The demand for private health care

Table 1 Proportion of sample using private health services in British Household Panel 1990–1995 Proportion 1990r 1991r 1992r 1993r 1994r 1991 1992 1993 1994 1995 Private dental care 9.0 8.5 1.1 11.6 12.7 Private eye care 8.6 8.4 10.5 10.0 10.7 Private physiotherapy, 4.0 4.2 4.9 5.2 5.0 chiropody or health visitor Private inpatient stay 1.0 0.9 0.8 0.9 0.9 Any private health 18.4 17.5 20.9 21.3 22.4 service use Ž . Base weighted 9911 9458 9021 9054 8816 availability of such care is being reduced by suppliers without any explicit change in government policy. Table 1 shows the patterns in private demand for the different services for each of the first 5 years of the survey. Approximately 16 of individuals in the BHPS sample use either private dental, eye, hospital or outpatient services. The table indicates a general rise in private service usage across the 5 years of the survey, with only private inpatient use showing no growth.

3. The demand for private health care

3.1. The economic model Ž . Goddard and Smith 1998 outline a simple model of demand for medical care where public and private care exists. This can be used to explore the impact of income, price and quality of the public alternative, attitudes to the role of state in the provision of health care, and past use on the current demand for private care. Ž . The model follows Goddard et al. 1994 in recognising that an individual has three discrete choices: to seek no medical care, to use private care, and to use public care. These choices will be affected by the severity of illness, the costs and the quality of NHS care, and the costs and quality of private care. For any individual, indexed by i, let V be the benefit of private health care and i Ž . p be the cost relative to income . The existence of the public sector constrains i the suppliers of private care to providing a service which is of at least as high a quality as the public alternative: no one will buy the service if it is of lower quality than the public sector. Let this quality difference be represented by a single parameter t . At its simplest, this can be waiting time but can be thought of more Ž generally as encompassing other quality dimensions such as the provision of . information, or the ability to choose the exact date and location of treatment . Individuals vary in their valuation of this quality by the parameter g . If t is i thought of as waiting time, g can be thought of as the rate at which the value of i Ž treatment decays because it is received later rather than sooner Lindsay and . Ž . Feigenbaum, 1984 . The value of NHS treatment for individual i is V exp yg t . i i NHS care has no direct user price but the individual has to access NHS facilities. Let this cost be c . i Faced with these costs and benefits, the potential user is indifferent between private and NHS care when V y V exp yg t s p y c 1 Ž . Ž . i i i i i is indifferent between private care and no care when V s p 2 Ž . i i and is indifferent between NHS care and no care when V exp yg t s c . 3 Ž . Ž . i i i Ž . Ž . Eqs. 1 – 3 can be used to examine the impact of changing parameters on the choice of the three alternatives. Decreasing p will increase the use of the private i alternative and decrease the use of the NHS and no care alternatives. Decreasing c will increase the use of NHS treatment and reduce the use of private and no i care. An increase in g will increase the use of private care, reduce the use of i NHS care and increase the use of no care. The effect of an increase in t is the same. An increase in V will increase the use of private care and NHS care. i These parameters will vary across individuals. The value of the benefits of medical treatment V will obviously be related to the severity of illness. In i addition, as V is the perceived benefit to the user of care, it may also be a i function of the importance of good health to the individual. This is often argued to be positively associated with education. Where t is taken to be the waiting time for NHS treatment, g is the individual’s valuation of time. This can be expected i to be a positive function of income and of type of employment. For example, those who are self-employed are less likely to get paid whilst waiting for medical treatment. 3 Where t is the quality of the NHS, g is the valuation of NHS quality. i This might be expected to be a function of the individual’s general attitudes to the 3 Ž . For an empirical estimate of the value of time in a medical context see Propper 1995 . NHS. For example, those who strongly value state provision of health care may be more tolerant of poorer quality. Then those who hold such attitudes will use more NHS services for a given value of V and t . i The value of p will be lower for those individuals who have medical insurance i which covers the cost of private care and lower for those with higher income Ž . since p is defined relative to all other goods . Since c is the access cost to NHS i i facilities, it will be lower, leading to greater availability of NHS facilities. The nature of medical care means these prices will also be a function of past use of services. First, the price of care in each sector includes the costs of search. In health care, consumers cannot always tell good from poor quality. As a result, regulation limits advertising and consumers are reliant on their own knowledge and that of friends, relatives etc. As a consequence, search costs for a supplier of care may be high. Second, information asymmetries mean trust is an important component of the relationship between suppliers and demander of care. Third, an individual invests effort in communicating their medical history to a supplier of health care. So once an individual has found one supplier, they may be less likely to change to another. A consumer who has used the private sector in the past will have lower search costs for private sector use and will have made an investment with a private supplier. Similarly, a consumer who has recently used the NHS will have lower NHS care search costs and will have made an investment with an NHS supplier. So the costs of care in each sector are likely to be a function of past use, and current use is likely to depend upon past use. 3.2. The econometric model At any time t, the individual chooses between private care, public care and no care. She thus has J alternatives. Let j s 1 denote no care, j s 2 denote NHS Ž . Ž . care, and j s 3 private care. From the economic model of Eqs. 1 – 3 , the latent net valuations associated with each alternative will depend on the characteristics of Ž . the individual her valuation of health states, of time, of price etc. . The weight attached to each characteristic will vary by alternative. Allowing for random error, the latent net valuations of the jth choice can be written as: m s b X z q e 4 Ž . ji j i ji Ž where z is a vector of attributes of the individual which may be allowed to vary i . 4 by alternative as well and e is a random error term. ji 4 Ž . This model is of the form of the model of occupational choice of Schmidt and Strauss 1975 . If the individual is observed making choice j, it is assumed that m has the ji highest net valuation. The statistical model is driven by the probability that choice j is made, which is Pr m m for all k j. Ž . ji k i If the e are assumed to be i.i.d. with Weibull distribution, then the statistical ji model that results for the choice of alternative is the multinomial logit model: J X X b z b z j i k i Pr m s j s e r e 5 Ž . Ž . Ý i ks1 where m is the observed choice of individual i. To estimate the parameters of this i model it is necessary to normalise and impose the restriction that one of the b s 0. In estimation of this model, I impose the restriction that b s 0. j 1 From the discussion above, the specific characteristics that will influence the Ž . choice the individual makes i.e. the variables in z include income and health i Ž status, the costs of accessing each service which will be a function of the costs of . private care, past use, and the availability of services and the perceived quality of Ž the public service which will be a function of waiting lists and possibly attitudes . to the public sector role in financing health care . In addition, there may be regional effects. So the latent net valuation of alternative j will depend on: m s f X ,Q , m , R;e , j s 1,2,3 6 Ž . Ž . ji i r ji ty1 ji where X is a set of personal and household demographics, socio-economic i variables including income, and measures of the beliefs individuals hold about the appropriate role for the private sector in the provision of health care, Q a set of r indicators of the quality of NHS provision in the region in which the individual lives, R a vector of regional dummies, m is use of alternative j last period, ji ty1 and e is white noise error. ji The data set records use at five time points and of several health services, Ž . denoted s defined below . So the data permit examination of the impact of past use of one service on another and of past use of one alternative on another. To estimate these effects I pool the panel data and expand the set of variables in Eq. Ž . 6 to include these ‘cross-service’ and ‘cross-alternative’ effects and also to allow for time effects. The latent utility for alternative j of service s at time t is thus: m s s b X q b Q q b m q b m X q b R q b T q e , ji t i t 1 r t 2 ji ty1 3 j i ty1 4 5 ji t j s 1,2,3 7 Ž . where m and m X are now vectors of past service use, j X indexes the other ji ty1 j ity1 alternatives, T is a vector of time dummies, and e is white noise error. From Eq. ji t Ž . 5 the parameter estimates will differ across the two alternatives but for simplicity of notation we have dropped the subscripts on these. As the restriction that the Ž . parameter estimates are 0 for alternative 1 no care is imposed, parameters are estimated only for alternatives 2 and 3. In pooling the data and conditioning on information at time t, the assumption made is that the errors are not correlated over time for an individual. In addition, the MNL model imposes the assumption of independence of the errors across options. 5 3.3. The data From the BHPS data I define the use of three services or sets of services. The Ž first set of services is defined as use of any of the services given in Table 1 dental 6 . care, eye care, inpatient stays, certain community services . If the respondent used any of these services in year t and any of her use was private, she is recorded as having positive private use in year t. If she used any of these services and none of her use was private, she is recorded as having positive NHS use in year t. The second service is an inpatient hospital stay. If this stay is privately paid for, the respondent is recorded as having a private stay. If the stay is not privately paid for, then she is recorded as an NHS user. The third service is dental care, where if the respondent has dental care and any of that care is private, she is recorded as having private dental care. If she has some dental care and none of it is private, she is recorded as having NHS dental care. 7 In addition to measures of health service use and standard socio-economic and demographic data for each individual in the household, the BHPS includes measures of current self-assessed health status, and detailed measures of longer term health status. Individuals are asked whether they are limited in their daily Ž . activities including work and leisure ADLs . If they have any limitations they are asked whether these limit their ability to work. They are also asked what specific limitationsrconditions they have. The data also record whether the individual is a smoker, and if the answer is positive the number of cigarettes consumed per day. Earlier research on the demand for private health care in the UK has either used no Ž . measures of health status Besley et al., 1999 , or rather more limited measures Ž . Propper, 1993 . Here the BHPS data are used to control for a large number of measures of health status. In addition, conditioning on health status enables a 5 While this is an unattractive feature of the MNL model, the rich set of regressor variables should reduce correlation between the errors. 6 These community services are all services for which NHS provision exists. The BHPS records use of other private services for which there is no NHS alternative. As we are interested in the choice between no use, NHS use and private use, we do not examine these services. 7 This coding gives the broadest definition of private use, which fits our interest in any use of the private sector. reduction in the potential contamination of the income coefficient, which arises from the correlation between health and income. To these data are matched regional indicators of the quality of the NHS. These are the length of NHS waiting Ž lists over- and under-12 months deflated by the regional population relevant for . Ž inpatient care and an index of dental service availability relevant for dental . 8 care . The prices paid by individuals are not observed so the impact of price is not examined. 9 As price at point of use for hospital inpatient services will depend on insurance coverage but the BHPS does not record private insurance cover, occupational dummies are included as instruments for corporate cover. 10

4. The determinants of private, NHS and no demand