Transcript
Income Inequality: Measures, Estimates and Policy
Illustrations
Focus of the Discussion:
• Framework: Kuznets’: explain inequality in terms of inter-sectoral disparities & intra-sectoral inequalities
• Final outcome measures:– Income generation:
• Sectoral perspective at the macro as well as disaggregate regional (district) level
– Income distribution • Proxy: consumption distribution - macro (state),
regional and district levels by rural/urban sectors
2
Inequality Measures & Welfare Judgments
• Inequality measures have implicit normative judgments about inequality and the relative importance to be assigned to different parts of the income distribution.
• Some measures are clearly unattractive:– Range: measures the distance between the poorest and richest; is y
unaffected by changes in the distribution of income between these two extremes.
Simpler (statistical) measures
• (normalised) Range
• Relative mean deviation• (Shows percentage of total income that would need to be transferred to make all incomes are the same.)
• Coefficient of variation = standard deviation/mean
• 75-25 gap, 90-10 gap
n
i
i mynm 1
||1
2
1
Inequality measurement: Some attractive axioms• Pigou-Dalton Condition (principle of transfers): a transfer
from a poorer person to a richer person, ceteris paribus, must cause an increase in inequality.– Range does not satisfy this property.
• Scale-neutrality: Inequality should remain invariant with respect to scalar transformation of incomes. – Variance does not satisfy this is property.
• Anonymity: Inequality measure should remain invariant with respect to any permutation.
Gini coeficient• Gini coeficient: The proportion of the total area under the Lorenz curve.
• Discrete version:
• Interpretation: Gini of “X” means that the expected difference in income btw. 2 randomly selected persons is 60% of overall mean income.
• Restrictive:• -- The welfare impact of a transfer of income only depends on “relative rankings” –
e.g., a transfer from the richest to the billionth richest household counts as much as one from the billionth poorest to the poorest.
)...2(11
1 212 nnyyyynn
G
The Atkinson class of inequality measures
• Atkinson (1970) introduces the notion of ‘equally distributed equivalent’ income, YEDE.
• YEDE represents the level of income per head which, if equally shared, would generate the same level of social welfare as the observed distribution.
• A measure of inequality is given by: IA = 1- (YEDE/μ)
The Atkinson class of inequality measures
• A low value of YEDE relative to μ implies that if incomes were equally distributed the same level of social welfare could be achieved with much lower average income.; IA would be large.
• Everything hinges on the degree of inequality aversion in the social welfare function.
• With no aversion, there is no welfare gain from edistribution so YEDE is equal to μ and IA = 0.
The Atkinson class of inequality measures
• Atkinson proposes the following form for his inequality measure:
1
1
1)(1 ii
iA f
Y
YI
Atkinson’s measure
• This is just an iso-elastic social welfare function defined over income (not utility) with parameter e, normalised by average income
e
i
ei
y
yA
1
11
)(1
The Atkinson class of inequality measures
• A key role here is played by the distributional parameter ε. In calculating IA you need to explicitly specify a value for ε.
• When ε=0 there is no social concern about inequality and so IA = 0 (even if the distribution is “objectively” unequal).
• When ε=∞ there is infinite weight to the poorer members of the population (“Rawls”)
Inequality measurement and normative judgements• Coefficient of variation:
– Attaches equal weights to all income levels– No less arbitrary than other judgments.
• Standard deviation of logarithms:– Is more sensitive to transfers in the lower income
brackets.• Bottom line: The degree of inequality cannot
in general be measured without introducing social judgments.
Theil’s Entropy Index
Formally, an index I(Y) is Theil decomposable if:
Ni
i
HH iYIiweNmeimIYIN
1
)(),...,()(1
Theil’s Entropy Index:
Ni
i
ii
y
y
y
y
NNYT
1
)log(log
1)(
Where Yi is a the vector of incomes of the Hi members of subgroup i, there are N subgroups, and mieHi is an Hi long vector of the average income (mi) in subgroup i. The terms wi terms are subgroup weights.
Recommendations• No inequality measure is purely ‘statistical’: each embodies
judgements about inequality at different points on the income scale.
• To explore the robustness of conclusions:• Option 1: measure inequality using a variety of inequality
measures (not just Gini).• Option 2: employ the Atkinson measure with multiple values
of ε.• Option 3: look directly at Lorenz Curves, apply Stochastic
Dominance results.
The Lorenz Curve• To compare inequality in two distributions:
– Plot the % share of total income received by the poorest nth percentile population in the population, in turn for each n and each consumption distribution.
– The greater the area between the Lorenz curve and the hypotenuse the greater is inequality.
• Second Order Stochastic Dominance (Atkinson 1970):– If Lorenz curves for two distributions do not intersect, then they can
be ranked irrespective of which measure of inequality is the focus of attention.
– If the Lorenz curves intersect, different summary measures of inequality can be found that will rank the distributions differently.
Inequality Measures
• Shortcomings of GDP can be addressed in part by considering inequality
• Common measures of inequality– Distribution of Y by Decile or Quintile
Income Distribution by Decile Group: Mexico, 1992
DECILE INCOME SHARE(%)
I 1.3II 2.4III 3.2IV 4.2V 5.1VI 6.4VII 8.3VIII 11.0IX 16.1X 42.1
Inequality Measures
• Shortcomings of GDP can be addressed in part by considering inequality
• Common measures of inequality– Distribution of Y by Decile or Quintile– Gini Coefficient
• most commonly used summary statistic for inequality
Gini Coefficient
Cumulative Income Share
Cumulative Population Share (poorest to riches)
0
100
100
Lorenz Curve
Gini Coefficient
Cumulative Income Share
Cumulative Population Share
0
100
100
Lorenz Curve 1 Lorenz Curve 2
Gini Coefficient
Cumulative Income Share
Cumulative Population Share
0
100
100
Lorenz Curve
A
B
Gini = A / A + B
Gini Coefficient
• Gini varies from 0 - 1• Higher Ginis represent higher inequality• The Gini is only a summary statistic, it doesn’t
tell us what is happening over the whole distribution
Inequality Measures
• Shortcomings of GDP can be addressed in part by considering inequality
• Common measures of inequality– Distribution of Y by Decile or Quintile– Gini Coefficient
• most commonly used summary statistic for inequality
– Functional distribution of income
Inequality: Policy Instrument
• Illustrate How Policy Strategies are made Little Realizing that the Very Framework used does not permit such an Approach
• Illustrate How Wrong Inferences are drawn on Empirical Estimates of Inequality, which finally form the basis for theoretically implausible Strategies for Poverty Reduction
DOES SPECIFICATION MATTER?
• CHOICE OF STRATEGIES
• ESIMATES OF MAGNITUDES
• EVALUATION OF POLICY CONSEQUENCES
• ILLUSTRATED WITH REFERENCE TO THE INDIAN EXPERIENCE ON POLICIES FOR POVERTY REDUCTION, ESTIMATES & EVALUATION
CHOICE OF DEVT STRATEGIES
• GROWTH WITH REDISTRIBUTION
• FORMULATED AND PURSUED INDEPENDENTLY
• BASED ON THE PREMISES OF SEPARABILITY AND INDEPENDENCE
• EXAMPLES: FIFTH & SIXTH FIVE YEAR PLANS
INDIAN SIXTH PLAN STRATEGY
• RURAL INDIA:• BASE YEAR (BY): 1979-80• BY POVERTY 50.70 % • TERMINAL YEAR (TY): 1984-85• REDUE TY POVERTY TO 40.47 % BY GROTH (15.44 %) • FURTHER DOWN TO 30 % BY REDISTRIBUTION (BY
REDUCING INEQUALITY FROM 0.305 TO 0.222)
INDIAN SIXTH PLAN STRATEGY
• URBAN INDIA:• BASE YEAR (BY): 1979-80• BY POVERTY 40.31 % • TERMINAL YEAR (TY): 1984-85• REDUE TY POVERTY TO 33.71 % BY GROTH (11.32 %) • FURTHER DOWN TO 30 % BY REDISTRIBUTION (BY
REDUCING INEQUALITY FROM 0.335 TO 0.305)
Base Year
Terminal Year: 1984-85
HCR (%)
Growth (%)
HCR (%)
Inequality change (%)
HCR (%)
Rural India
50.7 15.4 40.5 -27.4 30
Urban India
40.3 11.3 33.7 -8.8 30
Growth with Redistribution
HOW VALID ARE THE PREMISES?
• THE STRATEGIES ARE NEITHER SEPARABLE NOR INDEPENDENT
• GROWTH WILL REDUCE POVERTY
• AT AN INCREASING RATE IF HCR < 50%
• AT A DECREASING RATE IF HCR > 50%
• MAXIMUM IF HCR = 50%
RELATION BETWEEN GROWTH & POVERTY
P*
ln x*
1/2
AN INCREASE IN INEQUALITY WILL:
• INCREASE POVERTY AT A DECREASING RATE IF HCR < 50%
• DECREASE POVERTY AT AN INCREASING RATE IF HCR > 50%
• NEUTRAL WHEN HCR = 50%
RELATION BETWEEN INEQUALITY & POVERTY
0
1/2
1
P*
For ln x* <
For ln x* >
GROWTH vs. REDISTRIBUTION
• GROWTH ALWAYS REDUCES POVERTY
• PACE OF REDUCION VARIES BETWEEN LEVELS OF DEVT.
• REDISTRIBUTION REDUCES POVERTY ONLY WHEN THE SIZE OF THE CAKE ITSELF IS LARGE ENOUGH & POVERTY < 50%
What are the Bases for Indian Devt. Strategy?
• GROWTH & REDUCTION IN INEQUALITY• INEQUALITY, AS MEASURED BY LORENZ RATIO,
DECLINED AT THE RATE OF 0.38 % PER ANNUM IN RURAL INDIA DURING 1960-61 AND 1977-78
• INEQUALITY DECLINED AT THE RATE OF 0.59% PER ANNUM IN URBAN INDIA DURING THE SAME PERIOD
Lorenz ratios (at current prices)Year Rural Urban1960-61 0.3205 0.34771961-621962-63 0.313 0.53661963-64 0.2974 0.35961964-65 0.2936 0.34921965-66 0.2972 0.33851966-67 0.2934 0.33681967-68 0.2908 0.33241968-69 0.3051 0.32921969-70 0.2928 0.34031970-71 0.2831 0.32651971-721972-73 0.2993 0.3411973-74 0.2758 0.30131974-751975-761976-771977-78 0.3053 0.3349
T R E NDS IN INE QA U L IT Y IN IND IA
0
10
20
30
40
50
60
Y ear
Rural
Urban
How Valid are the Estimates?
• ESTIMATES ARE BASED ON THE NATIONAL SAMPLE SURVEY (NSS) DATA ON CONSUMER EXPENDITURE
• NSS DATA ARE AVAILABLE ONLY IN GROUP FORM, THAT IS, IN THE FORM OF SIZE DISTRIBUTION OF POPULATION ACROSS MONTHLY EXPENDITURE CLASSES
• LORENZ RATIOS ARE ESTIMATED USING THE TRAPEZOIDAL RULE
Lorenz Ratio
k
iiiii QQPPLR
111 ))((1
Limitations:
• UNDERESTIMATES THE CONVEXITY OF THE LORENZ CURVE;• IN OTHER WORDS, IGNORES INEQUALITY WITHIN EACH
EXPENDITURE CLAS• HENCE, UNDERESTIMATES THE EXTENT OF INEQUALITY• THE EXTENT OF UNDERESTIMATION INCREASES WITH THE
WIDTH OF THE CLAS INTERVAL
NSS Monthly Per Capita Expenditure (PCE) Classes
Expenditure Class Population (%) PCE(Rs)
< 8
8 – 11
11 – 13
13 – 15
15 – 18
18 – 21
21 – 24
24 – 28
28 –34
34 - 43
43 – 55
55 –75
> &5
Consumption Distribution: Metros 91961/62 & 1970/71)
Expenditure Class 1961/62 1970/71
< 8 - (-) - (-)
8 – 11 0.89 (0.18) - (-)
11 – 13 1.21 (0.31) 0.24 (0.04)
13 – 15 1.44 (0.43) - (-)
15 – 18 5.79 (1.99) 1.09 (0.27)
18 – 21 6.24 (2.53) 2.41 (0.70)
21 – 24 8.16 (3.86) 1.77 (0.60)
24 – 28 8.79 (4.82) 6.55 (2.50)
28 –34 12.55 (8.02) 7.38 (3.43)
34 - 43 13.15 (10.36) 17.55 (9.88)
43 – 55 14.31 (14.47) 13.61 (9.52)
55 –75 12.30 (16.36) 18.27 (17.91)
> &5 15.17 (36.67) 31.13 (55.15)
Lorenz Curve: Indian Metros 1961/62 (current unadjusted)
0
10
20
30
40
50
60
70
80
90
100
0 20 40 60 80 100
Series1
Series2
Lorenz Curve: Indian Metros 1970/71 (Current unadjusted)
0
10
20
30
40
50
60
70
80
90
100
0 20 40 60 80 100
Series1
Series2
Estimates of Lorenz Ratios: All India )(current: unadjusted)
Year Rural Urban Metros1961-62 31.36 35.7 34.51962-631963-64 29.7 36 34.11964-65 29.1 34.9 34.71965-66 29.7 33.9 33.11966-67 29.3 33.7 31.11967-68 29.1 33.2 29.91968-69 30.5 32.9 30.11969-70 29.3 34 30.21970-71 20.3 32.7 291971-72 1972-73 29.9 34.1 35.71973-74 27.6 30.1 34.61974-751975-761976-771977-78 34 34.6 NA
Estimates of Lorenz Ratio: All India (Current; unadgusted)
0
5
10
15
20
25
30
35
40
1961-62 1962-63 1963-64 1964-65 1965-66 1966-67 1967-68 1968-69 1969-70 1970-71 1971-72 1972-73 1973-74 1974-75 1975-76 1976-77 1977-78
Y ear
Loren
z Rati
o (%)
Rural
Urban
Metros
Estimates of Lorenz Ratios: All India)( adjusted & onstant)+B7
Year Rural Urban Metros1961-62 31.6 36.2 36.71962-631963-64 30.2 36.9 36.91964-65 30.9 37.1 39.91965-66 31.2 36.2 381966-67 31.4 37.1 37.91967-68 32 38 38.41968-69 32.9 36.6 38.11969-70 31.8 38.5 40.71970-71 30.4 37.3 40.11971-721972-73 32.4 37.8 401973-74 30.5 34.4 39.71974-751975-761976-771977-78 35.2 36 NA
ESTIMATES OF LORENZ RATIO: ALL-INDIA (CONSTANT; ADJUSTED)
0
5
1 0
1 5
20
25
30
35
40
45
1 961 -
62
1 962-
63
1 963-
64
1 964-
65
1 965-
66
1 966-
67
1 967-
68
1 968-
69
1 969-
70
1 970-
71
1 971 -
72
1 972-
73
1 973-
74
1 974-
75
1 975-
76
1 976-
77
1 977-
78
Y ear
Lore
nz R
atio
(%) Rur al
Ur ban
Metr os
top related