Вы находитесь на странице: 1из 7

The Economic Journal, 114 (March), C110–C116.  Royal Economic Society 2004.

Published by Blackwell
Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

THE EMPLOYMENT EFFECTS OF THE NATIONAL


MINIMUM WAGE*

Mark B. Stewart

This paper estimates the employment effects of the introduction of the UK National Minimum
Wage in April 1999 and subsequent upratings in 2000 and 2001. It uses a difference-in-dif-
ferences estimator based on position in the wage distribution. For the upratings an adjusted
estimator is also introduced to allow for the possible differential response to the change in the
macro employment trend. No significant adverse employment effects are found for any of these
events for any of the estimators for any demographic group.

Do minimum wages reduce employment? This question is central to the debate on


minimum wages and has been hotly contested for many years. The standard
textbook labour demand model predicts that the introduction or raising of a
minimum wage will lead to firms reducing employment. In contrast a range of
monopsony, efficiency wage, search and other models that have been proposed do
not predict a reduction and show that in some circumstances employment may
even rise.1 The empirical evidence has also been mixed, with some studies showing
negative effects and others showing positive or zero effects.2 The recent intro-
duction of the National Minimum Wage rekindled the debate in the UK.
This paper estimates the employment effects on low-wage workers of the
introduction of the National Minimum Wage in 1999 and of the subsequent up-
ratings in 2000 and 2001. The adult rate at introduction in April 1999 was set at
£3.60/hour, and was then increased to £3.70 in October 2000 and £4.10 in
October 2001.3 In addition to the necessary policy evaluation, the UK experience
also provides an important test-bed in the context of the international minimum
wage debate. Because its introduction followed a period without any minimum, it
provides a cleaner test than those for the countries on which most of the testing
has been done, since behaviour in the absence of a minimum wage is directly
observed. 4
This paper uses difference-in-differences estimation on individual-level longi-
tudinal data from matched Labour Force Surveys to estimate the impact of the
minimum wage introduction and upratings on the employment prospects of low

* The author is grateful to the ESRC for financial support (under award R000223440). The LFS data
were provided via the Data Archive.
1
For a description and discussion of these models in the context of minimum wages see, for
example, Brown (1999, section 2), Card and Krueger (1995, chapter 11) and Dickens et al. (1999,
section 2).
2
See Brown (1999) for a survey to that date, Stewart (2003) for a description of more recent
evidence.
3
The youth rate (for those aged 18–21 inclusive) was initially set at £3.00/hour and then increased to
£3.20 in June 2000 and £3.50 in October 2001. (There was also a development rate for adults in the first
6 months of a new job with accredited training at £3.20/hour, although it was not greatly used.)
4
Additional advantages are that other evidence suggests both a very high compliance rate and a lack
of spillover effects onto the wages of higher paid workers; see Dickens and Manning (2002) on latter.
[ C110 ]
[ M A R C H 2004 ] THE EMPLOYMENT EFFECTS . . . C111
wage workers whose wages would have had to be raised to comply with the new
minimum – the group directly affected. Their subsequent employment is com-
pared with a similar group from slightly further up the wage distribution, who act
as the control group.

1. Estimation Approach
Consider first the evaluation of the introduction. The subsequent employment
probabilities of employees whose wages would have had to be raised to comply with
the new minimum (i.e. those initially below the minimum) can be compared with
that of a group from higher up the wage distribution. However direct comparison
of the two groups would not be appropriate to identify any causal effect since, even
in the absence of a minimum wage, those at the bottom end of the wage distri-
bution have lower subsequent employment probabilities than those from higher
up. This makes the difference-in-differences approach a natural one to take.5 The
difference between the two groups in a period affected by the minimum wage can
be compared with the equivalent difference in an earlier period when no mini-
mum wage was in place.
Let the binary indicator e0it denote the employment status of individual i in time
period t in the absence of a minimum wage. Define two groups of employees,
indexed by g: those in group g ¼ 1 are affected by the minimum wage introduction
because their wages are initially below the minimum, while those in group g ¼ 2
are not directly affected because their wages are already above the minimum. In
addition, suppose that a minimum wage is introduced at t* and that for observa-
tions prior to t* no minimum wage is in place. The simplest form of the estimator
would use just two time periods: t1 and t2 such that t1 + 1 < t* £ t2 + 1. Compari-
sons across g and t are used to estimate the counterfactual employment proba-
bilities. Suppose that:
Pðe0itþ1 ¼ 1je0it ¼ 1Þ ¼ ag þ ct : ð1Þ

The first component is a group-specific effect fixed over time, the second a
macro time effect common across groups. A key identifying assumption is that
there are no interactions in the absence of the minimum, i.e. parallel growth paths
across the groups. This is tested below.
Define a ‘treatment’ interaction: Dit ¼ 1 if individual i is affected by the intro-
duction of the minimum wage, i.e. i is in group g ¼ 1 and t + 1 ‡ t*; Dit ¼ 0
otherwise. If there is a homogeneous impact in group 1 and no impact in group 2,
then the difference-in-differences estimator (given by double differencing the
sample means) consistently estimates this impact. The estimator is also given by
OLS estimation of a linear equation with group and time effects and with Dit as the
sole interaction term. This can also be augmented by additional control variables:
Pðeitþ1 ¼ 1jeit ¼ 1Þ ¼ x 0it b þ hDit þ ag þ ct : ð2Þ

5
See Meyer (1995), Angrist and Kreuger (1999) and Blundell and Dias (2000) for good discussions
of this type of estimator.
 Royal Economic Society 2004
C112 THE ECONOMIC JOURNAL [MARCH
The formulation easily incorporates additional groups and time periods. Group
g ¼ 1, those below the minimum, remains as the ‘treatment’ group and group
g ¼ 2, those slightly above the minimum, as the ‘control’ or ‘comparison’ group.
The latter is defined here as those between the age-specific minimum and 10%
above.6

2. Results for the National Minimum Wage Introduction


The model is estimated using matched LFS data covering the period March 1997 –
March 2000 (from 13 quarterly LFSs). It uses wage and characteristic information
from wave 1 and employment status information from wave 5 (12 months later).
This determines the start date: earnings questions were added to the wave 1
questionnaire in March 1997 (prior to this they had only been asked of the
(outgoing) wave 5 respondents). Advantages of the LFS over other nationally
representative longitudinal datasets for the current purpose are a better repre-
sentation of low earnings workers than the NES and a much larger sample than the
BHPS.
Estimates are presented in Table 1 for four demographic groups – male and
female adults (22–59) and youths (18–21). The dependent variable is the indi-
vidual’s employment status at time t + 1. Two alternative wage measures are used.7
The first is constructed as gross pay last time paid (converted to a weekly basis)
divided by the number of paid hours usually worked (both referring to main job
only). This is the ONS construction but suffers from a potential mismatch between
the hours measure used and the number of hours worked in the period covered by
the reported gross pay. The second measure uses hours actually worked in the
reference week (in the main job). This is likely to provide a more accurate measure
for those with low wages (Stewart, 2003), who are the focus of attention here, while
the measure based on usual hours is argued by ONS to provide a more accurate

Table 1
Estimated Effects of the Introduction of the National Minimum Wage
on the Probability of Subsequent Employment
(Difference-in-differences estimates, matched Labour Force Survey data)

Wage variable Controls? Adult men Young men Adult women Young women
WU No 0.020 (0.68) 0.012 (0.15) )0.006 (0.32) 0.067 (0.79)
WA No 0.015 (0.49) 0.059 (0.72) 0.011 (0.60) 0.079 (0.86)
WU Yes 0.014 (0.93) 0.073 (0.88) )0.010 (0.93) 0.119 (1.15)
WA Yes 0.011 (0.74) 0.155 (1.36) 0.006 (0.51) 0.065 (0.78)

Notes:
Wage variables: WU ¼ wage based on usual hours, WA ¼ wage based on actual hours.
Control variables as listed in Stewart (2003).
Data from March 1997 to March 2000 inclusive, from 13 Quarterly Labour Force Surveys. Sample
size ¼ 54,165, made up of 26,312 men (1,087 young) and 27,853 women (942 young).
Absolute ‘robust’ t-ratios on model coefficients in parentheses.

6
The robustness of the results to varying this latter threshold is examined in Stewart (2003).
7
The real wage variables (April 1999 prices) are constructed using the Retail Price Index (all items).
 Royal Economic Society 2004
2004 ] THE EMPLOYMENT EFFECTS OF THE NMW C113
measure for those higher up the wage distribution. Results for both are presented
in Table 1.8
The first two rows of Table 1 give linear difference-in-differences estimates
without additional control variables. Absolute ‘robust’ t-ratios are given in paren-
theses. The estimates are insignificantly different from zero for both wage variables
for all four demographic groups. In addition seven of the eight estimates are also
positive. The model with control variables uses a logit form.9 Table 1 presents
estimates of the ‘marginal effect’ of the dummy variable of interest. These are
probability difference-in-differences. These estimates are also insignificantly dif-
ferent from zero for both wage variables for all four demographic groups. For the
wage variable based on actual hours, the estimate is also positive for all four
demographic groups. For the wage based on usual hours it is positive for three
of the groups. Only for adult women and using this latter wage variable is there
a negative (although insignificant) effect. The absolute t-ratio in this case is less
than 1.10
An important identifying assumption here is that there are no interaction terms
in the absence of the minimum wage. An advantage of the UK introduction for this
approach is that it was preceded by a period with no minimum, allowing this
assumption to be tested. Estimation of the model using only the period without a
minimum and with a full set of interaction terms added finds them to be indi-
vidually and jointly insignificant. In addition the macro employment trend in the
period surrounding the minimum wage’s introduction was similar to that in this
pre-introduction phase, adding support to the assumption that there would have
been no interactions in this later period if the minimum wage had not been
introduced.

3. Adjusting the Estimator


The similarity of the macro trend in the two phases is however not the case when
considering the minimum wage upratings in either 2000 or 2001. This is illustrated
by a plot of the employment rate in Figure 1.11 The trend growth rate is very
different in 2000 and beyond to that in the no-minimum phase.12 The difference-
in-differences estimator requires that growth rates in the ‘treatment’ and ‘control’
groups would have been the same if the minimum wage had not existed. Despite
the test on the pre-introduction phase described above, the macro employment
downturn post-2000 casts doubt on this assumption for this later period.

8
Both show significantly greater wage growth for the ‘treatment’ group than the ‘control’ group.
Specifically, a wage growth equation using the equivalent estimator gives a significant positive effect.
9
Separate models are estimated for men and women and the main effects and intercept are allowed
to vary within these across the two age groups.
10
These findings are robust to an extensive range of specification modifications, to the use of a ‘wage
gap’ estimator and to the use of NES or BHPS data in place of the LFS (Stewart, 2003).
11
3-month moving average of ONS series MGSU: employment rate for all of working age, from LFS.
12
The comparison period for the estimator is a moving 12-month window (t, t + 1) such that t is
before the uprating and t + 1 after. The limits are 12 months before the uprating and 12 months after.
The start limit for the first uprating and end limit for the second are shown as vertical dashed lines in
Figure 1.
 Royal Economic Society 2004
C114 THE ECONOMIC JOURNAL [MARCH

Fig. 1. Employment rate

The estimator may therefore require some modification to evaluate the effects
of these upratings. A generalised specification needs to allow the macro effects to
have a differential impact across the groups. Following Blundell and Dias (2000),
the specification may be generalised to:
Pðe0itþ1 ¼ 1je0it ¼ 1Þ ¼ ag þ kg ct : ð3Þ

where the kg term allows for the differential impact of the macro effects across the
groups. Under this assumption the difference-in-differences estimator provides a
consistent estimator not of h, but of plim ^ hDID ¼ h þ ðk1  k2 Þðct2  ct1 Þ, and the
second term may be non-zero. (Note however that if the c-terms are the same for t1
and t2, the estimator remains consistent despite differential impact across the
groups.) Bell et al. (1999), to evaluate the New Deal programme for unemployed
youth using an older age group as control group, propose finding a second time
period, over which a similar macro trend has occurred13 and removing the bias by
taking a further time difference. If a suitable period can be found, this
‘differential-trend-adjusted’ difference-in-differences estimator will provide a
consistent estimator of h.
For the evaluation of the impact of the minimum wage on employment, the
problem is the lack of a suitable earlier time period. For the period for which
suitable individual-level data is available, there is not an earlier time interval with
macro employment trends to match those in the interval being investigated (see
Figure 1). An alternative is to adjust for differential trends by differencing in the
group dimension rather than the time dimension. Suppose instead that we can
find groups g ¼ 3 and g ¼ 4 so that the differences in their responsiveness to the
13
They suggest an earlier period at a similar point in the previous cycle.
 Royal Economic Society 2004
2004 ] THE EMPLOYMENT EFFECTS OF THE NMW C115
macro trend is the same as that between the original two groups, i.e.
k3 ) k4 ¼ k1 ) k2 This too would provide a consistent estimator. The two new
groups need to be such that, in the absence of a minimum wage, their employment
trends would differ from each other by the same amount as the original treatment
and control groups. This assumption then replaces the more restrictive one that
employment trends are the same in the absence of the minimum.

4. Results for the National Minimum Wage Upratings


The construction of suitable additional comparison groups requires an assump-
tion on how the responsiveness of employment growth to macro shocks varies
across the wage distribution (or at least the lower end of it). Estimates under a
number of alternative assumptions are considered here. The additional groups
need to be adjacent to the original control group to ensure similar supply char-
acteristics and then to be the same distance apart, in some suitable metric, as the
original two groups. Three estimators are considered. One uses equal width
comparison groups to the original. (For adults the new groups are therefore 1.1 m
to 1.2m and 1.2m to 1.3m, where m is the value of the minimum wage.) This might
be suggested if the underlying relationship was linear and there was suitable dis-
persion within groups. A second estimator constructs the groups so that the
medians are equi-distant, i.e. the median wage for group g ¼ 4 is the same distance
above that for g ¼ 3 (and g ¼ 3 relative to g ¼ 2) as that for g ¼ 2 is above g ¼ 1
(the treatment group).14 The third estimator constructs the additional comparison
groups to be contiguous to and of equal sample size to the original comparison
group.
Results for all three of these estimators and for the unadjusted difference-in-
differences estimator are presented in Table 2. The cell sizes are considerably
smaller for the upratings. To combat this the two youth groups are combined and
the upper limit for the first comparison group for youths is increased to the
minimum + 15%. However for the 2001 uprating even these two modifications still
leave the youth group cell sizes too small to be useful and estimates for the 2001
uprating are presented only for the two adult groups.
The unadjusted difference-in-differences estimates are insignificant for both
upratings and all demographic groups. The only one that even gets close to sig-
nificance is that for adult women for the 2000 uprating using the wage based on
usual hours. However it only has an absolute t-ratio of 1.41 and even this falls
sharply if the wage based on actual hours is used. For the 2001 uprating (a much
larger increase) the effect for this group is much smaller and less significant. The
adjusted difference-in-differences estimates show if anything less evidence of an
adverse employment effect. The majority of them are positive. Among those that
are negative the absolute t-ratio never rises above 0.97.

14
For this estimator results could only be obtained for the adult groups. For youths, even if the whole
of the rest of the wage distribution is used for group g ¼ 4 the median is not high enough to match.
 Royal Economic Society 2004
C116 THE ECONOMIC JOURNAL [ M A R C H 2004]
Table 2
Estimated Effects of the 2000 and 2001 Minimum Wage Upratings
on the Probability of Subsequent Employment
(Difference-in-differences estimates, matched Labour Force Survey data)

2000 Uprating 2001 Uprating


Wage
Estimator variable Adult men Adult women Youths Adult men Adult women
Unadjusted difference- WU 0.007 (0.40) )0.017 (1.41) )0.025 (0.44) 0.022 (1.09) )0.007 (0.40)
in-differences WA 0.013 (0.82) )0.008 (0.65) 0.022 (0.39) 0.018 (0.77) 0.001 (0.04)
Equal size comparison WU 0.000 (0.02) )0.004 (0.19) 0.011 (0.14) )0.001 (0.04) 0.002 (0.07)
groups WA 0.013 (0.82) 0.003 (0.28) 0.021 (0.38) )0.035 (0.82) 0.006 (0.22)
Equal width comparison WU 0.009 (0.37) )0.009 (0.45) 0.008 (0.10) 0.023 (0.84) -0.007 (0.24)
groups WA 0.030 (1.28) )0.015 (0.74) 0.006 (0.07) 0.017 (0.51) 0.013 (0.43)
Equi-distant medians WU 0.017 (0.85) )0.021 (0.97) n.a. 0.025 (1.02) 0.016 (0.38)
WA 0.033 (1.70) )0.007 (0.35) n.a. 0.016 (0.59) 0.002 (0.06)

Notes:
Wage variables: WU ¼ wage based on usual hours, WA ¼ wage based on actual hours.
Control variables as used in Table 1.
Sample sizes: for 2000 uprating: 24,098 adult men, 25,809 adult women, 1,973 youths; for 2001 uprating:
17,316 adult men, 18,565 adult women, 1,406 youths (these are for the wage based on usual hours,
slightly smaller for wage based on actual hours).
Absolute ‘robust’ t-ratios on model coefficients in parentheses.

5. Conclusions
This paper estimates the impact of the 1999 UK introduction of a minimum wage,
after a period without a minimum, and of the upratings in 2000 and 2001, on the
probability of subsequent employment among those whose wages would have had
to be raised to comply with the new minimum, using a difference-in-differences
approach. No significant adverse effect on employment is found for either the
introduction or the upratings for any of the demographic groups considered.

University of Warwick

References
Angrist, J. and Kreuger, A. (1999). ‘Empirical strategies in labor economics’, in (O. Ashenfelter and
D. Card, eds.) Handbook of Labor Economics, vol. 3, Amsterdam: Elsevier.
Bell, B., Blundell, R. and Van Reenen, J. (1999). ‘Getting the unemployed back to work: the role of
targeted wage subsidies’, International Tax and Public Finance, vol. 6, pp. 339–60.
Blundell, R. and Dias, M. C. (2000). ‘Evaluation methods for non-experimental data’, Fiscal Studies,
vol. 21, pp. 427–68.
Brown, C. (1999). ‘Minimum wages, employment and the distribution of income’, in (O. Ashenfelter
and D. Card, eds.) Handbook of Labor Economics, vol. 3, Amsterdam: Elsevier.
Card, D. and Krueger, A. B. (1995). Myth and Measurement: The New Economics of the Minimum Wage,
Princeton NJ: Princeton University Press.
Dickens, R., Machin, S. and Manning, A. (1999). ‘The effects of minimum wages on employment: theory
and evidence from Britain’, Journal of Labor Economics, vol. 17, pp. 1–22.
Dickens, R. and Manning, A. (2002). ‘Has the national minimum wage reduced UK wage inequality?’,
CEP, DP No. 533.
Meyer, B. (1995). ‘Natural and quasi-experiments in economics’, Journal of Business and Economic
Statistics, vol. 13, pp. 151–61.
Stewart, M. B. (2003). ‘The impact of the introduction of the UK minimum wage on the employment
probabilities of low wage workers’, mimeo, University of Warwick, forthcoming in Journal of the
European Economics Association.

 Royal Economic Society 2004

Вам также может понравиться