For what services do general practitioners induce demand? Economic incentives and professional norms Lotte Bøgh Andersen & Søren Serritzlew Paper presented at the Department of Economics, University of Copenhagen, 16. November 2007 Abstract Whether general practitioners (GPs) induce demand for their own services, when they experience a shortage of patients, is a point of disagreement in the literature. Likewise, the literature does not agree upon the existence of rationing in case of too many patients. We argue that the GPs induce demand and ration if no firm professional norm regulates the use of the specific service. GPs with few patients give their patients more of services without professional norms than GP with many patients, while there is no difference for services governed by professional norms. This indicates that both economic incentives and professional norms are important. The conclusions are based on register data concerning the GPs’ use of different services (n= 257 practices in the County of Aarhus), six qualitative interviews and documentary material. Introduction The income of general practitioners (GPs) in most western countries depends on the number (and type) of services supplied to the patients, because many payment systems include fee-for-service elements (Gosden et al. 1999 & 2001; Groenewegen et al 1991; Hoffmeyer & McCarthy 1994). As expected by standard agency theory, fee-for-service contracts lead to higher service production than salary and capitation contracts (Sørensen & Grytten, 2003; Krasnik et al 1990). Letting the GPs income depend on the number of services produced can, however, have unintended and undesirable effects if the GPs in fee-for-service systems prioritize their own pecuniary interests higher than public economy and patient welfare. The GPs might provide too many services per patient if they have few patients on the GPs lists. ‘Supplier-induced demand’ happens when the suppliers (here the GPs) try to produce more services than initially demanded by patients or ordered by society. If the inconvenience of producing the marginal service exceed the fees (due to many patients), they might on the other hand provide fewer services than desirable seen from the societal perspective (and especially seen from the patient perspective). This is called ‘rationing’. There is, however, 1 considerable controversy among health care researchers about the existence of supplier-induced demand and rationing. Some researchers find evidence of induction/rationing (Evans 1974; Richardson & Peacock 2006; Bech et al. 2007), and others find no relationship between the list size and the number of services (Davis et al. 2000; Madden et al., 2005; Grytten & Sørensen, 2007). We argue that this is because the existence of firm professional norms conditions the relationship. We hypothesize that if a specific medical service is firmly regulated by professional norms, the number of services per patient does not depend on the list size. Otherwise, we expect the number of services per patient to be higher, the fewer patients the GPs have. This proposition is tested on the very reliable Danish health insurance data which registers the exact use of 70 different GP-services. The structure of the paper is as follows: We first discuss the literature on supplier-induced demand followed by a brief introduction to the Danish GP payment system. We then use six qualitative interviews and documentary material to map out the relevant professional norms, which enables us to put forward the testable hypotheses. After a short description of the data and the methods, we test the hypotheses, and the paper concludes with a discussion about the interpretation of the results and the implications for further research. The literature on supplier-induced demand More than 35 years ago, Newhouse claimed that health care providers could create the demand for their own services (Newhouse 1970). The validity of this claim soon became one of the most controversial topics in health economics (Richardson & Peacock, 2006, 2). The central concepts in the supplier-induced demand (SID) literature are supplier-induction and rationing, which are physicians’ use of their market power to respectively increase and decrease the demand for their services. The market power of GPs is based on the asymmetric distribution of information. GPs have much more information about the service production than the patients. Patients seldom know how many (and which) medical services they need. Part of the SID-literature expects the GPs to take advantage of this information asymmetry and affect the demand if this is lower than the GPs want. This implies that the number of potential patients affects the service production: The fewer patients (that is, the lower demand for services), the more will the GP try to induce extra demand. In a list system (where the each GP has a list of patients attending only this GP) the SID-logic implies that GPs with short lists (low initial demand) try the hardest to increase demand and therefore tries to provide more services per patient than GPs with long lists. Physicians with very long lists might, 2 on the other hand, ration their services, because the disutility of providing the marginal service exceeds the benefits (the fee). Both supplier induction and rationing would lead to a negative relationship between the number of services per patient and the number of listed patients per GP (the list size). Several studies have identified this relationship empirically (e.g. Evans 1974; Richardson & Peacock 2006). One should not, however, leap from this association to the existence of supplier-induced demand. At least three alternative explanations of the negative association between list sizes and services per patient are mentioned in the literature (see Carlsen & Grytten 1998 for a more detailed discussion): First, if the fees are not fixed, and if the patients pay the services themselves, it can be difficult to distinguish a price effect from an inducement effect. Few patients per GP can drive the prices down, which increases the demand for services. Second, the availability of doctors (e.g. how long before the patients can get an appointment) can affect the demand. Patients might not want (or need) to go to the doctor as frequently if they must wait many days or drive long distances. Third, the number of patients per physician might be endogenous. If the GPs determine the list sizes themselves, doctors with very care-demanding patients might choose to have a short list (which also implies a negative relationship between the list size and the number of patients on the list, but with reversed time order). Even in literature considering these factors, the existence of supplier-induction is still a much contested issue in the literature. On one hand, Richardson and Peacock (2006: 14) for example argue that “SID [supplier induced demand] provides a satisfactory explanation of the observed pattern and change in the demand for Australian medical services”, and Delattre and Dormont concludes that ”Econometric results give a strong support for the existence of PID [physician-induced demand]” (2003: 741). On the other hand, Grytten and Sørensen (2007; 2001) for example find that long patient lists in Norway do no lead to rationing, and that short lists do not increase service production (for other studies, which find similar results for other countries, see for example Davis et al. 2000 and Madden et al. 2005). We argue these results conflict, because the relationship between list size and services per patient is conditioned by the existence of professional norms. Professional norms can be defined as prescriptions for the acceptable actions under given conditions (e.g. specific patient symptoms) applying to and sanctioned within a given occupation (Andersen, 2005: 25).Grytten, Skau, Sørensen and Aasland argue that the professional and medical norms might control the behavior of GPs, and that GP therefore do no allow their own “greed” to influence the production (2003: 66). They claim 3 that “to reduce the desirable or actual treatment is not in accordance with medical ethics and professional norms” (ibid: 52, our translation). This implies that profession norms govern the provision of all the medical services of GPs. On the contrary, we argue that professional norms govern the use of some medical services, whereas no firm professional norms apply for other services. The medical occupation is usually seen as a strong profession (Freidson 1970; Dent 2003: 175-178; Saks 1995) with strong professional norms, but services without professional norms can be identified also for occupations such as doctors and dentists (Serritzlew & Andersen, 2006; Andersen & Blegvad, 2003). Further, two notions of the medically correct level of service provision exist. The optimal level for the patient is when marginal medical benefit is higher than the price paid by the patient (often zero), and the optimal level for society is when marginal medical benefit is higher than marginal costs. The level prescribed by the professional norms (if any) is probably between these two standards, and for some services it seems to be professional acceptable to provide more services (and more time consuming care) to patients, when doctors have the capacity to do so (Richardson & Peacock 2006, 13). Although sometimes pictured as an exact (and omniscient) science, medical decision-making is often both complex and uncertain (Richardson & Peacock, 2006: 8-9). Even if the professional norms of the medical occupation counteract some of the potential drawbacks of the GP market power, strategic action is probably possible (for some medical services) without going against the norms. Richardson and Peacock claim that for many types of services no professionally defined level exists (ibid). This implies that the GPs have large discretion in the provision of these services, and several studies have shown that the GPs vary much with regard to the level of service provision (ibid; Vedsted et al. 2005). The general picture is that health professionals primarily affect the number of services strategically when no professional norms apply (Goodrick & Salancik 1996; Stano 1985; Dranovo & Wehner 1994; Grytten & Sørensen 2001; Carlsen, Grytten & Skau 2003). This indicates that the general supplier-induction hypothesis must take the service type into account. The theoretical proposition of this paper thus is: If a specific medical service is firmly regulated by professional norms, the number of services per patient does not depend on the list size. Otherwise, the number of services per patient is higher, the fewer patients the GPs have on their lists. Despite the increasing attention to the importance of professional norms within the SID literature, few have tested this proposition. Iversen and Lurås show (with data from Norway) that when ”professional opinions differ”, doctors with few patients make ”longer and more frequent consultations and more laboratory tests per listed person” (2000: 447), and Davis et al. (2000: 407) 4 argues that clinical factors rather than economic incentives explain the variation in clinical practice in New Zealand. We do not, however, know of any studies, which systematically examine the professional norms beforehand and test whether the level of supplier induction is higher for services without professional norms compared to services with professional norms. This paper provides such a test. After a short description of the Danish system, we analyze the norms. Primary physician services in Denmark In Denmark, the regions (before 2007: the counties) have responsibility for planning, organizing and running primary health services, including GP services. GPs are all self-employed specialists in general medicine with a contract with the National Health Insurance (Sygesikringens Forhandlingsudvalg & Praktiserende Lægers Organisation 2006). The pay system includes fee-peritem, fee-per-patient and a fixed amount per doctor. The fees are fixed in the agreement between the GPs’ organizations and the National Health Insurance. Fee-per-item comprises about 75 % of the GPs’ gross income. Only very few services involve user payment (e.g. medical certificates). The agreement between the GPs’ organization and the National Health Insurance specifies the availability conditions. A patient with an acute need must be given an appointment the same day, while non-acute treatments must be within 5 days (Sygesikringens Forhandlingsudvalg & Praktiserende Lægers Organisation 2006, § 39.1.c+d). Further, central planning has ensured that the nearest GP is very seldom far away. This makes availability as an alternative explanation of the association between the number of patients and the number of services per patient less relevant. As the bulk of services are totally fee of charge, and the fees are fixed, another alternative explanation can be eliminated (that the association is due to fee reductions when patients are sparse). The last alternative explanation cannot, however, be eliminated that easy. We cannot conclude that list sizes are not endogenous in Denmark. If the number of patients per GP exceeds 1600, they can choose to close their list for more patients. In case of more than 2411 patients the list is closed automatically. In the investigated county, 133 practices are open, 130 are closed voluntarily, while only one practice exceed the limit for automatic closure. Due to the voluntary closure of half of the GP practices, the last alternative explanation of the association between list size and services per patient (that patient morbidity affects both services per patient which then influence the GPs decisions regarding list size) thus remains a methodological problem and is discussed in the section on data and methods. 5 The mentioned agreement between the GPs’ organization and the National Health Insurance differentiates between base services and supplementary services. Base services are the different types of consultations while the supplementary services consist of add-ons to the consultations (e.g. puncture of the ear drum) and laboratory tests (e.g. urine test). The most used service is the ordinary consultation. Telephone and email consultations are faster alternatives (with lower fees) to the ordinary consultation, while visits, cognitive therapy and preventive consultations1 are more timeconsuming alternatives with higher fees. The Danish remuneration system thus opens for two types of supplier induction/rationing. First, the GP can provide more or less supplementary services, and second, they can choose strategically between three types of base services: The short ones with low few (telephone and email consultations), the normal one (ordinary consultation) and the more time consuming ones with higher fees (primarily visits and cognitive therapy). This raises another question: What services are lucrative (in terms of the trade off between fee and work load/time used)? This will be dealt with in the next section together with the professional norms. Different services: Professional norms and lucrativeness To put forward precise hypotheses on the relationship between list size and services per patient, we need two types of information about the services: The professional norms governing them (if any) and whether they are perceived to be lucrative compared to the alternative. We therefore conducted six qualitative interviews with general practitioners in the county of Aarhus. We stratified on gender and practice size (single or partnership) and sampled randomly within these strata. Further, we gathered clinical guidelines from the home page of the Danish Medical Association. The interviews and documents were coded in NVivo 7 for find evidence on the existence of professional norms (see http://www.ps.au.dk/lotte/praksis_kvali.doc for the interview guide and the categories from the coding). The analysis showed unambiguously that the GPs’ perceive visits to be unprofitable and cognitive therapy to be lucrative compared to the ordinary consultation. Both (are supposed to) take more time than the ordinary consultation. The findings on the supplementary services (add-ons and lab tests) did not indicate any substantial differences in lucrativeness: The GPs generally considered the fee for these services to be in line with the ordinary consultation, considering the used time. We 1 Before April 2006, a preventive consultation gave the same fee as the normal consultation, excluding prevention of ischemic heart disease which was better paid. In April 2006 a pre-arranged prevention consultation was introduced, but it was not fully implemented at the time of the data collection. 6 have chosen to focus on the services listed in table 1. They are all rather frequently used. Among these services, especially cognitive therapy is lucrative, especially for GPs with short lists (it is rather time consuming. Table 1: Professional norms and lucrativeness of the analyzed medical services Medical service Professional norm Fee in relation to time used Ordinary consultation Strong Medium Cognitive therapy Very weak Time-consuming, high fee Visit Strong concerning elderly Time-consuming, not high fee Urine test Medium Quick, but low fee INR test Medium Medium Index of ordinary add-ons Medium Medium Index of ordinary lab tests Medium Medium Note: only add-ons and lab tests, which the GP used on average one time or more in a two month period, are part of the indexes. The interviews clearly show that professional norms require that the GP must see a patient if he/she contacts the GP (unless the demand for a consultation is obviously unfounded). Thus, the norms governing the ordinary consultation are rather strong. The more specialized base services are, however, less firmly governed. Cognitive therapy was recently introduced, and we do not find any firm norms concerning the use of this service. The norms require visits to fragile elderly patients, but no norms regulate the use of visits to other groups of patients. The supplementary services all seem to be regulated by norms of medium strength. The main result is that cognitive therapy stands out as the service with the weakest norms. Being lucrative and without firm norms, it is the ‘most likely’ service in term of the SID hypothesis. The ordinary consultation is, on the other hand, the ‘most unlikely’ service for SID to happen. The other services lie between these extremes. Based on this analysis, the following hypotheses can be formulated: 1. The number of cognitive therapies per patient is higher for GP with relatively few listed patients. 2. The number of ordinary consultations per patient does not depend on the list size per GP. 7 3. The negative relationship between the number of cognitive therapies per patient and the number of listed patients per GP is significantly stronger than for ordinary consultations, urine tests, INR, visits, lab tests in general and add-ons in general. After a short presentation of the data and methods, these hypotheses will be tested quantitatively. Data and methods Predominantly, the paper is based on statistical analysis of register data, but (as discussed above) we included qualitative interviews with six general practitioners (GPs) and documentary material to investigate the professional norms. The primary data source was the Danish Health Insurance Register, which contains the number and type of services provided by GPs. Based on this register, we investigated services provided in April and May 2006 in the County of Aarhus. The units of analysis were 257 of the 264 practices in the county. Newly established and near-terminate practices were excluded, but we included both single practices and group practices. For the group practices, the number of patients was divided by the number of GPs working in the practice. We investigated only day-time services, because the patients attend different doctors at night. As suggested in the hypotheses, the dependent variable is the number of services per patient. We used three versions of this variable: The simple number of services per patient, the number of services per patient controlled for the socio-demographic characteristics of the patients (the county did the standardization) and a version where we calculated the relative deviation from the average use of the service. All three operationalizations gave roughly the same results. We present the lastmentioned and comments on the other measures if they gave other results. The primary explanatory factor was the list size (the number of listed patients per GP). The type of service is an important control variable (of course, the number of services provided per patient differs from service to service), and the interaction term between the type of service and the list size measures differences in supplier-induction between the services. As mentioned in the literature review, the time order of the variables might be problematic. With the available data, there was no simple way to determine whether list size affected service production or the other way around. The SID-theory expects list size to affect service production, but list size may also be determined by the service needs of the patients. That is, GPs with healthy patients might try to increase the number of patients on their list. Like we mentioned above, we made all the analyses using a version of the list size variable, which was standardized for the specific composition of patients. Further, the GPs can essentially only affect their list size by 8 opening and closing their lists. Thus, the status of the practices (open or closed) was coded and controlled for. We also controlled statistically for the following socio-demographic variables: The proportion of inhabitants from third world countries, the proportion of social housing, the population density, the unemployment, and the social index. These data come from the Ministry of the Interior’s key figures. Another solution to the endogeneity problem could have been to use instrument variables and a two-stage least square estimation, but no adequate instrument variable was available. As an additional control variable, we included the average age of the GP, because the qualitative interviews indicated that young and old GPs differ. This can be due to dept from the practice purchase or different priorities. The average age of the GPs is highly correlated with the seniority of the GPs. The results gave similar results when we included the seniority instead of the age. Grytten and Sørensen (2007: 724) suggest that the relationship should be non-linear and that the total number of consultations should be constant for varying list sizes (namely around the optimal total number of consultations for the GP). We find it more plausible that the GPs all the time adjust at the margin: That the elasticity is lower than one for the whole spectrum (a reduction of 1 % in the number of patients leading, for example, only to a reduction in the total number of services of 0.95 %). As nothing in our data indicates that the relationship is non-linear, we use linear models in the analyses of the number of services per patient and list size. We use three types of OLS regressions. The first type includes only one service, and the practices are the units of analyses. But to compare the use of different services systematically, the second type of regression analyzes the practices’ use of different services. In these analyses, the number of service1 per patient for GP1 is one observation, the number of service2 per patient is a second observation, the number of service2 per patient for GP2 is a third observation etc. The third type of analysis is a fixed effect model with dummies for the practices, in which the units of analysis are the same as in the second type of analysis. Results The structure of this section follows the three hypotheses. Firstly, we analyze the relationship between the number of cognitive therapies per patient and the number of listed patients per GP. Secondly, the same test is performed for ordinary consultations. Finally, we investigate whether the 9 number of services per patient depends more on the list size for cognitive therapy than for other services. Hypothesis 1 expects that number of cognitive therapy services per patient is higher, the fewer listed patients the GPs have. This is tested in model 1, 2 and 3 in table 2. All these analyses indicate that the list size affects the number of cognitive therapies per patient negatively. The inclusion of controls in model 2 and 3 reduce the size of the effect a little, but it is still highly significant. This confirms hypothesis 1. (Table 2 around here) According to hypothesis 2, the list size should not affect the number of ordinary consultations per patient. This is tested in model 4 in table 2. As expected, the effect is not significant, and this does not change if more controls are introduced (not shown). This supports hypothesis 2. Hypothesis 3 is about the interaction between the type of service and the list size. It expects the relationship between the number of services per patient and the list size to be more strongly negative for cognitive therapy compared to urine tests, INR, ordinary consultations, visits, lab tests and add-ons. It does, in other word, expect that the interaction term between these services and the list size is positive in the regressions with cognitive therapy as the reference service (a high negative association means a lot of SID). In accordance with the tests of hypothesis 1 and 2, the results show that the list size does affects cognitive therapy services more than normal consultations. The positive interaction term in model 6 in table 2 between the dummy for ordinary consultation and the list size shows this. The comparisons between cognitive therapy and the other services consistently show a positive interaction term between the list size and the dummies for the services other than cognitive therapy (urine tests and INR in model 5 and all the analyzed groups of services in model 7, 8 and 9). Except for visits, the regression coefficients of the interaction variables are significantly higher than zero. The professional norms for visits do, as mentioned, demand this service for fragile elderly patients, but otherwise the norms concerning the service is weak. This might explain why the association between list size and services per patient is not significantly weaker for visits compared to cognitive therapies. The results are generally in accordance with the expectations, except for one thing: When the practice dummies are introduced in model 9, the effect of the list size on the number of cognitive therapy services per patient is no longer significant and changes the sign (cognitive 10 therapy is the reference service in model 9, so the regression coefficient for list size is the effect to this specific service). There is much multicollinearity in this model (the tolerance for the list size is, for example, 0.0184), and this analysis might not be meaningful with the available data. We would be grateful for comments on this issue. Discussion and conclusion The results generally confirm the hypotheses (expect for the fixed effect model which was discussed above). The use of cognitive therapy, which is a well-paid, time-consuming alternative to the ordinary consultation unregulated by professional norms, depends on the number of listed patients per GP, whereas the number of ordinary consultations per patient does not depend on the list size. Further, the use of cognitive therapy depends significantly more on the number of listed patients than it is the case for ordinary consultations, urine tests, INR, ordinary lab tests and ordinary add-ons. The visit service, which is another time-consuming alternative to the ordinary consultation, is only regulated by professional norm for elderly patients, and its use depends less on the list size than the use of cognitive therapy do, but the interaction is not statistically significant. This fits with the results from the qualitative interviews, which show that the norms governing visits are the second-weakest (next to cognitive therapy). The result thus indicates that the degree of supplier induction decreases with increasing firmness of the norms governing the specific service. The analysis generally supports the theoretical proposition. If a specific medical service is firmly regulated by professional norms, the number of services per patient does not depend on the list size, and if no firm norm applies the number of services per patient decreases with the number of patients on the GPs list. The answer to the question posed in the title is thus: General practitioners induce demand for services which are not regulated by professional norms. The paper proposes a possible synthesis conflicting results in the SID literature: The degree of supplier induced demand depends on firmness of the professional norms governing the specific services. As the results come from only one county in one country, more evidence is needed, before the results can be generalized. We therefore call on studies where professional norms are included in the analyses and not only used as ex-post explanations. This paper indicates that economic theory can fruitfully be combined with the sociology of professions, as both economic incentives and professional norms seem affect the behavior of GPs. 11 References Andersen, Lotte Bøgh (2005) Offentligt ansattes strategier Århus: Politica. Andersen, Lotte Bøgh & Marianne Blegvad (2003) "Normer eller Egennytte? Professionelle og økonomiske incitamenter i dansk børnetandpleje", Politica, 35: 2, pp. 125-135. Anderson, Richard K., Donald House & Michael B. Ormiston (1981). "A Theory of Physician Behavior with Supplier-Induced Demand", Southern Economic Journal, 48, 1, pp. 124-133. Bech, M., Hansen, B.B., Lauridsen, T. & Lauridsen, J. (2007) Can supplier induced demand and availability effects be identified for Danish general practitioners? Presented 28th NHESG Meeting, Tartu, Estonia, August 2007. Carlsen F., J. Grytten, I. Skau (2003) “Financial incentives and the supply of laboratory tests”. The European Journal of Health Economics, 4:279–285 Carlsen, F. & J. Grytten (1998). “More physicians: Improved availability or induced demand?” Health Economics 7, 495-508. Davis, P., B. Gribben, A. Scott, & R. Lay-Yee (2000). “The supply hypothesis and medical practice variation in primary care: testing economic and clinical models of inter-practitioner variation”. Social Science and Medicine 50: 407-418. De Jaegher, Kris & Marc Jegers (2000). "A model of physician behaviour with demand inducement", Journal of Health Economics, 19, 2, pp. 231-258. Delattre, Eric & Brigitte Dormont (2003). “Fixed fees and physician-induced demand: A panel data study on French Physicians” Health Economics. 12: 741-754. Dent, Mike (2003) Remodelling hospitals and health professions in Europe. Medicine, nursing and the state, Basingstoke: Palgrave Macmillan Donaldson, Cam & Karen Gerard (1993). Economics of Health Care Financing: The Visible Hand, London: Macmillan. Dranove, D. & P. Wehner (1994) “Physician-induced demand for childbirths”. Journal of Health Economics; 13: 61-73. Evans, Robert G. (1974) “Supplier-induced demand: Some empirical evidence and implications” pp. 162-173 i Mark Perlman (red.) The economics of health and medical care. London & 12 Basingstoke: Macmillan. Freidson, Eliot (1970). Professional Dominance: The Social Structure of Medical Care, New York: Altherton Press. Goodrick, Elizabeth & Gerald R. Salancik (1996). "Organizational Discretion in Responding to Institutional Practices: Hospitals and Cerarean Births", Administrative Science Quarterly, 41, pp. 1-28. Gosden, T., L. Pedersen & D. Torgerson (1999). "How should we pay doctors? A systematic review of salary payments and their effects on doctor behavior", QJM, 92, pp. 47-55. Gosden, Toby, Frode Forlan, Ivar Kristiansen, Matthew Sutton, Brenda Leese, Antonio Giuffrida, Michelle Sergison & Lone Pedersen (2001). "Impact of Payment Method on Behaviour of Primary Care Physicians: A Systematic Review", Journal of Health Services Research and Policy , 6, pp. 44-55. Groenewegen PP, J. van der Zee & R. van Haaften (1991) Remunerating general practitioners in Western Europe. Athenaeum: Aldershot Grytten J & R. Sørensen (2001) “Type of contract and supplier-induced demand for primary physicians in Norway”. Journal of Health Economics 20: 379-93. Grytten, J., I. Skau, R. Sørensen & O. G. Aasland (2003). Fastlegereformen - en analyse av fastlegenes arbeidsbelastning og tjenestetilbud, Forskningsrapport nr. 11. Handelshøyskolen BI. Grytten, Jostein & Rune Sørensen (2001). “Type of contract and supplier-induced demand for primary physicians in Norway”. Journal of Health Economics. 20: 379-393. Grytten Jostein & Rune Sørensen (2007) “Primary physician services-list size and primary physicians' service production”. Journal of Health Economics 26(4):721-41. Hoffmeyer, U.K. & T.R. McCarthy (1994) Financing health care. Kluwer: London Iversen, Tor & Hilde Lurås (2000). ”Economic motives and professional norms: the case of general medical practice” Journal of Economic Behavior & Organization. 43: 447-470. Krasnik, Allan, Peter P. Groenewegen, Poul A. Pedersen, Peter v. Scholten, Gavin Mooney, Adam Gottschau, Henk A. Flierman & Mogens T. Damsgaard (1990). ”Changing remuneration systems: effects on activity in general practice“, British Medical Journal, 300: 1698-1701. Madden, D., A. Nolan & B. Nolan (2005). ”GP reimbursement and visiting behaviour in Ireland”. 13 Health Economics 14: 1047-1060. Nassiri, Abdelhak & Lise Rochaix (2005). “Revisiting physicians’ financial incentives in Quebec: a panel system approach” Health Economics 15: 49-64. Newhouse, Joseph P. (1970) “A Model of Physician Pricing”, Southern Economic Journal, Vol. 37, No. 2., pp. 174-183. Reinhardt, Uwe (1985). “The Theory of Physician-Induced Demand. Reflections after a Decade” Journal of Health Economics. 4: 187-193. Rice, Thomas & R.J. Labelle (1989). “Do physicians induce demand for medical services?” Journal of Health Politics, Policy and Law 14: 587-600. Richardson, Jeff & Stuart Peacock (2006). Reconsidering theories and evidence of supplier induced demand. Research paper 2006 (13). Centre for Health Economics, Monash University, Australia. http://www.buseco.monash.edu.au/centres/che/pubs/rp13.pdf. Saks, Mike (1995). Professions and the public interest. Medical power, altruism and alternative medicine, London & New York: Routledge. Serritzlew, Søren & Lotte Bøgh Andersen (2006), "Økonomiske incitamenter i praksissektoren", Politica, vol. 38 nr. 4, s. 392-409. Sørensen, Rune & Jostein Grytten (2003). ”Service production and contract choice in primary physician services”. Health Policy. 66: 73-93. Stano, M. (1985) “An analysis of the evidence on competition in the physician services markets”. Journal of Health Economics 4: 197-211. Stano, Miron (1987). "A Clarification of Theories and Evidence on Supplier-Induced Demand for Physicians' Services", The Journal of Human Resources, 22, 4, pp. 611-620. Sygesikringens Forhandlingsudvalg & Praktiserende Lægers Organisation (2006). Landsoverenskomst for almen lægepraksis (jf. www.laeger.dk/). Vedsted, Peter, Frede Olesen, Hanne Hollnagel, Flemming Bro & Finn Kamper-Jørgensen (2005). Almen Lægepraksis i Danmark. Om funktion, love, administration og udvikling, Århus: Tidsskrift for Praktisk Lægegerning & Forskningsenheden i Århus. 14 Table 1: Regression of services per 100 patients. April and May 2006. Unstandardized regression coefficients. Relative difference from mean. Model 1 : Only cognitive therapy) Intercept List size Open practice Average age of the GPs 0.0210 -0.0198 ** Model 2: Cognitive therapy with practice controls 0.118 -0.0167 ** -0.00268 -0.00189 *** Model 3: Cognitive therapy with all controls 0.143 -0.0152** -0.00465 -0.00187*** Model 4: Ordinary consultation with practice controls 0.0526 0.0050 -0.00215 -0.00129 *** Urine test Urine test * List size INR INR * List size Ordinary consultation Ordinary consultation * List size Visits Visits * List size Ordinary laboratory services Ordinary lab. services * List size Ordinary add-on services Ord. add-ons * List size Population density Proportion social housing Proportion unemployed Proportion third world immigrants Social index R2 N Model 5: Model 6: Cognitive therapy Cognitive therapy and compared to urine ordinary test and INR consultations 0.108 0.125 -0.0181** -0.0157*** -0.00146 -0.00392 -0.00137*** -0.00157*** Model 7: Cognitive therapy and all groups of services 0.0827 -0.0177*** -0.0000296 -0.00122*** Model 8: Model 7 with controls 0.0985 -0.0175*** -0.00241 -0.00127*** Model 9: Model 8 with dummies for all practices 0.0936 0.0210 0.0392*** -0.00196*** -0.0347*** 0.0228*** -0.0158 0.0105 -0.0424*** 0.0271*** -0.0217* 0.0152* -0.0347*** 0.0228*** 0.0158 0.0105 -0.0424*** 0.0271*** -0.0217* 0.0152* -0.0347*** 0.0228*** -0.0158 0.0105 -0.0424*** 0.0271*** -0.0217** 0.0152** 0.065 1285 0.00000151 0.0000870** 0.00622*** 0.00000296 0.0700*** 0.093 1285 0.0000213*** 0.0000994 -0.0104*** 0.0000349*** Dropped 0.57 1285 -0.0390** 0.0254** -0.0580*** 0.0360** -0.0347*** 0.0228*** 0.025 257 0.065 257 0.00000420 0.00109 0.00440 -0.000000649 -0.0784 0.092 0.12 257 257 0.000000268 0.0000543 0.00702** 0.00000452 -0.0705** 0.083 771 0.00000173 0.00114 0.00403 0.00000393 -0.0781** 0.13 514 Note: * p<0.10 ** p<0.05 *** p<0.01. All numbers are measured in 2006, except the unemployment (2003). Tolerance for list size in model 8 is 0.0184. Results for un-standardized data and data controlled for the gender and age of the patients are approximately the same, expect that the list size do not significantly affect the number of services in the last models. 15