Centre for Market and
Public Organisation
Can pay regulation kill? Panel data evidence on
the effect of labor markets on hospital
performance
Emma Hall, Carol Propper John Van Reenen
Feb 2008
1
Motivation
• Unintended consequences of wage regulation
– Pay setting (e.g. public sector) often has
“geographical equity” despite different local labor
markets. Implies problems of labour supply - and
poor performance - when outside labour mkts strong
• How do labour markets affect firm performance?
– Hard to identify as wages reflect equilibrium
outcomes of demand and supply shocks.
– In our design, pay regulation help identification
• Policy issue in hospital performance
– What are causes of large performance variation (note
also large productivity dispersion in other industries)
2
Our Design
• Wages for nurses (and doctors) in UK National Health
Service centrally set by National Pay Review Body.
NPRB “Mandates” wage rates for doctors and nurses by
grade. Uprated each year.
• Very little local variation in regulated pay despite
substantial local variation in total private sector
– E.g. 65% private sector pay gap between North-East
England and Inner London but only 11% in NPRB
regulated pay
– Use exogenous variation in “outside wage” and
examine impact on hospital outcomes (quality, prody)
• Institutional setting one in which selection of patients to
hospitals is limited
3
Our Results
• Main Finding: Hospitals in high outside wage
areas have lower hospital quality (higher AMI
death rates) and lower output per head.
• Not result of general UK labour market
conditions
– Placebo experiments on similar sectors: no evidence
of negative effect of outside wages on productivity
• One mechanism: greater reliance on lower
quality temporary/agency staff.
4
Geographical variation in:
Outside wages Agency nurses In-hosp AMI
deaths
5
OUTLINE
1. Models: What is the effect of pay regulation?
2. Empirical models
3. Data
4. Results
5. Conclusions
6
1. Effects of high outside wage
relative to regulated wage
• Employers
– try to circumvent by “over-promoting” (grade drift) and increasing
non-wage benefits. Limited by regulation/union enforcement
– Substitution to other factors: health care assistants, maybe
capital. But limited due to nature of needed expertise.
– Substitute temporary agency staff. Lower job-specific human
capital so less productive/lower quality (cf Autor & Houseman,
2006)
• Employees
– Lower participation, higher vacancies for permanent staff
– More likely to become agency staff.
– Permanent staff also less motivated, lower relative quality
compared to low outside wage areas
Implication: Worse hospital performance in high outside wage areas
7
Implications
• In high outside wage areas
– Problems of labour supply for permanent staff
• higher vacancies
• lower participation in nursing
• Greater reliance on agency nurses
– Worse health outcomes
• Lower quality (AMI death rate)
• Lower productivity
– See this in raw data at regional level
8
2. Empirical Models
1. Hospital quality equation
dit 1d Sit 2d Sit
PHYS NURSES
d wit d wit d zit id rid id it
O d d
For hospital i in year t:
d = 30 day death rate from emergency AMI admission for 55+ year
olds
SPHYS = share of clinical workforce who are physicians
SNURSES= share of clinical workforce who are nurses (and AHPs)
(base group is health care assistants)
wO = ln(outside wage)
Z = controls for casemix, area mortality rates, hospital size,
teaching status
w = ln(inside wage)
η = hospital dummies
τ = time dummies, r=regional dummies 9
2. Hospital productivity equation
ln(Y / L)it 1 Sit 2 Sit
PHYS NURSES
wit wit zit i ri i it
O
Ln(Y/L) = ln(Finished Consultant Episodes per clinical worker)
SPHYS = share of clinical workforce who are physicians
SNURSES= share of clinical workforce who are nurses (and AHPs)
(base group is health care assistants)
wO = ln(outside wage)
Z = controls for casemix, area mortality rates, hospital size,
teaching status
w = ln(inside wage)
r = regional dummies
τ = time dummies
η = hospital dummies
10
3. Placebo productivity equation
ln( R / L)it 1 Sit
QUAL
wit wit zit i ri i it
O
Ln(R/L) = ln(revenues/worker)
SQUAL = share of workforce who are qualified (nursing homes: with
nursing quals; ln (cap/labor) ratio other industries)
wO = ln(outside wage)
Z = total staffing (+ gender mix, age of staff for nursing homes)
w = ln(inside wage)
r = regional effects
τ = time dummies
η = firm fixed effect
Run for 42 industries + nursing homes
11
Issues
• Unobserved heterogeneity: OLS, long
differences and “System GMM”
• Endogeneity of wages and shares:
– Outside wage: hospitals are a small % of local
labor market
– Skill shares: GMM-SYS (Blundell-Bond,2000;
Bond and Soderbom, 2006)
• Standard errors allow for heteroscedacity,
autocorrelation and clustering by region
12
Issues
• Endogeneity of patient quality
– Selection of hospitals
– Association of illhealth and economic activity
• Hospital selection limited by inst. structure
– AMI patients sent to nearest hosp.
– Hospitals not monitored on quality; in theory financial
incentives exist but no systems to implement
• Upswings less associated with increase in hrs
(due to higher labor protection); also undertake
extensive checks to ensure no rel. between
community health and ‘good times’
13
3. Data
• Hospital level panel data
• 3 groups of clinical workers: Physicians, nurses
(AHPs) and Health Care Assistants. Total
employment. From Medical Workforce Statistics
• Agency staff – hospital financial returns
• Hospital quality: 30 day in-hospital death rates
for Emergency admissions for Acute Myocardial
Infarction (AMI) for over 55 year olds. From HES
(Hospital Episode Statistics).
• Productivity: Finished Consultant Episodes
(HES) per worker
14
Wage Data
• Outside wage
– New Earnings Survey (NES) 1% sample of all
workers
– Use travel to work area (78 in England)
– Compare results with 9 main regions
– Female non-manual wage
• Inside Wage
– Average wage in hospital (but can just reflect grades)
– Predicted wage based on NPRB regulation including
regional allowances (Gosling-Van Reenen, 2006)
15
Final Dataset
• 211 hospitals between 1996-2001
• 907 observations
16
OUTLINE
1. Models: What is the effect of pay regulation?
2. Empirical models
3. Data
4. Results
5. Conclusions
17
Table 2: Death Rates from AMI
Dependent variable Ln(AMI Death Rate) Ln(AMI Death Rate) Ln(AMI Death Rate)
Estimation technique OLS 3 year annual Long GMM-SYS
Differences
(1) (2) (3)
Ln (Area outside wage) 0.407*** 0.766** 0.460***
(0.124) (0.386) (0.175)
Physicians share -0.856*** -0.654 -2.629**
(0.316) (0.616) (1.258)
Qualified Nurses share -0.480** -0.288 -1.416
(0.227) (0.467) (0.959)
(omitted base is unqualified nurses/ health care assistants)
Hospital fixed effects No No Yes
Casemix controls (14) Yes Yes Yes
Year dummies (6) Yes Yes Yes
Region dummies (10) Yes No Yes
SC(1) p-value 0.000
SC(2) p-value 0.142
Hansen-Sargan p-value 0.923
No of Hospitals 210 133 210
Observations 901 345 901 18
Magnitudes (col 3)
• From 90th to 10th percentile of area outside wage
difference is a fall of 33%. Associated with
– a 14% fall in death rates (a quarter of the 62% 90-10 spread)
• Increase in physician share from 10th to 90th percentile
is 7 percentage points. Associated with
– 37% fall in AMI death rates (60% of 90-10 diff)
• Effect on AMI death rates of outside wage not dissimilar
magnitude to drug based medical interventions (aspirin,
beta blockers)
– 10% increase in outside wages leads to 1 pp increase in AMI
fatality
– Heidenrich and McClellan (2001) increase use of aspirins by
70% resulted in 3.3 p.p fall in AMI mortality
19
Table 3: Productivity (FCEs per employee)
Dependent variable Ln(Productivity) Ln(Productivity) Ln(Productivity)
Estimation technique OLS 3 year annual Long GMM-SYS
Differences
(1) (2) (3)
Ln (Area outside pay) -0.662*** 0.252 -0.551***
(0.145) (0.279) (0.181)
Physicians share 3.837*** 0.248 3.909***
(0.360) (0.411) (0.898)
Nurses share 0.386* 0.006 1.736***
(0.201) (0.216) (0.627)
(omitted base is unqualified nurses/health care assistants)
Hospital fixed effects No No Yes
Casemix controls (39) Yes Yes Yes
Year dummies (6) Yes Yes Yes
Region dummies (10) Yes No Yes
SC(1) p-value 0.004
SC(2) p-value 0.462
Hansen-Sargan p-value 0.042
No of Hospitals 210 133 210
Observations 901 345 901 20
Placebo tests
• Nursing homes
– Provide medical care and other care services
to elderly
– Wages not regulated
– 649 randomly selected homes: data for 1998
and 1999
– No evidence from OLS regression that
outside pay associated with lower output
(beds) per hour of staff time
21
Other placebo tests
• 42 service industries
• Dependent variable ln(revenues/worker)
• Only in 7/126 regression was outside
wage neg. and significant
• Inside wage significant in almost all
• Suggests our finding of neg. effect of
outside wages is a result of regulated pay
maxima
22
A possible mechanism: Agency nurses
• Higher outside wages associated with
significantly greater use of agency staff
• Doubling of agency staff increases AMI death
rates by 5%; no remaining effect of outside
wages
• Agency nurses disproportionately in A and E
wards
• Less effect on outside wages in productivity
equation, but agency use still significant
• Use of agency staff related to MRSA rates (for
2001-2002)
23
Robustness checks
Upswings lead to poorer health in local labour market (e.g.
Ruhm)
• Case-mix and local wages
– AMI severity (HRG category) not related to outside wages
– controls for HRG not significant for AMI deaths; total case-mix
not significant for prody
• Are outside wages associated with higher community
death rates?
– Our model implies weakly so
– Ruhm type argument – strong positive relationship
– We find weak n.s. positive relationship
– Also find no relationship between two key drivers of poor health-
upswing relationship (pollution, smoking)
24
Robustness checks
Outside labor market affecting ambulance care
• More economic activity – slower road speeds (‘floor to
door’)
– Control for ambulance speeds
• Poorer quality of ambulance crew (door to needle time’)
– Ambulance crew have no autonomy over which hospital to go to;
administration of reperfusion (to stop clotting) by crews under
0.6%.
Other tests
– Financial pressure
– Dynamics
– Regional heterogeneity in impact outside wage
25
Conclusions
• Regulated pay costs lives (and productivity) in
high outside wage areas
– Higher death rates (and lower productivity) in areas
where labour markets are tight
– Some of this affect seems to operate through greater
reliance on temporary agency staff
– Not a feature of other UK service industries where
(maximum) pay regulation does not operate
• Labour markets important for health on supply
side of medical care as well as demand side
• Policy solution – allow wages to reflect local
labour market conditions?
26
Back Up Slides
27
Next Steps
• Other explanations – e.g. technology
adoption (Acemoglu and Finkelstein,
2006)?
28
Underlying structural model
• Hospitals choose mix of factors depending on
environment and adjustment costs
• Factor with high adjustment costs changed more
slowly
• Implies that lagged values predict future values
• Empirical identification requires that adjustment
costs be sufficiently different across the factors
to avoid weak instruments problems
29
System GMM
Equation of interest
yit xit ait ; ait tt i uit
1) Difference equation eliminates firm fixed effects
Moment conditions allow use of suitably lagged levels of the variables as
instruments for the first differences (assuming levels error term serially
uncorrelated, see Arellano and Bond, 1991)
E[ xi ,t s uit ] 0
for s > 1 when uit ~ MA(0), and for s > 2 when uit ~ MA(1), etc.
Test assumptions using autocorrelation test and Sargan
Problem of weak instruments with persistence series…..
30
System GMM
2) Use lagged differences as instruments in the levels equation
additional moment conditions (Arellano and Bover, 1998; Blundell and Bond, 2000):
E[xi ,t s (i uit )] 0
for s = 1 when uit ~ MA(0), and for s = 2 when uit ~ MA(1)
Requires first moments of x to be time-invariant, conditional on common year dummies
Can test the validity of the additional moment conditions
We combine both sets of moments for difference and levels equations to
construct “System GMM” estimator
We assume all firm level variables are endogenous, while industry level variables are
exogenous in main specifications (relax in some specifications)
31
Alternative to regulation
• Avoiding permanent pay increases
(Houseman et al, 2003)
– Pay more observable than in US
– Differences in pay and quality across regions
are persistent
32
Big spread in productivity between hospitals (Fig 3)
Note: productivity measured by finished consultant episodes per worker 33
Sample characteristics
Mean Standard deviation Min Max
AMI Variables
AMI death rate (55 plus) 21.14 4.483 7.454 36.941
Total AMI deaths (55 plus) 79.99 33.83 13 294
Total AMI admissions (55 plus) 385.02 160.84 151 1,348
Productivity and FCE (finished Consultant Episodes)
Productivity (total FCEs/ total clinical staffing) 31.17 7.57 12.09 65.12
Total FCEs 58,664.58 24,515.83 13,490 138,984
Staffing Variables
Total clinical staffing (physicians + nurses + Allied 1675.79 692.25 398.61 4010.70
Health Professionals + Health Care Assistants)
Physicians share of staffing 0.148 0.030 0.058 0.270
Qualified Nurses (plus qualified Allied Health 0.597 0.037 0.476 0.741
Professionals) share
Health Care Assistants share 0.246 0.046 0.121 0.393
Hospital Expenditure Variables
Share of expenditure on agency staff as a proportion of 0.034 0.028 0.001 0.163
total expenditure (“Agency”)
Retained Surplus (£K) (745 obs) -206.1 1313.4 -11487 8505
34
Sample characteristics cont
Wages
Ln(Area outside wage) 9.60 0.140 9.27 9.99
Ln(nurse inside wage) 9.99 0.152 9.52 10.50
Ln(area inside wage) 10.09 0.110 9.53 10.45
Other variables
Directly Standardized Mortality rate in local area (per 723.43 77.13 518.73 944.21
100,000)
Teaching trust 0.111 0.341 0 1
Proportion of emergency admissions (to total admissions) 0.411 0.082 0.224 0.808
Proportion of transfer admissions (to total admissions) 0.160 0.066 0 0.448
Proportion of AMI admissions with HRG code E11 .162 0.075 0 0.667
HRG case mix index (892 obs) 93.98 9.08 75.49 175.89
MRSA rate (216 obs) 0.169 0.088 0.02 0.55
35
Large spread in death rates from AMI between hospitals
Worst 10%
Best 10%
• Improvements over time (cf. TECH Investigators)
• 1996: 10 percentage point (60%) difference between top and bottom (90th =27%,10th =17%)
36
Simple model
• 2 areas: high outside wage “South” and
low outside wage “North”
• Regulated wage the same in both areas
• Regulated wage lower than equilibrium
wage
37
Wages
Labour Supply,
South
Labour Supply,
North
Labor Demand
Regulated Wage
NSOUTH NNORTH N, employment
38
Wages
Labour Supply,
South
Labor Demand
Regulated Wage
NSOUTH N, employment
39
Wages
Labour Supply,
South
Labor Demand
Agency Wage
Regulated Wage
Agency staff
NPERMANENT NTOTAL N, employment
40
Table 4: Controls for inside wages
Dependent Ln(AMI Ln(AMI Ln(AMI Ln(AMI Ln(Producti Ln(Producti Ln(Producti Ln(Producti
variable Death Rate) Death Rate) Death Rate) Death Rate) vity) vity) vity) vity)
Estimation OLS Long GMM-SYS GMM-SYS OLS Long GMM-SYS GMM-SYS
technique Differences Differences
(1) (2) (3) (4) (5) (6) (7) (8)
Ln (Area 0.406*** 0.765** 0.431** 0.431** -0.659*** 0.244 -0.547*** -0.548***
outside (0.122) (0.384) (0.172) (0.172) (0.144) (0.282) (0.172) (0.180)
pay)
Average -0.286*** -0.126 -0.334** 0.071 0.097 0.241**
inside wage (0.101) (0.161) (0.168) (0.115) (0.128) (0.125)
Predicted -0.371 0.264
ln(inside (0.716) (0.342)
wage using
NPRB IV)
Physicians -0.498 -0.544 -1.787 -2.145* 3.750** 0.201 4.130*** 3.979***
share (0.342) (0.641) (1.236) (1.286) (0.390) (0.394) (0.930) (0.904)
Nurses -0.313 -0.253 -0.910 -1.002 0.347* 0.004 1.680*** 1.734**
share (0.224) (0.471) (0.822) (0.856) (0.207) (0.212) (0.607) (0.628)
(omitted base is unqualified nurses/health care assistants)
SC(1) p- 0.000 0.000 0.002 0.004
value
SC(2) p- 0.162 0.173 0.436 0.485
value
Hansen- p- 0.795 0.716 0.81 0.32
value
41
Table 6: Placebo experiments: nursing homes
Dependent Ln(revenues/hour) Ln(revenues/hour) Ln(revenues/hour) Ln(revenues/hour) Ln(output/hour)
variable
Estimation OLS OLS OLS OLS OLS
technique
(1) (2) (3) (4) (5)
Ln (Area outside -0.009 0.095 0.125 -0.084 -0.075
pay) (0.191) (0.171) (0.364) (0.228) (0.201)
Ln(Inside Pay) 0.166*** 0.179*** 0.166*** 0.179*** 0.049*
(0.031) (0.030) (0.037) (0.044) (0.028)
Ln (average hours) -0.466***
(0.056)
Nursing Home No No Yes Yes No
fixed effects?
Year dummies (1) Yes Yes Yes Yes Yes
Region dummies Yes Yes Yes Yes Yes
(10)
Number of 649 649 443 513 649
Nursing Homes
Observations 1,054 1,054 886 513 1,068
42
A possible mechanism: Agency nurses
43
Figure 5: Agency Nurses, outside wages and AMI death rates
Dependent Ln(Agen Ln(AMI) Ln(AMI) Ln(AMI) Ln Ln Ln
variable cy) (productivi (productivi (productivi
ty) ty) ty)
(1) (2) (3) (4) (5) (6) (7)
Ln (Area 2.851** 0.314* 0.175 -0.805*** -0.729***
outside pay) (1.138) (0.170) (0.202) (0.182) (0.194)
Ln(Inside 0.077 -0.494*** -0.477*** 0.219 0.296**
Pay) (1.045) (0.153) (0.161) (0.134) (0.141)
Ln(Agency) 0.057** 0.046* -0.106*** -0.057***
(0.026) (0.024) (0.027) (0.018)
No. of 176 176 176 176 176 176 176
hospitals
Observations 523 520 520 520 520 520 520
44
All regressions include hosp fixed effects, region dummies, year effects.
Robustness checks: coefficient on outside wage
Dependent variable Ln(AMI) Ln(Productivity) Obs.
(1) (2) (3)
1 Baseline 0.460** -0.547*** 901
(0.175) (0.172)
2 Additional casemix controls 0.427*** -0.556*** 900 (for AMI)
(0.170) (0.153) 892 (for prody)
3 Include hospital financial surplus 0.399** -0.516*** 745
(0.182) (0.184)
4 Include lagged dependent variable: long-run [p- 0.508*** -0.572*** 901
value] [0.008] [0.020]
5 Drop Inner and Outer London 0.304** -0.383** 776
(0.156) (0.173)
6 Drop big jumps in outside wage 0.530** -0.622*** 885
(0.197) (0.167)
7 Balanced Panel 0.600*** -0.612*** 582
(0.207) (0.163)
8 Regional outside wage 0.609 -0.445 901
(1.022) (0.587)
9 Regional outside wage (drop regional dummies) 0.520*** -0.493** 901
(0.172) (0.169)
10 Include alternative total hospital employment 0.404** -0.540** 901
measure (0.160) (0.170)
11 Include higher order and cross product terms in 0.541*** -0.637*** 901
skill shares (0.200) (0.181) 45
Cost effectiveness
• Effect on AMI death rates of outside wage not dissimilar
magnitude to drug based medical interventions (aspirin,
beta blockers)
– 10% increase in outside wages leads to 1 pp increase in AMI
fatality; Heidenrich and McClellan (2001) increase use of
aspirins by 70% resulted in 3.3 p.p fall in AMI mortality
• Cost of a life year saved by an 1% increase in (inside)
nurse wages to all staff and an 1 p.p. increase in
physician and nurses skill shares
– Increasing inside wages: $100,000
– physician share: $60,000
– nurse share: $36,000
– Value of QALY c $60,000
• Comparison with greater use of drug based medical
technology, increasing wages for nurses and skill shares
in hospitals expensive, but cheaper than the current cost
of AMI treatment in the US (Skinner et al 2006) 46
Higher nurse vacancy rates1 in stronger labor markets (fig 4)
Vacancy Rates for nurses predicted vacancy rate
5
Inner Lo
4 Outer Lo
South Ea
3
2
West Mid
East of
South We
Yorkshir
East MidNorth We
1
North Ea
9.4 9.6 9.8 10
mean ln(outside wage)
1 Percentage of nurse posts that have been vacant for 3 months or more
47
Higher use of agency nurses in stronger labor markets (Fig 6)
Intensity of using agency nurse predicted Agency rate
6
Inner Lo
4 Outer Lo
2 South Ea
East of
West Mid
South We
North We
East Mid
Yorkshir
North Ea
0
9.4 9.6 9.8 10
mean ln(outside wage)
48
Higher death rate from AMI admissions in stronger labor markets (fig 7)
AMI Rate AMI = 1.96*W -0.10W2
Yorkshir
23
Inner Lo
Outer Lo
North We
South Ea
22
South We
East of
21
North Ea Mid
East
20
West Mid
9.4 9.6 9.8 10
mean ln(outside wage)
49
Changes in AMI death rates and changes in outside wages
AMI growth pa 1996-2001
0
West Mid
East Mid
Outer Lo North Ea
South Ea
-.02
Inner Lo
Yorkshir
North We South We
-.04
East of
-.06
.044 .046 .048 .05 .052
av. outside wage gr 1996-2001
50
Magnitudes
• From 90th to 10th of area outside wage
difference is a fall of 33%, associated with:
– a 16% increase in productivity (a quarter of
the 90-10 productivity difference)
• Increase in physician share from 90th to
10th is 7 percentage points
– 35% increase in productivity (58% of the 90-
10 diff)
51