Back to basics: does decentralization improve health system

performance? Evidence from Ceará in north-east Brazil

Sarah Atkinson1 & Dave Haran2

Objective To examine whether decentralization has improved health system performance in the State of Ceará, north-east Brazil.
Methods Ceará is strongly committed to decentralization. A survey across 45 local (município) health systems collected data on
performance and formal organization, including decentralization, informal management and local political culture. The indicators
for informal management and local political culture were based on prior ethnographic research. Data were analysed using analysis
of variance, Duncan’s post-hoc test and multiple regression.
Findings Decentralization was associated with improved performance, but only for 5 of our 22 performance indicators. Moreover, in
the multiple regression, decentralization explained the variance in only one performance indicator; indicators for informal management
and political culture appeared to be more important influences. However, some indicators for informal management were themselves
associated with decentralization but not any of the political culture indicators.
Conclusion Good management practices in the study led to decentralized local health systems rather than vice versa. Any apparent
association between decentralization and performance seems to be an artefact of the informal management, and the wider political
culture in which a local health system is embedded strongly influences the performance of local health systems.

Keywords Delivery of health care/organization and administration; Community health services/organization and administration; Quality
indicators, Health care; Efficiency, Organizational; Informal sector; Politics; Regression analysis; Brazil (source: MeSH, NLM).
Mots clés Délivrance soins/organisation et administration; Service public santé/organisation et administration; Indicateurs qualité
santé; Efficacité fonctionnement; Secteur informel; Politique; Analyse régression; Brésil (source: MeSH, INSERM).
Palabras clave Prestación de atención de salud/organización y administración; Servicios de salud comunitaria/organización e
administración; Indicadores de calidad de la atención de salud; Eficiencia organizacional; Sector informal; Política; Análisis de regresión;
Brasil (fuente: DeCS, BIREME).

Bulletin of the World Health Organization 2004;82:822-827.

Voir page 826 le résumé en français. En la página 826 figura un resumen en español. .827

Introduction organizational arrangements, processes of policy implementa-

Decentralization is widely promoted in health system reform. tion and performance indicators are uncommon (9). The few
In many countries of the Southern Hemisphere, this is asso- that exist use a systems approach linking the system’s formal
ciated with change in the wider political sphere (1). This fits inputs and procedures to desired output: normally the coverage
Brazil well. A return to civilian government in the mid-1980s rates of basic preventive services (10–12). However, organiza-
followed more than 20 years of military dictatorship. A new con- tional research indicates a variety of ways in which organizations
stitution (2) promised major restructuring of the health system, can function and can be viewed, with important implications
including fully devolved management (3) combined with local for the variables through which researchers might explore how
participatory decision-making (4). The health system in Brazil this affects system performance (7, 13–15).
is decentralized to the third administrative tier after the federal In particular, the informal aspects of systems need to be
and state levels: the município. explicitly incorporated into evaluation studies. This term covers
The possible pitfalls of decentralization have been how participants in a system behave within that system in
described extensively (5–7), but few question the benefits if response to incentives from outside that system rather than to
appropriately implemented and it remains a ubiquitous policy incentives consciously built into the system. A second element
measure. Despite this widespread popularity, empirical stud- is how the norms and values of the wider society influence the
ies assessing whether the promise of decentralization has been operations of that system (16).
realized are surprisingly rare and mostly depict the variation Thinking in terms of formal incentives, let alone informal,
in the formal structures implemented from the national has only entered mainstream debates on health systems develop-
scale (8). Studies at the regional or district scales that can link ment in recent years (9). Health systems researchers have barely

1
Health Systems Development Programme, University of Manchester, School of Geography, Mansfield Cooper Building, Oxford Road, University of Manchester, M13
9PL, England; and Escola de Saúde Pública, Fortaleza, Brazil (email: [Link]@[Link]). Correspondence should be sent to this author.
2
Tropical Health Division, Liverpool School of Tropical Medicine, Liverpool, England.
Ref No. 02-0242
(Submitted: 18 September 2002 – Final revised version received: 19 February 2004 – Accepted: 25 February 2004 )

822 Bulletin of the World Health Organization | November 2004, 82 (11)

 Research
Sarah Atkinson et al. Decentralization and health system

engaged with the influence of the wider cultural context at all. employment and morale), the management style of the secre-
Prescriptions for improving the effectiveness of decentralization tary of health (behaviour of good management practice (18))
repeatedly return to the formal sphere and highlight the need and the extent of engagement with the population (population
for stronger processes of capacity-building, resource allocation, awareness of community-based activities and the local municipal
incentives and procedures for accountability (10). health council).
This study builds on a comparative ethnography of three
local health systems in north-east Brazil. This demonstrated how Political culture
local informal factors influence local policy implementation (17). Our ethnographic study found that rural and urban districts
These insights are converted, where possible, into quantifiable differed greatly. This spatial distinction is very recognizable to
indicators to explore relationships with the performance of local populations, managers and policy-makers although it captures
health systems across an extensive sample. The study aimed to a broad range of factors. Three socioeconomic indicators (edu-
examine whether decentralization improves the performance of cation and two measures of housing quality) aimed to separate
local health systems. This was investigated by analysing whether these effects from the spatial indicator. Latin American society
a broad range of indicators for formal organization, informal is traditionally based on vertical relations of clientelism (19).
management and political culture are associated with improved Although often covert, we tried to capture its influence at two
performance of local health systems and, if so, their relationship scales. First, the political affiliation of a município to the state
with decentralization. The study was performed in the State of government may facilitate access to state resources for addi-
Ceará, north-east Brazil, which has a strong political commit- tional programmes and activities. Secondly, the local exercise
ment to the national reform agenda (18). of patronage may be acceptable, presenting an obstacle to local
demands for change. Following the model of a large survey in the
Methods United Kingdom (20), five questions assessed the acceptability
Study design to health staff of grey areas of practice by either themselves or
The study defined four groups of variables: performance; formal local politicians. Three of these explicitly explored the accept-
organization; informal management; and local political culture. ability of politicians using the health system for political gain
Table 1 (web version only, available at: [Link] through clientelism. The other two explored the acceptability
bulletin) presents the indicators and the data sources. of practices involving material gain for health staff. Finally,
the commitment of key management figures to the district is
Performance captured by whether the prefect practises his or her original
Performance variables describe health outcomes (clinical out- profession locally and by whether the município secretary of
come and population satisfaction) and health system outputs health was born locally.
(service productivity and coverage and population utilization and
accessibility). Many of these data are available from routine infor- Secondary data
mation systems; primary data are collected on the population’s The State Secretariat of Health for Ceará (SESA) has data for
perspectives on the care provided and their clinical outcomes, each município health system on the productivity of the health
utilization and accessibility. Population perspectives on the care facilities and on financial budgets and expenditure. These data
provided were accessed from two angles: 15 questions about the were made available either directly from the appropriate SESA
aspects of care of a specific ill-health episode (called quality here) departments or through a district-level database (21). Data are
and five questions about the same aspects of health services in available for all municípios.
general (called satisfaction here). Aspects were defined from prior
ethnographic research, and the question responses were scored Questionnaires
on a five-point Likert scale. Composite scores were constructed The primary data were collected through three questionnaires
across the questions (Cronbach’s alpha > 0.8). Four indicators in each município:
were built from these composite scores: the mean scores for • to the município secretariat of health, including general data
quality and for satisfaction and the percentage of the popula- on the município;
tion in each município scoring above the 75th percentile of the • to health staff at each of three health facilities visited; the
total sample on quality and on satisfaction. number of staff sampled at each facility depended in part
on the staff size of the health facility, but a limit of five per
Formal organization facility was set so that the maximum number was 15;
Indicators of formal organization describe inputs (physical and • to women sampled from two neighbourhoods, a total of 100
human), the stage of decentralization (none, partial or full) and per município.
the capacity to process resources (here indicated by the percent-
age of the planned budget actually spent). These data are mostly The questionnaires were piloted and modified to ensure that
available through the routine information system. The quality the questionnaire was both comprehensible to interviewees and
of human resource inputs was assessed through a series of simple of reasonable duration, resulting in 100% response to the com-
questions, based on Brazilian guidelines, about good practice in munity-based questionnaire. Health staff and the município
health care related to antenatal care, respiratory infections among secretariat of health filled in the questionnaire themselves, result-
children, high blood pressure and treatment regimens. ing in lower but reasonable response rates. Data from secondary
sources were available for all municípios.
Informal management
Informed by the previous ethnographic studies (17), the aspects Sampling
of informal management included were characteristics of health The primary data were collected from 45 municípios. The 45
staff in relation to the local health system (continuity of local study sites were intentionally selected to represent stages of

Bulletin of the World Health Organization | November 2004, 82 (11) 823

 Research
Decentralization and health system Sarah Atkinson et al.

decentralization, urban and rural districts and geographical and decentralization against significant variables emerging from
spread across Ceará. the regression analysis. The relationships between decentraliza-
tion and performance indicators were explored using one-way
Health centres analysis of variance, with decentralization entered as a three
In each district three health centres were visited, selected by category variable. Regression analysis was performed with the
these criteria: regression procedure, employing the stepwise method of entering
• one facility in the município centre, preferably a hospital for variables. Associations between decentralization and significant
outpatients or, if none existed, a health centre; variables from the regression analyses were identified using one-
• one facility in a neighbourhood that was not the município way analysis of variance and Duncan’s post-hoc test for significant
centre nor near it and that was not close to major roads, differences between the three levels of decentralization (22–24).
preferably a health centre;
• one facility in a remote neighbourhood relatively under- Results
served by public services in general, preferably a health post. Decentralization and performance
Table 2 presents the mean values for the performance indicators
Community that proved to significantly differ across the three categories of
In each município a sample of 100 women was interviewed, decentralization by analysis of variance. For three of these five
sampled by these criteria: indicators, the difference with decentralization was in the direc-
• 50 women from neighbourhoods relatively near health tion aimed for by health reform. Coverage of preventive care
facilities; (antenatal care and vaccination) increased and utilization of
• 50 women from neighbourhoods far from any health facility hospital-based care decreased. The gains in increased antenatal
in either this or neighbouring municípios; care coverage and decreased use of a hospital facility progressed
• the neighbourhoods selected should represent those con- from not decentralized to partly and then fully decentralized
sidered to be poorest. municípios. Vaccination coverage was significantly higher in
fully decentralized municípios than the others. The percentage
In addition, in rural municípios women were not sampled from change in productivity for basic clinical services was ambivalent
the district centre, and in urban municípios women were not in relation to the reform agenda. Municípios not decentralized or
sampled from any rural areas of the município. Finally, field partly decentralized showed rapid rates of increased productivity
workers were to exclude the houses of affluent people in any compared with a very low percentage change in those fully decen-
neighbourhood visited, but otherwise to sample women on an tralized. Those not fully decentralized may have been building
ad hoc availability basis. their capacity for productivity, whereas those fully decentralized
The primary data were collected between October and were already working near capacity and consolidating service
December 1997. provision rather than expanding. However, the productivity
of basic clinical services was not higher in fully decentralized
Analysis municípios than in other municípios. On the contrary, the partly
The data were entered into the Statistical Package for the Social decentralized municípios had the highest productivity rates,
Sciences (version 10.0). Three stages of analysis are presented: whereas nondecentralized and fully decentralized municípios
decentralization against performance indicators; regression of performed similarly. The gain in performance with decentraliza-
performance indicators against formal organization (including tion was neither progressive nor maintained between the partial
decentralization), informal management and political culture; stage and the full stage.

Table 2. Decentralization and performance

Performance indicator Not decentralized Partly decentralized Fully decentralized

Outcome – user assessment
Quality, mean rating (1 = high, 5 = low) 2.42 2.37 2.30
Satisfaction, mean rating (1 = high, 5 = low) 2.79 2.77 2.56
% rating quality as high 15.6 21.3 28.1
% rating satisfaction as high 21.9 24.4 34.4
% rating the performance of the community health worker as high 24.0 30.6 33.2
Output – preventive care
% of pregnant women attending antenatal care 69.5 78.6 a 83.3
% of infants younger than 11 months vaccinated 74.8 75.2 85.0
Output – clinical care
Productivity per 1000 population for basic clinical services 0.65 1.08 0.48
% change in productivity of basic clinical services, 1996–97 129.4 137.5 41.0
Utilization and access
% attending hospital 76.2 53.4 43.2
a
Figures in italics were significant at the P  0.05 level in Duncan’s post-hoc test for significant differences.

824 Bulletin of the World Health Organization | November 2004, 82 (11)

 Research
Sarah Atkinson et al. Decentralization and health system

These results indicate that decentralization leads to gains ing the nearest health facility. In all these cases, the analysis
in performance in all categories except clinical outcome. However, produces a model involving only one independent variable or,
these gains in local health system performance with decentral- for the percentage attending antenatal care, no model at all.
ization seem rather limited and erratic in expression. Are other Across the performance indicators for which the regres-
factors influencing or mediating these associations? sion variables explained more than 65% of their variance, a mix
of formal inputs, informal management and political culture
Performance, formal organization, informal provided the explanatory variables. This result reinforces the find-
management and political culture ings from organizational studies in general and from previous
We used regression analysis to explore the relative contribution qualitative studies of local health systems in north-east Brazil in
of the variables for formal inputs (including decentralization), particular that aspects of the formal organization are only one
informal management and political culture to each performance influence on organizational performance. Here, the aspects of
indicator (Table 3: web version only, available at [Link] organizational reality captured by the indicators of informal man-
[Link]/bulletin). agement and political culture consistently emerge as important
Twenty-two indicators of performance were tested with influences on all measures of performance. Decentralization was
15 indicators of formal organization, 15 indicators of informal an important contributing variable for only one performance
management and 14 indicators of political culture. A few inde- indicator (percentage rating satisfaction as high). This implies
pendent variables explained a high proportion of the variance that, although decentralization is associated with some indica-
for most performance indicators. Seven or fewer independent tors of better performance of local health systems, it is not a major
variables predicted more than 75% of the variance for 12 per- explanatory factor of the variation in performance between local
formance indicators: mean quality; percentage rating satisfaction health systems compared with other indicators of formal inputs,
as high; percentage rating the performance of the community informal management and political culture.
health worker as high; percentage of infants with low birth
weight; perinatal mortality rate; infant mortality rate; percentage Decentralization and significant regression variables
of children younger than two years old growing well; produc- The independent variables that emerged as significant in the
tivity of basic clinical services, percentage of children younger regression analysis were examined for association with decen-
than two years old weighed; percentage attending any health tralization. Table 4 presents those proving significant (Duncan’s
facility; percentage attending a health facility in this município; post-hoc test) (23). Decentralization was significantly associ-
and percentage who had a problem requiring care. ated with two indicators of formal organization: poorer manage-
Surprisingly, given the complexity of health outcomes, ment of financing for basic and advanced clinical services in fully
the significant variables in the multiple regression explained decentralized municípios. However, this probably reflects how
more than 90% of variance for all the município-scale health financial records are kept. Once fully decentralized, a local health
outcome indicators except for the percentage of users reporting system no longer has to return productivity bulletins to the
that they got better after their consultation. For a further five state secretariat of health for monthly payments to be released;
indicators, more than 65% of the variance was predicted by the expenditure can be more flexible. Thus, an association between
independent variables: percentage who report getting better poorer management of financing and full decentralization is
after consultation; productivity of advanced clinical services; largely a mutual relationship rather than one of influence.
percentage change in productivity of both basic and of advanced Decentralization was significantly associated with five
clinical services; and percentage attending a hospital. indicators of informal management. Four times more people
Only in five cases did the regression explain variance of have heard of the Family Health Programme in fully and partly
40% or less: mean satisfaction; percentage rating quality as decentralized municípios than in those not decentralized. Fully
high; percentage attending antenatal care; percentage of infants decentralized municípios have more people who have been asked
younger than 11 months old vaccinated; and percentage attend- for their opinions by a member of the local health council

Table 4. Decentralization and the significant regression variables against performance: significant relationshipsa

Variables Not decentralized Partly decentralized Fully decentralized

Formal organization
Financial management capacity:
for basic services (0 = good, 8 = poor) 3.92 3.71 6.68
for advanced services 4.15 3.93 7.22
Informal management
Management style
Município secretary of health has unreliable office hoursb 1.95 2.04 1.41
Município secretariat holds meetings with staff 0.55 0.61 0.82
Population awareness of community health activities
% who had heard of Family Health Programme 5.3 19.9 21.1
Population awareness of participatory spaces
% who knew a member of the local health council 3.1 3.1 6.8
% who had a local health council member ask for their opinions 0.5 1.2 3.4
a
Duncan’s post-hoc test for significant differences, P  0.05; figures in italics indicate the stages at which significance operates.
b
% of staff who reported these occurrences.

Bulletin of the World Health Organization | November 2004, 82 (11) 825

 Research
Decentralization and health system Sarah Atkinson et al.

(CMS) and twice as many people who know a member of the performance of local health systems. Trying to quantify com-
CMS as other municípios. Fully decentralized municípios also plex social relations is an inherently reductionist exercise that
have more staff reporting that the município secretary of health can provoke more criticism and disagreement than consensus.
keeps reliable office hours and holds meetings between health Nevertheless, exploring the complex world of health systems
facility staff and the município secretariat of health. across a more extensive sample is useful.
These findings imply that decentralization was associated Does decentralization improve health system perfor-
with improvement in informal management. However, this was mance? No, not per se. Importantly, decentralization was never
not, in turn, always positively associated with improved perfor- associated with worse performance; any association with better
mance. This is specifically true with respect to greater awareness performance appeared to be mediated by aspects of informal
of participatory spaces: a higher percentage of the population management. Other formal inputs played a part in improving
knowing a member of the CMS was associated with fewer chil- health system performance, but this was consistently moderated
dren under two years old being weighed or growing well, and by informal management and political culture. Even where in-
a higher percentage of the population having a member of the formal management practices were desirable, this did not ensure
CMS ask for their opinions was associated with higher peri- a positive impact on performance, as everything is mediated
natal mortality rates. The creation of participatory spaces is a through the political and social context embedding the local
major component of Brazil’s health reform agenda, as in many health system. Decentralization per se is not associated with and
countries. Although associations do not necessarily indicate does not influence this. Thus, formal inputs are insufficient to
influence, these results imply ambivalence in the relationship improve the performance of local health systems. Health system
between desirable management practice in the health system researchers and planners need to discuss tackling management
reform agenda and performance. culture and the relationship of the local health system to the
Finally, decentralization was not associated with political local political culture. Our research demonstrates the value of
culture, whereas many of the indicators of political culture were further detailed local analysis of influences on health system
significantly associated with performance. performance, the potential and limitations of formal inputs
to improve health system performance and the need for policy
Discussion and conclusions initiatives to identify different kinds of complementary inputs
A common-sense interpretation of the association between for the varying contexts of local health systems. O
decentralization and informal management is that good manage-
ment practices lead to municípios achieving a more advanced Acknowledgements
stage of decentralization rather than decentralization leading to The research was funded by a grant from the United Kingdom
better local informal management. Decentralization shows no DFID. Staff from the ESP, Fortaleza, Brazil organized the data
association with local political culture, whereas political culture collection and provided logistical support, information and hos-
substantially influences local health system performance. pitality during the research.
A complex interplay between formal inputs, informal
management and political culture influences variation in the Conflicts of interest: none declared.

Résumé
Revenir à l’essentiel : la décentralisation améliore-t-elle l’efficacité des systèmes de santé ? L’expérience
de l’Etat de Ceará dans le nord-est du Brésil
Objectif Examiner si la décentralisation a amélioré l’efficacité des de l’efficacité des systèmes de santé, mais seulement pour 5 de nos
systèmes de santé dans l’Etat de Ceará, dans le nord-est du Brésil. 22 indicateurs de performance. De plus, dans l’analyse par régression
Méthodes L’Etat de Ceará est fortement impliqué dans la multiple, la décentralisation n’expliquait la variance que pour un seul
décentralisation. Une enquête sur 45 systèmes de santé locaux indicateur ; les indicateurs de gestion informelle et de culture politique
(décentralisés au niveau du município) a permis de réunir des ont paru exercer une influence plus importante. Cependant, certains
informations sur l’organisation officielle des systèmes et sur leur indicateurs de gestion informelle étaient eux-mêmes associés à la
efficacité, notamment en ce qui concerne la décentralisation, la décentralisation, mais aucun des indicateurs de culture politique.
gestion informelle et la culture politique locale. Pour ces deux Conclusion Les bonnes pratiques de gestion observées dans
derniers éléments, les indicateurs étaient basés sur les résultats l’étude ont conduit à décentraliser les systèmes de santé et non
d’une étude ethnographique préalable. L’analyse des données a l’inverse. Toute association apparente entre la décentralisation et
été effectuée par analyse de la variance, test post-hoc de Duncan l’efficacité des systèmes semble un artefact de la gestion informelle,
et analyse par régression multiple. et la culture politique plus large dans laquelle s’inscrivent les
Résultats La décentralisation était associée à une amélioration sytèmes de santé locaux influence fortement leur efficacité.

Resumen
Replanteamiento de una cuestión básica: ¿mejora la descentralización el desempeño de los sistemas de
salud? Evidencia procedente de Ceará, en el noreste del Brasil
Objetivo Determinar si la descentralización ha mejorado el Métodos Ceará ha apostado firmemente por la descentralización.
desempeño de los sistemas de salud en el Estado de Ceará, en el Mediante una encuesta realizada en 45 sistemas locales (município)
noreste del Brasil. de salud se recogieron datos sobre el desempeño y la organización

826 Bulletin of the World Health Organization | November 2004, 82 (11)

 Research
Sarah Atkinson et al. Decentralization and health system

formal, en particular sobre la descentralización, la gestión informal importante. Sin embargo, algunos de los indicadores relativos a la
y la cultura política local. Los indicadores de la gestión informal y gestión informal estaban asociados a la descentralización, lo que no
la cultura política local se basaron en investigaciones etnográficas ocurría con ninguno de los indicadores sobre la cultura política.
anteriores. Los datos fueron sometidos a análisis de varianza, Conclusión En el contexto de este estudio, las buenas prácticas
prueba de Duncan post hoc y análisis de regresión múltiple. de gestión condujeron a unos sistemas locales de salud
Resultados La descentralización se asoció a un mejor desempeño, descentralizados, en lugar de a la inversa. Cualquier relación
pero sólo en 5 de nuestros 22 indicadores de desempeño. Además, aparente entre descentralización y desempeño parece ser un
en la regresión múltiple, la descentralización explicó la varianza de artefacto de la gestión informal, y el marco general de cultura
sólo un indicador del desempeño; los indicadores sobre la gestión política en el que se inscriben los sistemas locales de salud influye
informal y la cultura política parecían tener una influencia más muy marcadamente en el desempeño de esos sistemas.

References
1. Decentralization and health systems change: a framework for analysis. 11. Barnabas G-A. Local government, equity and primary health care in Ethiopia
Geneva: World Health Organization; 1995. WHO document WHO/SHS/ [dissertation]. London: University of London, London School of Hygiene and
NHP/95.2. Tropical Medicine; 1997.
2. Constituição da república federativa do Brasil: de 5 outubro de 1988 (3a 12. Gonzalez-Block M, Layva R, Zapata O, Loewe R, Alagón J. Health service
edição) [Constitution of the Federal Republic of Brazil: 5 October 1988. 3rd decentralization in Mexico: formulation, implementation and results of
ed.]. São Paulo: Editora Atlas; 1988. In Portuguese. policy. Health Policy and Planning 1989;4:301-15.
3. Rondinelli D. Implementing decentralization programmes in Asia: a 13. Shafritz JM, Ott JS. Classics of organization theory. 3rd ed. Wadsworth:
comparative analysis. Public Administration and Development 1983;3:181-207. Belmont; 1992.
4. de Carvalho GI, Santos L. Sistema Único de Saúde: comentários à Lei 14. Abrahamsson B. The logic of organizations. Newbury Park, CA: Sage; 1993.
Orgânica da Saúde (Leis 8.080/90 and 8.142/90) [Single health system: 15. McKevitt D, Lawton A. Public sector management: theory, critique and
commentaries on the Health Act – Laws 8.090/90 and 8.142/90]. practice. Newbury Park, CA: Sage; 1994.
São Paulo: Editora Hucitec; 1992. In Portuguese. 16. North DC. Understanding the process of economic change. London:
5. Collins C. Decentralization. In: Janovsky K, editor. Health policy and systems Institute of Economic Affairs; 1999.
development. Geneva: World Health Organization; 1995. WHO document 17. Atkinson S, Medeiros RLR, Oliveira PHL, Almeida RDD. Going down to the
WHO/SHS/NHP/96.1. p. 161-78. local: incorporating social organisation and political culture into
6. Collins C, Green A. Decentralization and primary health care: some negative assessments of decentralised health care. Social Science and Medicine
implications in developing countries. International Journal of Health 2000;51:619-36.
Services 1994;24:459-75. 18. Tendler J. Good governance in the tropics. Baltimore: Johns Hopkins
7. Atkinson S. Restructuring health care: tracking the decentralization debate. University Press; 1997.
Progress in Human Geography 1995;19:486-503. 19. Eisenstadt SN, Roniger L. Patrons, clients and friends. Cambridge:
8. Mills A, Vaughan JP, Smith DL, Tabibzadeh I. Health systems decentralization: Cambridge University Press; 1984.
concepts, issues and country experiences. Geneva: World Health 20. Sheaff R, West M Marketization, managers and moral strain: chairmen,
Organization; 1990. directors and public service ethos in the National Health Service. Public
9. Savedoff WD. Social services viewed through new lenses. In: Savedoff WD, Administration 1997;75:189-206.
editor. Organization matters: agency problems in health and education in 21. IPLANCE. Ranking dos municípios [Ranking of the municípios]. Fortaleza:
Latin America. Baltimore: Johns Hopkins University Press; 1998. p. 3-35. Fundação Instituto do Planejamento do Ceará; 1997. In Portuguese.
10. Thomason JA, Newbrander WC, Kolehmainen-Aitken RL, editors. 22. Kinnear PR, Gray CD. SPSS for Windows made simple. 3rd ed. Hove:
Decentralization in a developing country: the experience of Papua New Psychology Press; 1999.
Guinea and its health service. Canberra: National Centre for Development 23. SPSS. User’s guide SPSS Base 8.0 for Windows. Chicago: SPSS Inc.; 1988.
Studies, the Australian University; 1991 (Pacific Research Monograph 25). 24. Sokal RR, Rohlf FJ. Biometry. 2nd ed. New York: W.H. Freeman; 1981.

Bulletin of the World Health Organization | November 2004, 82 (11) 827

 Research
Sarah Atkinson et al. Decentralization and health system

Table 1. Indicators used in the survey of local health systems

Indicators of performancea
Outputs Data sources
Productivity of services Routine productivity data provided by the
Productivity of clinical care coverage (basic and advanced) State Secretariat of Health to the Out-
Change in the productivity of clinical care coverage (basic and advanced) from 1995 to 1996 patient Information System of the Brazilian
Productivity of preventive care coverage (antenatal care, infants younger than 11 months old Unified Health System (SIA/SUS) (21)
vaccinated, children younger than two years old weighed)
Utilization and accessibility for the population (women) b Community-based survey, data collected
% attending any health facility October–December 1997 in 45 districts,
% attending a hospital n = 100 women in each
% attending a health facility in their own district
% attending the nearest health centre
% having a problem requiring care at a health facility
Outcomec
Clinical outcome State Secretariat of Health, Ceará, print-
Low birth weight, children younger than two years old growing well, perinatal and out of 1996 community health worker data
infant mortality rate
% reporting they got better after consulting the health services
Community evaluation Community-based survey, data collected
Population (women) rating of satisfaction and quality of services d (based on five-point October–December 1997 in 45 districts,
Likert scales): e n = 100 women in each
% rating high satisfaction with care; % rating high quality on specific questions of provision;
mean score on satisfaction questions; mean score on specific questions on quality of provision;
and % rating the performance of community health workers as high
Indicators of formal organization
Management structure
Stage of decentralization (none, partial, full) Escola de Saúde Pública, Fortaleza,
Two categories are used for analysis: not decentralized versus decentralized: not decentralized Ceará records
and basic decentralization versus full decentralization
Resource inputs
Health facilities per 1000 population IPLANCE (21)
Beds per 1000 population
Public facilities per 1000 population District secretariat of health survey, data
Private facilities per 1000 population collected October–December 1997
Public staff per 1000 population
Private staff per 1000 population Health centre survey, data collected
Public–private ratio of facilities October–December 1996
Public–private ratio of staff
Staff knowledge of correct procedures (based on national and state protocols) for: antenatal
care, respiratory infections among children, high blood pressure and treatment regimens
Financial management capacity
Management of financing (for basic and advanced services) State Secretariat of Health, Ceará, print-
Capacity scores = % of planned budget spent (for basic and for advanced services) in each out of 1996 budget and expenditure data
district and then categorized by difference from 100%e
90–110%: +1; 80–90% or 110–120%: +2; etc. to 0–30% or 170%: +8
Indicators of informal organization of the health system
Staff characteristics (of those interviewed)
Mean time spent working in this district Health Centre Survey data collected
Mean staff satisfaction scoree October–December 1996
District management style (staff perceptions)
% who think município secretary of health keeps reliable office hourse
% who have health facility staff meetings
% who have staff meetings with the district secretariat
% of staff who think that secretary of health makes decisions together with others
% staff who think that secretary of health asks for advice from others
% staff who think that secretary of health shares information with others
Who makes the decisions on the local health council according to health facility staff

Bulletin of the World Health Organization | November 2004, 82 (11) A

 Research
Decentralization and health system Sarah Atkinson et al.

(Table 1, cont.)
Indicators of performancea
Population awareness of community health activities
% who know the community health worker Community-based survey, data collected
% who had a community health worker come to their house October–December 1997 in 45 districts,
% who had heard of the Family Health Programme n = 100 women in each
Population awareness of participatory spaces
% who had heard of the local health council
% who knew a member of the local health council
% who had a member of the local health council ask for their opinions
Indicators of political culture
Geographical
Zone: urban versus rural IPLANCE (21)
Political affiliation
Prefect (mayor) is a member of the same party as the state governor District secretariat of health survey, data
collected October–December 1997
Management’s commitment to district
Whether the prefect was born in the district District secretariat of health survey, data
Whether the prefect exercises his or her profession in the district collected October–December 1997
Whether the secretary of health was born in the district
Whether the secretary of health lives in the district
Norms and values of health staff
Acceptability to staff of certain practicese Health centre survey data collected
Politicians help clients to gain preferential access to health resources October–December 1996; secretariat of
Community health workers involved in political campaigns health survey, data collected October–
Politicians keep drugs in their homes for distribution December 1997
Staff refer patients to their own private clinics
Staff get informal material gain from the health services
Socioeconomic status of the population
% of women with any education Community-based survey, data collected
% of houses made of mud October–December 1997 in 45 districts,
% of houses with a mud floor n = 100 women in each
a
Data availability varies from 48% to 66% for data collected from the município secretariat of health.
b
In the context of the aims of the health reforms, indicators of utilization are interpreted as being positive if districts have a higher % seeking a health facility for
care, seeking one that is not a hospital, seeking one in their own district and seeking the nearest one. Accessibility issues are also indicated by whether seeking
health care involves transport costs, payment at the chosen health facility or any problem experienced in being consulted.
c
The figures collected by the community health workers refer only to those registered by them and not to the whole district population. However, the community
health workers operate in the poorer neighbourhoods of the districts, which is where the impact of reforms is of most interest. These figures are by far the most
reliable of any available.
d
Community satisfaction with services is determined from two slightly different angles: questions on aspects of care identified through the qualitative work as
being important to users, with specific reference to two stages of the care process — the preconsultation and the consultation processes — referred to here as
quality. Questions on aspects of care identified through the qualitative work as important to users, with reference to the process as a whole, are referred to here
as satisfaction. The difference is in method, and the language distinction facilitates clarity in analysis rather than representing any conceptual difference.
e
A lower score indicates better results for: mean quality and mean satisfaction; financial management capacity for basic and for advanced clinical services; staff
morale; the município secretary of health keeps reliable office hours; and acceptability to staff of certain practices.

B Bulletin of the World Health Organization | November 2004, 82 (11)

 Research
Sarah Atkinson et al. Decentralization and health system

Table 3. Regression of performance indicators on decentralization, informal organization and political culture

Standardized t P-value
coefficient (β)
Outcomes — user assessment
Dependent variable: mean quality rating (1 = high)
Variable
Município secretary of health lives there –0.72 –6.5 < 0.0001
Settlement type –0.51 –4.9 0.001
House made of mud +0.36 +3.5 0.006
Município secretary of health shares information –0.33 –2.9 0.016
R 2 = 0.901
Dependent variable: mean satisfaction rating (1 = high)
Variable
Management of financing for advanced clinical services –0.63 –2.95 0.011
R 2 = 0.401
Dependent variable: % rating quality as high
Variable
Município secretary of health lives there +0.614 2.80 0.015
R 2 = 0.377
Dependent variable: % rating satisfaction as high
Variable
Settlement type +0.760 4.95 <0.0001
Decentralization +0.490 3.28 0.007
Município secretary of health consults others +0.366 2.39 0.036
R 2 = 0.756
Dependent variable: % rating the performance of the
community health worker as high
Variable
Private facilities per 1000 population –0.838 –8.42 <0.0001
Acceptability to staff that politicians store drugs –0.424 –4.01 0.002
Município secretary of health lives there +0.460 +4.48 0.001
Public facilities per 1000 population –0.299 –2.82 0.018
R 2 = 0.909
Outcomes — clinical
Dependent variable: % reporting that they got better
after consultation
Variable
Floor made of mud +0.597 3.41 0.005
% who had heard of local health council –0.394 –2.25 0.044
R 2 = 0.676
Dependent variable: % with low birth weight
Variable
Public facilities per 1000 population –0.629 –13.704 <0.0001
Ratio of public to private staff +0.599 +10.43 <0.0001
Município secretary of health lives there +0.501 +13.03 <0.0001
Public staff per 1000 population –0.595 –11.58 <0.0001
Unreliability of office hours of município secretariat of health +0.346 +7.83 <0.0001
Acceptability to staff that community health workers are involved in +0.188 +4.36 0.003
political campaigns
Settlement type +0.150 +2.67 0.032
R 2 = 0.991
Dependent variable: perinatal mortality rate
Variable
% who had a member of the local health council ask for their opinions +0.749 7.04 <0.0001
% who know a community health worker +0.406 4.15 0.002
% who had heard of local health council +0.363 3.64 0.005
% who know that the Family Health Programme exists –0.267 –2.55 0.029
R 2 = 0.910

Bulletin of the World Health Organization | November 2004, 82 (11) C

 Research
Decentralization and health system Sarah Atkinson et al.

(Table 3, cont.)
Standardized t P-value
coefficient (β)
Dependent variable: infant mortality rate
Variable
Meetings with município secretary of health –0.568 –5.65 <0.0001
Staff satisfaction +0.589 +5.80 <0.0001
Prefect practises profession in município +0.411 +4.04 0.002
Public facilities per 1000 population –0.341 –3.37 0.007
R 2 = 0.902
Dependent variable: % of children younger than two years
old growing well
Variable
Floor made of mud –1.058 –11.43 <0.0001
% who know a member of the local health council –0.554 –5.82 <0.0001
% who know a community health worker +0.509 +5.99 <0.0001
Acceptability to staff that politicians store drugs +0.727 +5.78 <0.0001
Acceptability to staff that community health workers are involved in –0.450 –3.74 0.006
political campaigns
Private facilities per 1000 population +0.237 +2.70 0.027
R 2 = 0.951
Outputs — preventive care
Dependent variable: % of infants younger than 11 months
old vaccinated
Variable
Acceptability to staff that politicians store drugs +0.620 2.851 0.014
R 2 = 0.385
Dependent variable: % of children younger than two years
old weighed
Variable
Acceptability to staff that politicians store drugs +0.790 19.08 <0.0001
Staff knowledge of treatment regimens +0.484 8.67 <0.0001
% who know a member of the local health council –0.456 –11.05 <0.0001
Ratio of public to private facilities –0.161 –3.59 0.007
Prefect practises profession in the município –0.252 –5.47 0.001
% of women with any education –0.179 –2.90 0.020
R 2 = 0.990
Outputs — clinical care
Dependent variable: productivity of basic clinical services
per 1000 population
Variable
Management of financing for basic services –1.119 –7.66 <0.0001
Settlement type –0.587 –5.55 <0.0001
Acceptability to staff of participating in political campaigns +0.803 +5.20 0.001
% who know that the Family Health Programme exists +0.312 +2.92 0.017
R 2 = 0.914
Dependent variable: productivity of advanced clinical
services per 1000 population
Variable
Private staff per 1000 population +0.727 4.16 0.002
Meetings with município secretary of health +0.427 2.45 0.032
R 2 = 0.666
Dependent variable: % change in productivity of basic
clinical services, 1996–1997
Variable
Floor made of mud +0.666 4.13 0.002
Management of financing for basic clinical services –0.494 –3.06 0.011
R 2 = 0.714

D Bulletin of the World Health Organization | November 2004, 82 (11)

 Research
Sarah Atkinson et al. Decentralization and health system

(Table 3, cont.)
Standardized t P-value
coefficient (β)
Dependent variable: % change in productivity of advanced
clinical services, 1996–1997
Variable
Acceptability to staff of referral to a private clinic +0.649 +3.50 0.006
% who had heard of the local health council –0.471 –2.55 0.029
R 2 = 0.657
Utilization and access
Dependent variable: % attending any health facility
Variable
Unreliability of the office hours of the município secretariat of health –0.431 –2.75 0.021
Staff satisfaction –0.900 –4.80 0.001
Public staff per 1000 population –0.723 –3.82 0.003
Style of decision-making of the município secretary of health –0.457 –2.48 0.032
R 2 = 0.820
Dependent variable: % attending hospital
Variable
Ratio of public to private facilities –0.545 –3.22 0.008
% who know that the Family Health Programme exists –0.447 –2.64 0.023
Floor made of mud +0.422 +2.49 0.030
R 2 = 0.685
Dependent variable: % who attended facility in this município
Variable
Management of financing for advanced clinical services 0.850 4.98 <0.0001
House made of mud 0.494 3.47 0.005
Average time staff worked in município –0.381 –2.23 0.047
R 2 = 0.777
Dependent variable: % attending nearest facility
Variable
% who had heard of local health council 0.603 2.73 0.017
R 2 = 0.364
Dependent variable: % who had a problem that required
health care
Variable
Município secretary of health consults others +0.650 4.15 0.002
Município secretary of health lives there –0.580 –3.79 0.003
Whom staff consider makes local health council decisions in reality –0.407 –2.59 0.025
R 2 = 0.745

Bulletin of the World Health Organization | November 2004, 82 (11) E