Fiscal Governance and Public Services:
Evidence from Tanzania and Zambia
Barak D. Hoffman
Clark C. Gibson
Department of Political Science
University of California, San Diego
September 2005
Abstract
Does a government’s source of revenue explain its policies? The predominate view in
development studies contends that policy variation results directly from institutional
variation. Building on a literature which we label fiscal theories of governance, we argue
that a government’s sources of revenue strongly affect its public expenditures,
independent of institutions. Using data from local government budgets in Tanzania and
Zambia, we find that local governments in both countries produce more public services as
their budget’s share of local taxes increases. Alternatively, revenue that local
governments receive from sources outside their boundaries -- transfers from the central
government and foreign assistance -- increases the share of local budgets consumed by
employee benefits and administrative costs. Because there is no variation in the powers
of local governments in Tanzania and Zambia, the effects of revenue sources on public
expenditure that we find are independent of political institutions. The results suggest that
fiscal accountability is an important factor to consider when designing policies to
enhance local government capacity.
Barak Hoffman
UCSD
9500 Gilman
La Jolla CA 92093-0521
858.248.9087
bdhoffma@ucsd.edu
Clark Gibson
UCSD
9500 Gilman
La Jolla CA 92093-0521
858. 822.5140
ccgibson@ucsd.edu
I. Introduction
Arumeru District and Monduli District in Tanzania share a number of similarities. The
adjacent districts have identical political institutions, are overwhelmingly agricultural,
and are relatively free from ethno-religious conflict. In fact, on nearly every dimension
that economists, political scientists, and development practitioners have claimed as
important for development - endowments, political institutions, geography, climate,
economic structure, and ethno-religious relations - Arumeru and Monduli are practically
identical. How then are we to explain the enormous variation in levels of development
that exist between the two districts? The literacy rate in Arumeru is 72% whereas the
literacy rate in Monduli is 39%; the net primary enrollment rate is 79% in Arumeru and
45% in Monduli; 70% of households have a radio in Aruemru but only 35% have a radio
in Monduli.
The gulf between Arumeru and Monduli’s development levels is not unique. Within
Tanzania, district literacy rates range from 27% to 92% and net primary school
enrollment rates range from 37% to 93%. In neighboring Zambia these puzzling
variations also exist: life expectancy ranges from 34 years to 59 years and infant
mortality range from 7% to 20%. How do we explain such different levels of
development within these countries?
In this study we explore one of the causes of these disparities. Building on a literature
which we label fiscal theories of governance, we argue that a local government’s revenue
source is central to explaining its policy choices. We find that local governments in
Tanzania and Zambia produce more public services as a share of total local government
expenditure as local revenue increases. Alternatively, as central government transfers
and foreign aid increase, public services as a share of the local government budget falls.
Not only does this effect remain after controlling for socioeconomic factors but, more
importantly, the influence of sources of revenue on local government expenditure that we
find is independent of political institutions because no variation in political institutions at
the local level in either country.
This study furthers our understanding of the politics of development in several important
ways. First, unlike the larger set of studies that link political institutions to policy and
development outcomes, we examine the effects of revenue on public expenditure under
identical political institutions. The main thrust of the political institutional approach is to
tie policy outcomes to political institutions; our approach demonstrates that there is
another source of political accountability independent of formal political institutions: the
fiscal link between the government and the governed. Understanding these fiscal links,
we argue, is central to understanding development in general and the political effect of
foreign aid in particular.
Second, our research design allows us to test directly the influence of revenue streams on
government choices. We use district level government budget data from Tanzania and
Zambia, two countries where there is no variation in political institutions at the local
level. Our research design is a methodological advance over most quantitative studies of
how revenue sources affect policy choices that use cross national data and rely on control
variables to account for the differences in political institutions across countries. Further,
because local level political institutions do not vary within either country but local level
political institutions vary widely between the two countries - local governments in
Tanzania are dominated by the center while Zambia’s district governments enjoy greater
powers – we are also able to explore the strength of revenue effects on policy outcomes
under two different sets of institutions. We believe this is the first study to implement
such a research design.
Third, we use the entire local government budget in our research. Existing studies that
attempt to explain sub-national variation in public policy outcomes tend to examine
single sectors. Such research could produce misleading results because demand for
different types of public services varies widely and this variation manifests itself in wide
differences in expenditures across different types public services. Using the entire local
budget provides a comprehensive analysis of how sources of revenue affect policy
choices
The results of this study bear directly on current development policy. Through programs
of institutional reform, donors are placing considerable effort into increasing the
accountability of local governments in these two countries. Our results predict that the
success or failure of these policies depends partly on the structure of local government
revenue. Specifically, institutional reform that results in substituting local taxes for donor
and/or government transfers, as recently occurred in Tanzania, could undermine efforts to
increase local government accountability.
We present our study in six parts. We review the theoretical foundations of our approach
in section II. In section III, we provide background information on the countries we use
to test our hypotheses, Tanzania and Zambia. Section IV presents the data and methods
that we use in testing our fiscal theory of governance. We discuss the results in section
V. Section VI concludes.
II. Theoretical Foundations
Political Institutions
The provision of public services is a central topic in political science. The vast majority
of this work explores how political institutions affect government policies. Much of this
research is cross-national and centers around how differences in types of democratic
institutions such as political systems (e.g. presidential versus parliamentary), electoral
rules (e.g., first past the post versus run-offs), electoral systems (single-member districts
versus multimember districts), the degree of separation of powers (veto points), and the
relationship between local governments and central governments affect public policy
outcomes (see Bardhan 2001, Bardhan, and Mookherjee 2005b, Carey and Shugart 1995,
Cox 1997, Linz 1990 and 1994, Lijphart 1999, McIntyre 2003, Persson and Tabellini
2003; Perrson et al., 2003, Perrson et al., 2005, Shugart and Carey 1992, Shugart and
Mainwaring 1997, Tsebelis 2002).
2
At the sub-national level, scholars have developed a large body of research that examines
the consequences of decentralization for political accountability.1 Studies of
decentralization generally explore how changes in the political power of local
governments shape public policy outcomes (Rondinelli, et al. 1989, Inman and Rubinfeld
1996, Ferejohn and Weingast 1997). Studies that focus on the effect of decentralization
on the provision of public services generally seek to adjudicate between two hypotheses.
Proponents of decentralization argue it should increase accountability because local
governments are more accountable that distant central governments. Opponents of
decentralization argue that decentralization may reduce local provision of public services
because local elites can divert public funds more easily through a decentralized system of
than a centralized one.2
Evidence exists to support both hypotheses; many of the studies find evidence to support
both hypotheses within the same country and even within the same village (e.g., Bardhan
and Mookherjee 2004; Galasso and Ravillon 2005). Reinikka and Svensson (2004), for
example, examine the effect of a school capitation program in Uganda and found that
central government grants to schools benefited students mainly in well-off districts while
local elites were able to divert the capitation grants in relatively poorer districts. Galiani
et al. (2005) present similar results in their study of Argentine school decentralization.
According to their study, decentralization facilitated the ability of elites to divert funds in
poor towns in poorly-run provinces while decentralization led to more active oversight of
schools in non-poor towns in well-run provinces. Most surprisingly, Bardhan and
Mookherjee (2004) and Galasso and Ravillon (2005) found evidence of elites diverting
funds and increased accountability within villages.3
These studies are an important addition to our understanding of the links between public
policy, institutions, and accountability. Generally the studies find that local factors, such
as level of development, inequality, and the ability of citizens to participate in local
politics, are important intervening factors that determine how decentralization affects the
distribution of public services. Such findings point to the limitations of institutions as
explanatory variables for public policy outcomes.
Fiscal Theories of Governance
Political institutions are not the only mechanisms that political scientists have employed
to explain public policy outcomes. A smaller but important set of studies explores how
the fiscal relationship between the government and the governed can help explain public
1
See Ahmed et al. (2005) and Bardhan and Mookherjee (2005b) for a review of these studies.
See Bardhan (2001), Bardhan and Mookherjee (2005a), and Bardhan and Mookherjee (2005b) for a
comprehensive summary of this debate.
3
While Galasso and Ravillon (2005) find that the larger the proportion of poor people in the village the
greater the share given to the poor, they also find that as village inequality rises, the proportion going to the
poor falls. Along the same lines, Bardhan and Mookherjee (2004) found that while village governments in
West Bengal, India were far better at targeting the poor than were higher levels of government but also
found some evidence of elite capture.
2
3
policy outcomes; we call these fiscal theories of governance. Fiscal theories of
governance argue two central points. First, the shape of political institutions reflects a
government’s need for revenue (e.g., Bates and Lien 1985; Levi 1988; Moore 1995).
This line of thinking is found most prominently in the work of scholars seeking to explain
the evolution of state structure. In general, this approach holds that a government has
incentives to defer to its citizens’ policy preferences when it is dependent on its citizens
for revenue. Alternatively, when a government is not dependent on its citizens to raise
revenue, the government has fewer incentives to defer to its citizens’ policy preferences
(Moore 1998). Second, taxpayers benefit from government policies roughly in
proportion with the share of government revenue they finance (e.g. Bates and Lien 1985;
Boix 2002; Lindert 2004).
Fiscal theories of governance have considerable power in explaining political outcomes.
Levi (1988), for example, uses evidence from ancient Rome, England and France in the
Middle Ages, 18th Century Britain, and modern Australia to demonstrate that since tax
payments are to a certain extent voluntary, governments need to create compliance (or
cede policy making power) in order to generate revenue. Other scholars have used fiscal
theories of governance to develop compelling arguments to account for the rise of
democratic political institutions in Europe (e.g., Bates 2001, Downing 1992, North and
Weingast 1989, Root 1992, Tilly 1992). According to Tilly (1992), for example, as wars
in Europe became more expensive, raising revenue and troops through coercion became
an increasingly inefficient strategy. As coercion became a less effective strategy for
raising resources needed to fight wars, monarchs sought to generate revenue through
policy concessions and policies that facilitated economic development. Bates and Lien
(1985) use a formal model to predict a similar outcome and make the important point that
the more mobile capital is the more policy making power a government must cede in
order to generate revenue. These works help us to understand that origins of public
revenue can have a dramatic impact on the structure and policies of government, as well
as the developmental outcomes associated with those policies.
Studies from politics in rentier economies extend this logic to assert that external funds
(i.e., funds not raised from the domestic population) should impede the development of
democracy since valuable external resources, such as oil, reduce the dependence of the
government on the governed (e.g., Karl 1997; Ross 2001; Tornell and Lane 1998).
According to Ross (2001), there are three causal mechanisms that lead from oil and
mineral dependence to reduced political accountability. The first is the rentier effect.
The rentier effect inhibits democratic accountability because governments that derive
their revenue from easily exploitable natural resources do not need to tax their
populations and therefore face no fiscal pressure from their citizens to produce public
services. The second mechanism is the repression effect where external sources of
revenue give the state greater coercive power. The third effect is the modernization
effect. Because a rentier economy tends to inhibit industrial development (see Sachs and
Warner 1998), resource-based economies are not subject to economic forces that often
catalyze citizen demand for democratic governance, such as education, industrialization,
and urbanization.
4
A number of scholars have extended the logic of rentier economics to suggest that foreign
aid may generate the same political incentives as rentier commodities, especially if
conditions on aid are weakly enforced (Brautigam 2000, Coolidge and Rose-Ackerman
1997, Knack 2000, Moore 1995, and Svennson 2000). Moore (1995) argues that foreign
aid, like easily exploitable natural resources, reduces the need for the government to
collect taxes and as a result, reduces fiscal accountability. Similarly, Brautigam (2000)
argues that long-term dependence on foreign aid undermines the quality of governance,
reduces pressure for reform and accountability, and diminishes effort to collect taxes.
Triesman (2000) extends this logic to transfers from the central government to local
governments.
Our Approach
We borrow from fiscal theories of governance to explain patterns of local government
expenditure in Tanzania and Zambia. Unlike studies that examine the effect of political
institutions on public policy outcomes, political institutions in our sample do not vary at
the local level. Moreover, we also diverge from decentralization studies that seek to
determine the effect of a change of institutions (e.g., from more centralized to less
centralized) on the supply of public services. Finally, we also depart from studies using
fiscal theories of the governance to investigate how sources of public revenue influence
political institutions. Rather, we examine how sources of revenue affect public policy in this case public expenditure - when there is no variation in political institutions.
Following theories of fiscal governance, we hypothesize that the more local revenue a
local government collects, the larger the share of public services in the local government
budget. Alternatively, we hypothesize that the more transfers and foreign aid a local
government receives, the smaller the share of public services in the local government
budget. As long as there are local elections, we expect this relationship to hold
independent of political institutions.
Using district level data from Tanzania and Zambia offers a major advantage over crossnational studies that have attempted to understand the effects of revenue streams on
public policy outcomes.4 Unlike cross-national studies, we do not need to rely on control
variables to account for differences in political institutions across countries because the
political institutions for our unit of analysis - district governments - do not vary within
each country.5 Because political institutions are orthogonal to our variables of interest
(sources of revenue for local governments), we can be certain that our results do not
reflect any correlation between revenue sources and political institutions.
Finally, one benefit of our study compared to the studies we have cited that examine the
effect of decentralization on public service delivery is that we examine the entire local
budget, not just a specific program or policy area. This is an advance because studies that
examine the effect of decentralization on one program or area face the difficulty that
demand for different public services varies widely. For example, Azfar, et al. (2001)
4
5
We study each country separately.
See Snyder (2001)
5
found that in Uganda, sub-national preferences were far stronger for primary education
than for immunization. Local leaders, in turn, were far more sensitive to the quality of
primary education than to the quality of immunization. As a result, separate studies of
each program would lead to drastically different results. By using the entire local budget,
we can make the far more simple and plausible assumption that the majority of the public
would prefer that the local government allocate funds to public services rather than to
salaries and administrative costs.
III. Tanzania and Zambia
Tanzania and Zambia share many important similarities that make them excellent cases
for this study (see table 1). First, both countries are in the same agro-climactic region.
Second, both countries exhibit quite similar economies and levels of development. Out
of 174 countries, Tanzania is ranked 162 and Zambia is rankled 164 on the UNDP’s
Human Development Index. Given their poverty it is not surprising that both countries
are very large aid recipients. Third, the two countries have a broadly similar political
history. Both countries are former British colonies and attained independence at about
the same time. Moreover both countries were one-party quasi-Socialist regimes in the
1970s and 1980s, democratized in the early 1990s, and are relatively free of ethic and/or
religious conflict.6 Currently, Tanzania and Zambia have very similar ratings in terms of
the two most popular measures of democracy, Freedom House and Polity. The critical
political difference between the two today is that the ruling party in Tanzania during the
single-party regime has won the first two multi-party elections (1995 and 2000) while
Zambia’s single party was replaced through multiparty elections in 1991. Fourth,
Tanzania and Zambia have similar electoral systems at the national level and local level.
Both countries have direct elections for president, elect members of parliament in single
member districts using a first-past-the-post system, elect councilors at the local level, and
allow for reelection for president, parliament, and local councilors.
6
The Minorities at Risk (MAR) database does not view any sub-national group within mainland Tanzania
or Zambia as an immediate risk for rebellion. However, it is important to note that MAR data suggest
Zanzibar in Tanzania and the Lozi in Zambia could be future threats.
6
Table 1: Economic, Political and Social Comparison between
Tanzania and Zambia
Tanzania
National
Population
Urban Population as a Share of Total Population
Life Expectancy
Fertility Rate
Infant Mortality Rate
Child Mortality Rate
Adult HIV Infection Rate
Adult Literacy Rate
Net Primary Enrollment
Primary Completion Rate
Per Capita GDP (Nominal)
Per Capita GDP (PPP)
Agriculture as a Percent of GDP
UNDP Human Development Rank (out of 174)
Polity Score
Freedom House Rating
Democratic Transition
Corruption (Transparency International)*
ICRG Composite Investment Risk**
Index of Economic Freedom (Heritage Foundation)
Aid/GDP
Aid/Per Capita
Local
Average Population Per District
District Employee Benefits Per Capita
District Other Charges Per Capita
District Government Consumption Per Capita***
District Government Consumption a Share of GDP
Probability Recurrent Costs are Equal (p-value)
Zambia
36 million
34%
43
5
10%
17%
9%
78%
68%
58%
300
580
44%
162
2
Partly Free
1995
2.8
58
Mostly Not Free
13%
47
10 million
40%
37
5
10%
18%
16%
80%
69%
58%
380
840
22%
164
1
Partly Free
1991
2.6
48
Mostly Not Free
18%
54
280,000
$11.43
$6.28
$17.71
5.9%
140,000
$7.35
$4.64
$11.99
3.2%
<.01
* Scale 1-10 (worst to best)
** Scale 0-100 (highest risk to lowest risk)
*** Government Consumption equals Employee Benefits plus Administrative Costs
The key institutional difference for the purposes of our study between the two countries is
that while local governments in Tanzania are almost wholly reliant on the central
government for funding, local governments in Zambia are much more fiscally
autonomous. About 90% of district government revenue in Tanzania comes from central
government transfers whereas only about 15% of local government revenue in Zambia
comes from central government transfers. Not surprisingly, local governments have
more political independence from the central government in Zambia than in Tanzania.
This difference is important for our analysis because our results demonstrate that fiscal
incentives of local governments operate independent of their link to the central
government and independent of the share of local the local budget financed by local
revenue.
7
To be clear, we are not comparing Tanzania to Zambia. We are looking for similar
patterns in local government behavior across local governments in two countries with
different political institutions. We study two countries because we want to be certain that
our results are not idiosyncratic to one country or one set of institutions. Moreover, if
local governments react the same way irrespective of the power of local government, then
we will have evidence or a more general phenomenon.
IV. Data and Methods
Unit of Analysis and Data Sources
Our unit of analysis is the district. Districts in Tanzania and Zambia are roughly the
equivalent of counties (or cities that are their own county) in the United States. There are
116 districts in mainland Tanzania and 72 districts in Zambia. On average, the
population of districts in both Tanzania and Zambia is about 200,000.
In Tanzania the source of the data are the Medium Term Expenditure Frameworks
(MTEFs) for each district in the country. MTEFs typically include detailed budget
information that identify the source of revenue (local, central government transfer, and
external assistance) and expenditures by individual components. MTEFs are completed
annually and use a three-year rolling budget form of planning.7 In Zambia our data
sources are local government budgets. Currently, local governments in Zambia produce
detailed budgets on district revenue and expenditure.
Budgets in Tanzania and Zambia report three general categories of local government
expenditure. The first is personal emoluments which are salaries and other direct
employee benefits. The second is administrative costs, such as supplies, maintenance,
and vehicles. The sum of personal emoluments and administrative costs are total
recurrent costs or government consumption. The third category of expenditure is public
services. Public services are all expenditures that are aimed at district improvement.
While public services includes public-type goods, such as education and health, public
services also include narrowly-targeted benefits, such as bus shelters and grants to certain
community groups, as well as services that benefit the public but have a low social return,
such as maintenance of parks.
It is important for us to note that we face no problem separating expenditures financed by
local revenue from expenditures financed by central government transfers and donor
funds. Local governments, in general, have discretion only over funds they collect from
their own constituents because almost all transfers from donors and the central
government are earmarked for specific expenditures and the local government budgets
identify those earmarked expenditures.
7
The MTEFs also review past budget performance by examining actual expenditure compared to estimated
expenditure and whether the district fulfilled its project objectives as stated in under the estimated
expenditures. MTEFs vary in quality. The two areas with the least detail tend to be (1) detailed local
government expenditures from own-source revenue; and (2) donor/NGO expenditures.
8
To test our hypotheses, we use budget estimates, as passed by district councils, rather
than actual budget expenditures. Using budget estimates instead of actual expenditures is
advantageous for two reasons. First, actual expenditures often are quite different from
budget estimates. While some of this is due to behavior by the district council, variation
in transfers from donors and the central government often is a consequence of factors
beyond the control of district governments, such as budget cuts for donor agencies by
their home governments. Second, elected members of the district council vote on
revenue and expenditure estimates and must justify these estimates to their constituents,
donors, and the central government. As a result, budget estimates are a clear public
declaration of the priorities of elected district councils.
Dependent Variables
Our key dependent variables measure local government expenditures funded from
locally-generated revenue; we exclude external sources of revenue (aid and transfers)
from our dependent variables. Recall that we have no trouble identifying expenditure
from different sources of revenue (local taxes, central government transfers, and foreign
aid) because local government budgets identify expenditure from sources of revenue.
Our ideal dependent variable is public services provided from locally-generated revenue.
In Zambia, constructing this variable is straightforward as local budgets contain complete
information on uses of local government sources of revenue. In Tanzania, measuring how
governments spend own-source revenue is more difficult. Specifically, while most
districts report the amount of own-source revenue used for government consumption,
most districts do not report how much of their own-source revenue they use for public
services. While it is tempting to infer that the share not used for recurrent costs must be
used for public services, such a conclusion is not entirely accurate because the central
government encourages districts to maintain budget surpluses and districts can run small
budget deficits if they can borrow money. For these reasons, we use the share of ownsource revenue used for government consumption to test our hypotheses. We also use
administrative costs as additional dependent variable because one can argue that
employee compensation is partly an investment in development (e.g., a school is useless
without a teacher and a health clinic is useless without a doctor or a nurse). Fortunately,
using government consumption as our dependent variable is not problematic. Our main
hypotheses are that we expect (1) a positive relationship between local revenue and local
services and (2) a negative relationship between external sources of revenue and local
services. We easily can restate these hypotheses as we expect (1) a negative relationship
between local revenue and local government consumption and (2) a positive relationship
between external sources of revenue and local government consumption.
For both countries we need to exercise considerable caution with how we express our
dependent variable. The most obvious way of testing our hypotheses would be to use
aggregate or per capita budget expenditures as our dependent variable. However because
we have no reason to believe that expenditures on any budget line-item will decrease as
taxes and/or transfers rise, using aggregate or per capita expenditure would be
9
inappropriate. Instead we express our dependent variable as a share of total own-source
expenditure. As a consequence, for Tanzania, our dependent variables are the share of
locally-generated revenue used for (1) total government consumption; and (2) other
charges (i.e., government consumption minus employee benefits). For Zambia our
dependent variable is public services as a share of locally-generated revenue. For this
reason, we expected the sign of our key explanatory variables will be different in each
country as the figure below shows. It is important to reiterate that we are testing the same
hypotheses in each country; the reason we predict different signs is because data
constraints force us to use different dependent variables in each country.
Figure 3: Hypotheses
Dependent Variable
Locally-Funded
Locally-Funded
Government Consumption/
Public Services/
Total Locally-Funded
Total Locally-Funded
Expenditure
Expenditure
(Tanzania)
(Zambia)
Explanatory
Variable
Local Revenue
-
+
External Revenue (Central
Government and Donors)
+
-
Explanatory Variables
The key explanatory variables for our study are the three sources of revenue for local
governments: locally-generated revenue, transfers from the central government, and
foreign aid. In both countries, locally-generated revenue is the easiest to identify.
Central government transfers are somewhat more problematic to identify because donors
partially finance funds that central governments in both countries use for transfers. Our
criterion for categorizing central government transfers is if parliament, not donors,
decides on the allocation of funds across districts. We categorize these items as central
government transfers, even if they are financed partially by donors, because from the
point of view of the local government, the transfers and any conditions associated with
those transfers originate from the central government.
Identifying foreign aid is more difficult. In Tanzania, local governments record two
types of donor flows. The first are direct transfers for local projects. The second
includes direct transfers plus programs that are financed by donors but are distributed
through independent government agencies that parliament does not control. This would
include programs like the Tanzania Social Action Fund and the Local Government
Capital Developments Program. Because classifying the latter category as donor could
be a debatable point, we us both measures (in separate models). In Zambia, local
governments do not comprehensively record donor transfers. Thus our measure of
foreign assistance to local governments in Zambia is disbursements from the Zambia
Social Investment Fund (ZAMSIF). Although ZAMSIF is only a subset of donor flows
into districts, using ZAMSIF data has two advantages: first, the data are comprehensive
and second, to receive ZAMSIF funds, local governments must develop projects with
10
community involvement and the projects must be directed in part at strengthening the
capacity of the local government (ZAMSIF 2004). Because two of the primary
objectives of ZAMSIF are to increase the capacity of local governments and to increase
community participation in development projects, ZAMSIF is explicitly designed to
increase local government accountability. As a result, using ZAMSIF data is a
particularly rigorous test of our hypothesis: if ZAMSIF projects which are specifically
designed to increase local government accountability result in larger government
consumption, foreign aid with fewer conditions on accountability is likely lead to a
similar outcome.
Because our dependent variables are expressed as a share of expenditure, it would seem
logical to express each source of revenue as a share of total revenue. However, because
the three shares must sum to one by definition, we would be unable to test the three
hypotheses simultaneously if we expressed our three explanatory variables as a share of
total local revenue. Although testing different sources of revenue separately would solve
the immediate statistical problems, we would be unable to say for certain which of the
three sources of revenue had the greatest impact on local government expenditure.8
Because we are interested in testing all three sources of revenue simultaneously and
because it is our objective to understand which source of revenue has the greatest impact
on local government expenditure, we use the three sources of revenue measured in per
capita terms.9
We also use an array of control and structural variables in our model, such as level of
development (i.e., poverty rates, mortality rates, life expectancy), size of district (total
population and total area), and electoral data from the most recent elections in Zambia
(2001).10 Structural economic variables are especially important because a plausible
alternative hypothesis could be that more economically developed districts have more
accountable governments, as Modernization Theory might predict. We can only test this
argument by including economic variables in our model. For Tanzania the
control/structural variables come from the 2000 census and for Zambia the source of
these variables is the 2000 census and the 2000 living conditions survey.
The two tables below summarize the variables we are using in our models. Table two
identifies our dependent and explanatory variables. Table three provides budget
definitions for the budget variables we use.
8
For the sake of simplicity, consider the case where we have two sources of expenditure, internal and
external), Because the share of internal funds equals 1 minus the share of external funds, in separate
regressions, the coefficients and standard errors on these variables would be exactly the same; the only
difference would be the sign. As a result, there would be no way to determine the relative importance of
each factor.
9
The correlations among sources of revenue (local, national, and foreign) are sufficiently low that we can
include all three in our models. In Tanzania, the highest correlation is between own sources of revenue and
transfers (0.4). In Zambia, the highest correlation is between local taxes and aid transfers (-0.2).
10
We do not use electoral data from Tanzania because of the ruling party’s overwhelming majority: in
mainland Tanzania over 90% of the seats in Parliament are held by the ruling party.
11
Table 2: Dependent Variables and Central Explanatory Variables
Tanzania
1. Government Consumption
(i.e., employee benefits plus
administrative costs)
2. Administrative Costs
(i.e., government consumption
minus employee benefits)
Zambia
Public Services
Explanatory Variable: Own Source
Revenue
Local Revenue
Local Revenue
Explanatory Variable: Aid
1. Direct donor aid
2. Aid disbursed by independent
government agencies + direct
donor aid
Zambia Social Investment Fund
Explanatory Variable: Transfers
Total government transfers
Total government transfers
Dependent Variable
Table 3: Budget Definitions
Total Expenditure = Own Source Expenditure + Central Government Expenditure +
Aid Expenditure
Own Source Expenditure = Government Consumption (Recurrent Expenditure) +
Public Services + Net Surplus
Government Consumption (Recurrent Expenditure) = Salaries and Employee Benefits +
Administrative Costs
Total Revenue = Local Revenue + Government Transfers + Aid
Method
We use OLS with robust standard errors to test our hypotheses since we have only one
year of data and thus can only conduct a cross-sectional analysis. We also do not use
instrumental variables since we have no evidence to suggest that our dependent variables
are endogenous. One, our control variables are from the 2000 census while the budget
data is from FY 2005. Two, while it is possible that the central government and donors
could base their transfers on local government budgets, in practice the opposite occurs:
local governments pass their budgets after receiving information on how much they will
receive from donors and the central government.
V. Results
Tanzania
Because about 90% of local government budgets in Tanzania come from the central
government and donors before we test our hypotheses, we first attempt to predict
12
transfers from the central government and donors to districts. The ability to predict
transfers is important because local governments that rely heavily on transfers, like
district governments in Tanzania, are only able to generate realistic multi-year budget
plans if they can reasonably predict future sources of revenue. To predict transfers, we
use a number of obvious “measures of need.” We use obvious “measures of need” to
determine if donors and the central government distribute funds to district governments
using fairly formulaic methods or of if they use more idiosyncratic methods.
Consequently, our explanatory variables are the size of the population, various measures
of household development (e.g., literacy rates, infant mortality rates, percent of
population involved in agriculture), and various measures of district development (e.g.,
percent of households with electricity and kilometers of road). The results we show
below show a sub-set of these regressions using log population, literacy rates, and log of
kilometers of road per district.11
These “measures of need” do a surprisingly good job of explaining government transfers
to districts as tables 4A and 4B show. Population, literacy rates, and kilometers of road
are able to explain between 76% and 98% of transfers. For all transfers - employee
benefits, administrative costs, and development - population is by far the strongest
predictor of transfers. Level of development (as measured by literacy rates and
kilometers of road) has a small and positive effect on transfers for employee benefits and
development transfers while level of development has a small but negative effect on
transfers for administrative costs.
Whereas the central government uses a transparent process for transfers to local
governments, donors appear to follow no such logic. While we are able to explain about
85% of the variance in all government transfers using population, level of development,
and kilometers of road, these same variables explain only 13% of the variance in donor
transfers. One interpretation is that the much smaller number of districts that report
donor transfers suggests our results using all districts may be biased if we compare them
to districts that report donor funding. When we restrict the sample to those districts that
report donor flows (table 4B) the results for government transfers stay the same
suggesting that we do not have selection bias. What is more likely is that districts are not
reporting donor flows in full and/or that local governments do not have control over
donor flows because donors bypass district governments. Either way, the data suggest
that a district’s inability to predict donor flows is a much larger impediment to multi-year
budgeting than its inability to predict government flows.12
11
The results of the regressions change minimally with regard to which measure of household and district
development we employ.
12
This is consistent with what we heard from district government officials; in most cases where we heard
complaints about the inability to forecast flows the government officials complained about donor opacity
not government.
13
Table 4a: Determinants of Transfers to All District Governments
in Tanzania
Employee Administrative Development
Benefits
Costs
0.672
0.617
0.952
Log Population
(15.95)***
(17.16)***
(55.28)***
0.011
-0.004
-0.002
Literacy Rate
(3.90)***
(2.59)**
(1.48)
0.079
-0.073
0.100
Log Road (KM)
(1.51)
(2.23)**
(7.66)***
5.080
6.645
-1.747
Constant
(11.74)***
(13.67)***
(5.65)***
Observations
113
113
113
R-squared
0.83
0.76
0.98
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Table 4a: Determinants of Transfers to District Governments in Tanzania that
Report Donor Funds
Employee Administrative Development
Benefits
Costs
0.607
0.659
0.956
Log Population
(14.04)***
(12.73)***
(41.23)***
0.012
-0.003
-0.001
Literacy Rate
(4.19)***
(1.44)
(0.35)
0.128
-0.048
0.119
Log Road (KM)
(2.69)***
(0.88)
(3.75)***
5.555
5.947
-1.994
Constant
(8.99)***
(7.58)***
(4.50)***
Observations
56
56
56
R-squared
0.88
0.82
0.97
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Donors
0.536
(2.06)**
-0.014
(1.37)
0.207
(0.65)
6.209
(1.99)*
53
0.13
While the results above suggest that there is no bias to limiting our analysis to only those
districts that report donor flows, is important to examine more thoroughly if there are any
clear differences between districts that report donor funds and those that do not. The
table below shows that across a broad range of indicators - enrollment rates, literacy
rates, recurrent transfers, population, area, percent of households with electricity,
agriculture as a share of the labor force, and total kilometers of roads - there are no
statistically significant differences between districts that report donor funds and those that
do not. For these reasons, there does not appear to be any bias between districts that
report donor funds and those that do not.
14
Table 5: Differences Between Districts that Report and Don't Report Donor Funds
Donor Funds
Reported
59
297,000
$14
7,646
62%
7%
70%
67%
619
Number of Districts
Population
Recurrent Transfers Per Capita
Area (Square Kilometers)
Literacy Rate
Houses with Electricity
Agriculture/Total Labor
Net Enrollment Rate
Kilometer of Roads
Donor Funds Not Reported
54
275,000
$12
7,471
60%
8%
70%
68%
608
Tables six and seven show the results using data from Tanzania for both measures of our
dependent variable (administrative costs as a share of total expenditure and total
government consumption as a share of total expenditure). Several points bear
highlighting. First, and most important, as local taxes increase, the share of the local
budget for government consumption falls. Second, higher central government transfers
increases the share of the local budget for government consumption. Third, donor flows
have no systematic effect on government expenditure. Although on the surface this is
evidence against our hypothesis, evidence from Zambia (that we show in the next subsection) as well as the evidence presented in table 4B suggests that either (1) we are not
capturing aid flows very well; or (2) aid flows are not sufficiently predictable for
government to react to them. Fourth, table seven shows that development indicators
(literacy rates and kilometers of road per district) have no effect on local government
expenditure even when we remove local taxes (table seven, columns two and four). This
is important because we can reject the hypothesis that the effect of locally-generated
revenue on local government expenditure is a proxy for the effect of level of development
on local government expenditure.
Table 6: Tanzania Results
Local Taxes Per Capita
Total Government Transfers Per Capita
Direct Donor Transfer Per Capita
Total Donor Flows Per Capita
Log Population
Constant
Administrative Costs/
Total Expenditure
-0.600
-0.230
(3.20)***
(2.38)**
0.135
0.080
(3.73)***
(3.16)***
-0.004
(0.15)
-0.026
(1.33)
1.037
0.972
(2.92)***
(3.57)***
-13.201
-11.504
(2.77)***
(3.08)***
50
76
0.23
0.18
Observations
R-squared
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
15
Government Consumption/
Total Expenditure
-0.747
-0.258
(3.30)***
(1.98)*
0.162
0.096
(3.72)***
(3.14)***
0.008
(0.25)
-0.026
(1.12)
1.258
1.234
(2.97)***
(3.75)***
-15.993
-14.679
(2.80)***
(3.26)***
50
76
0.24
0.18
Table 7: Robustness Check for Tanzania
Local Taxes Per Capita
Total Government Transfers Per Capita
Donor Transfer Per Capita
Log Population
Log Road (KM)
Adult Literacy Rate
Constant
Administrative Costs/
Total Expenditure
-0.697
(2.84)***
0.125
0.100
(3.52)***
(2.85)***
0.009
-0.019
(0.29)
(0.71)
0.994
0.931
(3.06)***
(2.69)**
0.118
0.632
(0.29)
(1.34)
0.012
0.007
(0.76)
(0.43)
-13.915
-16.352
(2.35)**
(2.42)**
49
49
0.24
0.14
Observations
R-squared
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Government Consumption/
Total Expenditure
-0.863
(2.88)***
0.148
0.116
(3.41)***
(2.69)**
0.023
-0.011
(0.63)
(0.36)
1.195
1.118
(3.10)***
(2.70)***
0.193
0.829
(0.38)
(1.44)
0.016
0.009
(0.80)
(0.46)
-17.064
-20.080
(2.39)**
(2.47)**
49
49
0.25
0.14
Zambia
The results from tests using Zambian data follow the striking pattern of those we found in
Tanzania. As locally-generated revenues increase, local government expenditure on
public services as a share of the local budget also increases while the opposite occurs
with government transfers. The last column in table six shows that local revenue is not a
proxy for wealth: literacy rates (our proxy for development) have no effect on
government expenditure even when we remove local taxes.
One key difference between the results from the two countries is that unlike donor
transfers in Tanzania, ZAMSIF transfers decrease services as a share of the government
budgets, consistent with our hypothesis. One hypothesis that may explain these divergent
results is that local governments in Zambia are able to anticipate ZAMSIF transfers with
a much higher degree of certainty compared to the average donor program because
ZAMSIF announces how much funds are available for each district and districts
subsequently propose projects to draw on those funds. Since ZAMSIF is designed to
increase local government capacity and because ZAMSIF projects must involve
community participation, our finding that ZAMSIF funds actually decrease local
government expenditure on services (as a share of the local government budget) is
powerful support for our hypothesis.
16
Table 8: Zambia Results and Robustness Check
Local Taxes Per Capita
Transfers Per Capita
ZAMSIF Transfers Per Capita
Adult Literacy
Log Population
Log Area (Square KM)
Log Population Density
Poverty
Constant
Services as a Share of Total Local Government Expenditure
0.004
0.005
0.008
0.008
(2.20)**
(2.65)**
(2.84)***
(4.30)***
-0.021
-0.023
-0.021
-0.021
(2.39)**
(2.30)**
(2.13)**
(2.04)**
-0.024
-0.020
-0.020
-0.019
(1.90)*
(2.44)**
(2.73)***
(2.53)**
-0.003
-0.004
-0.004
0.000
(1.31)
(1.52)
(1.91)*
(0.42)
-0.059
-0.058
-0.048
(1.65)
(1.60)
(1.29)
0.037
(1.50)
-0.042
-0.006
(2.37)**
(0.77)
0.003
(0.83)
1.022
1.425
0.981
0.873
0.474
(1.79)*
(4.00)***
(2.08)**
(5.96)***
(4.02)***
46
46
46
46
50
0.19
0.20
0.22
0.22
0.04
Observations
R-squared
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Table nine reports on whether political parties or different types of local taxes help
explain the public services share of local government expenditure. Columns one and two
show the logic of fiscal governance holds even in the face of partisanship. Unlike in
Tanzania, which is dominated by one party, Zambia has a competitive multiparty
electoral system. The two main political parties are the Movement for a Multiparty
Democracy (MMD), the ruling party, and the United Party for National Development
(UPND), the leading opposition party (there are a number of other smaller parties as
well). The first two columns of table nine test whether the parties affect local
government expenditure by adding each party’s seat share on local councils.13 There is
some evidence that districts with a higher share of MMD seats devote more of their
budget to government consumption while districts with a higher share of UPND seats are
more concerned with public services. More interesting, including seat shares of political
parties reduces the significance of central government transfers, not local revenue,
suggesting that partisanship may affect central government transfers.14 The last column
in table nine examines whether governments are sensitive to different sources of revenue
streams. Local governments in Zambia report three main types of local revenue: taxes
(e.g., income taxes and levies), charges (e.g., fees and rent), and other income.
Interestingly, given that other income is the most diverse source of local revenue it
nevertheless has the strongest effect on expenditures.
13
Because of the fairly strong negative correlation between the share of local government votes from each
party, we could not put both parties in the same equation.
14
Simple correlations provide tentative support for this hypothesis. While there is no correlation between
local taxes and political parties, there is a positive correlation between MMD seat shares and national
transfers while there is a small negative correlation between UPND seat shares and transfers.
17
Table 9: Additional Zambia Tests
Services as a Share of Local Government Expenditure
0.009
0.008
Total Revenue Per Capita
(4.33)***
(4.08)***
-0.005
MMD Seat Share
(1.85)*
0.002
UPND Seat Share
(2.16)**
0.007
Taxes Per Capita
(3.18)***
0.04
Charges Per Capita
(1.25)
0.018
Other Revenue Per Capita
(2.95)***
-0.010
-0.018
-0.021
Transfers Per Capita
(0.86)
(1.79)*
(2.06)**
-0.020
-0.020
-0.017
ZAMSIF Transfers Per Capita
(2.60)**
(2.58)**
(1.90)*
-0.004
-0.005
-0.004
Adult Literacy
(1.80)*
(2.12)**
(1.87)*
-0.066
-0.037
-0.041
Log Population Density
(2.90)***
(2.00)*
(2.28)**
1.052
0.845
0.861
Constant
(6.54)***
(5.50)***
(5.53)***
Observations
46
46
48
R-squared
0.22
0.29
0.24
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
The magnitude of the effects we find are economically significant as well as statistically
significant. For example, in Tanzania, model three in table four predicts that increase in
per capita taxes from the 25th percentile to the 75th percentile should decrease government
consumption as a share of the local budget from the 50th percentile to the 15th percentile.
Similarly, for Zambia, model five in table six predicts that an increase in per capita taxes
from the 25th percentile to the 75th percentile should increase public services as a share of
the local budget from the 50th percentile to the 75th percentile. The Graph 1 below shows
the effect of changes in per capita taxes on the service share of the budget in Zambia.
The graph shows that an increase in per capita taxes from the 25th to the 75th percentile
(roughly a three-fold increase in local taxes) increases the share of the budget going to
services from 47% to 57% (roughly a 20% increase).
18
Graph 1: Effect of Local Taxes on Services as a Share of
Local Expenditure in Zambia
Share of Budget for Services
60%
55%
50%
45%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
Per Capita Taxes (Percentile)
VI. Conclusion
In this paper we have employed insights from fiscal theories of governance to examine
how sources of revenue affect local government expenditures. First, independent of the
level of external assistance, local governments in Tanzania and Zambia devote a larger
share of locally-generated revenue to public services as the amount of taxes constituents
pay increases. Second, external funds encourage district governments to use locallygenerated revenue for government consumption. Third, the results on the impact of
donor funds on local government budgets are mixed. While local governments in Zambia
treat ZAMSIF transfers and central government transfers the same, donor flows to local
governments in Tanzania appear to have no impact on local government expenditure.
However, we also have some evidence that the inability of local governments in Tanzania
to predict donor funds may explain why, in aggregate, donor funds appear to have no
systematic impact on the expenditure decisions of local governments in Tanzania.
Our results speak directly to current development policy in Tanzania and Zambia.
Recently, in Tanzania, for example, donors and the central government replaced the
largest source of local government revenue with a central government transfer. At the
same time, donors currently are spending hundreds of millions of dollars every year to
increase the accountability of district governments in Tanzania. Over the past year, the
World Bank alone has agreed to spend approximately $75 million for local government
reform in Tanzania. The majority of the funds, however, are devoted to increasing
accountability from the supply side through institutional reform. Our results suggest that
increasing the accountability of local governments can come from the demand side as
well by strengthening the fiscal link between a district government and its citizens.
19
References
Ahmad, Junaid, Shantayana Devarajan, Stuti Khemani, and Shekhar Shah (2005),
“Decentralization and Service Delivery”, World Bank Policy Research Working Paper
3603, Washington, DC: World Bank.
Akin, John, Paul Hutchinson and Koleman Strumpf (2001), “Decentralization and
Government Provision of Public Goods: The Public Health Sector in Uganda”, Chapel
Hill, NC: Working Paper 01-35, Measure Evaluation, Carolina Population Center
University of North Carolina at Chapel Hill.
Azfar, Omar, Satu Kähkönen, and Patrick Meagher (2001), Conditions for Effective
Decentralized Governance: A Synthesis of Research Findings, College Park, MD:
Working Paper, IRIS Center, University of Maryland.
Bardhan, Pradhan (2002), “Decentralization of Governance and Development”, Journal
of Economic Perspectives, 16(4), 185-206.
Bardhan, Pradhan and Dilip Mookherjee (2004), “Pro-Poor Targeting and Accountability
of Local Governments in West Bengal”, Working Paper, Institute for Economic
Development, Boston University.
Bardhan, Pradhan and Dilip Mookherjee (2005a), “Decentralizing Delivery of AntiPoverty Programs in Developing Countries”, Journal of Public Economics, forthcoming.
Bardhan, Pradhan and Dilip Mookherjee (2005b), Decentralization and Local
Governments in Developing Countries: A Comparative Perspective, Cambridge, MA:
MIT Press, forthcoming.
Bates, Robert (2001), “Prosperity and Violence”, New York: Norton.
Bates, Robert and Dau-Hsiang Lien (1985), “A Note on Taxation and Representative
Government”, Politics and Society 14(l).
Bish, R. and V. Ostrom (1973), “Understanding urban government:
Metropolitan Reform Reconsidered”, Washington, DC: American
Enterprise Institute for Public Policy Research, Domestic Affairs Study
20.
Boix, Carles (2003), “Democracy and Redistribution”, New York: Cambridge University
Press.
Brautigam, Deborah (2000), “Aid, Dependence, and Governance”, Stockholm: Almqvist
and Wiksell.
20
Bueno de Mesquita, Bruce, Randolph Siverson, Alastair Smith, and James D. Morrow.
(2003), “The Logic of Political Survival”, Cambridge MA: MIT Press.
Carey, John and Matthew Shugart (1995), “Incentives to Cultivate a Personal Vote: A
Rank Ordering of Electoral Formulas”, Electoral Studies 14 (1).
Coolidge, Jacqueline and Susan Rose-Ackerman (1997), “High-Level Rent-Seeking and
Corruption in African Regimes”, World Bank Working Paper 1780, Washington, DC:
World Bank.
Cox, Gary (1997), “Making Votes Count”, New York: Cambridge University Press.
Cox, Gary and Mathew McCubbins (2001), ‘The Institutional Determinants of Economic
Policy Outcomes’ in Haggard and McCubbins (eds.), “Presidents, Parliaments, and
Policy”, New York: Cambridge University Press.
Downing, Brian (1992), “The Military Revolution and Political Change”, Princeton, NJ:
Princeton University Press.
Ferejohn, J, and B. Weingast (1997), “The New Federalism: Can the States Be Trusted?”,
Stanford (CA): Hoover Institution Press.
Galasso, Emanuela and Martin Ravillon (2005), “Decentralized Targeting of an
Antipoverty Program”, Journal of Public Economics, 89, 705-727.
Galiani, Sebastian, Paul Gertler and Erenesto Schargrodsky (2005), `School
Decentralization: Helping the Good Get Better, but Leaving the Poor Behind’, Working
Paper, Buenos Aires: Universidad de San Andres.
Inman, R., and Rubinfeld, D. (1996), “Designing Tax Policy in Federalist Economies: an
overview”, Journal of Public Economics 60, 307–334.
Karl, Terry Lynn (1997), “The Paradox of Plenty”, Berkeley, CA: University of
California Press.
Knack, Stephen (2000), “Aid Dependence and the Quality of Governance”, World Bank
Working Paper 2396, Washington, DC: World Bank.
Levi, Margaret (1988), “Of Rule and Revenue”, Berkeley: University of California Press.
Lijphart, Arendt (1999), “Patterns of Democracy”, New Haven, CT: Yale University
Press.
Lindert, Peter (2004), “Growing Public: Social Spending and Economic Growth Since
the Eighteenth Century”, New York: Cambridge University Press.
21
Linz, Juan (1990), “Perils of Presidentialism”, Journal of Democracy, 1, p. 50-69.
Linz, Juan (1994), ‘Presidential or Parliamentary Democracy: Does It Make a
Difference’, in Linz and Valenzuela (eds.), “The Failure of Presidential Democracy”,
Baltimore, MD: Johns Hopkins University Press.
Mainwaring, Scott and Matthew Shugart, eds. (1997), “Presidentialism and Democracy in
Latin America”, New York, Cambridge University Press.
McIntyre, Andrew (2003), “The Power of Institutions” Cornell, NY: Cornell University
Press.
Moore, Mick (1995) “Promoting Good Government by Supporting Institutional
Development?”, Institute for Development Studies, University of Sussex, Brighton UK:
IDS Bulletin 26 (2)
Moore, Mick (1998), “Death Without Taxes”, in Robinson and White (eds.), “The
Democratic Developmental State”, Oxford: Oxford University Press, p. 84-124.
Ndulu, Benno and Stephen O’Connell (1999), “Governance and Growth in Sub-Saharan
Africa”, Journal of Economic Perspective 13(1).
North, Douglass and Barry Weingast (1989), “Constitutions and Commitment”, Journal
of Economic History, 49(4), p. 803-832.
Persson, Torsten, Gerard Roland, and Guido Tabellini (2003), “How Do Electoral Rules
Shape Party Structures, Government Coalitions, and Economic Policies?”, NBER
Working Paper 10176, Cambridge, MA: NBER
Persson, Torsten, Gerard Roland, and Guido Tabellini (2005), “Electoral Rules and
Government Spending in Parliamentary Democracies”, mimeo.
Persson, Torsten, and Gerard Roland (2003), “Constitutional Rules and Fiscal Policy
Outcomes”, American Economic Review, 94 (1).
Przeworski, Adam, Michael E. Alvarez, José Antonio Cheibub, and Fernando Limongi
2001) “Democracy and Development: Political Institutions and Well-Being in the World,
1950-1990”, New York: Cambridge University Press.
Reinikka, Ritva and Jakob Svensson (2004), “Local Capture: Evidence from a Central
Government Transfer Program in Uganda”, Quarterly Journal of Economics, 119 (2).
Republic of Zambia (1996), “Local Government Act 1996”, Lusaka, Zambia:
Government of the Republic of Zambia.
22
Rondinelli, D., J. McCullough and R. Johnson (1989), “Analyzing Decentralization
Policies in Developing Countries: A Political-Economy Framework,” Development and
Change 20, p.57-87.
Root, Hilton (1992), “The Fountain of Privilege”, Berkeley, CA: University of California
Press.
Ross, Michael (2001), “Does Oil Hinder Democracy?”, World Politics, 53 (3), p. 325361.
Sachs, Jeffrey and Andrew M. Warner (1997), “Natural Resource Abundance and
Economic Growth”, Center for International Development, Harvard University.
Shugart, Matthew and John Carey (1992), “Presidents and Assemblies”, New York:
Cambridge University Press.
Svensson, Jakob (2000), “Foreign Aid and Rent-Seeking”, Journal of International
Economics, 51 (2), p. 437-461.
Tiebout, Charles (1956), “A Pure Theory of Local Public Expenditures," Journal of
Political Economy, 64: 416-24.
Tornell, Aaron and Philip Lane (1998), “Are Windfalls a Curse?”, Journal of
International Economics 44 (1).
Treisman, Daniel (2000), “The Causes of Corruption”, Journal of Public Economics, 76
(3), p. 399-457.
Tsebelis, George (2002), “Veto Players”, Princeton, NJ: Princeton University Press.
United Republic of Tanzania (2000), “Local Government Laws Principal Legislation
(Revised Edition) 2000”, Dar es Salaam, Tanzania: Government of the United Republic
of Tanzania.
Wiseman, J.A., (1995), “Democracy and Political Change”, London: Routledge, 1995.
ZAMSIF (2004), “Zambia Social Action Fund: ZAMSIF in Brief”, Lusaka, Zambia:
ZAMSIF.
23