Racial Discrimination and Occupational Attainment in Britain
Author(s): Mark B. Stewart
Source: The Economic Journal, Vol. 93, No. 371 (Sep., 1983), pp. 521-541
Published by: Wiley on behalf of the Royal Economic Society
Stable URL: http://www.jstor.org/stable/2232393
Accessed: 23-03-2018 12:59 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Royal Economic Society, Wiley are collaborating with JSTOR to digitize, preserve and
extend access to The Economic Journal
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
The Economic Journal, 93 (September I983), 521-541
Printed in Great Britain
RACIAL DISCRIMINATION AND OCCUPATIONAL
ATTAINMENT IN BRITAIN*
Mark B. Stewart
The last few years has seen rising concern about the relative economic position
of racial minority groups in Britain. Despite this growing concern empirical
analysis has been scarce, particularly in the area of earnings and employment
and particularly at a national level.' This continual lack of empirical evidence
itself acts as a barrier to the elimination of racial discrimination in the labour
market. If effective anti-discrimination policy is to be formulated, it is necessary
to know the 'form' that racial discrimination takes.
This lack of evidence contrasts markedly with the position in the United
States where a considerable literature on the occupational and earnings position
of various minority groups has accumulated.2 The comparative neglect of this
area by British economists may have been partly due in thepast to a feeling that racial
discrimination was less of a problem in the British labour market, but has mainly
been due to a lackof adequate data. The recentNational Training Survey provides
a timely opportunity to examine racial differentials and their interaction with
various factors, particularly education, labour market experience and training.
This paper is an attempt to provide empirical evidence on differentials in
occupational positions between black immigrants and white UK-born individuals
equal in other occupation-determining characteristics. Only discrimination
within the labour market is examined and it should be noted that this may be
less than total discrimination, since some of an individual's characteristics may
themselves be the result of discrimination before the market. In comparing these
two groups it should also be noted that any differentials ascribed to discrimination
will be the result of a combination of discrimination on the grounds of
colour and discrimination on the grounds of country of birth. The extremely
small proportion of the labour force who are black and UK-born is reflected in
the data set used, which contains an insufficient number of such individuals to
separate out these two factors. The use of a sample of white immigrants for this
purpose requires untenable assumptions and is not undertaken.
The National Training Survey was conducted on behalf of the Manpower
* The author is grateful to the Manpower Services Commission for permission to use the National
Training Survey data. They are, however, not responsible for the views expressed herein, which are
solely those of the author. The initial computer programming to unravel the complexities of the NTS
tapes was done by Ben Knox, to whom I owe a debt of gratitude. In addition, I am grateful to Christine
Greenhalgh, Richard Layard, David Metcalf, Steve Nickell and a referee and editor of this JOURNAL
for helpful comments on an earlier draft. This substantially revised version was finally completed while
I was visiting the Industrial Relations Section at Princeton University to whom I am most grateful.
Finally I gratefully acknowledge the expert secretarial assistance of Raquel Balestena.
1 The main national level evidence is provided by the PEP survey, the results of which are presented
in Smith (1976). The Community Relations Commission evidence to the Royal Commission on the
Distribution of Income and Wealth (1978), for example, relies heavily on this source.
2 See, for example, Ashenfelter (1970, 1972), Freeman (I973, 1974), Hall and Kasten (1973),
Strauss and Horvath (1976) and Weiss (I970).
[ 521 ]
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
522 THE ECONOMIC JOURNAL [SEPTEMBER
Services Commission in late I975 (see Manpower Services Commission (I978)
for details) and involved detailed interviews with about 54,000 individuals. Its
main advantages for the present investigation are twofold. Firstly, the provision
of information on occupation histories allows investigation both of individual
occupational movements and of changes in the average differential over time.
Secondly, the size of the survey means that a sufficiently large sample of black
immigrants can be generated for the analysis (about five times the size of that
available from the General Household Survey for example).
In addition to estimating the average differential in occupational attainment
it is also important to know how this differential varies across individuals with
different characteristics and in particular how the differential varies over the
life cycle. These questions are investigated in Section II using a simple model for
estimating differentials that is laid out in Section I. Given the differences in
occupational level attained by the two groups a key question to examine is how
these compare with those in earnings. What is the relative importance in the
earnings differential of differences in occupational attainment and differences
in earnings given the occupational level attained, i.e. within occupations? Are
differences in earnings the result of black immigrants being paid less for doing
similar jobs or the result of their not gaining access to the better paid jobs?
This question is investigated in Section III.
The I9 70's saw dramatic changes in labour market conditions in Britain and it
seems pertinent to ask how these affected the relative economic position of black
immigrants. The picture outlined in Sections II and III is a snap-shot of a point
in time (I975). In Section IV fixed age samples for each year from I970 to I975
are analysed to investigate whether or not that picture changed over the period.
In contrast to this examination of aggregate changes, individual mobility is
considered in Section V. The longitudinal nature of the occupation data is
utilised to investigate the key determinants of individual occupational progress
over a five-year period and to examine whether or not black immigrants differ
from UK-born white workers in this respect.
I. A MODEL FOR ESTIMATING DIFFERENTIALS
This section presents a methodology suitable for the estimation of racial
differentials, based on that used by Oaxaca (I 973) and others since. The
of direct discrimination experienced by individual i may be defined as
yi-ytw
where Yi is the occupational position attained by individual i measured in some
suitable way and Yw is the occupational position that individual i would have
attained were he white, but otherwise had identical characteristics. Since for
blacks Ylw is not directly observable, a model of the occupational attainm
process must be used. Assume that the position attained is generated for the two
groups in the following way
Ln YB = x' ,B + ui,
Ln Ylv = x? PW + u2V,
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
I983] RACIAL DISCRIMINATION 523
where xi is a vector of characteris
meters and uli, u2i are random err
process may be written as
LnJ? = x' BD +xPWv(I -Di) +ui
= xt W? + X (pB _PW) Di + ui,
where Di is a binary variable indicating whether the individual is black and
Ui = Diuli + (i -Di) u2i. From the definition of Ai
Ln Yi = Ln YIw + Ai Di,
where Ai = Ln (I + Ati). Thus having estimated occupational attainment
equations for the two groups, the differential for the ith individual can be
estimated from
X B= x W) = XiA^,
and the mean differential in the black sample from
A - XB'( B _W) = -B A
where xB is the vector of means in the black sample. Alternatively from the
differential
Ai = Ln Y, - Ln Ylv,
where Yl" must be estimated since it is not observable for blacks. Thus
A ___ __ _ _ _
A = Ln Y-Ln Yw = Ln YxB0W =x
as above. Of course there is an index number problem with this: A could be
evaluated at jjW instead of at xB. However, to measure the extent of discrimination
against blacks it seems intuitively more reasonable to consider the average
over the black sample. In addition since blacks are such a small proportion of the
population in Britain it is reasonable to use the coefficients from the white
equation to estimate the occupational attainment of the black sample in the
absense of discrimination. Having estimated Ai for an individual with a certain
set of characteristics, the estimate of Ai is given by
= *p (XA) - I.
The same model can be used to estimate differentials in earnings and, modified
slightly, to analyse occupational progress.
II. DIFFERENCES IN OCCUPATIONAL ATTAINMENT
Measurement of differences in occupational position requires the construction
of a numerical measure, since occupational position is not naturally cardinal or
even ordinal. Two main approaches are possible. The first is to measure an
occupation's position by the average hourly earnings received by those in that
occupation. The second is to use a 'status' scale such as that constructed by
Goldthorpe and Hope (I974) from an interview sample. The first approach h
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
524 THE ECONOMIC JOURNAL [SEPTEMBER
several advantages and is the way that occupational position is measured in this
paper. Unlike with the alternative approach, it is abundantly clear what is being
measured. Since this measurement is in money units, it also allows easy interpretation
of the results. In addition, it provides for an easy comparison with
current earnings differentials thus permitting an examination of the relative
importance of within-occupation earnings differences and differences in occupational
position. The main disadvantage of the approach is that it takes no
account of the non-pecuniary aspects ofjobs which are an important component
of their desirability.
Whilst the alternative approach may take some account of the non-pecuniary
attributes of an occupation, it suffers from the inherent difficulties in the interpretation
of the concept of 'status.' In addition, there is some degree of arbitrariness
in the cardinalisation involved in the construction of the GoldthorpeHope
scale from the interviewee rankings.
Despite these differences, Phelps Brown (I977) and others have pointed out
that there is broad agreement between the grading of occupations by status and
by average earnings. Although this does not necessarily imply that estimated
differentials in occupational attainment as measured by average earnings will be
similar to those measured on the status scale, results based on a projection of the
Goldthorpe-Hope scale presented in an earlier version of this paper do in fact
indicate considerable similarity.
Turning to the details of the construction of the measure, the National
Training Survey uses the KOS (Key Occupations for Statistical Purposes)
classification currently used by the Department of Employment which identifies
396 occupations. Average hourly earnings by OUG (Occupation Unit Group)
were calculated from the I975 General Household Survey tapes and mapped
into the KOS groups.' Other sources, such as the New Earnings Survey,
classified directly by KOS provided insufficient coverage of occupations.2
The samples used are of white male workers born in the United Kingdom and
black male workers born outside the United Kingdom.3 Both samples include
those not currently in employment who have worked at some time before. These
people are allocated the value for their most recent occupation. This is an
attempt to avoid the problem of sample bias that may result from omitting this
group.4
The specification of the determinants of the occupational level attained by an
individual is based on the conventional Human Capital model. Training (both
formal and informal), as the means by which skills are acquired, is clearly a key
determinant of occupational position. A number of modifications are made to the
1 The mapping was performed by David Metcalf and the resultant variable is the same as that used
in Metcalf and Nickell (I982) and Nickell (I982).
2 Even after addition of unpublished information, provided by the Department of Employment, the
New Earnings Survey only affords average hourly earnings data on about half the KOS occupations.
3 The term black is used throughout this paper to refer to the members of the sample assessed by the
survey interviewer to be 'non-white'. Roughly 30 % were born in the West Indies, 40 % in the Indian
subcontinent and 30 % elsewhere.
4 There may instead be a measurement error problem if those currently not working have potential
occupational levels systematically lower (or higher) than their most recent actual occupational levels.
However, this is thought likely to be a less serious problem than the sample bias problem.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
1983] RACIAL DISCRIMINATION 525
standard earnings-function specification of Mincer (I974) and others. First,
binary variables are used to represent age on completion of full-time education
(before starting work) since there is generally evidence of non-linearity in the
returns.
Age on completion of full-time education is used as the measure of the amount
of education obtained in preference to a scale or set of dummy variables for the
educational qualifications obtained because of a serious problem associated with
the treatment of foreign qualifications, which is particularly pertinent in the
present context. All foreign qualifications, at whatever level, are coded under
one heading in the National Training Survey. It is therefore not possible to
allocate individuals with such qualifications to the category of the equivalent
U.K. qualification as would be necessary to estimate the differential defined in
the previous section.
The possible solutions to this problem are most unsatisfactory. A separate
dummy variable could be included to represent possession of a foreign qualification.
However, since the level of qualifications that this would represent is
unknown, it would not be possible to assess what the return would be in the
absence of discrimination, and hence not possible to measure the desired
differential.
Alternatively the possession of foreign qualifications could be ignored altogether.
However, this will tend to result in an understatement of the human
capital of those with foreign qualifications, which will be particularly serious
here since 72% of those with foreign qualifications have no U.K. qualifications.
As a result there will be a tendency to underestimate the required differentials,
since 22 % of black immigrants have foreign qualification whereas only o.I %
of the white U.K.-born sample do.
A further complication would arise if dummy variables or a scale for highest
qualification were used in deciding where to rank foreign qualifications. Since
75 % of black immigrants with these foreign qualifications have no U.K.
qualifications this is important to the estimation of the differential and would
further exacerbate the problem. Given these various considerations the use of
education completion age is prefered to the use of educational qualifications for
the present purposes. The variables Si6 to S2I + represent completion of
full-time education at these ages.
The appropriate measure of labour market experience also requires careful
consideration. Two alternative specifications are used. In the first a quadratic
in years of total labour market experience is used, whilst in the second a distinction
is made, for immigrants, between experience before arrival in the United
Kingdom and experience since arrival in the United Kingdom. Employers and
craft unions may not regard skills and general experience obtained abroad as
being comparable with the equivalent obtained in the United Kingdom. This
may be partly due to the skills being less applicable or the experience being less
relevant to the job undertaken in the United Kingdom. It may also be partly
due to discrimination on the part of employers and/or trade unions. Hence
whilst the use of total experience, X, may overstate the extent of discrimination,
the partitioning of experience into that before arrival, A, and that after arrival,
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
526 THE ECONOMIC JOURNAL [SEPTEMBER
Table I
Estimates of Occupational Attainment Equations
(Dependent variable = Ln (1975 occupational position).)
Variable White Black Black
Si6 0-1736 (0o0053) -0-0146 (0-0325) -o-oI66 (o0o324)
S17 0-2770 (o0oo84) 0-0912 (0-0380) o-o858 (0-0380)
Si8 0-3501 (o oio8) 0-1111 (o0o423) o-ioo8 (0-0424)
Si9 0-3285 (o-oi8o) 0-1114 (0o0541) 0-1033 (0o0542)
S20 o03703 (0-0231) 0-2I96 (0-0532) 0-2126 (0-0531)
S21+ 0-5539 (0o0102) 0-3124 (0o0441) 0-2996 (0o0439)
MAR 0o0557 (0o0075) 0-0028 (0-0352) 0-0137 (0-0355)
WDS 0-0290 (0o0123) -o-o568 (0o0725) -00-36I (0-0719)
SWKR -0 0230 (o-oo8o) -o-oi88 (0-0373) -0-0311 (00359)
PSE -0-0222 (0.028i) - o-o885 (0-0253) - 0-0937 (o0o256)
FTT 0-0749 (00042) 0-0103 (0-0260) o-oo84 (0-0259)
EVT o-o812 (0o0059) o-Io62 (0-0402) o-io85 (0o0400)
X 0-0123 (o-ooo6) -0-0027 (0o0044)
X2 - 0-00022 (o ooooI) o-oooo6 (o-oooog) A
_ 0-00021 (O-OOOII)
A2 - o-oos6 (0o0034)
X-A - --0072 (o-oo6 i)
(X- A)2 -00023 (0.00021)
Constant 4-3130 4 5479 4.5823
SEE 0-2728 0-2608 0.2595
R 2 00234 0-200 0-211
F 476- I 9 5 8-9
Sample size 2 I,85o 548 548
Estimated - i i-6 % -9-6%
average (o*oI5I) (0-0152)
differential
Notes:
(i) For definitions of variables see text.
(ii) Standard errors in parentheses.
(iii) Average differential evaluated at black sample means (see text).
(iv) Standard error given below the estimated average differential is that for A = Ln (i +A).
(X- A), may understate the extent of discrimination. Results are presented for
both specifications in what follows and might be regarded as providing
respectively overestimates and underestimates of the, degree of direct
discrimination.
Whilst the effects of informal on-the-job training are assumed to be captured
in this way, it may be expected that there will be additional returns to any spells
of formal training. Two extra variables are included in an attempt to capture
these effects. FTT indicates those who have undertaken a spell of full-time
training in the last IO years and EVT indicates those who have undertaken a
spell of evening vocational training in the last IO years.1
Further binary variables are included to indicate those who are married
(MAR), those who are widowed, divorced or separated (WDS) and those who
1 Thus only the 'incidence effects' of training are measured. No allowance is made for 'duration
effects' since the general evidence is that for post- I 965 vocational training, spells of longer duration do
not provide higher returns than shorter ones. See Nickell (1982) and Greenhalgh and Stewart (I982)
on this point.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
I983] RACIAL DISCRIMINATION 527
are 'secondary' workers (SWKR). 'Secondary' workers are defined to be those
who are neither heads of households nor the chief wage earner in a household.
Finally, a binary variable is included to indicate those with poor spoken
English (PSE), since this may well reduce their chances of access to a number
of jobs. Thus, when the extent of discrimination is considered below, it is to be
understood as being differences in attainment between men alike in these
various characteristics.
The results for the occupational attainment equations are presented in Table I.
The average differential is estimated to be - I I -6 % when total experience is
used and - 96 % when experience is partitioned into that before and that after
arrival, argued above to be over- and under-estimates respectively. Both are
highly significant and represent a sizeable differential in occupational position
between men equal in other observed occupation-determining characteris-
tics.'
The coefficient estimates in the separate equations are of interest in themselves.
The returns to education are higher at all levels for white workers than for comparable
black workers. These returns rise to an estimated 74 % for whites with
a full-time education completion age of 2 I or above, which is roughly double the
comparable figure for black workers.2 For the black sample there is no significant
return to leaving at I6 relative to leaving at I5 or below, whereas for the white
sample all education coefficients are significant. It would appear that there may
be significant barriers to entry facing black workers in regard to those jobs for
which higher education is the entry route for whites.
The marital status variables are both significant for whites. Married men
reach levels roughly 6 % higher than comparable single men and for the widowe
divorced and separated the effect is roughly half this. These effects may be a
combination of demand-side and supply-side effects. Employers may regard
being married as a signal of stability or reliability and married men with their
additional responsibilities may work harder or seek promotion or other advancement
more vigorously. In either case these effects do not appear to be present for
black workers. Similarly, being a 'secondary' worker reduces the occupational
attainment of white workers by about 2 % other things being equal, again
possibly a combination of supply and demand effects. This effect also is not
present for black workers.
The quality of spoken English is, as might be expected, an important determinant
of occupational attainment for black immigrants. Other factors held
constant those with poor spoken English have occupational levels 8-9 % lower
than those with good spoken English. The proportion of U.K.-born whites in
the sample with poor spoken English is very small and the effect insignificant.
The estimated experience profile for white workers has the customary shape
and reaches a maximum after about 28 years in the labour market. In contrast
the estimated black profile is flat: the two terms in the total experience framework
1 It is unlikely that all factors relevant to the determination of an individual's occupational position
have been included in these equations, or indeed ever could be. Hence throughout this paper it is
important to remember that differentials are calculated holding constant only those factors which can
be observed in this data set.
2 Differentials such as this are calculated as exp (,j) - i.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
528 THE ECONOMIC JOURNAL [SEPTEMBER
are both individually and jointly insignificant. The four terms in the partitioned
experience framework are also jointly insignificant. Whereas white workers
progress up the occupational ladder during the first part of their working lives,
it would appear that the same is not true for black immigrants. In the case of fulltime
training also the effect of having had a spell in the last I o years is insignificant
for black workers whilst it carries a return of about 8 % for whites. In both cases
the indication seems to be, as with the results on education, that barriers reduce
entry by black immigrants into those higher-level jobs normally accessible to
whites with education/training/experience.
The results for evening vocational training courses show a markedly different
picture. For white workers the effect is similar to that for full-time training at
about 8 %. For black immigrants however the significant effect of about I I % is
considerably larger than their insignificant returns to full-time training. There
are a number of possible reasons for this. Firstly, evening training may be more
productivity enhancing than full-time training for black immigrants. One possible
reason for this would be heterogeneity of training with black immigrants discriminated
against in terms of the type of training that they receive. This would
tend to make the estimates of the extent of discrimination given above underestimates.
Other possible explanations are that evening training acts as a proxy
for motivational factors or that employers treat it as a signal of these or of
reliability or other characteristics that they seek.
Whilst the average differential is of considerable interest, additional information
can be provided by calculating predicted differentials for the individual
workers in the sample of black immigrants (as described in Section I). The
differential exhibits considerable dispersion over this sample: the upper and
lower quartiles are roughly -3 and - I9 %, the median differential at about
-I 2 % is similar to the mean. However, examination of the distribution for this
sample suggests that this may not be a particularly representative figure since
there appears to be peaks at about -I 7 % and close to zero. All these differential
are calculated from the equations using total experience. Those calculated using
partitioned experience produce numbers smaller in absolute value but tell a
similar story.
The main reasons for this pattern have already been suggested: barriers to
entry into higher-level jobs resulting in lower educational returns and flatter
experience profiles for black immigrants. To illustrate, an individual who completed
full-time education at age I5 or below and has other characteristics equal
to the averages for the black sample has a predicted differential of + O 7 % whi
a similar individual who completed full-time education at age i6 or above has
an average predicted differential of - I 7.8 %.1
Another source of dispersion in the differential is labour market experience.
Those with little labour market experience have lower differentials since their
white indigenous counterparts are also largely in lower-level occupations at this
stage. The differences widen when white workers move up the occupational
ladder with experience while the black immigrants do not. Of those with x <
1 The schooling effects are evaluated at the conditional means (given S > i6) for the black immigrant
sample.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
I983] RACIAL DISCRIMINATION 529
Table 2
Variation in the Differential
Differential Differential
without with
partitioned partitioned
experience experience
(%) (%)
Average differential -I i -6 -9-6
Man with basic set of characteristics - 174 - I5.6
Deviations from basic characteristics:
School leaving age < 15 -0-3 + 2-0
= 17 -17-2 -15-7
= I8 -2 1I5 -20-5
= 19 -I9-8 I8-5
= 20 -I143 -12-9
>3 21 -2I7 -20-9
Single -12-9 -12.0
Widowed/divorced/separated -24-2 - 17.6
Secondary worker - 17 - I6.3
Poor spoken English -22.7 - 215
Full-time training in last Io years -22-6 -2 I1
Evening training in last I0 years - 153 - 133
Years of experience = o o-6 + 3-8
= I0 -120 -14.1
= 30 -179 -12-9
= 40 -13-6 -I.5
Experience prior to arrival (A) = 5 -I 6 I
= 10 --3-6
= 15 -8-i
(Comparison with white with (20-A) years of experience)
Experience prior to arrival (A) = 5 - -179
= I0 -I8-3
= 15 -I6.9
(Comparison with white with 20 years of experience)
Notes:
(i) A man with the basic set of characteristics has 20 years of experience (all in the United Kingdom
in the case of the partitioned experience equation), left school at I6 and is married. All other variables
in the equation take the value zero.
(ii) All differentials calculated from the estimated equations given in Table i.
(iii) Deviations from the basic set of characteristics are considered singly.
only 5 I % have -A > I9 % while 44 I % have -A < 4 %. In the remainder
of the sample (x I IO) 294%/, have -A > i9 % and 26-9 %have -A <4 %.l
These findings, expressed in a slightly different way, can also be seen in Table 2
where predicted differentials for various typical individuals are considered. First
the predicted differential for a man with the basic set of characteristics is presented
and then the effects of various deviations from this basic set are considered.
A man with the basic set of characteristics has 20 years of labour market
experience (all in the United Kingdom in the case where the partitionedexperience
equation is used), completed full-time education at age I6 and is
1 These calculations are based on the total experience equations. The overall picture is similar when
partitioned experience is used.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
530 THE ECONOMIC JOURNAL [SEPTEMBER
married. All other variables in the equation are set to zero. Hence he is not a
secondary worker, has good spoken English and has not undertaken any formal
training in the preceeding IO years. Such a man has a predicted differential of
-I 7.4 %, roughly equal to the average given education to I6 or above quoted
earlier, although of course rather larger than the overall average. The rest of the
table is largely self-explanatory. It can be seen again that education and
experience are the main causes of variation in the differential as discussed above;
however a number of lesser causes are also evident.
III. EARNINGS AND OCCUPATIONAL ATTAINMENT
This section compares the already estimated differentials in occupational attainment
with the comparable differentials in current earnings. In particular it
examines whether in addition to differences in occupational attainment there are
also differences in earnings given the occupational level attained, i.e. within
occupations.
Some consideration of the appropriate method of estimating such earnings
equations using this data set must be made at this point since the information on
earnings is incomplete.' Each individual in the survey was asked to place his
current earnings in one of a number of consecutive ranges listed on a card.2 Such
a procedure is thought to increase both the response rate and the number of
correctly classified incomes.
The latent structure of the equations to be estimated may be assumed to be
given by
Yi = x + ui (i- I...n),
where yi is the unobserved dependent variable (in this case
current weekly earnings), xi a vector of non-stochastic regresso
of unknown parameters. The ui are assumed to be independently identically
normally distributed random variables with mean zero and variance o.2. Hence
the distribution of the unobserved dependent variable is given by
Yi -N(xt P, .) (i =I., n)
The observed information concerning the dependent variable is that it falls
into a certain range. Let Ak be the upper boundary of the kth range (of the
logarithm of earnings). Then the information on earnings is of the form
Ak7- < yi < Ak.
Since the end ranges are open-ended, AO = - oo and AK= + cc, where K is the
number of groups (IO in the case of the NTS variable).
The likelihood of the observed sample is given by
L =f 171[F(AkX%k ) X F(AkP-1 )]X
where F is the cumulative distribution of the standard normal. Maximum
1 A more detailed investigation of this problem can be found in Stewart (1982).
2 In addition the data is censored (in the statistical sense) in that both end-ranges are open-ended.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
I983] RACIAL DISCRIMINATION 531
likelihood estimates can be obtained by a least-squares algorithm as outlined
below. To obtain consistent estimates of the parameters by a least-squares
regression on the x's would require the use of the conditional expectation,
E(yilAkl < Yi < Ak), as dependent variable. Given the latent structure this
expectation is given by
E(yilAk1 i < Ak) = xt4+ f0Zikf)-f(Zik]
where Zik = (A,, - x P)/o- andf is the standard normal density function. Hen
estimation of p and o- (by least squares) requires estimation of the conditional
expectations and vice versa. This suggests that iterative estimation between the
two might be appropriate. In fact it can be shown that such a procedure (with a
suitable correction to the least-squares estimate of o at each iteration) will converge
to the maximum-likelihood estimates (see Stewart (1982) for further
details).' Maximum-likelihood estimates, obtained in this way, of the equations
for current weekly earnings are presented, together with their asymptotic
standard errors, in the first three columns of Table 3.2 The sample is reduced to
those in employment at the time of interview and answering the earnings
question. The estimates for occupational attainment using this sample are
presented in the remaining columns of the table for purposes of comparison.
The average differential in earnings is estimated to be -I 7.2 % when total
experience is used and -9. I % when experience is partitioned into that before
and that after arrival. These compare with estimated average differentials of
- I3-o and - IO-9 % in occupational attainment. Since the earnings andc
occupation equations contain the same variables, consideration of regressions
for the logarithm of earnings relative to occupational position permits testing
of these differences. When partitioned experience is used the estimated average
differential in earnings is insignificantly different from that in occupational
position. However, when total experience is used, this difference is significant.
In this case roughly 3 of the earnings differential is due to differences in the
occupational level attained and about W to differences in earnings given occupation.
If it is appropriate to think of the two specifications bracketing the degree
of direct discrimination, then the conclusion is that between 75 and IOO % of
the differential in earnings is due to differences in occupational attainment. This
of course suggests that the major policy problem concerns occupation entry rather
than pay within an occupati6n.
Further interesting results are evident in the individual coefficients. The
experience profile in earnings for black immigrants is significant, unlike that in
occupational attainment. When the differences between the earnings and
occupation equations are examined, it is interesting to note that the experience
profile in earnings relative to occupational level for black immigrants is identical
to that for whites both in its initial slope and its curvature (both differences are
I Initial estimates were obtained by fitting a lognormal distribution to the sample distribution of
earnings and then allocating to each individual in an earnings group the estimated marginal conditional
expectation.
2 Asymptotic standard errors are obtained by inversion of the information matrix (see Stewart (i 982)
for details). The value of ' R2 is computed as the square of the correlation between the predictions and
the final estimates of the conditional expectations (given k).
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
532 THE ECONOMIC JOURNAL [SEPTEMBER
~~~~~~~~~~~~~~~~~~~~~~O coCO .N a) Lo00 o C50) CO L o - OC) CQ
L 0(O( 0 C0 00 C4 'q OO0
co o - o0 od o11 oC o~ o o o 0 o
0 oz oOo o o o o o o o 0 cJ o0 0000
Y b ~~~~~~~~~~~~~co
O00 0000000o 0 C' 00
CO 00 N r- t- OC 41 - OLQO Oc O Oc O It LO )COC) o _ i4 _- 0 0 cO . 0 -O- 0 0 0 (.0 LO00 .. 400 0 0 0 00 LO NL.)
0
0
Cc) UL Q C4 L 0'14 C r-0(0 tr-C4 Lo
o o Lo ~o "o o) o) " -4 o- o0 o 8Lo
999999 00 00 0 00 0 0
0~~~~~~~~~
Cd~~~~~~~~~~~CC
-1 Y ~ o 4 oo o oLo , o ~o o oi oLo o cco Q0 N 00 r- X s r- C) r- > oo
t; CO Lo dz OtC) LO*0C Of Q 0 C'l Lo0t,0 00 - -CO LcO Oi - -.0 C'Cl c
O co 0 CO co LO L OLO OO- ?O
Lo0) C4 t0 CO0) 0 4 Q-0C4(
0 '.0 (0 0 ~ ic
.UciCoCoCoLO99999~~L
_~~~~~i 1
0 LO 0> co co d 0 n Co Co o 0 0
99999900009 " "V C r 99
0 .oo o o o ooo oo
00Co~~~~t~ci-~~~)0~~0Z 00~~~~0C o00
S~~~~~~~~~~~0 C4 f' o r- 0 00 0 CO "o Lo no
. - ( 0C - 0 - " LOO Lo CO O O 0 i 0 C 0 c b
LO LO O O O O OLO O LO I I
Cd)Co
a _sCo 00O/)0CLOc(0
4. L coC( LO0C LOCr- CoLOOM o "4o
O0 c00 00 -00 0 99 - 0 - CO
C?)~~~~~~~~~~~~L Q0 0) - l- (. O L OO -. CO LO ) x 0 00 0 VL
d<?nONX~~~~~~~OCXQOOQ
4.) 4. co o o) o o o o o o oo
H~~~~~~C CO " o. 0 Lo C) 0 " L N n + moLO LO r- o - o ~ 0o LO CO CO u' O nc
cd L~~~~~~~~~~O
> dz ow~~~~~~~~~~ CO dQ0 - tn co C) CO a) co 'I O 0od OC
s -o __ oo c c_o 0-0 o It.0 o o o O c co
k~~~~~~~~~~L oco co o0 of o ~o o0 ot Lo o. o o 1? o o C! d
~~~~~ 0~~~~~~~cCoo 990 9 0 1 on 0 co z
0
CO LO 0 co CO LO (0C 04 Ci 0)0 0) 0) 0
3 Lo o Lo Co 0 - 04 LO c-n 0 0
0 iCO 00 0 - + G
ciCoCo 0 0 00 0 0 0
k~~~~~~~~~~~"t O O O4 O- O O O- O O O LO O Oo
o)~~~~~~~~~~~C cc) CO O ,00o o O O O4 O OOo
bI)
N
*.r, Co 4 - -o
Cd C': o
> D .6Ec
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
1983] RACIAL DISCRIMINATION 533
Table 4
Changes in the Differential Over Time
Average differential Average differential
estimated using estimated using
Year total experience partitioned experience
I 970 -I i *66 % -9-53%
(o-o i 65) (oo i 66)
I97I i I82 % -9-86%
(0-0 I 62) (0o0I63)
I972 -I3.I4% -I I.32%
(0-0I59) (o-oi6o)
I973 -12-68% -IO9I %
(O I57) (o-OI58)
I974 -II-89% -IO-23%
(0-0154) (ooI 55)
I975 -I 174% - IO-02 %
(O-OI52) (0-0I52)
Notes:
(i) Each differential is estimated on the basis of occupations at January I st of the appropriate year.
(ii) The sample for a given year consists of those aged less than 6o at that date who had entered both
the labour market and the United Kingdon by that date.
(iii) The standard errors given in parentheses are those for A = Ln (i + A).
insignificant). Whilst black immigrants do not appear to move up the occupational
ladder with age, they move up the earnings scales within occupations
in the same way as white workers do. Again the policy problem involves access
to higher-level occupations.
IV. MOVEMENT IN THE DIFFERENTIAL OVER TIME
The evidence presented in the previous two sections is for a single point in time:
January I 975. This section examines how the relative economic position of black
immigrants was affected by the changing labour market conditions of the I 970's.
In particular the average differential is estimated for each year from I 970 to I 975
inclusive, each differential being estimated on the basis of the occupational level
attained at January Ist of the appropriate year. This is made possible by the
longitudinal information available in this data set. The sample for a given year is
restricted to those aged less than 6o at January ist of that year who had entered
both the United Kingdom and the labour market by that data. This created
comparable fixed-age samples across the years. The results of this analysis are
presented in Table 4, both on the basis of equations using total experience
variables and of equations using partitioned experience variables.
The two main questions of interest in such an analysis are whether there is any
secular trend in the differential and how it is affected by changes in the aggregate
unemployment rate. The results provide no evidence of either a worsening or
improving trend in the average degree of direct discrimination over the period.
The average differential in I975 is almost identical to that in I970.
I9 ECS 93
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
534 THE ECONOMIC JOURNAL [SEPTEMBER
During the 24 monthsJanuary Ist, I969 to January Ist, I97I male unemployment
in Britain was generally at or below the 500,000 level with rates of between
3 and 3-5 %. The 24 months January Ist, I97I to January Ist, I973 showed
considerably higher levels of male unemployment at between 6oo,ooo and
80o,ooo with rates of 4-5-5-5 %. Finally in the period January Ist, I973 to
January Ist, I975 the level of male employment was generally lower again at
around 500,000 once more. As is generally the case these changes were largely
due to changes in the unemployment outflow rate rather than in the inflow rate.
This is borne out by looking at engagement rates. Examination of the May
4-week counts for manufacturing industries done by the Department of Employment
shows rates of 2-2 per IOO persons employed for I97I and I972 compared
with considerably higher rates for the other 4 years which have an average
of 28.1 Such an economic climate might be expected to be particularly disadvantageous
to the black immigrant labour force and indeed the average differentials
for January Ist, I972 and I 973 are approximately one percentage point greater
in absolute terms than those for the remaining years. However, the standard
errors for the individual years are at least as large as this, and hence it must be
concluded that there is little or no evidence of the average differential widening
during periods of worse general labour market conditions.2
V. OCCUPATIONAL PROGRESS AND INDIVIDUAL
TEMPORAL CORRELATION
This section is concerned with individual mobility and utilises the longitudinal
nature of the occupation data to investigate the key determinants of individual
occupational progress over a five-year period and to examine whether or not
black immigrants differ from U.K.-born white workers in this respect. Since
it is occupational progress within the U.K. labour market that is being
considered the sample is restricted to those who entered both the labour market
and the United Kingdom prior tO I970. The shorter the period considered the less
occupational change is exhibited in the sample whilst the longer the period the
greater the reduction of the sample required. The latter is of course particularly
important for the sample of black immigrants, which is reduced by 23 % when
this five-year period is chosen.
To model the determinants of individual occupational progress two simple
and intuitively appealing formulations suggest themselves.
AYi = Zi Yi + Vi, (I)
AYi = AYiL+?zY2+Vi (2)
where Auy is the change in the logarithm of occup
being considered and YiL is its value at the start of
variables that remain unchanged over the period and start and finish values
of variables that change.
1 Figures are from Department of Employment - British Labour Statistics Yearbooks for I969-74.
2 This is in contrast to much evidence on the position in the United States. See, inter alia, Ashenfelter
(2970) and Freeman (1973).
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
1983] RACIAL DISCRIMINATION 535
One possible interpretation of formulation (I) is as the result of differencing
the original occupational position equation. If this is given by
Yit = Xtt Pt+ Uit)
then differencing gives Y = x* - X?L L + AUi,
or alternatively = xi iA + x L + t.
Thus for variables that do not change over the period ylj = -,jL,
variables that do change start-of-period values and either end-of-period values or
differences should be included and both ,8j and /jL can be estimated. In addition
the original equation may be permitted to contain an unobserved individual
effect
Yit = X14 Pt +fi + Uit)
which is then eliminated in the estimating equation by the differencing. (See
Nickell (I982) for such an interpretation of an equation of this type.) An advantage
of formulation (I) relative to (2) is that even if v is serially correlated, a
consistent estimate of yT can still be obtained by OLS.
A possible defect of formulation (I) is that it does not explicitly account for
the effect on occupational progress of the occupational position attained at the
start of the period. If such an effect exists then estimation of (I) will produce
biased estimates of Y2 in formulation (2). However, these two formulations
should not be regarded as alternative specifications of the model. Rather they
are to be regarded as measuring different quantities. One way of characterising
the difference between the two formulations isin terms of conditional expectations.
Formulation (2) specifies the determination of the expected value of Ay conditiona
on YL and z whereas (I) specifies the determination conditional only on z. As a
model of the former (I) is misspecified but as a model of the latter it need not be.
Y2 is to be regarded as the vector of 'direct' effects on zvygivenyL. Formulation (I
may be regarded as the result of substituting the determining equation for YL int
formulation (2) and hence yT is to be regarded as the vector of 'direct plus
indirect' effects on Ay including those via YL.
A possible problem with the estimation of formulation (2) occurs if the vi are
serially correlated. However, one possible interpretation of (2) is as the result of
transforming an original occupational position equation with a serially correlated
error term assumed to follow a first order autoregressive process with
parameter p. This gives
Ayi = (P-I) YiL+Xt(P-PLP) +AXiPLP +ei
Unless it is desired to impose p = PL, which it is not, this interpretation does not
imply any restrictions on the parameters. Since ci is serially independent, and in
particular not correlated with YiL, the equation can be consistently estimated by
OLS. Of course one can only estimate P3j and PiL for variables that change over
the period.
Unobserved individual effects are not allowed for in formulation (2). Hence
formulation (I) may be thought of as the model with fixed effects and formulation
(2) as the model with serial correlation. Nickell (I982) investigates a model in
which both serial correlation and individual effects are allowed for. However,
19-2
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
536 THE ECONOMIC JOURNAL [SEPTEMBER
4 to "t " 00 r- C co to 00s N O O Ch o ?c 0) of _ di
" L 3O CW 4 " a) -4CO ci u> m . r- co CI tr bco ? en
Q of XCO 0 1 of -4 LO : Oi d co + co 3 C Q cn
>~~~~~r C) o - o o - o t ooo C4 C oLo co Lo o in o oo.0,
o co o- o - co o - o oo o o o 4 o o o G cO
~~~~~~~~~0 _s- b O - b +r c - CO LOCO ?C0 Moo 0 ?-W~~~~ rc c^ o NO - h ocXo an e o cs Lor O co
Q N *4 C: 4M N o V 't CS C4 N - co Ch 4 C4 mo tt4 0 L O
C;S 0 O 0 CO 0 C0 'CO Lo O 0 .0 0 ,0 .0
A:1 c O O O0 of O - C O, a) O t LO VO 0 M Ot ? ? O?
co cO co 00O b o m 0 O Ch r-o LO C 0 C4C 0) " m 00o
N r~ N co N - t- - , 01 U-Q Cl 01 CO C- V- U") NN V*
._,~~~~~~ - 0 N O CN O C? O C)O OOOOOO,.0 0 .0 0
QQOQQOQOQQOOO$~~~~~QoQ -"o OQ
c o N- c 0N
.4 to 1)~ C 00 to N 0 a CO CO S- C)o - -cl C) 0c N C) -0 C:) 0Q Ci N - 01 - ) - ,I to3 C 4 CO - N Q-t)C N t - C N N
LOOOOO tO O O O O 99 t? ? t? ? O ?O
V> c o f ow r- CO 00 to dz CO LOdCOcNO to ) M 0 . 0Lo0 0 toc
> 24 Ir) + td 0~v0 v 8? 1 ; h : - C) -- N 0 T) Ct 0 C1 CD C ?
--o fC (o N co CO -c cf C-. O - LO 0-0o - - to C-I
COCO o c -t NOt CO c - o cC-O CO 0CI t- 0) d C c CO
> < < Q o O co oloNoco o o oo 0 en co o o ooo cl
* to CNO C 1 i u d - OCCO CC ) c--C)O o 1:0 cC 0.~~~~~~~C
< a 4; - ?t- o - . r- C) M :.0 i r- co 0 CO cst
10 * co - COO O CO COO - CO - O O-- O Q 0. NO N - O
ts m > = , 0 0 .0 .0 .0 0 .00 p,0 .0 ,0 .0 .0 .0 Q O O O O O
CO co o o o o ? ?t
V l~~~~4- - en co- t oo - oco0L
- J4 + X C-Co t- 0 Nt) CO O OCO N0 OC5
% ~ ~ ~ ~ ~ - ~tC C) c-- CC) N ~ COC LOCO -t 10 COCd C C O C -) 10 C- I CC)ON CO
Co a < O) 0 0 0 0 0 0 0 0C 0 C 0 _ 0 o0 0 0 0C 0 0 o
COC-I C C COO o oo oo o o o Co o Co- o o-o oCo
r>
~ ^
C'S Nt *4 - t 4 -*4 - *4 r )Q C--
C ;3OCO CO O " C- O CO OC9 O-O C O 0 0 C C O O O
V1 c- O COC OC - OOCO COCOO 0--C-)C 0 0 COCO 0 10' 0)10C0
CO CO - .0 , 0 . N 0 , 00 . 0 , 0 . 0 .
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
I983] RACIAL DISCRIMINATION 537
s. 6X oo :; m o o oo t-! - LO ^>- 0 -.4 ~C O 0) (O~0 - Of CO M c o
LO LO - O CO .0 0 0 0 00 CO 00 0 OC 0N
o0c) 1 1. )9 0 000 LLOci 1O 4 i
?? 0000000 L o ? 0
O O I O 9 O . . 9 I I I 0
O O O O Os ) O O | O ' I I 0 i.O O0 :O
ot ~ ~~cr-. -
LO
cis - co 0) - -SgHmo 00
H m cn LO) c 0 OC OHt CO of<.
cO LO 0 o 0o O ? O O o o o
0 0 0 ~0 0t'.0
9 I C CI
0go m X bbb m o C00 CO - LO
00 $ 0 8 Lo o oD o 4 ?o "Q t- QQ | l | o
..oooN~CO^
LO "44 Lr) cr) r- C4~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~0) 1-0t
0 0 0 _1 0 - - N 0 t
^8 w 00999O
99999999 o > (0 tO 1-0{oi C?O0
0)0~0 I L ~ c 0 ""0?0 N00 00 ~
N 0 0 C) N( 00) c O co O O " oo ? ? co of NCOC N '-0 00 1$ ) 0 O O O O O O O s 0N
o 0 o o o o ., .0 0 o N ce
O o oio o tr~00o o o
O ? C[U) LO LO O O C s d
o o o oI o o o o I
0 Nt O Ci O O 0 511 0)1 0
. 0 000000.0 o I C
4-4 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ i
oo 00 of )(0COf le) '00
O 10t-)C i)Lr t-0 0 -C- N14 c) o )
CO0 0 0 0 0000000,0001 ~
0(O CO o o a)0 2o U) 0LO
Cld ~ ~ ~ O
O O O O 0 L O O ? O O LO
_N 0 - LO CO 0 CC c c 0 0 (0
-0 00 0 000 o0 0 - o C
99999999 Of ~cI Ico C co-co0
0 m of Of Lo tt 0 0 ~~ci-'
ct)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.
't Nt, r- ~~c 1~~14 LO 0
00 LO r- n ~c LO 0 0 0 00 LO 0 N
" 0 N 0 N 0 0 0 0 cn t- 0 11V
o4 - p 080 OC
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
538 THE ECONOMIC JOURNAL [SEPTEMBER
limitations on computing resources force him to restrict drastically the number
of independent variables used and to assume that the coefficients on all
variables remain constant over time (i.e. , = PL) thereby absorbing those
variables which are time independent into the individual effects. These restrictions
are felt to be undesirable in the context of the present investigation
and thus Nickell's more general model is not considered here. It should also be
noted that the extension of the formulation to include a second order lag on y
would considerably reduce the size of the black sample and hence is also not
investigated.
The results from estimating formulations (i) and (2) are given in Table 5,
together with the corresponding estimates with y as the dependent variable using
the same sample. Estimates are presented using both total and partitioned
experience variables.
One of the main findings of Section II where only a single cross-section was
examined was the relative flatness of the experience-occupation profile for black
immigrants. This is supported by the longitudinal analysis in this section. When
individual movements over time are considered black immigrants can be seen to
experience less upward movement than comparable white U.K.-born workers.
Consider first the ceteris paribus differentials in the ratio
I975 occupational position
I970 occupational position'
since these differentials are calculated as outlined in Section I. When YL iS
excluded, the average differential is -22 I % (using total experience). This is
composite of a direct effect given YL and an indirect effect via YL. The two
opposite signs and the composite effect is smaller in absolute terms than the
direct effect alone. This is because upward movement for both groups is greater
ceteris paribus the lower the starting level and black immigrants have starting
levels that are ceteris paribus lower than those of whites. The average differential
givenYL is -4-4 %Alternatively
the differentials in the absolute or proportional change, which
are perhaps more usual concepts, can be considered. It should be noted in these
cases however that there is a difference between the estimated differential for an
individual with characteristics equal to the black sample mean, which will be
examined here, and the mean differential for the black sample. (For the differentials
examined up to this point they were the same.) Considering first the
unconditional differentials (based on the equations without YL), the estimated
differential for an individual with these average characteristics is -5 I 9 % in the
proportional increase and - 57.3 % in the absolute increase.' In each case the
I (a) The mean value of Ay for the sample of black inmigrants is 0-020I. The predicted Ar" for an
individual with characteristics equal to the means of the black inmigrant sample is 0-0413. (b) The
difference between the geometric means of Y and YL for the sample of black immigrants is I -92. The
predicted difference for an individual with characteristics equal to the means of the black sample is
calculated from A
(yWA YW) = [I-e
where A is the geometri
levels for the sample used in table 5. The calculated value is 4-50.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
1983] RACIAL DISCRIMINATION 539
Table 6
Predictions of Percentage Increase in Occupational Position
1970-75 for Men with Various Characteristics
Initial
occupational Predicted increase (%)
position White Black
Basic set of characteristics Unconditional 4-5 - 3.1
73P 12-5 2-I
96p 6-6 -3*7
I27p -9 - 94
School leaving age = I8 Unconditional 5-9 o-6
73p 174 8.7
96p 112 2-5
127p 52 -3 5
Full-time training spell during Unconditional 6-8 -0-I
period 73P 15.6 4.8
96p 9-6 -I-2
I27p 37 -7-0
Notes:
(i) A man with the basic set of characteristics has 20 years of experience, left school at 16 and is
married. All other variables in the equation take the value zero.
(ii) All predictions calculated from the estimated equations in Table 5 using total experience
variables.
(iii) The two deviations from the basic set of characteristics are considered singly.
actual increase observed is less than half that predicted for comparable white
U.K.-born workers.
Turning to the conditional differentials, given YL equal to the mean for the
black sample, the estimated differential for an individual with average characteristics
is - 69-8 % in both the proportional and the absolute increase.1 The
actual increase observed is less than one-third that predicted for comparable
white U.K.-born workers.
Additional information is provided by examining such differentials for different
types of individual. Accordingly predicted movements for black and white
workers with the same characteristics are examined next for various sets of
characteristics. These predictions are of the percentage increases in occupational
position over the period I970-5. They are presented in Table 6 for men with the
basic set of characteristics (as used in Section II) and two variants on this set. For
each of these the calculations are performed on the unconditional equations and
also conditional on three levels of initial occupational position. These three levels
are the geometric mean of the black sample and one standard deviation (in
logarithmic terms) above and below this level.
1 (a) The predicted Alw for an individual with characteristics (including I970 occupational position
equal to the means of the black inmigrant sample is o-o61I. (b) The predicted difference for an individu
with characteristics (including I970 occupational position) equal to the means of the black sample is
calculated from
(YwY.FL) = [exp (A'y )] YL,
where YL is the geometric mean for the black sample. Hence the differential in the absolute change is
identical to that in the proportional change.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
540 THE ECONOMIC JOURNAL [SEPTEMBER
Clearly in all cases the predicted extent of occupational progress of a worker is
greater, other things equal, if he is white and U.K.-born. The difference
between the predicted increases for the two groups varies between 5-3 and
Io 8 percentage points in the illustrations given. It is also noticeable that in many
of the cases the prediction for black immigrants is of a decrease in occupational
position over the 5-year period. As before the analysis indicates that black
immigrants make considerably less progress in occupational terms as their length
of time in the labour market increases than do comparable white U.K.-born
workers.
Finally, it is relevant to note that given the temporal correlation of occupational
level for black immigrants, variations from this are virtually random.
In particular the undertaking of training during the period does not appear to
increase the expected extent of occupational progress. This suggests that the
effects of evening training found in Section II may possibly be the result of the
selection (possibly self-) for training of those with higher occupation-paths rather
than any positive impact of the training itself.
VI. CONCLUSIONS
This paper has presented a body of evidence on the degree of racial discrimination
in occupational attainment in Britain. The average differential was
estimated at between - 9 % and- I2 %. Considerable dispersion around this
mean was found with lower returns to education and a flat experience profile
being the main contributory factors. It appears that barriers reduce entry by
black immigrants into those higher-level jobs normally accessible to whites with
education/training/experience.
The average differential in earnings was also estimated and suggested that
between 75 and ioo % of it is caused by differences in occupational attainment,
indicating that the policy problem concerns occupational entry rather than pay
within an occupation. Despite changing economic conditions, no evidence was
found of movement in the average differential over the period I970-I975.
Finally it was found that during this period individual black immigrants made
considerably less occupational progress than comparable white U.K.-born
members of the labour force. In particular a black immigrant with average
characteristics had a proportional increase in occupational position less than half
that of a comparable white U.K.-born worker.
Whilst this paper has presented much evidence on the 'form' that racial
discrimination takes in the labour market in Britain, many questions remain
unanswered and much further information is required if appropriate policies are
to be designed.
University of Warwick and Princeton University
Date of receipt offinal typescript: March 1983
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms
1983] RACIAL DISCRIMINATION 541
REFERENCES
Ashenfelter, 0. (1970). 'Changes in labour market discrimination over time.' Journal of Human
Resources, vol. 5, pp. 403-30.
(1972). 'Racial discrimination and trade unionism.' Journal of Political Economy, vol. 8o, pp. 435-64
Community Relations Commission (1978). Evidence to the Royal Commission on the Distribution of
Income and Wealth. In Selected Evidence for Report No. 6 on Lower Incomes, HMSO.
Freeman, R. B (I 973). 'Changes in the labour market for black Americans, I 948-72.' Brookings Papers on
Economic Activity, pp. 67-I 20.
(1974). 'Labour market discrimination: analysis, findings and problems.' In Frontiers in Quantitative
Economics, vol. ii (ed. M. D. Intrilligator and D. A. Kendrick). North-Holland.
Goldthorpe, J. H. and Hope, K. (I974). The Social Grading of Occupations: A New Approach and Scale.
Oxford University Press.
Greenhalgh, C. A. and Stewart, M. B. (I982). 'The effects and determinants of training.' Warwick
Economic Research Papers, no. 213.
Hall, R. E. and Kasten, R. A. (I 9 73). 'The relative occupational success of blacks and whites.' B
Papers on Economic Activity, pp. 781-95.
Layard, R., Piachaud, D. and Stewart, M. (I978). The Causes of Poverty. Royal Commission on the
Distribution of Income and Wealth Background Paper, no. 5. H.M.S.O.
Manpower Services Commission (1978). People and Their Work, 1975. Technical Report.
Metcalf, D. and Nickell, S.J. (I982). 'Occupational mobility in Great Britain.' In Research in Labor
Economics, v, (ed. R. Ehrenberg), pp. 319-57.
Mincer, J. (I 974). Schooling, Experience and Earnings. NBER.
Nickell, S. J. (I982). 'The determinants of occupational success in Britain.' Review of Economic Studies
vol. 49, pp. 43-53Oaxaca,
R. (1973). 'Male-female wage differentials in urban labor markets.' International Economic
Review, vol. 14, pp. 693-709.
Phelps Brown, E. H. (I977). The Inequality of Pay. Oxford University Press.
Smith, D. J. (I 976). The Facts of Racial Disadvantage: A National Survey. PEP.
Stewart, M. B. (i 982). 'On least squares when the dependent variable is grouped.' Warwick Economic
Research Papers, no. 207. Review qf Economic Studies, forthcoming.
Strauss, R. P. and Horvath, F. W. (1976). 'Wage rate differences by race and sex in the U.S. labor
market: I960-I970.' Economica, vol. 43, pp. 287-98.
Weiss, R. (1970). 'The effects of education on the earnings of blacks and whites.' Review qf Econo
Statistics, vol. 52, pp. 150-9.
This content downloaded from 147.251.55.81 on Fri, 23 Mar 2018 12:59:29 UTC
All use subject to http://about.jstor.org/terms