You are on page 1of 125

Spatial analysis of livelihoods

of smallholder farmers in
Striga-infested maize-growing areas
of Eastern and Southern Africa
H. Bouwmeester, V.M. Manyong, K.D. Mutabazi, C. Maeda,
G. Omanya, H.D. Mignouna, and M. Bokanga

www.iita.org
Spatial analysis of livelihoods
of smallholder farmers in
Striga-infested maize-growing areas
of Eastern and Southern Africa

H. Bouwmeester1, V.M. Manyong1, K.D. Mutabazi2, C. Maeda1,


G. Omanya3, H.D. Mignouna3, and M. Bokanga3

1
International Institute of Tropical Agriculture, IITA
2
Sokoine University of Agriculture
3
African Agricultural Technology Foundation, AATF

January 2009

i
© International Institute of Tropical Agriculture and African Agricultural Technology
Foundation 2009

The International Institute of Tropical Agriculture (IITA) and the African Agricultural
Technology Foundation (AATF) hold the copyright to this publication but encourage
duplication of these materials for noncommercial purposes. Proper citation is requested
and modification of these materials is prohibited. Permission to make digital or hard
copies of part or all of this work for personal or classroom use is hereby granted without
fee and without a formal request provided that copies are not made or distributed for
profit or commercial advantage and that copies bear this notice and full citation on the
first page. Copyright for components not owned by IITA and AATF must be honored
and permission pursued with the owner of the information. Prior specific permission is
required to copy otherwise, to republish, to post on servers, or to e distribute to lists.

International mailing address:


IITA, Carolyn House
26 Dingwall Road, Croydon CR9 3EE, UK

PMB 5320, Oyo Road


Ibadan, Oyo State

ISBN 978-131-328-5

Publication layout and design by IITA

Correct citation: Bouwmeester, H., V.M. Manyong, K.D. Mutabazi, C. Maeda, G. Omanya,
H.D. Mignouna, and M. Bokanga. 2009. Spatial analysis of livelihoods of smallholder
farmers in Striga-infested maize growing areas of Eastern and Southern Africa.
International Institute of Tropical Agriculture, Ibadan, Nigeria, and African Agricultural
Technology Foundation (AATF), Nairobi, Kenya. 114 pages.

ii
Contents
Acknowledgments...............................................................................................................vi
Executive summary............................................................................................................vii
Introduction.........................................................................................................................vii
How to read this report........................................................................................................ 1
The study area.................................................................................................................... 1
Methodology........................................................................................................................ 2
Sampling strategy................................................................................................................ 2
Data collection and pre-analysis management................................................................... 2
Character of data................................................................................................................. 2
Cleaning of data.................................................................................................................. 3
Plotting the data.................................................................................................................. 4
Background data................................................................................................................. 4
Relation of households and administrative units................................................................. 6
Spatial pattern..................................................................................................................... 8
Ordinary Kriging................................................................................................................ 10
Kriging accuracy................................................................................................................ 11
Results I. Spatial analysis in administrative units ............................................................. 12
Natural capital................................................................................................................... 12
Elevation............................................................................................................................ 12
Rainfall ............................................................................................................................. 14
Roads................................................................................................................................ 16
Area of land ...................................................................................................................... 16
Share of owned and cultivated land.................................................................................. 16
Physical capital.................................................................................................................. 19
Household productive asset index.................................................................................... 19
Human capital................................................................................................................... 20
Number of yearly extension visits/household.................................................................... 21
Ill-health index................................................................................................................... 22
Financial capital................................................................................................................. 23
Composite liquidity index................................................................................................... 23
Livestock ownership.......................................................................................................... 24
Household income............................................................................................................. 25
Maize and Striga............................................................................................................... 26
Maize yield........................................................................................................................ 26
Share of land under improved maize varieties.................................................................. 26
Share of land under intercropping..................................................................................... 26
Share of land infested by Striga........................................................................................ 30
Share of land infested by Striga 10 years ago.................................................................. 31
Livelihood outcomes.......................................................................................................... 31
Body mass index............................................................................................................... 31

iii
Country wealth index......................................................................................................... 32
Regional wealth index....................................................................................................... 34
Results II. Spatial analysis by interpolation....................................................................... 35
Striga Infestation............................................................................................................... 35
Maize yield........................................................................................................................ 37
Country wealth index......................................................................................................... 40
Conclusions and recommendations.................................................................................. 43
References........................................................................................................................ 45

Tables
Table 1. Statistics of the main indicators and variables. ..................................................... 3
Table 2. Coordinates can be written in several formats....................................................... 3
Table 3. Points A and B can be written in different formats................................................. 4
Table 4. Secondary data files used in the analysis.............................................................. 6
Table 5. The different administrative units........................................................................... 6
Table 6. Distribution of 880 sampled households................................................................ 7

Figures
Figure 1. Distribution of surveyed households.................................................................... 5
Figure 2. Administrative units.............................................................................................. 8
Figure 3. Distribution of administrative units....................................................................... 9
Figure 4. Clustered distribution of households in Handeni district, Tanzania....................... 10
Figure 5. High variability of households at short distance................................................. 11
Figure 6. Example of map................................................................................................. 13
Figure 7. Histogram of altitude of all households.............................................................. 13
Figure 8. Altitude in the area of interest. ........................................................................... 14
Figure 9. Histogram of mean annual rainfall..................................................................... 15
Figure 10. Annual rainfall in the area of interest. .............................................................. 15
Figure 11. Location of roads in relation to the surveyed points......................................... 17
Figure 12. Histogram of distance from households to the nearest road. . ........................ 18
Figure 13. Area of cultivated land owned by the average household in Tanzania............. 18
Figure 14. Share of owned, managed land in Uganda...................................................... 19
Figure 15. Household productive asset index (PAI) for Malawi. ....................................... 20
Figure 16. Number of extension visits/household/year in Tanzania.................................. 21
Figure 17. Distribution of ill-health index (IHI) for Uganda................................................ 22
Figure 18. Distribution of composite liquidity index (CLI) in Malawi.................................. 23
Figure 19. Distribution of livestock (Tropical Livestock Units) in Tanzania........................ 24
Figure 20. Household income acquired from various enterprises in Malawi..................... 25
Figure 21. Maize yield (t/ha) in Uganda............................................................................ 27
Figure 22. Share of land under improved maize varieties in Tanzania.............................. 28
Figure 23. Share of land under intercropping in Malawi.................................................... 29
Figure 24. Striga infestation as a percentage of total land under maize in Malawi........... 30

iv
Figure 25. Striga infestation 10 years ago in Malawi......................................................... 32
Figure 26. Distribution of body mass index (BMI) of adult women in Tanzania................. 33
Figure 27. Distribution of country wealth index in Tanzania.............................................. 33
Figure 28. Distribution of regional wealth index in the three countries.............................. 34
Figure 29. Histogram of Striga infestation in Malawi......................................................... 35
Figure 30. Predicted Striga infestation in Malawi.............................................................. 36
Figure 31. Striga infestation in Malawi.............................................................................. 36
Figure 32. Confidence level of the predicted Striga infestation......................................... 37
Figure 33. Histogram of maize yield in Tanzania............................................................... 37
Figure 34. Predicted maize yield in Tanzania.................................................................... 38
Figure 35. Maize yield in Tanzania.................................................................................... 39
Figure 36. Confidence level of the predicted maize yield in Tanzania............................... 39
Figure 37. Histogram of country wealth index in Uganda.................................................. 40
Figure 38. Predicted country wealth index in Uganda....................................................... 40
Figure 39. Country wealth index in Uganda...................................................................... 41
Figure 40. Confidence level of the predicted country wealth index in Uganda.................. 42

Annexes
Annex I. Distribution of surveyed points ................................................................... 46
Annex II. Frequency of households per administrative unit........................................ 50
Annex III. Distribution of administrative units.............................................................. 51
Annex IV. Altitude of surveyed households................................................................. 55
Annex V. Mean annual rainfall, 1951–2005................................................................ 56
Annex VI. Area of land (acre)...................................................................................... 60
Annex VII. Share of owned cultivated land................................................................... 63
Annex VIII. Productive asset index................................................................................ 66
Annex IX. Number of extension visits.......................................................................... 69
Annex X. Ill-health index............................................................................................. 72
Annex XI. Overall composite liquidity index................................................................. 75
Annex XII. Tropical livestock units................................................................................ 78
Annex XIII. Enterprise income per capita (US$/yr)........................................................ 81
Annex XIV. Overall maize yield (t/ha)............................................................................ 84
Annex XV. Share of land under hybrid maize............................................................... 87
Annex XVI. Share of land under intercropping............................................................... 90
Annex XVII. Striga infestation as percentage of total land area...................................... 93
Annex XVIII. Striga infestation as percentage of total land area 10 years ago................ 96
Annex XIX. Body mass index......................................................................................... 99
Annex XX. Country wealth index................................................................................. 102
Annex XXI. Regional wealth index............................................................................... 105
Annex XXII. Predicted Striga infestation....................................................................... 109
Annex XXIII. Predicted maize yield................................................................................ 112
Annex XXIV. Predicted country health index . ............................................................... 114

v
Acknowledgments

The authors would like to thank the International Institute of Tropical Agriculture (IITA) for
providing secondary data from the livelihoods project in Malawi, Tanzania, and Uganda
and from the GEO-spatial laboratory at Ibadan. The livelihoods project was funded by
the African Agricultural Technology Foundation (AATF) and this financial contribution is
acknowledged.

Acronyms and abbreviations

AATF African Agricultural Technology Foundation

ADM Administrative District

BMI Body Mass Index

CLI Composite Liquidity Index

DEM Digital Elevation Model

EPA Economic Planning Unit

GEO Geographic

GIS Geographic Information System

GPS Global Positioning System

IHI Ill-health Index

IITA International Institute of Tropical Agriculture

TLU Tropical Livestock Unit

vi
Executive summary

This report presents results from a spatial analysis of selected data generated through a
livelihoods project in Striga infested areas of Malawi, Tanzania, and Uganda. In addition
to mapping spatial patterns on livelihood indicators using Global Information Systems
(GIS), the study also compared two interpolation techniques (ordinary Kriging and
averaging) of measured values to surrounding locations. Livelihood indicators considered
and spatially mapped in this report are related to natural capital, human capital, financial
capital, maize growing Striga infestation and livelihood outcomes. Results show that
many variables and indicators are clearly related to space. This is especially true in
Malawi where many maps show a clear gradient from the “poor” south to the “rich” north.
Many other maps in Tanzania and Uganda seem to suggest a similar correlation in space
as nearby administrative units tend to have similar values on indicators. Although the
survey that generated data used for this report was set up according to socioeconomic
criteria and not so much on spatial criteria, the findings show that any economical study
can profit from spatial analysis. The report also makes recommendations on how to
improve on the collection and recording of geo-referenced data in the farmers’ fields.

The livelihood project was designed to understand the effects of Striga on the livelihoods
of the poor. Therefore, the sampled households were always located in areas known to
be heavily infested with Striga. Expansion of areas of interest to areas not heavily infested
to assess the effects on the researched indicators is recommended. This study indicates
the power of GIS in exposing the socioeconomic consequences of a biological threat
(Striga in this case) on smallholder farmers via a set of quantifiable indicators. Therefore,
it can be said that databases designed for socioeconomic purposes can be very useful
in spatial analysis. Two methods of interpolation were applied that allow socioeconomic
properties to be predicted for unvisited sites. The results indicate that applying the two
methods generate a spatial correlation in many of the economic indicators.

vii
Introduction
This report is part of the results of a comprehensive study of livelihoods based on a
baseline study (Manyong et al. 2008a). It describes how Geographical Information
Systems (GIS) can be used in analyzing outcomes and should be read in relation to the
regional and country technical reports of the Livelihood project (Manyong et al. 2008a,
b, c, d). The aim is to demonstrate the strength of GIS in visualizing and analyzing
livelihood surveys (Arbia 1993, Fais et al. 2005, Johansson 2005, Legg et al. 2005). It is a
systematic constellation of spatial maps with critical livelihood indicators across the region
covered in the survey.

GIS offers many benefits that make it valuable in any agricultural survey.

●● GIS allows the visualization of large tables.

●● GIS can help to identify errors.

●● By clear and logic presentation of data GIS adds to the presentation of data. GIS
turns tables into attractive maps.

●● Through GIS, results can be presented in a convincing way.

●● Spatial analysis allows the conversion from point-values to area-values

●● Interpolation can be used to identify gaps in surveys.

●● Interpolation can save costs in surveys by suggesting ways of optimizing sampling


design.

How to read this report


The second part of this report describes the methodology used and details on the input
data, how they were imported into a GIS, what other geographical data were used in
the analysis, and how the analysis was done. The results of this report are divided into
three sections. The first set is given in Results I where the survey results are compared
to GEO-physical data and the most important research themes within the administrative
units are mapped. Results II shows a limited selection of these research themes that
are subjected to spatial interpolation. Finally, the last section provides conclusions and
recommendations. The annexes show all resulting maps.

The study area


This study was conducted in Uganda, Tanzania and Malawi. They are part of the Eastern
Striga belt of Africa, a maize growing area where Striga is a major biological threat to
production. Based on data reported by Agricultural Technology Foundation (AATF 2006),
the three countries account for 46% of the infested area, around 1,120,000 ha of the
Eastern Striga belt and 22% of the total area (2,355,000 ha infested by Striga in Africa
(Manyong 2008a). The fact that the three study countries are Striga hotspots in Eastern
and Southern Africa makes them ideal for this study. More details on the region and on
the surveyed districts in each country are given in the regional report and in the individual
country reports (Manyong et al. 2008a, b, c, and d).

1
Methodology
This section describes the targeted area, the sampling strategy, and how the data were
collected. In addition it discusses the quality of the data and the techniques used that led
to the results.

Sampling strategy
In each country, four districts were chosen and in each of these districts a random
sample of 75 households was taken. Through literature and consultation with experts,
these districts were purposively chosen as they rank maize as an important crop and are
regarded as Striga hotspots (Manyong et al. 2008a). The villages within districts were
then listed, based on the high importance of maize and high ranking of Striga as a major
constraint to maize production. Using an inbuilt ‘sample [%]’ command in STATA software,
five villages from each district were randomly selected. Within each sampled village, the
village register from the village government office was used to list all the households.
Then, trained enumerators used a random numbers table to select 15 households for
interview and another five households for replacement if any of the households could not
take part in this survey (Manyong et al. 2008a). The enumerators interviewed members
of the household extensively. The coordinates of the house were recorded and several
measurements relating to farm households were made. A comprehensive description of
the sampling strategy is found in the country reports where the exact procedures followed
may vary slightly (Manyong et al. 2008b, c, and d).

The spatial location of each sampled household in the survey was determined with a
handheld global positioning system (GPS). These units are generally accurate within 100
m (horizontal) in the worse result, and 10 m in the best result. In the region some 900
households, 300 in each of the three countries, were geo-positioned.

Data collection and pre-analysis management


During country-based methodology workshops, enumerators were trained on how to use
GPS for taking coordinates and measuring areas of fields. A minimum of 5 GPS units
were distributed in each country, at least one unit in each district. The shared use of the
units in the field was overseen by the district-based extension officer and the IITA country
research supervisor. The industrial serial number of each unit was properly recorded
when the unit was handed to a particular user. The coordinates of the households and the
various field areas were manually entered into the database.

Character of data
Altogether there were 901 questionnaires, all of which had a unique Questionnaire-ID.
The coordinates of all questionnaires were determined using handheld GPS devices and
were added manually on the questionnaires. The questionnaires consisted of a great
quantity of variables or indicators. From this collection the most important were selected
and used for further analyses (Table 1). The map datum used was WGS-84 and the
coordinates are recorded in the degree-decimal format.

2
Table 1. Statistics of the main indicators and variables. All indictors are explained and
mapped in Results I.

Mean

Median

Mode

St. Deviation

Skewness

Minimum

Maximum

Count
land owned (acre) 1.8 2.0 1.0 0.9 0.8 1.0 4.0 856
share of cultivated
86.3 100.0 100.0 30.3 –2.1 0.0 100.0 856
land owned (%)
productive asset
14.2 12.0 12.0 10.6 2.5 0.0 102.0 856
index
extension visits
6.1 0.0 0.0 11.5 3.6 0.0 96.0 845
(no./yr)
ill health index 0.0 0.0 0.0 0.1 8.0 0.0 1.0 856
tropical livestock
0.9 0.2 0.0 2.1 6.9 0.0 31.6 856
units (no.)
household income
77.9 35.2 0.0 131.3 4.5 0.0 1464.3 856
(US$/yr)
maize yield (t/ha) 1.1 0.8 0.8 1.0 2.2 0.0 7.0 856
share of land under
33.0 0.0 0.0 0.5 0.7 0.0 100.0 856
improved maize (%)
share of land under
32.2 0.0 0.0 0.5 0.8 0.0 100.0 856
intercropping (%)
share of land under
42.1 50.0 0.0 28.6 –0.1 0.0 100.0 856
maize with Striga (%)
share of land under
13.3 5.0 0.0 17.8 1.7 0.0 100.0 856
maize with Striga 10 years ago (%)

body mass index 21.0 21.9 0.0 7.5 –1.1 0.0 56.4 856
country wealth index 0.0 -1.2 –7.5 7.9 1.8 -16.2 44.1 856

region wealth index 0.0 -2.1 –11.2 9.0 1.5 -11.2 40.2 856

Cleaning of data
As the coordinates were collected by many interviewers with different GPS-units, there
was bound to be some confusion. As the coordinates were manually copied onto the
questionnaires, the method of how a coordinate was written depended on the settings
of each individual GPS unit. Table 2 lists valid and acceptable ways to write geographic
coordinates. The table shows that the same location can be written in several different
ways, resulting in large spatial differences of location X. In Tanzania, for instance, 1
degree corresponds roughly with 110 km of distance.The degree-values are the same
in all systems but what comes after the degree varies. For instance, 26 minutes in the
degree°minute.decimal-format would be written as .44 in the degree°decimal-format. If
interpreted incorrectly, this could cause a spatial difference of 0.18 degree (0.44 – 0.26)
or roughly 20 km.

Table 2. Coordinates can be written in several formats.


Item X,Y-coordinates Format
Location X 40°26’21”N, 79°58’36”W Degree°minute’second
Location X 40d 26’ 21” N, 79d 58’ 36” W Degree°minute’second
Location X 40.44619N, 79.94886W Degree°decimal
Location X 40.44619, -79.94886 Degree°decimal
Location X 40° 26.772, -79° 56.931 Degree°minute.decimal

3
Table 3. Points A and B can be written in different formats.
item x- x- y- y-
deg°minute deg°decimal deg°minute deg°decimal
Location A 37°30.000 E 37°50.000 E 6°15.000 S 6°25.000 S
Location B 34°10.000 E 34°16667 E 14°20.000 S 14°33.333 S

Of the 901 questionnaires six had no coordinates recorded and were deleted. Most of
the remaining 895 questionnaires were written in degree°minute.decimal-format, as
this is the default setting of the GPS-units used in this survey. This assumption was
backed because from all 895 Y-coordinates only 15 had decimal values greater than
60000. Statistically this points to the degree°minute-format since the decimal value
cannot exceed values higher than 60000, as one degree is made up out of 60 minutes.
If the survey points were randomly selected, one would expect about 60% of all values
between 0 and 60.000 and 40% between 60.000 and 99.999. Only in degree°decimal-
format can the decimal values be anywhere between 0 and 99999, as 1 degree is divided
into 100000 decimals (Table 3).

For the previously described reasons, it was assumed almost all of the coordinates were
written in the degree-format except for 15, which were converted to the same system as
the others. This assumption, however, remains a possible source of error. The plotted
households were compared with the shape-files of the district to explore this source of
error in more detail.

Plotting the data


As the coordinates of the questionnaires were now in the same format they could
be plotted on a map using GIS software. Of the total 895 questionnaires 14 were
geographically speaking far away from the researched locations. They were verified on
the hard copy of questionnaires to exclude the possibility of typing errors. There appeared
to be no typing errors and it was assumed these outliers were a result of writing errors
and were deleted from the dataset. In Figure 1, the remaining 881 households are
plotted as black dots. This figure shows the regional map and the map of Tanzania as
an example; the full-sized maps can be viewed in Annex I, including the country maps of
Malawi and Uganda.

Background data
A variety of background data was used for the various analyses and the necessary
visualization. Table 4 lists these data, the purpose of use and their source. Items 1
through 6 were used to represent the contours of the different administrative units. Items
7 to 9 were used to represent the roads. Item 10 was used to extract mean rainfall/district
and item 11 to extract the altitude of the households.

4
5
Figure 1. Distribution of the households in the entire region (left) and in Tanzania (right).
Table 4. Secondary data files used in the analysis.
Item Filename Format Purpose Source
1 Adm3 shape-file contours of districts in IITA, Ibadan,
Malawi Nigeria
2 districts_2005 shape-file contours of districts in IITA, Ibadan,
Tanzania Nigeria
3 uganda_parish_july2006 shape-file contours of parishes and IGAD, Kenya
districts in Uganda
4 national_boundaries shape-file contours of countries IITA, Ibadan,
Nigeria
5 tz_wards_2005_new shape-file contours of wards in IITA, Ibadan,
Tanzania Nigeria
6 Malawi_central shape-file contours of EPA in IITA, Ibadan,
Malawi Nigeria
7 uganda_ads_roads shape-file roads in Uganda IITA, Ibadan,
Nigeria
8 Roads shape-file roads in Tanzania IITA, Ibadan,
Nigeria
9 Roald_line shape-file roads in Malawi IITA, Ibadan,
Nigeria
10 Pptnmean5105 Grid-file mean rainfall from 1951 IITA, Ibadan,
to 2005 Nigeria
11 tandem, ugdem, Grid-files Altitude in meters IITA, Ibadan,
maldem Nigeria

Table 5. The different administrative units; adm2 is not used as it stands for province.

Item New name


Country adm1
District adm3
Ward (Tanzania) adm4
Parish (Uganda) adm4
EPA (Malawi) adm4

Relation of households and administrative units


To allow further analysis the surveyed households had to be linked to shape-files
describing the various administrative units. In principle, three administrative units can be
distinguished: (1) country, (2) district, and (3) ward for Tanzania, parish for Uganda, and
economic planning units (EPA) for Malawi (Table 5). This paragraph describes the match
between the household’s location and these three administrative units.

On the country level a simple visual inspection revealed that all households appeared
to spatially match the target countries as illustrated by the shape-file. No editing was
necessary and further analysis on country-level seemed justified.

To validate the households on a district level, the following procedures were undertaken.
One household in Kalulu, Malawi, was deleted because it was not part of the target
districts. Table 6 lists the number of households/district in the three target countries.

6
Table 6. Distribution of 880 sampled households.
Country (ADM1) District (ADM3) No. households
Uganda Busia 75
Uganda Namutumba 74
Uganda Pallisa / Budaka 75
Uganda Tororo 72
Tanzania Handeni 74
Tanzania Mkinga / Muheza 75
Tanzania Morogoro Rural 70
Tanzania Mvomero 71
Malawi Dedza 70
Malawi Kasungu 75
Malawi Lilongwe 75
Malawi Mchinji 74

The number of households seems evenly distributed over the districts and, after the
dataset was cleaned, sufficient points remained to allow further analysis.

To quantify the geographic match between the survey and the shape-files used to portray
the districts, a spatial join between the two files was done. The districts of the three
countries were merged into one shape-file. A frequency test was done to quantify how
many of the household-locations were actually located within the appropriate districts.
The match was quite good, with only seven out of 880 points not being in one of the 12
targeted districts. The mismatch of these seven points can be the result of recording
errors, typing errors, or errors in interpreting the coordinates and they were removed from
the dataset. All in all, this seemed a satisfactory result that provided sufficient reason to
do analyses on a district level.

To validate the household dataset on the 4th administrative level, the following procedure
was undertaken. All 4th administrative units (adm4) were appended into one shape-file.
The resulting shape-file was spatially joined with the households. Of the 873 households
369 did not match with the adm4 of the shapefile. Extensive cleaning needed to be
done to improve this match. Some of the major reasons for the mismatch were spelling
mistakes and households that were close to the border of two adjacent adm4s.
In some cases in the shape-file an adm4 was split in two, each with a different name,
while the households used only one name for the adm4. This is illustrated in Figure 2. In
the shapefile the adm4 Kaphuka in Malawi was split in two, one called Kaphuka and one
called Mayani. In the households, the adm4 had only the name Kaphuka. In this example,
the adm4 Mayani was renamed Kaphuka.

As mentioned before, there were 369 mismatches where the names of the households’
adm4 did not match the name of the shape-file’s adm4. All these were handled
individually and appropriate action was taken for them to match. Sometimes, the adm4 of
the households was renamed and in others, the adm4 of the shape-file was renamed. To
save the reader a very long list, not all instances are individually described.

7
Figure 2. Administrative units might be split in one file but not in another file. The black
dots represent households. One cluster of households is located in the north of Kaphuka
District while another cluster is located just across the border in the adm4 formerly known
as Mayani.

After extensive cleaning, 17 out of the 369 households were left with no clear solution
to the mismatch and were therefore deleted from the household survey. What remained
after this cleaning were 856 households in a total of 66 different adm4. The minimum
amount of households/adm4 was 1 and the maximum, 44. About two-thirds of all adm4
had 10 or more households within their borders (see table in Annex II). Theoretically,
there should not have been adm4 with so few or so many points but the average should
have been about 15. However, this was not the case and could have resulted from
recording errors, copying errors, the removal of points, the renaming during cleaning, or
because the coordinates of the points were in a different format than was assumed.
The results are shown graphically for Malawi in Figure 3 and at full size for all countries
in Annex II.

Spatial pattern
In most maps in this report different themes of interest are displayed within an
administrative unit. The map implies that the value of the theme reflects the average
value of the administrative unit. Depending on the objective, a variety of sampling designs
are proposed that can be used to describe the most probable average (Arbia 1993,

8
Figure 3. Distribution of surveyed administrative units in Malawi. Adm1 stands for country,
adm3 for district, and adm4 for EPA in Malawi, for Ward in Tanzania, and Parish in Uganda.

Diggle et al. 2007, Smith et al. 2007). The maps of Annex I reveal that the sampling
pattern of the surveyed households is highly clustered. The reason for the spatial
distribution is the sampling strategy adopted in the Livelihood project. This strategy
aims to sample households in areas with both maize and Striga concentration. Within
these areas individual households were selected through a stratified random sampling
technique (Manyong et al. 2008a).

The clustering caused by this strategy is intensified by the clustering of settlements and
by the clustering of households within the settlements. To illustrate this, Figure 4 shows
32 out of the total 74 households in the district of Handeni, Tanzania.

9
Figure 4. Clustered distribution of households in Handeni district, Tanzania.

All points are located within a 6-km range while some are as close as 40 m apart. The
other 44 households in the district are not shown because they are located about 45 km
to the east and would therefore obscure the figure.

Within the dataset there is a gradation of clustering. In Morogoro Rural and Mvomero in
Tanzania, selected households are spatially confined to a very small proportion of the
district. In some districts such as Busia and Namatumba in Uganda, selected households
are relatively evenly distributed over the district (Annex I). It is therefore expected that
variability/region will differ significantly and thus the relation between the value of each
household and the district’s average. The same accounts for the smaller administrative
units (adm4), where the sampled households are often clustered in a small area within
an adm4. While there were always at least 70 households in a single district, there might
be only a very limited number of households within an adm4. An example of this is given
in Annex II, where only one household appears to be located in adm4 Sapiri, Uganda. In
this particular instance, the average value for the entire adm4 is identical to the value of
that one household. Although in all other examples the average value is based on more
than one observation, it should be noted that the robustness of the average values varies.

Ordinary Kriging
Ordinary Kriging (OK) assumes that the distance or direction between sample points
reflects a spatial correlation that can be used to explain variation in the surface. It
assumes a constant but unknown mean and fits a mathematical function to a specified
number of points to determine the output value for all surrounding locations.

z(x,y) = µ(xy) + e(x,y)

Here Z denotes the realization at location x,y. with µ as the fixed but unknown mean and
e being the variation around this mean (Smith et al. 2007).

10
Figure 5. High variability of households at short distance of the country wealth index
in Tanzania, with values as low as –8.54 and as high as 7.71 within a distance of 1 km
(standard deviation is 7.9).

Ordinary Kriging is an exact interpolator, meaning that at the location of the households
the predicted value is equal to the measured value. This can cause strange looking
maps as there often will be jumps between predicted and measured values, especially in
datasets with a high clustered spatial distribution. Figure 5 illustrates this phenomenon
where a large difference in the wealth index occurs within a short distance (mean is 0 and
standard deviation is 7.9). To limit the importance of individual values the neighborhood
search radius has been increased to a search radius of 50 points with a minimum of 5
points. For the remaining settings the default Arc-GIS settings were used.

Kriging accuracy
Kriging also predicts the accuracy of its prediction by calculating a level of confidence
of every location on the map. With each prediction map, a map showing the standard
error map was created using the same Kriging method and parameters that were used to
generate the prediction map. The standard error is the variation of the prediction, thus a
small error corresponds with a small variation of the predicted value and a large error with
a large variation. In other words, the smaller the variation gets, the better the prediction.
The bright yellow colors on the maps (Figures 32, 36, and 40) indicate areas where the
prediction standard error was low or, to state it another way, the level of confidence in the
results was high. The dark brown symbolizes areas of low confidence. It may not come
as a surprise that all maps clearly show that the level of confidence decreases as the
distance from the surveyed households increases.

11
Results I. Spatial analysis in administrative units

The survey consists of many quantitative physical, social, and economic data from 856
households divided over three countries. In this section these households are averaged
within a certain administrative unit. This allows the visualization of the data as they
are transformed from point (household) to surface (administrative unit). The reader
should bear in mind that the resulting maps suggest a certain value for a large area
that is sometimes based on the knowledge of only a few households. The first chosen
administrative unit is the district. Each of the 12 districts has about 70 households (Table
6). The second chosen administrative unit is the adm4. It is based on a much lower
sample size, ranging from 1 to 44 households (Annex II).

The various spatial analyses are focused on socioeconomic indicators developed through
the Livelihood study following the Livelihood framework. A more elaborate explanation
on each indicator is given in the region and country reports (Manyong et al. 2008a, b, c,
and d). To allow the reader to capture the full extent of the indicators, the order in which
the themes are presented is based on these reports. Usually, one theme of research
is represented in four different maps. The first map is region-based and shows the
entire targeted region. The second map covers the research area in Malawi, the third in
Tanzania, and the fourth in Uganda.

On the maps the average values of the district are shown. These district-values are
overlaid by the average values of the smaller adm4. This shows how the adm4 is related
to the bigger district by having the same legend. The legend is also the same for the
different countries, and this allows comparisons among these countries.

Figure 6 is an example of the procedure followed. It shows classes of the average area
of land owned by households in the Eastern part of Tanzania. The district of Handeni
is represented by the average value of 72 households at 2.65 acres of land/household.
Within Handeni, three adm4s are situated: Chanika with 15 households, Kwedizinga
with 13 households and Vibaoni with 44 households. In Chanika, the average household
owns 2.20 acres (below average for the district where it belongs), in Vibaoni 2.61 acres
(average), and in Kwedzinga 3.08 acres (above average).

Natural capital
Natural capital entails the stock of assets embodied in natural endowments, such as
elevation, annual rainfall, infrastructure, and land quantity and quality.

Elevation
To determine the altitude of the households a digital elevation model (DEM) of a 90 m
resolution was used. Figure 7 and Figure 8 show the altitude (m) of all households (Annex
IV at a larger scale). The households in Uganda are positioned between 1.000 and 1300
m in altitude with a majority of the points at around 1100 m. The households in Tanzania
are more or less evenly distributed between 150 and 750 m. The households in Malawi
are situated between 1000 and 1350 m and, as in Uganda, most points are at 1100 m.

12
Figure 6. In the district of Handeni three adm4s are included. The average area of land
owned for the whole district is 2.65 acres of land/household. In the adm4 within this
district, an average household owns 2.20 acres in Chanika, 2.61 acres in Vibaoni, and
3.08 acres in Kwedzinga.

Figure 7. Altitude (m) of all households with max = 1348 m and min = 147 m.

13
Figure 8. Altitude (m) in the area of interest. The black dots represent the households.
Source: Distributed Active Archive Center (http://edcdaac.usgs.gov).

Rainfall
Within the studied area farmers mostly rely on rainfall as their source of water for
agriculture because possibilities of irrigation are often limited. The amount of rainfall/
household is extracted from a 1 km resolution grid displaying the average annual
precipitation in the period 1951 to 2005. Figures 9 and 10 give an idea of the distribution
of rainfall in the area of interest. In Annex V, a map is shown for each targeted country.
In Malawi, the average annual rainfall of the households in that period was between 870
and 1050 mm. In Tanzania the range is a little wider with minimum values of 850 mm and
maximum values of 1120 mm. The eastern part of Uganda appears to be much wetter
with annual minimum values of 1230 mm and maximum values of 1670 mm.

14
Figure 9. Histogram of mean annual rainfall in mm in the period 1951 to 2005 of all
households, with a mean of 1104 mm, a max of 1628 mm and a min of 851mm.

Figure 10. Annual


rainfall in the
area of interest.
The black dots
represent the
households.
Source: Food
and Agriculture
Organization of
the UN (FAO), the
Climatic Research
Unit (CRU), and the
Global Historical
Climatology
Network (GHCN)
(http://www.
worldclim.org).

15
Roads
In principle roads are positively correlated with farmers’ welfare as this type of
infrastructure improves market access (Staal et al. 2000). Getting good and updated
geo-referenced data on road location remains a major problem in this region of Africa. To
save time and costs a standard road shape-file is used. The roads of the three countries
(Table 4) were merged into one shape-file. The attributes of the road-shape-files did
not allow a differentiation in road type. Therefore all roads were regarded as being of
the same quality. As becomes apparent from Figure 11, there seems to be a strong
spatial correlation with the surveyed points and the location of the roads, where most
of the households are near roads. This could be the result of policies that encouraged
households to settle near roads for ease of access to social and health infrastructure.
Figure 11 shows maps of Tanzania and Malawi.

It can be assumed that many of the indicators are somehow related to the roads, as
infrastructure coincides with market access. To verify this assumption the distance from
each household to the nearest road was determined. This distance between a household
and the nearest road ranges from 0 to 11.5 km. Figure 12 shows a very skewed
distribution with a mean distance of 2.1 km. In Malawi, the average distance is 3.2 km, in
Tanzania, 1.6 km and in Uganda, 1.4 km.

Area of land
The area of land/household represents the total area of land used for agricultural
purposes owned by the household. Figure 13 illustrates the distribution of this land area
in Tanzania, Annex VI shows the entire region of interest. In Tanzania the households in
the district of Handeni seem to own most of the land; the average is raised by the ward
Kwedizinga where households own more than 3 acres of farmland. In Malawi households
in the north own up to 2.5 times more land than in the south. In the south, the districts of
Dedza and Lilongwe score in the lowest category. In Uganda, the average size of land
owned in all four districts is 1.5 to 2.0 acres and no obvious trend becomes apparent as
some parishes are below and some above this average.

Share of owned and cultivated land


The share of owned and cultivated land shows how much of the land that is cultivated
during the reference season is actually owned by the household. This indicator may
explain the shortage of land where ideally a household owns 100% of the land it
cultivates. This share is depicted for Uganda in Figure 14 and for all countries in Annex
VII. In Uganda, only a few parishes in Namutumba own less than 70%. Tanzania seems
not to score well in this category, where farmers generally own less than 80% of their
farmed land. Handeni is noteworthy, as most farmers own more than 90% in this district.
Farmers in Malawi seem to own most of their farmed land; only the EPA Mkanda scores
are very low.

16
17
Figure 11. Location of roads in relation to the surveyed points in Tanzania (left) and Malawi (right). Source: ESRI’s Chart of the world,
(http://www.maproom.psu.edu/dcw).
Figure 12. Histogram of distance from households to the nearest road.

Figure 13. Area


of cultivated land
(acres) owned by the
average household
in Tanzania.

18
Figure 14. Share of owned, managed land in Uganda.

Physical capital
Physical capital refers to man-made assets such as productive assets, housing qualities,
and consumable durables. As opposed to natural capital, physical capital can be influenced
by the household.

Household productive asset index


A composite productive asset index (PAI) was developed by combining the number and
the working status of all combined productive assets/household. The index therefore
expresses the tools and their quality at hand. Figure 15 shows how this index varies
within Malawi. Annex VIII shows the resulting maps of the entire region at a bigger scale.

19
Figure 15.
Household
productive
asset index
(PAI) for
Malawi.

On a regional level Tanzania has the most even distribution of the PAI with all households
scoring on or above average. Uganda seems to be worse off, with two entire districts
below average. The difference between the south and the north of Malawi seems
prominent where the PAI in the north is generally higher than in the south.

Human capital
Human capital concerns the people who are both the objects and subjects of
development. Since this study was on smallholder farmers, sources and levels of
accessibility to know-how and human quality were quantified.

20
Number of yearly extension visits/household
The number of extension visits is defined as the annual number of visits to a household
by an extension officer. Visits are expected to introduce more know-how to farmers and
bring more productivity and market information which might result in higher income.
Figure 16 illustrates the distribution in Tanzania and Annex IX, the distribution for all
countries.

In Uganda, households appear to have little contact with extension officers. Only in
the west of Mamutumba do farmers make use of these services where the parish of
Nabitula looks like a hotspot. In Tanzania, only in the district of Mvomero and especially
in the ward of Melela do households frequently consult extension officers. In Malawi, the
distribution is irregularly divided and no clear pattern emerges.

Figure 16. Number of extension visits/household/year in Tanzania.

21
Ill-health index
Health as an indicator of human capital was conceptualized through a morbidity composite ill-
health index (IHI). The IHI includes 10 diseases, fever/malaria, dysentery/diarrhea, respiratory
system-related illnesses, measles, typhoid fever, undernutrition, tuberculosis, HIV/AIDS,
accidents causing injury and lifetime diseases/disorders. The more health problems the
interviewees experienced during the reference period of one year preceding the survey, the
higher the value of IHI. An example of the distribution of this index for Uganda is presented in
Figure 17 and for all three countries in Annex X.

On a regional scale, Malawi scores comparatively well with the majority of the EPAs below
average. In Tanzania, only the farmers in the district of Handeni appear to be in a comparatively
healthy condition. In Uganda the IHI looks more or less randomly distributed. Isolated low-
scoring parishes seem to have a big negative influence on the district’s average, for example,
Naboa, Nawansagwa, and Masaba.

Figure 17. Distribution of ill-health index (IHI) for Uganda.

22
Financial capital
This section describes the financial resources that people use to achieve their livelihood
objectives and is a function of cash or equivalent resources. For the spatial analysis three
indicators were considered: the composite liquidity index (CLI), the livestock ownership,
and household income.

Composite liquidity index


The CLI describes access to a variety of liquidity assets, the ordinal ranks of their
magnitude and ease of raising, commanding and spending cash. These are the most
important factors underlying any form of financial/liquidity capital in the context of
livelihood security. Relative ease in commanding and spending explains, in a given
situation, how easily and quickly money can be raised and spent to meet a financial
obligation. This index is mapped for Malawi in Figure 18 and for all three countries in
Annex XI.

Figure 18. Distribution of the composite liquidity index (CLI) in Malawi.

23
In Tanzania, despite the occasional ward with a high CLA, all districts score on or below
average. In Malawi access to cash appears to be easier; here only about 25% of the
EPAs score below average. There seem to be large differences, however, while in
Tanzania, the distribution seems to be more even. In Uganda also, the distribution of the
CLA appears to be widely dispersed.

Livestock ownership
Farmers can easily sell their livestock to acquire financial resources. To measure this
asset a common unit to describe livestock numbers of various species in a single figure
has been developed. To do this, the concept of an “Exchange Ratio” has been created,
where different species of different average sizes can be compared and described in
relation to a common unit, the Tropical Livestock Unit (TLU). The distribution of this TLU
in Eastern Tanzania is illustrated in Figure 19 and for all three countries in Annex XII.

Figure 19.
Distribution
of livestock
(Tropical
Livestock Units)
in Tanzania.

24
Farmers in Malawi own comparatively few cattle. In the south of the studied area this
becomes even more apparent than in the north. In Tanzania, only in the district of
Handeni does the TLU rise above 1 although this can be entirely contributed to the ward
of Vibaoni. The TLU in both countries is significantly lower than in Uganda where about
50% of the farmers in the parishes have TLUs above 1.

Household income
The household income is the estimated income the household gets from various
enterprises. Ten enterprises were considered in the computation of the total enterprise
income through simple summation. These were crop production, livestock, business,
salaried employment, casual wage-work, technical work, artisan work/handcrafts, natural
resources, traditional medicine/healing and resource rent out. All enterprise incomes
were converted into US dollars. The differences in this income are shown in Figure 20 for
Malawi and for all three countries in Annex XIII.

Figure 20.
Household
income acquired
from various
enterprises
in Malawi.

25
In Tanzania, all districts with the exception of Morogoro have a yearly enterprise income
of more than 50 USD. In Morogoro, income is generally below 50 USD and in the ward of
Mikese, even below 25 USD. The Handeni district, on the other hand, does relatively well.
In Malawi, more than half of the adm4 score below average. The northern part of the study
area seems to be better off than the south. In Uganda only the farmers in the district of
Namutumba have a comparatively high income while Busia had a very low
enterprise income.

Maize and Striga


Maize ranks first of the major cereal grains in many countries of Eastern and Southern
Africa. It is a very important staple food for the entire population as well as a source of
income. The crop is mainly produced by smallholder farmers on small-scale farms of
less than 3 ha. Striga is a root-parasitic flowering plant that causes a considerable loss
in growth and yield of many food and fodder crops. In general, low soil fertility, nitrogen
deficiency, well-drained soils, and water stress accentuate the severity of Striga damage
to the hosts. Striga has a greater impact on human welfare than any other parasitic
angiosperm as its hosts are subsistence crops in marginal agricultural areas (Manyong et
al. 2008a).

Maize yield
The yield of maize, expressed as hectare/ha, fluctuates greatly in the study area.
The distribution is illustrated in Figure 21 for Uganda and in Annex XIV for all countries.

On the whole, the average yield of maize is 1.0 to 1.5 t/ha. This is especially true for
Tanzania where only the district of Mkinga/Muheza has a higher yield. In Malawi, the
pattern seems more erratic with EPAs scoring above and below this average. Uganda
seems to have consistently lower production. Only the district of Budaka/Pallisa has
average yields, but this is the result of only two out of six parishes.

Share of land under improved maize varieties


Farmers produce local and improved maize varieties. Improved varieties are expected to
produce higher yields than local varieties. The proportion of land under improved maize
varieties is calculated as the area where improved varieties are grown, divided by the area
where all maize varieties are grown. Figure 22 shows how households in Malawi adapt this
technology, and Annex XV shows this adaption in the studied countries.

From the maps it becomes clear that there are large regional differences in the adoption
of improved maize varieties. In Malawi, it appears to be an accepted and widespread
technique, where more than 40% of farmers make use of these varieties in the western
part of the country. In Tanzania, only the farmers in the district of Mvomero frequently use
improved seeds. In Uganda, the distribution appears dispersed, although the improved
varieties appear to be unpopular in the district of Busia.

Share of land under intercropping


Another technique used to solve the Striga problem is intercropping. Intercropping involves
planting legumes, not susceptible to Striga, in rows alternately with maize. As the legumes
are not a host to Striga, no parasitism is expressed and the Striga seedbank is decreased.

26
Figure 21. Maize yield (t/ha) in Uganda.

27
Figure 22. Share of land under improved maize varieties in Tanzania.

28
The share of land under intercropping is calculated by determining the area of land under
intercropping divided by the total land under maize. The adoption of this technique is
illustrated in Figure 23 for Malawi and in Annex XVI for all three countries.

In Tanzania, the technique is generally disliked except in the district of Morogoro where
more than 25% of cultivated land is farmed using intercropping. This could be related to
the presence of Striga as it appears to show the same spatial pattern and might confirm
expectations that the parasite is combated with intercropping. In Malawi, intercropping
emerges as very popular in the south of the researched area where more than 75% of

Figure 23. Share of land under intercropping in Malawi.

29
farmed land is intercropped. Also in Uganda the method seems popular with all districts
on average and the parishes scoring below or above are more or less equally distributed.

Share of land infested by Striga


The share of land infested by Striga is calculated by determining the area of land under
maize affected by Striga divided by the total land under maize. Figure 24 shows the
distribution of Striga in Malawi and Annex XVII shows this for the entire region.

In Malawi, the situation is rather severe where almost the entire targeted area is facing
Striga on more than 40% of land under maize plots. In the south there are a few EPAs
that seem to escape the threat, but considering the fact that they are surrounded by
more diseased EPAs they too will probably be swiftly infected, provided no appropriate

Figure 24. Striga infestation as a percentage of total land under maize in Malawi.

30
action is taken. It is noticeable that Striga is not as common in Tanzania as in Malawi and
Uganda. The problem seems to be concentrated around the district of Morogoro where
up to 60% of farmland under maize is infected. Ten years ago only the ward of Kisemu
within this district had a problem. It might therefore well be that Striga spread from that
ward or a neighboring ward not sampled. In Uganda the situation seems very serious
with more than 60% of land infected in two districts. In the southern district of Busia, two
parishes even have infection values higher than 80%. Nevertheless these farmers do not
seem tempted to use combating methods, such as improved varieties and intercropping,
probably because they have not been exposed to these technologies.

Share of land infested by Striga 10 years ago


Farmers were asked to estimate the percentage of their land area under maize that
was infested with Striga 10 years ago. As becomes clear from Figure 25 for Malawi and
from Annex XVIII for all countries, Striga was a relatively limited problem 10 years ago.
Nevertheless, this legend was used to allow comparisons with the Striga situation at this
moment. Striga infestation was already a problem in Malawi 10 years ago, especially in
the central and northern parts of the researched area where values between 20% and
40% are not uncommon. In Tanzania, Striga occupied more than 20% of total farmland
only in the ward of Kisemu. In Uganda, only the eastern part of the area of interest seems
to experience problems with Striga. When compared to the situation at this moment the
spread of Striga is very serious and has increased strongly in virtually all
administrative units.

Livelihood outcomes
Different forms of capital ultimately result in a series of outcomes. The exact composition
of these outcomes is discussed in the regional report (Manyong et al. 2008a). For the
spatial analysis, the body mass index and the wealth index of the households
were considered.

Body mass index


The body mass index (BMI) is a measure of the nutritional status of adults, expressing
the health effects of body weight relative to height. A BMI score between 22 and 24 is
considered normal. Below 22, an individual is underweight and possibly malnourished.
Above 24, an individual is overweight or obese. Both underweight and overweight
individuals have increased relative risks relative to morbidity and mortality compared to
those of “normal” weight. The BMI was recorded for the mother or the guardian of each
household. Results for Tanzania are shown in Figure 26 and in Annex XIX for
all countries.

In Tanzania, women are overweight in Handeni but underweight in two of the other
districts. In Mvomero, the district average is normal, despite the fact women in the wards
of Hembeti and Mvomero are underweight. In Malawi, women are generally underweight
as the BMI is at normal levels only in four adm4. This pattern also arises in Uganda
where the women in all districts are clearly underweight, despite the fact that a few adm4
have normal values.

31
Figure 25. Striga infestation 10 years ago as a percentage of total land under maize in Malawi.

Country wealth index


A wealth index was computed by aggregating the various asset ownerships and housing
characteristics variables, based on the method of principal components. The asset
variables considered in the analysis were related to main building quality (roofing,
wall, floor, toilet, and extra house), consumable durables (iron/wooden bed, iron, sofa,
spongy mattress and watch/wall clock), communication means (television, cell phone,
landline, radio) energy and water source (energy for cooking, energy for lighting and
source of water) and transport means (car, motorbike and bicycle) making the total of 20
variables. The index is mapped for Tanzania in Figure 27 and for all countries in Annex
XX. The index is normalized for each country so comparisons are possible only within
the three countries.

32
Figure 26. Distribution of
body mass index (BMI) of
adult women in Tanzania.

Figure 27. Distribution of


country wealth index in
Tanzania.

33
In Malawi, the northern part of the studied area is clearly wealthier than the south. In
Tanzania, Morogoro and Mkinga/Muheza appear to be the poorest districts. In Uganda,
wealth also seems rather dispersed with poorer and richer parishes divided over the
country. Unlike in Malawi, the richer part of the studied area appears to be in the south
of Uganda.

Regional wealth index


The regional wealth index is the same as the wealth index described earlier; the only
difference is that the index is normalized over the entire area of interest. This allows
comparisons between the three researched countries (Figure 28 and Annex XXI).

Tanzania is clearly the wealthiest of the three countries. Even the poor district of
Morogoro scores above average in this respect. Malawi is undoubtedly the poorest,
with the southern part of the country in the worst position. In Uganda, the wealth index
seems to differ over the country, although three out of four districts score above the
regional average.

Figure 28.
Distribution of
regional wealth
index in the three
countries.

34
Results II. Spatial analysis by interpolation
In the previous section, a great number of research themes were converted from points
to areas by averaging households over administrative units (Results I). This method
shows the spatial spread of the researched themes but does not say anything about the
distribution and density of points within the administrative units. In this section, the spatial
interpolation technique Ordinary Kriging will be used to assess the spatial distribution
of Striga infestation, maize yield and the country wealth index. The technique itself is
described in Methodology.

Striga infestation
In this section Striga infestation will be explored using the data collected in 287
households in Malawi. The histogram (Fig. 29) shows that between 0 and 100% of
farmland is infested with Striga in the targeted area. The distribution appears to be normal
and there are no values that are considered to be outliers.

Figure 30 shows the results of applying the spatial interpolation technique ordinary
kriging on 287 households in Malawi. For cosmetic reasons the map extent taken into
consideration is the entire country of Malawi. The households themselves are shown on
the map as black dots. It appears farmers in the Northwestern part of Malawi suffer more
from Striga than those in the Southeast.

Figure 31 shows the Striga infestation based on the values of the same 287 households
averaged over the administrative units. Although the legends are slightly different, the
same pattern emerges as in Figure 30.

In Figure 32, the level of confidence clearly decreases as the distance to the households
increases. A darker color means a lower level of confidence, or (in other words) a higher
error in prediction. The map also shows that the level of confidence is higher in between
households than at the outer edges of household clusters. This suggests the sampling
design is important and a lower prediction error can be reached by optimizing this design.
Annex XXII shows both the prediction and its confidence.

Figure 29. Histogram of the percentage of farmland infested with Striga of 287 households
in Malawi (mean = 47.3 and skewness = -0.2).

35
Figure 30. Predicted Striga
infestation in Malawi based on 287
households (black dots).

Figure 31. Striga infestation


in Malawi based on 287
households (black dots).

36
Figure 32.
Confidence level
of the predicted
Striga infestation.

Figure 33. Histogram showing the distribution of maize yield (t/ha) of the 271 households in
Tanzania (mean = 1.1 and skewness = 0.9).

Maize yield
This explores maize yield using the data collected in 281 households in Tanzania. Ten
households have unrealistically high maize yields of over 7.0 t/ha and are therefore
omitted from further analysis. The histogram (Fig. 33) shows that the maize yield of the
remaining 271 households in Tanzania ranges from 0 to 7.0 t/ha.

37
Figure 34. Predicted maize yield (t/ha) in Tanzania.

Figure 34 shows the predicted maize yield of the target area in Tanzania in t/ha. The
prediction is based on 271 households, shown on the map as black dots (Annex XXIII).
There appears to be a region of high production in the center of the studied area.

Figure 35 shows the maize yield based on the values of the same 271 households
averaged over the administrative units. The pattern that appeared in the predicted map
(Fig. 34) can also be distinguished here. The center area with high production might be
the result of the households in Mvomero that have relatively high yields.

The map (Fig. 36) shows that the level of confidence depends on the distance to the
households. As in Figure 32, the prediction error increases with distance from the
households. It also suggests that the prediction error is lower in a location surrounded
by households.

Annex XXIII shows the predicted maize yield and its confidence level at a larger scale.

38
Figure 35. Maize yield (t/ha)
in Tanzania based on 271
households (black dots).

Figure 36. Confidence level


of the predicted maize yield
in Tanzania.

39
Figure 37. Histogram showing the distribution of the country wealth index of 287
households in Uganda (mean = 0.0 and skewness = –0.2).

Country wealth index


In this section the country wealth index is analyzed using the data collected from 287
households in Uganda. The histogram (Fig. 37) shows the wealth index ranges from
–12.0 to 14.0 and is normally distributed.

Figure 38 predicts the country wealth index of the Southeastern part of Uganda based
on 287 households, shown on the maps as black dots (Annex XXIV). It seems that the
Southern part of the country (in particular the Southwest) is richer than the Northern part.

Figure 38. Predicted


country wealth index in
Uganda based on 287
households (black dots).

40
Figure 39. Country wealth index based on 287 households in Uganda.

Figure 39 shows the distribution of the country wealth index in Uganda. As in Figure 38,
the Southwest appears to be richer. The influence of the high average values in the south
of the district of Namutumba is clearly visible on the predicted map.

Figure 40 illustrates the level of confidence of the predicted country wealth index. The
error in prediction increases with increasing distance from the households and is strongly
influenced by the spatial design of the study.

41
Figure 40. Confidence level of the predicted country wealth index in Uganda.

42
Conclusions and recommendations
The aim of this report is to demonstrate the strength of GIS in visualizing and analyzing
economic surveys and to link them to geophysical datasets. The theme- and indicator-
specific conclusions are drawn in Results I and II. Several general conclusions can be
made.

The report proves that it is certainly possible to expose patterns and trends spatially in
a database that is designed for socioeconomic purposes. Therefore, spatial statistics
can be used to predict socioeconomic properties of unvisited locations by using the
interpolation techniques presented. Many of the maps allow a quick impression of the
status of the studied area without extensive study of literature and tables. Many variables
and indicators are clearly related to space. This is especially true in Malawi where many
maps show a clear gradient from the ”poor” south to the ”rich” north. Many other maps
in Uganda and Tanzania seem to suggest a similar correlation in space, as nearby
administrative units tend to have similar values or indicators.

In this study, two methodologies are presented that allow visualizing and analyzing
socioeconomic data spatially. In the first method, the household values are averaged
within an administrative unit. In the second method, Kriging is used to predict the values
of locations not visited. The methods are comparable since they both interpolate from the
point-level (households) to the surface level (area of interest). A visual inspection seems
to suggest the two interpolation methods (averaging and Kriging) produce similar results.

From the original 901 households, 45 were omitted from the analysis because the
locations of these households were improbable. The remaining 856 households were
assumed to be at the correct location.

It is likely that the same number of surveyed households and/or of a larger area could
have produced similar results with greater accuracy. This could have been achieved
by spatially optimizing the survey design. The confidence maps generated in Results II
provide a useful tool to optimize future surveys.

This report produced results interesting enough to justify further research. The
methodology of Kriging seems especially promising (Results II). This report can be
viewed as a pre-analysis of the dataset that fits very well in a multistep process, such
as Kriging.

An important subject that has not adequately been covered is the spatial correlation of
features and indicators. For instance, the relationship between roads (Results I) and the
country wealth index seems of interest. Another example is a quantification of the spatial
relationship between maize yield and Striga infestation.

The maps of Malawi show an attached area; the maps of Tanzania and Uganda show
isolated districts. The advantage of an attached area is that it looks better and it exposes
a gradient (trend) of an indicator across the whole area of interest. This trend can be used
to forecast something of a neighboring, not sampled, area, especially because of the
apparently strong link between socioeconomics and GEO-statistics.

43
To avoid misinterpretation of coordinates and the resulting spatial error, a course on the
proper use of GPS units is strongly recommended for enumerators before they go into the
field, especially on the subject of spatial projection.

Coordinates should always be recorded using the WGS-84 map datum and in the
degree°decimal-format.

In this survey the coordinates of the households were copied manually by the
enumerators from the GPS to the questionnaires. To decrease possible errors, a unique
ID should be designated for each household and the coordinates of this household should
be stored on the GPS. At a later stage, the coordinates can be exported via a cable from
the GPS to a computer. Because of the unique ID, the coordinates can be linked to the
questionnaires. Coordinates in doubt can be validated on the appropriate questionnaires.

Every household owned one cultivated field or more. The coordinates of these fields
were recorded only in Tanzania and Malawi. The measurement of the area of the fields
was not possible on all the GPS units. Ideally, up-to-date GPS devices should be used
that can record not only the area of separate fields but can also store the shape of these
fields. Via the unique ID, the fields could be easily linked to the households. This would
allow many additional analyses.

The findings of this report show that the survey was set up more according to
socioeconomic criteria rather than spatial criteria. Nevertheless, any economic study
can profit from spatial analysis. It might even be said that if extra spatial data had been
collected, such as market locations and road networks on the village level, these could
have added value to this socioeconomic study. Therefore, a more multidisciplinary
approach is emphasized in which the sampling design allows spatial analysis.
This study aims to understand the effects that Striga has on the livelihoods of the poor.
The sampled households were, however, always located in areas known to be heavily
infested with Striga. It would be interesting to expand the area of interest to locations that
are not heavily infested and see if this has any effect on the researched indicators.
Mapping Striga distribution in the Eastern Striga belt would be worthwhile.

44
References
AATF (African Agricultural Technology Foundation). 2006. Empowering African farmers
to eradicate Striga from maize croplands. AATF, Nairobi, Kenya.
Arbia, G. 1993. The Use of GIS in Spatial Statistical Surveys, International Statistical
Review, Great Britain.
Diggle, P.J., P.J. Ribeiro. 2007. Model-based Geostatistics, Department of Mathematics
and Statistics, Lancaster University, Lancaster, UK.
Fais, A., P. Nino, and A. Giampaolo. 2005. Microeconomic and GEO-Physical Data
Integration for Agri Environmental Analysis, GEO-referencing FADN Data: A Case
Study in Italy, National Institute for Agricultural Economics, Italy.
Johansson, J. 2005. Improving Access to Geographic Information Systems,
University of Umea, Sweden.
Legg, C., P. Kormawa, B. Maziya-Dixon, et al. 2005. A Report on Mapping Livelihoods
and Nutrition in Nigeria using data from the National Rural Livelihoods Survey
and the National Food Consumption and Nutrition Survey, International Institute of
Tropical Agriculture (IITA), Ibadan, Nigeria.
Manyong, V.M., K.D. Mutabazi, A.D. Alene et al. 2008a. Livelihoods of smallholder
farmers in Striga-affected maize growing areas of Eastern and Southern Africa,
AATF/IITA Baseline Study, IITA, Tanzania. Regional report.
Manyong, V.M., K.D. Mutabazi, C. Maeda, et al. 2008b. Livelihoods of smallholder
farmers in Striga-infested maize growing areas of Eastern Tanzania, AATF/IITA
Baseline Study, IITA, Tanzania. Country report.
Manyong, V.M., K.D. Mutabazi, A.D. Alene, G. Omanya, H.D. Mignouna, and
M. Bokanga. 2008c. Livelihoods of smallholder farmers in Striga-infested maize
growing areas of Central Malawi, AATF/IITA Baseline Study, IITA a. Country report.
Manyong, V.M., K.D. Mutabazi, E. Rutto, A.D. Alene, G. Omanya, H.D. Mignouna, and
M. Bokanga. 2008d. Livelihoods of smallholder farmers in Striga-infested maize
growing areas of Eastern Uganda, AATF/IITA Baseline Study, IITA, Tanzania. Country
report.
Smith, M.J., M.F. Goodchild, and P.A. Longley. 2007. Geospatial Analysis —
a comprehensive guide to principles, techniques and software tools, Second edition.
Issue version: 2.16, online version - http://www.spatialanalysisonline.com.
Staal, S.J., C. Delgado, I. Baltenweck, and R. Kruska. 2000. Spatial aspects of producer
milk price formation in Kenya: A joint household GIS-approach, ILRI/IFPRI, Kenya.

GEO-data
Elevation model: Land Processes Distributed Active Archive Center (LP DAAC), Aster
Digital Elevation Model, resolution 30m, http://edcdaac.usgs.gov, supplied by the
GIS-laboratory, IITA, Ibadan, Nigeria.
Parishes Uganda: Intergovernmental Authority on Development (IGAD), http://ergodd.
zoo.ox.ac.uk/igadweb/tiki-index.php?page=Data+Archive Uganda.
Rainfall data: Food and Agriculture Organization of the UN (FAO), the Climatic Research
Unit (CRU) and the Global Historical Climatology Network (GHCN), http://www.
worldclim.org, resolution 1 km, supplied by the GIS-laboratory, IITA, Ibadan, Nigeria.
Roads of Malawi, Tanzania and Uganda: created by ESRI from US Operational
Navigation Chart (ONC) series, Chart of the world (DCW), http://www.maproom.psu.
edu/dcw, supplied by the GIS-laboratory, IITA, Ibadan, Nigeria.

45
Annex I. Distribution of the surveyed points

46
47
48
49
Annex II.
II. Frequency
Frequencyofofhouseholds
householdsper
peradministrative
administrative unit
unit
This table lists the amount of points which are located in each adm3. These points determine the average
of the district and are therefore of great importance. The value of Vibaoni in Tanania is for example the
average of 44 households while the adm3 Mpingu in Malawi only has 3 households within its boundary.

country (adm1) district (adm3) adm4 FREQUENCY


MALAWI DEDZA CHAFUMBWA 8
MALAWI DEDZA KABWAZI 15
MALAWI DEDZA KAPHUKA 15
MALAWI DEDZA LINTHIPE 14
MALAWI DEDZA LOBI 15
MALAWI KASUNGU CHULU 14
MALAWI KASUNGU KALULUMA 15
MALAWI KASUNGU KASUNGU 13
MALAWI KASUNGU LISASADZI 15
MALAWI KASUNGU SANTHE 15
MALAWI LILONGWE CHILEKA 15
MALAWI LILONGWE CHITEKWER 14
MALAWI LILONGWE CHITSIME 15
MALAWI LILONGWE KAWAMBA 14
MALAWI LILONGWE MLOMBA 2
MALAWI LILONGWE MPINGU 14
MALAWI MCHINJI CHIOSHYA 15
MALAWI MCHINJI KALULU 14
MALAWI MCHINJI MIKUNDI 15
MALAWI MCHINJI MKANDA 15
MALAWI MCHINJI MSITU 15
TANZANIA HANDENI CHANIKA 15
TANZANIA HANDENI KWEDIZINGA 13
TANZANIA HANDENI VIBAONI 44
TANZANIA MKINGA/MUHEZA LUSANGA 29
TANZANIA MKINGA/MUHEZA MARAMBA 29
TANZANIA MKINGA/MUHEZA MUHINDURO 17
TANZANIA MOROGORO RURAL KISEMU 11
TANZANIA MOROGORO RURAL MIKESE 42
TANZANIA MOROGORO RURAL MKUYUNI 14
TANZANIA MVOMERO HEMBETI 14
TANZANIA MVOMERO MELELA 26
TANZANIA MVOMERO MVOMERO 12
TANZANIA MVOMERO MZUMBE 15
UGANDA BUSIA BUSIKHO 7
UGANDA BUSIA BUTANGASI 4
UGANDA BUSIA JINJA 2
UGANDA BUSIA KUBO 7
UGANDA BUSIA LUMINO 7
UGANDA BUSIA LUNYO 14
UGANDA BUSIA MASABA 10
UGANDA BUSIA NAGABITA 5
UGANDA BUSIA NALWIRE 7
UGANDA BUSIA SAPIRI 1
UGANDA BUSIA SIKUDA 8
UGANDA NAMUTUMBA BUWAGA 8
UGANDA NAMUTUMBA IVUKULA 6
UGANDA NAMUTUMBA KISOWOZI 7
UGANDA NAMUTUMBA NABITULA 15
UGANDA NAMUTUMBA NAKALOKWE 8
UGANDA NAMUTUMBA NAMUTUMBA 7
UGANDA NAMUTUMBA NAWANKOFU 7
UGANDA NAMUTUMBA NAWANSAGWA 15
UGANDA PALLISA CHALI 6
UGANDA PALLISA JAMI 7
UGANDA PALLISA KAMONKOLI 18
UGANDA PALLISA LYAMA 21
UGANDA PALLISA NABOA 7
UGANDA PALLISA SAPIRI 14
UGANDA TORORO AMONI 6
UGANDA TORORO KALAIT 8
UGANDA TORORO KUITANGIRO 13
UGANDA TORORO MAGOLA 16
UGANDA TORORO MELLA 7
UGANDA TORORO NYAMALOGO 7
UGANDA TORORO PABONE 13

50

50
Annex III. Distribution of administrative units

51
52
53
54
Annex IV. Altitude of surveyed households

55
Annex V. Mean annual rainfall in the period 1951 to 2005

56
57
58
59
Annex VI. Area of land (acre)

60
61
62
Annex VII. Share of owned cultivated land

63
64
65
Annex VIII. Productive asset index

66
67
68
Annex IX. Number of extension visits

69
70
71
Annex X. Ill-health index

72
73
74
Annex XI. Overall composite liquidity index

75
76
77
Annex XII. Tropical livestock units

78
79
80
Annex XIII. Enterprise income per capita (US$/yr)

81
82
83
Annex XIV. Overall maize yield (tonnes/ha)

84
85
86
Annex XV. Share of land under hybrid maize

87
88
89
Annex XVI. Share of land under intercropping

90
91
92
Annex XVII. Striga infestation as percentage of total land area

93
94
95
Annex XVIII. Striga infestation as percentage of total land area
10 years ago

96
97
98
Annex XIX. Body mass index

99
100
101
Annex XX. Country wealth index

102
103
104
Annex XXI. Regional wealth index

105
106
107
108
Annex XXII. Predicted Striga infestation

109
110
Annex XXIII. Predicted maize yield

111
112
Annex XXIV. Predicted country health index

113
114

You might also like