The Causes and Consequences of Political Interference in

Bureaucratic Decision Making: Evidence from Nigeria

Daniel Rogger∗

August 2013

JOB MARKET PAPER

Abstract

Both politicians and bureaucrats are claimed as crucial to development and growth. Thispaper investigates key margins of interaction between these groups. Using data from theFederal Government of Nigeria, I study how political incentives influence which organiza-tions politicians delegate to. Second, I collect surveys of a representative sample of civilservants from these organizations, and use them to investigate whether politicians followup this delegation by directly engaging with bureaucrats. I find that politicians facing polit-ical competition delegate to more effective organizations, and engage with bureaucrats moreintensively. Finally, I use evaluation data on the implementation of 4493 public projects im-plemented by these organizations to assess how delegation and engagement affects projectimplementation. I find that delegation to agencies isolated from political interference im-proves project completion rates by roughly a third. The impacts of political engagementvary by the extent to which politicians face political competition. When politicians facelow levels of competition, their engagement with bureaucrats retards project implementa-tion significantly. When they face high political competition, the excesses of their negativebehavior are constrained analogously to how market competition constrains the excesses offirm activity. The findings provide micro-level evidence of the changes in politician behaviorthat are induced by political competition, and the subsequent impacts on project implemen-tation.

Keywords: politicians, bureaucrats, public goods, decentralization

JEL Classification: D72, D73, H00, H11, H41, O20

∗Department of Economics, University College London and Centre for the Evaluation of Development Policy, Institute forFiscal Studies. E-mail: d.rogger@ucl.ac.uk. Web: www.danrogger.com. I gratefully acknowledge financial support from theEconomic and Social Research Council [ES/G017352/1], the Federal Government of Nigeria, the Institute for Fiscal Studies,the International Growth Centre [RA-2009-11-018], the Royal Economic Society and University College London. I thank thePresidency and the Office of the Head of the Civil Service of Nigeria for their support throughout data collection. I am gratefulto the many government officials who have assisted me during this project but are too numerous to name individually, LauraAbramovsky, Beatriz Armendariz, Britta Augsburg, Wendy Carlin, Pedro Carneiro, Christian Dustmann, Silvia Espinosa, BenFaber, Emla Fitzsimons, Jonathan Phillips and Uta Schoenberg for valuable comments. Particular gratitude is reserved for myacademic supervisors, Imran Rasul and Orazio Attanasio, for their support and encouragement throughout this project. All errorsremain my own.

1

1 Introduction

Local public goods underpin development and growth. They are often left uncompleted or delivered to

a poor quality (World Bank, 2004). Failure to deliver these goods undermines citizen welfare and leads

to the loss of up to US$40 billion per year in public resources (World Bank, 2010).

Politicians and bureaucrats are claimed as crucial to the effective delivery of public goods. High levels

of competition may increase the pressure on politicians to see that public goods are delivered effectively.

Consequently, they may seek to overcome barriers such as bureaucrats’ inefficiency, laziness or corrup-

tion. However, political interference may also undermine the bureaucracy’s ability to deliver public

goods by misallocating public funds. Our understanding of the interactions between politicans and

bureaucrats is very limited, both in terms of their causes and their consequences. How does political

competition impact on politician’s interference in the bureaucracy and what are its consequences?

This paper lays out a framework for thinking about the causes and consequences of interaction between

these two groups and explores its empirical implications in a Nigerian context. We explore how politi-

cians distribute resources across government, by delegating projects to different tiers of government

that best serve their political interests. We assess whether they then follow up this delegation with

differential levels of direct engagement with bureaucrats. 1

To investigate these issues, I have assembled a data set that follows public projects from their initiation

in Congress through the organizations that implement them to evaluations of the implementation of

the projects themselves. Our data contains details of all members of Nigeria’s 5th National Assembly

including those that make the key delegation decisions, surveys of a representative sample of bureau-

crats at each of the organizations they delegate to, and evaluations of how effectively each of 4493 public

projects were implemented at these organizations.

This allows us to test how political competition impacts on delegation and engagement by looking at

how members of congressional standing committees delegate relative to non-members. The analysis is

based on the fact that membership of the committees seems exogenous with respect to constituency-

level characteristics that might effect public good provision but endows a member with additional

power to delegate and engage. I find that in constituencies in which the politician faces significant

competition for her seat, politicians delegate to the most effective organizations in government, the

decentralized agencies. When a politician faces little political competition in her constituency, she cen-

tralizes to ministries. I also find that politicians who face political competition engage more intensively

1To do so, we fix the policy environment, staff and resource distribution, implying that politicians lack the power to recruit,dismiss, demote or change the formal wages of appointed bureaucrats, or modify the legislation governing how their organi-zation functions. We will see that these are reasonable assumptions in our context. Public sector recruitment is delegated to anindependent organization that fiercely guards its independence. The modification of legislation governing an organization wouldrequire the agreement of a majority of politicians who are likely to have distinct political preferences across the constituenciesserved by an organization.

2

with the bureaucracy.

These findings imply that the distribution of resources across government is a function of political in-

centives. It also indicates that the working environment of bureaucrats differs across the constituencies

in which they work depending on the level of political competition there.

We look at how delegation and engagement impact on project implementation and find that delegation

to decentralized organizations isolates projects from political vulnerability. Decentralized organizations

are effective in both politically competitive and uncompetitive constituencies.

The impacts of political engagement vary by the level of political competition of the constituency in

which a project is implemented. In politically uncompetitive constituencies, a politician’s engagement

of the bureaucracy has large and significant negative consequences for project implementation. How-

ever, political competition constrains the excesses of politician behavior analogously to how market

competition constrains the excesses of firm activity. In politically competitive constituencies, the nega-

tive impacts of politician engagement is significantly reduced.

One interpretation of these patterns is that politicians delegate to those organizations they know to be

most effective when public good provision has the highest electoral returns. Decentralized agencies

have significantly higher completion rates than centralized ministries. They then engage organizations

when they require public goods to be provided or when they feel they can misuse the funds without

cost to their political careers. The corresponding impacts on project implementation ensue.

These results provide some of the first evidence of how political competition impacts on the choices of

politicians in their interaction with the bureaucracy. They allow us to explore how political competition

impacts on different types of organization, thereby offering an assessment of the merits of decentral-

ization. And they provide a platform for understanding whether political competition itself drives

superior government performance, or whether it requires the response of politicians.

An emerging literature links political competition to improved bureaucratic performance.2 For exam-

ple, Ferraz and Finan (2011) find that re-election incentives push Brazilian mayors to reduce misappro-

priation of resources in their municipality by 27% compared to those without such incentives. Foster

and Rosenzweig (2004) use panel data on governance and local public goods in India to show that

politicians facing greater political pressures seem to make government policy more pro-poor in their

constituencies.3 However, we provide evidence of the mechanisms through which political competi-2Existing work on delegation in economics is substantial (for an overview, see Laffont and Martimort, 2002). The Political Sci-

ence literature has attempted to use these models to characterize some of the constraints that define the politician-bureaucrat re-lationship. Huber and McCarty (2004) model delegation and policymaking when bureaucratic capacity is low; Huber et al (2001)assess when to use statutory control to regulate agencies; Volden (2002) investigates the degree of discretion to give bureaucrats;Bendor et al (1987) look at the role of unertainty about organization productivity in delegation; and, Banks and Weingast (1992)at the optimal design of agencies. None of these authors look directly at the delegation and monitoring framework pervasive ineconomics.

3The underlying assumption is that, in areas of high political competition, there may be significant incentives for public goodprovision as a source of votes. This has been explored for the US by Levitt and Snyder (1997) and for Africa by Weghorst andLindberg (2012).

3

tion impacts on bureaucratic performance. Once we have controlled for these mechanisms, the positive

impact of political competition is no longer significant.

Similarly, a number of papers have shown that politicians may be incentivized to affect the provision

of public goods for personal consumption. For example, Besley et al. (2004) and Besley et al. (2011)

find that politicians in charge of distributing fiscal transfer programs and public goods are prone to

directing resources towards themselves or their ethnic group. Pande (2008) provides an overview of the

determinants of political corruption in low income countries. We provide direct evidence of politicians

negative impacts on the provision of public goods, and show how political competition constrains the

excesses of this behavior.

Micro-level quantitative evidence on the interaction between politicians and bureaucrats has been rel-

atively scarce given the paucity of data.4Iyer and Mani (2012) use administrative data on the careers of

Indian civil servants to show how politicians affect the process of bureaucratic assignment across public

organizations. Balla (1998) finds little evidence of political control of the bureaucracy in the context of

the United States Medicare programme. Krause et al (2006) look at the impacts of politicized staff on

the efficacy of revenue forecasting in the United States.5 This paper adds further empirical evidence to

this potentially significant relationship.

One of the final results of the paper indicates the possibilty that political competition without appro-

priate mangement by a politician may in fact retard project implementation. This opens up important

questions of the roles of elites in the interactions between bureaucrats and citizens. By doing so, it pro-

vides micro-level evidence of the relationship between elite incentives and public good provision much

discussed in the new institutions literature (Acemoglu and Robinson, 2008; Besely and Persson, 2009).6

The rest of the paper is organized as follows. Section 2 overviews the relevant aspects of the Nigerian

government. Section 3 describes the data collected. Section 4 outlines my empirical strategy. Section 5

outlines the results. Section 6 concludes. The Appendix presents further data description.

4Incentives for bureaucrats instituted by politicians can be said to be ‘top-down’ in nature. The literature on ‘bottom-up’incentives for bureaucrats could be seen as a natural counterpart, for which there is a more extensive literature. Examples ofthis literature are Besley and Burgess (2002) which documents the benefits of media development on Indian local governmentperformance, Reinikka and Svennsson (2005), which studies a newspaper campaign that empowered Ugandan citizens to monitorlocal officials, and Olken (2007), which contrasted the impacts of community and audit monitoring efforts on the quality of ruralroad implementation in Indonesia.

5Our focus is the core civil service of the executive branch. There is also a literature on political engagement of judges. Politicalinfluence on the judiciary is obviously related to, but not the focus of, this study since the means of interaction are likely to bedistinct. Ramseyer and Rasmusen (2001) examine the impact of politically salient judicial decisions on the careers of judges inJapan. Besley and Payne (2003) present evidence that judicial selection methods affect the nature of judgements by US judges.Park and Somanathan (2004) document explicit links between Korean politicians and public prosecutors and their impact onbureaucratic assignments. Similarly, a number of papers study political influence on the private sector. Khwaja and Mian (2005)document how government banks in Pakistan lend more to politically connected firms and these firms are significantly morelikely to default on their loans. Sukhtankar (2012) finds evidence from India of politicians embezzling funds from politcallyconnected sugar mills during election years.

6Additional results in the paper allows us to touch on a number of literatures such as pork-barrel politics (Shepsle and Wein-gast, 1981; Lizzeri and Persico, 2001; Aghion et al, 2005, 2007, 2010) and decentralization (Bardhan and Mookherjee, 2000; Fismanand Gatti, 2002).

4

2 Institutional Background

Nigeria is the most populous country in Africa, with a population of 160 million people, or 20% of

the population of sub-Saharan Africa. United Nations (2013) predicts that Nigeria’s population will be

larger than that of the United States by 2050. It represents a leading setting in which to understand the

determinants of public sector productivity in the developing world.

Nigeria also shares important features of its economy and polity with other developing countries. Its

income per capita is roughly equivalent to that of India, or to that of sub-Saharan Africa as a whole

(World Bank, 2012). Its government makes up a similar proportion of economic activity as those of

many other developing countries, representing 26% of gross domestic product.7 Its political history

is marked by colonial origins preceding a string of military dictatorships, much like other developing

nations. Thus, Nigeria presents a window into the workings of government in the developing world.

The country returned to civilian rule with Presidential elections and a new constitution in 1999. Its

constitution has many similarities to the United States and its Congress shares many of the functional

components of the United States Congress. Nigeria is a Federal Republic, with an elected bicameral

National Assembly composed of the Senate and the House of Representatives. Its three branches are

the legislative, executive, and judiciary, and its three tiers are the federal, state, and local government

levels. Our study will focus on public organizations at the federal government level only.

The National Assembly is made up of 109 senators and 360 representatives. Each senator represents her

own constituency, whilst of the 360 representatives, 10 representatives share a constituency leaving 355

federal constituencies. Both senators and representatives serve four-year terms, and there are no limits

on re-election. I study the representatives of the 5th National Assembly which was elected in 2003 and

lasted until 2007. Representatives are elected by plurality vote so that each politician can be associated

with public projects implemented in their constituency.8

Political relations in the Senate and House are governed by separate bodies of rules called the standing

orders. These rules place many of the most important constraints on political activity in the country.

They define the duties and functions of officers of the chamber, the procedures by which chamber

business is done, and the standing committees in which much of the work of the legislative arm takes

place. Our focus in this study will be on the House of Representatives, and thus the standing orders

and committees of the House.7According to the International Monetary Fund World Economic Outlook Database (October 2012), government expenditures

as a percentage of GDP are 21% in China, 27% in Kenya, 28% in India and 30% in South Africa.8Of the 10 representatives that share a federal constituency, they split the constituency geographically in terms of responsibil-

ities.

5

2.1 Role and composition of standing committees

A core feature of the House of Representatives as dictated by the standing orders are the standing com-

mittees. Standing committees are permanent legislative panels that consider bills and issues relevant

to their sector of expertise (such as water, health, etc.). For each sector, the relevant standing committee

defines the public projects to be implemented in the coming year and determines who should imple-

ment each one. For example, the House Committee on Health will consider all issues relevant to health

in Nigeria, including how many health centres should be built within a fiscal year and by whom. The

standing orders define the list of committees, their jurisdiction, and their broad rules of operation.

The standing committees play a crucial role in the design of the federal budget. The projects we study

were all established in law by budget appropriation bills passed in 2006 or 2007. The passage of appro-

priation bills goes as follows. Having received inputs from the executive branch of government, a draft

appropriation bill is presented to the National Assembly. The draft bill is then split into sectors (water,

health etc.) and sent to the sectoral standing committees of the House and Senate. These committees

are delegated to hold hearings with relevant parties, scrutinize the proposals and define budgets for

each of the organizations we study. The committees then recommend a budget for the sector to an Ap-

propriation Committee which merges the recommendations into a single budget. This unified budget

is then voted on by both houses to form that year’s budget appropriation bill.

Membership of a standing committee provides a congressperson with significantly greater capacity to

influence the details of that sector’s budget than a non-member.9 Whilst there is a complex congres-

sional bargaining game that defines the broad features of a sector budget, committee members deter-

mine which projects should be implemented and by whom, and thus are most able to influence which

organization implements projects in their constituencies.

It is unlikely that members will have free reign to direct resources to their constitutiency, as this reduces

funds available for other representatives. However, members are likely to face little resistance to del-

egation decisions for projects in their constituencies, as this does not affect the constituencies of other

representatives nor the total quantum of resources available to other politicians. Much of the political

discourse and congressional bargaining in Nigeria revolves around equitable revenue sharing, in the

sense that each representative should receive a relatively equal share of national resources.10 There

is little, if any, discourse around delegation. It does not matter to a representative how another dele-

gates, as it does not have impacts across constituencies. As such, representatives on a standing committee9The significant power of standing committees has long been recognized for the US Congress. Woodrow Wilson asserted

that committees dominate congressional decision making, stating that "we are ruled by a score and a half of ‘little legislatures’"(Wilson, 1885). Richard Fenno, in his majesterial book on committees in congress, states that committee decisions are usuallyaccepted and ratified by the other members of the chamber, giving members of a committee significant influence over the sectorthey represent (Fenno, 1968). Shepsle and Weingast (1987) and Krehbiel et al. (1987) elaborate theories of why the congressionalcommittees are so powerful.

10For example, a number of groups of projects were entered into the 2006 and 2007 budgets as relating to ‘one in every localgovernment of the federation’ or ‘one in every state of the federation’.

6

should be little constrained in their ability to delegate to their preferred organization. Therefore, we

are studying a margin of political choice where we might expect to observe variation. Non-members

would somehow have to influence the committee process for their constituency, adding a significant

transaction cost to their delegation abilities. Similiarly, membership of a standing committee legitimizes

engagement with the bureaucracy in ways that are not available to non-members.11

The delegation and engagement powers conferred by membership of House committees is at the heart

of the analysis undertaken in this paper. I will compare the delegation and engagement decisions of

members to those of non-members. In doing so, I aim to compare politicians with greater delegation

and engagement power to those with less. This allows us to observe what politicians do with these

additional powers. How members of the standing committes are selected is thus an important element

of my identification strategy.

Membership of the standing committees is determined by the Committee on Selection. The process of

selection is a key difference between the US and Nigerian congressional procedures. As described in

Aghion et al. (2005, 2007, 2010), in the US Congress committee membership is driven by ensuring the

committees reflect state-level distributions of Democrats and Republicans. For example, if Florida’s Re-

publicans occupy a large proportion of Florida’s seats, they will take up a large proportion of Florida’s

committee places, with the most senior Republican first in line for a new committee place.

To understand the process in the Nigeria context, I gathered information from the rules governing

the House, from academic and committee secretary assessments of committee composition, and from

newspaper reports.

The Standing Orders of the House of Representatives lays down that the Committee on Selection is the

single authority for the determination of committee composition. It also determines the membership of

the Committee on Selection to be the principal officers of the House.12 The process of committee selec-

tion is dominated by a political requirement that each party and geo-political zone is fairly represented,

and a guiding principle that committee members are qualified in the sector in which the committee

specializes.

The Committee on Selection must balance the numerical strength of the parties in the House and the

need to represent each of the geo-political zones with a guiding principle that members should be

allocated to sectors in which they have relevant qualifications or experience. Qualifications seem to

play a critical role in selection.13 This implies that politicians who qualified as doctors should be placed11Nigeria’s bureaucrats state that politicians who are members of a standing committee often visit them under the pretence of

a ‘courtesy call’ on behalf of the committee they sit on.12These are the Speaker, the Deputy Speaker, House Leader, House Whip and Leaders of the other political parties. The ultimate

person that determines who belongs to which committee is the chair of the Committee on Selection, the Speaker, but the processis said to be consensual amongst the Committee on Selection members (David, 2011).

13As the Speaker of the House has confirmed in the House Hansard, “In the composition of Membership and Leadership ofCommittees, special attention will be paid to the skills and relevant experiences of Members in order to achieve greater efficiency”(House Hansard, 2011).

7

on the health committee, educationalists on the education committee, and so on.

I utilise the rules for selecting members of standing committees in my empirical strategy to investi-

gate delegation and engagement in the House of Representatives. My argument is as follows. Macro-

political factors, made up of a need to represent geo-political zones and parties in proportion with their

size in the Congress, are not correlated with micro-level characteristics of the constituency. Similarly,

we can expect that the decision a politician made decades ago as to which sector into which to spe-

cialise is not correlated with the capacity of a government organization to implement projects in their

constituency today.14 I will show empirically that this is the case. Since the determinants of committee

membership are exogenous with respect to constituency-level characteristics, members are allocated

additional bargaining and engagement power relative to non-members in the sector of their commit-

tee. I will therefore utilise this exogenous allocation of power by congressional procedure to investigate

delegation and engagement.

How closely this congressional procedure is followed is important for the empirical strategy. The poten-

tial for politicians lobbying across committees is clearly a concern, as they may lobby based on political

considerations in their constituency. I therefore consider six pieces of evidence that relate to committee

selection. All of them seem to indicate the three factors noted above play the core roles in determining

committee membership.

First, we are guided by what members of the Committee on Selection state that they do. They state that

their decisions are based on the geo-political factors and qualifications I have described. In a number of

fora they go on to note that the paramount importance of these macro-political concerns provides little

room for manouevering based on constituency concerns.15

Second, I can test this assertion quantitatively by coding the sector/s in which a politician has been

selected to serve on a committee. I can then regress these on the sector/s of her qualifications and

a host of constituency-level characteristics. Almost across the board we find evidence to support the

assertion. I will present the data and regression evidence together in the next section.

Third, we may be concerned that constituency-level characteristics determine the sectors into which

politicians specialised. Thus, indirectly such characteristics are determining who is selected into a par-

ticular committee. Again we can use regression evidence to test this. I find almost no evidence of such

a relationship.

Fourth, we can enquire as to what the political environment was in which politicians made their career

decisions. Could they have selected in to a career with the long-term strategy of taking up a place

14As will be stressed below, since Nigeria only became a democracy in 1999, politicians would not have known that the com-mittee structure or House standing orders would be designed as they were during the time that they made their career decisions.

15For example, the vice chairman of the Committee on Selection states, as reported in This Day newspaper, “The SelectionCommittee ... considered cognate experience, areas of specialization and zonal representation in order to ensure that the chairmanand vice chairman of a committee do not come from the same geopolitical zone.” (This Day, 2007)

8

on the relevant House committee thirty years down the line? Since Nigeria only became a democracy

in 1999, almost all of the qualifications on which the Selection Committee made their decisions were

gained well before the standing orders were even conceptualised. Politicians would not have known

that the committee structure or House standing orders would be designed as they were during the time

that they made career decisions, nor perhaps whether Nigeria would have a National Assembly at all.

Fifth, if constituency-level characteristics determined the sectors into which representatives selected,

we should find that senators from the same areas have selected into the same sectors. I find this is rare.

Only 13% of representatives are on a committee of the same sector as their senator. For any particular

committee, that figure is at most 3%.

Sixth, we can ask the administrators who have the most detailed understanding of the committees but

are not themselves politicians. I undertook a series of qualitative interviews with the secretaries of the

house committees we study. They were relatively uniform in their opinion that the members of their

committees all had relevant sectoral qualifications or experience and were chosen principally on that

basis.

The totality of the evidence points towards the House Selection Committee determining committee

membership based on factors that are exogenous to constituency-level characteristics that might have

significant impacts on project implementation. Thus, as Payne (2001) and Feller (2002) find substantial

evidence for arbitrariness in the makeup of US congressional commitees, I find a similar phenomenon

in Nigeria with respect to local political conditions. Relative to local factors that determine project

implementation, the granting of committee membership is an arbitrary allocation of additional power

to delegate public projects.16

2.2 Delegation to executive organizations

Members of the standing committees must decide which organizations within the executive to dele-

gate to. Each of the projects funded by the federal budget must be assigned to an organization of the

Executive. The National Assembly itself does not implement social sector public projects.

To analyze this delegation decision, I split the Executive into centralized ministries and their decentral-

ized agencies. This is the major categorical division of public organizations in Nigeria. Discussion of

implementing agencies is typically split in terms of ministries and agencies.

A centralized organization is a ministry, the central organizing authority for a sector. For example, the

Ministry of Health is the central organizing authority for the Health sector. It defines the long-term

16More discussion on the process utilized by the Committee on Selection to select committee members is provided in theAppendix.

9

strategy for the sector and interacts regularly with Parliament, and its Standing Committee on Health,

on the legislative aspects of healthcare in Nigeria.

The agencies of a ministry are independent bodies, established by law as self-accounting entities with

their own budget line in the federal budget. Agencies are run by a chief executive with boards of

governors chosen by the President. The membership of these boards is relatively stable, with most

members of the board spending many years in their position.

The position of these organizations in the hierarchy of the Federal Government implies that the min-

istries are closer to the National Assembly and the politicians of the House of Representatives. This is

true both in terms of government organization as well as geography. The centralized ministries are all

based in Abuja and are determined by the Public Service Rules, the rules that govern the public sector,

as the first point of contact for politicians interested in a sector. The decentralized agencies are on av-

erage hundreds of kilometers from the capital and are separated from the National Assembly in terms

of government organization by their parent ministry and their board of governors. Whilst they are ac-

countable to the National Assembly for the use of public funds, these organizational and geographic

constraints imply they are more independent of the National Assembly than the ministries. I will look

at the empirical evidence that supports this in the next section.

The relationship between the National Assembly and the two sets of organizations we study is relatively

stable. This is not only due to the stability of the organization of the Federal Government. There is little

change in the structure of many of the organizations we study. For example, the staff of ministries and

agencies are rarely moved. The data I collect from a representative sample of staff at the organizations

we study indicates that the mean tenure at the organizations we study is 13 years, with centralized

organizations having a slightly lower mean tenure of 11 years.17 There is little movement into the civil

service during the study period, nor much movement around it. The bureaucrats we study are subject

to public service rules which govern their appointment to and exit from the service, discipline, salaries,

management of public funds, and other major aspects of official assignments. Changing these rules, as

well as changing any other aspect of the civil service such as the design of public organizations, is a

slow and bureaucratic process. It is not amenable to political manipulation.18

Politicians faced by these slow changing structures therefore have the opportunity to learn about the

capacities of the two types of organization. They also have a choice between a centralized ministry

and a decentralized agency for every project we study. Since the ministries are national in scope, they17Given that Nigeria became a democracy in 1999, the majority of officials joined their organization under military rule, when

political competition of local constituencies would have not been a factor in their decision as to which organization to work for.18For another research project based on federal government organizations, I surveyed a number of managers at Federal Gov-

ernment organizations on the rate at which management changed. It could be that whilst staffing and formal rules do not changerapidly, the informal rules of organization that make up management change in response to political interference. However, hereto it was reported that there was only very slow change. The drivers of this change were said to be foremost changes in the publicservice rules, second the management preferences of the history of managers at an organization, and third external shocks typi-cally related to trade union activity. Overall, these drivers were not associated with short-term political interests of constituencypoliticians.

10

provide a centralized option for every project in their sector. There are also decentralized options for

every project, since the decentralized agencies provide a contiguous coverage of Nigeria within each

sector.19

Whilst each of the projects we study is implemented by a single tier, ‘like’ projects are implemented

by distinct tiers. For example, the Federal Ministry of Water Resources implements projects in districts

throughout the areas covered by the (decentralized) River Basin Development Authorities. In many

districts, ‘like’ water projects are implemented by distinct tiers. Similar arguments can be made for

other sectors. To ensure we are not picking up differences in project characteristics across tiers, we use

variables that reflect the budget, existing investment, and complexity of a project. It is to the body of

data collected on each aspect of the implementation of public projects that we now turn.

3 Data

To test the delegation and engagement margins discussed above, we require data from across govern-

ment. Based on the Federal Government of Nigeria’s 2006 and 2007 budgets, we have assembled a data

set that combines characteristics of politicians with their positions in Parliament. We have details of the

projects they delegate to organizations, surveys of the organizations that implement these projects and

evaluations of how effectively they do so. Overall, we have a snapshot of the delivery of public goods

from initiation to implementation.

We build up to the description of our right-hand side variables. Our core left-hand side variables relate

to politicians, their constituencies, and the characteristics of projects that are implemented in those

constituencies. We first elaborate these, and then describe the dependant variables in our final sub-

section.

3.1 Politicians and the National Assembly

Nigeria’s 5th National Assembly was inaugurated in May 2003, and consisted of 109 senators and 360

representatives. Our focus will be on the House of Representatives. For each of the 360 representatives

of the 5th National Assembly, we have assembled biographies that outline their demographic and polit-

ical characteristics (such as their party affiliation), their educational qualifications and work experience,

and the results of their election in 2003. The Appendix details the construction of these biographies

and Table A1 describes the politicians we study. The vast majority of our politicians are men, with a

19Another important feature of the public organizations we study is that the vast majority are serving multiple constituen-cies. Very few would only have the interests of a single politician as a dominant concern, and thus they are unlikely to changeorganization-level characteristics to suit the whims of a single politician in a short time-frame.

11

mean age of 48, and 16 years of education (equivalent to an undergraduate degree). Only a quarter of

politicians are serving a second term.

Table 1A describes the constituencies we study. The sample of public projects in our data are imple-

mented in 349 of Nigeria’s 355 federal constituencies.20 We provide descriptive statistics for the subset

of 349 we study. The descriptive statistics for Nigeria as a whole and our sub-set are almost identical,

indicating that the constituencies we study are representative of Nigeria as a whole.

The average population of a constituency is 380,000. This population is split, on average, between two

local governments, the most basic administrative unit of government in Nigeria. The local nature of

congressional politics implies that local public good provision is central to the success of a representa-

tive’s time in office.

The constituencies, reflective of Nigeria as a whole, are generally poor, with a high proportion of their

population in extreme or relative poverty. The average years of education is 5 years, equivalent to less

than primary school completion. Only 41% of Nigerians have potable water, and the average number

of hours of electricity available per day is 4.5. There is plenty of opportunities for local public goods to

have significant welfare impacts across the constituencies we study.

To understand the political dynamics of our constituencies, I collected data from the Independent Na-

tional Electoral Commission on the returns for each of the elections in 2003 (in which the politicians

we study were voted into power).21 This is data that was published as the official roll of the national

elections.

The political competition faced by representatives is relatively heterogeneous across constituencies.

Whilst the mean difference between the winning vote share and that of the runner up is 0.32, there

is substantial heterogeneity across the country. As Figure 1A shows, we observe constituencies subject

to the full distribution of levels of political competition, providing sufficient variation from which to es-

timate our parameters. This also allows us to make sensible interpretations of coefficients that describe

competition at very low and very high levels.

The heterogeneity in competition is not concentrated in one area of Nigeria. As Figure 1B indicates,

there are large differences in competition between neighbouring constituencies. The average differential

in vote share across constituencies is ???% [I am waiting for an ArcGIS licence from UCL before I can

do this.]

We utilize two different methods for measuring political competition that are popular in the economics

literature. Both are based on the voting data I obtained from the Independent National Electoral Com-20Of the 360 representatives, 10 representatives share a constituency leaving 355 federal constituencies. The characteristics and

winning vote shares of the pairs who share constituencies are very similar. We therefore allocate the constituency to that one ofthe two representatives who is first on the nominal roll. Our results are robust to using the second of the two representatives.

21Specifically, I obtained copies of the data published in the Independent National Electoral Commission’s ‘Compendium ofResults of the 2003 General Elections: Vol.1: Presidential and National Assembly Elections’.

12

mission. The first is utilized by Lee (2001), Besley and Burgess (2002) and Da Silveira and De Mello

(2011). It is defined as the winner’s vote share minus the runner’s up vote share, and represents the

proximity of the runner up to the winner. We then subtract this number from 1 so that the measure is

increasing in competition. Thus, a value of this measure of competition of 0.4 implies that the winner

of the constituency election has a vote share 60% larger than that of her opponent. The second, which

we will use as a robustness check on our results, is utilized by Besley and Case (2003) and Besley et

al. (2010). It is defined as the deviation of the winner’s vote share from 0.5, subtracted from 0.5 so it is

again increasing in political competition. Thus, a value of 0.4 implies that the winner of the constituency

election has a vote share of either 40% or 60%. These two measures of competition have a correlation of

0.94. Whilst they are highly correlated, they represent quite distinct political concepts and thus act as

the best robustness check we have available.22

Table 1B looks at the impacts of political competition on a range of variables of interest. We see that

political competition does raise the rate of completion of public projects. There is no evidence from

these simple correlations of political competition leading to delegation.

One concern in Nigeria is that politicians of the ruling party have very different behaviours to those

of other parties. Roughly two-thirds of constituencies are governed by a politician of the ruling party.

The second half of Table 1B looks at the extent to which politicians of the ruling party dominate politics

in Nigeria. We see that they are not able to secure higher funds per constituency or more projects, at

least in the OPEN data. There does seem to be a mild negative impact on delegation and completion

of public projects. However, the correlations indicate that ruling party politicians are not operating in a

highly distinct environment and thus we can assess the body of politicians as a whole.

From the set of politicians voted into power in 2003, the Committee on Selection chose members for the

standing committees. For each of the relevant committees of the 5th National Assembly, we coded the

membership of the committee using data from Nigerian Congress, a web site that provided details of

all the House Committees set up in 2003. We also gained information on membership for the relevant

period from secretaries of the relevant committees. The committees we study are agriculture, appropri-

ation, education, environment, Federal Capital Territory, finance, health, housing, power, water, women

and youth. Where there are small changes to the committees over time, we use the membership rele-

vant to the 2006 and 2007 budgets. We then matched each representative to the committee membership

records, and noted their position as member, chair, or vice-chair.23

22Table A1 compares the politicians and constituencies at the top and bottom 10% of competitive constituencies as defined byour core measure of political competition. We can see that the politicians are not particularly different, with politicians in the mostcompetitive constituencies having roughly half a year more education and being 7 percentage points more likely to have a secondterm. Similarly, there are no clear patterns of difference between the constituencies, although the most competitive constituenciesseem to have a little less access to water and a slightly longer walk to the nearest primary school. This would imply that thepatterns we observe are primarily incentive rather than selection effects.

23The US literature on standing committees I take as the guide for this study (Aghion et al. (2005, 2007, 2010) and Cohen et al.(2012)) conditions quantitative regression specifications on the ‘grade’ of the standing committee. Congressional committees are

13

The mean and median number of committees under study that a representative sits on is 1. All repre-

sentatives in the House are expected to serve on a committee although we only study the committees of

the social sectors and therefore do not observe membership of all committees in the House. We there-

fore observe some members in our data who do not sit on any of the committees we study, although

they are likely to be sitting on one of those committees we do not study. Some representatives serve on

multiple committees, and the maximum number of committees a representative sits on is 4.24

To further characterize a representative’s constituency, we utilize the largest household survey ever

undertaken in Nigeria, the 2005 Core Welfare Indicators Questionnaire (CWIQ).25 This was a cluster-

randomized household survey run by Nigeria’s National Bureau of Statistics, representative at the local

government level. Implemented in the months immediately preceding the implementation period of

our projects, it provides a baseline profile of the constituencies we study.

In much of the literature that assesses government performance, studies assess a community after the

public goods have been provided. However, as stated in Banerjee et al. (2007)’s assessment of the chal-

lenges to research on public good provision, “community characteristics may respond to the availability

of public goods”. The CWIQ survey was implemented at the end of 2005, approximately at the base-

line of our study and at the same time as politicians were making delegation decisions in the standing

committees.

Using this survey, we create constituency-level averages for indicators of poverty, access to existing

public goods and local economic dynamics. We expect politicians to delegate to areas of need, where

deprivation is highest or where there are significant degrees of inequality. We therefore construct

constituency-level means and standard deviations of the following indices: the proportion of poor in

the constituency, measured by both a national poverty index and a self-rating index, the average years

of education of the household head, the proportion of constituents with access to potable water, the

average number of hours constituents had electricity in the 24 hours preceding the household ques-

tionnaire, the average time in minutes to the nearest primary school, and the average journey time in

minutes to the nearest secondary school. These indices reflect the sectors under which the majority of

the projects in our data fall.

We may also expect politicians to respond to recent investments in their constituency. For example, a

politician may feel less inclined to invest in an area that has recently received substantial public goods

typically seen as having a hierarchy of importance, and the grade of the committee reflects the standing of the committee withinthat hierarchy. The rationale for following this practice is that in politically important committees, the dynamics of delegationmay be distinct from other committees. I follow this practice by including a binary indicator of the grade of the committee underwhich the project falls in all specifications. This is a dummy that takes the value 1 if the committee is perceived to be of highpolitical weight or 0 otherwise. We follow Ojeifo’s (2007) delineation of grade A (agriculture, water, education, power, housing,environment, Federal Capital Territory) and grade B (health and women) committees. Our core results are all qualitatively thesame when we do not include this variable.

24To understand how this reconciles with the need for sectoral knowledge, some politicians have qualifications/experience inmultiple sectors. For example, they may be a qualified doctor who is a professor of medicine at a university. This individualwould likely serve on both the health and education committees.

25The CWIQ survey targeted 774,000 households, 100 in each local government area.

14

investments. To reflect the frequency with which constituents have benefited from a public project

of the named type in the five years preceding the survey, we also construct indices of whether con-

stituents have received: construction of electrification infrastructure, rehabilitation of electrification in-

frastructure, a well/borehole, construction of piped water infrastructure, rehabilitation of piped water

infrastructure, sanitation, school construction project, school rehabilitation, health facility construction,

health facility rehabilitation, road construction, tarring/grading of roads, transportation services, and

agricultural-inputs schemes. Again, these indices correspond to the sectors of the projects we study.

Finally, there may be complementarities between the economic environment and public investments.

For example, greater access to credit may lead citizens to demand public goods that will facilitate their

use of that credit. Politicians may therefore respond to the economic dynamics of a constituency in

their delegation decisions, and so we create indicators of changes in opportunities for employment, the

availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension

services, the availability of credit facilities, and the availability of consumer goods.

Together, the politician characteristics and these indices provide a relatively comprehensive set of con-

trols at the constituency level. They also allow us to test a key element of our empirical specification.

We argued in section 2.1 that committee assignment was determined by the qualifications and relevant

experiences of politicians rather than local political and socio-economic conditions. We now have the

data to test that assertion. We can do this in two ways.

We first investigate the direct determinants of committee membership. Table A2 investigates these de-

terminants by estimating a seemingly unrelated regression (SUR) model across indicators of committee

membership. For each constituency-level regression in the SUR system, the dependent variable is a

binary variable reflecting whether a representative is a member of the committee for the named sector.

We regress these indicator variables on the politician and constituency controls described above, and

a series of dummy variables that indicate the politician’s geo-political region.26 We then display the

coefficients for those variables relevant to all the sectors. Table A2 indicates that for all bar one of the

committees, there is strong evidence that qualifications play the major role in determining which politi-

cians are allocated to a particular committee. There is also little correlation in error structure, implying

there is no evidence of underlying unobservables selecting politicians across sectors. There is evidence

for the agriculture committee that other factors play a role, and we will check the robustness of our

results for this deviation.

Second, we may be concerned that constituency characteristics indirectly determine committee mem-

bership by determining the sector of a representative’s qualifications and experience. For example, a

politician from an arid region of Nigeria may enter into the water sector. It may then be more chal-

26Where we could not find details of a politician’s biography, we took the most conservative approach possible and coded thatindividual as having no relevant experience in any sector. Thus, the regressions here are based on that conservative approach.

15

lenging to implement water projects in that region and so the politician experienced in the water sector

delegates to decentralized organizations who better understand her constituency’s needs. Table A3

attempts to investigate the determinants of a politician’s sector of qualifications and experience by es-

timating a SUR model of sector expertise on constituency characteristics. We create dummies for each

of the sectors into which a politician could have specialised. These take the value 1 when the politician

has specialised in that sector, and zero otherwise. Whilst we do not have constituency characteristics

for the periods in which politicians made their career decisions, we would only be concerned if these

characteristics drove career decisions and were correlated with constituency characteristics today.27 We

present coefficients on variables relevant to all sectors, as well as those variables most relevant to each

sector from within our set of controls. We see that in general there is little evidence that constituency

characteristics determine the sector into which a politician specializes. Almost none of the coefficients

are significant at the usual levels.

3.2 Public projects

In 2006 and 2007, the 5th National Assembly legislated a Federal Budget of US$12.7 billion and US$15.1

billion respectively, or US$27.8 billion cumulatively. This total amount contains expenditures on debt,

defense, and the police. Our interest is social sector capital projects, which account for roughly 35%

of the total, or 73% of capital expenditures. We do not have data on all the projects in each of these

budgets, but a representative sub-set.

In both of these years, Nigeria’s Presidency undertook a unique monitoring and evaluation initiative

that tracked the implementation of a representative sample of social sector projects.28 The ‘Overview

of Public Expenditure in NEEDS’ (OPEN) monitoring initiative arose out of Nigeria’s receipt of debt

relief in 2005. As a result of sweeping reforms across major organs of government (for an overview see

Nkonjo-Iweala and Osafo-Kwaako, 2007), Nigeria received cancellation of its external debt to the tune

of US$18 billion from the Paris Club.29 At the federal level, the annual savings from debt interest were

channeled into the social sectors (health, education, water etc.) that are our focus. The Presidency saw

this as an opportunity to track the effectiveness of government expenditures, and so in 2006 and 2007

the Nigerian Government traced, at a project level, the use and impact of 10% of all federal Government

social sector expenditures.

The OPEN projects were designed to be representative of existing social sector expenditures, but also27As is stressed in section 2.1, Nigeria’s current democracy only began in 1999 and thus its political details were not known at

the times when Nigeria’s politicians were choosing which sectors to specialise into.28A comparison of the distribution of funds in the OPEN program with that in the Federal Government budget as a whole

indicates that the sample is representative across sectors.29In 2003, the Nigerian Government began a reform program that was partially motivated by the promise of debt relief. Years

of military rule had undermined the country’s public institutions and the newly-elected President began his second term aimingto strengthen Nigeria’s economic position. A fiscal rule was introduced to de-link public expenditures from volatility in oil-revenues, state institutions were privatized, and a number of sectors deregulated to encourage private sector participation.

16

to be informative for those projects that were most needed to be scaled-up nationwide. The set of

OPEN projects can therefore be seen as a representative set of government expenditures. Since we are

interested in delegation, we are only interested in those projects that can be delegated to either tier of

government. In other words, those projects that could be implemented by either ministries or their

agencies. We therefore exclude all projects from the full set of OPEN expenditures that have national

or inter-jurisdictonal scope, such that a single decentralized organization does not have the mandate

to implement such a project. Examples of these projects are those that require engagement with inter-

national organizations or that are implemented nationwide. This leaves us with a representative set of

social sector projects that could be delegated to both ministries and their agencies.30

For each of our projects, we have evaluations of how effectively they were implemented. Under the

OPEN initiative, expert teams were sent to visit the selected projects and identify the extent to which

they had been implemented as planned in the Federal Budget, and embodied in each project’s technical

documentation. The Presidency contracted national and regional teams to undertake the monitoring

process outside of the institutions of the civil service. Hence the public sector projects were not evalu-

ated by potentially biased civil servants, but rather by teams of independent engineers and civil society.

The engineers evaluating the projects were not those working on the project sites and the civil society

groups were recognized third sector organizations.

Evaluations of the OPEN process indicate it successfully achieved its aims (Eboh 2010, Dijkstra et al.

2011). To ensure the accuracy of monitoring reports, the Presidency put in place a system of checks

and balances. First, a centralized team of technocrats monitored the evaluation teams, providing them

with training and opportunities for standardization of their methods at national conferences. Second,

evaluators were asked to provide material, photographic, or video evidence to support their reports.

Third, the national teams and Presidency performed random checks on evaluated sites, all of which

were consistent with the findings of OPEN monitors.

The reports of OPEN evaluators describe the fate of projects budgeted for execution in the 2006 and

2007 federal budgets (Federal Government of Nigeria 2008, 2009). I hand coded the material from all

projects recorded in OPEN initiative reports. Taken together, the coverage of projects in our sample

traces 7% of all Federal Government social sector expenditures in 2006/7 budget years, corresponding

to 4493 projects from 57 organizations, with an aggregate budget of around US$560 million.31

The OPEN evaluation teams coded: (i) whether the project had started; (ii) its stage of completion; (iii)

the quality of the inputs and work. Our main outcome variable is a continuous measure, from zero to

30More detailed information on how we determined whether projects were delegatable or not is provided in the Appendix.31We consider projects traced under the OPEN initiative that were approved in either the 2006 or 2007 federal budgets. For

projects funded in the 2006 (2007) federal budget, monitoring teams visited the relevant project sites around June 2007 (2008).Therefore, project implementers were given roughly 18 months from the time the project was centrally approved until when itcould be utilized by the community. All the projects we study had twelve month completion schedules, so that even accountingfor any delay in the disbursement of funds, it is feasible for these projects to be completed by the time of the monitoring survey.

17

one, of project completion rates: zero refers to the project never having been started, one corresponds

to the project being completed as specified in the original project description, and intermediate scores

reflect part completion. We have this ‘proportion completed’ variable for all of our projects. To maxi-

mize data coverage on project quality, we are forced to utilize the most aggregate formulation of quality

reporting. A project was either of insufficient quality, satisfactory, or commended for an ‘above average

or high’ quality level. With this definition, we obtain 1911 quality observations. We then define a project

quality indicator equal to one if the project is of satisfactory quality or above.

A politician’s delegation decision may be a function of a project’s characteristics. I therefore hand-

coded data on project-level characteristics such as the budget allocated to the project, whether it was a

rehabilitation project, and a brief summary of its technical specifications from project documentation.

I also coded which of 11 project types the projects fell into, with categories in both construction (local

electrification infrastructure, boreholes, buildings, and so on) and non-construction fields (procurement,

financial projects, training, and so on).

The project technical specifications were utilized to form engineer-approved measures of the complex-

ity of each project. In collaboration with a pair of Nigerian engineers familiar with the OPEN initiative

and a group of international scholars with research interests in project complexity, I drew up a series

of measures of the technical characteristics of the projects we study. These ‘complexity indicators’ were

based on the detailed technical specifications specified for each project, and are constructed follow-

ing engineering practice that emphasizes multiple dimensions of complexity (Remington and Pollack

2007). The Appendix: (i) details the construction of these indices, and presents descriptive statistics for

them; (ii) describes checks we put in place, using multiple engineers, to establish the validity of these

complexity measures.

These measures of complexity allow us to condition all our specifications on the aggregate complexity

of the project, which are likely to be important determinants of project completion. They also allow

us to define indices of local and national information needs. Given how important information has

been to the study of delegation (Moe, 2005; Mookherjee, 2006), it is important to understand how we

are controlling for the informational demands of each project. The literature on delegation emphasizes

the possible importance of superior information at a tier of government as a rationale for delegation.

Some projects require a lot of local information to be implemented. For example, they may require

sourcing of materials from the local area or be characterised by a high degree of uncertainty that requires

local information to respond to. Similarly, some project may require a lot of information more readily

available at the national level. For example, sourcing international expertise is something national

organizations are likely to be better at procuring than local organizations. The Appendix describes how

we then aggregated our complexity variables into indices of national and local information. We include

18

these in our project controls.

The project characteristics that have been assembled for each of the projects we study control for the

cost and scale of a project, any existing investment, it’s type, it’s complexity, and its informational

needs at both the local and the national level. Together they represent a relatively broad set of the

features one would assume important for project completion. Whilst we are making a conditional

independence assumption in our empirical work, that we have captured key determinants of project

characteristics that might influence delegation decisions, the breadth of our controls gives us some

reason for confidence in that assumption.

Table A4 provides descriptive evidence on each project type studied. Typically they are small-scale

projects that serve basic local needs such as the provision of a basic health center, the improvement

of local electrification infrastructure, or the drilling of a well to supply drinking water and irrigation

for agriculture. These projects often constitute the ‘nuts-and-bolts’ of rural infrastructure development.

Column 3 of Table A4 indicates the budget of most projects is under US$100,000. Local electrification

infrastructure is the most common type, corresponding to 33% of the OPEN projects we study. Table

A4 also details the number of organizations that are engaged in providing each project, and indicates

that all project types are implemented by both centralized and decentralized organizations.

We can see from the second half of Table A4 that completion rates vary significantly across project

types, providing a rationale for conditioning on project type fixed effects in our specifications. Whilst

only 14% of the canal’s in our data were ever built, 53% of roads were completed. Similar variation

can be observed across the proportion of projects fully completed. Whilst only 5% of canals were ever

completed, 58% of procurement was done.

3.3 Public organizations

The 4493 projects we study in this paper were implemented by one of 57 organizations. These are split

into 8 centralized ministries and 49 decentralized agencies. Our first set of specifications will assess the

delegation across these types of organization using a binary variable that will take the value 1 when a

project is implemented by a centralized organization.

For each of these organizations we collected data on their budgets, staffing, and location. Table 2 pro-

vides descriptives for our organizations. We see that on all the margins presented, centralized organi-

zations differ from decentralized ones. They are larger in terms of budget, staffing, and the number of

federal constituencies they serve.

For 55 of these organizations, we undertook surveys of a representative sample of their staff. Logistical

19

reasons meant that we were not able to survey two of the organizations on which we have data.32

These organizations accounted for 736 projects between them.33 Our core interest in surveying the

organizations that implemented the projects we study is to understand how closely they are engaged

by politicians. To construct the dependant variable for the second of set of specifications focused on

engagement, we use a question from this surveys of civil servants.

Each of the officers we surveyed was asked the following question, which we utilize as our core measure

of engagement, “Think about recent projects and/or programmes you worked on for this organisation.

How often, if at all, do you personally engage with [member(s) of the National Assembly] in the work

that you do?” [italics in original]

Table 3 describes means and standard deviations for the average answers to this question. We see that

on 10% of projects a member of the National Assembly personally engages with a bureaucrat on the

implementation of a public project. Compared with the Weberian separation of powers, that is a sig-

nificant proportion. The mean level is higher for centralized organizations, indicating that centralized

organizations are more closely integrated with politicians from the National Assembly.34 The difference

between the two tiers of government is significant at the 10% level.

To check the robustness of this question, we also asked officials, “Think about recent projects and/or

programmes you worked on for this organisation. In what proportion of the projects have [member(s)

of the National Assembly] intervened in the implementation of a project?”

I present descriptive statistics for the answers to this question in Table 3. This reconfirms the pattern we

observe with our engagement question. Politicians intervene more frequently in the projects of central-

ized organizations, although it is not significantly different at the usual levels. The level of intervention

by politicians on this measure rises to one-quarter of all projects. Again, this is remarkably high if our

benchmark is a division of powers.

To illustrate the sorts of activities implied by the terms used in these questions, I draw on the qualitative

surveys I undertook in conjunction with the quantitative surveys from which these questions are taken.

After all the quantitative surveys had been filled and collected, we split managers and non-managers

and asked both groups about the key challenges they faced in the implementation of public goods. Po-

litical interference was at the top of the list in both groups. A number of examples from our discussions

illustrate the nature of political engagement in the Nigerian civil service.

What these examples show is that politicians engage with bureaucrats along a number of dimensions.

Initial contact was in person, through a designated representative, or by phone. A key characteristic32One of the organizations we could not visit was being restructured and the other was missed due to administrative errors by

the survey team.33The 736 projects that we drop in the analysis that utilizes the civil servants surveys are a little different to the 3757 projects

that we keep. They have slightly higher budgets and are 4% more complex.34We did not ask separately about senators and representatives. We therefore have to assume in this paper that our data is

representative of the behavior of members of the House. Qualitative interviews imply it is.

20

is that the politician seemed to have existing knowledge of the project on which he was engaging the

officer. Typically bureaucrats would personally meet the relevant politician at some point during the

project cycle. However, the initial contact was intended to provide some steer as to project design, lo-

cation, or completion. After that, the engagement was based around monitoring the designated plan.35

4 Empirical strategy

To assess political delegation and engagement in Nigeria, we need to observe differences across politi-

cians in terms of the power they have to delegate projects and to engage with bureaucrats. We can

then compare the delegation and engagement decisions of those politicians with more power to those

with less. Given that more powerful politicians face fewer constraints on their actions, the magnitude

of their delegation and engagement activities will be greater than those with less power.36 To interpret

the differences we observe in delegation and engagement as representative of underlying behavior, we

require the asymmetry in powers to be exogenous with respect to our variables of interest. In short, we

would like to give a random set of politicians delegation and engagement power and see how they use

it.

Membership of standing committees in the Nigerian House of Representatives seems to be driven by

macro-political factors and politician qualifications. Both of these are uncorrelated with constituency-

level characteristics that might determine project completion rates. This implies that membership of

the relevant standing committee for some projects and not others in a constituency is exogenous with

respect to project implementation. Having identified a feature of the chain of public good implementa-

tion that is plausibly exogenous, we can observe what committee members do with the projects under

their remit versus projects in the constituencies of non-members.

Conditional on the assumption that committee selection is exogenous with respect to constituency char-

acteristics, our empirical task is relatively straightforward. For each of the outcomes of interest (dele-

gation and engagement), we can look at the conditional impact of committee membership by level of

political competition. We have seen in Table 2 that decentralized organizations have significantly higher

project completion rates than centralized organizations. Using committee membership as a binary in-

dicator of additional delegation power, we can observe whether members are more prone to delegating

to decentralized organizations of government when they face political competition. Similarly, using

committee membership as a binary indicator of additional engagement power, we can observe whether

members are more prone to engagment of the bureaucracy when they face political competition.

35It was more difficult to ascertain what the reward structure of the agreement was.36In the language of microeconomics, we are comparing politicians at different points of their common expansion path in a

commodity space in which delegation/engagement is traded off with other uses of power.

21

We operationalize this strategy as follows, where our analysis is at the project level. A representative is

said to be on the relevant committee for a project if it is of the same sector as the committee of which

she is a member. In other words, a representative on the health committee is on the relevant committee

for all the health projects in her constituency. We assess the direct impact of committee membership

by including a dummy variable that takes the value 1 if a project is implemented in a constituency in

which the representative is a member of the relevant committee. For example, the dummy would take

the value 1 for a health project implemented in the constituency of a member of the health committee.

We are motivated to assess delegation and engagement differentially by levels of competition. We there-

fore include the measure of political competition described above and an interaction between committee

membership and political competition. With an interaction of this type, the levels effect of the mem-

bership variable measures the impact of membership for constituencies with zero political competition.

The interaction measures the impact of membership for higher levels of political competition.

In summary, we use the following specification:

yijscn = γ1membershipsc + γ2competitionc + γ3membershipsc ∗ competitionc + γ4PCij + γ5CCc + λj + εijscn (1)

where we estimate for the ith project, of jth type, in sector s, implemented in the constituency c by

organization n. When studying delegation, yijscn is a dummy variable that takes the value 1 if the

project is implemented by a decentralized organization and 0 otherwise. When studying engagement,

yijscn is a continuous measure on the unit interval that signifies the extent of politician engagement

of the organization in which the project is implemented. Membership is a dummy variable indicating

whether the congressperson of the constituency in which the project is implemented is a member of the

sectoral committee relevant to the project; competition is a continuous measure on the unit interval of

one minus the margin of victory in the constituency; membership*competition is an interaction between

membership and competition and thus also measured on the unit interval; project controls (PCij) are

the key project characteristics described above; constituency controls (CCij) are the key congressperson

characteristics and socio-economic characteristics of the constituency described above; and, project type

fixed effects (λj) absorb 11 project type level effects.

Since a zero value of the competition measure in our interaction corresponds to zero competition, γ1 es-

timates the delegation decisions of members who face low levels of competition in their constituency. γ3

estimates the delegation decisions of members who face high levels of competition in their constituency.

These are our core coefficients of interest for specification (1).37

37Using equation (1) we can outline the bias that would arise if committee membership was allocated in a way that correlatedwith unobserved constituency-level characteristics. Suppose some politicians face high levels of some unobserved accountability

22

These specifications compare projects implemented in a constituency with a politician on the relevant

standing committee for that project with projects for which that is not true. Since we control in the

majority of specifications for project type fixed effects, these comparisons are within project type (which

typically cross sectors). Thus, our identifying variation arises from comparing projects of a particular

type for which the constituency politician is on the relevant committee to those for which the politician

is not on the relevant committee.38 We stress that the majority of the specifications we estimate control

for a wide variety of project, politician, and constituency controls.

Having determined whether patterns of delegation and engagement exist, we want to estimate the

impact of these choices on project implementation at high and low levels of political competition. We

therefore turn to project implementation outcomes. For the majority of our specifications, yijscn is a

continuous measure of proportion completed across the unit interval. We also estimate specifications

for which yijscn is a binary measure of project quality that takes the value 1 when the project is of

satisfactory quality or above.

To explain the variation in these outcomes, we include our measure of political competition, the decen-

tralization dummy we used to assess delegation and the continuous measure of political engagement

we used in (1), as well as an interaction between these and political competition. In summary, we use

the following specification:

yijscn = θ1competitionc + θ2decentralizationn + θ3 political engagementn + θ4interactionsc+ (2)

+ θ8PCij + θ9CCc + λj + εijscn

where we again estimate for the ith project, of jth type, in sector s, implemented in the constituency c

by organization n, with yijscn a measure of project implementation (either the proportion completed or

the quality of the project). Decentralization is a dummy variable that takes the value 1 if the project is

implemented by a decentralized organization and 0 otherwise; political engagement is a continuous mea-

sure on the unit interval that signifies the extent of politician engagement at the organization in which

the project is implemented; competition is a measure of one minus the margin of victory in the con-

mechanism that is not strongly correlated with any of our constituency controls or our political competition measure but createdsector-specific demands for delegation or engagement. If this caused politicians to maneuver their way on to committees andchange their behavior on these committees, membership would be correlated with the error term and we would have biasedcoefficients in our specifications. For such a bias to explain our results it would be necessary to be uncorrelated with politicalcompetition (otherwise, it’s effects would be absorbed by our competition variable) but would have opposing effects by level ofpolitical competition. The accountability mechanism would have to be unaffected by political competition but have differingeffects by it. On the other hand, if such an accountability mechanism was a key driver of our political competition measure,we would have to re-interpret our results as politicians facing different levels of political accountability rather than politicalcompetition.

38Since constituencies contain projects governed by multiple committees, we are comparing projects from the same sectoracross members and non-members, as well as within a constituency across projects governed by different committees, some ofwhich the constituency politician is a member of, and some of which she is not.

23

stituency; interaction is an interaction term between political competition and either the decentralization

dummy or our measure of political engagement; project controls (PCij) are key project characteristics;

constituency controls (CCij) are key congressperson characteristics and socio-economic characteristics

of the constituency; and, project type fixed effects (λj) absorb project type level effects.

The second set of specifications compares the outcomes of projects across the key variables of interest,

conditional on a host of project, politician and constituency controls. Our coefficients from specification

2 are intended to be interpreted in the light of our results from specification 1.

For clarity, to identify our parameters we make a conditional independence assumption with respect

to the delegation of our projects. We assume that our project controls sufficiently characterize the dele-

gatable aspects of the projects that remaining unobservables are not strongly correlated with decentral-

ization or our measures of competition or political engagement. Similarly, we assume our competition

term describes some equilibrium (though not necessarily exogenous) level.

5 Results

We split our results between the causes and consequences of political interference. We begin by assess-

ing how politicial competition affects the delegation of public goods to different tiers of government

and how politicians then engage with those tiers. This analysis provides us with a framework to un-

derstand how decentralization and engagement impact on project implementation at different levels of

political competition.

5.1 Causes

5.1.1 Delegation

Table 4 presents our baseline results on how politicians delegate to the two tiers of government un-

der study, following specification (1). The dependent variable in all specifications is a binary variable

reflecting whether a project is decentralized or not, which takes the value 1 when the project is im-

plemented by a decentralized organization. Thus, a positive coefficient implies a greater likelihood to

delegate a project to a decentralized organization. The baseline proportion of projects in our data that

are decentralized is 0.65.

Column 1 includes the ‘relevant committee’ dummy, project characteristics, and project type fixed ef-

fects only, and none of the other controls. The relevant committee dummy takes the value 1 when a

project from a particular sector is implemented in a constituency in which the politician is on the rele-

24

vant sectoral committee. For example, it would take the value 1 for a health project in the constituency

of a member of the health committee, and 0 for health projects in other constituencies.

The coefficient on committee membership is a precisely estimated zero. This implies that the mean

effect of committee membership on delegation is zero. Since we have motivated our analysis by the

differential effects of levels of political competition, this is not unexpected. Politicians’ electoral con-

cerns imply that politicians facing low and high levels of competition will make opposing delegation

decisions.

We test this idea in column 2 by including a measure of political competition and an interaction of

this measure with our indicator of committee membership. Our interaction variable is centered at zero

competition, so the coefficient on committee membership describes the behavior of committee members

faced by zero political competiton.

We can now see a distinctive pattern of sorting across tiers of government. The coefficient on committee

membership is significantly negative (γ1 < 0), implying that when competition is low in a constituency,

politicians centralize projects. However, the coefficient on the interaction term is significantly positive

(γ3 > 0), implying that when a politician faces strong competition in her constituency, she is more

prone to decentralization. Our coefficients on committee membership, political competition, and their

interaction are all significant at the 1% level.

These findings imply that the distribution of projects, and therefore resources, across government is

determined in part by political factors. The tendency to decentralize in response to pressures of political

competition seems to be of a similar magnitude to the tendency to centralize when faced with limited

political competition. A test of the null that the coefficient on committee membership, γ1, is of equal

and opposite magnitude to the sum of the coefficients on political competition and on the interaction

term, γ2 + γ3, has a p-value of 0.10. It is therefore on the margin of being rejected at the normal levels.

This fits with our finding of a zero coefficient in column 1.

Column 3 extends the specification in column 2 to include politician controls comprising a politician’s

sex, age, years of education and a binary variable that takes the value 1 if the politician has already

served a term in congress. We may believe that politician characteristics are important for the delegation

decisions they make. We find that they have little effect on the coefficients, although the magnitude

of the interaction is slightly strengthened. This implies that the pattern of delegation we observe is

relatively common across politicians with respect to the characteristics we condition on.

Column 4 of Table 4 then extends the specification in column 3 to include constituency controls. A key

concern in the analysis of delegation is the extent to which politicians are delegating to areas of greatest

need. It may be that political competition is correlated with deprivation and it is this that we are picking

up in our coefficients. To assuage this concern, we note that we are conditioning directly on the level

25

of political competition. However, it may be that constituency-level characteristics are correlated with

γ3 beyond this. We therefore include a range of constituency-level characteristics relating to poverty,

existing public goods investments, and recent economic dynamics constructed from a baseline survey

of households representative at the local government level.39

Column 4 controls for this battery of constituency-level controls. We see no change in our core parame-

ters.40 This indicates that our results are driven by political, rather than socio-economic, concerns. This

specification, which controls for project characteristics and project type fixed effects, politician charac-

teristics, and constituency characteristics, is our preferred specification. It provides us with a perspec-

tive on delegation by politicians that controls for many of the features we might think important in their

decision process.41

The implication of the results in Table 4 is that if you are a member of a standing committee and you

face zero political competition in your constituency, you are 11 percentage points more likely to central-

ize a project than a non-member. On the other hand, if you are a member of a committee and you face

significant political competition in your constituency, you are more likely to decentralize a project. The

coefficient on the interaction implies that a 1 percentage point increase in political competition leads to

a 0.16 percentage point increase in the probability of decentralization. Together, these coefficients imply

that members in marginal constituencies will decentralize 10% more projects than those in constituen-

cies with 100% of the vote. We observe the full distribution of competition levels in our data, and so

this is a valid comparison. For the projects we study, this implies a shift of resources across tiers of

government of US$160,000 per constituency.

One way to interpret this pattern is that when public good provision matters for a politician’s future

prospects, she delegates to the organization she knows is most effective. Conditional on project charac-

teristics and project type fixed effects, decentralized organizations complete 32% more of projects than39It is important to note the relative scarcity of public goods in Nigeria. It is a generally poor country with two-thirds of

households assessed as in poverty. Whilst our interest is the relative levels of need, there is a broad demand for public goodsacross the nation reflected in the household survey we use for this study. A reading of the OPEN evaluation reports implies thatcommunities almost universally demanded the projects we study.

40How do our control variables affect the delegation decision? Constituency controls are mixed in their impacts on delegation,and there is no systematic trend in how deprivation impacts on the decision. Similarly, politician controls do not seem to play asignificant role in determining the tier of government a politician delegates to. Whilst the age of the representatiove is significantlynegatively associated with decentralization at the 10% level, other characteristics are not. In contrast to these last two sets ofcontrols, the particular characteristics of a project are important predictors of delegation. For example, as would be expected,high local information needs increase the probability of decentralization. Conversely, high centralized information needs increasethe likelihood of centralization. Both of the coefficients on these variables are significant at the 1% level.

41The analysis in this paper focuses on the delegation and monitoring decisions of politicians, keeping constant the resourcesavailable to each constituency. We can test the validity of this assumption by undertaking a similar analysis as that undertakenin this section on the volume of resources a representative is able to secure for her constituency. Table A6 provides resultsof constituency-level regressions of total resources and number of projects similar to the analysis in our baseline specification,reflecting equation (1). Specifically, we regress the sum of project budgets in a constituency or the number of projects we observe,on the proportion of projects for which the congressperson is a member of the relevant committee, political competition, andtheir interaction, as well as project controls and fixed effects. We do this regression for both measures of competition althoughonly present those for our core measure as the results are qualitatively the same. We see that in both specifications there islittle evidence that a congressperson is able to secure greater resources for their constituency by being a member of a standingcommittee. Neither of the coefficients corresponding to γ1 or γ3 are significant at the usual levels. In contrast to Aghion et al.(2005, 2007, 2010), we find that members of the Nigerian House of Representatives have influence over delegation, rather thanresource control, when they are members of a committee. This may be due to the intense focus on an equal division of resourcesamong representatives in the Nigerian Congress. This analysis supports the approach taken in this paper.

26

centralized tiers.42 When the provision of public goods is of less importance because a politician is safe

in her seat, she delegates to centralized organizations over which she may have greater influence. We

shall turn to the impacts of this delegation in the next section, and investigate its effects over differing

levels of political competition.

The results in Table 4 also indicate a consistently negative coefficient on the levels variable of political

competition (γ2 < 0). This implies that political competition amongst non-members leads to the cen-

tralization of projects in their constituencies. The distribution of projects across tiers reflects a political

bargaining game that we are only partially able to observe. A potential explanation for this negative

coefficient is that committee members centralize projects they believe are of greater bargaining value

amongst their non-member colleagues.

To assess the robustness of our baseline results in column 4 of Table 4, we undertake a number of checks

of our preferred specification in Table A7. First, we may believe that decisions made by a politician on

a particular committee relating to her constituency are correlated. We are therefore motivated to cluster

our results at the sector-constituency level (or equivalently, the committee-politician level). Column 2

of Table A7 reruns our preferred specification clustering the standard errors at the sector-constituency

level. We see that the core coefficients are both still significant at the 5% level.

Now we turn to concerns that particular elements of the Congress are driving our results. In their

studies of standing committees in the US Congress, Aghion et al. (2005, 2007, 2010) and Cohen et al.

(2012) emphasize the importance of the chair to the decisions of a committee. They argue that a chair

has additional power to implement their political agenda through the committee. We can assess the

extent to which this is true in our context, as well as whether the chairs of our committees are driving

our results.

First, we can include a chair dummy, which takes the value 1 when a project is implemented in the

constituency of a chair of the relevant sectoral committee. We include such a dummy in column 3 of

Table A7. We can see that the chair does indeed use their additional power to centralize their projects

beyond the pattern of delegation we have already documented. However, our core results continue to

stand. We include our baseline specification in column 1 of Table A7 for reference. Comparing this to

column 3, we see that our core parameters do not change when we include a chair dummy.

We can also test for the impact of the chair by excluding projects from constituencies in which the

politician is chair of the relevant sectoral committee. Column 4 of Table A7 repeats our core specification

excluding all projects in which the constituency politician is chair of the committee corresponding to

that project’s sector. We can see that there is little change in our coefficients of interest.42A valid question is whether politicians could distinguish between the performance of the two tiers of government that we

study. In our data, the unconditional impact of decentralization is a 13% increase in project completion, a difference that issignificant at the 1% level. Thus, without having to make any adjustments for project characteristics, it is quite feasible thatpoliticians could identify the superior performance of decentralized organizations.

27

Next we address the concern that the ruling party of the National Assembly may have greater power to

delegate in sectoral committees than members from minority parties. We can again address this concern

by adding a dummy variable to our baseline specification. The ruling party dummy takes the value 1

when a project is implemented in a constituency governed by a politician from the ruling party. We add

such a dummy to our baseline specification in column 5 of Table A7. We see that in a similar fashion

to the chair, members of the ruling party use their additional powers to centralize projects. Being a

member of the ruling party makes a politician 4% more likely to centralize a project. Once again our

core results change little relative to the baseline specification.

Second, we rerun our baseline specification, but excluding projects implemented in constituencies gov-

erned by the ruling party. As column 6 of Table A7 shows, in this specification the pattern of delegation

is as in our baseline. Whilst the magnitude of the coefficients is larger, the standard errors are also larger,

possibly due to the reduction in sample size. This leaves the coefficient on the committee membership

variable significant at the 10% level and that on the interaction significant at the 5% level.

Having observed the significance of different aspects of political power in our previous robustness

checks, we may be keen to assess whether our results are driven by the most powerful politicians in the

House of Representatives. Following the political economy literature for the US, and echoing similar

work for Nigeria, an often used proxy for the relative power of a congressperson is the number of terms

she has served. Given that democracy was instated in Nigeria in 1999, we can observe at most two-

term representatives in the National Assembly in 2006/7.43 We therefore define a dummy for projects

implemented in constituencies of two-term representatives.

Adding this dummy to our baseline specification in column 7 of Table A7, we see that being a two-term

politician does not impact on the pattern of delegation. The coefficient on this variable is a precisely es-

timated zero. We can also exclude projects in constituencies of two-term politicians, as we do in column

8 of Table A7. We see the coefficients on our core variables are similar to the baseline specification.

Finally, we return to the discrepancy identified in our analysis of the determinants of committee mem-

bership. We saw in section 3.1 that constituency-level characteristics had limited impact on committee

selection for all sectors bar agriculture. However, there seemed to be indications of selection for the

agriculture committee. We therefore exclude from our analysis agriculture projects in all constituencies

where a politician does not have relevant qualifications or experience in the agriculture sector. The re-

sults of this regression are reported in column 9 of Table A7. We see that our core coefficients continue

to be significant at the 1% level.44

43Nigerian congresspersons serve 4 year terms and there are no term limits on the number of times a politician can serve.44We may also be concerned that our results are peculiar to the specific measure of political competition that we have utilized.

As described in section 3, I have created a secondary measure of political competition used in Besley and Case (2003) and Besleyet al. (2010), the deviation of the winning vote share from 0.5. The correlation between this measure and our standard measureof political competition is 0.94 implying a limited robustness check. However, it is worth emphasising that the two measures

28

5.1.2 Engagement

We now turn to the impact of political competition on the level of engagement between politicians and

the bureaucrats to whom they delegate. Table 5 estimates specifications similar to those in Table 4,

but with the proportion of projects on which politicians directly engage bureaucrats as the dependant

variable. These continuous proportions vary between 0 and 1 and are organization-level averages of

responses from a representative sample of organization staff. We are therefore estimating the likeli-

hood of a project being implemented by an organization with high levels of politician engagement. On

average across projects, politicians personally engage with bureaucrats on 14% of projects.45

The dependant variable was constructed from interviews with the organizations that implemented the

projects under study. As described above, for logistical reasons we could not survey two of these orga-

nizations. Together these two organizations implemented 736 projects. This leaves us with 3757 projects

implemented at 55 organizations.46 To ensure that our baseline results on delegation continue to hold,

we re-estimate our baseline specification described in column 4 of Table 4 for this restricted sample.

The results are provided in column 10 of Table A7. Our parameters are substantively the same as in our

baseline specification and still significant at the 1% level. The pattern of delegation we observe in the

full sample holds in this restricted sample.

Column 1 of Table 5 estimates the impact of a politician being a member of the relevant committee on

project engagement, conditional on project characteristics and project type fixed effects. We see that

simply being a member of the committee increases the likelihood of a politician directly engaging with

a bureaucrat on a project in her constituency. The coefficient is significant at the 10% level.

However, our interest is whether engagement varies with the degree of political competition that a

politician faces. In column 2, we introduce our measure of political competition and an interaction of

this measure with an indicator of committee membership. Our interaction variable is centered at zero

competition, so the coefficient on committee membership describes the behavior of committee members

faced by zero political competiton.

We now see a pattern of engagement across projects that echoes our findings on delegation. The co-

efficient on committee membership is negative (γ1 < 0), implying that when competition is low in a

test different conceptions of competition, with the deviation measure assessing the threat from the coalition of competitors in aconstituency. We re-estimate the results in table 4 using this alternative measure of political competition. Although the coeffi-cients differ in size, they are identical in sign and significance within the usual categories. Thus, using two conceptually distinctmeasures of political competition, we find very similar patterns of delegation.

45Survey data on politician engagement with bureaucrats around project implementation is rare. The closest numbers we havefor comparison to this paper come from the World Bank’s ‘Public Officials’ Surveys’ (Manning et al, 2000). These surveys covered16 countries, and there is significant heterogeneity in the average level of bureaucrat’s perception of politicians’ interference.Evidence from Bangladesh shows that when politicians stop interfering in day-to-day decisions, the perception of corruptionin that organization will fall by 31% (Mukerjee et al., 2001). A similar impact of political interference was found for Bolivia(Manning et al., 2000). In Guyana, 42% of officials interviewed stated that politicians’ interference is frequently or very frequentlya “significant problem” (Gokcekus et al., 2001). Thus, whilst these figures reflect varying contexts and survey questions, they areindicative that the magnitude of the issues studied here are of widespread significance.

46The 736 projects that we drop are a little different to the 3757 projects that we keep. They have slightly higher budgets andare 4% more complex.

29

constituency, politicians are less likely to engage with bureaucrats on project implementation. However,

the coefficient is only significant at the 10% level.

The coefficient on the measure of political competition is positive (γ2 < 0), indicating that at higher

levels of political competition, non-members are more likely to engage with project implementation.

We find similar result for members of standing committees. The coefficient on the interaction term is

positive (γ3 > 0), implying that when a politician faces strong competition in her constituency, she is

more prone to engaging with bureaucrats on projects. The coefficient is significant at the 5% level.

Comparing among committee members, we can see that a politician in a marginal constituency is 3

percentage points more likely to engage a bureaucrat on a project than a politician who wins 100% of

the vote. Since the mean level of engagement is 14% of projects, the coefficients here imply that political

competition can explain 20% of the degree of politician’s engagement with the bureaucracy.

Comparing between committee members and non-members, we see that the magnitude of the coeffi-

cients is relatively small. A committee member faced by zero competition is only 0.6 percentage points

less likely to engage a bureaucrat on a particular project than a non-member. It is likely that in our com-

parison of delegation, the additional power of a standing committee member to delegate is far higher

than that of a non-member. A member’s additional power of engagement with bureaucrats, however,

is likely to be much smaller. Whilst there is likely to be some advantage to committee members in their

capacity to engage the bureaucracy, and it is this which is driving our results, the effect is likely to be

small relative to that of delegation. Any comparison across Tables 4 and 5 should therefore condition

on the different power differentials on which they are estimated.

Column 3 of Table 5 extends the specification in column 2 to include politician controls. As before, we

may believe that politicians with different characteristics make different choices, here with respect to the

degree of engagement of bureaucrats on projects in their constituency. As in table 4, we find that they

have little effect on the coefficients, although the magnitude of the interaction is slightly strengthened.

This implies that the pattern of engagement we observe is relatively common across politicians with

respect to the characteristics on which we condition.

Column 4 extends the specification in column 3 to include constituency controls. Areas with differing

needs may require differing degrees of politicians’ attention to execute a project to the same quality. It

may be this that is driving the pattern of delegation we observe. The constituency controls used in our

analysis of delegation seem a natural fit for this purpose. We see that our coefficients do not change sign

but are smaller and lose some of their significance. The membership variable becomes insignificant at

the usual levels, whilst the impact of political competition and our interaction terms are significant at

the 5% and 10% levels respectively.47

47As with the delegation decision, we may be interested in how control variables impact on the level of engagement. Once

30

Together, the results in Table 5 indicate that political competition also affects a politician’s engagement

with bureaucrats on public projects. Taking politicians as motivated by electoral concerns, a natural

interpretation of these findings is that in marginal constituencies, politicians are incentivized to engage

bureaucrats and ensure that projects are delivered. Conversely, where they face little competition, such

effort is wasteful. Whether this interpretation finds support in project implementation data will be

assessed in the next section.48

Before that, we turn to testing the robustness of our results on engagement. Taking a similar approach

to Table A7, Table A8 provides specifications that assess potential drivers of our baseline results.49

Column 1 repeats the preferred specification of Table 5 for comparison. In column 2, we check whether

our results are robust to clustering at the sector-constituency level, and see that they are not. The smaller

differential in power between members and non-members is likely to make the results more sensitize

to our assumptions on the standard errors.

Column 2 includes a dummy that indicates whether the project is in the constituency of the relevant

chair. Chairs may have additional authority to engage bureaucrats that other members may not, and it

may be important to understand their role in the patterns we observe. The coefficient on the dummy is

not significant, although when we exclude projects in chairs’ constituencies in column 4, the coefficients

of interest both reduce in size.

Members of the ruling party may be less constrained in personally engaging with the bureaucracy, so

we include a dummy variable to our baseline specification that takes the value 1 when a project is

implemented in a constituency governed by a politician from the ruling party. Column 5 of Table A8

shows that the coefficient on the dummy is negative, and significant at the 1% level. A member of

the ruling party is 1.2% less likely to engage with bureaucrats on a project. This may be because the

executive is more likely to pander to ruling party members’ interests in general, and so they have to put

less effort into direct engagement. This interpretation is bolstered by excluding projects implemented in

constituencies governed by the ruling party. As column 6 of Table A8 shows, the coefficients are higher

for non-ruling party constituencies.

again, constituency level variables have significant impacts on the level of engagement, but do not provide a clear direction ofthe link between deprivation and engagement. Politician characteristics are more significant for engagement than for delegation.An additional year of education increases the level of engagement by a politician by 0.2%, with the corresponding coefficientsignificant at the 1% level. Being a two-term politician reduces engagement by 0.4%. Finally, project characteristics play a signif-icant role in determining the level of political engagement on a project. Across all three measures of complexity, higher levels ofcomplexity induce higher levels of political engagement, with the corresponding coefficients all significant at the 1% level.

48One could take the perspective that decentralized organizations naturally require greater engagement than centralized orga-nizations. This goes against the descriptive statistics provided in Table 3, where centralized organizations have higher engage-ment with politicians across both margins of engagement. However, under that perspective we cannot separately identify theimpact of political competition increasing the extent of delegation and the driver of increased engagement with decentralizedorganizations. It may be that having delegated projects to decentralized organizations, politicians then automatically increasetheir engagement of those projects.

49The discussion here has taken a bureaucrat as a relatively homogenous agent. However, we can assess our results by lookingat engagement with managers and non-managers separately. We find that the results for these two groups are roughly equivalent,implying that politicians are engaging officials throughout the organizations we study. This is in accordance with the qualitativeinterviews we did with officials in conjunction with the quantitative surveys.

31

Politicians with experience in the National Assembly may better understand how to work with the

bureaucracy and affect the change they desire. We therefore would like to test whether our results

are driven by those politicians who have more experience. We therefore define a dummy for projects

implemented in constituencies of two-term representatives. Adding this dummy to our baseline spec-

ification in column 7 of Table A8, we see that being a two-term politician is also associated with lower

levels of engagement with the bureaucracy but the size of the effect is only 0.04%. Excluding projects

in constituencies of two-term politicians, as we do in column 8 of Table A8, we see a similar pattern of

coefficients to our baseline specification.

Finally, we return again to the discrepancy identified in our analysis of the determinants of committee

membership. We saw in section 3.1 that constituency-level characteristics had limited impact on com-

mittee selection for all sectors bar agriculture. However, there seemed to be indications of selection for

the agriculture committee. We therefore exclude from our analysis agriculture projects in all constituen-

cies where a politician does not have relevant qualifications or experience in the agriculture sector. The

results of this regression are reported in column 9 of Table A8 and the results are similar to our baseline

specification.50

5.2 Consequences

Section 5.1 provides evidence that the distribution of public projects across tiers of government, and the

political pressures faced by bureaucrats implementing them, is a function of the political competition

faced by politicians. We interpret the patterns of delegation observed at the margin for committee

members to be representative of wider trends in political delegation in the House. We now turn to the

effects this behavior has on project implementation.

We can assess the consequences of this politically-motivated interference in bureaucratic function due to

the information I have gathered on the effectiveness of project implementation. Our analysis continues

to be at the project level, and our dependant variable is now the proportion of project completion, a

continuous variable that takes values between 0 and 1. A project that never started has a completion

level of 1, one that was half-way completed of 0.5, and one that is fully completed of 1.51 The mean

level of completion in our data is 0.52.52

We will build up to the full estimation of (2) in a series of steps. This allows us to assess the impacts of

50In these sections, I have analyzed the delegation and monitoring decisions separately. We could also take the perspective thatthe decisions are correlated. To check the robustness of our results to this possibility, I estimated a seemingly unrelated regressionmodel of our baseline model with dependant variables of decentralization and political engagement. The coefficients on our coreparameters estimated in a SUR system are very close, and at similar levels of statistical significance, to those estimated separately.

51The utility of a project at various degrees of completion will vary across project types. A borehole is useless until virtuallycomplete, whilst a primary health centre without a compound wall or any other external features provides a building for care.We take the perspective that on average an increase in the proportion completed or in quality increases citizen utility.

52The distribution of project completion levels has significant density at 0 and 1, with roughly a third of the distribution atintermediate levels. The 25th and 75th percentiles are 0 and 1 respectively.

32

political competition whilst providing a coherent framework for us to extend our analysis to investigate

the role of delegation and direct politician-bureaucrat engagements in the delivery of public goods.

We motivate our analysis by looking at the direct impact of political competition on the delivery of

public goods. Column 1 of Table 6 estimates a regression of project completion rates on the level of

political competition in the constituency in which the project is implemented. In all specifications we

include project, politician and constituency controls and project type fixed effects as in our baseline

specifications. We stick to the sub-sample of 3757 projects for which we have engagement information.53

We see that without controls for delegation or engagement, political competition has a positive impact

on project completion rates that is significant at the 5% level.

The implication is that political competition is good for bureaucratic performance. A 1 percentage point

increase in political competition leads to a 0.07 percentage point increase in completion rates. Since the

mean completion rate is 52%, a marginal constituency will have 13% higher completion rates than one

in which the politician faces zero political competition.

We can compare the finding of this specification to the existing literature on the impacts of political

competition on government performance. Besley et al (2010) show how a shift in political competition of

30% in the southern states of the US during the 20th century led to an increase in a state’s infrastructure

spending of roughly 10%. De Janvry et al (2011) show that Brazilian mayors improve the performance

of a conditional cash transfer scheme by 36% when they face re-election incentives. Similarly, Ferraz

and Finan (2011) find that re-election incentives push Brazilian mayors to reduce misappropriation of

resources in their municipality by 27% compared to those without such incentives. The question these

studies leave unanswered is precisely how politicians facing political competition facilitate improved

executive performance.

In column 2 of Table 6, we introduce the margin of delegation studied here. To the specification in

column 1 we add a binary variable that represents the degree of decentralization of the organization

implementing the project, which takes the value 1 when the project is implemented by a decentralized

organization. The coefficient on political competition continues to be positive, but is no longer sig-

nificant at the usual levels. The coefficient on the decentralization dummy is positive, implying that

decentralization increases project completion rates. It is large, explaining 31% of project completion

rates, and significant, at the 1% level.

These numbers imply that for a project of a particular type, and of a given budget and complexity,

the impact of being implemented by a decentralized organization rather than a centralized one is an

increase in the completion rate of 31 percentage points. Given the detail of our data set, this figure pro-

vides relatively novel micro-level evidence on the positive impacts of decentralization. We can there-53The results of the first two specifications in Table 6, for which we have sufficient data for all projects, are almost identical

whether we utilize the full sample or the sub-sample.

33

fore compare this figure to the existing literature on decentralization. Fisman and Gatti (2002) study

decentralization across countries, defining it as the ratio of local government expenditures over total

government expenditures. They find that a one standard deviation in their measure of decentraliza-

tion leads to a 30% reduction in corruption. Galasso and Ravallion (2013) study a Bangladeshi transfer

program and find that the maximum targeting differential between tiers of government is 20%. Both of

these figures are of the same order magnitude as the impacts we find here.

In column 3 of Table 6, we introduce our measure of political engagement, the proportion of projects at

an organization in which a member of the National Assembly directly engages with staff. The coefficient

on this variable is negative, and significant at the 1% level. The coefficient is large, at -0.94, but the mean

level of engagement is only 0.14. Thus, the average impact of political engagement with bureaucrats

implementing a project is a reduction in that project’s completion rate of 13%. However, the range of our

engagement measure extends to 27% of projects, implying an average reduction in project completion

of one quarter in those organizations at the top of the range of politician engagement.

Having looked at the levels effects of each of our core areas of interest, we now turn to how the impacts

of delegation and engagement vary with political competition. I postulated above that politicians were

delegating to more effective organizations when they faced political competition. That decentralized

organizations are more effective can be seen by our results in columns 2 and 3 of Table 6. However,

it is still to be seen whether decentralized organizations are even more effective for projects in political

constituencies with a lot of competition.

Column 4 of Table 6 adds an interaction between our measure of political competition and the decen-

tralization dummy to the specification in column 3. This is a version of (2) where the interaction is

between political competition and decentralization. In this specification, θ1, the coefficient on politi-

cal competition, estimates the impact of political competition for projects implemented by centralized

organizations. It is insignificant at the usual levels and under half the magnitude of the coefficient in

column 1. Political competition is therefore not an effective means of increasing project completion by

centralized organizations.

The coefficient on decentralization, θ2, estimates the impact of decentralization in constituencies with

little political competition. As in the other specifications of Table 6, it is strongly and significantly

positive, implying that decentralization is an effective means of improving project completion in un-

competitive constituencies.

The coefficient on the interaction of decentralization and engagement, θ4, indicates the additional im-

pact of decentralization in politically competitive constituencies beyond θ2. It is insignificant at the

usual levels, implying that decentralization is as effective in politically competitive constituencies as

in politically uncompetitive constituencies. The interaction and θ1 together imply that political com-

34

petition itself is not significant for project completion when we control for delegation. A potential

interpretation of the results in column 4 is that political competition drives politicians to improve their

delegation decisions, and once this is accounted for, competition itself is no longer a direct driver of

completion rates.

Together with the results from section 5.1.1, the findings from table 6 indicate that when politicians

delegate to decentralized organizations, they are gaining a greater degree of public good provision

regardless of the political competition they face. We saw that they then engaged differently across

organizations depending on the level of political competition they face. I interpreted the increased en-

gagement of politicians in competitive constituencies as a means to secure higher levels of public good

provision that would boost an incumbent politicians election chances. The fact that we find politician

engagement to be negatively correlated with project completion rates goes against this theory. How-

ever, we are yet to see whether the impacts of politician engagement are different for uncompetitive

and competitive constituencies.

Column 5 of Table 6 adds an interaction between our measure of political competition and the propor-

tion of projects with engagement of a politician to the specification in column 3. This is a version of

(2) where the interaction is between political competition and politician engagement. The coefficient

on the proportion of projects with engagement of politician variable, θ3, now represents the impact of

politician engagement on projects implemented in constituencies with low levels of competition. It is

negative, large in magnitude, and significant at the 1% level. It implies that a 1 percentage point increase

in politician engagement leads to a 1.55 percentage point decrease in project completion. Interpreting

this at the mean of politician engagement, on average politicians in low-competition constituencies re-

tard public good completion rates by 22%. At the upper bound of politician engagement, this coefficient

implies politicians retard completion rates by 42%.

On the flip side, political competition constrains this influence. In this setting, the coefficient on the

interaction term, θ4, describes the additional impact of political engagement in politically competitive

constituencies. The coefficient is positive, and significant at the 5% level. It implies that political com-

petition constrains the negative influences of politician engagement with bureaucrats. A stylized inter-

pretation is that political competition constrains the excesses of politician behavior analogously to how

market competition constrains the excesses of firm activity.54

The total effect of political engagement for projects implemented in settings of high levels of poltical

competition is θ3 + θ4. A null hypothesis of these being of equal but opposite value is rejected at the 1%

level. Political engagement has strong negative effects on the implementation of public goods across

54The null hypothesis that the coefficients on political competition, θ1, and the interaction, θ4, are of equal and opposite magni-tude is rejected at the 1% level.

35

the political spectrum, but these effects are constrained by political competition.55

We have focused the analysis in this section on project completion rates, as we have information on the

level of project completion for all the projects we study. We can also assess the impact of political com-

petition on the quality of projects.56 The drawback of using this outcome measure is that information

on project quality is only available for roughly half the projects for which we have project completion

data, originating in 45 civil service organizations. For this subset of projects, we re-estimate Table 6 us-

ing as the dependant variable a binary indicator of satisfactory quality of project implementation. The

dummy takes the value 1 when the project was rated by the evaluators as having satisfactory quality or

above.

Table A9 presents the results of a set of regressions following equation (2) where yijscn is a binary in-

dicator of project quality. Column 1 shows that political competition does not have significant effects

on project quality. Column 2 indicates that decentralization has at best weak impacts on project qual-

ity (with decentralization significant at the 10% level in the remaining specifications in the table). The

coefficient on politician engagement, θ3, is negative but insignificant at the usual levels.

Column 5 of Table A9 tests whether politician engagement has differing effects by levels of political

engagement. It adds an interaction between political competition and politician engagement. This is

a version of (2) where the interaction is between political competition and politician engagement. The

coefficient on the proportion of projects with engagement of politicians variable, θ3, now represents the

impact of politician engagement on projects implemented in constituencies with low levels of compe-

tition. It is negative, large in magnitude, and significant at the 1% level. It implies that a 1 percentage

point increase in politician engagement leads to a 2.52 percentage point decrease in the probability of

satisfactory quality of project implementation. Interpreting this at the mean of politician engagement,

on average politicians in low-competition constituencies reduce the probability of project quality being

satisfactory by 35%. At the upper bound of politician engagement, this coefficient implies that engage-

ment reduces the probability of project quality being satisfactory by 69%.

As we saw for project completion rates, political competition constrains this negative impact. The coef-

ficient on the interaction term, θ4, describes the impact of political engagement in politically competitive

constituencies. It is positive, and significant at the 1% level. The total effect of political engagement for

projects implemented in settings of high levels of poltical competition is θ1 + θ4. A null hypothesis of55Our analysis in section 5.1 was based on comparisons of the actions of members of standing committees in Nigeria’s National

Assembly with the actions of non-members. We can extend the analysis outlined here to investigate whether committee membersact in a distinct way to other politicians. If we estimate a triple-way interaction based on the analysis described here that takesinto account committee membership, we find no evidence of heterogeneity in effects by membership status.

56Looking at the margins of project completion and quality separately do not take into account that some projects may be ofhigh quality but little completed, and others may have high completion rates but be of low quality. To confront this concern, wecan combine our measures into a ‘quality-adjusted completion rate’, which is the product of the proportion of project completedand the binary indicator of satisfactory project quality. This measure deletes any project, however far completed, that is notof satisfactory quality. I have re-estimated the results of Table 6 using this measure and qualitatively we see identical results,although their significance is strengthened. This implies that politicians facing political competition are not trading off qualityfor quantity or vice-versa.

36

these being of equal but opposite value has a p-value of 0.61, implying the forces of low- and high-

political competition have similar impacts. Again, evidence from quality data indicates that political

competition restrains the negative influences of politician engagement on project implementation.57

The results in Tables 6 and A8 provide evidence that political competition incentivizes politicians to

make choices that improve project implementation. A less robust but intriguing result can also be dis-

cerned from the tables. In the specification reported in column 5 of Table 6, θ1, the coefficient on political

competition, estimates the impact of political competition in zero-engagement settings. It is insignifi-

cant at the usual levels but is now negative. This hints at the potential that whilst political competition

is good for project implementation when managed by politicians, it may be negative without political

engagement. Whilst we have no significant evidence for this in Table 6, in table A9 we do. There, θ1, the

coefficient on political competition, estimates the impact of political competition on project quality in

zero-engagement settings. It is negative and significant at the 1% level. Together, these findings indicate

that political competition may have negative effects on project completion and quality at low levels of

politician engagement. For political competition to have positive effects, we require politicians to man-

age the process of implementation. One potential explanation is that in areas of political competition

and low politician engagement, bureaucrats may find little constraints to their own corrupt activities

but more opportunities due to the lack of a dominant political owner of the project.

A final message from our results relates to the mechanisms through which political competition im-

pacts on bureaucratic function. Evidence from Table 6 implies that without controlling for mechanisms

through which politicians respond to political competition, political competition is significantly asso-

ciated with project implementation at the 5% level. However, evidence from tables 6 and A8 imply

that once we condition on the mechanisms of bureaucratic interference we study in this paper, that

correlation becomes insignificantly different from zero, or perhaps even negative. One interpretation

of the change in significance and sign of the coefficient on political competition is that the channel for

positive effects of political competition on constituency outcomes works through improved functioning

of government. That would explain the results described here, as well as those presented in Besley et

al. (2010) and Ferraz and Finan (2011). To improve the delivery of basic public goods may be as much

about changing the nature of political interaction with the bureaucracy as changing the bureaucracy

itself.57We motivated our analysis by utilising membership of committees. We can extend the analysis outlined in this section

to investigate whether committee members act in a distinct way to other politicians. If we estimate a triple-way interactionbetween decentralization, politician engagement, and committee membership, we find no evidence of heterogeneity in effects bymembership status for either project completion or quality.

37

6 Discussion and conclusions

This paper assesses the influence of politicians on the bureaucratic implementation of local public sector

projects. Using data from across the Federal Government of Nigeria, we are able to assess how political

competition affects the decisions made by politicians as to which bureaucrats to delegate to and the

degree to which to personally engage those bureaucrats. Using independent evaluations of project

implementation, we assess the subsequent consequences of these decisions on project completion rates

and quality.

This provides us with a window into the public sector and the underlying determinants of public sec-

tor productivity. Studying delegation provides a better understanding of the distribution of resources

across government. Studying engagement allows us to understand the political environments in which

different bureaucrats work. Together, our results explain a significant proportion of government out-

puts, critical to citizen welfare, particularly in the developing world.

We find that political competition incentivizes politicians to delegate to more effective organizations

than they would do if they faced limited competition. This provides quantitative evidence that the dis-

tribution of resources across government is therefore a function of the political incentives of politicians.

The sub-set of politicians chosen to oversee a sector by being selected for the relevant sectoral commit-

tee is chosen for macro-political reasons and their qualifications. Thus, within a sector, resources are

skewed by the political interests of a relatively ad hoc group of politicians.

Political competition also incentivizes politicians to engage more intensively with the bureaucrats they

have delegated to. Therefore, the bureaucratic environment implementing a project is disturbed by

politician engagement differentially across projects and organizations. Typical evaluations of public

sector organizations ignore the potential for the differential political constraints they face.

In the qualitative survey that I undertook to complement the quantitative work described here, bu-

reaucrats stressed that the most significant barrier to effective delivery of public projects was political

interference (Federal Government of Nigeria, 2011). Similar claims have been made by bureaucrats in

many other countries (Manning et al, 2000). We find that politician engagement with bureaucrats has

negative effects on project completion and project quality. This is particularly stark in constituencies

with little political competition. There, we find that politicians engagement can reduce project comple-

tion rates by up to 42%, and reduce the probability that project quality is of satisfactory quality by up

to 69%.

Our most significant result is that political competition constrains the excesses of politician behavior

analogously to how market competition constrains the excesses of firm activity. The ill effects of politi-

cian engagement are significantly reduced when the politician faces significant competition for her

38

seat. I argue that the mechanisms of delegation and politician engagement studied here may play an

important role in understanding why political competition has positive effects on bureaucratic imple-

mentation of public projects. Once we condition on these mechanisms, the coefficient on politician

competition becomes insignificant.

This finding highlights the contribution of this paper. A growing literature identifies the positive im-

pacts of political competition on government performance (Besley et al., 2010; Ferraz and Finan, 2011).

However, the mechanisms through which political competition affects bureaucratic implementation are

little understood. This paper provides micro-level evidence of the changes in politician behavior that

are induced by political competition and the subsequent impacts on their interactions with bureaucrats.

My identification strategy is grounded in the claim that a congressional procedure exogenously allo-

cates power to a subset of politicians for each of the social sectors. This paper therefore adds to the

political economy literature that investigates the behavior of politicians on standing committees, and

extends recent work by Aghion et al (2005, 2007, 2010) and Cohen et al. (2012) to the developing world.

In contrast to their findings, we do not find evidence that politicians are able to secure higher levels

of resources for their constituency as members of a standing committee. This may be due to the sig-

nificant expectation of equal resource distributions across constituencies that characterizes Nigerian

political bargaining. However, we do find that politicians utilize their additional power to delegate to

tiers of government that better suit their political interests, and engage them differentially depending

on the political competition they face in their constituency. By exploring standing committees in a dif-

ferent context from the US Congress, this paper adds variation in the rules we use to study politicians

bargaining over public resources.

The political determinants of delegation imply that we must be careful in assessing the impacts of

decentralization, the margin of delegation studied here, as purely a function of decentralized organiza-

tions superior performance. The results of this paper imply that decentralized organizations function

in different political environments to centralized ones. However, this paper also provides micro-level

evidence, conditional on project and constituency characteristics, that decentralized organizations are

more effective than centralized ones in both politically competitive and uncompetitive constituencies.

In the majority of specifications we estimate, decentralized organizations have completion rates that are

roughly a third higher than centralized organizations.

Understanding the determinants of the resource distribution across government and the drivers of

project completion and quality is important for our ability to improve government effectiveness. This

is especially true in the developing world, where government organizations face other constraints on

their productivity, such as inadequate resources or limited citizen accountability. The findings of this

paper do hint at the welfare benefits to higher levels of political competition as public resources are

39

used more effectively.

However, there is some evidence that political competition on its own, without management by a prop-

erly incentivized politician, may have negative consequences for public sector productivity. Such a

finding might link to the need for effective political institutions to manage the relationship between

citizens and bureaucrats, as described by Acemoglu and Robinson (2006, 2012).

Many questions remain. The most obvious is whether, if politicians are the drivers of policy change,

they would design institutions that would constrain their interactions with the bureaucracy towards im-

proved public sector performance (Robinson, 2003). Where that has happened, politicians have played

an important role in improved bureaucratic performance (Robinson et al, 2003). Where it has not, they

have undermined the capacity of the public sector to serve the people they have been elected to repre-

sent (Besley and Persson, 2009).58

Data Appendix

A.1 Project selection

The Overview of Public Expenditure in NEEDS (OPEN) monitoring and evaluation process forms the

basis of the sample of projects selected for this study. The OPEN evaluation was set up in 2006 and

“was adopted as the mechanism to monitor and evaluate public expenditure” (Federal Government of

Nigeria, 2009). The scheme intended to monitor the implementation of projects to be funded by debt

relief savings and evaluate their outcomes. The evaluation reports from the first two rounds of this

process act as the basis for our data set.

The President created an ‘OPEN office’ with a Presidential mandate to track and report on the expen-

diture of the debt relief gains.59 Rather than set up a parallel organisation to spend debt relief savings,

as had been done elsewhere, OPEN was seen as an opportunity to “find out where the most signifi-

cant barriers to public expenditure lay” (Federal Government of Nigeria, 2007). Thus, it was decided

to channel the funds through standard institutions of government - the ministries, departments, and58What are the policy ramifications of the results presented here? Khemani (2003) documents how constitutional rules can play

a role in limiting the extent to which politics can effect public resources in India. Robinson (2003) presents options for ‘politician-proofing’ policy. However, the discussion here indicates that we require policies that manage the relationship between politiciansand bureaucrats, mitigating the negative effects and supporting the positive effects. This is particularly true in a developingcountry contexts where bureaucratic corruption and inefficiency may be reduced by political pressures. One potential avenue isto link politicians more closely with the public goods provided in their constituencies. For example, Banerjee et al (2010) assessthe provision of newspaper reports on legislator performance in India. They find that access to such reports increases turnout,reduces cash-based vote buying, and increases electoral gains for better performing incumbents. Another route is to design robustmonitoring systems that manage the behavior of political elites. Olken (2007) describes how Indonesian bureaucrats are moresuccessful than citizens in keeping a check on village elites managing the provision of public road projects. Defining a coherentsystem of incentives across politicians, bureaucrats and citizens is at the heart of effective public good provision.

59The ‘OPEN office’ was otherwise known as the ‘Office of the Senior Special Assistant to the President on Millennium Develop-ment Goals’. I was closely associated to the OPEN office, and the OPEN process, having been sent by the Overseas DevelopmentInstitute to work as an economist in the OPEN office for the period 2005-7. For more information about the scheme that placedme there, see www.odi.org.uk/fellows.

40

agencies of the Federal Government. This enables us to use the OPEN evaluation as a window into the

workings of Nigeria’s government.

As background, it is worth understanding a little about the context in which the OPEN initiative was

started. In 1999, Nigeria transited to a democratic government under President Olusegun Obasanjo

after over a decade and a half of military dictatorship. The new administration inherited a huge external

debt portfolio.60 Based on the thrust of the government’s reform agenda, the Paris Club granted Nigeria

debt relief to the tune of US$18billion in September 2005. This translated to annual debt-service savings

of roughly US$1billion, US$750 million of which would accrue to the Federal Government. The OPEN

evaluation reports evaluate the effectiveness of the federal portion of these savings.

The President directed that debt relief expenditures go to “core projects and programs in the social

sector” (Federal Government of Nigeria, 2007). The OPEN office helped direct funds to a relatively

representative sample of the nation’s small-scale social-sector projects. All were suppose to take roughly

12 months to complete. This implies they are not representative of the entire budget, which includes

much recurrent expenditure (salaries, materials and supplies, and so on) and the funding of large scale

dams, oil refineries and so on. However, they are representative of social-sector capital expenditures.

I hand-coded the information from the 21,000 documents and project files that made up the monitor-

ing and evaluation report of the OPEN initiative for 2006/7. This makes up our set of representative

evaluations of Federal Government public projects.

From this representative set, I defined a dummy variable that indicated whether a project was purely

national in nature, or could be implemented by organizations at different tiers of government. Our

analysis focuses on projects that are delegatable, rather than being specialized so that they could only

be implemented at a single tier of government. The projects excluded are those projects whose scope is

national or multi-jurisdictional (across many states for example). This was relatively clear cut in almost

all cases.

The guidelines for excluding projects on the basis that they were not able to be decentralized were

the following. A project cannot be decentralized if: (i) it contains components that require access to

international policy inputs; (ii) it contains components that require engagement with stakeholders at

the national or international level; or, (iii) the scope of the project crosses multiple jurisdictions beyond

the mandate of any single decentralized organization.

An example of a project that was not delegatable or decentralizable is from the Ministry of Women

Affairs. It was entered into the budget as ‘Development and production of 2000 copies of a National

Gender Policy (NGP) and 2000 copies of its Strategic Implementation Framework for the sustenance of

60Nigeria’s Debt Management Office estimated that the nation owed external creditors US$36billion at the end of 2004, whichwas roughly twice the value of annual government expenditures (Debt Management Office, 2006).

41

gender equality perspective in all sectors’. This project requires international-level technical assistance,

inputs from multiple sectors, and national-level engagement with international donors and Nigerian

stakeholders. Thus, given the constraints on national and international engagements in the Public Ser-

vice Rules, it is infeasible that this would be implemented by any organization but the Ministry of

Women Affairs.

On the other hand, a borehole to be provided in Borno state could equally be implemented by the

Ministry of Water Resources or the Chad Basin River Basin Development Authority. Such a project is

specific to a single jurisdiction, has technical specifications that can be handled by either organization,

and is permissible for implementation at either tier of government in the Public Service Rules.

A.2 Politician characteristics

To effectively characterize the politicians in our data, I constructed biographies of each of the 360 rep-

resentatives in the 5th (2003-7) National Assembly. I followed the following process.

I drew the list of winning politicians from the Independent National Electoral Commission’s ‘Com-

pendium of Results of the 2003 General Elections: Vol.1: Presidential and National Assembly Elections’.

Where there were electoral tribunals, I followed the judicial process for each and noted where there was

a change of representative and the date on which the successful petitioner took up office. I also utilized

National Assembly web site records to identify any deaths amongst the congresspersons. This defined

a complete set of representatives relevant to the 2006 and 2007 budget processes.

For each politician, I then hand-coded basic demographic information (sex, age and education) from

the National Assembly’s ‘Nigeria Legislature 1861-2011: Compendium of Members and Officials’. In

the very small number of places where age was missing, I either confirmed this using other sources or

replaced their age with the mean of all representatives and include a dummy indicating that age was

missing in all specifications that include representative age.

I then aimed to build up a profile of the career of each representative and code their relevant experience

into the sectors of the standing committees in our data. These represent all the major social sectors.

Only substantive experience in a sector over a sustained period of time was counted as relevant expe-

rience. To give a brief overview of how I sorted candidates into sectors, those with a training in finance

or who had been a financial officer at a large private firm or public organization were coded as having

experience in finance. Doctors or other health professionals such as nurses, pharmacists, or affiliates of

medical institutions were coded as having experience in health. Electrical or other relevant engineers

and those involved in contracting power facilities were coded as having experience in the power sector.

Mechanical engineers or those with experience in river basins management were coded as having ex-

perience in the water sector. Farmers and those involved in the agro-processing industry were coded as

42

having experience in the agriculture sector. Women and those who have engaged with gender-focussed

organizations were coded as having experience in the womens sector. Civil engineers or those with ex-

perience in building large-scale urban infrastructure were coded as having experience relevant to to the

Committee on the Federal Capital Territory. Any one who had qualifications in environmental manage-

ment or worked for an organization with experience implementing environmental projects were coded

as having experience relevant to the environment sector. Architects and those involved in small-scale

urban development projects were coded as having experience in the housing sector.

Our information for the biographies came first from the National Assembly web site and where relevant

from the publication, ‘Nigeria’s 4th Republic Handbook 1999-2003’. Where these did not yield sufficient

detail, a comprehensive search of the AllAfrica archive of all newspaper articles from major Nigerian

newspapers was used to collect biographical information. Finally, when this was insufficient, simple

google searches yielded biographical details. In roughly 15% of cases this did not yield an appropriate

biography of the individual. I therefore took the most conservative approach and coded that individual

as having no relevant experience.

I then used education regulations from Nigeria’s education sector to define years of education variables

from the collection of qualifications each representative had earned. For the very small number of rep-

resentatives for which educational qualifications were not available, I replaced their years of education

with the mean of all other representatives and include a dummy indicating that years of education is

missing in all specifications that include representative years of education.

A.3 Process of defining committee membership

The Selection Committee of the House of Representatives selects those politicians they deem fit to be

members of the standing committees. Section XVII of the House Standing Orders states that, “There

shall be a Committee to be known as the Committee on Selection appointed at the commencement of

every Assembly ... The Committee’s jurisdiction shall cover nominating Members to serve on Standing

and Special Committees” amongst other duties.

The Selection Committee is required to select committee members such that each committee is repre-

sentative of Nigeria in terms of its 6 geo-political regions (North Central, North East, North West, South

East, South South, South Central) and the strength of the parties in the House. For example, Order

XIV of the House Standing Orders states, “Members of Committees shall be nominated by the various

political parties and appointed by the Committee on Selection in accordance with their strength in the

House.”

The Selection Committee have a guiding principle that aims to match representative’s qualifications

with the committees they sit on. For example, the vice chairman of the Committee on Selection states,

43

as reported in This Day newspaper, “The Selection Committee ... considered cognate experience, areas

of specialization and zonal representation in order to ensure that the chairman and vice chairman of a

committee do not come from the same geopolitical zone.” (This Day, 2007)

For each committee, the House Standing Orders outline the list of topics to which the House delegates

oversight responsibility. For example, the Health Committee’s jurisdiction covers specialist hospitals,

teaching hospitals, medical research, federal medical centres, and a host of other topics including ‘health

matters generally’. The Selection Committee therefore determine, for each committee, those members

that have relevant qualifications to provide appropriate oversight of the sector. They then choose in-

dividuals from this set within each geo-political zone and within the appropriate proportions of the

relative weights of the parties in the House.

As we have seen in section 2.1 of the main text, there is relatively strong evidence that the Selection

Committee does this based on the three factors of geo-political and party representation and relevant

qualifications and experience. In a number of fora, members of the Selection Committee have stated

that once they must find a doctor within the North-East zone from the ruling party, there is typically

very little room for other factors to play a role. This was confirmed by interviews with the secretaries

of the standing committees and with external academics.

A.3 Defining complexity indicators

Data on the complexity of government projects is not collected by the Nigerian Government nor is a

part of the OPEN data set. I thus worked with a pair of Nigerian engineers familiar with the OPEN

projects and a number of international researchers working on technical complexity to define a relevant

set of indicators. We followed the perspectives on complexity suggested by Remington and Pollack

(2007), by asking the engineer-assessors to individually assess projects along the following five topics,

each with their own set of indicators.

Structural complexity stems from the scale of different interconnected tasks and activities. The indica-

tors associated with this topic capture structural aspects such as project size and the number of inputs

required for production. They also capture issues in raw material and labour supply, and the ease with

which any necessary specialized skills and equipment can be sourced. Temporally complex projects are

those whose production involves uncertainties. Hence there are indicators for uncertainties in design

and implementation. Technically complex projects are those whose production have ambiguous risks,

namely their uncertainties are not well understood. Hence some indicators capture ambiguities in de-

sign and implementation. Directional complexity refers to the potential for preferences over the project

to diverge. The engineer assessors are thus asked to rate the managerial complexities of the project.

44

Finally, there is a subjective assessment as to the overall complexity of the project. This allows any

unassessed aspects of complexity to be measured and provides a coherent picture of project complexity.

Two qualified and independent Nigerian engineers were then contracted to assess each project in the

OPEN data set along these margins. The process of aggregation between engineers used in this project

aimed to build a consensus. The first engineer coded indicators for the entire data set. The codings of

the first engineer were then provided to the second engineer who then constructed his own codings

with reference to the codings of the first. The aim was to anchor the coding of the second engineer in

that of the first but give him freedom to disagree where he felt the coding was incorrect. Other methods

would have been to have them code independently and average the two data sets or to have them work

together. We decided our approach was a balance between consensus and subjectivity.

The two engineers were provided with project details and documents and asked to code a value for

each indicator. The documents only contained information available before implementation such that

there was no bias from the coding being done after the projects were implemented.

Table A5 provides descriptive statistics for all 16 indicators from which the complexity index is con-

structed, as well as how each is correlated with the other indicators. Aggregate complexity is a subjec-

tive assessment of the overall complexity of the projects by the two engineers, that includes ‘all factors

that might influence the difficulty of implementing the project, not only those assessed [by the other

indicators]’. We asked the engineers to take the distribution of complexity in the OPEN data set as a

whole, with the least complex project in the data having an aggregate complexity of zero and the most

complex project having an aggregate complexity of 100, and place each project within this distribution.

We undertook a number of measures to check the complexity of the OPEN indicators coded by the

engineers. First, we inserted 200 randomly chosen repeated projects into the data set provided to the

engineers. Since the project characteristics of the original and repeat projects are identical, we would

expect that the codings of the two sets of projects would be similar. Reassuringly, we find that in gen-

eral the original and duplicate projects are coded in similar ways. We compare the differences between

these two sets by looking at group and paired means, and distributional tests for each variable. The

differences are only statistically significant at conventional levels in a few cases, and the magnitude of

the differences are relatively small. For example, the only variable that is statistically significantly dif-

ferent below the 10% level in the mean-comparison t-test relates to raw material storage. Here, despite a

standard deviation of 0.2 in the originals, the difference is 0.07 between the originals and the duplicates.

Second, we looked at the similarity of the codings of the two engineers. We find that the second engi-

neer’s codings are not dramatically different from the first engineer’s efforts. Whilst there are a small

number of differences, they are relatively small and rarely significant, indicating that the re-coding left

the overall picture relatively stable.

45

Finally, over a year after he had completed the prompted codings, we asked the second engineer to re-

code a sub-sample of projects from scratch, this time without prompting. The differences between these

independent codings and the consensus data we rely on are again relatively minor. It seems that once

he had become accustomed to the broad parameters of the coding framework, the second engineer’s

coding was not dissimilar to the consensus generated by the two engineers working one after the other.

We therefore have evidence of similar projects within the data set being coded in a similar way, of the

two engineers coding in similar ways both when prompted and unprompted, and when there were

deviations, of the deviations not being particularly quantitatively large. Taken together these checks

reassure us that the complexity measures pick up meaningful variation across projects, rather than

merely picking up noise that should have led to the multiple reports (either across engineers or over

time) being uncorrelated.

With the validated complexity indicators, we required indices of the information demands of each

project. Some projects require a lot of local information to be implemented. For example, they may

require sourcing of materials from the local area or be characterised by a high degree of uncertainty

that requires local information to respond to. Similarly, some project may require a lot of information

more readily available at the national level. For example, sourcing international expertise is something

national organizations are likely to be better at procuring than local organizations.

I asked one of the engineers with whom I’d worked to define the complexity data to allocate the com-

plexity variables to one of three indices: (i) indicative of the project requiring local information for

successful implementation; (ii) indicative of the project requiring national information for successful

implementation; or, (iii) indicative of neither. This process led to three complexity indices being gen-

erated using z-scores of the underlying variables. These indices were: (i) localized information in-

dex (containing the variables ‘Storage of raw materials’,‘Requires local labor’,‘Access to construction

equipment’,‘Design uncertainty’,‘Implementation uncertainty’,‘Design ambiguity’ and ‘Implementa-

tion ambiguity’); (ii) national information index (containing the variables ‘Access to raw materials’

and ‘Requires skilled labor’); and (iii) neither information index (containing the variables ‘Project size’,

‘Number of inputs’, ‘Number of methods’, ‘Interdependencies’, ‘Difficulty to manage’ and ‘Number of

agencies involved’).

46

Means and standard deviations

Constituencies

Number of constituencies 349

-

Number of local governments in a constituency 2.09

(0.80)

Population (2006) 380,000

(130,000)

Winning vote share (2003 elections) 0.62

(0.15)

Runner's up vote share (2003 elections) 0.29

(0.12)

Proportion of constituencies run by ruling party 0.62

-

Proportion of constituencies ruling party is runner up 0.34

-

OPEN funds per constituency (US$) 1,600,000

(2,026,667)

Number of OPEN projects by constituency 13

(9.2)

Number of OPEN project types by constituency 3.98

(1.64)

Average project budget (US$) 130,667

(132,667)

Average project complexity (percent) 26.51

(8.21)

Proportion rehabilitation 0.15

(0.16)

Proportion construction 0.90

(0.15)

Proportion never started 0.34

(0.23)

Proportion completed 0.49

(0.23)

Proportion completed conditional on being started 0.73

(0.20)

Proportion fully completed 0.31

(0.25)

Proportion with satisfactory quality 0.80

(0.26)

Proportion implemented by decentralized organization 0.67

(0.20)

Notes: Standard deviations are in parentheses. In the OPEN data, we do not observe 6 of Nigeria's federal

constituencies, and so the descriptives provided here are for this restricted set of constituencies only.

Population data is from the 2006 Census. Election data is from the Independent National Electoral

Commission official record for the 2003 election. Centralized organizations are main ministries which act

as the central organising authority for the sector. Decentralized organizations are 'parastatals', with day-to-

day running largely independent of the central authority. Budget figures originally in Nigerian Naira are

converted to US dollars at a rate of US$1:N150. Figures rounded to two decimal places where relevant.

We have weighted constituencies's equally.

OPEN projects

Table 1a: Descriptive Statistics of Nigeria's Political

Constituencies

Correlations

OPEN funds per constituency -0.01

OPEN funds per constituency per capita -0.02

Number of OPEN projects by constituency 0.08

Number of OPEN project types by constituency 0.01

Proportion of projects implemented by decentralized org. -0.04

Proportion completed 0.16

OPEN funds per constituency -0.01

OPEN funds per constituency per capita 0.00

Number of OPEN projects by constituency 0.03

Number of OPEN project types by constituency 0.06

Proportion of projects implemented by decentralized org. -0.13

Proportion completed -0.07

Table 1b: Descriptive Statistics of the Relationship

Between Constituency Characteristics and Resource

Distribution

Notes: In the OPEN data, we do not observe 6 federal constituencies, and so the descriptives provided here

are for this restricted set of constituencies only. Population data is from the 2006 Census. Election data is

from the Independent National Electoral Commission official record for the 2003 election. Figures rounded

to two decimal places. We have weighted constituencies's equally.

Correlations between political competition and ...

Correlations between ruling party being in power and ...

Table 2: Descriptive Statistics of Nigeria's Federal Organizations, by TierMeans, standard deviations, and p-values of differences with centralized organizations

(1) Centralized (2) Decentralized

p-value of

difference

between (1)/(2)

Number of organizations 8 49 -

- -

Total budget (millions of Naira) 33,704 1,977 0.02

(30,819) (2,349)

Capital budget (millions of Naira) 26,307 1,379 0.04

(27,503) (2,404)

Personnel budget (millions of Naira) 2,280 531 0.04

(1,981) (413)

Overheads budget (millions of Naira) 1,537 68 0.13

(2,459) (56)

Number of staff at organization 5,059 714 0.09

(6,225) (472)

Distance to Abuja 0 301 0.00

(0) (174)

Federal constituencies served in OPEN data 69 19 0.00

(101) (47)

Number of projects 1570 2923 -

- -

Average project budget (millions of Nigerian Naira) 23.08 16.28 0.00

(60.09) (63.07)

Average project complexity (percent) 30.61 22.47 0.00

(19.78) (15.36)

Proportion built in 2006 (rather than 2007) 0.76 0.48 0.00

(0.43) (0.5)

Proportion rehabilitation 0.14 0.13 0.32

(0.35) (0.34)

Proportion construction 0.91 0.88 0.02

(0.29) (0.32)

Proportion never started 0.33 0.28 0.00

(0.47) (0.45)

Proportion completed 0.44 0.57 0.00

(0.4) (0.43)

Proportion completed conditional on being started 0.66 0.79 0.00

(0.31) (0.28)

Proportion fully completed 0.22 0.39 0.00

(0.42) (0.49)

Proportion with satisfactory quality 0.82 0.83 0.54

(0.39) (0.38)

Notes: Standard deviations are in parentheses. Budget data are an average of organization budget figures for the years 2006-7. Data on number of

staff are mainly from administrative data. In the few cases we do not have the staff numbers explicitly, we estimate them from the personnel

expenditures, which have are correlated with staff numbers with a coefficient of over 0.9. Centralized organizations are main ministries which act as

the central organising authority for the sector. Decentralized organizations are 'parastatals', with day-to-day running largely independent of the central

authority. Figures rounded to two decimal places. T-tests are performed under the assumption of equal variances when the chi-squared test statistic

is less than or equal to 3.84 and under the assumption of unequal variances otherwise. We have weighted organization's equally.

Organization characteristics

OPEN projects

Means, standard deviations, and p-values of differences with centralized organizations

(1) All

organizations(2) Centralized (3) Decentralized

(4) p-value of

difference

between (1)/(2)

Proportion of projects with engagement of member of National Assembly 11.2 15.4 10.4 0.09

(7.6) (5.7) (7.7)

Proportion of projects with intervention by member of National Assembly 18.7 23.4 17.9 0.20

(11.2) (6.7) (11.7)

Correlation between intervention and engagement measures 0.86 0.69 0.87

Notes: Standard deviations are in parentheses. Decentralized organizations are 'parastatals', with day-to-day running largely independent of the central authority. Figures rounded to two decimal places.

T-tests are performed under the assumption of equal variances when the chi-squared test statistic is less than or equal to 3.84 and under the assumption of unequal variances otherwise. We have

weighted organization's equally.

Table 3: Descriptive Statistics of the Relationship Between Decentralization and Political Engagement

Table 4: Delegation by Politicians to Centralized and Decentralized Organizations

Dependent Variable: Indicator of decentralization [decentralization=1]

Robust Standard Errors

OLS Estimates

(1) (2) (3) (4)

Politician member of relevant committee [yes=1] 0.00 -0.11*** -0.11*** -0.11***

(0.01) (0.03) (0.03) (0.03)

Level of political competition -0.08*** -0.08*** -0.06**

(0.02) (0.02) (0.03)

Politician member of relevant committee x level of political competition 0.15*** 0.16*** 0.16***

(0.04) (0.04) (0.04)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) No No Yes Yes

Constituency Controls (poverty and education indices, schooling, water

and electricity access, recent government investments, and indicators of

economic dynamics)

No No No Yes

Fixed Effects (project type) Yes Yes Yes Yes

Mean of dependant variable 0.65 0.65 0.65 0.65

R-squared 0.53 0.53 0.54 0.55

Observations 4493 4493 4493 4493

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. All columns report OLS estimates. The dependant variable in all specifications is a

binary variable reflecting whether a project is decentralized or not, which takes the value 1 when the project is implemented by a decentralized organization. Decentralization refers to whether the

organization is an organization, with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central organising authority for the sector. The level of

political competition is measured as one minus the difference between the winner's vote share and the runner's up vote share. Project controls comprise project-level controls for the project budget,

whether the project is new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a binary variable indicating whether it was implemented in 2006 rather than 2007.

We also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a project requires local information to implement and one that indicates a

project requires national information to implement. Project Type fixed effects relate to whether the primary classification of the project is as a financial, training, advocacy, procurement, research,

electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant

congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and include a

dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured

by both a national poverty index and a self-rating index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of

hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest primary school, and the average time in minutes to the nearest

secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five

years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of piped water

infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services,

and agricultural-inputs schemes. Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in

opportunities for employment, the availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability

of consumer goods. All specifications include an indicator of the 'grade' of the committee under which the project falls, which is a dummy that takes the value 1 if the committee is perceived to be of

high political weight or otherwise. Figures rounded to two decimal places.

Table 5: Engagement of Politicians with Implementing Organizations

Dependent Variable: Proportion of projects on which politician directly engages bureaucrat

Robust Standard Errors

OLS Estimates

(1) (2) (3) (4)

Politician member of relevant committee [yes=1] 0.002* -0.006* -0.007* -0.004

(0.001) (0.004) (0.004) (0.004)

Level of political competition 0.015*** 0.015*** 0.006**

(0.003) (0.003) (0.003)

Politician member of relevant committee x level of political competition 0.012** 0.014*** 0.010*

(0.005) (0.005) (0.005)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) No No Yes Yes

Constituency Controls (poverty and education indices, schooling, water

and electricity access, recent government investments, and indicators of

economic dynamics)

No No No Yes

Fixed Effects (project type) Yes Yes Yes Yes

Mean of dependant variable 0.14 0.14 0.14 0.14

R-squared 0.72 0.72 0.72 0.75

Observations 3757 3757 3757 3757

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. All columns report OLS estimates. The dependant variable in all specifications is a

binary variable reflecting whether a project is decentralized or not, which takes the value 1 when the project is implemented by a decentralized organization. Decentralization refers to whether the

organization is an organization, with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central organising authority for the sector. The level of

political competition is measured as one minus the difference between the winner's vote share and the runner's up vote share. Project controls comprise project-level controls for the project budget,

whether the project is new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a binary variable indicating whether it was implemented in 2006 rather than 2007.

We also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a project requires local information to implement and one that indicates a

project requires national information to implement. Project Type fixed effects relate to whether the primary classification of the project is as a financial, training, advocacy, procurement, research,

electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant

congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and include a

dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured

by both a national poverty index and a self-rating index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of

hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest primary school, and the average time in minutes to the nearest

secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five

years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of piped water

infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services,

and agricultural-inputs schemes. Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in

opportunities for employment, the availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability

of consumer goods. All specifications include an indicator of the 'grade' of the committee under which the project falls, which is a dummy that takes the value 1 if the committee is perceived to be of

high political weight or otherwise. Figures rounded to two decimal places.

Table 6: Impact of Delegation and Engagement at Different Levels of Political Competition

Dependent Variable: Proportion of Project Completed

Robust Standard Errors

OLS Estimates

(1) (2) (3) (4) (5)

Level of political competition -0.05* -0.03 -0.05 0.03 -0.06

(0.03) (0.04) (0.09) (0.04) (0.06)

Decentralization [decentralized=1] 0.38*** 0.65*** 0.33*** 0.36***

(0.03) (0.06) (0.04) (0.03)

Proportion of projects with engagement of politician 1.15*** -0.95*** -1.55***

(0.43) (0.22) (0.35)

Level of political competition x Decentralization 0.04

(0.05)

Level of political competition x Proportion of projects with engagement of politician 0.86**

(0.38)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling, water and electricity

access, recent government investments, and indicators of economic dynamics)Yes Yes Yes Yes Yes

Fixed Effects (project type) Yes Yes Yes Yes Yes

Mean of dependant variable 0.52 0.52 0.52 0.52 0.52

R-squared 0.27 0.31 0.31 0.31 0.31

Observations 3757 3757 3757 3757 3757

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. The dependant variable in all specifications is the proportion of the project completed (that is a continuous measure

between zero and one). Decentralization refers to whether the organization is a 'parastatal', with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central organising authority for the

sector. The proportion of projects with engagement of member of National Assembly is an organizational-average of the buearucrat responses to the question 'How often, if at all, do you personally engage with [members of the

National Assembly] in the work that you do?' Project controls comprise project-level controls for the project budget, whether the project is new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a

binary variable indicating whether it was implemented in 2006 rather than 2007. We also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a project requires local

information to implement and one that indicates a project requires national information to implement. Project Type fixed effects relate to whether the primary classification of the project is as a financial, training, advocacy,

procurement, research, electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant congressperson.

In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and include a dummy variable to indicate the missing data. Constituency

characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty index and a self-rating index, the average years of education of the

household head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest

primary school, and the average time in minutes to the nearest secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project

of the named type in the five years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of piped water

infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes.

Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of agricultural inputs, number of

buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of consumer goods. All specifications include an indicator of the 'grade' of the committee under which the project

falls, which is a dummy that takes the value 1 if the committee is perceived to be of high political weight or otherwise. Figures rounded to two decimal places.

Table A1: Investigating the Differences in Constituency Characteristics by Level of Competition

Means, standard deviations and p-values of differences

(1) All

constituencies

(2) 10% most

competitive

constituencies

(3) 10% least

competitive

constituencies

(4) p-value of

difference

between 2/3

Sex [female=1] 0.05 0.06 0.12 0.38

(0.22) (0.24) (0.33)

Politician age 47.45 46.71 49.17 0.13

(6.74) (4.34) (8.13)

Politician years of education 16.24 16.26 15.62 0.06

(1.51) (1.09) (1.61)

Tenure in house [second term=1] 0.27 0.23 0.15 0.39

(0.44) (0.43) (0.36)

Proportion of constituency population in extreme poverty 0.24 0.26 0.23 0.47

(0.14) (0.14) (0.15)

Standard deviation of poverty index 0.39 0.41 0.38 0.22

(0.09) (0.08) (0.11)

Average years of education of household head 4.78 6.50 4.76 0.01

(2.62) (2.42) (3.05)

Standard deviation of education of household head 4.94 5.01 4.81 0.41

(1.02) (0.58) (1.3)

Proportion of constituents with access to potable water 0.45 0.37 0.52 0.01

(0.25) (0.24) (0.24)

Standard deviation of access to potable water 0.42 0.41 0.44 0.17

(0.09) (0.09) (0.07)

Average hours of electricity in a day 4.48 3.69 4.80 0.21

(3.74) (3.03) (4.19)

Standard deviation of hours of electricity 5.69 5.50 5.48 0.98

(2.27) (2.5) (2.49)

Time in minutes to nearest primary school 20.65 22.25 19.43 0.07

(7.05) (5.86) (6.6)

Standard deviation of time to nearest primary school 14.88 14.88 14.78 0.92

(3.98) (3.6) (4.53)

Time in minutes to nearest secondary school 34.65 35.05 33.15 0.40

(9.13) (7.66) (10.71)

Standard deviation of time to nearest secondary school 17.06 16.95 17.21 0.72

(2.96) (2.41) (3.43)

Subjective poverty index (defined as poor by household) 0.66 0.74 0.68 0.17

(0.21) (0.16) (0.21)

Standard deviation of subjective poverty index 0.41 0.40 0.41 0.59

(0.1) (0.1) (0.09)

Observations 349 349 349

Notes: Standard deviations are in parentheses. Politician characteristics are a dummy variable indicating the sex of a politician, which takes the value 1 when the poltician is

female, the age in years and years of education of the relevant congressperson, and whether they were in congress in the 1999-2003 Assembly (before which Nigeria was ruled by

a military government). In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the

representatives and include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the

proportion of poor in the constituency, measured by both a national poverty index and a self-rating index, the average years of education of the household head, the proportion of

constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in

minutes to the nearest primary school, and the average time in minutes to the nearest secondary school. T-tests are performed under the assumption of equal variances when the

chi-squared test statistic is less than or equal to 3.84 and under the assumption of unequal variances otherwise. Figures rounded to two decimal places where relevant. We have

weighted constituencies's equally.

Politician characteristics

Constituency characteristics

Table A2: Investigating the Determinants of Committee Membership (House)

Dependent Variable: System of Eleven Equations in Membership of Sectoral Committees

Robust Standard Errors

Estimates by Maximum Likelihood to Fit a SUR Model

Member of

Finance

Committee

Member of

Appropriation

Committee

Member of

Power

Committee

Member of

Water

Committee

Member of

Agriculture

Committee

Member of

Health

Committee

Member of

Education

Committee

Member of

Environment

Committee

Member of

Housing

Committee

Member of

Women/Youth

Committee

Member of

FCT

Committee

Politician has relevant qualifications/experience [yes=1] 0.31*** 0.38*** 0.88*** 0.56*** 0.49*** 0.73*** 0.48*** 0.71*** 0.69*** 0.53** 0.80***

(0.05) (0.05) (0.05) (0.08) (0.06) (0.07) (0.07) (0.08) (0.07) (0.25) (0.07)

Level of political competition 0.12 0.04 -0.01 -0.03 -0.18*** 0.09 -0.05 0.01 0.01 0.08 0.07

(0.08) (0.08) (0.07) (0.09) (0.07) (0.07) (0.08) (0.08) (0.06) (0.05) (0.05)

Index of poverty 0.38 -0.09 0.23 0.03 -0.42** -0.33 -0.05 -0.28 0.09 0.08 -0.09

(0.24) (0.21) (0.19) (0.26) (0.18) (0.21) (0.22) (0.2) (0.23) (0.18) (0.19)

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling,

water and electricity access, recent government investments, and

indicators of economic dynamics)

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Correlation of residuals in SURE system:

Appropriation -0.19

Power 0.06 0.01

Water -0.02 0.07 0.09

Agriculture -0.09 0.01 -0.07 -0.10

Health -0.04 -0.06 0.08 0.01 -0.03

Education 0.02 -0.08 -0.16 -0.11 -0.18 -0.05

Environment 0.03 0.04 0.02 0.03 -0.05 0.05 0.04

Housing 0.02 -0.04 0.02 0.06 -0.04 0.03 -0.03 -0.05

Women/Youth 0.01 0.13 -0.01 0.09 0.03 -0.02 -0.01 0.10 -0.07

FCT 0.01 -0.05 -0.09 0.00 -0.07 0.04 -0.03 0.02 0.15 0.12

Observations 349 349 349 349 349 349 349 349 349 349 349

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. The dependant variable in all specifications is a binary variable reflecting whether a representative is a member of the committee for the named sector, taking the value 1 when the politician is a member.

Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and

include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty index and a self-rating index, the average years of education of the household

head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest primary school, and the average time in minutes to the nearest secondary school. Means and

standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped

water infrastructure, rehabilitation of piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes. Finally, constituency characteristics

include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of

consumer goods. Columns report maximum likelihood estimates to fit a SUR model for the eleven sectoral committees. All specifications include dummies for five of the six geo-political zones and an indicator of the 'grade' of the committee under which the project falls, which is a dummy that takes the value 1 if the

committee is perceived to be of high political weight or otherwise. Figures rounded to two decimal places.

Table A3: Investigating the Determinants of Politician's Qualifications and Experience

Dependent Variable: System of Ten Equations in Sector of Politician's Qualifications and Experience

Robust Standard Errors

Estimates by Maximum Likelihood to Fit a SUR Model

Finance

SectorPower Sector Water Sector

Agriculture

SectorHealth Sector

Education

Sector

Environment

Sector

Housing

Sector

Women/Youth

Sector

FCT (city

building)

Sector

Level of political competition 0.04 0.03 0.03 -0.05 -0.02 -0.11 -0.12 -0.13* 0.00 0.08

(0.11) (0.06) (0.08) (0.09) (0.07) (0.08) (0.07) (0.07) (0.03) (0.08)

Index of poverty -0.37 0.07 0.01 -0.08 0.28 -0.29 -0.07 0.06 0.09 0.34

(0.3) (0.14) (0.01) (0.24) (0.28) (0.24) (0.15) (0.28) (0.1) (0.26)

Average hours constituents have electricity each day 0.00 -0.01 0.00

(0.01) (0.01) (0)

Proportion of constituents with access to potable water -0.07 -0.06 -0.03 -0.10*

(0.09) (0.1) (0.07) (0)

Time in minutes to nearest secondary school 0.00 0.00

(0) (0)

Politician Controls (sex, age, years of education) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling,

water and electricity access, recent government investments, and

indicators of economic dynamics)

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Correlation of residuals in SURE system:

Power 0.01

Water 0.07 0.17

Agriculture -0.12 0.00 0.13

Health -0.13 -0.04 -0.05 -0.07

Education 0.07 -0.04 0.00 -0.05 -0.04

Environment -0.03 0.00 0.14 0.21 -0.13 -0.04

Housing 0.03 0.15 0.27 0.08 -0.06 -0.06 0.11

Women/Youth 0.05 -0.01 -0.07 -0.04 -0.03 -0.01 -0.05 0.08

FCT 0.06 0.10 0.29 0.05 -0.07 0.02 0.06 0.24 0.09

Observations 349 349 349 349 349 349 349 349 349 349

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. The dependant variable in all specifications is a binary variable reflecting whether a representative has qualifications and/or experience in the named sector, taking the value 1 when

the politician has relevant qualifications and/or experience. Politician controls comprise constituency-level controls for the sex, age and years of education of the relevant congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing

value with the mean of the rest of the representatives and include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty

index and a self-rating index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the

nearest primary school, and the average time in minutes to the nearest secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five years preceding 2006:

construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility

rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes. Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for

employment, the availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of consumer goods. Columns report maximum likelihood estimates to fit a SUR model for the eleven sectoral

committees. Figures rounded to two decimal places.

Table A4: Descriptive Evidence on Project Types

Project Type

(1) Number of

Projects

[Decentralised]

(2) Number of

Implementing

Organizations

[Decentralised]

Electrification 1466 [718] 3 [1]

Borehole 1188 [1144] 18 [15]

Building 769 [529] 33 [28]

Dam 292 [124] 14 [11]

Procurement 252 [145] 34 [29]

Road 213 [3] 4 [2]

Financial project 124 [120] 6 [3]

Canal 74 [63] 12 [9]

Training 63 [32] 15 [12]

Advocacy 41 [38] 16 [14]

Research 11 [7] 9 [6]

Notes: The “project type” classification refers to the primary classification for each project. Other project classifications exist. The budget allocation in Column 3 is in millions of Nigerian Naira. The sample of projects covers those which have a

positive budget allocation, for which the proportion completed evaluation variable and a location is available and which is decentralisable, in that it could be implemented by multiple tiers of government. The project quality variable in Column 8 is

not available for all projects. Standard deviations are in parentheses. Figures are rounded to two decimal places where relevant.

Table A4: Descriptive Evidence on Project Types

(3) Budget

Allocation

(Millions of

Nigerian Naira)

[standard dev.]

(4) Proportion

Never Started

(5) Proportion

Completed

14.9 [20.6] 0.15 0.57

10.4 [55.1] 0.37 0.53

24.7 [24.7] 0.35 0.51

27.2 [41.3] 0.58 0.32

17.4 [24.9] 0.31 0.64

30.5 [25.5] 0.11 0.53

10.6 [47.6] 0.43 0.46

82.9 [12.5] 0.69 0.14

63.8 [39.8] 0.37 0.55

5.5 [11.1] 0.27 0.63

9.2 [8.0] 0.09 0.79

Notes: The “project type” classification refers to the primary classification for each project. Other project classifications exist. The budget allocation in Column 3 is in millions of Nigerian Naira. The sample of projects covers those which have a

positive budget allocation, for which the proportion completed evaluation variable and a location is available and which is decentralisable, in that it could be implemented by multiple tiers of government. The project quality variable in Column 8 is

not available for all projects. Standard deviations are in parentheses. Figures are rounded to two decimal places where relevant.

(6) Proportion

Completed

Conditional on

Being Started

(7) Proportion

Fully Completed

(8) Proportion

With

Satisfactory

Quality

0.67 0.26 0.85

0.84 0.42 0.85

0.78 0.34 0.81

0.76 0.22 0.49

0.92 0.58 0.82

0.59 0.22 0.79

0.81 0.35 0.79

0.45 0.05 0.92

0.87 0.40 0.83

0.86 0.49 0.96

0.86 0.73 1.00

Notes: The “project type” classification refers to the primary classification for each project. Other project classifications exist. The budget allocation in Column 3 is in millions of Nigerian Naira. The sample of projects covers those which have a

positive budget allocation, for which the proportion completed evaluation variable and a location is available and which is decentralisable, in that it could be implemented by multiple tiers of government. The project quality variable in Column 8 is

not available for all projects. Standard deviations are in parentheses. Figures are rounded to two decimal places where relevant.

Table A5: Correlation of Subcomponents of the Project Complexity Indicator

MeanStandard

deviation

Pro

jec

t s

ize

Nu

mb

er

of

inp

uts

Nu

mb

er

of

me

tho

ds

Inte

rde

pe

nd

en

cie

s

Ac

ce

ss

to

ra

w m

ate

ria

ls

Sto

rag

e o

f ra

w m

ate

ria

ls

Re

qu

ire

s lo

ca

l la

bo

r

Re

qu

ire

s s

kille

d la

bo

r

Ac

ce

ss

to

co

ns

tru

cti

on

eq

uip

me

nt

De

sig

n u

nc

ert

ain

ty

Imp

lem

en

tati

on

un

ce

rta

inty

De

sig

n a

mb

igu

ity

Imp

lem

en

tati

on

am

big

uit

y

Dif

fic

ult

y t

o m

an

ag

e

Nu

mb

er

of

ag

en

cie

s in

vo

lve

d

Ag

gre

ga

te c

om

ple

xit

y

Project size 0.26 0.44 1.00

Number of inputs 6.65 4.03 0.10 1.00

Number of methods 5.18 2.16 0.36 0.62 1.00

Interdependencies 0.72 0.45 0.00 0.05 0.05 1.00

Access to raw materials 0.38 0.49 -0.20 -0.25 -0.11 0.14 1.00

Storage of raw materials 0.04 0.19 0.16 -0.07 0.05 0.08 -0.09 1.00

Requires local labor 0.61 0.49 0.32 -0.06 0.47 0.08 0.28 0.14 1.00

Requires skilled labor 0.63 0.48 -0.25 -0.07 -0.34 0.65 0.23 -0.05 -0.17 1.00

Access to construction equipment 0.38 0.49 -0.05 -0.29 0.02 0.42 0.67 0.09 0.53 0.49 1.00

Design uncertainty 0.79 0.41 0.23 0.10 0.13 0.78 0.02 0.07 0.29 0.49 0.34 1.00

Implementation uncertainty 0.86 0.35 0.14 0.22 0.22 0.56 -0.10 0.00 0.40 0.44 0.26 0.73 1.00

Design ambiguity 0.74 0.44 -0.03 -0.03 -0.04 0.85 0.09 0.10 0.04 0.72 0.39 0.68 0.58 1.00

Implementation ambiguity 0.72 0.45 0.05 -0.03 0.01 0.88 0.11 0.12 0.08 0.67 0.41 0.73 0.51 0.90 1.00

Difficulty to manage 0.45 0.50 0.14 -0.21 0.21 0.43 0.52 0.22 0.64 0.28 0.82 0.38 0.24 0.48 0.49 1.00

Number of agencies involved 3.72 0.45 -0.08 0.12 -0.16 0.27 0.11 -0.10 0.09 0.45 0.12 0.42 0.60 0.36 0.28 0.09 1.00

Aggregate complexity 25.31 17.47 0.47 0.21 0.56 0.18 -0.18 0.21 0.42 -0.19 0.01 0.24 0.28 0.21 0.24 0.33 -0.15 1.00

Observations (projects) 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493 4493

Notes: The sample used is those projects in our core analysis for which we have complexity and project completion data. 'Project size' is a binary variable that aims to gauge the physical size of the

project. It takes the value 1 if it is classified as equivalent to a medium scale build or larger. Number of inputs counts the number of distinct product classes the finished project contains. Number of

methods counts the number of distinct disciplines or methods involved in implementing the project. 'Interdependencies' is a binary variable reflecting the extent of interdependencies between the

activities involved in the project. It takes a value of 1 if the project is classified as highly interdependent. 'Access to raw materials' is a binary variable that takes the value 1 if raw materials could not be

sourced within the state of implementation. 'Storage of raw materials' is a binary variable that takes the value 1 if some of the raw materials could not be easily stored or transported. 'Requires local

labor' is a binary variable that takes the value 1 if local labor was useful or critical. 'Requires skilled labor' is a binary variable that takes the value 1 if specialized skills were necessary and difficult to

obtain. 'Access to construction equipment' is a binary variable that takes the value 1 if the equipment required is difficult to obtain, heavy duty, or difficult to transport to the site. 'Design uncertainty' is a

binary variable that takes on the value 1 if the design of the project is context specific. 'Implementation uncertainty' is a binary variable that takes on the value 1 if there are substantial risks involved in

implementation. 'Design ambiguity' is a binary variable that takes on the value 1 if there is a risk of redesign late on in the project. 'Implementation ambiguity' is a binary variable that takes on the value 1

if the technical risks of the project cannot be fully understood at implementation. 'Difficulty to manage' is a binary variable that takes the value 1 if the project is seen have elements that require project

management skills of above average level. 'Number of agencies involved' is simply a count of the estimated number of agencies involved in the project cycle. 'Aggregate complexity' is a subjective

assessment as to the overall complexity of the project by the coding engineers. This variable is an assessment of the interaction of the other variables as well as any unassessed aspects of complexity

and provides a coherent picture of the complexity of the projects by a specialist. The variables 'interdependencies', 'access to raw materials', 'requires local labor', 'requires skilled labor', 'access to

construction equipment', 'design uncertainty', 'implementation uncertainty', 'design ambiguity', 'implementation ambiguity' and 'difficulty to manage' are binary variables reflecting the variation in these

previously categorical variables. Figures are rounded to two decimal places.

Table A6: Impact of Committee Membership on Constituency Resources

Robust Standard Errors

OLS Estimates

(1) Total

constituency

resources

(millions of Naira)

(2) Number of

projects in

constituency

Politician member of relevant committee [proportion] -42.80 2.72

(98.00) (2.81)

Level of political competition 8.12 3.70

(86.20) (2.26)

Politician member of relevant committee x level of political competition 141.00 -1.71

(150.00) (4.18)

Project Controls (log budget, existing investment, complexity, year) Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes

Constituency Controls (poverty and education indices, schooling, water

and electricity access, recent government investments, and indicators of

economic dynamics)

Yes Yes

Fixed Effects (proportions of project types within the constituency) Yes Yes

Mean of dependant variable 240 13

Observations 349 349

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. The dependant variable in Column

1 is the total resources allocated to a constituency in the OPEN data. In Column 2, the dependant variable is the total number of projects allocated

to a constituency in the OPEN data. The committee membership variable is defined as that proportion of projects within the constituency where the

congressperson sits on the relevant committee. Project controls comprise constituency-level averages for the project budget, whether the project is

new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a binary variable indicating whether it was

implemented in 2006 rather than 2007 for all the OPEN projects in the constituency. Project Type fixed effects indicate the fraction of projects in a

constituency whose primary classification is as a financial, training, advocacy, procurement, research, electrification, borehole, dam, building, canal

or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant

congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of

the rest of the representatives and include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and

standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty index and a self-rating

index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of

hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest primary school,

and the average time in minutes to the nearest secondary school. Means and standard deviations of the following indices are also included to

reflect the frequency with which constituents benefit from a public project of the named type in the five years preceding 2006: construction of

electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of

piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road

construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes. Finally, constituency characteristics include a set of

indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of

agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the

availability of consumer goods. Figures rounded to two decimal places.

Dependent Variable: Column 1: Sum of project budgets within relevant federal constituency; Column 2: Count of

projects within relevant federal constituency

Table A7: Robustness of Delegation Findings

Dependent Variable: Indicator of decentralization [decentralization=1]

Robust Standard Errors in All Columns Bar 2 and Clustered at the Constitutency-Sector Level in Column 2

OLS Estimates

(1) Baseline

specification

(2) Clustering at

sector-

constitutency

(3) Chairperson

dummy

(4) Excluding

chairperson

projects

(5) Ruling party

dummy

(6) Excluding

ruling party

projects

(7) Two-term

dummy

(8) Excluding two-

term projects

(9) Excluding

unqualified

agriculture

(10) Projects with

organisation

survey

Politician member of relevant committee [yes=1] -0.11*** -0.11** -0.11*** -0.11*** -0.11*** -0.23* -0.11*** -0.16*** -0.09*** -0.10***

(0.03) (0.05) (0.03) (0.03) (0.03) (0.12) (0.03) (0.04) (0.03) (0.03)

Level of political competition -0.06** -0.06 -0.06** -0.06** -0.08*** -0.06 -0.06** -0.06 -0.06* -0.01

(0.03) (0.05) (0.03) (0.03) (0.03) (0.09) (0.03) (0.04) (0.03) (0.03)

Politician member of relevant committee x level of political competition 0.16*** 0.16** 0.16*** 0.17*** 0.16*** 0.37** 0.16*** 0.21*** 0.14*** 0.15***

(0.04) (0.07) (0.04) (0.04) (0.04) (0.15) (0.04) (0.05) (0.04) (0.04)

Chairpersons [yes=1] -0.08*

(0.04)

Ruling party [yes=1] -0.04***

(0.01)

Politician in second term at congress [yes=1] 0.00

(0.01)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling, water

and electricity access, recent government investments, and indicators of

economic dynamics)

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Fixed Effects (project type) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations (clusters) 4493 4493 (1169) 4493 4422 4493 1659 4493 3283 4464 3757

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses. They are robust in all columns bar 2 and clustered by sectors within each constituency in Column 2. The dependant variable in all specifications is a binary variable reflecting whether a project is decentralized or not, which takes the

value 1 when the project is implemented by a decentralized organization. Decentralization refers to whether the organization is a 'parastatal', with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central organising authority for the sector. The level of political competition is measured

as one minus the difference between the winner's vote share and the runner's up vote share. In Column 3 we include a dummy for whether the project is located in a constituency represented by the chair of the relevant sectoral committee for that project. Column 4 restricts our specification to those projects that are not implemented in

constituencies in which the representative is the chair of the relevant sectoral committee. In Column 5 we include a dummy for whether the project is located in a constituency in which the representative is a member of the ruling party. Column 6 restricts our specification to those projects that are not implemented in constituencies in

which the representative is a member of the ruling party. In Column 7 we include a dummy for whether the project is located in a constituency in which the representative is in their second term at congress. Column 8 restricts our specification to those projects that are not implemented in constituencies in which the representative is in

their second term of congress. Column 9 restricts our specification to those projects that are not implemented in constituencies in which the representative did not have relevant qualifications or experience in agriculture but was a member of the agriculture committee. Column 10 restricts our specification to those projects implemented

in organizations for which we have Civil Servants Survey data. Project controls comprise project-level controls for the project budget, whether the project is new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a binary variable indicating whether it was implemented in 2006 rather than 2007. We

also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a project requires local information to implement and one that indicates a project requires national information to implement. Project Type fixed effects relate to whether the primary classification of the project is as a

financial, training, advocacy, procurement, research, electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant congressperson. In the very small number of cases in which age or years of education are

missing, we replace the missing value with the mean of the rest of the representatives and include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty index and a

self-rating index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in minutes to the nearest primary school, and the average time in minutes

to the nearest secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole,

construction of piped water infrastructure, rehabilitation of piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes. Finally, constituency characteristics

include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of consumer goods. All

specifications include an indicator of the 'grade' of the committee under which the project falls, which is a dummy that takes the value 1 if the committee is perceived to be of high political weight or otherwise. Figures rounded to two decimal places.

Table A8: Robustness of Engagement Findings

Dependent Variable: Indicator of decentralization [decentralization=1]

Robust Standard Errors in All Columns Bar 2 and Clustered at the Constitutency-Sector Level in Column 2

OLS Estimates

(1) Baseline

specification

(2) Clustering at

sector-

constitutency

(3) Chairperson

dummy

(4) Excluding

chairperson

projects

(5) Ruling party

dummy

(6) Excluding

ruling party

projects

(7) Two-term

dummy

(8) Excluding two-

term projects

(9) Excluding

unqualified

agriculture

Politician member of relevant committee [yes=1] -0.004 -0.004 -0.004 -0.003 -0.005 -0.036*** 0.004 -0.013*** -0.002

(0.004) (0.006) (0.004) (0.004) (0.004) (0.013) (0.004) (0.005) (0.004)

Level of political competition 0.006** 0.006 0.006** 0.006* 0.000 0.015* 0.006** 0.001 0.007**

(0.003) (0.005) (0.003) (0.003) (0.003) (0.009) (0.003) (0.004) (0.003)

Politician member of relevant committee x level of political competition 0.010* 0.010 0.010** 0.008 0.011** 0.054*** 0.010* 0.026*** 0.009*

(0.005) (0.01) (0.005) (0.005) (0.005) (0.017) (0.005) (0.006) (0.005)

Chairpersons [yes=1] -0.003

(0.006)

Ruling party [yes=1] -0.012***

(0.001)

Politician in second term at congress [yes=1] -0.004**

(0.001)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes Yes Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes Yes Yes Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling, water

and electricity access, recent government investments, and indicators of

economic dynamics)

Yes Yes Yes Yes Yes Yes Yes Yes Yes

Fixed Effects (project type) Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations (clusters) 3757 3757 (1127) 3757 3697 3757 1443 3757 2722 3728

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. They are robust in all columns bar 2 and clustered by sectors within each constituency in Column 2. The dependant variable in all specifications is a binary variable reflecting whether a project is decentralized or not, which takes the value 1 when

the project is implemented by a decentralized organization. Decentralization refers to whether the organization is a 'parastatal', with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central organising authority for the sector. The level of political competition is

measured in as one minus the difference between the winner's vote share and the runner's up vote share. In Column 3 we include a dummy for whether the project is located in a constituency represented by the chair of the relevant sectoral committee for that project. Column 4 restricts our specification to those

projects that are not implemented in constituencies in which the representative is the chair of the relevant sectoral committee. In Column 5 we include a dummy for whether the project is located in a constituency in which the representative is a member of the ruling party. Column 6 restricts our specification to

those projects that are not implemented in constituencies in which the representative is a member of the ruling party. In Column 7 we include a dummy for whether the project is located in a constituency in which the representative is in their second term at congress. Column 8 restricts our specification to those

projects that are not implemented in constituencies in which the representative is in their second term of congress. Column 9 restricts our specification to those projects that are not implemented in constituencies in which the representative did not have relevant qualifications or experience in agriculture but was a

member of the agriculture committee. Project controls comprise project-level controls for the project budget, whether the project is new or a rehabilitation, an assessment of its aggregate complexity by Nigerian engineers and a binary variable indicating whether it was implemented in 2006 rather than 2007. We

also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a project requires local information to implement and one that indicates a project requires national information to implement. Project Type fixed effects relate to whether the primary classification of

the project is as a financial, training, advocacy, procurement, research, electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant congressperson. In the very small number of

cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and include a dummy variable to indicate the missing data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in

the constituency, measured by both a national poverty index and a self-rating index, the average years of education of the household head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the

average time in minutes to the nearest primary school, and the average time in minutes to the nearest secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit from a public project of the named type in the five years

preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility

rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs schemes. Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of

agricultural inputs, number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of consumer goods. All specifications include an indicator of the 'grade' of the committee under which the project falls, which is a dummy that takes the value 1 if the

committee is perceived to be of high political weight or otherwise. Figures rounded to two decimal places.

Table A9: Impact of Political Delegation and Engagement on Quality

Dependent Variable: Binary indicator of satisfactory quality of project implementation [satisfactory or above=1]

Robust Standard Errors

OLS Estimates

(1) (2) (3) (4) (5)

Level of political competition -0.07 -0.08* -0.07 -0.04 -0.39***

(0.04) (0.05) (0.05) (0.06) (0.1)

Decentralization [decentralized=1] 0.07 0.08* 0.12* 0.08*

(0.04) (0.04) (0.07) (0.04)

Proportion of projects with engagement of politician -0.52 -0.50 -2.52***

(0.34) (0.34) (0.65)

Level of political competition x Decentralization -0.06

(0.08)

Level of political competition x Proportion of projects with engagement of politician 2.72***

(0.77)

Project Controls (log budget, existing investment, complexity, year) Yes Yes Yes Yes Yes

Politician Controls (sex, age, years of education, tenure in congress) Yes Yes Yes Yes Yes

Constituency Controls (poverty and education indices, schooling, water and electricity

access, recent government investments, and indicators of economic dynamics)Yes Yes Yes Yes Yes

Fixed Effects (project type) Yes Yes Yes Yes Yes

Observations 1911 1911 1911 1911 1911

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses. The dependant variable in all specifications is a dummy variable that takes the value one if project quality is reported as

satisfactory or higher, and zero otherwise. Decentralization refers to whether the organization is a 'parastatal', with day-to-day running largely independent of the central authority, rather than a centralized ministry, the central

organising authority for the sector. The proportion of projects with engagement of member of National Assembly is an organizational-average of the buearucrat responses to the question 'How often, if at all, do you personally engage

with [members of the National Assembly] in the work that you do?' Project controls comprise project-level controls for the project budget, whether the project is new or a rehabilitation, an assessment of its aggregate complexity by

Nigerian engineers and a binary variable indicating whether it was implemented in 2006 rather than 2007. We also include two additional project complexity indices, one that is an aggregate of all complexity variables that indicate a

project requires local information to implement and one that indicates a project requires national information to implement. Project Type fixed effects relate to whether the primary classification of the project is as a financial, training,

advocacy, procurement, research, electrification, borehole, dam, building, canal or road project. Politician controls comprise constituency-level controls for the sex, age, years of education and tenure in congress of the relevant

congressperson. In the very small number of cases in which age or years of education are missing, we replace the missing value with the mean of the rest of the representatives and include a dummy variable to indicate the missing

data. Constituency characteristics comprise the means and standard deviations of the following indices: the proportion of poor in the constituency, measured by both a national poverty index and a self-rating index, the average years

of education of the household head, the proportion of constituents with access to potable water, the average number of hours constituents had electricity in the 24 hours preceding the household questionnaire, the average time in

minutes to the nearest primary school, and the average time in minutes to the nearest secondary school. Means and standard deviations of the following indices are also included to reflect the frequency with which constituents benefit

from a public project of the named type in the five years preceding 2006: construction of electrification infrastructure, rehabilitation of electrification infrastructure, well/borehole, construction of piped water infrastructure, rehabilitation of

piped water infrastructure, sanitation, school construction project, school rehabilitation, health facility construction, health facility rehabilitation, road construction, tarring/grading of roads, transportation services, and agricultural-inputs

schemes. Finally, constituency characteristics include a set of indicators of the economic dynamics of the constituency, comprising indicators of improvements in opportunities for employment, the availability of agricultural inputs,

number of buyers of agriculture produce, the availability of extension services, the availability of credit facilities, and the availability of consumer goods. All specifications include an indicator of the 'grade' of the committee under which

the project falls, which is a dummy that takes the value 1 if the committee is perceived to be of high political weight or otherwise. Figures rounded to two decimal places.

Figure 1A: Histogram of Difference Between Winner and Runner Up Vote Share

Notes: This is a histogram of the difference between the winning vote share in a constituency and that of the runner up. The sample used to

construct the histogram is those constituencies for which we observe the implementation of public projects.

Figure 1B: Map of Difference Between Winner and Runner Up Vote Share

Notes: This is a choropleth map of the difference between the winning vote share in a constituency and that of the runner up. The sample

used to construct the histogram is those constituencies for which we observe the implementation of public projects.

02

46

810

Per

cent

0 .2 .4 .6 .8 1Difference between winner and runner up vote share