Bad Neighbors? How Co-located Chinese and World Bank Development Projects

Impact Local Corruption in Tanzania

Please Cite as:

Brazys, Samuel; Elkink, Johan A.; Kelly, Gina (2017) 'Bad Neighbors? How co-located Chinese and World Bank Development Projects Impact Local Corruption in Tanzania'. Review of International Organizations, 12 (2):227-253.

Samuel Brazys

Johan A. Elkink

Gina Kelly1

1 Dr. Samuel Brazys is corresponding author (samuel.brazys@ucd.ie) and assistant professor of international relations in the School of Politics and International Relations at University College Dublin. Johan A. Elkink (jos.elkink@ucd.ie) is assistant professor of social science research methods in the School of Politics and International Relations at University College Dublin. Gina Kelly (gikelly@tcd.ie) is a Policy Analyst at the Commission for Energy Regulation and a graduate of the Trinity College Dublin-University College Dublin Masters in Development Practice Programme.

1

1 Introduction

The rise of China as a development partner has been one of the most important phenomena in

the international development field over the past decade. As a “non-traditional” donor that

has, to date, eschewed the OECD’s Development Assistance Committee (DAC), and its

classifications of “Official Development Assistance” (ODA), China is forging a new path in

development assistance (Kim and Lightfoot 2011; De Haan 2011). While China’s rise creates

space for “South-South” or “Triangular” modes of development cooperation, the implications

of this emergence are not yet fully understood. The lack of transparency in Chinese aid

programs and the apparently uninterested stance towards the governance implications of

development lead many to wonder if Chinese engagement will contribute to or undermine the

efforts of traditional development partners (Zimmermann and Smith 2011; Abdenur 2014; De

Haan 2011; Strange et al. 2015).

These concerns bring the Chinese development ascendancy directly into the ongoing debate

over the relationship between good governance and positive development outcomes. As

noted by Bräutigam and Knack (2004), the World Bank (1989: 60) has argued that

“underlying the litany of Africa’s development problems is a crisis of governance”. Many

African states are characterised by high levels of corruption, a lack of accountability, poor

institutions and a weak rule of law (Bräutigam 2011). A substantial recent literature has

considered how development assistance impacts this governance. Scholars have argued there

is a “curse” of development efforts on institutions, that there is no such political curse at all,

or that there is a more nuanced, non-linear, relationship between development flows and

governance (Djankov et al 2008; Altincekic and Bearce 2014; Brazys 2016). Some of the

ambiguity in this relationship perhaps rests on the fact that the interaction between

development projects and good governance may depend not only on which recipient is

2

receiving the development flow, but also on which donor is providing it (Schudel 2008).

Particular donors have been explicit in their efforts to use development flows to improve

governance. One of the leaders in this effort is the World Bank with its system of

“performance-based allocation” (Hout 2007). Indeed, recent evidence suggests that flows

from the World Bank are both allocated to, and associated with, countries with better

governance (Winters 2010; Okada and Samreth 2012). Therefore, it may be that while

development flows from some donors promote good governance (or at least do not hinder it),

flows from other donors undermine governance and institutional quality. Yet little work has

considered what happens if the impacts of development efforts on governance from different

donors are at cross-purposes. Additionally, we are aware of no published work that considers

the micro-level relationship between development flows and governance at the local level.

Accordingly, in this paper we add to a growing literature that focuses on the emerging

development actor China in three ways: first, by asking whether Chinese development flows

undermine local good governance; second, by evaluating whether the type of development

flow affects the impact of Chinese development efforts on local governance; and third, by

asking what the impact is on local governance when donors with different stances towards

good governance have projects that are co-located.

In order to examine these questions, this paper takes advantage of recent innovations in

development data to conduct a micro-level study relating perceptions of corruption to

proximity to development projects. Corruption is a particularly insidious form of poor

governance, with evidence suggesting that increases in corruption both reduce growth and

increase income inequality (Gyimah-Brempong 2002; Ugur 2014). Corruption is recognised

internationally as a significant problem, with ‘grand’ corruption – the diversion of public

funds meant for development – having significant impacts on welfare and political and

3

economic reform and with ‘petty’ corruption placing a greater burden on the most vulnerable

in society (Richmond and Alpin 2013). As with broader studies on the relationship between

development flows and governance, the evidence on the relationship between aid and

corruption is inconclusive, ranging from findings that suggest ODA reduces corruption, to

findings that ODA increases corruption, to studies that again suggest the result depends on

the donor, with multilateral donor efforts associated with reduced corruption and bilateral

efforts showing no relationship (Okada and Samreth 2012; Asongu and Nwachukwu 2014;

Charron 2011).

In order to evaluate the relationship between development flows and local corruption we

combine geo-located data on Chinese and World Bank development projects from the

AidData initiative with similarly geo-referenced data from household surveys in Tanzania to

employ a spatial identification strategy. These citizen surveys allow for a direct investigation

of how proximity to projects is correlated with citizen perceptions of, and experiences with,

local corruption. We focus our study on Tanzania for two reasons. First, Tanzania is one of

the few countries in Africa where the history and scope of Chinese engagement is roughly

comparable to that of the World Bank. Both have been important actors in Tanzania’s

development since the early 1960s and as such we can avoid any bias as a result of

sequencing or primary or “lead” donorship (Steinwand 2015). Second, focusing on one

country allows us to narrow in on issues of donor and project heterogeneity that are crucial to

our theoretical expectations and empirical investigation.

We argue that the relationship between development flows and local corruption is nuanced.

We do indeed find evidence that Chinese projects are associated with increased experiences

and, to a limited extent, perceptions, of corruption when controlling for co-location with

4

World Bank projects, but, crucially, that this finding is predicated on the type of Chinese

development flow. Only “other official flows” (OOF) projects are associated with increased

corruption, while “ODA-like” projects show no relationship with worsening governance.

Finally, while we find that World Bank projects, on their own, are associated with reduced

corruption experiences, we find that when Chinese and World Bank projects are co-located,

both contribute to higher levels of experienced corruption. These findings suggest that the

impact of development flows on local corruption may depend on the donor, flow type and the

spatial proximity to other projects.

2 Theorizing Development Flows and Local Corruption

We take our theoretical cues from recent scholarship on the determinants of corruption that

has focused on the relationship between experiences and perceptions of corruption (Olken

2009; Olken and Pande 2011; Gutmann et al. 2015; Belousova et al. 2016; Donchev and

Ujhelyi 2014). This research concludes that perceptions and experiences of corruption are not

analogous and that the latter only weakly predict the former. As discussed by Donchev and

Ujhelyi (2014) and Gutmann et al. (2015), individual corruption perception is the more

encompassing of the two phenomena, as it is a function not only of individual corruption

experience, but also of awareness of corruption experiences of others or reported corruption

in the media, as well as individual and aggregate characteristics such as religion, ethnicity,

and economic status that may bias perceptions vis-à-vis actual experience. Accordingly, the

pool of individuals who may perceive corruption is substantially broader than those who

experience corruption, as the latter may be a sufficient, but not necessary, cause of the former

(Belousova et al. 2016).

5

Recognizing this difference, we build and evaluate theory of development flows on local

corruption experiences and perceptions. We argue that different causal mechanisms underlie

these two effects. Further, drawing on empirical literature that suggests development flows

may both increase or decrease corruption (Okada and Samreth 2012; Charron 2011), we

develop theoretical mechanisms with two-sided expectations that are dependent on donor and

project characteristics. That the corruption effects of development projects are donor-

dependent draws on the literature that examines donor heterogeneity in foreign aid

approaches and outcomes (Berthelemy 2006; Kilby and Dreher 2010; Brazys 2013). We first

develop our general theoretical perspective before suggesting the implications in Tanzania for

the two donors in this study.

2.1 Development Flows and Local Corruption Experience

We suggest that development projects may influence corruption experiences via three

mechanisms: direct experience, general growth effects, and normative change (Gutmann et al.

2015; Knutsen et al. 2016; Isaksson 2015). In this text, we consider bribes as synonymous

with the corruption experience outcome as household surveys indicate it is this type of

behaviour which is most universally understood as corruption.2 Projects may influence

citizens’ direct experience with corruption by creating opportunities for bribes via

involvement and/or access. First, if citizens are locally involved in the construction or supply

of a project as sub-contractors they may experience corruption as they may be required to pay

2 A survey conducted by the Front Against Corrupt Elements in Tanzania (FACEIT) and the Tanzanian

Prevention and Combating of Corruption Bureau (PCCB) in 2009 found that most citizens understand

corruption as demand for unofficial payment (92.5%), as opposed to demand for sex (29.4%) or abuse of power

(25.9%). Perception beyond these facets of corruption was limited, with respondents failing to perceive informal

payments for services or embezzlement and fraud as corrupt practices (PCCB, 2009).

6

a bribe to secure contracts or provide materials. Second, the project may develop or improve

a local private or toll good such as a utility (water and electric supplies, information and

communications infrastructure, etc.); or school, hospital, stadium, or transportation facilities,

which may facilitate the extraction of bribes for access to these goods. In essence, the

presence of these goods potentially increases the “demand” for corruption, similar to the logic

in Knutsen et al.’s (2016, p. 4) analysis of local corruption and mines that suggests that the

presence of mines may attract a higher number of local officials who are then demanding

bribes. A development project may bring a water main or internet connection to a village, or

build a school, but this creates an opportunity for a bribe to be demanded for a household to

access the good. However, if the project is replacing or improving existing infrastructure it

may also have the opposite effect if the donor implements transparency and accountability

mechanisms that were not present in the existing infrastructure. While an existing purely

local utility may have extracted bribes, officials working on a project developed in

conjunction with a donor partner may not be able to extract bribes if donor monitoring is

effective.

The effect of development aid on corruption experiences by individuals can also be more

indirect, via change in the prevailing corruption norms among officials. As suggested by

Isaksson (2016, p. 81), if donors engage in “high level” corrupt practices this may alter the

norm for local officials who then become more inclined to engage in corrupt practices of their

own. Likewise, if donors simply turn a blind eye to the behaviour of partner officials with

whom they work, this may be enough to allow corruption to flourish. Alternatively, if donors

actively seek to implement or institutionalize anti-corruption norms through education or

conditionality, this may lead to normative feedback that leads to an environment of reduced

corruption (Isaksson and Kotsadam 2016).

7

An alternative indirect effect relates to the supply side of bribes as a result of new resources

resulting from development projects. The logic is that increased local resources, for example

as a result of mining, can put upward pressure on wages, which creates wealth that officials

can tap into through bribing (Knutsen et al. 2016). While Knutsen et al. are sceptical of this

mechanism, we think the logic remains applicable to development projects. Increased local

resources may increase the ability of citizens to pay bribes. Unlike the previous two

mechanisms, this mechanism would serve only to increase corruption experience. Thus,

development projects may either increase or decrease the experience of corruption via a

direct effect, a norm effect, or increase corruption via a growth effect.

2.2 Development Projects and Local Corruption Perception

We suggest that related but conceptually distinct causal mechanisms drive the relationship

between development projects and perceptions of corruption. Indeed, many of the ways by

which development projects might be related to corruption are unlikely to involve direct

experience for citizens. Actual corruption related to the locating of projects, permit issues for

projects, or the use of a preferred contractor are all unlikely to involve general citizen

engagement (Dreher et al. 2015). However, as noted by Olken and Pande (2011), perceptions

of corruption can also be based on hearsay of someone else’s direct experience of corruption

with the project or via a stereotype regarding the donor’s aptitude for corruption irrespective

of any actual corruption. Thus, perceptions of corruption may include perceptions both of

“high level” maleficence such as nepotism, cronyism, or coercion; or of the prevalence of

8

low-level or personal corruption among a respondent and/or her associates (Afrobarometer,

2006)3.

The likelihood of hearsay of someone else’s experience with corruption should be

monotonically related to the amount of actual corruption, although not necessarily

homogenous in degree one, as this hearsay of actual corruption can be enhanced or mitigated

by the extent of information dissemination. If an instance of corruption is widely reported via

formal media or informal information networks, perceptions of corruption may increase by a

factor greater than the increase in actual corruption. Conversely, in a low-transparency

environment, perceptions may increase by a factor less than experience.

However, projects implemented with low levels of transparency may lead to an impression of

corruption, especially when there are preconceptions about a donor. These stereotypes are

likely to be rooted in the donor’s history in a particular country. In order for these donor

impressions to lead to a change in perceptions of local corruption, citizens would need to be

able to accurately identify the donors that are operating in their locality. However, recent

work suggests that citizens may not have a high degree of information about the branding of

development projects in their vicinity (Dietrich et al. 2015; Baldwin and Winters 2016).

Therefore, beyond “direct corruption” experience driving corruption perception, as evidenced

in Gutmann et al. (2015), there are two additional causal mechanisms to link development

projects to perceptions of corruption: “hearsay” of actual experience and “impressions” based

on donor characteristics or stereotypes.

3 Afrobarometer briefing paper (2006) Combating Corruption in Tanzania: perceptions and experience.

Available at: http://www.afrobarometer.org/publications/bp33-combating-corruption-tanzania-

perception-and-experience Accessed April 5, 2016.

9

2.3 Heterogeneous Donor Effect Hypotheses

The two sided expectations regarding donor impact on corruption may stem from whether a

donor’s projects are subject to “aid capture” or “donor control” (Milner et al. 2016). With

regard to the former, recipient governments and officials that already have a proclivity for

corruption can use development projects to maintain power via inefficient use of the aid

resources. In respect to the latter, donors may use a higher degree of oversight to both limit

fungibility, but also to achieve policy objectives, which can include good governance (Milner

et al. 2016, p. 223-224). Milner et al. (2016) indeed suggest heterogeneity in donor

approaches and the empirical literature supports this assertion (Berthelemy 2006).

Accordingly, the extent to which a particular donor enables “aid capture” or promotes “donor

control” will influence the direction of the relationship for the different causal mechanisms

linking development projects and corruption discussed above. The implication is that

expectations need to be donor-specific.

With respect to the two donors in this study, China and the World Bank, there are strong

empirical and theoretical grounds for thinking the former’s projects may facilitate “aid

capture” while the latter’s are likely to manifest “donor control.” China’s approach to

engagement as a development partner has differed significantly from that of the DAC donors,

with efforts often focusing on infrastructure building and providing concessional loans to

countries without conditionality (Wang and Elliot 2014). Chinese involvement in

development has also largely eschewed “best practice” principles which have been

established in the DAC over decades. In particular, China has paid little heed to the

principles developed in the High-Level Fora on Aid Effectiveness outcomes in the Paris

10

Declaration and Accra Accords and has been, at best, a tepid participant in the Busan cum

Global Partnership for Effective Development Cooperation (Mawdsley et al. 2014). Beyond

limited rhetorical support for the ideals of the aid effectiveness movement, evidence suggests

that China has not adhered to these principles (Strange et al. 2015), instead advocating for

‘South-South’ cooperation without the limitations imposed by the OECD and others. While

this alternative focus has been appreciated by countries who have felt unduly constrained by

DAC conditionality, the approach has also met with both local and international backlash

(Zhao 2014). Several authors note that Chinese aid may be easier to exploit, i.e. subject to

“aid capture”, than that provided by other donors, for example by politicians involved in

patronage politics, due to the Chinese principle of non-interference in the domestic affairs of

recipient countries (Brautigam 2009; Tull 2006; Dreher et al. 2016; Isaksson and Kotsadam

2016). Many authors have expressed concern that China’s principle of non-interference,

providing unconditional aid and investment regardless of human rights or governance

considerations, is obstructing reforms to governance and accountability in African countries

(Xiaobing and Ozanne 2000; Collier 2007; Pehnelt 2007; Strange et al. 2013). Beyond a

principled non-interference stance, Chinese non-interference may also be a practical matter.

As China is still perfecting its development efforts, China may have not yet developed the

capacity to effectively monitor and oversee its development projects4. Non-interference,

for principled or practical reasons, is likely to be key in driving the likelihood of “aid

capture” for Chinese projects.

Thus, with respect to corruption experience, even if China does not directly engage in any

corruption itself, non-interference means they may not monitor or sanction officials who

engage in corrupt practices related to their projects. For instance, a Chinese project may bring

a utility to an area, but Chinese officials may then be indifferent to local officials charging a 4 We thank an anonymous reviewer for this insight.

11

bribe for connection to that utility. Non-interference may also lead to an increase in

experienced corruption, as China is unlikely to promote anti-corruption norms and indeed

may implicitly sanction corrupt behaviour. Finally, the new resources combined with non-

interference facilitate the growth mechanism outlined above.

Our expectation that Chinese projects will contribute to the experience of corruption means

that the projects should also contribute to perceived levels of corruption. As noted by

Muchapondwa et al. (2014), emerging donors, for the most part, do not take part in aid

reporting regimes such as the International Aid Transparency Initiative (IATI) or the OECD’s

Creditor Reporting System (CRS). While this opaqueness may reduce the likelihood of

hearsay, it leaves open the possibility of an impression that Chinese development projects

increase corruption. Indeed, as Zhao (2014, p. 1041) explains, “the lack of transparency in

China’s business deals facilitates corruption” and, indeed, China has been accused of

engaging with corrupt states and elites in exchange for access to resources (Pehnelt 2007).

Impressions of China in Tanzania are likely to be based on historical interactions that are

some of China’s oldest and deepest on the continent. China first established a diplomatic

relationship with Tanganyika (now mainland Tanzania) in 1961 (Sigalla 2014). Following

independence, under the rule of Julius Nyerere, the country initially aimed to follow a

socialist path of development and to lessen its dependence on the West, while developing a

closer relationship with China (Moshi et al. 2008). Today Tanzania is one of China’s top ten

preferred investment countries on the African continent (Hinga and Yiguan 2013). In 2011,

there were over 400 Chinese enterprises and 20,000 individuals from China operating in

Tanzania, and trade between China and Tanzania increased by 40% in 2010 (Sigalla 2014).

12

China has directed significant development flows to Tanzania for over 40 years, providing

over two billion dollars for a large number of projects, including the Urafiki Textile Factory,

the Benjamin Mkapa National Stadium, the Mwalimu Nyerere International Convention

Centre and the Zambia and Tanzania railway which linked Dar es Salaam and Kapiri Moshi,

one of the largest overseas projects China has ever undertaken (Bailey 1975; Furukawa

2014). In addition to this infrastructure, China has provided hundreds of medical teams,

constructed advanced medical facilities and built large scale state farms and farmer training

stations to promote agricultural development. It has built a number of rural primary schools

and granted scholarships to hundreds of Tanzanian university students (Furukawa 2014).

Chinese military assistance to Tanzania is also significant, through provision of equipment

and training to the Tanzanian army and the construction of military bases (Moshi et al. 2008).

Tanzania’s importance to China is due not only to its extensive endowment of natural

resources but also to the gateway function it serves via the Indian Ocean to the rest of Africa.

Impressions of China will be influenced by the national and international media reporting of

corrupt practices and bribes related to procurement and natural resources in relation to

Chinese officials and projects in Tanzania5. Beyond this, Round 6 of the Afrobarometer

survey was analysed by Mwombela (2015)6 to see if Chinese engagement in Tanzania is 5 E.g. LaFraniere, S. and J. Grobler “China Spreads Aid in Africa, With a Catch” New York Times, September

21, 2009; “Report: Chinese smuggled ivory out of Tanzania during state visit” Al Jazeera, November 6, 2014;

“Chinese bribes in Dar, admits China Envoy”, The Citizen, July 15, 2014.

6 Mwombela, Stephen (2015) What shapes Tanzanians’ image of China? Findings from the Afrobarometer

Round 6 Survey in Tanzania, Repoa, Policy Research for Development. [ONLINE] Available at:

http://www.repoa.or.tz/images/uploads/TAN_R6_dissemination_2_Chinese_Engagement_25Feb2015

.pdf [Accessed 15 August 2015]

National Bureau of Statistics, Statistics for Development. 2015. [ONLINE] Available

at :http://www.nbs.go.tz/. [Accessed 25 August 2015].

13

considered as positive or negative by Tanzanians. He finds that China’s economic and

political influence in Tanzania is viewed mostly positively, and China is perceived as having

more influence on Tanzania than the USA, UK, India, South Africa, the UN or the World

Bank (Mwombela, 2015). The factors that contribute to a positive impression of Chinese

engagement in Tanzania are its investment in infrastructure and the low cost of Chinese

products, while Chinese economic activities taking jobs and business from local people and a

perceived low quality of Chinese products leads to negative perceptions (Mwombela, 2015).

Likewise, a recent survey experiment in Uganda found that opinions of Chinese development

projects are no worse than those of World Bank and United States projects (Milner et al.

2016). Beyond this, there is little quantitative evidence that China and other emerging

partners have worsened governance in Africa (Collier 2007; Alves 2013). Thus, while we

expect Chinese development projects to increase perceptions of corruption, we suspect that

this happens more via the mechanisms of experience (both direct and hearsay), than by

inference based on existing perceptions or stereotypes of Chinese engagement.

Hypothesis 1: A higher number of local Chinese aid projects will correlate with a higher

likelihood of local corruption perceptions and experiences in Tanzania.

Conversely, we expect that the World Bank is more reflective of “donor control” theory

(Milner et al. 2016), and indeed Annen and Knack (2016) formally lay out the mechanism by

which this might take place. As Charron (2011) explains, since 1997, a number of

multilateral aid donors, including the World Bank, shifted the focus of their aid to encourage

good governance practices. The World Bank has explicitly allocated its aid based on

“Country Policy and Institutional Assessment” (CPIA) scores that consider, among other

dimensions, transparency, accountability and corruption (Molenares et al. 2015; Smets and

14

Knack 2016). Rather than non-interference, these efforts indicate an explicit focus by the

World Bank to actively engage with partner countries to reduce corruption. As such, the

World Bank is likely to condemn and sanction any corruption it finds within its partner

projects. Indeed, the World Bank has been explicitly involved in a “fight against corruption”

in Tanzania since 1995 (World Bank 1998, p. iv; Leeuw et al. 1999). These efforts have

consisted of building national integrity systems that included engagement across a range of

actors and issues to directly stem corrupt practices, but also to alter prevailing corruption

norms (Leeuw et al 1999). More broadly, previous research has shown that aid projects from

multilateral donors, and in particular the World Bank, correlate with reduced levels of

corruption, particularly since 1997 (Okada and Samreth 2012; Charron 2011)7. The World

Bank’s project oversight should avoid the increase in experienced corruption that can result

from development without such oversight and indeed could reduce experienced corruption if

the project substitutes for existing structures that facilitated corruption. Furthermore, World

Bank policies may stimulate a local normative environment that promotes good governance,

further reducing experienced corruption.

As with China above, once again the relationship between World Bank projects and

perceptions of corruption is nuanced. We expect that the World Bank’s hypothesized impact

on corruption experience will lead to lower perceptions of corruption based on direct or

hearsay experience. However, it remains difficult to assess the extent to which existing views

about the World Bank will lead to a positive impression effect. Despite the importance of the

World Bank as an aid donor there is surprisingly little research on perceptions of the World

Bank in the developing world. One exception is Breen and Gillanders (2015), who find that

those who had experienced corruption held less positive views of the World Bank. While this

7 However, as Winters (2014) shows, this aim is not always achieved, due to heterogeneous project effects that are further elaborated below.

15

finding poses an endogeneity challenge, one interpretation is that perceptions of the World

Bank are associated with corruption.

With regard to the historical experience, the World Bank’s breadth and depth of engagement

in Tanzania is similar to that of China, as it has been active in what is now Tanzania since

1959 (Payer 1983). As Payer notes, the early relationship was focused largely on agricultural

projects and was tepid due to Tanzania’s socialist leanings, which were at times at odds with

the World Bank’s preferred development approach8. This low-level antagonism culminated

with a struggle beginning in 1979-1980 over the implementation of a World

Bank/International Monetary Fund (IMF) structural adjustment program that ultimately

ended with a Tanzanian “capitulation” in 1985 as a result of World Bank “coercion” (Holtom

2005, p. 549). While the relationship has smoothed in more recent years, this somewhat rocky

history may suggest mixed attitudes towards the World Bank’s presence. Looking at

descriptive data from Round 6 of the Afrobarometer, the World Bank, explicitly mentioned

but only as an example (along with the UN) of an international organization, was rated as the

“most influential” actor by only 31 of the 2,072 respondents (1.5%) who had an opinion

(Afrobarometer 6). This suggests that regardless of the direction of the impression about the

World Bank and corruption, the salience of the World Bank in defining a respondent’s

overall perception of corruption is likely to be low. Finally, as discussed above, respondents

in Uganda had impressions of World Bank aid indistinguishable from Chinese aid (Milner et

al. 2016). Thus, as with China, we expect that the direct experience and hearsay mechanisms

will dominate the impression mechanism for the relationship between World Bank projects

and perceptions of corruption and we hypothesize that:

Hypothesis 2: A higher number of local World Bank aid projects will correlate with a

decreased likelihood of local corruption experiences and perceptions in Tanzania. 8 Indeed, the World Bank refused to fund the Tanzania-Zimbabwe railroad subsequently financed by China.

16

2.4 Heterogeneous Project Effects and Co-Location Hypotheses

The theory above suggests that some types of projects may be more or less prone to act as

vehicles for linking development flows and corruption. Brazys (2016, p. 298) notes that aid

can be used to provide either “public goods” or “selective benefits” and that these differing

provisions in aid will have opposite effects on governance. Milner et al. (2016) link this

variation in types of benefits to that in aid capture. These findings imply there are likely to be

heterogeneous project effects in the relationship between development projects and local

corruption. We would expect projects that supply private or toll goods (i.e. goods that are

excludable, or provide “selective benefits”) are more likely to be related to changes in

corruption perception and experience vis-à-vis projects that supply intangible or non-

excludable goods. Excludability is necessary in order to provide the opportunity for extra-

legal exchange. As such, pure public goods projects, such as national debt forgiveness or

technical assistance based policy projects are unlikely to provide an opportunity for an

exacerbation or mitigation of individual-level corruption experience. Conversely, projects

that provide direct cash or in-kind assistance, or those that provide excludable infrastructure,

are more likely to be linked to corruption, as individuals may have to engage in extra-legal

exchange to gain or maintain access to the good. Beyond this, Winters (2014, p. 394) shows

how more “delimited” aid projects, or those that are more precisely targeted, are less likely to

lead to corruption. A targeted project, say, to build a single school in a village, is likely to

have greater transparency and accountability mechanisms in place than a broad national roads

or electrification project, spread over a number of sites, that is necessarily more difficult to

oversee and thus may be prone to facilitating local corruption. Accordingly, we hypothesize:

17

Hypothesis 3: “Selective benefit” development projects are more likely to have an

association with experienced and perceived corruption than “public good” development

projects.

If our expectations above are correct, then Chinese and World Bank development projects

will have opposite effects on corruption. What then happens when projects with inducing and

mitigating effects on corruption are co-located? The logic in Winters (2014, p. 393) is again

instructive when considering the consequences of development project co-location. Co-

located projects may have overlapping implementers, stakeholders, responsibilities and,

indeed, outcomes, which contribute to an increasingly complex information environment.

This may make monitoring efforts increasingly difficult as individuals may be implementers,

officials and/or beneficiaries of multiple projects, but a donor may only have remit for

oversight on its particular project.

Therefore, the World Bank will be less able to implement its good governance policies

around aid. Note that for the mechanisms outlined above, all development projects create new

opportunities for corruption, regardless of the donor, but that the effects for the World Bank

differ through conditionality on local conditions, monitoring, and establishing good

governance norms. However, where projects from China and the World Bank co-locate, the

World Bank will have decreased ability to implement and enforce these policies. As the

prospect of a sanction becomes less likely, the cost of corruption decreases and corruption is

more likely to occur. Indeed, recent cross-country empirical work by Hernandez (2015) finds

that World Bank conditionality is less stringent in countries that receive assistance from

China, Kuwait or the United Arab Emirates because the World Bank must “compete” with

these donors to maintain a presence in the countries. This mechanism may also operate at a

18

local level where local World Bank projects have less oversight so they can remain attractive

to participation from local stakeholders in locations that also have Chinese projects.

Accordingly, our expectation with respect to the co-location of development projects and

corruption is that:

Hypothesis 4: An increasing number of co-located development projects between China and

the World Bank will increase the likelihood of corruption experience and perceptions as a

result of either donor’s projects.

One final consideration of project heterogeneity rests on Dreher et al’s (2015) important

delineation of the types of development flows. As China eschews DAC principles on

classification and reporting, the distinction between different types of Chinese development

flows is often murky. Using the classification codes from the AidData database, Dreher et al

(2015) find that Chinese “ODA-like” flows are driven more by foreign policy goals, and

indeed are linked to recipient needs, while “OOF-like” flows are more closely linked to

commercial interests, in particular resource endowments. Reflecting on these findings, it

seems plausible that “ODA-like” flows may be less influential on corruption perceptions and

experience vis-à-vis their “OOF-like” counterparts, as much of the evidence surrounding

Chinese corrupt practices is linked to commercial endeavours. Zhu’s (2016) recent analysis

finds that the presence of multinational corporations (MNCs) in China increases corruption.

Central to the argument is that the presence of MNCs creates (local) rents, the capture of

which provide incentives for corruption. By extension, the commercial and rent-creating

Chinese “MNC-like” activity abroad (i.e. OOF) may operate in a similar fashion, leading to

an increased impact on corruption vis-à-vis ODA-like flows.

19

Hypothesis 5: Chinese “ODA-like” flows are less likely to be associated with corruption

experiences and perceptions than Chinese “OOF-like” flows.

3 Evaluating Development Projects and Local Corruption

In order to identify a relationship between citizen perceptions and experiences of corruption

and Chinese and World Bank development projects we employ a spatial strategy. This spatial

identification approach has been used in several recent studies on aid (Dreher et al. 2016;

Manda et al. 2014) and allows for a micro-level evaluation of aid impact. Proximity and

distance are often utilized in the broader social sciences as a proxy measure to represent

different social processes based on Tobler’s “First Law of Geography” (Manda et al. 2014;

Tobler 1970). We expect that perceptions and experiences of corruption from particular

projects or endeavors will be stronger for those nearer to the events. Indeed, Lopez-Valcarcel

and Jimenez (2014) recently found that local corruption contagion is inversely related to

distance.

3.1 Dependent Variables

To gain an understanding of the impact of proximity to aid projects on perceptions of and

experiences with corruption, we make use of individual level survey data. In particular, we

make use of the fact that the most well-known survey in Africa, the Afrobarometer, allows

for geo-locating the respondents. This provides an opportunity to link this survey data to the

geo-referenced and project-level AidData database – similar to, for example, Milner et al.

(2016), Findley et al. (2015) or Isaksson and Kotsadam (2016). In other words, we have

ward-level measures of proximity to aid projects – as well as a number of ward-level control

20

variables – which we then link to individual-level data on perceptions and reported

experience – as well as a number of individual-level control variables.

Our primary outcome variables are measures of perceptions and experiences of corruption

from Round 6 of the Afrobarometer survey, conducted in August and September of 2014.9

Afrobarometer applies a clustered sampling strategy, typically sampling eight individual

respondents within each town or village visited. We geo-code this data in a manner similar to

that used in Knutsen et al. (2016), by taking the centroid of the ward reported in the

Afrobarometer data as the geo-location of the respondent. Wards are a relatively small spatial

delineation in the Tanzanian structure, often referring to segments of larger cities or rural

parts of larger regions, with a total of 3,644 wards in the country. The ward thus provides a

relatively fine-grained measure of geo-location. Afrobarometer 6 covers 209 wards with

typically one or two villages per ward.

9 “Summary of Results. Afrobarometer Round 6 Survey in Tanzania, 2014” available at :

http://afrobarometer.org/sites/default/files/publications/Summary%20of%20results/tan_r6_sor_en.pdf accessed

22/10/2016. We utilize only the 6th round of the Afrobarometer for several reasons. Practically, many projects,

including the 2nd phase of the national fiberoptic project (discussed further below) have completion dates after

earlier rounds of the survey, including the 5th round. Additionally, while AidData has made excellent strides

towards capturing the timing of the projects, we remain sceptical about the potential for close temporal

identification due to the fact that the data often include the same start or end dates at all project locations for a

project at multiple locations. Moreover, it is unclear to us that the timing for corruption would necessarily

immediately follow a start or end date. Accordingly, we are most comfortable using the latest round of surveys

and taking the cumulative projects over the period, with the idea that our theoretical mechanisms are likely to

lead to changes that endure or repeat over time. A project that could convincingly tackle issues of temporal

identification would be a significant step forward in this literature.

21

Our primary indicator of corruption perceptions is based on Question 53 of the survey: “How

many of the following people do you think are involved in corruption, or haven’t you heard

enough about them to say?” across a number of types of officials. In the first instance, in

Table 1, we create a binary indicator that equals “1” if the respondent thought any amount of

people were involved in corruption across any of the categories and equals “0” otherwise. We

only code this variable for respondents who did not respond “Don’t know/Haven’t heard” for

all categories. However, some categories are a better operationalization of perceptions of

local corruption than others, which may be more strongly influenced by perceptions of

national or global corruption. In particular, response category “D”, which asks the question in

the context of “local government councillors”, response category “F”, which relates to “local

government tax collectors” and category “H” for “traditional leaders”, who tend to be local,

have explicit local orientation. Accordingly, we consider ordered responses to these and other

categories in Table 2. As a robustness check, we also take advantage of an independent data

source to formulate the dependent variable for corruption perceptions. The 2013 REPOA

Citizen’s Survey, which was carried out across 1,203 households in 44 village locations and

six case councils, asked the question: “Is corruption a serious problem in your council?”10

Results using this data are presented in Section A3 of Online Appendix I.

Our preferred metric of corruption experience is based on Question 55, which is asked in two

parts. In the first part, the question asks if the respondent had an interaction with a number of

different government institutions. The second part then asks: “And how often, if ever, did you

have to pay a bribe, give a gift, or do a favour…?” As paying a bribe is always local, if the

respondent answers “yes” we construct a binary indicator across seven institutions that equals

10 The response categories to this question were ordinal, with “yes,” “no,” “maybe” and “don’t know” as the

permissible options. We recode the data as a binary variable coded as 1 for “yes” answers and as 0 for answers

of “maybe” and “no”, dropping the “don’t know” responses.

22

“1” if the respondent had paid any bribe to any institution and equals “0” otherwise. We only

code this variable for respondents who had an interaction with at least one of the institutions.

As with our perceptions question, we also take advantage of the fact that we can disaggregate

across the different institutions, which include utilities, permits, police, courts, hospitals and

schools, to check our results on each subset of corruption experience in Table 4.

3.2 Independent Variables

Our primary indicator of Chinese and World Bank involvement in Tanzania are instances of

Chinese and World Bank Development projects as captured by the AidData datasets.

Historically it has been difficult to assess the extent of Chinese aid efforts (in Africa and

elsewhere) due both to a lack of transparency and of conceptual clarity over the types of

activities that may be constituted as aid, trade or investment (Bräutigam 2009; Alves 2013;

Sun 2014; Dreher et al 2016). However, this difficulty is being overcome via AidData’s

Tracking Under-reported Financial Flows (TUFF) effort that has compiled a database on

Chinese project locations11 in Africa between 2000 and 2013. Their geocoded dataset details

1,673 project locations across 50 African countries for this time period and includes a

classification scheme of Chinese official and unofficial finance (Strange et al. 2015). We

code 297 total Chinese project locations over the period, however, only 139 are coded at a

sufficient level of precision to be used in our spatial analysis12. Similarly, we utilize

AidData’s “World Bank IBRD-IDA, Level 1” dataset that codes 1,035 project locations from

1995 to 2014, although once again only 275 of these are at a sufficient level of geographic

11 Importantly, this data maps project locations such that the same project may be listed in multiple locations.

We discuss this in more depth below.

12 We only include projects coded with precision code “1” or “2” by AidData.

23

precision for our analysis13. For both donors, we utilize the “total projects” measures, as some

projects that were not completed may have been halted due to corruption issues, which would

contaminate our analysis. We employ these total counts as the variables China and World

Bank in the tables below. To evaluate Hypothesis 3, we code a subset of these projects as the

variables China Infrastructure and World Bank Infrastructure based on whether the record

refers to an excludable infrastructure project. Finally, to evaluate Hypothesis 5, we take

advantage of the granularity in the AidData database and separate Chinese infrastructure

projects that are ODA-like, China ODA Infrastructure, from those that are OOF-like, China

OOF Infrastructure.

We calculate the Euclidean distance from each ward to the nearest respective location of each

development project and subsequently count the number of projects within a specified

distance. We have no a priori theoretical rationale for a specific radius other than the notion

that projects must be sufficiently close for villagers to be aware of their existence and any

surrounding issues of corruption. The results presented in Table 1 below utilize a 40

kilometer (km) radius, but we consider the model at other radii as discussed further below,

expecting that increased proximity to the project increases the likelihood of the project

influencing experience and perception of corruption. Utilizing ArcGIS we represent the

projects graphically in Figure 1 below. The map shows considerable co-location between

Chinese and World Bank projects. While this artifact is useful for our hypothesis test, it does

raise a concern about multicollinearity, and indeed the simple correlation coefficient between

our China and World Bank measures is 0.85. However, checking the variance inflation

factors (VIFs) across these two and the other geographically-based measures in our models

suggests multicollinearity is unlikely to be a problem.14 Additionally, as the main

consequence of multicollinearity is large standard errors, which is a conservative bias in 13 Full lists of both Chinese and World Bank projects used in the analysis can be found in Appendix 1.

24

terms of confirming our expectations, we make no further corrections. As a robustness check

we also provide estimates for the impact of Chinese and World Bank projects on both

perceptions and bribes for a range of different radii, from 10 to 150 km distance from the

survey respondents. These results, based on Model 2 in Tables 1 and 3, for perceptions and

experiences respectively, are presented graphically in Figures 4 and 5 below.

Figure 1: Chinese Aid and World Bank Aid Projects and Respondent Wards

14 The mean VIF between China, World Bank, Resources, the Urban indicator and the Government Party

indicator is 2.40 with a maximum VIF of 4.55. The condition number for these variables is 7.88.

25

3.3 Control Variables

We incorporate a number of baseline control variables, largely drawing from Attila (2009)

and Gutmann et al. (2015). In the corruption perception models we control for those who

indicated they had paid a bribe. In all models we use measures of the respondent’s age, sex,

education, income, occupation as well as indicators of whether the respondent’s location is

urban or rural, in a constituency whose parliamentarian is from the ruling party Chama Cha

Mapinzundi (CCM) as of the 2015 election, or near to natural resources. The latter two

variables are not found in Gutmann et al. (2015) or Attila (2009), but recent empirical

findings suggest these are important control variables. Knutsen et al. (2016) find significant

evidence that the presence of mining increases local corruption in Africa, and Dettman and

Pepinsky (2016, p. 2) find that local resources are associated with decreased “accountability

of politicians and bureaucrats”. With regard to local political control, Dreher et al. (2015)

find that Chinese development projects are more likely to be allocated to a leader’s birth

regions. If we fail to control for either the presence of resources or the local political affinity,

we may omit potential confounding factors on the relation between development aid and

local corruption. Source information and summary statistics on all data can be found in

online Appendix I.

3.4 Estimation Strategy and Results

Our estimation strategy uses discrete choice models because of the binary and ordinal nature

of our outcome variables. However, there are also important reasons to expect strong spatial

autocorrelation in our observations. Corruption will be affected by aid projects and socio-

economic factors, but also by corruption in geographically proximate areas. One possibility to

26

address this spatial autocorrelation is to allow for a spatial autoregressive component in the

error term, the so-called Spatial Error Model (SEM) (McMillen 1992; LeSage 2000). For

models with a binary dependent variable there are significant complications in estimating the

model, although recent implementations allow for the estimation of the SEM probit (Fleming

2004; Calabrese and Elkink 2014; Wilhelm and Godinho de Matos 2013). In our analysis,

however, the effect of the clustering will swamp the impact of a potential diffusion effect of

corruption such that a more straightforward multilevel modelling will more appropriately

address the concern of a lack of independence in the observations ( Case 1991; Case 1992).

We therefore add random intercepts for each ward within our sample and use a mixed-effects

(ordered) probit model, which is our preferred specification and is used in the tables below.

We evaluate perceptions of corruption in Tables 1 and 2, and experiences of corruption in

Tables 3 and 4.

TABLES 1 AND 2 ABOUT HERE

FIGURE 1 ABOUT HERE

Table 1: Perceptions of Corruption (Binary)

Model 1 Model 2 Model 3China 0.0190

(1.16)World Bank -0.0168

(1.40)China Infrastructure 0.0285

(0.86)0.0114(0.21)

World Bank Infrastructure -0.0157(0.80)

-0.0205(0.90)

China*World Bank 0.0017(0.41)

Baseline Controls Yes Yes YesWard Mixed Effects Yes Yes YesN 1836 1836 1836Prob > χ2 0.0000 0.0000 0.0000

Absolute value of z-scores in parentheses. † p>0.10, *5% p>0.05, **1% p>0.01

27

0.5

11.

5H

isto

gram

of C

ount

Wor

ld B

ank

-.1-.0

50

.05

.1C

oeffi

cien

ts a

nd 9

5% C

Is

0 2 4 6 8 10 12 14 16Count World Bank Aid

Corruption PerceptionsFigure 2: Coefficient on China by World Bank

0.5

11.

52

His

togr

am o

f Cou

nt C

hina

-.1-.0

50

.05

.1C

oeffi

cien

ts a

nd 9

5% C

Is

0 1 2 3 4 5 6 7 8 9 10 11 12Count Chinese Aid

Corruption ExperienceFigure 3: Coefficient on World Bank by China

28

Table 2: Perceptions of Corruption (Ordered)

Model 1Local

Model 2President

Model 3Parliament

Model 4Bureaucrat

Model 5Judge

Model 6Police

Model 7Tax

Model 7.1Tax

Model 8Traditional

Model 8.1Traditional

China Infrastructure

0.0232(1.08)

-0.0024(0.09)

-0.0011(0.05)

0.0026(0.12)

-0.0006(0.03)

-0.0059(0.29)

0.0530*(2.40)

0.0474†(1.84)

World Bank Infrastructure

0.0031(0.24)

-0.0040(0.24)

0.0061(0.48)

0.0090(0.70)

-0.0031(0.25)

0.0084(0.69)

-0.0296*(2.23)

-0.0337*(2.51)

-0.0415**(2.66)

-0.0482**(3.10)

China ODA Infrastructure

-0.0079(0.19)

-0.0680(1.35)

China OOF Infrastructure

0.0976**(2.91)

0.1283**(3.37)

Baseline Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes YesWard Mixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes YesN 1749 1612 1664 1727 1741 1805 1664 1664 1494 1494Prob > χ2 0.0000 0.0000 0.0006 0.0004 0.0039 0.0000 0.0014 0.0008 0.0000 0.0000

Absolute value of z-scores in parentheses. † p>0.10, *5% p>0.05, **1% p>0.01

Table 4: Experiences of Corruption (Ordered)

Model 1(Utility)

Model 2(Utility)

Model 3(Permit)

Model 4(Police)

Model 5(Court)

Model 6(Hospital)

Model 7(School)

China Infrastructure 0.1684*(2.54)

0.0423(0.63)

0.2042**(3.12)

0.1551*(2.01)

0.0587†(1.71)

0.0832*(2.07)

World Bank Infrastructure -0.0671(1.63)

-0.0327(0.91)

-0.0043(0.11)

-0.1361**(3.23)

-0.1080*(2.13)

-0.0188(0.91)

-0.0066(0.26)

China Fiber (OOF) Infrastructure 0.3322*(2.37)

China Water (ODA) Infrastructure -0.1327(0.47)

Baseline Controls Yes Yes Yes Yes Yes Yes YesWard Mixed Effects Yes Yes Yes Yes Yes Yes YesN 353 353 464 396 265 1668 1137Prob > χ2 0.0005 0.0013 0.0001 0.0001 0.0249 0.0000 0.0000

29

Absolute value of z-scores in parentheses. † p>0.10, *5% p>0.05, **1% p>0.01

30

3.4.1 Results on Perceptions of Corruption

We present the results on our key variables in Tables 1 and 2 which show qualified support

for our hypotheses. Full model results, including a baseline model of the controls, can be

found in Section A2 of Online Appendix I. In Model 1, the aggregated binary specification,

there is no statistically significant relationship between proximity to either Chinese or World

Bank projects and perceptions of corruption, either when considering all projects (Model 1)

or only infrastructure projects (Model 2). As neither World Bank nor Chinese projects have a

significant effect, it is unsurprising that the co-location interaction (Model 3) is also

insignificant.

There is, however, some support for Hypotheses 1 and 2 in Table 2. While the coefficients

for the local officials (Model 1) are insignificant, the coefficients on tax officials (Model 7)

and traditional leaders (Model 8) are in the expected direction and statistically significant.

This finding is worth further exploration, but suggests that different types of officials may be

differently prone to being seen as engaging in corruption as a result of local development

projects.15 Moreover, there is considerable support for Hypothesis 5, the relative effect of

Chinese ODA-like versus OOF-like projects, in Models 7.1 and 8.1, which both find a strong

and statistically significant effect of OOF-like projects increasing perceptions of corruption

while finding no statistically significant relationship between ODA-like projects and

perceptions of corruption.

15 There is also support for Hypotheses 1, 2 and 4 for corruption perceptions when using the REPOA data. These

results are presented and discussed in section A3 of Online Appendix I. There remains no support, however, for

Hypothesis 4 in the Afrobarometer data as shown in Figure A1 in Online Appendix I.

31

Table 3: Experiences of Corruption (Binary)

Model 1 Model 2 Model 3 Model 4China 0.0378*

(2.51)World Bank -0.0268*

(2.32)China Infrastructure 0.0672*

(2.23)-0.0082(0.16)

World Bank Infrastructure -0.0309†(1.68)

-0.0343†(1.78)

-0.0522*(2.41)

China ODA Infrastructure -0.0369(0.50)

China OOF Infrastructure 0.1311**(2.79)

China*World Bank 0.0074†(2.89)

Baseline Controls Yes Yes Yes YesWard Mixed Effects Yes Yes Yes YesN 1902 1902 1902 1902Prob > χ2 0.0000 0.0000 0.0000 0.0000

Absolute value of z-scores in parentheses. † p>0.10, *5% p>0.05, **1% p>0.01

TABLES 3 AND 4 ABOUT HERE.

FIGURE 2 ABOUT HERE

3.4.2 Results on Experiences of Corruption

Turning to corruption experiences, we find more consistent support for our hypotheses. In the

binary models in Table 3 all of our hypotheses are supported, Chinese projects are associated

with more experiences of corruption (Model 1), supporting Hypothesis 1, and this effect is

more pronounced when considering only infrastructure projects (Model 2), supporting

Hypothesis 3. This relationship only holds, however, for OOF-like projects (Model 3), as

Chinese ODA-projects show no statistically significant relationship with increased

experiences of corruption (Hypothesis 5). Conversely, World Bank projects are associated

with fewer experiences of corruption (Models 1-3) (Hypothesis 2) and, in particular, have

their strongest impact on corruption in the absence of any co-located Chinese projects (Model

32

4). Indeed, co-location of projects increases the impact on corruption for either World Bank

or Chinese projects (Model 4) (Hypothesis 4). Plotting this relationship in Figure 3 shows

that World Bank projects are negatively associated with experiences of corruption at a

statistically significant level for 0 or 1 co-located Chinese projects, but this relationship

becomes statistically insignificant at higher numbers of Chinese projects16.

Table 4 similarly shows broad support for our hypotheses, with the sign and significance

holding across a number of different institutions.17 Model 2 is particularly noteworthy as we

are able to further exploit the detail of the AidData dataset to identify specific Chinese

development projects that closely map to one of the Afrobarometer corruption experience

questions, 55H, which asks if respondents have had to pay a bribe in order to get utilities

from the government. Two major Chinese development efforts in our dataset include a

national fiberoptic project that was implemented in two phases at 55 project locations and

regional water-supply projects, implemented in 35 project locations. Interestingly for our

purposes, the fiber project is classified by AidData as “Vague (Official Finance).” The

project is a 200 million USD endeavour between the Chinese International

Telecommunication Construction Corporation (CITCC) and Chinese telecommunications

giant Huawei Technologies and the Tanzania Telecommunications Company TTCL.18 The

project is clearly commercial in nature and has already come under criticism both as a result

of contract disputes with subcontractors and for the absence of price decreases with the 16 All interactions plots were generated modifying code used to produce similar figures in Copelovitch and Ohls

(2012).

17 Interaction models of those presented in Table 4 also show general support for Hypothesis 4, as shown in

Figure A2 in Online Appendix I.

18 http://www.ictworks.org/2011/05/09/why-tanzanian-internet-access-prices-havent-decreased-arrival-seacom/

Accessed 28-10-2016

http://allafrica.com/stories/201411040466.html Accessed 28-10-2016.

33

arrival of the service.19 This absence of price decreases suggests a capturing of local rents,

which may be facilitating the corruption experience, as suggested by our theory above. Model

2 in Table 4 shows a large positive and statistically significant effect between the location of

these fiber projects and increased experience of paying bribes to access government utilities.

Conversely, the water projects are all classified as “ODA-like.” Chinese contributions include

support to water projects, including the Chalinze Water Supply Project (CWSP) and the

Dodoma City Water project that are part of the broader Water Sector Development Program

that is supported by the Bank of Arab Development in Africa, the Japanese International

Cooperation Agency (JICA), the African Development Bank, the UK Department for

International Development, and the World Bank, among others.20 As shown in Table 4,

Model 2, there is no statistically significant relationship between the presence of these

Chinese water projects and an increase in experience of paying bribes for utilities. This result,

combined with the fiber result above, is strong micro-level evidence in support of Hypothesis

5 that Chinese ODA-like projects are less likely to be associated with increased corruption

compared to OOF-like projects. This result is consistent with the theory as the water project

is much more of a public utility rather than a commercial good.

As noted above, our selection of a 40 km radius for determining proximity to projects and

resources is based on a rationale that proximity to projects influences local perceptions and

19 http://www.ictworks.org/2011/05/09/why-tanzanian-internet-access-prices-havent-decreased-arrival-seacom/

Accessed 28-10-2016

http://allafrica.com/stories/201608290091.html Accessed 28-20-2016.

20 http://www.chaliwasa.go.tz/index.php/en/about-us/overview Accessed 28-10-2016

http://tanzaniainvest.com/construction/world-bank-boosts-tanzania-water-sector-development-project-with-usd-

449-million Accessed 28-10-2016.

34

experiences of corruption. Our data allows us to be reasonably certain of our respondent and

project locations to a precision of no worse than 25 km, and often at a higher level of

precision.21 Accordingly, while we are confident that we can co-locate respondents and

projects at 40 km, we would not be confident in our geo-locating respondents and projects at

very small radii. However, we would also expect that as the radius increases, the effects of

development projects on local perceptions and experiences of corruption will become

increasingly diffuse until they become negligible. To evaluate the robustness of our results to

different radii we run Model 2 from Tables 1 and 3 across radii from 10 km to 150 km at 1

km intervals. Figures 4 and 5 below show the coefficients on China and the World Bank, with

their 95% confidence intervals, for corruptions perceptions and experiences, respectively.

Figure 4: Coefficients by Radius for Corruption Perceptions

21 Where Knutsen et al (2016) find that the method used to geo-code the Afrobarometer respondents is precise to

roughly 15km, while AidData projects at precision code “2” are precise to 25km.

35

Figure 5: Coefficients by Radius for Corruption Experiences

Figures 4 and 5 show that while the impact of local projects on corruption does indeed

depend on the distance to the project, this relationship behaves in an orderly way, essentially

showing the strongest relationships when projects are nearby, with the effect diminishing in

increasing distance. While the impacts are smaller at radii under 25km, this could well be

driven by the lack of geographical precision in our data as discussed above. For both the

World Bank and China, and for both perceptions and experience, the impacts are largest at a

radius of roughly 25km with relatively smooth decays as the distance increases. The

confidence intervals in Figure 5 suggest that the corruption experience results have the

highest levels of statistical significance when utilizing radii between roughly 20 and 40km.

36

Finally, it is worthwhile to emphasize that all of the results above account for the

confounding influence of the co-location of natural resources22, which were shown to strongly

correlate with local corruption in Knutsen et al. (2016). This suggests that the institutional

“resource curse” and “aid curse” (or “blessing”) may be separate phenomena. In other words,

development projects may induce or mitigate local corruption even in the absence of local

resources.

4 Limitations and Avenues for Further Research

There are two major limitations with our research approach that, if resolved, would be

important steps forward from this work. The first concerns the external validity of our results.

While focusing on Tanzania has allowed us to avoid heterogeneous country-level effects and

focus on micro-evaluations such as the fiber and water cases, it means that our results may

not be generalizable across the experience of developing countries. An important next step

would be to conduct a cross-national evaluation of development projects and local corruption.

A promising step in this direction is the recent working paper by Isaksson and Kotsadam

(2016) which considers questions similar to ours across almost 100,000 respondents in 29

African countries, although at the expense of more detailed analysis of different types of

projects and of co-location of projects from different donors.

The second major limitation with our study is that it is a cross-sectional instead of

longitudinal design. In the absence of being able to randomly assign the treatment – the

presence of development projects within the spatial proximity of the respondent – the best

observational research design would be to have pre- and post-treatment observations on the

22 Our results available in Online Appendix I also suggest a weakly significant (α=0.10) positive relationship

between resources and local experiences of corruption.

37

key dependent variables for all individual respondents, some proximate to the projects and

some not. While the Afrobarometer data consists of several waves over time, it is not panel

data, which makes it impossible to measure the dependent variable both pre- and post-

treatment for each individual survey respondent. Thus, we cannot exclude the possibility that

non-randomly distributed exogenous factors affect both the treatment assignment and the

outcome variable, or indeed, that there is a reverse effect. It is likely that project locations are

selected for any number of socio-political-economic reasons, as suggested by Dreher et al.

(2015). Thus, there is likely to be some endogeneity where Chinese and World Bank projects

locate to areas that already have higher or lower perceptions and experiences of corruption.

From a causal inference perspective, our research design is therefore limited, and the

possibility of confounding factors or endogeneity affecting our results cannot be denied, both

in terms of selection bias and in terms of differential treatment effect bias (Pearl 2000;

Morgan and Winship 2007).

Our findings are therefore based on strong theoretical expectations, from which observable

implications in a cross-sectional study can be derived, which are confirmed by our statistical

analysis. From an empirical perspective, a quasi-experimental design could be a useful

avenue forward for this research, one that engages in both pre- and post-treatment surveys of

corruption with the same respondents, where the carefully measured timing of additional

development projects are the treatment. (Quasi-)experimental research designs have been

gaining increased use in studies of development flows and could well be applicable here

(Dietrich et al. 2015; BenYishay et al. 2016).23 It should be noted that while Isaksson and

23 Where BenYishay et al. (2016) are able to incorporate temporal identification by combining “active years” of

a project with a panel of forestry satellite data, allowing for observation of changes in the same forest area over

a number of years.

38

Kotsadam (2016) address the first caveat, these empirical design challenges are also present

in their study.

A final interesting avenue for further research would be to compare the results of this study to

a similar study using geo-located DAC donor aid or foreign direct investment (FDI) flows, or

aid projects or FDI from other emerging donors in Tanzania or across a broader range of

developing countries.

5 Conclusions

Chinese development flows bring both opportunities and challenges to Tanzania, the broader

African continent, and the entire developing world, directly but also via interaction with other

development actors. Yet our results suggest that there are no blanket generalizations to be

made about the relationship between Chinese engagement and local corruption in Tanzania.

While there is evidence that proximity to Chinese development projects is associated with

increased local corruption experience, and to some extent perceptions, this finding appears to

be limited to non-ODA-like Chinese projects. Chinese OOF-like projects are much more

similar in nature to pure commercial flows like FDI, and as such, an “apples to apples”

comparison would need to consider how FDI relates to local corruption. Indeed, such

heterogeneity may even extend to the project-level, as we found substantially different effects

for an OOF-like fiber-optic project compared to an ODA-like water project. Moreover, we

provide evidence that co-location increases the association of local corruption experience

with both Chinese and World Bank projects. Our theory suggests that this may be due to

something akin to a “fog of aid”, wherein an increasing number of local development projects

make it difficult to monitor and sanction corruption. While there may be important reasons

39

for co-locating development projects, our finding may serve as a warning that there could be

negative externalities from this type of clustering.

Much of the literature on Chinese development finance, in particular, is founded on untested

assumptions, case studies and incomplete data (Strange et al. 2013). Our study has

contributed to an increasingly nuanced understanding of this large, complex and important

development actor. As China continues to increase its development presence it is crucial to

understand how its efforts and its policy of non-interference complement or hinder the

development and governance efforts of other development actors.

40

Bibliography:

Abdenur, A. E. (2014). China and the BRICS Development Bank: Legitimacy and

Multilateralism in South–South Cooperation. IDS Bulletin, 45(4), 85-101.

Altincekic, C., & Bearce, D. H. (2014). Why there Should be No Political Foreign Aid Curse.

World Development, 64, 18-32.

Alves, A. C. (2013). Chinese economic statecraft: A comparative study of China’s oil-backed

loans in Angola and Brazil. Journal of Current Chinese Affairs, 42(1), 99-130.

Annen, K., & Knack, S. (2015). On the delegation of aid implementation to multilateral

agencies. World Bank Policy Research Working Paper, (7455). Washington, DC.

Asongu, S. A., & Nwachukwu, J. (2014). Foreign Aid and Governance in Africa.

Attila, J. (2009). Individual attitudes toward anti-corruption policies in Sub-Saharan Africa:

Microeconometric evidence. Economics Bulletin, 29(3), 1927-1933.

Bailey, M. (1975). Tanzania and China. African Affairs, 74(294), 39-50.

Baldwin, K. & Winters, M. (2016). Can International Aid Change the Politics of Service

Delivery? Working Paper, August 2016.

Belousova, V., Goel, R. K., & Korhonen, I. (2016). Corruption perceptions versus corruption

incidence: Competition for rents across Russian regions. Journal of Economics and

Finance, 40(1), 172-187.

BenYishay, A., Parks, B., Runfola, R., Trichler, R. (2016) Forest Cover Impacts of Chinese

Development Projects in Ecologically Sensitive Areas. Paper presented at SAIS CARI

2016 Conference. October 13-14. Washington, DC.

Berthélemy, J. C. (2006). Bilateral donors’ interest vs. recipients’ development motives in aid

allocation: do all donors behave the same? Review of Development Economics, 10(2),

179-194.

Brautigam, D. (2009). The dragon's gift: the real story of China in Africa: OUP Oxford.

41

Bräutigam, D. (2011). Aid ‘With Chinese Characteristics’: Chinese Foreign Aid and

Development Finance Meet the OECD‐DAC Aid Regime. Journal of international

development, 23(5), 752-764.

Bräutigam, D. A., & Knack, S. (2004). Foreign Aid, Institutions, and Governance in Sub‐Saharan Africa*. Economic development and cultural change, 52(2), 255-285.

Brazys, S. R. (2013). Evidencing donor heterogeneity in Aid for Trade. Review of

International Political Economy, 20(4), 947-978.

— (2016). Aid and governance: Negative returns? The European Journal of Development

Research, 28(2), 294-313.

Breen, M., & Gillanders, R. (2015). Political Trust, Corruption, and Ratings of the IMF and

the World Bank. International Interactions, 41(2), 337-364.

Case, A. (1992). Neighborhood influence and technological change. Regional Science and

Urban Economics, 22(3), 491-508.

Case, A. C. (1991). Spatial patterns in household demand. Econometrica: Journal of the

Econometric Society, 953-965.

Charron, N. (2011). Exploring The Impact Of Foreign Aid On Corruption: Has The “Anti‐Corruption Movement” Been Effective? The Developing Economies, 49(1), 66-88.

Collier, P. (2007). The bottom billion. ECONOMIC REVIEW-DEDDINGTON-, 25(1), 17.

Copelovitch, M. S., & Ohls, D. (2012). Trade, institutions, and the timing of GATT/WTO

accession in post-colonial states. The Review of International Organizations, 7(1), 81-

107.

De Haan, A. (2011). Will China change international development as we know it? Journal of

international development, 23(7), 881-908.

42

Dettman, S. & Pepinsky, T. (2016). Resource Shocks and Public and Local Public Goods: A

Tale of Two Districts. Working paper. Accessed at http://cpd.berkeley.edu/wp-

content/uploads/2016/02/paper-for-berkeley-presentation.pdf 29-10-2016.

Dietrich, S., Mahmud, M., & Winters, M. S. (2015). Foreign Aid, Foreign Policy, and

Domestic Government Legitimacy: Experimental Evidence from Bangladesh.

AidData Working Paper No. 16. Williamsburg, VA.

Djankov, S., Montalvo, J. G., & Reynal-Querol, M. (2008). The curse of aid. Journal of

Economic Growth, 13(3), 169-194.

Donchev, D., & Ujhelyi, G. (2014). What do corruption indices measure? Economics &

Politics, 26(2), 309-331.

Dreher, A., Fuchs, A., Hodler, R., Parks, B., Raschky, P. A., & Tierney, M. J. (2016). Aid on

demand: African leaders and the geography of China's foreign assistance. AidData

Working Paper No. 3 Revised. Williamsburg, VA.

Dreher, A., Fuchs, A., Parks, B., Strange, A. M., & Tierney, M. J. (2015). Apples and Dragon

Fruits: The Determinants of Aid and Other Forms of State Financing from China to

Africa. AidData Working Paper No. 15. Williamsburg, VA.

Findley, M. G., Harris, A. S., Milner, H. V., & Nielson, D. (2015). ELITE AND MASS

SUPPORT FOR FOREIGN AID VERSUS GOVERNMENT PROGRAMS:

EXPERIMENTAL EVIDENCE FROM UGANDA.

Furukawa, M. (2014). Management of the International Development Aid System Aid

System and the Creation of Political Space for China: The Case of Tanzania.

Gutmann, J., Padovano, F., & Voigt, S. (2015). Perception vs. experience: explaining

differences in corruption measures using microdata. Experience: Explaining

Differences in Corruption Measures Using Microdata (September 11, 2015).

Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2659349

43

Gyimah-Brempong, K. (2002). Corruption, economic growth, and income inequality in

Africa. Economics of Governance, 3(3), 183-209.

Hernandez, D. (2015). Are “New” Donors Challenging World Bank Conditionality?.

University of Heidelberg Department of Economics Discussion Paper Series No. 601.

Heidelberg, Germany.

Hinga, J., & Yiguan, Q. (2013). China-Africa Cooperation–An Outstanding Relationship

Built on Mutual Respect and Common Benefits: A Review. International Research

Journal of Social Sciences, 2(9), 26-32.

Holtom, D. (2005). Reconsidering the power of the IFIs: Tanzania & the world bank, 1978-

1985. Review of African Political Economy, 32(106), 549-568.

Hout, W. (2007). The politics of aid selectivity: Good governance criteria in World Bank, US

and Dutch development assistance: Routledge.

Isaksson, A. S. (2015). Corruption along ethnic lines: A study of individual corruption

experiences in 17 African countries. The Journal of Development Studies, 51(1), 80-

92.

Isaksson, A. S., & Kotsadam, A. (2016). Chinese aid and local corruption. University of

Gothenburg Working Papers in Economics No. 667. June 2016. Goteborg, Sweden.

Kilby, C., & Dreher, A. (2010). The impact of aid on growth revisited: Do donor motives

matter? Economics Letters, 107(3), 338-340.

Kim, S., & Lightfoot, S. (2011). Does ‘DAC‐Ability’Really Matter? The emergence of non‐DAC Donors: Introduction to Policy Arena. Journal of international development,

23(5), 711-721.

Knutsen, C. H., Kotsadam, A., Olsen, E. H., & Wig, T. (2016). Mining and local corruption

in Africa. American Journal of Political Science. Early View 13 September 2016.

DOI: 10.1111/ajps.12268

44

Leeuw, F. L., Van Gils, G. H., & Kreft, C. (1999). Evaluating Anti-Corruption Initiatives

Underlying Logic and Mid-Term Impact of a World Bank Program. Evaluation, 5(2),

194-219.

LeSage, J. P. (2000). Bayesian estimation of limited dependent variable spatial autoregressive

models. Geographical Analysis, 32(1), 19-35.

Lopez-Valcarcel, B. G., Jiménez, J. L., & Perdiguero, J. (2014). Danger: Local Corruption Is

Contagious!. Unpublished manuscript.

http://www.alde.es/encuentros/anteriores/xviieea/trabajos/g/pdf/179. pdf.

Manda, C., Knowles, J., Connors, J., & Mwombela, S. (2014). Borders: Social Interaction

and Economic and Political Integration of the East African Community.

Mawdsley, E., Savage, L., & Kim, S. M. (2014). A ‘post‐aid world'? Paradigm shift in

foreign aid and development cooperation at the 2011 Busan High Level Forum. The

Geographical Journal, 180(1), 27-38.

McMillen, D. P. (1992). Probit with spatial autocorrelation. Journal of Regional Science,

32(3), 335-348.

Milner, H. V., Nielson, D. L., & Findley, M. G. (2016). Citizen preferences and public goods:

Comparing preferences for foreign aid and government programs in Uganda. The

Review of International Organizations, 11(2), 219-245.

Molenaers, N., Dellepiane, S., & Faust, J. (2015). Political conditionality and foreign

aid. World Development, 75, 2-12.

Morgan, S. L. & Winship, L. (2007). Counterfactuals and causal inference. Methods and

principles for social research. New York: Cambridge University Press.

Moshi, H., Mtui, J., & es Salaam, D. (2008). Scoping studies on China-Africa economic

relations: The case of Tanzania. Nairobi: AERC (mimeo).

45

Muchapondwa, E., Nielson, D., Parks, B., Strange, A. M., & Tierney, M. J. (2014). Ground-

truthing'Chinese development finance in Africa: Field evidence from South Africa

and Uganda. WIDER Working Paper.

Okada, K., & Samreth, S. (2012). The effect of foreign aid on corruption: A quantile

regression approach. Economics Letters, 115(2), 240-243.

Olken, B. A. (2009). Corruption perceptions vs. corruption reality. Journal of Public

economics, 93(7), 950-964.

Olken, B. A., & Pande, R. (2011). Corruption in developing countries (No. w17398).

National Bureau of Economic Research.

Payer, C. (1983). Tanzania and the World Bank. Third world quarterly, 5(4), 791-813.

Pearl, J. (2000). Causality: Models, reasoning, and inference. Cambridge: Cambridge

University Press.

Pehnelt, G. (2007). The political economy of China's aid policy in Africa. Jena Economic

Research Paper(2007-051).

Richmond, S., & Alpin, C. (2013). Governments falter in fight to curb corruption: the people

give most a failing grade. Afrobarometer Policy paper(4).

Schudel, C. J. W. (2008). Corruption and bilateral aid: a dyadic approach. Journal of Conflict

Resolution.

Sigalla, H. L. (2014). Changing trends in the Tanzania–China relationship. Österreichische

Zeitschrift für Soziologie, 39(1), 61-78.

Smets, L., & Knack, S. (2016). World Bank Lending and the Quality of Economic

Policy. The Journal of Development Studies, 52(1), 72-91.

Steinwand, M. C. (2015). Compete or coordinate? Aid fragmentation and lead

donorship. International Organization, 69(2), 443.

46

Strange, A. M., Dreher, A., Fuchs, A., Parks, B., & Tierney, M. J. (2015). Tracking

Underreported Financial Flows China’s Development Finance and the Aid–Conflict

Nexus Revisited. Journal of Conflict Resolution, 0022002715604363.

Strange, A. M., Park, B., Tierney, M. J., Fuchs, A., Dreher, A., & Ramachandran, V. (2013).

China's development finance to Africa: A media-based approach to data collection.

Center for Global Development working paper(323).

Sun, Y. (2014). Africa in China’s foreign policy. The Brookings Institution.

Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit region.

Economic geography, 46, 234-240.

Tull, D. M. (2006). China's engagement in Africa: Scope, significance and consequences.

The Journal of Modern African Studies, 44(03), 459-479.

Ugur, M. (2014). Corruption's Direct Effects on Per‐Capita Income Growth: A Meta‐Analysis. Journal of Economic Surveys, 28(3), 472-490.

Wang, F.-L., & Elliot, E. A. (2014). China in Africa: presence, perceptions and prospects.

Journal of Contemporary China, 23(90), 1012-1032.

Wang, X., & Ozanne, A. Two approaches to aid in Africa: China and the West. In Prepared

for the International Conference “Ten Years Of ‘War Against Poverty’: What Have

We Learned Since, 2000 (pp. 2010-2020)

Winters, M. S. (2010). Choosing to target: what types of countries get different types of

World Bank projects. World Politics, 62(03), 422-458.

— (2014). Targeting, accountability and capture in development projects. International

Studies Quarterly, 58(2), 393-404.

World Bank (1998) Support to the Government of Tanzania's Anti-Corruption Program.

World Bank Poverty Reduction and Social Development Unit Africa Region. October

1998. Available at:

47

http://www1.worldbank.org/publicsector/anticorrupt/TanzaniaGCA.pdf accessed 29-

10-2016.

Zhao, S. (2014). A neo-colonialist predator or development partner? China's engagement and

rebalance in Africa. Journal of Contemporary China, 23(90), 1033-1052.

Zhu, B. (2016). MNCs, Rents, and Corruption: Evidence from China. American Journal of

Political Science. Early Views. DOI: 10.1111/ajps.12259

Zimmermann, F., & Smith, K. (2011). More actors, more money, more ideas for international

development co‐operation. Journal of international development, 23(5), 722-738.

48

APPENDIX I

A1 Data Sources and Descriptive Statistics

Data on our main independent variables came from the AidData project, with the base data for the Chinese Counts from china.aiddata.org version 1.2 available at http://china.aiddata.org/datasets/1.2 Williamsburg, VA and Washington, DC: AidData. Accessed on [18-04-2016]. http://aiddata.org/research-datasets. and the base data for World Bank counts from AidData. 2016. WorldBank_GeocodedResearchRelease_Level1_v1.4.1 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [18-04-2016]. http://aiddata.org/research-datasets.

Our outcome variables, and many of our control variables including age, education, female, urban and occupation come from the AfroBarometer 6 Survey. Most of our variables came from the publicly available version of this data available at http://www.afrobarometer.org/data/tanzania-round-6-data-2015 Accessed on 15-09-2016. However, in order to get access to the respondent Ward data we needed to file an application to AfroBarometer for restricted data. This request was granted on 10-10-2016 and we used the Ward information to geo-locate the respondents in order to formulate the count variables of Chinese and World Bank Projects.

As mentioned in the text, we also use survey data based on research carried out by Repoa (http://www.repoa.or.tz/) which initially focused on citizens’ views of local government reform in Tanzania. The Local Government Reform Programme (LGRP) began in 1997 as a process by which to devolve powers from the central government to the local level, and improve citizen participation, service delivery and governance (Fjeldstad et al, 200524). REPOA (Research for Poverty Alleviation), an independent research-based NGO, followed the process of reform over a ten-year period with the Citizen’s Survey (Fjeldstad et al, 2005). This study uses a number of data points from REPOA’s Citizen’s Survey, which was carried out across six case councils in and 2013.

Data on the location of natural resources utilizes the United States Geological Survey Mineral Resources Online Spatial Data at http://mrdata.usgs.gov/minfac/select.php?place=fTZ&div=fips accessed 22-11-2015.

Data on the party of the local parliamentarian as of the 2015 election was sourced from https://web.archive.org/web/20151126095523/http://www.nec.go.tz/uploads/documents/

24 Fjeldstad et al (2005), Local governance, finances and service delivery in Tanzania, a summary of findings from six councils, CHR. Michelsen Institute.

49

1448025692-MALIMA-WABUNGE.pdf accessed 10-10-2016 and matched to the wards in the Afrobarometer 6 data.

Regressions in the main text were run in STATA 13. Regressions in Appendix I were run with STATA 13 and Matlab R2014a. Do files available upon request.

Summary statistics are presented available in table A1.1 below.

50

Table A1.1: Summary Statistics

Mean Variance Min Max NCorruption Perception 0.79 0.41 0 1 2304Corruption Experience 0.23 0.42 0 1 1956

Corruption Perception Bureaucrat 1.19 0.64 0 3 2178Corruption Perception Judge 1.31 0.73 0 3 2195Corruption Perception Local 1.17 0.67 0 3 2202

Corruption Perception Parliament 1.13 0.67 0 3 2110Corruption Perception Police 1.55 0.75 0 3 2267

Corruption Perception President 0.93 0.71 0 3 2051Corruption Perception Tax 1.36 0.74 0 3 2102

Corruption Perception Traditional 0.71 0.80 0 3 1832Corruption Experience Utility 0.32 0.72 0 3 359Corruption Experience Permit 0.20 0.54 0 3 471Corruption Experience Police 0.56 0.93 0 3 404Corruption Experience Court 0.52 0.85 0 3 272

Corruption Experience Hospital 0.31 0.73 0 3 1715Corruption Experience School 0.16 0.51 0 3 1165

China 3.78 6.60 0 22 2386China Infrastructure 1.93 3.20 0 12 2386

China ODA Infrastructure 0.50 1.20 0 12 2386China OOF Infrastructure 1.42 2.20 0 8 2386China Fiber Infrastructure 0.96 1.30 0 4 2386

China Water Infrastructure 0.13 0.75 0 12 2386World Bank 6.43 8.44 0 24 2386

World Bank Infrastructure 3.40 4.99 0 16 2386Age 38.39 14.33 18 93 2373

govtParty 0.74 0.44 0 1 2340Female 0.50 0.50 0 1 2386Urban 0.83 0.37 0 1 2386

Resources 0.28 0.45 0 1 2386Education

No Formal Schooling 254Primary 1192

Secondary 524Some Schooling 262

Tertiary 154Occupation

Private 146Public 131

Self-Employed 1811Unemployed 284

Income (Gone without cash Income)0 (Never)

1 (Just Once or Twice)2 (Several Times)3 (Many Times)

4 (Always)

45342560983067

51

52

A2 Full Regression Results

In the tables below we present the full regression results for the tables in the main text.

53

Table A2.1 “Table 1: Perceptions of Corruption (Binary)”Model 0 Model 1 Model 2 Model 3

bribe 0.134 0.128 0.131 0.129(1.40) (1.34) (1.37) (1.35)

age -0.004 -0.004 -0.004 -0.004(1.43) (1.33) (1.40) (1.38)

No schooling 0.000 0.000 0.000 0.000Primary 0.022 0.009 0.017 0.014

(0.15) (0.06) (0.11) (0.10)

Secondary -0.010 0.009 -0.001 0.001(0.06) (0.06) (0.00) (0.01)

Some schooling 0.107 0.093 0.104 0.101(0.60) (0.52) (0.58) (0.56)

Tertiary -0.070 -0.070 -0.072 -0.075(0.31) (0.31) (0.32) (0.34)

Private 0.000 0.000 0.000 0.000Public 0.193 0.197 0.203 0.204

(0.77) (0.79) (0.81) (0.81)

Self-Employed -0.394 -0.394 -0.388 -0.389(2.28)* (2.28)* (2.24)* (2.25)*

Unemployed -0.174 -0.173 -0.167 -0.168(0.84) (0.84) (0.81) (0.81)

govtParty 0.030 0.019 0.022 0.027(0.26) (0.16) (0.19) (0.23)

female -0.068 -0.066 -0.067 -0.067(0.90) (0.88) (0.90) (0.89)

urban 0.868 0.855 0.863 0.862(8.63)** (8.47)** (8.56)** (8.54)**

Resources -0.013 -0.040 -0.053 -0.067(0.11) (0.29) (0.38) (0.47)

0b.income 0.000 0.000 0.000 0.0001.income 0.339 0.338 0.340 0.342

(2.69)** (2.67)** (2.69)** (2.70)**

2.income 0.076 0.076 0.078 0.078(0.65) (0.65) (0.67) (0.67)

3.income 0.445 0.443 0.449 0.449(3.73)** (3.69)** (3.74)** (3.74)**

4.income 0.104 0.097 0.106 0.104(0.41) (0.39) (0.42) (0.42)

China 0.019(1.16)

World Bank -0.017(1.40)

China Infrastructure 0.028 0.011(0.86) (0.21)

World Bank Infrastructure -0.016 -0.021(0.80) (0.90)

China*World Bank 0.002(0.41)

_cons 0.241 0.240 0.240 0.241(4.19)** (4.19)** (4.18)** (4.19)**

1,836 1,836 1,836 1,836* p<0.05; ** p<0.01

54

Table A2.2 “Table 3: Experiences of Corruption (Binary)”Model 0 Model 1 Model 2 Model 3 Model 4

Age -0.006 -0.005 -0.005 -0.005 -0.005(2.00)* (1.79) (1.92) (1.93) (1.82)

No schooling 0.000 0.000 0.000 0.000 0.000Primary 0.102 0.085 0.095 0.102 0.086

(0.77) (0.65) (0.72) (0.77) (0.65)

Secondary 0.085 0.096 0.093 0.111 0.094(0.56) (0.63) (0.61) (0.72) (0.62)

Some schooling 0.242 0.222 0.238 0.242 0.223(1.52) (1.39) (1.49) (1.52) (1.40)

Tertiary 0.050 0.034 0.034 0.044 0.020(0.24) (0.16) (0.16) (0.21) (0.09)

Private 0.000 0.000 0.000 0.000 0.000Public 0.155 0.180 0.190 0.207 0.194

(0.77) (0.89) (0.94) (1.02) (0.96)

Self-Employed -0.329 -0.321 -0.308 -0.294 -0.309(2.39)* (2.33)* (2.23)* (2.13)* (2.24)*

Unemployed -0.328 -0.318 -0.306 -0.293 -0.305(1.90) (1.84) (1.78) (1.70) (1.77)

govtParty 0.117 0.088 0.092 0.092 0.112(1.09) (0.83) (0.86) (0.86) (1.05)

female -0.184 -0.181 -0.183 -0.183 -0.181(2.67)** (2.63)** (2.66)** (2.66)** (2.63)**

urban -0.247 -0.260 -0.253 -0.251 -0.257(2.47)* (2.58)** (2.52)* (2.50)* (2.56)*

Resources 0.343 0.232 0.214 0.154 0.158(3.39)** (1.86) (1.71) (1.19) (1.24)

0b.income 0.000 0.000 0.000 0.000 0.0001.income 0.097 0.095 0.099 0.102 0.103

(0.80) (0.79) (0.82) (0.84) (0.85)

2.income 0.187 0.189 0.192 0.193 0.191(1.66) (1.67) (1.70) (1.71) (1.69)

3.income 0.350 0.352 0.361 0.360 0.359(3.19)** (3.19)** (3.28)** (3.26)** (3.26)**

4.income 0.510 0.500 0.512 0.510 0.503(2.26)* (2.21)* (2.26)* (2.26)* (2.23)*

China 0.038(2.51)*

World Bank -0.027(2.32)*

China Infrastructure 0.067 -0.008(2.23)* (0.16)

World Bank Infrastructure -0.031 -0.034 -0.052(1.68) (1.78) (2.41)*

China ODA Infrastructure -0.037(0.50)

China OOF Infrastructure 0.131(2.79)**

China*World Bank 0.007(1.89)

_cons 0.199 0.187 0.190 0.188 0.182(3.92)** (3.81)** (3.85)** (3.85)** (3.77)**

1,902 1,902 1,902 1,902 1,902* p<0.05; ** p<0.01

55

Table A2.2 “Table 2: Perceptions of Corruption (Ordered)”Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 7.1 Model 8 Model 8.1

Local President Parliament Bureaucrat Judge Police Tax Tax Traditional TraditionalChina Infrastructure 0.023 -0.002 -0.001 0.003 -0.001 -0.006 0.051 0.050

(1.08) (0.09) (0.05) (0.12) (0.03) (0.29) (2.29)* (1.92)

World Bank Infrastructure 0.003 -0.004 0.006 0.009 -0.003 0.008 -0.027 -0.031 -0.041 -0.048(0.24) (0.24) (0.48) (0.70) (0.25) (0.69) (2.05)* (2.33)* (2.60)** (3.03)**

bribe 0.213 0.254 0.089 0.191 0.117 0.277 0.186 0.180 0.172 0.165(3.20)** (3.55)** (1.32) (2.85)** (1.80) (4.33)** (2.80)** (2.72)** (2.28)* (2.20)*

age -0.003 -0.001 -0.002 -0.003 -0.003 -0.003 -0.004 -0.004 -0.000 -0.000(1.37) (0.44) (0.88) (1.38) (1.18) (1.57) (1.72) (1.74) (0.15) (0.19)

No schooling 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000Primary -0.113 0.060 -0.099 -0.114 0.017 0.199 -0.158 -0.156 -0.069 -0.065

(1.09) (0.52) (0.93) (1.07) (0.17) (2.00)* (1.45) (1.44) (0.60) (0.56)

Secondary -0.047 0.129 0.076 -0.026 0.046 0.301 -0.203 -0.194 -0.210 -0.191(0.40) (1.00) (0.64) (0.22) (0.40) (2.67)** (1.67) (1.60) (1.58) (1.44)

Some schooling -0.164 -0.120 -0.141 -0.160 0.043 0.243 -0.150 -0.151 -0.176 -0.180(1.28) (0.84) (1.05) (1.23) (0.34) (1.98)* (1.13) (1.14) (1.19) (1.23)

Tertiary -0.039 0.135 -0.017 -0.156 0.112 0.431 -0.186 -0.184 -0.211 -0.207(0.24) (0.76) (0.11) (0.96) (0.71) (2.79)** (1.13) (1.12) (1.14) (1.13)

Private 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000Public -0.191 -0.000 -0.068 -0.007 -0.090 -0.188 0.022 0.038 -0.375 -0.343

(1.15) (0.00) (0.40) (0.04) (0.55) (1.17) (0.13) (0.23) (1.95) (1.79)

Self-Employed -0.179 -0.035 -0.228 -0.199 -0.233 -0.277 -0.033 -0.021 -0.406 -0.378(1.59) (0.28) (1.97)* (1.75) (2.09)* (2.50)* (0.29) (0.18) (3.08)** (2.87)**

Unemployed -0.073 0.141 -0.050 -0.143 0.005 -0.361 0.008 0.018 -0.074 -0.043(0.52) (0.94) (0.36) (1.01) (0.03) (2.66)** (0.06) (0.13) (0.46) (0.27)

govtParty -0.183 -0.265 -0.183 -0.156 -0.103 -0.057 -0.136 -0.136 -0.062 -0.061(2.39)* (2.74)** (2.40)* (2.05)* (1.38) (0.78) (1.74) (1.75) (0.67) (0.68)

female -0.100 -0.108 -0.120 -0.136 -0.140 -0.075 -0.191 -0.191 0.026 0.026(1.80) (1.82) (2.11)* (2.41)* (2.58)** (1.42) (3.44)** (3.44)** (0.42) (0.42)

urban -0.194 -0.392 -0.237 -0.187 -0.076 0.167 -0.047 -0.049 -0.374 -0.376(2.40)* (4.62)** (2.91)** (2.28)* (0.97) (2.14)* (0.58) (0.61) (4.38)** (4.41)**

Resources 0.017 0.097 -0.021 0.030 0.149 -0.019 0.031 -0.013 0.237 0.150(0.19) (0.84) (0.23) (0.33) (1.71) (0.22) (0.33) (0.14) (2.21)* (1.38)

0b.income 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.0001.income -0.050 0.095 0.163 0.047 -0.070 0.014 0.029 0.033 0.025 0.032

(0.54) (0.96) (1.73) (0.50) (0.77) (0.15) (0.32) (0.36) (0.25) (0.31)

2.income -0.017 0.065 0.131 0.023 -0.016 0.001 -0.058 -0.056 0.025 0.029(0.19) (0.68) (1.46) (0.26) (0.19) (0.01) (0.67) (0.64) (0.25) (0.29)

3.income 0.041 0.038 0.132 0.146 -0.023 0.279 -0.137 -0.136 -0.207 -0.205(0.47) (0.40) (1.48) (1.64) (0.26) (3.34)** (1.57) (1.56) (2.10)* (2.09)*

4.income 0.214 -0.256 0.038 0.106 0.120 0.037 -0.205 -0.204 -0.334 -0.332(1.13) (1.20) (0.19) (0.53) (0.66) (0.20) (1.03) (1.03) (1.60) (1.59)

China ODA Infrastructure -0.008 -0.070(0.20) (1.35)

China OOF Infrastructure 0.093 0.133(2.78)** (3.47)**

1,749 1,612 1,664 1,727 1,741 1,805 1,663 1,663 1,493 1,493* p<0.05; ** p<0.01

56

Table A2.4 “Table 4: Experiences of Corruption (Ordered)”Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Utility Utility Permit Police Court Hospital SchoolChina Infrastructure 0.168 0.042 0.204 0.155 0.059 0.083

(2.54)* (0.63) (3.12)** (2.01)* (1.71) (2.07)*

World Bank Infrastructure -0.067 -0.033 -0.004 -0.136 -0.108 -0.019 -0.007(1.63) (0.91) (0.11) (3.23)** (2.13)* (0.91) (0.26)

age 0.013 0.012 0.014 -0.006 -0.001 -0.004 0.004(1.73) (1.60) (1.95) (0.94) (0.15) (1.42) (0.74)

No schooling 0.000 0.000 0.000 0.000 0.000 0.000 0.000Primary 0.391 0.403 0.166 0.383 0.504 0.114 0.004

(1.23) (1.25) (0.51) (1.46) (1.58) (0.79) (0.02)

Secondary 0.314 0.369 -0.133 0.303 0.312 0.002 -0.246(0.84) (0.97) (0.36) (1.02) (0.86) (0.01) (1.00)

Some schooling 0.130 0.127 0.290 0.294 0.173 0.307 -0.105(0.26) (0.25) (0.67) (0.86) (0.45) (1.78) (0.38)

Tertiary 0.173 0.217 -0.325 -0.033 0.331 -0.046 -0.584(0.34) (0.42) (0.69) (0.08) (0.64) (0.20) (1.65)

Private 0.000 0.000 0.000 0.000 0.000 0.000 0.000Public -0.205 -0.201 0.586 0.175 0.729 0.254 0.737

(0.44) (0.42) (1.45) (0.46) (1.41) (1.13) (2.35)*

Self-Employed -0.271 -0.254 -0.424 -0.237 0.390 -0.255 -0.296(0.90) (0.84) (1.39) (0.87) (1.15) (1.69) (1.40)

Unemployed 0.653 0.647 0.091 -0.096 0.633 -0.121 0.536(1.82) (1.80) (0.26) (0.31) (1.70) (0.64) (2.14)*

govtParty 0.407 0.431 0.394 -0.405 -0.200 0.130 0.203(1.58) (1.64) (1.89) (2.54)* (1.01) (1.06) (1.26)

female -0.156 -0.156 -0.504 -0.221 0.086 -0.162 -0.153(0.86) (0.85) (2.73)** (1.61) (0.51) (2.14)* (1.35)

urban -0.444 -0.477 -0.690 -0.663 -0.514 -0.380 -0.372(1.88) (2.00)* (3.27)** (3.54)** (2.41)* (3.59)** (2.61)**

Resources 0.244 0.179 0.623 0.231 0.481 0.200 0.195(0.81) (0.56) (2.41)* (1.30) (2.29)* (1.41) (1.08)

0b.income 0.000 0.000 0.000 0.000 0.000 0.000 0.0001.income -0.265 -0.271 -0.342 -0.112 0.139 0.037 -0.176

(0.83) (0.83) (1.26) (0.47) (0.48) (0.28) (0.89)

2.income 0.676 0.647 0.061 0.134 -0.175 0.022 0.029(2.50)* (2.36)* (0.26) (0.58) (0.64) (0.18) (0.16)

3.income 0.553 0.525 0.280 0.171 0.429 0.454 0.373(1.95) (1.82) (1.09) (0.77) (1.63) (3.78)** (2.15)*

4.income 0.847 0.846 0.725 0.407 0.250 0.475 -0.168(1.68) (1.66) (1.55) (1.03) (0.52) (1.95) (0.45)

China Fiber Infrastructure 0.332(2.37)*

China Water Infrastructure -0.133(0.47)

N 353 353 464 396 265 1,668 1,137* p<0.05; ** p<0.01

57

Figure A1 – Two-way interactions (China*World Bank) for Table 2 “Perceptions of Corruption Ordered”

58

Figure A2 – Two-way interactions (China*World Bank) for Table 3 “Perceptions of Corruption Experience (Ordered)”

59

A3 REPOA Regression Results and Discussion

Table A.3: Perceptions of Corruption (Binary) REPOA

Model 1 Model 2 Model 3China 0.0384**

(3.33)World Bank -0.0128

(0.69)China Infrastructure 0.0314*

(2.11)0.0111(0.70)

World Bank Infrastructure 0.0182(0.60)

-0.0837†(1.72)

China*World Bank 0.0092*(2.52)

Age 0.0050(1.54)

0.0052(1.60)

0.0056†(1.74)

Resources -0.2348(0.68)

-0.1817(0.48)

-0.7902†(1.86)

Education 0.0236(0.52)

0.0288(0.63)

0.0305(0.66)

Occupation Factor Variables Yes Yes YesWard Mixed Effects Yes Yes Yes

N 1233 1233 1233Prob > χ2 0.0028 0.0290 0.0003

Absolute vale of Z score in parentheses. † p>0.10, *5% p>0.05, **1% p>0.01

60

Table A.3 and Figure A.3 suggest support for hypotheses 1, 2 and 4 for corruption perceptions.

Chinese projects are associated with higher levels of corruption perceptions for all projects (Model

1) and restricting to infrastructure projects (Model 2), although this result suggests a rejection of

Hypothesis 3 as the coefficient in Model 1 is both larger and of greater statistical significance than

the coefficient when just considering infrastructure projects in Model 2. Hypothesis 2 has weak

support in Model 3, suggesting a negative relationship at the 10% level of significance between

World Bank projects and perceptions of corruption when there are no co-located Chinese projects.

However, Model 3 suggests support for Hypothesis 4 as suggested both by the positive and

significant interaction term, but also by Figure A.3 which shows that at high levels of Chinese

projects the impact of World Bank goes from a relationship with perceptions of corruption that is

negative and weakly significant to one that is positive and significant.

61