summer school on multidimensional poverty analysis · poverty analysis 1 –13 august 2016 ......
TRANSCRIPT
Summer School on Multidimensional
Poverty Analysis
1–13 August 2016
Beijing, China
Changes over Time
Ana Vaz
OPHI
Outline
Descriptive Analysis using Repeated Cross-Sectional Data
Basic Concepts
Example: MPI Reduction in India
Analysis of Dynamic Subgroups using Panel Data
Descriptive Analysis using Repeated
Cross-Section Data
• Notation:
- 𝑡1 and 𝑡2 denote initial and final periods
- 𝑋𝑡1 and 𝑋𝑡2 are the achievement matrices for both periods
• The same set of parameters is used across the two periods
(deprivation cutoffs, weights, poverty cutoff)
• Expressions are equally applicable to:
- incidence (H),
- intensity (A),
- censored headcount ratios (ℎ𝑗 𝑘 ), and
- uncensored headcount ratios (ℎ𝑗).
Notation
• Absolute Rate of Change: is the difference in levels
between two periods.
• Relative Rate of Change: is the difference in levels across
two periods as a percentage of the initial period.
• Why use both absolute and relative?
Changes in M0, H and A
∆𝑀0 = 𝑀0 𝑋𝑡2 −𝑀0 𝑋𝑡1
𝛿𝑀0 =𝑀0 𝑋𝑡2 −𝑀0 𝑋𝑡1
𝑀0 𝑋𝑡1× 100
• Annualized Absolute Rate of Change: is the difference in
levels across two periods divided by the difference in the two
time periods.
• Relative Rate of Change: is the compound rate of reduction
per year between the initial and the final periods.
Annualized Changes
∆ 𝑀0 =𝑀0 𝑋𝑡2 −𝑀0 𝑋𝑡1
𝑡2 − 𝑡1
𝛿 𝑀0 =𝑀0 𝑋𝑡2
𝑀0 𝑋𝑡1
1𝑡2−𝑡1
− 1 × 100
Example
Year 1 Year 2
Statistical
Significance of
the Change
Annualized Change
Absolute Relative
Panel I: Multidimensional Poverty Index (MPIT)
Nepal 2006-2011 .350 (.013) .217 (.012) *** -.027 -9.1%
Peru 2005-2008 .085 (.007) .066 (.004) * -.006 -8.0%
Rwanda 2005-2010 .460 (.005) .330 (.006) *** -.026 -6.4%
Senegal 2005-2010/11 .440 (.019) .423 (.010) -.003 -0.7%
Panel II: Multidimensional Headcount Ratio (HT ,%)
Nepal 2006-2011 64.7 (2.0) 44.2 (2.0) *** -4.1 -7.4%
Peru 2005-2008 19.5 (1.5) 15.7 (.8) * -1.3 -6.9%
Rwanda 2005-2010 82.9 (.8) 66.1 (1.0) *** -3.4 -4.4%
Senegal 2005-2010/11 71.3 (2.4) 70.8 (1.5) -0.1 -0.1%
Panel III: Intensity of Poverty (AT ,%)
Nepal 2006-2011 54.0 (.6) 49.0 (.7) *** -1.0 -1.9%
Peru 2005-2008 43.6 (.5) 42.2 (.4) ** -0.5 -1.1%
Rwanda 2005-2010 55.5 (.3) 49.9 (.3) *** -1.1 -2.1%
Senegal 2005-2010/11 61.7 (1.0) 59.7 (.7) * -0.4 -0.6%
Example
Year 1 Year 2
Statistical
Significance of
the Change
Annualized Change
Absolute Relative
Panel I: Multidimensional Poverty Index (MPIT)
Nepal 2006-2011 .350 (.013) .217 (.012) *** -.027 -9.1%
Peru 2005-2008 .085 (.007) .066 (.003) * -.006 -8.0%
Rwanda 2005-2010 .460 (.005) .330 (.006) *** -.026 -6.4%
Senegal 2005-2010/11 .440 (.019) .423 (.010) -.003 -0.7%
Panel II: Multidimensional Headcount Ratio (HT ,%)
Nepal 2006-2011 64.7 (2.0) 44.2 (2.0) *** -4.1 -7.4%
Peru 2005-2008 19.5 (1.5) 15.7 (.8) * -1.3 -6.9%
Rwanda 2005-2010 82.9 (.8) 66.1 (1.0) *** -3.4 -4.4%
Senegal 2005-2010/11 71.3 (2.4) 70.8 (1.5) -0.1 -0.1%
Panel III: Intensity of Poverty (AT ,%)
Nepal 2006-2011 54.0 (.6) 49.0 (.7) *** -1.0 -1.9%
Peru 2005-2008 43.6 (.5) 42.2 (.4) ** -0.5 -1.1%
Rwanda 2005-2010 55.5 (.3) 49.9 (.3) *** -1.1 -2.1%
Senegal 2005-2010/11 61.7 (1.0) 59.7 (.7) * -0.4 -0.6%
Based on this information can we say that the number of
poor people is decreasing over time in Rwanda?
• In order to reduce the absolute number of poor people,
the rate of reduction in the headcount ratio needs to be
faster than the population growth.
• So, don’t forget to also check if the number of poor
people is decreasing over time!
Change in Number of Poor
Population MPI Poor
Year 1 Year 2 Annual
Growth
Year 1 Year 2 Absolute
Reduction
(in thousands) (in thousands)
Nepal 2006-2011 25,634 27,156 1.2% 16,585 12,003 -4,582
Peru 2005-2008 27,723 28,626 0.6% 5,406 4,494 -912
Rwanda 2005-2010 9,429 10,837 2.8% 7,817 7,163 -654
Senegal 2005-2010/11 11,271 13,141 3.1% 8,036 9,304 1,268
Interpreting Dimensional Changes
-8.0
-7.0
-6.0
-5.0
-4.0
-3.0
-2.0
-1.0
.0
An
nu
al
ab
solu
te c
han
ge (
p.p
.)
Nepal 2006 - 2011
Raw Headcount (Shaded) Censored Headcount
What indicator had the biggest contribution to
poverty reduction?
• The (annualized) absolute rate of change in 𝑀0 can be
expressed as the weighted average of the (annualized)
absolute rates of change in censored headcount ratios.
• When different indicators have different weights, the
effects of their changes on the change in 𝑀0 reflect these
weights.
Dimensional Changes
∆ 𝑀0 = 𝑤𝑗∆ ℎ𝑗 𝑘
𝑑
𝑗=1
-8.0
-7.0
-6.0
-5.0
-4.0
-3.0
-2.0
-1.0
.0
An
nu
al
ab
solu
te c
han
ge (
p.p
.)
Nepal 2006 - 2011
Raw Headcount (Shaded) Censored Headcount
Dimensional Changes
What indicator had the biggest contribution to
poverty reduction?
• Interpreting the real on-the-ground contribution of each
indicator to the change in 𝑀0 is not so mechanical.
• A reduction in censored headcount of j may reflect two
different situations:
Interpreting Dimensional Changes
- A poor person became non-deprived in indicator j
- A poor person who has been deprived in j became non-
poor due to reduction in other indicators, even though she
is still deprived in j.
-8.0
-7.0
-6.0
-5.0
-4.0
-3.0
-2.0
-1.0
.0
An
nu
al
ab
solu
te c
han
ge (
p.p
.)
Nepal 2006 - 2011
Raw Headcount (Shaded) Censored Headcount
Dimensional Changes
Why the censored headcount reduced more
than the uncensored one?
-8.0
-7.0
-6.0
-5.0
-4.0
-3.0
-2.0
-1.0
.0
An
nu
al
ab
solu
te c
han
ge (
p.p
.)
Nepal 2006 - 2011
Raw Headcount (Shaded) Censored Headcount
Dimensional Changes
It seems some people became non-poor but
remain deprived in fuel & flooring.
• Compare changes in censored and uncensored headcounts
to analyse the relation between the dimensional changes
among the poor and the society-wide changes in
deprivations.
• In repeated cross-sectional data, this comparison will also
be affected by migration and demographic shifts, as well as
changes in the deprivation profiles of the non-poor.
Dimensional Changes
Subgroup Decompositions
Nepal
-0.055
-0.045
-0.035
-0.025
-0.015
-0.005
0.005
-0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65
An
nu
al
Ab
solu
te C
han
ge i
n M
PI T
Multidimensional Poverty Index (MPIT) at initial year
Reduction
in MPIT
Size of bubble is proportional
to the number of poor in first
year of the comparison.
Subgroup Decompositions
Nepal
Eastern Mountain
Central Mountain
Western Mountain
Eastern Hill
Central Hill
Western Hill Mid-Western Hill
Far-Western Hill
Eastern Terai Central Terai
Western Terai
Mid-Western Terai
Far-Western Terai
-0.055
-0.045
-0.035
-0.025
-0.015
-0.005
0.005
-0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65
An
nu
al
Ab
solu
te C
han
ge i
n M
PI T
Multidimensional Poverty Index (MPIT) at initial year
Reduction
in MPIT
Size of bubble is proportional
to the number of poor in first
year of the comparison.
Population Shifts
• The interpretation of changes in regional poverty
estimates can be hugely influenced by populations shifts.
- Different rates of population growth
- Rural-urban migration
- Internal and International Migration
Example:
MPI Reduction in India
Alkire & Seth World Development 2015
22
• Data: National Family Health Surveys (NFHS)
- NFHS-2 (1998-99)
- NFHS-3 (2005-06)
• Indicators are strictly harmonised
Inter-temporal Multidimensional
Poverty in India
23
Uncensored Deprivations (raw)
Significant reduction in all deprivations. Highest reductions in housing,
sanitation, water and electricity deprivations.
24
Change in MPII Nationally for Different
Poverty Cut-offs
Poverty Cutoff (k) 1999 2006 Change
Union (>0) M0 0.366 0.320 -0.046 ***
H 92.9% 88.9% -4.0 ***
A 39.4% 36.0% -3.4 ***
One-fifth (0.2) M0 0.343 0.293 -0.050 ***
H 73.8% 65.5% -8.2 ***
A 46.5% 44.7% -1.7 ***
One-third (0.33) M0 0.300 0.251 -0.050 ***
H 56.8% 48.5% -8.3 ***
A 52.9% 51.7% -1.2 ***
Half (0.5) M0 0.197 0.156 -0.041 ***
H 30.6% 24.4% -6.2 ***
A 64.5% 64.1% -0.4 *
Where and How?
• Where poverty has been reduced?
- Across geographic regions, social groups and household
characteristics
• How poverty has been reduced?
- By reducing incidence or intensity?
- By improving which indicators?
25
Where and How?
• Where poverty has been reduced?
- Across geographic regions, social groups and household
characteristics
• How poverty has been reduced?
- By reducing incidence or intensity?
- By improving which indicators?
26
Absolute Reduction in Poverty across
Large States
27
Green = faster.
Stronger
reductions in
southern states
-0.110 -0.090 -0.070 -0.050 -0.030 -0.010
Andhra Pradesh (***) [0.299]
Kerala (***) [0.136]
Tamil Nadu (***) [0.195]
Karnataka (***) [0.255]
Jammu (***) [0.226]
Gujarat (***) [0.248]
Orissa (***) [0.381]
Maharashtra (***) [0.226]
West Bengal (***) [0.339]
Himachal Pradesh (***) [0.154]
Eastern States (***) [0.315]
Madhya Pradesh (***) [0.368]
Haryana (**) [0.19]
Uttar Pradesh (***) [0.348]
Rajasthan (**) [0.341]
Punjab (***) [0.117]
Bihar (**) [0.442]
Absolute Change (99-06) in MPI-I
Sta
tes
(Sig
nif
ican
ce)
[MP
I-I
in 1
99
9]
Absolute Reduction in Poverty across
Large States
28
We combined Bihar and Jharkhand, Madhya Pradesh and Chhattishgarh,
and Uttar Pradesh and Uttarakhand
Significant
reduction in all
states
Strongest Reductions
Absolute Reduction in Poverty Across
Sub-Groups
29
-0.110 -0.090 -0.070 -0.050 -0.030 -0.010
Urban (***) [0.116]
Rural (***) [0.368]
General (***) [0.229]
OBC (***) [0.301]
SC (***) [0.378]
ST (***) [0.458]
Sikh (***) [0.115]
Christian (***) [0.196]
Hindu (***) [0.306]
Muslim (*) [0.32]
Absolute Change (99-06) in MPI-I
Su
b-G
rou
ps
(Sig
nif
ican
ce)
[MP
I-I
in 1
99
9]
Significant
reduction for all
sub-groups
But slowest for
the poorest
Absolute Reduction in Acute Poverty Across
Household Characteristics
-0.110 -0.090 -0.070 -0.050 -0.030 -0.010
Female () [0.275]
Male (*) [0.302]
12 Years or More (*) [0.055]
11–12 Years (*) [0.114]
6–10 Years (*) [0.188]
1–5 Years (*) [0.31]
No Education (*) [0.448]
1–3 (*) [0.248]
4–5 (*) [0.265]
6–7 (*) [0.321]
10 or More (*) [0.332]
8–9 (*) [0.34]
Absolute Change (99-06) in MPI-I
Sta
tes
(Sig
nif
ican
ce)
[MP
I-I
in 1
999]
30
HH
Siz
e
Head
’s E
du
c.
Head
’s
Gen
der
Slower progress
for female headed
households and
larger households
Where and How?
• Where poverty has been reduced?
- Across geographic regions, social groups and household
characteristics
• How poverty has been reduced?
- By reducing incidence or intensity?
- By improving which indicators?
31
Improvement in Poverty: H &/or A?
32
Andhra Pradesh
Arunachal Pradesh
Assam Bihar
Goa Gujarat
Haryana Himachal Pradesh
Jammu & Kashmir
Karnataka
Kerala
Madhya Pradesh
Maharashtra Manipur
Meghalaya
Mizoram
Nagaland
Orissa
Punjab
Rajasthan
Tamil Nadu
Tripura
Uttar Pradesh
West Bengal
-1.0%
-0.8%
-0.6%
-0.4%
-0.2%
0.0%
0.2%
0.4%
-3.4% -2.9% -2.4% -1.9% -1.4% -0.9% -0.4% 0.1% 0.6%
An
nu
al
Ab
solu
te V
ari
ati
on
in
In
ten
sity
(A
)
Annual Absolute Variation in Headcount Ratio (H)
Reduction in
Intensity of
Poverty (A)
Bad/Good
Bad/Bad Reduction in Incidence of Poverty (H)
Good /Good
Good/ Bad
Performance
consistently strongest in
Kerala, TN, & AP.
Improvement in Poverty: H &/or A?
33
Scheduled Castes
Scheduled Tribes
Other Backward Classes
General
-0.4%
-0.3%
-0.3%
-0.2%
-0.2%
-0.1%
-0.1%
0.0%
0.0%
-2.5% -2.0% -1.5% -1.0% -0.5% 0.0%
An
nu
al A
bso
lute
Vari
ati
on
in %
In
ten
sity
(A
)
Annual Absolute Variation in % Headcount Ratio (H)
Reduction in Intensity of Poverty (A)
Bad/Good
Bad/BadReduction in Incidenceof Poverty (H)
Good /Good
Good/ Bad
Caste
34
Hindu
MuslimChristian
Sikh
-0.3%
-0.3%
-0.2%
-0.2%
-0.1%
-0.1%
0.0%
0.1%
0.1%
0.2%
-2.0% -1.5% -1.0% -0.5% 0.0%
An
nu
al A
bso
lute
Va
ria
tio
n in
% I
nte
nsi
ty (
A)
Annual Absolute Variation in % Headcount Ratio (H)
Reduction in Intensity of Poverty (A)
Bad/Good
Bad/BadReduction in Incidenceof Poverty (H)
Good /Good
Good/ Bad
Religion
Improvement in Poverty: H &/or A?
Where and How?
• Where poverty has been reduced?
- Across geographic regions, social groups and household
characteristics
• How poverty has been reduced?
- By reducing incidence or intensity?
- By improving which indicators?
35
Changes among people who are poor and
deprived in each indicator
36
Where and How?
• Where poverty has been reduced?
- Across geographic regions, social groups and household
characteristics > Population subgroup decomposability
• How poverty has been reduced?
- By reducing incidence or intensity? > H-A breakdown
- By improving which indicators? > Dimensional breakdown
37
Analysis of Dynamic Subgroups
using Panel Data
With Panel data we can identify 4
type of poor people
1. Chronic poor – across time periods
2. Churning (in and out)
3. Falling into poverty
4. Moving out of poverty.
1. How do the four groups differ – either demographically or in
the structure of their poverty?
2. Poverty traps? Are any dimensions in which chonic poor are
always deprived?
3. Does the composition of poverty for chronic poor change?
Does chronic poverty decrease over time?
4. How does poverty evolve across different ages? For different
social groups and household types?
Panel data enables new analyses:
• Did some people exit poverty?
• Did some exit poverty, and others become newly poor?
• Did some go in and out of poverty various times?
• Were the people that exited poverty among the poorest, or
the less poor in the previous period(s)?
How did poverty change?
• Assuming two period panel data with n individuals.
• Consider 4 mutually exclusive groups:
N: 𝑛𝑁 people who are non-poor in both periods
O: 𝑛𝑂 people who are poor in both periods (ongoing)
E_: 𝑛𝐸−
people who are poor in 𝑡1 but exit poverty
E+: 𝑛𝐸+
people who are non-poor in 𝑡1 but enter poverty
Dynamic Subgroups – Panel Data
• Change in 𝑀0 can be decomposed as follows:
∆𝑀0 =
=𝑛𝑂
𝑛𝑀0 𝑋𝑡2
𝑂 −𝑀0 𝑋𝑡1𝑂
−𝑛𝐸−
𝑛𝑀0 𝑋𝑡1
𝐸−
+𝑛𝐸+
𝑛𝑀0 𝑋𝑡1
𝐸+
Change in 𝑴𝟎
• It is impossible to decompose ∆𝑀0 with the empirical
precision as when using panel data.
• Theory-based approaches to decomposing ∆𝑀 between
incidence (∆𝐻) and intensity (∆𝐴)
- Based on assumptions regarding the intensity of those who
exited or remained poor…
• However, these give precise estimates that might be very
innacurate.
Dynamic Subgroups –
Repeated Cross-Sectional Data
References
• Alkire, S. et al. (2015). Multidimensional Poverty Measurement
and Analysis, Oxford: Oxford University Press, ch. 9.
• Alkire, S., Roche, J. M., and Vaz, A. (2015). “Changes Over
Time in Multidimensional Poverty: Methodology and Results
for 34 Countries.” OPHI Working Papers 76, University of
Oxford.
• Alkire, S. and Seth, S. (2015). “Multidimensional Poverty
Reduction in India between 1999 and 2006: Where and How?”,
World Development, 72, 93-108.
Thank you