PSE January 2007
The Distribution of
Individual Earnings in
Historical Perspective
A B Atkinson, Nuffield College, Oxford and
PSE
1
Introduction
1. Taking Today’s Sources back to 1945
•
New Earnings Survey in UK
•
DADS data in France
•
CPS in the US
2. Before the Second World War
•
US and Canadian Population Censuses
•
DADS data in France
•
Irish Census of Production
3. Before the First World War: UK
•
Official earnings surveys
•
Williamson estimates
•
New evidence from the income tax schedules
Conclusions
2
Figure 1 Earnings Dispersion in US 19734.8
40
Decile ratio (LH axis)
Decile ratio
4.6
39
4.4
38
4.2
37
4.0
36
3.8
35
3.6
34
3.4
33
3.2
32
Gini coefficient (RH axis)
3.0
31
2.8
30
1973
1978
1983
1988
1993
1998
2003
3
“Possibly the most striking phenomenon in the British
labour market over the last couple of decades has been
the massive rise in wage inequality. Wage differentials
have risen to a degree that pay inequality is now higher
than at any time over the last century” (Dickens, 2000,
page 27).
“Since the late 1970s wage inequality increased very
dramatically in the United Kingdom. After showing
relative stability for many decades (and a small
compression in the 1970s) there has been an inexorable
upward trend in the gap between the highest and lowest
earners in the labour market.” (Machin, 1996, page 62)
4
Earnings parade in UK
2
LN(Earnings/median)
Top percentile
1.5
3.0
1
4.5
Top decile
Pareto coefficient
0.5
LN(1/(1-F))
0
0
-0.5
Bottom
-1
decile
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
2005
1979
5
Introduction
1. Taking Today’s Sources back to 1945
•
New Earnings Survey in UK
•
DADS data in France
•
CPS in the US
2.
3.
Before the Second World War
Before the First World War: UK
Conclusions
6
Top and bottom earnings deciles in United Kingdom 1954-2005
200
180
75
Top decile
LH axis
70
% median
170
65
Employer survey data
160
Income tax
data
60
150
55
140
50
Bottom decile
RH axis
130
45
120
40
1954
1959
1964
1969
1974
1979
1984
1989
1994
1999
2004
7
% median
190
80
Pareto-Lorenz coefficient for UK: NES, Schedule E Earnings
calculated from share of top 1% in share of top 10%
5.0
Pareto-Lorenz coefficient
4.5
NES
4.0
3.5
Schedule E
3.0
2.5
2.0
1954
1959
1964
1969
1974
1979
1984
1989
1994
1999
8
Top and bottom earnings deciles in France 1947-2002
220
100
210
90
80
190
70
180
60
170
Bottom decile
RH axis
19
99
19
94
19
89
19
84
19
79
19
74
19
69
19
64
19
59
19
54
19
49
19
44
19
39
19
34
19
29
40
19
24
160
50
9
% median
200
19
19
% median
Top decile
LH axis
Pareto-Lorenz Coefficients France
calculated from share of top 1% in share of top 10%
4.5
Pareto Lorenz coefficient
4
UK NES
UK Schedule E
3.5
3
2.5
FRANCE
2
1.5
1946
1951
1956
1961
1966
1971
1976
1981
1986
1991
1996
2001
10
Top earnings decile in United States 1947-2005
250
240
230
Current Population Survey
ALL workers
EPI ALL hourly earnings
of full-time workers
% median
220
210
200
190
Current Population Survey
Men
180
EPI Men hourly earnings
of full-time workers
170
Census Bureau Men annual earnings of fullyear full-time workers
160
150
1939
1944
1949
1954
1959
1964
1969
1974
1979
1984
1989
1994
1999
11
2004
Bottom decile (lower quartile) earnings in United States 1947-2005
65
Current Population Survey
Men LOWER QUARTILE
60
EPI ALL hourly earnings
of full-time workers
% median
55
50
45
Census Bureau Men annual
earnings of full-year full-time
workers
40
35
EPI Men hourly earnings
of full-time workers
Current Population Survey
ALL workers LOWER
QUARTILE
30
1939
1944
1949
1954
1959
1964
1969
1974
1979
1984
1989
1994
1999
2004
12
Pareto-Lorenz Coefficients France, UK and US, calculated from
share of top 1% in share of top 10%
4.5
Pareto Lorenz coefficient
4
UK NES
UK Schedule E
3.5
US from Piketty and Saez
3
2.5
FRANCE
2
UK and US
UK and France
UK and US
1.5
1946
1951
1956
1961
1966
1971
1976
1981
1986
1991
1996
13
2001
Conclusions:
• Three distinct periods: “Golden Age”, Post-1968(1964),
and ICT Age.
• Different country experiences.
• Importance of looking at top as distinct from decile
ratio.
• Importance of looking at annual (frequent) data.
14
Introduction
1.
Taking Today’s Sources back to 1945
•
New Earnings Survey in UK
•
DADS data in France
•
CPS in the US
2.
Before the Second World War
• US and Canadian Population Censuses
• DADS data in France
• Irish Census of Production
3.
Before the First World War: UK
Conclusions
15
US and Canada Population Census Data
250
100
Census of Canada
225
Top decile
LH axis
175
150
80
Census of US
from Goldin and
Margo
% median
% median
200
"The Great
Compression"
Lower quartile
RH axis
Census of Canada
60
125
Census of US
100
40
1931
1936
1941
1946
1951
1956
1961
1966
1971
16
Top and bottom earnings deciles in France 1919-2002
220
Top decile
LH axis
80
190
70
180
60
170
Bottom decile
RH axis
19
99
19
94
19
89
19
84
19
79
19
74
19
69
19
64
19
59
19
54
19
49
19
44
19
39
19
34
19
29
40
19
24
160
50
17
% median
90
200
19
19
% median
210
100
Ireland: Census of Production Data 1937-1968
(joint work with Brian Nolan)
200
100
Highest decile
LH axis
180
% median
160
90
Upper quartile
LH axis
80
140
120
70
Lower quartile
RH axis
60
100
50
Lowest decile
RH axis
80
40
1937
1942
1947
1952
1957
1962
1967
18
Introduction
1.
Taking Today’s Sources back to 1945
•
New Earnings Survey in UK
•
DADS data in France
•
CPS in the US
2. Before the Second World War
•
US and Canadian Population Censuses
•
DADS data in France
•
Irish Census of Production
3. Before the First World War: UK
• Official surveys
• Williamson estimates
• New evidence from the income tax schedules
Conclusions
19
Earnings surveys in Great Britain
Normal
160
Highest decile
140
Last week
% median
120
Upper quartile
100
Lower quartile
80
60
Lowest decile
40
1886
1896
1906
1916
1926
1936
1946
1956
1966
20
Problems
• Not in fact stable: top decile changes by
more than 10 percent
• Coverage limited: male manual workers
in certain sectors
• Data of dubious quality
• Nothing on intervening years
21
Source of the UK 1886 data
“The rates of wages paid, or of average pieceearnings, in a normal week without overtime were
asked for each occupation in each industry in each
locality. … An attempt was made to describe the
distribution of wages by the assumption that the
wages for the same occupation in one district fell in
the same five shillings grade for all operatives,
distinguishing men, women, boys and girls.”
(Bowley, 1937, pp 100 and 101).
22
Williamson evidence for Britain "confirms that earnings inequality passed
through a "Kuznets curve" during the nineteenth century"
40
36
Gini coefficient %
32
28
24
20
4
16
12
8
4
Distance between gridlines = recent US increase
0
1827 1832 1837 1842 1847 1852 1857 1862 1867 1872 1877 1882 1887 1892 1897
J G Williamson, Did British Capitalism Breed Inequality?, 1985.
23
er
al
Lo
-4
ca
l
G
e
ov
e
s
es
tic
ou
re
r
do
m
la
b
M
es
M
in
Pr
er
in
s
se
tin
ng
g
tra
er
s
de
an
s
d
po
rt
er
N
s
at
Te
io
ac
na
he
lG
rs
ov
er
nm
en
t
Pr
of
C
es
le
si
rk
on
s
al
s
C
(p
le
r iv
rg
y
at
e
se
ct
or
)
no
n-
s
rn
m
fa
en
rm
t
la
bo
M
ur
an
er
ua
s
lw
R
a
ilw
or
Pu
ke
ay
rs
bl
s
ic
in
ut
co
ili
m
tie
m
s
od
ity
B
pd
ui
ld
tn
in
g
tr
ad
es
G
en
rm
M
al
Fa
Difference in Gini coefficient percentage points
Difference if remove occupational/industrial group
8
4
0
-8
24
Income tax Schedule E and Schedule D (income from
employment)
Data exist for 1845-1876 and 1898-1913
Typically 15 ranges
Cover small fraction of employed population (1% in 1875),
but use control totals from Feinstein.
Some years used by Wiliamson but not for this purpose.
25
Shares in total earnings in UK 1845-1913
4.0
8
3.5
7
Top 1%
RH axis
6
Top 0.5%
RH axis
2.5 Top 0.1%
5
2.0
4
1.5
3
LH axis
Top 0.05%
LH axis
1.0
2
0.5
1
Top 0.01%
LH axis
0.0
1845 1850 1855 1860 1865 1870 1875 1880 1885 1890 1895 1900 1905 1910
0
26
% total earnings
% total earnings
3.0
Pareto-Lorenz coefficients for UK 1845-1913
2.8
2.6
2.4
Calculated from
share of top 0.01% in
top 0.1%
Pareto coefficient
2.2
2
1.8
1.6
Calculated from share of top 0.05% in top 0.5%
1.4
1.2
1
1845
1850
1855
1860
1865
1870
1875
1880
1885
1890
1895
1900
1905
1910
27
Conclusions:
• Sequence of “episodes” rather than grand swings
• Differences at different points in the distribution
• Differences across countries
• To study rich pattern, need:
• Frequent observations
• “Real” distributions
There are data that have not yet been exploited!
28