Machin&Manning-2002
Machin&Manning-2002
Abstract
This paper examines the structure of wages in a very specific labour market, for care assistants in
residential homes for the elderly on England's "sunshine coast". This sector corresponds closely
to economists' notion of what should be a competitive labour market as: (i) there are a large
number of small firms undertaking a very homogeneous activity in concentrated geographical
areas; and (ii) the workers they employ are not unionized, nor are they covered by any minimum
wage legislation so that there are effectively no external constraints on the wage-setting process.
We find that the structure of wages does not, in important respects, resemble what we
would expect in a competitive labour market. We find there is a small amount of wage dispersion
within firms and a correspondingly large amount between firms. And, the wage dispersion that is
present does not seem to be closely related to the productivity related characteristics of workers.
We propose a test of the hypothesis that unobserved labour quality can explain our findings and
reject it. The paper concludes with a discussion of other possible explanations of the patterns in
our data.
0
1. Introduction
This paper examines the structure of wages in a very specific labour market, for care assistants in
residential homes for the elderly on England's "sunshine coast". It analyses the results of a postal
survey of all such homes conducted in April 1992 with a follow-up a year later and a few pieces
of information added two years after that. Our interest in this sector arises from the fact that it
corresponds closely to economists' notion of what should be a competitive labour market. There
are a large number of small firms undertaking a very homogeneous activity in a concentrated
geographical area. The workers they employ are not unionized, nor were they covered by any
minimum wage legislation at this time (the UK’s National Minimum Wage was introduced only
in April 1999 – see Machin, Manning and Rahman (2000) for an analysis of the impact on this
sector) so that there are effectively no external constraints on the wage-setting process. We think
it reasonable to argue that most economists, asked a priori, would think that this market was very
competitive.
market wage for workers of a given quality. This has two important implications. First, workers
of identical quality should receive the same wage in different firms. Secondly, workers of
different quality should receive different wages even if they work in the same firm. Yet, when
We find that there is remarkably little dispersion of wages within firms and a surprisingly
large amount between firms. Of all the variables on which we have individual variation, the
wage typically has a proportion of dispersion that is inter-firm that is twice as high as any other
variable. In fact, something like 1 in 4 workers work in firms where all care assistants get paid
the same hourly wage, and another third of workers are in firms where only two different hourly
1
wages are paid. These firms do not have significantly less variation in observable characteristics
Of course, a believer in the relevance of the competitive model could (plausibly) argue
that our measures of worker quality are far from perfect and that the distribution of unobservable
worker quality could be such as to make the data consistent with the competitive model. We
present a test of this hypothesis and reject it. The basis of the test is as follows. We show that
the correlation between observed characteristics and wages is very different in the firms with and
without wage dispersion but the correlation between the prices charged to residents (an indirect
We conclude, from the examination of this data, that the competitive model is not
particularly helpful for understanding of the structure of wages in this labour market. But, this
prompts the question: what should be put in its place? We review a number of alternative
theories of the wage structure, discussing their strengths and weaknesses for the purpose of
explaining our data. No-one theory emerges as the explanation but we do come to a number of
general conclusions. First, frictions in this labour market are substantial enough to accommodate
considerable and long-lasting heterogeneity in the wage policies of employers without some
firms suffering a catastrophic reduction in profits. These frictions can account for the
heterogeneity in wages across employers as was emphasized by earlier micro studies of labour
markets (e.g. Lester, 1946; Slichter, 1950; Reynolds, 1951). Secondly, the lack of wage
dispersion within firms is probably driven by two factors: worker dislike of wage heterogeneity
on grounds of ‘fairness’ and employer dislike of wage heterogeneity in order to keep worker
demands for wage increases to a minimum. However, these conclusions, while consistent with
our data, must remain somewhat tentative. It seems plausible that, for the small employers in
2
this sector, there is an opportunistic aspect to wage policy, with wages being determined on an
ad hoc basis as events evolve. For example, the threat of a particularly valued worker to leave
may cause an employer to break a ‘one-wage’ policy that is otherwise followed. If this is the
case, then outside observers are always likely to have a hard time explaining why a particular
Data Description
The data set used in this paper was obtained from a survey undertaken by us in April
1992 (plus a follow up a year later) of all (2036 in total) private-sector residential homes for the
elderly located on England's "sunshine coast".1 We were able to sample the entire population of
homes by obtaining information on all homes within each county that we considered (Devon,
Dorset, Cornwall, Kent, Somerset and Sussex) as they all have to register with the relevant local
the responses that we received and, in terms of region and size, they seemed very representative
This sector was chosen because it closely corresponds to economists’ a priori ideas about
a labour market that should be well approximated by the perfectly competitive model. It consists
of a large number of small employers doing a relatively homogeneous activity (caring for old
people) and which are geographically concentrated (in some streets in some towns on the south
coast of England, almost every second house is a residential care home). Furthermore, most
workers in these homes need no formal qualifications: the old people in the homes we consider
1
See Machin, Manning and Woodland (1993) and Woodland (1993) for more details.
3
do not need specialist medical care and, as will be seen below, few workers have a formal
nursing qualification.
One other unusual feature of this data set is that we have information on all workers
within a large number of firms. This allows us to address issues like the extent of wage variation
within and between firms that cannot be considered with most data sets. It is this feature of the
The principal occupation of workers employed in these nursing homes is that of Care
Assistant and we focus specifically on the wages of Care Assistants in this paper. The reason for
doing so is that we are interested in the structure of wages in a very tightly defined labour market
One should note that, even once we restrict attention to Care Assistants alone there is still
a choice about whether to include only those labelled as day-care assistants or also those
recorded as being senior or junior day-care assistants. This is not trivial as different occupational
titles may simply be a way of paying different wages to different individuals and may not signify
any real difference in job content (see Baker, Gibbs and Holmstrom, 1994a, 1994b, for a
statement of this idea and Manning, 1994, for a working-out in the context of a search model).
Most of the results in this paper include those workers labelled as being senior and junior day
care assistants, but we also report some results using the more narrow definition of day care
In Table I we report some summary statistics on the distribution of wages in our sample.
We have data on 3221 Care Assistants in 434 homes in 1992 and, from the follow up of those
who responded in 1992 that we conducted a year later, 1826 Care Assistants in 236 homes in
4
1993. In our matched sample of 213 homes that we obtained responses from in both years we
Average wages are very low in this sector. In 1992 the mean wage was £2.97 per hour
and in 1993 was £3.07 per hour. This lies beneath the lowest occupational wage reported in the
New Earnings Survey in each of these years and is well below the £3.40 that the Labour Party
was advocating as a National Minimum Wage had it been elected in the General Election of
April 1992. Despite the fact that we are focusing on a very specific occupational group, there is
considerable dispersion in hourly wages among these workers. For example, in both years the
range between the tenth and ninetieth percentiles of the hourly wage is over one third of the
median wage. We also report information on the standard deviation of the log hourly wage. As
the wage distribution for the matched sample is very similar to that for the sample as a whole, we
An interesting question is how much of the wage variation described above is between
employers (different employers paying different average wages to their workforce) or within the
same employer (the same employer paying different workers different wages). In Table II we
present information on the proportion of total variation in the log hourly wage of all care
assistants that is inter-firm, the remaining proportion being intra-firm. In the upper panel of the
table one can see that for all care assistants almost 2/3 of log wage variation is between firms,
with only one-third being within firms. Part of the measured inter-firm wage dispersion is
because of the variation of wages across different regions so we also present measures of the
importance of inter-firm wage dispersion after introducing geographical controls. The controls
2
We have not used the data from a handful of single-employee firms as there is obviously no meaningful difference
5
for area involve 16 regions that we will use as our regional controls in the regressions below, and
the controls for town involve controls for the postal address (this is a very disaggregated measure
as there are then 129 towns in our sample). As one would expect, introducing finer regional
dummies reduces the measured importance of inter-firm wage dispersion but even with the town
dummies, the proportion of inter-firm wage dispersion remains close to 50%. As one might be
concerned that the results are driven by the existence of many small firms, we also present the
variance decomposition for workers who are in homes with more than 5 workers: the results are
very similar.
In order to put these figures into some kind of perspective, we also computed the
proportion of the observed variance in other personal characteristics that is inter-firm and intra-
firm. We have information on age, job tenure, and hours so we use a variance decomposition of
the log of all these variables. These results are also reported in Table II. What is striking is that,
whatever geographical controls are used, a much higher proportion of wage variation (typically
twice) is inter-firm than for any other variables. So, of all the variables on which we have data,
wages have the smallest proportion of total variance within firms. The finding that there is a lot
of wage variation between employers in a given labour market is one on which there has been a
lot of research (e.g. the older papers of Lester, 1946, Reynolds, 1946, 1951, and Slichter, 1950,
and the more recent Krueger and Summers, 1988), but, as far as we are aware, this has not been
tied to the lack of wage variation within employers3. Our result on how there is extraordinarily
little wage variation within firms is even stronger when we restrict attention to day care
between firms and workers in these cases and it is this difference on which we want to focus.
3
One should note that this is not likely to be a result that is robust to looking at the dispersion in wages across
occupations as the wage gap between managers and care assistants in every firm far exceeds the dispersion in pay in
care assistants across firms. So, for example, the findings of Abowd et al (1999) who consider all occupations and
conclude that individual-specific effects are more important than firm effects is not inconsistent with the findings
reported here.
6
assistants only (the results are presented in the bottom panel of the table).
To reinforce the point that there is surprisingly little wage dispersion within firms we
now present some further information on the structure of wages within firms. These results are
reported in Table III. First, about 25% of workers working in about a third of firms have no
within-firm variation in wages i.e. all Care Assistants receive exactly the same wage. Another
third of firms, employing about a third of workers are in firms where there are only two different
wages. Only one firm pays all its workers different wages. These wage policies seem very
stable. Of the 213 firms in the matched sample, 52 changed from having wage dispersion to not
or vice versa (exactly 26 moving in each direction), but of these 29 involved the change in the
wage of only a single worker and there are only a handful of cases of large changes in wage
structures.
If one restricts attention to larger firms (those with more than 5 workers) one finds that
the proportion of both workers and firms with no wage dispersion falls. Looking at these figures
on the proportion of firms for which there is no wage dispersion one might be tempted to
conclude that there is more wage dispersion in larger firms. But, in some sense, this is the wrong
conclusion as there are simply more opportunities for wage dispersion in larger firms. One
would like to have some way to normalise the measure of wage dispersion by size of firm.
One way of trying to normalise the information on wage dispersion is the following.
Suppose that all existing workers in the firm are paid the same wage and, conditional on this fact,
the probability that an extra worker is paid the same wage is ρ. Then, the probability that a firm
with N workers will have no wage dispersion is given by ρN-1. We used the information on the
existence of wage dispersion or not to estimate ρ4. The results are reported in the last row of
4
Note that this is not using all the information about the number of different wages paid in the firm, nor is it
7
Table III. For all firms the spot estimate of ρ is 0.76 so that, given that all existing workers are
paid the same wage, the probability that an additional worker will be paid that wage is 76%.
Once we restrict attention to firms with more than five workers, the estimate of ρ rises to 0.83, a
difference that is statistically significant. So, there is a sense in which there is less wage
The notion of wage dispersion that we have considered so far is a very strong one. If a
firm pays even one hour of labour at a wage different from the rest of its hours, that firm will be
classed as having wage dispersion. We would like to have a measure that is less sensitive to this
kind of effect. Table III therefore also presents a number of alternative measures of wage
dispersion. We report the proportion of total hours worked by care assistants that are paid the
modal hourly wage. As one can see, about 75% of hours are paid the modal rate, a proportion
that seems extremely high. However, this statistic tells us little about the extent of the variation
in wages, so we also present data on the hours weighted standard deviation of log hourly wages.
So far, we have documented that there is surprisingly little wage dispersion within firms,
but that there is considerable heterogeneity across firms. Our initial reaction to looking at the
data is to think that it is a long way from the ‘law of one wage’ predicted by competitive labour
markets: there seems to be ‘too much’ wage variation across firms and ‘too little’ within them.
But, we have not presented a formal test of the hypothesis that the observed distribution of wages
is the outcome of a competitive labour market i.e. one in which all workers are paid their
wage dispersion must all have the same marginal product. In some ways this is surprising as
there is variation within these firms in observable characteristics that we might expect to be
related to worker quality. In our data set the available ‘quality’ variables are age, sex, tenure and
qualifications. Table IV presents descriptive statistics on these variables at both individual and
firm level both for the whole sample and for the sample divided according to whether the firm
has any wage dispersion or not. If the competitive model was correct we might expect to see less
variation in observable characteristics within firms with no dispersion, but as can be seen from
the lower panel of Table IV there is very little evidence for this.
However, this does not clinch the case against the competitive model as our measures of
worker quality are inevitably imperfect and it seems likely that there is an important component
of worker quality that is observable to employers but not to us. In general this is an intractable
problem but we can hope to make some progress if we are prepared to assume that we have a
second measure of worker quality. In the particular market considered here, it is natural to
consider that the advantage of higher quality workers is that, other things equal, a higher quality
of care can be provided and the price charged to residents can also be higher. So, we propose to
Let us denote the quality of a worker by q and assume that q can be written as:
q = β x+ε (1)
where βx represents the effect of observable characteristics and ε the effect of unobservable
the competitive model is to be able to explain the lack of wage dispersion in some firms then it
5
As we only have price information at firm level, this means that we can only see if the variation in worker
characteristics across firms is associated with variation in prices: we cannot examine whether wage dispersion in
9
cannot be. If the labour market is competitive then w=q where w is the measured wage (one
could also allow measurement error in this). When one runs a regression of the wage on the
E ( w x) = β x + E (ε x) = β * x (2)
Table V presents estimates of earnings equations both at the individual level and the firm level,
for the whole sample and dividing the sample into those firms with wage dispersion and those
firms without it. We also include the log of patients per worker hour as a measure of the
intensity of worker effort, the log of the number of residents as a measure of the size of the home
and a dummy variable for whether the home is part of a larger organisation. For the whole
sample, the estimated wage equation is very familiar: wages are a concave function of age,
increasing in job tenure, higher for qualified workers and higher in larger firms. There is no
premium for male workers but there are very few men in the sample. Large firms and homes
with high numbers of patients per worker hour are found to also pay higher wages. When the
sample is restricted to firms with wage dispersion, one finds similar results. But, when one
estimates a wage equation for workers in firms without wage dispersion one finds that tenure and
qualifications are no longer significant6, and that age, while significant, has only about a sixth of
the effect on wages of the effect in the firms with wage dispersion. One should think of the return
to age in this sub-sample as implying that firms that tend to pay higher wages tend to end up with
workers of a particular age. Only the effects of the firm-level variables seem to be the same (or
Furthermore, these differences between firms with and without wage dispersion are
significant: a Chow test for the equality of the coefficients in the two regressions leads to
F(25,4407)=11.57 for the individual equations and F(25,543)=2.20 for the firm-level equations,
both of which are convincing rejections of the null hypothesis. If the competitive model is
correct, these findings imply that the correlation of observed characteristics with unobserved
characteristics must be different in the two segments i.e. that, if we define a binary variable,
DISP, that takes the value 1 if there is wage dispersion and zero if there is not, then
E ( ε x, DISP = 1) ≠ E ( ε DISP = 0 ) .
Now let us turn to an analysis of the determinants of the price. Suppose that the
p = γ 0z + γ 1q + v (3)
where z is a vector of observed factors that affect price apart from worker quality (some or all of
which may be contained in x) and v is unobserved factors that affect the quality of care. Now,
E ( p x, z ) = γ 0 z + γ 1 E ( q x, z ) + E ( v x, z ) = γ 0 z + γ 1 β * x + E ( v x, z ) (4)
The basis of the test is that if we estimated price equations across the segments of the market for
which we have shown from the wage equation that the correlation of quality with characteristics
(i.e. β*) is different, we would also expect to find corresponding differences in the price
equations.
The results of this exercise are reported in Table VI. The first column estimates a price
equation for the whole sample and the next two columns divide the sample into the firms with
wage dispersion and those without. In this sector, the Department of Social Security pays a
11
subsidy for the care of many residents up to a maximum of £175 per week.7 One consequence of
this is that there is a spike in the price distribution at this price and very few homes charging
lower prices. Accordingly, we treated the price equation as an equation for the desired price and
then estimated a tobit model with £175 as the lower censoring point. Looking at the results for
the whole sample (column 1), price does seem to be significantly related to the log of patients per
worker hour (a measure of quality of care), whether the home is part of a larger organisation, the
size of the home and (among the worker characteristics) the average age. This is consistent with
a casual reading of job advertisements in this sector, which emphasize that employers prefer
older workers.
The second and third columns estimate separate price equations for those firms with and
without wage dispersion: what is striking is that the coefficients (on age in particular) are very
similar in the two sub-samples. A formal test cannot reject the hypothesis of equality of
coefficients with a likelihood ratio test yielding χ2(25)=19.26 (the critical value at the 5% level is
38). This is inconsistent with the competitive model which, given the evidence on wages, would
predict that there should be significant differences between the two segments. So, we conclude
that unobservable worker quality cannot reconcile the observed wage data with the perfectly
competitive model, and that the evidence from the price equations suggests that the correlation of
unobserved with observed worker quality would seem to be very similar in homes with and
By comparison of the wage equations in Table V and the price equations in Table VI one
can also see that the characteristics of workers that are associated with higher wages are not
7
The actual system of subsidy is more complicated than this as it involves means-testing, but it is this upper bound
on payments that seems to have the most effect on the market.
8
The one possible case where this will not work is where E(vz,x) also differs across the segments in a way that
12
necessarily associated with higher prices. In particular, job tenure is associated with
significantly higher wages but significantly lower prices. This is consistent with empirical
findings like those reported in Medoff and Abraham (1980, 1981) and Klein, Spady and Weiss
(1991) that the worker characteristics associated with higher wages are not necessarily associated
with higher worker quality. One potential way of explaining these results while retaining a
competitive view of the labour market would be to appeal to the existence of specific human
capital. For then, the wage paid is determined by productivity in the next-best alternative and
there is no reason why worker characteristics should not affect worker quality in this firm
independent of the wage paid. The main reason why we do not find this explanation of our
findings plausible is that the traditional measure used of the extent of firm-specific human capital
is job tenure. The Becker (1975) argument is that workers capture some but not all of the returns
to specific human capital so that wages rise with job tenure (as shown in the All workers
columns of Table V) but not as fast as productivity. However the estimates in Table VI shows
In this section, we have explored further whether the structure of wages we observe is
consistent with the competitive model. We have argued that it is not, and that traditional ‘get-
outs’ like unobserved labour quality are implausible explanations of what is observed in the data.
Given the results reported above, we do not think the competitive model is particularly
helpful for understanding the particular labour market analysed here. The ‘law of one wage’ in
which there is a given market wage for each quality of worker does not seem to hold. But, is
exactly off-sets the worker quality effect: this seems a thin straw at which to clutch.
13
there a more successful alternative? Any theory which successfully explains our data must
explain why there is so little wage dispersion within firms, why the wage dispersion that does
exist within firms does not seem to be closely related to productivity, and why apparently
identical firms seem to have different wage policies that are stable over time.
In this section, we consider a number of alternative models of the wage structure that
have been proposed, namely rent-sharing models, monopsony models, incentive models, fairness
models and implicit contract models. Not all of them are non-competitive (the implicit contract
models and some fairness models) in that there may be a single wage for labour of a given type
but they all imply some deviation from a competitive spot market. There is considerable overlap
among the ideas behind many of these theories so one should not think of them as necessarily
mutually exclusive.
Rent-Sharing Models
These models have become popular in recent years as a way to explain the dispersion in
wages between firms (see, for example, Christofides and Oswald, 1992). The argument is that
there are quasi-rents in all employment relationships and that workers (whether in unions or not)
have the ability to extract a share of these rents. As there is likely to be heterogeneity in quasi-
rents across firms, the result will be heterogeneity in wages. This could obviously explain the
inter-firm wage variability but does not seem persuasive as an explanation of the structure of
The reason is the following. There are no trade unions in any of the firms in our sample
so that any bargaining that does occur must be at an individual level. But, given the
considerable wage dispersion within firms that is not in the data. It is simply not credible to
14
think of wages in the firms with no wage dispersion as being negotiated individually with each
worker: it seems beyond reasonable doubt that the single wage paid is determined unilaterally by
the firm. However, it is quite possible that in some circumstances in some firms, a valued
worker gets a raise when they threaten to leave so some of the wages we observe are probably
Incentive Models
There has also been a considerable amount of literature emphasizing how, in the presence
of problems of worker moral hazard or shirking, firms may pay wages that diverge from
marginal products. Examples of this type of theory are Lazear (1981) and Akerlof and Katz
(1989). In this type of model, upward-sloping wage profiles are offered to workers because this
provides incentives for workers to put in effort and not to lose their jobs.
A fundamental problem with using this type of model to explain our data is that they have
generally been used to explain why there might be more variation in wages within firms than in
marginal products, which seems to be the case in, for example, the firms considered by Medoff
and Abraham (1980,1981). But, in our data set there is the opposite problem: employers who
have workers who we presume to differ in productivity choose to pay all their workers the same
wage. These ideas cannot begin to explain this important feature of our data set.
One obvious candidate for explaining why there is so little wage dispersion within firms
is that workers dislike wage dispersion and believe that all workers doing the same job should be
paid the same wage even though some of them may do the job more effectively than others. This
type of theory has a long tradition and has recently been suggested by a number of authors (e.g.
Akerlof and Yellen, 1990, or Frank, 1984) and the usefulness of this type of theory in explaining
15
labour market outcomes has been suggested by Bewley (1999). This type of idea is generally
these considerations are important. It seems plausible that this is an important factor behind the
Another type of economic theory that might be used to explain the lack of wage
dispersion within firms is implicit contract theory (see Rosen, 1986, for a survey). The basic
idea is that risk-averse workers are unable to insure against various employment risks in the
insurance market and buy insurance from their employers who are generally assumed to be risk-
averse. In this case the ex ante labour market is competitive but observed wages and
employment will not generally be the equilibrium of a spot market. The model was generally
used to explain the lack of wage variation over time but could also conceivably be used to
explain the lack of variation in wages across workers who are being insured against variation in
their productivity (see Harris and Holmstrom, 1982, for a more formal model of this type).
There are a number of reasons why we are sceptical about the relevance of this sort of
model to the labour markets we are considering here. First, there is no explicit wage contract
guaranteeing insurance so any insurance contract must be implicit and enforced on the side of the
firm by reputation effects. Yet, these are small firms for whom we would not expect reputation
effects to be very important. And, average job tenure in this sector is only something like three
years so that there is a limited amount of insurance that firms can offer. Secondly, workers
should only be able to purchase insurance against variation in productivity that is ex ante
unobservable. Yet the estimated wage equations of Table V suggest that workers in firms with
no wage dispersion also manage to obtain insurance against their age, which should not be
16
possible. Thirdly, the fact that owner-managed firms are very common in this sector means that
Our basic problem with this story is that we do not feel that implicit contract models are
the right way to think about the structure of wage policies in this type of labour market where the
Monopsony
There are quite a number of labour market models that have been designed to explain the
existence of equilibrium wage dispersion between firms (e.g. Albrecht and Axell, 1984; Burdett
and Mortensen, 1998; Lang, 1991; Montgomery, 1991). All of these models make some
assumption that the labour supply curve facing a firm is not perfectly elastic so that they have
some feature of monopsony (see Manning, 2002, for more extensive discussion of monopsony in
labour markets). In these models high and low wage firms can coexist in equilibrium because
high wage firms have larger workforces in equilibrium or lower turnover costs. This prediction
finds support in our data in the sense that there is a robust positive correlation between wages
and firm size. But, in all of these models, it is an assumption that all workers within firms are
In fact, there are good reasons why we would expect to see the emergence of wage
dispersion within firms in this sort of model. The reason is that appropriately chosen wage
dispersion can increase profits, essentially because it allows firms to act as a discriminating
This might be done through a formal structure e.g. an explicit wage scale relating wages
to tenure which will tend to bind workers to the firm (see Ioannides and Pissarides, 1983, for an
example of this form of argument). Or it could be done more informally by paying low wages to
17
those with bad outside opportunities, and by raising the wages of workers who receive outside
offers. For example, in the Burdett and Mortensen (1998) model workers are paid wages that
are below marginal products and leave when they receive a better wage offer. There are obvious
incentives for the firm to pay a higher wage to a worker who has just received an outside wage
offer and is about to quit. However, the structure of wages within the firm is determined not just
by productivity so this kind of model can explain why the wage dispersion that does exist is often
unrelated to productivity.
The problem that remains is to explain why it is that there is so little wage dispersion
within firms. One possible line of explanation for this is suggested by thinking about the
consequences of a firm adopting a strategy of matching outside wage offers. Workers in a firm
that adopts this strategy have an obvious incentive to generate or even invent outside wage
offers. It is likely that this means that a firm that responds to outside wage offers will find that
their workers have more outside wage offers than does a firm that does not vary wages and hence
average wages will be higher. Hence, while wage dispersion offers the possibility of reducing
turnover of valued workers, it also will tend to raise average wages. A similar sort of idea is
behind Ellingsen and Rosen (1997). In that model firms have a choice of paying a fixed wage to
all workers or to negotiating wages with individual workers. The disadvantage of the first
strategy is that some workers will not work for the firm (because the wage is too low) even
though it would be efficient for them to do so (i.e. at a higher wage workers would want the job
and they would still be profitable). On the other hand, if wages are negotiated individually,
wages are higher on average but all efficient matches are consummated. Ellingsen and Rosen
present a model in which both strategies co-exist in equilibrium (i.e. offer the same level of
18
5. Conclusions
Most microeconomic data sets do not have sufficient detail to permit a close examination
of the structure of wages in a specific labour market. In this paper we have used data on a
reasonably large sample of workers and firms in a very particular labour market which, given its
structure, we feel most economists would expect to be competitive. But, when one looks at the
data it is hard to avoid the conclusion that there are very serious limitations to the usefulness of
the competitive model in explaining the data. In particular we feel that the competitive model
cannot explain one of the most striking features of our data, the very small amount of wage
dispersion there is within firms and the correspondingly large amount between firms. And, what
wage dispersion there is does not seem to be closely related to the characteristics of workers that
We think it is more helpful to think of firms in the labour market as having considerable
discretion in the setting of wages, a discretion that has its roots in labour market frictions. They
seem to use this discretion to have very simple wage structures, probably because of worker
dislike of wage variation among workers doing the same job and reluctance on the part of
19
Table I:
Description of the Structure of Hourly Wages for Care Assistants in
Residential Nursing Homes, 1992 and 1993
Number of individuals 3221 2514 1826 1463 1571 1213 1647 1318
Number of firms 434 246 230 141 213 121 213 124
Average wage 2.96 2.98 3.06 3.07 2.97 2.99 3.07 3.08
Standard deviation of log .16 .16 .16 .16 .16 .16 .16 .16
hourly wages
10th percentile wage 2.45 2.50 2.50 2.50 2.45 2.50 2.50 2.50
25th percentile wage 2.70 2.70 2.75 2.75 2.70 2.70 2.75 2.75
50th percentile wage 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00
75th percentile wage 3.20 3.25 3.30 3.30 3.20 3.25 3.30 3.34
90th percentile wage 3.50 3.50 3.60 3.60 3.50 3.55 3.60 3.60
Number of individuals 2878 2246 1603 1271 1363 1057 1441 1154
Number of firms 434 246 235 141 212 121 210 124
Average Wage 2.97 2.99 3.07 3.09 2.98 2.99 3.07 3.09
Standard deviation of log .15 .15 .15 .15 .15 .16 .15 .15
hourly wages
10th percentile wage 2.50 2.50 2.50 2.55 2.50 2.50 2.50 2.50
25th percentile wage 2.70 2.73 2.77 2.77 2.70 2.70 2.75 2.75
50th percentile wage 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00
75th percentile wage 3.20 3.25 3.25 3.30 3.20 3.20 3.25 3.30
90th percentile wage 3.50 3.50 3.60 3.60 3.50 3.55 3.60 3.60
Notes.
1. Wages are hourly rates defined in pounds per hour.
20
Table II:
Proportion of Dispersion that is Inter-Firm
21
Table III
Measures of Intra-Firm Wage Dispersion
Notes.
1. The final five rows of this table are means across firms. One could of course present the means across individuals but they are
very similar so are not reported here.
22
Table IV:
Descriptive Statistics
1992 1993
All Firms with no Firms with All Firms with no Firms with
wage wage dispersion wage wage dispersion
dispersion (DISP > 0) dispersion (DISP > 0)
(DISP = 0) (DISP = 0)
Individual-level
Number of workers 3221 827 2394 1826 452 1374
Hourly wage 2.96 (.49) 2.97 (.42) 2.96 (.51) 3.06 (.50) 3.05 (.42) 3.07 (.52)
Age 36.5 (14.1) 38.7 (13.3) 35.7 (14.2) 36.9 (14.0) 38.6 (13.2) 36.4 (14.3)
Tenure 2.5 (2.7) 2.5 (2.5) 2.6 (2.7) 2.7 (2.6) 2.6 (2.4) 2.7 (2.7)
Proportion male .03 .02 .03 .03 .01 .04
Proportion with nursing .05 .03 .05 .05 .05 .05
qualification
DISP .74 .00 1.00 .75 .00 1.00
PROPMOD .73 1.00 .64 .74 1.00 .65
Number of workers 10.1 (5.7) 8.4 (4.3) 10.7 (6.1) 10.1 (5.1) 7.8 (3.7) 10.9 (5.2)
Number of residents 17.1 (9.6) 13.8 (6.4) 18.2 (10.2) 17.1 (9.2) 12.8 (5.7) 18.6 (9.6)
Patients per worker hour .091 (.050) .075 (.070) .089 (.042) .086 (.045) .089 (.044) .086 (.046)
Part of larger organisation .077 .075 .078 .056 .091 .045
Price of bed 195 (29) 194 (33) 195 (28) 208 (34) 204 (34) 209 (33)
Firm-level
Number of firms 432 135 297 231 74 157
Average hourly wage 2.97 (.39) 2.95 (.40) 2.97 (.39) 3.09 (.42) 3.00 (.39) 3.13 (.43)
Within firm standard .026 (.039) .00 .039 (.042) .024 (.036) .00 .036 (.039)
deviation of log hourly wages
Average age 37.4 (7.8) 39.6 (8.2) 36.4 (7.4) 37.2 (7.3) 38.5 (8.2) 36.6 (6.8)
Within firm standard 4.9 (2.8) 5.0 (3.1) 4.8 (2.7) 4.8 (2.8) 4.7 (3.0) 4.8 (2.8)
deviation of age
Average tenure 2.6 (1.6) 2.5 (1.7) 2.6 (1.6) 2.7 (1.5) 2.7 (1.6) 2.8 (1.4)
Within firm standard .76 (.79) .74 (.85) .77 (.76) .76 (.68) .69 (.58) .79 (.72)
deviation of tenure
Average proportion male .03 .03 .03 .03 .02 .03
Average proportion nursing .06 .04 .06 .07 .06 .07
qualification
DISP .68 .00 1.00 .68 .00 1.00
PROPMOD .76 1.00 .65 .76 1.00 .64
Number of workers 7.1 (4.4) 6.0 (3.8) 7.6 (4.5) 7.5 (4.3) 6.0 (3.3) 8.2 (4.5)
Number of residents 13.8 (7.3) 11.6 (5.5) 14.7 (7.8) 14.2 (7.9) 11.0 (5.5) 15.7 (8.5)
Patients per worker hour .111 (.070) .131 (.097) .102 (.051) .103 (.060) .104 (.053) .102 (.062)
Part of larger organisation .049 .045 .050 .047 .067 .038
Price of bed 190 (25) 188 (28) 191 (24) 201 (29) 197 (28) 204 (30)
Notes.
1. Means are presented with standard deviations in parentheses.
23
Table V
Estimated Wage Equations
Notes.
1. Standard errors reported in parentheses. All regressions also include a year dummy and 16 area dummies.
2. The firm-level wage equations are weighted by the number of residents for comparison with the price equations in the next
Table.
24
Table VI
Estimated Price Equations
Notes.
1. Standard errors are reported in parentheses. All regressions also include a year dummy and 16 area dummies.
2. The price equations have a tobit specification with a lower cut-off at ln(175).
25
References
Abowd, J., F. Kramarz and D. Margolis (1999) “High Wage Workers and High Wage
Firms”, Econometrica, 67, 251-333.
Abraham, K. and H. Farber (1987) “Job Duration, Seniority, and Earnings.” American
Economic Review , 77, 278-97.
Akerlof, G. and L. Katz (1989) “Workers' Trust Funds and the Logic of Wage Profiles”,
Quarterly Journal of Economics, 104, 525-36.
Akerlof, G. and J.Yellen (1990) "The Fair-Wage Effort Hypothesis and Unemployment",
Quarterly Journal of Economics, 105, 255-283.
Altonji, J. and R. Shakotko (1987). “Do Wages Rise with Job Seniority?” Review of
Economic Studies, 54, 437-59.
Baker, G., M. Gibbs and B. Holmstrom (1994a) “The Internal Economics of the Firm:
Evidence from Personnel Data”, Quarterly Journal of Economics, 109, 881-919.
Baker, G., M. Gibbs and B. Holmstrom (1994b) “The Wage Policy of a Firm”, Quarterly
Journal of Economics; 109(4), 921-55.
Becker, G. (1975) Human Capital, 2nd edition, Chicago: University of Chicago Press.
Bewley, T. (1999) Why Wages Don't Fall During a Recession, Cambridge Mass.:
Harvard University Press.
Frank, R. (1984) "Are Workers Paid Their Marginal Product", American Economic
Review, 74, 549-571.
26
Harris, M. and B. Holmstrom (1982) “A Theory of Wage Dynamics”, Review of
Economic Studies. 49, 315-33.
Ioannides, Y and C. Pissarides (1985) “Monopsony and the Lifetime Relation Between
Wages and Productivity”, Journal of Labor Economics, 3, 91-100.
Klein, R., R. Spady and A. Weiss (1991) “Factors Affecting the Output and Quit
Propensities of Production Workers”, Review of Economic Studies,58, 929-53.
Krueger, A. and L. Summers (1988) “Efficiency Wages and the Inter-Industry Wage
Structure”, Econometrica, 56, 259–293.
Lester, R. (1946) “Wage Diversity and Its Theoretical Implications,” The Review of
Economic Statistics, 28, 152–159.
Machin, S., A. Manning and L. Rahman (2000) “Care Home Workers and the
Introduction of the UK National Minimum Wage”, unpublished, CEP, LSE.
Machin, S., A. Manning and S. Woodland (1993) Are workers paid their marginal
product? Evidence from a low wage labour market, Discussion Paper No.93-09,
University College London.
Manning, A. (1994) “Labour Markets with Company Wage Policies”, CEP Discussion
Paper.
Marshall, R. and G. Zarkin (1987). “The Effect of Job Tenure on Wage Offers.” Journal
of Labor Economics, 5, 301-24.
Medoff, J. and K. Abraham (1981) “Are Those Paid More Really More Productive? The
Case of Experience”, Journal of Human Resources, 16, 186-216.
27
Reynolds, L. (1946) “The Supply of Labor to the Firm”, Quarterly Journal of
Economics, 60, 390–411.
Reynolds, L. (1951) The Structure of Labor Markets; Wages and Labor Mobility in
Theory and Practice, Westport, Conn.: Greenwood Press.
Slichter, S. (1950) “Notes on the Structure of Wages” The Review of Economics and
Statistics, 32, 80–91.
Topel, R. (1991). “Specific Capital, Mobility, and Wages: Wages Rise with Job
Seniority”, Journal of Political Economy; 99, 145-76.
Woodland, S. (1993) “Who Cares for the Workers?”, LSE CEP Working Paper, No. 426.
28