Elsevier

Journal of Health Economics

Volume 22, Issue 6, November 2003, Pages 881-913
Journal of Health Economics

The demand for physician services: Evidence from a natural experiment

https://doi.org/10.1016/S0167-6296(03)00047-XGet rights and content

Abstract

This study exploits a natural experiment in Belgium to estimate the effect of copayment increases on the demand for physician services. It shows how a differences-in-differences (DD) estimator of the price effects can be decomposed into effects induced by the common average proportional price increase (income effects) and by the change in relative prices (substitution effects). The price elasticity of a uniform proportional price increase is relatively small (−0.13 for men and −0.03 for women). Substitution effects are large, especially for women, but imprecisely estimated. Despite the substantial price increases, the efficiency gain of the reform, if any, is modest.

Introduction

On 1 January 1994, the health authorities in Belgium increased the copayment rates of three types of physician services in Belgium: office visits to general practitioners (GPs), GP home visits and specialist visits. In this study, we analyse the impact of this measure on the demand for these outpatient health care services. We decompose this impact into the effect of a common average proportional price increase (an income effect) and into the effects induced by changing relative prices (substitution effects). On the basis of these measured impact effects, we calculate the (gross) social benefit of this policy and its determinants.

In the literature, numerous studies have estimated the price elasticity of the demand for health services.1 However, unsatisfactory treatment of methodological problems results often in unreliable estimates (see Newhouse et al. (1980) for a review of these problems). In order to cope with these difficulties, the federal authorities of US initiated a social experiment in the 1970s: the Rand experiment Manning et al., 1987, Newhouse, 1993. The results of the experiment confirm that an increase in the individual’s cost sharing (from 0 to 95% for instance) implies a reduction in the average health care expenditures (of 46% in the example), in the probability of any medical use (27.5%) and in the unconditional probability of inpatient use (33.8%). The corresponding price elasticity of the demand for health care services is not very large, ranging from −0.2 to −0.1, but significantly different from zero. The price elasticity of the demand for ambulatory care is higher, ranging from −0.17 for copayments between 0 and 25% to −0.31 for higher copayment rates.

More recently, some studies have attempted to estimate the price sensitivity of the demand for medical services in Europe. Nolan (1993) studies the Irish health care system. Using data from a national household survey carried out throughout Ireland in 1987, the author estimates that the individuals benefiting from free access to primary health care services are more likely to consume outpatient and inpatient services and, conditional on positive consumption, report a significantly higher number of medical visits. This study faces, however, an endogeneity problem common to cross-section studies. Rather than being a consequence of a change in behaviour induced by a higher coverage (a lower copayment rate), the increased consumption could result from a positive correlation between coverage and a higher (unobserved) propensity to consume, such as predicted in an insurance market with adverse selection Chiappori et al., 1998, Newhouse et al., 1980.

Chiappori et al. (1998) exploit a natural experiment2 to estimate the effect of cost sharing on the demand for physician services in France. In France, employees can buy additional insurance on the private market in order to bring the cost sharing rate down to 0%. However, on 1 July 1993, a point in time when the government’s health insurance became less generous, some, but not all, private insurance companies decided to increase the copayment rate from 0 to 10%. The study estimates the effect of cost sharing on the demand by comparing a group of employees for whom the copayment increased to a group for whom it did not.3 On the basis of a panel probit model, contrasting the control and treatment groups, the study finds that the participation rate in GP home visits is significantly affected by the copayment level. No significant effect, however, is found for GP office visits.

Our study relies on a similar natural experiment. On 1 January 1994, the copayment rates for physician services of the mandatory public health insurance scheme in Belgium increased substantially for all but one category of individuals: in real terms, the rate increased by 48% for GP office visits, by 35% for GP home visits and by 60% for specialist visits. As in the French study, we can therefore contrast the impact of the change on the treatment group to a control group.4 Since, for reasons of confidentiality, we have only access to grouped data, a simple linear differences-in-differences (DD) estimator implements such a contrast.

In an independent work, Van de Voorde et al. (2001) apply such DD estimators on similar grouped data for Belgium.5 We argue, however, that the price elasticities deduced from this study are incorrect, since they implicitly ignore the substitution effects induced by the relative price variations of these physician services.6 In this paper, we explicitly allow for substitution effects within the group of three physician services considered. To this purpose, we estimate a system of demand equations as derived from the classic theory of consumer demand and the principles of two-stage budgeting Barten and Böhm, 1982, Deaton and Muellbauer, 1980a.

A second objective of this paper is to evaluate the (gross) efficiency gain of this policy reform. The efficient copayment rate in health insurance trades-off the efficiency gains from risk sharing and the efficiency costs induced by moral hazard (see Arrow, 1963, Pauly, 1968, Zeckhauser, 1970). In the US, substantial research efforts have been undertaken to determine this efficient rate. Until recently, this research concluded that the cost of moral hazard is large relative to benefits and that higher copayment rates were warranted Feldman and Dowd, 1991, Feldstein, 1973, Feldstein and Friedman, 1977, Manning and Marquis, 1996. However, the analysis in this research was based on a partial equilibrium framework. Such a framework ignores that (uncompensated) price effects decompose in income and substitution effects and that only the latter affect the efficiency of an insurance scheme. In a recent paper, Nyman (1999) argues that income effects are important and their neglect in previous studies have led to prescribe cost sharing rates in the US at too high levels.7 In addition, in a theoretical contribution, Besley (1988)8 demonstrates that not only the own compensated price effect, but also cross-price effects matter in the design of the optimal cost sharing rule. These are generally not allowed for in the empirical literature.9 Nevertheless, one type of interaction effect among different health services has retained interest. It has been argued that by reducing coverage of outpatient services, preventive care might be deterred thereby inducing more expenditure on inpatient (curative) services. However, Manning et al.’s (1987, p. 271) report that the findings of the Rand experiment suggest, if anything, that outpatient and inpatient services are complements, not substitutes.

In this study, we do not estimate the net, but the gross efficiency gain of the price reform. For, data limitations do not allow evaluating the cost of increased risk induced by the lower insurance coverage. Moreover, for the same reason, even if we decompose the price effect into income and substitution effects within the group of the three higher mentioned physician services, we can account neither for the income effect of the increased copayments on total expenditures nor for substitution effects between the three physician services and other (health) goods and services. We can therefore only calculate an upper bound for the gross efficiency gain.

The paper is organised as follows. The following section describes the Belgian health care system. In Section 3, we describe the data. In Section 4, we present the standard DD estimator and propose an alternative one that can be decomposed in income and substitution effects. In Section 5, we formulate the Rotterdam demand system Barten, 1966, Theil, 1965 assuming two-stage budgeting. We show how the alternative DD can be decomposed. On the basis of the estimation results, we calculate the gross proportional efficiency gain of the increase in the copayment rates and decompose this gain into its determinants. The final section concludes.

Section snippets

The Belgian health care system

In Belgium, almost all individuals are covered by a (semi-)public insurance system. All workers with a professional activity must contribute to the scheme. Premiums are set proportionally to earnings. Competition is introduced in that individuals are offered a choice between a number of non-profit sickness funds

The data

The analysis relies on administrative data originating from a Belgian sickness fund (the ‘Mutualités Chrétiennes’). They contain only data on individuals entitled to health insurance within the ‘general scheme’. For reasons of confidentiality, the sickness fund did not authorise access to individual data. We thus acquired grouped data on the average

Differences-in-differences estimators

We could calculate the DD on the basis of the average evolutions of physician visits of the treatment and the control group as a whole. However, this does not exploit all available information and is therefore not efficient. We can calculate, for each type of physician visit, many such DD estimators, since we can distinguish sub-groups within these control and treatment groups. By crossing the indicator variables reported in Table 2, one can deduce that M0=10 of such sub-groups can be formed

The demand system

We now propose a method to decompose the price effects in income and substitution effects. To that purpose, we rely on the classic theory of consumer demand and assume that consumers behave according to the principles of two-stage budgeting. In the first stage, the consumer decides upon the budget to allocate to expenditures on physician services. In the second stage, the consumer decides which type of physician service he will buy taking the budget allocated in the first stage as given.

Since

Conclusion

This study analysed the effect of a substantial increase of the copayments of three types of physician services on 1 January 1994 in Belgium: GP office and home visits and office visits to the specialist. We proposed a DD estimator of the effect of the price increase on the demand for these services and showed how it could be decomposed into one induced by the uniform proportional increase of copayments for all three services (i.e. an income effect) and into a number of substitution effects

Acknowledgements

The authors are grateful to M. Marchand, D. Weiserbs, I. Bardoulat, J. Nyman and an anonymous referee for their comments, though they fear they have not always agreed with what they have had to say. Bart Cockx especially thanks A. Barten for teaching him econometrics and demand analysis. We acknowledge the ‘Mutualités chrétiennes’ for supplying the data and the Fonds National de la Recherche Scientifique (F.N.R.S.) for the financial support. This research was also part of a programme supported

References (50)

  • J.A. Nyman

    The economics of moral hazard revisited

    Journal of Health Economics

    (1999)
  • J.A. Nyman

    Response. The income transfer effect, the access value of insurance and the Rand health insurance experiment

    Journal of Health Economics

    (2001)
  • R. Parks

    On the estimation of multinomial logit models from relative frequency data

    Journal of Econometrics

    (1980)
  • R. Zeckhauser

    Medical Insurance: a case study of the tradeoff between risk spreading and appropriate incentives

    Journal of Economic Theory

    (1970)
  • Adriaenssen, I., De Graeve, D., 2000. Economic determinants of medical care in Belgium. Mimeo, Department of Economics,...
  • Alliance Nationale des Mutualités Chrétiennes, 1995....
  • T. Amemiya et al.

    A modified logit model

    Review of Economics and Statistics

    (1975)
  • T. Amemiya

    Qualitative response models: a survey

    Journal of Economic Literature

    (1981)
  • Amemiya, T., 1985. Advanced Econometrics. Basil Blackwell,...
  • K.J. Arrow

    Uncertainty and the welfare economics of medical care

    American Economic Review

    (1963)
  • Barten, A.P., 1966. Theorie en empirie van een volledig stelsel van vraagvergelijkingen (Theory and empirics of a...
  • A.P. Barten

    Evidence on the Slutsky conditions for demand equations

    Review of Economics and Statistics

    (1967)
  • Barten, A.P., Böhm, V., 1982. Consumer theory. In: Arrow, K.J., Intriligator, M.D. (Eds.), Handbook of Mathematical...
  • J. Berkson

    Application of the logistic function to bio-assay

    Journal of the American Statistical Association

    (1944)
  • Besley, T.J., Case, A., 2000. Unnatural experiments? Estimating the incidence of endogenous policies. The Economic...
  • Cited by (46)

    • The fintech gender gap

      2023, Journal of Financial Intermediation
    • The effect of copayments on the utilization of the GP service in Norway

      2018, Social Science and Medicine
      Citation Excerpt :

      Several studies from a variety of geographic contexts have exploited situations where consumers experience a sudden change in the incentive structure (i.e. lowering or increasing copayments) to estimate the price elasticity of demand. The results from these studies overwhelmingly agree that copayments cause demand for medical care to decline (Cockx and Brasseur, 2003; Contoyannis, 2005; Kim et al., 2005; Li et al., 2007; Simonsen et al., 2015; Van de Voorde et al., 2001). A common problem of studies based on observational data, however, is that they typically cannot establish causality.

    • Nurse practitioner independence, health care utilization, and health outcomes

      2018, Journal of Health Economics
      Citation Excerpt :

      Hooker and McCaig (2001) show that there is little difference in the number of diagnostic tests ordered by NPs and physicians. Studies that have focused on primary care providers, such as Grytten et al. (1995), Van De Voorde et al. (2001), Cockx and Brasseur (2003), and Madden et al. (2005), do not find evidence of supplier-induced demand in the number of visits to primary care providers. The available empirical evidence suggests that there is at most a minimal effect of supplier-induced demand in primary care services.

    • Copayments and physicians visits: A panel data study of Swedish regions 2003–2012

      2016, Health Policy
      Citation Excerpt :

      From a European context there are only a few studies with a quasi-experimental approach. For example, a study using a difference-in-difference approach evaluating an increase in patient charges in Belgium during the 1990s estimated price elasticities of demand between −0.03 and −0.13 [11]. And, two studies with a similar empirical approach in a French setting found no price elasticity at all (general population) for office General Practitioner (GP) visits and a very small effect for GP home visits among a general population [12] and no effect at all in a sub-set of the poorest households [13].

    • Introducing out-of-pocket payment for General Practice in Denmark: Feasibility and support

      2014, Health Policy
      Citation Excerpt :

      Will there be any healthcare consequences of OOP payment? Will OOP payment reduce the use by some groups of the population more than by others? [5–21]. The purpose of this article is not to make conclusions about the effects of OOP payment, but to investigate and discuss the actors, arguments and the decision-making context in the health care policy debate in order to provide insight into why OOP payment for GP is introduced or not.

    View all citing articles on Scopus
    View full text