THE QUALITY OF 
GOVERNMENT INSTITUTE
 
  
Protected Areas and  
Spillovers on Corruption 
20 
24 
Working paper series 2024:16  
 
Aksel Sundström  
Stephen Dawson 
Sharon Pailler 
 
 
Protected Areas and Spillovers on Corruption   
Aksel Sundström    
Stephen Dawson   
Sharon Pailler  
  
WORKING PAPER SERIES 2024:16 
QoG THE QUALITY OF GOVERNMENT INSTITUTE 
Department of Political Science 
University of Gothenburg 
Box 711, SE 405 30 GÖTEBORG 
December 2024 
ISSN 1653-8919 
© 2024 by Aksel Sundström, Stephen Dawson and Sharon Pailler. All rights reserved. 
 
 
 
Protected Areas and Spillovers on Corruption   
Aksel Sundström    
Stephen Dawson   
Sharon Pailler   
QoG Working Paper Series 2024:16 
December 2024 
ISSN 1653-8919 
 
 
 
Abstract 
Do nature protected area (PA) establishments change local corruption levels? This article presents a theory 
of spatial spillovers, predicting (a) less corruption in areas inside protection, through increased government 
oversight and (b) more corruption outside of PAs because of corrupt activities’ displacement. To test these 
expectations we match geo-spatial information on the timing of the establishment of PAs in Africa with 
over 200,000 geo-coded Afrobarometer survey respondents. Our difference-in-differences approach finds 
little evidence of reductions in corruption inside PAs. However, we report an increase in experiences with 
bribery for respondents living in adjacent areas. Mechanism tests demonstrate an increased presence of 
police officers in these areas as well as larger effects in areas around PAs that attract tourists. We contribute 
to insights on linkages between conservation and government quality, a relationship with increasing rele-
vance due to the expansion of PAs around the world. 
 
 
 
 
   
Aksel Sundström Stephen Dawson   
The Quality of Government Institute The Quality of Government Institute 
Department of Political Science Department of Political Science 
University of Gothenburg University of Gothenburg 
aksel.sundstrom@pol.gu.se    
 
Sharon Pailler   
Research Director, ASPCA 
 
  
 
 
Acknowledgment 
The authors are thankful for comments from the EPG online seminar, the APSA annual conference, the 
ICSID workshop, the SWEPSA annual conference, QoG seminars, the CECAR conference and the SCAS 
in Uppsala. We also thank Aseem Prakash and several other colleagues for their feedback on an earlier draft. 
We are grateful for Erik Elwing and Magnus Åsblad, in their invaluable research assistance. 
 
Co-funded by Sundström’s ERC Starting Grant [project no. 101117889] and grants from the Swedish Re-
search Council [project no. 2020-03155, 2019-03218 and 2016-02119]. 
 
 
 
1 Introduction
Environmental regulations, such as protected areas (PAs), are crucial to meet global
biodiversity preservation targets and described as the ‘last line of defense’ for con-
serving nature (Joppa, Loarie, and Pimm 2009). Yet, PAs are repeatedly discussed in
relation to corruption in government. Robbins’ (2000) theory of ‘rotten institutions’
suggests that behavior such as bribery corrodes PAs, as it makes officials lax in en-
forcing rules to hinder violations. This matters since fear of misappropriation of
funds can discourage international donors to finance conservation schemes in the
Global South (Packer and Polasky 2018). The presence of corruption in government
institutions and conserving biodiversity through PAs are therefore both two policy
goals of high relevance. However, knowledge of how the two processes are related
is underdeveloped (Hanson and McNair 2014). Similarly, economists and political
scientists such as Ferraro (2005) and Barrett et al. (2006) note that links between
corruption and environmental regulations are more complex than often assumed.
This article provides a new perspective to this discussion by analyzing the impact
from the establishment of PAs on the spatial distribution of local corruption.
We aim to gain a further understanding of spillover effects, in the form of
changes in local corruption, from PAs. We ask: To what extent do PAs affect local
levels of corruption experiences and perceptions? We outline a theoretical frame-
work of how PA establishment can reduce local corruption inside the area due to (i)
the inflow of staff that are motivated to uphold rules and not entrenched in local
networks, (ii) the perceived spotlight from central authorities on officials in sectors
beyond conservation and (iii) a disruption of how resources are extracted. A fur-
1
ther spatial implication of our argument is that the benefits within PAs might be
coupled with negative changes outside these areas, as some corrupt behavior could
move to nearby locations with less scrutiny and new opportunities for illicit pay-
ments. Our reasoning has observable implications with regards to heterogeneity of
these effects that depend on park governance structures, the displacement of police
activities, and the level of tourism a PA attracts.
This argument is relevant beyond the discussion of nature protection and ties
into literature which holds that experiences with state institutions can affect peo-
ple’s view of government (Kumlin and Rothstein 2005; Martinangeli et al. 2024).
The larger question in this article therefore speak to a larger discussion related to
challenges associated with combating illicit behavior through institutional design:
whether regulations give rise to ‘red tape’ and new room to maneuver for rent-
seeing bureaucrats (Kaufman 1977) or, to the contrary, if regulations result in less
misbehavior (Levi 1988; Ostrom 1990).
We investigate this question focusing on the context of the African continent, a
region where many countries have invested considerably in conserving biodiver-
sity through PAs, whilst at the same time battling bribery in local authorities. We
match the geo-spatial data on the establishment of PAs with the location of close
to 215 000 Afrobarometer survey respondents in 30 countries, between 2002-2018.
Our focus is on changes in perceived and experienced local corruption. We use
a generalised difference-in-differences (DiD) approach with inverse probability of
treatment weighting (IPTW) and PA fixed effects to compare individuals inside and
close to PAs with those living far away. We also compare individual living inside
or close to PAs with the same areas prior to PA establishment. To test the mech-
2
anism of how PAs can affect local corruption, we develop novel data approaches
regarding the location of tourist attractions and their proximity to PAs using Open-
StreetMap (OSM) data.
Contrary to our first expectation, the results from our DID estimations suggest
little difference between those living inside a PA and those living far away from a
PA. However, we note a significant increase in local corruption in areas outside of
PAs, and increases are stronger for those who live within 10km of a PA. These re-
sults are particularly robust regarding experiences with having paid a bribe to the
police in those areas. While we find little evidence that increased central scrutiny
can deter corruption in PAs, we do detect a displacement of the police to the areas
surrounding PAs and stronger effects for perception and experience-based mea-
sures in areas around PAs with a considerable – but not very large – level of tourism
activity.
This article contributes to three different veins of research. First, our focus
complements the small but growing number of studies on the politics surround-
ing PA establishments (Alger 2023; Beacham 2023, 2024; Mangonnet, Kopas, and
Urpelainen 2022), by providing the novel insight that such establishments may also
impact local institutional quality. Second, our finding that the introduction of PA
institutions can increase bribery in areas outside state authorities’ attention speaks
to work in the corruption literature on displacement effects from anti-corruption
programs (Dávid-Barrett and Fazekas 2020). Third, this article provides a new per-
spective to a scholarly discussion on the influence from natural resource conser-
vation, which largely has focused on the broader societal impact from PAs (West,
Igoe, and Brockington 2006) and the socio-economic impact from such establish-
3
ments (De Sherbinin 2008; Andam et al. 2010; McKinnon et al. 2016; Naidoo et al.
2019). Finally, this article’s findings has also relevance for policy, given recent inter-
national commitments in the UN global biodiversity targets to protect 30 percent
of the Earth’s surface by the year 2030 (CBD 2021; Michaelsen, Sundström, and
Jagers 2024). There is, hence, a pressing need for a range of actors to understand
the nuances of how PA establishments may impact local institutional quality.
2 Prior research on corruption, natural resources and PAs
PA regulations are instrumental to achieve sustainable outcomes, where harvesters
are limited in their resource use (Dietz, Ostrom, and Stern 2003). If efficiently man-
aged, PAs may be a successful strategy to uphold biodiversity (Nolte et al. 2013;
Muñoz Brenes et al. 2018). This insight matters since governance of reserves in
low-income countries are often malfunctioning in practice (Gibson, Williams, and
Ostrom 2005; Halpern 2014), especially due to corruption, the misuse of public
power for private gain (Robbins 2000).1 Studies suggest that large-scale corrup-
tion reduces decision-makers’ tendency to designate areas as protected (Amacher
2006).2 Our focus is instead on small-scale transactions which can take a myriad of
shapes, such as favors and kickbacks, but often is thought of as bribes.3
Bribery cripples conservation regulations and studies across regions illustrate
how this takes numerous shapes; First, it increases access to PAs since people may
1There are national-level correlations between corruption levels and, for example, deforestation
(Koyuncu and Yilmaz 2009).
2E.g. large farmers in Latin America that bribe politicians for subsidies to convert lands (Barbier,
Damania, and Léonard 2005; Bulte, Damania, and López 2007).
3Termed petty corruption, involving officials (to be contrasted to grand corruption that targets
decision makers), divided into collusive acts (where officials turn a blind eye to law violations) and
non-collusive, when officials ask for payments for services citizens are entitled to (Smith et al. 2003).
4
collude with officials to gain entry (Hanson and McNair 2014). Second, it affects us-
age intensity, as bribes can allow harvesting exceeding permitted limits (Sundström
2013). Third, such payments help officials to lure landowners from land which is
sold to companies seeking to extract resources (Chayes 2017). Fourth, bribery en-
ables industrial actors to initiate activities in PAs (Beevers 2015). Fifth, it may make
resources harvested inside PAs appear legal through the falsification of documents
(Smith et al. 2003). Sixth, bribery assists the smuggling of illegally harvested goods
across borders (Gore, Ratsimbazafy, and Lute 2013). A final illustration is how con-
servation funds are pocketed, through e.g., selling monitoring equipment or hiring
‘ghost employees’ as rangers that in reality does not exist (Cavanagh 2012), or how
funds for benefit-sharing only reach local leaders (Smith and Walpole 2005).
This suggests that corruption erodes the effectiveness of PA management. Yet,
as Katzner (2005) notes, the relationship between corruption and conservation is
highly complex and Barrett et al. (2006) describe these links as “many and tangled”
(p. 1365). To date, we lack a clear understanding of spatial spillovers from PA in-
stitutions on local levels of corruption. This hints at a blind-spot in the literature.
Regarding definitions, spillover effects is a term used to describe unintended effects
from individual behavior (Truelove et al. 2014). We extend this concept to be un-
derstood as unintended effects on local levels of corruption from PA regulations.4
4Our use of this term differs from conservation scientists, denoting spillovers as effects on animal
populations outside a PA (Stamoulis and Friedlander 2013). There is also a discussion in management
studies on spillovers between organizational levels, e.g. when the public’s perception of legitimacy
of one actor has an impact on the legitimacy of other actors (Haack, Pfarrer, and Scherer 2014).
5
3 Theory
3.1 Changes in corruption norms
Corruption is a ‘wicked problem’, both difficult to spot and change (Rose-
Ackerman 1999). It matters how widespread it is locally and how ingrained it is
in norms that structure citizens’ and officials’ interactions (Bardhan 1997). When
corruption is rife, changes are difficult as citizens and officials often infer that
most other actors will engage in bribery (Dong, Dulleck, and Torgler 2012). Anti-
corruption scholars grapple to understand how such equilibriums can be shifted
and whether incremental reforms can set in motion virtuous circles of reduced cor-
ruption or if only overhauls of change across sectors is what works (Rothstein 2011).
Attempts to tackle corruption often center on shifting the costs and benefits of
bribery, and has involved external efforts to break up established networks. An ex-
ample is staff-rotation schemes, instigated to hinder officials to establish ingrained
ties with locals (Fjeldstad 2003). Importantly, changes in corruption in one specific
sector might create ripples, leading to transformations across spheres in the com-
munity. Isaksson and Kotsadam (2018) theorizes on how events that lead to change
in corruption can have effects beyond the directly affected sector. If external events
or actors manage to “establishing standards of conduct that delegitimize and stig-
matize corrupt practices” (p. 147), these notions can travel in the locality. They
argue that such events “could potentially influence social norms and thereby insti-
gate institutional change” (p. 147). A further way to understand local diffusion of
norms about bribery draws on the perspective where corruption is seen as a collec-
tive action problem: When people think that a sufficiently large portion of officials
6
and neighbours are engaged in – or reject engagement in – bribery, it shapes their
overall impression of state authorities (Persson, Rothstein, and Teorell 2013). We
believe such considerations to be in play when people’s corruption perceptions in
a community are affected by institutional change.
3.2 How PAs could affect perceptions and experiences of corruption
The establishment of PAs may shape a range of attitudes of people in their vicinity,
including; motivations to preserve nature (Agrawal, Chhatre, and Gerber 2015),
perceptions of state authorities (Macura et al. 2011), and trust towards conserva-
tion management (Gillingham and Lee 1999). We deem it likely that people’s as-
sessments of officials’ tendency to engage in bribery are also malleable to events
following the establishment of a PA. With regards to the area inside of PAs, we
suggest that we could witness a decrease in corruption and that this could happen
because of one or several of three different processes, outlined below.
(i) Inflow of new and educated staff
When PAs are introduced to a locality, there might be an inflow of new personnel
tasked to manage the area, such as rangers, service staff or people handling tourists.
Given that these people come from the outside, they are possibly not as embedded
in local networks of bribery as residents. This introduces uncertainty among actors
(of whom might tell authorities if bribes are suggested and established patterns of
favors) and therefore could disrupt local equilibria of corrupt practices. This no-
tion mirrors the idea that bureaucrats that are new to a locality are less likely to ask
for bribes because they are not yet familiar with entrusted networks (Abbink 2004).
7
Moreover, the personnel in PAs could very well be trained in conservation prac-
tices and uphold certain professional norms of combating poaching (and factors
facilitating poaching, such as corruption). In other words, there might be a differ-
ent ‘ethos’ among staff with conservation education and training. If PA institutions
that are established include staff that are well educated and manage to uphold an
esprit de corps — i.e., the feeling of loyalty among civil servants to an organiza-
tion’s goal (Rauch and Evans 2000) -– with a norm of protecting resources, this
could make them committed not to take bribes when interacting with officials or
citizens. This idea resonates with Kaufman (1960), who studies how forest rangers
uphold a culture where employees internalize certain professional values and are
not captured by local interests. An ideal outcome after the insertion of staff from
central authorities could therefore be one of personnel that abstain from taking or
paying bribes, setting an example of how impartial servants are meant to act. In
such a scenario, PAs could function as ‘pockets of efficiency’ in otherwise dysfunc-
tional settings (Evans 1989).
(ii) Perceived spotlight from central authorities
With investments in protecting an area from exploitation, there will be an in-
crease in monitoring and surveillance from central authorities. This can be about
rangers employed in the PA, but also about more fiscal controls, as these invest-
ments are likely coupled with instruments such as audits and other means to ensure
that funds are not siphoned off. In sum, this could create the perception that the lo-
cality is now under the ‘spotlight’ of central authorities, with more scrutiny of what
happens in an otherwise distant locality. Such increased scrutiny could plausibly
8
affect practices in sectors beyond conservation. For instance, a policeman that reg-
ularly ask for fees from drivers (to make them evade charges), could now perceive
more risks with their practices. Behavior that was previously normalized could
now be associated with a higher risk of reprimands from central authorities. The
feeling of increased surveillance from the center in PAs could make officials more
cautious in everyday practices in these areas. This is in line with ideas that institu-
tional reforms hinder corruption if they introduce uncertainty among bureaucrats
that change the calculus that one can violate rules without being sanctioned (Rose-
Ackerman 1999). We suggest that changes in corruption practices will create ripples
in the local community, where PA establishments affect local officials’ and citizens’
tendency to be involved in bribery in areas becoming protected. The process we
envision is one of norm transmission: seeing that the newly-arrived conservation
staff are less corrupt could have an effect on peoples’ perception of state authorities
in general. That is, we build on the suggestion that external events can affect local
standards of whether the use of bribery is condemned or not (Knutsen et al. 2017;
Isaksson and Kotsadam 2018) and we assume that people’s assessment of whether
to engage in local corruption or not will be affected by their perception of how other
people act (Dong, Dulleck, and Torgler 2012; Persson, Rothstein, and Teorell 2013).
(iii) Disruptions to resource extraction
The literature suggests that introducing PAs is often a process far from friction-
less, as they present local residents with restrictions on harvesting activities they
rely upon for their livelihood and, in a similar fashion, disturb activities of poach-
ers in the area (Duffy and St John 2013; Gore 2017). Corrupt resource management
9
institutions are ‘rotten’ (Robbins 2000). In such conditions, local networks can ham-
per attempts to conserve nature and resources are often accessed through corrupt
deals with local power players that benefit from extraction, such as hunting game
or collecting timber (Sundström 2016). Officials, such as policemen or clerks, can
be involved in these deals by asking for money or favors in return for access to re-
stricted areas. In dysfunctional local institutions, the role of entrenched and long-
lasting relations is important, as predictability in people’s tendency to engage in
bribery lowers transaction costs in corrupt deals (Lambsdorff 2002). Having a PA
established in such settings could disrupt these activities, by setting up a stricter
regime, and a system of monitoring. With the presence of rangers, there will likely
be a stark change in how people can extract resources and officials opportunities
to benefit from this. If the PA institution manages to uphold standards, it might
affect people in the area not being able to engage in rule violations, facilitated by
payments to officials.
We acknowledge that the above reasoning could build on an idealized view
of PA governance, and we hold it as plausible that there are many cases on the
African continent where rangers are recruited locally, where these establishments
are not meaningful enough to alter local status quo and where resource extraction is
not significantly altered and could even give rise to new opportunities for corrupt
behavior. Still, all things alike, we expect areas inside PAs to witness a change and
formulate the following expectation:
Hypothesis 1 Areas where PAs are established experience a decrease in local corruption
perceptions and experiences.
10
An implication of our spatial framework is that the benefits of reduced practices
of corruption within PAs might be coupled with negative changes outside these
areas. We suggest that this could take place because of processes of ’displacement.’
If actors were to perceive that the increased spotlight from state authorities’ atten-
tion reduce their income from bribery, it is possible that they for rational reasons
would relocate their practices to areas with less perceived scrutiny from the cen-
ter. From this follows that we would see a displacement of some behaviors that are
associated with corruption. This should refer to corruption-related activities that
are spatially flexible: police can set up random checks along roads (“shakedowns”
to collect an extra toll on travelling residents) in areas where central authorities are
less visible. This might push e.g. aspects of police corruption out from areas turned
to PAs. It could also refer to people in the locality choosing to work elsewhere (e.g.
the teacher wishing to hold a position where absenteeism is not perceived as a
risky behavior). In other words, officials in area X that are accustomed to use cor-
rupt behaviors to increase their income (e.g., clerks that take extra fees to provide
their public services) are incentivized to relocate such practices to neighboring ar-
eas outside of X after it has become protected. This reasoning has similarities to the
argument about unintended effects from anti-corruption reforms by Dávid-Barrett
and Fazekas (2020), suggesting that those wishing to engage in corrupt activities
have reasons to avoid sectors that are tightly controlled by authorities. We build on
this insight and propose that these considerations could be geographical in nature.
In sum, we suggest that we should observe an increase in local levels of corruption
experiences and perceptions outside of PAs, that decrease with the distance from
these borders.
11
Besides displacement of activities, the areas outside of PAs might also function
as ’honey pots’, in the words of Knutsen et al. (2017), where officials such as police
officers are attracted because of new opportunities of side payments. It is plausible
that areas outside of PAs pull actors that seek lucrative ways to get illicit side pay-
ments. Anecdotally, it is illustrated that areas near PAs are considered ’plum posts’
(highly desired positions) that are profitable for police officers, because they enable
a toll on economic activities (Petchkaew 2023). As such, this reasoning would im-
ply that we should expect an increase in police presence in areas outside of PAs.
Further, there is important heterogenity between protected areas. PAs that attract
domestic and international flows of tourism (because of e.g., scenic views, safari
tours or hunting opportunities) stand in contrast to those that do not, because the
presence of tourists and surrounding economic activities also enable opportunities
for bribery. This points to the expectation that PAs that attracts visitors and tourism
investments are likely those where we would expect to see larger increases in cor-
ruption in the buffer zones outside of PAs. We therefore expect:
Hypothesis 2 Areas outside those where PAs are established experience an increase in
local corruption perceptions and experiences.
4 Data and approach
4.1 Data
4.1.1 Survey data
We examine the effects of PAs on local corruption in Africa. To do so, we use
geocoded Afrobarometer data, linking clusters of survey respondents (enumera-
12
tion areas) to GPS-points (BenYishay et al. 2017).5 The Afrobarometer Data (2016)
survey collects information about public attitudes surrounding governance and
corruption, as well as basic demographic information. We use data from rounds
2-7 (2002-2018), as some of the variables of interest in round 1 were not consistent
with the other rounds. This results in a final sample of 228,360 survey respondents
across 22,173 enumeration areas.
We focus on two items that estimate local corruption. One question captures
corruption perceptions. Respondents were asked: “How many of the following
people do you think are involved in corruption, or haven’t you heard enough about
them to say?” Here, we focus on the question that relates to the police. Answers are
coded on a 0-3 scale, with higher values indicating more widespread corruption.6
A second question gauges experiences with corruption and reads: “In the past year,
how often (if ever) have you had to pay a bribe, give a gift, or do a favor to gov-
ernment officials in order to avoid a problem with the police?”.7 We recode these
answers to a dummy variable which equals ’1’ if an individual has paid a bribe to
police in the past year, and ’0’ for those that responded ’Never’.
4.1.2 Proximity to protected areas
For protected areas (PAs), we rely on the World Database on Protected Areas
(WDPA) (UNEP-WCMC et al. 2019). The WDPA is part of a collaboration between
the UN Environment Programme (UNEP) and the International Union for the Con-
servation of Nature (IUCN) and stores the characteristics of protected areas (such
5The Afrobarometer survey draws a clustered, stratified, multi-stage, area probability sample. See
Appendix A2 for information of the countries sampled in each survey round.
6Respondents could respond ”All of them”, ”Most of them”, ”Some of them” and ”None”.
7Respondents could respond ”Never”, ”Once or twice”, ”A few times”, and ”Often”.
13
as their name, year of establishment, size, governance structure, etc.) as well as
their precise geographic location. PAs are recorded as spatial polygons or points
that approximate PA centroids. As we are interested in differences between those
who live on either side of the border of a PA, we consider only polygon geome-
tries in this study.8 The WDPA is compiled by a combination of governmental and
non-governmental agencies who report information about PAs in their territory to
Protected Planet. The completeness of the data for PAs therefore varies consider-
ably between countries, though each country in the Afrobarometer is also present
in the WDPA.
We capture the proximity of respondents to PAs in several ways. First, and for
the purposes of testing hypotheses, we create a series of variables for each survey
enumeration area (typically a city, town or village) that indicate whether that enu-
meration area lies within a PA, or within a series of distance ranges outside of at
least one PA (10km and 50km) within the same country. In cases where an enu-
meration area lies within a certain range of several PAs, we also record the count of
PAs within this range. In addition, we also calculate the proximity of respondents
to the single nearest PA within the same country, as well as the characteristics of
this PA as recorded by the WDPA.
8Examples of the few type of features we do not capture are, thus, monuments or certain rare
heritage sites, which in all only a very small portion of the data.
14
Figure 1: Proximity to protected areas
0.4
(A) (B)
0.15
0.3
0.10
0.2
0.05 0.1
0.00 0.0
0 10 20 30 40 50 60 70 80 90 100 Inside <10km 11−50km 50km+
Distance from PA (KM) Distance from PA − binned
Note: Light grey lines in Panel A represent individual countries, while the thick black line represents
the pooled density.
Figure 1 illustrates the distribution of our sample across space relative to pro-
tected areas, using the proximity to the closest PA in the same country. We consider
everyone who lives inside a PA equally, regardless of their proximity to the border
from the inside. Considering those outside of PAs on a continuous scale and those
on the inside categorically may introduce bias to our estimates. We therefore also
present the data in distance bins to compare groups who live in or to varying de-
grees outside protected areas. This is illustrated in panel B in Figure 1. The figure
demonstrates that around 4% of the sample (close to 10,000 respondents) live in-
side an existing PA. Panel A in Figure 2 maps protected areas and the geo-location
of respondents from the Afrobarometer, colour-coded by their proximity to their
nearest PA. Panel B magnifies the Zimbabwean case to more clearly demonstrate
15
Density
Proportion of sample
the structure of the data.
Figure 2: Survey respondents proximity to protected areas: mapped
40°N
N
(A)
30°N
20°N
PA proximity
10°N Inside
<10km
11−50km
 0°
>50km
(B)
10°S
Area protected
20°S
30°S
200 km
2000 km
20°W  0° 20°E 40°E
Note: Panel A shows the spatial distribution of protected areas (green) in Africa as of 2023 for
countries in our sample. Afrobarometer enumeration areas are plotted as points, colored by their
proximity to PAs at the time the survey was taken. Panel B shows a magnification of Zimbabwe as
an illustrative example.
4.2 Empirical strategy
Our baseline regression estimates the linear effect of residing inside a protected
area on perceptions and experiences of local corruption in the following way:
corruptionicy = β1protected + αj + ρy + γXi + ϵicy (1)
where the measure of corruption for individual i in enumeration area cluster
c in year y is regressed on a dummy indicator of whether or not they live in a
16
protected area, as determined by geocoded Afrobarometer data and the WDPA. In
the case of H2, we are also interested in the areas immediately outside of PAs. We
therefore estimate the following equation:
corruptionicy = β1protected + β2<10km + β311-50km + αj + ρy + γXi + ϵicy (2)
where the measure of corruption for individual i in enumeration area cluster c
in year y is regressed on a series of dummy indicators of whether or not they live
in a PA, within 10km on the outside of a PA, or between 11-50km outside of a PA.
While in equation 1 the comparison group is all who do not reside inside a PA,
whereas in the case of equation 2 the comparison group is those who live far away
(>50km) away from a PA.
To increase causal inference, we use a generalized difference-in-differences de-
sign in our baseline models using country (αj) and year (ρy) fixed effects. Country
fixed effects control for time-invariant differences between countries, such as polit-
ical institutions, whereas year fixed effects control for unobserved factors that vary
over time but are constant across countries, such as international agreements or the
value of natural resources. In all regressions we also control for a vector of indi-
vidual level control variables (Xi), specifically age, education, employment status,
gender and whether they live in an urban area. We also employ a more conserva-
tive approach in some models by including PA fixed effects – effectively limiting
comparisons to individuals who live inside and outside the same PA.
17
4.3 Addressing endogeneity
Inherent in this approach is a risk of endogeneity since the location of PAs is non-
random and and likely subject to not only biological and geological considerations,
but also political and social. For example, protected areas may be less likely to be
established in highly corrupt (though biodiverse) regions, as corruption can ob-
struct the enforcement of protection regulations and restrict economic benefits of
conservation such as tourism. We therefore attempt to navigate this particularly
acute risk by considering the likelihood of respondents (or more accurately enu-
meration areas) being protected in the first place.
To do this, we use propensity score matching to generate inverse probability of
treatment weights (IPTW). In this procedure, we consider our treatment variable –
residing inside a protected area – as the dependent variable that we regress on a
series of predictors that may be responsible for the endogenous placement of PAs.
Specifically, we estimated a logistic regression of local protection on local levels of
biodiversity, the local human development, the proximity from the EA to the near-
est worked valuable natural resource (gems, diamonds, gold and oil), whether the
EA is in an urban or rural area, and whether the EA is in an ethnic homeland of
a politically included group, as PA allocation can be subject to ethnic favoritism
(Dawson et al. 2024). This model was used to generate propensity scores that to-
gether with actual treatment status are used to calculate the IPTW. This weight is
then simply added to the regression equations outlined above. This approach ef-
fectively brings greater balance to treatment and control groups, by giving more
influence to observations in protected areas that are most similar to observations in
18
non-protected areas, and vice versa.9
5 Results
5.1 Main results
In Table 1, Models 1-2 and 5-6 present the results from equation 1. Models 3-4
and 7-8 investigate the spatial nature of the relationship outside of PAs referred to
in equation 2. Results are presented for the two corruption measures: individual
perceptions and experienced bribes. All models include country-year fixed effects,
and models 2, 4, 6 and 8 include more conservative PA fixed effects.
Table 1: Main Analysis
Perceptions Bribes
(1) (2) (3) (4) (5) (6) (7) (8)
Protected Area -0.007 -0.019 0.031 -0.009 0.006 0.000 0.011 0.007
(0.018) (0.021) (0.020) (0.023) (0.008) (0.009) (0.009) (0.010)
<10km 0.050*** 0.004 0.008* 0.010*
(0.010) (0.011) (0.003) (0.004)
11-50km 0.047*** 0.022* 0.005+ 0.007*
(0.009) (0.009) (0.003) (0.003)
Num.Obs. 201148 201148 201148 201148 220791 220791 220791 220791
Country-year FE Yes Yes Yes Yes Yes Yes Yes Yes
PA FE No Yes No Yes No Yes No Yes
R2 0.116 0.147 0.117 0.147 0.085 0.107 0.085 0.108
R2 Within 0.008 0.005 0.008 0.005 0.014 0.013 0.014 0.013
Note: The table reports OLS estimates. Robust standard errors clustered by enumeration area in parentheses. Unit
of observation is the individual survey respondent. All models include inverse probability of treatment weights
(IPTW). Significance levels: ∗∗∗p < .001; ∗∗p < .01; ∗p < .05; +p < .1.
Across the two measures, contrary to our expectations outlined in hypothesis 1,
we observe no anticipated negative effect of living inside as opposed to outside
a protected area on corruption levels. This is the case both for measures relating
9For the analysis and further discussion on the matching process, see Appendix C.
19
to perceptions of corruption in the police, and experiences with paying bribes to
the police. When we introduce proximity range dummy variables, however, we
observe indications of increases of corruption perceptions and experiences in the
areas immediately outside PAs, relative to areas far away from PAs. Even in the
more conservative specifications in models 4 and 8, there is a significant indica-
tion that areas outside the border of PA experience increases in local corruption.
More specifically, model 8 demonstrates a larger increase in experiences with pay-
ing bribes to police the closer one lives to a protected area. Living witin 10 kilome-
ters of a protected area increases the chances of an individual having paid a bribe
to the police in the last year by roughly 1 percentage point which, considering the
average likelihood of reporting having paid a bribe of 10.7%, is a substantial effect.
Thus, while there is little difference in corruption perceptions and experiences be-
tween those living inside vis-á-vis outside PAs, we do find evidence that PAs can
have negative side-effects on local corruption outside their borders, supporting our
hypothesis 2.10
5.2 Robustness tests
In addition to making our primary comparison across space (those living varying
distances away from PAs), we can also consider comparisons of the same areas at
different points in time relative to the establishment of PAs. In this section, we
use an alternative approach which directly compares individuals in each proximity
range before and after the establishment of a PA (Isaksson and Kotsadam 2018;
Knutsen et al. 2017). Each model presented in Table 2 also contains PA fixed effects,
10For an analysis of alternative corruption items, see Appendix Table A.3.
20
constraining the comparison to individuals who lived inside, or close to the same
PA before and after its establishment. The number of observations is significantly
reduced with this approach – particularly in the case of models 1 and 3 – due to
the limited number of observations in areas pre-PA establishment. The results of
this alternative approach are very similar to those presented in Table 1, and in fact
demonstrate a greater magnitude effect of living outside PAs on local corruption
relative to the same areas before those PAs existed.
Table 2. Alternative approach
Perceptions Bribes
(1) (2) (3) (4) (5) (6)
Active PA 0.034 -0.011
(0.097) (0.027)
Active PA <10km -0.004 0.017*
(0.025) (0.008)
Active PA 11-50km 0.055* 0.015+
(0.026) (0.009)
Num.Obs. 9627 79305 156030 10580 85551 169478
Country-year FE Yes Yes Yes Yes Yes Yes
PA FE Yes Yes Yes Yes Yes Yes
R2 0.196 0.153 0.140 0.170 0.120 0.112
R2 Within 0.005 0.004 0.004 0.012 0.013 0.013
Note: The table reports OLS estimates. Robust standard errors clustered by enumeration area
in parentheses. Unit of observation is the individual survey respondent. Each model directly
compares areas around active and inactive PAs. All models control for PA and country-year
fixed effects. Significance levels: ∗∗∗p < .001; ∗∗p < .01; ∗p < .05; +p < .1.
5.3 Mechanism test: Central oversight and state officials
In addition to these baseline results, we also consider several mechanisms through
which we contend that living inside or close to PAs may affect local corruption.
21
First, we suggest the effect on the area within a PA to be particularly acute when
the PA is centrally governed by national or federal agencies. The inflow of edu-
cated staff, perceived spotlight from central authorities, and disruptions to resource
extraction should all be more pronounced in PAs governed by the state (e.g., na-
tional parks). Because these parks tend to be associated with much more invest-
ments from government – often being denoted a national park because of more
precious biodiversity and visitor potential – there is much higher stakes at mak-
ing sure funds invested are allocated correctly. As such, it is possible that state
governed PAs will be associated with more spotlight from central authorities than
those managed by NGOs or private actors.
To investigate this aspect, we replicate Models 1 and 5 from Table 1 and create
a new set of distance dummy variables that indicate whether a survey respondent
lives inside or close to a centrally governed or non-centrally governed PA, respec-
tively. This information is derived from the WDPA. We collapse the classifications
detailed in the WDPA and consider here the distinction between those governed
by federal/national agencies and otherwise. The second group contains PAs gov-
erned by subnational agencies, individual/private organizations, non-state bodies,
and those for which information is not recorded. Federal PAs comprise around 50%
of protected areas in our sample. In this set-up, our expectation is that corruption
will decrease inside federal PAs relative to non-federal PAs.
Figure 3 presents a coefficient plot of the heterogenous effects of PAs with dif-
ferent governance structures on perceptions and experiences of corruption in the
police. Results do not suggest a significant difference between the effects of federal
and non-federal PAs for those living inside those areas. This is the case for both cor-
22
ruption measures. We do therefore not find support for the idea that governance
structure is responsible for the effects of PA establishment on local corruption.
Figure 3: Disaggregating effects by park governance type
Perception Bribery
0.04
0.00
Not federal
Federal
−0.04
−0.08
PA proximity: Inside
5.4 Mechanism test: PAs as honeypots
An alternative mechanism proposed to explain the increase in corruption levels
outside of PAs is that PA establishment brings with it the creation of economic op-
portunities and development in conservation and tourism-related industries. In
this section we test the viability of this mechanism by investigating two observable
implications: (1) there should be an increase in police presence in the areas sur-
rounding PAs and (2) the effect of PAs on corruption should be moderated by the
extent of tourism related to the PA.
23
Coefficient estimate
First, corrupt officials such as police officers should be attracted to operate in
the areas surrounding PAs to illicit bribes in return for access to them and/or the
extraction of resources from them. We therefore anticipate a greater presence of
police officials in the areas surrounding PAs. We test this by utilizing questions an-
swered by the Afrobarometer survey enumerators on whether they witnessed any
police officers or police vehicles in the enumeration area. In Table 2, we use a sim-
ilar approach to our main analysis, though the unit of analysis is the enumeration
area-year. We present two models: one with country-year fixed effects, and one
which adds district (administrative area level 2) fixed effects for a more conserva-
tive specification, effectively comparing the effect of proximity to protected areas
on police presence within districts. We also control for whether the enumeration
area is in an urban area.
In line with expectations, Table 2 indicates that police officials are more present
in areas immediately outside PAs. Moreover, enumeration areas within 10km of
the borders of a PA are the most likely to see an increase in the levels of police
officials on the ground. Based on a comparison within districts in model 2, we
report approximately a 4 percentage point increase in the likelihood of the police
being present in an enumeration area within 10km of a PA compared to enumera-
tion areas over 50km away from a PA. The baseline likelihood of police presence in
enumeration areas in our sample is 30.9%.
We test the moderation effect of tourism by considering heterogenous effects
across PAs that attract different levels of tourism. While some PAs such as strict
conservation areas are to varying extents closed off to visitors, others - such as
national parks - attract tourists in large numbers, generating considerable economic
24
Table 3: Police Presence
(1) (2)
Protected Area -0.014 0.017
(0.027) (0.031)
<10km 0.022* 0.039*
(0.010) (0.017)
11-50km 0.019* 0.032*
(0.009) (0.015)
Num.Obs. 26171 26022
Year FE Yes Yes
Country FE Yes No
District FE No Yes
R2 0.169 0.367
R2 Within 0.075 0.056
Note: The table reports OLS estimates. Robust standard errors clustered
by enumeration area in parentheses. Unit of observation is the enu-
meration area-year. All models include inverse probability of treatment
weights (IPTW). Significance levels: ∗∗∗p < .001; ∗∗p < .01; ∗p <
.05; +p < .1.
activity within and around those parks. To measure the level of tourism relating to
PAs in the African continent, we utilize tourist attraction geo-referenced point data
from OpenStreetMap (OSM) and carefully created a set of original variables for
this purpose. We identify any OSM point entry that includes a tourism ”tag”, the
most common of which in our case being ”hotel”, ”guest house”, or ”viewpoint”
(see a description and discussion of this data collection and construction of these
variables in Appendix A2). We merge these tags with the WDPA polygon data by
adding a 10km buffer around each PA and counting the number of tourism tags
that fall within each of these buffers.11 This results in a number of tourism tags
for each PA, which we equate to more or less tourism. We then recode this into an
ordinal variable with four levels: PAs with 0 tourism tags (40.2% of PAs), those with
11We use a 10km buffer as tourist-related institutions are frequently located outside rather than
inside PAs. Nevertheless, we analyse a variety of buffer ranges in the appendix, and correlate these
with a subsample of PAs where tourist visitor numbers are recorded and available. The 10km buffer
correlated the highest with these visitor numbers (R=0.85).
25
1-10 tags (35.9%), those with 11-50 tags (17.3%), and those with >50 tags (6.5%).
Figure 4 presents a coefficient plot of the effect of these different groups of PAs
on the two measures of corruption. These models replicate the approaches taken in
models 3 and 7 in Table 1. While the effect does not appear to be linear across levels
of tourism, there is indication that PAs with 11-50 tourism tags increase corruption
to a greater extent than other levels, and this is fairly consistent across different
corruption measures. Interestingly, however, the most touristic PAs (those with
more than 50 OSM tags) are seemingly not conducive to increases in local corrup-
tion - and even appear to be associated decreases in bribery relative to areas over
50km away from these areas. This is perhaps indicative of a qualitative difference
between PAs with very high levels of tourism and those where tourism is develop-
ing.
Figure 4: Disaggregating effects by the number of tourism attractions
Perception Bribery
0.2
0.1
0.0
−0.1
Inside <10km 11−50km Inside <10km 11−50km
PA proximity
OSM tourist attraction tags 0 1−10 11−50 50+
26
Coefficient estimate
6 Conclusion
This study takes a first step in considering the socio-political spillovers of nature
protection. By considering the perspectives of citizens who live in, or close to, pro-
tected areas in Africa, we present evidence that the establishment of PAs can have
unsavory consequences for levels of local corruption in adjacent areas. We present
a spatial theory of the societal consequences of conservation institutions, and test
derived expectations about areas inside and outside of PAs. Our findings point to
little effect inside PAs themselves, but a marked increase in experiences of bribery
in the areas surrounding PAs. Furthermore, we find evidence for an increase in
police presence around PAs, and that areas around PAs with a developing tourism
presence are at greater risk of corruption increases.
These findings provide new insights to our knowledge on the intricate linkages
between institutional quality and nature protection areas. Specifically, we highlight
that PAs may bring negative spillover effects -– possibly by displacing corrupt ac-
tivities out from these protected areas. This resonates with the caution voiced in
the United Nations Anti-Corruption Toolkit: “reforms that increase uncertainties
and the risk of criminal punishment or financial losses tend to reduce corruption..
[but] reforms must be broad-based and systemic, or corrupt conduct may simply
be displaced into other areas or activities” (UNODC 2024, p. 20). It also speaks to
the findings of Dávid-Barrett and Fazekas (2020), suggesting that actors engaged
in corruption behave strategically to avoid tightly-controlled sectors. Our theory
and empirical findings have novel implications and suggest that such strategies
can also be geographical, as increased oversight in certain areas can create spatial
27
spillovers to peripheral areas outside of government scrutiny.
While our tests are conservative, there are restrictions to inference that we must
also be aware of with the data used. First, our data comes from several Afrobarom-
eter rounds that sample different enumeration areas, so we are unable to consider
the same respondents or even EAs (in most cases) at different points in time. We
are also unable to corroborate whether the composition of EAs is comparable over
time which can be problematic for over-time comparisons. For example, one of the
primary human costs associated with protected areas is that they often result in the
(oftentimes forced) displacement of local inhabitants. PA establishment is therefore
likely associated with demographic shifts in the population. While this does not
preclude our theoretical approach (rather we theorize about the relocation of cor-
rupt individuals/practices to areas outside of PAs), it may restrict claims of causal
inference. Nevertheless, we attempt to navigate these issue by basing our conserva-
tive estimations on comparisons between individuals living inside/close to and far
away from the same PAs (while controlling for the likelihood of PA establishment
in the area), as well as people living in the same areas before and after PA establish-
ment. We therefore urge more research on this topic, using complementary types
of methods and materials.
Understanding the effects from PAs on local corruption have high relevance,
given recent trends in policy. The Kunming-Montreal Biodiversity Framework, rat-
ified by 196 countries in 2022, sets the ambitious target to protect 30 percent of
Earth’s surface by the year 2030 (CBD 2021). On the African continent, such an
expansion would translate to an average increase of about 50 percent of the area
under protection. This highlights the pressing need to understand whether the es-
28
tablishment of conservation institutions may bring spillover effects on institutional
quality and local corruption in government conduct.
29
References
Abbink, Klaus. 2004. “Staff rotation as an anti-corruption policy: an experimental
study.” European Journal of Political Economy 20(4): 887–906.
Afrobarometer Data. 2016. http://www.afrobarometer.org “South Africa.
Round 2-6. 2002-2016.”.
Agrawal, Arun, Ashwini Chhatre, and Elisabeth R. Gerber. 2015. “Motivational
Crowding in Sustainable Development Interventions.” American Political Science
Review 109(3): 470–487.
Alger, Justin. 2023. “The Political Economy of Protected Area Designations: Com-
mercial Interests in Conservation Policy.” Global Environmental Politics 23(2): 54–
73.
Amacher, Gregory S. 2006. “Corruption: A challenge for economists interested in
forest policy design.” Journal of Forest Economics 12(2): 85–89.
Andam, Kwaw S., Paul J. Ferraro, Katharine R. E. Sims, Andrew Healy, and Mar-
garet B. Holland. 2010. “Protected areas reduced poverty in Costa Rica and Thai-
land.” Proceedings of the National Academy of Sciences 107(22): 9996–10001.
Barbier, Edward B., Richard Damania, and Daniel Léonard. 2005. “Corruption,
trade and resource conversion.” Journal of Environmental Economics and Manage-
ment 50(2): 276–299.
Bardhan, Pranab. 1997. “Corruption and Development: A Review of Issues.” Jour-
nal of Economic Literature 35(3): 1320–1346.
Barrett, Christopher B., Clark C. Gibson, Barak Hoffman, and Mathew D. McCub-
bins. 2006. “The Complex Links between Governance and Biodiversity.” Conser-
vation Biology 20(5): 1358–1366.
Beacham, Austin. 2023. “Extraction, Contestation, and Conservation: Natural Re-
source Dependence and Protected Area Designation.”.
Beacham, Austin. 2024. “Conserving What’s Left: The International Political Econ-
omy of Protected Areas.”.
Beevers, Michael D. 2015. Large-scale mining in protected areas made possible
through corruption: Options for donors. Technical Report U4 Brief 2015:7 Chr.
Michelsen Institute Bergen: .
30
BenYishay, Ariel, Renee Rotberg, Jessica Wells, Zhonghui Lv, Seth Goodman, Lidia
Kovacevic, and Dan Runfola. 2017. http://geo.aiddata.org “Geocoding
Afrobarometer Rounds 1-6: Methodology & Data Quality.”.
Bulte, Erwin H., Richard Damania, and Ramón López. 2007. “On the gains of com-
mitting to inefficiency: Corruption, deforestation and low land productivity in
Latin America.” Journal of Environmental Economics and Management 54(3): 277–
295.
Cavanagh, Connor. 2012. Unready for REDD+? Lessons from corruption in Ugan-
dan conservation areas. Technical Report U4 Brief 2012:3 Chr. Michelsen Institute
Bergen: .
CBD, UN. 2021. “First draft of the post-2020 global biodiversity framework.” Paper
presented at the Convention on Biological Diversity,.
Chayes, Sarah. 2017. https://www.washingtonpost.com/outlook/
a-hidden-cost-of-corruption-environmental-devastation/
2017/06/16/03f93c1e-52b8-11e7-b064-828ba60fbb98_story.html
“A hidden cost of corruption: environmental devastation.”.
Dawson, Stephen, Felix Haass, Carl Müller-Crepon, and Aksel Sundström. 2024.
“The Ethnic Politics of Nature Protection: Ethnic Favoritism and Protected Areas
in Africa.” Quality of Government Working Paper Series 2024:4 .
De Sherbinin, Alex. 2008. “Is poverty more acute near parks? An assessment
of infant mortality rates around protected areas in developing countries.” Oryx
42(01): 26–35.
Dietz, Thomas, Elinor Ostrom, and Paul C. Stern. 2003. “The Struggle to Govern
the Commons.” Science 302(5652): 1907–1912.
Dong, Bin, Uwe Dulleck, and Benno Torgler. 2012. “Conditional corruption.” Jour-
nal of Economic Psychology 33(3): 609–627.
Duffy, Rosaleen, and Freya St John. 2013. Poverty, poaching and
trafficking: what are the links? Technical report Evidence on
Demand. https://www.gov.uk/dfid-research-outputs/
poverty-poaching-and-trafficking-what-are-the-links
Dávid-Barrett, Elizabeth, and Mihály Fazekas. 2020. “Anti-corruption in aid-
funded procurement: Is corruption reduced or merely displaced?” World De-
velopment 132: 105000.
31
Evans, Peter B. 1989. “Predatory, developmental, and other apparatuses: A com-
parative political economy perspective on the Third World state.” Sociological Fo-
rum 4(4): 561–587.
Ferraro, Paul. 2005. “Corruption and conservation: the need for empirical analyses.
A response to Smith & Walpole.” Oryx 39(3): 257–259.
Fjeldstad, Odd-Helge. 2003. “Fighting fiscal corruption: lessons from the Tanzania
Revenue Authority.” Public Administration and Development 23(2): 165–175.
Gibson, Clark C., John T. Williams, and Elinor Ostrom. 2005. “Local Enforcement
and Better Forests.” World Development 33(2): 273–284.
Gillingham, Sarah, and Phyllis C. Lee. 1999. “The impact of wildlife-related benefits
on the conservation attitudes of local people around the Selous Game Reserve,
Tanzania.” Environmental Conservation 26(3): 218–228.
Gore, Meredith L., ed. 2017. Conservation criminology. Wiley-Blackwell.
Gore, Meredith L., Jonah Ratsimbazafy, and Michelle L. Lute. 2013. “Rethinking
Corruption in Conservation Crime: Insights from Madagascar.” Conservation Let-
ters 6(6): 430–438.
Haack, Patrick, Michael D. Pfarrer, and Andreas Georg Scherer. 2014. “Legitimacy-
as-Feeling: How Affect Leads to Vertical Legitimacy Spillovers in Transnational
Governance.” Journal of Management Studies 51(4): 634–666.
Halpern, Benjamin S. 2014. “Making marine protected areas work.” Nature
506(7487): 167–168.
Hanson, Jonathan H., and David McNair. 2014. “Should conservationists continue
to dodge the issue of tax dodging?” Oryx 48(3): 313–314.
Isaksson, Ann-Sofie, and Andreas Kotsadam. 2018. “Chinese aid and local corrup-
tion.” Journal of Public Economics 159: 146–159.
Joppa, Lucas N., Scott R. Loarie, and Stuart L. Pimm. 2009. “On Population Growth
Near Protected Areas.” PLoS ONE 4(1): e4279.
Katzner, Todd E. 2005. “Corruption–a double-edged sword for conservation? A
response to Smith & Walpole.” Oryx 39(3): 260–262.
Kaufman, Herbert. 1960. The forest ranger: a study in administrative behavior. Balti-
more: John Hopkins University Press.
Kaufman, Herbert. 1977. Red tape, its origins, uses, and abuses. Washington, D.C.:
The Brookings Institution.
32
Knutsen, Carl Henrik, Andreas Kotsadam, Eivind Hammersmark Olsen, and Tore
Wig. 2017. “Mining and Local Corruption in Africa.” American Journal of Political
Science 61(2): 320–334.
Koyuncu, Cuneyt, and Rasim Yilmaz. 2009. “The Impact of Corruption on Defor-
estation: A Cross-Country Evidence.” The Journal of Developing Areas 42(2): 213–
222.
Kumlin, Staffan, and Bo Rothstein. 2005. “Making and Breaking Social Capital: The
Impact of Welfare-State Institutions.” Comparative Political Studies 38(4): 339–365.
Lambsdorff, Johann Graf. 2002. “Making corrupt deals: contracting in the shadow
of the law.” Journal of Economic Behavior & Organization 48: 221–241.
Levi, Margaret. 1988. Of rule and revenue. Number 13 in “California series on social
choice and political economy” Berkeley: University of California Press.
Macura, Biljana, Francisco Zorondo-Rodrı́guez, Mar Grau-Satorras, Kathryn
Demps, Marie Laval, Claude A. Garcia, and Victoria Reyes-Garcı́a. 2011. “Local
Community Attitudes toward Forests Outside Protected Areas in India. Impact
of Legal Awareness, Trust, and Participation.” Ecology and Society 16(3): art10.
Mangonnet, Jorge, Jacob Kopas, and Johannes Urpelainen. 2022. “Playing Politics
with Environmental Protection: The Political Economy of Designating Protected
Areas.” The Journal of Politics 84(3): 1453–1468.
Martinangeli, Andrea F.M., Marina Povitkina, Sverker Jagers, and Bo Rothstein.
2024. “Institutional Quality Causes Generalized Trust: Experimental Evidence
on Trusting under the Shadow of Doubt.” American Journal of Political Science
68(3): 972–987.
McKinnon, Madeleine C., Samantha H. Cheng, Samuel Dupre, Janet Edmond,
Ruth Garside, Louise Glew, Margaret B. Holland, Eliot Levine, Yuta J. Masuda,
Daniel C. Miller, Isabella Oliveira, Justine Revenaz, Dilys Roe, Sierra Shamer,
David Wilkie, Supin Wongbusarakum, and Emily Woodhouse. 2016. “What are
the effects of nature conservation on human well-being? A systematic map of
empirical evidence from developing countries.” Environmental Evidence 5(1): 8.
Michaelsen, Patrik, Aksel Sundström, and Sverker Jagers. 2024. “Mass Support to
Conserve 30 % of Earth by 2030: Experimental Evidence from Five Continents.”
Quality of Government Working Paper Series 2024:14 .
Muñoz Brenes, Carlos L., Kelly W. Jones, Peter Schlesinger, Juan Robalino, and Lee
Vierling. 2018. “The impact of protected area governance and management ca-
pacity on ecosystem function in Central America.” PLOS ONE 13(10): e0205964.
33
Naidoo, R., D. Gerkey, D. Hole, A. Pfaff, A. M. Ellis, C. D. Golden, D. Herrera,
K. Johnson, M. Mulligan, T. H. Ricketts, and B. Fisher. 2019. “Evaluating the
impacts of protected areas on human well-being across the developing world.”
Science Advances 5(4): eaav3006.
Nolte, Christoph, Arun Agrawal, Kirsten M. Silvius, and Britaldo S. Soares-Filho.
2013. “Governance regime and location influence avoided deforestation success
of protected areas in the Brazilian Amazon.” Proceedings of the National Academy
of Sciences 110(13): 4956–4961.
Ostrom, Elinor. 1990. Governing the Commons: The Evolution of Institutions for Col-
lective Action. Political economy of institutions and decisions Cambridge: Cam-
bridge University Press.
Packer, Craig, and Stephen Polasky. 2018. “Reconciling corruption with conserva-
tion triage: Should investments shift from the last best places?” PLOS Biology
16(8): e2005620.
Persson, Anna, Bo Rothstein, and Jan Teorell. 2013. “Why Anticorruption Re-
forms Fail—Systemic Corruption as a Collective Action Problem.” Governance
26(3): 449–471.
Petchkaew, Kannikar. 2023. https://news.mongabay.com/2023/01/
corruption-scandal-in-thai-parks-agency-has-far-reaching-impacts-activists-say/
“Corruption scandal in Thai parks agency has far-reaching impacts, activists
say.”.
Rauch, James E, and Peter B Evans. 2000. “Bureaucratic structure and bureaucratic
performance in less developed countries.” Journal of Public Economics 75(1): 49–71.
Robbins, Paul. 2000. “The rotten institution: corruption in natural resource man-
agement.” Political Geography 19(4): 423–443.
Rose-Ackerman, Susan. 1999. Corruption and Government: Causes, Consequences, and
Reform. New York: Cambridge University Press.
Rothstein, Bo. 2011. The Quality of Government: Corruption, Social Trust, and Inequality
in International Perspective. Chicago: University of Chicago Press.
Smith, J., K. Obidzinski, Subarudi Subarudi, and I. Suramenggala. 2003. “Illegal
logging, collusive corruption and fragmented governments in Kalimantan, In-
donesia.” International Forestry Review 5(3): 293–302.
Smith, Robert J., and Matthew J. Walpole. 2005. “Should conservationists pay more
attention to corruption?” Oryx 39(3): 251–256.
34
Stamoulis, Kostantinos A., and Alan M. Friedlander. 2013. “A seascape approach to
investigating fish spillover across a marine protected area boundary in Hawai‘i.”
Fisheries Research 144: 2–14.
Sundström, Aksel. 2013. “Corruption in the commons.” International Journal of the
Commons 7(2): 454–472.
Sundström, Aksel. 2016. “Understanding illegality and corruption in forest gover-
nance.” Journal of Environmental Management 181: 779–790.
Truelove, Heather Barnes, Amanda R. Carrico, Elke U. Weber, Kaitlin Toner Raimi,
and Michael P. Vandenbergh. 2014. “Positive and negative spillover of pro-
environmental behavior: An integrative review and theoretical framework.”
Global Environmental Change 29: 127–138.
UNEP-WCMC, Heather C Bingham, Marine Deguignet, Edward Lewis, Jes-
sica Stewart, Diego Juffe-Bignoli, Brian MacSharry, Amy Milam, and Naomi
Kingston. 2019. http://wcmc.io/WDPA_Manual “User Manual for the World
Database on Protected Areas and world database on other effective area- based
conservation measures: 1.6.”.
UNODC. 2024. The Global Programme Against Corruption: UN Anti-Corruption
Toolkit 2nd edition. Technical report United Nations Office on Drugs and Crime
Vienna: . https://www.unodc.org/documents/corruption/Toolkit_
ed2.pdf
West, Paige, James Igoe, and Dan Brockington. 2006. “Parks and Peoples: The
Social Impact of Protected Areas.” Annual Review of Anthropology 35(1): 251–277.
35
Supplementary Material
Table of Contents
A Additional Data . . . . . . . . . . . . . . . . . . . . . . . . A1
A.1 Afrobarometer . . . . . . . . . . . . . . . . . . . . . . . . A1
A.2 OpenStreetMap data . . . . . . . . . . . . . . . . . . . . . A2
A.3 Alternative corruption measures . . . . . . . . . . . . . . . . A4
B Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . A5
B.1 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . A5
B.2 Application . . . . . . . . . . . . . . . . . . . . . . . . . A5
C References (Appendix) . . . . . . . . . . . . . . . . . . . . . A8
A Additional Data
A.1 Afrobarometer
Table A1: Afrobarometer sample by country and round
Country R2 R3 R4 R5 R6 R7
Algeria 0 0 0 1204 1200 0
Benin 0 1198 1200 1200 1200 1200
Botswana 1200 1200 1200 1200 1200 1198
Burkina Faso 0 0 1200 1200 1200 1200
Burundi 0 0 0 1200 1200 0
Cameroon 0 0 0 1200 1182 1202
Cape Verde 1268 1256 1264 1208 1200 1200
Côte d’Ivoire 0 0 0 1200 1199 1200
Egypt 0 0 0 1190 1198 0
Eswatini 0 0 0 1200 1200 1200
Gabon 0 0 0 0 1198 1199
Gambia 0 0 0 0 0 1200
Ghana 1200 1197 1200 2400 2400 2400
Guinea 0 0 0 1200 1200 1194
Kenya 2398 1278 1104 2399 2397 1599
Lesotho 1200 1161 1200 1198 1200 1200
Liberia 0 0 1200 1199 1199 1200
Madagascar 0 1350 1350 1200 1200 1200
Malawi 1200 1200 1200 2408 2400 1200
Mali 1283 1244 1232 1200 1200 1200
Mauritius 0 0 0 1201 1200 1200
Morocco 0 0 0 1196 1200 1200
Mozambique 1400 1198 1200 2400 2400 2304
Namibia 1199 1200 1200 1200 1200 1200
Niger 0 0 0 1199 1200 1200
Nigeria 2428 2363 2324 2400 2400 1600
Senegal 1200 1200 1200 1200 1200 1200
Sierra Leone 0 0 0 1190 1191 1200
South Africa 2400 2400 2400 2399 2390 1840
Sudan 0 0 0 1199 1200 888
São Tomé & Prı́ncipe 0 0 0 0 1196 1200
Tanzania 1223 1304 1208 2400 2386 2400
Togo 0 0 0 1201 1200 1200
Tunisia 0 0 0 1200 1200 1199
Uganda 2400 2400 2431 2400 2400 1200
Zambia 1198 1200 1200 1200 1199 1200
Zimbabwe 1104 1048 1200 2400 2400 1200
A1
A.2 OpenStreetMap data
This section details the validation process of the OpenStreetMap (OSM) tourism
tag data. Our approach is to rely on the data of Balmford et al. (2015) who estimate
the number of visitors to PAs around the world 1998-2007. However, their analysis
is based on around 500 PAs globally and 95 in Africa (compared to over 5500 in our
African sample). We therefore look to OpenStreetMap: geo-spatial data based on
user self-reporting of points of interest. In this instance, we consider all points that
include a tourism ’tag’. Below, table A2 demonstrates the 10 most frequent tourism
tags in the OSM data in Africa.
Table A2: OSM tourism tag frequency: top ten
OSM tag Frequency
hotel 21768
guest house 11293
viewpoint 8589
attraction 8426
information 8041
apartment 6398
camp site 5560
artwork 4057
hostel 3045
picnic site 1617
Our intuition is that protected areas with the most tourism tags inside – or
in close proximity – on average attract more visitors. To validate this, we create
buffers around PAs of varying distances (no buffer, 10km, 25km, 50km) and count
the numer of OSM tourist tags within each PA + buffer zone. We then correlate this
PA-specific figure to the Balmford et al. (2015) data for PAs where tourism numbers
are available. Plots illustrating the correlation between tourism numbers and OSM
tourism tags for each buffer are presented below. We use the 10km buffer as our
preferred measure as it has the highest correlation with the number of PA visitors.
This buffer is the basis of the tourism analysis presented in Figure 4 in the main
paper.
A2
Figure A1: Correlations between PA tourist visitor figures and OSM tourism tags
in Africa across different buffers
No buffer 10km
900 R = 0.59, p < 0.001 R = 0.85, p < 0.001
600
300
0
25km 50km
900 R = 0.77, p < 0.001 R = 0.66, p < 0.001
600
300
0
0 500000 1000000 15000000 500000 1000000 1500000
Mean visitors
A3
OSM tourism tags
A.3 Alternative corruption measures
Table A2: Replication of main results using alternative corruption measures
Perceptions Bribes
Police Local council Nat. gov Police Doc/permit Service
Protected Area 0.031 0.039* 0.013 0.011 0.002 0.001
(0.020) (0.020) (0.020) (0.009) (0.007) (0.008)
<10km 0.050*** 0.053*** 0.040*** 0.008* 0.004 -0.001
(0.010) (0.009) (0.009) (0.003) (0.004) (0.003)
11-50km 0.047*** 0.029*** 0.031*** 0.005+ 0.000 -0.005
(0.009) (0.008) (0.009) (0.003) (0.003) (0.003)
Num.Obs. 201148 187627 193503 220791 220818 220517
Country-year FE Yes Yes Yes Yes Yes Yes
R2 0.117 0.095 0.087 0.085 0.081 0.065
R2 Within 0.008 0.007 0.006 0.014 0.013 0.006
Note: The table reports OLS estimates. Robust standard errors clustered by enumeration area in
parentheses. Unit of observation is the individual survey respondent. All models include inverse
probability of treatment weights (IPTW). Significance levels: ∗∗∗p < .001; ∗∗p < .01; ∗p < .05; +p <
.1.
A4
B Matching
B.1 Approach
Given that the placement of protected areas is plausibly subject to similar human
and societal factors as perceptions and experiences of corruption, we attempt to
strengthen the causal inference of our approach by improving the balance between
treatment and control group with regard to the likelihood of treatment. To do this,
we implement propensity score estimation to generate the inverse probability of
treatment weights (IPTW). This process effectively estimates the likelihood of a
given enumeration area-year being treated (protected) based on a series of pre-
defined covariates that could potentially influence both treatment assignment (PA
establishment) and the outcome (local corruption). The propensity scores produced
by this model are then used to calculate a weight that is assigned to each enumera-
tion area-year for the main analysis. Effectively, more weight is attributed to obser-
vations that are less likely to be treated according to the PS model, and vice versa.
This approach reweights the population to create a pseudo-population where the
treatment assignment is independent of covariates. To guard against influential ob-
servations (those close to a PS of 0 or 100), we follow Cole and Hernán (2008) and
calculate a stabilized IPTW as follows:
P (T = t)
Stabilized IPTW = (A1)
P (T = t | X)
where P (T = t | X) is the predicted probabilities of a denominator model in-
cluding covariates and P (T = t) is the is the predicted probabilities of a numerator
model (intercept only).
B.2 Application
As treatment is at the enumeration area-year level, we select covariated for the
propensity score generation model based on their potential to influence both the
spatial allocation of treatment and corruption perceptions and experiences. The
denominator model also includes country and year fixed effects. The covariates
selected are as follows:
• Biodiversity, as measured by the number of recorded species at a geographic
location by the IUCN.
• Distance from the enumeration area to the nearest mineral resource (gems ,
diamonds, gold, and petroleum). Governments may avoid placing PAs in
areas rich with mineral resources in order to facilitate extractive industries,
which may also be more liable to corruption. These data come from the
A5
Gemstone Dataset, Diamond Dataset, Goldata, and Petrodata, respectively
(Balestri and Maggioni 2014; Lujala, Ketil Rod, and Thieme 2007).
• Whether the enumeration area is in an urban or rural area, as large PAs are
unlikely to be in urban areas which may also be more corrupt. This variable
is recorded by Afrobarometer enumerators.
• Whether the area is is the ethnic homeland of politically included/excluded
area. While ethnicity is closely related to patronage networks in some com-
munities in Africa, research also demonstrates an ethnic bias in the allocation
of PAs (Dawson et al. 2024). This data is taken from the Ethnic Power Rela-
tions (EPR) geo-spatial dataset (Vogt et al. 2015).
• Human development. The development of a region is likely to relate closely
to both the state’s willingness and ability to establish protected areas and lev-
els of corruption in the ares. This is measured by the Subnational Human De-
velopment Index (SHDI) at the admin 2 (district) level (Smits and Permanyer
2019).
A6
Table A3. Binomial regression analysis used to generate propensity scores for
matching
(1) (2)
Biodiversity 0.002***
(0.000)
Distance to resource 0.002***
(0.000)
Urban -0.536***
(0.086)
Ethnic inclusion -0.452***
(0.097)
SHDI -7.411***
(0.821)
Intercept -3.810*** -3.229***
(0.828) (0.032)
Country FE Yes No
Year FE Yes No
Num.Obs. 26323 26323
AIC 7494.0 8523.8
Log.Lik. -3691.003 -4260.894
Note: The table reports results of a binomial regression analysis. The unit of analysis
is enumeration area-years. The dependent variable is whether an enumeration area is
protected or not (0-1). Robust standard errors clustered by enumeration area in paren-
theses. Significance levels: ∗∗∗p < .001; ∗∗p < .01; ∗p < .05; +p < .1.
A7
C References (Appendix)
Balestri, Sara, and Mario A. Maggioni. 2014. “Blood Diamonds, Dirty Gold and
Spatial Spill-overs Measuring Conflict Dynamics in West Africa.” Peace Eco-
nomics, Peace Science and Public Policy 20(4): 551–564.
Balmford, Andrew, Jonathan M. H. Green, Michael Anderson, James Beresford,
Charles Huang, Robin Naidoo, Matt Walpole, and Andrea Manica. 2015. “Walk
on the Wild Side: Estimating the Global Magnitude of Visits to Protected Areas.”
PLOS Biology 13(2): e1002074.
Cole, Stephen R., and Miguel A. Hernán. 2008. “Constructing Inverse Probabil-
ity Weights for Marginal Structural Models.” American Journal of Epidemiology
168(6): 656–664.
Dawson, Stephen, Felix Haass, Carl Müller-Crepon, and Aksel Sundström. 2024.
“The Ethnic Politics of Nature Protection: Ethnic Favoritism and Protected Areas
in Africa.” Quality of Government Working Paper Series 2024:4 .
Lujala, Päivi, Jan Ketil Rod, and Nadja Thieme. 2007. “Fighting over Oil: Introduc-
ing a New Dataset.” Conflict Management and Peace Science 24(3): 239–256.
Smits, Jeroen, and Iñaki Permanyer. 2019. “The Subnational Human Development
Database.” Scientific Data 6(1): 190038.
Vogt, Manuel, Nils-Christian Bormann, Seraina Rüegger, Lars-Erik Cederman,
Philipp Hunziker, and Luc Girardin. 2015. “Integrating Data on Ethnicity, Ge-
ography, and Conflict: The Ethnic Power Relations Data Set Family.” Journal of
Conflict Resolution 59(7): 1327–1342.
A8