African traditional healers and outcome-contingent contracts in health care

Journal of Development Economics 71 (2003) 1 – 22 www.elsevier.com/locate/econbase

African traditional healers and outcome-contingent contracts in health care Kenneth L. Leonard Department of Economics, Columbia University, Mail Code 3308, 420 W. 118th Street, New York, NY 10027, USA Received 1 November 2000; accepted 1 April 2002

Abstract Even with the expansion of modern medicine, African traditional healers remain popular. This paper advances an economic perspective of healers to contribute to an explanation of this phenomenon. An important element of their practice has previously been ignored: healers use and are able to enforce outcome-contingent contracts. This, in turn, allows them to credibly deliver high quality care. Data on patient choice of health facility from Cameroun shows that patients choose healers over modern facilities for reasons that can be directly traced to the advantages inherent in the use of outcome-contingent contracts. D 2002 Elsevier Science B.V. All rights reserved. JEL classification: D8 I1 Keywords: Traditional medicine; Traditional healers; Asymmetric information; Outcome-contingent contracts; African health care

Traditional healers are a source of health care for which Africans have always paid (Van der Geest, 1992) and even with the expansion of modern medicine, healers are still popular and command fees exceeding the average treatment cost at most modern practitioners.1 A possible explanation is that healers have access to valuable and effective therapies unavailable to modern providers. Another view—more widely held among public health policy-makers—is that they are charlatans who consistently dupe their clients. This paper advances an economic perspective of healers that resorts to neither ignorant patients nor

E-mail address: [email protected] (K.L. Leonard). 1 In Kenya ‘‘[t]he average patient treatment cost per visit (in cash) reported by healers was 46 Ksh, far more than the mean charges even in the private health facilities’’ (Mwabu et al., 1993). In Cameroun, the average cost of a visit to a healer is larger than that at either government or church-run clinics (Leonard, 2000b). 0304-3878/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. doi:10.1016/S0304-3878(02)00131-1

2

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

extraordinary medical ability to explain their persistent popularity. An important element of their practice has previously been ignored: healers use and are able to enforce outcomecontingent contracts. This allows them to credibly deliver medical effort and therefore, high quality care. Data on patient choice of health care practitioner collected in Cameroun shows that patients choose healers over government and church-run (mission) clinics and hospitals for reasons that can be directly traced to the economic features of their practices. The attraction of the healer derives from the fact that health care is not a consumption good, but a set of inputs to the production of health. Health care is sought because it increases the probability of being cured and patients value cures. Medical effort is an important input to health, however, it is also a classic example of hidden action.2 Even when patients can observe the activities of practitioners, they cannot evaluate their appropriateness. A patient can leave a consultation without knowing whether she received high or low quality care. Because patients cannot observe inputs and do not value them directly, paying for outcomes (not inputs) is intuitively attractive. Nonetheless, the outcome-contingent contract remains an intellectual curiosity (Arrow, 1963; Dranove and White, 1987, p. 409; Mooney and Ryan, 1993, p. 126) The African traditional healer is unique in his general use of such contracts.3 Non-contractibility is frequently cited as a reason why outcome-contingent contracts are not more commonly observed (McGuire, 2000, p. 499). However, in Africa, patients believe healers are the agents of higher powers, and that they can therefore observe and verify outcomes to all illness conditions. This allows healers to use contracts that are not otherwise implementable. Another important aspect of the practice of healers is the attention they pay to the effort of patients (which can include following treatment regimes, changing diet, avoiding exercise, etc.). This is a natural extension of the fact that they use outcome-contingent contracts. Patient effort is another example of hidden effort. Just as patients cannot evaluate the effort of practitioners, so the practitioner cannot observe the effort of patients because patients exert effort out of their presence. If a practitioner is paid on the basis of his effort, or simply a fixed fee, he need not care if patients exert effort. When practitioners are paid on the basis of outcomes, patient shirking will impact the probability of getting paid. In the context of joint or team production (a` la Ho¨lmstrom, 1982) not only will healers care that patients exert effort, but their own choice of effort will depend on their expectation of patient effort. The characteristics of the outcome-contingent contract suggest that healers have highpowered incentives to exert hidden effort and that there should be some interplay between the efforts of patients and healers. These conclusions are tested by comparing healers to

2 Hidden action in health care is also referred to as agency. However, McGuire (2000, p. 499) suggests that there are at least two classes of models of physician agency: those where physicians withhold or disclose information about the patient’s condition and those in which physicians exert or do not exert effort to diagnose the patient’s condition. This paper discusses unobservable diagnostic quality or effort. The term hidden action is a more precise term, and is closely aligned with Harris and Raviv (1978), Arrow (1963, 1986), Dranove and White (1987), Ma and McGuire (1997), where terms such as quality, shirking, or slacking-off mirror the concerns of this paper. 3 Outcome-contingent contracts are in limited use in the US for areas such as reproductive medicine and corrective eye surgery (New York Times, 1999; Robertson and Schneyer, 1997).

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

3

the other available health care providers. Illness conditions (the collection of symptoms and characteristics of an illness that are visible to patients before they seek care) can be seen as production functions. Each illness condition can respond differently to medical effort, patient effort or other medical inputs. The healer represents one form of contract among many. The results show that patients choose a particular provider in part because that contract delivers levels of unobservable inputs that are appropriate to their illness condition. In the southwest Province of Cameroun, patients can choose between healers, churchrun (mission) hospitals and clinics and government hospitals and clinics. These other facilities take advantage of the fact that medical effort can be imperfectly evaluated by other medically trained personnel. Government and mission health care systems attempt to insure quality effort by regulating practitioners and providing direct incentives to exert effort. However, these institutions do not observe outcomes and practitioners’ compensation is not dependent on outcomes. Therefore, patient effort does not impact compensation and these practitioners do not have a direct incentive to care about the effort of patients. In addition, practitioners in government facilities have weak incentives to provide effort, whereas mission facilities are more closely regulated. This paper hypothesizes that healers provide high levels of effort compared to government facilities and provide particularly appropriate levels of effort when patient effort is important. Healers should be more popular for illness conditions that require high levels of effort and for illness conditions in which both medical and patient effort are required. The paper uses a mixed multinomial/conditional logit regression of the choice of these five practitioners (traditional, government clinics and hospitals and mission clinics and hospitals) against individual characteristics, illness condition characteristics and travel costs. Patients choose healers over government providers when their illness conditions require large amounts of medical effort and they choose healers over mission providers when they suffer under conditions that require large amounts of both medical and patient effort. These results can be directly traced to the economic characteristics of healers. This paper follows Eswaran and Kotwal (1985), which deals not with health care but with agricultural tenancy. Agents choose between a series of contracts, none of which are fully optimal, but among which one will be second-best optimal for some set of conditions. The output of interest is produced through the contribution of unobservable effort on the part of both participants. In this paper, the patient plays the role of landowner (since she is the owner of the productive input—her physical self) and the health care practitioner is the tenant. The contract available at healers is a sharecropping contract in which the patient (landowner) lets her physical self (her land) to the healer (sharecropper) and both share in the value of the outcome. The contract available at modern providers is similar to the wage contract, although with the addition of hierarchical supervision. There is no counterpart to the rental contract. Unlike Eswaran and Kotwal, the key source of exogenous variation driving contract choice is in the characteristics of the illness condition (the crop), not in the distribution of abilities among agents. A stylized description of the practices of healers follows. Section 2 looks at the incentives to provide unobservable medical effort under both outcome-contingent and effort-based regulation. Section 3 introduces the data set and preliminary observations.

4

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

Section 4 presents the results of a mixed multinomial/conditional logit estimation using illness characteristic data as well as individual characteristics. Section 5 concludes.

1. Traditional healers The way that patients consult, receive treatment and pay for a typical visit to a healer contrasts strongly with a typical visit to a modern provider. Data from the survey (discussed in greater detail in Section 3), interviews with healers in Cameroun, Tanzania and Ethiopia and anthropological studies are used to present a stylized view of their practices. The contrast in payment method is often noted in the literature (Stauga˚rd, 1985; Korse et al., 1989, for example), but the implications of this practice are not discussed. Healers ask for an initial payment, usually small or token. In addition, the healer will often negotiate with the patient over a payment to be made in the future. In all cases, if the treatment did not result in improvement, the patient paid nothing beyond the initial payment. This was the case even when patients lived with, were cared for and fed by the healer during treatment. Healers do not charge for medicines administered.4 In some cases, healers and patients did not reach explicit agreements over the future payment, and occasionally healers said they did not expect the final payment even if the patient was cured. However, healers were clear that patients always made additional payments when they were cured. The data set used in this paper contains information on how much patients paid healers. Total payments were twice as high when patients were cured than when they were not ( p-value 0.077), and payments made after consultation were three and a half times larger if the patient was cured ( p-value 0.047). In addition, healers have the reputation of poisoning or cursing patients with whom they were displeased. When asked about this practice, many healers were adamant that they never engaged in the practice5, though almost all admitted that their ancestors, or specifically parents, had done so. The practice operated as follows: when a patient refused to pay, the healer would either invoke a curse on the patient or revoke the cure. Neither of these actions took place in the presence of the patient. They are considered to be among the strongest forms of magic. This practice invokes near universal fear in rural populations, and most non-healers believed that if they failed to pay, they would be poisoned. This is a useful belief for healers, since they need only wait until the patient eventually falls sick of anything. All healers told stories of patients leaving without paying and then returning, sometimes years later, begging to be allowed to pay. That patients believe poisoning is still practiced allows healers to wait until after the treatment to collect payment without fear that the patient will refuse to pay.

4 Some healers in Ethiopia insist on a fixed payment for a particular type of medicine and never received payments after medicines were administered. However, these healers dispensed only two or three types of drugs and did not diagnose. Healers in Ethiopia who diagnose are paid primarily after the outcome of treatment. 5 Not surprising, since this practice is what transformed a healer into a witch doctor in the eyes of colonial officers.

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

5

The practices of healers are steeped in tradition and mystique. All healers see their practices within the context of culture and tradition and many were reluctant to talk about payment. Nonetheless, they are all paid for their services, whether in-kind or cash. The anthropological literature highlights the significantly greater time healers spend on diagnosis (Conco, 1972, pp. 291 – 294; Stauga˚rd, 1985, pp. 74 –81, 112, 124; Baerts, 1989, pp. 31– 32) This paper suggests the payment method and time spent on diagnosis are not unrelated. The form of payment and the ability to enforce it form a pattern in which healers face very different incentives to care for their patients than other providers.

2. Effort and capacity Patients visit a health care practitioner to benefit from the intervention of health care in the course of an illness. The impact of health care can be simplified into two components; effort and capacity. Capacity (skill) should be thought of as a location-specific constant, derived from the availability of equipment, medicines, training, etc. Capacity is observable: hospitals have more capacity than clinics and clinics more capacity than healers. Effort is a variable input and cannot be directly observed, but is at least as important as capacity. Being in the best equipped hospital in the world is of no use to a patient if no one will care for her. Modern medicine has superior capacity to traditional medicine; the comparative advantage of traditional medicine is in the provision of effort. 2.1. Medical effort at traditional healers The outcome-contingent contract offered by the healer creates incentives for the practitioner to exert effort. A combination of capacity (s), medical effort (m) and patient effort ( p) will lead to some gain in health, H = h(s, m, p). Though represented as a cardinal measure of the benefit of health, it is more accurate to think of h as the probability of being cured. Inputs increase the probability of being cured. The value of the cure is the difference in the utility of the patient when sick and when well, V. The expected gain to seeking health care is VH. Patients agree to share some of the benefits of being cured with healers. The payment to the healer when the patient is cured is a share (r) of the gain in utility plus some fixed payment ( F + rV).6 When the patient is not cured the healer is paid F. His expected payment is F + rVH. The exertion of effort has a cost that is normalized to a per unit cost of 1. The healer’s expected gain from treating a patient is Y ¼ F þ rVhðs; m; pÞ  m

ð1Þ

By increasing his effort, the practitioner increases the probability of a cure and therefore increases his expected payment. The healer will choose to exert effort at the point where 6 From the perspective of a particular patient with a particular illness condition the fact that the contingent payment to the healer is a share of the gain in utility (rV) is not relevant. From her perspective, there is a payment she will make if she is cured and a payment she will make if she is not cured.

6

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

the marginal benefit is equal to the marginal cost. As is well known in the sharecropping contract, this level of effort is always less than the optimal level of effort when r < 1. The patient also retains incentives to exert effort and her utility will be the share of the value of health retained, minus the disutility of effort exerted ( p), minus the fixed fees ( F), minus the travel costs (TC). U ¼ ð1  rÞVhðs; m; pÞ  p  F  TC

ð2Þ

Patients will exert effort until the marginal benefit is equal to the marginal cost. Again, this level of effort is less than the optimal level when (1  r) < 1. The value of the contract available at healers will vary between illness conditions. It is useful, therefore, to introduce some definitions through which to compare illness conditions. Definition 1 (Responsiveness to medical effort). An illness condition A is more responsive to medical effort than illness condition B if, ceteris paribus, the return to medical effort for condition A is greater than for condition B. BhA ðs; m; pÞ BhB ðs; m; pÞ > Bm Bm

bs; m; p

Definition 2 (Responsiveness to patient effort). An illness condition A is more responsive to patient effort than illness condition B if, ceteris paribus, the return to patient effort for condition A is greater than for condition B. BhA ðs; m; pÞ BhB ðs; m; pÞ > Bp Bp

bs; m; p

Assume that h(s, m, p) has the properties of a standard production function (decreasing marginal productivity of inputs). Therefore, if one condition has a higher marginal benefit of effort for every level of effort, the optimal level of effort will be higher for that condition. Proposition 1. For the outcome-contingent contract and two diseases with otherwise identical production functions, if illness A is more responsive to medical effort than illness B, more medical effort will be provided for illness A. Proposition 2. For the outcome-contingent contract and two diseases with otherwise identical production functions, if illness A is more responsive to patient effort than illness B, more patient effort will be provided for illness A. Despite the fact that patients cannot observe medical effort, healers will provide generally high (though not optimal) levels of effort that are increasing in the responsiveness of the disease to medical effort.

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

7

2.2. Medical effort at other practitioners If this contract helps to align incentives, why then are outcome-contingent contracts so rare in health care? A contingent-fee scheme depends on the practitioner’s knowing the outcome of the disease. Outcomes are difficult to observe and more difficult to verify. Often, when the patient leaves a clinic or hospital, the outcome is not yet known by anyone. If government centers were paid on the basis of outcomes, patients would have strong incentives to lie or not return.7 The healer is able to use outcome-contingent contracts because he maintains a cloud of mystery over his practice that encourages people to tell the truth about their condition. Although many different mechanisms are in use world-wide (including professional associations, referral networks, redress through courts (malpractice) as well as hierarchical incentive systems (Ellis and McGuire, 1986; Dranove, 1988; Blomqvist, 1991), for example) the institution of quality assurance commonly observed in Africa is direct monitoring of the practitioner (effort-based regulation). The employer of the practitioner does not seek to know the outcome of treatment but does observe other outcomes that give information about the effort of the practitioners. Practitioners produce both health for the patient and what can be called organizational quality. This second output is observed by the employer. Records are kept of the various activities that go into producing health. Typically, a selection of records are examined during a site visit. The patients’ symptoms and complaints are part of all records and therefore procedures and records should follow protocols developed for each set of complaints. If a particular record or collection of records is determined to be in violation of standards, the practitioner is punished in accordance with the gravity of the deviation. These practitioners provide quality effort because if they do not, they will be punished. The penalty varies between organizations. Supervisors of government facilities do not have the power to fire, promote, demote or give bonuses. Church-based supervisors, on the other hand, can use these tools. Mliga (2000) reports that in Tanzania, where he studied four different health care provision systems, those organizations that had the power to use these forms of incentives provided significantly superior quality of care. Mission personnel will exert more effort and adhere closer to protocols than government personnel. Organizational quality is a measure specifically designed to be highly correlated with outcomes. It can be seen as a production function ( q(m)) that is similar to the production function for health itself. Practitioners seek to maximize their utility by choosing m, which equates the marginal change in q with the marginal disutility of effort and exert more effort when the responsiveness of organizational quality to medical effort is larger. Assuming q is correlated with h, medical effort should increase when the responsiveness of h to medical effort is high, i.e. protocols are appropriately designed. However, since q is not a measure of outcomes and supervisors cannot observe patient effort, the change in q with respect to p must be zero (Bq/Bp = 0). Examination of records would show that correct 7 There are some health events that produce observable outcomes, for example, the outcome of a normal delivery—the health of mother and child are known before either returns home. Ndeso-Atanga (2000) reports that, in Cameroun, outcome-contingent appreciation (tipping) is the cultural norm for childbirth services. Observability appears to lead to exactly the contracts described at healers.

8

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

diagnostic procedures were used and that correct medicines were prescribed but they would be unable to show whether or not the patient took her medicines correctly, for example. 2.3. Joint determination of medical and patient effort Both the outcome-contingent contract and effort-based regulation can deliver high levels of medical effort. However, if both medical and patient effort are necessary, the outcome-contingent contract cannot achieve the full information solution (Ho¨lmstrom, 1982). On the other hand, effort-based regulation can, in theory, achieve the full information solution. It will not do so for a variety of reasons; regulation is imperfect and neither governments nor missions are pure social welfare maximizers. Since neither regulation nor welfare maximizing play a role in the healer’s contract, these failures in effort-based regulation will increase the relative advantage of healers. In addition, the fact that patient effort is important in health care suggests a more fundamental difference between the two types of contracts. For outcome-contingent contracts, the level of effort provided by either the practitioner or the patient will depend on the level of effort provided by the other. Changes in the patients’ expectation of medical effort will change the optimal level of patient effort. If medical and patient effort are complements, then the expectation of higher medical effort will lead to higher patient effort and, in parallel, the expectation of higher patient effort will lead to higher medical effort. Proposition 3. For the outcome-contingent contract, if the cross-partial of health with respect to medical and patient effort is positive, an increase in patient effort will lead to an increase in medical effort. Define m*( p) to be the function describing the optimal level of medical effort given the practitioner’s expectation of patient effort (the reaction function). Implicit differentiation yields B2 h Bm*ðpÞ BmBp ¼ 2 Bp B h Bm2

ð3Þ

With decreasing marginal returns, if the cross-partial is positive, then the change in the optimal m with respect to p is positive. The healer has a direct incentive to encourage patients to exert effort; if patient effort increases the expected outcome, it increases the practitioner’s expected payment. Proposition 3 shows that there is a second effect of patient effort which depends on whether patient and medical effort are complements. If efforts are complements, the more effort the patient exerts, the more useful is the effort of the practitioner. The effect of these incentives can be seen in the degree to which healers concentrate on encouraging or forcing patients to exert effort. The nature of health care is such that medical and patient effort can be characterized as complements. The patient seeks the care of a practitioner precisely because she cannot do what the practitioner does. Though there are some areas in which patients could substitute

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

9

for practitioner effort, these have already been exploited in Africa. Patients and their families already provide a number of medical services typically provided by nurses in the west.8 In addition, the contribution of patients to their own health can rarely be replaced by medical effort. Medical and patient effort are specialized and complementary inputs. Though the cross-partial of health with respect to patient and medical effort is positive, the cross-partial of organizational quality is zero, because patient effort is not observed by regulators. This does not prevent the regulator from achieving the optimal solution; he can act as a Stackelberg leader if he has information about the patient’s reaction function. However, experience suggests that health regulators, far from knowing the reaction function of patients, see patients as passive players in their own health, and do not take into account the effort of the patient at any level.9 In a separate investigation10, of 450 consultations by practitioners at government and mission hospitals, in only 16% did the doctor tell the patient what the diagnosis was or what medicine was being prescribed. In only 6% of cases did the doctor inform the patient of any activity that she could undertake to increase the chance of recovery or to avoid a similar illness in the future. Of cases in which a dispensing nurse gave the patient drugs requiring that the patient know how or when to take the drug, in only 32% did the nurse check to see if the patient had any idea how to do so. Patient effort is not a priority in either mission or government organizations (there were no significant differences in these measures between mission or government facilities). They do not even take into account the effect of effort on outcomes, much less react to patient provision of effort in their own decision about how much effort to exert. 2.4. Capacity Capacity is important in the provision of health but is more readily observable. Each location has a fixed level of capacity and illness conditions can respond differently to the capacity present at each provider. Thus, not only does the benefit of capacity change for different illness conditions, but the difference between different levels of capacity can also change. For example, when a patient suffers from malaria, the superior capacity of hospitals over clinics offers the patient no particular benefit—everything necessary for a cure is available at both centers. However, if she suffers from appendicitis, the capacity present at a hospital offers a distinct advantage over a clinic. Definition 3 (Responsiveness to capacity). An illness condition A is more responsive to the capacity at a particular provider (sj where j is the provider index) than illness condition 8 Van der Geest and Sarkodie (1998) provide an eye-opening description of life as a patient in a typical African hospital, including the extensive reliance on family to provide important nursing functions. 9 Since patients retain strong incentives to exert effort on their own behalf, medical personnel might feel justified leaving patient effort as the concern of patients. However, patient effort is increasingly discussed and medical personnel are beginning to take responsibility for managing and reacting to patient effort. This is especially important with the complex pharmaceutical cocktails required to suppress the advance of AIDS in industrialized countries. 10 Research was sponsored and carried out by Dr. Mpuya under the auspices of the District Medical Office of Iringa, Tanzania with the support of Collegio Universitario Aspiranti e Medici Missionari.

10

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

B if, ceteris paribus, the return to the capacity at that provider for condition A is greater than for condition B. BhA ðsj ; m; pÞ BhB ðsj ; m; pÞ > Bsj Bsj

bsj ; m; p

In this sample, modern providers operate both clinics and hospitals. A mission clinic is similar in capacity to a government clinic and mission hospitals are similar to government hospitals. Clinics are always at least as skilled as healers and that hospitals are always at least as skilled as clinics. Thus, for each illness condition, there are three possible responsivenesses to capacity. Proposition 4. Conditions in which the difference in responsiveness to capacity between types of centers is greater lead to a greater difference in the benefits of visiting those types of centers. 2.5. Patient selection of practitioners Imperfect regulation in modern providers and outcome-contingent contracts at healers suggest the following differences between centers. Traditional healers provide high levels of effort and encourage patients to do so as well. Mission clinics and hospitals also provide high levels of effort but do not encourage patients to do so. Government hospitals and clinics, on the other hand, do not provide high levels of effort. Clinics provide higher levels of capacity than healers and hospitals provide higher levels of capacity than clinics. These differences are known to patients and should lead to distinct patterns in the choice of practitioner. Proposition 1 stated that as the responsiveness to medical effort increased, the amount of medical effort provided also increased. Some practitioners have higher powered incentives to provide effort and this difference will be most important when medical effort has the greatest impact on outcomes. Since all effort provision is sub-optimal (none of the contracts achieve the full information solution) an increase in effort is welfare improving. Thus, for two different illness conditions, A and B, and two different practitioners, I and II, Hypothesis 1. If condition A is more responsive to medical effort than condition B and practitioner I faces greater incentives to provide unobservable effort than practitioner II, then the probability of a patient’s choosing provider I over II will be higher for condition A than condition B.11 From Proposition 4, Hypothesis 2. If condition A is more responsive to capacity at a given level of practitioner (untrained, clinic or hospital) than condition B, the probability of a patient choosing that level over other levels will be higher for condition A than condition B. 11 Formal proofs of this and the following hypotheses require specification of functional forms and are contained in Leonard and Graff Zivin (2001).

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

11

Fig. 1. Four combinations of responsiveness to medical and patient effort and the medical effort that would be provided at healers and institutional providers with high-powered incentives.

Proposition 3 stated that for healers, an increase in p leads to an increase in m. It is unlikely that this effect exists for organizational practitioners (missions and the government). If medical effort under effort-based regulation does not increase with the responsiveness to patient effort, then the outcome-contingent contract is more likely to be superior to effort-based regulation when the responsiveness to patient effort is high. Fig. 1 shows four types of conditions with combinations of low and high responsiveness to medical and patient effort. Shown are the levels of effort provided by the healer and an institutional practitioner with high-powered incentives. Both practitioners provide more effort for conditions with high responsiveness to medical effort than for conditions with low responsiveness to medical effort (compare [2] to [1] and [4] to [3] in both tables). However, only healers provide more effort when Ep is higher (compare [4] to [2] in both tables). High effort at a healer cannot be directly compared to high effort at an organizational practitioner, however, the best a healer can do when compared to an organizational practitioner is when both practitioners provide high levels of effort. This corresponds to illness conditions where the responsiveness to both medical and patient effort are high. This intuition leads to the following hypothesis: Hypothesis 3. If condition A is more responsive to patient effort than condition B and the responsiveness to medical effort is high, the probability of choosing a healer over a mission clinic or hospital is higher for condition A than condition B. Note that, even if the responsiveness to patient effort is high, a patient might prefer to visit a modern provider since the patient retains the full incentive to exert effort on her behalf. It is only when she expects that her effort and medical effort could significantly complement each other that she will seek an outcome-contingent contract. These hypotheses permit investigation of patient perceptions of the level of effort provided by healers, government practitioners and mission practitioners. If healers are providing high levels of effort, this should be detected by comparing the conditions reported at government, mission and traditional practitioners.

3. Data To examine the role of incentives in the choice of practitioner, we use data collected on patient behavior in the face of illness in Mbonge Sub-Division, in the southwest province of Cameroun in 1994.12 Mbonge sub-division is entirely rural. This area was chosen 12

The data collection effort was joint with Sylvester Ndeso-Atanga.

12

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

Table 1 Average characteristics of patients (and caregivers) differentiated by practitioner visited Variable

Government clinic

Mission clinic

Government hospital

Mission hospital

Traditional healer

Patient (count) Adults ( z 16) (count) age female (%) education (years) incomea Children ( < 16) (count) age female (%) education (years) Caregiver age female (%) education (years) incomea Household predicted family incomea,b

154 61 36.93 59 4.02 3.26 93 5.06 43 1.26

140 59 40.98 47 3.90 7.32 81 4.23 43 0.68

127 73 42.58 60 3.60 3.38 54 5.98 37 1.91

53 46 48.50 46 2.54 10.76 7 7.14 71 1.29

61 38 38.00 50 4.18 3.37 23 7.65 61 1.78

38.31 53 4.32 5.22

38.70 51 4.41 8.46

40.58 50 4.96 6.37

42.42 42 4.19 10.52

38.57 39 4.62 5.58

6.05

9.58

7.21

11.40

6.12

a

Units are 100,000 CFA franc/year c $200 US. b Predicted family income is obtained by regressing actual total household income on observable characteristics of the household (roofing material, type of floor, ownership of durables, productive animals, etc.). This is a more general measure of household resources.

because of the presence of a German aid project which insured a consistent, reasonably priced drug supply in all government health centers and hospitals, permitting the claim that factors other than the availability of drugs are driving patients’ choices. This fact also suggests that patients can be relatively sure of receiving some treatment at any location. Forty villages were randomly chosen and twenty randomly selected households from each village were interviewed. Data were collected on all members of the household: 4489 individuals were thus polled, of which 681 illness episodes were reported within the month previous to the survey; 548 of these episodes resulted in first visits to one of the five types of practitioners examined in this paper13 (with complete data for 533). Villages sampled were an average of 28 km from a government clinic, 51 km from a government hospital, 87 km from a mission clinic and 212 km from a mission hospital. Table 1 compares patients and caregivers by their choice of practitioner. When patients are young or infirm, the caregiver is the person who takes them to the health care facility and manages therapy (the patient is her own default caregiver). The characteristics of patients are subdivided at 16 to clarify differences in education and income. By adult income, caregiver income and household income, visitors to healers are among the poorest in the sample. On the other hand, education levels do not reflect this pattern. This is because travel costs are a significant factor in the choice of a practitioner. Access to a road both reduces travel costs and increases incomes. Since healers are located 13 Other practitioners included drug peddlers, pharmacists, neighbors, private hospitals, private clinics and parastatal hospitals.

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

13

Table 2 Outcomes of illness episodes differentiated by practitioner visited Result

Government clinic

Mission clinic

Government hospital

Mission hospital

Traditional healer

Total

Died Cured Well enough Sought other care Went home (uncured) Referred Continued treatment

1.92 69.87 9.62 1.28 3.85 2.56 10.90

2.70 56.76 10.14 2.70 4.05 3.38 20.27

1.47 55.88 16.18 2.21 5.15 0.74 18.38

3.23 41.94 17.74 4.84 9.68 0.00 22.58

4.55 40.91 12.12 4.55 4.55 3.03 30.30

2.46 56.69 12.50 2.64 4.93 2.11 18.66

The percentage of all visits to a practitioner is reported, resulting in given outcome.

throughout the sample and government clinics are the most widely distributed of the modern health care practitioners, choosing a healer or government clinic out of geographical isolation from other centers is correlated with poverty. Since education levels do not have the same geographical distribution as income and family income, the same pattern is not observed. When travel costs are taken into account, poorer individuals still prefer healers, however, there is no effect of caregiver income or family income. Table 2 shows evidence of patient selection of providers. This table reports the outcomes at different practitioners. Almost 5% of people who visited healers died and healers have the lowest cure rate (looking at both ‘cured’ and the sum of ‘cured’ and ‘well enough’). At first glance, this supports the idea that these practitioners are charlatans. However, such a conclusion has to assume passive patients who visit providers without taking into account any information about their illness condition. By this assumption, the second worst practitioner is the mission hospital. Mission hospitals are, beyond question, the highest quality centers. The best cure rates are at government clinics and these are the worst of the modern practitioners. The findings do not reflect quality, but rather selfselection by illness condition. Information about their illness condition drives patients to incur significant cost in the search for care. The survey was designed to elicit information that patients had before they sought care; all of the symptoms they experienced, the self-declared severity of the disease, the number of days sick before seeking care, and the number of those days in which the patient was bedridden. A table of symptom prevalence at each provider is included in Appendix A as Table 5. Mwabu (1986) shows similar symptoms/health care provider matches in Kenya. This table shows strong patterns of selection by illness condition, but does not provide any rationale for this behavior. With the characteristics of the disease plus the age and sex of the individual and information about endemic diseases in the area (but not information on the choice of practitioner or the diagnosis), scores were created by examining basic medical references (Griffith, 1985; Strickland, 1984; Werner, 1977).14 14 These references provided information about the medical effort necessary to differentiate between alternate diagnoses given conditions presented, the treatment regimes required, the participation of patients, the medicines and diagnostic tools required and the possible outcomes, among other things.

14

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

The following measures were created: the responsiveness to medical effort (Em), the responsiveness to patient effort (Ep) and the responsiveness to capacity (Es). This latter consists of the responsiveness to capacity at untrained or informally trained practitioners (Eus), practitioners at clinics (Ecs), and practitioners at hospitals (Ehs). In addition, each illness condition was scored according to the range of possible outcomes (r). For example, if death is a possible outcome, r is larger. From the scores for Ep and Em, a third score was created: the joint responsiveness to medical and patient effort (Epd  Em ). This variable reflects whether or not the responsiveness to medical and patient effort are simultaneously high.15 To validate this method, two doctors and one nurse (all experienced in rural tropical medicine) independently scored all the cases using the same terms. These scores are broadly correlated with the scoring from medical references and the econometric results they suggest are similar.

4. Empirical analysis Eq. (4) is the reduced form equation of the net benefit to seeking health care, with indices added for clarity. Uij ¼ f ðj; Eim ; Eip ; Eijs ; ri ; Zi ; TCij ; Fj Þ;

ð4Þ

with the following definitions;

Variable

Definition

i j Uij Eim Eip Eijs ri Zi TCij Fj

index of individuals and their illness conditions index of providers, J={TH,GC,MC,GH,MH} net expected utility for patient i at practitioner j responsiveness of patient i’s illness condition to medical effort responsiveness of patient i’s illness condition to patient effort responsiveness of patient i’s illness condition to practitioner j’s capacity for that illness condition range of possible outcomes for patient i’s illness condition individual characteristics travel cost for individual i to practitioner j fixed fees at practitioner j

There are five practitioners between which patients choose; traditional healers (TH), government clinics (GC), mission clinics (MC), government hospitals (GH) and mission hospitals (MH). In addition, since regulation is provided by the same institution, the effort provided at mission clinics is the same as at mission hospitals and the effort provided at 15 E pd  Em is defined as the residual vector obtained from regressing EmEp on both Em and Ep. Therefore,  Em is not correlated with either Em or Ep but is correlated with EmEp. E pd

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

15

government clinics is the same as at government hospitals. Thus we introduce the terminology of type indexed by l. 8 8 < GC < MC l ¼ T if j ¼ TH l ¼ G if j ¼ l ¼ M if j ¼ : : GH MH Coefficients are obtained by maximizing the following log likelihood with respect to c, q, and g. logL ¼

n X X

dij logPij

i¼1 jaJ

expðcVxij þ ql Vyi þ gj Vzi Þ Pij ¼ X expðcVxim þ ql Vyi þ gmVzi Þ maJ

dij = 1 if the ith individual visits practitioner j and 0 otherwise. xij is a vector of information about the potential practitioners ( j) for the individual and her illness condition, i. It includes the travel cost and the responsiveness of illness condition i to provider j’s skill. There is one vector of coefficients (c) corresponding to x. These are referred to as the conditional coefficients because they enter the estimation as would the regressors of a conditional logit estimation. yi is a vector of information about the patient’s illness condition and is one vector of information with three sets of coefficients (ql) for the index l. ql are multinomial coefficients with the term for government clinics and hospitals restricted to be equal and the term for mission clinics and hospitals restricted to be equal (called hybrid coefficients for clarity). yi are the responsiveness to medical and patient effort and the joint responsiveness. zi is a vector of characteristics of the individual. There is one vector of characteristics per individual and five sets of coefficients (gj), one for each practitioner. gj are multinomial coefficients. zi are the age, gender, education level, and income (log) of both the patient and her caregiver, predicted household income and the range of outcomes (r). There is a dummy variable for each provider (to contain information on fees). In order to solve the model, gTH and qT are normalized to zero. The regression is a specific case of the more general conditional logit model (Maddala, 1983, p. 44) and therefore has the required properties for obtaining a solution. 4.1. Results Table 3 reports the coefficients and z-tests of the regression. The base choice is the healer, so, for example, the negative coefficient on the age of the patient in the column for the government clinic implies that increasing the age of the patient increases the probability of a visit to a healer over a government clinic. Variations in the age, education

16

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

Table 3 Discrete choice regression of choice of five providers against individual and illness condition characteristics Multinomial coefficients

Government clinic

Mission clinic

Government hospital Mission hospital

Coefficient z-test

Coefficient z-test

Coefficient

z-test

Constant Age (individual) Female (individual) Education (individual) Log income (individual) Age (c  g) Female (c  g) Education (c  g) Log income (c  g) Log wealth r (range of outcomes)

2.204  0.034  0.097

2.58  3.00  0.29

3.429  0.034  0.090

3.79  2.87  0.26

0.259  0.005  0.029

0.29  0.46  0.08

2.472 0.017 0.339

2.18 1.45 0.78

 0.076

 1.65

 0.110

 2.30

 0.066

 1.45

 0.073

 1.20

0.044

1.82

0.064

2.50

0.021

0.89

0.054

2.03

0.022 0.406 0.011  0.014  0.197 0.832

1.53 1.05 0.23  0.63  1.16 3.19

0.017 0.426  0.001 0.005  0.008 0.872

1.14 1.08  0.01 0.19  0.05 3.27

0.029 0.597 0.078  0.029  0.025 1.171

1.94 1.49 1.63  1.24  0.14 4.45

0.012 0.179 0.021  0.038 0.263 0.714

0.66 0.36 0.32  1.36 1.29 2.40

Hybrid coefficients

Government Coefficient z-test

Coefficient z-test

Ep (patient responsiveness) Em (medical responsiveness) E pd  Em

 0.179

 1.11

 0.206

 1.18

 0.555

 2.36

 0.400

 1.64

 0.306

 2.05

 0.530

 3.19

Conditional coefficients

Coefficient z-test

Travel cost s E{u,c,h}

 0.750 0.342

Coefficient z-test

Mission

 7.95 2.17

log likelihood =  701.84. 44.7 of predictions correct. Mixed multinomial/conditional logit regression where the multinomial coefficients for healer are normalized to zero. Patients choose between healers, mission clinic and hospitals and government clinics and hospitals, with the p E m, are restricted to be the same for clinics and healer as the base choice. Multinomial coefficients for Em, Ep, Ed hospitals of the same organization.

and income of patients and the age of the caregiver are the only significant predictors of visits among individual and caregiver characteristics. Visitors to healers are not less educated than visitors to other practitioners and they are more educated than visitors to government and mission clinics. They earn less income than visitors to government clinics and mission centers, but are not from poorer families. The collection of illness condition characteristics play a significant role in the choice of practitioner. The responsiveness of the illness condition to patient effort does not appear to play any direct role in the choice of practitioner. The responsiveness to medical effort is significantly negative for visits to government centers, and negative, but not significant for visits to mission centers. Hypothesis 1 stated that if one practitioner exerted more effort than

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

17

another, patients are more likely to visit the high-effort practitioner when they suffer from a condition that is responsive to medical effort. Thus, patients choose practitioners as if they believed that healers provide more effort than government facilities and similar levels of effort to mission facilities. In addition, the joint responsiveness is significant for both government and mission practitioners. As suggested by Hypothesis 3, patients are more likely to visit healers when the responsiveness to both medical and patient effort are high. The range of outcomes is a significant determinant of the choice not to visit a healer for every pair. The coefficient for capacity is significant and positive, meaning that as the benefit of capacity (for the patient’s illness condition at a given practitioner) increases the likelihood of a visit to that practitioner also increases. In addition, the coefficient for travel is significant; patients prefer to travel shorter distances. Table 4 shows the marginal effects of the independent variables on the choice of practitioner derived from the four sets of coefficients of the logit. Each element can be read as the percentage change in the probability of visiting a practitioner given a 1% change in the respective variable from its average value. Elasticities for the responsiveness to medical effort are reported at three different values of the responsiveness to patient effort, and reflect both the direct effect through responsiveness to medical effort and the indirect effect through the joint responsiveness to medical and patient effort. Elasticities are not reported for individual characteristics with no significant coefficients. Increasing the level of education has a strong positive impact on the probability of a visit to a healer, with the difference coming from visitors to mission clinics. It is unlikely that education drives these results and more likely that it is a proxy for an omitted variable. However, clearly it is not only the uneducated who visit healers. Increasing the income of a patient does reduce the overall probability of a visit to a healer as well as government hospitals (though the elasticity is not significantly negative for healers). Healers appear to serve the poor. This could be through the implied credit and subsidy available in the contract at a healer. Poor patients are asked to pay less and even

Table 4 Elasticity of the probability of visiting a given practitioner with respect to significant explanatory variables from Table 3 Traditional healer Age (individual) Education (individual) Income (individual) Age (c  g) r (range) Em for Ep low Em for Ep medium Em for Ep high

Government clinic

Mission clinic

0.045 (0.054) 0.018 (0.001)

 0.110 (0.017)  0.001 (0.129)

 0.090 (0.018)  0.023 (0.049)

0.093 (0.006) 0.006 (0.023)

0.062 (0.002) 0.000 (0.017)

 0.045 (0.048)

0.017 (0.083)

0.073 (0.025)

 0.058 (0.029)

0.012 (0.011)

 0.075 (0.004)  0.082 (0.023)  0.093 (0.037) 0.034 (0.030) 0.182 (0.043)

0.030  0.003  0.132  0.036 0.058

(0.028) (0.017) (0.087) (0.026) (0.086)

 0.025 0.008 0.267 0.026  0.228

Government hospital

(0.014) (0.041) (0.133) (0.031) (0.133)

0.092 (0.020) 0.086 (0.076)  0.113 (0.073)  0.031 (0.022) 0.049 (0.074)

Mission hospital

 0.022  0.009 0.071 0.007  0.061

(0.015) (0.026) (0.038) (0.008) (0.036)

Reported value is the percent change in probability of visit with a 1% change in explanatory variable from its mean value. Standard error of elasticity (bootstrap method) in parentheses. Explanatory variables without significant coefficients are omitted. Low is 30th percentile and high is 70th percentile.

18

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

that is in-kind and when the patient is able. Mission centers attempt to reduce fees for the poor, but a major component of the cost of a visit to a mission center is travel costs, which cannot be reduced or forgiven. Age plays an important role, but there is no particular hypothesis about the role it should play. Age could contain some uncaptured information about income earning opportunities. It might contain information about variations in illness conditions but this should have been taken into account in the codings. When the range of possible outcomes is larger, the probability of a visit to a government hospital or mission clinic is increased, whereas the probability of a visit to a mission hospital, government clinic or healer is decreased. The effect is strong at healers. This variable captures information about the urgency of care. If there is a health crisis, patients will always find someone present at government hospitals or mission facilities. Traditional healers are not on call and do not have stocks of medicine on hand. Mission hospitals are all distant and thus inappropriate in emergencies. The effect of an increase in the responsiveness to medical effort on the probability of visiting a healer is negative when the responsiveness to patient effort is low, but large and positive when the responsiveness to patient effort is high. When the joint responsiveness is high, patients are drawn from all four other locations to healers, but mostly from government centers. The analysis of the advantages of visiting different providers suggested three hypotheses: patients visit high-effort providers when they need medical effort; they seek outcome-contingent contracts when they need both medical and patient effort; and they seek high-capacity providers when they need capacity. In addition, evidence was presented that government facilities are low-effort facilities, missions are high-effort facilities, and healers provide more effort than government facilities and might provide more effort than mission facilities. The results suggest that healers do provide more effort than government centers and effort on par with mission centers. More significantly, an important feature of the outcome-contingent contract and the practices of healers is confirmed in the data: healers work with patients better than modern providers. The fact that healers face high-powered incentives is a direct outcome of the type of contracts that healers use. The paper proposed that their comparative advantage is in illnesses in which both patient and medical effort matter and this hinges on the fact that medical and patient effort are complements. However, in the reduced form specification, joint responsiveness identifies illness conditions for which medical opinion sees medical and patient effort as complementary. This does not require that medical and patient effort are complements for every illness condition. The reduced form offers evidence that healers specialize in those conditions for which medical and patient effort are complements. 4.2. Sensitivity tests Two additional tests were performed on the data. A binomial specification was tested in which patients choose between healers and all other providers grouped together. This would be the first stage of a model in which patients choose first between modern and

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

19

traditional and then between modern providers. Responsiveness to medical effort and the joint responsiveness to medical and patient effort are both significant and show that patients prefer healers when the responsiveness to medical effort is high and also when the joint responsiveness to medical and patient effort is high. These results are not different from those shown above and suggests that simultaneous choice over all providers is not an unreasonable view of behavior. The second specification test uses information from the illness codings given by the two doctors and one nurse. A regression was run in which a composite score was created by adding together each of the four scores. The regression simultaneously found maximum likelihood coefficients for each composite score (responsiveness to medical effort, patient effort, capacity, joint responsiveness to medical and patient effort and the range of outcomes) as well as the optimal weights for each of the four sources. The results are almost identical to the results shown above and the medical references scoring has the highest weighting. Thus, the score from medical references is the best match to patient behavior (in the context of this model) and including information from other sources does not change the results discussed above. 4.3. Endogeneity concerns The health status of individuals is not generally considered exogenous (Strauss and Thomas, 1995). If the location of patients determines the illnesses from which they are likely to suffer as well as the distance to facilities, it is possible that the patterns derive from reasons other than those discussed above. The statistical methods used control for such variation, but if the geographical variations in illness condition are large, then our model is potentially misspecified. The same could be true if particular types of individuals suffer from particular illness conditions and also have unique reasons to visit providers. If this were true, location or individual characteristics should have explanatory power over the characteristics of illness conditions used in this analysis. The poor (who work as rural laborers) are more likely to be injured, but are they more likely to suffer from illness conditions that are responsive to joint effort? Regressions of illness condition characteristics on location and individual characteristics have been run and show no statistically significant patterns: location and individual characteristics do not explain the characteristics of illness conditions. The illness condition is not determined by characteristics of location or the patient.16

5. Conclusions This paper explains part of the continued popularity of healers by using an economic view of their practices. Healers can enforce contracts that depend on outcomes and therefore can use the economically advantageous outcome-contingent contract, and credibly deliver medical effort. The value of this contract shows up in comparison to

16

These results and those mentioned above, are available from the author on request.

20

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

modern providers who cannot use this type of contract. Healers contrast directly with modern providers in their attention to the effort of patients and patients in this data set choose healers disproportionately when they suffer from conditions that are responsive to both medical and patient effort. Traditional healers also provide more effort than the poorly regulated government providers and are chosen over them for conditions that are responsive to medical effort alone. This paper advances an image of patients that are active in the production and pursuit of their own health. This has important implications for health policy, where there is currently a strong bias towards the ‘passive patient’. The definition of the word patient includes ‘one that is acted upon’. Indeed Sen (1995, p. 11) used this definition in a different context as a contrast to agents: ‘‘To see (them) as patients rather than as agents can undermine the exercise. . . Not to focus on the fact that they think, choose, act, and respond is to miss something terribly crucial.’’ The patients interviewed for this paper are not passive in either treatment or choice of practitioner. Healers are successful because they understand this. Traditional healers are ‘traditional’. Their success depends on a belief that is based on an indigenous and pre-modern set of beliefs. Ironically, this belief allows them to be more economic, and therefore modern, than the modern providers in our sample. Unfortunately, despite the gains this ‘tradition’ offers to both patients and practitioners, it is not transferable. On the other hand, there is evidence that the incentive-aligning mechanism used at modern providers can deliver high-quality effort if it is properly used. The fact that regulation is weak at government providers is an economic inefficiency that should be addressed. Quality at government facilities could be significantly increased and even if fees had to be increased to pay for it, it should increase total welfare, since effort is currently sub-optimal (Leonard, 2002, 2000a). In addition, modern medicine does not take an adequate view of patient effort. This paper does not offer a way forward on either point, but it does suggest that progress could be welfare enhancing.

Acknowledgements This work was funded NSF grant #94-22768 and a University of California Rocca Fellowship. I am grateful to Mr. Mwela in Tanzania and Mr. D. Phinheas in Ethiopia for their assistance in interviewing healers; to Dr. D. Mfonfu and his staff in Kumba for their assistance in the household survey; and to Dr. D. Haile Mariam, Dr. G. Djomand and Ms. T. Pouani for their expertise in tropical medicine. This paper has benefited from the comments of David Bronkema, Joshua Graff Zivin, Lena Edlund and Rodney Ramcharan and two referees.

Appendix A . Illness Condition Characteristics Table 5 shows the prevalence of symptoms reported to each practitioner. Reported are the number of episodes in which that symptom was found as well as the deviation in percentage points of the observed against the predicted percentage of visits. If the

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

21

Table 5 Symptom prevalence by provider count and deviation in observed percentage from expected percentage Symptom

Location Government clinic

Fever Headache Cough Extremity pain Abdominal pain Malaise Vomiting Stomach ache Diarrhea Chest pain Appetite Extremity Swelling Eye problem Abscess Rash Constipation Short of breath Runny nose Lesions Convulsions Fracture Deep cut General injury Lower back pain Epilepsy Coughing blood

Mission clinic

Government hospital

Mission hospital

Traditional healer

Total

#

dev.

#

dev.

#

dev.

#

dev.

#

dev.

#

95 33 21 10 9 12 11 6 12 5 9 5 3 9 10 8 3 9 4 7 2 4 3 2 1 1

6* 1 3  12*  12* 5 4  13* 5  13 1 9  16* 10 19* 18*  11 27* 2 30*  11 11 1 9  18  15

77 24 27 18 9 16 16 13 13 7 12 4 10 3 4 6 6 6 3 1 2 1 0 3 2 4

2 4 7 5 9 5 10 7 10 4 14*  10 16*  13 7 9 9 11 3  18 8  16  26* 4 4 31*

66 33 20 13 10 14 15 9 10 13 7 7 7 4 2 2 4 1 1 2 4 5 4 2 0 0

0 7* 1 1 5 4 10* 1 4 18* 0 4 5 6  14  12 0  17*  16 7 13 26* 16 4  24*  24

16 8 6 9 13 5 1 8 1 6 1 2 3 1 4 0 3 0 4 0 1 0 0 1 1 2

 4* 3 3 6 15* 0  8* 11* 7 9* 7 2 3 6 9  10 8  10 21*  10 1  10  10 0 1 19*

20 11 7 8 12 4 1 3 0 0 1 7 1 6 1 1 1 0 1 2 2 0 3 2 5 0

 4* 1 3 2 11* 4  9* 4  11*  11* 8 17* 7 15* 7 6 6  11 4 5 7  11 19* 9 44*  11

274 109 81 58 53 51 44 39 36 31 30 25 24 23 21 17 17 16 13 12 11 10 10 10 9 7

* Indicates v2 test of independence rejected at 10% significance level with the null that all symptoms are found in the same proportion at all practitioners.

proportion of visits with a given symptom is equal to the total proportion of patients visiting that practitioner, the deviation is zero. References Arrow, K.J., 1963. Uncertainty and the welfare economics of medical care. American Economic Review 53 (5), 941 – 973. Arrow, K.J., 1986. Agency and the market. In: Arrow, K.J., Intriligator, M.D. (Eds.), Handbook of Mathematical Economics, vol. 3. Elsevier, Amsterdam, pp. 1183 – 1195. Baerts, M., 1989. Guerisseurs et plantes medicinales de la region des cretes Zaire – Nil au Burundi, number 18 in Annales Science Economiques, Koninklijk Museum voor Midden-Afrika, Tervuren, Belgie. ˚ ., 1991. The doctor as double agent: information asymmetry, health insurance, and medical care. Blomqvist, A Journal of Health Economics 10 (4), 411 – 432. Conco, W.Z., 1972. The African Bantu practice of medicine: some preliminary observations. Social Science & Medicine 6 (3), 283 – 322.

22

K.L. Leonard / Journal of Development Economics 71 (2003) 1–22

Dranove, D., 1988. Demand inducement and the physician/patient relationship. Economic Inquiry 26 (2), 281 – 298. Dranove, D., White, W.D., 1987. Agency and the organization of health care delivery. Inquiry 24, 405 – 415. Ellis, R.P., McGuire, T.G., 1986. Provider behavior under prospective reimbursement: cost sharing and supply. Journal of Health Economics 5 (2), 129 – 151. Eswaran, M., Kotwal, A., 1985. A theory of contractual structure in agriculture. American Economic Review 75, 352 – 367. Griffith, H.W., 1985. The Complete Guide to Symptoms, Illness and Surgery. The Body Press, Los Angeles, CA. Harris, M., Raviv, A., 1978. Some results on incentive contracts with applications to education and employment, health insurance, and law enforcement. American Economic Review 68 (1), 20 – 30. Ho¨lmstrom, B., 1982. Moral hazard in teams. Bell Journal of Economics 13, 324 – 340. Korse, P., et al., 1989. Sorcellerie, maladie et chasseur de sorciers chez les Mongo (Rep. du Zaire), number 103 in CEEBA, series 11, Publications (CEEBA), Bandundu, Zaire. Leonard, K.L., 2000a. Agency in health care: an empirical investigation in rural Cameroun, mimeo (available online), Columbia University. Leonard, K.L., 2000b. Incentives and rural health care delivery. In: Leonard, D.K. (Ed.), Africa’s Changing Markets for Human and Animal Health Services. Macmillan, London. Chapter 4. Leonard, K.L., 2002. When states and markets fail: asymmetric information and the role of NGOs in African health care. International Review of Law and Economics 22 (1), 61 – 80. Leonard, K.L., Graff Zivin, J., 2001. How to compensate physicians when both patient and physician effort are unobservable, mimeo (available on-line), Columbia University. Ma, C.A., McGuire, T.G., 1997. Optimal health insurance and provider payment. American Economic Review 87 (4), 685 – 704. Maddala, G., 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge Univ. Press, Cambridge, UK. McGuire, T., 2000. Physician agency. In: Culyer, A., Newhouse, J. (Eds.), Handbook of Health Economics. Elsevier, Amsterdam, pp. 461 – 536. Mliga, G.R., 2000. Decentralization and the quality of health care. In: Leonard, D.K. (Ed.), Africa’s Changing Markets for Human and Animal Health Services. Macmillan, London. Chapter 8. Mooney, G., Ryan, M., 1993. Agency in health care: getting beyond first principles. Journal of Health Economics 12, 125 – 135. Mwabu, G.M., 1986. Health care decisions at the household level: results of a rural health survey in Kenya. Social Science & Medicine 22, 315 – 319. Mwabu, G.M., Ainsworth, M., Nyamete, A., 1993. Quality of medical care and choice of medical treatment in Kenya: an empirical analysis. Journal of Human Resources 28 (4), 838 – 862. Ndeso-Atanga, S., 2000. Health care quality and the choice of care providers in rural Cameroon. In: Leonard, D.K. (Ed.), Africa’s Changing Markets for Human and Animal Health Services. Macmillan, London. Chapter 5. New York Times, 1999. Vital signs: fertility; help for vasectomy reversals that fail. The New York Times. Robertson, J.A., Schneyer, T.J., 1997. Professional self-regulation and shared-risk programs for in vitro fertilization. Journal of Law, Medicine and Ethics 25 (4), 283 – 291. Sen, A., 1995. The political economy of targeting. In: van de Walle, D., Nead, K. (Eds.), Public Spending and the Poor. Johns Hopkins Univ. Press, Baltimore, MD. Stauga˚rd, F., 1985. Traditional Medicine in Botswana, vol. 1. Ipelegeng Publishers, Gaborone, Botswana. Strauss, J., Thomas, D., 1995. Human resources: empirical modeling of household and family decisions. In: Behrman, J., Srinivasan, T. (Eds.), Handbook of Development Economics. Elsevier, Amsterdam. Strickland, G.T. (Ed.), 1984. Hunters’ Tropical Medicine, 6th ed. W.B. Saunders, Philadelphia. Van der Geest, S., 1992. The cultural acceptability of paying for health care. Social Science & Medicine 34 (6), 667 – 673. Van der Geest, S., Sarkodie, S., 1998. The fake patient: a research experiment in a Ghanian hospital. Social Science & Medicine 47, 1373 – 1381. Werner, D., 1977. Where There is No Doctor; a Village Health Care Handbook. The Hesperian Foundation, Palo Alto, CA.