Performance Pay and Ethnic Earnings Differences in Britain Colin P. Green Department of Economics, Lancaster University John S. Heywood Department of Economics, University of Wisconsin-Milwaukee Nikolaos Theodoropoulos Department of Economics, University of Cyprus Abstract In the first study using British data, we show that the ethnic earnings gap among performance pay jobs is smaller on average than that among time rate jobs. Moreover, quantile estimates and decompositions show that this narrowing is driven by bonus payments in the upper middle portion of the earnings distribution. The results partially reflect sorting in which the more able take performance pay jobs but the results persist with diminished magnitude in fixed effect estimates. These findings differ dramatically from those for the United States in which performance pay has been associated with larger negative racial differentials especially at the top of the earnings distribution. KEYWORDS: Performance Pay; Ethnic Earnings Differentials JEL CODES: J33, J71 Corresponding Author: Prof. J.S. Heywood, Department of Economics, University of Wisconsin- Milwaukee, Milwaukee, WI 53201 USA; [email protected]: (414)229-4310
Transcript
Performance Pay and Ethnic Earnings Differences in Britain
Colin P. Green
Department of Economics, Lancaster University
John S. Heywood
Department of Economics, University of Wisconsin-Milwaukee
Nikolaos Theodoropoulos
Department of Economics, University of Cyprus
Abstract
In the first study using British data, we show that the ethnic earnings gap among performance pay
jobs is smaller on average than that among time rate jobs. Moreover, quantile estimates and
decompositions show that this narrowing is driven by bonus payments in the upper middle portion
of the earnings distribution. The results partially reflect sorting in which the more able take
performance pay jobs but the results persist with diminished magnitude in fixed effect estimates.
These findings differ dramatically from those for the United States in which performance pay has
been associated with larger negative racial differentials especially at the top of the earnings
Corresponding Author: Prof. J.S. Heywood, Department of Economics, University of Wisconsin-Milwaukee, Milwaukee, WI 53201 USA; [email protected]: (414)229-4310
1
1. INTRODUCTION
Performance pay mitigates inherent agency problems in labour contracts by linking
worker compensation to worker performance. While this link generates greater wage
dispersion within the firm (Lazear 2000), it may also influence the broader earnings
distribution. Lemieux et al. (2009) argue that performance pay provides an important
mechanism for translating differences in skills into differences in earnings. They show that
performance pay stretches out the US earnings distribution. Yet, Heywood and Parent
(2012) demonstrate that this is unique to white workers. Performance pay does not stretch
out the top of the black distribution as it does for the white distribution. As a consequence,
the racial earnings gap is larger in the performance pay sector and especially so at the top
of the earnings distribution.
The recognition that performance pay may influence racial and ethnic earnings
differences has been developed in a series of North American based studies but we
provide the first examination of its influence in Britain. We estimate the relative ethnic
earnings gap in performance pay jobs and in time rate jobs. We draw data from the British
Household Panel Survey (BHPS) which identifies performance pay and provides detailed
information on employees’ personal and parental characteristics as well as on job
characteristics and work conditions. We observe two measures of performance pay:
performance related pay and bonuses. Further, the panel element of the data allows us to
remove time invariant unobserved worker heterogeneity.
The British results differ substantially from those presented for the United States.
First, performance pay (and especially bonuses) is associated with narrower rather than
expanded ethnic average wage differentials. Second, while this association with smaller
differentials varies over the wage distribution, quantile decompositions make clear there is
2
no evidence that performance pay expands differentials at the top of the earnings
distribution. Indeed, the racial earnings gap actually turns modestly positive (in favour of
minorities) at the top of the earnings distribution and the largest contribution by bonuses
to improving the relative position of minorities is in the upper middle portion of the
distribution.
We find evidence of sorting as individual fixed effect estimates are smaller in
magnitude. Nonetheless, bonuses continue to be associated with smaller ethnic earnings
differentials even in the fixed effect estimates. We examine the observables associated
with the sorting and show that it is the more educated and those minority groups typically
associated with labour market success that move into bonus jobs.
There appears to be heterogeneity in the pattern of results. The relatively smaller
differential is associated with bonuses but not with performance related pay. Moreover
the smaller differential associated with bonuses appears somewhat more prominently for
Asians than for Blacks although our sample size limits the precision of estimates. In the
end, we identify a differentiated pattern of results unique to Britain that emphasize the
contrast with the US. We find little evidence that performance pay is driving larger ethnic
earnings differences in Britain or that ethnic minorities select away from performance pay
or that the influence of performance pay is particularly adverse to minorities at the top of
the distribution as suggested in the North American studies.
The structure of the paper is as follows. Section 2 provides background on the
underlying theoretical conjectures, the North American evidence and the British context.
Section 3 presents the data and reports descriptive analysis. Section 4 describes the
estimation methodology and section 5 presents the results. Section 6 summarizes the
findings and suggests future research.
3
2. BACKGROUND
There exist conflicting theories about the influence of performance pay on racial
and ethnic wage differentials. Heywood and O'Halloran (2005) present a model in which
objective based performance pay such as piece rates or commissions provide information
to all workers and potential enforcement officials about the pay and productivity of each
worker. This information makes it more likely that a discriminating employer would be
detected, and face financial penalties. In turn, this reduces the incentive to discriminate
suggesting that performance pay should be associated with reduced measured earnings
differentials. Alternatively, performance pay is often set based on a managerial appraisal in
which there may be large elements of subjectivity (MacLeod 2003). This subjectivity
increases the latitude for managers who set, for example, the bonuses of their
subordinates. As a consequence, the presence of performance pay may facilitate earnings
set in accordance with the objective functions of managers rather than in accordance with
profit maximization (Becker 1971). In their theoretical modeling, Meon and Szafarz (2011)
show that in the face of agency problems between owners and managers, performance
pay will not eliminate discrimination calling the modern corporation "a safe haven for
taste-based discrimination".
In seeming accord with this distinction between formulaic and subjective bonuses,
Heywood and O'Halloran (2005; 2006) use the US National Longitudinal Survey to show
that while measured racial wage differentials are significantly smaller among those paid
objective piece rates than among those paid time rates, they are actually largest among
those paid subjective bonuses. Elvira and Town (2001) confirm that supervisors'
performance appraisals are influenced by the race of their subordinates. A White
4
supervisor typically gives non-White subordinates worse ratings even when holding
constant all other demographic variables and available (to the researcher) objective
measures of productivity. This raises the possibility that equally productive Black and
White workers are paid differently but have supervisory evaluations that support the
differences. Castilla (2008; 2012) also uses US data going further to show not only that race
influences appraisal ratings but that Blacks receive smaller raises than Whites even with
equal ratings. Canadian evidence shows a substantial ethnic earnings gap among those on
time rates which include bonuses but no evidence of an ethnic earnings gap among those
receiving formulaic performance pay: tips, commissions or piece rates (Fang and Heywood
2006).
While there may be a difference between subjective and formulaic performance
pay, this difference may not always determine the extent of discrimination. Bureaucratic
rules or reputation concerns may limit discrimination in subjective schemes (Prendergast
and Topel 1993) and objective schemes may be manipulated to reflect managerial
preferences. Thus, Madden (2008) uses data from two large US financial service firms to
show that the size of stockbrokers' formulaic bonuses depends critically on
"complementary inputs" provided by managers (such as the quality of the accounts
managers distribute to brokers). She shows that differences in these complementary
inputs, rather than differences in success with otherwise equal inputs, explain the lower
earnings of female stock brokers who have male managers.
Heywood and Parent (2012) return to the issue of the influence of performance pay
on racial wage differences with special attention paid to the distributional aspects. They
use the US Panel Study of Income Dynamics and show that in the time rate sector the
Black-White earnings differential shrinks when moving up the earnings distribution. On
5
the other hand, in the performance pay sector the Black-White earnings differential grows
when moving up the earnings distribution and explodes at the very top of the distribution.
The authors highlight two potential explanations. Unmeasured ability at the top of the
performance pay distribution may be greater for whites than for blacks and if measured
might explain the large differential. Alternatively, the top of the performance pay
distribution may be dominated by subjective evaluation schemes increasing the latitude
for discrimination. While unable to distinguish between these explanations, the evidence
of sorting was clear with high ability blacks tending to sort out of performance pay
schemes as high ability whites tended to sort into performance pay schemes.
Despite the policy importance of earnings differentials in many countries and the
growth of performance pay in the labour markets of many countries, the vast majority of
the empirical literature stems from North America. 1 In particular, we know of no
corresponding study for Britain. Yet, such a study seems called for.
Both British academics and policy makers remain concerned about the relative
labor market position of ethnic minorities and the causes of that relative position (see the
report of the National Equality Panel, 2010). One measure of this relative position is the
ethnic earnings differential which we investigate. It is widely accepted that ethnic earnings
differentials have declined far less than gender differentials over the period since the equal
opportunities legislation were introduced in Britain starting with the Race Relations Act of
1965. Indeed, ethnic minorities have become increasingly concentrated in the lower
percentiles of the pay distribution (Bell 1997; Blackaby et al. 1994; 2002). Numerous
studies have also found evidence of more limited employment prospects and wages (Clark
and Drinkwater, 2009; Dustmann and Theodoropoulos, 2010), fewer training and
promotion opportunities (Pudney and Shields, 2000) and lower occupational attainment
6
(Stewart, 1983) for ethnic minority groups compared to similar white natives. Specifically,
(Blackaby et al. 1998; 2002) estimate that non-white ethnic males earned 11 percent less
in the mid 1990s and that differences in characteristics explain very little of this gap. They
also show great heterogeneity with the gap largest for Pakistanis and lowest for Indians.
This wage disadvantage exists even for British born ethnic minorities (Blackaby et al. 2005).
Dustmann and Theodoropoulos (2010) use more recent data suggesting that ethnic
minority men have better observed characteristics than white men, and that once those
differences are taken into account the wage differential becomes larger. Yet, their
corrected differential is of roughly the same size as the earlier estimates, approximately 9
percent. At issue, for us, is the role performance pay plays in generating the British pattern
of ethnic wage differentials.
Focusing on the influence of performance pay on ethnic wage differentials is timely
as such pay has taken center stage both among academics in the economics of personnel
and in the minds of firm managers.2 Bloom and Van Reenen (2010 p. 4) show a substantial
increase in the incidence of performance pay in the UK. The incidence increased from 41
percent of firms using performance pay in 1984 to 55 percent of firms by 2004. While this
increase was largely among the private sector, UK government policy has increasingly
called for greater pay for performance within the public sector (for a review see Graham et
al. 2007). The increased attention paid to performance pay by both academics and
policymakers is driven by the view that it has the potential to make earnings more closely
correspond with productivity. Yet, the consequences of greater use of performance pay on
ethnic labor market outcomes have not been given sufficient consideration. We now turn
our effort to providing this consideration.
7
3. DATA AND DESCRIPTIVE STATISTICS
We draw data from the BHPS which has been frequently used for exploring
performance pay issues (see for instance Booth and Frank, 1999; Green and Heywood,
2008). The BHPS surveys households annually and is representative of the British
population. From 1991 to 1998, 5,500 households were interviewed providing information
on 10,000 adults. From 1999 15,000 individuals participate on average in the survey. The
panel is unbalanced as there is entry and exit.
The key advantage of the BHPS for our purposes is that it identifies performance
pay, provides a rich set of controls and tracks individuals over multiple time periods. Thus,
while the Labour Force Survey includes many more ethnic minorities, it has low response
rates on its performance pay questions (see Green and Heywood 2011), doesn't have the
depth of controls such as indicators of parental education and non-cognitive human capital
and it interviews individuals regarding their wages at most twice (potentially limiting the
reliability of fixed-effect estimates). We use eleven BHPS waves from 1998 to 2008.3
Earlier waves fail to identify performance related pay but from 1998 two questions are
asked regarding payment type, “does your pay include performance related pay?” and “in
the last 12 months have you received bonuses or profit sharing bonuses?” These generate
two indicators, performance related pay and bonuses.
The data does not separately distinguish those bonuses associated with profit
sharing. Such bonuses might influence ethnic differentials if minorities are more or less
likely to be in firms providing profit sharing but seem less likely to influence ethnic
differentials within firms.4 Broader evidence for the UK suggests that profit sharing receipt
is less than 1/5th as common among workers as are non-profit sharing bonuses and that
many of the latter are based on individual performance (Green and Heywood 2011). Thus,
8
it seems incorrect to classify one type of payment method as based on group or firm level
performance and the other as based on individual performance. The distinction, instead,
may be the number of factors that go into determining the payment. Performance related
pay may tend to be more formulaic, like piece rates, while bonuses may tend to reflect a
wider variety of both objective and subjective measures of performance. As we will show,
bonuses do tend to be more concentrated at the top of the earnings distribution where
jobs reflect more varied tasks. Nonetheless, the indicators provide no details on the
nature of the payment scheme or on the share of a worker's earnings they comprise. In the
end, we simply view both payment schemes as broadly indicative of attempting to more
closely link earnings and performance and are interested in how their use influences ethnic
wage differentials. More detailed information on scheme characteristics might allow a
more nuanced view but is unavailable in the BHPS. We initially combine the indicators into
a single category (they are not mutually exclusive) and then break them out separately in
our examination.
We limit our sample to all male workers aged 16 to 65 who are British private
sector employees.5 To further increase homogeneity and make comparisons of wages
more clear cut, we exclude the self-employed. Indeed, the measures of performance pay
are essentially defined only for the employed. This yields a total of 25,841 annual
observations across the eight waves. While the self-employed might be seen as receiving
the ultimate form of performance pay, including them moves away from our focus on the
employment relationship and concerns with labour market discrimination.6 Moreover, the
earnings of the self-employed are difficult to reconcile with those of the employed as much
of their earnings appear not as labour income but as retained profits.
9
The BHPS contains relatively detailed information on ethnicity. Individuals provide
responses including “black” ethnicities (i.e. black-Caribbean, black-African) and “Asian”
ethnicities (Indian, Pakistani, Bangladeshi and Chinese) and other non-white ethnicities. In
our primary analysis we contrast workers who identify as white with all those who identify
as non-white. 7 We appreciate that this categorisation is broad and contains much
heterogeneity (see for instance Longhi et al. 2013 and Blackaby et al. 2002) but it is driven
by the sample size and the desire to gain more precise estimates. As a consequence, our
primary estimates represent aggregate ethnic differences. Nonetheless, we do provide
more disaggregated robustness checks. Moreover, in recognition of the relatively small
number of minority observations, our decomposition methodology reweights the
characteristics of whites to match those of ethnic minorities as described in the next
section.
The BHPS contains a variety of wage and income variables. In order to best match
with the bonus question we use the measure of annual labour income. The alternative of
using the earnings in the last pay period likely excludes low frequency performance pay
receipt. We use the log of annual earnings as the dependent variable and control for total
hours worked which is the product of usual weekly hours (including overtime) and the
weeks worked in the year. Annual earnings are deflated to a 2005 base year using the CPI.
INSERT TABLE 1
Table 1 presents descriptive information on selected variables for our sample. It
shows that a higher percentage of ethnic minorities are in jobs offering performance
related pay compared to whites. Yet, essentially the same percentage of whites and ethnic
minority employees are in bonus jobs. Ethnic minorities in our sample have substantially
higher levels of education than whites. This higher educational achievement of ethnic
10
minorities fits with results in Dustmann and Theodoropoulos (2010) and Dustmann et al.
(2011). The sample has a smaller proportion of non-white observations (3%), compared to
for instance Dustmann et al. 2011 (Figure 15.1) who report that 4-5% of the working age
population over the 1998-2008 period are from ethnic minorities. This may reflect our
focus on private sector employees (i.e. not public or self-employed) and the oversampling
of Scotland and Wales in the BHPS. In our single equation estimates we use sample weights
as robustness checks and in the subsequent decompositions we reweight our white worker
distribution to match that of non-whites (more details further). We find no evidence that
our pattern of results is driven by the under-sampling of minorities in the BHPS.
INSERT TABLE 2
Table 2 presents the prevalence of the two payment methods across the earnings
distribution separately for Whites and ethnic minorities. Bonuses are more common than
performance related pay. The incidence of bonuses increases when moving up the
earnings distribution for both Whites and minorities. The increase is steeper for minorities
with the incidence for minorities starting well below that of Whites but ending above that
of Whites at the top of the distribution. As the incidence of performance pay differs
substantially across the distribution, it makes sense to explore its influence on earnings
across the distribution as well.
INSERT TABLE 3
Table 3 presents mean earnings separately by payment method and ethnic group.
Among those on time rates, ethnic minorities earn significantly less. This difference of
around 18 percent does not persist among those on performance pay. Among those in
either category of performance pay, there exists a small and insignificant difference in
earnings. Among those receiving performance related pay, the earnings differential
11
remains insignificant. Among those receiving bonuses, ethnic minorities earn substantially
more. Seen another way, the difference in earnings associated with bonus pay is greater
for ethnic minorities than for whites. While whites receiving bonuses earn .356 log points
more than whites receiving time rates, minorities receiving bonuses earn .663 log points
more than minorities receiving time rates. These differences do not hold constant earnings
determinants or unmeasured worker characteristics and we next describe estimates
designed to do so.
4. ESTIMATION METHODOLOGY
The initial results will be single equation log-linear estimates using typical earnings
determinants including the performance pay variables. These focus on the difference in
ethnic differentials using interactions with the payment variables:
(1)
Here the time element has been suppressed but accounted for in the estimation by year
dummies and standard errors that are clustered by individual. In these pooled estimates
the coefficient indicates any differential return to performance pay, PP, by ethnic status,
E. These pooled estimates will be supplemented by otherwise similar single equation fixed
effect estimates that cause E to drop out as invariant but allow estimation of and
from the within worker variation in performance pay. Yet, much of our interest is how the
racial differential may be influenced by performance pay at different parts of the earnings
distribution. The concern is that estimates at the average may hide a different influence at
the top of the distribution as found in the US. Thus, we present single equation (ethnic and
white workers together) quantile estimates that mimic the interaction specification above.
12
These preliminaries set up an examination that allows for separate estimated
equations for ethnic minorities and for whites. The role of earnings determinants is well
known to differ across the two groups (see Dustmann and Theodoropoulos 2010 among
others) and so relying on single equation methods likely biases the coefficients on the
simple interaction of ethnicity and performance pay in (1). The obvious difficulty comes in
combining the desire to estimate separate equations by ethnic status with isolating the
influence of performance pay on the ethnic earnings differential at different points in the
earnings distribution. This is easy at the point of means as a typical Blinder-Oaxaca
decomposition divides the influence of each observable (including performance pay) into a
composition and wage structure effect. Thus, at the point of means one can estimate what
share of the unexplained ethnic earnings difference is driven by unexplained differences in
returns to performance pay. Until recently, this could not be done at points away from the
means as the law of iterated expectations does not apply. We make use of recent
advances to provide the desired evidence.
Firpo et al. (2009) provide a method for computing unconditional quantiles and
decomposing the effect of each covariate at different parts of the distribution. While
readers are referred to the original work, their method uses the influence function as the
dependent variable at different quantiles.8 In the case of quantile q(τ) and observation ,
the Recentered Influence Function (RIFi) is
(2)
Where is the density and 1( ) indicates that the earnings observation is at or above the
quantile q(τ). It is recentered as is added and as a consequence, the expected value
13
of the RIF will be itself. Indeed, Firpo et al. (2009) show this property extends to the
conditional-on-controls RIF. In practice, the estimation of on the observables
becomes a variation on the linear probability model of whether the observed earnings
exceed the quantile of interest. The variation is merely that the coefficients are divided by
the density evaluated at that quantile.
Thus, while one cannot directly decompose the earnings gap at , say the
median earnings for ethnic minorities, one can decompose the probability gap by
estimating linear probability models that first minorities and then whites are above the
minority median earnings. The technique exploits the fact that the earnings gap and the
probability gap are ratios of each other with the ratio being the slope of the cumulative
distribution function, the density. So imagine the earnings gap (minority minus white) at
quantile :
(3)
The technique then uses a reweighting methodology (DiNardo et al. 1996) to construct a
counterfactual wage distribution, the distribution of earnings for white workers if they had
the same distribution of observables as minorities. This generates the counterfactual
earnings and allows decomposing the quantile earnings gap:
(4)
The first term gives the portion of the quantile wage gap due to the difference in returns as
the distribution of observables is constant. This is the unexplained portion of the gap. The
second term gives the portion of the quantile earnings gap due to differences in the
composition as the earnings structure is the same. This is the explained portion of the gap.
14
The RIF regressions allocate the difference due to return and to composition among
the individual covariates including performance pay. Thus, for each quantile three
separate RIF regression are estimated for , for and for . Following this,
the usual Blinder-Oaxaca decomposition is performed on each of the two portions of the
gap. Specifically for our purposes this provides the portion of the explained difference at
any quantile that is attributable to differences in performance pay prevalence and the
portion of the unexplained difference at any quantile attributable to differences in the
return to performance pay.
We emphasize that the wage distribution of white workers is reweighted to make it
similar to that of non-whites using the method in DiNardo et al. (1996). We do not perform
the reverse exercise of reweighting the non-white distribution to make it look like the
white distribution. Although equally reasonable in theory, the relatively small sample size
of the minorities suggests we could face a common support problem. While it is not
difficult to find white workers comparable to nonwhites at any point of support in the
distribution of observables for nonwhites, it is more difficult to find non-white workers
who are comparable to whites. Reweighting in this fashion helps control for the possibility
that the distribution of observables for whites and non-whites could be profoundly
different (Fortin et al. 2011).
Finally, after examining the decomposition across the earnings distribution we
return to fixed effect estimates and provide quantile fixed effect results. While not
allowing the decomposition, these allow us to explore whether the pattern we isolate in
the earlier quantile estimates differs when holding constant worker fixed effects. In the
end, the picture painted by exploring this wide variety of distributional issues will be
broadly consistent in pointing out the importance of bonuses in providing larger returns to
15
minorities and that this influence appears to be concentrated in the upper middle portion
of the distribution.
4. Results
Table 4 summarizes our initial regression estimates. The indicator of performance
pay is a single dichotomous variable that takes the value of one if the worker receives
either performance related pay (PRP) and/or a bonus. The estimates follow the simple
assumption that a single equation summarizes the earnings determinants for both White
and minority workers, an assumption we later relax. The controls include those shown in
Table 1 as well as occupation, industry, region and time dummies. The full estimates are
available upon request but the table focuses on the interaction between the indicators for
performance pay and for being an ethnic minority. The estimates pool across years and
estimate robust standard errors.
INSERT TABLE 4
Looking at Column 1, nonwhites not in performance pay jobs face an earnings
penalty of 21.6 percent while those in performance pay jobs face a significantly smaller
earnings penalty of 11.0 percent.9 Alternatively, for whites performance pay is associated
with 12.1 percent higher earnings while for nonwhites it is associated with 27.1 percent
higher earnings. Thus, the initial estimates suggest that performance pay may help close
the ethnic earnings gap by increasing the earnings of minorities more than those of whites.
Our performance pay measure puts together PRP and bonuses. The third column
disaggregates these into two indicators. The estimated results show that the pattern in the
first column is driven by bonuses. There is a large and highly significant interaction
indicating that the measured ethnic earnings gap is smaller among those receiving
16
bonuses. On the other hand, the interaction of ethnic status with PRP is both insignificant
and small. In what follows, we will continue to consider PRP but much of the attention will
be on the indicator of bonuses.
While we have included hours as an independent variable, the pattern is not a
function of that choice. We created a measure of hourly wages that takes the actual
earnings and dividing by the hours worked. Using this imputed wage measure, the results
appear largely unchanged with a positive and statistically significant coefficient for the
interaction of bonuses and ethnic minority status. We retain our specification as it is
inherently more flexible. We also note that estimates using population weights do not
materially change these results and are available from the authors.
A key issue is that workers are not randomly assigned into performance pay jobs.
Thus, what appears to be a differential attributable to performance pay may instead reflect
workers with greater unobserved ability sorting into performance pay jobs (Lazear 2000).
The measured return to performance pay reflects the unmeasured characteristics of the
workers who sort into performance pay. The first order treatment for such unmeasured
characteristics is to estimate a fixed effects model that holds constant time invariant
worker specific earnings determinants. At issue is whether or not the greater return to
performance pay for ethnic minorities reflects greater unmeasured ability among those
minorities that sort into performance pay jobs.
Columns 2 and 4 of Table 4 report these fixed effect estimates. Only those
respondents who change payment method identify the estimates. As there is no change in
minority status, this and other unchanging characteristics play no role in the estimate.
Nonetheless, we can identify the influence of performance pay on earnings separately for
Whites and ethnic minorities even though we will be unable to identify a direct ethnic
17
minority wage gap. Column 2 of Table 4 shows that the coefficient on output pay remains
statistically significant but shrinks. When accounting for the time invariant fixed effects,
the coefficient implies a 5.2 percent increase in earnings associated with performance pay
which is roughly half of that implied by the pooled estimate. The coefficient on the
interaction remains statistically significant suggesting a difference by ethnicity in the
premium associated with performance pay. The sum of the coefficients suggests that for
minorities the return to performance pay is 13.9 percent. The fourth column separates the
two pay components revealing again that the difference in returns to performance pay is
driven by a significant interaction for bonuses but not for performance related pay which
takes an insignificant coefficient.10 It is well recognized that panel estimates of this sort
increase the role of measurement error and that increased standard errors can be the
cause of a lack of statistical significance (Solon 1985). There is some evidence of this as the
standard error is much larger in the fixed effect estimates. Nonetheless, the pattern and
basic significance remains.
Before moving to the distributional analysis, we describe our efforts to examine
heterogeneity by ethnic group. We focus on three non-white ethnicities: Asian, Black and
other nonwhites. 11 We reproduce the estimates from Table 4 now replacing the single
non-white indicator with these three group indicators and their interactions with the
performance pay variables. Column 1 of Appendix Table 1 shows the pooled estimates
with Asians and other nonwhites having virtually identical negative significant differentials
but with blacks earning the same as whites. The interactions suggest that both Asians and
other nonwhites receive larger returns to bonuses than do whites. Indeed, the differential
return is so large for Asians that it eliminates the ethnic differential. There appear to be no
significant differences in returns to PRP by ethnicity. The second column of Table A1
18
reproduces the fixed effect estimate with the disaggregated ethnic categories and shows
fewer significant differences. While bonuses continue to be associated with higher
earnings, PRP becomes insignificant. Moreover, although the interactions on bonus pay
and three ethnic categories remain positive, both Asian and Black are now insignificantly
different from zero. The only observed difference is a weakly significant difference for
other non-Whites suggesting that their return to bonuses may be larger than that of
whites.
We recognize the heterogeneity of ethnicity even within the groups we have
broken out. Longhi et al. (2013) demonstrate marked variation among Asians in labour
market performance by detailed ethnicity, religion and whether they are first or second
generation. Our data are not sufficient to precisely estimate such differences and as a
consequence we continue to present only broad differences between whites and
nonwhites as we move into our distributional analysis.
Distributional Differences
In the US it has been claimed that the negative racial wage differential shrinks for
those at the top of distribution who have more education and training (Lang 2007). Those
receiving performance pay are an exception as the differential for them increases at the
top of the distribution (Heywood and Parent 2012). To test for similar distributional claims
in Britain we estimate quantile regressions of the basic specification from Table 4. These
are presented in Table 5. We show the quantiles for the 10th, 25th, 50th, 75th and 90th
percentiles. The ethnic earnings differentials clearly shrink as moving up the earnings
distribution in a mirror of the US generalization. The non-white coefficient takes a
coefficient in the 90th percentile that is less than 40 percent of what it was in the 10th
19
percentile. Critically, however, the evidence that bonuses reduce that differential
attenuates only modestly across the distribution. The interaction of bonuses with minority
status retains approximately the same size and significance in all but the lowest quantile.
As a consequence, this shows the opposite of the US pattern of performance pay
increasing differentials at the top of the distribution. Instead, ethnic minorities receive
essentially a similar differential return on bonuses over whites throughout most of the
distribution.
INSERT TABLE 5
The quantile regressions indicate three further patterns of interest. First, at the top
of the distribution, the ethnic differential has shrunk sufficiently that while still statistically
significant for those earning time rates it is non-existent for those receiving bonuses (the
sum of the coefficients on ethnicity and its bonus interaction). Second, in the lower half of
the earnings distribution minorities receive a significantly lower return on performance
related pay. Alternatively, ethnic wage differentials are larger for those in the lower half of
the distribution who receive performance related pay than for those who receive time
rates. Third, the return on bonuses itself seems to attenuate as one moves up the
distribution even as the interaction with minority status does not.
As we suggested in our methodology section, the structure of the estimates so far
limits the coefficients on every variable except performance pay to be identical for ethnic
minorities and whites. This is known not to be the case and likely causes bias to the
estimated interactions. Table 6 presents the estimates of the Firpo et al. (2009) quantile
decomposition. In each case the quantile wage gap is divided into an explained and
unexplained portion and those portions can be further assigned to each specific covariate.
While the full estimation is available from the authors, we have highlighted the results
20
critical to our interest. The first row shows the full predicted ethnic earnings differential at
each quantile. The next two rows show the total explained and unexplained portion of
that differential, that due to differences in characteristics and due to differences in returns
to characteristics. The next two rows show the portion of the explained and unexplained
differences that are attributable to each of the two performance pay variables.
INSERT TABLE 6
The results indicate large total unadjusted negative ethnic earnings differentials at
the bottom of the distribution. These shrink and eventually turn positive at the top of the
distribution. At the 10th percentile the unadjusted differential is over 0.4 log points and it
is, in aggregate, almost entirely unexplained by differences in observables. The differences
in the incidence of performance related pay plays a very small but statistically significant
role in providing explanation. If minorities and white had the same incidence of
performance related pay the observed earnings gap would close very slightly. This will
emerge as a general finding throughout the distribution. Neither of the payment variables
contributes significantly to the very large unexplained differential at the 10th percentile.
While the portion due to the difference in coefficients on bonuses is relatively large it
misses statistical significance at the 10% level.
The 25th percentile reveals a pattern that persists throughout the remainder of the
distribution. The unexplained differential is more negative than the unadjusted differential.
Thus, controlling for observables implies a larger earnings disadvantage to ethnic
minorities than that actually observed. This mimics the findings of Dustmann and
Theodoropoulos (2010). At the median, the unexplained differential is attributable partly
to differences in the returns to payment methods. The influence associated with the two
variables approximately offset each other. With differences in returns to performance
21
related pay making the differential larger (more negative) and differences in returns to
bonuses making the differential smaller (less negative).
The influence of the returns to bonuses becomes more dramatic in the 75th
percentile. The overall unadjusted differential is now essentially zero but is composed of
roughly offsetting explained and unexplained portions. Critically, the negative unexplained
differential includes the now large positive influence from the difference in returns to
bonuses. Indeed, the size of the bonus influence is such that if it did not exist the overall
unexplained difference would be twice as large (more negative). The role of differences in
returns to performance related pay evident at the median has now vanished. Overall, the
evidence provides greater richness to our earlier estimates by indicating that the larger
returns that minorities receive to bonuses significantly reduces the adjusted minority
earnings gap in this portion of the distribution.
Finally, at the 90th percentile, the unexplained minority earnings differential
vanishes and the higher earnings of ethnic minorities are completely explained by the
observables. Again, in the 50th and 75th percentiles the difference in returns to bonuses
narrows the unexplained ethnic minority earnings gap. As a consequence, allowing for
separate earnings regimes for minorities and whites continues to reveal an influence for
bonuses in closing the gap although perhaps not as uniformly as suggested by single
equation methods. Moreover, it continues to provide no indication that performance pay
causes the unexplained earnings gap to expand at the top of the distribution.
INSERT FIGURE 1
Repeating the exercise from Table 6 at every five percentage point quantile allows
us to more fully chart the distributional implications in Figure 1. The top series shows the
total ethnic earnings gap which starts negative but becomes modestly positive in roughly
22
the upper quartile. The third series shows the unexplained portion of the gap. It starts
above the total gap but quickly crosses and remains below for the remainder of the
distribution. At the top of the distribution it reaches a value of approximately zero. Here
the higher wages of ethnic minorities are fully explained by their characteristics and the
wage returns received by Whites and minorities are, in aggregate, the same. In the bulk of
the distribution, however, there remains a negative unexplained differential which is often
taken as consistent with discrimination. The other two series allow us to see the different
roles played by performance related pay and bonuses in the unexplained differential. The
second series removes the differences in returns to performance related pay from the
unexplained differential. This series moves slightly above the total unexplained differential
indicating that the unexplained differential is modestly more negative because of the
difference in the returns to performance related pay. The fourth series retains the role of
performance related pay but removes the difference in returns to bonuses from the
unexplained differential. This fourth series moves well below the unexplained gap and
especially so from approximately the 60th to 85th percentile. Thus, the unexplained
differential is less negative (closer to zero) because of the higher return for minorities to
bonus receipt. The role of this narrowing is most pronounced in what we see as the upper
middle portion of the distribution.
Again, a concern may be that the differences apparent in the quantile regressions
reflect differences in unobserved worker productivity across pay types and across the
distribution. To examine this one could include a worker specific shift for each individual
for every quantile. This might be termed a conditional (on the quantile) fixed effects
quantile model. However, as discussed by Koenker (2004) including this quantile
dependent distributional worker effect is difficult in a setting such as ours where there is a
23
large number of cross-sectional units but few time observations per unit. Instead he
suggests the inclusion of a single individual specific effect across the quantiles. To do this
requires the joint estimation of all τ quantile regressions simultaneously. This
unconditional fixed effect quantile regression is implemented using the public domain
package "R" and the program by Koenker (2004).
INSERT TABLE 7
Table 7 shows the fixed effect quantile estimates and presents similarities with and
differences from the patterns without the fixed effects. First, it continues to be the case
that bonus receipt remains a significant positive determinant of earnings throughout the
distribution and that the magnitude of the coefficient clearly declines at the top of the
distribution. As anticipated, the magnitude of the fixed effect coefficients is typically
smaller than those without fixed effects. Second, the fixed effect estimate of the
coefficient on the interaction of bonus with minority status remains positive but grows
over the distribution and is significant only at the 75th and 90th percentile. This pattern of
the interaction concentrated at the top of the distribution was not evident in the quantile
estimates without fixed effects (Table 5) but was at least partially reflected in the
decompositions which emphasised the 60th to 85th percentile. The fixed effect result
continues to differ from the evidence on US black-white earnings differences. Heywood
and Parent (2012) show blacks on performance pay doing increasingly worse relative to
their whites at the top of the earnings distribution. As a consequence, more able blacks
sort out of performance pay. Our results suggest that bonuses improve the relative
earnings of ethnic minorities in the upper or at least upper-middle portion of the
distribution. Thus, our implication for Britain would seemingly be the opposite: more able
ethnic minorities should sort into bonuses. We will explore this finding before concluding.
24
We emphasize that the patterns we show are not dependent upon our particular
choice of independent variables or our particular choice of how to trim the sample. In
robustness exercises we have added educational levels of the respondent’s parents,
whether or not the respondent is foreign born, self-perception of being overweight (as
proxy for non-cognitive skills) and being paid by the hour.12 None of these alter the basic
results. Similarly, we have excluded the top and bottom one percent of respondents in the
hours of work distribution and also eliminated all workers less than 25 years of age.
Neither modification materially changes the results.
While we obviously cannot identify all the components that drive sorting, we can
compare the composition of the sectors by observables. Table A2 presents two probit
estimates showing marginal effects on the probability of receiving a bonus. The first is for
Whites and the second is for non-whites. These are not meant to imply causation but
rather simply document the patterns of sorting across the observables.
Comparing the estimates suggests differential ability sorting. The influence of
possessing a university degree or more is nearly five times larger for ethnic minorities.
Thus, highly educated minorities are more likely to sort into bonus jobs. Managers and
supervisors who are minorities are also more likely to sort (or be sorted) into bonus jobs.
Expanding the estimates to include the unique indicators available in the BHPS reveals that
parental education also plays a larger role for minorities sorting into performance pay than
for whites. Thus, if we anticipate that unobservable measures of ability are correlated with
observed measures, the pattern of sorting is consistent with high ability minorities sorting
into performance pay jobs and consistent with the fixed effect estimates being smaller
than the pooled estimates. While minorities earn more from being in performance pay jobs
than whites, this partially reflects that high ability minorities disproportionately appear in
25
performance pay jobs. This contrasts with the finding from the US that high ability blacks
sort out of performance pay jobs.
6. Conclusions
Performance pay is thought to more closely link worker productivity and earnings.
We have explored how performance pay influences the size of the measured ethnic
earnings differential. We are, to our knowledge, the first to do this using data from Britain.
We show that in pooled estimates performance pay is associated with a smaller average
ethnic wage differential. This is driven by a smaller ethnic differential for those receiving
bonuses. This pattern of results remains in fixed effect estimates but the size of the
influence is smaller than in the OLS suggesting sorting. Asians and Black ethnicities present
a contrasting pattern in pooled estimates but that contrast does not carry-over to the fixed
effect estimates.
The quantile estimates suggest that the tendency of bonuses to shrink ethnic wage
differentials is concentrated at the top (or upper middle) of the earnings distribution. It is
here that the return to performance pay is significantly larger for ethnic minorities. This
general finding is seemingly robust and carries over to quantile estimates holding constant
the time-invariant individual characteristics. The descriptive estimates suggest that high
ability minorities (using the observables) are disproportionately sorting into the
performance pay sector. If the unobservables are correlated with the observables, the
fixed effect estimates are consistent with sorting on unobserved ability. Thus, ethnic
minorities may anticipate that they will more likely be able to reveal this unobserved (to
the researcher) ability and be rewarded for it in the performance pay sector.
26
It is valuable to contrast one last time the tenor of our results for Britain with those
for the United States. The latter show that typically only very formulaic performance pay is
associated with reduced racial earnings gaps. Broader forms of performance pay and
especially bonuses have been associated with increased differentials. Moreover, the
results from the US show blacks on performance pay doing far worse than whites at the
top of the earnings distribution (Heywood and Parent 2012). The results of the current
paper differ. Performance pay and especially bonuses tend to reduce the size of the ethnic
earnings differential. This tendency is evident in the upper middle portion of the
distribution and there is no evidence that bonuses hurt the relative position of minorities
at the very top of the distribution.
Future work should examine alternative data sources for Britain. Linked employer-
employee data may allow new insights by offering information about the firms that
provide performance pay schemes and the characteristics of those schemes. Larger
samples of minorities might be able to explore more successfully the heterogeneity of
which our results could only hint even if they largely forego the longitudinal element.
Moreover, many countries other than the US and Britain have ethnic minorities and they
may serve to generalize the findings. In each case, the issue would be the influence of
performance pay on the ethnic earnings differential. While these alternative sources may
provide new or different results, the estimates in this paper suggest that the North
American results do not easily translate in similar patterns for Britain.
All controls as per Table 4. Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
36
Table A2 - Probability of Bonus Receipt, Probit Estimates (Marginal Effects)
(1) (2)
whites non whites
Age 0.0129*** 0.00287
(0.00224) (0.0167)
Age2/100 -0.0199*** -0.00970
(0.00280) (0.0205)
A Level 0.0552*** 0.0206
(0.00886) (0.0773)
Diploma 0.00534 -0.0108
(0.0138) (0.0862)
Degree or Higher 0.0450*** 0.207***
(0.0121) (0.0702)
Married 0.0256*** 0.0898
(0.00924) (0.0672)
Child -0.00229 -0.203***
(0.00892) (0.0617)
Health Limits Work -0.0491*** 0.110
(0.0153) (0.0992)
Union Member 0.0737*** -0.0550
(0.00961) (0.0728)
Temporary Job -0.288*** -0.299***
(0.0155) (0.0771)
Large Firm 0.0736*** 0.0531
(0.00772) (0.0498)
Tenure
0.00382*** 0.0105**
(0.000713) (0.00460)
Tenure2/100 -0.00478*** -0.0105*
(0.00121) (0.00782)
Manager/Supervisor 0.0780*** 0.159***
(0.00882) (0.0542)
Observations 21,024 659
Controls included but not reported 10 year dummies, 8 Industry Dummies, 8 Occupation
Dummies and 11 region dummies. Robust standard errors in parentheses *** p<0.01, **
p<0.05, * p<0.1
37
References:
Becker Gary, S. (1971) The Economics of Discrimination (2nd Edition). Chicago: University of Chicago Press.
Bell Brian, D. (1997) “The Performance of Immigrants in the United Kingdom: Evidence
from the GHS,” Economic Journal, 107, 333-44. Blackaby David, Derek Leslie, Phil Murphy and Nigel O’Leary (1994) “Black-White Male
Earnings and Employment Prospects in the 1970s and 1980s: Evidence for Britain,” Economics Letters, 46, 273-79.
Blackaby David, Derek Leslie, Phil Murphy and Nigel O’Leary (1998) “The Ethnic Wage
Gap and Employment Differentials in the 1990s: Evidence for Britain,” Economics Letters, 58, 97-103.
Blackaby David, Derek Leslie, Phil Murphy and Nigel O’Leary (2002) “White/Ethnic
Minority Earnings and Employment Differentials in Britain: Evidence from the LFS,” Oxford Economic Papers, 54, 270-97.
Blackaby David, Derek Leslie, Peter Murphy and Nigel O’Leary (2005) “Born in Britain:
How are Native Ethnic Minorities Faring in the British Labour Market?” Economics Letters, 88, 370-75.
Bloom, Nicholas and John Van Reenen (2010) “Human Resource Management and
Productivity” NBER Working Paper Series, No. 16019. Booth, Alison and Jeff Frank (1999) “Earnings, Productivity and Performance-Related
Pay” Journal of Labor Economics, 17, 447-463. Castilla, Emilio (2012) “Gender, Race and the New (Merit-Based) Employment
Relationship,” Industrial Relations, 51, 528 – 62. Castilla. Emilio (2008) “Gender, Race and Meritocracy in Organizational Careers,”
American Journal of Sociology, 113, 1479-526. Clark, Ken and Stephen Drinkwater (2000) “Pushed Out or Pulled In? Self-Employment
Amongst Britain’s Ethnic Minorities,” Labour Economics, 7, 603-28. Clark, Ken and Stephen Drinkwater (2009) “Dynamics and Diversity: Ethnic Employment
Differences in England and Wales,” Research in Labor Economics, 29, 299-333. DiNardo, John, Nicole Fortin and Thomas Lemieux (1996) "Labor Market Institutions and
the Distribution of Wages," Econometrica, 64, 1001 – 44.
38
Dustmann, Christian, Tomasso Frattini and Nikolaos Theodoropoulos (2011) Ethnicity
and Second Generation Immigrants, Chapter 15 in “The Labour Market in Winter: The State of Working Britain”, edited by J. Wadsworth and P. Gregg, Oxford University Press, 220-39.
Dustmann Christian and Nikolaos Theodoropoulos (2010) “Ethnic Minority Immigrants
and Their Children in Britain,” Oxford Economic Papers, 62, 209-33. Elivira Marta and Robert Town (2001) “The Effects of Race and Worker Productivity on
Performance Evaluations,” Industrial Relations, 40, 571-90. Fang Tony and John Heywood (2010) “Immigration, Ethnic Wage Differentials and
Output Pay in Canada,” British Journal of Industrial Relations, 48, 109-30. Fang Tony and John Heywood (2006) “Output Pay and Ethnic Wage Differentials:
Canadian Evidence,” Industrial Relations, 45, 173-94. Firpo Sergio, Nicole Fortin and Thomas Lemieux (2009) "Unconditional Quantile
Regressions," Econometrica 77, 953 – 73. Fortin Nicole, Thomas Lemieux and Sergio Firpo (2011) "Decomposition Methods in
Economics," in Handbook of Labor Economics Volume 4A edited by Orley Ashenfelter and David Card, North-Holland: Amsterdam.
Graham Prentice, Simon Burgess and Carol Propper (2007) Performance Pay in the
Public Sector: A Review of Issues and Evidence, Office of Manpower Economics, November.
Green, Colin and John Heywood (2008) “Does Performance Pay Increase Job
Satisfaction?”, Economica, 75, 710-728. Green, Colin and John Heywood (2011) “Profit Sharing, Separation and Training”, British
Journal of Industrial Relations, 49, 623-642. Heywood John and Patrick O’Halloran (2005) “Racial Earnings Differentials and
Performance Pay,” Journal of Human Resources, 40, 435-53. Heywood John and Patrick O’Halloran (2006) “Market Structure, Payment Methods and
Racial Earnings Differences,” in “Product Market Structure and Labor Market Discrimination”, edited by J.Heywood and J.Peoples, University of New York Press, Albany NY, 187-208.
39
Heywood John and Daniel Parent (2012) “Performance Pay and the White-Black Wage Gap,” Journal of Labor Economics, 30, 249-90.
Jirjahn Uwe and Gesine Stephan (2004) “Gender, Piece Rates and Wages: Evidence from
Matched Employer–Employee Data” Cambridge Journal of Economics, 28, 683-704. Johnson, William and Derek Neal (1996) “The Role of Premarket Factors on Black-White
Wage Differences,” Journal of Political Economy, 104, 683-704. Koenker Roger (2004) “Quantile Regression for Longitudinal Data,” Journal of
Multivariate Analysis, 91, 74-89. Lazear Edward P (2000) “Performance Pay and Productivity,” American Economic
Review, 90, 1346-61. Lang Kevin (2007) Poverty and Discrimination, Princeton NJ: Princeton University Press. Lemieux Thomas, W. Bentley MacLeod, and Parent Daniel (2009) “Performance Pay and
Wage Inequality,” Quarterly Journal of Economics, 124, 1-49. Longhi, Simonetta, Cheti Nicoletti and Lucinda Platt (2013) “Explained and Unexplained
Wage Gaps Across the Main Ethno-Religious Groups in Great Britain”, Oxford Economic Papers, forthcoming.
MacLeod W. Bentley (2003) “Optimal Contracting with Subjective Evaluation,” American
Economic Review, 93, 216-40. Madden Janice F. (2008) “Performance Pay and Gender Discrimination: Commissions to
Stockbrokers,” Working Paper: Population Studies Center, University of Pennsylvania.
Meon Pierre-Guillaume and Ariane Szafraz (2011) “The Modern Corporations as a Safe
Haven for Taste-based Discrimination: An Agency Model of Hiring Decisions,” Labour Economics, 18, 487-97.
National Equality Panel Report (2010) An Anatomy of Economic Inequality in the UK.
Available from http://eprints.lse.ac.uk/28344/1/CASEreport60.pdf OECD (2011) Divided We Stand: Why inequality Keeps Rising, OECD Publishing: Paris. Prendergast Canice and Robert Topel (1993) “Discretion and Bias in Performance
Evaluation,” European Economic Review, 37, 355-65.
40
Pudney Stephen and Michael Shields (2000) “Gender, Race, Pay and Promotion in the British Nursing Profession: Estimation of a Generalized Ordered Probit Mode,” Journal of Applied Econometrics, 15, 367-99.
Solon, Gary (1985) “Benefits and Limitations of Panel Data: Comment,” Econometric
Reviews, 4, 183-86. Solon, Gary (1988) “Self-selection Bias in Longitudinal Estimation of Wage Gaps,”
Economic Letters, 28, 285-90. Stewart, Mark (1983) “Racial Discrimination and Occupational Attainment in Britain,”
Economic Journal, 93, 521-43. Stewart, Mark (2011) “The Changing Picture of Earnings Inequality in Britain and the
Role of Regional and Sectoral Differences,” National Institute Economic Review, 218, R20-R32.
41
Endnotes
1 Additional North American based studies of ethnic earnings differentials and
performance pay include Fang and Heywood (2006, 2010). There also exist studies of
the gender earnings differential and performance pay and while many are also based on
North American data, Jirjahn and Stephan (2004) use German data to demonstrate a
smaller gender earnings differential among piece rate workers than among time rate
workers.
2 Ethnic earnings differentials may be thought of as a specific form of earnings inequality
and research suggests that general earnings inequality is growing quickly in Britain
(OECD 2011) and that this growth is uniquely tied to the use of performance bonuses by
the financial services industry (Stewart 2011).
3 Our data goes through the end of the BHPS which was replaced with the
Understanding Society Survey. While BHPS households can be traced into the new
survey, the performance pay questions were not carried over.
4 Although we note that some profit sharing plans are reserved only for executives and
do not include all workers.
5 In a robustness check we show that focusing on prime age males does not change the
results.
6 See Clark and Drinkwater (2000) for evidence on the relative performance of self-
employed ethnic minorities in Britain.
7 We include workers with mixed ethnicity, those who identify one or more non-white
ethnicities, as non-white.
8 The influence function measures how robust a distributional statistic is to outliers.
Thus, when considering the mean, , the influence function is .
9 The second estimate comes from adding the relevant coefficients (-.243+.126) and
taking e to this power and subtracting one.
10 Those moving into and out of performance pay may represent a selected sample
causing the fixed effect estimate to not be representative (Solon 1988). As a check, we
limited our sample to only those observed moving between performance pay and time
42
rates and re-estimated the pooled OLS from Table 4. The smaller sample size (2,350)
resulted in less precision but mimicked the entire sample with the bonus-minority
interaction remaining positive and significant.
11 Asians included those who identify as Indians, Pakistanis, Bangladeshi, Chinese or
Other Asian. Blacks are those who identify as Black-Caribbean, Black-African or Black
Other.
12 Johnson and Neal (1996) show that controlling for parental characteristics causes the
