WPS8595


Policy Research Working Paper                    8595




  Management and Bureaucratic Effectiveness
           Evidence from the Ghanaian Civil Service

                               Imran Rasul
                              Daniel Rogger
                             Martin J.Williams




Development Economics
Development Research Group
September 2018
Policy Research Working Paper 8595


  Abstract
 A burgeoning area of social science research examines                              practices is partly explained by bureaucrats having to multi-
 how state capabilities and bureaucratic effectiveness shape                        task, interactions with their intrinsic motivation, their
 economic development. This paper studies how the man-                              engagement in influence activities, and project character-
 agement practices of civil service bureaucrats correlate to the                    istics such as the clarity of targets and deliverable outputs.
 delivery of public projects, using novel data from the Ghana-                      The paper discusses the interplay between management
 ian Civil Service. This paper combines hand-coded progress                         practices and corruption, alternative methods by which to
 reports on 3,600 projects with a management survey in                              measure management practices in organizations, and the
 the government ministries and departments responsible for                          external validity of the results. The findings suggest that the
 these projects. The analysis finds that management matters:                        focus of many civil service reform programs on introducing
 practices related to autonomy are positively associated with                       stronger incentives and monitoring may backfire in some
 project completion, yet practices related to incentives/mon-                       organizations, and that even countries with low levels of
 itoring of bureaucrats are negatively associated with project                      state capability may benefit by providing public servants
 completion. The negative impact of incentives/monitoring                           with greater autonomy in some spheres.




 This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the
 World Bank to provide open access to its research and make a contribution to development policy discussions around the
 world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors
 may be contacted at drogger@worldbank.org.




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
  management and bureaucratic effectiveness:
    evidence from the ghanaian civil service⇤
            imran rasul                  daniel rogger                  martin j.williams†




          JEL Classification: J33, O20.


   ⇤
     We gratefully acknowledge financial support from the International Growth Centre [1-VCS-VGHA-VXXXX-
33301] and the World Bank’s i2i trust fund financed by UKAID. We thank Julien Labonne, Anandi Mani, Arup
Nath, Simon Quinn, Ra↵aella Sadun, Itay Saporta-Eksten, Chris Woodru↵ and seminar participants at Oxford,
Ohio State, PMRC, and the EEA Meetings for valuable comments. Jane Adjabeng, Mohammed Abubakari, Julius
Adu-Ntim, Temilola Akinrinade, Sandra Boatemaa, Eugene Ekyem, Paula Fiorini, Margherita Fornasari, Jacob
Hagan-Mensah, Allan Kasapa, Kpadam Opuni, Owura Simprii-Duncan, and Liah Yecalo-Tecle provided excellent
research assistance, and we are grateful to Ghana’s Head of Civil Service, Nana Agyekum-Dwamena, members
of the project steering committee, and the dozens of Civil Servants who dedicated their time and energy to the
project. This study was approved by UCL’s Research Ethics Committee. All errors remain our own.
   †
     Rasul: University College London and the Institute for Fiscal Studies [i.rasul@ucl.ac.uk]; Rogger: World Bank
Research Department [drogger@worldbank.org]; Williams: Blavatnik School of Government, University of Oxford
[martin.williams@bsg.ox.ac.uk].
1       Introduction
A burgeoning area of social science research examines how state capabilities shape economic de-
velopment [Besley and Persson 2011, Acemoglu and Robinson 2012, Pepinksy et al. 2017]. Much
attention has been placed on understanding the e↵ectiveness of government bureaucracies, a key
component of state capability. Bureaucratic e↵ectiveness matters for macroeconomic outcomes
such as growth and inequality, and for microeconomics given the presumption that successful
micro-evaluations of interventions can lead to them being e↵ectively scaled-up by government.
    Despite the importance of bureaucratic e↵ectiveness, economic analysis of public sector agents
has largely focused on the selection, retention, and response to incentives of frontline public sec-
tor workers, or ‘street-level’ bureaucrats. In contrast, we contribute to a nascent body of work
studying the vital middle-tier of bureaucrats working in central government civil services, who
are responsible for policymaking, administrative functions and for supervising frontline workers
[Bertrand et al. 2017, Finan et al. 2017]. Specifically, we study whether the management practices
that this professional class of civil servant bureaucrats operate under correlate to the e↵ective de-
livery of projects their Ministries and Departments are responsible for.1 We do so in the context of
a lower middle-income country, Ghana, at a time when many developing countries are engaged in
reforming bureaucracies in line with the ‘good governance’ agenda of the World Bank and United
Nations [Goldfinch et al. 2012, Hasnain et al. 2012].
    Our analysis focuses on the professional grades of technical and administrative bureaucrats
within 45 Ministries and Departments. To quantify the delivery of public projects, we exploit the
fact that each organization is required to provide quarterly and annual progress reports. These
detail targets and achievements for individual projects the organization is charged to deliver.
Progress reports cover the entire range of bureaucratic activity, including project types that have
been much studied, such as procurement and infrastructure, but also areas of activity that have
been far less subject to quantitative study, such as policy development, advocacy, human resource
management, budgeting and regulatory design. We use these progress reports to identify 3628
projects underway during 2015, and to hand-code each project’s initiation and full completion.
    To measure the management practices bureaucrats operate under, we follow the methodologi-
cal approach of Bloom and Van Reenen [2007], and Bloom et al. [2012] (henceforth BSVR). We
adapt their management surveys to the Ghanaian public sector setting, taking account of insights
from the public administration literature [Rose-Ackerman 1986, Wilson 1989]. We elicited de-
tails on management practices in an individual survey administered to bureaucrats. The surveys
    1
     There are well-documented reasons why management practices could have di↵erent impacts on middle-tier
bureaucrats than for lower-tier frontline public sector workers [Dixit 2002]. The selection, objectives and motivations
of middle-tier of bureaucrats might di↵er. The nature of work for middle-tier bureaucrats might also di↵er: they
might need to multi-task, and the mapping between e↵ort inputs and observable output is perhaps more uncertain.
Finally, there can be specific labor market rigidities applying to middle-tier bureaucrats, leading to di↵erent dynamic
selection e↵ects than for frontline workers.



                                                          2
enumerated senior bureaucrats in managerial roles, as well as the less senior bureaucrats they
manage. In contrast to much of the earlier work on management, we can therefore reconstruct
management practices based on alternative sets of respondents in the organization. The measure
of management practices used for our baseline analysis averages management scores over the most
senior divisional-bureaucrat reports. We later consider alternative constructs, based on all senior
bureaucrats in an organization (a top-down view of management), or as elicited from lower-tier
bureaucrats being managed (a bottom-up view of management).
     For each organization we construct two dimensions of management practice: (i) the autonomy
provided to bureaucrats; (ii) the provision of incentives and monitoring of bureaucrats. The
essence of the BSVR approach is to capture a holistic set of practices along each dimension,
rather than coding very formal and specific rules or compensation scheme structures in place.
Importantly, this approach aims to capture the actual management practices being used in practice,
not an idealized version of what the organization’s practices are supposed to be on paper.
     The autonomy index captures the extent to which bureaucrats of all levels are empowered
to make meaningful contributions into policy formulation and implementation processes, and the
flexibility with bureaucrats can use their discretion in responding to project peculiarities and
introducing innovations. There are long-standing views in the public administration literature on
the importance of autonomy. As Rose-Ackerman [1986] describes, at one extreme lies the view that
public agencies ought to delegate decision making to bureaucrats, relying on their professionalism
and resolve to deliver public services [Simon 1983]. At the other extreme lies the Weberian view
that, because bureaucrats are self-interested, only an entirely rules-based system that leaves little
to the individual judgement of bureaucrats can ensure consistent and acceptable levels of public
service.
     The incentives/monitoring management index captures the extent to which an organization
collects indicators of project performance, how these indicators are reviewed, and whether bu-
reaucrats are rewarded for achievements reflected in these indicators. The use of performance
incentives are a central part of ‘New Public Management’ agenda that has swept through govern-
ment bureaucracies over the past three decades, yet the evidence base on such reforms remains
thin, and mostly based on evidence from frontline workers rather than the kinds of middle-tier
civil servant working in central Ministries that we focus on here.2 A priori the correlation between
bureaucratic output and the provision of such incentives in public sector settings is uncertain be-
cause: (i) bureaucrats might need to exert multiple e↵ort types, not all of which are measurable;
(ii) the process by which inputs are converted to outputs is uncertain; (iii) there can be competing
views on the right way to implement bureaucratic outputs; (iv) bureaucratic objectives are not
clear cut; and (v) performance incentives might crowd out the intrinsic motivation of those self-
selected into the public sector [Gneezy and Rustichini 2000, Benabou and Tirole 2006, Francois
   2
    Perry et al. [2009] review 57 studies on pay for performance in the public sector and conclude ‘pay-for-
performance continues to be adopted but persistently fails to deliver’. Finan et al. [2017] overview recent evidence.


                                                         3
and Vlassopoulos 2008, Besley and Ghatak 2018].
    Our research design exploits the fact that multiple organizations conduct similar project ac-
tivities. We thus measure the partial correlation of management practices with public service
delivery within project type, namely, conditioning on project type fixed e↵ects and so accounting
for unobserved heterogeneity in bureaucracies arising from the composition of projects they are
tasked to implement. Our key results are as follows.
    First, codifying the progress reports reveals the importance of non-infrastructure projects in the
work of bureaucracies. The most common project type in Ghanaian central government bureau-
cracies relates to human resource management. Comprising 29% of all projects, this reinforces the
importance of understanding whether the management practices bureaucrats operate under cor-
relate to bureaucratic e↵ectiveness. 23% of projects relate to policy advocacy and development,
while the two traditional areas of quantitative study, infrastructure and procurement, together
comprise around a third of projects.
    Second, management practices robustly correlate with the project completion success of civil
service bureaucracies. The two dimensions of management practice emphasized by the public
administration and economics literatures, autonomy and incentives/monitoring, robustly correlate
to project completion rates. However, they have opposing correlations with project delivery: a one
standard deviation increase in management practices related to providing autonomy to bureaucrats
is associated with a 25% increase in the likelihood a project is initiated and a 28% increase in
the likelihood it is fully completed; in contrast, a one standard deviation increase in management
practices related to the provision of incentives or monitoring to bureaucrats is associated with a
28% decrease in the likelihood a project is initiated, and a decrease of 18% in the likelihood it is
fully completed. These magnitudes are of economic as well as statistical significance, against the
backdrop that 21% of public sector projects are never started and only 34% are fully completed.
    A chief concern in interpreting results is that management practices might be endogenous to
organizational performance, so that the negative result could in part pick up the fact that if an
organization has low project completion rates, it increases the provision of incentives/monitoring
to its bureaucrats. We address such reverse causality using an instrumental variables strategy that
exploits the fact that similar interviews on management practices had previously been undertaken
for a subset of the same civil service organizations in 2013, by one of us in a separate and indepen-
dent study [Williams 2015]. We thus instrument current practices related to incentives/monitoring
with each organization’s historic practices along the same dimension, exploiting the persistence of
management practices within organizations [Gibbons and Henderson 2013]. The exclusion restric-
tion requires that past management practices related to incentives/monitoring only impact current
project delivery through their persistent e↵ects on contemporaneous management practices related
to incentives/monitoring. These IV estimates continue to find a statistically significant negative
e↵ect of incentives/monitoring on project completion, and suggest that the OLS results may ac-
tually underestimate the negative conditional e↵ects of incentives/monitoring. We consider this

                                                  4
suggestive evidence that our core results are not driven by reverse causality.
    To understand the mechanisms driving this negative e↵ect, we then investigate four channels
suggested by incentive theory through which the detrimental impacts of management practices
related to incentives/monitoring for bureaucrats might occur. First, bureaucrats might need to
multi-task in their work environment, needing to exert some types of e↵ort that can be labelled
as ‘processing’ and does not directly increase project completion rates, while also needing to exert
more productive types of e↵ort that do raise project completion rates. Our management practice
measure might then capture an incentive system that places excessive regulatory burden or ‘red
tape’ on bureaucrats that has been argued to cause bureaucrats to mis-allocate e↵ort towards
processing activities [Kelman 1990, Baker 2002]. Consistent with this, we find the negative partial
correlation between incentives/monitoring and project completion rates is even more negative when
bureaucrats need to coordinate/negotiate internally with other bureaucrats in their organization,
or need to coordinate with external stakeholders.
    Second, incentives/monitoring might crowd out the intrinsic motivation of bureaucrats. To
investigate this, as part of our civil servants survey we measured the public service motivation
(PSM) of bureaucrats, using an abbreviated version of the standard Perry scale [Perry 1996]. We
find the negative impact of management practices related to incentives/monitoring are partly ame-
liorated when bureaucrats in the organization score higher on the PSM dimensions of ‘compassion’
and ‘public interest’. As in Ashraf et al. [2014], our evidence thus suggests incentives/monitoring
crowd-in the e↵ort of intrinsically motivated bureaucrats, partly (but not entirely) ameliorating
the negative e↵ects of practices related to incentives/monitoring.
    Third, management practices related to incentives/monitoring might be subject to subjective
performance evaluation (SPE). While SPE has the benefit of being based on a more holistic set of
assessments, such subjective assessments also give rise to other biases and dysfunctional responses,
especially the desire of agents to engage in influencing activities to curry favor with superiors
[Milgrom 1988, Milgrom and Roberts 1988]. If so, the increased use of such mis-targeted incentives
or key performance indicators can lead bureaucrats reallocating e↵ort towards non-productive
tasks, reducing project completion rates. We examine this hypothesis using two measures of
the social connections between non-senior bureaucrats to their senior managers: whether they
overlapped in time at university, and whether they belong to the same ethnic group. Along both
dimensions of connectedness, we find the negative impact of incentives/monitoring on project
completion becomes significantly worse.
    Fourth, there might be specific project characteristics, such as the clarity of their target or
output, that impact the ability of organizations to put together a well-designed set of incen-
tive/monitoring practices. We find there is a far more detrimental impact of practices related to
incentives/monitoring for projects that have poorly defined targets, that might well reflect incen-
tive misalignment. On the clarity of the output to be delivered, we find the absolute impacts of
both dimensions of management practice are far larger for projects with high output clarity. This

                                                 5
is again consistent with incentive misalignment: when outputs are well defined ex post, granting
bureaucrats discretion over how to implement projects is likely to be more e↵ective since abuse of
that discretion will be easier to detect ex post. The negative association of incentives/monitoring
with output, conditional on autonomy, is then likely to be greater.
    Finally, we discuss three remaining important issues related to management practices in bu-
reaucracies: (i) the interplay between management practices and corruption in service; (ii) alter-
native methods by which to measure management practices in organizations; and (iii) the external
validity of our findings linking management practices and project delivery.
    On corruption, a long-standing literature in public administration emphasizes that civil ser-
vants might pursue their own self-interest [Wilson 1989]. This more negative view of bureaucrats
spurs us to explore how the partial correlations between project completion rates and management
practices are mediated through perceptions of corruption. We find that for multiple measures of
corruption, there is no evidence that the provision of autonomy to bureaucrats leads to a signif-
icantly lower likelihood of project completion. This suggests there are few bureaucrats on the
margin of being corrupt, for whom small changes in management practices related to autonomy
lead to large changes in corrupt behavior. At the same time, we find some evidence that the
negative consequences of corrupt practices are partially o↵set by the provision of incentives or
monitoring of bureaucrats.
    On measuring management practices, we find that there are di↵erences between top-down and
bottom-up views of management. While our core results are robust to these di↵erent measure-
ment approaches, our results highlight that there remains scope for research to understand why
bottom-up views of management do not coincide with top-down views, and whether organizational
discord measured this way might capture important elements of organizational functioning and be
predictive of their performance.
    The final issue we turn to is that of the external validity of our findings. To do so it is most
natural to compare our findings from Ghana to our earlier work on management practices for
middle-tier bureaucrats in the Federal Civil Service of Nigeria [Rasul and Rogger 2018, henceforth
RR]. A common set of results emerge across the Ghanaian and Nigeria contexts: (i) the provision of
autonomy is robustly positively correlated with project initiation, full completion and completion
rates; and (ii) incentives/monitoring of bureaucrats is robustly negatively correlated with project
initiation, full completion and completion rates in both settings. Moreover, the estimates show
similar e↵ect sizes of both dimensions of management practice on the initiation and full comple-
tion margins, in which the two settings are most comparable. While not an exact replication,
establishing robust findings across similar contexts underpins the external validity of any given
study, and so moves the knowledge frontier closer to establishing stylized facts.
    Overall, our results point to new directions for theoretical research to better understand the
contracting environment in public bureaucracies [Dixit 2002], and lay out an agenda for future
research using field experiments to establish causal impacts of management practices in bureaucra-

                                                6
cies on public service delivery and state capabilities more broadly. Our findings also suggest that
the overwhelming focus of many civil service reform programs on introducing stronger incentives
and monitoring may backfire in at least some public sector organizations, and that even countries
with low overall levels of bureaucratic e↵ectiveness or state capability may benefit by providing
their public servants with greater autonomy in some spheres.
    The paper is organized as follows. Section 2 describes the Ghanaian context, data sources and
key measures of project completion and management practices. Section 3 presents our empiri-
cal method and main results. Section 4 unpacks the drivers behind the robust negative partial
correlation we find between project completion rates and management practices related to incen-
tives/monitoring. Section 5 further discusses the interplay between management practices and
corruption, alternative approaches to measuring management practices, and the external validity
of our findings. Section 6 concludes. The Data Appendix provides details on data sources and
further robustness checks.


2       Context and Data
Ghana is a West African state home to 28 million individuals. Its central government bureaucracy
is structured along lines reflecting its British colonial origins, where Ministries are the central
coordinating authority. We study 45 Ministries and Departments in the Civil Service. These are
all located in Accra, but have remit over public projects implemented nationwide.3 Ministries and
Departments are overseen by the O ce of the Head of Civil Service (OHCS), that is responsible for
personnel management and performance within the civil service. OHCS coordinates and decides on
all hiring, promotion, transfer, and (in rare circumstances) firing of bureaucrats across the service.
While OHCS develops and promulgates o cial management regulations and processes, Ministries’
and Agencies’ compliance with these is imperfect, with the result that actual management practices
are highly variable across organizations.
     Our analysis of bureaucrats focuses on the professional grades of technical and administrative
o cers within these Ministries and Departments. We therefore exclude grades that cover cleaners,
drivers, most secretaries, etc. On average, each organization employs 64 bureaucrats of the type we
study (those on professional grades), although there is a considerable degree of variation in their
size: the organization at the 25th percentile of employee size employs 33 bureaucrats, while the
organization at the 75th percentile employs 72 bureaucrats. We designate bureaucrats as being at
either a senior or non-senior level. Seniors are those that classify themselves as a ‘Director (Head
of Division) or Acting Director’ or as a ‘Deputy Director or Unit Head (Acting or Substantive)’.
    3
    Ghana makes a distinction between the Civil Service and the broader Public Service, which includes dozens
of autonomous agencies under the supervision but not direct control of their sector ministries, as well as frontline
implementers such as the Police Service, Education Service etc. Our sample is restricted to the Civil Service and
excludes non-Civil Service government organizations.



                                                         7
By this definition, the span of control of senior bureaucrats over non-seniors is around 4.52, but
again there is considerable variation across Ministries.4
    Around 45% of bureaucrats are women, 70% have a university education, and 31% have a
postgraduate degree (seniors are more likely to be men, and have a postgraduate degree). As in
other state organizations, civil service bureaucrats enjoy stable employment once in service: the
average bureaucrat has 14 years in service, with their average tenure in the current organization
being just under 9 years. This reflects some of the rigidities in the labor market for bureau-
crats: appointments are made centrally by OHCS, bureaucrats enjoy secure tenure and transitions
between bureaucracies are infrequent.
    Our analysis is based on two data sources. First, we hand-coded quarterly and annual progress
reports from Ministries and Departments, covering projects ongoing between January and Decem-
ber 2015. As detailed below, these reports enable us to code the individual projects under the
remit of each organization, and the extent to which they are initiated or successfully completed.
Second, we surveyed 2971 bureaucrats from all 45 civil service organizations over the period August
to October 2015. Our civil servant survey covers 75% of all professional bureaucrats in service,
with 20% of interviewees being seniors. As detailed below, civil servants were questioned on top-
ics including their background characteristics and work history in service, job characteristics and
responsibilities, engagement with stakeholders outside the civil service, perceptions of corruption
in the service, and their views on multiple dimensions of management practices. It is this last
survey module from which we derive measures of management practice for each organization.


2.1     Coding Projects and Completion
Worldwide, civil service bureaucracies di↵er greatly in whether and how they collect data on their
performance. Unlike data related to the macroeconomy, households, firms, schools or labor mar-
kets, central statistical agencies are typically not involved in measuring government e↵ectiveness,
and few international standards exist to aid cross country comparisons. To therefore quantify
the delivery of public sector projects in our context, we exploit the fact that each Ghanaian civil
service organization is required by OHCS to provide quarterly and annual progress reports. Orga-
nizations di↵er in their reporting formats and coverage, and some either did not produced reports
for this time period or produced them in a format that was infeasible to code. We are thus able
to use the progress reports of 30 Ministries and Departments (and our civil servant survey covers
2247 bureaucrats in these 30 organizations). Figure A1 provides a snapshot of a typical progress
report and indicates the information coded from it, and the Data Appendix discusses the coding
   4
    In Ghana, grades of technical and administrative bureaucrats are o cially referred to as ‘senior’ o cers while
grades covering cleaners, drivers etc. are referred to as ‘junior’ o cers, regardless of their tenure or seniority.
While we restrict our sample to ‘senior’ o cers in the formal terminology, throughout we use the terms senior and
non-senior in their more colloquial sense to refer to hierarchical relationships within the professional grades.




                                                        8
process in detail.5
    Progress reports cover the entire range of bureaucratic activity. While some of these projects
are public-facing, others are purely internal functions or intermediate outputs. We were able to
use quarterly and annual progress reports from bureaucracies to identify 3628 projects underway
during 2015. The projects undertaken by each organization in a given year are determined through
an annual planning and budgeting process jointly determined between: (i) the core executive,
mainly the Ministry of Finance and the sector minister representing government priorities; (ii) the
organization’s management, based in large part on consultatively developed medium-term plans;
and (iii) ongoing donor programs. This schedule of projects is formalized in the organization’s
annual budget (approved by Parliament) and annual workplan. The quarterly and annual reports
that we use to code project completion thus detail the projects that the organization’s workplan
committed the organization to working on during the time period of study.
    Progress reports detail targets and achievements for individual projects the organization is
charged to deliver. The Appendix describes how we hand-coded and harmonized the information
to measure projects and project completion across the Ghanaian civil service. Three key points are
of note in relation to this process. First, each quarterly progress report was codified into project
line items using a team of trained research assistants and a team of civil servant o cers seconded
from the Management Services Department (MSD), an organization under OHCS tasked with
analyzing and improving management in the civil service. MSD o cers are trained in management
and productivity analysis and frequently review organizational reports of this nature, making them
ideally suited to judging project characteristics and completion.
    Second, coders were tasked to record project completion on a 1-5 scoring grid, where a score
of one corresponds to, “No action was taken towards achieving the target”, three corresponds
to, “Some substantive progress was made towards achieving the target. The output is partially
complete and/or important intermediate steps have been completed”, and a score of five corre-
sponds to, “The target for the output has been reached or surpassed.” Projects can be long-term
or repeated (e.g. annual, quarterly) projects. There were at least two coders per project. Given
the tendency for averaging scores across coders to reduce variation, we use the maximum and
minimum scores to code whether projects are fully complete/never initiated respectively (we later
show robustness of our main result to alternative algorithms to aggregate scores).
    Third, as progress reports are self-compiled by bureaucracies, an obvious concern is that low
performing bureaucracies might intentionally manipulate their reports to hide the fact. To check
the validity of progress reports, we matched a sub-sample of 14% of projects from progress reports
to project audits conducted by external auditors through a separate exercise undertaken by OHCS.
Auditors are mostly retired civil servants, overseen by OHCS, and they obtain documentary proof
of project completion. For matched projects, 94% of the completion levels we code are corroborated
  5
    Where an organization produced multiple reports during this time (e.g. a mid-year report and an annual
report), we selected the latest report produced during the year to include in our sample.


                                                    9
based on the qualitative descriptions of completion in audits.6


2.2     What Do Bureaucrats Do?
The data reveals the importance of non-infrastructure projects in the work of bureaucracies. Figure
1A shows the most common project type in Ghanaian central government bureaucracies relates
to human resource management (‘monitoring, training and personnel management’). Comprising
29% of all projects, this reinforces the importance of understanding whether the management
practices bureaucrats operate under correlate to bureaucratic e↵ectiveness. The Figure also shows
that 23% of projects relate to policy advocacy and development, while the two traditional areas of
quantitative study, infrastructure and procurement, together comprise around a third of projects.
    In Figure 1B, each bar corresponds to a project type, and within-bar colors signify projects
conducted by a given organization. This reveals that: (i) the same project type is implemented
by multiple organizations; (ii) each organization is tasked to implement multiple project types.
Hence a lack of specialization is a fundamental feature of the Ghanaian civil service. Such a lack of
specialization was also an inherent feature documented in RR in the context of the Nigerian Civil
Service. We thus exploit a research design that measures the partial correlation of management
practices with public service delivery within project type, so accounting for unobserved hetero-
geneity in bureaucracies arising from the composition of projects they are tasked to implement.
If project types vary in the optimal set of management practices, this lack of specialization leaves
more scope for management practices to matter on the margin (even conditional on project type
fixed e↵ects). This situation can persist given the rigidities in the labor market for bureaucrats,
which slows down the di↵usion of information on best management practices.
    Table 1 and Figure 2 show how bureaucratic output varies by project type. Table 1 reiterates
that all project types are implemented by the majority of organizations. On completion rates,
the first bar in Figure 2 shows that 21% of projects are never started (i.e. are recorded as a one
on the scoring card); 34% are fully completed. The variation on the extensive margin of project
completion varies by project type. For example, procurement projects are more than twice as likely
not to be initiated as permits and regulation projects. However, there is considerable variation in
average completion rates even within project types (Column 5, Table 1).
    Figure 3 focuses on the variation in completion rates across civil service organizations. To
quantify this variation, we note that the 75th percentile organization has an average completion
rate 22% higher than 25th percentile organization. This variation occurs despite the fact that
multiple organizations engage in similar project types, they are assigned hires from the same pool
of incoming bureaucrats, and they are located close to each other in Accra.7 Figure 3 further
   6
     Among the handful of non-corroborated projects, the lowest “true” completion rate was a 3, indicating that
the rare instances of misreporting were relatively minor.
   7
     As we use the minimum and maximum of reports for the extensive margin of project output, it is possible that
the percentage of initiated projects is below that for completed projects, as occurs in one organization.


                                                       10
highlights the weak link between the proportion of projects initiated by an organization, and
the proportion of projects actually completed. Our empirical analysis considers the impacts of
management practices on the initiation, completion and progression of projects.
   Table 2 presents descriptive evidence on the public service delivery of the 10 organizations that
implement the most projects. This reiterates the lack of specialization in the kinds of projects
that each organization is tasked to implement (among these 10 organizations, they are tasked
with between 4 and 7 unique project types). The Table also reiterates that there is huge variation
across organizations in their measured bureaucratic e↵ectiveness. This all suggests there might be
important organizational factors correlating with this variation in e↵ectiveness. Our focus is on
one such factor: the management practices civil service bureaucrats operate under.


2.3    Measuring Management
We follow BSVR’s approach to measuring management practices, but first adapt their procedures
and survey to the Ghanaian public sector setting. Survey team leaders were recruited from the
private sector, with an emphasis on previous experience of survey work in Ghana. We worked
closely with the team leaders to give them an appreciation and understanding of the practices
and protocols of the public service. We then collaborated with OHCS to recruit junior public
o cials with pre-existing experience of public sector work to act as our enumerators. The Head
of Service ensured their commitment to the survey process by stating the research team would
monitor enumerator performance and that these assessments would influence future posting op-
portunities. We trained the team leaders and public o cials jointly, including intensive practice
interview sessions, before undertaking the first few interviews together. This approach allowed us
to capitalize on both the experience of our private-sector team leaders and the knowledge of our
public sector enumerators.
    Over the period to August to November 2015, our enumerators interviewed 2971 bureaucrats
employed at 45 organizations. This constitutes 98% of all eligible sta↵ in these organizations,
with the remainder mostly having been out of the o ce during the survey period. Interviews were
conducted in person, but were double-blind in the sense that interviewers had never worked in the
organizations in which they were interviewing and did not know the names of their interviewees,
and likewise interviewees did not know the names of their interviewers.
    We adapted BSVR’s methodology to cover six dimensions of management practice: roles, flex-
ibility, incentives, monitoring, sta ng and targets. Table A1 details each management related
question, by topic, as well as the 1-5 scoring grid used by our enumerators for each question.
There are 14 questions in total over the six topics.
    To provide a sense of the holistic nature of these questions, a question on practices related
to bureaucratic flexibility was, “Does your division make e↵orts to adjust to the specific needs
and peculiarities of communities, clients, or other stakeholders?”. Following BSVR, enumerators


                                                11
would then probe respondents’ responses and ask for examples, and then score responses on a
continuous 1-5 scale, where for indication the scoring grid described a score of one as corresponding
to circumstances where, “The division uses the same procedures no matter what. In the face
of specific needs or community/ client peculiarities, it does not try to develop a ‘better fit’ but
automatically uses the default procedures.”; a score of three corresponded to, “The division makes
steps towards responding to specific needs and peculiarities, but stumbles if the specific needs are
complex. Often, tailoring of services is often unsuccessful.”; and a score of five corresponded to,
“The division always redefines its procedures to respond to the needs of communities/ clients. It
does its best to serve each individual need as best as it can.”
    We elicited information on management practices in individual surveys administered to bu-
reaucrats. The surveys enumerated both those in managerial roles, as well as bureaucrats being
managed by seniors. In contrast to much of the earlier work on management, we can therefore
reconstruct management practices based on alternative sets of respondents in the organization.
The measure of Ghanaian management practices we use for our core analysis averages manage-
ment scores over the most senior divisional-bureaucrat reports. The median (mean) number of
senior managers per organization is 13 (20). This retains the closest similarity to the way in
which management practices were elicited in RR. However, we later show robustness to alterna-
tive constructs for management practices in organizations, based on all senior bureaucrats in an
organization (a top-down view of management), or as elicited from lower-tier bureaucrats being
managed (a bottom-up view of management).
    The answers to questions on roles and flexibility are then combined to produce a measure
of management practices related to autonomy, denoted CS-autonomy. Similarly, the answers to
questions on incentives and monitoring scores are combined to produce a CS-incentives/monitoring
measure of management practices. The answers to questions on sta ng and targeting topics are
combined into a CS-other measure of management practices. The scores on each practice are
converted into normalized z-scores by taking unweighted means of the underlying z-scores (so are
continuous variables with mean zero and variance one by construction), where both are increasing
in the notion of ‘better management’. For the CS-autonomy index, we assume greater autonomy
corresponds to better management, and for the CS-incentives/monitoring measure we also assume
the provision of incentives/monitoring corresponds to better management practices (as suggested
to be so in private sector settings [Prendergast 1999]).8
    The CS-autonomy and CS-incentives/monitoring management scores are positively correlated
(⇢ = .67) suggesting complementarity of these practices, but this did not have to be so. For
example, substitution between practices could occur if bureaucrats have career concerns, and so
performance incentives are not required once autonomy is provided. Alternatively, if bureaucrats
   8
    These z-scores are constructed using data for all 45 organizations we can construct management practices for
based on our civil servant survey. To address concerns over sample selection, we later show robustness of our main
finding to redefining the z-scores based only on the 30 organizations for which project completion data exists.


                                                       12
are intrinsically motivated they might need only to be provided autonomy, and the provision of
explicit incentives might crowd out their intrinsic motivation. Importantly, the partial correlations
of these two dimensions of management can still be separately estimated from each other and from
the CS-other practices index.9


2.4     Management Practices Across Civil Service Organizations
Figure 4 shows the across-organization variation in management practices, using our preferred
aggregation of survey responses over the most senior divisional-bureaucrats. As with bureaucratic
performance on projects, there is high variation in the management practices bureaucrats are sub-
ject to across organizations. For practices related to the provision of autonomy, the 75th percentile
organization has a CS-autonomy score that is 145% higher than 25th percentile organization. On
management practices related to incentives/monitoring, the 75th percentile organization has a CS-
incentives/monitoring score that is 97% higher than 25th percentile organization. To reiterate, this
variation occurs despite the fact that all organizations share the same colonial and post-colonial
structures, are governed by the same civil service laws and regulations, are overseen by the same
supervising authorities, are assigned new hires from the same pool of incoming bureaucrats each
year, and are located proximately to each other in Accra.10
    It is this variation in management practices along the dimensions of the autonomy given to
bureaucrats, and the extent to which they are provided incentives or monitoring, that we now link
to variation in project completion rates.
   9
    The CS-autonomy (CS-incentives/monitoring) index has a correlation of .59 (.70) with CS-other.
  10
    To provide practical detail on what drives this variation in management scores, we go back to consider the
raw management scores in Table A1. In relation to the component of the autonomy score based on the question,
“When senior sta↵ in your division are given tasks in their daily work, how much discretion do they have to carry
out their assignments? Can you give me an example?”, we note that 45% of organizations score between 3.75 and
5 (and so are closer to practices where, “O cers in this division have a lot of independence as to how they go
about their daily duties”, while the remaining 55% of organizations score below 3.75 (and so are closer to practices
where, “O cers in this division have no real independence to make decisions over how they carry out their daily
assignments. Their activities are defined in detail by senior colleagues or organizational guidelines.” Similarly, in
relation to the component of incentives/monitoring that asks, “Does your division use performance, targets, or
indicators for tracking and rewarding (financially or non-financially) the performance of its o cers?”, we note that
52% of organizations score between 1 and 3.75 (and so are closer to the practice where, “O cers in the division are
rewarded (or not rewarded) in the same way irrespective of their performance. The evaluation system awards good
performance in principle (financially or non-financially), but awards are not based on clear criteria/processes”, and
the remaining 48% of organizations score between 3.75 and 5 (and so are closer to practices where, “Rewards are
given as a consequence of well-defined and monitored individual achievements.”).




                                                        13
3      Empirical Method and Main Results
3.1      Method
The unit of observation is project i of type j in organization n. We estimate the following OLS
specification,

yijn =   1 CS -autonomyn + 2 CS -incentives/monitoringn + 3 CS -othern + 1 P Cijn + 2 OCn + j +✏ijn

                                                                                              (1)
where yijn is either an indicator of whether the project is initiated, whether it is fully com-
pleted (both extensive margin outcomes), or a continuous measure of the project completion
rate (the intensive margin). Management practices are measured using the CS-autonomy, CS-
incentives/monitoring and CS-other indices, and P Cijn and OCn are project and organizational
controls.11 As Figure 1B highlighted, many organizations implement the same project type j , so
we can control for project type fixed e↵ects j in (1), as well as fixed e↵ects for the broad sector
the implementing organization operates in.12
    The partial correlations of interest are 1 and 2 , the e↵ect size of a one standard devi-
ation change in management practices along the respective margins of autonomy and incen-
tives/monitoring. To account for unobserved shocks, we cluster standard errors by organization
(n), the same level of variation as management practices. In the Appendix we show robustness of
our main results to alternative levels of clustering.


3.2      Main Results
Table 3 presents our main results. Columns 1 to 4 focus on the outcome of full project completion.
A robust set of findings emerge across specifications: (i) management practices providing bureau-
crats more autonomy are robustly positively correlated with the likelihood of project completion
(b1 > 0); (ii) management practices related to the provision of incentives or monitoring to bureau-
crats are robustly negatively correlated with the likelihood of project completion (b2 < 0). The
  11
      Project controls comprise project-level controls for whether the project is regularly implemented by the or-
ganization or a one o↵, whether the project is a bundle of interconnected outputs, and whether the division has
to coordinate with actors external to government to implement the project. Capital controls comprise a count of
the number of interviews undertaken, which is a close approximation of the total number of employees. General
controls comprise organization-level controls for the share of the workforce with degrees, the share of the workforce
with postgraduate qualifications, and the span of control. Following BVSR, we condition on ‘noise’ controls related
to the management surveys. Noise controls are averages of indicators of the seniority, gender, and tenure of all
respondents, the average time of day the interview was conducted and of the reliability of the information as coded
by the interviewer.
   12
      Project type fixed e↵ects relate to whether the primary classification of the project is listed as Advo-
cacy and Policy Development, Financial and Budget Management, ICT Management and Research, Monitor-
ing/Training/Personnel Management, Physical Infrastructure, Permits and Regulation, or Procurement. Sector
fixed e↵ects relate to whether the project is in the administration, environment, finance, infrastructure, secu-
rity/diplomacy/justice or social sector.


                                                         14
remaining Columns show similar partial correlations of both dimensions of management practice
with project initiation (Column 5) and the continuous measure of project completion rate (Col-
umn 6). On magnitudes, taking our preferred specification (1) shown in Column 3, a one standard
deviation increase in management practices related to CS-autonomy increases the likelihood a
project is initiated by 25%, and it increases the likelihood it is fully completed by 28%; a one stan-
dard deviation increase in management practices related to CS-incentives/monitoring decreases
the likelihood a project is initiated by 28%, and it decreases the likelihood it is fully completed
by 18%. These magnitudes are of economic as well as statistical significance: recall the backdrop
here is that 21% of projects are never started and only 34% are fully completed.13
    Our core finding thus confirms the two dimensions of management practice emphasized by the
public administration and economics literatures do indeed robustly correlate to e↵ective public
service delivery in the Ghanaian context. The positive correlation of CS-autonomy with project
completion supports the notion bureaucracies could delegate some decision making to civil ser-
vants, relying on their professionalism and resolve to deliver public services. The evidence is less
supportive of the notion that when bureaucrats have more agency, they are more likely to pursue
their own, potentially corrupt, objectives that diverge from societal interests. We return to the
issue below, when we discuss the interplay between management and corruption in service.
    The negative partial correlation between project completion rates and management practices
related to the provision of incentives and monitoring of bureaucrats, runs counter to a body of
evidence from private sector settings. As described earlier, evidence on the impacts of performance-
related incentives in public sector settings is mixed (and often based on frontline workers or street-
level bureaucrats) [Perry et al. 2009, Banerjee et al. 2014, Khan et al. 2018]. Moreover, Finan et
al. [2017] review the evidence base on the impacts of monitoring in public organizations, including
studies highlighting how this can often lead to gaming/circumventing monitoring systems. Our
findings reinforce the evidence base suggesting the possibility that management practices related
to incentives and monitoring negatively correlate to outputs of the professional tier of civil service
bureaucrats across contexts.
    Appendix Table A2 provides a battery of checks on our baseline estimates. These show the
results to be robust to alternative codings of completion rates, samples, estimation methods,
and fixed e↵ects specifications.14 Appendix Table A3 further shows the results to be robust to
  13
      Given the management practices are positively correlated to each other, this also implies that only controlling
for incentives/monitoring leads b2 to be biased upwards as CS-autonomy is omitted. Similar biases arise from a
traditional public administration perspective that might only focus on the provision of autonomy.
   14
      More precisely, we first redefine project completion rates to be the average of the codings of the two enumerators
designated to each progress report. The result in Column 2 shows the baseline results continue to hold even when
we reduce the variation in completion rates this way. Column 3 excludes projects implemented by the largest orga-
nization in terms of number of projects; Column 4 excludes the five smallest organizations by number of projects.
Columns 5 and 6 exclude organizations at the top and bottom of the CS-autonomy and CS-incentives/monitoring
management scales respectively. Column 7 uses only the 30 organizations for whom we have coded project comple-
tion data for, to define the management z-scores (and so does not define the z-scores based on all 45 organizations
for which these scores are available based on our civil servant survey). Column 8 reports the result of estimation


                                                         15
alternative clusterings of the standard errors, including robust standard errors, allowing them to
be clustered by project type within organization (so at the jn level), by project type within sector,
and by sector. In all cases, the coe cients of interest, 1 and 2 , remain precisely estimated and
statistically di↵erent from zero at conventional significance levels (p < .05 throughout).


3.3     IV
Before digging deeper into what drives the negative partial correlation between project completion
rates and incentives/monitoring of bureaucrats, we first present additional evidence that this might
represent a causal relationship. A chief concern is that management practices are endogenous to
organizational performance, so the negative result could reflect that if an organization has low
project completion rates, it increases the provision of incentives/monitoring to its bureaucrats.
    We address such reverse causality using an IV strategy that exploits the fact that similar
interviews on management practices had previously been undertaken for a subset of the same
organizations in 2013, by one of us in a separate and independent study [Williams 2015]. While
these interviews were primarily qualitative and on a much smaller scale (based on one or two senior
bureaucrats in each organization), they also took a similar semi-structured format and adapted
the approach of BSVR, coding the quality of each management practice on a 1-5 scale, with overall
organization z-scores compiled in the same way.
    Table A4 details the questions on management practices administered in the 2013 interviews.
While the same six topic areas were covered as in our 2015 survey, there is a far more limited set
of questions on the topic of bureaucrat roles, which meant it was not possible to reconstruct a
measure of CS-autonomy from the 2013 data. We therefore focus on exploiting the earlier measure
of management practices related to the provision of incentives/monitoring, as measured two years
prior to the main management and project completion measures used for the core analysis. We
instrument current practices related to incentives/monitoring with the historic practices along the
same dimension of management. The exclusion restriction requires that past management practices
related to incentives/monitoring only impact current project delivery through their persistent
e↵ects on management practices on this dimension.
    Table 4 shows the results. Column 1 replicates our baseline OLS specification in the subsample
of 18 organizations for which management practices related to incentives/monitoring are available
from 2013 (and so can be used for the IV strategy).15 Our baseline results continue to hold,
although the e↵ect sizes are smaller than in the full sample (b1,OLS = .04, b2,OLS = .09), but
are both estimated with similar if not greater precision. Column 2 shows the first stage and
demonstrates the strength of the instrument: the first stage F-statistic is over 100. There is high
a specification analogous to (1) but using a fractional regression to account for the fact that project completion
rates lie between zero and one. Finally, in Column 9 we control for project-sector level fixed e↵ects (so allowing for
sector specific impacts of project types).
  15
     As we only have 18 organizations in these specifications, we do not control for general and capital controls.


                                                         16
persistence in management practices related to incentives/monitoring over time: the first stage
correlation is .44 (p < 0.01). Column 3 then shows the second stage IV estimates, using project
completion as the outcome. The IV estimate on CS-incentives/monitoring is larger in absolute
value than the OLS estimate (b2,IV = .27), suggesting OLS is upwards biased and attenuated.
The remaining Columns show the IV results are also supportive of a causal interpretation in
relation to the other outcome margins of bureaucratic e↵ectiveness considered: project initiation
(Column 4) and the continuous measure of project completion (Column 5). Contrary to the
concern that reverse causality could be driving or exaggerating the negative association between
incentives/monitoring and project completion, the IV results suggest that (if anything) the OLS
estimates understate the true negative e↵ect.


4     Unpacking the Negative E↵ect of Incentives/Monitoring
Incentive theory provides many explanations why incentives/monitoring might be ine↵ective or
backfire in public sector contracting environments [Dixit 2002, Besley and Ghatak 2005, Finan et
al. 2017]. Our data allows us to study four mechanisms in more detail. First, bureaucrats might
need to multi-task in their work environment, where they need to exert some types of e↵ort that can
be labelled as ‘processing’, and do not directly increase project completion rates, while also needing
to exert more productive types of e↵ort that raise project completion rates. Our management
practice measure might then capture an incentive system that places excessive regulatory burden
or ‘red tape’ on bureaucrats that has been argued to cause bureaucrats to mis-allocate e↵ort
towards processing activities [Kelman 1990, Baker 2002]. Second, incentives/monitoring might
crowd out the intrinsic motivation of bureaucrats [Perry and Wise 1990, Benabou and Tirole
2006, Francois and Vlassopoulos 2008, Besley and Ghatak 2018]. Third, our holistic management
practices related to incentives/monitoring might pick up subjective performance evaluation (SPE).
While SPE has the benefit of being based on a more holistic set of assessments, such subjective
assessments also give rise to other biases and dysfunctional responses, especially the desire of agents
to engage in influencing activities to curry favor with seniors [Milgrom 1988, Milgrom and Roberts
1988]. If so, the increased use of such mis-targeted incentives or key performance indicators can
lead bureaucrats reallocating e↵ort towards non-productive tasks, reducing project completion
rates. Fourth, there might be specific project characteristics, such as the clarity of their target,
that impact the ability of organizations to put together a well-designed set of incentive/monitoring
practices.


4.1    Multi-tasking
Table 5 examines whether bureaucrats might face multi-tasking concerns. We do so by consid-
ering two organization-level measures of environments in which bureaucrats will need to exert

                                                  17
multiple kinds of e↵ort to push forward projects. We first examine whether the impact of in-
centives/monitoring practices varies with the ethnic fractionalization of bureaucrats in the orga-
nization. Such ethnic di↵erences internal to an organization capture di↵ering policy preferences
over how projects should be designed, prioritized, or implemented [Rasul and Rogger 2015]. In
such environments, bureaucrats might then need to exert multiple types of e↵ort to negotiate
around these concerns and get projects completed. Column 1 shows the marginal impact of in-
centives/monitoring on project completion rates is indeed even more negative if there is greater
ethnic fractionalization of bureaucrats in the organization.16
    On multi-tasking concerns relating to external e↵ort, we next examine whether the negative
impact of incentives/monitoring becomes exacerbated in organizations where a greater share of
their projects require engagement with stakeholders outside of their organization. This set of
stakeholders includes members of civil society, Ministers/Members of parliament, Member(s) of
the Metropolitan/Municipal/District Assemblies (MMDAs - local government units), the private
sector, traditional authorities, community or religious group(s), and the media. The result in
Column 2 shows the marginal impact of incentives/monitoring on project completion rates is even
more negative in organizations where there is a greater need to coordinate with such external
stakeholders.
    Both results suggest that when the environment faced by bureaucrats requires them to ex-
ert multiple forms of e↵ort to push forward projects - due either to a greater need to coordi-
nate/negotiate internally or to coordinate with external stakeholders - it becomes harder to design
and instigate e↵ective forms of incentives/monitoring. In such multi-tasking environments where
not all e↵orts can be measured or monitored to the same extent, organizations tend to have worse
outcomes because practices related to incentives/monitoring might be poorly designed and not be
holistic enough to encourage all forms of e↵ort required for successful project completion.


4.2     Intrinsic Motivation
A long established literature suggests those that self-select into public service might be relatively
more intrinsically motivated than those working in the private sector [Crowley and Smith 2014].
Performance incentives or monitoring might then be detrimental if such practices crowd out such
intrinsic motivation. To measure civil servant’s intrinsic motivation, As part of our civil servants
survey, we obtained individual measures of the public service motivation (PSM) of bureaucrats, us-
ing an abbreviated version of the standard Perry scale [Perry 1996]. This includes four sub-indices
related to motivations related to compassion, public interest, policy making and sacrifice. We then
examine whether the negative impacts of management practices related to incentives/monitoring
  16
    An established macroeconomic literature documents a negative correlation between societal diversity and
economy-wide outcomes [Easterly and Levine 1997, Alesina and La Ferrara 2005]. In Kenyan private sector set-
tings, Hjort [2014] and Macchiavello and Morjaria [2015] show ethnic divisions impact productivity due to worker
discrimination.


                                                      18
interplay with these dimensions of intrinsic motivation.17
    The results are shown in Table 6. Column 1 shows the negative impacts of incentives/monitoring
are reinforced in organizations whose bureaucrats on average score more highly on the ‘policy mak-
ing’ dimension of PSM (this sub-index measures intrinsic interest in the structures and procedures
of policymaking). Columns 2 and 3 show the negative impact of management practices related
to incentives/monitoring is partly ameliorated when bureaucrats in the organization score higher
on the PSM dimensions of ‘compassion’ and ‘public interest’. This runs counter to the notion
that incentive provision crowds out e↵orts of intrinsically motivated individuals: if anything, as in
Ashraf et al. [2014], our evidence suggests intrinsically motivated bureaucrats work harder in the
face of poorly designed incentives/monitoring practices (so as to ameliorate the negative e↵ects
of these practices). However, in neither case does this interaction fully o↵set the negative overall
e↵ect of incentives/monitoring except for a small handful of bureaucrats at the extreme top end
of the PSM distribution.18 Finally, Column 4 shows that the ‘self-sacrifice’ dimension of PSM has
no interactive e↵ect with management practices.
    Taken together our results shed new light on an old issue: the interplay between intrinsic and
extrinsic motivation. We show that there are some dimensions of intrinsic motivation that are
crowded out by practices related to incentives/monitoring, other dimensions that are crowded in,
and others that are independent. We leave a fuller exploration of the nature of this heterogeneity
to future research.


4.3     Subjective Performance Evaluation
As described in Section 2, Ghanaian bureaucrats enjoy long tenure. On the one hand, longer
serving bureaucrats might learn how best to respond to incentives by exploiting other flexibilities.
On the other hand, if bureaucrats are subject to SPE they might learn how best to engage in
influencing activities. We investigate this using two measures based on the social connectedness
between non-seniors and their managers: (i) the proportion of non-senior bureaucrats in an orga-
nization that overlapped in time at university as an undergraduate, with their most senior civil
servant; (ii) the proportion of non-senior bureaucrats in an organization in the same ethnic group
as their most senior civil servant. As Table 7 shows, along both dimensions, we find the negative
impact of incentives/monitoring on project completion becomes significantly worse. This strongly
suggests that these management practices might be capturing schemes in place that e↵ectively
  17
     In the public administration and economics literatures, the PSM scale developed in Perry [1996] is often found
to be positively associated with various measures of individual commitment, pro-social behavior, and performance
[Belle 2013, Dal Bo et al. 2013, Warren and Chen 2013, Perry 2014].
  18
     We have explored whether there are within-sample values of the interactions at which the marginal e↵ect of
CS-incentives/monitoring becomes positive. Generally, this is not the case. In the case of PSM-compassion, it is
only individuals who score at the 95th percentile of the index or above for which CS-incentives/monitoring would
have a positive e↵ect on project completion. For the PSM-public interest measure, the same is true again only
individuals who score at the 95th percentile or above.



                                                        19
allow for subjective performance evaluation or influencing activities to take place between socially
tied senior and non-senior bureaucrats [Milgrom 1988, Milgrom and Roberts 1988].


4.4     Project Characteristics
The fourth dimension we examine focuses in on the interplay between project characteristics and
management practices. As we exploit di↵erences in individual project characteristics (rather than
characteristics of organizations or bureaucrats as in Tables 5 to 7), we present the results splitting
the sample by the relevant project characteristic, and so allow the partial correlation between all
covariates (including both dimensions of management practice) and project completion to vary
with project characteristics. These characteristics are derived from the progress reports of each
organization (as detailed in Figure A1 and the Appendix).
    We first consider projects for which the quarterly targets set out in progress reports are more
or less clear. In Columns 1 and 2 of Table 8 we see that the partial correlation of CS-autonomy
to be similar across target clarities but that there is a far more detrimental impact of practices
related to incentives/monitoring for projects that have poorly defined targets (i.e. are below the
median clarity). More precisely, b2 = .23 for projects with poorly defined targets, and the point
estimate for 2 is less than half this magnitude (but still di↵erent from zero) for projects with well
defined targets. This is intuitive: for projects with poorly defined targets, the design of incentives
and monitoring schemes is harder, all else equal. Hence the more detrimental impacts on projects
with poorly defined targets reflects incentive misalignment.
    We next consider the clarity of the description of output to be delivered. Columns 3 and 4 show
the absolute impacts of both management practices are far larger for projects with high output
clarity (b1 = .37, b2 = .31), while the impacts are attenuated (but still statistically significant)
for projects with low output clarity (b1 = .17, b2 = .09). This is again consistent with incentive
misalignment: when outputs are well defined ex post, granting bureaucrats discretion over how to
implement projects is likely to be more e↵ective since abuse of that discretion will be easier to
detect ex post. Hence the negative association of incentives/monitoring with output (conditional
on autonomy) is likely to be even greater.19
  19
     We have also examined whether the complexity of projects impacts how management practices relate to project
completion. Unfortunately, our ability to precisely codify the complexity of projects from the progress reports is far
more limited than in RR. As Figure A1 and the Data Appendix make clear, we end up with a crude 0-1 measure of
complexity. We find that the partial correlation of CS-autonomy with project completion is the same for more or
less technically complex projects. The negative partial correlation of incentives/monitoring and project completion
is of similar magnitude across both project types, but is more precisely estimated for more complex projects. Again
this is in line with e↵ective incentive/monitoring schemes being harder to design when projects are more complex.




                                                         20
5      Discussion
We discuss three remaining important issues: (i) on the interplay between management practices
and corruption in service; (ii) on alternative approaches to measuring management practices; (iii)
on the external validity of our findings.


5.1     Corruption
While the recent economics literature has emphasized the importance of the intrinsic motivation
of bureaucrats, a long-standing literature in public administration emphasizes that civil servants
might pursue their own self-interest [Wilson 1989]. This more negative view of bureaucrats spur
us to explore how the correlations between project completion rates and incentives/monitoring
are mediated through perceptions of corruption among civil service organizations. Corruption in
public bureaucracies is an issue in Ghana, and in other countries at similar stages of development
(although as Figure 2 shows, the fact that 34% of projects are fully completed also suggests
corruption is not all-pervasive).20
    We use two approaches to elicit information on perceptions of corruption in service. First,
we focus on corruption by senior bureaucrats by measuring the proportion of received corrup-
tion rents (kickbacks) that are claimed to have gone to seniors in the organization. Second, we
build a measure based on bureaucrats’ reports of the proportion of recent projects and/or pro-
grams on which o cials report observing others engaging in corrupt practices. We then examine
whether the partial correlation of management practices related to autonomy and the provision
of incentives/monitoring vary with each measure of corruption.21
    Table 9 shows the results: for neither measure of corruption do we find evidence that the
provision of autonomy to bureaucrats interacts with corruption to lead to a significantly lower
likelihood of project completion. For the second measure, the negative impact of practices related
to incentives/monitoring is ameliorated if there is a greater proportion of recent projects on which
o cials report observing others engaging in corrupt practices. This suggests that the negative
consequences of corrupt practices can be partially o↵set by the provision of incentives/monitoring
of bureaucrats.
  20
     In 2015 Ghana scored at the 53rd percentile worldwide on the World Governance Indicators’s Control of
Corruption measure; Nigeria scored at the 13th percentile [World Bank 2018].
  21
     The first measure is based on asking what proportion of uno cial payments are shared with the superior
in the hypothetical scenario that, ‘Imagine that a corrupt bureaucrat extracts uno cial payments. Typically in
your organization, what proportion of the uno cial payments does s/he share with the following types or groups
of people.’ An organization-level average from across all interviews is constructed and for our whole sample, the
organization-average was 11%, but wide a relatively wide dispersion reflected in a standard deviation of 8%. The
second measure used a similar question structure, but asked o cials the proportion of projects on which they
‘observed others breaking service rules for their own benefit’. O cials are more likely to report the corrupt
behavior of others, and 16% of o cials have observed such acts by others in their organization.




                                                       21
5.2    Alternative Measures of Management Practice
Our measures of management practice at the organization level are all aggregated from individual
responses on management to our civil servant survey. The measure used for our core analysis
averages management scores over the most senior divisional-bureaucrat reports. We now consider
alternative constructs for management practices, based on all the senior bureaucrats in an orga-
nization, or as elicited from non-senior bureaucrats, namely those that are being managed. We
can thus compare and contrast top-down versus bottom-up views of management practices, an
underexplored element of the BSVR approach to measuring management.
    The results are in Table 10, where we focus on project completion. Column 1 repeats the
baseline specification from Column 3 in Table 3. Column 2 shows the findings to be robust to
constructing management practice scores from all senior bureaucrats in the organization, while
Column 3 shows them to be weaker if bottom-up views of management are used based on all
non-senior bureaucrats only. Column 4 constructs management scores based on all bureaucrats
in each organization, and these closely replicate the baseline findings in Column 1 in terms of the
magnitudes of the coe cients of interest ( 1 , 2 ) and the precision with which they are estimated.
    Given these results, we probe further where the top-down and bottom-up views of man-
agement diverge. Figure 5 shows the CDF of management scores for CS-autonomy and CS-
incentives/monitoring elicited from senior and non-senior bureaucrats. We see that non-seniors
generally report there being more autonomy than seniors (the average of the CS-autonomy score
for non-seniors is .021, while for seniors it is .003); and they also report their being more in-
centives/monitoring (the average of the CS-incentives/performance score for non-seniors is .004,
while for seniors it is .018 and seniors often report the lowest scores in this dimension).
    We then examine what individual characteristics of non-senior bureaucrats predict the dif-
ference between their raw individual management score, along each dimension of autonomy and
incentives/monitoring, and the corresponding organizational average derived from seniors. We
create a dependent variable of the di↵erence between the individual and senior scores, where the
unit of observation is bureaucrat i in organization n. We then regress this di↵erence on: (i)
bureaucrat characteristics (Xi ); and (ii) organization n fixed e↵ects.
    These results are shown in Table 11 and highlight that: (i) women bureaucrats report higher
autonomy, and a greater use of incentives/monitoring; and (ii) older bureaucrats and those with
an undergraduate degree report lower levels of autonomy and incentives/monitoring. We also
control for other bureaucrat characteristics such as their tenure, Raven test score of cognition,
their aggregate PSM score and the extent to which they share an undergraduate institution or
ethnicity with their manager/most senior bureaucrat in their division. These other characteristics
do not systematically correlate with di↵erences in management scores with seniors. These findings
all hold: (i) controlling for organization fixed e↵ects (Columns 2 and 5); and (ii) limiting attention
to those 30 organizations for which we have project completion data for (Columns 3 and 6).


                                                 22
    These results open new avenues for research, highlighting that: (i) in terms of top-down views
of management, it might be preferable to use a small number of individual surveys of senior
managers to measure management practices if it is more cost e↵ective than the consensual approach
followed in RR; and (ii) there remains scope to understand why top-down and bottom-up views
of management do not coincide, and whether organizational discord measured this way captures
elements of organizational functioning and is predictive of their performance.


5.3    External Validity
The final issue we turn to is that of the external validity of our findings linking management
practices that central civil servants are subject to, and the quantity of public services delivered.
To do so it is most natural to compare our findings from Ghana to our earlier work linking
management practices for middle-tier bureaucrats and public sector output in the Federal Civil
Service of Nigeria [Rasul and Rogger 2018]. In the Nigerian context, RR hand-coded independent
engineering assessments of project completion rates for 4700 public projects. For each of 63 civil
service organizations tasked to implement these projects (including 10 ministries and 53 other
Federal civil service organizations), RR conducted a survey among senior bureaucrats to elicit
management practices in place for middle-tier bureaucrats, also building on the methods pioneered
by BSVR.
    The comparison across Ghanaian and Nigeria settings exploits obvious similarities in terms
of civil service structures and economic environment. At the same time, the two sets of analysis
di↵er in the measurement of key variables. How projects and project completion is measured di↵ers
across contexts, and this is necessitated by the fact that civil service bureaucracies worldwide di↵er
greatly in whether and how they collect data on their own performance. Second, the precise way
in which management practices are measured also di↵ers across settings, although each approach
is anchored in the BSVR methods. The Appendix details these key data features from Nigeria.
    Table 12 repeats our core results for Ghana and Nigeria. Columns 1 to 3 refer to Ghana;
Columns 4 to 6 refer to Nigeria. A common set of results emerge across contexts: (i) CS-autonomy
is robustly positively correlated with project initiation, full completion and completion rates;
and (ii) CS-incentives/monitoring is robustly negatively correlated with project initiation, full
completion and completion rates in both settings. Moreover, the estimates show similar e↵ect
sizes of both dimensions of management practice on the initiation and full completion margins, in
which the two settings are most comparable.
    We have to be cautious in further comparing project types across contexts. As described earlier,
the bulk of tasks conducted by the Ghanaian bureaucracy relate to non-physical projects whereas
the bulk of projects RR have data on from Nigeria relate to physical infrastructure (boreholes,
roads etc.). However, overall the results suggest the impacts of management practices do seem



                                                 23
broadly similar for their respective samples.22
    Being able to replicate findings in this nascent literature is valuable because: (i) each individual
study is nearly always limited to a small number of bureaucratic organizations, especially when
examining middle-tier civil servants working in central ministries; (ii) establishing robust findings
across similar contexts underpins the external validity of any given study, and so moves the
knowledge frontier closer to establishing stylized facts; (iii) establishing similar findings using
alternative methodologies/measurement tools helps researchers collect data more cost-e↵ectively;
and (iv) where di↵erences in results have emerged, this helps focus researchers’ future attention
on such sources of heterogeneity across contexts.23


6      Conclusions
There is growing recognition that bureaucrats and bureaucracies play a key role in determining
state capability. This in turn has implications for macroeconomic outcomes and the scaling up
of e↵ective microeconomic interventions. Our work highlights the role that management plays in
driving pockets of good governance within the structure of political institutions in low-income and
relatively weak states [Leonard 2010].
    Our findings provide support to the notion that public agencies might delegate some decision
making to bureaucrats [Simon 1983]: even countries with low overall levels of bureaucratic e↵ec-
tiveness or state capability may benefit by providing their public servants with greater autonomy
in some spheres. The robust negative correlation documented between project completion rates
and management practices related to the provision of incentives and monitoring of bureaucrats,
is surprising and counter to a large body of evidence from private sector settings. We find that
the negative partial correlation of incentives/monitoring is related to bureaucrats engaging in
multi-tasking, interactions with their intrinsic motivation, influence activities and other project
characteristics such as the clarity of targets and deliverable outputs. As such, our results sound
a word of caution to the good governance agenda: the simple import of incentive/monitoring
practices from the private sector might backfire in bureaucratic settings.
    Our research points to areas for methodological advance for the measurement of management
practices in organizations. Going forward, the remains a rich agenda to study the frictions pre-
  22
      One project type for which there is reasonable overlap in definition and number across both contexts relates to
monitoring/training and personnel management (this is the modal project type in Ghana). For this project type,
we find that in Ghana the provision of autonomy to bureaucrats increases project completion and there are weak
impacts of management practices related to incentives/monitoring. This is exactly in line with the results from RR
[c.f. Figure 1] where for those kind of human resource project conducted by the Nigerian Civil Service, there was
no detrimental impact of management practices related to incentives/monitoring.
   23
      Many social and natural sciences are actively discussing the replicability of research, although there remains
no consensus on the precise meanings of reproducibility, replicability and robustness [Hamermesh 2007, Clemens
2015]. Attention is being placed on scientific replicability in the social sciences, including economics where there
has been a long-standing debate on the transparency, reproducibility and credibility of research, heightened by
evidence of p-hacking [Christensen and Miguel 2016].


                                                        24
venting organizations adopting optimal management practices. These might relate to the lack of
specialization in project types assigned to organizations, there being large fixed costs of adopting
better practices, frictions in the labor market for bureaucrats that prevent best practice from
spreading, a lack of competitive pressure enabling poorly managed organizations to survive, and
the fact that bureaucracies are driven by objectives other than maximizing project completion.
    While our results across Ghana and Nigerian settings are similar, going forward, it will be
important for researchers to understand similarities and di↵erences across such state organizations
in order to advance the literature. Bureaucracies di↵er in terms of their selection and retention
policies for bureaucrats [Dal Bo et al. 2013, Deserranno 2017, Ashraf et al. 2018], as well as
mechanisms for the public and politicians to hold public sector organizations accountable [Olken
2007, Bjorkman and Svensson 2009].
    Building on the literature examining cross-country di↵erences in bureaucratic e↵ectiveness,
our analysis pushes forward the frontier to understand within-country variation in e↵ectiveness.
Indeed, a nascent body of work has now started to examine the impacts of autonomy and incen-
tives/monitoring in bureaucracies using experimental variation in these practices [Banerjee et al.
2014, Bandiera et al. 2018].24 We hope our work further helps establish a picture of what find-
ings on bureaucratic e↵ectiveness replicate over settings and what the sources of within-country
heterogeneity driving e↵ectiveness might be.


A      Data Appendix
A.1      Measuring Bureaucracy Projects and Output
In Ghana each civil service organization is required to provide quarterly and annual progress
reports. These detail targets and achievements for individual projects. The process of measuring
projects and output for each organization then comprised two steps. First, extracting the data
from organizations’ reports (which di↵ered slightly in their formats) into a standardized template.
Second, coding variables based on the standardized data.
   Figure A1 shows a snapshot of a typical quarterly progress report. The unit of observation is
the project or task, defined as the most disaggregated output reported. For each quarterly progress
  24
    Banerjee et al. [2014] report on a field experiment with the state police of Rajasthan, where a series of treat-
ments involved middle managers providing more autonomy to frontline police o cers. They find such reforms to be
poorly implemented and ine↵ective in rasing the e↵ectiveness of policing. This opens up the issue of understanding
how exactly to reallocate autonomy within organizations without creating constituencies of individuals that would
prefer to sabotage such reforms or see them weakly implemented. This also echoes recent findings from private
sector settings: Atkin et al. [2017] document that in the context of manufacturing firms in Pakistan, productivity
enhancing innovations were blocked by lower-tier workers who stood to loose from the adoption of such techniques.
In ongoing work, Bandiera et al. [2018] vary the autonomy and incentives/monitoring of purchasing managers in
procurement decisions across 500 public bodies in Pakistan. Echoing our results, they find providing such bureau-
crats greater autonomy allows them to perform better (they procure lower prices) while incentives have weak or
perverse e↵ects (leading them to procure goods at higher prices).


                                                        25
report, we codified project line items using a team of trained research assistants and a team of civil
servant o cers seconded from the Management Services Department in the Civil Service. Each
project was thus assigned to an organization (ministry or department) and to a division within
that organization. For organizations in which reporting formats did not specify which division
was responsible for a particular project, coders were supplied with information about the divisions
in the organization and assigned each project to the division that was most likely responsible
for it, in consultation with civil servants and/or research assistants who were familiar with the
organization. In cases where the two coders assigned a project to di↵erent divisions, a manager
made a judgment about which division to assign the project to.


A.2     Extracting and Standardizing
Although organizations’ reports di↵ered in their format and variable coverage, we extracted the
following standard variables for each organization (leaving them blank where the variable was
missing).
Output Level 1 The name or short description of the output specifying the action to be taken
during the time period, at the most disaggregated or fine-grained level available. For instance, in
Figure A1, this is ‘Develop draft competition policy’. This variable defines the unit of observation,
and by definition, cannot be missing.
Output Level 2 The name or short description of the output, aggregated to one level higher than
in Output level 1. Many organizations reported outputs that were nested into broader outputs, or
whose completion required multiple sequential or simultaneous smaller outputs to be completed.
For example, in Figure A1 the Output level 2 for ‘Develop draft competition policy’ is ‘Competition
Policy Developed and Approved.’ Multiple outputs can thus share the same Output level 2.
Output Level 3 The same as Output level 2, but one level of aggregation higher. As in Figure
A1, this level of aggregation was frequently unreported, but was extracted where relevant.
Budget Allocation/Cost The budgeted cost of the project. This was reported infrequently.
Baseline Completion Level Where reported, the level of attainment on the output at the start
of the time period.
Actual Output The actual attainment or work done during the time period. Together with the
target level of achievement for the time period (from Output level 1) and (where relevant) the
baseline level of completion, this is used to code project completion (as described in more detail
below).
Remarks Where reported, the organization’s comments about the output. These often explain
why the target level of attainment was not achieved during the time period.




                                                 26
A.3     Coding
After extracting the data, our team of civil servants and project research assistants coded a fixed
list of variables for each output (at the most disaggregated level, Output level 1 ). As the variables
to be coded required coders to interpret and judge the information being reported by each orga-
nization, coding was undertaken by two independent coders, with reconciliation led by managers
where necessary. Below is a list of all variables coded for each output.
Output Type (primary) Which category best describes this output? Coders had to select one
of the following: (i) Advocacy, outreach and stakeholder engagement/relations; (ii) Financial &
budget management; (iii) ICT management and/or development; (iv) Monitoring, review, & au-
dit; (v) Permits and regulation; (vi) Personnel management; (vii) Physical infrastructure – o ce
& facilities; (viii) Physical infrastructure – public infrastructure and projects; (ix) Policy develop-
ment; (x) Procurement; (xi) Research; (xii) Training.
Output Type (secondary) If output covers more than one category, select the secondary cate-
gory here. Coders had to select one of the same twelve categories as above.
Period/Regular vs. One-o↵ Is the output repeated (e.g. weekly, quarterly, annually) or one-o↵
(no planned repetition)? Coders had to select one of: (i) Periodic/ regular (e.g. weekly, quarterly,
annually); (ii) One-o↵ (no planned repetition).
Output Scope How narrowly is the output defined? Does it include multiple tasks, or even mul-
tiple outputs? Coders had to select one of: (i) Single activity (one step in a larger activity, has no
value on its own; e.g. hold a meeting about writing a policy); (ii) Single output (multiple steps,
has value on its own; e.g. write a policy); (iii) Bundle of outputs (multiple outputs that each have
their own value; e.g. write four policies)].
Technical Complexity Does the output require specific technical or scientific knowledge, beyond
the level most civil servants would have? Coders had to select one of: (i) No technical knowledge
required (any senior civil servant could do this); (ii) Technical knowledge is required (special ed-
ucation or training needed).
Coordination Required Does the division have to coordinate or interact with other actors in
order to achieve the output? Coders could select any of the following that applied: (i) Requires
action from other divisions in the organization; (ii) Requires action from other government orga-
nizations; (iii) Requires action from stakeholders outside government.
Target Clarity How precise, specific, and measurable is the target? Coders had to answer on a
1-5 scale (where integers and half values were both permitted) using the following scoring guide-
lines. Score 1: Target is undefined or so vague it is impossible to assess what completion would
mean; Score 3: Target is defined, but with some ambiguity; Score 5: There is no ambiguity over
the target – it is precisely quantified or described.
Output Clarity How precise, specific, and measurable is what the division actually achieved?
Coders had to answer on a 1-5 scale (where integers and half values were permitted) using the


                                                  27
following scoring guidelines. Score 1: Output information is absent or so vague it is impossible to
assess completion; Score 3: Output information is given but there is some ambiguity over whether
the target was met; Score 5: Output information is clear and unambiguous.
Completion Status How did actual achievement compare to the target? Coders had to answer on
a 1-5 scale (where integers and half values were permitted) using the following scoring guidelines.
Score 1: No action was taken towards achieving the target; Score 3: Some substantive progress
was made towards achieving the target. The output is partially complete and/or important in-
termediate steps have been completed; Score 5: The target for the output has been reached or
surpassed.
Completion Remarks Were any challenges/ obstacles mentioned? Coders could select all that
applied from the following: (i) awaiting action from another division, organization or stakeholder;
(ii) 2 = Procurement/sourcing delay or problem; (iii) Sequencing issue (can’t start until another
output has been completed); (iv) Lack of technical knowledge to complete activity; (v) Delayed/
non-release of funds; (vi) Unexpected event; (vii) Activity not due.
    There are at least two coders per project. Given the tendency for averaging scores to reduce
the measured variation, we use the maximum and minimum scores to code whether projects are
fully complete/never initiated respectively. We show robustness of our main result to alternative
methods by which to combine codings.

A.3.1     Nigeria: Measuring Projects and Project Completion

The output data for the RR study based on the Federal Civil Service in Nigeria exploited the
following historical event. In 2006/7 the Nigerian Government undertook the Overview of Public
Expenditure in NEEDS (the ‘OPEN initiative’), in which it traced, by project, the use and impact
of a representative sample of 10% of federal social sector expenditures approved in 2006/7. Under
the OPEN initiative, expert teams visited projects to record their implementation. This moni-
toring process was independent of civil servants: projects were evaluated by teams of independent
engineers and civil society.25 Monitoring teams visited project sites 18 months after projects were
approved. The projects studied in RR had 12 month completion schedules. RR hand-coded the
material from all projects recorded in OPEN initiative reports from 63 federal civil service organi-
zations, covering 4721 projects. 11 project types are covered (road, borehole, training etc.) with
boreholes being the modal project type, and 75% of projects relating to small-scale infrastructure.
  25
    A system of checks and balances were further put in place to underpin the credibility of the initiative. First, a
centralized team of technocrats monitored the evaluation teams, providing them with training and opportunities for
standardization of their methods. Second, evaluators were asked to provide material/photographic/video evidence
to support their reports. Third, random checks were performed on evaluated sites.




                                                         28
A.3.2    Nigeria: Measuring Management Practices

In the Nigerian context, RR also followed BSVR’s approach to measuring management practices,
adapting their survey tool to the Nigerian public sector setting. Management practices were elicited
from senior management in each organization. While each manager filled in their own question-
naire, enumerators looked for a consensus and recorded that in their own questionnaire. This is the
information used to construct management practice indices for each organization. Enumerators
held double blind interviews in the 63 organizations with project completion data. The manage-
ment survey covered nine topics: roles, flexibility, incentives, monitoring, culture, targeting, facili-
ties, skills and sta ng. RR combined answers to the roles and flexibility questions to construct an
index of management practices capturing bureaucrats’ autonomy (CS-autonomy). RR combined
answers on the incentives and monitoring topic areas to construct an index capturing the incen-
tives/monitoring management practices bureaucrats operate under (CS-incentives/monitoring).
RR combined answers on all remaining topics to construct an ‘other’ management practices in-
dex (CS-other). The responses to each practice were converted into normalized z-scores by taking
unweighted means of the underlying z-scores (so are continuous variables with mean zero and vari-
ance one by construction), where, as in Ghana, both are increasing in the commonly understood
notion of ‘better management’.


References
 [1] acemoglu.d and j.a.robinson (2012) Why Nations Fail, Crown Business, New York.

 [2] alesina.a and e.la ferrara (2005) “Ethnic Diversity and Economic Performance,” Jour-
     nal of Economic Literature 43: 762-800.

 [3] ashraf.n, o.bandiera and k.jack (2014) “No Margin, No Mission? A Field Experiment
     on Incentives for Pro-social Tasks,” Journal of Public Economics 120: 1-17.

 [4] ashraf.n, o.bandiera and s.s.lee (2018) Losing Prosociality in the Quest for Talent?
     Sorting, Selection, and Productivity in the Delivery of Public Services, mimeo LSE.

 [5] atkin.d, a.chaudhry, s.chaudhry, a.khandelwal and e.verhoogen (2017) “Organi-
     zational Barriers to Technology Adoption: Evidence from Soccer-Ball Producers in Pakistan,”
     Quarterly Journal of Economics 132:

 [6] baker.g.p (2002) “Distortion and Risk in Optimal Incentive Contracts,” Journal of Human
     Resources 37: 728-51.

 [7] bandiera.o, m.c.best, a.q.khan and a.prat (2018) Incentives and the Allocation of
     Authority in Organizations: A Field Experiment with Bureaucrats, mimeo LSE.

                                                  29
 [8] banerjee.a, r.chattopadhyay, e.duflo, d.keniston and n.singh (2014) Improving
     Police Performance in Rajasthan, India: Experimental Evidence on Incentives, Managerial
     Autonomy and Training, mimeo, MIT.

 [9] belle.n (2013) “Experimental Evidence on the Relationship Between Public Service Moti-
     vation and Job Performance,” Public Administration Review 73: 143-53.

[10] benabou.r and j.tirole (2006) “Incentives and Prosocial Behavior,” American Economic
     Review 96; 1652-78.

[11] bertrand.m, r.burgess, a.chawla and g.xu (2017) The Glittering Prizes: Career In-
     centives and Bureaucrat Performance, mimeo, LSE.

[12] besley.t and m.ghatak (2005) “Competition and Incentives with Motivated Agents,”
     American Economic Review 95: 616-36.

[13] besley.t and m.ghatak (2018) “Prosocial Motivation and Incentives,” Annual Review of
     Economics 10: 411-38.

[14] besley.t and t.persson (2011) Pillars of Prosperity, Princeton University Press, Prince-
     ton, New Jersey.

[15] bjorkman.m and j.svensson (2009) “Power to the People: Evidence from a Randomized
     Experiment on Community-Based Monitoring in Uganda,” Quarterly Journal of Economics
     124: 735-69.

[16] bloom.n and j.van reenen (2007) “Measuring and Explaining Management Practices
     Across Firms and Countries,” Quarterly Journal of Economics 122: 1351-408.

[17] bloom.n, r.sadun and j.van reenen (2012) “The Organization of Firms Across Coun-
     tries,” Quarterly Journal of Economics 127: 1663-705.

[18] christensen.g.s and e.miguel (2016) “Transparency, Reproducibility, and the Credibility
     of Economics Research,” Journal of Economic Literature, forthcoming.

[19] clemens.m, (2015) The Meaning of Failed Replications: A Review and Proposal, CGD WP.

[20] cowley.e and s.smith (2014) “Motivation and Mission in the Public Sector: Evidence from
     the World Values Survey,” Theory and Decision 76: 241-63.

[21] dal bo.e, f.finan and m.rossi (2013) “Strengthening State Capabilities: The Role of
     Financial Incentives in the Call to Public Service,” Quarterly Journal of Economics 128:
     1169-218.


                                             30
[22] deserranno.e (2017) “Financial Incentives as Signals: Experimental Evidence from the
     Recruitment of Health Workers,” American Economic Journal: Applied Economics, forth.

[23] dixit.a (2002) “Incentives and Organizations in the Public Sector: An Interpretive Review,”
     Journal of Human Resources 37: 696-727.

[24] easterly.w and r.levine (1997) “Africa’s Growth Tragedy: Policies and Ethnic Divi-
     sions,” Quarterly Journal of Economics 112: 1203-50.

[25] finan.f, b.a.olken, and r.pande (2017) “The Personnel Economics of the State,” in
     A.V.Banerjee and E.(eds.) Handbook of Field Experiments, Elsevier.

[26] francois.p and m.vlassopoulos (2008) “Pro-social Motivation and the Delivery of Social
     Services,” CESifo Economic Studies 54: 1-33.

[27] gibbons.r and r.henderson (2013) “What Do Managers Do?,” in R.Gibbons and
     J.Roberts (eds.) Handbook of Organizational Economics, Princeton: Princeton University
     Press.

[28] gneezy.u and a.rustichini (2000) “Pay Enough or Don’t Pay At All,” Quarterly Journal
     of Economics 115: 791-810.

[29] goldfinch.s, k.derouen and p.pospieszna (2012) “Flying Blind? Evidence for Good
     Governance Public Management Reform Agendas, Implementation and Outcomes in Low
     Income Countries,” Public Administration and Development 33: 50-61.

[30] hamermesh.d.s (2007) “Viewpoint: Replication in Economics,” Canadian Journal of Eco-
     nomics 40: 715-33.

[31] hasnain.z, n.manning and j.h.pierskalla (2012) Performance-related Pay in the Public
     Sector, World Bank Policy Research Working Pa2per 6043.

[32] hjort.j (2014) “Ethnic Divisions and Production in Firms,” Quarterly Journal of Economics
     129: 1899-946.

[33] kelman.s (1990) Procurement and Public Management: The Fear of Discretion and the
     Quality of Public Performance, Washington DC: American Enterprise Initiative.

[34] khan.a.q, a.i.khwaja and b.olken (2016) “Tax Farming Redux: Experimental Evidence
     on Performance Pay for Tax Collectors,” Quarterly Journal of Economics 131: 219-71.

[35] leonard.d.k (2010) “‘Pockets’ of E↵ective Agencies in Weak Governance states: Where are
     they Likely and Why does it Matter?,” Public Administration and Development 30: 91-101.


                                              31
[36] macchiavello.r and a.morjaria (2015) “The Value of Relationships: Evidence from a
     Supply Shock to Kenyan Rose Exports,” American Economic Review 105: 2911-45.

[37] milgrom.p.r (1988) “Employment Contracts, Influence Activities, and E cient Organiza-
     tion Design”, Journal of Political Economy 96: 42–60.

[38] milgrom.p.r and j.roberts (1988) “An Economic Approach to Influence in Organiza-
     tions”, American Journal of Sociology 94: S154-79.

[39] olken.b (2007) “Monitoring Corruption: Evidence from a Field Experiment in Indonesia,”
     Journal of Political Economy 115: 200-49.

[40] pepinsky.t.b, j.h.pierskalla and a.sacks (2017) “Bureaucracy and Service Delivery,”
     Annual Review of Political Science 20: 13.1-13.20.

[41] perry.j.l (1996) “Measuring Public Service Motivation: An Assessment of Construct Relia-
     bility and Validity,” Journal of Public Administration Research and Theory 6: 5-22.

[42] perry.j.l (2014) “The Motivational Bases of Public Service: Foundations for a Third Wave
     of Research,” Asia Pacific Journal of Public Administration 36: 34-47.

[43] perry.j.l, t.a.engers and s.y.jun (2009) “Back to the Future? Performance-related Pay,
     Empirical Research, and the Perils of Persistence,” Public Administration Review 69: 39-51.

[44] perry.j.l and l.r.wise (1990) “The Motivational Basis of Public Service,” Public Admin-
     istration Review 50: 367-73.

[45] prendergast.c (1999) “The Provision of Incentives in Firms,” Journal of Economic Liter-
     ature 37: 7-63.

[46] rasul.i and d.rogger (2015) “The Impact of Ethnic Diversity in Bureaucracies: Evidence
     from the Nigerian Civil Service,” American Economic Review, Papers and Proceedings 105:
     457-61.

[47] rasul.i and d.rogger (2018) “Management of Bureaucrats and Public Service Delivery:
     Evidence from the Nigerian Civil Service,” Economic Journal 128: 413-46.

[48] rose-ackerman.s (1986) “Reforming Public Bureaucracy Through Economic Incentives?,”
     Journal of Law, Economics and Organization 2: 131-61.

[49] simon.w (1983) “Legality, Bureaucracy, and Class in the Welfare System,” Yale Law Journal
     92: 1198-269.




                                              32
[50] warren.d.c and l.t.chen (2013) “The Relationship Between Public Service Motivation
     and Performance,” in E.J.Ringquist (ed.), Meta-analysis for Public Management and Policy,
     San Francisco, CA: Jossey-Bass.

[51] williams.m.j (2015) From Institutions to Organizations: Management and Informality in
     Ghana’s Public Bureaucracies, mimeo, LSE.

[52] wilson.j (1989) Bureaucracy: What Government Agencies Do and Why They Do It, New
     York: Basic Books.

[53] world       bank      (2018)    World   Governance       Indicators,     available    at:
     http://info.worldbank.org/governance/WGI/#reports.




                                             33
                              Figure 1A: What Do Ghanaian Bureaucrats Do?
                                                         Procurement
                                                             6%

                                                   Permits and
                                                    Regulation
                                                       4%                      Physical infrastructure
                                                                                         27%



                                  Monitoring, Training
                                    and Personnel
                                    Management
                                         29%




                                                                         Advocacy and Policy
                                                                            Development
                                                      Financial & Budget        23%
                                   ICT Management and    Management
                                        Research              4%
                                           7%



                           Figure 1B: Project Types by Implementing Organization
              1,000  800
    Number of projects
     400    600
              200
              0




                                             e




                                                                                      t
                                                                        h
                              l




                                                                                                                         l
                                                                                                    its
                                                          y
                           ne




                                                                                    en




                                                                                                                      ia
                                           ur




                                                       lic




                                                                      rc




                                                                                                  rm




                                                                                                                    nc
                                                                                  em
                         on




                                                                    ea
                                         ct



                                                     Po




                                                                                                                 na
                                       ru




                                                                                                Pe
                       rs




                                                                  es



                                                                                ur
                                     st




                                                                                                               Fi
                     Pe




                                                                 R



                                                                              oc
                                  fra




                                                                            Pr
                                In




Notes: The “project type” classification refers to the primary classification for each project. Other project classifications exist. Each
colour in a column represents an organization implementing projects of that project type, but the same colour across columns may
represent multiple organizations.
   Figure 2: Bureaucratic Performance by Project Types
 0.80
 0.70
 0.60
 0.50
 0.40
 0.30
 0.20
 0.10
 0.00




                                    (3) Proportion Never Started (Min Report)   (4) Proportion Fully Completed (Max Report)



    Figure 3: Bureaucratic Performance by Organization
                1          .8
    Proportion of projects
        .4      .2
                0  .6




                                0                          10                      20                                         30
                                               Organisation ranking by proportion of projects initiated

                                                      Proportion started          Proportion completed


                                              Mean Completion Proportion: 0.38
Notes: Multiple coders assessed a project such that here we take the minimum assessment of initiation and
the maximum assessment of completion. The 'Mean Completion Proportion' is an average of the
organizational averages of proportion completed.
     Figure 4: Variation in Management Practices Across
                         Bureaucracies
                                       A. CS-Autonomy Score
               2          1
      Raw management score
    -1        0-2




                              0            10                    20                              30
                                         Autonomy ranking of organisation


                                  B. CS-Incentives/Monitoring Score
              2          1
     Raw management score
   -1         0
              -2




                              0            10                       20                           30
                                         Incentives ranking of organisation

Notes: CS-Autonomy (CS-Incentives) is an aggregate index of management questions relating to roles and
flexibility (monitoring and incentives). Responses by senior bureaucrats are averaged across all sub-topics
and converted into z-scores before averaged into a single index of management.
   Figure 5: Management Practices Score by Senior and
                 Non-Senior Bureaucrats
                                         A. CS-Autonomy Score
               1         .8
   Cumulative probability




                                                                                       Senior
       .4       .6
               .2




                                                           Non-senior
               0




                               -2        -1               0                   1                     2
                                                   Management Score

                                    B. CS-Incentives/Monitoring Score
                                        Monitoring Management Indices
               1         .8
   Cumulative probability
                .6




                                                                                  Senior
       .4      .2




                                                                               Non-senior
               0




                              -4              -2                       0                             2
                                                   Management Score

Notes: CS-Autonomy (CS-Incentives) is an aggregate index of management questions relating to roles and
flexibility (monitoring and incentives). Responses by senior/non-senior bureaucrats are averaged across all
sub-topics and converted into z-scores before averaged into a single index of management. Officials are
designated senior based on whether they classify themselves as a 'Director (Head of Division) or Acting
Director' or as a 'Deputy Director or Unit Head (Acting or Substantive)'.
                                 Table 1: Variation in Bureaucratic Performance by Project Types
                                                          (1) Number of             (2) Number of               (3) Proportion              (4) Proportion                (5) Average
Project Type                                                 Projects               Implementing                Never Started              Fully Completed              Completion Rate
                                                           [Proportion]             Organizations                (Min Report)                (Max Report)                     [CV]

All Project Types                                             3,628 [1.00]                    30                         0.21                        0.34                      3.23 [0.38]

Physical Infrastructure                                       1,010 [0.27]                    23                         0.17                        0.28                      3.20 [0.36]
All non-Physical Infrastructure Projects                      2,610 [0.73]                    30                         0.23                        0.37                      3.24 [0.39]

Advocacy and Policy Development                                833 [0.23]                     29                         0.23                        0.34                      3.20 [0.38]
Financial & Budget Management                                  137 [0.04]                     19                         0.27                        0.44                      3.29 [0.44]
ICT Management and Research                                    265 [0.07]                     24                         0.18                        0.35                      3.23 [0.38]
Monitoring, Training and Personnel
                                                              1,044 [0.29]                    30                         0.23                        0.40                      3.30 [0.38]
Management
Permits and Regulation                                         141 [0.04]                     20                         0.14                        0.32                      3.28 [0.31]
Procurement                                                    190 [0.05]                     22                         0.31                        0.36                      3.02 [0.46]

Notes: The “project type” classification refers to the primary classification for each project. Other project classifications exist. The 'Average Completion Rate' is a categorical variable taking
the value 1 if the project was not started and 5 if it was fully completed. It is accompanied in Column 5 by the coefficient of variation. The 'Proportion Never Started' and 'Proportion Fully
Completed' columns are based on dummies that take the corresponding values of the completion status variable. The corresponding columns with minimum and maximum reports take the
lowest and highest reports by project assessors respectively. Figures are rounded to two decimal places where relevant.
                 Table 2: Variation in Bureaucratic Performance by Largest Civil Service Organizations
                                                                                   (2) Number of                                                   (4) Proportion                 (5) Average
                                                                     (1) Number of                                      (3) Proportion
Civil Service Organization                                                         Unique Project                                                 Fully Completed               Completion Rate
                                                                        Projects                                        Never Started
                                                                                       Types                                                        (Max Report)                      [CV]

All Organizations (Average)                                                   117                    5.60                       0.21                        0.34                      3.23 [0.38]

Ministry of Trade and Industry                                                495                      7                        0.26                        0.27                       2.97 [0.4]
Ministry of Finance                                                           374                      7                        0.37                        0.26                       2.75 [0.5]
Department of Feeder Roads                                                    315                      6                        0.20                        0.32                       3.2 [0.42]
Ministry of Energy and Petroleum                                              285                      7                        0.31                        0.29                      2.99 [0.44]

Ministry of Gender, Children and Social Protection                            274                      7                        0.16                        0.43                      3.51 [0.32]

Controller and Accountant-General's Department                                254                      7                        0.27                        0.54                      3.49 [0.42]
Environmental Protection Agency                                               193                      6                        0.28                        0.17                       2.86 [0.4]
Department of Urban Roads                                                     188                      4                        0.15                        0.21                      3.11 [0.34]
Ministry of Environment, Science, Technology, and
                                                                              172                      6                         0.10                       0.34                      3.45 [0.23]
Innovation
Office of the Head of Civil Service                                           158                      7                         0.11                       0.49                      3.87 [0.24]

Notes: The sample covers the ten largest civil service organizations for which we had data on activities ranked by number of projects from our overall sample of projects. The “project type”
classification refers to the primary classification for each project. Other project classifications exist. The 'Average Completion Rate' is a categorical variable taking the value 1 if the project was not
started and 5 if it was fully completed. It is accompanied in Column 5 by the coefficient of variation. The 'Proportion Never Started' and 'Proportion Fully Completed' columns are based on dummies
that take the corresponding values of the completion status variable. The corresponding columns with minimum and maximum reports take the lowest and highest reports by project assessors
respectively. Figures are rounded to two decimal places where relevant.
Table 3: Management of Bureaucrats and Public Service Delivery
Dependent Variable Cols 1-4: Completion Binary [yes=1]
Dependent Variable Col 5: Initiation Binary [yes=1]
Dependent Variable Col 6: Completion Index
Standard Errors: Clustered by Organization
OLS Estimates
                                                                    (2) Noise and                                    (4) Fixed            (5) Binary       (6) Completion
                                                 (1) Noise                                  (3) Baseline
                                                                       General                                        Effects              Initiation           Status

CS-Autonomy                                         0.00                 0.28***               0.28***                0.23***               0.25***             0.18***
                                                   (0.01)                 (0.04)                (0.04)                 (0.03)               (0.04)               (0.03)
CS-Incentives/Monitoring                            0.01                -0.17***               -0.18***               -0.11**              -0.28***             -0.16***
                                                   (0.05)                 (0.04)                (0.05)                 (0.06)               (0.04)               (0.03)
CS-Other                                            0.02                -0.08***               -0.07***              -0.09***              -0.09***             -0.04***
                                                   (0.03)                 (0.01)                (0.02)                 (0.02)               (0.02)               (0.01)

Noise Controls                                      Yes                    Yes                   Yes                    Yes                   Yes                 Yes
General and Capital Controls                         No                    Yes                   Yes                    Yes                   Yes                 Yes
Project Controls                                     No                     No                   Yes                    Yes                   Yes                 Yes
                                              Project Type,          Project Type,          Project Type,                               Project Type,        Project Type,
Fixed Effects                                                                                                     Project Type
                                                 Sector                 Sector                 Sector                                      Sector               Sector
Observations (clusters)                          3620 (30)             3620 (30)              3620 (30)             3620 (30)             3620 (30)            3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization throughout. All columns report
OLS estimates. In Columns 1 through 4, the dependent variable is a dummy variable that takes the value 1 if the project is fully completed and 0 otherwise. In Column 5,
the dependent variable is a dummy variable that takes the value 1 if the project is initiated and 0 otherwise. In Column 6, the dependent variable is an index of project
completion (that is a continuous measure between zero and one). Project Type fixed effects relate to whether the primary classification of the project is 'Advocacy and
Policy Development', 'Financial & Budget Management', 'ICT Management and Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure',
'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to whether the project is in the administration, environment, finance, infrastructure,
security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the project is regularly implemented by the organization or a one off,
whether the project is a bundle of interconnected outputs, and whether the division has to coordinate with actors external to government to implement the project. Capital
controls comprise a count of the number of interviews undertaken, which is a close approximation of the total number of employees. General controls comprise
organization-level controls for the share of the workforce with degrees, the share of the workforce with postgraduate qualifications, and the span of control. Noise controls
are averages of indicators of the seniority, gender, and tenure of all respondents, the average time of day the interview was conducted and of the reliability of the
information as coded by the interviewer. Figures are rounded to two decimal places.
Table 4: IV
Dependent Variable Cols 1, 3: Completion Binary [yes=1]
Dependent Variable Col 2: CS-Monitoring/Incentives Management Index, 2015
Dependent Variable Col 4: Initiation Binary [yes=1]
Dependent Variable Col 5: Completion Index
Standard Errors: Robust
                                  OLS          First Stage           IV: Second Stage
                                            (1) Baseline [IV                                    (3) Project              (4) Binary       (5) Completion
                                                                      (2) Monitoring
                                                Sample]                                         Completion                Initiation           Status

CS-Autonomy                                        0.04**                                           0.11**                 0.14***              0.35***
                                                   (0.02)                                            (0.05)                 (0.02)              (0.10)
CS-Incentives/Monitoring                          -0.09***                 0.44***                  -0.27**                -0.38***            -0.95***
                                                   (0.02)                   (0.04)                   (0.11)                 (0.06)              (0.21)
CS-Other                                           -0.06*                                            -0.03                   0.01                -0.04
                                                   (0.03)                                            (0.03)                 (0.02)              (0.07)

First Stage F Statistic                                                      107
Noise Controls                                      Yes                      Yes                      Yes                    Yes                  Yes
Project Controls                                    Yes                      Yes                      Yes                    Yes                  Yes
                                              Project Type,            Project Type,                                   Project Type,        Project Type,
Fixed Effects                                                                               Project Type, Sector
                                                 Sector                   Sector                                          Sector               Sector
Observations (clusters)                         2,463 (18)                  2463                     2,463                  2,463                2,463

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Robust standard errors are in parentheses, and are clustered by organization in
Column 1. All columns report OLS estimates. In Columns 1 and 3, the dependent variable is a dummy variable that takes the value 1 if the project is fully
completed and 0 otherwise. In Column 2, the dependent variable is the monitoring management index defined using baseline data. In Column 4, the
dependent variable is a dummy variable that takes the value 1 if the project is initiated and 0 otherwise. In Column 5, the dependent variable is an index of
project completion (that is a continuous measure between zero and one). Project Type fixed effects relate to whether the primary classification of the
project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT Management and Research', 'Monitoring, Training and Personnel
Management', 'Physical infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to whether the project is in the administration,
environment, finance, infrastructure, security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the project is
regularly implemented by the organization or a one off, whether the project is a bundle of interconnected outputs, and whether the division has to
coordinate with actors external to government to implement the project. Capital controls comprise a count of the number of interviews undertaken, which is
a close approximation of the total number of employees. General controls comprise organization-level controls for the share of the workforce with degrees,
the share of the workforce with postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender, and
tenure of all respondents, the average time of day the interview was conducted and of the reliability of the information as coded by the interviewer. Figures
are rounded to two decimal places.
Table 5: Multi-tasking
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
Interactions in Deviation from Mean
OLS Estimates                                                                      Internal                  External
                                                                                 (1) Ethnic               (2) Stakeholder
                                                                              Fractionalization             Engagement

CS-Autonomy                                                                          0.31***                   0.28***
                                                                                     (0.04)                     (0.04)
CS-Incentives/Monitoring                                                            -0.18***                   -0.18***
                                                                                     (0.04)                     (0.05)
CS-Other                                                                            -0.08***                   -0.07***
                                                                                     (0.02)                     (0.02)
CS-Incentives/Monitoring x Ethnic Fractionalization                                  -0.05*
                                                                                     (0.03)
Ethnic Fractionalization                                                              0.06*
                                                                                     (0.03)
CS-Incentives/Monitoring x Stakeholder Engagement                                                              -0.06**
                                                                                                                (0.02)
Stakeholder Engagement                                                                                          0.05**
                                                                                                                (0.02)

Noise, General and Capital Controls                                                   Yes                        Yes
Project Controls                                                                      Yes                        Yes

Fixed Effects                                                               Project Type, Sector Project Type, Sector

Observations (clusters)                                                            3620 (30)                  3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by
organization throughout. All columns report OLS estimates. The dependent variable in all columns is a dummy variable that
takes the value 1 if the project is completed and 0 otherwise. The variable with which management scores are interacted is z-
score defined across the 30 organizations we study. In Column 1, 'Ethnic Fractionalization' is the organization-level
Herfindahl index that is equal to 1 minus the sum of squares of ethnicity proportions in an organization. In Column 2,
'Stakeholder Engagement' is the proportion of projects officials at an organization report that they interact with stakeholders
outside of their organization. Sector fixed effects relate to whether the project is in the administration, environment, finance,
infrastructure, security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the
project is regularly implemented by the organization or a one-off, whether the project is a bundle of interconnected outputs,
and whether the division has to coordinate with actors external to the government to implement the project. Capital controls
comprise a count of the number of interviews undertaken, which is a close approximation of the total number of employees.
General controls comprise organization-level controls for the share of the workforce with degrees, the share of the workforce
with postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender,
and tenure of all respondents, the average time of day the interview was conducted and of the reliability of the information as
coded by the interviewer. Figures are rounded to two decimal places.
Table 6: Intrinsic Motivation of Bureaucrats
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
Interactions in Deviation from Mean
OLS Estimates
                                                                    (1) PSM: Policy           (2) PSM:            (3) PSM: Public (4) PSM: Self-
                                                                         Making              Compassion               Interest       Sacrifice

CS-Autonomy                                                               0.28***                0.34***                0.34***              0.28***
                                                                           (0.04)                 (0.04)                (0.06)               (0.06)
CS-Incentives/Monitoring                                                  -0.14***              -0.19***               -0.20***             -0.16***
                                                                           (0.05)                 (0.04)                (0.05)               (0.05)
CS-Other                                                                  -0.09***              -0.11***               -0.11***             -0.08***
                                                                           (0.03)                 (0.01)                (0.03)               (0.03)
CS-Incentives/Monitoring x PSM - Policy Making                            -0.05**
                                                                           (0.02)
PSM - Policy Making                                                         0.02
                                                                           (0.02)
CS-Incentives/Monitoring x PSM - Compassion                                                      0.08***
                                                                                                  (0.02)
PSM - Compassion                                                                                 -0.08***
                                                                                                  (0.01)
CS-Incentives/Monitoring x PSM - Public Interest                                                                        0.12***
                                                                                                                        (0.04)
PSM - Public Interest                                                                                                  -0.09***
                                                                                                                        (0.03)
CS-Incentives/Monitoring x PSM - Sacrifice                                                                                                    0.04
                                                                                                                                              (0.04)
PSM - Sacrifice                                                                                                                             -0.04***
                                                                                                                                              (0.01)

Noise, General and Capital Controls                                         Yes                    Yes                    Yes                  Yes
Project Controls                                                            Yes                    Yes                    Yes                  Yes
                                                                      Project Type,          Project Type,          Project Type,        Project Type,
Fixed Effects
                                                                         Sector                 Sector                 Sector               Sector
Observations (clusters)                                                  3620 (30)             3620 (30)              3620 (30)            3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization throughout.
All columns report OLS estimates. The dependent variable in all columns is a dummy variable that takes the value 1 if the project is completed and 0
otherwise. The variable with which management scores are interacted is z-score defined across the 30 organizations we study. In Columns 1 through
4, the variables represent organization averages of aggregate Perry Public Service Motivation scores on the sub-categories of attraction to policy
making, compassion for underprivileged people, commitment to the public interest, and self-sacrifice respectively. Project Type fixed effects relate to
whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT Management and
Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects
relate to whether the project is in the administration, environment, finance, infrastructure, security/diplomacy/justice or social sector. Project controls
comprise project-level controls for whether the project is regularly implemented by the organization or a one off, whether the project is a bundle of
interconnected outputs, and whether the division has to coordinate with actors external to government to implement the project. Capital controls
comprise a count of the number of interviews undertaken, which is a close approximation of the total number of employees. General controls comprise
organization-level controls for the share of the workforce with degrees, the share of the workforce with postgraduate qualifications, and the span of
control. Noise controls are averages of indicators of the seniority, gender, and tenure of all respondents, the average time of day the interview was
conducted and of the reliability of the information as coded by the interviewer. Figures are rounded to two decimal places.
Table 7: Subjective Performance Evaluation
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
Interactions in Deviation from Mean
OLS Estimates
                                                                                      (1) Shared
                                                                                                             (2) Same Ethnicity
                                                                                    University Time

CS-Autonomy                                                                                0.28***                   0.28***
                                                                                           (0.03)                     (0.03)
CS-Incentives/Monitoring                                                                  -0.22***                   -0.10***
                                                                                           (0.04)                     (0.04)
CS-Other                                                                                  -0.06***                   -0.09***
                                                                                           (0.01)                     (0.03)
CS-Incentives/Monitoring x Shared University Time                                         -0.11***
                                                                                           (0.02)
Proportion of Officials In Organization Who Share Time In                                  0.10***
Undergraduate With Senior Unit Civil Servant                                               (0.02)
CS-Incentives/Monitoring x Same Ethnicity                                                                            -0.10***
                                                                                                                      (0.03)
Proportion of Officials In Organization Who Are the Same                                                             0.10***
Ethnicity As The Senior Unit Civil Servant                                                                            (0.02)

Noise, General and Capital Controls                                                          Yes                       Yes
Project Controls                                                                             Yes                       Yes

Fixed Effects                                                                     Project Type, Sector Project Type, Sector

Observations (clusters)                                                                  3620 (30)                  3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by
organization throughout. All columns report OLS estimates. The dependent variable in all columns is a dummy variable that takes
the value 1 if the project is completed and 0 otherwise. The variable with which management scores are interacted is z-score
defined across the 30 organizations we study. In Column 1, 'Shared University' refers to the proportion of officials in an organization
who share an undergraduate university with the senior civil servant in the unit. In Column 2, 'Same Ethnicity' refers to the proportion
of officials in an organization who share an ethnicity with the senior civil servant in the unit. Project Type fixed effects relate to
whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT
Management and Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure', 'Permits and Regulation'
or 'Procurement'. Sector fixed effects relate to whether the project is in the administration, environment, finance, infrastructure,
security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the project is regularly
implemented by the organization or a one off, whether the project is a bundle of interconnected outputs, and whether the division
has to coordinate with actors external to government to implement the project. Capital controls comprise a count of the number of
interviews undertaken, which is a close approximation of the total number of employees. General controls comprise organization-
level controls for the share of the workforce with degrees, the share of the workforce with postgraduate qualifications, and the span
of control. Noise controls are averages of indicators of the seniority, gender, and tenure of all respondents, the average time of day
the interview was conducted and of the reliability of the information as coded by the interviewer. Figures are rounded to two decimal
places.
Table 8: Project Characteristics
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
OLS Estimates
                               Target Clarity                                            Output Clarity

                                             (1) Below          (2) Above           (3) Below          (4) Above
                                              median             median              median             median


CS-Autonomy                                    0.30***            0.25***             0.17***            0.37***
                                                (0.03)             (0.05)              (0.03)             (0.06)
CS-Incentives/Monitoring                      -0.23***            -0.10**             -0.09**            -0.31***
                                                (0.03)             (0.04)              (0.03)             (0.09)
CS-Other                                        -0.01             -0.11**              -0.01             -0.16***
                                                (0.01)             (0.04)              (0.02)             (0.03)

Noise Controls                                   Yes                Yes                 Yes                Yes
General and Capital Controls                     Yes                Yes                 Yes                Yes
Project Controls                                 Yes                Yes                 Yes                Yes
                                           Project Type, Project Type,            Project Type, Project Type,
Fixed Effects
                                              Sector        Sector                   Sector        Sector
Observations (clusters)                      1851 (29)          1769 (29)           2177 (29)          1443 (27)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are
clustered by organization throughout. All columns report OLS estimates. The dependent variable is a dummy
variable that takes the value 1 if the project is fully completed and 0 otherwise. Project Type fixed effects relate to
whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget
Management', 'ICT Management and Research', 'Monitoring, Training and Personnel Management', 'Physical
infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to whether the project is in
the administration, environment, finance, infrastructure, security/diplomacy/justice or social sector. Project
controls comprise project-level controls for whether the project is regularly implemented by the organization or a
one off, whether the project is a bundle of interconnected outputs, and whether the division has to coordinate with
actors external to government to implement the project. Capital controls comprise a count of the number of
interviews undertaken, which is a close approximation of the total number of employees. General controls
comprise organization-level controls for the share of the workforce with degrees, the share of the workforce with
postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority,
gender, and tenure of all respondents, the average time of day the interview was conducted and of the reliability
of the information as coded by the interviewer. Figures are rounded to two decimal places.
Table 9: Corruption
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
Interactions in Deviation from Mean
OLS Estimates
                                                                                          (1) Superior        (2) Proportion of Projects that
                                                                                           Shares In          Bureaucrats Report Observing
                                                                                           Corruption              Corrupt Practices On

CS-Autonomy                                                                                  0.28***                          0.33***
                                                                                              (0.06)                           (0.04)
CS-Incentives/Monitoring                                                                     -0.13**                          -0.20***
                                                                                              (0.07)                           (0.07)
CS-Other                                                                                     -0.07***                         -0.14***
                                                                                              (0.02)                           (0.03)
CS-Autonomy x Superior Shares In Corruption                                                   -0.01
                                                                                              (0.03)
CS-Incentives/Monitoring x Superior Shares In Corruption                                       0.02
                                                                                              (0.05)
Average Proportion of Corruption Rents Received By Superior                                   0.05**
                                                                                              (0.02)
CS-Autonomy x Proportion of Projects that Bureaucrats Report Observing Corrupt                                                  0.00
Practices On                                                                                                                   (0.02)
CS-Incentives/Monitoring x Proportion of Projects that Bureaucrats Report                                                      0.11**
Observing Corrupt Practices On                                                                                                 (0.04)
Proportion of Projects that Bureaucrats Report Observing                                                                        0.04
Corrupt Practices On                                                                                                           (0.03)

Noise, General and Capital Controls                                                            Yes                              Yes
Project Controls                                                                               Yes                              Yes
                                                                                         Project Type,
Fixed Effects                                                                                                         Project Type, Sector
                                                                                            Sector
Observations (clusters)                                                                     3620 (30)                        3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization throughout.
All columns report OLS estimates. The dependent variable in all columns is a dummy variable that takes the value 1 if the project is completed and 0
otherwise.The variable with which management scores are interacted is z-score defined across the 30 organizations we study. In Column 1, 'Superior
Shares in Corruption' represents the average proportion of corruption rents received by a unit superior within an organization. In Column 2, 'Other
Breaking' refers to the proportion of recent projects and/or programmes on which officials report observing others engaging in corrupt practices.
Project Type fixed effects relate to whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget
Management', 'ICT Management and Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure', 'Permits and Regulation'
or 'Procurement'. Sector fixed effects relate to whether the project is in the administration, environment, finance, infrastructure,
security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the project is regularly implemented by the
organization or a one off, whether the project is a bundle of interconnected outputs, and whether the division has to coordinate with actors external to
government to implement the project. Capital controls comprise a count of the number of interviews undertaken, which is a close approximation of the
total number of employees. In both countries, general controls comprise organization-level controls for the share of the workforce with degrees, the
share of the workforce with postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender, and
tenure of all respondents, the average time of day the interview was conducted and of the reliability of the information as coded by the interviewer.
Figures are rounded to two decimal places.
Table 10: Different Measures of Management Practices and Public Service Delivery
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Project Type Within Organization
OLS Estimates
                                                         (1) All Senior              (2) Most Senior               (3) Non-senior
                                                                                                                                            (4) All Respondents
                                                         Bureaucrats                   Bureaucrat                   bureaucrats

CS-Autonomy                                                  0.28***                      0.31***                       0.12***                      0.30***
                                                              (0.04)                       (0.04)                       (0.05)                        (0.07)
CS-Incentives/Monitoring                                     -0.18***                     -0.14***                       -0.02                       -0.17**
                                                              (0.05)                       (0.04)                       (0.04)                        (0.07)
CS-Other                                                     -0.07***                      -0.08**                       -0.04                        -0.01
                                                              (0.02)                       (0.03)                       (0.04)                        (0.06)

Noise, General and Capital Controls                            Yes                          Yes                           Yes                          Yes
Project Controls                                               Yes                          Yes                           Yes                          Yes
Fixed Effects                                        Project Type, Sector         Project Type, Sector         Project Type, Sector          Project Type, Sector
Observations (clusters)                                     3620 (30)                    3620 (30)                    3620 (30)                    3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by project type within organization
throughout. All columns report OLS estimates. The dependent variable in all columns is a dummy variable that takes the value 1 if the project is completed and 0
otherwise. All senior bureaucrats referes to that set of individuals who identify themselves as Director (Head of Division) or Acting Director, or Deputy Director or
Unit Head (Acting or Substantive). The most senior bureaucrat refers to the official of the highest grade that we interviewed within a division. Non-senior
bureaucrats are those officials who do not identify themselves as senior bureaucrats. Project Type fixed effects relate to whether the primary classification of the
project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT Management and Research', 'Monitoring, Training and Personnel
Management', 'Physical infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to whether the project is in the administration,
environment, finance, infrastructure, security/diplomacy/justice or social sector. Project controls comprise project-level controls for whether the project is regularly
implemented by the organisation or a one off, whether the project is a bundle of interconnected outputs, and whether the division has to coordinate with actors
external to government to implement the project. Capital controls comprise a count of the number of interviews undertaken, which is a close approximation of the
total number of employees. General controls comprise organization-level controls for the share of the workforce with degrees, the share of the workforce with
postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender, and tenure of all respondents, the average
time of day the interview was conducted and of the reliability of the information as coded by the interviewer. Figures are rounded to two decimal places.
Table 11: Different Measures of Management Practices and Public Service Delivery
Dependent Variable: Difference between bureaucrat and organizational management scores
Standard Errors: Clustered by Organization
OLS Estimates                                  Autonomy                          Incentives/Monitoring
                                                                        (2) Organization         (3) Output                              (5) Organization          (6) Output
                                                     (1) Baseline                                                      (4) Baseline
                                                                          Fixed Effects         Organizations                              Fixed Effects          Organizations

Gender [female=1]                                        .089**               .121***                 .094**              .120***               .140***                .143***
                                                          (.036)               (.039)                 (.044)               (.042)               (.043)                 (.047)
Age                                                      -.006**              -.004**                 -.002               -.009***             -.008***                -.007*
                                                          (.002)               (.002)                 (.002)               (.003)               (.003)                 (.004)
Undergraduate Education                                 -.143***              -.139***               -.100**              -.166***             -.196***               -.211***
                                                          (.048)               (.044)                 (.044)               (.051)               (.048)                 (.051)

Other Bureacrat Controls                                   Yes                  Yes                    Yes                  Yes                   Yes                    Yes
Fixed Effects                                             None             Organization           Organization             None             Organization           Organization
Observations (clusters)                                1359 (45)             1359 (45)              1039 (30)            1361 (45)            1361 (45)              1041 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization throughout. All columns report OLS
estimates. The dependent variable in Columns 1 to 3 (4 to 6) is the difference in CS-Autonomy (CS-Incentives/Monitoring) index score between the individual public official and
the aggregate score for her organisation. Other Bureaucrat Controls relate to the tenure of the official in the organization, their cognitive score on the Raven cognition test, their
aggregate score on the Perry Public Service Motivation index and the extent to which they share an undergraduate institution or ethnicity with their manager/the most senior
bureaucrat in their division. Figures are rounded to two decimal places.
Table 12: External Validity
Standard Errors: Clustered by Project Type Within Organization
OLS Estimates
                                               Ghana                                                                                Nigeria
                                                                       (2) Full (3) Completion                                    (5) Full (6) Completion
                                                   (1) Initiation                                             (4) Initiation
                                                                     Completion      Rate                                       Completion      Rate

CS-Autonomy                                            0.13***           0.23***             0.12***              0.15***           0.16***             0.18***
                                                        (0.05)           (0.03)              (0.02)               (0.03)            (0.02)              (0.03)
CS-Incentives/Monitoring                              -0.16***           -0.11**            -0.10***             -0.16***          -0.10***            -0.14***
                                                        (0.06)           (0.06)              (0.04)               (0.02)            (0.02)              (0.02)
CS-Other                                              -0.07***          -0.09***             -0.04**              0.06**            0.06**              0.08***
                                                        (0.02)           (0.02)              (0.01)               (0.03)            (0.03)              (0.02)

Noise, General and Capital Controls                      Yes               Yes                 Yes                  Yes               Yes                 Yes
Project Controls                                         Yes               Yes                 Yes                  Yes               Yes                 Yes
Fixed Effects                                      Project Type Project Type             Project Type         Project Type Project Type             Project Type
Observations (clusters)                              3620 (30)         3620 (30)           3620 (30)           4721 (201)        4721 (201)          4721 (201)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by project type within organization
throughout. All columns report OLS estimates. The dependent variable in Columns 1 and 4 is a dummy variable that takes the value 1 if the project is initiated and 0
otherwise. The dependent variable in Columns 2 and 5 is a dummy variable that takes the value 1 if the project is completed and 0 otherwise. The dependent
variable in Columns 3 and 6 is an index of project completion (that is a continuous measure between zero and one). In Nigeria, 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. In Ghana, Project Type fixed effects relate to whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget
Management', 'ICT Management and Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure', 'Permits and Regulation' or
'Procurement'. In Ghana, sector fixed effects are also included, and relate to whether the project is in the administration, environment, finance, infrastructure,
security/diplomacy/justice or social sector. In Nigeria, project controls comprise project-level controls for the project budget, whether the project is new or a
rehabilitation, and an assessment of its aggregate complexity by Nigerian engineers. In Ghana, project controls comprise project-level controls for whether the
project is regularly implemented by the organization or a one off, whether the project is a bundle of interconnected outputs, and whether the division has to
coordinate with actors external to government to implement the project. In Nigeria, capital controls comprise organization-level controls for the logs of number of
employees, total budget, and capital budget. In Ghana, capital controls comprise a count of the number of interviews undertaken, which is a close approximation of
the total number of employees. In both countries, general controls comprise organization-level controls for the share of the workforce with degrees, the share of the
workforce with postgraduate qualifications, and the span of control. In Nigeria, noise controls are four interviewer dummies, indicators of the seniority, gender, and
tenure of the managers who responded, the day of the week the interview was conducted, the time of day the interview was conducted, a dummy variable indicating
whether the interview was conducted during Ramadan, the duration of the interview, and an indicator of the reliability of the information as coded by the interviewer.
In Ghana, noise controls are averages of indicators of the seniority, gender, and tenure of all respondents, the average time of day the interview was conducted and
of the reliability of the information as coded by the interviewer. Figures are rounded to two decimal places.
Table A1: Defining Management Practices Using the CS Indices
Management Practice    Topic                  Question                             Score 1                             Score 3                              Score 5


                                                                        Senior staff do not have                                                 It is integral to the
                                    Can most senior staff in your                                           Substantive contributions can
                                                                        channels to make substantive                                             organisation’s culture that any
                                    division make substantive                                               be made in staff meetings by
                                                                        contributions to organisational                                          member of senior staff can
CS-Autonomy             Roles       contributions to the policy                                             all senior staff but there are no
                                                                        policies, nor to the                                                     substantively contribute to the
                                    formulation and implementation                                          individual channels for ideas to
                                                                        management of their                                                      policies of the organisation or
                                    process?                                                                flow up the organisation.
                                                                        implementation.                                                          their implementation.

                                                                        Officers in this division have no
                                    When senior staff in your
                                                                        real independence to make           Officers in this division have
                                    division are given tasks in their
                                                                        decisions over how they carry       some independence as to how Officers in this division have a
                                    daily work, how much
                                                                        out their daily assignments.        they work, but strong guidance lot of independence as to how
                                    discretion do they have to carry
                                                                        Their activities are defined in     from senior colleagues, or from they go about their daily duties.
                                    out their assignments? Can
                                                                        detail by senior colleagues or      rules and regulations.
                                    you give me an example?
                                                                        organisational guidelines.

                                    Is the burden of achieving your
                                                                                                                                                 Each member of the division
                                    division’s targets evenly           A small minority of staff           A majority of staff make
                                                                                                                                                 provides an equally valuable
                                    distributed across its different    undertake the vast majority of      valuable inputs, but it is by no
                                                                                                                                                 contribution, working where
                                    officers, or do some individuals    substantive work within the         means everyone who pulls
                                                                                                                                                 they can provide their highest
                                    consistently shoulder a greater     division.                           their weight.
                                                                                                                                                 value.
                                    burden than others?
                                                                       Often tasks are not staffed by
                                                                                                            Most jobs have the right staff
                                                                       the appropriate staff. Staff are
                                    Would you say that senior staff                                         on them, but there are
                                                                       allocated to tasks either                                                 The right staff are always used
                                    try to use the right staff for the                                      organisational constraints that
                                                                       randomly, or for reasons that                                             for a task.
                                    right job?                                                              limit the extent to which
                                                                       are not associated with
                                                                                                            effective matching happens.
                                                                       productivity.

                                                                        The division uses the same
                                                                                                            The division makes steps             The division always redefines
                                    Does your division make efforts     procedures no matter what. In
                                                                                                            towards responding to specific       its procedures to respond to
                                    to adjust to the specific needs     the face of specific needs or
                                                                                                            needs and peculiarities, but         the needs of communities/
                      Flexibility   and peculiarities of                community/ client peculiarities,
                                                                                                            stumbles if the specific needs       clients. It does its best to serve
                                    communities, clients, or other      it does not try to develop a
                                                                                                            are complex. Often, tailoring of     each individual need as best as
                                    stakeholders?                       ‘better fit’ but automatically
                                                                                                            services is often unsuccessful.      it can.
                                                                        uses the default procedures.

                                                                                                            New ideas or practices are           Seeking out and adopting
                                                                        There is no effort to incorporate
                                                                                                            sometimes adopted but in an          improved work practices is an
                                    How flexible would you say          new ideas or practices. When
                                                                                                            ad hoc way. These are                integral part of the division’s
                                    your division is in terms of        practice improvements do
                                                                                                            sometimes shared informally or       work. Improvements are
                                    responding to new and               happen, there is no effort to
                                                                                                            in a limited way, but the division   systematically disseminated
                                    improved work practices?            disseminate them through the
                                                                                                            does not actively encourage          throughout the division and
                                                                        division.
                                                                                                            this or monitor their adoption.      their adoption is monitored.
Table A1 Continued: Defining Management Practices Using the CS Indices
Management Practice           Topic                Question                          Score 1                           Score 3                           Score 5

                                                                                                          Poor performance is
                                                                                                          addressed, but on an ad hoc         Repeated poor performance is
                                                                           Poor performance is not
                                                                                                          basis. Use of intermediate          systematically addressed,
                                                                           addressed or is inconsistently
                                         Given past experience, how                                       interventions, such as training,    beginning with targeted
                           Performance                                     addressed. Poor performers
CS-Incentives/Monitoring                 would under-performance be                                       is inconsistent. Poor               intermediate interventions.
                           Incentives                                      rarely suffer consequences or
                                         tolerated in your division?                                      performers are sometimes            Persistently poor performers
                                                                           are removed from their
                                                                                                          removed from their positions        are moved to less critical roles
                                                                           positions.
                                                                                                          under conditions of repeated        or out of the organisation.
                                                                                                          poor performance.
                                                                                                           An officer may break the rules
                                                                                                           infrequently and not be            Any officer who breaks the
                                         Given past experience, are        Breaking the rules of the civil punished. An officer who           rules of the civil service is
                                         members of [respondent’s          service does not carry any      regularly breaks the rules may     punished; the underlying driver
                                         organisation] disciplined for     consequences in this division. be disciplined, but there would     is identified and rectified. On-
                                         breaking the rules of the civil   Guilty parties do not receive   be no other specific actions       going efforts are made to
                                         service?                          the stipulated punishment.      beyond this. The underlying        ensure the issue does not
                                                                                                           drivers of the behaviour can       arise again.
                                                                                                           persist indefinitely.

                                                                                                                                              The evaluation system
                                         Does your division use
                                                                                                         The evaluation system awards         rewards individuals (financially
                                         performance, targets, or          Officers in the division are
                                                                                                         good performance in principle        or non-financially) based on
                                         indicators for tracking and       rewarded (or not rewarded) in
                                                                                                         (financially or non-financially),    performance. Rewards are
                                         rewarding (financially or non-    the same way irrespective of
                                                                                                         but awards are not based on          given as a consequence of
                                         financially) the performance of   their performance.
                                                                                                         clear criteria/processes.            well-defined and monitored
                                         its officers?
                                                                                                                                              individual achievements.

                                                                        Measures tracked are not
                                                                                                             Performance indicators have
                                                                        appropriate or do not indicate                                        Performance is continuously
                                                                                                             been specified but may not be
                                         In what kind of ways does your directly if overall objectives are                                    tracked, both formally with key
                                                                                                             relevant to the division’s
                                         division track how well it is  being met. Tracking is an ad                                          performance indicators and
                            Monitoring                                                                       objectives. The division has
                                         delivering services? Can you hoc process and most                                                    informally, using appropriate
                                                                                                             inclusive staff meetings where
                                         give me an example?            processes aren’t tracked at all.                                      indicators and including many
                                                                                                             staff discuss how they are
                                                                        Tracking is dominated by the                                          of the divisional staff.
                                                                                                             doing as division.
                                                                        head of the division.
Table A1 Continued: Defining Management Practices Using the CS Indices
Management Practice    Topic                Question                             Score 1                            Score 3                             Score 5


                                  Do you think about attracting                                                                              The division actively identifies
                                                                                                         Having top talent throughout
                                  talented people to your            Attracting, retaining and                                               and acts to attract talented
                                                                                                         the division is seen to be a key
                                  division and then doing your       developing talent throughout                                            people who will enrich the
                                                                                                         way to effectively deliver on
CS-Other              Staffing    best to keep them? For             the division is not a priority or                                       division. They then develop
                                                                                                         the organisations mandate but
                                  example, by ensuring they are      is not possible given service                                           those individuals for the
                                                                                                         there is no strategy to identify,
                                  happy and engaged with their       rules.                                                                  benefit of the division and try
                                                                                                         attract or train such talent.
                                  work.                                                                                                      to retain their services.

                                  If two senior level staff joined                                       There is some scope for high
                                                                                                                                             The division would certainly
                                  your division five years ago       The division promotes people        performers to move up through
                                                                                                                                             promote the high-performer
                                  and one was much better at         by tenure only, and thus            the service faster than non-
                                                                                                                                             faster, and would rapidly move
                                  their work than the other,         performance does not play a         performers in this division, but
                                                                                                                                             them to a senior position to
                                  would he/she be promoted           role in promotion.                  the process is gradual and
                                                                                                                                             capitalise on their skills.
                                  through the service faster?                                            vulnerable to inefficiencies.

                                                                                                                                             Targets are defined for the
                                                                The division’s targets are very
                                                                                                         Targets are defined for the         division and individuals
                                  Does your division have a     loosely defined or not defined
                                                                                                         division and its individual         (managers and staff) and they
                                  clear set of targets derived  at all; if they exist, they are
                                                                                                         officers (managers and staff).      provide a clear guide to the
                                  from the organization’s goals rarely used to determine our
                      Targeting                                                                          However, their use is relatively    division and its staff as to what
                                  and objectives? Are they used work schedule and our
                                                                                                         ad hoc and many of the              the division should do. They
                                  to determine your work        activities are based on ad hoc
                                                                                                         division’s activities do not        are frequently discussed and
                                  schedule?                     directives from senior
                                                                                                         relate to those targets.            used to benchmark
                                                                management.
                                                                                                                                             performance.
                                                                                                         To some extent, or at least on
                                  When you arrive at work each                                                                               Yes. It is always clear to the
                                                                     No. There is a general level of     some days. The
                                  day, do you and your                                                                                       body of staff what the
                                                                     confusion as to what the            organisation’s main goals and
                                  colleagues know what their                                                                                 organisation is aiming to
                                                                     organisation is trying to           individual’s roles to achieve
                                  individual roles and                                                                                       achieve with the days activities
                                                                     achieve on a daily basis and        them are relatively clear, but it
                                  responsibilities are in                                                                                    and what individual’s roles and
                                                                     what individual’s roles are         is sometimes difficult to see
                                  achieving the organisation’s                                                                               responsibilities are towards
                                                                     towards those goals.                how current activities are
                                  goals?                                                                                                     that.
                                                                                                         moving us towards those.
Table A2: Robustness
Dependent Variable: Completion Binary [yes=1] in Columns 1 and 3 through 9; Completion Index in Column 2
Standard Errors: Clustered by Organization
OLS Estimates
                                                                                                                                                           (6) Excl. Orgs. (7) Defining
                                                                                                     (4) Excl. Five        (5) Excl. Orgs. Below
                                                             (2) Average       (3) Excl. Org.                                                                Below 5%      Management
                                                                                                      Orgs. With             5% or Above 95%                                            (8) Fractional (9) Alternative
                                          (1) Baseline       Completion          With Most                                                                or Above 95% of Scores Using
                                                                                                    Smallest No. of          of CS-Autonomy                                              regression     Fixed Effects
                                                                 Rate             Projects                                                                 CS-Inc./Monit.   Orgs with
                                                                                                        Projects                   Scale
                                                                                                                                                                Scale      Project Data

CS-Autonomy                                   0.28***           0.12***             0.24***              0.30***                     0.35***                    0.30***               0.26***             1.25***             0.26***
                                              (0.04)             (0.03)             (0.04)                (0.04)                      (0.02)                     (0.04)                (0.04)              (0.19)              (0.04)
CS-Incentives/Monitoring                     -0.18***           -0.11***           -0.19***              -0.23***                   -0.21***                    -0.21***              -0.14***            -0.80***            -0.16***
                                              (0.05)             (0.03)             (0.05)                (0.05)                      (0.03)                     (0.05)                (0.04)              (0.22)              (0.04)
CS-Other                                     -0.07***            -0.03**           -0.08***              -0.06***                   -0.06***                    -0.07***              -0.06***            -0.35***            -0.08***
                                              (0.02)             (0.01)             (0.02)                (0.02)                      (0.01)                     (0.02)                (0.02)              (0.09)              (0.02)

Noise Controls                                 Yes                Yes                 Yes                  Yes                         Yes                        Yes                   Yes                 Yes                 Yes
General and Capital Controls                   Yes                Yes                 Yes                  Yes                         Yes                        Yes                   Yes                 Yes                 Yes
Project Controls                               Yes                Yes                 Yes                  Yes                         Yes                        Yes                   Yes                 Yes                 Yes
                                          Project Type, Project Type,           Project Type,         Project Type,                                         Project Type,         Project Type,       Project Type,       Project Type
Fixed Effects                                                                                                                Project Type, Sector
                                             Sector        Sector                  Sector                Sector                                                Sector                Sector              Sector           Within Sector
Observations (clusters)                     3620 (30)          3620 (30)          3125 (29)             3593 (26)                  3379 (27)                   3585 (29)             3620 (30)          3620 (30)            3620 (30)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization throughout. All columns report OLS estimates. Column 8 reports estimates from a fractional
regression model. In Columns 1 and 3 through 9, the dependent variable is a dummy variable that takes the value 1 if the project is fully completed and 0 otherwise. In Column 2, the dependent variable is an index of project
completion (that is a continuous measure between zero and one). Column 3 excludes projects implemented by the largest organization in terms of number of projects. Column 4 removes the 5 smallest organizations by number of
projects. Columns 5 and 6 exclude organizations at the top and bottom of the CS-autonomy and CS-incentives/monitoring management scales respectively. Column 7 uses the 30 focal organisations to define the management z-
scores. Project Type fixed effects relate to whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT Management and Research', 'Monitoring, Training and
Personnel Management', 'Physical infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to whether the project is in the administration, environment, finance, infrastructure, security/diplomacy/justice or
social sector. Project controls comprise project-level controls for whether the project is regularly implemented by the organization or a one off, whether the project is a bundle of interconnected outputs, and whether the division has to
coordinate with actors external to government to implement the project. Capital controls comprise a count of the number of interviews undertaken, which is a close approximation of the total number of employees. General controls
comprise organization-level controls for the share of the workforce with degrees, the share of the workforce with postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender,
and tenure of all respondents, the average time of day the interview was conducted and of the reliability of the information as coded by the interviewer. Figures are rounded to two decimal places.
Table A3: Differential Clustering
Dependent Variable: Completion Binary [yes=1]
Standard Errors: Clustered by Organization
OLS Estimates
                                                       (1) Baseline:                             (3) Project Type             (4) Project Type
                                                                              (2) Robust                                                                (5) Sector
                                                       Organization                             Within Organization            Within Sector

CS-Autonomy                                                0.28***               0.28***                 0.28***                    0.28***               0.28***
                                                            (0.04)               (0.05)                   (0.05)                     (0.06)                (0.05)
CS-Incentives/Monitoring                                  -0.18***              -0.18***                -0.18***                    -0.18**               -0.18***
                                                            (0.05)               (0.05)                   (0.07)                     (0.07)                (0.06)
CS-Other                                                  -0.07***              -0.07***                 -0.07**                    -0.07**               -0.07***
                                                            (0.02)               (0.02)                   (0.04)                     (0.04)                (0.02)

Noise, General and Capital Controls                          Yes                   Yes                     Yes                        Yes                   Yes
Project Controls                                            Yes                   Yes                      Yes                       Yes                   Yes
                                                       Project Type,         Project Type,                                      Project Type,         Project Type,
Fixed Effects                                                                                    Project Type, Sector
                                                          Sector                Sector                                             Sector                Sector
Observations (clusters)                                 3620 (30)                3620                  3620 (167)                3620 (41)              3620 (6)

Notes: *** denotes significance at 1%, ** at 5%, and * at 10% level. Standard errors are in parentheses, and are clustered by organization in Column 1, by project
type within organization in Column 3, by project type within sector in Column 4, and by sector type in Column 5. In Column 2, robust standard errors are reported. All
columns report OLS estimates.The dependent variable is a dummy variable that takes the value 1 if the project is fully completed and 0 otherwise. Project Type fixed
effects relate to whether the primary classification of the project is 'Advocacy and Policy Development', 'Financial & Budget Management', 'ICT Management and
Research', 'Monitoring, Training and Personnel Management', 'Physical infrastructure', 'Permits and Regulation' or 'Procurement'. Sector fixed effects relate to
whether the project is in the administration, environment, finance, infrastructure, security/diplomacy/justice or social sector. Project controls comprise project-level
controls for whether the project is regularly implemented by the organization or a one off, whether the project is a bundle of interconnected outputs, and whether the
division has to coordinate with actors external to government to implement the project. Capital controls comprise a count of the number of interviews undertaken,
which is a close approximation of the total number of employees. General controls comprise organization-level controls for the share of the workforce with degrees,
the share of the workforce with postgraduate qualifications, and the span of control. Noise controls are averages of indicators of the seniority, gender, and tenure of
all respondents, the average time of day the interview was conducted and of the reliability of the information as coded by the interviewer. Figures are rounded to two
decimal places.
Table A4: Ghana 2013 Management Survey
Management Practice    Topic         Ghana 2013 questions                  Score 1 (2013)                     Score 3 (2013)                     Score 5 (2013)


                                                                    Procedures for undertaking          Procedures for routine
                                                                    routine operations are              operations are defined and
                                                                    undefined or are rarely             generally followed, but may be
                                                                    followed in practice. The           inefficient or arbitrary. Where    Routine operations procedures
                                                                    organization relies almost          responsibility for work is         have been consciously
                                    Examines how well procedures entirely on top-down                   assigned to individuals, follow-   structured to improve quality
CS-Autonomy             Roles
                                    for routine work are structured assignment to get work done,        up and checks sometimes            and efficiency, and staff
                                                                    so execution depends on the         occur but are inconsistent.        consistently implement these
                                                                    individuals involved. Main          There is some consideration of     procedures.
                                                                    rationale for current practice is   how best to undertake day-to-
                                                                    “that’s the way it’s always been    day operations but this is
                                                                    done”.                              mainly on an ad hoc basis.

                                                                                                   Problems are sometimes
                                                                                                   detected before they have
                                                                                                   negative impacts, but there is          Exposing problems in a
                                                                                                   no structured, proactive                structured way is integral to
                                                                    Problems are only detected
                                                                                                   process to do so.                       individuals’ responsibilities and
                                                                    after they have affected work.
                                                                                                   Improvements are made on an             resolution occurs as a part of
                                    Examines processes for and      No process improvements are
                                                                                                   ad hoc basis, but depend                normal business procedures
                      Flexibility   attitudes towards continuous    usually made when problems
                                                                                                   entirely on individuals taking          rather than by extraordinary
                                    improvement                     occur. Frontline staff do not
                                                                                                   initiative. Resolution of               individual effort or ad hoc
                                                                    contribute to improving
                                                                                                   problems involves some, but             teams. Resolution of problems
                                                                    processes.
                                                                                                   not all, of the appropriate             involves all appropriate
                                                                                                   individuals and staff groups, or        individuals and staff groups.
                                                                                                   includes non-relevant
                                                                                                   individuals or groups.
                                                                                                   New ideas or practices are              Seeking out and adopting
                                                                    There is no effort to
                                                                                                   sometimes adopted but in an             practice improvements is an
                                                                    incorporate new ideas or
                                    Examines whether practice                                      ad hoc way. These are                   integral part of the
                                                                    practices. When practice
                                    improvements are effectively                                   sometimes shared informally             organization’s regular work.
                                                                    improvements do happen,
                                    disseminated throughout the                                    or in a limited way, but the            Improvements are
                                                                    there is no effort to
                                    organization                                                   organization does not actively          systematically disseminated
                                                                    disseminate them through the
                                                                                                   encourage this or monitor their         throughout the organization
                                                                    organization.
                                                                                                   adoption.                               and monitored for adoption.
Table A4 Continued: Ghana 2013 Management Survey
Management Practice           Topic       Ghana 2013 questions                 Score 1 (2013)                     Score 3 (2013)                     Score 5 (2013)

                                                                                                            Poor performance is
                                                                                                            addressed, but on an ad hoc        Repeated poor performance is
                                                                         Poor performance is not
                                                                                                            basis. Use of intermediate         systematically addressed,
                                                                         addressed or is inconsistently
                                         Examines whether the                                               interventions, such as training,   beginning with targeted
                           Performance                                   addressed. Poor performers
CS-Incentives/Monitoring                 organization is able to deal                                       is inconsistent. Poor              intermediate interventions.
                           Incentives                                    rarely suffer consequences or
                                         with underperformers                                               performers are sometimes           Persistently poor performers
                                                                         are removed from their
                                                                                                            removed from their positions       are moved to less critical roles
                                                                         positions.
                                                                                                            under conditions of repeated       or out of the organization.
                                                                                                            poor performance.
                                                                                                      Responsibility for follow-up             Responsibility for follow-up is
                                                                                                      may be allocated, but there              clearly defined and
                                         Examines how well the          Failure to achieve agreed     would be no other specific               accompanied by other
                                         organization follows up on and objectives does not carry any actions beyond this. Failure to          complementary steps. Failure
                                         enacts plans to fix identified consequences. Responsibility achieve agreed results is                 to achieve agreed objectives
                                         problems (department/          for following up is poorly    tolerated for a period before            is quickly addressed and may
                                         process-level)                 defined.                      action is taken and sanctions            lead to sanctions. Ongoing
                                                                                                      are rare. Difficult problems can         efforts are made to fix even
                                                                                                      persist indefinitely.                    difficult problems.

                                                                                                                                               There is an evaluation system
                                                                                                       There is an evaluation system           which rewards individuals
                                                                                                       which in principle awards good          (financially or non-financially)
                                         Examines whether good          People are rewarded in the     performance (financially or             based on performance.
                                         performance is rewarded        same way irrespective of their non-financially), but awards            Rewards are given as a
                                         proportionately and influences performance. Promotions are are not based on clear criteria/           consequence of well-defined
                                         career progression             solely based on tenure.        processes. Performance can              and monitored individual
                                                                                                       sometimes influence career              achievements. Performance is
                                                                                                       progression.                            a key criterion for promotion
                                                                                                                                               decisions.

                                                                         Measures tracked are not           Performance indicators have
                                         Examines whether
                                                                         appropriate or do not indicate     been specified but may not be      Performance is continuously
                                         organizational performance is
                                                                         directly if overall objectives     relevant to the organization’s     tracked, both formally and
                            Monitoring   tracked using meaningful
                                                                         are being met. Tracking is an      objectives. Some important         informally, using appropriate
                                         metrics and with appropriate
                                                                         ad hoc process and most            performance indicators are         indicators.
                                         regularity
                                                                         processes aren’t tracked at all.   tracked formally.
                                                                       Performance is reviewed
                                                                       infrequently or in an un-            Performance is reviewed
                                                                                                                                               Performance is continually
                                                                       meaningful way (e.g. only            periodically with appropriate
                                                                                                                                               reviewed, based on the
                                                                       success or failure is noted).        data, and both successes and
                                                                                                                                               indicators tracked, with a
                                                                       The right information for a          failures identified. Objectives
                                         Examines whether                                                                                      focus on problem solving and
                                                                       constructive discussion is           of meetings are clear to all
                                         organizational performance is                                                                         addressing root causes.
                                                                       often not present.                   participating, but results are
                                         meaningfully reviewed with                                                                            Purpose and follow-up steps
                                                                       Conversations are ‘one-way’,         communicated only to senior
                                         appropriate frequency and                                                                             are clear to all. Meetings are
                                                                       focus overly on data that is not     staff and conversations often
                                         communicated to staff                                                                                 an opportunity for constructive
                                                                       meaningful, and exclude most         do not drive to the root causes
                                                                                                                                               feedback and coaching.
                                                                       staff. The purpose of the            of problems. Next steps are
                                                                                                                                               Results are communicated to
                                                                       review is not explicitly stated.     mentioned but not well
                                                                                                                                               all staff.
                                                                       Next steps are not clearly           defined.
                                                                       defined.
Table A4 Continued: Ghana 2013 Management Survey
Management Practice    Topic       Ghana 2013 questions                   Score 1 (2013)                       Score 3 (2013)                       Score 5 (2013)


                                                                                                        The organization occasionally
                                                                 The organization does not go           tries to recruit applicants with
                                                                 out of its way to recruit talented     particular skills, but
                                                                                                                                             The organization actively
                                                                 applicants or applicants with          identification of needed skills is
                                                                                                                                             recruits talented applicants and
                                                                 particular skills, so hiring is        not systematic and recruitment
                                  Examines the emphasis put on                                                                               applicants with needed skills,
                                                                 driven entirely by staff numbers       is only occasionally based on
                                  attracting talented applicants                                                                             based on regular assessments
CS-Other              Staffing                                   and availability of funds.             identified skill needs. The
                                  and applicants with particular                                                                             of skill needs. Training
                                                                 Training opportunities are not         organization rarely goes out of
                                  skills                                                                                                     opportunities are sought and
                                                                 actively sought, and when              its way to recruit talented
                                                                                                                                             allocated to the staff who can
                                                                 available are not allocated            applicants. Training
                                                                                                                                             most benefit from them.
                                                                 based on capacity and/ or              opportunities are occasionally
                                                                 need.                                  sought, but driven mostly by
                                                                                                        availability.

                                  Examines whether the             Objectives and targets are           Some relevant objectives and
                                                                                                                                             A balanced range of objectives
                                  organization defines an          vague, only externally imposed,      targets are defined, but these
                                                                                                                                             and targets is defined for the
                                  appropriate range of             or non-existent. The only            mainly consist of lists of
                      Targeting                                                                                                              organization, including process
                                  meaningful objectives and/ or    targets are lists of activities      activities and related outputs.
                                                                                                                                             targets and other performance-
                                  targets to operationalize its    drawn up as part of the budget       These may be loosely linked to
                                                                                                                                             related targets.
                                  goals                            process.                             outcome targets.


                                                                                                        Objectives may be formally
                                                                                                        defined but are vague or
                                                                   Objectives are undefined or
                                                                                                        inconsistently applied. Staff
                                                                   very vague. Staff are expected                                            Individuals understand clearly
                                                                                                        have a general sense what
                                                                   simply to “do what they’re told”.                                         what they are responsible for
                                                                                                        they are responsible for, but
                                  Examines how well the            Tasks are allocated to                                                    and what they should be doing
                                                                                                        completion of most tasks relies
                                  organization allocates tasks     individuals on an ad hoc basis,                                           at all times. Objectives are
                                                                                                        mainly on the giving/ following
                                  and defines individual           with little or no consideration of                                        meaningful and are defined
                                                                                                        of direct orders. There is some
                                  objectives                       individual competences or                                                 through a two-way process.
                                                                                                        consideration of which tasks
                                                                   workloads. Staff are often                                                Tasks are allocated based on
                                                                                                        should be allocated to certain
                                                                   unsure what they should be                                                identified competences.
                                                                                                        individuals, but this is partial,
                                                                   doing.
                                                                                                        inconsistent, or based only on
                                                                                                        an officer’s formal position.


                                                                Goals do not cascade down the           Goals do cascade, but only to        Goals increase in specificity as
                                                                organization, are unclear, and          some staff and/ or division          they cascade, ultimately
                                  Examines how easily
                                                                are not clearly understood.             heads. Individuals have only a       defining individual expectations
                                  understandable objectives/
                                                                Beyond just following orders,           general sense of how their           for all staff groups. Individuals
                                  targets are and how well they
                                                                individuals do not know how             work contributes to                  understand clearly how their
                                  cascade down the organization
                                                                their work is related to overall        performance and meeting              work contributes to overall
                                                                performance.                            overall objectives.                  performance.
                             Figure A1: Quarterly Report, an Example



Division name




Expected output                                                                                                                     Actual output




                  Notes: Key information used in coding is highlighted; Appendix A provides details on all output data variables.