ArticlePDF Available

Abstract and Figures

This Paper exploits a unique micro-level data set on primary health care facilities in Uganda to address the question: What motivates religious not-for-profit (RNFP) health care providers? We use two approaches to identify whether an altruistic (religious) effect exists in the data. First, exploiting the cross-section variation, we show that RNFP facilities hire qualified medical staff below the market wage; are more likely to provide pro-poor services and services with a public good element; and charge lower prices for services than for-profit facilities, although they provide a similar (observable) quality of care. RNFP and for-profit facilities both provide better quality care than their government counterparts, although government facilities have better equipment. These findings are consistent with the view that RNFP are driven (partly) by altruistic (religious) concerns and that these preferences matter quantitatively. Second, we exploit a near natural experiment in which the government initiated a program of financial aid for the RNFP sector, and show that financial aid leads to more laboratory testing of suspected malaria and intestinal worm cases, and hence higher quality of service, and to lower user charges. These findings suggest that working for God matters.
Content may be subject to copyright.
Available online at:
No. 4214
Ritva Reinikka and Jakob Svensson
ISSN 0265-8003
Ritva Reinikka, Development Research Group, The World Bank
Jakob Svensson, Institute for International Economic Studies, Stockholm
University, Development Research Group, The World Bank, and CEPR
Discussion Paper No. 4214
January 2004
Centre for Economic Policy Research
90–98 Goswell Rd, London EC1V 7RR, UK
Tel: (44 20) 7878 2900, Fax: (44 20) 7878 2999
Email:, Website:
This Discussion Paper is issued under the auspices of the Centre’s research
programme in PUBLIC POLICY. Any opinions expressed here are those of
the author(s) and not those of the Centre for Economic Policy Research.
Research disseminated by CEPR may include views on policy, but the
Centre itself takes no institutional policy positions.
The Centre for Economic Policy Research was established in 1983 as a
private educational charity, to promote independent analysis and public
discussion of open economies and the relations among them. It is pluralist
and non-partisan, bringing economic research to bear on the analysis of
medium- and long-run policy questions. Institutional (core) finance for the
Centre has been provided through major grants from the Economic and
Social Research Council, under which an ESRC Resource Centre operates
within CEPR; the Esmée Fairbairn Charitable Trust; and the Bank of
England. These organizations do not give prior review to the Centre’s
publications, nor do they necessarily endorse the views expressed therein.
These Discussion Papers often represent preliminary or incomplete work,
circulated to encourage discussion and comment. Citation and use of such a
paper should take account of its provisional character.
Copyright: Ritva Reinikka and Jakob Svensson
CEPR Discussion Paper No. 4214
January 2004
Working for God?*
This Paper exploits a unique micro-level dataset on primary health care
facilities in Uganda to address the question: What motivates religious not-for-
profit (RNFP) health care providers? We use two approaches to identify
whether an altruistic (religious) effect exists in the data. First, exploiting the
cross-section variation, we show that RNFP facilities hire qualified medical
staff below the market wage; are more likely to provide pro-poor services and
services with a public good element; and charge lower prices for services than
for-profit facilities, although they provide a similar (observable) quality of care.
RNFP and for-profit facilities both provide better quality care than their
government counterparts, although government facilities have better
equipment. These findings are consistent with the view that RNFP are driven
(partly) by altruistic (religious) concerns and that these preferences matter
quantitatively. Second, we exploit a near natural experiment in which the
government initiated a program of financial aid for the RNFP sector, and show
that financial aid leads to more laboratory testing of suspected malaria and
intestinal worm cases, and hence higher quality of service, and to lower user
charges. These findings suggest that working for God matters.
JEL Classification: H39, I11 and L31
Keywords: altruism, financial aid, natural experiment and religious not-for-
profit health care providers
Ritva Reinikka
Development Research Group
The World Bank
1818 H Street NW
Washington DC 20433
Tel: (1 202) 473 3328
Fax: (1 202) 522 1154
For further Discussion Papers by this author see:
Jakob Svensson
Institute for International Economic
Studies (IIES)
Stockholm University
S-106 91 Stockholm
Tel: (46 8) 163 060
Fax: (46 8) 161 443
For further Discussion Papers by this author see:
*We are grateful for comments and suggestions by James Adams, Markus
Goldstein, Daniele Guisti, Michael Kremer, Magnus Lindelöw, Peter Okwero,
Per Pettersson-Lidbom, Tomas Philipson, Halsey Rogers, Rob Yates, and
seminar participants at the World Bank and IIES, Stockholm University. We
also thank Makerere Institute of Social Research for their assistance in survey
implementation. Financial support from the Japanese Policy and Human
Resources Development Fund (PHRD) grant is gratefully acknowledged. The
findings, interpretations, and conclusions expressed in this Paper are those of
the authors and do not necessarily represent the views of the World Bank, its
Executive Directors, or the governments they represent.
Submitted 15 August 2003
1 Introduction
In man y dev eloping coun tries, particularly in Sub-Saharan Africa and Latin
Am er ica, religiously b ase d no t-for-pr ot organizations play an important role in
the provision of social services. The stated goal of these provid ers are typically
altruistic in na ture. Howev er, in m any poor countries there is limited or no
regulation or mo nitoring of the not-for-prot sector, raising concern that the
actual situation on the ground may be quite dierent from the stated objectives.
In this paper w e exploit a unique data set on service delivery of go vern-
ment, priva te for-prot, and private not-for-prot (religious) pro vid ers of pri-
mary health care in Uganda. We use the data to distinguish between tw o
alternative theories of the religious not-for-pro t(RNFP)providerbehavior:
(i) work ers and managers of RNFP health facilities are intrin sica lly motivated
to serv e poor people; and (ii) RNF P pro viders are captured by their managers
and /or w orkers and beha ve like a for-prot actor, although it cannot directly
appropriate prots. Thus, any surplus must be used to nance perks (wages
and perquisites) for the management and/or sta.
To guide the empirical work, we set up a simple model on service provision
and solve the m odel under the two hypoth eses laid ou t above. The model yields
a joint set of predictions o n p rice setting, wages, service mix and qualit y choice
conditional on preferences. We also explore the eects of nancial aid in this
framework and sho w that the im pact on price setting and quality choice crucially
depends on the assumption of objectiv es.
We tak e t wo approac hes to iden tify whether religious aliatio n matters.
The rst builds on the presumption that we can identify the beha vior of the
RNF P pro viders b y comparing their perform ance in variou s dimensions with
private for-prot and go vernm ent pro vid ers. Specic ally, w e exploit the cross-
section variation across types of o w nersh ip, controlling for other confound ing
observable cha racteristics and unobserved location-speciceects. The idea is
that since the beha vior of private for-prot pro viders (presumably driv en by
prot maximization ) and government-oper ated units (regulated b y cen tral and
loca l authorities to deliver a minimum package of services) is generally quite
w ell understood, b y comparing outcomes, w e can learn about the objectives of
the RNFP.
One concern with the cross-section approach is that there ma y be unob-
Glaeser (2002) argues that weak board control may be just as important as dierential
tax privileges, donations, and nondistribution constraint in explaining the behavior of not-
for-prot rms. Th us capture by managers is not specic to not-for-prots in developing
countries, although it seems plausible that boards in general have stronger control in the U.S.
not-for-prot sector than in the Ugandan primary health care sector (see discussion in section
2). The capture argument is also close in spirit to the Pauly and Redisch (1973) view of
hospitals as physicians’ cooperatives.
serv ed (by the econometrician) quality dierences across owner s. The second
appro ach avoids this p rob lem b y exploiting a near natural policy experiment of
public nancial aid for the not-for-prot sector.
In scal y ear 1999/2000, the
survey year, the governme nt initiatedaprograminwhicheverynot-for-prot
primary health unit w as to receive an untied gran t. As this was a new and
unan ticipated program and due to poor comm unication from the go v ernment’s
part, some not-for-prot facilities did not receive their gra nt until the follow ing
y ear. This de facto phasing-in of the nancial aid program pro vides a source
of variation (across RN F P s) that w e can exploit to identify the objectiv es of
RN F P providers.
In the cross section, we nd that religious not-for-prot facilities hire quali-
ed medica l sta below the market wa ge. Moreo ver, RNFP are more lik ely to
provide pro-poor services and services with a public good element, and charge
strictly lo wer prices for services than fo r-pr ot units. Religious not-for -prot
and for-prot facilities both provide better qualit y care than their go vernm ent
counterparts, although government facilities hav e better equipment. These nd-
ings are consistent with t he hypothesis that RNF P wor kers and m a na gers h av e
intrinsic motivations to serve poor people.
The near natural experiment re-
veals that n an cial aid leads to more testing o f suspected malaria and intestinal
worm cases and lower prices in religious not-for -pro t facilities. Moreov e r, the
estimated eects are substan tial. Th us, working for God appears to matter!
This paper is related to a large literature on the behavior of not-for-prot
rms or organizations in the develo ped world, especially in the United States.
Our work diers in three dimensions. First, w e explicitly consider religious
not-for-prot pro viders, rather than the more comprehensive notion of not-for-
prots. Second, w e use quantitativ e survey data of dieren t aspects of ser-
Duggan (2000) also studies the diential response of not-for-protversusfor-prothealth
facilities (hospitals) to a natural experiment induced by a government subsidy program. He
examines hospitals aected by California’s Disproportionate Share program and shows that
the behavior of not-for-prot hospitals varies with the share of nearby hospitals organized as
for-prot rms: increased for-prot penetration makes not-for-prot hospitals more prot-
These intrinsic motivations may in turn be driven either by altruistic preferences and/or
a religious call to serve people in need. See further discussion in section 8.
The theoretical work has mainly evolved around three types of models; altruism models,
which have quantity and quality of output in the rm’s objective function; physician coop-
erative models that are analogous to earlier cooperative rm theories (Pauly and Redisch
1973); and non-contractible quality models, where for-prot rms have an incentive to shirk
on the quality of service to cut costs (for a review, see Malani, Philipson, and David 2002;
Lakdawalla and Philipson 2001). With respect to the U.S. health sector, where most services
are produced by the not-for-prot sector, the empirical evidence is mixed (Malani, Philipson,
and David 2002; McClellan and Staiger 2000; Philipson 2000; Rose-Ackerman 1996; Sloan
and others 1998).
vice delivery from a poor developing country. Finally, as not-for-protprimary
health care providers in Uganda are not regulated; have no ob vious tax adv an-
tages o ver private for-prot rm s and, until 1999/2000 (the scal year fo r whic h
we hav e data), beneted only marginally from don ations or other nancial sup-
po rt, w e circumv ent an importa nt identication problem that has rendered it
dicu lt to test altruistic models using U.S. data.
The rest of the paper is or gan ized as follow s. Section 2 describes the institu-
tional setting of health care in Uganda, including ow nership and the gov ernmen t
nancial aid program to not-for-prot health pro viders. Sections 3 and 4 present
a simple mod el of the behavior of a religious not-for-prot health facility, de-
velop two extensions of the mod el, an d la y out the inference p rocedure. Section
5 discusses the survey data used in the empirical analysis. Section 6 presents
the empirical evidence from the cross-section regressions. Section 7 explores the
impa ct of nancial aid on service delivery. Section 8 concludes.
2 In stitutional setting in health care
It is commonly held that Uganda had w ell-func tion ing health services in the
1960s. Health care w a s provided free of cha rge, and access to care was rela-
tively good. Steady improvements w e re experienced in most health indicators.
Ho wever, as a result of the political and military turmoil of the 1970s and 1980s,
the government de facto retreated from funding and providing public services.
In health care the burden w as taken up b y the private for-protsectorand
faith-based providers. The latter were able to mobilize external resources to
pro vide lim ited services (Republic of Uganda 2001a). D espite eorts by the
private for-prot and not-for-prot sectors, health indicators fell dramatically.
Follow ing restoration of peace in the late-1980s and subsequent economic re-
co v ery, the governmen t implemented a major program of health infr astructure
The problem is that the type of o wnership may be endogenous. A non-altruistic en-
trepreneur may choose a not-for-prot status and locate in a poor neighborhood if she expects
to benet from, for example, c haritable donations as a consequence of this ownership/location
choice. Th us, although the ownership/location choice will have adverse nancial consequences,
higher expected donations will compensate for them and make the ownership/location choice
optimal (i.e., the total expected cash value of perks is higher when taking donations into
account). Due to the absence of regulation and tax benets, and minimal donations, such
incentives do not play an important role in Uganda. Thus, prior to 1999/2000, there were
no obvious advantages for a nonaltruistic entrepreneur to choose a not-for-protstatus. Of
course, the lack of regulation and monitoring still raises the concern that the preferences of
the owner or the founder (for instance a Catholic parish) and the manager may dier. In par-
ticular, the f acility may be captured by a manager with dierent obje ctives from the owner.
This is one of the hypothesis we test.
rehabilitation in the public sector in the 1990s. This coincided with politi-
cal, administrative, and nancia l decentralization, which led to slow growth in
recurrent funding for health facilities, as districts prioritized area s other than
health care (Jeppson 2001). A s a result the qualit y of public services did not
improve at the same pace with health infrastructure, whic h is reected in the
con tin ued high demand for privately pro vided care (Hutchinson 2001). Some
health indicators ha ve improved, but others ha ve not. Specically, the infant
mortality rate stagnated during the latter half of the 1990s at 88 deaths per
1,000 liv e births (Republic of Uganda 2002, Moeller 2002).
The modern health sector in Ugand a has four types of facilities: hospitals,
health cen ters, dispensaries, and aid posts. These facilities can be o wn ed and
operated by the government, private for-prot, or not-for-prot sector. The
health facility surv ey w e exploit in this paper has the dispensary (w ith or with -
out a maternity unit) as the unit of observa tion. Dispensaries are the most
com m on health facilities in Uganda. Most dispensaries are rural (89 perc ent).
Accordin g to the governm ent health sector strategic plan, the standard for
dispensaries includes prev entive, promo tiona l, outpatient care, maternity, gen-
eral wa rd, and labora tory services (Republic of Uganda 2000). A dispensary is
suppose to ha ve eigh t beds for inpatien t care and to serve a population of 20,000.
Dispensaries are usually not expected to hav e a medical doctor (although som e
do), and are managed either by a clinical ocer or a compreh ensive or registered
n u rse.
2.1 Ownership of he a lth facilitie s
The private not-for-prot health sector in Uganda consists of religious and non-
religious providers. The rst ever census on the not-for-prot health care sector
in Uganda carried out in 2001 indicated that auto nom o us dioceses and pa rishes
o w n 70 percent of all private not-for-prot health facilities, which total 450
lo wer-level units and 42 hospitals (Republic of Uganda 2001b). The rest are
o wned b y nongov ernm ental organizations (16 percen t), some of which are also
religious, commun ity-based organization s (6 percent), and by district councils,
mosques, and individuals (8 percent). The census also show s that most pri-
va te not-for-pr ot health facilities (82 percent) are coo rdinated by one of three
national um brella organizations: Catholic, Protestant, and M uslim medical bu-
The rst religious not-for-prot health unit was established b y missionaries
in 1897 (Republic of Uganda 2001a). Thereafter local churc hes and missionar-
ies have set up hospital and health centers throughout the coun try.
At their
Since expatriates were not allowed to own xed assets, missionaries established the units
departure, missionaries handed over the management to the local c hurc h (dio-
cese or parish). In the last three decades, as new parishes were established,
they (usually initiated b y the parish priest) routinely set up their o w n socia l
services, particularly health services. In man y cases par ishioner s contributed
to the in v estment cost of these facilities, sometimes aided by donations from
the respective medial bureau or outside sources. The majorit y of dispensaries
o wned b y religious providers w ere built between 1960 and 1990. In our sample,
the median y ear of establishment is 1983.
Not-for-prot health care providers are self-governing. At the time of our
survey, there was no certication for not-for-prot status (either b y a medical
bureau or by the gov ernm ent). Hence, the manager in charge of the not-for-
prot health unit together with the unit-specic managemen t committee w ere
free to decide on the mix and price of services pro v ided b y the facility.
It is w o rth noting that the institutional str uc ture of the not-for-prot sector
is considerably dierent from the governmen t’s institutio na l framework . M ost
importan tly, the medical bureaux operated by various religious denominations
do not have administrative authority over the individual units or o wners (that
is, dioceses or parishes).
Private for-prot practice also began decades ago, rst with a few medical
practitioners in urban areas. Their numbers grew dramatically during th e eco-
nomic and political turmoil of the 1970s and 1980s (Republic of Uganda 2001a).
Private health care was provided by a mixture of licensed and unlicensed pri-
va te clinics, pha rm acies, drug shops, and home providers,
and little systematic
inform ation is a vailable on these providers. Many medical professionals w orking
in the public sector are believed also to ha ve a private practice to earn extra
incom e (McPake and others 1999), but factual evidence of the extent of this
pra ctice, partic ula rly in the case of dispen sar ie s, is limite d.
In the public sector, th e health secto r strategic p lan determines facility stan-
dards and the mix of services to be pro vided at each level (Republic of Uganda
2000). Both centra l and loca l go vernm ent authorities attempt to enforce these
standards b y con tro lling inputs; setting stang norms; supplying pharm aceu-
ticals, vaccines, and equipmen t; and directing transfers and investm ent fund-
ing. In additio n, they issue man agem ent and technical guid elines a nd supervise
health facilities. Public health facilities ha ve also a unit-specic managemen t
comm ittee to represent the loca l communit y. Due to a variety of factors, suc h
as diculties in recruitment of qualied medical sta, and the a vailability of
funding, the actual situation on the ground ma y vary considerably from the set
in the name of the local diocese or parish.
In principle, the national professional c ouncils are supposed to regulate both private for-
prot and not-for-prot facilities (but not governmen t facilities) by licensing them, setting
standards, and monitoring their premises. This regulatory system is not working in practice.
Finally, w h ile all health care providers are exemp t from the value added tax,
private for-prot providers are, in principle, expected to p ay income tax, as well
as the pa y -as-you-earn tax for their employ ees. But there are major problems
in com p lian ce. Thus, apart from a few p rivate clinics in Kampa la ( the capital),
private for-prot dispensaries are de facto tax exempt. As mention ed earlier,
religious not-for-prot pro viders are exempt from income tax.
2.2 Financial aid for nonprots
Two umbrella organizations for not-for-pr ot health pro viders–the Uganda
Protestant Medical Bureau and the Uganda Catholic Medical Bureau–were
established in the 1950s to coord inate disbursemen t of government grant s to
religious h ealth c are providers. While p u blic su bsidies continued after indepen -
dence, ov er time the relations betw een religious providers and the go vernm ent
deteriorated, as there w as competition and a perceived dierence in pay and
privileges (Republic of Uganda 2001a). During the decline in public service
delivery in the 1970s and 1980s, subsidies to not-for-prots dwindled and ev en-
tually ceased altogether. In response to the disappearing public support, not-
for-prots had to resort to user fees and external donations. The t wo bureaux
established a join t medical store to supp ly their alia ted facilities with drugs
and other medical con sum a bles and equipment.
In the early 1970s the Uganda
Mu slim Sup rem e Cou nc il also established a similar umbrella organization .
Over time, the importance of external donations declined. In our sample
of (religious) not-for-pro t facilities, only 3 out of 44 not-for-protdispensaries
receiv ed donations from private sou rces and only 2 out of 44 facilities received
funds from the donor community in 1999/2000.
In 1997 the government reinstated nancial aid to hospitals. In scal y ear
1999/2000 a new program extended a similar subsidy to lower-level health units.
The nancial aid program prescribed that every not-for-prot unit was to re-
ceiv e a xed-amoun t grant for the scal y ear. The amoun t of the gran t varied
accord ing to the lev e l of the health facility. Each dispen sar y was to receiv e the
same amount, namely 2.5 million shillings ($US 1,400) a year. Eac h dispensary
with a mater nity unit was to receiv e 3.4 million Ush ($US 1,900). Ad m in is-
trative problems in getting the program fo r low er-level units o the ground are
Today all types of health-care providers can purchase drugs from the joint medical store
and hence take advantage of its bulk purchase prices.
As stressed above, donations were m ore importan t in the 1970s and 1980s, as well as at
the start-up phase of a new health facility, when raising funds for construction. We have some
indirect evidence for the latter. Of the 29 not-for-prot facilities that had renovated their
facility in the past, 14 had received nancial support from private and/or donor sources.
discussed in section 7.
3 Conceptual framework
Next we lay out a simple framework to analyze the not-for-protbehavior. We
consider three models. The rst two models implicitly assume that the (a ltru is-
ticowners)not-for-prot facility is captured b y a nonaltr uistic man ag er(s). In
the rst model, the nondistribution constrain t is not binding so that the not-for-
prot provider acts exactly like a prot-maximizer. In the second version, we
assum e the constraint binds, in which case the not-for-pro t provider maxim izes
perquisites instead. The third model assumes that the religious not-for-prot
facilities maxim ize the total health impact of its activities, here conceptualized
We start by solving the simplest version of the three models, and then con-
sider two extensions: endogen ous quality and costs.
3.1 Basics
A manager for a not-for-prot health facility i m ust hire w ork ers to work in the
facilit y and agree on wages w
. Each worker j can perform one task or service.
There are S potent ia l services. Thus, a facility can at the most h ave S workers.
Thereisapoolofworkerswhodier with respect to the value placed on wo rk ing
in a not-for-prot facilit y. Specically, a worker j’s utility is u(w
where u(w) is a concave function, NFP is an indicator varia ble taking the
value 1 if the worker is emp loyed by an altruist ic not-fo r-prot facilit y (and
zero otherwise), and δ
is the non-pecuniary gains of w orker j of working in a
not-for-prot facility. We label δ
as the "religious premium". Each w orker can
get a job in the public sector, which pa ys the w age ¯w.
The manag er must also decide what services to provide and prices of these
services. The total cost of producing a give n service s S that x
will be buying is w
+ cx
is the wage cost of the w o rker assigned
to produce service s,andc is the (constant) marginal cost. We th us assume
Clearly, conceptualizing altruism in the health sector with the number of patien ts treated
is not uncontroversial. See Malani, Philipson, and David (2002) for a review of altruism
models that typically have quantity (and/or quality) of output in the not-for-prot’s objective
The assumption of excess demand of workers by the public sector at (the administratively
set) wage ¯w, is a good approximation of the health labor market for qualied sta in Uganda
given the economy wide shortage of qualied medical sta (Okello and others 1998).
that a worker w ill be pa id the sam e amount irrespective of th e nu mber pa tients
treated. The inverse-deman d function for health service s is p
= P(x
) where
is the price and P
) < 0.Welet
denote the elasticity of dema nd with
respect to price for service s. The facilit y is assum ed to be a loca l mono polist.
3.2 The prot-m a x imizing not-fo r -p r o t facility
The total cash prots of facility i is π =
[P (x
the set of services oered b y facility i. Weassumethatworkersdonotobtain
any additional non-pecu nia ry gains from working in a prot-maximizing not-for-
prot facilit y ; that is, δ
=0.Aprot-max im izin g facility can hire an unlimited
n u mber of workers at the w ag e ¯w. Its maximization problem is thu s,
[P (x
¯w cx
] . (1)
The rst order conditio n of activity s can be stated as,
P (x
c 0 . (2)
Equation (2) is a standard condition for prot maxim iza tion ; the price will
be set to equate the marginal revenue (rst term in (2)) with the (constant)
marg ina l cost. Equation (2) implicitly denes the optimal quantit y x
and b y
the inverse-dem and function the price p
that creates this demand. Since the
margina l cost is the sam e for ea ch service, the marginal revenue for each service
being provided m ust also be the same.
Equatio n (2) is a necessary condition for prot maxim ization. In addition,
eac h service m u st yield non-negative prot s. That is, the facility will provide
the service s only if
P (x
¯w cx
> 0. (3)
3.3 The perquisites-m aximizing not-for-prot facility
Follow in g Gla eser and Shleifer (2001), we assume that if the nondistribution
constraint binds, the manager is forced to spend prots on perquisites, denoted
by z. The utility of spending prots on perqu isites is v(z)=αz,whereα<1
is a constant.
As with the prot maximizer, w e assum e that w ork ers do not obtain any non-
pecu niary gains of wo rking in a perqu isites-maxim izing not-for-prot facilities;
that is, δ
=0. Its maximization problem is th us,
max α
[P (x
¯w cx
] . (4)
Clearly, the rst-order condition of activit y s, and the non-negative prot
constraint are iden tica l to (2) and (3). Thus, a perquisites-m axim izing not-for-
prot facility will set the same prices p
as a for-prot facilit y. Moreo ver, it
will pay workers the sam e wage as a for-prot facilit y, and it will also choose to
provide the same set of services.
If p r ivate-for-pro t and private not-for-prot facilities o nly dier in the ease
in with which a facility can app ropriate pro ts, and if facilities decision variables
are (i) which services to pro vid e, (ii) the pric es of these services, and (iii) wages
to their w orkers, we should not observe an y dierences between private-for-prot
and priva te not-for-pro t facilities.
3.4 The altruistic not-for-prot fac ility
The third assum es that private not-for-pr ot facilities maximize the total health
impa ct of its activities. Clea rly, the total health impact could be dened in a
va riety of ways. H er e we choose to operationalize it as the nu mber of (poor)
patients treated. That is, the private not-for-prot facilities maximize
subject to the constraint that
[P (x
w cx
] 0.
Consider rst the choice of workers. A m an ag er for an altru istic fac ility will
trytohireworkerswhohaveabiastowardworkinginanot-for-prot facility
(i.e., wo rkers with an in trinsic motivation to serv e poor people). To simplify
the exposition, assume there are two large group of w orkers, one with δ
and one with δ
= δ>0. The not-for-prot facilit y will hire w orkers with
= δ and pa y them the w age w = u
[uw) δ].Notethatw < ¯w.Thus,the
not-for-prot facilit y will exploit the workers’ non-pecuniary benets of w or king
in a not-for-prot facilit y by oering a low er wage. The w ag e is set so that at
the margin, a worker with a positive religious premiu m is indieren t to working
in a not-for-prot facilit y or a for-prot facility.
To solve the altruistic not-for-prot manager’s maxim ization problem we
formulate the Lagrange function ,
L =
+ λ
[P (x
w cx
, (5)
where λ is the Lagran ge m u ltiplier. Max im izing the Lagran ge function yields
the follow ing r st order conditions,
P (x
0 s S
[P (x
w cx
]=0. (7)
Equatio n (6) implies that as for the prot- or perquisites-m axim izing not-
for-prot facility, the marginal rev enue for eac h service being pro vide d will be
the same. Th us, higher prices will be ch ar ged for services with low elasticity of
demand. The intuition is straigh tforw ard. Given the zero-prot condition (7),
and given that dierent patient t y pes are perfect substitutes, if the marginal
rev en ues dier, the facility can pro v ide one less patient with the service with
the lowest marginal rev enue, and instead provide more than one extra patient
with the service with the highest rev enue. Thus, b y shifting types of patients
treated, the aggregate n umber of patien ts treated could be raised.
Note further that prices will be set suc h that the marginal return is strictly
lo wer th an the marginal cost. That is, an altruistic not-for-prot facility will
charge lower prices than a prot- or perquisites-m a xim izing no t-fo r-prot facil-
Finally, (6) and (7) imply that an altruistic provider may cross-subsidize
services and therefore will tend to prov ide a broader range of services. In par-
ticular, whereas a perquisites-maxim izing not-for-prot facility (or a private
for-prot facility) w o uld never prov id e a service it cannot make a positive prot
from; i.e., for whic h (3) does not hold , an altruistic provider ma y do so in order
to increase the total number of patien ts treated. The lower wag e cost will also
make it more likely that altruistic unit provides a broader range of services.
3.5 Q uality of ca r e
So far w e ha v e assumed that qualit y of care is exogenous. Assume no w in-
stead that before (or sim ultan eo usly) choosin g wha t services to pro vid e, the
manager/facilit y also mak es an eortchoicethatinuences the quality of the
services being pro vided. Let the inverse dem and function be p
= P (x
where q is eort and P
> 0 and P
> 0. We assume that higher quality ser-
vices imply both higher nancial and non-pecuniary (eort) costs. Total cash
protisnowπ =
[P (x
(q) > 0 and
(q) > 0. The manager m ust also bear a non-pecuniary cost of exerting eort
given by γ (q),whereγ
(q) > 0 and γ
(q) > 0.
Consider rst a for-protprovider. Theadditionalrst-order condition is
given in (8),
(q) 0 . (8)
The tw o rst-order conditions (8) and (2) dene the optimal price and qualit y
for service s.
The rst-order con d ition for th e qua lity choice for a perquisites-maxim izing
provider is
0 . (9)
Totally dierentiating the rst-ord er conditions (9) and (2), using the im-
plicit function theorem (see appendix), it is easy to show that the qualit y of
care of the for-prot facility ex ceeds that of the perquisites-m a ximizing facilit y.
Higher qualit y of services will also allo w the facility to dema n d a higher price.
This is an intuitiv e result. Providing higher qualit y services requires eort.
Since private for-prot rm s are more responsive to prots, a for-protprovider
has stronger incentives to put in high eort.
Conside r next the altruistic facility. The rst-order conditions of the altru-
istic prov ider’s maxim iza tion progr am are giv en in (6), (10) and (11).
(q) λ
0 (10)
[P (x
w cx
] C(q) 0 (11)
Higher qualit y will increase demand and allow the altruistic provider to
Withou t further restrictions on the model, howev er , we
cannot sa y if the altruistic facility will exert more or less eort than the for-
prot pro vider. Ho wev er, what w e can say is that only an altruistic pro vider
will tend to cross-subsidize services, and th u s can pro vid e a service it cannot
make a prot from. It will also pa y their w orkers less. Moreo ver, conditional on
the qua lity choice being sim ilar, an altru istic provider w ill charge strictly lower
In the standard (reduced form) altruism model, the provider cares about quantity and
quality. Obviously, if quality has its own value for the altruistic provider, this would provide
even stronger incentives to supply high-quality care.
3.6 Endogenizing cost
In the baseline model, (marginal) cost is constan t and exogenous. Howev er , it
is reasonab le to think that facilities can partly inuence their cost structure.
For example, a facility could reduce cost ex post b y shirking on quality. Belo w
we consider how such an extension wo uld aect the results.
Glaeser and Shleifer (2001), building on Hansmann (1980), argue that pri-
va te not-for-pr ot rms face softer incentives which protect consumers from ex
post appropriation. Since private for-prot rms are m ore responsive to pro ts,
they will hav e stronger incen tives to pursue cost and non veriable qualit y re-
ductions on th e service(s) pro vided. It is straightforward to incorporate Glaeser
and Shleifer’s mech anism in the model.
= P (x
) where q
is the expected quality of the service being provided , with P
> 0.Unitcost
is c = C(q),withC
> 0 and C
> 0. As in Glaeser and Shleifer (2001), the
manager m ust bear a non-nancial cost of β(q
q) of shirking on quality.
In this set up, when the manager chooses q, he has already collected rev enues
(th u s he takes the price and demand as giv en). The perquisites-maxim izing not-
for-prot facility’s optim al quality reducing c h o ic e is given by
+ β 0 . (12)
Ration al patients will anticipate the manager’s ex po st incentives. Thus, in
equilibrium q
= q
.Thefor-prot provider’s equilibr ium condition is the same
as in (12), with α =1.
Total dierentiating (12) yields,
< 0 .
Th us, the non v eriab le quality of the not-for-p rot facility exceeds that of
the for-prot facility. Lo wer quality (whic h is expected in equilibrium) w ill lead
to lo wer costs. Lo wer qu ality will also lead to lo wer dem an d . Both factors lead
to lower prices. Lower demand will ten d to reduce the n umber of services that
can be provided, although this force is counteracted b y lower cost. W itho ut
further restrictions on the model, it is unclear ho w service pro v ision will be
The altruistic facilit y will have no incentives to shirk ex post on quality,
since this will not aect (ex post) the number of patients that could be treated.
The predictions of the baseline and the extended versions of the model are
summarized in Table 1. The baseline model suggests that we could test for the
not-for-prot facilities’ objectiv e function by running the follo wing regression
on a sample of facilities with dieren t o wners,
= β
+ β
+ β
+ ε
where the dependen t variable y
is either s
, a indicator if service s is being
provided or not by facility i; p
, the price of service s c h arged by facility i; w
thewagepaidtoworkeroftypej in facilit y i;orq
, the qualit y of services.
is a dummy indicating if the facilit y is no t-for -pr ot, and FP
is a dummy
indicatin g if the facility is priva te for-pro t. The ow nership category exclu ded
in (13) is governm ent. The perqu isites-maxim izin g not-for-prot facility h y -
po thesis suggests th at β
= β
for y
= {s
}, wh er eas the altruistic
not-for-prot facility hy pothesis suggests that β
, β
Endogen izing cost and allo w ing facilities to also choose quality make it more
dicult, using observ ed prices, w ag es, and service prov ision, to distinguish be-
t ween the not-for-prot’s objectiv es. In particular, there are parameter con-
gu ration s for whic h we cannot reject either of the two hypotheses. However,
only the altruistic model, under all model specication s, is consisten t with the
predictio n that β
, β
, β
and β
implica tion therefore forms the basis for the empiric al analysis.
Wh ile the model pro vides a starting point to assess the behavior of not-
for-prot facilities, it is clearly based on a nu mber of simplifying assumption s.
Thus, the question is whether an association between ownersh ip a nd outcomes,
from a regression lik e (13), is a causal relationship. In particular, the dierent
t ypes may ha ve other c h aracter istics that are also associated with the dependent
variable y. For examp le, for-prot and not-for protprovidersmaylocatein
dieren t areas and thus face dierent demand.
We consider two strategies to iden tify a causal relationsh ip in the data:
con tro lling for other confounding observables (discussed below ) and exploiting
a near natural experiment of nancial aid to not-for-prots from government
(discussed in section 7).
In the cross-section analysis, identication is based on the assumption that
we can control for variables that are co nfou nd ed with o w nersh ip and the depen-
dent va riable. Th us, w e will estimate an equation of the follo w ing form,
= β
+ β
+ β
+ β
+ ε
where X
is a vector of other controls. Belo w we discuss the controls we use.
In the baseline regression , we pro xy for the degree of competition b y in-
cluding as a con tr ol the "number of competitors", i.e., number of dispen saries
and health cen ters in the facility’s catchment area. In the model, each facility
acts lik e a local monopolist. In reality, patients ha ve some c h oice about where
to seek health service (although the data suggest that proxim ity is the most
importan t fa ctor o verall for selecting a given facility), so the market structure
may be importan t.
Not-for-prot facilities receive (lim ited ) in-kind support (medicine and sta)
that may shift the marginal cost curve and thus inuence y.Weexplicitly
control for this by includ ing a measure of th e value of free d rug s received an d a
variable capturing the full-time equivalen t nu mber of sta w o rkin g in the facility
for free.
Because eac h fac ility’s location, in principle, is endogenous, determ ining
wheth er it is ow ner ship per se or location or some other fa ctor correlated with
loca tion that drives any observable dierences in o utcom e, could p resent a di-
cult identication problem. Howev er, in practice it is less of a concer n. First, as
discussed above, most not-for-prot facilities w ere established many y ea rs ago.
Giv en the large social and economic changes in Uganda during the last few
decades, the local situation may ha ve c hanged dramatically for man y facilities.
Second, empirically, w e can (to some extent) control for location by including
con tro ls such as distance to closest subcount y headquarters a nd district-specic
eects. Thus, we identify the ow nership eects from the within-district vari-
ation. Finally, and most importantly, it is possible to rein terp ret the mod el,
letting the c hoice of services to pro v ide and the prices of these services, really
be a choice of w h ere to locate. If not-for-prot facilities are driven by altruistic
concerns, they would ten d to locate in poor areas where they would not be able
to charge high prices. If not-for-prot facilitie s a re n ot drive n by altruistic con-
cerns, they would instead tend to locate in areas w here they, ju st lik e for-pro t
facilities, would maxim iz e prots. The reduced form approac h of studying the
relationship between ownership an d outcomes is va lid as long as w e attempt to
measure underlying objectiv es (preferences).
Tools to collect data and analyze service provid er behavior include facility mod -
ules in household surv eys and empirical studies to estimate facility (in particular
hospital) cost functions. The approac h used here, a quantitativ e service deliv-
ery survey (QS D S), is distinct from these other tools in a number of respects
(Dehn, Reinikka, and Svensson 2003). First, unlike most other surveys, the
service pro vider is the key unit of analysis. In household surveys that include
facilit y modules, th e perspectiv e is that of the household rather than the service
provider (Lindelöw and Wagsta 2003). Consequen tly, while nding pro xies for
service quality, they pay little attention to the question of why quality of services
is the wa y it is. This is reected in the type of data collected, which is mainly
on simple access indicators and the range of services oered. In other wo rds,
these surv eys largely ignore provider behavior and the processes and complexi-
ties through whic h pu blic spending is transformed into services. In most cases,
facilit y information is collected as a part of community questionnaires, whic h
rely on the knowledge of one or more informed individuals. Information sup-
plied by informants is not only heavily dependent on the perception of a few
individuals but also not detailed enough to form a basis for analysis of service
delivery. To the extent that the inform ation is based on perception s, there
ma y be additional problems due to the subjective nature of the data and its
sensitivit y to respondents’ expectations.
Second, the QSD S does not rely on budgeted costs, as m u ch of public ex-
pend itu re incidence an alysis does, but collects detailed data on actual spending
and services provided at the facilit y level.
Finally, the QS D S explicitly recogniz es that agents in the service delivery
system may ha ve strong incentives to misreport (or not to report) k ey data.
These incen tives derive from the fact that information provided by, fo r example,
a health facilit y may partly determine its public funding. Also, in case resources
(including sta time) are used for other purposes (for instance in the case of
shirking or corruption), the agen t involved in the activity will most lik ely not
report it truthfully. Moreover, certain t ypes of information, such as ocial
c h arges, may only partly capture what is intended to be measured (e.g., the
users’ costs of the service). The QSD S deals with these data issues in two ways.
First, data are collected using a multi-angular data collection strategy; that is,
a com bination of information from dierent sources. Specically, data on the
Ugan da n health facilities were collected both at the district and health facilit y
level, as w ell as from patients using an exit poll. Second, data sources that are
least inuen ced b y misreporting w ere identied. For this reason, the data are
obtained directly from the records k ept b y facilities for their o wn needs (such as
patien t registers, medical records) rather than administrative records submitted
to local gov ern m ent. Th e former, often ava ilab le in a highly disaggre gated
format, was considered to suer least from an y incentive problem s in record-
keeping (see Table A.1 for summ ar y statistics).
The su rvey d ata that we use in this paper consists of 155 randomly selected
prim ar y health care facilities dra wn from 10 random ly cho sen districts in all four
regions of Ugan da. A detailed description of the sample design and the survey
is provid ed in the appendix (see also Lindelöw, Reinikka, and Sv ensso n 2003).
The sample is restricted to dispensaries and dispensaries with maternit y units in
order to ensure a degr ee of homogen eity across facilities. T h e sample includes
facilities from the three main o wn ersh ip categories: governm ent, private not-
for-prot, and private for-prot. As described earlier, the priva te not-for-prot
health facilities in Ugand a are mostly operated by religious organization s, and in
our sample all nonprots have religious aliatio ns.
so that the proportion of facilities dra w n from dieren t regions and ow nersh ip
categories broadly mirrors th e population of facilities. However, as noted earlier,
no census of private for-prot health facilities is available in Ugand a, and it is
hence dicu lt to assess the exten t to wh ich the samp le is representative in this
Of the 155 facilities, 81 (52%) are go vernm ent o wne d, 44 (29%) are
o w n ed b y not-for-prot pro viders, and 30 (19%) are priv ately o w ned.
6 E mpir ic a l resu lt s
6.1 Sta remuneration
We start b y looking at the simple relationship bet ween sta remuneration and
o wnership (Table 2). We ha v e data for about 900 employ ees in a total of 130
facilities. We have information on position, skill level, and pay but no other
employ ee c h aracteristics.
Regression 1 repor ts a basic wage regression , with dummy variables for n ot-
for-prot and priv ate for-prot facilities. The depend ent variable is the full-time
equivalen t salary plus lunc h allo wan ces per month.
As evident, the religious
not-for-prot facilities pa y signicantly less than both the private for-prot
ones (F-test) and the go vernm ent operated units (t-test). The private for-prot
facilities also pa y signican tly less than gov ern m ent facilities. On av era ge, reli-
gious not-for-prots pa y rough ly 65,000 Ush per employ ee less per month than
the gov ern m ent operated facilities and 17,000 Ush less than for-prot facilities.
These are large dierences, considerin g that the av erage (unconditional) full-
time equiva lent salary plus allo wanc es per month is 1 09 ,000 Ush. In R egres sio n
1, the district eects are also highly signicant (LR-test). Facilities in more
Two of the 44 not-for-prot pro viders did not have a religious aliation. These facilities,
however, drop out of most regressions due to lack of data.
A sample of government and private not-for-prot facilities was drawn randomly from
the health facility register kept by the Ministry of Health. For-prots were identied on the
basis of information obtained from the sampled government facilities.
In 1999/2000, the lunch allowance for public sector employees was supposed to be 66,000
shillings per month for health care professionals and 44,000 for support sta. In the follo wing
year, the lunch allowance was formally rolled up into public sector salaries and is no longer
regarded as a separate item. The qualitative results are similar if we use the more narrow
measure for salary, excluding lunch allowances.
remote areas, i.e., where th e distance to the closest subcou nty center is greater,
the pa y on average is less, but the eect is not signicant. The number of
com petitors also en ter s insignicantly.
One explanation for the dierence in remuneration is that sta compositio n
diers across ownersh ip. If the a verage skill (education) level is correlated w ith
o wnership, and better-educated workers are paid more, the average eect cap-
tured in Regression 1 ma y simply be a composition eect. To con tro l for this,
Regression s 2-6 report the ndings from subsamples of the sta.Regression
2 considers only qualied sta.
The pattern is similar. Gov ernmen t facilities
pa y the most, and the religious not for-prots pay the least.
Regressio n 3 show s wa ge-setting condition al on o wn ersh ip for the highest
qualied sta, i.e., sta with a least A-lev el with subsequen t medical training.
For this group of w ork ers, on average, the religious not-for-prot pro viders pa y
60,000 Ush per emplo yee less per month than the government-operated facilities
and 56,000 Ush less than for-prot facilities. T h er e is no signicant dierence
in remuneration bet ween fo r-prot and gov ernm en t providers. The average
(unconditional) full-tim e equiv alent salary plus allowances per mon th for the
highest qualied sta is 212,000 Ush. Th us, on averag e, the highest qualied
sta are paid 28 percent less than both for-protandgovernmentsta in the
same category.
Regressio n 4 reports the results for the largest group amon g the qualied
sta, that is, enrolled nu rses. While we still observe a large dierence bet ween
private and go v ernm en t sta (enrolled nurses employed by private pro viders
receive 65 percent lower wag es than average), there is no signicant dierence
in rem uneration between for-prot and not-for-prot providers.
The same pattern holds for unqualied sta. Regression 5 depicts the re-
lationship between w ages and ow n er ship for nursing aides (the largest group
of w ork ers in the unqualied group). Private for-prot dispensar ie s pa y 41,000
Ush less per month (comp ared with th e go vernm ent facilities), while the not-
for-prot providers on a verage pay 49,000 less. These two coecient estimates
are not signicantly dierent at the 10-percent level.
The prelim inary an aly sis thus suggests that there exist a religious premium
but only for qualied sta, which makes it possible for religious not-for-prot
facilities to hire qualied w o rkers below market wage. This premium does not
show up in the sample of unqualied or less-qualied sta. One explanation
for this is simply that unqualied sta are paid a ve ry low salary. They may
Qualied sta include medical doctor, clinical ocer (A level and three years of medical
training), comprehensive nurse (A level and three years of medical training), registered nurse
(A level and two-and-half years of medical training), laboratory assistant (O level and three
years of medical training), and enrolled nurse and midwife (O level and two-and-half years of
medical training).
therefor e not be able to accept a lo wer pay.
One concern with these results is that w e are missing not only the usual
(in a w age regression) unobservables, but also some standard observables in de-
termining wages, suc h as experience. Unfortunately, information on experien ce
was not collected in the survey. A priori, it is not clear how this omitted vari-
able bias would inuence the results. If health sta in the public sector ha ve
longer tenu re and thus are more experienced tha n their counterparts elsew h ere,
we would overestimate the religious (altruistic) wage premium. Conv e rsely, if
the not-for-protproviders’sta is more experienced, the reverse would be true.
Fortunately, w e can quantify how important this experience bias might be, since
w e have inform ation on the salary scale for medical personnel in gov ernmen t
health fa cilities.
For a qualied nurse; i.e., a nu rse with at least A le vel with
three years of medical training, the maximum returnstoexperienceis12,000
Ush. T hus, in the extreme case where qualied sta in not-for-prot facilities
havelittleexperienceandqualied sta in government and private for-pro t
units are highly experienced , this would explain roughly one-fth of the dier-
ence in the observ ed wage dieren tial between go vernm ent (and for-prot) and
not-for-prot pro viders.
The nding that go v ernment pays higher salaries is surprising, as the com-
mon ly held view in Uganda is that public sector pa y is well below the private
sector, including health care professionals. For example, a pay com parator study
underpinning the government pay reform, puts the pa y of a clinical ocer and
an enrolled n urse employ ed by governm ent at 20-40 percent of that in the pri-
va te sector (Repu blic of Uganda 1999). Our nding s are in start contrast to
this, at least in the case of lower-level health facilities.
Regressio n 6 pools the sta similarly to Regression 2, but adds information
on the level of med ical train ing. The va riab le qualication takes the value 0
for enrolled nu rses and midw ives, 1 for laboratory assistan t, 2 for registered
and comp reh ensive n u rses, 3 for clinical ocers, and 4 for medical doctors (see
footn ote 8 fo r details of medical training in Uganda). We allo w the ow nersh ip
eect to be condition al on sta qualications b y interacting qualica tion with
the ownership dum m ies. The results are illustra ted in Figure 1. As before,
go v ernm en t pay is higher than that of both types of priv ate pro viders. Not
surprisingly, more qualied sta are generally be tter rewarded (positive coe-
cient of qualication). Ho wever, there are dierences in the marginal return to
medical training depending on the o wnership of the dispensary they w ork in.
More specically, the mar ginal return to medica l training is lowest in the go v -
ernment service (i.e., wages are th e m ost compressed in the governm ent sector)
and highest in the private for-prot sector. Highly qualied not-for-protsta
Salary schedule B for medical personnel species salaries for 10 categories of sta,witha
range of salaries for each category depending on the experience of the respective sta member.
Figure 1: Salaries in relation to government units in for-prot (thin line) and not-
for-prot (thick line) facilities conditional on qualication.
are paid signican tly less than their for-prot coun terparts. Hence, the religious
premium in pay, shown in Figure 1 as the v ertical dierence bet ween the t wo
po sitively sloped line, falls largely on the most qualied medical sta.
6.2 Mix of services
Tables 3a an d 3b report a series of regression, wh ere the dependent variable is a
0,1 in dicator if a given service is being provided (1), or not (0).
All facilities
are pro viding general outpatient services (OPD). From Table 3, it is po ssible
to identify two broad sets of services. The rst group includes in-patient care,
medical care, laborator y services, and immunization.
The religious not-for-
protandtheprivatefor-prot pro viders are as lik ely to provide these services.
For medic al care, all three facility ty pes are similar, while gov er nment facilities
are signicantly less likely to pro v ide laboratory services.
Our empirical evidence sho ws that governmen t units are the most likely
ones to carry out immunizations, follow e d b y the religious not-for-prot facil-
ities. However, this eect is solely driven by dieren ces in vaccine supply.
We focus on the most common health services. A handful of facilities also provide mental
care, eye care, and dental care.
The term "medical care" refers to (non-surgical) curative care.
Immunization is a special service from the individual facility perspective. The national
Controlling for the free supply of vaccines in Regression 5, w e nd no signi-
can t dierence between the three t ypes of fa cilities. Th is is also consistent with
the principles of the national (vertical) immunization program.
For the second set of services, the two private sector pro v iders dier. This set
includes outreac h, health education, training of n u rses and community health
workers, antenatal care, and family planning. All these services, except family
planning, are more likely to be provided by the not-for-prot than the for-
prot facilities. Not-for-prot facilities are less like ly to pro v ide family planning.
Comp aring the not-for-pr ot and governm ent facilities, the la tter is more likely
to do outreach (almost all did, 77 out of 80), although the religious not-for-
prot and go vernmen t facilities are similar in number of sta days per mon th for
outreach (Reg ression 7). The r eligious not-for-prot facilities are more likely to
run training programs for community health w orkers, while go vernm ent clinics
are more lik ely to provide antenatal care, but the eect is small.
How do w e interpret these results? Clearly, the services depicted in Table 3
dier in their prot potential, the exten t to which they benet the poor, and
in their public good nature. In general, inpatient care, medical care, an ten atal
care, and laboratory testing are services that are demanded b y a broad spectru m
of the population and are not typically pu blic goods. Most o f th ese services are
just as likely to be p r ov ided by fo r - p r ot facilities as th e r eligiou s no t-for -pro t
ones, an tenatal care being an exception.
It could be argued that outreach is a service that in general has a bias to ward
the poor. Fewer not-for-prots provide this service compared to go vernment but
those that do pro vide more of it. Health education and training of communit y
health w orkers ha ve a public good element. Therefore it ma y be dicult to
make a positive prot from these three services. The perqu isites-m axim izing
model predicts that suc h services w ould not be pro vided neither by for-prot
nor b y not-for-prot facilities. The data does not suppor t this prediction. In
fact, the pro-poor services (outreach) an d those with a pub lic good element are
signican tly more lik ely to be provided by the religious not-for-prot facilities
than the for-protones.
As can be seen from Table 3, family planning in an exception. A probable
explanation for this is that the not-for-prot facilities hav e religious motivations
for not providing this type of service, particularly as the Catholic Church is
important as a health care provider.
immunization program (UNEPI) sets countywide standards for immunization services and
manages the program vertically by providing supplies to health facilities. In fact UNEPI is a
monopoly supplier of vaccines in Uganda (both regular supplie s and for immunization da ys).
The program sets its targets for immunization based on population and pro vides vaccines
to meet those targets directly to the facilities. Not-for-prot dispensaries also receive their
vaccines from UNEPI as do some private for-prot providers.
6.3 Price setting
Table 4 shows that there are large dierences across types of facilities. For g en -
eral OPD , the gov er nm ent facilities c harg e almost 2,000 Ush less per rst visit
compared to p rivate for-pro t pro viders. The m edian paymen t in a go vernment
facilit y is 500 Ush. Th e religious not-for-pro t facilities charge signicantly
more than the go vernm e nt fa cilities, but signicantly less (roughly 600 Ush)
than the for-prot fa cilities. A simila r pattern mat ches user fees for the other
services, as can be seen from Regressions 2-5. Private pro v iders cha rge more
for mino r surgery, an tena tal care, medical care, and deliv ery-related services.
With the exception of antenatal care, the for-prot facilities char ge more than
the religious not-for-prot, ranging from around 600 Ush for minor surgery to
5,000 Ush for deliv ery.
The baseline model suggests that a perquisites-m a xim izing not-for-prot
facilit y will set the sam e prices as a for-prot facilit y, wh ile the altruistic mode l
suggests that the not-for-protwillsetpricesatwhichthemarginalreturnis
strictly lo wer than the marginal cost. In other w ords, an altruistic n ot-for-prot
facilit y will c ha rge strictly lo wer prices than a for-pro tunit. Thendings on
user-fee charg es are consistent with the altruistic model.
6.4 Q uality
In the mod el, a for-prot facility would ch oose to exert higher eort (ex ante) to
increase quality than a perquisites maximizing not-for-pro t facility. Withou t
further restrictions, w e cannot say if an altruistic clinic would c h oose to exert
more or less eort than a for-prot one. We can, howev er, draw the conclusion
that if qualit y of services is the same or higher in the not-for-pr otsectorthanin
the for-prot sector, this is inconsisten t with the perquisites-m aximizing mod el.
Mea s ur in g quality is dicult. We provide three complementary mea sures.
The rst measure is b ased on observ e d treatment practice. The second measure
is based o n observed supply (that is, availability of health in frastru ctu re), while
the last one is a qualitative indicator derived from the exit poll data.
One important component in prescribing the correct treatment for malaria
and in testin al worm cases is laborato ry testing. We ha ve informa tion on these
t wo types of test. The number of malaria blood slides carried out (for every
100 suspected malaria patients), and the nu mber of stool tests undertaken (for
ev er y 100 suspected intestinal worm cases). Table 5 reports the ndings with
respect to testing . In line with the nding on laborator y services, we nd that
private pro viders are signica ntly mo re likely to test patients for malar ia and
intestinal worms. The eect is large. For example, on av erage, the private
providers test 25 more patien ts of every 100 suspected malaria patients. It is
interestingtonotethatthisisnotduetodierences in health equipment and
sta (regressions 2 and 4). Adding these additional controls do not change the
results regarding o wnership. Having highly qualied sta and a microscope
increase the frequency of testing, how ever.
Table 6 reports the result on health infrastructu re. Government facilities are
more likely than private for-prot facilities to ha ve sterilization and refrigeration
equipment, and more lik ely than not-for-prot facilities to have equipme nt to
measure blood pressure. It is interesting to note th at private fo r-prot facilities
are as lik ely to have observable health equipment (inputs); that is, equipment
that is being used in the actual treatmen t process (such as protective clothes,
blood pressure equipment), while they are less lik ely to ha ve equipment that is
more dicult to observe (like sterilization equipment and refrigerators). O ne
explanation for this is that the private for-prot clinics are more responsive to
prots, and thus ha ve stronger incentiv es to cut costs and pursue non veriable
quality reductions on the services being provided.
The last result on qualit y c hoices is based on information on wh y the patient
had c h osen to visit the facility whe re she w as interviewed (from exit polls).
Patients reported that pro xim ity and good treatment and/or good sta were
the most important factors for selecting the fa cility. Proximity is the most
important factor o verall; this is partic ular ly true for go vernmen t facilities. In
contrast, patients are signicantly more likely to report good treatmen t and/or
good sta as a reason for visiting private facilities. Not surprisingly, facilities
without qualied sta are less lik ely to be visited for quality reasons.
appear to provid e better qu ality care. We cannot distinguish betw een the pri-
vate for-protandthenot-for-prot pro viders. These ndings are inconsistent
with the perqu isites-maximizing hypothesis.
7Theeects of nancial aid
A key question in a cross-section framework such as (14) is whether the selectio n-
on-observables assum ption is plausible. There migh t be unobserved variables
that a re related to both the dependent variable y and the o w nership indicators.
In particular, there may be unobserved (by the econometrician) qualit y dier-
ences across ow ners. Our second approach avoids this problem by exploiting a
near natural experim ent of go vernm ent nancial aid for the not-for-protsector.
As discussed in section 2.2, the nancial aid program for dispensaries w as
initiated by the go vernm ent of Uganda in 1999/200 0 and prescribed that every
Results are available upon request.
not-for-prot unit was to receiv e a xed-amoun t gran t for the scal ye ar. The
program wa s implemen ted b y the local go vernm ents (districts). Specically, the
Ministry of Finance transferred the funds mean t for lower-level units operated
by not-for-prots to the local governments (districts), which in turn distributed
the funds to the units concerned once they had submitted a w orkplan. In the-
ory, all not-for-prot facilities should have receive d the funds in 1999/2 000. In
practice, ho wev er, there was variation in receipts due to a number of idiosyn-
cratic factors, including not-for-prot dispensaries not submitting the necessary
documentation in time, uncertainty about w ha t the grants could be used for (it
w as meant to be an untied gran t), and generally poor comm unications and lack
of information. As the system of providing nancial aid for not-for-prot units
was new, this pattern w a s not surprising. The outcome for 1999/2000 was that
some units did not receive their entitlement. Instead their rst grant reached
them the follow ing scal year. Th us, de facto the gran t program was phased in.
It is this va riatio n in receipts over time that w e exploit .
A possible objection to this approach is that the de facto phasing-in was not
random , or more specically that the incidence of receipts could be correlated
with the error term in equation (14). In th at case, correlation between transfers
and outcomes m ay be spuriou s. Altho u gh we cannot empir ic ally fully reject
both the transfer and dependen t variable y), we can ch eck if the groups of g rant
recipien ts and nonrecipients dier on observables.
Tables 7a and 7b report a set of regressio ns using observable facilit y c har -
acteristics as dependent varia bles. The regressor is a dummy variab le takin g
the v alue 1 if the facility receiv ed the en titled grant, and 0 otherwise (denoted
receip t of aid).
Regression s 1 and 2 show that gran t recipien ts and nonrecip-
ien t s do not dier signicantly in age, measured either as the y ear the facility
was established (Regression 1), w hen the facility has been renovated last (year),
or whether the facility had been renovated (Regression 2). The recipien ts and
nonrecipien ts do not dier in access to communication infrastructure (R egres-
sion 3), that is, a nonrecipien t is as lik ely as a grant recipien t to ha ve access
to telephon e, newspa pers, and radio at the facilit y. We also do not nd any
dierence in distance to district or subcounty headquarters (R egressio n 4), size
of the facility (Regressio n 5), whether or not the facilit y wa s staed with at least
one qualied nurse or a doctor (Regression 6 ), or in access to h ealth infrastr uc-
ture (Regression 7-10). Th u s, there is no (observable) evidence suggesting that
the grant recipien ts and nonrecipients dier on observable chara cteristics (apart
from receiving the transfer or not).
We allow a 10 % variation between en titled and received funds. That is, a facility is
considered a recipient (dummy=1) if it receiv ed at least 90 % of its entitlement.
Although the groups of facilities do not dier in observables, they may still dier in some
The reason w e use the var iation in gran t receipts to identify the eects of
o wnership is that a prot or a perqu isites-ma xim izing not-for-prot pro vider’s
behavior would not be aected b y the ino w of aid. Un tied aid does not aect
the marginal cost or rev en ue schedules. Th us, it w ould set the same prices and
pro vide the same services as without aid.
Thealtruisticnot-for-prot facility’s maximization program would however
be aecte d. Forma lly, with aid, the facility max imizes,
L =
+ λ
a +
[P (x
where a is aid or nancial support. As shown in the appendix, for an altruistic
provider, aid will le ad to lower prices (and possibly more services) and to higher
quality care. These results are in tu itive. The altruistic provider cares abou t the
n u mber of (poor) people treated and this n umber can be increased by either
lo wering prices or increasing the quality of care. Both strategies are costly.
Aid relaxes the provider’s budget constraint and at the margin it is optimal to
increase the n u mber of peop le treated using bo th strategies.
Wh ile religious not-for-prot facilities receive na ncia l aid from public sources,
no for-prot facility did.
Conditiona l on receiving nancial assistance, the me-
dian receipt for not-for-prot d ispensarie s (with maternity unit) w as 3.2 million
Ugan da n shillings (Ush), which is close to the amount allocated and disbursed
by the central go vernm ent (3.4 million Ush). Ro ug hly 25 perce nt of the not-
for-prot facilities did not receive na n cial aid.
Wh en evalua ting the eects of nan cial aid, it is impo rta nt to iden tify whic h
potential variables might be aected b y the inow in a short time inter val. We
loo k at three sets of variables that facilities can easily adjust in the short run:
testing procedu res, prices, and sta rem u neration.
In Table 8, Regression s 1 and 2 , report the correlation be tween nancial aid
and laborato ry testing. Financial aid is positively correlated with testing for
malar ia and in te stina l wo rm s. Th e estimated eects are large. A not-for-prot
provider with the median grant receipt tests, on average, 24 more patients out
of every 100 suspected malaria case.
Wh en w e test the relationship betw een user fees and nancial aid for the
specic services we have in form atio n on (minor surgery, an tenatal care, m ed ical
care, and deliv eries), we nd no impact of nancial aid. H owever, as depicted
in Regr ession 3, nan cial aid is negativ ely correlated with OPD user c har ges.
unobserved dimension. However, this unobserved dimension m ust then be uncorrelated with
the set of observable characteristics reported in Table 7.
Dropping two suspected misrecorded observations, we have data from 152 of the 155
sample facilities.
A religious not-for-prot provider with the median gran t receipt charge, on
a verag e, 900 Ush less for general OPD .
Calculating the foregone revenues of the price cut and the increased cost of
testing for malaria and in testinal w o rm s suggests that the cost for the median
facilit y is approxima tely 2.3 million Ush, or 68 percent of the total grant.
Finally, we analyze the relationship between salaries and nancial aid. There
is no robust relationship in any sta category (w e report the results for qualied
sta and nursing aides only).
A concern with the regression results in Table 8 is that we explore both the
va riation created b y the de facto phasing-in of the grant program (i.e. bet ween
recipien ts in 1999/2000 and recipients in the following year) as well as the
va riation in actual receipts across gran t-receiving providers (although it seems
plausible to assum e that this second source of variation is also random). Table 9
depicts IV-regressions, using the binary receiptofaidvaria ble discussed above
as instrument. The coecients in the laboratory testing regressions increase
and are highly signican t in both the testing for malaria and intestin al worms
cases. The coecien t on user fees also remains signicant. There is still no
signican t relationship between aid and rem u nera tion .
To summarize, we nd evidence that n an cial aid leads to mor e testing of
suspected malaria and intestinal wor m cases and lower prices for OP D services,
but only in religious not-for-prot facilities. Since the var iation in nancial as-
sistance is, as we have argued abo ve, to a large exten t exogenous, these ndings
pro vide strong evidence in support of the altruistic h ypothesis.
8 Concluding remarks
In this paper we exploit a unique m icro-level data set on primary health care
facilities in Uganda to explore the motivation of religiou s not-for-prot(RNFP)
health care providers. To iden tify these objectiv es we use tw o strategies. The
rst builds on the assumption that w e can identify the behavior of the not-
for-prot providers b y comparing their performance in various dimensions with
for-protandgovernmentproviders. Thesecondapproachreliesonanear
natura l policy experim ent of public nancial aid for the not-for-protsector.
The ndings from both approac h es point in th e same direction and suggest that
RNF P workers and manag ers ha ve intrinsic motivations to serv e poor people.
The calculation is based on the assumption that the price cut resulted in a 10 percent
increase in the number of patients and that the grants were received with a three-month
delay. Assuming no delay in grant receipts, the reduced charges and additional cost of testing
account for 91 percent of the grant. Assuming six-month delay in grant receipts, the reduced
charges and additional cost of testing account for 46 percent of the grant.
It is worth poin ting out what we ha ve not measured other possible explana-
tions for the pattern we observe. First, we interpr et the evidence abov e in fav or
of the altruistic model. Howev er , as argued b y Glaeser (2002), it may still be the
case that the not-for-prot providers are captured by their work ers/ m ana gers,
but that their preferences are also altruistic and therefore they (partly) in ter-
nalize the stated goals of the o w n er. There is some qualitativ e evidence to
support this interp retation , as many practitioners in the eld report that the
working enviro nmen t in religious not-fo r-p rot facilities ar e considerably better
(i.e., rev e nues are spent on imp rovin g the working en v iron m e nt for the sta).
This in turn could also help explain why salaries in not-for-prots are lo wer (i.e.,
compensated by a better environ m ent). On the other hand, there are reports
that labor practices in religious not-for-prots are not alwa ys ideal–dismissing
single pregnan t work er s, compulso ry religious activities–and that these policies
can be resen ted b y the workers.
Second, since all the not-for-prots in our sample have religiou s aliations,
it is possible that the objectives are not altruistic, but to convert people. The
provision of services and the service delivery c hoices ma y therefore be guided
b y this goal. Distingu ishing between these t wo objectiv es would require data
on nonreligious not-for-prot providers and a model of a health pro vider max-
imiz in g the number of people con verted (for example by maxim izin g public
relations). Clearly, this is an im portant area for futu re research.
Third, it is possible to think of alternative explanations for each individual
n din g reported above. For example, the religious w ag e premium may be due
to rigidities in the labor market com b ined with recent increases in the pa y for
go v ernmen t emplo y ees. On the other hand, paid training (where a per diem
typically makes up a signica nt part of the mon th ly salary) and benets (e.g .,
pe nsions) are more preva lent in the public sector, whic h suggest that what w e
pick up is a lower bound . Also, this eect cannot explain the wage dieren tial
between private for-prot and not-for-prot pro viders. If the type of workers
(within a category of w orkers, say, n urses) diers across ownersh ip t y pes, this
may also explain the w ag e dierential. In particular, if w orkers in the not-for-
prot sector are less competent health care prov iders, this may explain why they
are paid signican t ly less. However, this interpretation is dicult to reconcile
withthefactthatnot-for-prot facilities pro v ide better quality care than their
gov ernmen t coun terp arts.
An alternative explanation for the nancial aid ndings is that better-run,
w ell-organized, not-for-prot pro viders managed to get the nancial aid soon er
than poorly functioning ones. But if these well-organized units also pay higher
wages, w e should observ e a positive relationship between monthly salaries and
aid. We do not. In addition, on observables, the early and late aid recipients
look similar. We believe the strength of the argumen t put forward in the paper
lies in the fact that w e nd consistent evidence across dierent aspects of service
delivery (price and w ag e setting, service mix, and qualit y cho ices) and across
empirical tec hn iques. We cannot think of one particular alternative explanation
that w o uld explain all these facts.
9.1 Sample design and the surv ey
The sample design was go verned by three principles. First, attention w as re-
stricted to dispensa ries and dispensaries with maternity units (i.e., health center
III) to ensure a degree of homo gen eity across sampled facilities. Second, subject
to securit y constraints, the sample was meant to capture regional dierences.
Finally, t he sample had to include facilities from the main ow n ersh ip categories:
go v ernmen t, private not-for-prot and private for-protproviders.
These three considerations lead us to choose a stratied random sample. The
samp le w as based on the Ministry of Health (MOH) facility register for 1999.
The register includes government, private not-for-prot, and private for-prot
facilities, but is known to be inaccurate with respect to the latter. A total of
155 health facilities were surveyed. On the basis of existing in formation, it w as
decided that the sample w ou ld in clud e 81 government fa cilities, 44 private non-
for-prot facilities, and 30 private for-prot facilities. The exit poll of clients
co vered 1,617 individuals. The eld work was carried out during October to
December 2000. For summar y statistics see Table A.1.
As a rst step in the samplin g process, 8 districts (out of 45) had to be
dropped fro m the sam ple frame due to security concerns.
From the remaining
districts, 10 districts, stratied according to geog raphical location with the size
distribution determined b y population shares, w ere random ly samp led in pro-
po rtion to district popu lation size. T hus, three districts w e re ch osen from the
Eastern and C en tral regions, and tw o from the Western and Northern regions.
From the selected districts, a sample of gov er nmen t and private nonprot
facilities was d rawn randomly from the MOH register. A reserve list of replace-
ment facilities was also dra w n from the sample frame. Due to the unreliability of
the register for private for-prot facilities, it was decided that for-prot facilities
w ould be identied on the basis of information from the governm ent facilities
samp led .
The adm in ist rative records for facilities in the original samp le were
review ed rst at the district headquarters, where some facilities that did not
meet the selection criteria and data collection requirements w er e dropped from
the sample. These we re replaced b y facilities from the reserve list. Overall 30
The eight districts were Bundibugyo, Gulu, Kabarole, Kasese, Kibaale, Kitgum, Kotido,
and Moroto.
The study districts were Mpigi, Mukono and Masaka in the Central region; Mbale, Iganga
and Soroti in the East; Arua and Apac in the North; and Mbarara and Bushenyi in the West.
Specically, the x private facilities in region y would be determined by the in-charge in
the rst x randomly drawn government facilities in region y, where each in-charge would be
asked to identify the closest private dispensary or dispensary with maternit y unit.
facilities were repla ced.
At the district lev el, the district director of health services was in terv iewed
to obta in information on health infrastructure, s ta, supervision arrangements,
and nan ce. Data were also collected from the district records on eac h health
unit included in the survey.
A t the facility lev el, the manager in charge of the health unit was interview ed
and data were collected from medical, patient, and nancial records, stoc k cards,
etc. An