Is

JC

a

ARRA

JI

KGIEPMM

1

tj[iitItbrt

0h

Health Policy 108 (2012) 140– 148

Contents lists available at SciVerse ScienceDirect

Health Policy

j ourna l ho me p ag e: www.elsev ier .com/ locate /hea l thpol

ntrinsic and extrinsic motivations in primary care: An explanatorytudy among French general practitioners

onathan Sicsic, Marc Le Vaillant, Carine Franc ∗

ERMES3, UMR8211, Inserm U988, Site CNRS, 7, rue Guy Moquet, 94801 Villejuif Cedex, France

r t i c l e i n f o

rticle history:eceived 3 February 2012eceived in revised form 23 August 2012ccepted 25 August 2012

EL classification:18

eywords:eneral practitioners

ntrinsic motivationxtrinsic motivationayment for performance

a b s t r a c t

Background: Like many other OECD nations, France has implemented a pay-for-performance (P4P) model in primary care. However, the benefits have been debated,particularly regarding the possibly undesirable effects of extrinsic motivation (EM) onintrinsic motivation (IM).Objective: To examine the relationship between French GPs’ IM and EM based on an intrinsicmotivation composite score (IMCS) developed for this purpose. If a negative relationship isfound, P4P schemes could have side effects on GPs’ IM that is a key determinant of qualityof care.Method: From data on 423 GPs practicing in a region of France, IM indicators are selectedusing a multiple correspondence analysis and aggregated from a multilevel model.Results: Several doctors’ characteristics have significant impacts on IMCS variability,especially group practice and salaried practice. Qualitative EM variables are negatively

ultiple correspondence analysisultilevel analysis correlated with the IMCS: GPs who report not being satisfied with their income or feel-

ing “often” constrained by patients’ requests in terms of consultations length and officeappointments obtain a lower mean IMCS than other GPs.Conclusion: Our results provide a cautionary message to regulators who should take intoaccount the potential side effects of increasing EM through policies such as P4P.

. Introduction

In the majority of OECD countries, general medical prac-ice is in crisis. Various symptoms include increasing GPob dissatisfaction, low morale and general unhappiness1,2]. General practice does not attract young physicians;n France, 33% of the positions offered in general practicen 2010 remained vacant [3]. Moreover, there are difficul-ies in recruiting and retention, particularly in rural areas.n this difficult context, one likely reason for the persis-ence of problems in general practice is the income gap

etween specialities. General medicine is one of the least-emunerated medical specialties, and this is particularlyrue in France [4–6].

∗ Corresponding author. Tel.: +33 149583727.

168-8510/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.healthpol.2012.08.020

© 2012 Elsevier Ireland Ltd. All rights reserved.

Pay-for-performance (P4P) payment schemes havebeen introduced in many OECD countries over the last 10years. These policies base a part of each doctor’s incomeon indicators designed to measure individual performance[7]. More recently, France engaged in this type of reformwith the CAPI1 in 2009, which has been generalised tothe entire population of GPs from January 2012. The ratio-nale for these policies is to provide monetary incentivesfor physicians to enhance preventive care provisions andchronic disease follow-up in an effort to improve the qual-ity of care [8]. P4P can also be viewed as a way to achieve

several other indirect goals, such as increasing GPs’ com-pensation and improving computer utilisation, informationsharing and feedback [9].

1 Contrat d’Amélioration des Pratiques Individuelles.

Policy 1

intrinsically satisfying and rewarding [20,27]. In the socialscience approach, individuals may experience IM to play asocial role in an organised system [23,28]. Thus, according

J. Sicsic et al. / Health

Nevertheless, the efficiency of such additional pay-ments is debated, and the need to implement targetpayment schemes is questionable because the relationshipbetween pay and performance has not been well estab-lished [10–12]. Additionally, there are potentially adverseeffects, such as patient or disease selection [13–15]. Severalstudies have shown that incentive schemes may decreasedoctors’ quality of life and have negative consequences onend-user satisfaction [16,17]. Such schemes have also beenassumed to be one of the reasons for increasing levels ofprofessional burn-out [1,17].

Another potential side effect of these policies, asaddressed in many theoretical papers, is that extrinsicrewards, such as P4P incentives, may crowd out IntrinsicMotivation (IM). IM emerges from activities that responddirectly and immediately to individual needs. Such moti-vations have positive impacts on both work engagementand quality of care [18,19]. Historically, the concept wasfirst defined by psychologists, who provided evidencefor a crowding-out effect of IM by Extrinsic Motivation(EM) in the educational field [20,21]. Subsequent studiesin the context of various social science approaches havesuggested a negative relationship between IM and EM,supporting the hypothesis of the motivation crowding-out[22–26]. This issue may constitute an important health pol-icy consideration, as the economic efficiency of any P4Pmechanism and the long-term effects of such paymentschemes are questionable.

Our aim in this exploratory study is to develop a com-posite score of GPs’ IM, examine its determinants throughdoctors’ characteristics and, finally, test the relationshipbetween this score and GPs’ EM. Because our study iscross-sectional, we focus on the substitutability or com-plementarity between GPs’ IM and EM to highlight thepossible side effects of increasing EM through policies suchas P4P. We find that highly intrinsically motivated GPs maybe less extrinsically motivated, although their turnover isnot significantly different. In addition, GPs’ practice charac-teristics explain the variability in the IM composite score,independently of their age or sex.

The paper is organised as follows. Section 2 describes thedata used and the methods to develop the Intrinsic Motiva-tion Composite Score (IMCS). Section 3 presents the results,which are discussed in Section 4.

2. Materials and methods

2.1. Data

A survey was conducted based on an auto-questionnaireof GPs working in community-based offices in south-west France (Midi-Pyrénées). Data were collected amongGPs registered in the database of the Regional Unionfor Private Practitioners (Union Régionale des MédecinsLibéraux—URML). In collaboration with the URML, an infor-mative letter presenting the study was sent in the middleof March 2010, two weeks before the anonymous postal

questionnaire was mailed. A recall letter was sent to allgeneral practitioners three weeks after the questionnaire.Due to budget and time constraints (questionnaires hadto be returned before the summer period), we decided to

08 (2012) 140– 148 141

collect and compensate up to 500 respondents. At the endof July, 438 questionnaires were returned, and 423 wereusable.

Considering that medical profession has a number ofspecific characteristics involving both the provision of awide range of technical tasks which require high level ofexpertise and responsibility (diagnosis, treatment selec-tion, patient monitoring, etc.) as well as non technical skillsduring patient–doctor relationship (listening of the patient,educating the patient, compassion, etc.), we chose to con-sider a specific questionnaire to measure IM among GPs.The questionnaire is composed of 68 items divided intofour parts. The first part concerns GPs’ practice organisa-tion (e.g., group/solo practice, partial salaried/only privatepractice, engaged in the continuity of care). The secondpart addresses the GPs’ working activities (e.g., preventionactivities, working hours, turnover). The third part char-acterises doctors’ satisfaction and professional constraints.Finally, the fourth part is dedicated to additional questionsthat are mainly socio-demographic in nature (e.g., age, gen-der, perceived state of health). The GPs were asked to reportinformation on medical activity described in the annualdocument from the Public Health Fund that summarisesindividual turnover and the details of services providedby each GP throughout the year (individual professionalactivities record2). For the majority of the variables relatedto their activities (parts 1 and 2), GPs were asked toanswer on a four-point Likert scale (“very often”, “often”,“sometimes”, or “rarely”). Concerning the satisfaction andprofessional constraints variables (part 3), they answeredon another four-point Likert scale (“often”, “sometimes”,“rarely”, or “never”). In other cases, they replied to binaryvariables (“yes/no”).

The sample is representative of the overall French GPpopulation according to gender (72% are male, as in thetotal French GP population), age (the mean age is 52.1 vs.50.1 years), participation in group practice (58% belong toa group vs. 54% in the total French GP population) and indi-vidual turnover (the average turnover free of charge is D71,364 vs. D 71,690 in the French GP population).

2.2. Variables

2.2.1. Selection of the IM variablesWe are interested in computing the IMCS for GPs. The

first step consists of selecting and validating IM indicatorsamong physicians. Because no standardised questionnairehas been developed to measure IM among GPs, the conceptof IM remains an unobserved characteristic of GPs, and ourselection of IM variables is based on explanatory methods.The variables are first selected based on previous results inthe literature. There are generally two conceptions of IM.In the psychological approach, individuals derive pleasurefrom performing an activity, and the activity is considered

2 “Relevé Individuel d’Activité Professionnelle” annually provided bythe Public Fund.

1 Policy 1

to

wataw

ftstfpwtavopEataeatnb“tmt“a(

iteoGshrtp(opcCc1tioGpw

rural or urban location of the practice (≤10, 000/>10, 000inhabitants) [31].

42 J. Sicsic et al. / Health

o Lindenberg [29], there can be more than one dimensionf IM.

The variables are selected based on hypotheses abouthether they at least partly reveal IM. Dichotomous vari-

bles are created to allow the same weight to be assignedo each variable. The candidate variables are coded 1 when

GP’s answer indicates IM (positive answer) and 0 other-ise. The relevant descriptive statistics are listed in Table 1.

A first group of selected candidates collects variablesor which we assume that GPs answer positively whenhey experience IM at work. Three variables are selected:atisfaction with their professional activity, the feelinghat the medical profession is personally rewarding andrequent exchanges with other GPs or specialists aboutatients’ situations. The first two variables are associatedith the concept of enjoyment at work [20]. It is known

hat two factors may impact satisfaction with professionalctivity (job satisfaction): intrinsic and/or extrinsic moti-ation [30]. In the questionnaire, GPs were also askedn the satisfaction with their income. To properly inter-ret job satisfaction as a candidate to reflect IM and notM, we tested the relationship between job satisfactionnd either income satisfaction or turnover and found thathere were no statistical relationship. Initially, the GPs weresked to respond to the statements on a four-point Lik-rt scale (“often”, “sometimes”, “rarely”, and “never”). Wessumed that only physicians who answered “often” tohese variables experienced IM, in order to have discrimi-ant indicators. The majority of physicians reported “often”eing satisfied with their professional activity (58%) andoften” finding it to be personally rewarding (63%). Thehird variable may indicate a commitment to the results of

edical practice. Initially, the GPs were asked to respondo the statements on a four-point Likert scale (“very often”,often”, “sometimes”, and “rarely”). We define a positivenswer as doctors who replied “very often” to this variable19% of GPs).

A second group of selected indicators represents activ-ties that indicate a motivation, an investment, to increasehe quality of care and/or to transfer and to share knowl-dge, such as teaching and tutoring activities (19% of GPs)r participation in the continuity-of-care system (55% ofPs). Theoretically, such activities could also be extrin-ically motivated if well remunerated, but we reject thisypothesis because our indicators represent activities thatequire time and for which payment is rather low. Addi-ionally, the questionnaire investigates the frequency ofhysicians’ accomplishment of three prevention activitiesalcohol, tobacco, and obesity) using four categories: “veryften”, “often”, “rarely”, and “never”. Alcohol and tobaccorevention activities are considered to be the same indi-ator because they are closely related (p < 0.0001 in thehi-square test of independence), and obesity prevention isonsidered separately. A first prevention indicator is coded

if the physician reported providing either alcohol orobacco prevention “very often” (15%). A second preventionndicator is coded 1 if the physician replied “very often” for

besity prevention (20%). Finally, indicators concerning thePs’ participation in professional practice evaluation (42%roviding positive answers) and involvement in a care net-ork (31% providing positive answers) are selected. Such

08 (2012) 140– 148

variables may also indicate the motivation to improve thequality of care utilising peer advice.

The third and last group of two selected variables con-cern physicians’ attitudes toward more informed patientsand toward health insurance control. The first variableindicate how often physicians feel constrained by moreinformed patients (“often”, “sometimes”, “rarely”, and“never”). Here, we assume that GPs experience IM whenthey answer “rarely” or “never” (40%) because theseanswers indicate that the GP is not unwilling to take timeeducating the patients on the rationale of his therapeu-tic decision. A second variable indicates whether they feelconstrained by health insurance. We hypothesise that anegative answer for this variable may indicate a higherlevel of IM (67% of GPs), because feeling of autonomy playsan important role in IM [27].

2.2.2. Extrinsic motivation variablesIn contrast to IM, EM refers to motivations that come

from “outside” the person, when the activity is performedto obtain an external reward (e.g., income or praise) [23].Five variables are selected as proxies for EM. Three vari-ables examine GPs’ monetary motivations: (a) the turnoverin 2008, (b) the turnover per service provided by the GP in2008 and (c) income satisfaction (“yes/no”). Indicator (b)may be more informative than indicator (a) because it canconstitute a proxy for the physician’s consultation fee orproductivity. For variable (c), we hypothesise that GPs whoare not satisfied with their income may be more interestedin money and thus have greater EM.

The other non-monetary variables considered as prox-ies for EM are the feeling of being constrained bydemanding patients in terms of either (d) consultationlength or (e) office appointments. Indeed the two vari-ables may reflect the doctor’s trade-off between work andleisure. GPs who report to feel “often” or “sometimes” con-strained by patients request in terms of consultation lengthor office appointments may not want to lose time, whichwould imply losing money or leisure time, appear to beextrinsically motivated; these answers are coded 1 and 0otherwise (i.e., when doctors report to “rarely” or “never”feel constrained) (see Table 1 for the descriptive statistics).

2.2.3. Doctors’ characteristicsThe individual GP variables describe GPs by using char-

acteristics that define them intrinsically as individuals (i.e.,age, gender, and health status), as practitioners (type ofpractice: solo or group practice sharing medical records3,private or salaried practice). The variables also describethe context of the activity (i.e., whether the GP encoun-ters difficulties with the patients’ profile), and, finally, the

3 Here, we employed a more precise definition of group practice, inwhich doctors were considered to be “true” members of a group if theyshared their patients’ medical records.

J. Sicsic et al. / Health Policy 108 (2012) 140– 148 143

Table 1Descriptive statistics for the intrinsic and extrinsic motivation variables and usual GPs’ variables (number = 423).

Variables Modalities

1—Intrinsic motivation (%)Participation in the continuity-of-care system Yes /no 54.6/45.4Exchanges with GPs/specialists about patients’ situations Very often/often–sometimes–never 19.4/80.6Teacher or tutor activities Yes/no 19.6/80.4Provision of alcohol or tobacco prevention Very often/often–sometimes–never 14.9/85.1Provision of obesity prevention Very often/often–sometimes–never 20.3/79.7Involvement in a care network Yes /no 31.2/68.8Satisfaction with professional activity Often/sometimes–rarely–never 57.9/42.1The job is considered personally rewarding Often/some times–rarely–never 62.9/37.1Feeling constrained by more informed patients Rarely–never/often–sometimes 47.9/52.1Feeling constrained by health insurance Yes/no 33.5/66.5Participation in professional practice evaluation Yes/no 41.6/58.42—Extrinsic motivation (%)Feeling constrained by demanding patients . . .

. . . on consultation duration Often–sometimes/rarely–never 55.3/44.7

. . . on office opening Often–sometimes/rarely–never 54.8/45.2Income satisfaction Yes/no 34.4/65.6Turnover Mean (± S.D.) 125,000 (± 55,000) (D )Turnover/number of acts Mean (± S.D.) 26.6 (±5.1) (D )3—GP characteristicsAge Mean (± S.D.) 52.1 (±7.6)Sex Men/women 71.8/28.2 (%)Status Solo/in group practice 42.3/57.7 (%)Health status Very good/poor 89.1/9.0 (%)Location City/rural 59.6/40.4 (%)Activity Hospital/only private practice 20.6/79.4 (%)Difficulties with patients’ characteristics Yes/no 24.1/72.8 (%)

examine the relationship between the IMCS and contin-

2.3. Methods for the step-by-step construction of theIMCS

2.3.1. IM indicatorsA Multiple Correspondence Analysis (MCA) [32] is per-

formed to examine the relationship between the selectedIM variables and to determine how many dimensions ofIM are captured by our variables. MCA extends PrincipalComponent Analysis (PCA) to categorical data and enablesthe selection of a majority of correlated variables associatedwith a latent axis of IM. When finally selected, the variablesare labelled as “indicators” of IM.

2.3.2. The IMCS construction methodA model was defined to aggregate the IM indicators

into a composite score, the IMCS. This score is com-puted using a multilevel approach using SAS software®(NLMIXED procedure). This model enables us to disen-tangle the components of the IM score variability arisingfrom between- and within-doctor variability [33,34]. Thegoals here are twofold. First, we account for the covariancebetween doctors, and second, we allow for the introductionof parameters at the doctor’s level to explain the variabilityin the composite score [31,35].

We assume that a latent variable of IM �, with realisation�j, determines the answer of physician j to all of our selectedindicators. Moreover, we assume that these answers are

independent for each GP, conditionally on �j. The latentvariable is theoretically normally distributed with mean 0and standard deviation �� .

The empty multilevel model is thus defined below.

Model 1: the empty model

logit pij = �0j + eij (1)

�0j = ˇ00 + �j, �j → N(0, ��) (2)

Logit pij is the log odds ratio associated with the proba-bility of doctor j answering positively to indicator i. �0j isthe random coefficient varying from one doctor to another,and eij and �j are two random residuals: eij is the residualfor indicator i of doctor j (within-doctor variability), and �jis the residual of doctor j (between-doctor variability).

Model 2: the doctor model

To explain the IMCS variability, we introduce individ-ual covariates to the model at the GP level. Therefore, themodel becomes

logit pij = �0j + eij (3)

�0j = ˇ00 +K∑

k=1

ˇkxk + �j �j → N(0, ��) (4)

where xk are covariates representing the doctors’ charac-teristics.

2.3.3. Interactions between EM and IMSpearman correlation coefficients are computed to

uous EM variables, such as the turnover in 2008 and theturnover per service provided over the year. For categoricalEM variables such as income satisfaction, Wilcoxon tests[36] are computed.

144 J. Sicsic et al. / Health Policy 108 (2012) 140– 148

Table 2The multiple correspondence analysis results: contributionsa and coordinates of the 11 intrinsic motivation variables on the first two axesb (number = 423).

Intrinsic motivation variables Axis 1 Axis 2

Contribution Coordinates Contribution Coordinates

Satisfaction with professional activitySometimes–rarely–never 12.63 −0.85 5.47 −0.5Often 7.25 0.49 3.14 0.28The job is considered personally rewardingSometimes–rarely–never 11.60 −0.76 5.97 −0.48Often 8.33 0.54 4.29 0.35Participation in the continuity-of-care systemNo 0.44 −0.14 0.80 0.17Yes 0.36 0.12 0.66 −0.14Exchanges with GPs or specialists about patients’ situationsOften–sometimes–never 1.56 −0.20 0.00 0.00Very often 6.41 0.82 0.00 0.00Teacher or tutor activitiesNo 2.22 −0.24 0.00 0.00Yes 8.85 0.96 0.00 0.01Provision of alcohol or tobacco preventionOften–sometimes–never 1.08 −0.16 5.00 0.31Very often 6.00 0.90 27.86 −1.73Provision of obesity preventionOften–sometimes–never 1.36 −0.19 6.95 0.38Very often 5.25 0.73 26.90 −1.46Involvement in a care networkNo 2.72 −0.29 0.19 0.07Yes 5.94 0.63 0.43 −0.15Feeling constrained by more informed patientsRarely–never 4.71 0.51 0.31 0.12Often–sometimes 2.75 −0.34 0.56 −0.14Participation in professional practice evaluationNo 3.72 −0.36 0.00 0.01Yes 5.13 0.50 0.00 −0.01Feeling constrained by health insuranceNo 0.77 0.23 6.62 0.57Yes 0.38 −0.11 3.31 −0.28

a The contribution of a variable to an axis represents the extent to which it “explains” the construction of the axis.rrespon

b ivation

3

3

3

iTiafitihttaaoBctNr

3.1.2. Distribution of the IMCSOnce selected, the 9 indicators are entered into the mul-

tilevel model 1 to compute the IMCS. The distribution and

Table 3Summary of the selected intrinsic motivation indicators and the statedhypotheses (number = 423).

Intrinsic Motivation (IM) indicators Modality associated toIM indicators

Exchanges with GPs or specialistsabout patients’ situations

Very often

Teacher or tutor activities YesProvision of alcohol or tobacco

preventionVery often

Provision of obesity prevention Very oftenInvolvement in a care network YesSatisfaction with professional activity OftenThe job is considered as professionally Often

b An axis is a latent/unobserved variable generated by the Multiple Coetween a set of variables representing a same concept (i.e., intrinsic mot

. Results

.1. The IMCS

.1.1. Identification of IM indicatorsFirst, the 11 candidate variables identified as potential

ndicators of IM are defined as active variables in the MCA.he results of the MCA are displayed in Table 2. Accord-ng to the Benzecri correction for eigenvalues [37], the firstxis preserved 71% of the total inertia in the data, and therst two axes conjointly preserve 95%. Thus, the interpre-ation of relationships between modalities of IM variabless limited to the first two axes. All variables are found toave significant contributions to the first axis (i.e., a con-ribution superior to the mean contribution), except forhe two variables “participation in the continuity of care”nd “feeling of being constrained by health insurance”. Inddition, all significant variables have the same coordinaten this axis, indicating that they are positively correlated.ased on these results, the 9 indicators with significant

ontribution to the first axis are selected to constructhe IMCS (Table 3 presents a summary of this selection).ote that the prevention activity indicators are less cor-

elated to the other indicators, especially the satisfaction

dence Analysis (MCA), which enables to summarize correlation patternsin our study).

and the feeling of personal reward from the activity(see Table 2, axis 2).

rewardingFeeling constrained by more informed

patientsRarely–never

Participation in professional practiceevaluation

Yes

J. Sicsic et al. / Health Policy 108 (2012) 140– 148 145

tivation

Fig. 1. Distribution of the intrinsic mo

cumulative distribution of the IM score are displayed inFigs. 1 and 2, respectively. The score is distributed from 0to 1: a doctor with a score of 0 is not at all intrinsicallymotivated, and a doctor with a score of 1 is intrinsicallymotivated at the highest possible level. These extrema areintuitively theoretical, and none of the doctors in our anal-ysis obtained scores less than 0.31 or greater than 0.81.Half of the GPs have scores greater than 0.5, which is alsothe mean IMCS (standard deviation = 0.12). For GPs whoanswered positively to less than one indicator, the mini-mum score is 0.31. Thus, the 10% of the GPs who are leastintrinsically motivated have an IMCS less than 0.37. Simi-larly, the maximum score of 0.81 is obtained by GPs whoanswered positively to more than 8 of the 9 indicators. The

10% of the GPs who are most highly intrinsically motivatedhave an IMCS greater than 0.68. For 50% of the GPs, the IMCSis between 0.44 and 0.57.

Fig. 2. Cumulative distribution of the intrinsic motivation composite score (blacurve) (number =423 GPs).

composite score (number = 423 GPs).

3.1.3. The GPs’ determinants of IMCS variabilityBased on the measured IMCS, we next examine which

characteristics of the doctors explain the score variabil-ity. The first step is to identify the effects that arise frombetween-doctor variability (Eq. 3 and Eq. 4). First, succes-sive univariate analyses are performed to select the mostrelevant physicians’ characteristics to be entered into thefinal multivariate model. The results of the univariate andmultivariate models are displayed in Table 4. In the uni-variate models, three GP characteristics appear to havesignificant impacts on between-doctor variability. First,“group practice” and “being salaried in addition to a privatepractice status“(mixed activities) play positive and signifi-cant roles (p < 0.0001 and 0.0001, respectively) in the level

of doctors’ IM. In other words, being salaried or workingin a group increases the probability of obtaining a highIMCS, illustrating a high level of IM. Moreover, the GPs’

ck curve) and adjustment with a normal cumulative distribution (grey

146 J. Sicsic et al. / Health Policy 108 (2012) 140– 148

Table 4Results of the multilevel model: impact of physicians’ characteristics on the probability of answering positively for one indicator (number = 423a).

Level Variableb Univariate analyses Multivariate analysis

Estimate p-value Estimate p-value

Fixed Constant −1.2501 <.0001Age (≤45 years, 45 < Age < 60) −0.08828 0.5044Age (45 < Age < 60, ≥60 years) −0.1018 0.4571Gender (male, female) 0.055 0.6216Practice (group, solo) 0.4433*** <.0001 0.3389*** <.0001Location (urban, rural) 0.1265 0.2515Health status (very good, poor) 0.4197** 0.0248 0.3701** 0.0388Status (salaried, only liberal) 0.6145*** <.0001 0.5807*** <.0001Difficulties with patients’ characteristics (yes, no) 0.0384 0.7439

Level Variance

RandomGP 0.4051

Level of statistical significance: *10%, **5%, ***1%.acteristi

c lts, becaty.

paiea

dedd

3

ctaia(

wef

rtctha(mcr

4

it

a We created separate categories of missing values for some GPs’ charategories were included into the model, but we did not report their resu

b Terms in italics between brackets correspond to the reference modali

erceived health status also explains between-doctor vari-bility; reporting an excellent or very good health statusncreases the estimated level of IM (p = 0.0248). It is inter-sting to note that neither age, gender, nor the location ofctivity has any effects on the IMCS.

When considering the different variables, the between-octor variability is reduced; it is equal to 0.5122 in thempty model and 0.4051 in the doctors’ model. Thus, theifferent covariates account for 24.5% of the between-octor variability.

.2. Interactions with EM variables

Because the IMCS is not perfectly normal, Spearman’sorrelation coefficients (rank tests) are computed to studyhe relationship between the IMCS and quantitative vari-bles of EM. The results are presented in Table 5. Theres no statistical relationship between the level of IMCSnd the turnover (p = 0.8322) or the turnover per servicep = 0.6586).

Note that 19% of the GPs did not report their turnover,hich calls into question the robustness of the results. Nev-

rtheless, non-respondents are not significantly differentrom respondents according to age, gender, location or IM.

Wilcoxon non parametric rank test are used to study theelationship between IMCS and other EM variables (quali-ative variables). GPs who reported rarely or never feelingonstrained by patients’ requests with respect to consulta-ion duration (or office appointments) have a significantlyigher IMCS . Moreover, GPs who obtain a higher IMCSre more often satisfied with their private practice incomep = 0.0309), which may testify to lower EM. This finding

ay also indicate that these GPs have lower expectationsoncerning their income, such that their target income iseached more often.

. Discussion

The first aim of this study is to construct a compos-te score for intrinsic motivation (IMCS) and to analysehe main determinants of this score among doctors’

c variables, so that the multivariate model was run with 423 GPs. Theseuse we are not interested in those effects.

characteristics. The second goal is to examine the relation-ship between the IMCS and various proxies for extrinsicmotivation (EM). Several indicators of IM are selectedthrough a Multiple Correspondence Analysis (MCA),revealing a latent axis of IM. These indicators mainly repre-sent GPs’ satisfaction with their professional activities andtheir commitment to public health objectives (e.g., involve-ment in a care network, prevention provision). There is atheoretical debate on whether “social intrinsic motivation”should be distinguished from “enjoyment-based intrinsicmotivation” and if “social intrinsic motivation” can be dis-tinguished from extrinsic motivation [23]. Deci and Ryan[27] did not assume different dimension of IM but showedthat there is a continuum of motivation, which is coher-ent with the results of our exploratory analysis: only onedimension of IM has been identified by the MCA. Never-theless, the two prevention activity variables were foundto be less associated with other variables. Moreover, furtheranalyses showed that GPs who report frequent provision ofprevention activities also report feeling constrained by thehealth insurance more often (p = 0.0484) and report theirturnover less often (p = 0.0161). We cannot exclude poten-tial declaration bias, such as memory or social desirabilitybias, due to the declarative nature of the data. Neverthe-less, these results are very interesting in that they questionthe relevance of indicators for the provision of preventionas proxies for IM. Indeed, these variables are often used asproxies for IM because prevention provision occupies timewithout financial compensation [38]. It would be inter-esting to collect more precise data concerning preventionprovision (time dedicated to prevention or specific actions).

The IMCS is centred on an average probability of 0.5for positively replying to the indicators and ranges froma probability of 0.31 to 0.81. For 50% of the GPs, the IMCSis between 0.44 and 0.57. We focused on the GPs’ charac-teristics that impacted between-physician IMCS variability.First, we find that GPs who practice in a group (shar-

ing medical records) have higher IMs than those in solopractice. This is an interesting result because prior stud-ies have often cited rationales for joining a group thatinclude more flexible schedules, reduced workload and

J. Sicsic et al. / Health Policy 108 (2012) 140– 148 147

Table 5Spearman’s correlation coefficients and Wilcoxon rank test between the intrinsic motivation composite score and extrinsic motivation variables.

Extrinsic motivation quantitative variables (number = 342) Spearman correlation coefficient p-value

Turnover −0.00942 0.8622Turnover/number of acts 0.02487 0.6586

Extrinsic motivation qualitative variables (number = 423) Mean rank suma Wilcoxon statistic p-value

Patients constraints on consultation duration 40,770** 0.0496Rarely–never 219Often sometimes 199Patients constraints on office openings 41,315** 0.0181Rarely–never 220Often–sometimes 196Income satisfaction (yes vs. no) 31,104** 0.0113Yes 222No 194

Level of statistical significance: *10%, **5%, ***1%.n his in

a The rank of physician j is calculated based on the difference betwee

database. The higher the difference, the higher is the rank.

shared office costs [39]. Such incentives may be interpretedin light of EMs. In our data, the more precise definitionof group practice (the doctors also share medical records)may lead to the selection of more intrinsically motivatedphysicians. Indeed, by sharing medical records, GPs may beconfronted with peer judgements and/or knowledge andmay be involved in collegial decision processes. Two othervariables at the doctor’s level impact between-doctors’variability: the diversified nature of the medical exercise(liberal and salaried) and, less surprisingly, excellent orvery good declared health status. Salaried exercise maydenote a willingness to pursue a diversified activity. Ulti-mately, 75% of the between-doctor covariance remainsunexplained by the model. This remaining variance maybe due to other psychological factors that influence thedoctors’ IM.

Overall, our study provides mixed results concerningthe substitutability between IM and EM. The results differdepending on the type of EM measures. For the mone-tary measures, the IMCS is not significantly correlated withturnover or turnover per service. According to our analy-sis, this result is not due to the method used to computethe score because more complex methods do not gen-erate scores significantly related to turnover.4 However,note that 19% of the GPs decided not to report their 2008turnover. Even if these non-respondents do not appearto differ significantly from the other doctors based onage, gender, or location, the non response may at leastpartially explain this result. Considering the qualitativeEM indicators, we find a significant negative relationshipwith the IMCS: a highly intrinsically motivated GP is less

likely to report feeling constrained by patients’ requestswith respect to consultation duration or office appoint-ments. Similarly, highly intrinsically motivated GPs are

4 We actually ran several scores using sum scoring and item responsemodels: one-parameter and two-parameter logistic models. The resultswere not significantly different according to the IM indicators aggregatingmethod.

trinsic motivation composite score (IMCS) and the average IMCS in the

more likely to report being satisfied with their privatepractice incomes. These variables appear to be signifi-cantly associated with the IMCS in the expected way, i.e.,assuming substitutability between IM and EM. For exam-ple, doctors who worry about consultation duration mayhave in mind the opportunity cost of work.

In conclusion, based on our IM scores, there may benegative relationships between GPs’ IM and EM. Thus, reg-ulators should be aware of the potential side effects ofpayments when implementing P4P schemes. Beyond thisstudy, important work remains to be done in France and inother countries to examine the potential perverse effectsof P4P, especially the erosion of IM by extrinsic rewards.Future studies should focus on the measurement of IMin the health care context through standardized question-naires. This study provides a path for further research onthe precise definition of IM indicators and ways to measureIM through statistical analysis. Furthermore, it would beinteresting to collect data over time to provide additionalevidence concerning the motivation crowding theory. Therecent implementation of the P4P policy in France andother countries should lead to a review of several IM indi-cators (e.g., prevention activities) because these activitieswill incorporate several forms of extrinsic motivation.

References

[1] Smith R. Why are doctors so unhappy. British Medical Journal2001;322:1073–4.

[2] Dowel C, Coster G, Maffey C. Morale in general practice: crisis andsolutions. The New Zealand Medical Journal 2002;115(1158).

[3] Fauvet. Les affectations des étudiants en médecine à l’issue desépreuves classantes nationales en 2010. DREES 2001. Etudes et Résul-tats n◦767.

[4] OECD. The remuneration of general practitioners and specialists in14 OEDC countries: what are the factors influencing variations acrosscountries? OECD Health Working Papers; 2008. no. 41.

[5] Bellamy V. Les revenus libéraux des médecins en 2007 et 2008. DRESS2010. Etudes et Résultats, n◦765.

[6] Kroneman MW, Van der Zee JV, Groot W. Income development ofGeneral Practitioners in eight European countries from 1975 to 2005.BMC Health Services Research 2009;9:26.

[7] Lindenauer PK, Remus D, Roman S, Rothberg MB, BenjaminEM, Ma A, et al. Public reporting and pay for performance in

1 Policy 1

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

48 J. Sicsic et al. / Health

hospital quality improvement. The New England Journal of Medicine2007;356:486–96.

[8] Hughes D. General practitioners and the new contract: pro-moting better health through financial incentives. Health Policy1993;25:39–50.

[9] Lanier D, Roland M, Burstin H, Knottnerus JA. Doctor performanceand public accountability. Lancet 2003;362:1404–7.

10] Croxon B, Propper C, Perkins A. Do doctors respond to financial incen-tives? UK family doctors and the GP fundholder scheme. Journal ofPublic Economics 2001;79:375–98.

11] Armour BS, Pitts MM, Maclean R, Cangialose C, Kishel M, Imai H, et al.The effect of explicit financial incentives on physician behaviour.Archives of Internal Medicine 2001;161:1261–6.

12] Li J, Hurley J, DeCicca P, Buckley G. Physician response to pay-for-performance: evidence from a natural experiment. National Bureauof Economic Research 2011:16909.

13] Mannion R, Davies H. Payment for performance in health care. BritishMedical Journal 2008;336:306–8.

14] Gosden T, Forland F, Kristiansen IS, Sutton M, Leese B, Griffrida A,et al. Impact of payment method on behaviour of primary care physi-cians: a systematic review. Journal of Health Services Research Policy2001;6(1):44–55.

15] Giuffrida A, Gosden T, Forland F, Kristiansen IS, Sergison M. Tar-get payments in primary care: effects on professional practiceand health care outcomes. Cochrane Database Systems Review 32000:CD000531.

16] Hadley J, Mitchell JM, Sulmasy DP, Bloche MG. PerceivedFinancial Incentives, HMO market penetration, and physicians’practice styles and satisfaction. Health Services Research 1999;34:1.

17] Badia JG, Escaramis-Babiano G, Sans-Corrales M, Sampietro-Colom L,Aguado-Menguy F, Cabezas-Pena C, et al. Impact of economic incen-tives on quality of professional life and on end-user satisfaction inprimary care. Health Policy 2007;80:2–10.

18] Benware C, Deci EL. Quality of learning with an active versuspassive motivational set. American Educational Research Journal1984;21:755–65.

19] Janus K. Managing motivation among health care professionals.Advances in health care management 2010;9:47–77.

20] Deci EL. Effect of externally mediated rewards on intrinsic moti-vation. Journal of Personality and Social Psychology 1971;18:105–15.

21] Deci EL, Koestner R, Ryan RM. Meta-analytic review of experiments:examining the effects of extrinsic rewards on intrinsic motivation.Psychological Bulletin 1999;125(6):627–68.

22] Gibbons R. Incentives in organizations. Journal of Economic Perspec-tives 1998;12(4):115–32.

[

08 (2012) 140– 148

23] Frey BS. Not just for money: an economic theory of personal motiva-tion. Cheltenham, UK/Brookfield, USA: Edward Elgar; 1997.

24] Bénabou R, Tirole J. Intrinsic and extrinsic motivation. Review ofEconomic Studies 2003;70(3):489–520.

25] Harvey S, James Jr. Why did you do that? An economic examinationof the effect of extrinsic compensation on intrinsic motivation andperformance. Journal of Economic Psychology 2005;26:549–66.

26] Bowles S. Policies designed for self-interested citizens may under-mine ‘the moral sentiments’: evidence from economic experiments.Science 2008;320(5883):1605–9.

27] Deci EL, Ryan RM. The “what” and “why” of goal pursuits: humanneeds and the self-determination of behavior. Psychological Inquiry2000;11(4):227–68.

28] Kreps DM. Intrinsic motivation and extrinsic incentives. The Ameri-can Economic Review 1997;87(2):359–64.

29] Lindenberg S. Intrinsic motivation in a new light. Kylos2001;54(2/3):317–43.

30] Goetz K, Campbell SM, Broge B, Dörfer CE, Brodowski M, Szecsenyi J.The impact of intrinsic and extrinsic factors on the job satisfaction ofdentists. Community of Dental and Oral Epidemiology; 2012. April,p. S1600-0528.

31] Pelletier N, Le Vaillant M, Hebbrecht G, Boisnault P. Determinantsof preventive services in general practice. A multilevel approachin cardiovascular domain and vaccination in France. Health Policy2007;81:218–27.

32] Greenacre MJ. Theory and applications of correspondence analysis.London: Academic Press; 1984.

33] Rice N, Jones A. Multilevel models and health economics. HealthEconomics 1997;6:561–75.

34] Snijders T, Bosker R. An introduction to basic and advanced multilevelmodelling. London: Sage Publication; 1999.

35] Duncan C, Jones K, Moon G. Context, composition and heterogene-ity: using multilevel models in health research. Social Science andMedicine 1998;46(1):97–117.

36] Conover WJ. Practical nonparametric statistics, 3rd ed. New York:John Wiley & Sons; 1999.

37] Benzécri JP. Sur le calcul des taux d’inertie dans l’analyse d’un ques-tionnaire. Cahiers de l’Analyse des Données 1979;4:377–8.

38] Videau Y, Batifoulier P, Arrighi Y, Gadreau M, Ventelou B. Le cyclede vie des motivations professionnelles en médecine générale: uneétude dans le champ de la prévention. Revue d’Epidémiologie et deSanté Publique 2010;58(5):301–11.

39] Bourgueil Y, Chambaretaud S, Marek A, Mousqès J. Les nou-velles formes de coopération entre professionnels de santé:premiers enseignements des expériences internationales. In Enjeuxéconomiques des coopérations entre professionnels de santé. ParisHAS; 2007/12. p. 46–59.