CHANGES IN THE DISTRIBUTION OF MALE AND
FEMALE WAGES ACCOUNTING FOR EMPLOYMENT
COMPOSITION USING BOUNDS
R
ichard Blundel
l
Amanda Goslin
g
Hidehiko Ichimura
Costas Meghir
THE INSTITUTE FOR FISCAL STUDIES
WP04/25
Changes in the Distribution of M ale and Female Wages
Accoun ting for Emplo ymen t Composition Using Bounds
∗
Richard Blund ell
†
Amanda Gosling
‡
Hidehiko Ichim ura
§
Costas Meghir
¶
April 2006
Abstract
This paper examines changes in the distribution of wages using bounds to allow for t he
impact of non-random selection into work. We show that worst case b ounds can b e infor-
mative. However, since employment rates in the UK are often low th ey are not inform a tive
about changes in educational or gender wage differentials. Thus we explore ways to tighten
these bounds using restrictions motivated from economic theory. With these assumptions
we find convincing evi den ce of an in creas e in inequality within e duc ation groups, changes in
educational differentials an d increa ses in th e relative wa ges of women.
∗
Acknowledgments: We would like to thank three anonymous referees, Jaap Abbring, Daron Acemoglou,
Manuel Arellano, Jam es H eckman, Jo el H orowitz, Chuck M anski, W hitney Ne wey, E lie Tamer for useful comm ents
and discussion as well as pa rticipants in numerous seminars including the Ec onome tric Society Europ ean m eeting
in M adrid 2004, the mee ting of the E urop ea n A sso ciation of Labour E conomists in Lisbo n 20 04, the S tanford
Institute for Theoretical Eco nom ics 2002 and seminars at UC L, Johns Ho pkins and Northwestern Universities,
the Eco nometric Study G ro up in Lond o n 2004 a s well as pa rticipants in the conference o n Inferrence in Partially
Identified M o dels, Northwestern University, Fall 2005. Funding for this resea rch was provided by the ESRC centre
for the analysis of F iscal Policy at the IF S. D ata from the F ES ma de available from the CS O through t he ES RC data
archive has been used by pe rmission o f the C ontroller of HM S O. We are respo nsible for all errors an d interpretation s.
†
UCL and IFS, r.blundell@uc l.ac.uk
‡
Essex U niversity, IFS and C EP R, agosling@essex.ac.uk
§
UCL and IFS, h.ichimura@ucl.ac.uk
¶
UCL, IFS an d CE PR, c.meghir@ucl.ac.uk
1
1 Introduction and motivation
Economies suc h as the US and the UK have seen large and unprecedented increases in wag e in-
equalit y amongst w orkers over the la st 30 to 40 years. This is illustrated in Figure 1 where we
show the wa y that the interquartile ranges of male and female log hourly wages have evolved for
thosewhoworkintheUK.
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
78 80 82 84 86 88 90 92 94 96 98
Men Women
Figure 1: Interquartile range of log wages
These increases in inequalit y have been associated with increased returns to education, cohort
effects and increases in the returns to unobserv ed skill.
1
A v ariety of interpretations have been
given as to why these events ha ve occ urred; these include skill biased technical change, globaliza-
tion induced increased in competition for low skill workers and changes in the supply of graduates.
Gosling, Mac hin and Meghir (2000) show that the increases in the UK can be attributed to per-
1
See Juhn, M urphy and Pierce (1993) for the US and G osling, M achin and M eghir (2000) for th e UK .
2
manent differences across cohorts and in ch anges in the returns to education. A t the same time
the gap between the wages of working men and working women have fallen.
2
In parallel with these momento us changes in the distribution of observed wages, employm ent
rates for males and females and the composition of the workforce have changed. Employment for
men aged 22 to 59 has decreased from 93% in 1978 to 83% in 2000. This decline is not confined
to older men and reflects more than the increase in early retirement. Women on the other hand,
especially those below 30, sa w an increase in their emplo yment. In Figure 2 we show the age
profile for employment for 1978 and for 2000 by gender.
Thedeclineinemploymentsincethelate1970shasitssourcesinanumberofpossiblecauses
that in teract with each other. The large demand shock of the early 1980s combined with the
w elfare system and the wage setting institutions at the time to cause persistent unemploym ent.
3
A
mechanism propagating such persistence and relating to older individuals was the steady increase
of those receiving sickness and disability benefits during the 1980s, from whic h there is little
incentive to drop out and return to work.
4
Blundell, Reed and Stok er (2003) also emphasize the
role of the reforms to the housing market and in particular the change since 1983 in rent setting for
the public sector , leading to a large and steady increase in rents over the 1980s and 1990s. Rents
for those on lo w income are subsidized through housing benefit, which carries an implicit tax rate
on earnings of 95% when combined with other welfare programmes.
5
The reform inadverten tly
led to a steady increase over time in the range of earn ings ove r which some groups of people
faced high marginal tax rates. This would ha ve affected both the level and the composition of
participan ts.
6
These institutional changes affected older cohorts disproportionately because they
were ov er-represented in the public sector housing, partly explaining the more rapid decline in
2
See Ha rkness (1996 ) for the UK and Blau (1998) and Blau and Kahn (1997) for the U S.
3
see Jackman , Laya rd and Nickell (1991).
4
see Disney and Webb (1991)
5
Housing b enefit on its own has a margin al withdrawal rate of 65%.
6
Tho se w ith children an d th ose in h igh re nt areas have ex perienced large relative increases in out of work benefit
incom e.
3
emplo yment for older peop le. The above factors, will ha ve affected older and unskilled indiv iduals
more than the rest, although not exclusively, because out of work b enefit income increased for all.
Figure 3 illustrates that the ch ange in employment has been heavily skill biased. In this
paper we define skill by three education groups: those leaving full time education at or before 16
(statutory schooling), those who completed education sometime between 17 and 18 (high school
graduates) and those who completed full time education after 18 (some college). For women most
of the increase in employment can be accounted for by the increase in employment of women
with more than the minimal level of education. The employment rate of the statutory schooling
group has shown a slight decline over the entire time period. Blundell, Duncan and Meghir
(1998) document changes in the incen tive structure. Moreover, it is possible that the gradual
implementation and enforcement of anti-discrimination practices, formally introduced in the 1970s
may have made career progression for skilled women more of a reality over this time period (see
Stewart and Greenhalgh, 1984) .
As the composition of the workforce changes so will in general the observed distribution of
wages for workers, whether the aggregate participation rate changes or not. This obscures the
ch a n ges in the actual/uncensored distribution of wages, preven ting us from understanding the
nature of the change in inequality and the associated changes in educational, age and gender w age
differentials. A recent example which shows how important such selection issues can be for wages
is the paper by Blundell, Reed and Stoker (2003).
Selection effects have been central to labour economics ever since the pioneering work of
Gronau (1974) and Heckman (1974, 1979). Approaches are either parametric
7
or semi/non-
parametric.
8
Unfortunately, structural economic models fall short of delive ring the assump tio ns
required for identification. Taking a stand on the labour mark et paradigm is generally not sufficient
7
See He ckman ( 1979) or Heckman and Sedlacek (1985).
8
See for exam ple, Powell (198 7), N ewey (1988), Das, Newey a nd Powell (200 3), Heckman (1990), Choi ( 1990),
Ichimura and Lee (199 1), Coslett (1991 ), Ahn and Powell (1993).
4
