OPHI Oxford Poverty & Human Development Initiative Department of International Development Queen Elizabeth House, University of Oxford www.ophi.org.uk
Multidimensional Child Poverty Measurement
UNICEF, September 2009
OPHI‟s research at a glance • Missing Dimensions • Multidimensional Poverty Measurement Future topics in economic theory & methods: • Sustainability Responsibility & Equity: incorporating plural principles in economics • Behavioural Economics: incorporating non-self interested motives, and informed preferences. • Multidimensional modelling: informing policy with better models, and better „participation.‟
Multidimensional Poverty Comparisons • How do we create an Index? – – – – – – –
Choice of Unit of Analysis (child or family) Choice of Dimensions (done) Choice of Variables/Indicator(s) for dimensions Choice of Weights across each dimension (indicators) Choice of Poverty Lines for each dimension (indicators) Identification method Aggregation method – within and across dimensions.
Why multidimensional not income poverty measures for children?
Not income poor Income Poor
Non-deprived in non-monetary dimension
Deprived in nonmonetary dimension
Group A
Group B (I)
Group C (II)
Group D Ruggieri-Laderchi 2007
If income/consumption poverty is used for policy & targetting purposes, Group B represents a targeting error I (omission of some poor) Group C represents a targeting error II (inclusion of some non-poor)
Why multidimensional not income poverty measures for children? Country
Education
Nutrition
Percentage of non-income poor children who ARE deprived in education or nutrition
India
43%
53%
Peru
32%
21%
Percentage of income poor children who are NOT deprived in education or nutrition
India
65%
53%
Peru
93%
66%
Source: Franco et al. (2002) cited in Ruggieri-Laderchi, Saith and Stewart.
Young Lives: Dimensions chosen • • • •
1) Nutritional status 2) Physical morbidity 3) Mental morbidity 4) Life skills (literacy, numeracy, work skills etc.) • 5) Developmental stage for age • 6) Perceptions of well-being and life chances
Composite measures on child poverty – shape dimensions • World summits and conventions on poverty and rights of the child: i.e. CRC 1989, WSC 1990, WFFC 2002 Data requirement for monitoring progress • More data available and a greater degree of consensus on multidimensionality e.g. UNICEF - MICS • Looking for a composite measure of child poverty One that can inform policy makers if the overall situation is improving or worsening, and allow to identify where it is changing (i.e. in which dimension, in which particular subgroup of population) • Range of different studies e.g. CDI-Save the Children , Bristol approach, SAIMDC
Where Poverty Measures may fit in
Intuitive Overview – no maths! Step 1: choose the unit of analysis
Relevant Options: - Child (in which case some data are on the individual such as nutritional status; and hh data – for example on drinking water, or sanitation – are applied to each member of the household) (also community data) - Household (in which case individual data from hh members need to be aggregated)
Intuitive Overview – no maths! Step 2: choose domains Step 3: choose indicators for each domain How to choose? Principles: Accuracy: as many indicators as necessary (so analysis can guide policy well) Parsimony: as few indicators as possible (Ease of analysis for policy purposes, transparency) In Practice: Correlation analysis (Factor analysis, PCA, and similar techniques can be used with care) Feedback from enumerators on quality of data Calculations with/without variables, sensitivity etc Explicit reasoned judgements of the team.
Intuitive Overview – no maths! Step 4: set a poverty cutoff for each indicator* •
Schooling: “How many years of schooling have you completed?” – –
•
Drinking Water: “What is the main water source for drinking for this household?” – – – – – – – – –
•
•
6 or more (bold is non-poor) 1-5 years (non-bold is poor) 9. Piped Water 8. Well/Pump (electric, hand) 7. Well Water 6. Spring Water 5. Rain Water 4. River/Creek Water 3. Pond/Fishpond 2. Water Collection Basin 1. Other
Sanitation: “Where do the majority of householders go to the toilet?” – – – – – –
11. Own toilet with septic tank 10. Own toilet without septic tank 9. Shared toilet 8. Public toilet 7. Creek/river/ditch (without toilet) 6. Yard/field (without toilet)
– – – – –
5. Sewer 4. Pond/fishpond 3. Animal stable 2. Sea/lake 1. Other
* different if using PCA/Factor analysis etc
Income: (the national or a nutrition-based poverty line is often used)
note: can test cutoff for robustness
Intuitive Overview Step 5: identify deprivations for each person (in this example) with respect to each indicator/dimen.: Dimen Nutrition sions
Physical Mental Life Skills Developmental Perception morbidity morbidity Stage of WB & life chances
Child 1 (or hh)
ND
D
ND
D
D
D
Child 2
ND
ND
D
ND
D
ND
Child 3
D
D
D
ND
ND
ND
Child 4
D
D
D
D
D
D
Note: you may add a twist • You could identify the child as deprived or nondeprived in a given dimension – On the basis of cardinal data plus a cutoff (one or more indicators – need internal aggregation also) – On the basis of ordinal/dichotomous data plus a cutoff – On the basis of clearly defined assessment criteria that draw on various kinds of qualitative data.
• The only important thing for the measure is that each child can be clearly and reliably identified as deprived or not with respect to each dimension.
Intuitive Overview Step 6 (assumption: equal weights) – Set a second cutoff: how many deprivations must a child or household have to be considered poor? • Must they be deprived in at least one area? union • Must they be deprived in all areas? intersection • Or maybe something in the middle Ex: UNICEF, Child Poverty Report, 2003 -Two or more deprivations for absolute pov Ex: Mack and Lansley, Poor Britain, 1985
-Three or more out of 26 • Note: you can test these for robustness or endogenize k to focus on poorest x% of kids
Intuitive Overview Step 7 (equal weights) – count the number of deprivations for each child/hh Dime Nutritio nsions n
Physical morbidit y
Mental morbidit y
Life Skills
Development al Stage
Perception Count of WB & life chances
Child 1 (or hh)
ND
D
ND
D
D
D
4
Child 2
ND
ND
D
ND
D
ND
2
Child 3
D
D
D
ND
ND
ND
3
Child 4
D
D
D
D
D
D
6
Identification – The problem empirically k= Union 1 2 3 4 5 6 7 8 9 Inters. 10
H 91.2% 75.5% 54.4% 33.3% 16.5% 6.3% 1.5% 0.2% 0.0% 0.0%
Poverty in India for 10 dimensions: 91% of population would be targeted using union, 0% using intersection Need something in the middle. (Alkire and Seth 2009)
Intuitive Overview Step 8: identify who is poor: Example: if k = 4, children 2 and 3 are non-poor Dime Nutritio Physica Mental Life nsion n l morbidi Skills s morbidi ty ty
Developmen Perceptio tal Stage n of WB & life chances
Count
Child 1 (or hh)
ND
D
ND
D
D
D
4
Child 2
ND
ND
D
ND
D
ND
2
Child 3
D
D
D
ND
ND
ND
3
Child 4
D
D
D
D
D
D
6
Intuitive Overview Step 9: focus on the poor children from now on. Dim Nutriti ensio on ns
Physica Mental l morbid morbid ity ity
Life Skills
Developme ntal Stage
Percepti on of WB & life chances
Count
Child 1 (or hh)
D
D
D
D
4
ND
ND
Child 2
0
Child 3
0
Child 4
D
D
D
D
D
D
6
Intuitive Overview Step 10: First, calculate the headcount H = the number of poor children divided by the number of children = 2 poor/4 children = 50%. Dim Nutriti ensio on ns
Physica Mental l morbid morbid ity ity
Life Skills
Developme ntal Stage
Percepti on of WB & life chances
Count
Child 1 (or hh)
D
D
D
D
4
ND
ND
Child 2
0
Child 3
0
Child 4
D
D
D
D
D
D
6
Intuitive Overview Step 11: Next, calculate the „average poverty gap‟ A. A is the average number of deprivations a poor child has. Sum the proportion of total deprivations each child suffers (4/6+6/6) and divide by the total number of poor children (2) so A = (4/6+6/6)/2 = 5/6. Dim Nutriti ensio on ns
Physica Mental l morbid morbid ity ity
Life Skills
Developme ntal Stage
Percepti on of WB & life chances
Count
Child 1 (or hh)
D
D
D
D
4
ND
ND
Child 2
0
Child 3
0
Child 4
D
D
D
D
D
D
6
Intuitive Overview Step 12: What is multidimensional poverty? If the data are ordinal, multidimensional poverty is H times A H = 2/4; A = 5/6 HA= 2/4 * 5/6 = 5/12 = 0.42. This is the adjusted headcount poverty measure, which we call M0 Now let us consider some properties of M0
Properties of M0: Dimensional Monotonicity • Dimensional Monotonicity: if a Di m person becomes deprived in a dimension in which they were earlier not deprived, M0 Chil increases. d1 • HA = 2/4[(5/6 + 6/6)/2] (or hh) = 11/24 = 0.46 Chil d2 • (previously it was 0.42) Chil • M0 has risen because d3 deprivations have Chil increased. d4 • Note: Headcount would be unchanged!
Nutrit ion
Physi cal morbi dity
Ment al morbi dity
Life Skills
Developm ental Stage
Percept ion of WB & life chance s
Cou nt
ND
D
ND
D
D
D
4 5 0 0
D
D
D
D
D
D
6
Properties of M0: Decomposability • Subgroup Decomposability: the national measure can be ‘decomposed’ by age, gender, region, ethnicity, rural/urban etc. • Dimension decomposability (after identification): You can easily see what dimensions are causing greater poverty in different groups or areas.
Extension 1: to reflect depth of poverty Possible if some dimensions have cardinal data. Follow steps 1-11 above. Then, in addition to calculating H and A, calculate a third variable: G = the poverty gap or S = the severity (the sum of the squared poverty gaps). How? For each cardinal domain in which a child is deprived, subtract their score y from the poverty cutoff z, and divide by z. (z-y)/z This is the normalized poverty gap. Sum the normalized poverty gaps for all children in all domains to obtain G. To obtain S, first square each normalized poverty gap (gap1 times gap1). S is the sum of squared normalized poverty gaps.
Calculate two more measures M1 = HAG or M2 = HAS These relate to P1 and P2 FGT measures. Intuition: M1 reflects the *depth* of deprivation in each dimension. M2 emphasises the poverty of the poorest children.
Extension 2: if some dimensions have a greater importance than others, you may easily apply weights. You must modify for weights at two places: 1) Identification (k is now a cutoff of the weighted sum of dimensions) 2) Aggregation (simply weight prior to summation) Both weights are easily applied, as described in Alkire and Foster 2008.
Critique of Bristol Approach The headcount ratio is not sensitive to the breadth of deprivation –acknowledged by Delamonica and Minujin (2007) • Alkire and Foster (2007): headcount ratio adjusted by breadth of deprivation, easy to compute and decomposable
Not clear how sensitive analysis is to the decision of threshold across dimension - for absolute poverty, and severe deprivation • Alkire and Foster (2007): specification of the dual cutoff, robustness tests, alternative calculations – v transparent.
Comparison with Bristol Approach Bristol approach uses i. list of eight dimensions: human rights based approach, e.g. food, safe drinking water, sanitation facilities, health, shelter, education, information, access to services ii. threshold within dimension to identify deprivations iii. identification of poverty according to a threshold across dimensions -In absolute poverty: at least two deprivations -In severe deprivation: one or more deprivations AF measure: flexible; normative
iv. computation of the headcount ratio H number only AF measure: also reflects breath, is decomposable, can show depth.
India: a new Deprivation Index? • ‘In the new system, poverty would be measured with reference to basic facilities like quality education, good health sectors, and clean drinking water…’ The new index, he said, will help the government formulate a ‘multidimensional approach’ to check deprivation. 18 Aug 08 Hindustan Times: Montek Singh Ahluwalia, Deputy Chair of the Planning Commission
Proposed Data Content (v tentative) Dimensions
Poverty line Cut-off (Deprived if)
1. Living Standard
Live in Kaccha (mud) House
2. Health
The minimum BMI of one woman in the household is less than 18.5 Kg/m2
3. Water & Sanitation
If the house has electricity
Uses Pit latrine – without slab, No facility/uses bush/field, Composting toilet, Dry toilet, other The sources of water are unprotected well and spring, river, dam, lake, ponds, stream, tanker truck, cart with small tank, bottled water, other
4. Air Quality
The sources of fuel are coal, lignite, charcoal, wood, straw/shrubs/grass, agricultural crop, animal dung, other
5. Assets
Has none of the following assets: refrigerator, motorcycle, car, phone, mattress, table, color TV, computer, thrasher, tractor. May have electric fan, B/W television, pressure cooker, radio
6. Education
Maximum year of education completed by any member is less than 5 years
7. Livelihood
The respondent and her partner both fall into the following occupation categories: unemployed, agricultural labourer, plantation labourers, simply labourers, and new workers seeking jobs
There is at least one incidence of child labor with the age group of 5-14
8. Child Status
The children in the age-group 5-18 do not go to school because they are required for household work, work on farm, outside work, care of sibling, costs too much, or ‘other’
9. Empowerment
Some woman in the household does not have right to go alone in the market, health facility, and somewhere outside of village
Weighting: 1) equal weights (1/12 each) 2) nested weights (1/9 per dimension): House
12/18
Electricity
12/18
Health
12/9
Sanitation
12/18
Water
12/18
Fuel
12/9
Asset
12/9
Education
12/9
Livelihood
12/9
Child Work Status
12/18
Child School Status
12/18
Empowerment
12/9
Dimensional Head Count Ratio N = 43,178 Dimensions
House
# of Deprived Household
Deprivation Head Count Ratio
7,994
0.19
Electricity
18,738
0.43
Health
18,186
0.42
Sanitation
32,996
0.76
Water
6,796
0.16
Fuel
13,261
0.31
Asset
14,075
0.33
Education
12,827
0.30
Livelihood
13,166
0.30
Child Status
8,689
0.20
23,099
0.53
Empowerment
Multidimensional Poverty Measures Poverty Cut-off (k)
Number of Persons Head Deprived Count (H)
M0
A = M0/H
1
39,437
0.913
0.349
0.382
2
32,370
0.750
0.326
0.435
3
28,979
0.671
0.309
0.460
4
20,220
0.468
0.247
0.528
5
12,195
0.282
0.170
0.602
6
8,863
0.205
0.131
0.640
7
3,644
0.084
0.061
0.727
8
1,044
0.024
0.020
0.815
9
497
0.012
0.010
0.857
10
48
0.001
0.001
0.952
11
7
0.000
0.000
1.000
Indian States
MD Poverty Ranking (NFHS 2005/06) Vs. Income Poverty Ranking (NSS 2004)
Jharkhand Madhya Pradesh Uttar Pradesh Orissa Rajasthan Chhattisgarh Bihar West Bengal Assam Arunachal Pradesh Andhra Pradesh Maharashtra Karnataka Gujarat Haryana Tamil Nadu Meghalaya Uttaranchal Jammu and Kashmir Tripura Nagaland Punjab Goa Manipur Himachal Pradesh Mizoram Sikkim Kerala
NSS Income Poverty MD Head Count
Deteriorated from Rank 1 to Rank 18
0.00
0.10
Improved from Rank 7 to 0.20 0.30 0.40 0.50 0.60 0.70 Rank 1 Poverty Rates
0.80
0.90
Poverty Decomposition by Dimensions Break Down
Sikkim
Break Down
India
Break Down
States
Kerala
House
0.013
2.57%
0.032
6.01% 0.149
3.35%
Electricity
0.038
7.21%
0.034
6.49% 0.311
6.99%
Health
0.037
14.16%
0.018
6.73% 0.284
12.79%
Sanitation
0.024
4.50%
0.051
9.62% 0.442
9.94%
Water
0.036
6.95%
0.036
6.73% 0.096
2.15%
Fuel
0.023
8.75%
0.042
15.87% 0.242
10.90%
Asset
0.046
17.76%
0.020
7.69% 0.274
12.32%
Education
0.007
2.57%
0.054
20.19% 0.241
10.84%
Livelihood
0.034
13.13%
0.015
5.77% 0.232
10.42%
Child Status
0.019
7.21%
0.022
8.17% 0.148
6.64%
Empowerment
0.040
15.19%
0.018
6.73% 0.304
13.65%
M0
0.029
100.00%
0.029
100.00% 0.247
100.00%
Bhutan: We decompose the measure to see what is driving poverty. In Bhutan the rank of the districts changed. The relatively wealthy state Gasa fell 11 places when ranked by multidimensional poverty rather than income; the state Lhuntse, which was ranked 17/20 by income, rose 9 places. Decomposing M0 by dimension, we see that in Gasa, poverty is driven by a lack of electricity, drinking water and overcrowding; income is hardly visible as a cause of poverty. In Lhuntse, income is a much larger contributor to poverty.
% Contribution of Each Indicator
Composition of Multidimensional Poverty in Two Districts - Mo with k=2 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
Gasa Income Drinking Water
District
Literacy Electricity
Lhuntse People per Room Santitation
Robustness With Respect to Cut-off (k) Spearman’s Rank Correlation Matrix for Different M0 Rankings Cut-off (k)
3
4
0.997
5
0.994
0.995
6
0.993
0.993
0.998
7 0.976 0.975 *Minimum correlation is 0.962 8 0.963 0.962
0.985
0.987
0.972
0.973
4
5
6
7
0.980
Rural Vs. Urban MD Poverty Rate 0.80
Multidimensional Headcount
0.70 0.60 0.50 0.40
Rural Urban
0.30 0.20 0.10 0.00
Scheduled Caste Scheduled Tribe Other Backward None of the Class above
India Various Castes
0.70
Comparison Between Various African Countries using DHS Data
M0 Poverty Rate
0.60 0.50 0.40 0.30 0.20 0.10 0.00
0.1
0.4
0.6
0.9
Benin
1.1
1.4
Burkina
1.6
1.9
Ghana
2.1
2.4
Kenya
2.6
2.9
Niger
3.1
3.4
3.6
3.9
Poverty Cut-offs Nigeria
Source: Batana Y. M. (2008) ‘Multidimensional Measurement of Poverty in Sub-Saharan Africa’, Working Paper No.13, Oxford Poverty & Human Development Initiative, Oxford University.
80 60 0
Main contrib: Sanitation, Shelter, Education of HH.
40
Sanitation, Education HH. Income increased its contribution Chile & Mexico:
20
Argentina & Uruguay:
100
Brazil: Main contrib: Education HH, Sanitation, Income.
1992 2000 2006 1995 2003 1992 2000 2006 1995 2003 1992 2000 2006 1995 2003 1995 2003 1992 2000 2006 1995 2003 1992 2000 2006 1995 2003 1992 2000 2006
Argentina
Brazil
Chile
Income Education HH Sanitation
El Salvador Mexico
Uruguay
Children in School Running Water Shelter
El Salvador: More similar contrib. across dimensions
Example of messages
Headcount = 37% are deprived in at least five of 10 dimensions.
M0 = 0.22 Breadth: most poor people are on average deprived in 6/10. The fullest bowls nationally were Education and Housing The emptiest bowls nationally were Employment and Health In the state with highest poverty, Ardenia, Headcount: 82% are poor Breadth: most poor people are on average deprived in 8/10
Strength:
Transparency
Why are we not poor? How do you decide who is poor? (example). Any household that suffers two or more deprivations is poor.
Income
III
Deprivation threshold
Example: Mexico 1. Income Poverty 2. Deprivation Index (6 d) 3 MD poverty = I, below
IV
n ivcó p e ld ra b m U
I
Income threshold
II
Deprivation index
Measure can decompose the change in deprivations across time – key M&E. (Example: China panel data. Employment
Cont r i but i on t o Mo
deprivation rising; income and resource deprivations shrinking) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 1993
Resour ces Secur i t y Empl oyment Heal t h Educat i on I ncome 1997
2000 Year
2004
Ongoing Studies Applications have been completed for: Sub-Saharan Africa (14 countries) Latin America (6 countries) China (2), India, (2) Pakistan, (2) Bhutan (2)
Ongoing Studies Other Applications to: Quality of Education Child Poverty Governance Fair trade Social Protection Gender
(Mexico, Argentina) (Bangladesh, Afghanistan, Italy) (Index of African Governance) (Human Rights – Benetech) (India, Mexico) (international index)
Workshop was 1-2 June, 2010 HDR
Preliminary Feedback: benefits a)
you can target the poor more accurately. By looking at the breadth and depth of deprivation in each dimension, we can zoom in, like a magnifying glass, on the extreme poor. b) you can see policy cues. This multidimensional measure, displays how the components of poverty vary. The same data tives you more relevant information. c) you can look at people not just households. Children‟s distinct needs can be seen directly, for example. d) you can make a measure that matches your needs. The dimensions, poverty cutoffs, etc can be standardized to ensure comparability. But in many cases it can be useful to tailor these to specific contexts and measurement needs.
OPHI’s whole team thanks you
