inequality increasing everywhere? evidence from a global ... · inequality of opportunity: a teaser...

Inequality increasing everywhere?

Evidence from a global database of

household surveys

Francisco H. G. Ferreira, Christoph Lakner and Ani Rudra Silwal

World Bank

Expert Group Meeting on Inequality Research, UN Headquarters, 12 September 2018

Overview

1. Motivation and context

2. Inequality within countries: Data assembly

3. Global and regional averages

4. Country trends (1993-2008; 2008-2013)

5. Robustness checks: alternative databases

6. Inequality of opportunity: a teaser slide

7. Conclusions

1

1. Motivation and context: global inequality

declining for first time since industrial revolution

2

Global Inequality, 1820–2010

Source: Based on figure 1 (p. 27) of The Globalization of Inequality by Francois Bourguignon

(Princeton University Press 2015).

Global inequality after 2008

3

Source: World Bank (2016): Taking on Inequality

2. Inequality within countries: Data Assembly

• Sources: PovcalNet + (LIS-based) ‘All the Ginis’ (Milanovic, 2016)

Final database: 9% of obs. from ATG, rest from PovcalNet

• 6 benchmark years: 5-year intervals from 1988 to 2013. Surveys are

within a two-year window.

• Measurement challenges:

Mix of income and consumption

Survey comparability (08-13 sub-sample ‘more comparable’)

• Inequality indicator: Gini coefficient of household per capita disposable

income or consumption expenditure among individuals.

4

Population coverage

5

World East AsiaEastern

EuropeL. America Middle East South Asia

Sub-Saharan

Africa

Industr.

Countries

A. Full sample

1988 73 79 90 93 91 42 96 10 75

1993 101 88 95 86 93 76 97 68 77

1998 106 71 95 82 95 70 22 71 75

2003 135 91 95 99 94 77 98 77 78

2008 137 92 96 93 95 72 98 70 95

2013 104 81 94 90 92 57 87 52 72

B. Sub-samples

Balanced (1988-2013) 43 47 88 14 79 22 11 4 69

Balanced (1993-2013) 58 53 93 45 83 22 12 14 71

Long-run trends

(1993-2008)91 84 95 83 87 62 97 53 76

Short-run trends

(2008-2013)81 54 81 87 90 32 12 21 69

C. Share of income surveys for full sample (percent)

2013 51 0 42 100 25 0 5 100

No. of

countries

Share of regional population covered by data (percent)

3. Global and regional averages

6

Remainder of presentation based on unweighted results.

34

35

36

37

38

39

40

41

42

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

de

x

Unweighted Weighted

Unweighted, balanced Weighted, balanced

Recent trends in national inequality are sensitive to population weights

National inequality in the average country

7

34

35

36

37

38

39

40

41

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

de

x

15

20

25

30

35

40

45

50

55

60

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

dex

10th percentile Median

Mean 90th percentile

Differences across regions

8

20

25

30

35

40

45

50

55

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

dex

East Asia and Pacific

Eastern Europe and Central Asia

Latin America and the Caribbean

Middle East and North Africa

South Asia

Sub-Saharan Africa

Industrialized Countries

Balanced panel of countries

9

34

35

36

37

38

39

40

41

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

dex

Mean (unbalanced sample1988-2013)

Mean (balanced sample1988-2013)

Mean (balanced sample1993-2013)

4. Looking directly at country-level trends

10

↑ +/-1pp ↓ Total 1993 2008 ↑ +/-1pp ↓ Total 2008 2013

E. Asia & Pacific 5 1 3 9 37.8 39.1 1 1 5 7 39.3 37.3

E. Europe & C. Asia 5 2 6 13 33.9 32.5 6 8 9 23 31.9 31.4

L. America & Caribbean 8 0 11 19 49.0 47.0 3 2 12 17 49.7 48.0

M. East & N. Africa 1 1 3 5 39.8 36.4 0 1 1 2 35.3 33.4

S. Asia 3 0 1 4 31.0 34.5 0 1 2 3 36.7 36.2

Sub-Saharan Africa 8 2 10 20 47.6 45.1 3 2 4 9 44.1 43.8

Ind. Countries 12 4 5 21 31.4 32.6 6 6 8 20 32.0 31.8

World 42 10 39 91 40.1 39.3 19 21 41 81 37.9 37.1

Long-run (1993-2008) Short-run (2008-2013)

Number of countries with: Mean Gini Number of countries with: Mean Gini

5. Robustness checks

1. Income and consumption surveys: Global results are robust to scaling

down income-based Gini indices (factor of 0.861 from Alvaredo and

Gasparini (2015))

2. Alternative inequality databases:

2015 special issue of the Journal of Economic Inequality: Review of

different databases; shows that country-trends can be quite different.

Our main source: PovcalNet (+ATG)

Comparison with other microdata-based databases: CEPAL, Eurostat,

LIS, OECD, SEDLAC; WID (incomplete).

Also compare with SWIID

11

Robustness check: level differences

12

1988 1993 1998 2003 2008 2013

Eurostat Average difference 1.88 2.93 2.23 1.59 1.66

N 7 18 24 30 31

LIS Average difference 3.02 2.77 2.57 2.53 2.97 3.01

N 19 24 27 25 10 12

OECD Average difference 2.33 1.16 2.21 2.05 1.79 1.81

N 6 11 11 23 29 18

SEDLAC Average difference 1.55 0.71 -0.75 0.56 0.35 0.14

N 9 13 19 15 18 17

CEPAL Average difference -0.57 -0.75 -0.84 -0.88 -0.92 -1.72

N 3 9 6 10 14 15

SWIID Average difference 2.26 1.86 1.80 1.24 1.41 2.21

N 72 101 105 129 120 65

PovcalNet total # of observations 73 101 106 135 137 104

Robustness check: average national Gini

Limited number of pairwise

comparisons possible

(1) Begin with baseline sample

(PovcalNet + ATG)

(2) Replace income-based Ginis

with observations from Eurostat,

LIS and SEDLAC (in this order

of preference)

Possible to replace around 40% of

observations from baseline

13

34

35

36

37

38

39

40

41

1988 1993 1998 2003 2008 2013

Ave

rage

wit

hin

-co

un

try

Gin

i in

dex

PovcalNet Using some alternative sources

Robustness check: country-trends (1993-08)

↑↑ both PovcalNet & alt. source show rise greater than 1 point

↑ one database shows rise of more than 1 point, while the other has

change within 1 point

? Gini changed by less than 1 point in both

Disagreement They go in opposite directions (change exceeds 1 point)

14

Database ↑↑ ↑ ? ↓ ↓↓ Disagreement Total

PovcalNet 42 10 39 91

PovcalNet vs. Eurostat 1 2 2 1 1 0 7

PovcalNet vs. LIS 4 1 0 0 2 0 7

PovcalNet vs. SEDLAC 6 0 0 0 6 0 12

PovcalNet vs. CEPAL 1 0 0 0 4 2 7

PovcalNet vs. SWIID 34 8 5 9 27 5 88

PovcalNet vs. (Eurostat + LIS + SEDLAC) 11 3 2 1 8 0 25

Robustness check: country-trends (2008-13)

↑↑ both PovcalNet & alt. source show rise greater than 1 point

↑ one database shows rise of more than 1 point, while the other has

change within 1 point

? Gini changed by less than 1 point in both

Disagreement They go in opposite directions (change exceeds 1 point)

15

Database ↑↑ ↑ ? ↓ ↓↓ Disagreement Total

PovcalNet (baseline) 19 21 41 81

PovcalNet vs. Eurostat 7 3 7 3 10 0 30

PovcalNet vs. LIS 1 0 1 0 0 0 2

PovcalNet vs. SEDLAC 3 0 1 1 12 0 17

PovcalNet vs. CEPAL 1 2 0 1 8 0 12

PovcalNet vs. SWIID 8 9 6 16 15 2 56

PovcalNet vs. (Eurostat + LIS + SEDLAC) 12 1 9 4 22 0 48

Comparison with WID: 1993-2008

Comparison with tax-record based data from WID, carried out in mid-2017.We

could probably expand the number of matches now.

Can match 15 or 16 countries. Different measures (top shares vs. Gini) and

different datasets (tax records vs. household surveys).

Notice how the overlap between our sample and WID lies firmly in our sub-

sample where inequality is increasing.

There are disagreements, but given all the differences in data sources and

summary measures, remarkably few disagreements.

16

Database ↑↑ ↑ ? ↓ ↓↓ Disaggrement Total


PovcalNet vs. WID Top 10% 9 2 1 2 0 1 15




Comparison with tax-record based data from WID, carried out in mid-2017.We

could probably expand the number of matches now.

Can match 15 or 16 countries. Different measures (top shares vs. Gini) and

different datasets (tax records vs. household surveys).

Notice how the overlap between our sample and WID lies firmly in our sub-

sample where inequality is increasing.

There are disagreements, but given all the differences in data sources and

summary measures, remarkably few disagreements.

17






What are the disagreements (in top 1%)?

Reading: Between 1993 and 2008, Italy’s Gini index fell from 35.3 to 33.7 (by

1.6 points), while the top 1% share increased from 7.9% to 9.7% (by 1.7

pp).

Country

Survey

year 1993

Survey

year 2008 Gini 1993 Gini 2008

Change

in Gini

Top 1% share

1993

Top 1% share

2008

Change in

Top 1% share

Ireland 1994 2008 0.369 0.309 -0.060 0.079 0.105 0.026

Italy 1993 2008 0.353 0.337 -0.016 0.079 0.097 0.017

Spain 1995 2008 0.364 0.348 -0.016 0.083 0.098 0.015


Fewer matches (only 8 or 9 countries); this was updated a year ago.

Even the WID largely shows a decline or no change over this period

19






Country

Survey

year 2008

Survey

year 2013 Gini 2008 Gini 2013

Change

in Gini

Top 5% share

2008

Top 5% share

2013

Change in

Top 5% share

South Africa 2006 2011 0.648 0.634 -0.014 0.419 0.442 0.024

Spain 2007 2012 0.339 0.359 0.020 0.246 0.213 -0.033

A final caveat: absolute inequality typically still rises,

even as relative inequality declines!

6. Inequality of opportunity

Source: www.Equalchances.org

21

A sizable share of income inequality represents unequal opportunities: and that

share rises with income inequality

Conclusions

1. Global inequality – as measured by HH surveys – has declined during this

century. This is driven primarily by a decline in inequality between countries.

2. Yet, for the average country, within-country inequality appears to have

stopped rising since around 1998. Indeed, when unweighted by population, it

has declined somewhat.

3. This trend is driven by developing countries. Increases are more frequent

among rich countries.

4. The trends are robust to comparisons with other cross-national datasets.

5. More work is needed in comparing it with datasets that make a special effort to

include top incomes (e.g. WID)

i. Sample selection bias in WID?

ii. Important to use inequality measures that satisfy continuity.

22

inequality increasing everywhere? evidence from a global ... · inequality of opportunity: a teaser...

Documents