- Email: [email protected]
Bribery in health care in Uganda
Journal of Health Economics 29 (2010) 699–707
Contents lists available at ScienceDirect
Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase
Bribery in health care in Uganda Jennifer Hunt a,b,∗ a b
McGill University, Canada NBER, United States
a r t i c l e
i n f o
Article history: Received 25 February 2009 Received in revised form 17 March 2010 Accepted 15 June 2010 Available online 25 June 2010 JEL classification: I1 K42 O17
a b s t r a c t I examine the role of household permanent income in determining who bribes and how much they bribe in health care in Uganda. I find that rich patients are more likely than other patients to bribe in public health care: doubling household expenditure increases the bribery probability by 1.2 percentage points compared to a bribery rate of 17%. The income elasticity of the bribe amount is about 0.37. Bribes in the Ugandan public sector appear to be fees-for-service extorted from the richer patients amongst those exempted by government policy from paying the official fees. Bribes in the private sector appear to be flat-rate fees paid by patients who do not pay official fees. I do not find evidence that the public health care sector is able to price discriminate less effectively than public institutions with less competition from the private sector. © 2010 Elsevier B.V. All rights reserved.
Keywords: Corruption Health care Bribery Governance
The empirical literature on corruption has identified consequences of corruption for countries, such as lower growth and foreign direct investment,1 and causes of corruption across countries, such as the legal, political and fiscal systems.2 It has made progress in suggesting remedies for corruption: some papers infer corrupt practices in particular industries, and examine how rule changes or audits affect business practices.3 In this paper, I contribute to a nascent empirical literature that seeks to understand bilateral interactions between public officials and clients as a stepping stone to devising policy.4 I do so by studying bribery in health care in Uganda, with particular emphasis on the role of household permanent income in determining who bribes and how much. Uganda is a low-income country (GNI per capita of US$1500) classified as one of the most corrupt countries in Transparency International’s Corruption Perceptions Index, with an excellent
∗ Correspondence address: Department of Economics, McGill University, 855 Sherbrooke Street West, Montreal, QC H3A 2T7, Canada. Tel.: +1 514 398 6866; fax: +1 514 398 4848. E-mail address: [email protected]. 1 Mauro (1995) and Wei (2000). 2 Fisman and Gatti (2002) and Treisman (2000). 3 Di Tella and Schargrodsky (2003), Ferraz and Finan (2008a,b), Olken (2006, 2007) and Yang (2008a,b). 4 See also Kaufmann and Wei (1999) and Svensson (2003) for firms, and Deininger and Mpuga (2004) and Thompson and Xavier (2002) for individuals. 0167-6296/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhealeco.2010.06.004
source of data on bribes by individuals in the 2002 Second National Integrity Survey.5 In earlier work, Hunt and Laszlo (2009) analyzed bribery mechanisms for samples pooling all institutions in Uganda and Peru. The health sector is worthy of separate study for several reasons. First, mechanisms could differ across institutions, and different mechanisms may require different solutions. Unlike many public institutions, the public health care system has competition from the private sector, which could influence bribery mechanisms. Second, the health sector is one where equitable access, and hence the link between permanent income and bribery, is of particular concern. Third, my data show that 37% of Ugandan bribes are paid in the health sector, due to widespread use of the health system. Fourth, a comparison between bribery in the public and private health care systems may be made for Uganda.6 Finally, the Ugandan data allow a richer set of covariates to be used in the study of health care than could be used with other institutions. Theory suggests that richer clients should be more likely to bribe a public official, due to a higher valuation of their time and
5 Perceptions from Transparency International (2004). Purchasing power parity and Gross National Income from siteresources.worldbank.org/DATASTATISTICS/ Resources/GNIPC.pdf. 6 Corruption is not by definition confined to the public sector: see Gambetta (2002). A payment in the private sector is a bribe if it is paid to an agent to induce behavior contrary to the interest of the principal.
700
J. Hunt / Journal of Health Economics 29 (2010) 699–707
higher demand for services. Richer clients should also pay more conditional on bribing. This stems from the official’s possessing a degree of monopoly power, and hence the ability to price discriminate amongst customers. If such discrimination is observed, it could reflect first-degree price discrimination, or, if the exact service being paid for by the client cannot be observed, third-degree price discrimination (the rich pay more and get more).7 Greater competition between service providers, whether private or public, should reduce the ability of officials to price discriminate, and indeed, under perfect competition bribe amounts should be bid down to zero. Lewis (2006) has proposed that infrequent bribery in health care in certain countries is explained by the presence of private-sector competition in those countries. Clarke and Xu (2004) find that in countries with competition amongst utilities there is less utility bribery than in countries with monopolies. As expected, I find that rich patients are more likely than other patients to bribe in public health care: doubling household expenditure in Uganda increases the bribery probability by 1.2 percentage points compared to a bribery rate of 17%. The income (expenditure) elasticity of the bribe amount is about 0.37 in the public sector: the rich pay more, but pay a smaller share of their expenditure. This elasticity is the same as that for official payments in both the public and the private sector. This could indicate that in all three cases the elasticity is determined by the same combination of fee-for-service, with the rich demanding more expensive services, and price discrimination. I also find that bribes in the public sector are disproportionately paid by the richer patients amongst those not paying official fees. The results, combined with anecdotal evidence, suggest that much public sector bribery represents a facility-level extortion policy to raise revenue from patients exempted from payment by government policy. This suggests that well-intentioned policies to reduce health care fees for the poor may be thwarted by health workers seeking to supplement their own incomes. In addition to undermining the goal of increased access to health care, the effect could be to contribute to a culture of bribery which helps bribery flourish in other institutions. The private-sector results provide an intriguing contrast. The probability of bribing is unrelated to household expenditure, and the income (expenditure) elasticity at 0.15 is much lower than for the public sector. Bribes in the private sector are therefore close to flat-rate fees, and as in the public sector are paid almost exclusively by patients not paying official fees. Bribes paid in the private sector are more effective than those paid in the public sector, with some bribers successfully raising the quality of the facilities and the level of service from good to very good. This suggests that bribes may not represent revenue-raising extortion as they do in the public sector. Although bribery rates and amounts are lower in public health care than in other public institutions, I do not find evidence that the public health care system in Uganda is able to price discriminate less effectively than other public institutions. The results therefore fail to support the hypothesis that competition reduces bribery. 1. Health, health care and corruption in Uganda
Table 1 Health and health care in Uganda.
Life expectancy at birth – males Life expectancy at birth – females Percent of population 15–49 HIV positive Percent of births attended by skilled personnel Doctors per 1000 population Nurses per 1000 population Health spending as % GDP Government spending as % total health spending
Year
Value
2004 2004 2003 2000 2004 2004 2003 2003
48 51 4.1 39.0 0.08 0.61 7.3 30.4
Source: WHO, http://www3.who.int/whosis/core/core select.cfm.
health spending. In 2000 there were 1226 public, 465 NGO and 49 private (non-NGO) health facilities in Uganda.8 Health insurance is essentially non-existent. During the period relevant for my data, health care at public facilities below the hospital level was free, while in public hospitals fees were based on the patient’s ability to pay.9 Although NGO and mission hospitals also make provision for fee exemptions for the poor, Amone et al. (2005) observe that in their sample of Catholic hospitals, only a minority of exempted patients were poor (the others were predominantly hospital and church staff, and teachers and pupils of the Catholic school). Most of these hospitals charged on a fee-for-service basis (each service had an associated fee), with the exception of treatment for tuberculosis and sexually transmitted diseases including HIV/AIDS, which was funded by the government.10 1.2. Corruption in health care As part of the collection of the data, described below, the consulting company commissioned by the government ran focus groups on bribery and availability of public services in 180 villages and towns. Almost every focus group notes that medical attention at public hospitals and health units can only be obtained in exchange for payment despite the official abolition of user fees at health units. They state that patients have to bribe to attract the attention of medical staff and pay for all medical supplies, no matter how small.11 The impression conveyed by the focus groups is less one of individual bad apples within a particular facility than of facility-wide policies to extort bribes. Focus group participants complain that the only drug available at Ugandan health facilities is Panadol (Tylenol). Other drugs must be purchased at pharmacies, drug shops or private practices with connections to the doctor recommending the drug, despite the fact that they should be available free in the public health units.12 Some groups note that the corruption and poor service in the public health sector lead people to use private clinics, despite their cost. Accounts of the health care system in periods when user fees existed describe widespread corruption similar to that deplored by the 2002 focus groups.13 McPake et al. (1999) estimate that 68–77% of revenues from official fees were misappropriated and that 76% of drugs at the facilities they studied disappeared before reaching patients. Given this, it seems highly unlikely that any bribe revenue
1.1. Health and health care I present in Table 1 statistics documenting the poor health outcomes and health care quality in Uganda. Life expectancy at birth is only 48 for men and 51 for women, and there are only 0.08 doctors per thousand population. Private provision of health care is significant, as government health spending represents only 30% of total
7
See Hunt and Laszlo (2009) and Olken and Barron (2009).
8 Uganda Investment Authority: http://www.ugandainvest.com/health.htm. Reinikka and Svensson (forthcoming) outline the post-colonial history of private and public health care in Uganda. 9 Nabynoga et al. (2005). 10 See also Uganda Ministry of Health Online (2000) at http://www.health. go.ug/budget.htm. 11 Jitta et al. (2003) observe that patients routinely bring their own syringes and must pay for the liquids used to mix the injection fluid. 12 The respondents in Jitta et al. (2003) make the same observation. 13 For example, Konde-Lule and Okello (1998).
J. Hunt / Journal of Health Economics 29 (2010) 699–707
701
Table 2 Health unit users in past three months. (1) Bribed
(2) Made official payment
(3) Amount bribe
(4) Amount official payment
(5) Observations
(A) Public All Bribe reported Official payment reported
0.17 – –
0.38 – –
– 6.06 (15) [1.95] –
– – 11.66 (59) [2.71]
4104 697 1550
(B) Private All Bribe reported Official payment reported
0.11 – –
0.83 – –
– 5.26 (10) [2.46] –
– – 12.23 (87) [4.06]
3467 398 2868
Note: Amounts in US dollars. Standard deviations are in parentheses. Medians in square brackets.
is used to fund health facility activities, rather than following the embezzled money into health workers’ pockets. The abolition of most user fees shortly before my study period may simply have led health workers to extract more bribes as a way of maintaining their illicit revenue. Based on a mix of quantitative and qualitative data from a period when user fees were becoming more widespread, McPake et al. (1999) conclude that there is some evidence that introducing official fees reduced bribery.
company to conduct this survey in 2002. All questions are asked of the household head or spouse. The core of the survey has a series of questions on usage, bribery, and service quality posed for each of 21 types of official or institution. These data reveal that 37% of bribes are paid in the health sector, due to households’ frequent need for health care. While the bribery rate in the health sector is half the rate for police, who have the second-largest share of bribes with 19%, 64% of households had used the health sector in the previous six months compared to only 16% using the police. In this paper I focus on the module devoted to bribery in the health sector. A series of questions is asked about the most recent health care visit of a household member in the last three months. Information gathered about this visit includes the type of facility, whether it was public or private, the age of the patient and the
2. Data I use information on the 12,000 household respondents to the Ugandan Second National Integrity Survey, which over-sampled urban areas. The Ugandan government commissioned a consulting Table 3 Health unit non-users and users in past three months. Public users (1) No bribery Household expenditure, monthly, in US$ Urban household Respondent education None Primary 1–4 Primary 5–7 Secondary 1–4 Secondary 5–6 Post-secondary Other
82 (124) [45] 0.37
Private users (2) Bribery 91 (125) [47] 0.37
(3) No bribery 109 (171) [60] 0.44
(5) Non-users (4) Bribery 101 (153) [56] 0.36
68 (128) [34] 0.42
0.11 0.15 0.32 0.28 0.03 0.09 0.03
0.10 0.16 0.35 0.25 0.03 0.08 0.02
0.08 0.13 0.30 0.29 0.06 0.11 0.03
0.10 0.12 0.34 0.26 0.05 0.08 0.05
0.14 0.14 0.30 0.24 0.05 0.11 0.03
36 (12) 19 (18)
35 (11) 19 (17)
34 (11) 17 (17)
35 (12) 17 (16)
35 (13) –
Patient illness Malaria/fever/headache Pregnant/gynecological Intestinal/stomach Respiratory
0.62 0.07 0.04 0.04
0.61 0.09 0.04 0.05
0.70 0.03 0.04 0.05
0.68 0.04 0.04 0.03
– – – –
Patient treatment Single Part of course Other
0.54 0.44 0.02
0.48 0.49 0.02
0.54 0.43 0.02
0.53 0.46 0.01
– – –
Health facility Public Private or other Mission
1 0 0
1 0 0
0 0.91 0.09
0 0.96 0.04
– – –
Health facility Hospital Health unit Dispensary Hospital and health unit/dispensary Pharmacy Clinic Drug store, other
0.53 0.28 0.15 0.03 0.00 0.01 0.00
0.58 0.24 0.10 0.06 0.00 0.01 0
0.14 0.44 0.13 0.02 0.03 0.22 0.01
0.06 0.54 0.15 0.03 0.03 0.17 0.01
– – – – – – –
Respondent age Patient age
Observations
3407
697
3069
Notes: Standard deviations are in parentheses. Medians in square brackets. Additional means are given in Appendix Tables 1 and 2.
398
2516
702
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table 4 Means of quality variables. Public users (1) Poor or very poor Facilities quality Staff quality Medicine availability Treatment quality
0.24 0.17 0.52 0.28
Observations
Private users (2) Good or very good 0.41 0.51 0.25 0.40
(3) Very good
(4) Poor or very poor
0.05 0.07 0.04 0.06
0.09 0.05 0.08 0.07
4104
nature of his or her ailment, whether the visit was part of a longer treatment, the amount of official payments and the amount of unofficial payments (which I call bribes), and qualitative questions about the quality and cost of treatment. The survey also gathers information on household expenditures. I use this as a proxy for permanent income, since households will attempt to smooth expenditures over time. The Appendix A provides further information. Uganda is divided into 56 districts, which in turn are divided into counties and sub-counties. In 206 of 216 sub-counties sampled, at least one respondent had obtained private health care in the previous three months, an indicator that competition in the health care sector is widespread.14 2.1. Samples and descriptive statistics I restrict my analysis to households who report in the health bribery module having used a health care unit in the previous three months (75% of households). I analyze patients in the public sector and the private sector separately (46% of visits are to private facilities). Column 1 of Table 2 shows that the bribery rate for the public sector (panel A) is a large 17%. The private bribery rate is lower, at 11% (panel B). In column 2, I show the share of patients who made official payments: 38% of public patients do so, but almost all private patients (83%) do so. The average and median (non-zero) bribe amounts are US$6.06 and US$1.95 respectively in the public sector, higher than the corresponding private-sector values of US$5.26 and US$2.46 (column 3). Official payments, for those paying, are similar in the public and private sectors, at about US$12 (column 4). For the public sector, the bribery rate of 17% is slightly below the bribery rate for all institutions together (20%) and the bribe amount is also lower compared to the average bribe of $13 (the numbers for all institutions are not reported in the table). This could be a sign that competition from the private sector reduces corruption in public health care. Table 3 contains the means of variables taken principally from the health module for users of the health system (columns 1–4) and non-users (column 5). Patients of the public sector (columns 1 and 2) are slightly poorer than patients of the private sector (columns 3 and 4). Patients of both sectors are better off than non-users (column 5). Public sector patients who did not bribe averaged US$82 in monthly household expenditure, while public sector bribers had higher average expenditure at US$91. Private-sector bribers are not richer than other private-sector patients. In each column, at least 61% of patients suffered from malaria/fever/headache. Slightly more than half of public patients visited a hospital, while this share was low for private patients: 14% for non-bribers and 6% for bribers (though private clinics perform many functions of a hospital). In Table 4, I present users’ ratings of the quality of facilities, the quality of service from staff, the availability of medicine and
14 The sub-counties with no respondents using private health care have unusually small single-digit samples.
(5) Good or very good 0.60 0.69 0.65 0.66
(6) Very good 0.09 0.12 0.14 0.14
3467
the quality of treatment. Respondents have a choice of five ratings, from very poor to very good. The table shows that public users (columns 1–3) are much less satisfied with their experience than private users (columns 4–6), and that the (non)-availability of medicine is the most serious concern of public patients. Additional means for users are in Tables A.1 and A.2. 3. Estimation My first outcome of interest is the probability that household i bribes conditional on using the health system, P(B | U)i . I estimate P(B|U)i = ˇ0 + ˇ1 Wi + ˇ2 Xi + i ,
(1)
where Wi is household expenditure, and Xi are other covariates. My second outcome of interest is the amount of the bribe, Ai . Using the sample of bribes, I estimate log Ai = ˇ3 + ˇ4 Wi + ˇ5 Xi + i .
(2)
I estimate both equations separately for users of the private and public health systems. In order to allow a comparison of how official and unofficial payments work to allocate resources, I also compare the results when the dependent variable refers to official payments, rather than bribes. Because household expenditure is constructed based on relatively few survey questions, I expect some bias towards zero in the estimates of ˇ1 and ˇ4 . Conversely, these coefficients could be biased upward if bribes are effective in obtaining better health care and hence better health, which in turn diminishes income losses from ill health. The detailed controls for the health condition and age of the patient mitigate this concern, although I cannot control for the severity of the condition, and information on illnesses for which treatment is sought does not fully capture health. Furthermore, if low-income patients have more severe illnesses, and more severely ill patients are more likely to bribe, the lack of controls for illness severity means that the estimates of ˇ1 and ˇ4 could be biased down.15 I investigate explicitly whether bribers obtain better health care. To do so, I estimate equations of the form P(Q k = j|U)i = ˇ6k + ˇ7k Bi + ˇ8k Wi + ˇ9k Xi + ik ,
(3)
Qk
where represents the user-reported quality of the k th aspect of the medical care (facilities, service, availability of medicine, treatment). The possible values of j I consider are poor or very poor; good or very good; and very good. The coefficients of interest are the ˇ7k , and as explained in detail and with a formal model in Hunt and Laszlo (2009), they are likely to be subject to endogeneity bias: patients encountering poor service and conditions are likely to consider bribing to improve them. As no suitable instrument for bribing
15 If it is true that unscrupulous people both become rich and bribe, the coefficients will also be biased upwards. However, I doubt that the willingness to pay small bribes for health care is an indicator of the sort of unscrupulousness that might bring business success.
J. Hunt / Journal of Health Economics 29 (2010) 699–707
703
Table 5 Users in past three months – who bribes, who pays? Public sample
Private sample
(1)
(2)
(3)
(4)
(5)
(6)
(A) Bribe Log household expenditure R-Squared
0.015 (3.0) 0.08
0.021 (3.9) 0.10
0.017 (3.2) 0.11
0.001 (0.1) 0.06
0.005 (0.7) 0.07
0.006 (0.8) 0.10
(B) Official payment Log household expenditure R-Squared
0.051 (5.8) 0.08
0.053 (5.6) 0.08
0.043 (4.4) 0.12
0.011 (1.6) 0.12
0.010 (1.4) 0.12
0.010 (1.3) 0.14
Observations Basic covariates Education, job Other non-medical covariates Medical covariates
Yes – – –
4104 Yes Yes – –
Yes Yes Yes Yes
Yes – – –
3467 Yes Yes – –
Yes Yes Yes Yes
Notes: Marginal effects from probits in columns 1–3, linear probability coefficients in columns 4–6. t-Statistics in parentheses, clustered by sub-county. All regressions include 54 district dummies, 13 household size dummies, and an urban dummy (basic covariates). Education is captured by dummies and job refers to dummies for respondent occupation and the main source of household income. The other non-medical covariates are the number of household males and females over 18, respondent sex, age and age squared, and status as head or spouse. The medical covariates are the patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment, and (in columns 4–6) a dummy for a mission facility.
is available, the likely bias must simply be borne in mind when interpreting the results. I estimate Eq. (1) using a probit for the public sample, but for the private sample and Eq. (3) I use linear probability, to avoid the problem of cells lacking any observation whose dependent variable equals one.16 I estimate Eq. (2) using OLS. In all regressions, the standard errors are clustered at the level of the smallest region in the data, the sub-county. Because of the oversampling of urban areas, all specifications include controls for district and urban location. All specifications also include controls for household size dummies. The other Xi covariates are those whose means are in Table 3 and Tables A.1 and A.2: respondent (not necessarily the patient) education and occupation dummies, sex, age, age squared, and status as head or spouse; the number of males and females over age 18 in the household and dummies for the principal source of household income; patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment and (for the private sector) mission facility. I do not include non-users of the health system in any of the regressions. As Table 3 shows, non-users are poorer than users, and may be deterred from seeking treatment by the combination of corruption (particularly in the public sector) and fees (particularly in the private sector). The analysis based on users only will tend to underestimate the share of the costs of corruption borne by the poor. 4. Results 4.1. Probability of bribing Panel A of Table 5 presents results for the probability of bribing conditional on using the health care system. Columns 1–3 contain marginal effects from probits for patients in the public sector. The preferred marginal effect of 0.017, with the full covariates including medical covariates in column 3, implies that a doubling of expenditure (an increase of two thirds of a standard deviation) increases the bribery probability by 1.2 percentage points. This compares to the overall bribery probability for public sector of 17%. The effects are similar to point estimates of 0.011–0.018 for all officials pooled in Hunt and Laszlo (2009). There is again no evidence that competition in health care reduces the role of expenditure in determining
16
A probit cannot be estimated for a sample containing such cells.
bribery. In contrast to the public sector, there is no significant effect of expenditure on the probability of bribing in the private sector, in the linear probability regressions of columns 4–6 (nor in unreported probit counterparts with fewer covariates). Hunt and Laszlo (2009), considering all institutions, reported that lower-expenditure Ugandan users were more likely than richer users to pay what they considered to be official payments, but receive no receipt. If these are unwitting bribes, it means that their omission leads to the role of expenditure in paying a bribe to be overstated. These unwitting bribes cannot be identified in the health module, but I have examined their relation to household expenditure in the main bribery module for health care alone. The unreported results show a positive but insignificant relation between expenditure and the payment of unwitting bribes, suggesting that their omission does not bias the health care bribery results as much as was the case for all officials pooled in Hunt and Laszlo (2009).17 I investigate the link between expenditure and the probability of an official payment in panel B of Table 5. In the public sector, the marginal effect of 0.043 in column 3, the preferred specification, implies that a doubling of consumption increases the probability of making an official payment by 3.0 percentage points, compared to a mean payment rate of 38%. The absolute effect is larger than for bribes, while the percent effect is similar. The positive effect is not a surprise, as there are exemptions from payment for the poor, and the rich may demand more of the services which are not free. By comparison, Hunt and Laszlo (2009) found a smaller marginal effect of 0.022 for all officials pooled.18 There is no significant link between expenditure and official payment probability in the private sector, where most people make official payments (83%), although the point estimate is 0.01 (panel B, columns 4–6). This is consistent with the observation of Amone et al. (2005) that patients exempted at Catholic hospitals are typically not poor. The patients paying bribes and the patients making official payments are to a large degree different people, as Table 6 indicates. In the public sector, while 22% of those making no official payment bribe, only 8% of those making an official payment bribe, and 81% of bribes are paid by patients not making an official payment.
17 This check is not perfect, as in the main bribery module the private and public sectors cannot be distinguished. 18 For official payments for which a receipt was given. The marginal effect including receiptless payments would be smaller.
704
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table 6 Bribes and official payments among health care users (%). Public No bribe
ing official payments, the procedure differs from that in the public sector, as the richer patients are not those paying the bribes.
Private Bribe
All
No bribe
Bribe
All
No official payment Official payment
78 92
22 8
100 100
45 98
55 2
100 100
No official payment Official payment
58 42
81 19
– –
9 91
83 17
– –
100
100
–
100
100
–
All
The contrast is even greater in the private sector: fully 55% of the small number of patients making no official payment bribe, while only 2% of patients who make an official payment bribe, and 83% of bribes are paid by patients not making an official payment. The strong negative correlation between paying bribes and making official payments holds up for both sectors in regressions with the controls used in Table 5. When the bribery regressions of panel A in Table 5 are repeated for samples of patients not paying official fees, the marginal effects of expenditure are larger for the public sector (though not significantly so), and remain insignificant for the private sector (these results are not reported). It is possible that people bribe to avoid official payments: one focus group mentioned this, although the bribe in question was paid to the local government official issuing the exemption certificate. However, most of the public sector bribes mentioned in the focus groups had a different purpose, which suggests that the causality may be the opposite: health officials may attempt to extract bribes from the richer patients who need not pay officially (because of the type of facility or because of their low income). It is possible that local private hospitals likewise try to circumvent exemption policies imposed on them by their national organization, such as the church, or the government, which at least in theory pays for the private treatment of patients with respiratory and sexually transmitted diseases. Private patients with these diseases do not appear more likely to bribe in the regressions, suggesting the purpose of the bribes is not to raise money from exempt patients (however, these patients appear more likely than malaria patients to pay official fees, raising the possibility that they are unaware their care should be free and are paying unwitting bribes). Bribes could also be paid to avoid official payments, or simply to obtain better health care, for which I provide evidence below. Whatever the reason for the bribes by private patients not pay-
4.2. Amount of bribe In Table 7, panel A, I examine the determinants of the amount of the (log) bribe for the sample of bribes. The preferred specifications are those with the maximum covariates including medical covariates: in the public sector (column 3), the income (expenditure) elasticity is 0.37, while is only 0.15 in the private sector (column 6). The rich pay more than the poor in both sectors, but pay a lower share of their expenditure, since the elasticity is less than one. These elasticities compare with elasticities of 0.25–0.33 for pooled officials in Hunt and Laszlo (2009): once again, competition from the private sector does not appear to reduce price discrimination in the public sector. If the panel A regressions are repeated for patients making no official payments, the (unreported) elasticities are similar. Panel B shows that the income (expenditure) elasticity for official payments is essentially the same as for bribes in the public sector, at 0.36, while the private-sector elasticity for official payments is much higher than for private bribes, at 0.38. Public sector bribes, public sector official fees, and private-sector official fees are extracted from payers in the same way based on their expenditure. Only private-sector bribes are extracted differently, and are closer to a flat-rate fee. 4.3. Discussion Together, Tables 5–7 suggest the following characterization of payments for health care in Uganda. In the public sector, richer patients are considerably more likely to make official payments, presumably in part because of official policies charging patients according to ability to pay, and in part because richer patients demand more of services that are not free. However, since some rich patients use free services, and some poorer patients not making official payments are willing to pay non-zero amounts, health facilities are able to induce these patients, particularly the richer ones, to make unofficial payments (bribes). The income (expenditure) elasticity for the bribe amounts is the same as for official payment amounts. The private-sector differs in that almost everyone makes an official payment, exemptions are not based on household expenditure, and bribes appear to be flat fees assessed on the exempted.
Table 7 Bribers – log amount of bribe, log amount of official payment. Public sample
Private sample
(1)
(2)
(3)
(4)
(5)
(6)
(A) Bribe Log household expenditure R-Squared Observations
0.427 (7.3) 0.24
0.431 (6.4) 0.27 697
0.373 (5.4) 0.37
0.257 (3.5) 0.30
0.251 (3.0) 0.37 398
0.153 (2.0) 0.51
(B) Official payment Log household expenditure R-Squared Observations
0.452 (9.7) 0.26
0.464 (9.3) 0.28 1550
0.357 (8.1) 0.40
0.433 (16.7) 0.23
0.440 (15.9) 0.25 2868
0.376 (13.9) 0.36
Basic covariates Education, job Other non-medical covariates Medical covariates
Yes – – –
Yes Yes – –
Yes Yes Yes Yes
Yes – – –
Yes Yes – –
Yes Yes Yes Yes
Notes: OLS coefficients with t-statistics in parentheses, clustered by sub-county. All regressions include 54 district dummies, 13 household size dummies, and an urban dummy. Education is captured by dummies and job refers to dummies for respondent occupation and the main source of household income. The other non-medical covariates are the number of household males and females over 18, respondent sex, age and age squared, and status as head or spouse. The medical covariates are the patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment, and (in columns 4–6) a dummy for a mission facility.
J. Hunt / Journal of Health Economics 29 (2010) 699–707
705
Table 8 Effectiveness of bribery. Public users
Facilities quality Staff service quality Medicine availability Treatment quality
Private users
(1) Poor or very poor
(2) Good or very good
(3) Very good
(4) Poor or very poor
(5) Good or very good
(6) Very good
0.080 (4.0) 0.064 (3.3) 0.156 (7.4) 0.120 (5.1)
−0.123 (−6.3) −0.120 (−5.5) −0.090 (−5.2) −0.104 (−4.4)
−0.029 (−4.1) −0.026 (−2.5) −0.017 (−2.3) −0.021 (−2.4)
0.025 (1.4) 0.031 (2.2) 0.055 (2.9) 0.041 (2.3)
−0.051 (−1.8) −0.035 (−1.3) −0.101 (−3.3) −0.068 (−2.5)
0.033 (1.9) 0.034 (1.7) 0.001 (0.1) 0.004 (0.2)
Observations
4104
3467
Notes: Coefficients on dummy for bribe payment from linear probability regressions with all covariates used in Table 7, including medical covariates. t-Statistics in parentheses, clustered by sub-county. The fifth quality category is “satisfactory”.
However, the income elasticity for official payment amounts is the same as for official payments and bribes in the public sector. This could indicate that in all three cases the elasticity is determined by the same combination of fee-for-service (and the rich demanding more expensive services) and price discrimination. Anecdotal evidence indicating that public sector bribes often consist of paying for supplies is consistent with an important role for fee-for-service bribes. If private-sector health facilities operate in more competitive environments than public facilities, this would be an explanation for less frequent and lower-value private-sector bribes, as well as the inability of the private facilities to price discriminate in bribes. However, the ability to price discriminate also depends on how well staff can judge patients’ ability to pay, which is in turn influenced by staff turnover, the distance patients travel to the hospital and procedures for assessing patient eligibility for official fee exemptions. A priori it is unclear whether private or public staff know their patients better. 4.4. Does bribery improve health care? In Table 8, I investigate what bribers obtain in return for their bribe. I present the coefficients on the bribe dummy from linear probability regressions with the full set of covariates. Columns 1–3 for the public sector show that bribery is associated with worse health care. For example, column 1 shows that bribers are 8.0 percentage points more likely to report that the facilities were poor or very poor, column 2 that bribers are 12.3 percentage points less likely to report that the facilities were good or very good, and column 3 that bribers are 2.9 percentage points less likely to report that the facilities were very good. Columns 4 and 5 for the privatesector paint a similar picture, and unreported ordered probits also indicate that bribery is associated with worse healthcare. These results presumably indicate that bad conditions push people to bribe to try to improve them, and it is difficult to tell whether they are successful in doing so. However, the final column of results for the private sector is different. Bribery raises the probability of reporting very good facilities or service, though these results are not quite statistically significant, and has no effect on the probability of reporting very good medicine availability or treatment. These results suggest that bribery is most effective for private patients who can make already good health care very good. 5. Conclusions In the public health care system in Uganda, richer patients are more likely to bribe. A doubling of household expenditure increases the bribery probability by 1.2 percentage points, compared to a bribery rate of 17%. The income (expenditure) elasticity of the bribe amount is 0.37. The link between bribery and consumption or expenditure is similar to or higher than the link for all institu-
tions pooled, despite the presence of competition from the private sector which might be expected to weaken this link. Additional results combined with anecdotal evidence suggest that the public sector health staff extort bribes particularly from the richer among the patients who officially need not pay. The amount of the bribe is then influenced by household expenditure in the same way as official charges in both the public and private sector, probably through a combination of price discrimination and fee-for-service. The results suggest that well-intentioned policies to reduce health care payments for the poor may be thwarted by health workers seeking to supplement their own incomes. In addition to undermining the goal of increased access to health care, the effect could be to contribute to a culture of bribery which helps bribery flourish in other institutions. Although private-sector bribes are even more strongly associated with not making official payments, the mechanism works differently. Those exempted in the private sector have similar household expenditures to those not exempted, and appear to pay close to a flat-rate bribe. Private-sector bribes are more effective in improving health care than public sector bribes. Acknowledgements I am grateful to Vincent Chandler for excellent research assistance, to the Instituto Nacional de Estadística e Información for the Peruvian data, and to Niels Ulrich, Paul Mpuga, Barbara Kasura Magezi Ndamira and K2 Consulting for the Ugandan data. I thank Lorena Alcázar, Leah Brooks, Janice Seinfeld, Dean Yang and seminar participants at McGill, UC Davis, UC Berkeley and the NBER Inter-American Seminar for comments on an earlier version including Peruvian data. I am also affiliated with the CEPR, IZA and DIW-Berlin, and I acknowledge the Social Science and Humanities Research Council of Canada for financial support. Appendix A. Data appendix A.1. Data The 2002 Second National Integrity Survey was conducted in 55 of 56 districts of Uganda. The subsequent non-random sampling of sub-counties led to the sub-county of the district headquarters always being chosen, which means that urban areas are over-sampled. The district’s sub-counties were divided into three categories based on availability of government services and infrastructure, and 20% of sub-counties in each category were randomly chosen. Within each of these sub-counties, the local council 1 areas were similarly divided into three categories, and one local council 1 area per category was chosen randomly. The selection of which households to interview within these local council 1 areas did not appear to be random, as it appeared to involve choosing households near the residence or office of the local council 1 chairperson. No weights are provided with the survey (Table A.1).
706
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table A.1 Means of Ugandan household main income source and respondent occupation. Public sample No bribery Farming – cash crops Farming – foods crops Farming – livestock Manufacturing, crafts, repair Trade – petty Trade – retail/shop/stall Trade – wholesale, crop buying Government – salaried or wage Private – salaried or wage Stipends from relatives Casual work Other Farmer – mainly crops Farmer – mainly livestock Trader Civil servant/armed forces Teacher Professional in private practice (doctor/lawyer) Craftsperson (carpenter/mechanic, etc.) Casual laborer Housewife Student Tailor/builder Bodaboda or taxi driver Repair and service jobs Unemployed Retired Observations
0.16 0.25 0.02 0.06 0.09 0.10 0.02 0.12 0.05 0.02 0.08 0.05 0.37 0.02 0.18 0.06 0.05 0.02 0.06 0.07 0.10 0.01 0.01 0.01 0.04 0.01 0.00 3407
Private sample Bribery
No bribery
0.15 0.25 0.01 0.09 0.07 0.11 0.03 0.10 0.04 0.01 0.11 0.03 0.38 0.01 0.19 0.04 0.04 0.02 0.07 0.08 0.09 0.00 0.01 0.01 0.04 0.01 0.00
Bribery
0.12 0.18 0.02 0.07 0.09 0.15 0.03 0.10 0.08 0.01 0.07 0.06 0.26 0.02 0.25 0.05 0.06 0.03 0.07 0.07 0.10 0.01 0.01 0.02 0.05 0.00 0.00
697
0.16 0.19 0.02 0.09 0.09 0.15 0.02 0.08 0.06 0.00 0.10 0.05 0.33 0.02 0.21 0.03 0.04 0.03 0.08 0.08 0.10 0.01 0.01 0.02 0.04 0.01 0
3069
398
Note: Households using the health care system in the previous three months.
Household expenditure is based on the question “What did you spend on the following items for your household/family in the past one week?” (food, household durables e.g. saucepans, soap/cleaning materials, entertainment/cigarettes/drinks, travel, other), and on the question “What did you spend on the following items for your household/family in the past one month?” (health/medicines/doctors, education, clothes). It is not possible to distinguish between zeroes and missing values in the components of expenditure, so I simply assign zeroes to all missing values and sum the nine components with appropriate weights. In the health module, there are some valid responses for households who had not used the health care system in the previous three months: I drop these observations. Also, some households gave information on more than one health care visit in the previous three months (contrary to the survey instructions): I retain only one visit per household. I focus on the special module on bribery in health care. The 22 agencies listed in the main part of the survey are: local primary school, Department of Education, health unit, police, traffic police, local council 1, local council 3, Agriculture Department, Veterinary Department, Fisheries Department, Forestry Department, Department of Cooperatives, Public Service (pensions), Water Department, Land Board, Magistrates Court, Ugandan People’s Defence Force, Local Defence Force, Uganda Revenue Authority (licencing), Uganda Revenue Authority (customs, anti-smuggling), Uganda Electricity Board and “other”. In this part of the survey, respondents are invited to report the amount of unofficial payments, the amount of official payments, and whether they received a receipt for their payments. A.2. Are the bribery rates plausible? There are a number of factors that make it likely that any survey will underestimate the incidence of bribery. People may be
reluctant to admit to illegal behavior, and ignorant clients may pay bribes unwittingly. I do not believe my underestimation to be severe, however (Table A.2). In high-bribery countries, bribery is viewed as inevitable and the fault of the system. The stigma attached to admitting paying a bribe is therefore low. Also, anti-corruption drives typically target officials, so the fear of prosecution from such an admission should be low. The high reported bribery rates in Uganda are consistent with this. The Ugandan survey gives respondents the option of giving their name along with the responses. I have checked the correlation
Table A.2 Means of Ugandan patient illnesses. Public sample No bribery Malaria/fever/headache Diarrhea Injury Pregnant/gynecological Immunization STDs Intestinal/stomach Heart Accident Dental Ear/nose/throat/eye Skin Backache Bone Other diseases Respiratory Tetanus, measles General check up Observations
0.62 0.03 0.03 0.07 0.03 0.01 0.04 0.02 0.00 0.01 0.02 0.01 0.00 0.00 0.03 0.04 0.02 0.00 3407
Private sample Bribery 0.61 0.02 0.04 0.09 0.00 0.01 0.04 0.01 0.00 0.02 0.01 0.01 0.00 0.00 0.03 0.05 0.03 0.00 697
No bribery 0.70 0.03 0.03 0.03 0.00 0.00 0.04 0.02 0.00 0.01 0.01 0.01 0.00 0.00 0.03 0.05 0.03 0.00 3069
Bribery 0.68 0.05 0.05 0.04 0 0.01 0.04 0.02 0.00 0.01 0.01 0.01 0 0.00 0.04 0.03 0.01 0 398
J. Hunt / Journal of Health Economics 29 (2010) 699–707
between bribing and providing a name. Neither providing a first name only nor both a first and last name is associated with a lower probability of bribing conditional on using an official than giving no name (conditional on other covariates), although oddly providing one name and one initial is associated with a higher bribery probability than giving no name. I are therefore not concerned that reluctance to report is a major issue. References Amone, J., Asio, S., Cattaneo, A., Kweyatulira, A.K., Macaluso, A., Maciocco, G., Mukokoma, M., Ronfani, L., Santini, S., 2005. User fees in private non-for-profit hospitals in Uganda: a survey and intervention for equity. International Journal for Equity in Health 4 (6). Clarke, G.R.G., Xu, L.C., 2004. Privatization, competition, and corruption: how characteristics of bribe takers and payers affect bribe payments to utilities. Journal of Public Economics 8 (9–10), 2067–2097. Deininger, K., Mpuga, P., 2004. Does greater accountability improve the quality of public service delivery? Evidence from Uganda. World Development 33 (1), 171–191. Di Tella, R., Schargrodsky, E., 2003. The role of wages and auditing during a crackdown on corruption in the city of Buenos Aires. Journal of Law and Economics 1, 269–292. Ferraz, C., Finan, F., 2008a. Electoral Accountability and Corruption: Evidence from the Audits of Local Governments. UC Berkeley Working Paper. Ferraz, C., Finan, F., 2008b. Exposing corrupt politicians: the effect of Brazil’s randomized anti-corruption program on electoral outcomes. Quarterly Journal of Economics 123 (2), 703–745. Fisman, R., Gatti, R., 2002. Decentralization and corruption: evidence across countries. Journal of Public Economics 83, 325–345. Gambetta, D., 2002. Corruption: an analytical map. In: Kotkin, S., Sajo, A. (Eds.), The Political Corruption of Transition: A Sceptic’s Handbook. Central European University Press, Budapest. Hunt, J., Laszlo, S., 2009. Is Bribery Really Regressive? Bribery’s Costs, Benefits and Mechanisms. McGill University Working Paper. Jitta, J., Whyte, S.R., Nshakira, N., 2003. The availability of drugs: what does it mean in Ugandan primary care. Health Policy 65, 167–179.
707
Kaufmann, D., Wei, S.-J., 1999. Does ‘Grease Money’ Speed Up the Wheels of Commerce? NBER Working Paper 7093. Konde-Lule, J.K., Okello, D., 1998. User Fees in Government Health Units in Uganda: Implementation, Impact and Scope. Partnerships for Health Reform Small Applied Research Paper No. 2. Lewis, M., 2006. Governance and Corruption in Public Health Care Systems. Center for Global Development Working Paper Number 78. Mauro, P., 1995. Corruption and growth. Quarterly Journal of Economics 110 (3), 681–712. McPake, B., Asiimwe, D., Mwesigye, F., Ofumbi, M., Ortenblad, L., Streefland, P., Turinde, A., 1999. Informal economic activities of public health workers in Uganda: implications for quality and accessibility of care. Social Science and Medicine 49, 849–865. Nabynoga, J., Desmet, M., Karamagi, H., Kadama, P.Y., Omaswa, F.G., Walker, O., 2005. Abolition of cost-sharing is pro-poor: evidence from Uganda. Health Policy and Planning 20 (2), 100–108. Olken, B.A., 2007. Monitoring corruption: evidence from a field experiment in Indonesia. Journal of Political Economy 115 (2), 200–249. Olken, B.A., Barron, P., 2009. The simple economics of extortion: evidence from trucking in Aceh. Journal of Political Economy 93 (7–8), 950–964. Olken, B.A., 2006. corruption and the costs of redistribution: micro evidence from Indonesia. Journal of Public Economics 90 (4–5), 853–870. Reinikka, R., Svensson, J., forthcoming. Working for god? Journal of the European Economic Association. Svensson, J., 2003. Who must pay bribes and how much? Evidence from a cross section of firms. Quarterly Journal of Economics 118, 207–230. Thompson, R., Xavier, A., 2002. Unofficial Payments for Acute State Hospital Care in Kazakhstan. A Model of Physician Behaviour with Price Discrimination and Vertical Service Differentiation. LICOS Working Paper 124. Transparency International, 2004. Corruption continues to deprive societies around the world. www.icgg.org/downloads/ (retrieved 5 July 2010). Treisman, D., 2000. The causes of corruption: a cross-national study. Journal of Public Economics 76 (3), 399–457. Wei, S.-J., 2000. How taxing is corruption on international investors? Review of Economics and Statistics 82 (1), 1–11. Yang, D., 2008a. Integrity for hire: an analysis of a widespread program for combating customs corruption. Journal of Law and Economics 51 (1), 25–57. Yang, D., 2008b. Can enforcement backfire? Crime displacement in the context of customs reform in the Philippines. Review of Economics and Statistics 90 (1), 1–14.
Contents lists available at ScienceDirect
Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase
Bribery in health care in Uganda Jennifer Hunt a,b,∗ a b
McGill University, Canada NBER, United States
a r t i c l e
i n f o
Article history: Received 25 February 2009 Received in revised form 17 March 2010 Accepted 15 June 2010 Available online 25 June 2010 JEL classification: I1 K42 O17
a b s t r a c t I examine the role of household permanent income in determining who bribes and how much they bribe in health care in Uganda. I find that rich patients are more likely than other patients to bribe in public health care: doubling household expenditure increases the bribery probability by 1.2 percentage points compared to a bribery rate of 17%. The income elasticity of the bribe amount is about 0.37. Bribes in the Ugandan public sector appear to be fees-for-service extorted from the richer patients amongst those exempted by government policy from paying the official fees. Bribes in the private sector appear to be flat-rate fees paid by patients who do not pay official fees. I do not find evidence that the public health care sector is able to price discriminate less effectively than public institutions with less competition from the private sector. © 2010 Elsevier B.V. All rights reserved.
Keywords: Corruption Health care Bribery Governance
The empirical literature on corruption has identified consequences of corruption for countries, such as lower growth and foreign direct investment,1 and causes of corruption across countries, such as the legal, political and fiscal systems.2 It has made progress in suggesting remedies for corruption: some papers infer corrupt practices in particular industries, and examine how rule changes or audits affect business practices.3 In this paper, I contribute to a nascent empirical literature that seeks to understand bilateral interactions between public officials and clients as a stepping stone to devising policy.4 I do so by studying bribery in health care in Uganda, with particular emphasis on the role of household permanent income in determining who bribes and how much. Uganda is a low-income country (GNI per capita of US$1500) classified as one of the most corrupt countries in Transparency International’s Corruption Perceptions Index, with an excellent
∗ Correspondence address: Department of Economics, McGill University, 855 Sherbrooke Street West, Montreal, QC H3A 2T7, Canada. Tel.: +1 514 398 6866; fax: +1 514 398 4848. E-mail address: [email protected]. 1 Mauro (1995) and Wei (2000). 2 Fisman and Gatti (2002) and Treisman (2000). 3 Di Tella and Schargrodsky (2003), Ferraz and Finan (2008a,b), Olken (2006, 2007) and Yang (2008a,b). 4 See also Kaufmann and Wei (1999) and Svensson (2003) for firms, and Deininger and Mpuga (2004) and Thompson and Xavier (2002) for individuals. 0167-6296/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhealeco.2010.06.004
source of data on bribes by individuals in the 2002 Second National Integrity Survey.5 In earlier work, Hunt and Laszlo (2009) analyzed bribery mechanisms for samples pooling all institutions in Uganda and Peru. The health sector is worthy of separate study for several reasons. First, mechanisms could differ across institutions, and different mechanisms may require different solutions. Unlike many public institutions, the public health care system has competition from the private sector, which could influence bribery mechanisms. Second, the health sector is one where equitable access, and hence the link between permanent income and bribery, is of particular concern. Third, my data show that 37% of Ugandan bribes are paid in the health sector, due to widespread use of the health system. Fourth, a comparison between bribery in the public and private health care systems may be made for Uganda.6 Finally, the Ugandan data allow a richer set of covariates to be used in the study of health care than could be used with other institutions. Theory suggests that richer clients should be more likely to bribe a public official, due to a higher valuation of their time and
5 Perceptions from Transparency International (2004). Purchasing power parity and Gross National Income from siteresources.worldbank.org/DATASTATISTICS/ Resources/GNIPC.pdf. 6 Corruption is not by definition confined to the public sector: see Gambetta (2002). A payment in the private sector is a bribe if it is paid to an agent to induce behavior contrary to the interest of the principal.
700
J. Hunt / Journal of Health Economics 29 (2010) 699–707
higher demand for services. Richer clients should also pay more conditional on bribing. This stems from the official’s possessing a degree of monopoly power, and hence the ability to price discriminate amongst customers. If such discrimination is observed, it could reflect first-degree price discrimination, or, if the exact service being paid for by the client cannot be observed, third-degree price discrimination (the rich pay more and get more).7 Greater competition between service providers, whether private or public, should reduce the ability of officials to price discriminate, and indeed, under perfect competition bribe amounts should be bid down to zero. Lewis (2006) has proposed that infrequent bribery in health care in certain countries is explained by the presence of private-sector competition in those countries. Clarke and Xu (2004) find that in countries with competition amongst utilities there is less utility bribery than in countries with monopolies. As expected, I find that rich patients are more likely than other patients to bribe in public health care: doubling household expenditure in Uganda increases the bribery probability by 1.2 percentage points compared to a bribery rate of 17%. The income (expenditure) elasticity of the bribe amount is about 0.37 in the public sector: the rich pay more, but pay a smaller share of their expenditure. This elasticity is the same as that for official payments in both the public and the private sector. This could indicate that in all three cases the elasticity is determined by the same combination of fee-for-service, with the rich demanding more expensive services, and price discrimination. I also find that bribes in the public sector are disproportionately paid by the richer patients amongst those not paying official fees. The results, combined with anecdotal evidence, suggest that much public sector bribery represents a facility-level extortion policy to raise revenue from patients exempted from payment by government policy. This suggests that well-intentioned policies to reduce health care fees for the poor may be thwarted by health workers seeking to supplement their own incomes. In addition to undermining the goal of increased access to health care, the effect could be to contribute to a culture of bribery which helps bribery flourish in other institutions. The private-sector results provide an intriguing contrast. The probability of bribing is unrelated to household expenditure, and the income (expenditure) elasticity at 0.15 is much lower than for the public sector. Bribes in the private sector are therefore close to flat-rate fees, and as in the public sector are paid almost exclusively by patients not paying official fees. Bribes paid in the private sector are more effective than those paid in the public sector, with some bribers successfully raising the quality of the facilities and the level of service from good to very good. This suggests that bribes may not represent revenue-raising extortion as they do in the public sector. Although bribery rates and amounts are lower in public health care than in other public institutions, I do not find evidence that the public health care system in Uganda is able to price discriminate less effectively than other public institutions. The results therefore fail to support the hypothesis that competition reduces bribery. 1. Health, health care and corruption in Uganda
Table 1 Health and health care in Uganda.
Life expectancy at birth – males Life expectancy at birth – females Percent of population 15–49 HIV positive Percent of births attended by skilled personnel Doctors per 1000 population Nurses per 1000 population Health spending as % GDP Government spending as % total health spending
Year
Value
2004 2004 2003 2000 2004 2004 2003 2003
48 51 4.1 39.0 0.08 0.61 7.3 30.4
Source: WHO, http://www3.who.int/whosis/core/core select.cfm.
health spending. In 2000 there were 1226 public, 465 NGO and 49 private (non-NGO) health facilities in Uganda.8 Health insurance is essentially non-existent. During the period relevant for my data, health care at public facilities below the hospital level was free, while in public hospitals fees were based on the patient’s ability to pay.9 Although NGO and mission hospitals also make provision for fee exemptions for the poor, Amone et al. (2005) observe that in their sample of Catholic hospitals, only a minority of exempted patients were poor (the others were predominantly hospital and church staff, and teachers and pupils of the Catholic school). Most of these hospitals charged on a fee-for-service basis (each service had an associated fee), with the exception of treatment for tuberculosis and sexually transmitted diseases including HIV/AIDS, which was funded by the government.10 1.2. Corruption in health care As part of the collection of the data, described below, the consulting company commissioned by the government ran focus groups on bribery and availability of public services in 180 villages and towns. Almost every focus group notes that medical attention at public hospitals and health units can only be obtained in exchange for payment despite the official abolition of user fees at health units. They state that patients have to bribe to attract the attention of medical staff and pay for all medical supplies, no matter how small.11 The impression conveyed by the focus groups is less one of individual bad apples within a particular facility than of facility-wide policies to extort bribes. Focus group participants complain that the only drug available at Ugandan health facilities is Panadol (Tylenol). Other drugs must be purchased at pharmacies, drug shops or private practices with connections to the doctor recommending the drug, despite the fact that they should be available free in the public health units.12 Some groups note that the corruption and poor service in the public health sector lead people to use private clinics, despite their cost. Accounts of the health care system in periods when user fees existed describe widespread corruption similar to that deplored by the 2002 focus groups.13 McPake et al. (1999) estimate that 68–77% of revenues from official fees were misappropriated and that 76% of drugs at the facilities they studied disappeared before reaching patients. Given this, it seems highly unlikely that any bribe revenue
1.1. Health and health care I present in Table 1 statistics documenting the poor health outcomes and health care quality in Uganda. Life expectancy at birth is only 48 for men and 51 for women, and there are only 0.08 doctors per thousand population. Private provision of health care is significant, as government health spending represents only 30% of total
7
See Hunt and Laszlo (2009) and Olken and Barron (2009).
8 Uganda Investment Authority: http://www.ugandainvest.com/health.htm. Reinikka and Svensson (forthcoming) outline the post-colonial history of private and public health care in Uganda. 9 Nabynoga et al. (2005). 10 See also Uganda Ministry of Health Online (2000) at http://www.health. go.ug/budget.htm. 11 Jitta et al. (2003) observe that patients routinely bring their own syringes and must pay for the liquids used to mix the injection fluid. 12 The respondents in Jitta et al. (2003) make the same observation. 13 For example, Konde-Lule and Okello (1998).
J. Hunt / Journal of Health Economics 29 (2010) 699–707
701
Table 2 Health unit users in past three months. (1) Bribed
(2) Made official payment
(3) Amount bribe
(4) Amount official payment
(5) Observations
(A) Public All Bribe reported Official payment reported
0.17 – –
0.38 – –
– 6.06 (15) [1.95] –
– – 11.66 (59) [2.71]
4104 697 1550
(B) Private All Bribe reported Official payment reported
0.11 – –
0.83 – –
– 5.26 (10) [2.46] –
– – 12.23 (87) [4.06]
3467 398 2868
Note: Amounts in US dollars. Standard deviations are in parentheses. Medians in square brackets.
is used to fund health facility activities, rather than following the embezzled money into health workers’ pockets. The abolition of most user fees shortly before my study period may simply have led health workers to extract more bribes as a way of maintaining their illicit revenue. Based on a mix of quantitative and qualitative data from a period when user fees were becoming more widespread, McPake et al. (1999) conclude that there is some evidence that introducing official fees reduced bribery.
company to conduct this survey in 2002. All questions are asked of the household head or spouse. The core of the survey has a series of questions on usage, bribery, and service quality posed for each of 21 types of official or institution. These data reveal that 37% of bribes are paid in the health sector, due to households’ frequent need for health care. While the bribery rate in the health sector is half the rate for police, who have the second-largest share of bribes with 19%, 64% of households had used the health sector in the previous six months compared to only 16% using the police. In this paper I focus on the module devoted to bribery in the health sector. A series of questions is asked about the most recent health care visit of a household member in the last three months. Information gathered about this visit includes the type of facility, whether it was public or private, the age of the patient and the
2. Data I use information on the 12,000 household respondents to the Ugandan Second National Integrity Survey, which over-sampled urban areas. The Ugandan government commissioned a consulting Table 3 Health unit non-users and users in past three months. Public users (1) No bribery Household expenditure, monthly, in US$ Urban household Respondent education None Primary 1–4 Primary 5–7 Secondary 1–4 Secondary 5–6 Post-secondary Other
82 (124) [45] 0.37
Private users (2) Bribery 91 (125) [47] 0.37
(3) No bribery 109 (171) [60] 0.44
(5) Non-users (4) Bribery 101 (153) [56] 0.36
68 (128) [34] 0.42
0.11 0.15 0.32 0.28 0.03 0.09 0.03
0.10 0.16 0.35 0.25 0.03 0.08 0.02
0.08 0.13 0.30 0.29 0.06 0.11 0.03
0.10 0.12 0.34 0.26 0.05 0.08 0.05
0.14 0.14 0.30 0.24 0.05 0.11 0.03
36 (12) 19 (18)
35 (11) 19 (17)
34 (11) 17 (17)
35 (12) 17 (16)
35 (13) –
Patient illness Malaria/fever/headache Pregnant/gynecological Intestinal/stomach Respiratory
0.62 0.07 0.04 0.04
0.61 0.09 0.04 0.05
0.70 0.03 0.04 0.05
0.68 0.04 0.04 0.03
– – – –
Patient treatment Single Part of course Other
0.54 0.44 0.02
0.48 0.49 0.02
0.54 0.43 0.02
0.53 0.46 0.01
– – –
Health facility Public Private or other Mission
1 0 0
1 0 0
0 0.91 0.09
0 0.96 0.04
– – –
Health facility Hospital Health unit Dispensary Hospital and health unit/dispensary Pharmacy Clinic Drug store, other
0.53 0.28 0.15 0.03 0.00 0.01 0.00
0.58 0.24 0.10 0.06 0.00 0.01 0
0.14 0.44 0.13 0.02 0.03 0.22 0.01
0.06 0.54 0.15 0.03 0.03 0.17 0.01
– – – – – – –
Respondent age Patient age
Observations
3407
697
3069
Notes: Standard deviations are in parentheses. Medians in square brackets. Additional means are given in Appendix Tables 1 and 2.
398
2516
702
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table 4 Means of quality variables. Public users (1) Poor or very poor Facilities quality Staff quality Medicine availability Treatment quality
0.24 0.17 0.52 0.28
Observations
Private users (2) Good or very good 0.41 0.51 0.25 0.40
(3) Very good
(4) Poor or very poor
0.05 0.07 0.04 0.06
0.09 0.05 0.08 0.07
4104
nature of his or her ailment, whether the visit was part of a longer treatment, the amount of official payments and the amount of unofficial payments (which I call bribes), and qualitative questions about the quality and cost of treatment. The survey also gathers information on household expenditures. I use this as a proxy for permanent income, since households will attempt to smooth expenditures over time. The Appendix A provides further information. Uganda is divided into 56 districts, which in turn are divided into counties and sub-counties. In 206 of 216 sub-counties sampled, at least one respondent had obtained private health care in the previous three months, an indicator that competition in the health care sector is widespread.14 2.1. Samples and descriptive statistics I restrict my analysis to households who report in the health bribery module having used a health care unit in the previous three months (75% of households). I analyze patients in the public sector and the private sector separately (46% of visits are to private facilities). Column 1 of Table 2 shows that the bribery rate for the public sector (panel A) is a large 17%. The private bribery rate is lower, at 11% (panel B). In column 2, I show the share of patients who made official payments: 38% of public patients do so, but almost all private patients (83%) do so. The average and median (non-zero) bribe amounts are US$6.06 and US$1.95 respectively in the public sector, higher than the corresponding private-sector values of US$5.26 and US$2.46 (column 3). Official payments, for those paying, are similar in the public and private sectors, at about US$12 (column 4). For the public sector, the bribery rate of 17% is slightly below the bribery rate for all institutions together (20%) and the bribe amount is also lower compared to the average bribe of $13 (the numbers for all institutions are not reported in the table). This could be a sign that competition from the private sector reduces corruption in public health care. Table 3 contains the means of variables taken principally from the health module for users of the health system (columns 1–4) and non-users (column 5). Patients of the public sector (columns 1 and 2) are slightly poorer than patients of the private sector (columns 3 and 4). Patients of both sectors are better off than non-users (column 5). Public sector patients who did not bribe averaged US$82 in monthly household expenditure, while public sector bribers had higher average expenditure at US$91. Private-sector bribers are not richer than other private-sector patients. In each column, at least 61% of patients suffered from malaria/fever/headache. Slightly more than half of public patients visited a hospital, while this share was low for private patients: 14% for non-bribers and 6% for bribers (though private clinics perform many functions of a hospital). In Table 4, I present users’ ratings of the quality of facilities, the quality of service from staff, the availability of medicine and
14 The sub-counties with no respondents using private health care have unusually small single-digit samples.
(5) Good or very good 0.60 0.69 0.65 0.66
(6) Very good 0.09 0.12 0.14 0.14
3467
the quality of treatment. Respondents have a choice of five ratings, from very poor to very good. The table shows that public users (columns 1–3) are much less satisfied with their experience than private users (columns 4–6), and that the (non)-availability of medicine is the most serious concern of public patients. Additional means for users are in Tables A.1 and A.2. 3. Estimation My first outcome of interest is the probability that household i bribes conditional on using the health system, P(B | U)i . I estimate P(B|U)i = ˇ0 + ˇ1 Wi + ˇ2 Xi + i ,
(1)
where Wi is household expenditure, and Xi are other covariates. My second outcome of interest is the amount of the bribe, Ai . Using the sample of bribes, I estimate log Ai = ˇ3 + ˇ4 Wi + ˇ5 Xi + i .
(2)
I estimate both equations separately for users of the private and public health systems. In order to allow a comparison of how official and unofficial payments work to allocate resources, I also compare the results when the dependent variable refers to official payments, rather than bribes. Because household expenditure is constructed based on relatively few survey questions, I expect some bias towards zero in the estimates of ˇ1 and ˇ4 . Conversely, these coefficients could be biased upward if bribes are effective in obtaining better health care and hence better health, which in turn diminishes income losses from ill health. The detailed controls for the health condition and age of the patient mitigate this concern, although I cannot control for the severity of the condition, and information on illnesses for which treatment is sought does not fully capture health. Furthermore, if low-income patients have more severe illnesses, and more severely ill patients are more likely to bribe, the lack of controls for illness severity means that the estimates of ˇ1 and ˇ4 could be biased down.15 I investigate explicitly whether bribers obtain better health care. To do so, I estimate equations of the form P(Q k = j|U)i = ˇ6k + ˇ7k Bi + ˇ8k Wi + ˇ9k Xi + ik ,
(3)
Qk
where represents the user-reported quality of the k th aspect of the medical care (facilities, service, availability of medicine, treatment). The possible values of j I consider are poor or very poor; good or very good; and very good. The coefficients of interest are the ˇ7k , and as explained in detail and with a formal model in Hunt and Laszlo (2009), they are likely to be subject to endogeneity bias: patients encountering poor service and conditions are likely to consider bribing to improve them. As no suitable instrument for bribing
15 If it is true that unscrupulous people both become rich and bribe, the coefficients will also be biased upwards. However, I doubt that the willingness to pay small bribes for health care is an indicator of the sort of unscrupulousness that might bring business success.
J. Hunt / Journal of Health Economics 29 (2010) 699–707
703
Table 5 Users in past three months – who bribes, who pays? Public sample
Private sample
(1)
(2)
(3)
(4)
(5)
(6)
(A) Bribe Log household expenditure R-Squared
0.015 (3.0) 0.08
0.021 (3.9) 0.10
0.017 (3.2) 0.11
0.001 (0.1) 0.06
0.005 (0.7) 0.07
0.006 (0.8) 0.10
(B) Official payment Log household expenditure R-Squared
0.051 (5.8) 0.08
0.053 (5.6) 0.08
0.043 (4.4) 0.12
0.011 (1.6) 0.12
0.010 (1.4) 0.12
0.010 (1.3) 0.14
Observations Basic covariates Education, job Other non-medical covariates Medical covariates
Yes – – –
4104 Yes Yes – –
Yes Yes Yes Yes
Yes – – –
3467 Yes Yes – –
Yes Yes Yes Yes
Notes: Marginal effects from probits in columns 1–3, linear probability coefficients in columns 4–6. t-Statistics in parentheses, clustered by sub-county. All regressions include 54 district dummies, 13 household size dummies, and an urban dummy (basic covariates). Education is captured by dummies and job refers to dummies for respondent occupation and the main source of household income. The other non-medical covariates are the number of household males and females over 18, respondent sex, age and age squared, and status as head or spouse. The medical covariates are the patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment, and (in columns 4–6) a dummy for a mission facility.
is available, the likely bias must simply be borne in mind when interpreting the results. I estimate Eq. (1) using a probit for the public sample, but for the private sample and Eq. (3) I use linear probability, to avoid the problem of cells lacking any observation whose dependent variable equals one.16 I estimate Eq. (2) using OLS. In all regressions, the standard errors are clustered at the level of the smallest region in the data, the sub-county. Because of the oversampling of urban areas, all specifications include controls for district and urban location. All specifications also include controls for household size dummies. The other Xi covariates are those whose means are in Table 3 and Tables A.1 and A.2: respondent (not necessarily the patient) education and occupation dummies, sex, age, age squared, and status as head or spouse; the number of males and females over age 18 in the household and dummies for the principal source of household income; patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment and (for the private sector) mission facility. I do not include non-users of the health system in any of the regressions. As Table 3 shows, non-users are poorer than users, and may be deterred from seeking treatment by the combination of corruption (particularly in the public sector) and fees (particularly in the private sector). The analysis based on users only will tend to underestimate the share of the costs of corruption borne by the poor. 4. Results 4.1. Probability of bribing Panel A of Table 5 presents results for the probability of bribing conditional on using the health care system. Columns 1–3 contain marginal effects from probits for patients in the public sector. The preferred marginal effect of 0.017, with the full covariates including medical covariates in column 3, implies that a doubling of expenditure (an increase of two thirds of a standard deviation) increases the bribery probability by 1.2 percentage points. This compares to the overall bribery probability for public sector of 17%. The effects are similar to point estimates of 0.011–0.018 for all officials pooled in Hunt and Laszlo (2009). There is again no evidence that competition in health care reduces the role of expenditure in determining
16
A probit cannot be estimated for a sample containing such cells.
bribery. In contrast to the public sector, there is no significant effect of expenditure on the probability of bribing in the private sector, in the linear probability regressions of columns 4–6 (nor in unreported probit counterparts with fewer covariates). Hunt and Laszlo (2009), considering all institutions, reported that lower-expenditure Ugandan users were more likely than richer users to pay what they considered to be official payments, but receive no receipt. If these are unwitting bribes, it means that their omission leads to the role of expenditure in paying a bribe to be overstated. These unwitting bribes cannot be identified in the health module, but I have examined their relation to household expenditure in the main bribery module for health care alone. The unreported results show a positive but insignificant relation between expenditure and the payment of unwitting bribes, suggesting that their omission does not bias the health care bribery results as much as was the case for all officials pooled in Hunt and Laszlo (2009).17 I investigate the link between expenditure and the probability of an official payment in panel B of Table 5. In the public sector, the marginal effect of 0.043 in column 3, the preferred specification, implies that a doubling of consumption increases the probability of making an official payment by 3.0 percentage points, compared to a mean payment rate of 38%. The absolute effect is larger than for bribes, while the percent effect is similar. The positive effect is not a surprise, as there are exemptions from payment for the poor, and the rich may demand more of the services which are not free. By comparison, Hunt and Laszlo (2009) found a smaller marginal effect of 0.022 for all officials pooled.18 There is no significant link between expenditure and official payment probability in the private sector, where most people make official payments (83%), although the point estimate is 0.01 (panel B, columns 4–6). This is consistent with the observation of Amone et al. (2005) that patients exempted at Catholic hospitals are typically not poor. The patients paying bribes and the patients making official payments are to a large degree different people, as Table 6 indicates. In the public sector, while 22% of those making no official payment bribe, only 8% of those making an official payment bribe, and 81% of bribes are paid by patients not making an official payment.
17 This check is not perfect, as in the main bribery module the private and public sectors cannot be distinguished. 18 For official payments for which a receipt was given. The marginal effect including receiptless payments would be smaller.
704
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table 6 Bribes and official payments among health care users (%). Public No bribe
ing official payments, the procedure differs from that in the public sector, as the richer patients are not those paying the bribes.
Private Bribe
All
No bribe
Bribe
All
No official payment Official payment
78 92
22 8
100 100
45 98
55 2
100 100
No official payment Official payment
58 42
81 19
– –
9 91
83 17
– –
100
100
–
100
100
–
All
The contrast is even greater in the private sector: fully 55% of the small number of patients making no official payment bribe, while only 2% of patients who make an official payment bribe, and 83% of bribes are paid by patients not making an official payment. The strong negative correlation between paying bribes and making official payments holds up for both sectors in regressions with the controls used in Table 5. When the bribery regressions of panel A in Table 5 are repeated for samples of patients not paying official fees, the marginal effects of expenditure are larger for the public sector (though not significantly so), and remain insignificant for the private sector (these results are not reported). It is possible that people bribe to avoid official payments: one focus group mentioned this, although the bribe in question was paid to the local government official issuing the exemption certificate. However, most of the public sector bribes mentioned in the focus groups had a different purpose, which suggests that the causality may be the opposite: health officials may attempt to extract bribes from the richer patients who need not pay officially (because of the type of facility or because of their low income). It is possible that local private hospitals likewise try to circumvent exemption policies imposed on them by their national organization, such as the church, or the government, which at least in theory pays for the private treatment of patients with respiratory and sexually transmitted diseases. Private patients with these diseases do not appear more likely to bribe in the regressions, suggesting the purpose of the bribes is not to raise money from exempt patients (however, these patients appear more likely than malaria patients to pay official fees, raising the possibility that they are unaware their care should be free and are paying unwitting bribes). Bribes could also be paid to avoid official payments, or simply to obtain better health care, for which I provide evidence below. Whatever the reason for the bribes by private patients not pay-
4.2. Amount of bribe In Table 7, panel A, I examine the determinants of the amount of the (log) bribe for the sample of bribes. The preferred specifications are those with the maximum covariates including medical covariates: in the public sector (column 3), the income (expenditure) elasticity is 0.37, while is only 0.15 in the private sector (column 6). The rich pay more than the poor in both sectors, but pay a lower share of their expenditure, since the elasticity is less than one. These elasticities compare with elasticities of 0.25–0.33 for pooled officials in Hunt and Laszlo (2009): once again, competition from the private sector does not appear to reduce price discrimination in the public sector. If the panel A regressions are repeated for patients making no official payments, the (unreported) elasticities are similar. Panel B shows that the income (expenditure) elasticity for official payments is essentially the same as for bribes in the public sector, at 0.36, while the private-sector elasticity for official payments is much higher than for private bribes, at 0.38. Public sector bribes, public sector official fees, and private-sector official fees are extracted from payers in the same way based on their expenditure. Only private-sector bribes are extracted differently, and are closer to a flat-rate fee. 4.3. Discussion Together, Tables 5–7 suggest the following characterization of payments for health care in Uganda. In the public sector, richer patients are considerably more likely to make official payments, presumably in part because of official policies charging patients according to ability to pay, and in part because richer patients demand more of services that are not free. However, since some rich patients use free services, and some poorer patients not making official payments are willing to pay non-zero amounts, health facilities are able to induce these patients, particularly the richer ones, to make unofficial payments (bribes). The income (expenditure) elasticity for the bribe amounts is the same as for official payment amounts. The private-sector differs in that almost everyone makes an official payment, exemptions are not based on household expenditure, and bribes appear to be flat fees assessed on the exempted.
Table 7 Bribers – log amount of bribe, log amount of official payment. Public sample
Private sample
(1)
(2)
(3)
(4)
(5)
(6)
(A) Bribe Log household expenditure R-Squared Observations
0.427 (7.3) 0.24
0.431 (6.4) 0.27 697
0.373 (5.4) 0.37
0.257 (3.5) 0.30
0.251 (3.0) 0.37 398
0.153 (2.0) 0.51
(B) Official payment Log household expenditure R-Squared Observations
0.452 (9.7) 0.26
0.464 (9.3) 0.28 1550
0.357 (8.1) 0.40
0.433 (16.7) 0.23
0.440 (15.9) 0.25 2868
0.376 (13.9) 0.36
Basic covariates Education, job Other non-medical covariates Medical covariates
Yes – – –
Yes Yes – –
Yes Yes Yes Yes
Yes – – –
Yes Yes – –
Yes Yes Yes Yes
Notes: OLS coefficients with t-statistics in parentheses, clustered by sub-county. All regressions include 54 district dummies, 13 household size dummies, and an urban dummy. Education is captured by dummies and job refers to dummies for respondent occupation and the main source of household income. The other non-medical covariates are the number of household males and females over 18, respondent sex, age and age squared, and status as head or spouse. The medical covariates are the patient age, dummies for patient illness, type of facility, ongoing versus once-off treatment, and (in columns 4–6) a dummy for a mission facility.
J. Hunt / Journal of Health Economics 29 (2010) 699–707
705
Table 8 Effectiveness of bribery. Public users
Facilities quality Staff service quality Medicine availability Treatment quality
Private users
(1) Poor or very poor
(2) Good or very good
(3) Very good
(4) Poor or very poor
(5) Good or very good
(6) Very good
0.080 (4.0) 0.064 (3.3) 0.156 (7.4) 0.120 (5.1)
−0.123 (−6.3) −0.120 (−5.5) −0.090 (−5.2) −0.104 (−4.4)
−0.029 (−4.1) −0.026 (−2.5) −0.017 (−2.3) −0.021 (−2.4)
0.025 (1.4) 0.031 (2.2) 0.055 (2.9) 0.041 (2.3)
−0.051 (−1.8) −0.035 (−1.3) −0.101 (−3.3) −0.068 (−2.5)
0.033 (1.9) 0.034 (1.7) 0.001 (0.1) 0.004 (0.2)
Observations
4104
3467
Notes: Coefficients on dummy for bribe payment from linear probability regressions with all covariates used in Table 7, including medical covariates. t-Statistics in parentheses, clustered by sub-county. The fifth quality category is “satisfactory”.
However, the income elasticity for official payment amounts is the same as for official payments and bribes in the public sector. This could indicate that in all three cases the elasticity is determined by the same combination of fee-for-service (and the rich demanding more expensive services) and price discrimination. Anecdotal evidence indicating that public sector bribes often consist of paying for supplies is consistent with an important role for fee-for-service bribes. If private-sector health facilities operate in more competitive environments than public facilities, this would be an explanation for less frequent and lower-value private-sector bribes, as well as the inability of the private facilities to price discriminate in bribes. However, the ability to price discriminate also depends on how well staff can judge patients’ ability to pay, which is in turn influenced by staff turnover, the distance patients travel to the hospital and procedures for assessing patient eligibility for official fee exemptions. A priori it is unclear whether private or public staff know their patients better. 4.4. Does bribery improve health care? In Table 8, I investigate what bribers obtain in return for their bribe. I present the coefficients on the bribe dummy from linear probability regressions with the full set of covariates. Columns 1–3 for the public sector show that bribery is associated with worse health care. For example, column 1 shows that bribers are 8.0 percentage points more likely to report that the facilities were poor or very poor, column 2 that bribers are 12.3 percentage points less likely to report that the facilities were good or very good, and column 3 that bribers are 2.9 percentage points less likely to report that the facilities were very good. Columns 4 and 5 for the privatesector paint a similar picture, and unreported ordered probits also indicate that bribery is associated with worse healthcare. These results presumably indicate that bad conditions push people to bribe to try to improve them, and it is difficult to tell whether they are successful in doing so. However, the final column of results for the private sector is different. Bribery raises the probability of reporting very good facilities or service, though these results are not quite statistically significant, and has no effect on the probability of reporting very good medicine availability or treatment. These results suggest that bribery is most effective for private patients who can make already good health care very good. 5. Conclusions In the public health care system in Uganda, richer patients are more likely to bribe. A doubling of household expenditure increases the bribery probability by 1.2 percentage points, compared to a bribery rate of 17%. The income (expenditure) elasticity of the bribe amount is 0.37. The link between bribery and consumption or expenditure is similar to or higher than the link for all institu-
tions pooled, despite the presence of competition from the private sector which might be expected to weaken this link. Additional results combined with anecdotal evidence suggest that the public sector health staff extort bribes particularly from the richer among the patients who officially need not pay. The amount of the bribe is then influenced by household expenditure in the same way as official charges in both the public and private sector, probably through a combination of price discrimination and fee-for-service. The results suggest that well-intentioned policies to reduce health care payments for the poor may be thwarted by health workers seeking to supplement their own incomes. In addition to undermining the goal of increased access to health care, the effect could be to contribute to a culture of bribery which helps bribery flourish in other institutions. Although private-sector bribes are even more strongly associated with not making official payments, the mechanism works differently. Those exempted in the private sector have similar household expenditures to those not exempted, and appear to pay close to a flat-rate bribe. Private-sector bribes are more effective in improving health care than public sector bribes. Acknowledgements I am grateful to Vincent Chandler for excellent research assistance, to the Instituto Nacional de Estadística e Información for the Peruvian data, and to Niels Ulrich, Paul Mpuga, Barbara Kasura Magezi Ndamira and K2 Consulting for the Ugandan data. I thank Lorena Alcázar, Leah Brooks, Janice Seinfeld, Dean Yang and seminar participants at McGill, UC Davis, UC Berkeley and the NBER Inter-American Seminar for comments on an earlier version including Peruvian data. I am also affiliated with the CEPR, IZA and DIW-Berlin, and I acknowledge the Social Science and Humanities Research Council of Canada for financial support. Appendix A. Data appendix A.1. Data The 2002 Second National Integrity Survey was conducted in 55 of 56 districts of Uganda. The subsequent non-random sampling of sub-counties led to the sub-county of the district headquarters always being chosen, which means that urban areas are over-sampled. The district’s sub-counties were divided into three categories based on availability of government services and infrastructure, and 20% of sub-counties in each category were randomly chosen. Within each of these sub-counties, the local council 1 areas were similarly divided into three categories, and one local council 1 area per category was chosen randomly. The selection of which households to interview within these local council 1 areas did not appear to be random, as it appeared to involve choosing households near the residence or office of the local council 1 chairperson. No weights are provided with the survey (Table A.1).
706
J. Hunt / Journal of Health Economics 29 (2010) 699–707
Table A.1 Means of Ugandan household main income source and respondent occupation. Public sample No bribery Farming – cash crops Farming – foods crops Farming – livestock Manufacturing, crafts, repair Trade – petty Trade – retail/shop/stall Trade – wholesale, crop buying Government – salaried or wage Private – salaried or wage Stipends from relatives Casual work Other Farmer – mainly crops Farmer – mainly livestock Trader Civil servant/armed forces Teacher Professional in private practice (doctor/lawyer) Craftsperson (carpenter/mechanic, etc.) Casual laborer Housewife Student Tailor/builder Bodaboda or taxi driver Repair and service jobs Unemployed Retired Observations
0.16 0.25 0.02 0.06 0.09 0.10 0.02 0.12 0.05 0.02 0.08 0.05 0.37 0.02 0.18 0.06 0.05 0.02 0.06 0.07 0.10 0.01 0.01 0.01 0.04 0.01 0.00 3407
Private sample Bribery
No bribery
0.15 0.25 0.01 0.09 0.07 0.11 0.03 0.10 0.04 0.01 0.11 0.03 0.38 0.01 0.19 0.04 0.04 0.02 0.07 0.08 0.09 0.00 0.01 0.01 0.04 0.01 0.00
Bribery
0.12 0.18 0.02 0.07 0.09 0.15 0.03 0.10 0.08 0.01 0.07 0.06 0.26 0.02 0.25 0.05 0.06 0.03 0.07 0.07 0.10 0.01 0.01 0.02 0.05 0.00 0.00
697
0.16 0.19 0.02 0.09 0.09 0.15 0.02 0.08 0.06 0.00 0.10 0.05 0.33 0.02 0.21 0.03 0.04 0.03 0.08 0.08 0.10 0.01 0.01 0.02 0.04 0.01 0
3069
398
Note: Households using the health care system in the previous three months.
Household expenditure is based on the question “What did you spend on the following items for your household/family in the past one week?” (food, household durables e.g. saucepans, soap/cleaning materials, entertainment/cigarettes/drinks, travel, other), and on the question “What did you spend on the following items for your household/family in the past one month?” (health/medicines/doctors, education, clothes). It is not possible to distinguish between zeroes and missing values in the components of expenditure, so I simply assign zeroes to all missing values and sum the nine components with appropriate weights. In the health module, there are some valid responses for households who had not used the health care system in the previous three months: I drop these observations. Also, some households gave information on more than one health care visit in the previous three months (contrary to the survey instructions): I retain only one visit per household. I focus on the special module on bribery in health care. The 22 agencies listed in the main part of the survey are: local primary school, Department of Education, health unit, police, traffic police, local council 1, local council 3, Agriculture Department, Veterinary Department, Fisheries Department, Forestry Department, Department of Cooperatives, Public Service (pensions), Water Department, Land Board, Magistrates Court, Ugandan People’s Defence Force, Local Defence Force, Uganda Revenue Authority (licencing), Uganda Revenue Authority (customs, anti-smuggling), Uganda Electricity Board and “other”. In this part of the survey, respondents are invited to report the amount of unofficial payments, the amount of official payments, and whether they received a receipt for their payments. A.2. Are the bribery rates plausible? There are a number of factors that make it likely that any survey will underestimate the incidence of bribery. People may be
reluctant to admit to illegal behavior, and ignorant clients may pay bribes unwittingly. I do not believe my underestimation to be severe, however (Table A.2). In high-bribery countries, bribery is viewed as inevitable and the fault of the system. The stigma attached to admitting paying a bribe is therefore low. Also, anti-corruption drives typically target officials, so the fear of prosecution from such an admission should be low. The high reported bribery rates in Uganda are consistent with this. The Ugandan survey gives respondents the option of giving their name along with the responses. I have checked the correlation
Table A.2 Means of Ugandan patient illnesses. Public sample No bribery Malaria/fever/headache Diarrhea Injury Pregnant/gynecological Immunization STDs Intestinal/stomach Heart Accident Dental Ear/nose/throat/eye Skin Backache Bone Other diseases Respiratory Tetanus, measles General check up Observations
0.62 0.03 0.03 0.07 0.03 0.01 0.04 0.02 0.00 0.01 0.02 0.01 0.00 0.00 0.03 0.04 0.02 0.00 3407
Private sample Bribery 0.61 0.02 0.04 0.09 0.00 0.01 0.04 0.01 0.00 0.02 0.01 0.01 0.00 0.00 0.03 0.05 0.03 0.00 697
No bribery 0.70 0.03 0.03 0.03 0.00 0.00 0.04 0.02 0.00 0.01 0.01 0.01 0.00 0.00 0.03 0.05 0.03 0.00 3069
Bribery 0.68 0.05 0.05 0.04 0 0.01 0.04 0.02 0.00 0.01 0.01 0.01 0 0.00 0.04 0.03 0.01 0 398
J. Hunt / Journal of Health Economics 29 (2010) 699–707
between bribing and providing a name. Neither providing a first name only nor both a first and last name is associated with a lower probability of bribing conditional on using an official than giving no name (conditional on other covariates), although oddly providing one name and one initial is associated with a higher bribery probability than giving no name. I are therefore not concerned that reluctance to report is a major issue. References Amone, J., Asio, S., Cattaneo, A., Kweyatulira, A.K., Macaluso, A., Maciocco, G., Mukokoma, M., Ronfani, L., Santini, S., 2005. User fees in private non-for-profit hospitals in Uganda: a survey and intervention for equity. International Journal for Equity in Health 4 (6). Clarke, G.R.G., Xu, L.C., 2004. Privatization, competition, and corruption: how characteristics of bribe takers and payers affect bribe payments to utilities. Journal of Public Economics 8 (9–10), 2067–2097. Deininger, K., Mpuga, P., 2004. Does greater accountability improve the quality of public service delivery? Evidence from Uganda. World Development 33 (1), 171–191. Di Tella, R., Schargrodsky, E., 2003. The role of wages and auditing during a crackdown on corruption in the city of Buenos Aires. Journal of Law and Economics 1, 269–292. Ferraz, C., Finan, F., 2008a. Electoral Accountability and Corruption: Evidence from the Audits of Local Governments. UC Berkeley Working Paper. Ferraz, C., Finan, F., 2008b. Exposing corrupt politicians: the effect of Brazil’s randomized anti-corruption program on electoral outcomes. Quarterly Journal of Economics 123 (2), 703–745. Fisman, R., Gatti, R., 2002. Decentralization and corruption: evidence across countries. Journal of Public Economics 83, 325–345. Gambetta, D., 2002. Corruption: an analytical map. In: Kotkin, S., Sajo, A. (Eds.), The Political Corruption of Transition: A Sceptic’s Handbook. Central European University Press, Budapest. Hunt, J., Laszlo, S., 2009. Is Bribery Really Regressive? Bribery’s Costs, Benefits and Mechanisms. McGill University Working Paper. Jitta, J., Whyte, S.R., Nshakira, N., 2003. The availability of drugs: what does it mean in Ugandan primary care. Health Policy 65, 167–179.
707
Kaufmann, D., Wei, S.-J., 1999. Does ‘Grease Money’ Speed Up the Wheels of Commerce? NBER Working Paper 7093. Konde-Lule, J.K., Okello, D., 1998. User Fees in Government Health Units in Uganda: Implementation, Impact and Scope. Partnerships for Health Reform Small Applied Research Paper No. 2. Lewis, M., 2006. Governance and Corruption in Public Health Care Systems. Center for Global Development Working Paper Number 78. Mauro, P., 1995. Corruption and growth. Quarterly Journal of Economics 110 (3), 681–712. McPake, B., Asiimwe, D., Mwesigye, F., Ofumbi, M., Ortenblad, L., Streefland, P., Turinde, A., 1999. Informal economic activities of public health workers in Uganda: implications for quality and accessibility of care. Social Science and Medicine 49, 849–865. Nabynoga, J., Desmet, M., Karamagi, H., Kadama, P.Y., Omaswa, F.G., Walker, O., 2005. Abolition of cost-sharing is pro-poor: evidence from Uganda. Health Policy and Planning 20 (2), 100–108. Olken, B.A., 2007. Monitoring corruption: evidence from a field experiment in Indonesia. Journal of Political Economy 115 (2), 200–249. Olken, B.A., Barron, P., 2009. The simple economics of extortion: evidence from trucking in Aceh. Journal of Political Economy 93 (7–8), 950–964. Olken, B.A., 2006. corruption and the costs of redistribution: micro evidence from Indonesia. Journal of Public Economics 90 (4–5), 853–870. Reinikka, R., Svensson, J., forthcoming. Working for god? Journal of the European Economic Association. Svensson, J., 2003. Who must pay bribes and how much? Evidence from a cross section of firms. Quarterly Journal of Economics 118, 207–230. Thompson, R., Xavier, A., 2002. Unofficial Payments for Acute State Hospital Care in Kazakhstan. A Model of Physician Behaviour with Price Discrimination and Vertical Service Differentiation. LICOS Working Paper 124. Transparency International, 2004. Corruption continues to deprive societies around the world. www.icgg.org/downloads/ (retrieved 5 July 2010). Treisman, D., 2000. The causes of corruption: a cross-national study. Journal of Public Economics 76 (3), 399–457. Wei, S.-J., 2000. How taxing is corruption on international investors? Review of Economics and Statistics 82 (1), 1–11. Yang, D., 2008a. Integrity for hire: an analysis of a widespread program for combating customs corruption. Journal of Law and Economics 51 (1), 25–57. Yang, D., 2008b. Can enforcement backfire? Crime displacement in the context of customs reform in the Philippines. Review of Economics and Statistics 90 (1), 1–14.