Download presentation
Presentation is loading. Please wait.
Published byΠιλάτος Δαγκλής Modified over 6 years ago
1
Using an asset index to assess trends in poverty in seven Sub-Saharan African countries
Frikkie Booysen, Servaas van der Berg, Ronelle Burger, Gideon du Rand & Michael von Maltitz Paper presented at IPC conference on The Many Dimensions of Poverty, August 2005, Brasilia, Brazil
2
Outline Background Data Method Findings Conclusions
How presentation is structured
3
Background Income-based cross-country poverty comparisons difficult due to price conversions / fluctuations Comparisons within countries across time often not possible due to insufficient or incomparable surveys Data reliability an issue for many African countries’ official statistics Worse for income/expenditure data because complexity of surveying Data = major obstacle for this type of analysis
4
Background Sahn and Stifel (2000) propose used of Demographic and Health Surveys (DHS) as solution to this problem Standardization of surveys ensures comparability across time and space Possession of assets, access to public services and characteristics of infrastructure easier to survey than income/expenditure Data = major obstacle for this type of analysis
5
Data Criteria for selection: three surveys available from late 1980s to early 2000s DHS conducted in different years for different countries, thus survey years are not matched To enable comparability over time: First wave/baseline: Second wave: Third wave:
6
Data Seven African countries in our sample: Ghana Kenya Mali Senegal
Tanzania Zambia Zimbabwe
7
Data Variables included in asset index
TV ownership Fridge ownership Radio ownership Bicycle ownership Type of toilet facility Type of floor material Source of drinking water Apart from a few peculiarities in access to slow-moving assets, data appears reliable… BUT there is an inherent urban bias?
8
Method Multiple correspondence analysis used for constructing an asset index More appropriate than PCA/factor analysis often used in literature Aim is to find a number of smaller dimensions to capture most of information contained in original space Each of these dimensions are the weighted sum of the original variables
9
Method Owns a radio 0.294 Does not own a radio -0.234 Owns a TV 1.568 Does not own a TV -0.103 Owns a fridge 1.630 Does not own a fridge -0.099 Owns a bicycle 0.022 Does not own a bicycle -0.006 Flush Toilet 1.147 Pit latrine -0.087 No toilet -0.308 Earth floor -0.270 Cement floor 0.359 Smart floor 1.830 Piped water 0.877 Public water -0.037 Surface water -0.223 Well water -0.229 MCA weights were allocated based on pooling of countries for the baseline (first) period, using mca command in Stata 8.2 Explain 94% of inertia Logical distribution of weights across response categories, excl. “other categories”
10
Method MCAPi = Ri1W1 + Ri2W2 + … + RijWj + … + RiJWJ
, where MCAPi is the ith household’s composite poverty indicator score, Rij is the response of household i to category j, and Wj is the MCA weight applied to category j Negative index values transformed into positive, non-zero values by adding to the index
11
Method
12
Method Given the arbitrary transformation required to make all index values non-negative and the arbitrary poverty line, it was not deemed appropriate to calculate P1 and P2 Poverty analysis confined to the poverty headcount ratio (P0) and the investigation of stochastic poverty dominance, using cumulative density curves or functions
13
Method Employed three poverty lines… 40th percentile of asset index
Absolute poverty line: weighted sum of categories that is deemed as representing an adequate standard of living: radio bicycle cement floor public water pit latrine no refrigerator no TV
14
Number of unique values per quintile
Findings Number of unique values per quintile Quintile 1 6 Quintile 2 18 Quintile 3 78 Quintile 4 128 Quintile 5 463 Total 693
15
Household consumption always or continuously in deficit
Findings Asset index rankings compared to household consumption rankings (Uganda 1995) Household consumption always or continuously in deficit 13-item asset index (40th percentile poverty line) Poor Non-poor 1,005 1,140 1,334 3,998
16
Household head has no education or primary education only
Findings Asset index rankings compared to rankings based on education of household head (Uganda 1995) Household head has no education or primary education only 13-item asset index (40th percentile poverty line) Poor Non-poor 2,007 117 3,604 1,574
17
Poverty headcount across countries
Findings Poverty headcount across countries Country Mean asset index Poverty headcount Asset index rank WDI $2 WDI rank Ghana 0.267 71.7 5 75.2 4 Kenya 0.187 76.2 3 62.3 7 Mali 0.147 85.3 2 90.6 1 Senegal 0.319 60.9 6 63.1 Tanzania 0.108 89.3 72.5 Zambia 0.217 73.2 90.1 Zimbabwe 0.308 60.8 83.0 To what extent is it possible, with the limited (dummy) variables available to us that are not already incorporated into the asset index, to statistically “explain” differences in asset welfare? Table shows a number of OLS regressions of MCA baseline-weighted assets index, regressed on location (urban-rural), country, and time (period), Particularly noticeable in these regressions are: The importance of urban location for asset wealth In Equation 1, a full 35% of the variation in asset wealth can be explained by spatial factor alone The poor performance of Tanzania, the reference country; compared to Tanzania, all other countries show positive and statistically highly significant coefficients, indicating better performance than Tanzania, in all the equations in which the country dummies enter The good performance of Zimbabwe and Ghana By far the largest part of what can be explained by these variables (around 40%) can be ascribed to the explanatory role of location. Variation between countries is a much smaller factor than variation between urban and rural areas Over time, there is improvement in the wealth index
18
Poverty headcount over time by country
Findings Poverty headcount over time by country Country Period 1 Period 2 Period 3 Asset index trend WDI trend Ghana 83.2 72.5 64.6 - Kenya 79.9 78.8 71.4 + Mali 95.6 88.8 80.9 Senegal 75.8 59.5 57.3 Tanzania 88.4 88.9 92.1 Zambia 69.6 74.3 75.2 Zimbabwe 63.5 63.7 57.0 To what extent is it possible, with the limited (dummy) variables available to us that are not already incorporated into the asset index, to statistically “explain” differences in asset welfare? Table shows a number of OLS regressions of MCA baseline-weighted assets index, regressed on location (urban-rural), country, and time (period), Particularly noticeable in these regressions are: The importance of urban location for asset wealth In Equation 1, a full 35% of the variation in asset wealth can be explained by spatial factor alone The poor performance of Tanzania, the reference country; compared to Tanzania, all other countries show positive and statistically highly significant coefficients, indicating better performance than Tanzania, in all the equations in which the country dummies enter The good performance of Zimbabwe and Ghana By far the largest part of what can be explained by these variables (around 40%) can be ascribed to the explanatory role of location. Variation between countries is a much smaller factor than variation between urban and rural areas Over time, there is improvement in the wealth index
19
Findings Discrimnatory power weak at lower levels due to high number of non-unique values
20
Findings Discrimnatory power weak at lower levels due to high number of non-unique values
21
Approach “In places the density curves are almost indistinguishable. In most cases therefore it is not possible to reach strong conclusions on trends and disparities in poverty, giving rise to uncertainty as to whether there has been progress in terms of the alleviation of poverty.”
22
Findings Poverty of what?
Lorenz dominance more robust than Theil and Gini,but not practical in most cases – lines crossing – thus use aggregated measures like Theil and Gini
23
Findings Now look at rural-urban differences to try and explain observed variation in poverty across countries /time
24
Findings OLS regression of country, time and place of residence on the asset index Equation 1 Equation 2 Equation 3 Equation 4 Urban 0.344** 0.334** Ghana 0.159** 0.122** 0.113** Kenya 0.079** 0.090** 0.081** Mali 0.039** 0.033** 0.018** Senegal 0.211** 0.152** 0.140** Zambia 0.109** 0.061** 0.054** Zimbabwe 0.200** 0.164** 0.154** Period 2 0.014** Period 3 0.044** R-squared 0.36 0.07 0.40 0.41 To what extent is it possible, with the limited (dummy) variables available to us that are not already incorporated into the asset index, to statistically “explain” differences in asset welfare? Table shows a number of OLS regressions of MCA baseline-weighted assets index, regressed on location (urban-rural), country, and time (period), Particularly noticeable in these regressions are: The importance of urban location for asset wealth In Equation 1, a full 35% of the variation in asset wealth can be explained by spatial factor alone The poor performance of Tanzania, the reference country; compared to Tanzania, all other countries show positive and statistically highly significant coefficients, indicating better performance than Tanzania, in all the equations in which the country dummies enter The good performance of Zimbabwe and Ghana By far the largest part of what can be explained by these variables (around 40%) can be ascribed to the explanatory role of location. Variation between countries is a much smaller factor than variation between urban and rural areas Over time, there is improvement in the wealth index
25
Conclusions Evidence that overall poverty declined in Ghana, Kenya, Mali, Senegal and Zimbabwe, but increased in Zambia over this period Evidence that urban poverty declined in Ghana, Kenya, Mali, Tanzania and Zimbabwe, but increased in Senegal Zambia over this period
26
Conclusions , BUT caution required in interpreting results, given caveats of asset index approach… Not a complete measure of welfare Sensitivity of results to choice of poverty line Urban bias of the asset index means that analysis of trends in rural poverty remains problematic Aggregation conceals divergent shifts in underlying variables and complicates policy recommendations, e.g. increased access to private assets versus decline in access to public assets Slow-moving nature of component variables: asset index not a good measure for assessing changes in welfare over short- to medium-term?
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.