range of between 41 and 76 rain events a season.
We should expect these figures to be slightly higher than the DMS values due to
the possibility of multiple rain events in one rain-day. The DMS identified the major

hail season as occurring in October, November and then declining sharply in

107

Table 4.1 Subjective probabilities indicating VR perceptions of the probability of the number
of rainfall events, of each rainfall type, occurring in any two week period of the

 

 

 

 

 

rainy season.
Month Tact Nov Bee Jan Feb Mar Apr
Week
Rain type 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2
# Events
1 .2 .a o o o o o o o o o o
2 o o o o o o o o o o o o
3 o o o o o o o o o o o o
4 o o o o o o o o o o o o
”a" 5 o o o o o o o o o o o o
5 o o o o o o o o o o o o
1 o .s 1 1 1 1 1 1 1 .5 o o
2 o .2 1 1 1 1 1 1 .7 .2 o o
3 o o .5 .5 .5 .5 1 .6 .3 .1 o o
4 o o .3 .2 .2 .3 .4 .5 .1 o o 0
H3131? 5 o o .1 o o o .1 .2 o o o o
6 o o o o o o o o o o o o
1 o o o o 1 1 .4 .4 .3 .2 o o
2 o o o o 1 1 .2 .2 .1 .1 o o
3 o o o o .a .5 o .1 o o o 0
mm“ 4 o o o o .5 .3 o o o o o o
5 o o o o .2 .1 o o o o o o
s o o o o o o o o o o o o
1 o o o o o o o 1 1 1 .5 .s
2 o o o o o o o 1 1 .7 .3 .2
3 o o o o o o o .s .9 .2 .2 0
”gm rain 4 o o o o o o o .4 .4 o .1 o
5 o o o o o o o .1 .1 o o o
5 o o o o o o o o o o o o

 

December (DMS, 1981). These expectations were consistent with the VR
perceptions of a low possibility of hail in late October and a high probability in
early November (Table 4.1, Figure 4.15). The VRs did not recognize any possibility
of hail occurring in December. VRs indicated a high probability of thunder activity

in late December through to the beginning of March (Table 4.1, Figure 4.15) which

 

 

 

 

 

 

Probabllttg
000053.099
Ox
l

.3"1 u *

 

I
I

‘

I

as: .

E \
0. 1
a ’5
a. ‘,
‘. t

0" Iifo :.:.':—_:'II::IZIIII
06/09 26/ 10 15/ 12 03/02 25/03 14/05

Date (dag/month)

 

“.O-
253‘.
'.:|-.
i:
1
I
1
1

0.1- ..
l

 

 

 

—B— Hall ............. Heavg W Thunder “’0' quht

Figure 4.15. Probability of a rainfall event of each of four rainfall types
occurring in each week of the rainy season.

109
is consistent with the DMS reports of high thunder activity in November through

to the end of February and then dropping rapidly in March (DMS, 1981).

The DMS reported the frequencies of rainfall storms lasting for particular
periods of time (DMS, 1975). These data may be used to compare the rainfall
event duration probabilities developed by the VRs for each rainfall type (Figure
4.16). The DMS indicated a 45% probability of a storm lasting for one hour or less,
26% probability of a storm lasting for two hours, a 14% probability of lasting three
hours, an 8% probability of lasting four hours, a 5% probability of lasting five hours
and a 1% probability of lasting 6 hours. The VRs indicated a 30% probability of a
heavy rainfall event lasting one hour or less, a 30% probability of it lasting two
hours, a 10% probability of it lasting three hours a 20% probability of it lasting four
hours, a 10% probability of it lasting five hours and no probability of it lasting as
long as six hours (Figure 4.16). VRs indicated that light rain events would, in
general last, longer than the heavy storms (Figure 4.16).

Masoka residents, VHS and a local expert, identified by the VHS, classified
previous rainfall seasons as being either good, average or poor. The VRs, working
as a group, and the local expert classified the previous ten seasons whilst in the
questionnaire survey presented in Chapter 3 all respondents were asked to
classify three seasons ending 1990/91. The results of these classifications, together

with the rainfall measured at Angwa Bridge are shown in Table 4.2.

Drobab111tg

110

 

 

 

 

 

 

 

0.
0.
0. "
0.
0. ..
0. 1- x
0 fl I T I I .“‘* I
0 1 2 3 4 5 6 7 8
Event duratlon (hrs)
+ Hall "'8'" Thunder W Heavq W quht

Figure 4.16. Probability of a rainfall event lasting at least a certain time for
each of the four types of rainfall.

111

Table 4.2 Classification of ten rainfall seasons ending 1991 [92 as good (G), average (A) or
poor (P) by village representatives (VRs), a local informant and questionnaire
survey (1988/89 to 1990/91 only).

 

 

 

 

Rainfall season Measured Source of season classifications1
rainfall (mm yr") , ,
at Angwa Bridge VRs Local Questionnaire
expert survey
% Respondents
G A P
1 982/83 561 P P
1 983/84 529 G A
1 984/ 85 1 303 G P
1985/86 ' 1057 A e
1 986/87 558 G G
1 987/88 579 A P
1938/89 91 1 A P 76 2 22
1 989/90 803 P nd 41 25 34
1 990/91 478 P nd 1 4 1 4 72
1 991 / 92 431 P Dd

 

1.Seasonclasses:G=good,A=averageandP=poor.nd=nodata.

These data show an inconsistent pattern of agreement between rainfall amount at
Angwa Bridge and the season classification. Some of this inconsistency is
attributable to the spatial variability of rainfall in the Zambezi Valley area; quite
possibly the rainfall at Masoka differed significantly from that at Angwa Bridge in
any one season. We could expect some overlap in seasons classed as good with
average or poor with average but we would not expect a consistent classification

system to confuse good and poor. We note one such conflict (for the 1984/85

112
season) between the VR classification and that of the local expert (Table 4.2). The

VR classification, based on a consensus of eight people, was consistent with the
rainfall data. if we accept that the VR classification is correct then the proportions
of good, average and poor seasons are .3, .3 and .4.

Consistency in results of the rainfall analysis that were obtained using
different methods (Table 4.1 and Figure 4.15) gave us confidence that the results
were not biased by the method of data collection. The good agreement between
the published records of rainfall patterns and the observations of the VHS gave us
further confidence in the validity of the methods used as well as in the accuracy
and precision of the VR observations. The positively skewed distributions of rainfall
event durations provided by the DMS (DMS, 1975) provide the theoretical
expectation with which to compare VR data. As with previous tests the VR data
were consistent with the theoretical distribution providing further support to the
validity of the methods as well as the accuracy and reliability of the results. In each
of these cases the results do not support the prediction that VRs are unable to
provide accurate descriptions of the structure or patterns of key elements in their

environment.

Labor requirements for each crop enterprise
Cotton has the highest labor demand per hectare (160 person days) of the
three crops under consideration and sorghum the lowest with 84 person days

(Tables 4.3 to 4.5). The most labor demanding activities were weeding, harvesting

113

Table 4.3 Labor requirements (days ha"), for each activity of maize production in each

 

 

month of the growing season.
Labor days ha'1 morith'1
Activity NOV Dec Jan Feb Mar Total
Digging holes 10 10
Planting 5 5
Weeding 12 25 37 12 86
Harvesting 17 17
Collecting maize 7 7
TOTAL 27 25 37 12 24 125

 

and land preparation in all three crops with pest management (scouting and
spraying) being demanding on cotton. Peak labor demands occurred in
November, December and January which are similar to the results reported by
Shumba (1985. 1988) for Mangwende CA. The Farm Management Research
Section of the Economics and Markets Branch of the Ministry of Lands, Agriculture
and Rural Resettlement provided labor input data input for maize, cotton and
sorghum collected by questionnaire survey in several CAs of Zimbabwe (MLARR,
1989, 1990). Not all of the categories of activity in the MLARR report were
comparable to those used by the VHS for Masoka but the general trends may be
observed. The hourly labor requirements presented in the MLARR report have
been divided by six to develop daily figures for comparison with the Masoka data.
Reynolds (1991) reported working days of between four and six hours and

Shumba (1986a) also uses a six hour working day in his analyses.

1 14
Table 4.4 Labor requirements (days ha") for each activity of cotton production for each

 

 

month of the growing season.
Labor days ha'1 month"
Activity Dec Jan Feb Mar Apr Total
Digging holes 12 12
Planting 5 5
Weeding 20 20 20 1o 70
Scouting 2 5 14
Spraying chemicals 5 10 7 2 24
Picking 35 35
TOTAL 44 35 32 14 35 160

 

Table 4.5 Labor requirements (days ha") for each activity of sorghum production, in each
month of the growing season.

 

Labor days ha" month"

 

Activity Oct Nov Dec Jan Feb Mar Total
Clearing 5 5
Digging holes 12 12
Planting 5 5
Weeding 20 17 10 47
Harvest 10 10
Carrying 5 5
TOTAL 5 1 7 20 17 1O 1 5 84

 

For maize production the MLARR reported labor inputs as being between
63 and 127 labor days per hectare for their sample sites in Natural Regions Ill and
IV. Due to the non-availability of animal draft power in Masoka and the limited

ability of households to hire a tractor for ploughing operations we might expect

1 15
Masoka labor inputs to be higher than those in the areas reported by the MLARR.

For the most part the VR data indicated a greater proportion of the labor allocated
to weeding than is reported by the MLARR and less on harvesting than reported
by the MLARR. Ascertaining from the MLARR report what activities are included in
their harvesting category, which accounts for between 50 and 70% of the labor
inputs reported for maize was not possible. Shumba (1986a) reported half the
harvest labor, at approximately the same yield level as do the MLARR. These great
differences suggest the MLARR harvest data included activities not used by the VR
nor were in the data reported by Shumba.

The MLARR data on weeding labor ranged from 15 to 27 days per hectare
in 1988/89. Possibility diagrams were developed by the VR depicting the
probability of one adult completing a task in a given number of days. For weeding
one hectare of maize the expected value (standard deviation in parentheses) for
task completion was 16.3 (3.4) days which is close to the 17.5 days implied by
Shumba (1986b). Shumba indicated that farmers weed their maize fields twice in
a season. The VR data indicate six weeding rounds per season for maize which
must be an ideal situation. if we assume two weeding rounds in Masoka then the
total labor requirements for maize are 72 days ha", much the same as Shumba’s
73 days. Mangwende farmers used slightly more labor for planting (7.4 days ha")
as well as for harvesting (30 days ha") (Shumba, 1986) than do farmers in
Masoka. The lack of draft power in Masoka increased the labor demand by 15%

for maize.

116
The MLARR data presented for cotton labor requirements ranged from 107

labor days for NR III to 172 labor days for NR N (MLARR, 1989). Once again
these overall totals were similar to those presented by the Masoka VRs (Table 4.4).
The weeding labor figures for one of the MLARR sites (Nyjena, NR N), 76 labor
days, were similar to those expressed by the VR for Masoka (7O labor days, Table
4.4). The mean number of days for all sites was however, 36 days ha'1 which
suggested about two weeding rounds a season. The harvest labor depicted by
MLARR for cotton, 42 to 91 labor days, (MLARR, 1989) was higher than the 35
labor days expressed for Masoka by the VRs. The VRs did not develop a
possibility scoring diagram for cotton picking but indicated that three picking
rounds would be required; the first requiring about 14 labor days, the second 10
and the last 6. These data were consistent with the 35 days depicted in Table 4.4.
It seems likely that VRs were representing an ideal situation in identifying four
weeding rounds on cotton. As with maize it seemed likely that only two would be
done. If this was the case then the weeding labor would be 40 days ha'1 and
make the total labor requirement for cotton, 130 labor days ha".

The MLARR labor input data for sorghum production indicate about 65 labor
days were required per hectare (MLARR, 1989) which were lower than the 84 days
indicated for Masoka by the VHS (Table 4.5). As with cotton and maize production
we might expect Masoka labor requirements to be higher because of hand tillage
requirements and the need to guard crops from'wild animals. Guarding labor data

were not included in the labor budgets and the labor required for digging holes

117
was 12 labor days per hectare. With these days removed the Masoka labor data

were much closer to those from the MLARR. As with cotton and maize however,
the distribution of labor among tasks differed greatly: weeding labor in the MLARR
sites (between 21 and 33 labor days ha") was less than in Masoka and that for
harvesting (29 labor days) much higher. As with the previous two crops we might
assume that households weed their sorghum fields only twice (Chiduza et al.,
1992; Reynolds, 1991) which would make the Masoka data (37 labor days ha")
comparable to the MLARR data.

When the data from the labor matrices (T ables 4.3 to 4.5) were compared
to those derived from possibility diagrams there was general agreement giving us
confidence that the methods were not distorting the results. The results for maize
presented by the VRs were very close to those described for Mangwende by
Shumba (1986a) as well as being close to the more general patterns of labor input
described by Reynolds (1991). The VRs appeared to be indicating optimal labor
inputs in order to answer the question posed. We might expect therefore, the
actual labor inputs to be more in line with those suggested by Shumba -
particularly with regard to weeding. The MLARR data exhibit great variability in
labor input data between sites and except for the harvest data, the Masoka labor
input values were well within the ranges reported by the MLARR. Given the close
agreement between VR harvest data for Masoka and that of Shumba (1986a) for
MangWende we may be reasonably confident that the labor input data reported

by the VHS were valid and accurate. These results do not support the prediction

118

that VRs were incapable of providing accurate descriptions of the structure or

patterns of key elements in their environment.

Soils and soil fertility

VRs were asked to define and then map the soils of Masoka and then to
select sites, representative of those soils, in which to dig soil pits and collect soil
samples for soil analysis. A pedologist from the University of Zimbabwe, Dr. W.
Verboem, was brought to Masoka to assist in mapping and classifying the local
soils. He identified essentially the same soil classes as had the VRs.

VRs were asked to describe the indicators they used to determine how
fertile or infertile a particular field was and how they would know when a fallowed
field's fertility had been restored. Loss of fertility was indicated by yield decreases
and the increasing cover of annual weed species commonly called Warimani and
Kambumbu1o (Table 4.6). They said this occurred after three to seven years on
jecha soils and was consistent with the findings of Reynolds (1991). Fertility was
indicated with the presence of tree and grass species, particularly Acacia tortillas
(Table 4.6). They believed full fertility was restored on jecha soils after 15 years of
being left fallow at which time A. tortillas reached a height of nine to 10 meters and
a maximum biomass (Figure 4.17) although few farmers leave their fields fallow for
that long. Grant (1981), in discussing the fertility of granitic sands in the CA5 of

Zimbabwe, suggested these soils could be cropped for three years and then left

 

10

Botanical names for these plants were not available at the time of writing.

119

Table 4.6 Relative importance weights (5 highest to 1 lowed) of tree and herb or grass
speciesafterajechafieldlsleftfallow.

 

Local species name Years after cultivation ceases
incl1 1 2 3 4 5 5 7 5+

 

Trees
Mzungu (Acacia F 1 2 3 4 5 5 5 5
tortillus)

Marowe F 1 2

Mhangara (Dalbergi'a 1 2 2
melanoxylon)

Mutohwe (Azanza F 1 2 2 2 2
garckaena)

Mupakasa 1 2 2 2 2 2 2
(Lonchocarpus
capassa)

Herbs and grasses
Warimani l 5

6050 (T richodesma 1 2 3 3 3 3 3 3
zeylancia)

Mhunhuniwa 1 2 3
Tsikinya F
Kambumbu l 4 3 2

Tsine (Heteropogon F 1 2 3 3
contortus)

Bande (Urochloa F 1 2 3 3
tribhopus)

1. ind = Fertility indicator. F = Indicator of fertile soil. l = Indicator of infertile soil.
fallow for 15 or more years until legumes of the genus Croteli'aria indicated that the
soil was again fertile. Attempts to establish normal fallowing practices in Masoka
proved fruitless. No consistent pattern was discemable either from discussions
with the VHS or from the people within the community that the VHS indicated were

particularly knowledgeable about Masoka. Many farmers stated that they

1mportance uelqht

Qelatlve

.—.

120

   
   
 

10=Mox biomass
1=M|n blomdss _;;§;§g§g§g§g
0 O=Does noioccur
7 iii???3253555353333???
5 .:i§i§§§§§§§§é§§§3"
4 2§§§§§§§§§§§§3§3§*
3 Mzunqa
2 Hpakasa
Marowe
1 .t..........«....... Mpanqara
I\I\I\/\/\/\/\/\/\/\l\I\I\/\I\/\/\/\I\I\
0 III/Ill/l/l/l/IIIII/ HUtOhue

11213 4 5 6 7 8 9 N31112131415
Years alter jecha field abandonment

Figure 4.17. Increase in total tree biomass with time after field abandonment
as a proportion of the maximum achievable in 15 years for major tree
species growing on jecha.

121

would leave a field fallow after five years of continuous cultivation but when asked
about the actual history of their fields few actually did what they said they should
do. The period that fields were left fallow was equally difficult to discern clearly.
One local expert, Mr. Dishon, who had been cropping in Masoka since 1964,
indicated that he cropped his jecha fields for four consecutive years and expected
slight yield reductions each year. In the fifth year he would plough the field with a
tractor and the yields would increase to the same as those expected from a virgin
field. Thereafter, they would steadily decline until the ninth year when he would
leave the field fallow. He left the field fallow for five years and then restarted the
cycle. Most Masoka residents that were questioned on this subject held
approximately this view of the fallow cycle however, their actions were more
diverse than this pattern indicates. This nebulous fallow practice was also reported
by Reynolds (1991) for areas further west in the Zambezi Valley.

The VRs indicated that households seldom, if ever, left bepe or river jecha
fields fallow. They knew of no one in the community who did so. Bepe soils were
considered to be endlessly fertile and the periodic flooding of the Angwa river was
considered to restore the fertility of the river jecha fields. None of the households
in the community had had sufficient experience with cultivation of either katondo
or shapi soils to know when fertility began to decline or how long to leave them
fallow for fertility restoration.

The understanding of soil fertility by VRs was consistent with the trends

reported in the literature but the practices of households differed considerably from

122

what was considered good practice. Some farmers, for example, followed more
or less regular fallowing cycles whilst in at least one case, a farmer had
continuously cultivated the same fields with sorghum for at least 20 years and
perhaps as much as 30 years. There was general agreement on the importance
of trees, particularly A. tortillas, in restoring soil fertility and several tree and grass
species were used as indicators of fertile soils. A. albida is considered a sacred
tree in Masoka and residents may not cut one down. No attempts were observed
or described by the VRs to enhance the nutrient restoration potential of these trees
by silvicultural or management practices. The clear descriptive understanding VRs
had of their soils, as well as the good relationship between the knowledge of VRs
on soil fertility and that in the published literature, do not support the prediction
that VRs were unable to provide accurate descriptions of the structure or patterns

of key elements in their environment.

Crop yields

The yield estimates derived from asking household respondents to map
household fields and asking them for their estimates of acreages and total yields
generated the yields shown in Table 4.7. Analysis of variance indicated yields
were different among seasons for maize (p<.05, F=4.468, n=52), for cotton
(p< .05, F=5.375, n=28) and for the two classes of season (average and poor) for
which there were sufficient sorghum yield data to make comparisons (p<.01,

F=8.439, n=26). These data were similar to the average yields reported for the

123
Table 4.7 Mean (standard deviation in parentheses) yield estimates (kg ha“ 1) for maize,
cottonandsorghum,derivedfromhouseholdfieldmapsandhousehold
respondent inforn'iation.

 

 

 

 

Season Maize Cotton Sorghum
< kg l‘la'1 >

Good 1334 (755) 755 (35) ND
n=2 n=2

Average 804 (515) 915 (458) 631 (445)
n=14 n =1 1 n: 8

Poor 511 (429) 459 (209) 235 (252)
n-36 n=15 n=18

Weighted Mean1 e45 (575) 700 (337) --
n=52 n=28

 

1. Weighted mean. Season averages weighted by the probability of that season
occurring. Good = .3. average = .3 and poor = .4. ND = No data.

Table 4.8 Maize. cotton and sorghum yields reported for the mid-Zambezi Valley.

 

 

 

Brunt‘ Harlziz Hawkins’ Jessat‘
Avg. Dry East Best Avg. 1983]
84
Maize 700 200 300- 1800 720 180 846 1340
1000 (793)
Cotton 700 nd 500- 1600 1200 950 981 1600
1800 (512)
Sorghum 600 450 400 1600 900 450 540 910
(364)
1. Brunt et al. (1986). Average, drought year and eastern Zambezi valley yields
1984/85 season. nd = No data.
2. Harizi (1985). Best, average and 1983/84 (drought) season yields for Muzarabani.
3. Hawkins Associates (1982). Mean (standard deviation in parentheses) yields for
multiple sites in the eastern midlambezi Valley area. 1981/82 season (low rainfall).
4. Jassat and Chakaodza (1986). Average yields, based on Agritex crop forecasts, for

Rushinga District (NR N and ill), some of which lies in the eastern Zambezi Valley.

124
mid-Zambezi valley by Brunt et al. (1986) and to the maize yields reported by

HaWkins Associates (1982) and Harizi (1985) as shown in Table 4.8. Cotton yields
reported by Hawkins Associates (1982) and Harizi (1985) were higher than those
presented in Table 4.7. Without good season yields for sorghum estimating the
weighted mean and hence comparing these results to published yield figures was
difficult. The values for an average season were however. reasonably close to the
mean yields reported by Hawkins Associates (1982) and Brunt et al. (1986), but
were much lower than those reported by Harizi (1985) as shown in Table 4.8.
None of these authors reported yields on different soil types.

The data collected from field maps show reasonable agreement with
published yields for this part of Zimbabwe and were within the range one could
expect from national data sources. There may, however, be problems with the
repeatability of these measurements; The mapping of household fields and
requests of yields was repeated twice each for two VRs (one male, one female).
This facilitated examining the repeatability of yield estimates derived in this manner.
In both cases the responses differed quite markedly from one mapping exercise
to another. Differences occurred in the statements of the acreages as well as in
the total yields obtained. These anomalies, as well as the notable variation in yields
reported for the Zambezi Valley (Table 4.8), prompted development of the
possibility scoring ’method for estimating yields.

The expected yields of each soil type (Table 4.9) were weighted by that soil

type expressed as a proportion of the total of cultivatable soils (bepe = .13, jecha

125

= .77 and river jecha = .1, when calculating cotton and maize values; bepe = .144
and jecha = .856 when calculating sorghum values). For jecha soils the average
of virgin and five year cultivated jecha expected yields were used to derive the
weighted means. These values are shown in the weighted mean column of Table
4.9. The weighted means across seasons were calculated by weighting the
expected yields of each season by the proportion of those seasons calculated
from Table 4.2 (good=.3, average=.3 and poor=.4).

The yields of all crops, on all soil types and for any season show
considerable variability (1' able 4.9) with co-efflcients of variation between 10 and
133%. Variability among seasons appeared greater than that among soil types.
VRs reported highest maize yields on river jecha soils in any season. The low per
household proportion11 of river jecha fields (p=.46 of households have river
jecha and p = 0.1 of the lands are river jecha) reduced the impact of river jecha
yields on household needs satisfaction. Maize yields on both bepe and river jecha
were slightly higher in an average season than in a good season. This is
consistent with what we expect because the high clay content of the bepe soils
makes them waterlogged in wet years, thus reducing yields (Mittra and Stickler,
1961). On river jecha soils the flooding of the Angwa river in wet years reduces
yields through waterlogging, lodging or being washed away. Poor season maize

yields were higher than what we might expect on reading the literature; Yields on

 

11

The proportion of river jecha fields was estimated using a digitized soil map. The procedure is
described in Chapter 5.

1 26
Expected values for maize, cotton and sorghum yields (kg ha“) on three soils
and for good, average and poor rainfall seasons. Source, possibility diagrams.

Table 4.8

 

 

 

 

 

 

 

 

Crop Season Bepe Virgin 5yr River Weighted means
class' jecha2 cult jecha 1 2
G 1800 2000 1100 3300 1750 1600
(475) (530) (351) (427) (797) (621)
Maize A 2000 1350 1000 3500 1500 1300
(629) (571) (424) (330) (885) (614)
P 700 550 250 1350 550 450
(218) (178) (109) (209) (362) (236)
Wtd. 1450 1200 750 2600 1200 1050
Mean‘5 (745) (748) (504) (1055) (883) (709)
G 850 1200 650 1200 950 900
(385) (395) (545) (395) (526) (531)
Cotton A 1100 1050 700 1400 950 900
(484) (466) (446) (370) (506) (494)
P 250 250 150 400 250 200
(129) (151) (174) (272) (139) (163)
Wtd. 650 750 450 950 650 650
Mean (508) (547) (467) (559) (524) (527)
e 500 1100 300 NA‘ NA 700
(149) (1 127) (181) (834)
Sorghum A 1000 850 250 NA NA 600
(285) (259) (323) (422)
P 50 150 50 NA NA 100
(26) (98) (25) (as)
Wtd. 550 650 200 NA NA 450
Mean (439) (752) (233) (583)
1. Season class: G = good; A = Average; P a Poor.
2. Virgin jecha Jecha soils that had not previously been cultivated.
3. 5yr cult jecha. Jecha soils that had been cultivated for five years continuously.
4. NA - not applicable. Sorghum was not grown on this soil type.
5. Weighted mean across soil types. The expected value for each soil type was

weighted by that soil type expressed as a proportion of all cultivatable soil. For jecha
the mean of virgin and 5 year cultivated expected values were used. 1 = All soil
types including river jecha. 2 = All soil types excluding river jecha

6. Weighted mean across seasons. The proportion of seasons in each class, G=.3,
A=.3 and P=.4, were used to weight the expected values for each season.

127
bepe were comparable with national averages for the CA5, 695kg ha".

(T attersfield, 1982)
and the yields on river jecha were well above the all-season averages reported for
the Zambezi Valley (Table 4.8). In contrast, the poor season yields of cotton and
sorghum were very low. These figures add support to the argument that sorghum
is not as resistant to drought as is often believed (Shumba, 1990) and suggests
why farmers continue to select maize as- their major food crop (Johnson, 1992)
even when they are considered to be outside the region of dryland farming, as
Masoka is (Whitlow, 1980). The very noticeable drop in expected yields from an
average to a poor season in each of these crops was remarkable. This implies a
steeply sloping sigmoid curve rising rapidly between a poor season and an
average season and then leveling out on (jecha soils) or declining slightly (on bepe
and river jecha) in good seasons, trends that have been shown in the literature for
jecha (Phia, 1992). There appeared to be little difference in expected yields among
soil types for cotton and for sorghum.

Expected yields were lower on cropped jecha soils than on virgin jecha soils
(Table 4.9) a trend that were consistent with the findings of Reynolds (1991).
Cotton yields also appeared to be more variable on cultivated soils than on virgin.

The most striking feature of the data derived from VR probabilities were the
great variability in yields, among soil types and among seasons (Table 4.9). This
variability was reflected in the literature for the eastern mid-Zambezi Valley (Table

4.8). Mean maize yields derived from VR knowledge were similar to those reported

128
in the literature (Table 4.8). If we take the cultivated jecha yields as a yardstick,

(these soils being the most comparable to those of the eastern mid-Zambezi Valley
as a whole) then the VR derived yields were entirely consistent with published yield
records (Table 4.8). VR derived mean yields were also similar to the yields
households achieved (Tables 3.2, 4.7). The average maize yields reported for
Masoka appear to be slightly higher than the averages reported for the eastern
mid-Zambezi Valley and were also slightly higher than the national average, 900kg
ha", for the period 1952 to 1999 (Ashworth, 1990).

Perhaps the most useful data with which to compare the yields derived from
VR probability estimates was the frequency distribution of maize yields reported
by the Farm Management Research Section of the Economics and Markets
Branch, Ministry of Lands, Agriculture and Rural Resettlement (MLARR, 1989). This
positively skewed distribution, with a mean (for NR III, N and V sites) of 1100 and
a standard deviation of 1060kg ha'1 was close to the VR estimated distribution, in
terms of the mean and spread of the data. The MLARR data for the 1989/90
season showed a similar distribution.

The published data for average cotton yields were also highly variable with
average yields between 500 and 1800kg ha'1 being reported (Table 4.8). The VR
derived yield estimates were similar to what households achieved in good and
average seasons (Table 4.7). The VR derived poor season cotton yields were
much lower than what households achieved in the drought season of 1990/91

(Table 3.2) and less than the drought season yields reported by Harizi (Table 4.8).

129

In general, the average cotton yields predicted by the VRs were lower than the
trends reported for the eastern mid-Zambezi valley (Table 4.8) but were close to
the national average of 700kg ha“, for the period 1982 to 1989 (Ashworth, 1990).
The MLARR (1989) data for cotton yield distributions are from a smaller sample
size than the maize data (n=46 vs. n=255) so the distribution shape was not as
clear. The mean of 630 and standard deviation of 509kg ha'1 were, however, very
similar to the VR derived yields.

The published data for average sorghum yields also showed considerable
variation (Table 4.8) with average yields ranging from 540 to 900kg ha". The
achieved, poor season yields (Tables 4.7 and 3.2) were higher than those
predicted by the VHS (Table 4.9), but the great variability made detecting true
trends difficult. The poor season yields achieved by Masoka households (Table
4.7) were closer to those predicted by the VRs (Table 4.9) than were the mean
yields published in the literature (Table 4.8). The average season sorghum yields
achieved by households and those predicted by VRs were very similar. In general.
the sorghum yields households were likely to achieve in Masoka were lower than
yields reported in the literature but were higher than the national average of 350kg
ha" for the period 1982 to 1999 (Ashworth, 1990).

Deriving statistically satisfying tests of our hypotheses from these yield data
was difficult. Inconsistencies in VR responses to field mapping yield estimates
indicated that this method may bias the results. The consistency in yield estimates

derived from two separate data collection exercises suggested that this method

1 30
did not bias the results. Whilst collecting these yield data, I brought to the attention

of the VHS the inconsistencies in yield estimates among methods. After much
debate and reflection, the VHS re-examined each of their yield possibility diagrams
and confirmed that these results represented their best estimates of yield
relationships. These factors made us confident that the method was unbiased and
provides reasonably accurate and precise information on yields under the
specified conditions. Over the range of conditions investigated, these results are

likely to be at least as accurate and precise as published data.

Performance measures

VRs were asked to develop a spidergram that would indicate what factors
determine a household’s ability to satisfy its needs (Figure 4.17). Knowledge was
considered the most important factor (RlW=.42) followed by the availability of labor
(RlW=.25) and access to good soils (RlW=.25). Wealth (RIW= .08) was considered
least important. Each of these factors was then further defined in secondary, and
in some cases tertiary, levels of indicators. Factors were included if households
could be observed to posses or not posses that factor. Whilst developing Figure
4.17 VRs focused much of their discussion on the importance of, and their ability
to observe or measure, a household’s knowledge. The harvest component of
knowledge involved establishing whether households had good harvests or not.
Households consistently identified as the community’s best farmers were identified

as such due to their high total production. When, as a group exercise, the yields

131

.mooo: 9mg mg 9.35:8 b.2525: o .o 5852. 55 mmmmmm 2 mosmamE 3:25.55 32.50 m> N 5. 2:9".

            

 

582» .8...
.80..
._ so“. >_ so“. Mm. \
unis“. BIS“. # mm. 8838
/ \ K 9358
mm. NO bogub 0cm \ £905
1 Lone. meg: .
52m .
cocom bopcooom
/ c_ 5.5.20
N F .

259 52a / /
m w.

m__0m
88 its. 583 .

 

  

mm.
9.52. \ .5282.
cozooamzom
mode:
.3525
895 8
.0 382m new . \
5522 3.55: ll: %
mm.
8.3 cm. .
baa maps—.5 Eco.
8.8 558.95 _

8582 55¢ 5 658.39
EoEmmocoE

132
per hectare of some of these respected farmers were evaluated, their per hectare

production was found to be much lower than many other producers. Most of the
VHS however, retained their belief that these were still the best farmers in Masoka.
Management of yields involved allocating points to a household for each month
it had grain in its granary - up to a maximum of 15 points. If households had
sufficient labor to weed their fields in good time, guard their fields from animals
and birds as well as harvest and store crop yields, then these factors contributed
importantly to that household’s ability to satisfy its basic needs. The VRs
considered a household having access to jecha soils as being more important
than access to other soils types in determining the household’s ability to satisfy its
basic needs. Education, hiring of labor and hiring a tractor for ploughing as well
as the qualities of the homestead building, were important indicators of household
wealth.

VRs were asked to assign each household in the community to a class
according to the household’s ability to satisfy its basic needs. Three classes were
defined: households that always satisfied their needs, those that sometimes
satisfied their needs and those that rarely satisfied their needs. Of the households
classified (n=98) 19% were classed as needs satisfiers, 45% were classed as
households who sometimes satisfied needs and 36% were classed as households
who seldom satisfied needs (Table 4.10). The numbers of adults differed among
these three classes (p< .01, n=98, F =6.349) as did the number of children (p< .01 ,

n=98, F =5.68) but the relationship with class was not clear. Households that

133
sometimes satisfied needs generally had fewer adults and children than either of

the other classes which had similar numbers of both adults and children. The
number of years households had been resident in Masoka was not statistically
different among household needs satisfaction classes. Households in the higher
well-being classes were more likely to employ local labor than households in the
lower well-being classes (p<.001, n=98, Spearman’s rank correlation r = -.271).
Households in the lower well-being classes were more likely to work for other
households than households in the higher well-being classes (p<.001, n=98,
Spearman’s rank correlation r = .35).

The VRs developed a list of all households in the community and identified
those who regularly employed members of other households as agricultural
workers, as well as which households were regularly employed. Twenty one
households (15%) were identified as regular employers of either school children
or members- of other households. This value is similar to the 13% of households
employing labor in Mangwende CA, reported by Shumba (1985). Thirty two
households (22%) were identified as being regularly employed by other
households within the community. Thirty seven households (26%) were considered
to be regular receivers of incomes from working for other households, from formal
employment within or outside Masoka or from both of these sources. Only nine
households (6%) were thought to receive remittances from relatives living outside
of Kanyurira Ward. The proportion of households regularly employed by other

households was almost double that reported by Jackson and Collier (1991) in their

134

 

 

 

Table 4.10 Mean (standard deviation in parentheses) number of adults, number of children
andyears residenceln Masokaforclassesofhouseholdwell-being.
Household characteristic Household well-being classes
Always Sometimes Seldom
satisfy satisfy satisfy needs
needs needs (Class 3)
(Class 1) (Class 2)

Number of adults 2.4 1.8 2.2
(0.61) (0.48) (0.79)
n=19 "=44 "=35

Number of children 5.4 2.8 4.0
(3.89) (1.82) (3.51)
n=19 n=44 n=35

Family size (adults + children) 7.8 4.6 6.2
(4.22) (2.02) (3.69)
n=19 n=44 "=35

Years resident in Masoka 15.9 10.5 9.6
(8.53) (10.91) (9.52)
n=16 n=41 n=35

Dependency ratio (children / adults) 2.2 1.5 2.1

(1.3) (1.07) (219)
n=19 n=44 n=35

Number employing local labor 10 6 5

Number employed as local labor 0 7 17‘

1. The total number of employed reported here are less than 32 households described

inthetextasthesetabledvaluesdo notinclude VaDomahouseholdsworkingfor
Masoka households but the totals reported in the text do.

national survey, whilst the proportion of households receiving remittances was very
much less than the 37% of households reported by these authors. Stanning (1989)
also reported much higher proportions of households receiving remittances in her
study in Hurungwe District (88% of households, Table 5, p12) as well as 64% of
households receiving local off-farm wages. Stanning, as well as Jackson and
Collier, reported that local off-farm incomes contributed little to total household

incomes. The 15% of regular employers reported by the VR was similar to the

135
findings of Jackson and Collier (1991) that around 10% of communal area

households dominated production and of Starining (1989) who reported 30% of
households in Hurungwe accounted for more than 75% of marketed maize
production. Households whose members were employed as agricultural workers
by other households were resident in Masoka for a shorter period of time than
households that were not regularly employed and had a higher proportion of
children to adults (dependency ratio) than households that were not regularly

employed (Table 4.11). Employing households had a slightly larger number of

 

 

 

Table 4.1 1 Mean (standard deviation in parentheses) for characteristics of households in
each of four employment classes.
Household employment classes
””8”“ °' “3mm" Employing Not- Employed Not-
emplovine employed

Number of adults 2.5a 2.09 2.0 2.1
(0.75) (0.58) (0.62) (0.68)

n=21 n=77 n=24 n=74

Number of children 4.0 3.6 4.7 3.4
(4.14) (2.78) (3.53) (29)

=21 n=77 n=24 n=74

Family size (adults + 6.6 5.6 6.8 5.5
children) (4.37) (3.05) (3.8) (3.18)
n=21 n=77 =24 n=74

Years resident in Masoka 11.3 11.0 7.6b 12.3b
(7.20) (10.84) (8.0) (10.60)

n=19 n=73 n=24 =68

Dependency ratio (children/ 1.6 1.9 2.5c 1.7c
adults) (1.5) (1.59) (2.16) (1.24)
n=21 n=77 n=24 n=74

1. Values followed by the same letter were significantly different (p<.05):

a) t=-3.766, DF=96
b) t=2.003, DF=90
c) t=-2.256, DF=96

1 36
adults per household than non-employing households (Table 4.11). These results

reconfirm the need to differentiate among households and incorporate household
interactions in our predictive analysis.

The cash and labor budget data provided by the VRs were used to develop
partial budgets for cotton, maize and sorghum (Tables 4.12 to 4.14). These tables
give a clear indication that households faced complex decision making processes
in selecting crop type, crop variety, soil and input mixes to satisfy their objectives.
In each of these budgets, the B scenario approximates what the wealthier
households might actually do. Scenario C showed the effect of changing producer
prices on this scenario for maize and sorghum. Scenario D for maize and sorghum
and C for cotton showed the effects of poor season yields. Scenario A was an all
factors base case. Using the government stipulated producer price in the maize
budget (Scenario C, Table 4.12) indicated quite clearly why no households sold
maize to the Grain Marketing Board (GMB) in the past few seasons; The returns
to. household labor were well below the local daily wage rate (232.19 day") and
households stand to make substantial losses in poor seasons. With poor season
yields and the Masoka maize price, household labor earned a little less than the
local daily wage (Scenario D, Table 4.12).

The returns to labor and to land for cotton were substantially higher than
those for maize. The returns to labor and land for cotton in a poor season were
however, very much less than those for maize. What was interesting to notice was

the relatively high return on investment for maize, in an average and poor season,

137

 

 

 

 

 

Table 4.12 Partial budgets for one hectare maize production enterprises with different input
and pricing scenarios. 1991/92 season prices.
Costs and returns ha’1
Scenario Scenario Scenario Scenario
A‘ 9 c o
FIXED COSTS
Transport to Guruve 22.00 22.00 22.00 22.00
Food 15.00 15.00 15.00 15.00
SUB-TOTAL 37.00 37.00 37.00 37.00
VARIABLE COSTS
Seed (25kg ha") 45.55 45.55 45.55 45.55
Seed transport (231.00 I bag) 2.47 247 2.47 2.47
Ploughing (23149.20 ha“) 149.20 0.00 0.00 0.00
Fertilizer (25197.50 ha") 197.50 0.00 0.00 0.00
Weeding labor (25219 day") 51.45 51.45 51.45 51.45
SUB-TOTAL 445.25 100.45 100.45 100.49
HARVEST COSTS
Bags (@3200 ea) 23.07 23.07 23.07 10.03
Total Costs 505.35 150.55 150.55 147.51
Returns 515.75 519.75 25242 355.90
NET MARGIN 31240 559.20 101.97 209.39
RETURNS TO OWN LABOR 3.22 5.15 0.95 1.95
RETURNS PER HECTARE 312.40 559.20 101.97 209.39
RETURNS TO iNVESTiiliENTz 0.52 4.10 0.09 1.41

 

1. Scenario A: Yield level 1050kg ha". hired labor for two weeding rounds, tractor
ploughed, 124 kg ha'1 each of Compound D and Ammonium Nitrate
fertilizers, household uses own labor for all activities on 0.405ha. Local
Masoka maize price used (230.78 kg").

Scenario B: As for Scenario A except hand cultivated instead of tractor cultivated, no
fertilizer and hired labor used for two weeding rounds.

Scenario C: As for Scenario 8 except producer price of maize used (230.25 kg").

Scenario D: AsforscenarioBexceptpoorseasonyield levelused(450kg ha' ).

2. Returns to investment = Gross returns / total costs.

138

 

 

 

 

 

Table 4.13 Partial budgets for one hectare cotton production enterprises with different input
scenarios. 1991/92 season prices
Costs and returns ha'1
Scenario Scenario Scenario
A B c
FIXED COSTS

Transport to Mahuwe 44.00 44.00 44.00

Food 30.00 30.00 30.00

Transport 8. cost of empty bales 15.00 . 15.00 15.00

Knapsack spray hire (for season) 15.00 15.00 15.00
SUB-TOTAL 104.00 104.00 104.00
VARIABLE COSTS

Seed (25kg ha") 30.95 30.55 30.99

Seed transport (231.00 I 10 kg bag) 247 2.47 2.47

Ploughing (25149.20 ha") 149.20 0.00 0.00

Fertilizer & chemicals (Agritex 1 acre packs 345.80 345.80 345.80

@23140.00/pack)

Weeding labor(@2$2.19 day") 10290 51.45 51.45
SUB-TOTAL 530.25 430.50 430.50
HARVEST COSTS

Harvest labor (@251900 bale") 59.13 59.13 19.23

Transport (@zsazbo bale") 105.12 105.12 3241
SUB-TOTAL 154.25 154.25 50.54

Total costs 947.05 599.95 555.23

Returns 1919.50 1919.50 591.42

NET MARGIN 1071.45 1219.55 5.19

RETURNS TO OWN LABOR 10.51 10.57 0.05

RETURNS PER HECTARE 1071.45 1219.55 5.19

RETURNS TO INVESTMENT 1.25 1.74 0.01

 

1. Scenario A: Yield level 650kg ha". Hired labor used for weeding two rounds, tractor
ploughed, cotton producer price 25292 ltg". Household labor used
exclusively for all activities on 0.405ha.

As for Scenario A except hand cultivated rather than tractor ploughed.

As for Scenario 8 except poor season yield used (200kg ha").

Scenario 8:
Scenario C:

139
compared to cotton. This suggested, that for poor households that are short of

money to invest in crop inputs, cotton was not as attractive an investment as

maize. Sorghum however, showed substantially higher returns to investment in an

 

 

Table 4.14 Partial budgets for one hectare sorghum production enterprises with different
input and price scenarios. 1991/92 season price data.
Costs and returns ha"
Scenario Scenario Scenario Scenario
A1 5 c 0
VARIABLE COSTS

Seed (7.41 kg ha") 3.71 3.71 3.71 3.71
Harvest labor (@25219 day") 27.05 27.05 27.05 27.05
Weeding labor (@25219 day") 51.45 0.00 0.00 0.00
Total costs 9250 30.75 30.75 30.75
Returns 247.50 247.50 100.39 55.53
NET MARGIN 159.75 220.20 59.53 24.55
RETURNS TO OWN LABOR 3.55 4.77 1.51 0.54
RETURNS PER HECTARE 159.75 220.20 59.53 24.99
RETURNS TO lNVESTMENT 2.05 7.15 225 0.91

 

1. Scenario A: Yield level 450kg ha". Hired labor used for harvest and two weeding

rounds, Masoka price (230.55 kg' 1) used. Household labor used exclusively
for all activities, except harvest, on 0.405ha.

Scenario 8: As for Scenario A except hired labor only used for harvesting.
Scenario C: As for scenario 8 except producer price (230.22 kg") used.
Scenario D: AsforScenarioB exceptpoorseasonyield (120kg he“) used.

average season than either maize or cotton and greater returns than cotton in a
poor season. We can clearly see from scenario C of sorghum why few households
sell grain sorghum to commercial buyers - their returns to labor, land and their

initial investment would be just a fraction of what they earn by trading within

140

Masoka. Returns to labor for sorghum were double the local daily wage in an
average season but only one quarter the local wage in a poor season.

If we take the expected yield for each season on a each soil type, multiply
it by the probability of a season type and then sum these for each season we
obtain the expected returns households fees for each crop on each soil (Table

4.15). Looking at the returns to land and labor, cotton was clearly a more attractive

Table 4.15 Expected returns (23 ha") to land, to labor and to initial investment for maize,
cotton and sorghum on each soil type.

 

Retums to land Returns to Returns to

 

 

 

(23 ha") labor (ZS investmertt
Crop Soil ha") (25 25")
Bepe 951.25 8.90 5.45
Maize Virgin jecha 802.34 7.51 4.67
Cultivated jecha 422.55 3.95 254
River jecha 1540.30 17.22 9.09
Bepe 1314.35 11.50 1.70
Cotton Virgin jecha 1559.68 13.64 1.93
Cultivated jecha 719.49 5.30 1.00
River jecha 2005.55 17.55 241
Bepe 250.79 5.09 9.14
Sorghum Virgin jecha 195.43 7.11 10.67

Cultivated jecha 72.15 1.57 2.35

crop on any soil type than either maize or sorghum and sorghum was by far the
poorest performer. The returns to labor for maize and sorghum were close for

comparable soils except maize out-performs sorghum on cultivated jecha. When

141

we look at the returns to initial investment however, a remarkably different picture
emerges: Overall, cotton was easily the worst performer and the least attractive
option. Returns to dollars invested in sorghum were far higher than maize except
on cultivated jecha where maize returns to investment were a little higher. These
findings had obvious importance for developing predictions as to what we might
expect different classes of household to grow and on what soils. We might expect
the poorest households to focus on sorghum production because they may still
achieve a respectable return on their investment. We also might expect strongly
risk averse households to focus on maize, because even in poor seasons on the
poorest soil, maize yields positive returns to land, labor and to initial investment.
Both cotton and sorghum however, might yield negative returns in poor seasons
on cultivated jecha soils - the soil type that comprised most of the average
household’s land holdings. Households that try to maximize their incomes and are
not labor constrained might focus on cotton as the returns to land and labor for
this crop are far higher than for either maize or sorghum.

To summarize, the performance criteria developed by the VR and extended
in the partial budget analysis, the VRs identified household knowledge, the
availability of labor to a household, a household’s access to good soils and the
wealth of a household as being key determinants of a household’s ability to satisfy
its basic needs. Of these factors, knowledge was considered almost twice as
important as any of the other factors. Households in the community were classed

into one of three well-being classes; those who always satisfied their needs, those

142
who sometimes satisfied their needs and those who seldom satisfied their

needs. Households in the first of these classes were more likely to be employers
of local labor whilst households in the last of these classes were more likely to be
local farm wage employees. Employed households had generally lived in Masoka
for fewer years and had a greater number of dependents than employing
households.

The data presented in Table 4.15 provide support to the VR assessment of
the importance of soils and labor; households that only had access to cultivated
jecha fields were less likely to be able to satisfy their needs than those who had
access to all soil types, particularly river jecha and bepe. Households that were
labor constrained might not be able to either clear new fields so that exhausted
jecha soils could be rested or they may not be able to meet the higher labor

demands of cotton or maize. Both factors would reduce their ability to satisfy

needs.

Problem formulation
In this section, the needs that VRs indicated households attempted to satisfy
were quantitatively presented. These data were important in that they were used
to set the household objectives to be incorporated in latter simulation analyses.
The detailed needs for an average woman, an average man and an average
child are presented in Tables 4.14, 4.15 and 4.16. The importance weights placed

on these needs indicated which needs a person cannot live without (weight = 5).

143
to which needs are essentially luxury (weight = 1). The weights assigned to each

month reflected the periods of greatest scarcity (weight = 1) to least scarcity
(weight = 5). For both groups maize was the principle food with chickens and
pumpkins also being important. There were some notable inconsistencies when
comparing these data and the RIW associated with needs identified in Figures 4.5
to 4.8. In their needs calendar, female VRs identified both fresh vegetables and
sorghum as being more important than pumpkins (Table 4.14) but in their needs
spidergram, female VRs identified vegetables and sorghum as being less important
than pumpkins (Figure 4.6). The spidergrams were developed in mid-March, a

period towards the end of the growing season but where pumpkins and pumpkin

12

leaves were readily available and fresh vegetables are becoming available. The

needs calendars were developed in late June when all forms of fresh food were
becoming scarce. The shifts in perceived relative importance of these items is
therefore, not surprising. This suggests that the importance weights that VRs
associated with needs may be seasonally biased, particularly in regard to
seasonally available inputs to needs satisfaction.

Female VRs suggested that women needed (used) a greater diversity of
food types and needed more of each food item. Males required about one third
more cash each year than women (Table 4.16). Children had approximately the

same food needs as men but required only about half quantity adult males

 

Pumpkin leaves are an important vegetable relish; a side dish used to garnish the staple
food of maize meal porridge.

144

3§§§39

2555855253551..o§§E255585umEmas85§E_d
55255925» iconocoumfimasaoaom;

 

OZ.
no.
0..

5
8'

mp

559:9:

9"“7 99* have as

88

0.. or

V 0..
9. mw

NM

88

BIDNIO

N

cw

ON

V

88 5101'-

0..

80

“’8

O
P

'88

ase~~v

9.
Op
on
m

N

mw

ON

mp
mp
m

Ni!)

88

mmIOIOQVVVVVV‘COGCONNv-v-v-Pv-

a... 5...:

a... moofiEo...
2095.530
25.0.2552,

a... 95.5
85a

mmmm

a... 8998 503m
maidEbm
.85.. 89m
.3. 8590

.5... E398
030E .520

a... seem

a... fies.

.923. 85.5.5
a... 8.3559, been
a... 5m

.52... do) 28.“.
ad :80

a... 555.2

 

45.0...

5:05

(It)

2|!)

LL

filo

E

.5803 0935.6 :3 .0. 92903 3.238 memo: 3.55:. ecu $550.03 0052.038. so... .800: 3.5.5.2

meooz
A- Is 339.com

m We 03o...

145

.Sozgzoncobucubaxa. u 9.965835 222.03: um £92535:an d
:9.» 22.2.3 .1. P .5305 u m .3903 323$ ..

 

5
mp

mu m m 9.8 m
h h:

IDIDQMNNNCOVNPPO

c O O O
ON ON ON ON ON ON ON ON ON ON
mp mp m.. m—

a... 5.“.
a... 5...:
38

.9. 8988 $95
.8. 850.0

.9: Suzuki
88:. :35

.9: was.

36:3. 89.0.5
a... 8.859.. 8.5
.295. a? 58.“.

.8. :80
a... gas.

 

45.0...

(’3

o

O m v
z 0 m < q. __. 2 <

'1;

O v

2m
LLID
'50)

5:0:

.:m:. 3926 cm .6. “3:92., 3.28» 88: 25:08 new «@5599: 3:2..an :35 .88: 25:05.

382

4:5. 2.28m

h ...v 030..

.52.; 8 c8 05 233. n F .20.. .8: 205.800 a 0 20.03 85:85. .0
:90 20525 n p .585 u 0 £0.03 E0000 .—

 

146

3000. 0:00 .0030
0:0..0:..0.0>>
0.005.030

6... 05:05.90
.0... 5...:

mmmw

.0... 08.08.. .0030
.8. 09.8.0

a... .002

0.2;. 08.0.5
.005. 00> :8...
.8. :80

.0... 3.0.2

O
O O
O». on

V800

0 0. 0.

IDIOVCONO‘JNNNNPNP

 

00002
0 0 0 0 0 0 0 0 v 0 0 c A-..>>2.280

“1;

.0... 0.00.. 0" v. 0.30 000.05 :0 .0. 05903 3.0.000 0000: 25:0... 0:0 00:55.03 00:00.09... .0000: 25:22 09.0 0.00...

147
required (Table 4.18).

The RIW given to needs in the male VR spidergrams (Figures 4.7 to 4.9)
appeared more consistent with the weights given to needs identified in Table 4.17
than those of the women. Maize, chickens. vegetables and sorghum were
consistently rated as important needs.

For the most part, the data presented in Tables 4.16 to 4.18 cannot be
validated by comparative reports as. with the exception of Stanning’s work (1987a)
there are no published accounts of Zimbabwean communal area household needs
or objectives. The maize needs identified in Tables 4.16 to 4.18 were similar to the
data published elsewhere. Jayne and Chisvo (1991) reported per capita grain
requirements of 219kg while Akwabi-Ameyaw (1990) reported mean per capita
maize retentions in Mufurudzi resettlement scheme for 1984/85 season of 232kg.
Stanning (1987a) reported per capita maize grain retentions, in Hurungwe, of
about 190kg for the 1983/84 and 1984/85 seasons but only 130kg per capita in the
1985/86 season (Stanning, 1987b). The sorghum retentions reported by Stanning
(1987a) however. were very much lower than those reported by Masoka VRs.

Periods of greatest food scarcity were the months of September through
January - the period immediately prior to the growing season through to the time
of the first harvests. The male VR needs matrix showed November to be the month
of greatest scarcity. Children have a greater proportion of months with high

scarcity weights than either adult males of females (Tables 4.18. 4.16, 4.17).

148
SUMMARY AND coucwsrous

An account of the methods of identifying key actors and groups, of
establishing the needs of client groups, of identifying the components of the
system that households use to satisfy needs as well as the criteria they use for
evaluating needs satisfaction is presented. The results of applying these methods
to an analysis of the Masoka agroecosystem are described and discussed. This
study differs from reports in the literature by using local farmers to state their
objectives, identify their production system and the controls they use to achieve
their objectives.

Households attempt to satisfy cash and food needs from a broad range of
agriculturally based, as well as some non-agricultural production activities. The
relative importance of each of these needs. as well as the inputs to satisfy these
needs, were identified. Major inputs to achieving good yields of maize, cotton and
sorghum were identified as rainfall, soil, crop variety and labor. Each of these
factors, except crop variety, were analyzed in greater detail. Household
differentiation. as well the interactions among households, were shown to be
important factors to include in our predictive analysis.

VRs used possibility diagrams to indicate the probability of households
achieving specific yield levels of each of three crops, on particular soil types in any
season class. Expected yield values were derived from these possibility diagrams
and these yields compared favorably to published yield levels under similar

conditions. Despite the great variability in the yields reported by the VHS, as well

149 _
as those reported in the literature, the yield functions developed from VR

information are considered to be reasonably accurate representations of the yields
households achieve.

These yields were combined with labor input and price data to develop
partial budgets for maize, cotton and sorghum. These budgets indicated the far
greater profitability of cotton when considered from the perspective of returns to
land and labor but when evaluated from the perspective of returns to initial
investment cotton was a poor performer. From this latter perspective, sorghum
was easily the best performer, except on cultivated jecha. Soil type had a notable
effect on the returns to crops; maize generated the best returns on river jecha,
followed by bepe; cotton generated best returns on river jecha and then virgin
jecha; and sorghum returns were best on virgin jecha. Declines in soil fertility on
jecha soils generally more than halved the returns from a particular crop.

VRs indicated that farmer knowledge, access to labor, access to good soils
and wealth were key determinants of a household’s ability to satisfy needs.

The measures used to record the VR responses to examination of their
knowledge were shown to be unbiased for all cases except in collecting yield
information. These biases were due to the field mapping method of eliciting yield
information. The possibility scoring method provided yield distributions that were
repeatable as well as being similar to published yield distributions. The pattern of
yields by season, on different soil types, were also found to conform to theoretical

expectations which increased our confidence in the methods and in the results.

150
Despite the difficulty of quantitatively testing the predictions posed earlier

in this chapter, the results of these analyses provide little justification for us to
accept the alternatives to the predictions:

1) That households are able to accurately describe the patterns or their
environment.

2) That households can correctly describe the results they are likely to
achieve using commonly used combinations of controllable inputs
with different resources and under commonly experienced
environmental situations.

We may therefore, conclude that our model of accurate and reliable local
knowledge being essential for survival in Masoka can, for the moment at least, be
accepted. We may then feel confident in using this information to develop a
simulation model with which to investigate various household, community and
policy level questions on the ability of households to satisfy their basic needs in

the future.

BIBLIOGRAPHY

Adams. J.M. 1991. Female wage labor in rural Zimbabwe. World Development, 19,
1 63-1 77.

Adelman, L 1992. Evaluating decision support and expert systems. John Wiley &
Sons, Inc., New York.

Akwabi-Ameyaw, K 1990. The political economy of agricultural resettlement and
rural development in Zimbabwe: The performance of family farms and producer
cooperatives. Human Organization. 49, 320-338.

Alreck. P.L and RB Settle. 1985. The survey research handbook. Richard D. Irwin
Inc., Homewood, IL, USA.

Anderson, I. 1986. Communal Land Physical Resource Inventory. Chemistry and
Soil Research Institute, Harare, Zimbabwe.

Ashworth, VA. 1990. In: Agricultural technology and the communal farm sector.
Background paper to the Main Report, Zimbabwe agriculture sector memorandum.
The World Bank, Washington. DC. USA.

Bawden. R.J. 1992. Systems approaches to agricultural development: The
Hawkesbury experience. Agricultural Systems. 40, 153-176.

Bawden. R.J. and R.L. lson. 1992. The purposes of field-crop ecosystems: Social
and economic aspects. In: C.J. Pearson (Ed.), Ecosystems of the world. Field
crop ecosystems. Elsevier, Amsterdam. Holland.

Boserup. E. 1965. In: The conditions of agricultural growth, George Allen & Unwin
Ltd, London, UK

Brokensha, D.W., D.M. Warren and 0. Werner. 1980. Indigenous knowledge
systems and development. University Press of America, Lanham, MD. USA.

Brunt, M.A., V.J. Clarke and C. de Greling. 1986. Tsetse area development and
land use planning in the Zambezi Valley, Zimbabwe. Food and Agriculture
Organization of the United Nations, Rome.

Campbell, 8M. and R.F. Du Toit. 1988. Relationships between wood resources

and use of species for fuel and construction in the communal lands of Zimbabwe.
Monographs in Systematic Botany of Missouri Botanic Gardens, 25, 331-341.

151

152

Campbell, B.M., R.F. Du Toit and CAM. Attwell. 1989. In: The Save Study.
Relationships between the environment and basic needs satisfaction in the Save
catchment, Zimbabwe. Supplement to Zambezia, University of Zimbabwe. Harare,
Zimbabwe.

Carter. S.A. Chidiamassamba, P. Jeranyama, B. Mafukidze, G. Malakela, Z. Mvena,
M. Muduhara, N. Nabane, S. Van Oosterhout-Campbell, l. Price and N. Sithole.
1993. Some observations on wealth ranking after an RRA looking at soil fertility
management in northeastern Zimbabwe. RRA Notes, 18, 47-59.

Chambers, R. 1983. Rural Development: Putting the Last First. Longman Scientific
and Technical, Harlow, England.

Chiduza, C.S.R. Waddington and M. Rukuni. 1992. On-farm evaluation of short
season sorghum and fertilizer for smallholder farmers in a semi-arid region of
Zimbabwe. Farming Systems Bulletin, 11, 25-33.

Chopak, C.J. 1991. Determinants of rural incomes in communal areas of
Zimbabwe: Household food security implications. PhD dissertation. Michigan State
University, East Lansing, Michigan.

Coudere. H. and S. Marijse. 1991. Rich and poor in Mutoko communal area. In:
Mutizwa-Mangiza, ND. and A.H.J. Helmsing (Eds). Rural development and
planning in Zimbabwe, Avesbury, Aldershot, UK.

Cutshall. CR. 1989. In: Masoka / Kanyurira Ward. A socio-economic baseline
survey of community households. Centre for Applied Social Sciences, University
of Zimbabwe, Harare, Zimbabwe.

Department of Meteorological Services. 1981. Climate handbook of Zimbabwe.
Government of Zimbabwe, Harare, Zimbabwe.

Department of Meteorological Services. 1975. Rainfall intensity in Rhodesia.
Government of Zimbabwe, Harare, Zimbabwe.

Derman, B. 1992. Careless development: The mid-Zambezi rural development
project. Culture and Agriculture. 44, 2-9.

Derman, B. 1990. In: The unsettling of the Zambezi Valley: An examination of the
mid-Zambezi Rural Development Project. Centre for Applied Social Sciences,
University of Zmbabwe, Harare, Zimbabwe.

Drinkwater, M. 1991. The state and agrarian change in Zimbabwe’s communal
areas. St. Martin’s Press. New York.

153

Elwell, HA 1985. An assessment of soil erosion in Zimbabwe. The Zimbabwe
Science News, 19, 27-31.

Elwell. HA. 1980. In: Design of safe rotational systems. Department of
Conservation and Extension (AGRITEX). Harare, Zimbabwe.

Elwell. HA. and MA Stocking. 1988. Loss of soil nutrients by sheet erosion is a
major hidden farming cost. The Zimbabwe Science News, 22, 79-82.

Elwell, HA and MA. Stocking. 1984. Estimating soil life-span for conservation
planning. Tropical Agriculture, 61, 148-150.

Farrington, J. and A. Martin. 1988. Farmer participation in agricultural research: a
review of concepts and practices. Agricultural Administration Occasional Paper No
9. Overseas Development Institute, London, UK

FSRU. 1985. Annual Report of the Farming Systems Research Unit. Department
of Research and Specialist Services, Government Printers, Harare, Zimbabwe.

Gil, G.J. 1991. But how does it compare with the real data? RRA Notes, 14.

Grandin, BE. 1988. Wealth ranking in smallholder communities: A field manual.
Intermediate Technology Publications Ltd. Rugby, UK.

Grant, PM. 1981. The fertilization of sandy soils in peasant agriculture. Zimbabwe
Agricultural Journal, 78. 169-175.

Harizi, KB. 1985. Zimbabwe. Mid-Zambezi Valley agricultural development project.
Agricultural production, FAO/ ADB Working Paper No. 2. FAQ, Rome.

Hasler. A.R.P. 1992. The cultural and political dynamics of Zimbabwean wildlife
resource use in the Zambezi Valley: A case study of Chapoto Ward. PhD
dissertation. Michigan State University, East Lansing.

Hawkins Associates. 1982. Mid-Zambezi Development Study. Agricultural
Development Authority, Harare. Zimbabwe.

Helmsing, A.H.J. 1991. Non-agricultural enterprises in the communal lands of
Zimbabwe. In: Mutizwa-Mangiza, ND. and A.H.J. Helmsing (Eds.), Rural
development and planning in Zimbabwe. Avesbury, Aldershot, UK.

IDS Workshop. 1989. Farmer’s knowledge, innovations, and relation to science.
In: Chambers, RA Pacey and LA. Thrupp (Eds), Farmer first: Farmer innovation
and agricultural research. Intermediate Technology Publications. London, UK

154

Jackson, J.C. and P. Collier. 1991. Incomes, poverty and food security in the
communal lands of Zimbabwe. In: Mutizwa-Mangiza, ND. and A.H.J. Helmsing
(Eds.) Rural development and planning in Zimbabwe. Avesbury, Aldershot, UK

Jayne, TS. and M. Chisvo. 1991. Unravelling Zimbabwe’s food insecurity paradox.
Implications for grain market reform in Southern Africa. Food Policy, August,
319-329.

Johnson, P.H.S. 1992. Why farmers grow maize in Natural Region V of Zimbabwe.
In: Proceedings of the second annual scientific conference of the SADCC-Land 8:
Water Management Research Programme. Mbabane, Swaziland, October 7-9,
155-163.

Khon Kaen University. 1987. Proceedings of the 1985 International Conference on
Rapid Rural Appraisal. Khon Kaen, Thailand, September 2-5, 1985.

Lan, D. 1985. Guns & rain. Guerrillas & Spirit Mediums in Zimbabwe. James Currey
Ltd, Islington, UK

Mascarenhas, J.P. Shah, S. Joseph, R. Jayakaran, J. Devararam, V.
Ramachandran, A. Fernandez, R. Chambers and J. Pretty. 1991. Participatory
Rural Appraisal. Proceedings of the February 1991 Bangalore PRA Trainers
Workshop, RRA Notes, No. 13. lntemational Institute for Environment and
Development (IIED), London, England.

Mehretu, A. and C. Mutambirwa. 1992. Gender differences in time and energy
costs of distance for regular domestic chores in rural Zimbabwe: A case study
from Chiduku Communal Area. World Development. 20, 1675-1683.

Ministry of Lands, Agriculture and Rural Resettlement (MLARR), Farm Management
Section of the Economics and Markets Branch. 1990. In: The second annual report

of farm management data for communal area farm units 1989/90 farming season.
MLARR, Harare, Zimbabwe.

Ministry of Lands, Agriculture and Rural Resettlement (MLARR), Farm Management
Section of the Economics and Markets Branch. 1989. The first annual report of
farm management data for communal area farm units 1988/89 farming season.
MLARR, Harare, Zimbabwe.

Mitchell, RC. and Carson, LT. 1989. Using surveys to value public goods: The
contingent valuation method. Resources for the Future, Washington, 0.0., USA.

Mittra, MK and PC Stickler. 1961. Excess water effects on crops. Transactions
of the Kansas Academy of Science, 64, 275-286.

155

Mudimu, G.G. Mbwanda, S. Chigume and J. Govere. 1988. Household food
insecurity in low rainfall areas of Zimbabwe: Initial findings in Mudzi, Mutoko and
Buhera communal areas. In: Southern Africa: Food Security Policy Options.
Harare, Zimbabwe, 1-5 November, 1987. 217-233.

Phia, MI. 1992. Efficient fertilizer and water use under variable rainfall conditions.
In: Proceedings of the second annual scientific conference of the SADCC-Land &
Water Management Research Programme, Mbabane, Swaziland, October 7-9,
241-257.

Redclift, M. 1987. Sustainable development. Exploring the contradictions.
Routledge, London, UK

Reh, l., Van Oosterhout, S.A.M., Campbell, EM. and R.F. Du Toit. 1989. Resource
use and agricultural production. 15-33 In: B.M. Campbell, R.F. Du toit and CAM.
Attwell (Eds.), The Save study: Relationships between the environment and basic
needs satisfaction in the Save Catchment, Zimbabwe. Supplement to Zambezia,
University of Zimbabwe, Harare, Zimbabwe.

Reynolds, P. 1991. Dance civet cat. Child labour in the Zambezi Valley. Boabab
Books, Harare, Zimbabwe.

Richards, P. 1980. Community environmental knowledge in African rural
development. In: Brokensha, D.W., D.M. Warren and 0. Werner (Eds.), Indigenous

knowledge systems and development. University Press of America, Lanham, MD,
USA.

Shumba, EM. 1990. Zimbabwe’s experience in agricultural research priority setting
for communal area households. In: Food security policies in the SADCC region.
Proceedings of the sixth annual conference of food security research in Southern
Africa. Harare, Zimbabwe, November, 1990.

Shumba, EM. 1988. Maize technology research in Mangwende, a high potential
communal area environment in Zmbabwe. Part 1: Developing a research agenda.
Farming Systems Newsletter, 34, 12-34.

Shumba, EM. 1985. Application of the farming systems research approach in
Mangwende communal area in the Murewe District. Zmbabwe Agricultural
Journal, 82, 31 -35.

Shumba, E.M. 1986a. A comparison of maize and groundnut husbandry practices
under communal area production. Zmbabwe Agricultural Journal, 83, 137-147.

156

Shumba, E.M. 1986b. Farmer maize production practices in a high potential
communal area environment: Implications for research. Zimbabwe Agricultural
Journal, 83, 175-179.

Shumba, E.M. 1984a. Reduced tillage in the communal areas. Zimbabwe
Agricultural Journal, 81, 235-239.

Shumba, E.M. 1984b. Yields of maize in the semi-arid regions of Zimbabwe.
Zimbabwe Agricultural Journal, 81, 91 -94.

Stanning, J.L. 1987. Household grain storage and marketing decisions in surplus
and deficit communal farming areas in Zimbabwe: Preliminary findings. Working
Paper AEE 1/ 87. Department of Agricultural Economics and Extension, Harare,
Zimbabwe.

Stanning, J.L 1987. Policy implications of household grain marketing and storage
decisions in Zimbabwe. In: Southern Africa: Food Security Policy Options, Harare,
Zimbabwe, 1-5 November, 1987. 329-358.

Stanning, J.L 1989. Smallholder maize production and sales in Zimbabwe: Some
distributional aspects, Working Paper AEE 2/89. Department of Agricultural
Economics and Extension, University of Zimbabwe, Harare, Zimbabwe.

Swift, J. 1979. Notes on traditional knowledge, modern knowledge, and rural
development. IDS Bulletin, 10, 41 -43.

Tattersfield, JR. 1982. The role of research in increasing food crop potential in
Zmbabwe. Zimbabwe Science News, 16, 610.

Thompson, J.G. and W.D. Purves. 1978. A guide to the soils of Rhodesia.
Rhodesian Agricultural Journal, Technical handbook No. 3.

Thompson, JG. 1965. The soils of Rhodesia and their classification, Department
of Research and Specialist Services, Harare, Zimbabwe.

Tversky, A. and D. Kahneman. 1981. The framing of decisions and the psychology
of choice. Science, 211, 453-458.

Tversky, A. and D. Kahneman. 1974. Judgement and uncertainty: Heuristics and
biases. Science, 185, 1124-1131.

Warren, D.M. 1991. In: Using indigenous knowledge in agricultural development,
World Bank Discussion Papers, No. 127. The World Bank, Washington, DC.

157

Whingwiri, E.E. and G. Harahwa. 1985. Maize varieties for the low yielding
environments in Zimbabwe. Zmbabwe Agricultural Journal, 82, 29-30.

Whitlow, JR 1980. Agricultural potential in Zimbabwe. A factorized survey.
Zimbabwe Agricultural Journal, 77, 97-106.

Zinyama, LM., D.J. Campbell and T. Matiza. 1987. Traditional household strategies
to cope with food insecurity in the SADCC region. In: Southern Africa: Food
Security Policy Options, Harare, Zimbabwe, 1-5 November, 1987. 183-205.

158

Chapter 5.

Simulatlon model: development, structure and testlng

INTRODUCTION

'All models are wrong, but some are useful.’ (Box, 1979).

To make meaningful evaluations of the performance of an agroecosystem
using performance criteria such as stability, sustainability, resilience, efficiency or
productivity, we need a thorough understanding of the structure and functioning
of the agroecosystem and the ability to predict agroecosystem behavior. In the
case of Masoka, and perhaps most communal area farming systems in Zimbabwe,
neither this level of understanding nor these predictive capabilities are, as yet, well
developed.

The development and implementation of models can be a useful method of
enhancing our understanding of the structure and functioning of complex systems
(Shannon, 1975; Dent and Anderson, 1971; Wright, 1971). They may also provide
hypotheses with which to test our understanding, both of the structure and of the

functioning of the agroecosystem. At this early stage of our investigations, our

objective is as full an understanding as is possible. We expect therefore, a model
of the Masoka agroecosystem to mirror the complexity of the real world system
(Wright, 1971).

The simulation of entire farming systems or agroecosystems is a complex
and difficult task. Difficulties arise from a) the lack of suitable biological data; b)
uncertainty in the input data and in the state of the system at any time; c) the
problem of obtaining the correct balance of detail in the various components of the
model; d) the propensity of farming systems to change; and e) uncertainty in the
influence of the controls farmers may use to achieve their objectives (Halfon, 1979;
Dent and Anderson, 1971; Wright, 1971 ). The difficulty of the task is reflected in the
literature where no published accounts could be found of farming systems
simulation models, that simultaneously simulate multiple households producing
multiple crops.

Whilst there are several examples of conceptual models of agroecosystems
in Zimbabwe (Swift et al., 1989; FSRU, 1985), as elsewhere (Fresco, 1986), none
of these models have been implemented as simulation or analytical models. Where
peasant farming systems have been simulated the models are of only a single
household or farm unit and most often use some form of mathematical
programming (Berdegue et al., 1989; Maino ef al., 1993; Ngambeki et al., 1992)
or a mixture of dynamic models and mathematical programming models
(Crawford, 1982). Dynamic models of single farms or farm enterprises have been

used in commercial agricultural analysis (Deybe and Flichman, 1991; Ungar, 1990)

159

160
but here again mathematical programming models dominate (Kingwell et al., 1992;

Kingwell and Pannell, 1987). When dealing with regional studies that require
information on micro-scale impacts Klein et al. (1989) used individual farm dynamic
simulation models to generate activities that were then fed into a regional input-
output model. Deybe and Flichman (1991) used a dynamic model (EPIC) to
generate activities that were subsequently incorporated into a linear programming
model for the region. Owsinski (1982) used a suite of linear programming models
for agricultural development planning in Poland. Despite their widespread use,
mathematical programming models are far from ideal for modeling
agroecosystems because they have only limited capabilities for simulating temporal
change in state variables and of incorporating the effects of feedback and feed-
forward processes on production (Dent and Anderson, 1971).

Missing from previous modeling studies of peasant farming systems is the
ability to examine the effects of household interactions on household needs
satisfaction and on the biophysical resources of the agroecosystem. In virtually all
of these studies, the independence of households is implicitly assumed. In
Zimbabwean communal area farming systems however, inter-household sharing
of equipment and labor is important (Steinfeld, 1988; FSRU, 1985). Equally
important is the inter-household provision of support in drought years (Zinyama
et al., 1988).

Commonly, analysts simplify the task of modeling farming systems by

assuming households to be homogenous. They are then justified in using

161
aggregated data (means) to predict household performance. The heterogeneity

of Zimbabwean communal area households has however, been well documented
(Steinfeld, 1988; Shumba, 1988; FSRU, 1985). Debrah and Hall (1989) clearly show
the dangers of using aggregated data for the development of farm plans;
aggregated data grossly underestimated the income variability faced by farmers
when selecting a farm plan. Individual farm yields were between two and a half and
eight times more variable than aggregated yields. Thus, whilst aggregated
analyses will be useful for some purposes they are unlikely to be as productive,
in terms of increasing our understanding of the Masoka agroecosystem, as
analyses that allow us to examine the behavior of individual households interacting
with other households.

Models may take many forms. There is a considerable body of literature
concerned with their classification (Naylor et al., 1966; Shannon, 1975; Steiner,
1989; Vemuri, 1978). When dealing with complex and large scale systems
simulation models are frequently identified as the most suitable approach to
anaan the system of interest (Anderson, 1974; Manetsch et al., 1971; Naylor et
al., 1966; Vemuri, 1978).

There is no generally accepted definition of simulation (Pritsker, 1979). For
the purposes of this study simulation is defined as "the process of designing a
model of a real system and conducting experiments with this model for the
purpose of either understanding the behavior of the system or of evaluating

various strategies (within the limits imposed by a criterion or set of criteria) for the

162
operation of the system" (Shannon, 1975).

Models represent a synthesis of our understanding ofthe system of interest.
The ultimate test of this understanding is our ability to accurately predict the
behavior of the system, given any set of initial conditions and system inputs. This
is typically called model validation. The terms verification and validation are
however, often used synonymously. In this analysis, verification is used to mean
the process of determining that the simulation model is working as intended whilst
validation is the process of determining whether the model is an adequate
representation of the real world system of interest (Sargent, 1982).

The problem of model validation is complex and has not been resolved
(Balci and Sargent, 1984; Naylor at al., 1966). Much of the difficulty stems from the
philosophical problems of proving that a hypothesis is true (Giere, 1991; Naylor
at al., 1966; Popper, 1968) but also from the objectives for which the model is
designed. Officer and Dillon (1968) argue that results based on classical
significance tests have no relevance to real world managerial decision making
because they are “absolutely devoid of any economic appraisal of real-world
consequences of alternative decisions". Mankin et al. (1975) suggest that, by
definition, all models are invalid (model behavior does not correspond to real
system behavior under all conditions of interest) and we should focus on questions
of usefulness, especially when the models are designed to further understanding.
These authors define a useful model as one that accurately represents some of the

system behavior under consideration and claim it is useless if it does not. In the

163

business literature, the issue of validity is extended to include credibility, or the
extent to which a simulation model and its results are accepted by the user and
are used to assist in making decisions (Law and Kelton, 1991). Law and Kelton
(1991) describe a comprehensive approach to validation based on the three stage
approach of Naylor and Finger (1967).

In this Chapter I describe a dynamic stochastic simulation model of
the Masoka agroecosystem. The model was developed to improve our
understanding of the structure and behavior of the agroecosystem and also as
part of a developing methodology for the analysis of complex agroecosytems.
More specifically, the model was designed to enable me to examine the
interactions among agroecosystem components and in so doing identify those
which significantly contributed to a household being able to satisfy needs. It was
hoped that, once fully validated, the model might assist in evaluating policy and

household decision options from both a household and a community perspective.

MODEL DESCRIPTION
Overall approach
There are an infinite number of possible models of the Masoka
agroecosystem. The model that was developed was the result of attempting to
satisfy a changing set of modeling objectives within the constraints imposed by the
abilities and world view of the modeler as well as by the available technology. In

this section, those aspects of the modeler's world view that were important

164

influences on development of the model are briefly discussed. The model was
designed to be used in Zimbabwe and therefore, was designed for use on a
personal computer.

The temporal and spatial scale at which the system was modeled was
expected to have a significant impact on the results that were produced and the
interpretations that could be made (Firbank, 1993; Meentemeyer and Box, 1987;
\Megart, 1988; Mans 1989; Wiens et al., 1986). The primary scale of interest was
the scale at which Masoka residents manage their resources. Essentially, this was
the scale of the field, usually between a half acre and one acre. Our hope was to
model the Masoka agroecosystem at three scales. Whilst the current model went
some way towards meeting this objective, the limited resolution within the
household, as well as within seasons, is less than what was originally desired.

Complexity and uncertainty were considered important properties of the
Masoka agroecosystem. A useful and valid representation of the Masoka
agroecosystem was expected to incorporate these characteristics. Complexity was
defined as the property of having interacting components with non-trivial detail at
several scales. Uncertainty was defined as the inability to know or predict the
outcome of some event or action, with accuracy, before the event had occurred
(Anderson, 1991).

Spatial heterogeneity and spatial interactions among agroecosystem
components were believed likely to be important determinants of agroecosystem

behavior (Baudry, 1993; Sauget and Balent, 1993). The model was therefore,

165
designed so that these agroecosystem attributes could be included where

appropriate. Spatial interactions were not however, included in this version of the
model. 1

As discussed in Chapter 4, knowledge of the goals, objectives and
experiences of Masoka residents were essential building blocks for our
understanding of the agroecosystem. The model was, therefore, developed to
incorporate both local farmer knowledge as well as formal scientific knowledge.

The models presented in this chapter were the result of numerous false
starts, the exploration of several paths of investigation and multiple iterations of
model development. These were essential precursors to the models presented in
this chapter. Not all of the capabilities that were originally planned were
incorporated into the model but where these were considered important, the

model was designed so that these aspects could be added in the future.

Model overview

The Zambezi Valley Agroecosystem Model, (ZVAM) comprised three main
sub-models (Figure 5.1): An initializing model (INITIAL), a rainfall generating model
(RAINGEN) and an agroecosystem simulation model (MASOKA1). Both INITIAL
and MASOKA1 used ASCII data layers (maps) developed using a geographical
information system (GIS). GIS data layers were in raster format with cells (pixels)

of 64m. INITIAL was used to generate a number of households, each with a

166

 

A ”Fan—#30 —'

.._2<>N. _0005. E0.0>00000.m< >0=0> _N0nE0N 05 .0 0.:0can00 .90.). Om 0.3.”.

0.00 0.9.003... 0.00 0.9.0030...
22.0. 0.00 0.0 0.0.0. 800 0.6

JMDOE .5002

 

 

 

 

F<¥Ow<2 l...<Ez.

_.aIIJL al.. \T/WAII—IIL am I
_l_ I“ a. WW“.

Juno:
zuGzim
5i i i 50.0

_ =55!"—

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

167

randomly generated number of adults, children, upland13 cells and riverland
cells.
INITIAL also randomly placed households on the Masoka landscape and located
household fields around the household site (non-riverland cells) and along the
river. Output files from INITIAL were used as input files to MASOKA1 (Figure 5.1).
RAINGEN was used to generate an annual rainfall amount based on a
gamma distribution. The rainfall was written to an ASCII file for use by MASOKA1.
MASOKA1 read in household and biophysical data from the files created by
RAINGEN and INITIAL as well as the slope file created in the GIS. Up to 300
households could be simulated at a time over an area of about 1500 ha and with
an annual time step. The population of households and the number of adults and
children in a household were kept constant within any simulation. The number of
fields households owned also remained constant in any simulation. Households
could vary the number of their fields they planted in any season and the crops
they planted. Values in each cell (n=3695) of the landscape (soil erodibility, crop
yield, yield potential, crop cover) were updated once each growing season
(October to April). MASOKA1 used an erosion model developed by Elwell (1980)

and estimated erosion for each cell.

 

13 Upland cells were those jecha cells inside the fenced area and on the older alluvial terraces.
Riverland cells were those lands outside the fence on the most recent alluvium.

168
Rainfall generating model (RAINGEN)

Fifteen years of rainfall data from Angwa Bridge were used to generate the
two parameters (alpha and k) required to describe annual rainfall with a gamma
distribution. The algorithm of Naylor at al., (1966) was used to generate a gamma
random variable with mean = 725mm and k = 10.9. '

Naylor et al. (1966) suggested estimating alpha by setting alpha equal to
the mean annual rainfall divided by the variance in the mean. For the Angwa
Bridge data this yielded an alpha value of 0.01 1866. These authors also suggested
setting k equal to the square of the mean annual rainfall divided by the variance,
yielding 8.6 for the Angwa Bridge data. Hutchinson and Unganai (1992) provide
a regression equation (r2 = 0.832) to estimate the k parameter of the gamma

distribution for annual rainfall in Zimbabwe:
k = 0.7004 + 0.0141 * MAR

Where k = the shape parameter for the gamma distribution and
MAR = the mean annual rainfall in millimeters.

This equation yields a R value of 10.9 for Angwa Bridge. Given the paucity
of rainfall data for Angwa Bridge this general function was considered preferable
to using the methods of Naylor et al., (1966) to derive the k parameter. The alpha
value used in RAINGEN was set equal to kl mean Ge. 0.015). In generating the

gamma variate the subroutine was coded to generate an integer number of

1 69
exponentially distributed variates. The integer was set equal to the rounded value

of k (is. 11).

Inltlallzlng model (INITIAL)

The purpose of INITIAL was to generate households with size and land
holding characteristics from known distributions and then to randomly place these
households on a GIS data layer of Masoka that could be read into MASOKA1 as
an input file. INITIAL also generated random cropping histories (the number of
years cultivated) and vegetation cover (96) for each cell in the landscape.
Randomly generated households patterns were used to facilitate more general
analyses. As simulation results had implications, both good and bad, for individual
households it was considered inappropriate to use real household data in

simulation experiments.

Household size generation and household field allocation

Prior to running INITIAL a household site suitability data layer and a field
suitability data layer were created. The suitability index of a site for household
residence was based on distance from the Angwa river (the primary source of
water for most households), the soil type and the slope. A data layer of distances
from the Angwa river was created with 250 meter distance classes scored from

one to 10 with increasing distance from the river. Slope was re-classed into 10

170

slope classes with class one representing areas with slopes between 0 and
1.999%, class two representing areas with slopes between 2 and 2.999% and so
on. The tenth class included all areas with slopes greater than, or equal to, 9%.
Riverland, riverine, bepe and sodic soils were re-classed as zero values. The
remaining soils (jecha, elevated and mopane soils) were re-classed as follows:
Jecha soil was classed as one, elevated soil as five and mopane soil as eight.
These values reflected the analyst’s judgement as to the relative desirability of
these soils for household sites with the lowest values being best. These three data
layers were then overlaid by multiplying the individual cell values for each of these
factors. The lowest, non-zero, scores in the resulting data layer reflected the best
household sites.

A data layer of a field suitability was generated based on soil type and
slope. Areas with slopes less than 6% were scored with a one and all other areas
with a zero. Bepe, jecha and elevated area soils were scored with a one, riverland
jecha scored with a two and all other soils scored with a zero. The field suitability
index data layer was created by multiplying these two data layers (slope and soil)
together (i.e. they were overlaid by multiplying the values). Sites were therefore
classed as being zero, one or two. Non-zero classes were suitable for cultivation.
By giving riverland and non-riverland cells different identifiers they could be
differentially allocated to households.

Once executed, INITIAL searched for the lowest scores of the household

site suitability scores and placed households on those cells. Then, in a clockwise

171

direction INITIAL would search for cells immediately adjacent to the household site
that had a cultivation suitability score of one (is. were non-riverland) and were not
already owned. These were allocated to the household being initialized by giving
the cell the household number in the HSESITEJMG file. If the household’s cell
quota (the number of hectares of non-riverland that a household was allocated
represented as a number of cells) was not fulfilled then the search was expanded
to the next ring of cells around the house site and once again INITIAL searched
clockwise for suitable cells. This procedure was repeated until a household’s quota
of cells were allocated or the number of search layers exceeded 20 (Le. equivalent
to 1.28km in all directions from the household site). For households that had
riverland fields, INITIAL searched row by row across the data layer until available
riverland cells (cultivation suitability score of two) were found to allocate to that
household. The data layer or map of fields for every household was saved
(HSESITEJMG) for use as an input file to MASOKA1.

The number of cells of land to be cultivated and the number of riverland
cells to be cultivated by each household were randomly generated from the
distributions observed in Masoka. The distributions of areas cultivated by number
of adults were taken from the questionnaire survey data described In Chapter 3.
The number of fields (hectares) of riverland cultivated per household were derived
from data provided by the VHS. These data were a list of all households that
owned or used riverland fields, the number of adults and children in the household

and the size of the riverland fields. The number of adults and children in each

172

household were randomly generated from distributions derived from the
questionnaire survey data described in Chapter 3. The number of non-riverland
cells that a household cultivated as well as the number of riverland cells that a
household had under its control could be increased or decreased using a

separate multiplier for each land type.

Cropping history

INITIAL allocated a randomly generated number of years of previous
cropping to each cell. These random variates were normally distributed and
generated using the function:

CULT = INT (1 + (UL - LL) * RND)

Where CULT = The number of years that cell had been previously
cultivated;
INT = A function returning a rounded integer;

UL = The maximum number of years cultivation was likely to
occur on that soil type;

Ll. = The minimum number of years cultivation was likely to
occur on that soil type;

RND

A pseudo random number from a standardized normal
distribution (zero mean and variance == 1).

The lower and upper limits were derived from the VR data.

173

Vegetation cover

Two measures were aggregated to develop a random vegetation cover
generator for each cell of the Masoka landscape. The first was an estimate of tree
cover and the second was an estimate of grass and herb cover. The following
procedure was used to estimate tree and shrub cover: A transparent, 2.5mm grid
was placed on 129750 air photographs of Masoka. Five samples each were taken
for jecha, bepe, elevated ground, mopane and hilly soil types. Four samples were
taken for riverine and two for sodic soils. Each sample comprised a block of
25*25, 2.5mm squares that were randomly selected so as to cover an area of the
soil type of interest. Within these squares the number of grid lines that intersected
over a tree or bush were counted to yield a percentage tree cover for the sample.
Mean and standard deviation tree cover values were calculated for each soil type
from these samples.

Estimates of grass and herb cover were based on field measurements
made in Masoka in October, 1992. Three transects were marked across a
cultivated, alluvial terrace and four across an uncultivated, alluvial terrace (both with
bepe and jecha soils). Transects on each alluvial terrace were between 300 and
350 meters long, parallel, and 100 meters apart. Point samples of the cover class
were made every five meters along each of these transects. Cover classes were
bare ground, litter, grass, tree or burnt ash. The means and standard deviations
of these measurements were used to develop the natural vegetation grass and

herb cover values.

174

Final cover values were generated by creating a function for each soil type

of the form:

C=GM+GS*RND+TM+TS*RND

Where C

FIND

GM
GS

TM
TS

Outputs from INITIAL

Total vegetation cover (%)

A pseudo random number from a standardized normal
distribution (zero mean and variance = 1)

The mean cover of grass and herbs (%)

The standard deviation of cover for grass and herbs
(96)

The mean cover of trees and shrubs (%)

The standard deviation of cover for trees and shrubs

(%).

INITIAL generated six output files (Figure 5.2). Three of these were GIS data

layers with values for cell of the landscape. The first of these data layer files

(HSEITEJMG) was a data layer of all household fields. The second of these images

(COVERJMG) was the vegetation cover data layer and the third (FERTILJMG) was

a data layer of the number of years each cell had been cultivated. Of the remaining

three files the first was an ASCII file (HOUSEHLDDAT) with the household number,

175

.50.m<mm.<

 

 

50.0.50 /

h<DdAImeOI _

:8

 

 

\oflfi.:.mu\
\0fi..0m>W0\

02_.w..._wm.._

I‘

 

m
\

 

 

 

00000.0 0.00» .0 50:5...

 

 

 

 

 

 

p.520

 

 

 

"00505.0 0.2.0030...

 

 

 

 

 

 

n..<_._._Z.

..._<Ez_. .89.. 83.0.... m... 5. 8... .398 9.0 59.. .00 9......

    

  

0:25.230...
0.00 0.2.0030...

     

 

.02.
a 00:0“.

 

 

 

4/\0co_m\

/ .50 ‘1
It“...
\Ocsmi

 

 

176

the number of adults and children in the household as well as the number of non-
riverland cells and riverland cells the household owned. The second output file
(STATS.OUT) was used to write goodness of fit test statistics for the number of
adults and children per household. The third output file (AREAS.OUT) was used
to record the total areas of cultivation, areas of riverland cultivation, the number of
cells cultivated, the number of riverland cells cultivated and the total number of

adults and children in the landscape.

Agroecosystem model (MASOKA1)

MASOKA1 read in five GIS data layer files (slope, soil type, field and
household sites, vegetation cover and the number of years of cultivation) and two
ASCII data files (household data and rainfall, Figure 5.3). Conceptually, the model
has seven major sub-systems or components (Figure 5.3): a STORAGE sub-
system, an ALLOCATIVE or decision making sub-system, a LABOR sub-system,
a CROP sub-system, an EROSION sub-system, an ECONOMIC sub-system and

an OUTPUT sub-system. The sequence of activities in MASOKA1 are outlined in

Figure 5.4.

The STORAGE sub-system
In the STORAGE sub-system household needs were established and
household production (maize, cotton and sorghum) was compared with these

needs. Production could be converted from one form to another based on the

m

.. 50.0052. .009... 80.0..0000050 05 .0. 00... 5050 0:0 50... .00 0.39".

202.5.
20.910030:
_ 0.202000 \%_J....Wu\

_ 20.0000 \ .
r0000 _ , I \ogdmi
_ _
0005
1 / I
fm>_»<0o._.._< _F / \g. mcoVw\

— wOéOhw

50.562205.

SO.>C§EDw
sown—Aw;
50.5008

.
SAVEDSS
SOdJImeOI
\0§.0So

\GENHDO
\08

 

 

 

 

 

 

 

, _ 5.50

 

 

 

 

 

 

l /////

 

 

 

 

 

 

 

 

 

i

 

 

“i
111

\i

 

 

. E50

AEE;

2_.mn.0m

 

 

 

  

 

 

 

 

 

_.ZO_.wO
_.mOmm...

, , 02.4.00 ‘
F<2llllo0<2 \ tlv.
02..m.:mm...

l‘

is

178

 

C START D

Initialization

 

 

 

 

 

 

 

MONTE CARLO LOOP

 

 

YEARLY LOOP
CELL LOOP

Allocate crops

l

Generate yield
Update cell and
household states

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimate erosion

 

 

 

 

 

 

 

 

Statistics

 

 

 

 

 

Household loop

Estimate labor needs ' m '1
Correct yields >
Budget calculations
Balance needs

 

 

 

 

 

 

 

 

 

 

 

Statistics

 

 

 

 

       

 

 

I
1"

 

 

 

Statistics

 

 

V

 

       

 

 

In
L
V

Figure 5.4. Sequence of major components of MASOKA1.

 

 

END

179

 

 

 

 

     

Production greater
than needs?

Yes—D Update stores

 

 

 

   
  

 

 

Store greater Update deficit and

Yes—b

 

  
 

 

 

 

 

 

 

than deficit? store
No
. Update deficit and
sufficuent to safisly Yes—D stores

   

 

 

 

 

T

 

Household classed as
deficit

 

 

 

 

 

 

 

Retum to main
program

Figure 5.5. Flow chart illustrating sequence of events in
STORAGE sub-system.

180

prices used in the model (Figure 5.5). For most simulations these were the local,
Masoka cash prices. Production in excess of needs was placed in storage, from
which there was no loss. If household needs could not be met from production an
attempt was made to meet them from storage (including using cash stores to
purchase needs). If this was not possible the household was considered deficit for

that particular need.

The ALLOCATIVE subsystem

The ALLOCATIVE subosystem allocated each household to a production
and to a tillage strategy. Fixed proportions of households were allocated to each
strategy at the beginning of each simulation. Households maintained the same
strategy for the duration of the simulation. The production strategy governed what
crops a household grew. the proportion of cells that the household planted to
each crop and household fallowing practices. Tillage strategies governed whether
a household used hand or tractor cultivation (tractor cultivation was only used on
cotton and had no impact on yields and whether a household planted with, or at
an angle to, the contour. These factors were important only in their effect on
erosion.

Four production strategies could be used in any simulation. These
strategies differed in the parameters that defined: a) the proportion of fields
households planted to each crop; b) the number of years a household cultivated

a field before leaving it fallow; and c) the proportion of fallowable fields households

181

 

C mm D

 

 

 

 

 

 

 

 

 

 

 
      
   
   
    

 

 

 

 

 

 

 

 

 

 

 

 

     
 

 

 

 

 

 

 

 

 

 

 

 

 

Set parameters
Randomly generate
test for whether or not
b have hum" For each strategy and
........................ 99'!!YR°.--.--------------
3 '1 eady for fallowin
I and not currently You Leave fallow? Yes—s»
: fallow?
No

M! v
3 Call yield
I subroutine.
. Plant sorghum? Yes—u Update household
I sorghum store.
I No
: can yield
. . subroutine.
I Plant maize? Yes—h” I t 9 household
. maize store.
I No
I Plant maize.
- . Call yield
I Fugdsozgfgrct’ent No—usubroutine.
' ' Update household
I maize store.
: Yes

i
' Plant cotton
I Call yield subroutine Update cell n
' Update household state
I cash store.

 

 

 

 

 

Retum to main
program

Figure 5.6. Flowchart of activities in the
ALLOCATIVE sub-system.

182

actually left fallow each year (Figure 5.6). These parameters were important control
variables that could be set for any simulation experiment.

Households planted a fixed proportion of their fields to each crop in each
year. These proportions were fixed for each production strategy. Whilst not a
realistic representation of household decision making the use of fixed proportions
simplified model development. The model was designed so that more complex
cropping decision patterns could be incorporated at a later date. For sensitivity
analyses only two strategies were used. income maximizers (INCMAX) and needs
satisfaction variance minimizers (MINVAR). INCMAX households produced a higher
proportion of cotton (50% of cells on all soil types), planted little sorghum (15% of
jecha cells) and used shorter fallow periods (minimum cropping period four years.
maximum of 10 years) than other households. MlNVAR households planted mostly
maize (50% of bepe and river jecha and 35% of jecha cells) with a little cotton
(50% of bepe and river jecha cells) and sorghum (35% of jecha cells) and used
longer fallow periods (minimum cropping period of four years and a maximum of
seven years).

The decision whether or not to leave a particular field fallow was only
pertinent on jecha soils; neither bepe nor riverland soils were left fallow. The
decision to leave a field fallow was a probability based on the household
production strategy and the number of years the field (cell) had been cultivated.
In Figure 5.7 the cumulative probability curves of a household leaving a jecha field

fallow are shown for each production strategy. An income maximizing (MAXINC)

183

household. for example, would not leave a field fallow before the field had been
cultivated for at least four years. Thereafter, the probability of leaving a field fallow
increased with increasing cultivation until, after ten years of cultivation the
probability of leaving the field fallow was one. Other production strategies used
different cumulative probability curves reflecting what i believed were reasoned

approximations of some likely approaches to fallowing (Figure 5.7).

The LABOR sub-system

In the LABOR sub-system available household labor for each month of the
growing season was estimated by multiplying the number of adults by 25 working
days in the month. Based on Reynold’s data (Reynolds, 1991) 44% of children
were considered to be in the age group that contributes to the labor force (10 to
20 years of age). Children were considered to provide labor equivalent to 42% of
an adult. This value was a mean of the proportions given by Reynolds (1991) for
clearing, planting, re-planting and weeding.

Once a household's fields had been planted to a crop the LABOR sub—
system identified the labor requirements (person days) in each month of
thegrowing season (October to March) based on the VR labor budgets presented
in Chapter 4. These requirements were compared to household labor availability.
if households were labor deficit and they had cash or grain balances they could
search for available labor from other households (Figure 5.8). Households that had

labor greater than their needs and were deficit in cash or maize could supply labor

184

 

 

 

Cumulatlve probabllltg
(300000000
01
l

 

 

 

0 I r r l

0 2 4 6 8 10 12
Years oi culttvatlon

+ MRXINC + MINURR + SRTIS —+— GREEN

Figure 5.7. The cumulative probability of leaving a jecha field fallow for
households using income maximizing (MAXINC), variance minimizing
(MINVAR), needs satisfying (SATIS) and environmentally protective
(GREEN) production strategies.

185

 

C sum D

Estimate household
labor needs and labor
availability

 

 

 

 

 

 

 

 

Calculate employment
potential.

 

 

 

 

Search all households
for available labor until

   

  

 

  
 

 

Sufficient deficit = 0.
cash to hire all Yes—t Update employee and
employer cash
balances and labor
stocks.

 

 

Search all households
for available labor until

 

     
    

 

 

 

 

deficit = O.
03:33:: 3:; to Yes-h First use cash and then
deficit? grain for payments.
Update cash and grain
balances and labor
No
i
Search allhouseho

for available labor until

 

 

 

cash and grain stores
are empty. , < Return to main >
First use cash and then program
grain for payments.
Update cash and grain

balances and labor

 

 

 

Figure 5.8. Flowchart of major activities in the EMPLOY
component of the LABOR sub-system.

186
to households that were labor deficit. Labor was hired from other households at

the local wage rate (232.54 day") in cash or grain equivalent calculated using the
Masoka prices for maize. Sorghum was not used to pay for hiring labor. When
labor was employed the labor balances of both the employing and the employed
households were updated.

if households were still labor deficit, either through there being no available
labor or through not having the resources to hire labor then the yields of each
crop were adjusted. Where labor deficits occurred the resulting yield reductions
were first allocated to sorghum, then to cotton and finally to maize (Figure 5.9).
Thus, if the total labor deficit was less than the total labor required for sorghum
then only the sorghum yields were corrected. If the total labor deficit was greater
than the labor required for sorghum then as much of the deficit as possible was
used on sorghum before yield reductions were made on cotton. The reduction in
yield due to labor deficits was calculated as the product of the proportion of labor

needs that were deficit times the yield reduction for that time period expressed as

a proportion.

 

  

deficit in month
'm'?

 

 
   
 

Yes

 

Establish month(s) of
labor defict(s) and call
appropriate subroutine

Reduce total
sorghum production.

Sorghum grown? Yes—h Update stores and
labor deficit.

No
Reduce total cotton
production.
Cotton grown? wk. Update stores and

labor deficit.

2 bor deficit = 0?

No
Reduce total maize
. production.
No labor deficit.

Update frequency of
labor defict households
in month.

Return to main
program

 

 

 

 

 

 

 

  
 

Yes-D

 

 

 

 

 

   

YOG-D

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.9. Flowchart of major activities in yield correction component of

the LABOR sub-system.

188

Final yield (Y,) was expressed as:

Y, = YI'W'*LD*YR)
Where Yi = the initial, uncorrected yield. LD =
the labor deficit expressed as a proportion of
the required labor.
YR = the yield reduction factor expressed as a proportion of

the uncorrected yields.

For sorghum the yield reduction factors were based on the results of
Burnside and WleS (1967). Sorghum was assumed to be planted in mid-
November. A labor deficit in December led to a 12% yield reduction. A deficit in
December and also in January resulted in a 36% yield reduction. A labor deficit in
January (but not in December) led to a 3% yield reduction.

Yield reduction factors for cotton were derived from Buchanan and Burns
(1970) and Schwerzel and Thomas (1971). Cotton was assumed to be planted at
the beginning of December. A labor deficit in December reduced yields by 13%.
A labor deficit in December and in January resulted in a 54% yield reduction.
Labor deficits in December, January and February reduced yields by 100% as did
labor deficits in December, January, February and March. A labor deficit in January
(but not in December) reduced yields by 56%.

Weld reduction factors for maize were derived from Hall et al. (1992). Maize

189

was assumed to be planted in November after the first rains. A labor deficit in
November had no effect on yields. A labor deficit in December reduced yields by
10%. A labor deficit in December and January reduced yields by 35%. If there was
also a deficit in February then yields were reduced 45%. A deficit in January (but
not in December) reduced yields by 25% whilst a deficit in February (but not in

December or January) reduced yields by 10%.

The CROP sub-system

Yields for three crops, (maize. cotton and sorghum) were generated in
MASOKA1. Households planted their fields (cells) to one of these crops (or left the
cell fallow) each year (Figure 5.10). Yields. for each of these three crops, were a
function of soil type, time since fallow (on jecha soils only) and rainfall season.
Yield functions were derived from the yield possibility diagrams developed, by the
VHS, for each crop, soil and season combination.

To derive a rainfall season from the annual rainfall amount generated by
RAINGEN membership functions for good, average and poor seasons were
derived, using fuzzy set membership functions. Good. average and poor seasons
were considered to be fuzzy sets. The membership functions described the
relationship between annual rainfall amount and the degree of membership in each
of these sets. Annual rainfall amounts had graded membership, from zero (not a

member) to one (full member) in one or more of these sets. The data on Masoka

 

190
(mm)
Determine degree
of membership in
season.
Generate p value

based on season
membership

 

 

 

 

 

 

 

Calculate yield as a
function of p.
Convert from per
hectare to cell value.

 

 

 

 

 

Update household
production store and
counter of cell
cultivation history.

Retum to main
program

Figure 5.10. Flowchart of major activities in CROP sub-
system.

 

 

 

 

 

191

resident classification of each season (Chapter 4. Table 4.2) were used to derive
the membership functions for good and poor season. The proportion of
respondents allocating each rainfall year to a particular season class (good.
average or poor. Table 4.2) were plotted against Angwa Bridge rainfall (Figure
5.11). Clear peaks were discemable for a good season (1000mm) and a poor
season (500mm) but the trends for an average season were not at all clear. These
peaks were used (to define the boundary of rainfall amounts that were definitely
members of the set of either good or poor seasons. Thus. for any rainfall
amountless than or equal to 500mm the membership of that rainfall amount in the
set of poor seasons was set equal to one (ux

poor
greater than or equal to 1000mm the membership of that amount in the set of

= 1). For any rainfall amount

good seasons equalled one (uxgood = 1). For any rainfall amount greater than
500mm but less than 600mm the degree of membership of that rainfall amount in
the set of poor seasons declined according to the membership function fitted to
the curve shown in Figure 5.12. Initial tests of model yield predictions found good
season yields to be consistently more than predicted yields when comparing the
simulated means with those derived from the VR distributions. To reduce this
deviation from the expected means the value at which uxgood = 1 was increased
to 1124mm. With this correction simulated mean and median yields were closer to
the expected values. For any rainfall amount less than 1124mm but greater than
767mm the degree of membership in the set of good seasons declined according

to the function fitted to the good season curve shown in Figure 5.12. These curves

192

 

QDIS

 

respond
0 C) C)
¢ 01 ox
KKK
Irratf
/‘

 

 

 

 

oi
. FD .
(.0
xx.
#/
M,”

on

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Proport
c3

 

400 600 800 1000 1200 140
500 700 900 1100 1300
Annual rainfall 1978/79 - 1991/92

 

+ Good —B— Hveraqe + Poor

Figure 5.11. The proportion of respondents allocating each growing season from
1978/79 to 19911/92 to good, average or poor seasons plotted
against the rainfall for that season measured at Angwa Bridge.
Rainfall data from the Department of Meteorological Services, Harare.

193

 

 

 

30:6 \ / \K
1... \/ \/

25K

K . .

400 600 800 1000 1200
500 700 900 1100
Ra1nial] (mm)

 

 

 

 

 

 

 

 

 

 

 

 

+ Good —5— Average + Door

Figure 5.12. Membership functions for good. average and poor seasons used in
NVUNDKA1.

194

describe the membership functions for good and poor seasons and were defined

as follows:

up”, = 5.592 - rain / 108.89 500 <= rain < 600
= 1 rain < 500
= 0 otherwise

ugood = 1 rain >= 1124
= raln / 357.14 - 2.15 767 <= rain < 1124
= 0 otherwise

Where rain = The annual rainfall in millimeters.

Because there was no clear trend with which to describe the average
season membership function the mean annual rainfall (725mm) was identified as
having a membership of one in the set of average seasons. The membership
function was then described so the lower and upper boundaries of annual rainfall
events, that were members of set of average seasons, occurred plus or minus one
standard deviation (247mm) from the mean (i.e. 478 and 972mm). The

membership function for the set of average seasons was therefore:

ummg. = rain / 247 - 1.935 478 < rain <= 725

3.935 - rain / 247 725 < rain <= 972

0 otherwise

195

The maximum membership value was used to class seasons when a rainfall
amount had membership values greater than one in more than one season.
Functions were fitted to each yield cumulative probability distribution

generated by the VHS (Chapter 4). The general form of these functions were as

follows:
YIELDi-ae '“Pwe ‘d‘P' (5)

Where YIELD = The yield in kg ha"

i,j,k = The ith crop grown on the jth soil in the kth was

season

a.b.c,d = parameters fitted to the function

e = the base of natural logarithms

p = a pseudo random number between zero and

one.

To derive the specific yield for a cell the following procedure was used. The
maximum seasonal membership value was identified and used to class the season
as good, average or poor. A random number (r) in the range (0.1) was generated
and compared to the membership value of that season. For a poor season the
probability (p) value used in the yield function equalled the minimum of the random
variable and the degree of membership of a poor season (MIN(r,upoo,)). For a

good season, p = MAX(1- gown-r). These derivations may be more clearly

196

explained with reference to Figures 5.12 and 5.13. As the p value increases the
generated yield decreases (Figure 5.13). In a poor season, as rainfall increased the
membership of that season decreased (Figure 5.12) and, logically. we expected
higher yields. Thus, by using the minimum of the membership value and r we
generated yields whose lower bound was the yield achievable for that membership
value and whose upper bound was the maximum yield achievable for that season.
In Figure 5.13. for example, if the membership of the poor season was 0.4 we
expected yields of between 2000 and 2900 kg ha". In a good season the reverse
was true. As rainfall increased. ugood increased and yields were expected to
increase to the maximum possible for that season. In this case, however, p needed,
to be smaller as ugood got bigger. I, therefore, took the maximum of one minus the
membership value or one minus r. In this case, and looking at Figure 5.13. if ugood
= 0.4 then expected yields would fall in the range 1100 to 1700 kg ha".

The logic behind the average season yield derivation is a little more
complex. With rainfall less than 725mm, um”ago increased with increasing rainfall
but then decreased again as rainfall increased beyond 725mm. Yields were,
however. expected to increase with increasing rainfall. This problem was dealt with
in the following way. Average season membership values were greater than the
poor and good season membership values between annual rainfall amounts of 568
and 873mm. A linear regression model (I2 = 0.9985. 12 DF) was developed to

describe the relationship between rainfall and k, where R was the membership

value of a dummy average season. Membership of this season was governed by

197

 

3000 R

2800 \
2600 \

'8 2400

_C \

\ 2200

2000 “F
1800 -\
1600 \\
1400

1200 \\l
1000

 

 

 

 

 

 

 

(kg

 

 

Yleld

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p

Figure 5.13. Example yield generation function from MASOKA1. Dependent

variable is yield (kg ha ) and independent variable is the random
variate p.

198
the equation (k = -1.70476 + 0.03069 * rainfall) between rainfall amounts of 550

and 875mm and was zero otherwise. After determining the R value for the average
season, the procedure for deriving a p value was much like that of deriving the p
value for a good season but with p = MAX(1-k,1-r).

For crops grown on jecha soils, soil fertility had an important effect on

yields. If a soil had been previously cultivated, yields were derived using the

following form:

YIELD = 5YR_YLD + (VIRG_YLD - 5YR_YLD) I YRSCLT

Where 5YR_YLD = The yield from a soil continuously cultivated for

five years derived from equation 1, above;

VIRG_YLD = The yield from a virgin soil derived from
equation 1, above;
YRSCLT = The number of years the soil had been

previously cultivated.

The EROSION sub-system
The Soil Loss Estimation Model for Southern Africa (SLEMSA) developed

by Elwell (1980) was used in ZVAM. SLEMSA calculates erosion (Z), in t ha'1 as

a function of three sub-models yielding K. C and X: t

Where 2

199
K*C*X

Predicted mean annual soil loss. t ha'1 yr‘1 from the
land under evaluation;

Mean annual soil loss (t ha'1 yr") from a standard tilled
field plot 30m * 10m at 4.5% slope, for a soil of known
erodibility (F) under a weed free, bare fallow;

The ratio of soil loss from a cropped plot to that from
a bare fallow;

the ratio of soil loss from a plot of length L meters and

slope percent (S), to that lost from the standard plot.

Elwell (1980) identified five control variables which were used in SLEMSA:

rainfall energy (E), soil erodibility (F), percent energy intercepted by the crop G),

slope percent (8) and slope length (L).

Rainfall energy (E) was calculated for non-guti14 areas using:

E = 18.846 * rainfall

Where Rainfall = the annual rainfall in millimeters.

 

‘4 Gutl is a light rainfall that occurs in parts of Zimbabwe but not In the study area.

200
The erodibility of soils in the study area was determined by using a rainfall

simulation method described in Elwell (1986). The erodibility factor is an estimate
of the soil loss from a bare soil at 4.5% slope. Soil samples of approximately 300kg
were collected from virgin and from cultivated sites representative of each soil type
in Masoka. Each soil sample was collected from the top 10 to 15 cm of soil in a
contiguous area or plot. Based on these simulations Elwell (personal

communication) estimated the Fb values used in the model (Table 5.1).

Table 5.1 Erodibility values for cultivated and uncultivated jecha and bepe soils in Masoka.
Source Elwell, personal communication.

 

Soil Fb
Cultivated Uncultlvated
Jecha 2

Bepe

Crop cover curves were based on measured data provided by Elwell
(personal communication) for maize. cotton and sorghum. For maize the minimum
cover values were doubled to reflect the under sowing with cucubits that was

common practice in Masoka. The following cover equations were used:

MC

48.02 + 0.00608 * malze_yleld

CC = 46.97 + 0.0043 * cotton_yleld

201

SC 69.1 + 0.005366 * sorghum_yleld

Where MC, CC and SC are maize, cotton and sorghum cover values (%) and the
yield values are expressed in kg ha".
From these crop cover curves the soil loss ratio (C) was estimated using

the following equations (Elwell. 1980):

c = e'°-°‘*' i < 50%
= (2.3 - 0.01*l) / 30 50% <= i
Where i = the crop cover value (MC. CC or SC) indicating the

percentage energy intercepted.

Slope values were determined for each cell using the lDRlSl GIS system.
Ten meter contours, over the entire study area, were digitized into IDRISI. This
vector file was converted into a raster file (with pixels 64 * 64 meters) and the
lDRlSl lntercon routine used to create a digital elevation model (DEM). A mean
(low pass) filter was used to smooth the image. The SURFACE routine in lDRlSl
was then used to calculate the slope of each 64 meter cell. Slope length was
considered to be the length of one side of a cell (64m).

The tillage practices of concern in Masoka were hand cultivation and tractor
ploughing. Elwell (1980) provided data used to update the Fb values based on

tillage type and planting and ploughing directions. For the EROSION sub-system

202

the soil loss from the previous year factor, the planting and ploughing direction
factors, the land rough-ploughed only factor and an estimate of hand cultivation
factor were used. According to Elwell (personal communication) hand cultivation
(planting of seeds into holes made with a hand hoe) would increase the soil
erodibility compared to a soil with a fine powdery tilth. The hand cultivation factor
was therefore set to -0.75.

From these values the F factor is estimated as:

F = Fb + tf + ppf + prvf
Where ti = the tillage factor set = 1 if tractor ploughed and to -0.75 if
hand cultivated;
ppf = the ploughing and planting factor, set to 0 if planting and

ploughing carried out on or level to the contour, set to -0.25
if these operations at an acute angle to the contour and set
to -0.5 if these operations at right angles to the contour;
prvf = the previous years erosion factor, set to 0 if the previous
years erosion was < 10 t ha", set to -0.5 if the previous years
erosion was between 10 and 20 t ha'1 and set to -1.0 if the

previous years erosion was greater than 20 t ha”.

203
The K value in the EROSION sub-system was thereafter estimated using (Elwell,

1980):

K = o((0.4631 + 0.7663*F) * In (E) + 2.884 - 3.1209 . 5

Where F = The soil erodibility factor defined above.

E = The rainfall energy defined above.

The final factor, X, was estimated using the following equation (Elwell. 1980):

x = L” * (0.76 + 0.53s + 0.07682) / 25.65 s > 4%
= s*i.°-5/ (10.7425 + 0.030) 1% <= 3
Where L = The slope length in meters:

S = ‘ The slope angle expressed as a percent.

Erosion was estimated for each cell in the Masoka landscape and then
summed to give a total landscape erosion (t ha"), a mean and standard deviation
(t ha“) as well as the mean, and standard deviation, per hectare of cultivated land
for each household. The sequence of activities in the EROSION model are shown

in Figure 5.14. Estimated soil loss did not affect soil properties or yields.

 

 

Yes

 

Estimate Fb

 

 

 

“Now—h
Yes Yes
i ii

 

Estimate cover as a
function of yield

 

 

 

 

 

 

 

Generate cover as
random variable

 

 

 

Read cover from
GIS file

 

 

 

 

 

 

 

Estimate tillage
practice effect and
previous erosion effect

 

 

 

l

Estimate F
Estimate K

 

 

 

 

 

 

Estimate % rainfall
energy intercepted (C)

 

 

 

 

 

Estimate X

 

 

 

 

 

Estimate erosion (Z)
Z a K * c ‘ X
Convert to cellular
erosion

 

 

Retum to main
program

 

 

D

Figure 5.14. Flowchart of major activities in EROSION sub-system.

 

 

205
The ECONOMIC sub-system

I assumed that households were able to sell all of the cotton they produced,
to the Cotton Marketing Board (CMB), as well as all of the maize and sorghum
they wished to sell. Maize and sorghum were only sold within Masoka and were
traded at the 1990/91 season prices used in Masoka: Z$0.78 kg'1 for maize and
Z$0.55 kg'1 for sorghum. Labor was valued at the local (Masoka) daily wage which
varied between Z$2.00 and Z$3.44 per day. The average value (Z$2.54 day") was
used as a constant in the model. Hired labor could be paid in cash or the grain
equivalent of the daily wage. All households were assumed to face the same costs
except INCMAX households that hired a tractor for ploughing had to pay the going
rate of Z$148.00 ha'1 for ploughing. Households growing cotton were assumed
to use the cotton packs supplied by Agritex which included all fertilizer and
chemicals for 0.4096 ha (one acre). In allocating crops to a cell (ALLOCATE sub-
system) a test was made to determine if the household had sufficient cash or
maize reserves to pay for the basic costs of producing one cell (0.4096ha) of
cotton. This value was set at Z$155.0015. Only the variable costs were used in
calculating this amount and it excluded weeding and ploughing costs. A zero / one
switch could be set in the model that would disallow / allow households to obtain
loans for this amount and up to a maximum number of cells. The number of cells

for which loans were obtained were recorded and later used to update the returns

 

15

This value was derived from the cotton partial budget presented in Chapter 4 (Table 4.13) and
consists of seed (2330.88) + seed transport (232.47) + fertilizer and chemical costs (25345.80)
multiplied by the hectare to cell conversion factor of 0.4096.

206

C enter 3

 

 

 

Set parameters

 

 

 

 

 

 

 

 

 

Estimate costs for
crop x using own
labor
Estimate costs for
crop x using own
labor and hired
labor

 

 

 

 

 

Estimate total returns
Calculate returns to
land, labor and
investment

 

 

 

 

 

 

 

Update household
cash store

Retum to main
program

Figure 5.11. Flowchart of major activities in ECONOMIC sub-system.

 

 

 

 

 

207

to cotton. No interest was charged on these loans. The general sequence of
activities in the ECONOMIC sub-system are shown in Figure 5.15.

A partial budget format similar to that of Chapter 4 was used to calculate
the returns to land, to labor and to initial investment for each crop. Returns to labor

were calculated as the returns to household labor and management.

The OUTPUT sub-system

The OUTPUT sub-system facilitated sub-sampling of individual households,
yields, Fb values or erosion periodically during any simulation experiment. Thirteen
output files were produced as required (Figure 5.3).

Data for each household in each season were written to the
HOUSEHLD.OUT file (T able 5.2). These data describe general household
characteristics, household needs and household production. Data for all
households or for a random selection of households could be written to this file.
If the model was being run in monte carlo mode then the same set of randomly
sampled households could be used in each monte carlo run.

Mean erosion over the simulation period, for each cell in the landscape, was
written to a GIS image file (T OTEROSJMG). The model could also write the mean
erosion of any season to a GIS image file (EROSIONJMG).

Yields, uncorrected for labor deficits, were written to file (YIELDS.OUT) for
each cell in each season. Data written to this file included the year identifier, the

season classifier, the soil type of the cell and the number of years the cell had

been cultivated.

A GIS data layer file (CROPSJMG) was generated for each season to show
what crops had been grown in each cell in that season. Each cell in the landscape
was assigned a crop identifier to indicate what crop had been grown in that
season.

If required, a GIS data layer (NEWFERTJMG) of the cropping history of the
landscape (how many years each cell had been cultivated) could be written at the
end of a simulation experiment.

The labor requirements, labor available and labor surpluses or deficits for
each month of the growing season could be written to a file (LABOUR.OUT) for
each household in each season or for randomly sampled households in randomly
sampled seasons. Also written to this file were each household’s production
strategy class and the number of cells of each crop planted by the household.

Returns to land, labor and to initial investment as well as information on the
number of cells of each crop the household planted, the number of labor days
hired out and hired in and the number of cells for which cotton loans were
received were written to the file BUDGET.OUT for each household in each season.
If required the data from a random sample of households could be written to the
file.

Two summary data files were generated (T able 5.3). In the first,
SUMMARY.OUT, data averaged over all households were produced each season.

These included season attributes as well as production, erosion and economic

209

Table 5.2 Summary of output variables written to household output file (HOUSEHLD.OUT).

 

 

VARIABLE DESCRIPTION

ADLTS Number of adults (is. 16yrs old or greater) in the
household.

CHLD Number of children (is. 16yrs old or greater) in the
household.

CELLS Number of nomriverland cells household owns.

RIV Number of riverland cells household owns.

PROD Household production strategy (MAXINC, MINVAR,
SATIS, GREEN).

TILL Household tillage strategy.

CASHND Household annual cash needs (25).

MZND Household annual maize needs (kg).

SORND Household annual sorghum needs (kg).

CASHBAL Household cash balance (23).

MAIZEBAL Household maize store balance (kg).

SORBAL Household sorghum store balance (kg).

CASH Total household cash generated for the year (23).

MAIZE Totalhouseholdmaize productionfortheyear (kg).

SORGHUM Total household sorghum production for the year (kg).

CASHDEF Surplus (deficit) of household cash production minus
cash needs (23).

MAIZEDEF Surplus (deficit) of household maize production minus
maize needs (kg).

SORDEF Surplus (deficit) of household sorghum production
minus sorghum needs (kg).

DAYEMPL Labor days hired in that season.

DAYWORK Labor days hired out that season.

LOANCEL Number of cells for which cotton loans received.

TEROS Total erosion over household fields (t he“).

CEROS Mean erosion (t ha") over household fields.

 

 

 

 

21 0
Table 5.3 Summary of output variables written to SUMMARY.OUT and to MONTECAR.OUT.

 

 

 

VARIABLE m S NT
WEAR Annual or seasonal idfiifier X‘—

MONTE Monte-carlo run identifier X

SEAS Season class (good, average. X
poor)

TOTEROS Average erosion over the X X
landscape (t ha'1)

ANCASHDEF Households that were cash deficit X X
(96)

ANMAIZEDEF Households that were maize X X
deficit (%)

ANSORDEF Households that were sorghum X X
deficit (%)

MZEHA, MZESD, COTHA, COTSD, Average and standard deviation of X X

SORHA, SORSD uncorrected yields for maize,

cotton and sorghum (kg ha")

MZCELLS. COTCELLS, SORCELLS Total number of cells planted to X
maize, to cotton and to sorghum

OCTDEF, NOVDEF, DECDEF, Percentages of households with X

JANDEF, FEBDEF, MARDEF, labour deficits in each month

APRDEF (October through April) (%)

MNMZREI', SDMZRET, MNCTRET, Mean and standard deviation of X

SDCTRET, MNSORET, SDSORET total returns to maize, cotton and
sorghum as household")

MNMZACR, SDMZACR, MNCTACR, Mean and standard deviation of X

SDCTACR, MNSORACR, per hectare returns to maize,
SDSORACR cotton and sorghum (2S ha")
MNMZLAB, SDMZLAB, MNCTLAB, Mean and standard deviation of X

SDCTLAB, MNSORLAB, SDSORLAB returns to labor for maize, cotton

and sorghum (25 ha“ 1)
MNMZINV, SDMZINV, MNCTINV, Mean and standard deviation of X
SDCTINV, MNSORINV, SDSORINV returns to initial investment for

maize, cotton and sorghum (25

ha")

MNEMPLDYS, MNWRKDYS Mean number of labor days hired X
in and hired out per household

MNCRCTMZ, SDCRCTMZ, Mean and standard deviation of X

MNCRCTCT, SDCRCTCT, corrected maize, cotton and

MNCRCTSR, socacrsn sorghum yields (kg ha")

GOOD, AVERAGE, POOR Frequencies of good, average and X
poor rainfall seasons

INITFRT, ENDFRT, CHGFRT Initial, final and change in years X X
cropped summed over all cells

TOTFB Total Fb value summed over all X X

cells

 

21 1
attributes.

In the second summary file, MONTECAR.OUT, essentially the same data
were presented except they were the averages over each year in a monte carlo
run. Data that were in the SUMMARY.OUT file but not in the MONTECAR.OUT file
were labor deficit data and household returns data. Data included in
MONTECAR.OUT but not In SUMMARY.OUT were the frequencies of good,

average and poor rainfall seasons.

METHODS

Verification, validation, usefulness

All model sub-systems and code segments were checked several times for
correct representation, logic, the accuracy of outputs, data correctness and the
correct sequencing of events. Wherever possible, analytical checks were made of
calculations made by the model. The maximum and minimum as well as mean and
standard deviation of output data for each module were used to identify incorrectly
functioning model components. For example, the ranges of possible yields derived
from VR cumulative probability distributions were be used to check that simulated
yields were within the correct range.

Given the objective of an enhanced understanding of the Masoka
agroecosystem, evaluation of model performance was more concerned with model
usefulness than with the accuracy of predictions. The model was considered to be

useful if it did any of the following:

. ‘ ‘7“
‘ "‘ -
._h.l

212
a) Demonstrated the utility (or lack thereof) of the approach to analyzing

complex agroecosystems. Did the model provide insights that would not

have been likely or possible with any other analytical approach;

b) Helped identify gaps in our understanding of the Masoka agroecosystem;

c) Enabled us to improve the management of the Masoka agroecosystem: or

d) Enabled us to make comparisons of different system configurations in a
design situation.

I assumed, a priori, that the model was an inaccurate representation of the
real world system. As far as was possible major components of the model were
tested to establish the degree to which they accurately represented the real world
processes of interest. This was difficult however, as there are no measured data
on production, erosion or on household needs satisfaction. A major part of model
validation that was not completed was the iterative evaluation of model results with
Masoka VRs and leaders as well as with development professionals actively
involved in the design, analysis and management of Zambezi Valley
agroecosystems.

The model presented here was not expected to withstand rigorous
deductive testing. Rather, it was developed as a working hypothesis with which to
focus discussion and further research. Given the complexity of the Masoka
agroecosystem the model was neither complete nor sufficiently accurate to
facilitate making reliable predictions of Masoka’s future. The model was expected

to provide useful information on the direction and perhaps relative magnitudes of

01—-.-

213
changes in the agroecosystem. The model was also expected to be able to

identify what known populations could be supported with the levels of technology
and needs used in the model.

Evaluation of the Masoka agroecosystem model broadly followed the
procedures described by Law and Kelton (1991): Input data were checked for
correctness and representativeness. The distributions of output variables
generated by the model were tested for goodness of fit to the original distributions
using Chi square, log-likelihood and Kolmogorov-Smimov tests. Model outputs
were compared to observed system data where these existed or to theoretical
distributions where appropriate.

The total area cultivated each year in a simulations was found to fluctuate
notably in the first few years of a simulation. For all simulations with MASOKA1 the
data from an initial warm up period of 20 years were discarded. The length of this
period was calculated following Welch (1983). This procedure requires estimating
and plotting moving averages, with different window sizes of the variable of
interest. The window size would be steadily increased until a relatively stable plot
of the variable of interest against time was achieved. For these calculations the
total number of cells cultivated in any year were used.

To verify that the crop yield component of MASOKA1 generated yields that
did not differ from expected yields (Table 4.9) the model was run for 50 years with
100 households on a landscape generated with INITIAL. In generating the initial

conditions the land allocation multipliers were set to one. Yield data, uncorrected

214

for labor shortages, were collected for each call over the 30 year period starting

from year 21. Mean rainfall over the 70 year period was 721mm with a standard

deviation of 243mm.

Sensitivity analysis

Sensitivity analyses were carried out to identify the sensitivity of model
response variables to changes in model parameters and inputs (factors). There
were more than 100 factors and decision variables that could have affected model
responses. Examining the effects of all of these was not possible. A 25 fractional

factorial design16 was used with 13 factors (T able 5.4). Where appropriate I

selected model parameter values so that the +1 level was the normal (standard
model) mean plus one standard deviation and the -1 level was the mean minus
one standard deviation. To facilitate the examination of a larger number of factors,
some factors were selected to represent a group of factors. The assumption being
made was that the interaction effects among these grouped factors would not
cancel each other. Where grouped factors were found to significantly influence
model response variables the effects of factors within the group could be

evaluated at a later data.

With INITIAL, two sets of initial conditions were generated. In the first

 

16 I am grateful to Professor D. Gilliland, of the Department of Statistics and Probability, Michigan

State University, for his able assistance In generating, and analyzing the results of. this design.

215

 

 

 

Table 5.4 Factors and factor levels used in sensitivity analysis of MASOKA1.
Factor +1 -1
Mean annual rainfall (mm) 972 478
Good season classification membership um s 1 um,“l a: 1 rainfall
function rainfall >2 1124 >= 1000
Average season classification Mean _-l; 1*SD Mean 1 0.5*SD
membership function
Poorseasonclassification' membership u I,-1 u r=1
function reiniall’°<°- 500 raimair°<°= 442
Percentage of children that work 59 29
Loanordividend Loan-26155 Loan=2$0
Dividend a: 230 Dividend a 23155
Households could employ labor Yes No
Maize price .00 23 kg" .26 23 kg"
Household cash needs base+.41*base base-.41*base
Proportion of households In production 80% MINVAR 20% MINVAR
strategies 20% INCMAX 80% INCMAX
Wage rate 233.07 day'1 Z3201 day'1
Land multiplier for riverland and non- 1.7 .63
riverland cells
Minimum and maximum years jecha Min 0 Min 5‘
fields cropped before leaving fallow Max 10 Max 7
1. The model included an error for these analyses. For the income maximizing
strategists the minimum number of years cultivated before a field was left fallow was

sixandnotfive.

(INIT63) the land multiplier was set to 0.63 for both riverland and non-riverland
cells and the number of households was set to 100. The vegetation cover and
cropping history maps produced in this run were used for all subsequent
sensitivity simulation experiments. In the second run (INIT 17) the land multiplier
was set to 1.7 for both riverland and non-riverland cells. The household data and

field map files were saved for use in subsequent sensitivity simulation experiments.

216

Two rainfall files were generated. For the first, the rainfall mean was set to 972mm.
Using the equation of Hutchinson and Unganai (1992) this mean yielded a k value
of 14.41. For the second rainfall file the mean was set to 478 with a k value of 7.44.
Two membership functions were used for classifying good season rainfall. In the
first, the membership function described earlier in this chapter was used. In the
second, the membership function was derived so that am = 1 when rainfall =
1000mm. Two membership functions were also used for each average and poor
season classification. For average seasons the first function was that described
earlier in this chapter. The second was fitted so that uavemge = 0 when rainfall was
less than 600mm (725 - 0.5*247) and when rainfall was greater than 850mm (725
+ 0.5*247). The poor season membership function described earlier in this
chapter was used for the +1 level of this factor and for the -1 level the function
was fitted so that upoor = 1 when rainfall = 600mm.

Reynolds (1991, Table 1.1) provides data on the number of children in each
of six classes (broken down by age and gender) from which the mean and
standard deviation of the percentage of children that worked were estimated.
Children in the 10 to 20 year-old class were considered likely to work.

The mean dividend received by households over the four year period 1989
to 1992 was Z$150 (Chapter 2). I decided to examine the relative effects of these
dividends compared to a loan of the same amount, but which was provided in the

form of inputs to cotton production. The +1 level of this factor was a loan of 23155

(sufficient for one cell) with no dividend and the -1 level was a dividend of 26155

217

and no loan.

The two levels of employment used were simply an on / off switch. At the
+1 level households were able to employ labor and at the -1 level they could not.

The maize price factor levels were set at the mean (Z$0.53 kg") plus one
standard deviation (260.27 kg") and the mean minus one standard deviation of
the government producer price (260.25 kg"), the consumer price of maize meal
(260.55 kg") and the Masoka maize price (Z$0.78 kg") were used to calculate a
mean and standard deviation maize price. The +1 level maize price was this mean
plus one standard deviation (Z$0.80 kg") and the -1 level was the mean minus
one standard deviation (230.26 kg").

To derive a standard deviation for household cash needs the data from the
questionnaire survey described in Chapter 2 were used to develop a regression
model of cotton needs (dependent variable) on family size (independent variable).
The standard error of the coefficient on family size (0.413) was used to derive the
range in household cash needs. The factor levels were set at the VR derived value
plus or minus 41%.

The VRs provided data on the number of households in each well-being
class (Chapter 4). The proportion of households in the class of households who
always satisfied needs (19%) were assumed to be income maximizers. The
remaining households were considered to be risk averse or variance minimizers.
These values were rounded to 20 and 80% and were used as the +1 factor level

for the proportion of households in each production strategy. For the -1 level these

218

proportions were reversed.

The levels of wage used in the sensitivity analysis were derived from the VR
data on the prices that were paid for weeding and the mean times to completing
a weeding task. Daily wage rates ranged from Z$1.77 to Z$3.04 day". The mean
(Z$2.54) and standard deviation (260.53) of these were calculated. Factor levels
were set at the mean plus and the mean minus one standard deviation.

The values used for the minimum and maximum number of years before a
field was left fallow were derived from my discussions with Vrs and other
community members. The ranges used in the sensitivity analysis were selected to
reflect a mean of 5.5 years with a large variance (+1 level) and a mean of 6.5
years with low variance (-1 level).

Ten monte-carlo runs, each of 50 years with a 20 year warm-up period,
were run for each combination of experimental factors. Data were collected from
all montecarlo runs. A random sample of 20% of the 100 households were saved
for subsequent analyses.

Mean effects due to each factor were computed using Yates algorithm
(Yates, 1937, described in Box at al., 1978). Tests of significance for each
response variable were made by comparing the following ratio to the t-distribution

with three degrees of freedom:

A B

7138-) iii-‘8—

 

 

219

 

where;
s. . (4302 + (00512 + @002
3
A, B... were the mean effects of factor A, B,...N on the response

variable of interest, (N=13);
ABE, CDE, 800 were the effects of runs 30,31 and 32 combined as error
terms.

The results of sensitivity analyses were presented as the mean effect of
each factor changing from the low (-1) to the high (+1) level. The sign on these
means indicated the direction of the effect. Also shown in the results tables were
the mean effects expressed as percentages of the overall mean for the response
variable of interest. To calculate the value of the response variable at the low level
one need only subtract half of the mean effect for the factor of interest from the
overall mean effect for the variable of interest. The value of the response variable

at the upper limit would be calculated by adding half of the mean effect to the

overall mean effect.

220
VERIFICATION AND VALIDATION

Rainfall and seasons

RAINGEN was used to generate 10,000 annual rainfall values. The mean
(725mm) and standard deviation (222mm) of these values were no different to
those observed for Angwa Bridge (t=-.004, DF = 9998, p>.99).

The season classifier in MASOKA1 was used to classify each season for
which there was data from Angwa Bridge (1977/78 to 19991/92). The difficulties of
comparing these classes to those derived from local knowledge have been
discussed in Chapter 4. Where we have the broadest data for classification
(1988/89 to 1990/91) from the questionnaire survey, the model and data
classifications were the same. In the period 1980/81 to 1987/88 the model and VR
classifications agree completely in five out of the eight seasons, differ by one
season class in one season and differ by two season classes in two seasons.
Given the good agreement between the survey and model classifications, and the
reasonably good agreement between VR and model classifications, the model
season classification system was accepted as being useful and sufficiently

accurate for the purposes of this study.

Initial conditions
Soil characteristics were assumed to be homogenous for specific types of
soils. The accuracy of the soil, the slope and the vegetation cover data layers were

not determined. The spatial patterns of household placement on the Masoka

221

landscape were similar to patterns observed on air photographs of the Masoka
agroecosystem.

To verify that INITIAL was working correctly it was used to generate 100
households in 10 separate and independent runs. The household data files
produced were saved and the distributions of adults, children and land analyzed.
The distributions of number of adults per household, produced by INITIAL, did not
differ from observed distributions (Mann-Whitney U = 36195.5, p>.937). The
distributions of numbers of children for each number of adults, generated by

INITIAL, did not differ from observed distributions (T able 5.5).

Table 5.5 Goodness of fit test statistics for the distribution of numbers of children per
number of adults.

 

Number of Adults Chi Square‘ '2

2.16
3.44
4.84
8.94
10.40
5.77
6.97
4.89
20.21

CDQNOTUI-hwmd

1. Critical value of the Chi square distribution for v=14, iii-.05 is 23.685. The
hypothesis that the generated data were from the specified distribution was rejected
if the Chi square values in the table were greater than this critical value.

2. Cells with fewer than five values were grouped.

222

The distributions of land for each household class (is. based on the
number of adults) were derived from only 70 data points. In many cases the data
in the tails were smoothed to give a continuous distribution of land area rather
than the discontinuous distributions described by the data. To compare the non-
riverland areas generated by INITIAL with those of the survey data, the 70 data
points were compared to the 1000 points generated by INITIAL Ouantile plots
were produced of the generated data and of the survey data (Figures 5.16a,b).
These plots show the proportion of the data that were in each quantile. The
similarity in these plots suggests that the land generation routines in INITIAL
produced land holdings from the same distribution as occurs in Masoka.

The proportion of households with riverlands (60%) was not very different
from the observed proportion (54%). The mean areas of riverland per household
generated by INITIAL were not different from those provided by the Vrs (Mann-
Whitney U = 33825.5, p>.33). The distributions of these two data sets were
however, somewhat different. The survey data were from a more positively skewed
distribution (G117 = 4.578) than the generated data (G1 = 1.353) and exhibited
a very much more platykurtic distribution (G2 = 27.712) than the generated
distribution (G2 = 0.956). With the exclusion of the one outlier in the survey data

these distributions (Figures 5.17a,b) were however, sufficiently similar to warrant

 

The G1 and 62 statistics refer to the third and fourth moments about the mean - the
skewness and kultosis of the distribution (Zar, 1984).

 

 

 

 

 

 

 

 

oa- - 06- I 4

< ( I

l- i-

305" ~ 306)- l .

8 8, |

30‘.- -1 20‘- I -I

02- - oz-l ~

OD 111 11 0011111111
01 204687 012046878

mm “W

Figure 5.16. The cumulative proportion of a) actual (n=70) and b) simulated
(n=1000) Masoka households cultivating a given area of non-
riverland.

224

 

 

 

 

 

 

 

 

 

a) b)
1.0 T I T I T ‘IO U I I V
as '- " 08 I- I -l
< < I
I- f-
3 on - . a on - ~
8 8
ga‘ " "1 20.4 I- cl
M «It 02 l- "
0.0 l 1 1 1 00 1 1
-2 O 2 4 8 8 -1 D 1 2 8
I-iarlvsrhdedlrated “WNW

Figure 5.17. The cumulative proportion of a) actual (n=70) and b) simulated
(n=1000) Masoka households cultivating a given area of riverland.

25

our acceptance of the INITIAL riverland generating routine.
In summary, the data generation functions built into the INITIAL model
produce distributions of adults, children and land that are similar to those

observed in Masoka.

Maize yields

The simulated yield statistics of maize grown on each soil for good and
average seasons (T able 5.6) were the same as the expected means derived from
VR probability distributions (T able 4.9). Poor season, simulated yields were
between 11 and 25% higher than expected means (Table 5.6). These differences
were expected. When, in the model, the season was classed as poor with
membership one, any yield value within the set of poor season yields was
possible. As the membership of a poor season declined (and rainfall increased)
the lower yields from the set of possible yields were excluded. This would result
in an increasing mean yield with increasing rainfall. The simulated yields for poor
seasons were, therefore, expected to have higher means than those of Table 4.9.

Statistics of yields for each season and for each soil type were confounded
by the proportions of seasons and each soil type used in the model compared to
those used in deriving statistics for Table 4.9. The proportion of good, average and
poor seasons in the verification rainfall data were .16, .56 and .27 compared to the
.3, .3 and .4 values used in Table 4.9. These probabilities of each season type

were used to calculate yield statistics for each soil type, based on the data of

226

 

 

 

Table 5.6 Simulated maize yields (kg ha") for each soil and in each season. Data are
means with standard deviations in parentheses, medians and the sample size.
Soil type Bepe Jecha River jecha Mean1

Season (kg ha") (kg ha" (kg ha ) (kg ha )
1800 1600 3350 2150
Good (884) (486) (401) (918)
1700 1500 3350 1900
n=131 n=433 n=248 n=812
1800 1050 3450 1950
Average (428) (423) (268) (1 139)
1750 950 3400 1550
n=473 n=1499 n=936 n=2908
850 500 1500 850
Poor (218) (191) (218) (494)
850 500 1500 700
n=212 n-674 n=421 n=1307
1550 1000 2900 1700
M98" (566) (522) (892) (1092)

1550 850 3300

n=816 n=2606 n-1605 n=5027

 

Table 4.9. The means derived for bepe, jecha and river jecha soils (1600, 1000 and
2900 kg ha") were no different to the simulated means (Table 5.6). These results
suggest that the approach to season classification and yield calculation used for
maize was appropriate and produced acceptable maize yield values.

Seasonal mean yields were expected to be more variable than soil means
owing to the dependence of jecha yields on cropping history. The mean yield of
virgin jecha and of jecha cultivated continuously for five years from Table 4.9 and
the probabilities of each soil type derived from the verifying simulations (bepe=.16,
jecha=.52 and river jecha=.32) were used to calculate seasonal mean yields. The
good and average season yields (2150 and 2050 kg ha") were essentially the

same as simulated yields whilst the poor season mean yield (750 kg ha"), as

227

expected, was lower than the simulated value.

Cotton yields

The simulated yield statistics for cotton (Table 5.7) were similar to those
derived from VR yield probability distributions. Mean yields for bepe, jecha and
river jecha (850, 700 and 1100 kg ha") calculated using the VR derived means

(Table 4.9) and the season probabilities used in the verifying simulations, were

 

 

Table 5.7 Simulated cotton yields (kg ha") for each soil and in each season. Data are
means with standard deviations in parentheses, medians and the sample size.
Soil Bepe Jecha River Mean
1 jecha1 1
(kg ha' ) (kg ha") (kg ha' i (kg ha )
1000 1000 1850 1150
Good (379) (476) (421) (471)
900 1 100 1300 1200
n=44 n=290 n=167 n=501
1050 800 1400 1050
Average (370) (430) (295) (469)
950 750 1300 1 100
n-=157 n=1015 n=558 n=1730
350 300 700 450
Poor (159) (180) (342) (308)
300 300 650 350
n=68 n=421 n=243 n=732
850 750 1200 900
Mean (452) (473) (448) (513)
800 650 1200 850
n=269 n=1726 n=968 n=2963

 

 

28

close to the simulated mean yields for each soil type (Table 5.7). As with maize
yields the simulated, poor season means were higher than VR derived means for
each soil type. The simulated yields on each soil type in a good season suggest
a bias toward yields higher than might be expected from the VR derived yield
probability distributions (T able 5.7).

The mean yields for each season were compared to those derived from the
VR yield probability distributions using the proportions of each soil type used in
the model simulations. Mean cotton yields for a good, an average and a poor
season were 1000, 1050 and 300 kg ha". Apart from the poor season mean which
we expect to be a little higher than the VR derived mean, these values are

satisfactorily close to the means derived from VR yield probability distributions.

Sorghum yields

The simulated yield statistics for sorghum, grown on jecha soils (T able 5.8)
were slightly higher than VR derived values in a good season (Table 4.9). Average
and poor season yields were the same as means derived from VR probability
distributions. The overall simulated mean for sorghum (500 kg ha") was not

different than the mean yield derived from the VR yield probability distributions

(450 kg ha").

229
In general, the simulated yields compared favorably with expectations.

These results increased our confidence in the season classification system and in
our ability to generate yield patterns that conform to the VR probability

distributions.

Table 5.8 Simulated sorghum yields (kg
ha") on jecha in each season.
Data are means, with standard
deviations in parentheses,
medians and the sample size.

 

SOII Jecha
Season (kg ha")

800
Good (337)
n=275

550

Average (268)
500

its 1 01 5

100

Poor (122)
50

n=438

500

Mean (351)
400

n=1728

 

 

As discussed in Chapter 4 there were no published measurements of actual
yields that could be used as empirical comparisons of these simulated yield

statistics. The yield values of Table 4.7 were all within the range of values that the

230
MASOKA1 model generated. The mean, poor season maize yields on jecha that

households claimed to have harvested (500 kg ha", Table 3.2) was the same as
that predicted by the model. Both the poor season cotton and sorghum yields
reported in Table 3.2 appear lower than those derived from the model but were
still within the range of yields predicted by the model. The yields of Table 4.7 were
difficult to compare with the model derived values because these statistics were
from situations where the proportions of soil types were unknown. Despite these
unknowns the model generated yield statistics for poor season maize and cotton
and average season cotton and sorghum that were reasonably close to the
observed yields presented in Table 4.7. The average season means derived from
the model, whilst appearing far higher than the average season data of Table 4.7,
were variable and we expected mean yields as low as 800 kg ha" with a high
probability. There were too few samples of good season yields presented in Table

4.7 to enable meaningful comparisons with the model.

Erosion

SLEMSA was validated in the high rainfall areas of Zimbabwe (Elwell, 1978)
but not in the low rainfall areas, such as the Zambezi Valley. The model was,
however, in general use throughout the SADC region.

Erosion results were perceived as an index of erosion risk rather than actual
erosion amounts that might be experienced under the conditions described in any

model simulation. Simulated erosion for cultivated areas of Masoka were between

281

1.4 and 9.3 t ha" year" on cropped lands. These values were reasonably close
to the erosion values estimated for each soil type by Elwell (1978), using a rainfall
simulator (0.8 and 9.3 t ha" for bepe with 3.8 and 5.7 t ha" on virgin and
cultivated jecha). The EROSION sub-system was the least closely examined of the
sub-systems in MASOKA1. In the original model design I planned to use a spatially
dependent erosion model. That aspect of MASOKA1 was not yet developed and

an unaltered SLEMSA was used as an Interim measure.

RESULTS
Sensitivity analysis

Model response variables were grouped as needs, yield, economic and
resource responses. Needs responses included the mean proportion of
households who failed to satisfy their cash (ACDF), maize (AMDF) or sorghum
(ASDF) needs each year (Table 5.9).

Yield responses were the mean yields (kg ha") of maize, cotton and
sorghum, from all fields, corrected for labour deficits, achieved over the 30 year
simulation. These means were then averaged over the 10 monte carlo simulations.
Economic response variables included, for maize, cotton and sorghum, mean total
returns to a household (MZRET, CTRET, SRREI'); mean returns per hectare
(MZHA, CTHA, SRHA); mean returns to own labour and management (MZKAB,
CTLAB, SRHA); and mean returns to initial investment (MZINV, CTINV, SRINV). As

with the yield data these were the means of all households averaged over the 10

232

monte carlo runs.

The resource response variables were the changes in the total
number of years that each jecha cell had been cultivated summed over the whole
landscape (CHGFRT), the mean erosion (t ha") over the landscape (MNERS), and
the change in the total Fb value of the landscape, or the Fb value of each cell
summed across the entire landscape at the beginning of the simulation minus the

sum of all cell Fb values at the end of the simulation. Lower Fb values indicated

a higher erodibility.

Household needs satisfaction

For the most part, the responses of ACDF, AMDF and ASDF were in the
direction and of magnitudes that were expected. The effects of rainfall and the
price of maize are obvious and require no explanation. The effects of factors such
as the land multiplier, the proportion of households in each production strategy or
the effects of the membership functions used were easily explained or the effects
were too small to warrant discussion. Increasing household cash needs had a
large impact on the proportion of households that satisfied cash and food needs.
The effect was contrary to what was expected and required a more elaborate
explanation.

At the +1 level of land availability, households attempted to cultivate, on
average, 70% more land than the real world households in Masoka cultivated.

Given that in the model households cultivated fixed proportions of the land

233

available to them, irrespective of their labor resources, it was not surprising that
more land would increase the levels of deficits as households would have had
greater labor shortages and consequently achieved lower yields. Cotton’s greater
sensitivity to weed cover was reflected in the greater response of ACDF to
increases in land availability (T able 5.9).

The mean percentage of households that were deficit for each need each
year decreased when the poor season classification changed from the low to the
high level (Table 5.9). This was expected because at the +1 level the season was
classified as poor (upoor > uaverage) at higher rainfall levels than at the -1 level.
The yield functions for all three crops showed higher yields as upoor approached
zero than when uavmae was small. Thus, at the +1 level of poor season
classification I expected higher mean yields and, therefore, a lower percentage of
households that were cash, maize or sorghum deficit. The classification of good
seasons had a similar trend for households that were cash deficit but the opposite
trend for the proportion of households that were maize deficit (T able 5.9). For the
negative response of the percentage of households that were cash deficit the
explanation was much the same as that for the poor season classification; the
yields at the high end of an average season were higher than those at the low end
of a good season. At the +1 factor level for good season classification, ugood is
dominated by uaverage over a wider range of rainfall amounts than was the case
at the -1 factor level. I expected, therefore, the mean yields to be higher at the +1

factor level and for there to be fewer cash deficit households. The explanation for

.96.$2.9seasoasesssossosacafiaooaosaaeecooesoofiaecssofiaao .0
see. .0 0.0.00 .0. as. 80 .02. 00.3. a 9% .02. 020.00 6.. 0:0 E

00.1 va a... 5.01 va 2:... 50.01 va .0 ecu 59. 0.0.00.0 so; 285. so. . 008030 00.0: 08.05.00 8:80.090 .0
seawataboaeaaoefiososs ..

 

 

 

 

0.: 0.3 0.00 000:. .065
e .0 02 02 0208.800 888 000.02
02 a: 0. 0.0. 02 000.505 30:0“.
2. 00 0.00 : cm 0.0 02 0:00.20 0 08..
wz N 0; m. .6. 2020050000 000000 0000
02 0. ha. 0 0.0 20.00.06 $32.2 .0 5082a
v. 0.0. n. E- 0. mi. £550.86 80000 .ooc
0.- 0...- 02 e 0.0. 80.. 0002
t. 09 0.00. 3+ 0.0. 0.- 0.0.- 008.. 0000
t R. «.8. 00- 0.00. 0.- 0.0.- 5.0.00
0 0.0 02 0. «.3 0.0.0.05 000..
.0000 6000
0.0 a. :8... 0.0 0. aces. «0.0 0.... 6000 :8...
00.288; 0... “.00... 00.9.83. .0... ”.052 000088.. 0... “.00.. been.
. so. .902 a. 80:05 c. .00..
coco .5000 50$ 5:093 0:0 .002... case 600$ :08 so; a... 000083.. a o... .o 08:. o... .0 0.20800 0.0 0.0:.

235

the reverse of this expectation for maize deficit households was not clear. Firstly,
the difference was slight (1%) and despite the significance (P<.05) could be due
to random error. Secondly, the difference in yields between a high uaverage
membership compared to a low ugood membership was not as great for maize as

it was for cotton.

As expected, when the proportion of households that were classed as
variance minimizers was low (level -1) the percentage of households with a cash
deficit was higher and the percentage of households that were maize deficit was
lower (T able 5.9). Income maximizers planted a greater proportion of their lands
to cotton and less to maize than did variance minimizers. At the low factor level
(80% income maximizers), therefore, I expected fewer households to have a cash
deficit but more to have a maize deficit.

The percentage of households that had a maize deficit and the percentage
that were sorghum deficit were sensitive to whether the Z$155 subsidy was paid
as a dividend or as a loan (Table 5.9). In both cases (AMDF and ASDF), the
provision of a loan resulted in a greater percentage of households with a deficit
than did the payment of a dividend, and for sorghum the provision of a loan
resulted in an almost 30% increase in deficit households. Loans were provided to
only those households that grew cotton and were provided only as inputs to
cotton. Dividends, on the other hand, were paid to all households and were not
tied to any productive activity and could therefore, be used to satisfy household

needs. With lower total inputs of cash to the community (not all households

236

receiving loans) and loans being tied to cotton production, we expected a lower
proportion of maize and sorghum deficit households with dividend payments than
with tied loans.

Household fallowing practices had a significant impact on the percentage
of households that had a maize deficit each year (Table 5.9). At the -1 factor level
the minimum number of years that a field would be cultivated before being left
fallow was five. All yields of maize grown on jecha at this factor level were based
on the five-year cultivated jecha function which had a much lower mean yield than
the uncultivated jecha function (Table 4.9). I expected, therefore, that at the +1
factor level, where yields could be based on the virgin and the cultivated functions,
the mean maize yields would be higher and fewer households have a maize deficit.
This expectation is consistent with the data of Table 5.9.

The effect of increasing household cash needs was to reduce the
percentage of needs deficit households (Table 5.9). Similarly, several other model
response variables responded to increasing household cash needs in ways that
were contrary to expectations. A tentative explanation is presented (Figure 5.18).
The argument had two components. The first was as follows: When household
cash needs were increased more cash was required to satisfy basic household

cash needs and less was available to meet the relatively high initial costs of

237

00000.00 2000 0:0 000:. 0.020000: :0
0000.. 2000 0.9.0000: 05000.0... .0 0.0020 0... .0 0.0.00.0 000. .00000 .90 0.00....

   

          
     

20.8000...
0~.0E .90...

     
     

l 0.5.00
0.00 .0.

050:.
00.0....w

      

 

        
   
   

muse:
:08 29.0030:
00020:.

0:00 .0200

 

      
           

:0 0200.0 0.2.9.096 0000.00 l
050.... .0 2000 2000
202.000.... Hi

I

 

00:0E00
.000.

  

00:00
0. 002.0...

 

0~.0E
0. 00.00....

  

- ..... - .v i
0.00000... . .
.- -- . {.0590} .

‘

0 ‘

 

- - - - . 00.0... 00000.. .

\ I

.. 00:00 .. 000.2 4....
...00_0..0000... .. .. 0.00.300... - - +
.. 5.600 .. ...--+a...

§

' ' - ‘
- ‘
I

I \
l 0 0 0

238

producing cotton- The per household area planted to cotton was therefore,
expected to decrease. This decrease would have three important consequences-
Firstly, household expenditure on agriculture would have decreased, leading to
higher household cash balances which could be used to satisfy cash needs.
Secondly, smaller areas of cotton would have resulted in lower labor demands, a
lower likelihood of labor deficits and lower labor deficit induced yield reductions.
The third. and probably most important consequence of reduced cotton planting
would have been the increased areas planted to maize, including areas of the
better soils. The greater area planted to maize, coupled with the higher average
maize yields would lead to higher total maize production and hence to lower
household maize deficits. Households with maize surpluses could sell their
surpluses to offset their cash deficits. To reduce the same cash deficit households
required between three and 10 times as much maize as they would cotton. Thus,
once household maize and cash deficits were satisfied there was not likely to be
a cash balance sufficient to support cotton production.

From this hypothesis a number of predictions were deduced. Based on this
hypothesis we would have predicted that average maize yields would be higher
when household cash needs were at the higher level. We would also predict that
there would be a lower percentage of maize deficit households when household
cash needs were at the higher level. Maize yields were higher when household
cash needs were at the higher level (Table 5.10). From Table 5.9 we had already

seen that higher household cash needs resulted in a lower percentage of

239

households that were maize deficit each year. Acceptance of this hypothesis was
contingent on acceptance of a few key assumptions. the most important of which
was that whilst households unable to produce cotton would increase the
probability of their satisfying household cash needs. they would not be earning
sufficient cash to enter into cotton production. These assumptions required further
analysis for me to have felt entirely comfortable with the hypothesis-

A second, and perhaps more plausible hypothesis (but not exclusive of the
first) was as follows (depicted by the dotted lines in Figure 5.18): Wlth increased
household cash needs, many of the more marginal cotton producers (those with
poor soils or who had a labour deficit) would be excluded from cotton production
due to their not having the capital to invest in inputs as described in the previous
hypothesis. These households would have to switch to maize production. The
remaining cotton producers would be those households that were more likely to
achieve high yields as they would not be labor constrained. They would also be
more likely to satisfy household cash needs. The marginal cotton producers (i.e.
those who were likely to be labor deficit) who had switched to maize were likely
to have been those households that were not satisfying household cash needs.
in part because of the high costs of cotton production. These households, relieved
of the burden of cotton production might have been more likely to satisfy
household needs through maize production owing to the lower costs of
production. I assume here, that these households were generally smaller (hence

the labor constraints) and therefore quite likely to satisfy household cash needs

240
from maize and sorghum production. As with the first hypothesis this one required

further investigation to be acceptable.

Although they may both be plausible, these hypotheses are related to
explaining the behavior of the model and did not necessarily provide acceptable
explanations of real world processes. It was clear that some aspects of the model
specification were contributing to behavior that we would not have expected in the
real world. In particular, households in the model planted a set proportion of their
fields to specific crops, irrespective of the labor they had available. This could lead
to households’ reducing their chances of satisfying needs by planting areas that
were larger than they could cope with. This was an unrealistic expectation which
was to be corrected in future versions of the model.

By proposing these hypotheses, I did not discount the possibility that the
model contained an error that resulted in the observed effects of increasing

household cash needs.

Crop yields

As expected. mean yields of maize, cotton and sorghum were all markedly
sensitive to changes in the mean annual rainfall (Table 5.10). The yields of maize
and cotton more than doubled with the change from a low mean annual rainfall to
a higher mean whilst the yields of sorghum more than tripled. The +1 level of
fallowing practices led to a reduction in maize yields but increased cotton and

sorghum yields. The effects on cotton and sorghum were what was expected; at

..0>0.$00.0..-0£Eo...00.0§.0.0.00.0500.0000020000000.00055000050526.3000 .0
0.00. .0 0.0.00 .0. 00. 000 .02. 00.00 .0 0.00 0.0.. 00.00... .00 000 .0
00." V0 43. 5.0" V0 .03... 50.0" V0 .0 Son 0.0.. 0.0.00.0 0.0.... 0.005 .80. . 0.000020 00.0: 0000.00.00 000000.090 .N

 

241

80008....o.0000...0.0>00£.o$ ...
000 can 003 :00... ..0.0>O
z. 0 0. z. 0 00 2. 0 00 020050000 000000 .000
:5 . 0 02 :3 0 am 00000 0000
t .v E 02 . . 0.: 00.000.00.00 000000 000.02
2:. 0. mo 02 0:. 0. 2. 002.080 00.39.00
2; 0: 0.0 :5 2 .00 2:. 00 08. .0000

 

0.0 0.. .00.... 000.2 0.0 0.. .0000 000.2 N0.0 .0. .0000 0002

 

.000 9.. 0.500 .000 9.. 0.50 .000 9.. 00.5.. .000...

 

.0003. .000. 0. 0000000 0. 6.5.0. 0500.00 000 6.50. 00000 .6052. 0000. .0 00.0... .8000 0000. .0 32.0000 3.0 0.0.»

242

the +1 level we would expect a larger number of fields cultivated for fewer years
resulting in higher yields. The reduction in maize yields at the +1 factor level was
difficult to explain. Despite the significance of the result, the response (70 kg ha")
was within the range of error I expected from model yield functions.

Mean yields for all three crops were sensitive to the average season and the
poor season membership functions used. The direction of the effects were all the
same with an increase in yields at the +1 factor level for both seasons. With the
poor season membership function at the +1 level we would expect higher yields
as at all but the lowest upoor levels the +1 level upoor dominated -1 level upoor
(area c of Figure 5.19). For the average season membership function the
explanation for higher yields at the +1 factor level was a little more difficult to
visualize but was nonetheless straightforward; when moving from the +1 level to
the -1 level the effect was to have a wider range of annual rainfall amounts
dominated by low membership values of the set of poor season yields and a wider
range of annual rainfall amounts dominated by low membership values of the set
of good season yields (areas a and b in Figure 5.19). At the +1 level areas a and
b of Figure 5.19 would have been dominated by low membership values of the
average season (the A+1 curve). Yields at low membership values of an average

season, on each soil type. were higher than yields at low membership values of

either poor or good seasons. Thus, when areas a and b of Figure 5.19

O 0 0 mm
QHLMLGDEQE Commwm

Figure 5.19

 

 

243

 

 

 

 

 

1.2
0 1 - ,r ’7-
H ' /
I
{7; d /
I /
L O a 8 A ’ /’
0’ I. r
o I /
E I ’/
(D 0. 6 ‘ ’l /
E .‘ .0 ./
.’ /
C l /’
O 0 I 4 - {-1. /’
Ul I '1.
010.2“ '
. \ b
0 1".
0

 

400 560 660 760 860 960 who 11b0 1200
Qalnlall (mm)

 

p+1 A+1 - ------- (3+1
—p-1--- 0-1- ------- 5—1

Figure 5.19. Good (G), average (A) and poor (P) season membership functions
used for +1 and -1 levels in sensitivity analysis.

(I'I

244

were dominated by the average season membership function (at the +1 level)
higher yields were achieved.

The responses of maize and sorghum yields to increases in household cash
needs have been described in the previous section and with reference to Figure

5.18.

Economic response variables

Apart from the consistent sensitivity to rainfall and household cash needs
there were no patterns of sensitivity to factor levels across all economic response
variables.

Not unexpectedly the returns to maize were sensitive to the change in maize
price (T able 5.11). Increasing the maize price from 250.26 kg'1 to 230.80 kg'1 (i.e.
208%) resulted in increases of 22% in returns to land. labor and initial investment
as well as an increase of 17% in total returns to households from maize.

Increasing household cash needs by 41% increased returns to all inputs for
maize. Given the explanatory model presented in Figure 5.18 these increases were
expected. The model of Figure 5.18 indicates higher maize yields with increased
household cash needs and therefore, higher returns to land, to labor, to initial

investment and total returns to households.

.TwNMN. .7000 0N. .~-0..|0.N. .xb.o..00:o0 0N.
>2.N.2 0<..Ns. (INS. KENS. .200...

 

. 0.059.: 000 «>2.NS¢ 0:00:80)... .325 O.
36.0 AmsN—Zv 0°95. 60‘:sz USS. 0. 00:03.05 .CrWKN—Zv 0:01.00: .GuOu 00 NCNOE 02. CO “@8530 .08. LOuONs 50 0000000 00s.». 0. .35 .sOIL.

245

.06. .+ 00.0. .-00.0.0..00.0.0..0.0.00.00.00.000000.00..0000.000E 50009005300000
0.00. .0 0.0.00 .0. 0.0. 000 .02. 00.00 .0 0.00 0.0.. 0.0.00.0 .00 000 .0.
00... v0 ..:. 5.0" v0 .2... 50.0» v... .0 0.00 0.0.. 0.0.00.0 0.0... 0.0000 .00. . 0.0.0030 00.00 0000.00.00 000000.090
00.00. .00. .0. .0000 0000. .0006 .0 0..

 

 

 

00.0 00.. 000 000 0000. ..0.0.6
.00 O. 5...? .00 0 Md. 0:. 0. _.V- c N up- 80:00.5 30.3“—
0000.00
$2 $2 a. N. o _.l a. V mm \ :81—
00.000.00.00
0.0 O hrd a N de c N mw 00 m 3 $088 08a.
: 0 000 02 . 0 0 z . 00 0.00.0... 000..
- 00.000000...
02 02 02 t v 00 000000 000.03.
0.050.000 m<>z=2
00 m hNd m2 0. N @w .25 GN :N 3 32:80.3.
0: 00 0.0 :0. k 00 z. 0. 000 :5 0.. 000 00.0.00
2; 00 00.. z. 0 0.0 2; 0 00 0: 00 00. 00000 0000
.2. «0. :0 z. .0. 00.0 2.. .0. 00. :5 t. 000. 00.... 00.0.2
.0000 .0000 .0000 .0000
0.0 0.. 000.2 0.0 0.. 000.2 0.0 0.. 000.2 00.0 .0.. 000....
...0~ 0N. .100 0N. .700 0N. .~.0.0..0000.. 0N.
>z_~s. 050.... $8... Emu... 0.00..

 

.0000... .0. $2.52. 0.0000000. .000. 0.

000 .32. .000. .252. 000. 0. 0.00.0. .5002. 0.00.0. .00. .0 00000. 00. 00 009.000 .000. .200. .0 0.00.... 0.: 2.0 0.00»

246

.000. .+ 00.0..-00.0.0..00.0.....0.0.00.00.00.000000.0000000.000.00.000000050600000 .0
0.00. .0 000.00 .0. .00. 000 .02. 00.3 .0 0.00 0.0.. 0.0.00.0 0.. 000 0..
00.» V0 0:. .0..." V0 0:... .80...- V0 .0 0.00 0.0.. 0.0.00.0 0.02. 0.0000 .00. . 0.00020 00.0.. 0000.00.00 000000.090 .0

 

 

 

00.002050200500030... ..
00.0 0.0. 000. 000 000.0 00.000
02 00 - 00... 02 wz EoEBEEw
. 0 00.0 02 0 0 0.. 02 0000 00.0.2
02 02 02 . 0 00 000....
$2 $2 wz 0 .. no. 3.50.0000... 50000 .0000
. 0 00.0 02 02 . 0 0.. 08000.0 30.0..
. 0- 00.0- 02 02 0. 0.- 00.- $000.00. 000..
z. 0. 0.0 : 0 00.0 : 0 00. z 0. 0... 0000000000 000000 .000
0.060.000
. 0- 00.? 0 0 00.0 02 z .0- 00.- m<>z..2 .0 00.0000...
. 0a 8.- 0 0 .00 02 2; 0.. 0.0... 00000 0000
2. 0.. ...0. 02 . 0 00 2: 00- 03. 0000.20 .0 000..
.2. .0 00.0 0: 00 00.0 .2. 00 000. z. 00. 000 00.0.00.
.0000 .0000 .0000 .0000
0.0 0. 000.2 0.0 0.. 000.2 0.0 .0 000.2 0m0 .0. 000.2
run 0N. .7000 0N. .00.. 0N. ..-0_0..00..0.. 0N.
>zFo 055 <15 BE... 0.00..

 

. 08 .0. 92.5. 000.000... 00.0.
0.000 .058. .000. .218. 000. 0. 00.0.0. .9050. 0.00.0. .0.0. .0 00000. 0... 0000000 .000. .900. .0 0.0000 00. 0.... 0.00.-

J- 1.1.5:: I «teen 2: 2 .m 32:

247

.65. fro—:9TofizfihggfibofifiofimsgeowcoamecaoE£ogcoon~BOcmmfio .m
.98. 8 £58 .2 5. com .Asz 8.3 a 22 52. Eemsu 8: can 8
Siva as... Ed" va 6.5L Soda Va a San Ea... 822% So; 385 .82 . 9.536 as? 35.52% 8:85:9m .u

 

 

 

:03 «we: 50* EOE =95>O *0 * .w
Nh 8.0 CNN war camp...— .3505
5%.".
cat 0 3N a: 0 N v.0 «Eu 0 m :5 0 V 838 gm
at. Gr 36 «an, E. 8.6 :5 MP mm :5 _.w m-. «00302 393“.
«cc w 8.9 cc. F 8.9 .2: w 0 :5 ON QN mg: 2&0
9!: hp 9 0.3 :5 h S. 0.? 9:. hr w @ rm :5 0:. hmw _aucam
«00:0 50% 900:0
gm :8: 9m x 8% :8: 9m x. :8: «gm 3.. 522
A73 «NV :38 3 A72 «N. $52832 mu
>Z_¢w 05mm (IN—fl hum—mm 58m“-

 

.neacaa» 9 92.me Enos. E.E.
2 as 655. .82 .5m@ 2m. 2 8.32 .Eme 25.2 .29 .o 885 2. 8 £55 .26. .28. .o 2005 9P 2.» oz.»

248

At the +1 factor level total returns (per household) more than doubled. providing
supporting evidence for the hypothesis of Figure 5.18 in which I suggested that
the areas under maize production would increase when household cash needs
increased. From greater areas I expected greater total returns. The unexpected
higher total returns to cotton with increased household cash needs did not fit
neatly into the explanatory model of Figure 5.18. One explanation was that, with
the reduced areas planted to cotton and the consequent reductions in labor
demands (and deficits) the yields of cotton (as well as those of sorghum and
maize) would have increased. Whether those increases would have been sufficient
to increase total returns by almost 50% required further investigation.

Rainfall had the expected effect of increasing total returns for maize, for
cotton and for sorghum, as well as to all inputs for each of these crops.

The effect of increasing the proportion of variance minimizers from 20%
(factor level -1) to 80% produced the results I had expected; variance minimizers
planted a greater proportion of their land to maize so I expected higher total
returns and slightly higher returns to land. labor and initial investment for maize.
Only the returns to labor were not significantly improved at the higher level of
variance minimizers (Table 5.11). With respect to cotton. the effect of changing the
proportion of variance minimizers was not as clear. Variance minimizers planted
less of their area to cotton so I expected. at the -1 factor level, the higher total
returns to cotton as were found (Table 5.12). With larger areas however, I

expected greater labor deficits and hence, lower yields per hectare with lower

249
returns per hectare. per labor day and per 23 invested. The data of Table 5.12

confirmed this expectation for returns to land and to labor but not for returns to
initial investment. The effect for the latter, was however, very low (1% of the overall
mean) and could be due to error. '

Economic response variables for all three crops were sensitive to the poor
season membership functions used, although. except for the effect of the poor
season membership function on cotton returns, these effects were small compared
to the factors discussed previously. Only total returns to maize were sensitive to
the average season membership function used and the effect was small (Table
5.11). For the most part the effects of changes in the seasonal membership
functions were due to changes in yields as discussed above. The increased yields
due to the +1 level of the poor season membership function probably accounted
for higher total returns and returns to land, labor and initial investment for cotton
(Table 5.12).

Maize and cotton returns Were notably sensitive to the land multiplier used
0' ables 5.11. 5.12). Maize returns (except returns to labor) increased with
increasing land availability whilst those for cotton decreased. These trends were
not unexpected. With larger land areas households would have been less likely to
have had sufficient capital to plant all of their normal quota of cotton fields than
with smaller land areas. This would have been due to households having to plant
a proportion of their fields to cotton. More maize and less cotton would, therefore,

have been planted, thereby increasing the total returns to maize and reducing

250
those to cotton.

Fallow practices had notable effects on sorghum returns and lesser effects
on maize and cotton returns (T ables 5.11. 5.12. 5.13). For maize the returns to
land. to labor and to initial investment decreased when fallow practices changed
from the -1 to the +1 level by between 3% (MZINV) and 6% (MZHA, MZLAB) of
the overall mean. Given the reduction in maize yields with the +1 level of fallow
practices (Table 5.10) these changes in maize returns were expected. Cotton
returns increased when fallow practices changed from the -1 to the +1 level. Total
returns increased by 8 % and returns to initial investment increased by 5 % (Table
5.12). Yields at the +1 level of fallowing practice were expected to be higher than
those at the -1 level; at the -1 level the cultivated jecha yield function would always
be used (with lower yields) whilst at the +1 level the uncultivated jecha yields

function would be used at least some of the time, leading to higher average yields.

Sorghum returns increased by about 12% when the fallow practice factor
changed from the -1 to the +1 level. This was equivalent to the increase in mean
sorghum yield (13%) due to the fallowing practice factor.

Total cotton returns increased by 7% as the wage rate increased from
232.01 to 233.07 (i.e. by 53%). I suspected that as the wage rate increased more
of the marginal cotton producing households (i.e. those likely to experience labor
shortages) would have gone out of cotton production. Expecting the mean total

returns to cotton production to increase would have been plausible. Mean total

returns r-
househo

Tc
sensitive
decreasi
Loans ir
constrair
cotton, t<
bepe or
cotton. T
PTOduce
Although

however,

ETOSiOn.

A
CHGFRT

examinin
hOW the l

the total
These Dr

r980“ rCe

251

returns represented the total returns to cotton averaged among cotton producing
households.

Total cotton returns, as well as returns to land and initial investment, were
sensitive to the form that subsidies took with the mean of total cotton returns
decreasing 50% when subsidies were paid in the form of dividends (Table 5.12).
Loans increased the area planted to cotton but also facilitated more labor
constrained households, that would otherwise not have been able to produce
cotton, to plant cotton. Labor constrained households or those without access to
bepe or riverlands faced relatively high risks of low or even negative returns to
cotton. The effect of increasing the number of these households that were able to
produce cotton would be to reduce the mean of the total returns to cotton.
Although the effect was small, the higher returns to land from cotton were

however, inconsistent with this model.

Erosion, erodibility and "fertility"

A number of the responses of the resource performance measures (MNEFl,
CHGFRT and CHGFB) were completely contradictory to my expectations. In
examining the model I realized that many of these responses were an artifact of
how the changes had been calculated as well as the large temporal variability in
the total number of years the landscape had been cultivated (and hence Fb).
These problems made interpretation of the results of factor level changes on

resource responses difficult.

The

landscape

Where

A sin
developed
WOW. The
total ”Umbl
”0mm ,
simulations
the ”limb er

Clear W are

practices W;

252

The change in the total number of years of cultivation summed over the

landscape (CHGFRT) was calculated as:

CHGFRT = ENDFR‘I’so - STRTFRT21

Where CHGFFlT = The change in the total number of years of

cultivation in each cell summed across the 30

year simulation;

ENDFRTso = The total number of years of cultivation in each
cell summed over the landscape at t=50;
STRTFFle = The total number of years of cultivation in each

cell summed over the landscape at t=21.

A simple simulation model that used the fellow functions of MASOKA1 was
developed and initialized, with the same functions used in INITIAL, for cultivation
history. The model was used, with only 100 cells, to examine the changes in the
total number of years of cultivation of each cell summed across the landscape
(T OTFRT) and over the same 30-year simulation period. Several independent
simulations were carried out and produced the same results. TOTFRT as well as
the number of cells left fallow each year were plotted in Figure 5.20. What became
clear were the very large fluctuations in TOTFRT when the -1 level of fallow

practices was used (5 year minimum and 7 year maximum). Despite the obviously

253

lower mean TOTFRT at the +1 factor level, by using the above equation to
calculate CHGFRT the +1 level of fallowing practices appeared to increase
CHGFFlT compared to the -1 level (Figure 5.20). These appearances were
deceptive; the mean of several runs with the fallow test model showed that with
0-10 fallow practice mean TOTFRT over 30 years was 50% less than with the 5-7
fallow regime and 60% less with the 6-7 fallow regime that was used. in error.
Unfortunately, the sensitivity analyses were conducted using the CHGFRT and
CHGFB values as calculated above (CHGFB was calculated in a similar manner)
and the results were therefore, of little use. In future versions of the model it was
planned to use more suitable analytical methods for these data, autoregressive
moving averages or spectral analyses were tools that were thought likely to prove
insightful.

Although mean erosion was calculated over the entire landscape and then
averaged for each year, its dependence on Fb suggests that interpretations of the
sensitivity of erosion (as well as Fb) to factor level changes should be made with

far more attention paid to the time series than to the average values.

SUI

a O 5 0 r3 PL
4: 4. 3 3 2 2
JLmeaL COHHM>HHH3U

P.» y C
l 1

HmVHOH

FiQUre 51

254

 

 

150 ‘ u

 

 

 

100 1 I
20 25 30 35 40 45 50

Slmulatlon gear

-*'—'O-10 Fallow -43-'5-7 Fallow

Figure 5.20. Total number of years each cell had been cultivated, summed over
the landscape for +1 fallow practice factor level (min = O, max = 10
years) and -1 factor level (min = 5, max = 7 years). Result of single
shnubfion.

 

pr l
str

de'

slrr
SOT
nee
SOU
retu
sum
to th
the r
Yield
relJre
and l
fUncfi
amoU

CE” in

255
SUMMARY AND CONCLUSIONS

The development and structure of simulation models of the Masoka
agroecosystem were described and tests of model validity and sensitivity were
presented. The model was developed to enhance our understanding of the
structure and functioning of the Masoka agroecosystem and as part of a
developing methodology for the analysis of complex agroecosystems.

Three models were described: The agroecosystem model (MASOKA1)
simulated household decisions regarding which of three crops (maize, cotton and
sorghum) to plant on each field in each year, compared household cash and food
needs to production, compared labor needs to available household labor and
sought to employ labor to make up deficits. The model also calculated household
returns to land, labor and initial investment for each of the three crops and
summary statistics for all performance measures on an annual time step. Inputs
to the agroecosystem model were generated by the initializing model (INITIAL) and
the rainfall generator (RAINGEN). Yields were generated, within MASOKA1, using
yield probability distribution functions fitted to yield probabilities provided by village
representatives for each of three soils and three classes of season (good, average
and poor). Elements of fuzzy set theory were used to develop continuous yield
functions. The degree of membership in each season of each annual rainfall
amount was used to determine the yields achieved by each household on each

cell in that season.

256
Household production was corrected if households experienced labor

deficits in key months of the growing season for each crop. Households that were
labor deficit could employ labor from households that had labor surpluses but
were food or cash deficit if the employing household had the cash or food to pay
for the hired labor. Once household production was corrected for labor deficits the
model used household yields, land areas planted and labor hired to calculate the
total returns to the household for each crop as well as the returns to land, to labor
and to initial investment. The model also estimated the erosion on each cell of the
landscape, using the SLEMSA model of Elwell (1980).

Outputs from the model were written to data files each year. Outputs of
individual household performance could be written to files or random samples of
household data could be written to data files. Household data were aggregated
each year and written to separate output files. These data in turn were aggregated
when the model was run in monte carlo mode and the output written to a monte
carlo output file. GIS data layers of soil, slope, household field holdings, cultivation
history and vegetation cover were key input files to the model. The model
produced an updated cultivation history.GlS data layer at the end of the simulation
as well as data layers of erosion and the change in soil erodibility over the
landscape. The latter data layers could be produced at up to three times during
the course of a simulation.

Outputs from the rainfall generator and initializing model were tested against

the original data distributions taken from Masoka and were found to generate

at

rai
ea
Co

for

257

rainfall and initial conditions that were acceptably close to the original distributions.
The yields generated by the agroecosystem model were compared to the yield
distributions developed by the VRs and were found to work adequately. Interactive
validation of model results with Masoka residents and VHS as well as Zimbabwean
development professionals was still required at the time of writing.

Preliminary sensitivity analyses of the model examined the effects of 13
factors on response variables. Crop yields were particularly sensitive to rainfall. The
percentage of households that were cash, maize and sorghum deficit each year
were sensitive to rainfall, the land area available to households, the cash needs of
households, the membership functions used for each season, whether subsidies
were provided as loans or dividends and fallowing practices. For the most part the
relationships between factor levels and response variables conformed to
expectations. The decrease in households experiencing needs deficits with
increased household cash needs was surprising. A tentative model, not yet tested,
was described to explain the effect.

The effects of factor levels on economic response variables were largely
attributable to the effects of rainfall on yields, and in the case of maize, to changes
in the maize price. Maize returns were particularly sensitive to the maize price,
rainfall, household cash needs and the proportions of households that were in
each of the two household production strategies used for these simulations.
Cotton returns were particularly sensitive to rainfall, the membership functions used

for each season, the proportion of households in each production strategy and

fallow
house

for ea

dklnc
ofthe
chang

invesfi

Masok

1)

2)

3)

258

fallowing practices. Sorghum returns were particularly sensitive to rainfall,

household cash needs, fallowing practices and to the membership functions used

for each season.

The model estimates of changes in cultivation intensity and soil erodibility

did not take the cyclic nature of these changes into consideration. Interpretation

of the results of sensitivity analyses on the responses of these variables to

changes in the 13 factors were therefore, ambiguous and required further

investigation.

In conclusion, I proposed that the models developed for analyzing the

Masoka agroecosystem, although incompletely evaluated, were useful:

1)

2)

3)

4)

The models enabled me to examine the responses of a very complex
system to changes both in exogenous inputs typical of policy decisions as
well as to endogenous changes typical of household production decisions.
The models enabled me to examine the feasibility of integrating indigenous
knowledge as well as formal scientific knowledge to provide a predictive
simulation model.

The models enabled me to examine, simultaneously, the effects of various

management or policy options on household and community performance

measures.

The models enabled us to use biophysical, social and economic

performance criteria, in our evaluations of policy or management options.

This capability would enable us to observe the relative trade-offs between

 

inl

so;
the
inc

rea

259

these criteria and could lead to more informed decision making.

As an aid to understanding the structure and behavior, the models proved
informative and represent a largely untapped source of insight and hypotheses.
The modeling methodology shows promise as an aid to decision making in data
scarce environments. We would do well however, to remember that "..in running
the models we investigate the consequences of the guesses and assumptions
incorporated in them - model output may tell us nothing new about empirical

reality." (Landsberg, 1987).

BIBLIOGRAPHY

Anderson, JR 1991. A framework for examining the impacts of climate variability.
In: Muchow, R. C. and J. A. Bellamy (Eds.), Climatic risk in crop production:
Models and management for the semiarid tropics and subtropics. C.A.B
lntemational, Wallingford, U.K. 3-17.

Anderson, JR. 1974. Simulation: Methodology and application in agricultural
economics. Review of Marketing and Agricultural Economics, 42, 3-55.

Balci, O. and R.G. Sargent. 1984. A bibliography on the credibility assessment and
validation of simulation and mathematical models. Simuletter, 15. 15-27.

Baudry, J. 1993. Landscape dynamics and farming systems: Problems of relating
patterns and predicting ecological changes. In: Bunce, R.G.H., L Ryszkowski and
M. G. Paoletti (Eds.), Landscape ecology and agroecosystems. Lewis Publishers,
Ann Arbor, USA. 21 ~39.

Berdegue, J.A., M. Installe, C. Duque, R. Garcia and X. Ouezada. 1989. Application
of a simulation software to the analysis of a peasant farming system. Agricultural
Systems, 30, 317-334.

Box, G.E.P. 1979. Robustness in the strategy of scientific model building. In:
Launer, R.L and GM Wilkinson (Eds.), Robustness in statistics. Academic Press,
New York, USA. 201 -236.

Box, G.E.P., W.G. Hunter and J.S. Hunter. 1978. Statistics for experimenters. An
introduction to design, data analysis, and model building. John Wiley & Sons, New
York, USA

Buchanan, GA and ER. Burns. 1970. Influence of weed competition on cotton.
Weed Science, 18, 149-154.

Burnside, 0.0. and GA. Wicks. 1967. The effect of weed removal treatments on
sorghum growth. Weeds, 15, 204-207.

Debrah, S. and H.H. Hall. 1989. Data aggregation and farm risk analysis.
Agricultural Systems, 31, 239-245.

Dent, J.B. and JR Anderson. 1971. Systems, management and agriculture. In:

Dent, J.B. and JR. Anderson (Eds.), Systems analysis in agricultural management.
John Vlfiley & Sons, Australasia, Sydney, Australia, 3-14.

260

261

Deybe, D. and G. Flichman. 1991. A regional agricultural model using a plant
growth simulation program as activities generator - an application to a region in
Argentina. Agricultural Systems, 37, 369-385.

Elwell, H.A. 1980. Design of safe rotational systems. Department of Conservation
and Extension (AGRITEX), Harare, Zimbabwe.

Elwell, H.A. 1986. Determination of erodibility of a subtropical clay soil: a laboratory
rainfall simulation experiment. Journal of Soil Science, 37, 345-350.

Elwell, H.A. 1978. Modelling soil losses in Southern Africa. Journal of Agricultural
Engineering Research, 23, 117-127.

Firbank, LG. 1993. The implications of scale on the ecology and management of
weeds. In: Bunce, R.G.H., L. Ryszkowski and MG. Paoletti (Eds.). Landscape
ecology and agroecosystems. Lewis Publishers, Ann Arbor, U.S.A. 91-103.

Fresco, LC. 1986. Cassava in shifting cultivation. A systems approach to
agricultural technology development in Africa. Royal Tropical Institute, Wageningen,
The Netherlands.

FSRU. 1985. Annual Report of the Farming Systems Research Unit, Department
of Research and Specialist Services. Government Printers, Harare, Zimbabwe.

Giere, RN. 1991. Understanding scientific reasoning. Holt, Rinehart and Winston,
Inc, Orlando, Florida, USA.

Halfon, E. 1979. The effects of data variability in the development and validation
of ecosystem models. In: Zeigler, B.P., M.S. Elzas, G.L. Klir and T.I. Oren (Eds).
Methodology in systems modelling and simulation. North-Holland Publishing
Company, Amsterdam, Holland. 335—343.

Hall, M.R., C.J. Swanton and G.L. Anderson. 1992. The critical period of weed
control in grain corn (Zea mays). Weed Science, 40, 441 -447.

Hutchinson, P. and LS. Unganai. 1992. The effect on rainfall frequencies of
changes in mean annual and monthly rainfalls. In: Kronen, M. (Ed.). Proceedings
of the Second Annual Scientific Conference of the SADCC-Land and Water

Management Research Programme. Mbabane, Swaziland, October 7-9, 1991. 42-
55.

Kingwell, R.S., D.A. Morrison and AD. Bathgate. 1992. The effect of climate risk on
dryland farm management. Agricultural Systems, 39, 153-175.

262

IGngwell, RS. and DJ. Pannell, (Eds.). 1987. MIDAS, a bioeconomic model of a
dryland farm system. Pudoc, Wageningen, Holland.

Klein, K.K., Kulshreshtha and SA. Klein. 1989. Agricultural drought impact
evaluation model: Description of components. Agricultural Systems, 30, 117-138.

Landsberg, J.J. 1987. Book review. Forest Ecology and Management, 18, 161-163.

Law, AM. and W.D. Kelton. 1991. Simulation modeling and analysis. McGraw—Hill,
Inc, New York, USA.

Maino, M.J. Berdegue and T. Rivas. 1993. Multiobjective programming - an
application for analysis and evaluation of peasant economy of the Vlllth Region of
Chile. Agricultural Systems, 41, 387-397.

Manetsch, T. J., M.L Hayenga, A.N. Halter, T.W. Carroll, M.H. Abkin, D.R. Byerlee,
K Chong, G. Page, E. Kellog and G.L Johnson. 1971. A generalized simulation
approach to agricultural sector analysis. Michigan State University, East Lansing,
Michigan, USA

Mankin, J.B., R.V. O’Neill, H.H. Shugart and SW. Rust. 1975. The importance of
validation in ecosystem analysis. In: Innis, G.S. (Ed.), New directions in the analysis
of ecological systems. Part 1. The Society for Computer Simulation, La Jolla,
California, USA. 63-71.

Meentemeyer, V. and CE. Box. 1987. Scale effects in landscape studies. In:
Turner, M.G. (Ed.), Landscape heterogeneity and disturbance. Springer-Veflag,
New York, USA. 15-34.

Naylor, T.H., J.L Balintfy, D.S. Burdick and K. Chu. 1966. Computer simulation
techniques. John Wiley & Sons, New York, USA.

Naylor, TH. and J.M. Finger. 1967. Verification of computer simulation models.
Management Science, 14, 8-92 - B-101.

Ngambeki, D.S., R.R. Deuson and RV. Peckel. 1992. Integrating livestock into
farming systems in northern Cameroon. Agricultural Systems, 38, 319-338.

Officer, RR. and J.L Dillon. 1968. Probability and statistics in agricultural research
and extension: A pro-Bayesian view of modern developments. The Journal of the
Australian Institute of Agricultural Science, September, 121-129.

Owsinski, J.W. 1982. Agriculture, population, environment and balanced regional
development: A modelling project in Poland. Agricultural Systems, 8, 87-104.

 

I“ CF

C\ C\ Q\

263

Popper, Sir KR. 1968. The logic of scientific discovery. Harper and Row, New
York, USA

Pritsker, A.A.B. 1979. Compilation of definitions of simulation. Simulation, August,
61-63.

Reynolds, P. 1991. Dance civet cat. Child labour in the Zambezi Valley. Boabab
Books, Harare, Zimbabwe.

Sargent, R.G. 1982. Verification and validation of simulation models. In: Cellier, F.E.
(Ed.), Progress in modelling and simulation. Academic Press, London, UK

Sauget, N. and G. Balent. 1993. The diversity of agricultural practices and
landscape dynamics: The case of a hill region in the southwest of France. In:
Bunce, R.G.H., L Ryszkowski and MG. Paoletti (Eds.), Landscape ecology and
agroecosystems. Lewis Publishers, Ann Arbor, U.S.A. 113-129.

Schwerzel, P.J. and P.E.L Thomas. 1971. Weed competition in cotton. PANS, 17,
30-34.

Shannon, RE. 1975. Systems simulation, the art and science. Prentice-Hall, New
Jersey, USA

Shumba, E. 1988. Maize technology research in Mangwende, a high potential
communal area environment in Zimbabwe. Part 1: Developing a research agenda.
Farming Systems Newsletter, 34, 12-34.

Steiner, J.L 1989. Modeling agroclimatic systems: Guidelines and Perspectives.
Soil, crop and water management systems for rainfed agriculture in the
Sudano-Sahelian zone. ICRISAT Sahelian Center, 11-16 January, 1987.

Steinfeld, H. 1988. Livestock development in mixed farming systems: a study of
smallholder livestock production systems in Zimbabwe. Wissenschaftsverlag Vauk
IGel, FR Germany.

Swift, M.J., P.G.H. Frost, B.M. Campbell, J.C. Hatton and KB. Wilson. 1989.
Nitrogen cycling in farming systems derived from savanna: Perspectives and
challenges. In: M. Clarholm and L Bergstrom (Eds.), Ecology of arable land.
Kluwer Academic Publishers, Holland. 63-76.

Ungar, ED. 1990. Management of agropastoral systems in a semiarid region.
Pudoc, Wageningen.

264

Vemuri, V. 1978. Modeling of complex systems. An introduction. Academic Press,
New York, USA.

Welch, PD. 1983. The statistical analysis of simulation results. In: Lavenberg, S.S.
(Ed.), Computer performance modeling handbook. Academic Press, New York,
USA. 267-329.

Wiegart, R.G. 1988. Holism and reductionism in ecology: Hypotheses, scale and
systems models. OIKOS, 53, 267-269.

Wiens, J.A., J.F. Addicott, T.J. Case and J. Diamond. 1986. Overview: The
importance of spatial and temporal scale in ecological investigations. In: Diamond,
J. and T.J. Case (Eds.), Community Ecology. Harper & Row, New York, USA.
145153.

Wiens, J.A. 1989. Spatial Scaling in Ecology. Functional Ecology, 3, 385-397.

Wright, A. 1971. Farming systems, models and simulation. In: Dent, J.B. and JR.
Anderson (Eds.), Systems analysis in agricultural management. John Wiley & Sons
Australasia, Sydney, Australia. 17-33.

Yates, F. 1937. The design and analysis of factorial experiments. Imperial Bureau
of Soil Science. Harpenden, Herts, England.

Zar, J.H. 1984. Biostatistical analysis. Prentice Hall, Englewood Cliffs, New Jersey,
USA.

Zinyama, L.M., D.J. Campbell and T. Matiza. 1988. Traditional household strategies
to cope with food insecurity in the SADCC region. Southern Africa: Food Security
Policy Options. Harare, Zimbabwe, 1-5 November, 1987.

Chapter 6.

Overview and key Issues for further research

INTRODUCTION
In this chapter the results of Chapters 3 through 5 are integrated and
discussed with respect to the research objectives defined in Chapter 1. In the last

section of the chapter, key issues requiring further research are identified.

OVERVIEW

In Chapter 1 it was suggested that two problems needed to be
overcome in conducting agroecosystems analyses. The first of these was the
problem of applying performance criteria such as sustainabilty, stability, resilience,
productivity and equitability. The second was the lack of a sound philosophical
and methodological basis for conducting the analysis. The checklist of Grimm et
al., (1992) was suggested as being a minimum set of definitions required when
using these performance criteria; these factors were a) the level of description; b)
the variable of interest; c) the referential behavior of the variable of interest; d) the

nature of the disturbance; e) the spatial scale; and f) the temporal scale.

265

266

The field data collection and modeling methodologies presented in the
preceding chapters not only required that I make explicit the primary level of
analysis but facilitated use of several levels of analysis simultaneously. With these
capabilities the modeling methodology in particular, could be used where
emergent properties and multi-objective evaluations were likely to be problems. By
using several levels of analysis, the household compared to the community for
example, the policy maker could examine the impacts of policies from the
household perspective. Given that several scales of analysis were possible in any
simulation experiment, the impacts of household practices or policies on emergent
properties could be examined.

The large number of performance measures in model output provided an
integrated assortment of social, economic and bio-physical variables that were
available to evaluate agroecosystem performance. This cross section of variables
facilitated evaluation of the trade-offs that might have to be made in the analysis
of real world agroecosystems.

The simulation models developed in this study were capable of generating
expected or referential behavior based on a wide variety of assumptions regarding
initial conditions, inputs and model parameter values. These referential behaviors
could be used in evaluating the behavior of the agroecosystem.

The large number of factors that were changed in the sensitivity analysis
were indicative of changes that could be made to examine the effects of particular

disturbances on agroecosystem behavior. The approach thus enabled the me to

In

n)

(I)

267
specify the nature, direction and magnitude of a disturbance and to observe the

impacts of these changes on the behavior of the agroecosystem using any of the
large number performance measures estimated by the model. Both spatial and
temporal scales were fixed parameters in the model with respect to some variables
but not to others. The simulation time step and cell size were fixed parameters.
The number of households was, however, a variable factor whose value might
affect the values of performance criteria. Household numbers were explich set in
each simulation.

Viewed as a whole, the field and modeling methodologies provide a useful
set of tools with which to address questions about agroecosystem behavior when
either static or dynamic performance criteria were to be used in evaluating
agroecosystem behavior. Some aspects the model expanded the criteria
suggested by Grimm et al., (1992); the model provided information to examine the
effects of changes in endogenous or exogenous control variables on
agroecosystem performance criteria at several levels simultaneously; the model
provided information that could be used to examine the effects of changes in
endogenous or exogenous control variables on several response variables and
several dynamic performance criteria. This capability would enable decision
makers to evaluate the relative trade-offs that would be made under different policy
or experimental conditions.

The development of a simulation model as part of the methodology would

enable me to conduct controlled simulation experiments which could be used to

C0

268
explore the consequences of policy or household level changes in control

variables.

The unavailability of data was a major constraint in planning the analysis
described in previous chapters. Methodologies were developed and used that
enabled me to develop useful insights in to the structure and behavior of the
agroecosystem using the best available data. Much of this data was in the form
of the knowledge of agroecosystem residents. The methodology presented was
sufficieme flexible to incorporate data from local knowledge sources as well as
more formally measured data. These data varied greatly in quantity and quality but
were integrated into a simulation model. Very little data was available on the
Masoka agroecosystem before the analysis was conducted. The methods
developed in this study enabled meaningful examinations of the agroecosystem
to be conducted in spite of the limited availability of quantitative data. In particular
the use of spidergrams facilitated quantitative analyses of the components and
relative importance of inputs to and outputs from these components for an entire
production system. The resulting graph of the production system could prove
useful for focusing further investigations. The spidergrams also helped select those
components of the production system that were likely to be most important and
needed to be included in the simulation model.

Although an approach to analyzing the behavior of the Masoka
agroecosystem was presented the analysis was far from complete. Few statements

could therefore, be made regarding the actual performance of the agroecosystem

w'rl

cle

or

269
with regard to sustainability, stability, resilience and productivity. What became

clear as the analysis proceeded was the dependence of these performance criteria
on those factors listed in the checklist of Grimm ef al. (1992). The widespread use
of these performance criteria belies the diffictu of actually applying them in the
real world. Although tentative, the analysis presented in Chapter 5 indicated that
the ability of households to satisfy cash and food needs were critically dependent
on rainfall, the land available to them both through physical availability as well as
through the labor available to manage a given area and also on the level of cash
needs that households expected to satisfy. The land available to each household
and the quality of life expected by households were key community and household
decisions. The analytical approach adopted, with the emphasis on community
participation, could assist the community and households to make these decisions
with at least some information on the trade-offs that needed to be made.

There was no doubt that aspects of the agroecosystem model lacked
realism; The fixed proportion of a household’s lands that were allocated to each
crop in each season was certainly not a reflection of reality. A more flexible
household decision making algorithm, in which household plantings were made
as a function of both cash and labor resources available was planned for future
versions of the model. The use of a constant number of households with
unchanging land holdings and number of adults and children were also not
realistic but considerably simplified the model. The use of constant prices and

wage rates was obviously an over simplification of what is likely to be a complex

as

of

 

270

and changing set of values. Within season rainfall distributions are likely to have
significant impacts on yields and erosion, yet the poor within season resolution of
the model made analysis of these effects impossible. Despite these many
shortcomings however, the model provided a useful first attempt at simulating a
very complex production system at several scales; the complex interactions that
were apparent in the results of the sensitivity analyses justified using the multi-
scaled systems approach; despite the scarcity of measured data I was able to
develop a meaningful model of the agroecosystem. l was also able to use the
model to examine the effects of changes in the policy environment (crop price
changes for example) as well as the household decision making environment
(switches in fallowing practices for example) on several performance measures.
The model could therefore, have utility both as a tool for planning extension
packages as well as for examining the effects of various agricultural policies.

In conclusion, whilst the analysis and evaluation of the methodology were
not complete, the methodology developed for analyzing the Masoka
agroecosystem was useful and enabled the development of a fuller understanding
of agroecosystem behavior. Some of the methods developed could prove useful

in analyses of similar problems in similar classes of agroecosystem.

271
KEY ISSUES FOR FUTURE RESEARCH

The approach to agroecosystem analysis described in this dissertation is

clearly far from adequate. Several issues emerged from the analysis and

subsequent writing.

1.

A methodological approach to validating expert knowledge has not been
reported in the literature. Validation of indigenous knowledge (Chapter 4)
was difficult and the methods of doing so were not well developed. If, as is
likely, indigenous knowledge is to be an important source of information
then the issue of validating this knowledge needed to be addressed.
Methods of analyzing the results of data collected from VRs were not readily
apparent and require further investigation. In this regard Bayesian
probability theory appeared to be a suitable starting point.

The simulation models developed for analyzing the Masoka agroecosystem
were quantitatively based. Although some use was made of fuzzy set
theory, greater use of these concepts could have resolved some of the
problems associated with using indigenous technical knowledge, such as
the conversion of total production to production per hectare.

Some analyses were not completed due to time constraints. Important
among those was the expansion of model sensitivity analyses to finer scales
of resolution. Were different classes of household, for example, differentially
sensitive to changes in factor levels? How sensitive were the same

performance criteria to simulation experiments conducted over longer

272

periods of simulated time?

A key issue that was not dealt with in the prototype agroecosystem model
was the effect of spatial dependence on erosion simulation. The erosion
modeling component of the version of MASOKA1 presented in these
chapters was far from adequate and perhaps the weakest aspect of that
model.

As was discussed earlier, the model and model results should be validated
against VR opinion. The opinions of WWF MAPS Project staff would also be
needed to establish the usefulness of the model and the approach.

It was likely that the spidergrams that were developed during the field data
collection were biased by conditions that were dominant at the time (the
drought for example). Investigations were needed that would indicate the
relative changes in the components and weights of components of
spidergrams if these analyses were conducted at different times and places.
The model as well as the overall approach to analyzing agroecosystems
have potential utility both for evaluating government policy (extension
packages or resettlement in the Zambezi valley for example), community
level policies of land allocation and expansion as well as for identifying
optimal household production practices. To utilize that potential the
approach and results needed to be communicated to audiences that
differed greatly in their technical understanding of the approach as well as

in their objectives and values. One of the major challenges for the future

273
would be to present the findings of the study to these different groups and

then to integrate their responses into future analyses.

Despite its complexity the model was a greatly simplified version of the
Masoka production system; only three of the 13 production activities listed
by female VRs were simulated in the model. It was quite possible that a
greater diversity of production activities could significantly affect the results
of analyses of the agroecosystem. Methods of evaluating the sensitivity of
model response variables to the way the model was specified needed to be
devised and tested. A potential solution tothis problem might have been to
examine the effects of allocating income to specific households at specific
times in a simulation to represent income generated by these other
activities. The income so generated could be either in the form of food (to
represent vegetables or poultry for example) or as cash generated by craft
or beer brewing. An advantage of the modeling approach used was the

ability to examine effects such as these on different classes of household.

274

BIBLIOGRAPHY

Grimm, V.E. Schmidt and C. Wissel. 1992. On the application of stability concepts
in ecology. Ecological Modelling, 63, 143-161.

"III’LIIIILIIIIIIIIIIIIIIIII