Ref: SSM-D-16-01597R1

Does the proportion of pay linked to performance affect the job satisfaction of general practitioners?

Thomas Allen*, Manchester Centre for Health Economics, University of Manchester

William Whittaker, Manchester Centre for Health Economics, University of Manchester

Matt Sutton, Manchester Centre for Health Economics, University of Manchester

*Corresponding author

Thomas Allen, Research Fellow, Manchester Centre for Health Economics, University of Manchester, M13 9PL, UK.Thomas.allen@manchester.ac.uk

1

Does the proportion of pay linked to performance affect the job satisfaction of general practitioners?

Abstract

There is concern that pay-for-performance (P4P) can negatively affect general practitioners

(GPs) by reducing their autonomy, increasing their wage dispersion or eroding their intrinsic

motivation. This is especially a concern for the Quality and Outcomes Framework (QOF), a

highly powered P4P scheme for UK GPs. The QOF affected all GPs but the exposure of their

income to P4P varied. GPs did not know their level of exposure before the QOF was

introduced and could not choose or manage it. We examine whether changes in GPs’ job

satisfaction before and after the introduction of the QOF in 2004 were correlated with the

proportion of their income that became exposed to P4P. We use data on 1920 GPs observed

at three time points spanning the introduction of the QOF; 2004, 2005 and 2008. We estimate

the effect of exposure to P4P using a continuous difference-in-differences model. We find no

significant effects of P4P exposure on overall job satisfaction or 12 additional measures of

working lives in either the short or longer term. The level of exposure to P4P does not harm

job satisfaction or other aspects of working lives. Policies influencing the exposure of income

to P4P are unlikely to alter GP job satisfaction subject to final income remaining constant.

Keywords

United Kingdom; pay-for-performance; job satisfaction; general practitioners; longitudinal data; difference in differences, continuous treatment

2

Introduction

In many healthcare systems, General Practitioners (GPs) are the primary contact for those

seeking health care and act as gatekeepers to hospital services. They are therefore crucial to

healthcare system performance. Several countries have introduced pay-for-performance

(P4P) reimbursement schemes to encourage practitioner behaviour to align with specific

objectives of the decision maker (Eijkenaar, 2012). The rationale of P4P schemes is based on

the premise that income is a key motivating factor for GPs. However, there is also a concern

that as financial incentives become more highly-powered, the unintended effects worsen

(Ariely, Gneezy, Loewenstein, & Mazar, 2009; Gravelle, Sutton, & Ma, 2010; Jacob &

Levitt, 2003). Also, as financial incentives become stronger, they can erode other forms of

intrinsic motivation or induce cheating (Benabou & Tirole, 2003; Frey & Jegen, 2001; James,

2005). There is therefore a balance to be found between incentives sufficient in size to

positively change behaviour, but not so large as to induce unintended effects.

The effects of P4P on practitioners can present in several forms. Practitioners may lose

autonomy (Freeborn, 2001; Young, Beckman, & Baker, 2012), performance heterogeneity

may lead to wage dispersion and relative income effects (Clark, Kristensen, & Westergård-

Nielsen, 2009; Frick, Prinz, & Winkelmann, 2003; Georgellis, Lange, Ileana Petrescu, &

Simmons, 2008; Pfeffer & Langton, 1993), and intrinsic motivation may be eroded by the

extrinsic motivation to attain targets (Deci, Koestner, & Ryan, 1999; Le Grand, 2003;

Prendergast, 1999; Siciliani, 2009). The response of practitioners to the introduction of risk in

payments is likely to affect practitioners differentially based on their attitudes to risk (Booth

& Frank, 1999; Eriksson & Villeval, 2008; Ganster, Kiersch, Marsh, & Bowen, 2011; Jensen,

2001; Lazear, 2000). Each avenue may manifest in changes to GP job satisfaction.

3

The job satisfaction of GPs is important for two reasons: the effect on GPs leaving the

workforce and the effects on the quality of care they provide. Studies have found lower job

satisfaction was related to GPs switching from public to private provision (Kankaanranta et

al., 2007), higher intentions to quit (Hann, Reeves, & Sibbald, 2011; Pathman et al., 2002;

Scott, Gravelle, Simoens, Bojke, & Sibbald, 2006), and working fewer hours (Williams et al.,

2001). The importance of GP job satisfaction extends beyond turnover and retention. Higher

GP job satisfaction is associated with higher patient satisfaction (Haas et al., 2000), fewer

missed appointments (Linn et al., 1985), and increased patient adherence (DiMatteo et al.,

1993).

The effect of financial incentives on GP job satisfaction has been studied in many settings.

Grembowski et al. (2003) found GPs in the US were indifferent to managed care (an

incentive to reduce healthcare costs) once models control for GP and practice characteristics.

Only GPs being paid by salary were associated with dissatisfaction. The study used cross-

sectional data in a setting where GPs were likely to self-select into practices based on their

own preferences. Gené-Badia et al. (2007) found financial incentives for GPs in Catalonia

had no impact on GP or nurse intrinsic motivation, such as job satisfaction and team support

over the two years analysed. In a study of GPs in France, Sicsic, Le Vaillant, & Franc (2012)

found a negative relationship between intrinsic and extrinsic motivation for a small cross-

section of GPs. Again in France, Saint-Lary et al. (2013) found many GPs were unwilling to

participate in a voluntary P4P programme due to concerns about the care of disadvantaged

patients, professional ethics and potential conflicts of interest.

One of the largest P4P schemes is the Quality and Outcomes Framework (QOF) introduced in

the National Health Service (NHS) in 2004 (Roland, 2004). The scheme rewards GPs based

on their performance on a range of clinical and non-clinical indicators. Approximately 25%

of practitioner income is linked to performance in the QOF and incomes rose sharply by 33%

4

within the first two years of the scheme (National Audit Office, 2008). Several recent articles

have cautioned against having a large proportion of income related to P4P and have

suggested that QOF incentives may be too large (Gillam & Steel, 2013; Raleigh & Klazinga,

2013; Roland & Campbell, 2014). These cautions come notwithstanding a 2011 Cochrane

systematic review of P4P that highlighted the lack of research on the appropriate ratio

between income linked to performance and other sources (Scott et al., 2011).

The effect of P4P exposure for GPs has become more relevant as the QOF has recently

undergone the most significant redesign since its inception. A third of performance income

has been shifted into capitation, reducing GPs’ exposure to P4P (BMA, NHS Employers, &

NHS England, 2014; Roland & Campbell, 2014). The lack of evidence of the relationship

between P4P exposure and GP job satisfaction means the expected effects of this change are

not known.

Evidence of the relationship between the QOF and GP job satisfaction is scant. A systematic

review of the impact of the QOF highlighted only three studies that assessed GP professional

wellbeing or GP job satisfaction (Gillam, Siriwardena, & Steel, 2012). An ethnographic study

of GPs and practice staff concluded that the internal motivation of GPs was not affected by

the financial incentives of the QOF (McDonald, Harrison, Checkland, Campbell, & Roland,

2007). The study benefits from repeatedly observing several healthcare workers over a five

month period but, as only two practices were sampled, the results may not be representative.

Semi-structured interviews of 21 GPs and 20 practice nurses in 2007 reported raised morale

and improved work-life balance for GPs resulting from increased income for work already

undertaken (Campbell et al., 2008). GPs also expressed concerns about extra income creating

negative public opinion and the development of a culture of monitoring and surveillance.

5

Semi-structured interviews with 12 GPs from 12 different practices found GPs had higher job

satisfaction with relation to higher incomes and reduced hours, but lower job satisfaction

with relation to additional managerial roles, administrative burden and the feeling that the

GPs’ professional identity was being eroded (Maisey et al., 2008).

This current study adds to the existing literature in a number of ways. First, we construct a

linked longitudinal panel dataset of GPs containing information on income, P4P income, P4P

exposure (P4P income as a percentage of total income), and job satisfaction. Our approach

enables longitudinal analyses on a larger sample of GPs than the current literature. Second,

we measure the relationship between P4P exposure and job satisfaction using a continuous

difference-in-differences (DID) method which provides the effect of increased P4P exposure

(Angrist & Pischke, 2008; Card, 1992). Third, we model a wider range of measures of job

satisfaction and working lives than previous studies. Fourth, we measure the immediate

effects and the longer-term effects of individual GP exposure to P4P and observe satisfaction

from before and after the introduction of P4P.

Data

The data for this study come from two sources: the QOF and the GP Worklife Survey (WLS).

The former provides a source of data used to estimate the extent of performance-related pay

at practice level. The latter provides a survey, conducted by The University of Manchester, of

the GP population focusing on job satisfaction and working conditions (Sibbald, Enzer,

Cooper, Rout, & Sutherland, 2000). A bespoke linked dataset is created by linking QOF data

with the GP WLS. Ethical approval was not required for secondary analysis of these

anonymised data.

GP WLS

6

We utilise surveys from 2004, 2005 and 2008 for this study. The 2004 GP WLS was

conducted in February 2004 and consisted of a random cross-section target sample of 1950

GPs and an additional longitudinal target sample of 2258 GPs who had responded to the

previous GP WLS in 2001 (Whalley, Bojke, Gravelle, & Sibbald, 2005). Response rates for

the 2004 samples were 53% for the cross-section and 54% for the longitudinal sample.

The 2005 GP WLS was conducted in September 2005 and consisted of a cross-sectional

target sample of 2000 GPs and a longitudinal target sample of 2122 GPs (Whalley, Gravelle,

& Sibbald, 2006). The response rate in 2005 was lower for the cross-sectional sample (45%)

but higher for the longitudinal sample (64%) than in 2004.

The 2008 GP WLS was conducted between September and November 2008 and had a target

sample of 3,000 GPs and 1,986 GPs for the cross-sectional and longitudinal samples

respectively (Hann, Goudie, Sutton, Gravelle, & Sibbald, 2009). The response rates were

44% for the cross-sectional sample and 70% for the longitudinal sample.

The GP WLS provides a measure of the overall job satisfaction for GPs as well as a number

of GP characteristics. Job satisfaction, the sub-domains of job satisfaction and life satisfaction

were measured on a 7-point scale from ‘extremely dissatisfied’ to ‘extremely satisfied’.

Several of the job satisfaction sub-domains focus on elements of a GP’s working life which

are likely to have been affected by a large-scale P4P scheme: choice of working methods,

remuneration and variety in job.

Intentions to quit are measured from the question: “what is the likelihood that you … will

leave direct patient care in within five years”. A binary scale is created from answers to this

question with 1 equalling considerable or high likelihood of leaving direct patient care and 0

equalling moderate, slight or no likelihood of leaving. GPs are also asked: “how many hours

per week do you typically work as a GP”.

7

The GP WLS uses a banded measure of GP income. GPs are asked: “what is your total

annual income from your practice? This is the amount you receive from your practice before

taxes but after deducting practice expenses”. Surveys from 2004 and 2005 use the same

bands while the survey from 2008 changed the bands to reflect the increases in GP income

due to the QOF.

QOF

The initial design of the QOF has been explained by academics (Roland, 2004; Smith &

York, 2004) and in policy documents (Department of Health, 2003a, 2003b). The scheme

rewarded practices with points based on their performance on indicators across four domains:

clinical; organisation; additional services; and patient experience. As the practice increases

their performance on these indicators, they are rewarded with more points. Each point was

worth £75 to the average practice in 2004/5 and £124.60 to the average practice in 2006/7.

The precise value of a point is determined by two adjustments: the adjusted disease

prevalence factor (ADPF) and the contractor population index (CPI). These features increase

the value of a point for practices with higher disease prevalence and larger lists of patients

(Guthrie, McLean, & Sutton, 2006).

The effect of the scheme on GP incomes was significant, due to practices scoring very well

across all domains. On average 95.5% of total points were achieved in the first year (Doran et

al., 2006). Between 2003/4 and 2005/6 GP incomes increase from £85,000 to £114,000

(National Audit Office, 2008).

We downloaded data for the first QOF year (2004/5) and the fourth QOF year (2007/8) as

these years correspond with the 2005 and 2008 surveys (Health & Social Care Information

Centre, 2015b). Data from these years were used to first measure the maximum QOF income

8

for each practice. This was then used to calculate the P4P exposure for each GP in the GP

WLS.

GP WLS – QOF linkage

The first year of the QOF started in April 2004 and ended in March 2005 but performance

results were published, and payments made, after the end of the financial year. Therefore, the

2004 GP WLS was conducted just prior to the start of the QOF and over 14 months before

practice payments were made, while the 2005 GP WLS was conducted after the first year

payments had been made (Whalley et al., 2005; Whalley, Bojke, Gravelle, & Sibbald, 2006).

We linked the 2005 survey data to the first year of the QOF payments and used the 2004

survey as our pre-QOF observation. The survey from 2008 provides an observation after the

fourth year of the QOF (Hann et al., 2009), allowing for analysis of the effect of P4P

exposure in the long run and also the effect of changes to exposure between the first and

fourth years of the QOF.

Practice characteristics

Supplementary data obtained from the Health and Social Care Information Centre were used

to control for practice characteristics (Health & Social Care Information Centre, 2015a). The

Low Income Scheme Index is a measure of income deprivation based on the proportion of the

practice population eligible for free prescriptions. Dispensing practices are those able to

dispense, as well as prescribe, prescriptions. Dispensing is a source of additional practice

income by providing a service more commonly provided by local pharmacists. Contract type

distinguishes practices on the two types of contract available in England which determine

their incomes. GPs per practice and practice list size are measures of the size of the practice

and reflect the workload of each GP. Black or minority ethnic group measures the proportion

9

of the practice population from these ethnic groups. Rural practice is used to distinguish

practices in rural areas from those in urban areas.

Methods

Measure of P4P exposure

We calculated the maximum income a practice could receive if they achieved all available

QOF points. Variations in this maximum income measure do not depend on practice

performance but on how the design of the QOF affects practices differently, which GPs could

not influence. Specifically, practices with more registered patients and with higher disease

prevalence rates have higher potential incomes. This distinction between achieved income

and potential income is important in creating a measure of exposure to P4P that GPs could

not influence directly. Achieved income would have been determined by the effort of the GP

which itself may have been determined, in part, by their job satisfaction. By measuring

maximum potential income we remove a possible source of endogeneity.

Algebraically, P4P exposure for GPs’ is expressed as:

P 4 Pijt=[ (QO F jt∗( FT E ijt

∑ FT E jt))

Y ijt]∗100

1

Where P 4 P is the P4P exposure for GP i in practice j at time t . QOF is net QOF income.

FT E ijt

∑ FT E jtis the full-time equivalent (FTE) of the individual GP divided by the sum of FTE

GPs in the whole practice. Y ijt is the predicted “no-QOF” counterfactual income for 2005 and

2008. The whole expression is multiplied by 100 to arrive at the percentage P4P exposure for

each GP in our sample. The individual components of P4P exposure are explained below.

10

In order to measure exposure we first predict GP income (Y ijt) if the QOF had not been

introduced. To predict income for individual GPs we use self-reported GP income as the

dependent variable and regress this on a range of GP and practice characteristics. As GPs

reported their income in bands, these models are estimated using interval regressions using

the income bands as thresholds (Wooldridge, 2009, p. 601). The first and last income bands

are open ended, for example less than £25,000 or more than £150,000. We assume the lower

bound of the first band was zero and the upper bound of the last band was infinity.

We estimate the following:

y ijt¿ =β0+β ' 1 X ijt+uijt ,

2

Where y ijt¿ is the income of GP i in practice j at time t , which is not observed. X ijt is a vector

of GP and practice characteristics. uijt is the error termi .i . d . N (0 , σ2). y ijt denotes the

observed banded income from the GP WLS:

y ijt=1 if y ijt¿ ≤ a1

y ijt=2 if a1< y ijt¿ ≤a2

⋮y ijt=J if aJ−1≤ y ijt

¿

3

Where a1…aJ−1 represent the income band thresholds (Sutton & Godfrey, 1995).

The variables within X ijt were: age, age2, patients per GP, partnership size, dispensing

practice, ethnic minority GP, practice contract, population ethnicity, population deprivation

11

and rural practice. The estimation method and choice of independent variables are consistent

with previous literature using the GP WLS (Morris et al., 2011).

We estimate the determinants of income in the 2004 survey and use the estimated coefficients

to predict the income that would have been received had the QOF not been introduced. These

predictions are made for 2005 and 2008. The assumption in this approach is that, without the

QOF, the effect of the determinants of income would have remained constant over the time

period 2004-2008. This method provides the denominator in Equation 1.

The maximum practice level QOF income is calculated and used to create the numerator in

Equation1.

Revenue¿=∑k=1

K

( π kidt∗ADP Fidt∗CP I ¿∗αt ) 4

Where π denotes the points available which varies over indicator k , practice i, disease d and

time t . Adjustments to revenue are made for the clinical indicators by the ADPF. All

indicators are adjusted by the CPI . α denotes the value of a QOF point which varies only

over time.

The self-reported GP income figures are net of expenses. In order to have a comparable

denominator and numerator, we adjusted the QOF income downwards to account for

expenses. In 2004/5 GPs in England had average gross earnings of £241,795 and average net

earnings of £103,654, giving a gross/net ratio of 2.33 (Health & Social Care Information

Centre, 2006). We divided gross QOF income by 2.33 to obtain a net figure. This rescales the

income figure but the relative variation across GPs and practices is maintained.

12

We also account for the fact that not all GPs in a practice will receive an equal share of QOF

income. We assume that the share received is determined by the FTE of the GP and how

many FTEs where at the practice. For example, a 0.5 FTE GP in a practice with seven FTE

GPs would receive 1/14th of the practice QOF income.

Analysis of effect of P4P exposure on job satisfaction

We use a continuous DID model to estimate treatment effects when all subjects are treated

but the treatment intensity varies across subjects (Card, 1992; Gaynor, Moreno-Serra, &

Propper, 2010). This is appropriate as P4P was introduced in all practices at the same time

but P4P exposure varies across GPs.

We estimate a continuous DID model using a random effects regression:

Y ijt=β0+β11 [ t=T ]+β2 P 4 P ij

+β3 P 4 Pij∗1 [t=T ]+β4 Dijt+β5' X ijt+αi+uijt

5

Y is the dependent variable, a measure of job satisfaction or working lives, for GP i in

practice j at time t . D is a set of dummy variables for income bands. X is a matrix of practice

and GP characteristics. α i is a random GP effect and uijt is the error termi .i . d . N ( 0 , σ2 ).

P 4 P measures P4P exposure. A year dummy is used to denote when exposure occurs:

1[ t=T ], where T takes the value 2005 or 2008. The year dummy is also interacted with P4P

exposure giving the treatment effect.

The coefficient on P4P exposure for 2004, the non-interacted term, is analogous to the

treatment dummy from a standard DID model and measures the effect on the treated before

treatment occurs. This variable also absorbs unobservable individual heterogeneity that is not

13

explained by the model and is therefore captured by the exposure variable. This unobserved

heterogeneity does not confound the estimated effect of post-QOF exposure if we assume it

to be time-invariant. The interaction term measures the effect of this exposure after the

introduction of the QOF.

The characteristics is X are selected based on previous research on the determinants of GP

job satisfaction (Scott et al., 2006). These determinants are not directly affected by the QOF,

therefore removing a potential source of endogeneity. Holding other factors constant,

increases in GP income would be expected to increase job satisfaction. Increased income is

also likely to be associated with greater P4P exposure. By controlling for GP income (D) in

each model, we ensure that P4P exposure does not capture the amount of income but only

captures the effect of the method by which this income is earned.

Since there are different income bands in 2004 and 2008 we include a set of income dummy

variables for each year. As this allows the effect of income to differ in each year, we also

include two sets of income dummy variables in the 2004 and 2005 samples.

This model is estimated on a sample combining observations from 2004 and 2005 and then a

separate regression combining observations from 2004 and 2008. It is not a requirement that

GPs appear in both years.

Results

Descriptive statistics on the GP WLS and practice level characteristics are shown in Table 1.

Job satisfaction increased substantially between 2004 and 2005, but fell between 2005 and

2008. A similar pattern is observed for all of the sub-domains of job satisfaction with the

exception of working conditions and fellow workers. The changes in likelihood of quitting

14

also reflect this pattern in working lives. Hours worked decreases in 2005 only to increase

again in 2008.

Comparing the proportion of GPs in each income band reveals large increases in income

between 2004 and 2005. The comparison between 2005 and 2008 is complicated by the

change in bands. However, it is clear that incomes also increased between these years.

GP workload, as measured by the number of patients per GP, decreased in 2008. This is

likely to be the result of increases in the use of salaried GPs as the number of practice

partners decreases. The other variables are largely static over the period.

QOF income and P4P exposure

The results in Table 2 are from the regression estimating the determinants of GP income

before the QOF. The estimated incomes from this regression are shown in Table 3. In 2004,

average estimated income was £73,800. Changes in the predicted incomes in subsequent

years reflect changes in the composition of the GP WLS sample.

Table 3 also contains descriptive statistics for P4P exposure. The mean value for 2005 is

14.6%. The 90th percentile for GP income exposure is 20% and there are 11 GPs with income

exposure in excess of 40%. Of these 11 outliers, 10 are from single partner practices where

the individual GP cannot spread QOF income exposure between multiple partners. The other

outlier comes from a practice with two partners but where the maximum QOF income is well

above average, resulting in the GPs being particularly exposed. The increase in the value of a

point in 2008 resulted in a large increase in P4P exposure for all GPs. Histograms in Figure 1

reveal the extent of variation in exposure of income to P4P.

Job satisfaction

15

Table 4 reports the results from the continuous DID models. The coefficient on P4P exposure

in 2004 is negative and statistically significant at 0.01%. As mentioned in the econometric

methods section, this variable captures unobserved individual heterogeneity. Therefore, the

association between exposure in 2004 and job satisfaction is the result of a correlation

between the variables used to measure exposure and job satisfaction. It does not measure the

effect of exposure since the observation is from before the exposure occurs.

The interaction term suggests a positive, but not statistically significant, effect of QOF

income exposure on job satisfaction in 2005 (t-ratio 1.74). In 2008 the coefficient is negative

and not statistically significant (t-ratio -0.14).

Measures of GP working lives

Table 5 reports regressions using 2004 and 2005 data as well regressions using 2004 and

2008 data. For brevity only exposure variables and year dummies are shown, but the full set

of controls were included. The pre-QOF term was statistically significant for hours worked

per week, life satisfaction, satisfaction with working conditions, choice of method of working

remuneration and hours of work. This suggests that P4P exposure in 2004 is capturing some

aspect of these dependent variables that our control variables could not.

The interaction term is not statistically significant in any model, with the exception of

satisfaction with physical working conditions in 2005. There is no relationship between P4P

exposure and any of the dependent variables for 2008.

Results not shown (but available on request) are qualitatively the same for models with

samples stratified by GP gender, GP age, practice size by patient population, practice size by

GP numbers and contract type. Salaried GP are not exposed directly to P4P and are removed

from our main sample. Results of models estimated including the salaried GPs do not differ.

16

Residuals from linear regressions were approximately normally distributed and the results

were robust to using non-linear ordered logistic regression models for job satisfaction. To

account for a non-linear effect of exposure we re-estimated the models placing GPs into

quartiles of exposure. The only statistically significant difference in the effect of exposure on

job satisfaction was found in 2005 and for the least exposed quartile. GPs in this quartile had

smaller increases in job satisfaction than GPs in other quartiles, proving weak evidence that

GPs exposed the least to P4P experienced temporary losses in job satisfaction compared to

other GPs.

Discussion

We aimed to uncover the effects on working lives and job satisfaction of income exposure to

P4P from a large P4P scheme introduced in the UK in 2004 using continuous DID methods.

Our findings suggest job satisfaction was not related to the rate of P4P exposure. This result

was found in the immediate year following the introduction of the P4P scheme, and under a

longer time frame of four years post introduction. GPs were also found to be insensitive to

P4P exposure with regards to: working hours; intentions to quit; life satisfaction; and nine

sub-domains of job satisfaction. Key sub-domains of satisfaction which are associated with

the design of the QOF (choice of method of working, remuneration and variety in job) were

also found not to be affected by P4P exposure.

Strengths and weaknesses

This study builds on an existing body of research using qualitative and quantitative methods.

Previous qualitative studies have been limited due to small sample sizes but have benefitted

from a greater focus on the specific subject of study. Quantitative studies had larger sample

sizes but were biased by self-selection into P4P schemes and cross-sectional data.

17

We have a large sample of GPs who could not self-select into or out of P4P. We have a

survey which asks a range of questions about job satisfaction and working lives. This survey

is linked to administrative data to create a unique linked dataset which has enabled us to

measure the amount of P4P exposure for each GP in our sample. We employed an

econometric method which estimates the effect of increased P4P exposure. In addition to job

satisfaction we also model the effects of P4P exposure on intentions to quit, hours worked,

life satisfaction and nine sub-domains of jobs satisfaction. We analysed the immediate effect

one year after the introduction of P4P and the effects four years after P4P was introduced.

However, despite the strengths of this study mentioned above we should discuss some

limitations. We had to estimate P4P exposure based on two assumptions. Firstly, that the

sharing of QOF income within a practice is related only to a GP’s FTE. This assumption is

likely to oversimplify a complicated set of income sharing rules. However, these sharing

rules are unknown, and our approach makes use of the information that is available. FTE is

likely to play an important role in the allocation and we have taken this into account. Other

factors which may influence the allocation include seniority, tenure and additional

responsibilities but data on these are not available.

Our second assumption is that our predictions of GP income had the QOF not been

introduced are accurate. We made our predictions of income based on the pre-QOF

determinants of income. These predictions assume that the effect of determinants of income

would not have changed over time in the absence of the QOF. As a robustness check we

compared the results when using practice level exposure, accounting for practice income in

place of GP income, and the results did not differ.

18

Finally, a single predicted value was used for each GP’s income. This did not account for the

variation in the values predicted by the interval regressions. However, accounting for this

variation could only lower the statistical significance of our findings further.

Our estimation method requires the same assumptions of a standard DID, specifically parallel

trends between control and treatment groups in the pre-treatment period. It is not possible to

test this assumption in our setting due to differences in survey questions in the GP WLS prior

to 2004. We also have a continuous treatment variable in place of distinct groups. However,

treatment intensity is determined by the design of the QOF, the specifics of which were

unknown prior to 2005. We do not feel it is likely that GPs who would later have different

exposure would be on different trends in terms of job satisfaction and working lives. The

model controls for differences in job satisfaction that are time-invariant and changes that

affect all GPs equally. For example, if GPs anticipated changes in working lives due to

knowledge of the new contract, this effect would be captured by the common year effects.

As GPs did not know their own P4P exposure before 2005, this anticipation effect is unlikely

to be correlated with exposure.

The GP WLS may not be representative of the total GP population. However, research has

shown that this response bias does not affect the determinants of job satisfaction (Gravelle,

Hole, & Hossaun, 2008). To account for potential confounding due to sample response we

included in the analysis those factors that have been associated with differential rates of

response. Further work could include additional waves of the GP WLS to better model

heterogeneity between GPs. However, our data cover a period of large changes to P4P

exposure, changes made after 2008 are smaller and less likely to impact on working lives.

Including more waves increases the risk that other unmeasured factors confound the

estimated relationship between exposure and job satisfaction and reduces the accuracy of the

estimated measure of exposure.

19

Our results are in relation to a specific, large scale P4P scheme in the UK. There may be

other factors that impact on the relationship between P4P exposure and job satisfaction in

other settings. For instance, the relationship may differ in health systems that are not centrally

funded or where the role of the GP differs.

Policy implications

Our findings suggest GP job satisfaction is insensitive to the proportion of income exposure

to P4P. Our results suggest therefore, that policymakers seeking to make changes to the

exposure of income to P4P are unlikely to alter GP job satisfaction subject to final income

remaining constant.

20

References

Angrist, J. D., & Pischke, J.-S. S. (2008). Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton, NJ: Princeton University Press.

Ariely, D., Gneezy, U., Loewenstein, G., & Mazar, N. (2009). Large Stakes and Big Mistakes. Review of Economic Studies, 76(2), 451–469. http://doi.org/10.1111/j.1467-937X.2009.00534.x

Benabou, R., & Tirole, J. (2003). Intrinsic and Extrinsic Motivation. Review of Economic Studies, 70(2), 489–520. http://doi.org/10.1111/1467-937X.00253

BMA, NHS Employers, & NHS England. (2014). 2014/15 General Medical Services (GMS) Contract Quality and Outcomes Framework (QOF). Retrieved May 1, 2015, from http://www.nhsemployers.org/~/media/Employers/Documents/Primary care contracts/QOF/2014-15/14-15 General Medical Services contract - Quality and Outcomes Framework.pdf

Booth, A. L., & Frank, J. (1999). Earnings, Productivity, and Performance Related Pay. Journal of Labor Economics, 17(3), 447–463. http://doi.org/10.1086/209927

Campbell, S. M., McDonald, R., & Lester, H. (2008). The experience of pay for performance in English family practice: a qualitative study. Annals of Family Medicine, 6(3), 228–34. http://doi.org/10.1370/afm.844

Card, D. (1992). Using regional variation in wages to measure the effects of the federal minimum wage. Industrial and Labor Relations Review, 46(1), 22–37. Retrieved from http://ideas.repec.org/a/ilr/articl/v46y1992i1p22-37.html

Clark, A. E., Kristensen, N., & Westergård-Nielsen, N. (2009). Job Satisfaction and Co-worker Wages: Status or Signal? The Economic Journal, 119(536), 430–447. http://doi.org/10.1111/j.1468-0297.2008.02236.x

Deci, E. L., Koestner, R., & Ryan, R. M. (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin, 125(6), 627–68; discussion 692–700. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10589297

Department of Health. (2003a). Delivering investment in general practice: implementing the new GMS contract. Retrieved November 1, 2016, from http://webarchive.nationalarchives.gov.uk/20130107105354/http://www.dh.gov.uk/prod_consum_dh/groups/dh_digitalassets/@dh/@en/documents/digitalasset/dh_4070231.pdf

Department of Health. (2003b). Investing in general practice The new General Medical Services Contract. Retrieved November 1, 2016, from http://www.nhsemployers.org/~/media/Employers/Documents/SiteCollectionDocuments/gms_contract_cd_130209.pdf

DiMatteo, M. R., Sherbourne, C. D., Hays, R. D., Ordway, L., Kravitz, R. L., McGlynn, E. A., … Rogers, W. H. (1993). Physicians’ characteristics influence patients' adherence to medical treatment: results from the Medical Outcomes Study. Health Psychology : Official Journal of the Division of Health Psychology, American Psychological Association, 12(2), 93–102. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/8500445

Doran, T., Fullwood, C., Gravelle, H., Reeves, D., Kontopantelis, E., Hiroeh, U., & Roland,

21

M. (2006). Pay-for-Performance Programs in Family Practices in the United Kingdom. New England Journal of Medicine, 355(4), 375–384. http://doi.org/10.1056/NEJMsa055505

Eijkenaar, F. (2012). Pay for Performance in Health Care: An International Overview of Initiatives . Medical Care Research and Review , 69 (3 ), 251–276. http://doi.org/10.1177/1077558711432891

Eriksson, T., & Villeval, M. C. (2008). Performance-pay, sorting and social motivation. Journal of Economic Behavior & Organization, 68(2), 412–421. http://doi.org/10.1016/j.jebo.2007.10.003

Freeborn, D. K. (2001). Satisfaction, commitment, and psychological well-being among HMO physicians. The Western Journal of Medicine, 174(1), 13–8. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1071220&tool=pmcentrez&rendertype=abstract

Frey, B. S., & Jegen, R. (2001). Motivation Crowding Theory. Journal of Economic Surveys, 15(5), 589–611. http://doi.org/10.1111/1467-6419.00150

Frick, B., Prinz, J., & Winkelmann, K. (2003). Pay inequalities and team performance. International Journal of Manpower, 24(4), 472–488. http://doi.org/10.1108/01437720310485942

Ganster, D. C., Kiersch, C. E., Marsh, R. E., & Bowen, A. (2011). Performance-Based Rewards and Work Stress. Journal of Organizational Behavior Management, 31(4), 221–235. http://doi.org/10.1080/01608061.2011.619388

Gaynor, M., Moreno-Serra, R., & Propper, C. (2010). Death by Market Power: Reform, Competition and Patient Outcomes in the National Health Service. National Bureau of Economic Research Working Paper Series, No. 16164. Retrieved from http://www.nber.org/papers/w16164

Gené-Badia, J., Escaramis-Babiano, G., Sans-Corrales, M., Sampietro-Colom, L., Aguado-Menguy, F., Cabezas-Peña, C., & de Puelles, P. G. (2007). Impact of economic incentives on quality of professional life and on end-user satisfaction in primary care. Health Policy (Amsterdam, Netherlands), 80(1), 2–10. http://doi.org/10.1016/j.healthpol.2006.02.008

Georgellis, Y., Lange, T., Ileana Petrescu, A., & Simmons, R. (2008). Human resource management practices and workers’ job satisfaction. International Journal of Manpower, 29(7), 651–667. http://doi.org/10.1108/01437720810908947

Gillam, S., Siriwardena, A. N., & Steel, N. (2012). Pay-for-Performance in the United Kingdom: Impact of the Quality and Outcomes Framework—A Systematic Review. The Annals of Family Medicine, 10(5), 461–468. http://doi.org/10.1370/afm.1377

Gillam, S., & Steel, N. (2013). The Quality and Outcomes Framework--where next? BMJ, 346(feb07 2), f659–f659. http://doi.org/10.1136/bmj.f659

Gravelle, H., Hole, A. R., & Hossaun, M. I. (2008). Response bias in job satisfaction surveys: English general practitioners. Retrieved July 20, 2015, from https://www.york.ac.uk/media/economics/documents/discussionpapers/2008/0824.pdf

Gravelle, H., Sutton, M., & Ma, A. (2010). Doctor Behaviour under a Pay for Performance Contract: Treating, Cheating and Case Finding? The Economic Journal, 120(542), 129–156. http://doi.org/10.1111/j.1468-0297.2009.02340.x

Grembowski, D., Ulrich, C. M., Paschane, D., Diehr, P., Katon, W., Martin, D., … Velicer,

22

C. (2003). Managed Care and Primary Physician Satisfaction. The Journal of the American Board of Family Medicine, 16(5), 383–393. http://doi.org/10.3122/jabfm.16.5.383

Guthrie, B., McLean, G., & Sutton, M. (2006). Workload and reward in the Quality and Outcomes Framework of the 2004 general practice contract. British Journal of General Practice, 56, 836–841. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1927091/pdf/bjpg56-836.pdf

Haas, J. S., Cook, E. F., Puopolo, A. L., Burstin, H. R., Cleary, P. D., & Brennan, T. A. (2000). Is the professional satisfaction of general internists associated with patient satisfaction? Journal of General Internal Medicine, 15(2), 122–8. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1495336&tool=pmcentrez&rendertype=abstract

Hann, M., Goudie, R., Sutton, M., Gravelle, H., & Sibbald, B. (2009). Working Conditions and Job Satisfaction of GPs: Findings from the Fifth National GP Worklife Survey. National Primary Care Research and Development Centre. Retrieved from http://www.population-health.manchester.ac.uk/primarycare/npcrdc-archive/Publications/WorkingConditionsandJobSatisfactionofGPsFindingsfromtheFifthNationalGPWorklifeSurvey.pdf

Hann, M., Reeves, D., & Sibbald, B. (2011). Relationships Between Job Satisfaction, Intentions to Leave Family Practice and Actually Leaving Among Family Physicians in England. The European Journal of Public Health, 21(4), 499–503. http://doi.org/10.1093/eurpub/ckq005

Health & Social Care Information Centre. (2006). GP Earnings and Expenses Enquiry 2004/05. Retrieved November 1, 2016, from http://content.digital.nhs.uk/pubs/gpearnex0405

Health & Social Care Information Centre. (2015a). Health & Social Care Information Centre. Retrieved November 1, 2016, from http://content.digital.nhs.uk/

Health & Social Care Information Centre. (2015b). The Quality and Outcomes Framework. Retrieved November 1, 2016, from http://qof.digital.nhs.uk/

Jacob, B. A., & Levitt, S. D. (2003). Rotten Apples: An Investigation of The Prevalence and Predictors of Teacher Cheating. Quarterly Journal of Economics, 118(3), 843–877. http://doi.org/10.1162/00335530360698441

James, H. S. (2005). Why did you do that? An economic examination of the effect of extrinsic compensation on intrinsic motivation and performance. Journal of Economic Psychology, 26(4), 549–566. http://doi.org/10.1016/j.joep.2004.11.002

Jensen, M. C. (2001). Paying People to Lie: The Truth About the Budgeting Process. SSRN Electronic Journal. http://doi.org/10.2139/ssrn.267651

Kankaanranta, T., Nummi, T., Vainiomäki, J., Halila, H., Hyppölä, H., Isokoski, M., … Rissanen, P. (2007). The role of job satisfaction, job dissatisfaction and demographic factors on physicians’ intentions to switch work sector from public to private. Health Policy (Amsterdam, Netherlands), 83(1), 50–64. http://doi.org/10.1016/j.healthpol.2006.11.010

Lazear, E. P. (2000). Performance Pay and Productivity. The American Economic Review, 90(5), 1346–1361. Retrieved from http://www.jstor.org/stable/2677854

23

Le Grand, J. (2003). Motivation, Agency, and Public Policy: Of Knights and Knaves, Pawns and Queens. Oxford: Oxford University Press. http://doi.org/10.1093/0199266999.001.0001

Linn, L. S., Brook, R. H., Clark, V. A., Davies, A. R., Fink, A., & Kosecoff, J. (1985). Physician and patient satisfaction as factors related to the organization of internal medicine group practices. Medical Care, 23(10), 1171–8. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/4058071

Maisey, S., Steel, N., Marsh, R., Gillam, S., Fleetcroft, R., & Howe, A. (2008). Effects of payment for performance in primary care: qualitative interview study. Journal of Health Services Research & Policy, 13(3), 133–9. http://doi.org/10.1258/jhsrp.2008.007118

McDonald, R., Harrison, S., Checkland, K., Campbell, S., & Roland, M. (2007). Impact of financial incentives on clinical autonomy and internal motivation in primary care: ethnographic study. BMJ (Clinical Research Ed.), 334(7608), 1357. http://doi.org/10.1136/bmj.39238.890810.BE

Morris, S., Goudie, R., Sutton, M., Gravelle, H., Elliott, R., Hole, A. R., … Skåtun, D. (2011). Determinants of general practitioners’ wages in England. Health Economics, 20(2), 147–160. http://doi.org/10.1002/hec.1573

National Audit Office. (2008, January 24). NHS Pay Modernisation: New contracts for general practice services in England. Retrieved November 1, 2016, from http://www.nao.org.uk/wp-content/uploads/2008/02/0708307.pdf

Pathman, D. E., Konrad, T. R., Williams, E. S., Scheckler, W. E., Linzer, M., & Douglas, J. (2002). Physician job satisfaction, dissatisfaction, and turnover. The Journal of Family Practice, 51(7), 593. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12160487

Pfeffer, J., & Langton, N. (1993). The Effect of Wage Dispersion on Satisfaction, Productivity, and Working Collaboratively: Evidence from College and University Faculty. Administrative Science Quarterly, 38(3), 382–407. http://doi.org/10.2307/2393373

Prendergast, C. (1999). The Provision of Incentives in Firms. Journal of Economic Literature, 37(1), 7–63. Retrieved from http://www.jstor.org/stable/2564725

Raleigh, V. S., & Klazinga, N. (2013). Future proofing the Quality and Outcomes Framework. BMJ, 346.

Roland, M. (2004). Linking Physicians’ Pay to the Quality of Care — A Major Experiment in the United Kingdom. New England Journal of Medicine, 351(14), 1448–1454. http://doi.org/10.1056/NEJMhpr041294

Roland, M., & Campbell, S. (2014). Successes and Failures of Pay for Performance in the United Kingdom. New England Journal of Medicine, 370(20), 1944–1949. http://doi.org/10.1056/NEJMhpr1316051

Saint-Lary, O., Bernard, E., Sicsic, J., Plu, I., François-Purssell, I., & Franc, C. (2013). Why did most French GPs choose not to join the voluntary national pay-for-performance program? PloS One, 8(9), e72684. http://doi.org/10.1371/journal.pone.0072684

Scott, A., Gravelle, H., Simoens, S., Bojke, C., & Sibbald, B. (2006). Job Satisfaction and Quitting Intentions: A Structural Model of British General Practitioners. British Journal of Industrial Relations, 44(3), 519–540. http://doi.org/10.1111/j.1467-8543.2006.00511.x

Scott, A., Sivey, P., Ait Ouakrim, D., Willenberg, L., Naccarella, L., Furler, J., & Young, D.

24

(2011). The effect of financial incentives on the quality of health care provided by primary care physicians. In Cochrane Database of Systematic Reviews. John Wiley & Sons, Ltd. Retrieved from http://onlinelibrary.wiley.com/doi/10.1002/14651858.CD008451.pub2/abstract

Sibbald, B., Enzer, I., Cooper, C., Rout, U., & Sutherland, V. (2000). GP Job Satisfaction in 1987, 1990 and 1998: Lessons for the Future? Family Practice, 17(5), 364–371. http://doi.org/10.1093/fampra/17.5.364

Siciliani, L. (2009). Paying for performance and motivation crowding out. Economics Letters, 103(2), 68–71. http://doi.org/DOI: 10.1016/j.econlet.2009.01.022

Sicsic, J., Le Vaillant, M., & Franc, C. (2012). Intrinsic and extrinsic motivations in primary care: an explanatory study among French general practitioners. Health Policy (Amsterdam, Netherlands), 108(2-3), 140–8. http://doi.org/10.1016/j.healthpol.2012.08.020

Smith, P., & York, N. (2004). Quality Incentives: The Case Of U.K. General Practitioners. Health Affairs, 23(3), 112–118. http://doi.org/10.1377/hlthaff.23.3.112

Sutton, M., & Godfrey, C. (1995). A grouped data regression approach to estimating economic and social influences on individual drinking behaviour. Health Economics, 4(3), 237–47. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/7550773

Whalley, D., Bojke, C., Gravelle, H., & Sibbald, B. (2005). 2004 National Survey of General Practitioner Job Satisfaction. Retrieved November 1, 2016, from http://www.population-health.manchester.ac.uk/primarycare/npcrdc-archive/Publications/GP Satisfaction REPORT.pdf

Whalley, D., Bojke, C., Gravelle, H., & Sibbald, B. (2006). GP job satisfaction in view of contract reform: a national survey. The British Journal of General Practice, 56(523), 87–92. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1828251/pdf/bjpg56-87.pdf

Whalley, D., Gravelle, H., & Sibbald, B. (2006). 2005 National Survey of General Practitioner Job Satisfaction. Retrieved November 1, 2016, from http://www.population-health.manchester.ac.uk/primarycare/npcrdc-archive/Publications/2005_GP_Satisfaction_Survey_Report.pdf

Williams, E. S., Konrad, T. R., Scheckler, W. E., Pathman, D. E., Linzer, M., McMurray, J. E., … Schwartz, M. (2001). Understanding physicians: Intentions to withdraw from practice: The role of job satisfaction, job stress, mental and physical health. In Advances in Health Care Management (Vol. 2, pp. 243–262). Emerald Group Publishing Limited. http://doi.org/doi:10.1016/S1474-8231(01)02029-8

Wooldridge, J. (2009). Introductory Econometrics. A Modern Approach (4th ed.). South-Western Cengage Learning.

Young, G. J., Beckman, H., & Baker, E. (2012). Financial incentives, professional values and performance: A study of pay-for-performance in a professional organization. Journal of Organizational Behavior, 33(7), 964–983. http://doi.org/10.1002/job.1770

25

Tables

Table 1: Descriptive statistics for GP WLS

Variable 2004 2005 2008Life satisfaction (1=low to 7=high) 4.649 5.095 5.008Overall satisfaction with job (1 to 7) 4.567 5.201 4.728Satisfaction with physical working conditions (1 to 7) 4.862 5.044 5.129Satisfaction with freedom to choose own method of working (1 to 7) 4.636 4.892 4.640Satisfaction with colleagues & fellow workers (1 to 7) 5.515 5.599 5.602Satisfaction with recognition you get for good work (1 to 7) 4.224 4.726 4.495Satisfaction with amount of responsibility you are given (1 to 7) 4.976 5.406 5.276Satisfaction with remuneration (1 to 7) 4.376 5.387 4.849Satisfaction with opportunity to use abilities (1 to 7) 4.787 5.147 5.074Satisfaction with hours of work (1 to 7) 3.914 4.802 4.205Satisfaction with amount of variety in job (1 to 7) 5.011 5.269 5.276High or considerable likelihood of quitting 0.256 0.218 0.251Hours per week typically work as a GP 44.540 40.509 42.738Male 0.662 0.636 0.633Married/living with spouse 0.920 0.914 0.911Number of children under 18 years old 1.418 1.284 1.302Black or minority ethnicity 0.155 0.121 0.123Age (years) 47.034 47.977 48.777No personal partner or personal partner does not work 0.261 0.304 0.203Personal partner works part-time 0.360 0.336 0.417Personal partner works full-time 0.379 0.360 0.381Net income per year <£25,000 0.009 0.003

£25,000-£50,000 0.164 0.099£50,000-£70,000 0.261 0.140£70,000-£85,000 0.266 0.166£85,000-£100,000 0.178 0.210£100,000-£120,000 0.086 0.232£120,000-£150,000 0.027 0.114£150,000+ 0.008 0.037<£25,000 0.004£25,000-£50,000 0.049£50,000-£75,000 0.154£75,000-£100,000 0.284£100,000-£125,000 0.310£125,000-3150,000 0.127£150,000-£175,000 0.042£175,000+ 0.031

Total number of GP partners in the practice 4.776 4.962 4.610Practice list size* 8975.922 9091.339 9357.244Patients per GP/1000* 1.850 1.831 1.591Time in current practice (years) 14.561 15.417 16.298Practice contract type 0.456 0.451 0.447Rural practice* 0.183 0.187 0.188Low income scheme index* 10.433 10.618 9.599Black or minority ethnic group population* 0.122 0.116 0.113Dispensing practice* 0.194 0.189 0.209

* Variables not from the GP WLS Income bands were changed in 2008.

26

Table 2: Interval regressions for the determinants of income for GP WLS respondents

2004Male 20891.1*** (19.29)Age (years) 796.4 (1.11)Age squared -7.166 (-0.95)Patients per GP/1000 8749.8*** (7.00)Partnership size: 2 -6100.2 (-1.89)Partnership size: 3 -2949.1 (-0.92)Partnership size: 4 -684.8 (-0.21)Partnership size: 5 -1420.5 (-0.44)Partnership size: 6 -1107.9 (-0.34)Partnership size: 7 738.2 (0.22)Partnership size: 8 441.8 (0.11)Partnership size: 9+ 6007.8 (1.41)Dispensing practice 12623.0*** (7.60)Non-white GP -192.6 (-0.12)Practice contract type 7414.7*** (6.95)Non-white population 9596.6* (2.35)Low income scheme index

-255.1** (-2.75)

Rural practice 3139.8* (2.04)Constant 19348.0 (1.14)Observations 1867McFadden’s R2 0.103

t statistics in parentheses (Standard errors clustered by practice)* p < 0.05, ** p < 0.01, *** p < 0.001Omitted category Partnership size=1

Table 3: GP incomes predicted from interval regressions and QOF exposure for GPs and

practices

Practices Mean sd p10 p90Predicted income 2004 1918 73827 13584 53670 90028Predicted income 2005 1956 73584 13814 53390 90211Predicted income 2008 2071 71360 13194 52198 87256

GP QOF income exposure 2005 1956 14.591 4.053 9.830 20.144GP QOF income exposure 2008 2069 25.644 8.936 16.764 35.770

Predicted incomes are based on determinants of income from Table 2 applied to each GP WLS sample

27

Table 4: The effect of P4P exposure on job satisfaction estimated by random effects

continuous DID

Job satisfaction 2004 & 2005

Job satisfaction 2004 & 2008

QOF income exposure * year

0.0161 (1.74) -0.000842 (-0.14)

QOF income exposure -0.0385*** (-3.43) -0.00572 (-0.92)

Year (2005 or 2008) 0.282 (1.78) -0.0880 (-0.39)

50-70K -0.0954 (-0.65) 0.226 (1.57)70-85K -0.181 (-1.25) 0.312* (2.07)85-100K 0.000763 (0.01) 0.374* (2.26)100-120K -0.132 (-0.88) 0.355 (1.90)120-150K 0.0795 (0.32) 0.464 (1.66)150+K 1.213* (2.17) -0.909 (-1.68)50-70K * 2005 0.236 (1.95)70-85K * 2005 0.281* (2.17)85-100K * 2005 0.445** (3.26)100-120K * 2005 0.596*** (4.02)120-150K * 2005 0.578* (2.41)150+K * 2005 -0.477 (-0.84)50-75K 0.306* (2.08)75-100K 0.217 (1.49)100-125K 0.439** (2.96)125-150K 0.671*** (4.08)150-175K 0.576** (2.80)175+K 0.892*** (3.59)Male -0.247*** (-3.48) -0.304*** (-4.38)Black or minority ethnic GP

-0.232* (-2.43) -0.0167 (-0.17)

Age (years) -0.174*** (-4.75) -0.194*** (-5.25)Age squared 0.00185*** (4.81) 0.00202*** (5.27)Patients per GP/1000 -0.215** (-2.87) -0.308*** (-4.21)Time in current practice (years)

-0.00245 (-0.46) 0.00702 (1.26)

Practice contract type 0.0672 (1.20) 0.0906 (1.58)Rural practice -0.00143 (-0.02) 0.0293 (0.38)Low income scheme index

0.00162 (0.38) 0.000290 (0.06)

Constant 9.484*** (10.83) 9.681*** (10.67)Observations 3079 2722R2 0.098 0.043Rho 0.401 0.408

t statistics in parentheses (robust standard errors clustered by GP)* p < 0.05, ** p < 0.01, *** p < 0.001Omitted category Income <50K, Rho (the intra-class correlation coefficient) shows the proportion of the error variance attributed to variation across GPs

28

Table 5: Random effects continuous DID estimation of the effect of P4P exposure on GP working lives in 2005 and 2008

High likelihood of quitting

Hours per week

Life satisfaction

Physical working conditions

Choose method of working

Colleagues

Recognition for good work

Amount of responsibility

Re-muneration

Opportunity to use abilities

Hours of work

Variety in job

QOF income exposure * 2005

-0.00283(-1.02)

-0.00520(-0.07)

0.0115(1.07)

0.0244*

(2.06)0.0117(1.04)

0.0159(1.56)

0.0154(1.39)

0.0103(0.84)

0.0168(1.55)

0.00556(0.55)

0.00734(0.63)

0.0146(1.37)

QOF income exposure

-0.000552(-0.17)

0.742***

(7.06)-0.0435***

(-3.63)-0.0590***

(-3.95)-0.0290*

(-2.21)-0.00634(-0.51)

-0.0212(-1.63)

-0.0154(-1.11)

-0.0434***

(-3.31)-0.0142(-1.10)

-0.0426**

(-2.84)-0.00311(-0.25)

2005 -0.0267(-0.64)

-4.011***

(-3.51)0.161(1.00)

-0.137(-0.70)

-0.00313(-0.02)

-0.154(-0.94)

0.302(1.64)

0.181(0.94)

0.802***

(4.08)0.131(0.76)

0.572**

(2.76)-0.00687(-0.04)

Observations 3069 3050 3083 3082 3080 3073 3077 3078 3081 3083 3083 3082R2 0.287 0.388 0.0579 0.0382 0.0395 0.0313 0.0471 0.0503 0.196 0.0537 0.113 0.0333Rho 0.397 0.555 0.397 0.465 0.378 0.329 0.406 0.299 0.318 0.381 0.344 0.414

QOF income exposure * 2008

-0.00289(-1.88)

0.0735(1.72)

-0.00400(-0.67)

0.00458(0.63)

-0.00248(-0.35)

0.00314(0.48)

0.000682(0.11)

-0.00869(-1.44)

0.00570(0.74)

0.00367(0.67)

-0.00364(-0.49)

-0.00351(-0.64)

QOF income exposure

0.00215(1.35)

0.0686(1.57)

-0.00518(-0.92)

-0.00887(-1.38)

0.000981(0.15)

-0.0108(-1.79)

-0.00610(-0.93)

0.000892(0.15)

-0.0134(-1.93)

-0.00806(-1.42)

-0.00695(-0.92)

0.000242(0.04)

2008 0.0762(1.20)

-5.046**

(-3.14)0.329(1.47)

0.155(0.56)

-0.109(-0.46)

-0.186(-0.85)

0.126(0.50)

0.426(1.80)

-0.218(-0.83)

-0.115(-0.52)

0.449(1.66)

0.263(1.23)

Observations 2709 2687 2722 2725 2717 2722 2716 2721 2723 2724 2721 2712R2 0.267 0.306 0.0340 0.0255 0.0198 0.0266 0.0329 0.0280 0.0932 0.0392 0.0436 0.0315Rho 0.221 0.438 0.357 0.252 0.306 0.250 0.376 0.337 0.287 0.335 0.295 0.320

t statistics in parentheses (robust standard errors clustered by GP)* p < 0.05, ** p < 0.01, *** p < 0.001Control variables included in all models, Rho (the intra-class correlation coefficient) shows the proportion of the error variance attributed to variation across GPs

29

Figure 1: P4P exposure histograms for 2005 and 20080

.05

.1.1

5D

ensi

ty

5 10 15 20 25Income exposure 2005

0.0

5.1

.15

Den

sity

10 20 30 40 50 60Income exposure 2008

1st and 99th percentiles have been dropped

30