Okwi et al. 10.1073/pnas.0611107104.

Supporting Information

Files in this Data Supplement:

Supporting Appendix

Poverty Mapping in Kenya.

The poverty mapping g exercise in Kenya makes use of two household data sets: the 1997 Welfare Monitoring Survey and the 1999 Population and Housing Census. Poverty mapping is one application of the method called small area estimation. The method is typically divided into three stages:

• Stage 0 involves identifying variables that describe household characteristics that may be related to income and poverty and that exist in both the household survey and in the census

• Stage 1 estimates a measure of welfare, usually per capita expenditure, as a function of these household characteristics using regression analysis and the household level data.

• Stage 2 applies this regression to the same household characteristics in the census data, generating predicted welfare for each household in the census. This information is then generated up to the desired administrative unit, such as district or Location, to estimate the incidence of poverty and the standard error of the poverty estimate.

Details of this approach and the resulting poverty estimates are explained in CBS and ILRI, 2003.

Diagnostic Tests for Spatial Dependence.

The results of the diagnostic tests show the presence of spatial dependence, as they are all highly significant (Table A1). Thus, we chose to use a spatial regression model to control for this spatial autocorrelation.

Table A1. Diagnostics for spatial dependence (OLS model) poverty rate

For weight matrix :(row-standardized weights)

Test

Value

Probability

Moran's I

2.97322

0.00295

Lagrange multiplier (lag)

0.77618

0.37831

Robust LM (lag)

6.87594

0.00874

Lagrange multiplier (error)

5.35721

0.02064

Robust LM (error)

11.45697

0.00071

 

In the spatial error model, a full set of variables hypothesized to have some spatial relationship with community level poverty is included. The model fit increases to 0.54, which is not a huge change from the OLS model, but by removing the nuisance caused by spatial autocorrelation, we can now have more confidence in our parameter estimates, and concentrate on the variables that are showing a strong spatial relationship to poverty prevalence at the location level.

Distance Bands.

To assess spatial autocorrelation, we must develop a spatial weights matrix to define exactly the "neighborhood" for each rural aggregated location. There are many ways to assign neighbor weights and the choice depends on the type of spatial application and on the research question. This specification requires a priori knowledge of the range and intensity of the spatial covariance between regions. Common methods include row standardization, length of common boundary and distance functions [Anselin L (2002) Agricult Econ 27:247-267]. In this study, we used inverse distance band weights for each province. Different distance bands (in meters) are used for each region: Central = 11,573, Coast = 46,775, Eastern = 46,775, North Eastern = 50,542.63, Nyanza = 9,160, Rift Valley = 9,160, Western = 9,160. We conducted sensitivity analysis of the results obtained by using different weighting schemes.

Table A2. Description of variables

Short description

Source

Explanation

Agroclimatological

Annual Rainfall (mm)

The WorldClim interpolated global terrestrial climate surfaces for the year 2000. Version 1.3.

The average annual rainfall within the location boundaries, calculated as the sum of all the monthly rainfall figures derived from the original Worldclim1.3 dataset of monthly layers.

Rainfall coefficient of variation

The WorldClim interpolated global terrestrial climate surfaces for the year 2000. Version 1.3.

The average coefficient of variation (CV) of rainfall between the months within 1 year within the location boundaries. This variable was derived from the worldclim1.3 dataset of bio-climatic information, which describes the "rainfall seasonality".

Distance and access to services

Travel time to municipality

- Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

- NASA, Shuttle Radar Topography Mission (SRTM) - 2000

- World Database on Protected Areas (WDPA -

sea.unep-wcmc.org/wdbpa) - based on data from 1963 - 2002

- Roads - ILRI 1999

- Settlements - 1997

This variable represents the average travel time from any place within the location to the nearest municipality (according to definitions of CBS). Travel time is a function of slope, road type and "impediments" (i.e. wetlands, water bodies and natural parks). The table below summarizes the travel times:

Travel time to town

Idem above

This variable represents the average travel time from any place within the location to the nearest town (according to definitions of CBS).

Travel time to trade centre

Idem above

This variable represents the average travel time from any place within the location to the nearest trade centre (according to definitions of CBS).

Travel time to market centre

Idem above

This variable represents the average travel time from any place within the location to the nearest market centre (according to definitions of CBS).

Travel time to type 1 road

Idem above

This variable represents the average travel time from any place within the location to the nearest road of type 1.

Type 1: Tarmac/All Weather Bound

Type 2: Murram/All Weather Loose

Type 3: Earth/Dry Weather

Travel time to type 1 or 2 road

Idem above

This variable represents the average travel time from any place within the location to the nearest road of type 1 or 2.

Travel time to type 1, 2 or 3 road

Idem above

This variable represents the average travel time from any place within the location to the nearest road of type 1, 2 or 3.

Travel time to type 1, 2 or 3 road

Idem above

This variable represents the average travel time from any place within the location to the nearest road of type 1, 2 or 3.

Land use

Percent of location under protected area

World Database on Protected Areas (WDPA -

sea.unep-wcmc.org/wdbpa) - based on data from 1963 - 2002

This variable represents the percent of location that is under the Protected Area.

Percent of location under wetlands

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The original land cover was produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999. Wetland areas are extracted on the basis of code1 of the original layer (considered to be wetland areas)

Percent of location arable land (I.e. LGP > 60 days)

Jones P.G., 2004. Report on preparation of growing season days coverages for Hadley 3 scenarios A2 and B2 for the year 2000. Consultant's report, ILRI

The variable describes the percentage of the location that is arable. Arable land was defined as land with a length of growing period of more than 60 days per year.

Arable land between 30 and 60 % (1=yes ; 0=no)

Jones P.G., 2004. Report on preparation of growing season days coverages for Hadley 3 scenarios A2 and B2 for the year 2000. Consultant's report, ILRI.

This variable takes a value of 1 if the arable land is 30-60% of the location's area, and 0 otherwise. Arable land was defined as land with a length of growing period of more than 60 days per year.

Percent of location under water

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The original land cover has been produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999. Water areas extracted on the basis of code1 of the original layer (considered to be water bodies: 7WP, 7WP-Y, 8WFP).

Percent of location that is built up

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The original land cover has been produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999. Build-up areas extracted on the basis of code1 of the original layer (considered to be build-up areas: 5U, 5UC, 5UR, 5I, 5A).

Percent of location under forest

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The original land cover has been produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999.

Forest areas extracted on the basis of code1 of the original layer (considered to be forested areas). The resulting shapefile was converted to a raster with the following values: 100 = forest (covering about 100% of the area); 65 = mixed forest (covering approx. 65% of the area; 0 = non-forest

Percent of location under farmland

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The variable contains the percentage of the location's area that is under agricultural land.

The original land cover has been produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999.

Farming areas were extracted on the basis of code1 of the original layer (considered to be agricultural areas). The resulting shapefile was converted to a raster with the following values: 100 = agriculture (covering about 100% of the area); 65 = mixed agriculture (covering approx. 65% of the area);

0 = non-agriculture

Percent of location under grass

Africover landcover multipurpose database (FAO) - based on LANDSAT TM 1999

The original land cover has been produced from visual interpretation of digitally enhanced LANDSAT TM images (Bands 4,3,2) acquired mainly in 1999.

Grass areas extracted on the basis of code1 and code2 of the original layer (considered to be grassland areas)

Natural factors

 

 

Arable land more than 60 % (1=yes ; 0=no)

Jones P.G., 2004. Report on preparation of growing season days coverages for Hadley 3 scenarios A2 and B2 for the year 2000. Consultant's report, ILRI.

This variable takes a value of 1 if the arable land is more than 60% of the location's area, and 0 otherwise. Arable land was defined as land with a length of growing period of more than 60 days per year.

Percent of location with Arid or Semi-Arid land (i.e. LGP <= 180 days)

Jones P.G., 2004. Report on preparation of growing season days coverages for Hadley 3 scenarios A2 and B2 for the year 2000. Consultant's report, ILRI.

This variable describes the percentage of the location that is arid or semi-arid (ASAL). ASAL was defined as land with a length of growing period of less than 180 days per year.

Elevation (masl)

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The average elevation in meters above sea level within the location.

Percent of location Steep land (I.e. > 10%)

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

This variable represents the percentage of the location's area that is defined as steep. Steep land was defined as having a slope of more than 10%. The slope was calculated based on the elevation and can be expressed in degrees or percent.

Percent of location with 0 - 4% slope

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The percentage of the location's area with a slope between 0 and 4 %.

Percent of location with 4 - 8% slope

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The percentage of the location's area with a slope between 4 and 8 %.

Percent of location with 8 - 15% slope

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The percentage of the location's area with a slope between 8 and 15 %.

Percent of location with 15 - 30% slope

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The percentage of the location's area with a slope between 15 and 30 %

Percent of location with over 30% slope

NASA, Shuttle Radar Topography Mission (SRTM) - 2000

The percentage of the location's area with a slope of more than 30 %.

 

Soil Classification.

The soil suitability was based on the classifications from morphological sequence where soils in the highland areas i.e., Andosols and Nitisols are classified as of good quality whereas those in the low lying areas comprising soils such as Gleysols and solonetz soils are classified as of poor quality due to saturation with base and poor drainage, thus poor in terms of plant growth. The soil suitability refers to the whole land surface within the country and is not restricted only to the arable lands and is based on the dominant type in a location.

Table A3: Descriptive statistics

Variable

Label

Mean

Std. Dev.

Min

Max

popden

Population density

205.64

226.05

0.12

2431.78

Elevation

Elevation

1345.19

659.38

2.82

3087.83

distance to forest

Distance to forest (m)

5658

8114

0.00

46756

Perc_water

Percent of location under water

0.44

2.63

0.00

60.20

Perc_built

Percent of location built

0.14

0.79

0.00

13.85

Perc_for

Percent of location under forest

4.42

11.13

0.00

84.66

Perc_farmland

Percent of location under farmland

28.28

27.26

0.00

97.95

Perc_grass

Percent of location under grassland

17.42

14.26

0.00

82.11

Perc_wooded

Percent of location under wooded

20.35

22.24

0.00

100.00

Perc_prota

Percent of location under protected area

1.60

9.39

0.00

100.00

Perc_wetlands

Percent of location under wetlands

1.60

6.13

0.00

97.43

Perc0_4slop

Percent of location with 0_4 slope

43.60

32.87

0.00

100.00

Perc4_8slop

Percent of location with 4_8 slope

24.08

16.18

0.00

73.76

Perc8_15slop

Percent of location with 8_15slop

17.14

14.79

0.00

59.27

Perc15_30slop

Percent of location with 15_30slope

10.79

13.26

0.00

62.13

Perc30_abovesl

Percent of location with 30_aboveslope

4.40

8.79

0.00

70.87

t_trav_munic

Travel time to municipality (minutes)

296.49

383.67

11.07

4417.17

t_trav_town

Travel time to town (minutes)

201.07

295.94

7.94

4323.98

t_trav_tcentre

Travel time to trading centre (minutes)

167.46

284.54

7.79

4323.98

t_trav_mrkt

Travel time to market (minutes)

128.81

254.57

7.53

3933.49

t_trav_road1

Travel time to road 1(tarmac) (minutes)

229.44

365.34

5.46

4308.73

t_trav_road12

Travel time to road 1 or 2 ( tarmac or murram) (minutes)

175.17

297.74

5.46

4275.04

t_trav_raod123

Travel time to 1or 2 or 3 (tarmac, murram or dirt)(minutes)

116.49

251.88

3.93

3939.50

t_trav_hc

Travel time to health centre (minutes)

131.64

136.65

8.87

1302.03

Flood

Flood potential (Dummy)

0.40

0.49

0.00

1.00

Cvrain

Coefficient of variation (rainfall)

63.96

27.98

30.00

131.58

NDVI

Normalized Difference Vegetation Index

0.70

0.10

0.37

0.86

av_rainfall

Average rainfall (mm)

961.42

512.05

0.00

1987.00

lgparidsemi180

Length of growing period (LGP) 180 days

20.10

38.53

0.00

100.00

lgp60days

Length of growing period (less than 60days)

95.56

19.21

0.00

100.00

d_dist_disthosp

Distance to district hospital (meters)

24983.52

28494.55

1508.18

160466.00

d_dist_dispen

Distance to dispensary (meters)

7574.55

7987.97

1022.44

64203.69

Goodsoil

Good soil (dummy)

0.44

0.50

0.00

1.00

d_dist_10k2

Distance to nearest town of 10,000 pple (meters)

29853.15

32467.06

1407.87

234269.70

d_dist_50k2

Distance to nearest town of 50,000 pple (meters)

70740.03

108082.70

1756.27

547139.10

d_dist_100k2

Distance to nearest town of 100,000 pple (ms)

92253.60

121926.80

2366.59

638788.10

d_dist_200k2

Distance to nearest town of 200,000 pple(m)

152382.00

152727.90

4892.93

798293.00

Table A4: Ordinary least squares (OLS) estimation

Dependent variable: poverty rate

 

 

Variable

Coefficient

t-statistic

Population density

-0.0001

(5.0228)**

Average elevation (meters above sea level)

0.0000

(2.5578)*

reg2 (Central)

-0.1400

(14.0785)**

reg3 (Coast)

0.0622

(4.5167)**

reg4 (East)

0.0983

(12.1229)**

reg5 (North Eastern)

0.1817

(13.3528)**

reg6 (Nyanza)

0.1418

(15.8019)**

reg8 (Western)

0.0998

(9.4043)**

Percent of location under grass

-0.0016

(6.0789)**

Percent of location under farmland

0.0002

-1.1856

Percent of location wooded

0.0005

(3.3277)**

Percent of location that is built up

-0.0125

(4.3257)**

Percent of location with 4-8% slope

0.0013

(6.4337)**

Percent of location with 8-15% slope

0.0001

-0.4054

Percent of location with 15-30% slope

-0.0001

-0.3696

Percent of location with >30% slope

0.0021

(5.7023)**

Percent of location with LGP <60 days

0.0005

(3.6689)**

Percent of location with LGP 180 days

-0.0006

(5.9913)**

Rangeland (dummy)

0.0109

-1.6311

Good soil (dummy)

-0.0106

(2.0023)*

Average travel time to type 1 or 2 road (minutes)

ff0.0000

-1.6445

Mean distance to district hospital

-0.0002

-1.3807

Constant

0.7901

(25.4316)**

Observations

2232

 

Adjusted R2

0.5114

 

Akaike info criterion : -3,818.37

Log likelihood : 1933.19

 

Absolute value of t-statistics in parentheses

 

 

*, significant at 5% level; **, significant at 1% level

 

 

 

Table A5. Diagnostics for spatial dependence by province

 

 

Spatial error:

 

Spatial lag:

 

Statistic

Moran's I

Lagrange

Robust

Lagrange

Robust

Province

 

 

multiplier

Lagrange multiplier

multiplier

Lagrange multiplier

Central

Statistic

1.341

8.742

8.111

0.725

0.394

 

p-value

0.180

0.003

0.004

0.094

0.760

Coast

Statistic

5.386

11.054

6.669

27.794

23.409

 

p-value

0.000

0.001

0.010

0.000

0.000

Eastern

Statistic

11.767

85.099

7.203

91.052

13.157

 

p-value

0.000

0.000

0.007

0.000

0.000

North Eastern

Statistic

1.639

0.420

0.077

0.653

0.311

 

p-value

0.101

0.517

0.781

0.419

0.577

Nyanza

Statistic

13.061

140.72

0.002

151.302

10.575

 

p-value

0.000

0.000

0.962

0.000

0.001

Rift Valley

Statistic

25.126

540.038

50.392

516.655

27.010

 

p-value

0.000

0.000

0.000

0.000

0.000

Western

Statistic

7.656

38.286

1.755

56.610

20.080

 

p-value

0.000

0.000

0.185

0.000

0.000

 

Table A6: Results of the spatial corrected models: Central Province

Variable name

Coefficient.

Std. Err.

P>z

Demographic

 

 

 

Population density

-0.00005

0.00002

0.04000

Distance and travel time

 

 

 

Distance to forest (km)

0.00279

0.00096

0.00400

Distance to district hospital (km)

0.01413

0.00380

0.00000

Distance to nearest town of 200,000 people (kms)

-0.00029

0.00043

0.50800

Travel time to road all road types (track, tarmac or murram) (minutes)

0.00026

0.00005

0.00000

Land use

 

 

 

Percent of location under bush

0.00348

0.00474

0.46200

Percent of location under wetland

-0.00643

0.00212

0.00200

Natural factors

 

 

 

perc4_8slop

-0.00643

0.00212

0.00200

Mean rain coefficient of variation

-0.00083

0.00078

0.28700

Elevation (km above sea level)

0.00245

0.00060

0.00000

Goodsoil (dummy)

0.00532

0.00150

0.00000

Rangeland (dummy)

0.02006

0.01532

0.19000

l

0.11285

0.01616

0.00000

Intercept

0.00122

0.00774

0.87500

Number of observations

164

 

 

Squared correlation

0.46

 

 

Log likelihood

220.78

 

 

Table A7: Results of the spatial lag models: Coast Province

Variable

Coefficient

Std.Error

Probability

Demographic variable

 

 

 

Popden

0.00004

0.00010

0.69000

Distance and travel time

 

 

 

d_dist_hc

0.00244

0.00216

0.25700

t_trav_road12

0.00639

0.00318

0.04500

d_dist_for2

0.00027

0.00054

0.61500

d_dist_50k2

0.00008

0.00032

0.79200

d_dist_200k2

-0.00009

0.00026

0.73000

Land use

 

 

 

perc_grass

-0.00272

0.00099

0.00600

perc_farmland

0.00037

0.00080

0.64900

perc_water

0.01323

0.00633

0.03700

Natural capital

 

 

 

perc4_8slop

0.00353

0.00094

0.00000

perc8_15slop

-0.00435

0.00125

0.00100

lgp60days

0.00445

0.00225

0.04800

Lgparids~180

-0.00101

0.00054

0.06300

Flood

-0.05425

0.02520

0.03100

_cons

-0.12354

0.23523

0.59900

Rho

0.52899

0.09460

0.00000

Number of observations

167

 

 

Squared correlation

0.689

 

 

Log likelihood

132.27

 

 

Table A8. Results of the spatial-lag model: Eastern Province

Variable

Coefficient

Std.Error

Probability

Demographic

 

 

 

Popden

-0.0001

0.0000

0.0430

Distance and travel time

 

 

d_dist_for2

0.0008

0.0002

0.0000

Land use

 

 

 

perc_grass

-0.0025

0.0007

0.0000

perc_farml~d

0.0005

0.0004

0.1830

perc_wooded

0.0000

0.0004

0.9320

perc_wetla~s

-0.0061

0.0013

0.0000

Natural capital

 

 

 

perc4_8slop

-0.0003

0.0004

0.4650

perc8_15slop

0.0007

0.0006

0.2430

lgp60days

0.0004

0.0006

0.4520

lgparids~180

-0.0005

0.0002

0.0120

Flood

-0.0116

0.0122

0.3390

Meanraincv

0.0017

0.0008

0.0290

Elevkm

-0.0123

0.0025

0.0000

_cons

0.2124

0.0896

0.0180

Rho

0.5881

0.09182

0.0000

Number of observations

416

 

 

Squared correlation

0.446

 

 

Log likelihood

384.467

 

 

Table A9. OLS model: North Eastern Province

Variable name

Coefficient.

Std. Err.

t

Demographic

 

 

Popden

-0.0002

0.0001

-2.3000

Distance and travel time

 

 

d_dist_disthosp

0.0000

0.0000

1.7400

d_dist_10k2

0.0002

0.0001

2.2400

d_dist_50k2

0.0000

0.0000

0.3800

d_dist_200k2

 

 

 

t_trav_road12

0.0000

0.0000

-0.6000

Land use

 

 

 

perc_built

0.0079

0.0052

1.5300

perc wooded

0.0003

0.0002

1.6200

Natural capital

 

 

 

perc0_4slop

-0.0005

0.0005

-0.9600

perc4_8slop

-0.0021

0.0011

-1.9600

Meanraincv

0.0003

0.0002

2.0900

Intercept

0.6899

0.0540

12.7800

Adj R2 =

0.1785

 

 

Number of obs =

202

 

 

The poverty estimates used for North Eastern are derived estimates from the model for Coast Province, because the Household Budget Survey for 1997, which was used to estimate location-level poverty levels for all of the other provinces, was not implemented in this province due to security-related reasons. Because most of the Coast Province has similar characteristics with the North Eastern Province, its first stage model was adopted and applied in North Eastern Province. Details of this procedure can be obtained from the Kenya National Bureau of Statistics.

Table A10. Results of the spatial-lag models: Nyanza Province

Variable

Coefficient

Std.Error

Probability

Demographic

 

 

Popden

-0.0001

0.0000

0.0280

Distance and travel time

 

 

d_dist_for2

0.0006

0.0003

0.0430

d_dist_disthospl

0.0000

0.0000

0.0300

d_dist_200k2

0.0001

0.0000

0.0010

Land use

 

 

 

Rangelandyes

-0.0168

0.0115

0.1460

perc_water

0.0021

0.0009

0.0270

perc_grass

0.0000

0.0006

0.9580

perc_farmland

-0.0002

0.0002

0.2760

perc_wetlands

-0.0011

0.0007

0.1100

Natural factors

 

 

 

perc4_8slop

0.0006

0.0003

0.0540

perc8_15slop

-0.0001

0.0004

0.8690

av_rainfall

0.0000

0.0000

0.3140

Elevation

0.0001

0.0000

0.0030

Goodsoil

-0.0168

0.0110

0.1280

_cons

0.6864

0.0818

0.0000

Rho

0.4938

0.0470

0.0000

Number of observations

305

 

 

Squared correlation

0.6150

 

 

Log likelihood

357.2417

 

 

Table A11. Results of the spatial-lag models: Rift Valley Province

Variable

Coefficient

Std. Error

Probability

Demographic

 

 

 

Popden

0.0000

0.0000

0.3965

Distance and travel time

 

 

 

D_dist_forest

-0.0001

0.0001

0.4393

T_trav_road12

0.0000

0.0000

0.3813

D_dist_disthsop

0.0000

0.0000

0.9310

D_dist_201

0.0000

0.0000

0.5695

land use

 

 

 

perc_water

-0.0014

0.0009

0.1345

perc_built

-0.0263

0.0034

0.0000

perc_grass

-0.0004

0.0003

0.2172

perc_farmland

-0.0003

0.0002

0.1163

perc_wetland

0.0003

0.0006

0.6496

natural factors

 

 

 

perc4_8slope

0.0005

0.0003

0.0758

perc8_15slope

0.0000

0.0002

0.8448

perc15_30slope

-0.0002

0.0003

0.5015

perc30_abo

0.0011

0.0003

0.0001

Flood

0.0184

0.0068

0.0066

lgp60days

-0.0002

0.0002

0.2267

Lgparidsem

-0.0009

0.0001

0.0000

Constant

0.5458

0.0463

0.0000

Rho

0.7621

0.0382

0.0000

Number of observations

785

 

 

Squared correlation

0.6016

 

 

Log likelihood

995.284

 

 

Table A12. Results of the Spatial lag Models: Western Province

Variable

Coefficient

Std. Error

Probability

Demographic

 

 

Popden

-0.0000

0.0000

0.773

Distance and travel time

 

 

t_trav_road12

-0.0001

0.0000

0.1080

d_dist_10k2

0.0014

0.0008

0.0960

d_dist_for2

0.0010

0.0004

0.0110

Land use

 

 

 

perc_grassland

-0.0017

0.0005

0.0000

Rangelandyes

0.0200

0.0187

0.2850

perc_farmland

0.0002

0.0003

0.4810

perc_protected area

0.0055

0.0020

0.0070

Natural factors

 

 

 

Elevation

0.0081

0.0054

0.1330

perc4_8slope

0.0070

0.0044

0.1110

perc8_15slope

-0.0002

0.0003

0.4340

perc15_30slope

0.0001

0.0004

0.8420

Goodsoil

-0.0133

0.0112

0.0440

_cons

0.1984

0.0671

0.0030

Rho

0.5860

0.0677

0.0000

Number of observations

193

 

 

Squared correlation

0.6190

 

 

Log likelihood

302.182

 

 

Simulations.

Using the estimated parameters of the model(s), we generate predictions of new poverty rates for every Location when the level of a particular independent variable xj is changed. Of course, not all of our independent variables are amenable to policy changes (e.g., rainfall or slope) thus we target those that can be influenced by investments, such as roads and soils. The changes in explanatory variables result in changes in the predicted probabilities, and these are taken to be the effect of the policy. We do not consider higher order changes in this study. Because the results of the simulations assume that the considered changes in the determinant variables do not affect the model parameters or other exogenous variables, these results need to be interpreted as indicative only. While this is a plausible assumption for incremental changes, it warrants a more cautious interpretation for simulations that involve "large" policy changes.

We simulate interventions aimed at reducing the proportion of poor people in a Location. When interpreting the simulation results, it is important to note that changes in poverty for each simulation will depend essentially on: (i) The magnitude and sign of the coefficients from the regression; (ii) The proportion of the population affected by the simulation; (iii) The size of the change considered in the determinants variable.

It is also important that we consider the resultant effects of the simulations as instantaneous because we estimate them from static models. In reality, the effects on community poverty realized from a change in an agricultural variable (say fertilizers for soil improvement) will only be observed in the next growing season, and the benefits from road construction will only be realized when the road is complete and market forces informed, perhaps 2 years later.

Table A13: Impact of changes in soils and travel time: an illustrative simulation

 

Travel time simulations

 

 

 

Central

Obs

Poverty rate

Overall effect

Base poverty rate before road improvement

164

31.3

 

Poverty rate after road improvement

164

30.5

Reduces poverty

Eastern

Base poverty rate before road improvement

416

57.7

 

Poverty rate after road improvement

416

56.9

Reduces poverty

Western

Base poverty rate before road improvement

193

59.2

 

Poverty rate after road improvement

193

58.9

Reduces poverty

 

 

 

 

Soil improvement

Western Province

 

 

 

Base poverty rate before soil improvement

193

59.2

 

Poverty rate after soil improvement

193

49.8

Reduces poverty

Livestock systems

Base poverty rate before soil improvement

1159

55.9

 

Poverty rate after soil improvement

1159

50.4

Reduces poverty

Poverty rate after road improvement

1159

48.3

Reduces poverty