Resolution Foundation held an event on job polarisation with guest speakers Craig Holmes, Economist at Pembroke College, Oxford and Andrea Salvatori, Research Fellow at the University of Essex.
1. Looking through the hourglass: Hollowing out of the
UK jobs market pre- and post-crisis
LauraGardiner, Resolution Foundation
Adam Corlett, Resolution Foundation
Craig Holmes, Pembroke College,Oxford
Andrea Salvatori, University of Essex
MatthewWhittaker, Resolution Foundation (chair)
Wifi: AvantaGuest // Passcode: LondonWorkspaces
#futurejobs // @resfoundation
2. Looking through the hourglass
Hollowing out of the UK jobs market
pre- and post-crisis
Laura Gardiner & Adam Corlett
March 2015
@resfoundation
3. • A large and growing body of research details the ‘hollowing out’
of developed labour markets
• Previous Resolution Foundation research has confirmed that these
trends continued in the UK in the early years of the crisis
• We update this picture to 2014, and discuss UK trends in the
context of broader debates on polarisation
3
Has the UK’s job structure polarised pre- and
post-crisis?
4. Since the early 1990s, mid-skilled occupations
have experienced falling employment shares
Using initial wages as
a proxy for skill levels,
mid-skilled
occupations have
declined 1993-2014
and high-skilled
occupations have
grown, with smaller
changes in low-skilled
occupations.This
leads to a ‘U-shaped’
graph
The picture is similar
when looking at hours
or headcount
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
4
5. Since the early 1990s, mid-skilled occupations
have experienced falling employment shares
We summarise the
trends in different
parts of the
occupational skill
distribution by
grouping together
skill deciles 1 and 2
(low-skilled), 3 to 7
(mid-skilled), and 8
to 10 (high-skilled)
5
Low-skilled Mid-skilled
High-skilled
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
6. Low-skilled occupations were growing in share
in the mid-1990s, but then declined
Low-skilled jobs
declined in share
through the late-
1990s and early
2000s, and have been
broadly flat since
6
Notes: The first quarter of 2001 and the final quarter of 2014 are not included due to missing variables or because data was
not available at the time of analysis. See annex for other methodological details. Source: Resolution Foundation analysis of
Labour Force Survey, ONS
7. The downturn may have ‘amplified’
polarising trends
Updating our starting
point to 2002 (to
reflect a decade of
changes to the
occupational wage
structure) gives a
similar picture
The crisis shows a
potential return to the
trends of the mid-
1990s, with growth in
high-skilled jobs,
slight growth in low-
skilled jobs, and
sharper relative
decline in mid-skilled
ones.These trends
then slow
7
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
8. The self-employed skew the picture slightly
towards low-skilled jobs
When including the
self-employed, we
find that low-skilled
jobs expanded
slightly, and high-
skilled jobs grew
slightly more slowly,
between 2002 and
2014
8
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
9. So what are these declining mid-skilled jobs?
manual trades and mid-skilled office workers…
The two occupations
experiencing the
largest decline in their
share of employment
since 1993 are
‘process, plant and
machine operatives’
and ‘secretaries’
There has been strong
growth in caring and
service occupations
across the
occupational wage
distribution, some of
which may reflect
demographic changes
9
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. Bubble size reflects the
average labour share between 1993 and 2014. See annex for other methodological details. Source: Resolution Foundation
analysis of Labour Force Survey, ONS
10. With similar trends enduring during the crisis
and recovery
The employment
share of construction
occupations declined
sharply after 2007 (in
contrast to the longer-
run view), likely
reflecting the collapse
in demand for these
skills during the crisis
10
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. Bubble size reflects the
average labour share between 2002 and 2014. See annex for other methodological details. Source: Resolution Foundation
analysis of Labour Force Survey, ONS
11. It is often assumed
that a polarising labour
market has been the
main driver of rising
wage inequality – with
more low- and high-
paid occupations
increasing the gulf
between the two
However, research has
demonstrated that
while a shift in theUK’s
job structure has
played a role in lower
wage growth for low-
and middle-earners,
this is only one part of
the story
11
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. 1993 analysis based
on SOC 1990 (3-digit); 2014 analysis based on SOC 2010 (4-digit).See annex for other methodological details. Source:
Resolution Foundation analysis of Labour Force Survey, ONS
But does this matter?There is limited evidence
of job polarisation driving wage polarisation
12. What lies behind hollowing out? Strong links to
the automation (or offshoring) of routine jobs
What do mid-skill jobs
have in common?
‘Routineness’ and
‘offshorability’ scores
assigned to each
broad occupation
group are a good
predictor of changes
in employment share
The strongest relative
declines in manual
trades and some
office jobs attest to
this – these are the
roles most at threat
from computerisation
12
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
13. But how do ‘routineness’ and hollowing out
relate? Low-/ mid-skill jobs are more routine...
As a mirror image of
falling employment
shares, jobs of above-
average ‘routineness’
are concentrated in
the middle and
bottom of the pay
distribution
13
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
14. …And it is these routine jobs which have been
lost, particularly from the middle
And the employment
share of these routine
jobs has fallen over
time, with the largest
absolute falls in the
middle, helping
explain the earlier ‘U-
shape’
14
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
15. However, there is some evidence that higher
paying routine jobs are the most at risk
But there is some
very tentative
evidence that middle
to high paying
routine jobs are most
at risk
Low paying routine
jobs will – all else
equal – be less
profitable to
automate, though
this theory requires
further exploration
15
Notes: The final quarter of 2014 is not included because data was not available at the time of analysis. See annex for other
methodological details. Source: Resolution Foundation analysis of Labour Force Survey, ONS
16. • The ‘rise of the robots’ hasn’t yet harmed overall employment
• And if ‘routine-biased technological change’ was the only factor
behind polarisation we would expect to see corresponding wage
polarisation (wages changing in line with employment shares) –
which we don’t
• Supply-side factors are also likely to be important – including
upskilling of the workforce, as explored by others
• As well as other more localised factors – such as demographic
changes and the cyclical collapse in the construction industry
16
Don’t just blame the robots – technology is not
the only factor in occupational polarisation
17. • An expanded slide pack and blog will be available on our
website following this event
• Further work to explore the real-world implications and what
the near future may hold
• Using the latest data, and potentially new ways of measuring
routineness
• A report later this year as part of our New Labour Market
research programme
17
Next steps
18. The anatomy of job
polarisation in the UK
Andrea Salvatori
University of Essex
23 March 2015
Resolution Foundation
London
@iseressex
@andysalvatori
19. @andysalvatori
@iseressex
The (mostly US-focused) literature supports a demand-centred story:
- Middling “routine” jobs easier to automate
- All education groups have lost shares in middling jobs
- Polarisation of occupational wages in 1990s
- Over time, stronger growth at the bottom
- No growth at the top the 2000s
Is polarisation in the UK different?
Among employees, since 1980:
% graduates triplicated
% immigrants doubled
Is there a role for these supply-side changes?
Polarisation and computerisation
20. @andysalvatori
@iseressex
Job polarisation in each decade, 1979-2012
-20
-15
-10
-5
0
5
10
15
20
1980s 1990s 2000s 1979-2012
Changeinemploymentshare
Bottom deciles (1-2) Middle deciles (3-8) Top deciles (9-10)
Occupational deciles based on the 1979 ocucpational median wage.
Growth at the top always larger than at the bottom:
Top has gained 16pp of the 19pp lost by middle
21. @andysalvatori
@iseressex
Polarisation is a non-graduate phenomenon (1979-2012)
3.9
9
15
3.1
8.3
16.6
0.8 0.7
-1.5
Graduates
Total contribution (1)+(2)
Explained by change in relative size of group (1)
Explained by reallocation across occupations (2)
-0.4
-28.3
0.7
-12 -15.5
-0.4
11.6
-12.7
1.2
Bottom Middle Top
Non-Graduates
Compositional changes:
- >50% non-graduate decline in
middle
- 100% graduate increase at the
top
Changes within groups:
- Non-grads moved to the bottom
At the bottom:
- Net growth is grads
- But reallocation of non-grads
offsets decline from educational
improvement
22. @andysalvatori
@iseressex
2000s: graduates and immigrants more important
Bottom occupations:
• Education upgrading continues while
• Reallocation of non-grads slows down
• Graduates shift towards the bottom
• Number of immigrants increases
Native graduates and immigrants are main contributors to growth of
bottom occupations
Overall contribution of natives is negative: in the aggregate,
educational upgrading stronger than reallocation to bottom.
Contribution of immigrants not limited to bottom occupations:
• (Graduate) immigrants account for 35% of growth at the top (up from 16% in
1990s)
23. @andysalvatori
@iseressex
Occupational wages have not polarised in any decade
No evidence of decline in wages in middling occupations in any decade
Performance of median wages in top occupations deteriorates over time –
and it is worst in the 2000s.
Points to importance of supply at the top
24. @andysalvatori
@iseressex
So, polarisation in the UK is different from the US
The findings on
1) importance of educational upgrading
2) occupational wages
are not consistent with a simple demand-based story and suggest that
supply-side changes played an important role in the UK.
particularly in the 2000s when growth at the top stalled in US
Impact of technology on labour market more complex than often
suggested.
Technology is certainly important, but it is its interaction with the skill
structure of the workforce that determines what happens to the quality and
quantity of jobs.
25. HAVE UK EARNINGS DISTRIBUTIONS
POLARISED?
Dr Craig Holmes
Research Fellow, Employment, Equity and Growth Programme, Institute for New
Economic Thinking at the Oxford Martin School
March 23rd 2015
Resolution Foundation
26. Introduction
• Polarisation towards high wage and low-wage work implies
increasing inequality
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
30%
35%
40%
0 0.2 0.4 0.6 0.8 1
Wagegrowth
Percentile
1987-2001 1994-2007 2006-2013
27. Introduction
• To put this in a different way:
– Low wage work: hourly wage < 2/3 median hourly wage
– High wage work: hourly wage > 1.5x median hourly wage
Low wage work High wage work
1987 20.2% 23.4%
2001 23.0% 25.6%
1994 22.6% 25.2%
2006 21.4% 25.6%
2013 22.3% 26.3%
28. Introduction
• Questions:
1. How important has the change in the occupational structure –
“hollowing out” - played in these trends?
2. Why has “hollowing out” not always accompanied increased pay
dispersion?
• In both cases, the structure of wages within occupations is key
29. Overview of approach
• Typical (OLS) regression predicts the mean of a variable (say,
wt), conditional on the explanatory variables, Xt:
• From this, we can calculate the unconditional mean of the
whole distribution:
• Changes in the mean wage over time can be broken down into
‘compositional effects’ and ‘wage effects’:
tttt Xw
ttt Xw
10101001 XXXww
30. Overview of approach
• Here, essentially doing the same thing except looking at
changes in wages at different points of the distribution
instead of the mean
• The approach I follow is that of Firpo, Fortin and Lemieux
(2009).
• Three time periods
• Real hourly wages
• Explanatory variables: occupational groups, education levels,
union membership, gender, part-time status