Understanding Wage and Productivity
Dispersion in the United Kingdom
by
Giulia Faggio, Kjell G. Salvanes
and John Van Reenen
Preliminary
Contents of the presentation
•
•
•
•
•
Motivations
Review of the literature
Data
Description of the trends
Decomposition of wage and productivity
dispersion
• Analysis of Total Factor Productivity (TFP)
• Conclusions
Motivations I
• Wage inequality has increased substantially in the US and
the UK over the last 30 years.
• A large proportion of this increase has been explained as a
within-group phenomenon (i.e. by education, skill,
occupation or industry group).
• Many theories have been proposed to explain this withingroup inequality. They assume that increasing inequality
takes place between firms within industries rather than
within firms.
• There is almost no evidence on the impact of these betweenfirm effects on the aggregate increase in wage inequality
and this paper tries to fill this gap.
Motivations II
• Looking at firm productivity, previous studies have analysed:
– the cross-sectional distribution of firm productivity (e.g. Haltiwanger,
Lane and Spletzer, 1999)
– the evolution of average productivity (e.g. Foster, Haltiwanger and
Krizan, 1998 and 2002)
• Very few studies have looked at the changes of the distribution
of firm productivity over time:
– Dunne, Foster, Haltiwanger and Troske (2002): they find a large
increase in wage and productivity dispersion between 1975-1992 across
US manufacturing plants.
– Haskel and Martin (2002): they look at the changes in productivity
dispersion focusing on UK manufacturing plants.
• This paper tries to carefully document changes in productivity
dispersion over time looking at both manufacturing and nonmanufacturing.
Literature
•
We consider three classes of theories that have tried
to explain the rising in within-group wage
inequality:
1. the first group emphasizes the impact of a technological
revolution that has caused differential adoption of new
technology by firm (e.g. Caselli, 1999)
2. the second group links increased inequality to increased
segmentation of workers by skill (e.g. Kremer and
Maskin, 1996)
3. the third group argues that institutional change (e.g.
union decline) has an impact on wage inequality and thus
potentially on productivity dispersion
Caselli (1999)
• He models the impact of a technological revolution
(e.g. the ICT revolution) on productivity and wage
dispersion.
• Implications:
– High skilled workers will benefit the most (if the
technology is skilled-biased), but all workers in the more
productive plants will tend to receive higher wages,
– A skilled-biased technological revolution leads to an
increase in the dispersion of wages and TFP across plants,
– These increases are modelled as between firm within
industry effects.
Kremer and Maskin (2000)
• They model the simultaneous existence of increased
wage inequality and increased segregation of workers
by skill across plants.
• Implications:
– If the distribution of skills across workers is dispersed &
because of the complementary between tasks, an increase
in the mean skill level leads to a separating equilibrium:
• High skill/high productivity firms and low skill/low
productivity firms can exist in equilibrium,
• Increased wage inequality is associated with increased
labour productivity across firms (non necessarily with
TFP),
• These rises are modelled as between firm with industry
effects.
Union decline
• Union declines allows low wage/low productivity
firms to enter the market.
• Under strong union power, these firms were nonviable because they could not satisfy union demands
over wages.
• The entrance of low wage firms raises wage
dispersion and thus potentially productivity
dispersion.
– If low wage firms employ mostly low skill workers, they
will be characterised by low productivity as well.
Our Strategy
• Descriptive analysis of trends in wage dispersion and
productivity dispersion
– There is limited evidence on the changes in productivity dispersion and ours is
one of the first attempts.
• Decomposition of trends in wage and productivity dispersion
– Is the rising wage dispersion a within- or between-firm phenomenon?
– Is the rising productivity dispersion a between-firm within-industry or
between-industry phenomenon?
• Analysis of Total Factor Productivity (TFP)
– Caselli (1999)’s model implies a positive link between wage and TFP
dispersion.
– Kremer and Maskin (2000)’s model suggests a positive link between wage and
labour productivity dispersion, but not necessarily TFP dispersion.
• Evaluation of the theoretical explanations of these changes
• Cross-country analysis
– Since theories suggest that these are global phenomena, we want to conduct an
international comparison between the UK, Norway, France and the US.
Description of the data
• We use 2 different data sets for the UK:
– Financial Analysis Made Easy (FAME) data set:
• Consolidated company account data for the UK
manufacturing and non-manufacturing firms
• Sample period: 1984-2002.
– New Earnings Survey (NES) data set:
• Employee data taken from the UK National Insurance
Database
• It samples 1 % of the UK population
• Sample period: 1975-1999.
Description of the trends
• the UK:
– Wage dispersion in private services and
manufacturing:
• using data at the individual level (log real annual wage)
• using data at the firm level (log real annual average firm
wage)
– Labour Productivity dispersion in private services
and manufacturing:
• using data at the firm level (log real value added per
worker)
Wage dispersion in the UK:
private services, individual data
Wage Dispersion in the UK
1
1.2
1.4
1.6
1.8
Private services
1984
1989
1994
1999
year
10th percentile (indexed)
90th percentile (indexed)
Log hourly wage: 16-64 men
50th percentile (indexed)
Wage dispersion in the UK:
manufacturing, individual data
Wage Dispersion in the UK
1
1.2
1.4
1.6
1.8
Manufacturing industries
1984
1989
1994
1999
year
10th percentile (indexed)
90th percentile (indexed)
Log hourly wage: 16-64 men
50th percentile (indexed)
Wage dispersion in the UK:
private services, firm level data
Wage Dispersion in the UK
1
1.2
1.4
1.6
1.8
Private services
1984
1989
1994
year
10th percentile (indexed)
90th percentile (indexed)
Log average firm wage
1999
50th percentile (indexed)
Wage dispersion in the UK:
manufacturing, firm level data
Wage Dispersion in the UK
1
1.2
1.4
1.6
1.8
Manufacturing industries
1984
1989
1994
year
10th percentile (indexed)
90th percentile (indexed)
Log average firm wage
1999
50th percentile (indexed)
Productivity dispersion in the UK:
private services, firm level data
Productivity Dispersion in the UK
1
1.2
1.4
1.6
1.8
Private services
1984
1989
1994
year
10th percentile (indexed)
90th percentile (indexed)
Log value added per employee
1999
50th percentile (indexed)
Productivity dispersion in the UK:
manufacturing, firm level data
Productivity Dispersion in the UK
1
1.2
1.4
1.6
1.8
Manufacturing industries
1984
1989
1994
year
10th percentile (indexed)
90th percentile (indexed)
Log value added per employee
1999
50th percentile (indexed)
Summary of the evidence
Using individual level data:
– Wage inequality was increasing in both private
services and manufacturing until the early 1990s;
then it continued to increase in private services but
not in manufacturing.
• Using firm level data:
– Wage inequality as well as productivity dispersion
is increasing substantially in private services.
Limited increase in manufacturing.
Decomposition of trends in wage and
productivity dispersion
• Following Davis and Haltiwanger (1991), we decompose wage
dispersion into between and within firm components:
– Within firm component
– Between firm within industry component
– Between industry component
 (W
j
i
k
 W )   [(Wkij  Wij )  (Wij  W )]
2
kij
j
i
k
2
k=worker
i=firm
j=sector
Where:
- Wkij is the wage for worker k at firm i in sector j,
- Wij is the average wage at firm i in sector j and
- W is the mean wage across all workers in all firms and sectors.
(1)
Variance decomposition
• Multiplying and dividing the first term of the right-hand side of equation
(1) by Nij (Nij = total number of workers k in firm i):
 (W
kij
j
i
 W )   Nij 
2
k
j
i
(Wkij  Wij ) 2
Nij
k
  Nij (Wij  W ) 2
j
(2)
i
(Wkij  Wij ) 2
2
Total
variance
Within
firm
component
Between
firm
component
(
W

W
)

N

N
(
W

W
)



kij
ij 
ij
ij
Nij
j i k
j i
k
j i
(2)

(3)
2
2


2
• Calling (Vij
variance
of wages
Wkij the
 W estimated
) 
N
N ij (Wij across
 W ) workers of firm i,
ijVij 
j
i
k (2) becomes: j
equation
 (W
j
i
k
i
j
i
 W )   N ijVij   N ij (Wij  W )
2
kij
j
i
j
i
2
(3)
Variance decomposition
• Dividing through by N, the total number of workers in the economy, we
obtain the decomposition of overall variation into within-firm (WF) and
between-firm (BF) components:
1
N
1
(
W

W
)


kij
N
j
i
k
2
1
N
V


ij ij
N
j
i
 N
j
ij
(Wij  W )
2
(4)
i
1
VW F    N ijVij , Within firm component
N j i
(5)
2
1
VBF    Nij (Wij  W ) , Between firm component
N j i
(6)
V  VW F  VBF
(7)
Variance decomposition – Between component
• It is possible to decompose the between-firm variation as:
– the sum of total variation of firm average wage relative to the sector
average (VBFI = between firm/within industry)
– the total variation of the sector average wage relative to the average of
the whole economy (VBI = between industry):
1
N
 N ij (Wij  W )  
2
j
i
j
VBFI  
j
VBI  
j
Nj
N
Nj
N
N ij
N
i
Nj
N
N ij
N
i
(Wij  W j ) 2 
j
(W j  W ) 2
1
N
(Wij  W j ) 2  
j
j
 N
j
ij
Nj
N
(W j  W ) 2
(Wij  W j ) 2
(8)
(9)
i
VBF = VBFI + VBI
V= VWF + VBFI + VBI
(10)
• The variance decomposition of the between component is also applied to
firm productivity
Variance decomposition of yearly wages
•
We combine the individual level data from the NES with the firm level
data from FAME at the 2-digit industry level:
– We estimate total wage variance (V) directly from the NES.
– We estimate the between components (VBFI & VBI) directly from FAME.
– The within firm component (VWF) can be estimated as a residual.
•
FAME data give a measure of total annual remuneration at the firm level
while NES data provide a measure of weekly gross pay. In order to make
the measures comparable:
1. We adjust total annual remuneration by a coefficient (0.92) taken from a
regression of log wage on log remuneration with industry, firm size and year
dummies.
2. We express both measures at the annual level and in real terms (deflated by the
RPI)
Variance decomposition of yearly wages
Decomposition of wage dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK manufacturing and private services
1984
1989
1994
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between component
1999
Variance decomposition of yearly wages
further decomposition of the between component
Decomposition of Wage Dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK manufacturing and private services
1984
1989
1994
1999
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between firm within industry component
between industry component
Variance decomposition of yearly wages
UK private services
Decomposition of Wage Dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK private services
1984
1994
1989
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between component
1999
Variance decomposition of yearly wages
UK private services
Decomposition of Wage Dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK private services
1984
1989
1994
1999
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between firm within industry component
between industry component
Variance decomposition of yearly wages
UK manufacturing
Decomposition of Wage Dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK manufacturing industries
1984
1989
1994
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between component
1999
Variance decomposition of yearly wages
UK manufacturing
Decomposition of Wage Dispersion
0
.1 .2 .3 .4 .5 .6 .7 .8
UK manufacturing industries
1984
1989
1994
1999
year
total variance
within firm component
NES and FAME Datasets 1984-1999.
between firm within industry component
between industry component
Variance decomposition of labour productivity
Decomposition of Labour Productivity Dispersion
0
.1
.2
.3
.4
.5
.6
UK manufacturing and private services
1984
1989
1994
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Variance decomposition of labour productivity
UK private services
Decomposition of Labour Productivity Dispersion
0
.1
.2
.3
.4
.5
.6
UK Private Services
1984
1994
1989
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Variance decomposition of labour productivity
UK manufacturing industries
Decomposition of Labour Productivity Dispersion
0
.1
.2
.3
.4
.5
.6
UK Manufacturing Industries
1984
1989
1994
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Summary of the results
• Evidence indicates that wage inequality and labour
productivity dispersion have increased in the UK over the
period 1984-99.
• The rises in wage dispersion and in productivity dispersion are
driven by increases in private services.
• We find that rising wage dispersion is largely a between-firm
within-industry phenomenon rather than a within firm
phenomenon.
• We find that rising labour productivity dispersion is largely a
between-firm within-industry phenomenon rather than a
between-industry phenomenon.
• These findings appear to give some support to explanations
(e.g. Caselli, 1999, Kremer and Maskin, 2000) that link rising
wage inequality to between firm effects.
Variance decomposition of TFP
We define TFP simply as:
TFP  ln(VA)  ln(L)  (1   )ln(K)
1. α = 0.7 as average share of labour (L) in the economy,
2. α = industry specific weights (i.e. ratio of total wage bill over
value added at the 2-digit industry level),
3. α = firm specific weights (i.e. ratio of total wage bill over
value added at the firm level).
A preliminary analysis uses α = 0.7.
Variance decomposition of TFP
Decomposition of TFP Dispersion
0
.1
.2
.3
.4
.5
.6
UK manufacturing and private services
1984
1989
1994
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Variance decomposition of TFP
UK private services
Decomposition of TFP Dispersion
0
.1
.2
.3
.4
.5
.6
UK private services
1984
1989
1994
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Variance decomposition of TFP
UK manufacturing industries
Decomposition of TFP Dispersion
0
.1
.2
.3
.4
.5
.6
UK manufacturing industries
1984
1989
1994
1999
year
between firm within industry comp.
UK FAME Dataset 1984-1999. NES weights.
between industry comp.
Summary of the results - TFP
• As for wage and labour productivity dispersion, the
rise in TFP dispersion is mostly driven by increases in
private services.
• We find that rising TFP dispersion is largely a
between-firm within-industry phenomenon rather
than a between-industry phenomenon.
• However, the between-industry component seems
more important in explaining TFP dispersion rather
than labour productivity dispersion.
• These findings appear to give some support to Caselli
(1999)’s model.
Next steps
• Distinguishing between theories:
– Caselli(1999) versus Kremer and Maskin(2000)
– Institutional change versus Caselli(1999)
• Measurement error issues:
– We do not have information about skills, gender and parttime/full-time workers at the firm level. Increased wage
dispersion might be due to:
• Changes in the composition of workers in the firm
• Higher female participation
• Larger share of part-time workers
– We use remuneration instead of hourly wages
– Analysis of TFP: we do not measure all inputs properly
– Using employer-employee data on Norway we can deal
with some of these issues
Institutional change versus Caselli(1999):
a simple framework
• UK/US: union decline and reduction in the share of workers covered by
collective agreements
• France/Norway: union decline but very high share (95%in france) of
workers covered by collective agreements
UK/US
Caselli (1999)
Institutional
change
France/Norway
σw ↑
σw -
σp ↑
σp ↑
σw ↑
σw -
σp ↑
σp -
Caselli’s channel is
from productivity to
wages.
Institutional channel
goes from changes
in wage to changes in
productivity.