Money Flows, Water Trickles:
Explaining the Disconnect between Spending and Improved
Access to Clean Water in Tanzania
⇤
Ruth Carlitz†
September 14, 2015
1 Introduction
Since the beginning of the 21st century, the number of African states holding multi-party
elections has increased dramatically. As of 2014, 43 of the region’s 48 countries had orga-
nized at least one round of multi-party elections (Bogaards, 2014). Given that authoritarian
regimes dominated African politics until the end of the 1980s, this trend has been hailed as
evidence of a democratizing continent.1 However, nearly half of these countries have settled
into a pattern of hegemonic party rule, where opposition parties may legally compete but
do not represent a true threat to the ruling party. This pattern of governance is not unique
to Africa – indeed, hegemonic party rule represents the most common type of authoritarian
rule in the post-World War II period.2 Understanding these regimes therefore represents
an important area of research.
Empirical work on this topic has primarily focused on how elections and legislatures
serve to bolster hegemonic party regimes. Scholars have shown how seemingly democratic
⇤Note: Figures at end of document best viewed in color.†[email protected]; www.ruthcarlitz.com1Sta↵an Lindberg is the main proponent of this argument. See, e.g. Lindberg (2006).2Author’s analysis of the GWF Autocratic Regimes 1.2 dataset (Geddes, Wright, and Frantz, 2012)
1
institutions help to co-opt the opposition (Gandhi and Przeworski, 2006), disseminate
information about the strength of the party (Magaloni, 2006), balance military power
(Geddes, 2005, 2008), and ease distributional conflicts (Blaydes, 2011). More recently, Diaz-
Cayeros, Estevez, and Magaloni’s (2012) book-length inquiry deftly shows how hegemonic
party regimes use social assistance programs to buy votes.
Less well understood is how the dynamics of hegemonic party rule play out at the local
level. In the wake of decentralization reforms across Africa and around the world, local gov-
ernments in hegemonic party regimes (and other polities) are increasingly responsible for
resource allocation. Advocates of decentralization argue that it will increase government
accountability and responsiveness to local needs (Faguet, 2012). However, these objectives
can be frustrated in hegemonic party regimes, where incumbent politicians have used de-
centralization reforms to consolidate their power (Green, 2011). Indeed, hegemonic party
regimes may decentralize in order to create subnational footholds (increasing electoral sup-
port and territorial penetration) and expand patronage networks at the local level (Riedl
and Dickovick, 2014).
This paper shows how decentralized service delivery not only fails to be responsive
but also serves as a tool of regime maintenance in the context of hegemonic parties. I
demonstrate how local politicians have skewed basic resource distribution in such a way
that favors their core supporters at the expense of demonstrably needier constituents.
While this pattern is similar to that documented at the national level in hegemonic party
regimes, the underlying logic of distribution at the local level is di↵erent.
At the national level, hegemonic parties typically institute “punishment regimes” –
distributing resources to citizens who remain loyal and withdrawing them from those who
defect (Magaloni and Kricheli, 2010: 128). This pattern of distribution is important for
obtaining supermajorities, which allow hegemonic parties to maintain control over national
2
electoral institutions and project an “image of invincibility” (Magaloni, 2006).
Local politicians in hegemonic party regimes face di↵erent incentives. Their political
survival depends largely on their ability to “deliver” the votes of their constituents to
party higher-ups. Such demonstrations of competence are rewarded by party bosses with
continued access to resources from the central government. Local politicians then distribute
these resources to their constituents in a manner intended to promote their reelection,
at the same time shoring up support for the ruling party. Hence, I argue that in the
context of hegemonic party regimes, local politicians function as political brokers – “local
intermediaries who provide targeted benefits and solve problems for their followers” in
exchange for votes (Stokes et al., 2013: 75).
The “targeted benefits” local politicians provide are not the private, material benefits
such as food, clothing, household items, cash handouts that typify the clientelist exchange
(Stokes et al., 2013). Rather, local politicians in hegemonic party regimes target the
distribution of public goods over which they have discretion. This is important since
vote-buying is not the primary mode of political survival in a number of hegemonic party
regimes. For instance, in the most recent (2014-2015) round of the Afrobarometer survey,
citizens in hegemonic party regimes3 reported significantly lower levels of vote-buying than
did residents of other political systems: 34% of Botswanans, 16% of Namibians, and 41% of
Tanzanians reported that voters are bribed in their countries’ elections. Compare this with
multi-party Kenya, where 56% of respondents reported vote-buying, or Senegal, where 68%
of respondents confirmed the existence of such behaviors. Further, “competitive elections
not only a↵ect average levels of vote buying across countries, they also amplify the general
tendency of political operatives to target and mobilize poor voters by trading money and
material rewards for votes. Consequently, vote markets seem to thrive when elections are
3Regime classification from Geddes, Wright, and Frantz (2012).
3
close” (Jensen and Justesen, 2014: 230).
My study focuses on the distribution of infrastructure for water provision in Tanzania,
a public good that local politicians have the power to allocate. This is a substantively
important topic for Tanzania, where the water budget has increased four-fold over the past
15 years, while access to clean water has risen by just one percentage point – from 54.4%
in 2000 to 55.6% in 2015.4
I show that local capture and politicized misallocation explain a considerable amount
of this disconnect. Specifically, local governments could have built over twice as many
water points with the money they received over the seven-year time period of my study.
Furthermore, the water points that did get built could have served over twice as many
people had they been allocated in a more e�cient manner.
My analysis proceeds in two stages. I first examine the distribution of money for water
from the central government to Tanzania’s local government authorities (LGAs). Next,
I look at how LGAs use these funds to distribute water infrastructure within their juris-
dictions, using detailed, geo-referenced data from a recent water point mapping exercise.
I find that the central government’s allocation of money to LGAs diverges to a some ex-
tent from the formula, with a disproportionate amount of funds going to the Minister of
Water’s home district. Political favoritism is much more pronounced at the local level.
Within LGAs, the distribution of new water infrastructure is skewed to favor localities
with higher demonstrated levels of support for the ruling party. In addition, wealthier and
better connected communities – those with the resources to e↵ectively express their de-
mands – are significantly more likely to experience improvements in water point coverage.
My analysis therefore shows that local politicians funnel the resources they get from the
central government to their supporters, as well as other vocal citizens they cannot easily
4Statistics from the World Health Organization/UNICEF Joint Monitoring Programme: http://www.
wssinfo.org/data-estimates/tables/
4
ignore.
A notable feature of my study is its use of finely grained, geo-referenced data on pub-
lic goods provision. While the literature on distributive politics in developing countries
has been expanding (Golden and Min, 2013; Stokes et al., 2013) studies that incorporate
such detailed data are still rare.5 Considering outcomes at the level of service delivery
strengthens the inferences I am able to make, particularly when compared with studies
that rely on blunter measures of resource distribution. Specifically, I uncover di↵erent pat-
terns of distribution for goods delivered at the local level vs. those allocated by the central
government, reflecting the di↵erent incentives and constraints these politicians face.
This paper proceeds as follows. The next section explains the basic logic of my approach
and derives hypotheses, Section 3 describes the context that I study, Section 4 describes
my empirical strategy, Section 5 presents my findings, and Section 6 concludes.
2 Local Public Goods Provision and Hegemonic Party Rule
In the context of decentralization, the allocation of local public goods typically proceeds in a
two-step process. First, the central government allocates funds to local jurisdictions. Next,
local politicians decide how and where to allocate public goods within their districts. I draw
on the distributive politics literature to understand the apparent logic of both central and
local government politicians. I then derive hypotheses to explain (1) financial allocations
from the central government to districts, and (2) the distribution of infrastructure for water
provision within districts.
As noted above, hegemonic party regimes are not only interested in winning elections
but also in obtaining supermajorities, which they typically maintain by instituting pun-
ishment regimes. Diaz-Cayeros, Estevez, and Magaloni (2012) explain that in settings
5Notable exceptions include Harris and Posner (2015); Burgess et al. (2015); Wilfahrt (2014).
5
characterized by limited political competition, “voters... are forced to support the incum-
bent party even when it fails to deliver any collective benefits, because they are likely to
be punished and removed from the government’s spoils system if they defect to the op-
position” (p. 235). Magaloni (2006) explains that the poorer the median voter, the more
e↵ective the punishment regime in deterring mass and elite defections and the less need
for electoral fraud.
Evidence of resources being allocated in a way that disproportionately favors supporters
of the hegemonic party, while punishing those that defect to the opposition has been
documented by, among others, Magaloni (2006) in Mexico, Blaydes (2011) in Egypt, and
Weinstein (2011) in Tanzania. This suggests the following:
Punishment/Favoritism Hypothesis (Districts): Money for water will be dispropor-
tionately channeled to districts that support the ruling CCM party with a higher
margin of victory, and will be reduced when the CCM’s margin falls.
I also consider whether the Minister for Water’s home district is favored when it comes
to the allocation of finance for water provision, reflecting a widespread belief that politicians
in Africa use their o�ces to favor their ethnic kin. In the Kenyan context, Kramon and
Posner (2014) find that coethnics of the minister of education acquire more schooling
than children from other ethnic groups. Similarly, Burgess et al. (2015) find that Kenyan
districts that share the ethnicity of the president benefit disproportionately when it comes
to new road construction. Empirical evidence of hometown favoritism with respect to
infrastructure provision has also been found in a broader sample of African countries (Ohler
and Nunnenkamp, 2014) and other authoritarian contexts such as Vietnam (Do et al.,
2013). While ethnic politics is a much less common phenomenon in Tanzania, accusations
of regional favoritism are not uncommon.
6
Hometown Favoritism Hypothesis (Districts): Money for water will be disproportion-
ately channeled to the Minister for Water’s home district.
Next, I consider the incentives facing the local politicians who decide how to spend the
funds they receive from the central government. While the legislative imperative to obtain
supermajorities is lacking at the local level, projecting invincibility still matters to deter
opposition parties from entering local politics. Furthermore, in clientelist systems where
access to funding is controlled by the central government, local governments must rely
heavily on the center, so local organizations, politicians, and voters have strong incentives
to a�liate with the national ruling party (Scheiner, 2006). In many African countries
resource allocation is understood to be strongly influenced by the nature of the political
relationship the receiving group has with higher tiers of government (Banful, 2011).
By allocating local public goods to regime supporters, politicians help to cultivate mass
support for the party, which is essential for regime survival (Magaloni, 2006). As Kramon
(2013) notes, voters in Africa, and especially in rural Africa, value the delivery of local
public goods. Furthermore, local public goods are often the only outputs of government
that rural voters can observe. In the context of district politics this suggests the following:
Punishment/Favoritism Hypothesis (Wards): Within districts, water infrastructure
will be disproportionately channeled to the CCM’s ‘core’ wards, and will be reduced
when the CCM’s margin in local races falls, or when a ward flips from supporting
the CCM to the opposition.
Another way local politicians target their supporters is by rewarding areas with high
levels of voter turnout. This interpretation is in keeping with scholarship on hegemonic
party regimes. For instance, research on communist elections in Eastern Europe has focused
7
on turnout, since to abstain or to spoil ballots was seen as an act of defiance in that
context (Gandhi and Lust-Okar, 2009). Blaydes (2011) notes that turnout was of utmost
importance to Egyptian regime under Mubarak, since it helped to legitimize the elections.
Similarly, Magaloni (2006) cites turnout as key in helping to create and sustain a hegemonic
party regime’s image of invincibility. This suggests:
Turnout Hypothesis (Wards): Within districts, water infrastructure will be dispro-
portionately channeled to wards with higher levels of voter turnout in the most recent
election.
Beyond electoral politics, another important factor motivating the distribution of re-
sources within districts is the degree to which communities can e↵ectively express their
demands. Decentralization of water provision in Tanzania reflects the so-called ‘demand-
responsive approach’ (DRA), which has emerged as the leading paradigm for rural water
provision in developing countries over the past two decades (Lockwood and Smits, 2011).
The shift from supply-driven centralized government programming to more demand-driven
approaches reflects increasing concerns about cost recovery and ine�ciency among water
sector donors (Wedgwood, 2005). This parallels a shift in thinking about water embodied
in the 1992 Dublin Principles,6 which recognize water as an “economic good” (Jimenez
and Perez-Foguet, 2010b). Major proponents of the approach include the World Bank, the
UK’s Department for International Development (DFID) and the Water and Sanitation
Program (WSP).
According to the DRA, water users are supposed to demand, own, and maintain their
water services and participate in their design. In practical terms, demand for water tends to
6The Dublin Principles, also known as the Dublin Statement on Water and Sustainable Development,were advanced at the International Conference on Water and the Environment (ICWE), Dublin, Ireland,organized on 2631 January 1992.
8
be understood as the willingness to pay for access (Rout, 2014). In Tanzania and a number
of other countries (e.g., Nigeria and Mozambique), this translates into mandatory cost-
sharing, with the government’s water policy requiring that a community must contribute a
given percentage of the total project cost before construction can begin (Wedgwood, 2005).
The notion of wealth facilitating demand is not unique to water. In their study of
community-driven development projects in Tanzania, Baird, McIntosh, and Ozler (2013)
uncover a regressive pattern on the demand side, with richer districts producing more
applications per capita and richer households more likely to be aware of the program.
Another prominent study of a central government transfer program in Uganda finds that
schools in better-o↵ communities received more of their entitlements than did schools in
poorer areas (Reinikka and Svensson, 2004). The authors interpret this as implying that
these schools had greater bargaining power vis-a-vis local governments to secure greater
shares of funding.
This leads to my final hypothesis regarding the distribution of water infrastructure
within districts:
E↵ective Demand Hypothesis (Wards): Within districts, water infrastructure will be
disproportionately channeled to wards with higher levels of income.
3 Background and Context
3.1 Tanzania’s Hegemonic Party Regime
Largely in response to external pressures, the Government of Tanzania legalized multi-
partyism in 1992. Since then, elections have been held regularly and are increasingly
viewed to be free and fair, with candidates at all levels of government respecting term limits
9
and transferring power peacefully. However, the country’s ruling Chama Cha Mapinduzi
(CCM) party has retained the firm grip on power it has held since Tanzania achieved
independence in 1961.7
Opposition parties in Tanzania have remained weak, due primarily to e↵orts by the
ruling party to impede potential competitors. As Ho↵man and Robinson (2009) note, biases
in the electoral formula give CCM more than its proportional share of seats in parliament.
Furthermore, the country’s National Electoral Commission lacks independence, campaign
finance rules overwhelmingly favor the CCM, and onerous party registration procedures
create barriers to entry for would-be challengers.
3.2 Local Politicians as Political Brokers
Beyond the party’s control of national institutions, this paper highlights the role of local
politicians in sustaining CCM dominance. Specifically, I show how the distribution of local
public goods can play a role that is akin to vote buying. The outcome I study, infrastruc-
ture for rural water provision, is arguably a public good that provides important positive
externalities, reducing the spread of disease, and lowering child and infant mortality. How-
ever, a new water point (e.g. a standpipe, borehole, or well) is fairly excludable, since it
only benefits people living in its immediate vicinity. This makes rural water infrastructure
highly targetable. Local politicians are in an ideal position to target this and other semi-
excludable public goods in such a way that promotes their reelection and contributes to
regime longevity.
I focus on the broker-like role of Tanzania’s ward councillors. Each of Tanzania’s 159
mainland local government authorities (LGAs) is governed by a district council, made up
7Tanzania’s ruling party at independence was called the Tanganyika African National Union (TANU);in 1977 TANU merged with the ruling party in Zanzibar to form the current CCM party.
10
of councillors elected from the district’s 20-40 wards.8 Ward councillor elections are held
concurrently with Parliamentary and Presidential contests every five years, with candidates
running in single-member constituencies.
In the wake of decentralization reforms beginning in 2000, the district council has be-
come a locus of development, particularly in rural areas. Districts are responsible for
delivering a number of key public services, including primary education, local health ser-
vices, agriculture extension and livestock, water supply, and local road maintenance. Such
services are funded and regulated by the central government but provision is devolved to
the district level (Venugopal and Yilmaz, 2010).
We may understand councillors as political brokers for three main reasons – their deep
local knowledge, their ability to mobilize their constituents, and their dependence on the
ruling party. I will discuss each of these aspects in turn. Note that this argument pertains
primarily to councillors from the ruling party, who accounted for 84% of all councillors as
of the most recent (2010) election.
Regarding local knowledge, candidates for ward councillor are required to be residents
of the ward in which they seek nomination and must be nominated by at least 10 reg-
istered voters from within the ward (National Electoral Commission, 2006). This means
that councillors tend to have broad knowledge about local issues (Mafuru et al., 2015).
Furthermore, wards are fairly small – about 250 square kilometers and comprising 13,000
people on average. Given Tanzania’s population distribution this translates into approxi-
mately 6,000 eligible voters. As Lange (2008) explains, the growth of a middle class in rural
areas has made patronage an increasingly aspect of local politics. She notes that the ward
councillor’s position is time consuming and unpaid, and tends to be occupied by retired
8Members of parliament representing serving the district also serve on the district council, as do femalerepresentatives appointed in proportion with their parties’ elected seats (Venugopal and Yilmaz, 2010).Each district comprises one to three Parliamentary constituencies.
11
civil servants or successful businessmen (very seldom women).9 The success of a political
leader is often measured by his or her ability to attract donor and/or district funding to
local development projects.
Although some ward councillors have further political ambitions, the position does not
tend to be a steppingstone post to higher o�ce.10 This represents a common pattern across
sub-Saharan Africa. As Hyden and Mmuya (2008: 44) explain, councillors “primarily
regard themselves as representatives and patrons of local communities.”
By serving as local patrons, ward councillors create local blocs of supporters. As
Harrison (2008: 178) explains, “throughout Tanzania’s post-colonial history, councillors
have needed to maintain legitimacy by being ‘development advocates’ on behalf of their
constituency.” Interviewing councillors in Lushoto, Harrison (2008) finds that one of their
key tasks as local politicians was arguing their case in full council meetings for support to
schools, health care, road improvement, irrigation and the improvement/construction of
wells.
Excepting those with considerable resources of their own, most ward councillors are
extremely dependent on their respective political parties. First and foremost, all candidates
for ward councillor must be members of an o�cially recognized political party. A�liating
oneself with the ruling party provides the longest coattails on which to ride. This is so both
in terms of campaign finance as well as the resources that councillors are able to provide
their citizens.
With respect to campaign finance, all political parties receive disbursements from the
government proportional to their share of seats in the National Parliament and local coun-
9Women who serve on the district council primarily occupy “special seats,” appointed by their respectivepolitical parties in proportion with the parties’ vote share in each district. This reflects a constitutionalamendment passed in 2000 requiring that women make up at least one-third of all seats on the districtcouncil (Simonen, 2010).
10Personal communication with Keith Weghorst, August 2015.
12
cils. Given the ruling party’s dominance this creates huge discrepancies. For instance,
in the run-up to the 2010 elections, the ruling CCM party received TZS. 2.3 billion (ap-
proximately USD $1.1 million) and the most prominent opposition party (CHADEMA)
received just TZS. 750 million ($350,000). The party with the smallest representation
(APPT-Maendeleo) received just TZS 100,000 (less than $50!)11 from government co↵ers.
Furthermore, although decentralization reforms have given ward councillors more re-
sources to control, the nature of the reforms have made local politicians more financially
dependent on the central government. This reflects the increase in the value of central
government funding alongside the stagnation of local authorities’ own revenue (Weinstein,
2011). Furthermore, remaining local revenue collections have largely been absorbed by
centrally appointed local bureaucrats to finance costs associated with the running of the
district headquarters (Pallotti, 2008).
A�liation with the ruling party promotes success at the polls in other ways, too. As
Babeiya (2011) explains, while a semi-independent National Electoral Commission super-
vises the country’s presidential and parliamentary elections, local government elections are
supervised by the ministry responsible for local government and regional administration.
The head of this ministry is an appointee of the president who is also the chairman of the
ruling party.
The broker-like role of ward councillors in Tanzania is likely not unique. Jensen and
Justesen’s (2014)’s analysis of survey data from 18 countries in Africa finds that encounters
with local councillors are positively associated with vote buying, while direct contacts with
Members of Parliament have little e↵ect.
This argument also has normative implications. As Yilmaz, Beris, and Serrano-Berthet
(2010) explain, an emerging literature has begun to compare partisan systems of local gov-
11Figures in local currency from TEMCO (2011).
13
ernment with non-partisan ones. Advocates of non-partisanship in local elections maintain
that local governments tend to concern themselves with issues on which there can be no
division along party lines. Hence, partisan local governments risk policy-making becoming
contaminated by patronage and clientelism instead of focusing on long-term benefits. In
Ghana, for example, parties have been outlawed in local elections. India’s panchayats also
operate on a non-partisan basis by law.
3.3 Policy Framework for Water Provision
Water has been high on the (stated) agenda of Tanzania’s ruling Chama cha Mapinduzi
(CCM) party since shortly after the country achieved independence. In 1965, the gov-
ernment took full responsibility for the funding of rural water supplies, and declared that
water at public distribution points (standpipes, boreholes, etc.) should be free (Jimenez
and Perez-Foguet, 2010a). At the end of 1970, the ruling party created an ambitious plan
which stated that by 1991 the entire population (both rural and urban) should have access
to safe water within easy reach of their homes (Gine and Perez-Foguet, 2008). However,
economic crisis throughout the 1970s and 1980s led to major declines in service delivery.
Foreign donors started developing water supply programs, largely bypassing government
structures and ultimately proving unsustainable. In response, the Government of Tanzania
launched a new national water policy in 1991, representing “the start of a long process of
reforms to address the shortcomings of the previous system and build donor confidence”
(African Ministers Council on Water, 2010).
In order to implement the revised policy, a coalition of donors12 worked with the Gov-
ernment of Tanzania to develop the Water Sector Development Program (WSDP). The
WSDP, spurred by the Millennium Development Goal to increase access to clean water,
12These include the World Bank, the African Development Bank, the UK Department for InternationalDevelopment and a handful of others.
14
was intended to enhance coordination among donors as well as across three sub-sectors
(rural water supply, urban water supply and sewerage, and water resources management)
under one comprehensive investment and regulatory regime (Ministry of Water, 2006a).
Along with increasing access to clean water, the WSDP also aims to promote decentral-
ization and encourage public participation – in keeping with the broader decentralization
reforms described above.
My analysis focuses exclusively on rural water provision. Despite rapid urbanization
over the past half century, most countries in Africa remain predominantly rural. Thus the
insights I generate from this study should travel widely. Furthermore, most Tanzanians
who lack access to clean water reside in rural areas. According to the most recent (2012)
statistics, over 19 million rural Tanzanians, or 56% of the rural population, lacked access,
compared with 2.8 million – or 22% of all urbanites.13 This reflects broader demographic
patterns (Tanzania is over two-thirds rural). In addition, urban residents are much more
likely to have water piped into their homes – 23% vs. just 4.1% of rural residents (World
Health Organization and UNICEF, N.d.) – or are able buy water from private vendors
(Banerjee and Morella, 2011: p. 48). People living in rural areas are thus more reliant
on government service provision to meet their needs. Finally, data availability guides my
focus. The criteria meant to guide resource allocation to rural areas is much more clearly
specified than that guiding the distribution of resources to urban areas, as funding for urban
water supply tends to be concentrated in a few large, earmarked projects (Oxford Policy
Management, 2013). In addition, the recent nationwide water point mapping exercise that
accounts for my infrastructure data was only conducted in rural areas.
13My calculations of unserved populations are based on estimates from World Health Organization andUNICEF (N.d.).
15
4 Empirical Strategy and Data
In this section I first explain how I will analyze variation in central government financial
allocations to LGAs, and then how I will account for the distribution of water infrastructure
within districts. For each level of analysis I first describe the empirical strategy and then
the data.
4.1 Where Does the Money Go? Financial Allocations Across Districts
The central government’s funding formula relies on two main criteria to allocate funds
for water infrastructure to districts - the proportion of the district population that lacks
access to an “improved” water source14 and the dominant extraction technology. The
latter criterion serves as a proxy for the geographic and hydrological factors that a↵ect the
di�culty of extraction, and hence, the cost of building appropriate infrastructure.
Table 1 below depicts the formula more precisely.
[Table 1 about here.]
I use this formula to construct ideal allocations to districts for water for each year over
the first phase of the WSDP (2007-2013), which I then compare to the actual amount
that each district received.15 I find that actual allocations diverge substantially from ideal
allocations in the majority of cases, as shown in Figure 1.16
[Figure 1 about here.]
From this figure we see that in 2007 the vast majority of districts received less than
their ideal allocations, while three districts (Hai, Serengeti, and Monduli) benefited to an
14These include piped water, public taps or standpipes, tubewells or boreholes, protected dug wells,protected springs, and rainwater.
15This time period saw the creation of 30 new districts in Tanzania. In the Appendix I explain how Ideal with new and split districts.
16Formula criteria is from 2006, which I assume is used to determine ideal allocations for 2007.
16
extremely disproportionate degree. Each of these districts received between 12 and 25
times their allocation as per the formula. Similar patterns persist in subsequent years,
with the majority of districts receiving less than what they should per the formula, while
a select few benefit disproportionately. The pattern of favoritism is not the same in each
year, supporting an analysis that looks both across districts and over time.
In order to understand what drives deviations from the formula, I consider a number
of factors, relating to the hypotheses presented above. I estimate regressions based on the
following model:
Proportionit = ↵it + �1Regime(Opposition)Supportit + �2MinisterHomeit
+�3Turnoutit + �4AuditOpinionit�1 + �5Xi + ✏i
where Proportioni,t is the actual allocation of funds for water as a proportion of the
ideal allocation to district i in year t. Regime(Opposition)Supportit is support for the
ruling party or opposition measured in various ways as I describe below, MinisterHome
is a dummy variable indicating the Water Minister’s home district, Turnout is district-
level turnout in the most recent election, AuditOpinionit�1 is the auditor’s opinion of the
district’s accounts in the previous year, and X is a vector of time-invariant controls.
The main specification I consider is a mixed-e↵ects linear regression model, since devia-
tions from the formula vary considerably both within and across districts.17 I also estimate
fixed e↵ects regressions to consider what drives changes within districts only.
17The intraclass correlation coe�cient of my dependent variable is .3, implying that one third of thevariation is accounted for by clustering within districts, while two-thirds reflects cross-district variation.
17
4.1.1 District-Level Data
Data on actual disbursements to rural districts for water projects for each year from 2007-
2013 comes from the Ministry of Water’s Management Information System18. In the Ap-
pendix I describe how I calculate ideal allocations and then divide ideal by actual allocations
to construct my dependent variable.
I operationalize my independent variables as follows: I measure regime support using
both the vote margin (percent) for the CCM Parliamentary candidate and total votes for
the CCM Presidential candidate in the most recent election. (There were two elections
during the study period: 2005 and 2010). I operationalize support for the opposition using
a dummy variable indicating whether the district was represented by an MP from an op-
position party following the most recent election, and another dummy variable indicating
whether the CCM lost dominance of the district (i.e. the district went from being repre-
sented only by CCM MPs to either CCM and opposition MPs or opposition MPs only in
the last election).19
I also construct a dummy variable for the Minister of Water’s home district, which
changed three times during the study period.20 Turnout serves as another proxy for regime
support.
According to more recent descriptions of the allocation formulae for water and other
sector block grants,21 districts must satisfy a set of minimum conditions related to financial
management, planning and budgeting, procurement, and other functional processes in order
to receive their full grant amounts. Councils that do not meet the minimum conditions
are supposed to receive only 50% of the development grant amount amount. As a proxy
for the quality of financial management, I consider the auditor’s opinion of the district’s
18http://www.mowimis.go.tz/
19Recall that each district contains one to three Parliamentary constituencies.20See Appendix for details on variable construction.21See, e.g. United Republic of Tanzania (2011).
18
accounts in the year preceding disbursement. The audit reports take into account much of
the same criteria as the annual assessments of financial management, which are not publicly
available for all years that I study. Each year, the National Audit O�ce (NAO) of Tanzania
subjects each district to an audit and then issues an overall opinion, which can be of three
main types: “Unqualified” (clean), “Qualified” (when there are material misstatements in
districts’ financial record-keeping), or “Adverse” (when the district’s financial statements
are not in accordance with the applicable financial reporting framework or accounting
standards). In each year, I code the audit opinions on a 3-point scale such that higher scores
correspond to better financial management (Adverse=1, Qualified=2, Unqualified=3).22
Diversions from formula may also be explained by district-level poverty, which I there-
fore include as a time-invariant control.23 I measure poverty using estimates from the
WorldPop high resolution poverty maps (Tatem et al., 2013). The WorldPop poverty
maps illustrate the proportion of people living in poverty (defined as less than $1.25 per
day) per square kilometer in 2010.24
In addition, I control for population density (logged). Given economies of scale and the
fixed costs associated public service delivery, I expect that more densely populated districts
will require less expenditure per capita. Finally, I control for depth to groundwater as a
proxy for how di�cult it is to extract water from a given district. My data on depth to
groundwater is from MacDonald et al. (2012)’s quantitative maps of groundwater resources
for Africa.
Table 2 depicts summary statistics for the district-level variables.
[Table 2 about here.]
22For more detail on the criteria corresponding to the di↵erent opinions, see United Republic of Tanzania(2013)
23While poverty rates arguably vary over time, I only have estimates for 2010.24The fact that my measure of poverty is predicted rather than observed suggests possible attenuation
bias. That is, the e↵ect of poverty that I predict on my dependent variable are likely to be smaller thanthe actual e↵ect.
19
If there were a one-to-one relationship between the amount of money a district receives
for water and improvements in district-level access to clean water, my analysis could end
here. However, this is not the case – for a number of reasons.
First, variation in money spent largely does not map onto variation in water points
built. Figure 2 plots waterpoints built over the first phase of the WSDP (2007-2013)
against money spent in each district during the same time period. The figures for water
points built are weighted such that more expensive schemes (e.g. diesel pumps) are given
a higher weight than less expensive schemes (hand pumps).25 The fitted line depicts a
positive relationship (districts that spend more money on water build more water points);
however, there are a large number of outliers.
[Figure 2 about here.]
Furthermore, when new water points are constructed, they are not always targeted at
the neediest areas. Over the first phase of the WSDP, a total of 21,978 new water points
were built.26 Given that each new water point could serve up to 250 people, this could
have represented an increase in 5.5 million people with access to clean water. However, the
addition of these water points only generated access for 2.6 million people – less than half
of the potential.27 This reflects redundancies in water point placement, as well as a failure
to allocate new infrastructure to needy areas.
Figure 3 illustrates this tendency with the example of Monduli district. The left-hand
panel of the figure shows the distribution of water points (in blue) in Monduli district as of
2006. The darker areas of the map indicate those that are more densely populated. We see
that there are substantial dark portions of the map with no water points, indicating that
25The weighting scheme is described in the Appendix.26This figure is restricted to the 127 districts for which I have complete information about water point
construction.27Access is defined as living within 1km of the water point. I calculate access using data from WorldPop
in QGIS.
20
many people lacked access to clean water. Indeed, as of 2006, only 16.7% of Monduli’s res-
idents had access to a water point.28 Between 2006 and 2013, Monduli district constructed
35 new water points, the placement of which is depicted in red in the right-hand panel of
the figure. We see that for the most part new infrastructure was not targeted to reach
unserved areas. Hence, the proportion of the population with access increased by less than
one percentage point, to 17.6%, over the seven-year period. This example illustrates that,
within districts, water points are not always targeted to benefit the neediest communities.
[Figure 3 about here.]
The following section describes how I test for the influence of factors other than need
in order to explain the distribution of new water infrastructure within districts.
4.2 How is the Money Spent?
Distribution of Water Infrastructure Within Districts
At the LGA level, local o�cials are supposed to allocate water funds to projects in specific
rural communities within the district, based on a combination of need, as demonstrated
by current levels of access, and demand, as demonstrated through a bottom-up planning
process. Figure 4 illustrates the ideal flow of funds. Money for rural water infrastructure
is supposed to be allocated from the central government (Ministry of Finance) to districts
according to the formula described above. Once it reaches the district level, the local
government has discretion to decide where to spend it.
[Figure 4 about here.]
While many local o�cials pay lip service to the bottom-up planning process that is
meant to guide resource allocation within districts, priorities expressed at the local level
28I define access here as living within 1 kilometer of a water point. This corresponds to the United Nations’physical accessibility criterion (http://www.un.org/waterforlifedecade/human_right_to_water.shtml)
21
are often displaced by priorities at higher levels of government (Tilley, 2013). Furthermore,
many local government o�cials complain that they typically do not receive the entire
budget from the central government in time or in full, making it di�cult to implement
their plans (Orgut Consulting, 2009; Quinn and Tilley, 2013). Finally, evaluations of the
WSDP note a lack of capacity at the local government level (Oxford Policy Management,
2013), which can prevent oversight of implementation and results in financial discretion
(Tilley, 2013).
Indeed, my first cut at the spending and construction data suggests substantial local
capture. The 93 rural districts for which I have complete spending and construction data
received a total of $170 million dollars over the first seven years of the WSDP, which they
used to construct 16,813 water points. Had all of this money been dedicated to water point
construction, these districts could have built over 40,000 water points.29
Beyond local capture, I have described evidence of substantial misallocation within
districts. In order to understand what a↵ects the placement of the water points that
do get built, I model the number of water points built in each ward over the first seven
years of the WSDP as a function of ward-level political variables and appropriate controls.
Both my dependent and independent variables vary over time within wards, so I estimate
a count model with ward-level fixed e↵ects. Specifically, negative binomial regression is
appropriate, since the dependent variable has a large number of zero values. Figure 5
depicts the distribution of the dependent variable.
[Figure 5 about here.]
29This estimate is based on unit costs for di↵erent water point types from Ministry of Water (2006b) andreflects a mix of technologies that is in keeping with Tanzania’s hydrology.
22
The model can be represented as follows:
WaterpointsBuiltit = ↵it + �1CCMSupportit�1 + �2Turnoutit�1 + �3Xit + ✏ij
where WaterpointsBuiltit refers to the number of water points built in ward i in year t,
CCMSupportit�1 is a measure of support for the ruling party in ward i in the most recent
election, Turnoutit�1 refers to that ward’s turnout in the most recent election, and Xit is
a vector of ward-level controls corresponding to the relevant time period.
I also estimate another set of models with ward-level coverage improvements over the
entire first phase of the WSDP as my dependent variable. Whereas the previous set of
models considered ward-year variation, in these models the ward is the unit of analysis. I
focus on improvements in coverage (the number of water points per person) rather than
access (the proportion of ward residents within 1 km of a water point) since the former
is what district o�cials use in their routine monitoring (Harris, 2012). Coverage improve-
ments at the ward level likely depend on district-level characteristics as well (such as the
financial resources the district council receives for water provision, or geographic factors).
It is thus important to consider the clustered nature of the data when estimating regression
models. I do this by estimating a two-level random intercept regression model (with wards
clustered into districts) of the following form:30
Coverage�ij,2005�2012 = ↵ij + �1CCMCouncillorMarginij,2005 + �2Turnoutij,2005 + �3Xij + ✏ij
Coverage�ij,2005�2012 refers to the improvement in water point coverage in ward i in
district j, CCMCouncillorMarginij,2005 is the CCM candidate for ward councillor’s 2005
margin of victory in ward i in district j, Turnoutij,2005 refers to that ward’s turnout in the
30I use Stata’s xtreg command with the mle option.
23
2005 election, and Xij is a vector of ward-level controls.
4.2.1 Ward-Level Data
My data on water point construction at the ward level is derived from a recent water
point mapping (WPM) exercise conducted by the World Bank and the Tanzanian Ministry
of Water. The WPM dataset includes observations of 75,000 public water points serving
rural communities in mainland Tanzania, with information on their year of construction,
source type, management scheme, functionality status and precise geographical location.
This information allows me to construct a time series of water point construction, using
Geographic Information Systems (GIS) software to map the water points into wards.
Given that each water point is assumed to serve 250 people (per the Ministry of Water’s
guidelines) I calculate coverage by multiplying the number of water points in each ward
by 250 and dividing by the ward’s population. I then calculate the percentage di↵erence
in coverage between 2005 and 2012.
My focal independent variable is support for the ruling party at ward level. I measure
this in multiple ways, including 1) a dummy variable indicating whether the ward elected a
ruling party councillor in the most recent (2005 or 2010) election, and 2) the CCM’s margin
of victory (where higher, positive margins indicate higher levels of support and negative
margins indicate support for the opposition). I also consider the CCM councillor’s vote
share and whether the ward councillor was aligned with a CCM in the previous election.
I use turnout as an additional measure of support for the ruling party and again consider
turnout in both 2005 and 2010. The o�cial election data from the Tanzanian National
Electoral Commission (NEC) do not include turnout figures so I estimate turnout by divid-
ing the total number of votes in a given ward by an estimate of the voting-age population
in each ward in each election year.31
31Voting-age population is not provided by the NEC, either, so I use population data from the
24
I control for existing ward point stock in the ward-year model of water point construc-
tion, and ward-level water point coverage as of 2005, as well as a proxy for ward-level rate
of water point functionality, in the model of coverage improvements. I cannot calculate
the latter measure directly since I only have information on functionality as of the date of
the water point mapping exercise. As a result, I examine the proportion of the 2005 water
point stock that is still functional as of the mapping exercise. Both of these measures
allow me to determine whether district councils are trying to equalize the allocation of
infrastructure across wards.
The model of coverage improvements also controls for the log of population density and
poverty. I expect that more densely populated districts will require fewer water points per
capita, and therefore will experience smaller increases in coverage. As above, I measure
poverty as the proportion of people living in poverty per square kilometer using estimates
from the WorldPop high resolution poverty maps (Tatem et al., 2013).
The latter model also controls for remoteness. This is important since water infras-
tructure is more built near major roads, whereas areas that are harder to reach may see
less construction. The fact that 68% of all water points in the sample are located within
1 kilometer of a road lends support to this idea. I construct my measure of remoteness
using roads data from OpenStreetMap. Using R and GIS I first determine the geographic
coordinates of each ward’s centroid, which then allows me to calculate the distance from
the center of each ward to the nearest road.
Finally, the coverage improvement model again controls for depth to groundwater using
MacDonald et al. (2012)’s quantitative maps of groundwater resources for Africa.
Table 3 depicts summary statistics for the district-level variables. We see that although
2012 Census, and scale it back to 2005 and 2010 levels using growth rates from the World Bank’sWorld Development Indicators. I calculate the proportion of the population that is of voting age usingdata from the International Institute for Democracy and Electoral Assistance (IDEA) Unified Database(http://www.idea.int/uid/).
25
the great majority of wards elect councillors from the ruling party, there is considerable
variation in support for the ruling party candidates (as measured by their vote shares) and
turnout.
[Table 3 about here.]
5 Findings
This section presents my findings for both allocations from the central government to
districts, and the distribution of new water infrastructure within districts. Comparing the
two levels, I find much stronger evidence of political interference at the local (district) level.
5.1 Explaining Financial Allocations to Districts
Table 4 shows that the only political variable that is significantly associated with deviations
from the formula is the dummy variable indicating the Minister of Water’s home district.
[Table 4 about here.]
I also estimate a fixed e↵ects regression, since the Minister for Water’s home district
changed over the study period. The results are reported in Appendix Table A4 and show
that none of the regressors included register significantly, with the exception of a negative
sign on votes for the CCM Presidential candidate and a positive sign on turnout. This
appears to be due to multicollinearity. These two variables are positively and significantly
correlated (r = 0.23) and when I exclude either one from the regression, the other loses
significance.
In addition, when the mixed e↵ects model is estimated and the variable indicating
Minister’s home district is excluded, none of the other regressors register significantly.
26
This is reported in Appendix Table A5. This suggests that deviations from the formula
are more or less random noise.
As a robustness check, I also include interactions between the political variables and
years until the election. None of these register significantly.
The Appendix also reports findings for regressions in which the dependent variable is
simply the actual allocation (logged) that districts receive in Table A6. Model 1 considers
the formula criteria only as regressors. We see that only audit opinion is significant in
predicting which districts will receive more money for water. In subsequent models, I
then consider a similar set of regressors to that analyzed above. The positive sign on the
Minister’s Home district persists in the mixed e↵ects model. I also report the results of a
fixed e↵ects model in Table A7.
It is important to note that I do not measure need using precisely the same data as the
government (due to lack of public availability). The government may be using worse or
di↵erent data. Thus any deviation from the ideal may be due to errors in the government’s
calculation. This suggests that the formula may in fact “work,” if imperfectly, as allocations
do not appear to be driven primarily by politics, with the exception of the Minister for
Water’s home district being favored.
In sum, while I find evidence that the formula does not appear to be followed to the
letter, political favoritism cannot explain most of the deviations.
5.2 Explaining Distribution of Infrastructure Within Districts
When we look at how local governments spend they money they receive within their dis-
tricts, the story changes. The results presented in this section suggest that political fa-
voritism accounts for much of the ine�cient placement of water points within districts.
Table 5 shows that wards in which a CCM candidate was elected in the most recent elec-
27
tion experience greater improvements in water point coverage, as do those where turnout
was higher. Alignment between the ward councilor and MP from the ruling party also
positively influences water point construction.
Wards with a higher level of existing water point stock are less likely to receive new
water points, suggesting some attempt to equalize distribution of water points within wards.
However, that equalizing tendency is largely o↵set by the political variables.
[Table 5 about here.]
As a robustness check I also estimate a series of models with new water points per
capita as the dependent variable. The results, shown in Appendix Table A8 depict similar
results to the above for turnout and existing water point stock.
In addition I estimate logit models where the dependent variable is a dummy indicating
whether water point construction occurred in each ward in each year. The results, shown
in Appendix Table A9, confirm the importance of turnout in predicting where new water
points will be built.
Turning to the determinants of improvements in water point coverage, Table 6 further
highlights the importance of local politics. Again we see that wards with higher levels of
turnout in both the 2005 and 2010 elections, as well as those where CCM councillors won
their elections in 2010 by higher margins of victory, experience greater improvements in
water point coverage.
[Table 6 about here.]
Table 6 also shows that proportionate change in coverage at ward level is negatively
associated with 2005 water point coverage. As above, this suggests an attempt by district
councils to equalize the distribution of water points across wards. However, such attempts
at equalization still fail to achieve allocative e�ciency in most cases, given that people
28
are not distributed evenly in most wards. Further compounding this, poorer and more
remote wards are less likely to see improvements in water point coverage, as indicated by
the negative signs on poverty and distance to the nearest main road.
Figure 6 shows that the e↵ects of turnout and poverty are substantively larger than
any other regressors.
[Figure 6 about here.]
As a robustness check I run a similar set of regressions with proportionate improvements
in access (as opposed to coverage) as my DV, where access is measured as the proportion
of a ward’s residents living within 1 kilometer of a water point. The results reported in
Appendix Table A10 show that access improvements are less likely in wards with higher
levels of preexisting access, similar to the results reported for improvements in coverage,
shown in Table 6. Appendix Table A10 also shows that improvements in access are harder
to achieve in poorer and more remote wards. None of the variables capturing support for
or opposition to CCM a↵ect improvements in access. This is unsurprising since access is
less visible than coverage, in terms of a figure that politicians are likely to track.
In sum, these results imply that while relatively wealthier communities may be better
at expressing their demands (as evidenced by the fact that poorer communities experience
more limited improvements to their water infrastructure), those demands are more likely
to be met in places with higher levels of voter turnout. Given the hegemonic CCM party’s
interest in obtaining supermajorities, voter turnout is likely a closely monitored metric of
regime support. While turnout is thought to be more important in Presidential contests
as a signal of regime invincibility (Gandhi and Lust-Okar, 2009), the hierarchical structure
of Tanzania’s hegemonic party regime suggests that it could very well matter at lower
levels of government too. While citizens elect leaders at various levels of government
(including village chairmen and the ward councillors who constitute the district councillor),
29
the ruling party also appoints administrators at each level (village and ward executive
o�cers, e.g.) who often share responsibilities – and sometimes o�ce space – with their
elected counterparts. Even elected o�cials at lower levels of government have often been
seen as more accountable to their higher-ups in the central government than to the citizens
that put them in o�ce (Venugopal and Yilmaz, 2010).
In Tanzania’s most recent (2010) elections, turnout was considerably lower than it
had been in any previous election since the transition to multipartyism, with just 39% of
registered voters turning out to vote in Parliamentary contests, compared with over 70%
in all three previous elections. Tripp (2012) suggests that voters believed the outcome to
be a foregone conclusion and/or lacked confidence in the electoral process.
Supporting this interpretation, I note that declines in turnout were significantly larger
in wards that elected ruling party candidates. In 2010, wards that elected CCM councillors
experienced a 51% proportionate decline in turnout compared to the 2005 elections, while
wards that elected opposition councillors saw turnout decline by 33% on average. A similar
pattern emerges at the constituency level. I also find a significant, negative correlation
between turnout decline and ruling party vote margin. That is, turnout declines were
more likely in wards with larger vote margins (less competitive wards). This suggests that
in wards where voters saw CCM victory as a foregone conclusion, they were less likely to
come to the polls.
Failing to vote may also represent a way of expressing opposition to the ruling party.
Overtly supporting an opposition party can have negative consequences, in terms of the
government withholding resources (the punishment regime described above) – or at least
many Tanzanians fear that it can. Furthermore, the ruling party has managed to paint
the opposition as dangerous and disruptive, serving to dissuade voters from supporting
opposition parties even if they are dissatisfied with CCM. Indeed, the CCM’s ability to
30
maintain civil peace is cited as one of the main factors explaining the party’s continued
dominance (Lofchie, 2014; O’Gorman, 2012). During elections the CCM has gone so far as
to show videos depicting the Rwandan genocide as example of what might happen should
the opposition win (Bakari and Whitehead, 2013).
Recent empirical work confirms the notion that turning out to vote in Tanzania tends to
be rewarded. In the only other study of within-district targeting of public goods in Tanzania
of which I am aware – Baird, McIntosh, and Ozler (2013)’s analysis of the Tanzanian Social
Action Fund (TASAF) – the authors find that higher levels of turnout at the district level
are associated with higher numbers of TASAF applications. Within districts, wards with
higher levels of voter turnout receive more TASAF funds per capita.
Beyond fear, poverty also constrains communities’ abilities to e↵ectively express their
demands. In the context of the WSDP, beneficiary communities are expected to raise
initial financial contributions for the capital costs involved in developing water supply and
sanitation facilities (Ministry of Water, 2006b: p. 28). Required community contributions
range from 2.5% of capital costs for gravity-fed or pumped and piped schemes, to 30%
in the case of spring protection. Such contributions can be substantial. For instance,
the average cost of a small, gravity-fed piped scheme was projected to be $76,300 USD
in 2006 (Ministry of Water, 2006b: p. 28). The community contribution in such case
would therefore amount to $1,907.50. While such a figure seems manageable when divided
amongst the 1,500 beneficiaries that ought to benefit from such a scheme, determining who
the beneficiaries will be and how to best raise money from them has proven challenging.
Even small sums can be di�cult when the majority of the population survives on less than
$1.25 per day, as is the case in many rural wards. Furthermore, many Tanzanians regularly
lack access to cash. According to the most recent (2012) Afrobarometer survey, over 80%
of rural respondents reported that in the past year they had gone without a cash income
31
several times or more during the past year.
The WSDP also requires beneficiary communities to open a bank account for their
water and sanitation funds. Given that banks tend to concentrate in urban areas, this
presents another barrier (and also explains the negative sign on distance to nearest road
in my regressions). Indeed, in their study of WSDP implementation in four rural districts
in Tanzania, Jimenez Fernandez de Palencia and Perez-Foguet (2011) also find that more
populated and well communicated villages (with easier access to the bank o�ces located
in the capital) were better able to express their demands (by making the needed cash
contributions).
In sum, my analysis of intra-district allocation of water infrastructure shows that local
politicians tend to reward their supporters, and that they direct resources to those best
able to articulate their demands. This pattern of allocation results in leaving much of the
neediest population behind.
6 Conclusion
This paper began with a striking disconnect between increased funding for Tanzania’s
water sector and limited improvements in access to clean water. I explain this disconnect by
following the money for water from the central government to the level of service provision.
At the national level, political interference in the allocation of money to districts is not
a striking factor. Although the home district of the Minister of Water appears to be
favored, the regime does not disproportionately reward loyal districts or punish districts
that the opposition has captured. The local level is where things get more interesting – and
problematic. First, my analysis suggests substantial capture by local governments. With
the funds they receive from the center for water provision, local governments construct
fewer than half of the water points that the money they received might have allowed.
32
Next, the water points that are constructed only reach half as many people as they could
have, had they been allocated more e�ciently.
The fact that local politicians skim o↵ their allocations is unsurprising, and has been
well-documented elsewhere in the region (Reinikka and Svensson, 2004). More interesting is
the way in which local politicians use public goods provision as a form of vote buying. This
serves to not only secure their careers but also the longevity of the ruling party. Document-
ing this phenomenon therefore represents an important contribution to our understanding
of how hegemonic party regimes survive and endure.
My study also highlights the limits of the ‘demand-responsive approach’ to public ser-
vice delivery. The Tanzanian water sector’s major donors appear to be recognizing this, as
they begin to experiment with results-based financing, paying local government authorities
for each additional well-maintained and functioning water point. Such e↵orts are promising,
though they ultimately serve to promote accountability by the government to Tanzania’s
donors rather than to the country’s citizens. That said, Tanzania’s dependence on tradi-
tional donors has been declining as domestic revenues have increased, Chinese investment
has grown, and a significant amount of natural gas has been discovered (Swedlund, 2013).
While this may limit the ability of donors to promote changes in government behavior,
these new resources in the hands of rural citizens may empower them to demand changes
themselves.
33
References
African Ministers Council on Water. 2010. “Water Supply and Sanitation in Tanzania.”Country Status Overview.
Babeiya, Edwin. 2011. “Electoral Corruption and the Politics of Elections Financing inTanzania.” J. Pol. & L. 4: 91.
Baird, Sarah, Craig McIntosh, and Berk Ozler. 2013. “The Regressive Demands of Demand-Driven Development.” Journal of Public Economics 106 (C): 27–41.
Bakari, Mohammed, and Richard Whitehead. 2013. “Tanzania: Nurturing Legacies of thePast.” In One-Party Dominance in African Democracies, ed. Renske Doorenspleet, andLia Nijzink. Boulder, CO: Lynne Rienner Publishers, Inc. pp. 93–116.
Banerjee, Sudeshna Ghosh, and Elvira Morella. 2011. “Africa’s Water and SanitationInfrastructure: Access, A↵ordability, and Alternatives.” Report for the World Bank’sDirections in Development (Infrastructure) Series.
Banful, Afua Branoah. 2011. “Do formula-based intergovernmental transfer mechanismseliminate politically motivated targeting? Evidence from Ghana.” Journal of Develop-ment Economics 96 (2): 380–390.
Blaydes, Lisa. 2011. Elections and distributive politics in Mubarak’s Egypt. CambridgeUniversity Press.
Bogaards, Matthijs. 2014. “Multiparty Elections in Africa: For Better or Worse.” In PartySystems and Democracy in Africa, ed. Renske Doorenspleet, and Lia Nijzink. New York,NY: Palgrave Macmillan pp. 22–45.
Burgess, Robin, Remi Jedwab, Edward Miguel, Ameet Morjaria, and Gerard Padro iMiquel. 2015. “The Value of Democracy: Evidence from Road Building in Kenya.”American Economic Review 2015 (6): 1817–51.
Diaz-Cayeros, Alberto, Federico Estevez, and Beatriz Magaloni. 2012. Strategies ofVote Buying: Democracy, Clientelism, and Poverty Relief in Mexico. Unpublishedmanuscript.
Do, Quoc-Anh, Kieu-Trang Nguyen, Anh N Tran et al. 2013. “One Mandarin Benefitsthe Whole Clan: Hometown Favoritism in an Authoritarian Regime.” Presented at theAnnual Congress of the European Economic Association in Toulouse.
Faguet, Jean-Paul. 2012. Decentralization and popular democracy: Governance from belowin Bolivia. University of Michigan Press.
34
Gandhi, Jennifer, and Adam Przeworski. 2006. “Cooperation, cooptation, and rebellionunder dictatorships.” Economics & Politics 18 (1): 1–26.
Gandhi, Jennifer, and Ellen Lust-Okar. 2009. “Elections under authoritarianism.” AnnualReview of Political Science 12: 403–422.
Geddes, Barbara. 2005. “Why parties and elections in authoritarian regimes?” Paperpresented at the annual meeting of the American Political Science Association.
Geddes, Barbara. 2008. “Party creation as an autocratic survival strategy.” ”Dictatorships:Their Governance and Social Consequences” Conference at Princeton University.
Geddes, Barbara, Joseph Wright, and Erica Frantz. 2012. “New Data on AutocraticRegimes.” Dataset.
Gine, Ricard, and Agusti Perez-Foguet. 2008. “Sustainability assessment of national ruralwater supply program in Tanzania.” Natural Resources Forum 32: 327–342.
Golden, Miriam, and Brian K Min. 2013. “Distributive Politics Around the World.” AnnualReview of Political Science 16.
Green, Elliott. 2011. “Decentralization and political opposition in contemporary Africa:evidence from Sudan and Ethiopia.” Democratization 18 (5): 1087–1105.
Harris, J. Andrew, and Daniel N. Posner. 2015. “(Under What Conditions) Do PoliticiansReward Their Supporters? Evidence from Kenya’s Constituency Development Fund.”Work in Progress.
Harris, Tim. 2012. “Review of Tanzania’s Water Statistics.” Report commissioned byAccountability in Tanzania.
Harrison, Graham. 2008. “From the global to the local? Governance and development atthe local level: reflections from Tanzania.” Journal of Modern African Studies 46 (02):169–189.
Ho↵man, Barak D, and Lindsay Robinson. 2009. “Tanzania’s Missing Opposition.” Journalof Democracy 20: 123–136.
Hyden, Goran, and Max Mmuya. 2008. “Power and Policy Slippage in Tanzania – Dis-cussing National Ownership of Development.” SIDA Studies No. 21.
Jensen, Peter Sandholt, and Mogens K. Justesen. 2014. “Poverty and vote buying: Survey-based evidence from Africa.” Electoral Studies 33 (2014): 220–232.
Jimenez, A, and A Perez-Foguet. 2010a. “Challenges for Water Governance in Rural WaterSupply: Lessons Learned from Tanzania.” International Journal of Water ResourcesDevelopment 26 (2): 235–248.
35
Jimenez, Alejandro, and Agustı Perez-Foguet. 2010b. “Building the role of local governmentauthorities towards the achievement of the human right to water in rural Tanzania.”Natural Resources Forum 34: 93–105.
Jimenez Fernandez de Palencia, Alejandro, and Agustı Perez-Foguet. 2011. “Implementingpro-poor policies in a decentralized context: the case of the Rural Water Supply andSanitation Program in Tanzania.” Sustainability Science 6 (1): 37–49.
Kramon, Eric. 2013. Vote Buying and Accountability in Democratic Africa PhD thesisUniversity of California, Los Angeles.
Kramon, Eric, and Daniel N. Posner. 2014. “Ethnic Favoritism in Primary Education inKenya.” Unpublished manuscript.
Lange, Siri. 2008. “The depoliticisation of development and the democratisation of politicsin Tanzania: parallel structures as obstacles to delivering services to the poor.” TheJournal of Development Studies 44 (8): 1122–1144.
Lindberg, Sta↵an I. 2006. Democracy and elections in Africa. JHU Press.
Lockwood, Harold, and Stef Smits. 2011. Supporting Rural Water Supply. Moving Towardsa Service Delivery Approach Practical Action Pub.
Lofchie, Michael F. 2014. The Political Economy of Tanzania: Decline and Recovery.Philadelphia: University of Pennyslvania Press.
MacDonald, A.M., H.C. Bonsor, B.E.O.Dochartaigh, and R.G. Taylor. 2012. “Quantitativemaps of groundwater resources in Africa.” Environmental Research Letters 7: 1–7.
Mafuru, Wilhelm, Deogratias Mpenzi, Idda Lyatonga, Oscar Tefurukwa, Orest Masue,Richard Ngowi, and Lameck Wilfred. 2015. “Politico-Administrative Relations in Tan-zanian Local Government Authorities.” Final Report 02nd February 2015.
Magaloni, Beatriz. 2006. Voting for Autocracy. Hegemonic Party Survival and its Demisein Mexico Cambridge University Press.
Magaloni, Beatriz, and Ruth Kricheli. 2010. “Political Order and One-Party Rule.” AnnualReview of Political Science 13: 123–43.
Ministry of Water. 2006a. “Water Sector Development Programme 2006-2025.” Policydocument.
Ministry of Water. 2006b. “Water Sector Development Programme (WSDP) ProgrammeImplementation Manual.” Annex 4: Formula Based Allocation of Financial Resources toLocal Government Authorities.
36
National Electoral Commission. 2006. The Report of the National Electoral Commissionon the 2005 Presidential, Parliamentary and Councillors’ Elections. United Republic ofTanzania.
O’Gorman, Melanie. 2012. “Why the CCM won’t lose: the roots of single-party dominancein Tanzania.” Journal of Contemporary African Studies 30 (2): 313–333.
Ohler, Hannes, and Peter Nunnenkamp. 2014. “Needs-Based Targeting or Favoritism?The Regional Allocation of Multilateral Aid within Recipient Countries.” Kyklos 67 (3):420–446.
Orgut Consulting. 2009. “Tanzania Country Report on Accountability in Local Councils.”Africa Local Council Oversight and Social Accountability Project (ALCOSA).
Oxford Policy Management. 2013. “Water Sector Development Programme 2007-2014:Evaluation of Phase I Final Report.” Assessment carried out for the Ministry of Water,Tanzania.
Pallotti, Arrigo. 2008. “Tanzania: Decentralising Power or Spreading Poverty?” Review ofAfrican Political Economy 35 (116): 221–235.
Quinn, Andrew, and Helen Tilley. 2013. “Rapid Budget Analysis of the Water Sector inTanzania.” Published by the Overseas Development Institute.
Reinikka, Ritva, and Jakob Svensson. 2004. “Local Capture: Evidence from a CentralGovernment Transfer Program in Uganda.” The Quarterly Journal of Economics 119(2): 679–705.
Riedl, Rachel Beatty, and J Tyler Dickovick. 2014. “Party systems and decentralization inAfrica.” Studies in Comparative International Development 49 (3): 321–342.
Rout, Satyapriya. 2014. “Institutional variations in practice of demand responsive ap-proach: evidence from rural water supply in India.” Water Policy 16 (4): 650–668.
Scheiner, Ethan. 2006. Democracy Without Competition In Japan: Opposition Failure ina One-Party Dominant State. New York, NY: Cambridge University Press.
Simonen, Saara. 2010. “Gender Equality in Local Governance in Tanzania.” The North-South Co-operation Programme of Association of Finnish Local and Regional Authori-ties.
Stokes, Susan C, Thad Dunning, Marcelo Nazareno, and Valeria Brusco. 2013. Brokers,Voters, and Clientelism: the puzzle of distributive politics. Cambridge University Press.
37
Swedlund, Haley J. 2013. “From Donorship to Ownership? Budget Support and DonorInfluence in Rwanda and Tanzania.” Public Administration and Development 33: 357–370.
Tatem, AJ, PW Gething, S Bhatt, D Weiss, and C Pezzulo. 2013. “Pi-lot high resolution poverty maps, University of Southampton/Oxford.” Avail-able for download at http://www.worldpop.org.uk/data/summary/?contselect=
Africa&countselect=United+Republic+of+Tanzania&typeselect=Poverty.
TEMCO. 2011. “The 2010 Tanzania General Elections.” Report of the Tanzania ElectionMonitoring Committee (TEMCO).
Tilley, Helen. 2013. “Unblocking Results: Rural Water in Tanzania.” ODI Case Study.
Tripp, Aili Mari. 2012. “Donor Assistance and Political Reform in Tanzania.” UNU - WiderWorking Paper No. 2012/37.
United Republic of Tanzania. 2011. “Annual Assessment of Minimum Conditions and Per-formance Measures for Local Councils under LGDG System for FY 2011/12.” NationalSynthesis Report.
United Republic of Tanzania. 2013. “General Report on LGAs 2011/2012.” Annual GeneralReport of the Controller and Auditor General.
Venugopal, Varsha, and Serdar Yilmaz. 2010. “Decentralization in Tanzania: An assess-ment of local government discretion and accountability.” Public administration and de-velopment 30 (3): 215–231.
Wedgwood, Alison. 2005. “Practical Challenges for Demand Responsive Approaches.”Discussion Paper presented at ESRC Seminar on Access, Poverty and Social Exclusion,March 1st 2005, Overseas Development Institute, London.
Weinstein, Laura. 2011. “The Politics of Government Expenditures in Tanzania, 1999–2007.” African Studies Review 54 (01): 33–57.
Wilfahrt, Martha. 2014. ““They should build us a school too”: The Local Politics ofPrimary School Delivery in Rural Senegal.” Dissertation chapter.
World Health Organization and UNICEF. N.d. “Joint Monitoring Programme for WaterSupply and Sanitation: Data and Estimates.” Available for download at http://www.wssinfo.org/data-estimates/.
Yilmaz, Serdar, Yakup Beris, and Rodrigo Serrano-Berthet. 2010. “Linking local govern-ment discretion and accountability in decentralisation.” Development Policy Review 28(3): 259–293.
38
Figure
1:Formula
Criteriaan
dActual
vs.IdealAllocations,
2007
Unserved
Pop
ulation
,20
06Dom
inan
tExtractionTechnolog
y,20
06
IdealAllocations,
2007
Actual
vs.IdealAllocations,
2007
39
Figure 2: Spending vs. Construction, 2007-2013
40
Figure
3:Allocationof
Water
Pointsin
Mon
duliDistrict,20
06-201
3
Water
Points,
2006
Additional
Water
Points,
2006
-201
3(inred)
Darkerareasofthemapindicategreaterpopulationdensity.
41
Figure 4: WSDP Flow of Funds
42
Figure 5: Distribution of Dependent Variable (Number of Water Points Built)
43
Figure 6: Coe�cient Plot, Proportionate Change in Ward-Level Water Point Coverage
44
Table 1: Scoring of Criteria and Subcriteria
Adopted Criteria Scores
Unserved population 70%
Less than 30% 10%Between 30% and 50% 20%More than 50% 40%
Technology 30%
Gravity scheme 20%Pumping scheme 8%Shallow well with hand pump 2%
Table reproduced from Ministry of Water (2006b: p. 10)
45
Table 2: Summary Statistics (District-Level Variables)
count mean sd min max
Funds Disbursed (Millions of TZS) 601 684.93 864.81 0.00 8735.80Actual Allocation as Proportion of Ideal 601 1.06 1.96 0.00 31.46CCM MP Margin 687 0.50 0.29 -0.28 1.00Minister for Water’s home district 687 0.01 0.12 0.00 1.00Turnout 651 0.65 0.16 0.27 0.93Audit Opinion 599 2.63 0.50 1.00 3.00Poverty Rate (% under %1.25/day) 687 0.82 0.07 0.65 0.93Population (thousands) 687 319.10 160.87 45.38 1009.94Area (km squared) 687 9381.45 8471.76 627.62 49601.80Depth to Groundwater (meters) 687 3.83 2.05 0.94 9.72
46
Table 3: Summary Statistics (Ward-Level Variables)
count mean sd min max
Number of water points built 19216 1.07 3.86 0.00 100.00Water point stock 19216 21.66 22.70 0.00 313.00CCM councillor won last election 18252 0.90 0.29 0.00 1.00CCM councillor margin in last election 15807 0.38 0.30 -0.72 0.97CCM councillor’s vote share in last election 15807 0.67 0.15 0.00 1.15Turnout in Last Election (Proxy) 18252 0.54 0.38 0.00 4.66Poverty Rate (% under 1.25 per day) 2402 0.81 0.09 0.26 0.95Population Density (People per Km. Sq.) 2402 181.84 459.54 0.10 9997.56Distance from Nearest Road (meters) 2402 3935.03 5896.54 0.00 56849.72Depth to Groundwater (meters) 2402 3.75 2.43 0.39 20.49Waterpoint Coverage Rate 19216 0.54 0.55 0.00 7.152006 Functionality Rate (Proxy) 2255 1.26 2.83 0.00 63.00
47
Table 4: DV = Actual Allocation to Districts as Proportion of Ideal, 2007-2013
(1) (2) (3) (4)Model Model Model Model
CCM MP Margin 0.03(0.52)
Minister for Water’s home district 2.88⇤⇤⇤ 2.88⇤⇤⇤ 2.86⇤⇤⇤ 2.86⇤⇤⇤
(0.97) (0.97) (0.97) (0.97)Turnout -0.21 -0.22 -0.26 0.02
(0.69) (0.68) (0.69) (1.07)Audit Opinion -0.04 -0.03 -0.03 -0.04
(0.16) (0.16) (0.16) (0.16)Poverty Rate (% under %1.25/day) -3.63 -3.57 -3.61 -3.64
(2.49) (2.49) (2.50) (2.50)Population (log) 0.34 0.33 0.36 0.45
(0.31) (0.30) (0.31) (0.52)Area (log) 0.17 0.16 0.16 0.17
(0.21) (0.21) (0.21) (0.21)Depth to Groundwater (meters) 0.13 0.13 0.13 0.13
(0.08) (0.08) (0.08) (0.08)District represented by opposition -0.28
(0.74)CCM lost dominance of district -0.25
(0.49)Votes for CCM Presidential Candidate (log) -0.11
(0.41)
Observations 489 489 489 489
Standard errors in parenthesesThe dependent variable is the actual allocation as a proportion of the ideal formula allocation.All models restricted to rural districts where year of construction is not missing.All models assume first-order autocorrelation within panels and are estimated using mixed e↵ects linearregression fit via maximum likelihood.⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
48
Table 5: DV = Number of Waterpoints Built (Negative Binomial Regression with WardFixed E↵ects)
(1) (2) (3) (4)Model Model Model Model
CCM councillor won last election 0.12⇤
(0.07)
Turnout in Last Election (Proxy) 0.37⇤⇤⇤ 0.37⇤⇤⇤ 0.37⇤⇤⇤ 0.37⇤⇤⇤
(0.06) (0.08) (0.08) (0.06)
L.Existing waterpoint stock -0.00⇤ -0.00⇤⇤ -0.00⇤⇤ -0.00⇤
(0.00) (0.00) (0.00) (0.00)
CCM councillor margin in last election 0.14(0.08)
CCM councillor’s vote share in last election 0.33⇤⇤
(0.16)
Councillor aligned with CCM MP in last election 0.13⇤⇤
(0.06)
Observations 12175 10468 10468 12175
Standard errors in parenthesesThe dependent variable is a count of waterpoints built.Fixed-e↵ects negative binomial regression.All models exclude urban wards and those where data on year of construction is missing.All models include time trend.⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
49
Table 6: DV = Proportionate Change in Ward-Level Water Point Coverage
(1) (2) (3) (4)Model Model Model Model
CCM Councillor Margin (2005) 0.00(0.03)
Turnout, 2005 (Proxy) 0.09⇤⇤⇤ 0.06⇤⇤⇤
(0.02) (0.02)
Waterpoint Coverage Rate (2006) -0.12⇤⇤⇤ -0.12⇤⇤⇤ -0.10⇤⇤⇤ -0.11⇤⇤⇤
(0.01) (0.01) (0.01) (0.01)
2006 Functionality Rate (Proxy) -0.01 0.01 -0.01 0.00(0.02) (0.02) (0.02) (0.02)
Poverty (2010, % under USD 1.25) -0.29⇤⇤ -0.15 -0.26⇤⇤ -0.20⇤
(0.13) (0.12) (0.13) (0.11)
Log of Population Density -0.03⇤⇤⇤ -0.02⇤⇤⇤ -0.03⇤⇤⇤ -0.03⇤⇤⇤
(0.01) (0.01) (0.01) (0.01)
Log of Distance to Nearest Road -0.01⇤⇤⇤ -0.00⇤⇤ -0.01⇤⇤⇤ -0.01⇤⇤⇤
(0.00) (0.00) (0.00) (0.00)
Depth to Groundwater (meters) 0.00 0.00 0.00 0.00(0.00) (0.00) (0.00) (0.00)
CCM Councillor Margin (2010) 0.04⇤
(0.03)
Turnout, 2010 (Proxy) 0.30⇤⇤⇤ 0.13⇤⇤⇤
(0.06) (0.03)
Councilor Aligned with CCM MP, 2005 0.01(0.03)
Councilor Aligned with CCM MP, 2010 0.02(0.02)
Observations 1628 1796 1789 2229
Standard errors in parenthesesThe dependent variable is improvement in waterpoint coverage, 2005-2013.Multi-level model with wards clustered into districts.⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
50
Money Flows, Water Trickles:
Explaining the Disconnect between Spending
and Improved Access to Clean Water in Tanzania
Appendix
September 14, 2015
Contents
A1 Variable Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-2
A1.1Deviation from Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-2
A1.2Minister of Water’s Home District . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-4
A2 Dealing with New/Split Districts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-4
A3 Assigning Weights to Different Types of Water Points . . . . . . . . . . . . . . . . . . . . . . APP-4
A4 Additional Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-7
A4.1Financial Allocations Across Districts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-7
A4.2Distribution of Waterpoints within Wards . . . . . . . . . . . . . . . . . . . . . . . . . . . APP-11
APP-1
A1 Variable Construction
A1.1 Deviation from Formula
The formula described in the WSDP Program Implementation Manual (Ministry of Water, 2006) allows for
the derivation of allocation coefficients for each district, which can then be applied to the entire development
budget allocation for water for a given year in order to determine the district’s ideal allocation according to
the formula. Specifically, the formula for allocation of development budget to the districts is equal to d ⇤ Y ,
Y is the total development budget allocation, and d is the coefficient to be applied, which is the sum of t
(the technology coefficient) and u (the coefficient for unserved population).
For gravity schemes t is 0.2/m1, where m1 is the total number of districts with gravity schemes and
protected springs are dominant by at least 60% of the coverage. For pumping schemes it is 0.08/m2 where
m2 is the total number of districts with pumping schemes dominant by at least 60% of the coverage. For
shallow wells it is 0.02/m3 where m3 is the total number of districts with shallow wells, deep wells with
hand pumps and traditional sources developed are dominant by 60% of the coverage.
u is the coefficient to be applied depending on the unserved population. For unserved population < 30%
it is 0.1/h1 where h1 is the total number of LGAs with unserved population less than 30%. For unserved
population between 30% to 50% it is 0.2/h2 where h2 is the total number of LGAs with unserved population
between 30% to 50%. For unserved population > 50% it is 0.4/h3 where h3 is the total number of LGAs
with unserved population >50%.
I derive the underlying formula criteria using the water point mapping (WPM) database. Given that the
WPM database has information on the year of construction for each water point, I can derive an estimate
of the existing water point stock for each district for each year from 2006-2012. I calculate the proportion
unserved as per the Ministry of Water’s guidelines: First, I estimate the ‘served’ population by taking the
stock of water points in a given year and multiplying it by 250, corresponding to the Ministry’s assumption
that each water point serves 250 people. I divide the ‘served’ population by the total population of each
district in each year to derive the proportion served. The proportion unserved is simply 1 minus this figure.
In order to determine the dominant extraction type, I once again leverage information from the WPM
database on each water point’s source ( sourceg in the WPM database) well as the extraction technology
(extractc in the WPM database). Table A1 below shows the equivalents between the classification of water
APP-2
points in the PIM and the classification I undertake by hand.
Table A1: Scoring of Criteria and Subcriteria
PIM Classification Author’s Classification
Shallow wells, deep wells with hand pumps orimproved traditional sources
sourceg == “shallowwell”|
extractc == “handpump”
Gravity schemes and protected springs extractc == “gravity”
Pumping schemes extractc == “motorpump”|“submersible”
I am unable to derive allocation coefficients for a number of districts for two reasons. First, some districts
have no one extraction technology that accounts for 60% or more of all water points. For instance, in 2006,
of the 64 districts for which I have data on the year of construction, there are 29 districts in which no single
extraction technology exceeds 60%. In light of this, I construct an alternate measure that considers which
of the extraction technologies makes up a majority of the water points (i.e., which exceeds 50%). Even
by that alternate measure, I still identify 9 districts where no single extraction technology holds a majority.
I therefore construct a third alternate measure that simply considers which extraction technology has the
plurality in each district for each year. This allows me to classify all districts, and is what I ultimately use
for variable construction.
Additionally, the year of construction is missing for 8,762 waterpoints (11.7%) of the 75,178 water
points in the WPM database. I therefore exclude districts where year of construction is missing for over
50% of all water points from my analysis, since in these districts I cannot distinguish between water points
that are part of the existing stock and those that were newly constructed between 2007 and 2013. Ultimately,
I construct allocation coefficients based only on the universe of districts for which I have data on year of
construction for the majority of water points I then determine ideal formula allocations by applying the
allocation coefficients for each district to the total amount of money allocated to this restricted universe of
districts.
My data on allocations is from the Ministry of Water’s Management Information System (MIS).1 The
MIS allows for the tracking of project funds to different components of the WSDP. I create reports for each
district from 2007-2013, showing how much money they are allocated for rural water supply. Given that1http://www.mowimis.go.tz/
APP-3
the MIS is a project tracking system, I assume that the funds allocated for water projects are essentially
equivalent to the development budget.
A1.2 Minister of Water’s Home District
The table below provides background information for use in my construction of the dummy variable indi-
cating whether a district is home to the current Minister of Water.
Table A2: Minister of Water’s Home District, 2006-2015
Minister Year Entered Year Left Home District Home Region
Shukuru Kawambwa 2006 2008 Bagamoyo CoastMark Mwandosya 2008 2012 Rungwe Mbeya
Jumanne Maghembe 2012 NA Mwanga Kilimanjaro
A2 Dealing with New/Split Districts
Between 2006 and 2013, Tanzania added over 30 districts – in keeping with a trend common to sub-
Saharan Africa, where almost half of the countries have increased their number of administrative units
by at least 20% since 1990, following decentralization reforms (Grossman and Lewis, 2014). Given that
Tanzania’s land mass has not expanded, all of the new districts have been carved out of existing districts.
Though I have not been able to obtain any official record denoting the timing and process of district creation,
I have been able to determine the ‘parent’ wards of all newly created districts by comparing election results
for 2005 and 2010, comparing shape files from the 2002 and 2012 Census, and conducting additional Internet
searches where necessary. In order to analyze changes over time since beginning of WSDP, I collapse all
new rural districts in with their "parents."
A3 Assigning Weights to Different Types of Water Points
The WSDP Program Implementation Manual (Ministry of Water, 2006) provides information on unit
costs for capital investment in new water systems, which I use to construct weights for the different water
point types in the WPM database. The unit costs are given in Table A3.
APP-4
Table A3: Unit Costs for Capital Investment in New Water Systems*
Technology Average cost/system (USD) Population served
Shallow well and handpump 2,100 250Borehole & Hand Pump 6,150 250Gravity Fed and Piped (Small) 76,300 1,500Gravity Fed and Piped (Large) 84,800 2,500Electric or Diesel Pumped and Piped (Small) 64,000 1,500Electric or Diesel Pumped and Piped (Large) 71,300 2,500Protected Spring 900 250Windmill 8,000 250Rainwater Catchment 4,335 500Charco Dam 15,600 1,500I reproduce Table 4 from the Program Implementation Manual in part.
APP-5
Using information from the on extraction-class and water source, I weight water points as follows:
• Shallow wells with handpump (weight = 1)
• Boreholes with handpumps (weight = 3)
• Motor pumps (weight = 4)
• Gravity schemes (weight = 5)
These four classes make up 87.8% of waterpoints in database. I weight all others the same as shallow wells
with handpumps.
APP-6
A4 Additional Tables
A4.1 Financial Allocations Across Districts
Table A4: DV = Actual Allocation to Districts as Proportion of Ideal, 2007-2013 (Fixed Effects Regression)
(1) (2) (3) (4)Model Model Model Model
Audit Opinion 0.04 0.04 0.04 0.07(0.13) (0.14) (0.14) (0.13)
CCM MP Margin -0.07(0.53)
Minister for Water’s home district 2.34 2.34 2.35 2.33(2.19) (2.19) (2.19) (2.20)
Turnout -0.30 -0.31 -0.43 2.85⇤
(0.34) (0.41) (0.33) (1.58)District represented by opposition 0.11
(0.24)CCM lost dominance of district -0.38
(0.57)Votes for CCM Presidential Candidate (log) -1.74⇤⇤
(0.73)
Observations 489 489 489 489R
2 0.013 0.013 0.015 0.020Standard errors in parenthesesThe dependent variable is the actual allocation as a proportion of the ideal formula allocation.All models include district fixed effects and time trend.All models restricted to rural districts where year of construction is not missing.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-7
Table A5: DV = Actual Allocation to Districts as Proportion of Ideal, 2007-2013 (Mixed Effects Regression,Excluding Minister’s Home District)
(1) (2) (3) (4)Model Model Model Model
CCM MP Margin 0.06(0.54)
Turnout -0.24 -0.25 -0.30 0.23(0.71) (0.69) (0.70) (1.10)
Audit Opinion -0.05 -0.05 -0.04 -0.05(0.16) (0.16) (0.16) (0.16)
Poverty Rate (2010, 1.25) -3.69 -3.63 -3.67 -3.71(2.64) (2.64) (2.65) (2.64)
Population (log) 0.34 0.33 0.36 0.56(0.32) (0.32) (0.33) (0.55)
Area (log) 0.14 0.13 0.12 0.14(0.22) (0.22) (0.22) (0.22)
Depth to Groundwater (meters) 0.13 0.13 0.13 0.12(0.09) (0.09) (0.09) (0.09)
District represented by opposition -0.31(0.76)
CCM lost dominance of district -0.32(0.50)
Votes for CCM Presidential Candidate (log) -0.23(0.43)
Observations 489 489 489 489Standard errors in parenthesesThe dependent variable is the actual allocation as a proportion of the ideal formula allocation.All models restricted to rural districts where year of construction is not missing.All models assume first-order autocorrelation within panels and are estimated using mixed effects linearregression fit via maximum likelihood.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-8
Table A6: DV = Log of Actual Allocation to Districts, 2007-2013 (Mixed Effects Regression)
(1) (2) (3) (4) (5)Model Model Model Model Model
L.% Unserved -0.20(0.21)
L.% gravity schemes -0.02(0.21)
L.Audit Opinion 0.31⇤⇤⇤
(0.11)CCM MP Margin -0.01
(0.21)Minister for Water’s home district 1.10⇤⇤ 1.09⇤⇤ 1.11⇤⇤ 1.10⇤⇤
(0.49) (0.49) (0.49) (0.49)Turnout -2.85⇤⇤⇤ -2.86⇤⇤⇤ -2.79⇤⇤⇤ -2.77⇤⇤⇤
(0.32) (0.32) (0.32) (0.46)Poverty Rate (% under %1.25/day) 0.14 0.18 0.12 0.15
(0.83) (0.83) (0.82) (0.83)Population (log) 0.25⇤⇤ 0.25⇤⇤ 0.24⇤⇤ 0.30
(0.11) (0.11) (0.11) (0.20)Area (log) -0.04 -0.05 -0.03 -0.04
(0.08) (0.08) (0.08) (0.08)Depth to Groundwater (meters) -0.01 -0.01 -0.01 -0.01
(0.03) (0.03) (0.03) (0.03)District represented by opposition -0.17
(0.33)CCM lost dominance of district 0.18
(0.21)Votes for CCM Presidential Candidate (log) -0.04
(0.16)
Observations 555 591 591 591 591Standard errors in parenthesesThe dependent variable is the log of the actual allocation to districts.All models restricted to rural districts; Models 2-5 restricted to those for which year of construction is not missing.All models assume first-order autocorrelation within panels and are estimated using mixed effects linearregression fit via maximum likelihood.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-9
Table A7: DV = Log of Actual Allocation to Districts, 2007-2013 (Fixed Effects Regression)
(1) (2) (3) (4) (5)Model Model Model Model Model
L.% Unserved -3.33⇤⇤⇤
(0.83)L.% gravity schemes -3.69
(2.79)L.Audit Opinion 0.32⇤⇤⇤
(0.11)CCM MP Margin 0.19
(0.46)Minister for Water’s home district 1.78 1.78 1.78 1.77
(1.59) (1.59) (1.59) (1.60)Turnout -2.25⇤⇤⇤ -2.22⇤⇤⇤ -2.21⇤⇤⇤ -0.96
(0.40) (0.35) (0.36) (0.94)District represented by opposition -0.25
(0.46)CCM lost dominance of district -0.05
(0.23)Votes for CCM Presidential Candidate (log) -0.67
(0.51)
Observations 555 503 503 503 503R
2 0.058 0.086 0.087 0.086 0.089Standard errors in parenthesesThe dependent variable is the log of the actual allocation to districts.All models restricted to rural districts; Models 2-5 restricted to those for which year of construction is not missing.All models assume first-order autocorrelation within panels and are estimated using mixed effects linearregression fit via maximum likelihood.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-10
A4.2 Distribution of Waterpoints within Wards
Table A8: DV = Log of New Waterpoints Per Capita (Regression with Ward Fixed Effects)
(1) (2) (3) (4)Model Model Model Model
CCM councillor won last election 0.00(0.00)
Turnout in Last Election (Proxy) 0.00⇤⇤⇤ 0.00⇤⇤⇤ 0.00⇤⇤⇤ 0.00⇤⇤⇤
(0.00) (0.00) (0.00) (0.00)
L.Water point stock -0.00⇤⇤⇤ -0.00⇤⇤⇤ -0.00⇤⇤⇤ -0.00⇤⇤⇤
(0.00) (0.00) (0.00) (0.00)
CCM councillor margin in last election 0.00(0.00)
CCM councillor’s vote share in last election 0.00(0.00)
Councillor aligned with CCM MP in last election 0.00(0.00)
Observations 16383 14111 14111 16383Standard errors in parenthesesThe dependent variable is the log of new waterpoints built per capita.Fixed-effects negative binomial regression.All models exclude urban wards and those where data on year of construction is missing.All models include time trend.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-11
Table A9: DV = Waterpoint Construction Occurred (Logistic Regression with Ward Fixed Effects)
(1) (2) (3) (4)Model Model Model Model
CCM councillor won last election 0.25⇤⇤
(0.12)
Turnout in Last Election (Proxy) 0.90⇤⇤⇤ 1.29⇤⇤⇤ 1.29⇤⇤⇤ 0.90⇤⇤⇤
(0.13) (0.18) (0.18) (0.13)
L.Existing waterpoint stock -0.12⇤⇤⇤ -0.12⇤⇤⇤ -0.12⇤⇤⇤ -0.12⇤⇤⇤
(0.01) (0.01) (0.01) (0.01)
CCM councillor margin in last election 0.14(0.17)
CCM councillor’s vote share in last election 0.21(0.33)
Councillor aligned with CCM MP in last election 0.12(0.10)
Observations 12147 10408 10408 12147Standard errors in parenthesesThe dependent variable is a dummy variable indicating whether waterpoint construction occurred.Fixed-effects logistic regression.All models exclude urban wards and those where data on year of construction is missing.All models include time trend.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-12
Table A10: DV = Proportionate Change in % of Ward Pop. within 1km of Water Point
(1) (2) (3) (4)Model Model Model Model
CCM Councillor Margin (2005) 0.13(0.23)
Turnout, 2005 (Proxy) -0.37⇤⇤ -0.34⇤⇤
(0.16) (0.14)
Ward residents within 1km of clean water, 2006 -0.00⇤⇤⇤ -0.00⇤⇤⇤ -0.00⇤⇤⇤ -0.00⇤⇤⇤
(0.00) (0.00) (0.00) (0.00)
2006 Functionality Rate (Proxy) -0.11 0.08 -0.13 -0.03(0.20) (0.20) (0.20) (0.17)
Poverty (2010, % under USD 1.25) -1.70⇤ -3.19⇤⇤⇤ -2.23⇤⇤ -2.54⇤⇤
(1.02) (1.08) (1.10) (0.99)
Log of Population Density -0.01 -0.11⇤ -0.01 -0.09(0.07) (0.07) (0.07) (0.06)
Log of Distance to Nearest Road -0.01 -0.03⇤ -0.00 -0.02(0.02) (0.02) (0.02) (0.02)
Depth to Groundwater (meters) 0.05 0.07⇤⇤ 0.03 0.04(0.03) (0.04) (0.03) (0.03)
CCM Councillor Margin (2010) -0.17(0.22)
Turnout, 2010 (Proxy) -0.94⇤ -0.01(0.52) (0.28)
Councilor Aligned with CCM MP, 2005 -0.09(0.23)
Councilor Aligned with CCM MP, 2010 -0.01(0.15)
Observations 1490 1637 1637 2038Standard errors in parenthesesThe dependent variable is improvement in access (proportion of residents within 1km of waterpoint), 2005-2013.Multi-level model with wards clustered into districts.⇤p < 0.10, ⇤⇤
p < 0.05, ⇤⇤⇤p < 0.01
APP-13
Appendix: References
Grossman, Guy, and Janet I. Lewis. 2014. “Administrative Unit Proliferation.” American Political Science
Review 108 (1): 196–217.
Ministry of Water. 2006. “Water Sector Development Programme (WSDP) Programme Implementation
Manual.” Annex 4: Formula Based Allocation of Financial Resources to Local Government Authorities.
APP-14