Content uploaded by Xavier Basurto
Author content
All content in this area was uploaded by Xavier Basurto on Feb 20, 2018
Content may be subject to copyright.
30 MARCH 2017 | VOL 543 | NATURE | 665
ARTICLE doi:10.1038/nature21708
Capacity shortfalls hinder the performance
of marine protected areas globally
David A. Gill1,2†, Michael B. Mascia3, Gabby N. Ahmadia4, Louise Glew4, Sarah E. Lester5, Megan Barnes6,7, Ian Craigie8,
Emily S. Darling9, Christopher M. Free10, Jonas Geldmann11,12, Susie Holst13, Olaf P. Jensen10, Alan T. White14, Xavier Basurto15,
Lauren Coad16,17, Ruth D. Gates18, Greg Guannel19, Peter J. Mumby20, Hannah Thomas21, Sarah Whitmee22,
Stephen Woodley23 & Helen E. Fox4,24
Awareness of human impacts upon global marine biodiversity has
spurred the largest expansion in the number and coverage of marine
protected areas (MPAs) in history
1,2
. As part of the 2011 Convention
on Biological Diversity (CBD) Aichi Targets, 193 countries agreed to
“effectively and equitably” manage 10% of coastal and marine areas
within marine protected areas and “other effective area-based conserva
-
tion measures” by 2020 (ref. 3). A 10% conservation target for MPAs has
also been included within Goal 14 of the United Nations Sustainable
Development Goals (SDGs)4. Yet despite recent advances towards these
coverage targets (currently 4.1% (ref. 2)), the efficacy and equity of
many MPAs remain uncertain
2
; evidence suggests that MPAs often fail
to deliver positive social and ecological outcomes5–7.
It is assumed that MPAs that are effectively regulated and actively
managed through equitable and inclusive decision-making approaches
are more likely to meet ecological and social goals than those that are
merely legislated on paper (‘paper parks’) and those with exclusionary
decision-making
8–10
. However, research linking the efficacy and equity
of MPA management processes to conservation outcomes lies mostly
in theory and select local-scale case studies
11
. This is largely due to a
lack of a globally representative dataset on MPA management12 and
lack of counterfactuals to infer conservation outcomes in the absence
of MPAs13,14.
We constructed a global database of management and ecological data
from 433 and 218 MPAs, respectively, to document and examine link-
ages between MPA management processes and conservation outcomes.
Our dataset included MPAs from every tropical and temperate ocean
basin, ranging in size from 0.006 to 989,836 km2, and spans diverse
social, political and biophysical contexts. First, to assess the efficacy
and equity of MPA management processes, we drew on empirically
supported governance and management theories10,15–17 (Supplementary
Table 1 and Extended Data Fig. 1) to identify key management pro-
cess indicators from 433 MPAs. We extracted data on these indicators
from three widely applied survey instruments (Supplementary Table 2)
that provided qualitative, Likert-scaled scores on questions posed
to MPA stakeholders concerning MPA management activities and
capacities18. From these, we defined binary thresholds for effective
management based on the scoring criteria and alignment with social
theory (Supplementary Tables 1 and 3). Second, to measure ecological
impacts (n = 218 MPAs), we compiled MPA outcome data extracted
from published studies5 (n = 40 MPAs) and transect- or site-level obser-
vations from unpublished regional and global datasets (Supplementary
Table 2 and Extended Data Fig. 2; n = 178 MPAs). For the unpublished
ecological data, we calculated logged response ratios (lnRR): the natu-
ral logarithm of the ratio of mean fish biomass per unit area inside an
MPA site relative to mean fish biomass in a statistically matched control
site (that is, pre-establishment and/or outside MPA; Methods). Finally,
we investigated the relationship between management processes and
ecological impacts in 62 MPAs where both management and ecological
data were available. We used random forest and linear mixed-effects
models to identify important management predictors of ecological
Marine protected areas (MPAs) are increasingly being used globally to conserve marine resources. However, whether many
MPAs are being effectively and equitably managed, and how MPA management influences substantive outcomes remain
unknown. We developed a global database of management and fish population data (433 and 218 MPAs, respectively) to
assess: MPA management processes; the effects of MPAs on fish populations; and relationships between management
processes and ecological effects. Here we report that many MPAs failed to meet thresholds for effective and equitable
management processes, with widespread shortfalls in staff and financial resources. Although 71% of MPAs positively
influenced fish populations, these conservation impacts were highly variable. Staff and budget capacity were the strongest
predictors of conservation impact: MPAs with adequate staff capacity had ecological effects 2.9 times greater than MPAs
with inadequate capacity. Thus, continued global expansion of MPAs without adequate investment in human and financial
capacity is likely to lead to sub-optimal conservation outcomes.
1National Socio-Environmental Synthesis Center (SESYNC), Annapolis, Maryland 21401, USA. 2Luc Hoffmann Institute, World Wildlife Fund International, 1196 Gland, Switzerland. 3Moore Center
for Science, Conservation International, Arlington, Virginia 22202, USA. 4World Wildlife Fund US, Washington DC 20037, USA. 5Department of Geography, Florida State University, Florida 32306,
USA. 6Centre for Biodiversity and Conservation Science, University of Queensland, St Lucia Campus, Brisbane, Queensland 4072, Australia. 7Department of Natural Resources and Environmental
Management, University of Hawaii, Honolulu HI 96822, USA. 8ARC Centre of Excellence for Coral Reef Studies, James Cook University, Townsville, Queensland 4811, Australia. 9Wildlife
Conservation Society, Bronx, New York 10460, USA. 10Department of Marine & Coastal Sciences, Rutgers University, New Brunswick, New Jersey 08901, USA. 11Conservation Science Group,
Department of Zoology, University of Cambridge, Downing Street, Cambridge CB2 3EJ, UK. 12Center for Macroecology, Evolution and Climate, Natural History Museum of Denmark, University
of Copenhagen, Universitetsparken 15, DK-2100 Copenhagen E, Denmark. 13NOAA Coral Reef Conservation Program, Silver Spring, Maryland 20910, USA. 14Indo-Pacific Division, The Nature
Conservancy, Honolulu, Hawaii 96817, USA. 15Nicholas School of the Environment, Duke University, Beaufort, North Carolina 28516, USA. 16Environmental Change Institute, University of Oxford,
South Parks Road, Oxford OX1 3QY, UK. 17Centre for International Forestry Research, Bogor (Barat) 16115, Indonesia. 18Hawaii Institute of Marine Biology, University of Hawaii at Manoa, Hawaii
96744, USA. 19The Natural Capital Project, Stanford University, 371 Serra Mall, Stanford, California 94305-5020, USA. 20Marine Spatial Ecology Lab, School of Biological Sciences and ARC Centre
of Excellence for Coral Reef Studies, The University of Queensland, St Lucia Campus, Brisbane, Queensland 4072, Australia. 21UNEP – World Conservation Monitoring Centre, Cambridge CB3 0DL,
UK. 22CBER – University College London, London WC1E 6BT, UK. 23WCPA-SSC Joint Task Force on Biodiversity and Protected Areas, International Union for the Conservation of Nature (IUCN),
Quebec J9B 1T3, Canada. 24National Geographic Society, Washington DC 20036, USA. †Present addresses: Moore Center for Science, Conservation International, Arlington, Virginia 22202, USA;
George Mason University, Fairfax, Virginia 22030, USA.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
666 | NATURE | VOL 543 | 30 MARCH 2017
ARTICLE
RESEARCH
outcomes, while accounting for other factors known to impact fish
responses to protection (for example, MPA age and size
7,19,20
; Methods
and Supplementary Information).
MPA management processes
MPA management processes varied widely, with many of the 433 MPAs
failing to meet thresholds for effective management (Fig. 1a). While
the majority of MPAs were legally gazetted (79%) and had appropri-
ate regulations regarding resource use (69%), very few MPAs (13%)
reportedly used results from scientific monitoring (biological, social
or management) to inform management. Many also reported limited
capacity, with 65% of MPAs reporting that their budget was inade-
quate for basic management needs and 91% stating that staff capacity
(sufficient (on-site) staff capacity/numbers) was inadequate or below
optimum.
Most MPAs were state-managed (80%), with the remaining either
co-managed or managed by non-state actors (for example, NGOs,
local communities; Fig. 1a). Inclusive decision-making arrangements
were reported in 51% of MPAs and were more common in shared/non-
state-managed MPAs than those managed solely by state agencies
(P < 0.001; Extended Data Fig. 3).
Management processes were largely consistent across geographic
contexts (Fig. 1b). In Oceania, however, devolved and inclusive manage-
ment was more common and relatively few MPAs were legally gazetted.
Where data were available for all indicators (excluding non-state
management; n = 277 MPAs), only 21% of MPAs met more than half of
the nine thresholds, and only five MPAs (2%) met all nine thresholds
(Supplementary Table 7). Twenty-two MPAs (8%) failed to meet any of
the threshold levels for effective and equitable management.
MPA ecological outcomes
MPAs on average had positive, but variable, impacts on fish popu-
lations. We observed positive responses to protection in 71% of the
218 MPAs with fish biomass data. On average, fish biomass was 1.6
times higher in MPAs than in matched non-MPA areas (average
lnRR = 0.47 + 0.96 s.d.). Positive responses were observed across almost
all geographies and habitats (Fig. 2), consistent with other analyses
5,20
.
Response ratios varied marginally by latitudinal zone (F = 2.963,
P = 0.087; Fig. 2b) and significantly among habitats (F = 6.403,
P < 0.001; Fig. 2c) and continental regions (F = 5.284, P < 0.001;
Fig. 2d). MPAs or MPA zones where all fishing was prohibited (no-take)
had higher response ratios than MPAs/zones where fishing was per-
mitted (multi-use) by almost twofold (t = 2.24, P = 0.026; Extended
Data Fig. 4). Nonetheless, on average, we observed positive response
ratios in both multi-use MPAs and MPA zones that prohibited fishing.
Responses in prohibited fishing areas were lower than in some previous
studies (for example, 82% increase in fish biomass in our study versus
387% reported elsewhere
5
), probably owing, in part, to the statistical
matching approach, which reduced the observable biases arising from
non-random MPA placement.
Linking MPA management and outcomes
We next explored the relationships between management processes
and ecological impacts in MPAs for which we had both management
and ecological data (62 MPAs in 24 countries), while accounting for
other significant MPA and contextual attributes (for example, MPA
age, size, ocean conditions; Supplementary Table 4). In these MPAs,
adequate staff capacity was the most important factor in explain-
ing fish responses to MPA protection (Fig. 3a). Budget capacity
was the second most important management variable and had
similar performance in other analyses (Supplementary Table 9);
however, budget data were only available in 43 MPAs. Clearly defined
boundaries, MPA age and size, location (ecoregion, country), mean
chlorophyll concentration, and mean shore distance were also
identified as important by the conditional inference forest models
(Fig. 3a).
Shared/non-state management (n = 419)
Inclusive decision-making (n = 388)
Adequate staff capacity/presence (n = 417)
Monitoring informs management activities (n = 395)
Clearly dened boundaries (n = 419)
Acceptable budget capacity (n = 375)
Acceptable enforcement capacity (n = 411)
Implementing management plan (n = 420)
Appropriate MPA regulations in place (n = 373)
Legally gazetted (n = 371)
100 50 500 100%
Indicator
Threshold
a
Africa
(n = 58)
Americas
(n = 241)
Asia
(n = 54)
Europe
(n = 55)
Oceania
(n = 25)
50% 0 50% 50% 0 50% 50% 0 50% 50% 0 50% 50% 0 50%
Shared/non-state management
Inclusive decision-making
Adequate staff capacity/presence
Monitoring informs management activities
Clearly dened boundaries
Acceptable budget capacity
Acceptable enforcement capacity
Implementing management plan
Appropriate MPA regulations in place
Legally gazetted
Per cent of MPAs
Indicator
Continent
b
Efcacy
Equity
Below threshold Above threshold
Figure 1 | Per cent of MPAs exceeding or falling below threshold values
for indicators of effective and equitable management processes.
a, b, Values shown for all MPAs (n = 433 MPAs) (a) and by continent (b).
Dark blue bars (right) indicate the proportion of MPAs with scores at or
above the threshold value, light blue bars (left) indicate the proportion
below the threshold. Details on indicators, scores and threshold values in
Supplementary Tables 1 and 3.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
30 MARCH 2017 | VOL 543 | NATURE | 667
ARTICLE RESEARCH
Our results demonstrate that effective biodiversity conservation is
not simply a function of environmental (for example, ocean conditions)
or MPA features (for example, MPA size, age, fishing regulations), but is
also heavily dependent on available capacity (Fig. 3). Staff capacity was by
far the most important explanatory variable in our study, accounting for
approximately 19% of the variation in ecological outcomes (n = 62 MP As;
t = 3.786; P < 0.001). Qualitative examination of the MPA man-
agement data indicated that additional staff resources were needed
to support monitoring, enforcement, administration, community
engagement and sustainable tourism activities (amongst other tasks).
Though specific capacity needs varied among MPAs, biomass response
ratios were on average 2.9 times greater in MPAs reporting adequate
Caribbean Southeast Australia
Biomass (lnRR)
–3.75
+3.75
0.00
Temperate
(24, 45)
Tropical
(102, 83)
0.0 0.5
Biomass (lnRR)
Latitudal zone
MPA zone
Multi-use Fishing prohibited
b
a
Coral
(113, 88)
Mangrove
(3, 0)
Other
(14, 7)
Rocky reef
(12, 40)
Seagrass
(5, 2)
–2 –1 01
212
Biomass (lnRR)
Habitat
c
Africa
(4, 10)
Americas
(67, 47)
Asia
(7, 14)
Europe
(5, 11)
Oceania
(43, 46)
0
Biomass (lnRR)
Continent
d
Figure 2 | MPA effects on fish populations (biomass). a, Global variation
in mean fish biomass response ratios (natural log scale; lnRR) for 218
MPAs. Positive response ratios (blue) indicate MPAs with greater biomass
inside MPA relative to matched non-MPA areas. Negative values are in red.
Base map sourced from ref. 29. b–d, Mean response ratios (dot) and 95%
confidence interval (error bars) for multi-use areas (light blue) and areas
where fishing is prohibited (dark blue) in 254 zones in 218 MPAs shown
by latitudinal zone (b), habitat (c) and continental region (d). Values in
parentheses on the y axes indicate the number of MPAs/zones that are
multi-use and those where fishing is prohibited, respectively.
Sea surface temp.
Market distance
Wave exposure
Proportion no shing
Human population
Depth
Country
MPA age
MPA size
Shore distance
Ecoregion
Chlorophyll
Monitoring
Enforcement
Inclusive decision-making
Management plan
Legally gazetted
Non-state management
MPA regulations
Clear boundaries
Budget capacity
Staff capacity
0.00 0.01 0.02
Relative importance
(n = 62 MPAs)
Predictor
Variable type
Management Other
a
Adequate
(8, 12)
Inadequate/
below optimal
(23, 17)
None
(6, 5)
0.00.5 1.0
Biomass (InRR)
(n = 58 MPAs)
Staff capacity
MPA zone
Multi-useFishing prohibited
b
Figure 3 | Relationship between MPA management processes and
ecological impact. a, Random forest variable importance measures for
management (dark blue bars) and other (non-management; light grey
bars) variables as they relate to ecological effects in 62 MPAs. Importance
measures exceeding the red dashed line are considered non-random.
b, Mean fish biomass response ratios (lnRR; dot) and 95% confidence
interval (error bars) for multi-use areas (light blue) and areas where
fishing is prohibited (dark blue) by reported staff capacity (excluding
MPAs with intermediate scores (n = 4)). Proportion no-fishing represents
the proportion of survey sites for an MPA sampled from within a
prohibited-fishing (no-take) zone (0, all multi-use; 1, all prohibited
fishing). Values in parentheses on the y axis indicate the number of MPAs/
zones that are multi-use and those where fishing is prohibited, respectively.
Additional bivariate plots in Extended Data Fig. 5.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
668 | NATURE | VOL 543 | 30 MARCH 2017
ARTICLE
RESEARCH
staff capacity than those MPAs reporting inadequate or no capacity
(Fig. 3b). Where data were available (n = 43 MPAs), we observed a
significant relationship between budget capacity and ecological impacts
(Supplementary Table 9), even after we removed potential outlying
data (Extended Data Fig. 5a; n = 42 MPAs; t = 2.55; P = 0.019). Budget
capacity was also significantly correlated with staff capacity (Spearman’s
ρ = 0.35, P < 0.001), and both capacity variables were positively cor-
related with many of the other management variables (Extended
Data Fig. 6). Thus, the effectiveness of many other key manage-
ment processes may be limited by available human and financial
capacity.
In addition to staff capacity, clearly defined boundaries and appropriate
regulations were significantly correlated with ecological outcomes
(Extended Data Fig. 7). However, the predictive strength of these two
variables was sensitive to the modelling approach. Other management
variables theorized to foster sustainable outcomes in common pool
resources (for example, inclusive decision making, monitoring of the
resource and users15) were not significantly related to ecological perfor-
mance (Fig. 3a and Supplementary Table 9), a finding consistent with
some previous studies
21,22
. Possible explanations for this are that these
described processes have stronger, more direct impacts on resource
users than on resource conditions22, or that the indicators used in
management assessments may imperfectly measure the governance
and management processes from common pool resource theory
23
(for
example, Ostrom’s design principles15).
In agreement with other studies, we found that non-management
factors such as MPA age and size also shape MPA ecological impacts
(Fig. 3a)7,19,20. Although we observed a significant difference in
ecological impacts between prohibited fishing and multi-use zones
(Extended Data Fig. 4), fishing regulations (defined as the propor-
tion of survey sites for an MPA sampled from within a prohibited-
fishing (no-take) zone) were not significant in our sample of 62 MPAs
while controlling for (or interacting with) other factors (Fig. 3a.
and Supplementary Table 9). Other variables, such as proximity to
shore and chlorophyll concentration (a potential proxy for ocean
productivity
24
, but also for reduced coastal water quality at extremely
high levels25), were negatively correlated with fish biomass. This
suggests that land-based stressors may be influencing effects inside
nearshore MPAs, as noted in other work25,26. Differences in variable
constructs among studies may partially explain observed differences
in our results from previous work. For example, a recent study that
found ‘enforcement’ to be a significant factor
7
measured the enforce-
ment construct as a combination of compliance, community support
and enforcement activities, whereas our study focused on manage-
ment inputs into enforcement activities.
Assessing MPA efficacy and equity
We drew on social theory (Supplementary Table 1) to identify aspects
of MPA management hypothesized to be important for ecological out-
comes, independent of many of the MPA and site features also known to
affect MPA performance (for example, MPA age and size7,19). Our theory-
based analytic framework (Extended Data Fig. 1 and Supplementary
Table 1) provides a robust, replicable approach to measuring the proce-
dural and substantive efficacy and equity of protected areas. In particular,
the integrated use of impact evaluation methodologies and indicators
derived from widely used MPA monitoring tools permits us to make
novel, evidence-based inferences of conservation effects at a global
scale
27
. Despite uneven geographic distribution and limited data on some
indicators, this study represents one of the most comprehensive assess-
ments of MPA management and ecological outcomes to date. While the
ecological data centre heavily on areas in the North Atlantic, US Pacific,
and Australia, the available management data are more dominant in
other geographies (for example, Africa, Europe, southeast Asia), particu-
larly in the developing world. These spatial incongruities limit the overlap
between our ecological and management datasets (n = 62 MPAs), but
collectively provide a broad view on global MPA performance.
Given data availability, our research focused on the efficacy and
equity of MPA management processes and, as an indicator of substan-
tive efficacy, the ecological impact of MPAs on fish populations. We
lacked sufficient data on other taxa to assess other ecological indicators
of substantive efficacy. We were also unable to measure the substantive
social impact of MPAs, particularly substantive equity; the spatial and
temporal resolutions of relevant data were too coarse or geographi-
cally limited to assess these impacts globally. Our research highlights a
need for contemporaneous social, ecological and management data in
order to fill these remaining knowledge gaps and explore synergies and
trade-offs among the procedural and substantive outcomes of conser-
vation. Also, to guide more effective and holistic conservation policy,
future research should examine interactions between MPAs and other
management measures (for example, fisheries management), as well as
site-specific MPA capacity needs.
Achieving global conservation targets
As we approach the CBD and SDG milestone year of 2020, the global
conservation community and many governments will continue to
invest heavily in MPA expansion1. Although many MPAs with low
management capacity in our sample had positive ecological impacts,
in general the magnitude of ecological effects was strongly linked to
the available human and financial capacity for MPA management.
Given the widespread shortfall in staff capacity that we document
worldwide (Fig. 4), inadequate capacity appears to compromise the
Figure 4 | Reported level of MPA staff capacity. MPAs reporting adequate (dark blue), inadequate or below optimum (blue) and no (light blue) staff
capacity in their most recent management assessments where spatial data were available (n = 243 MPAs; excludes MPAs with intermediate scores
(n = 5)). Base map sourced from ref. 29.
Staff capacity
None
Inadequate/below optimum
Adequate
Caribbean
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
30 MARCH 2017 | VOL 543 | NATURE | 669
ARTICLE RESEARCH
ecological performance of many MPAs. Adequate capacity is likely
to be even more critical in the future, as increasing anthropogenic
pressures on marine resources necessitate more resilient marine
ecosystems and corresponding management regimes. For effective
and equitable management to be achieved, increased investment in
MPA capacity is necessary. Rapid MPA expansion without increased
investment has the potential to dilute already scarce resources across
a larger management area, weakening management and leaving many
marine habitats and species at risk. With such a high dependence on
under-resourced MPAs to meet current and future conservation and
sustainable development goals3,4, investment in MPA capacity devel-
opment could potentially result in high returns on investment for both
people and nature28.
Online Content Methods, along with any additional Extended Data display items and
Source Data, are available in the online version of the paper; references unique to
these sections appear only in the online paper.
Received 12 July 2016; accepted 15 February 2017.
Published online 22 March 2017.
1. Lubchenco, J. & Grorud-Colvert, K. Making waves: The science and politics of
ocean protection. Science 350, 382–383 (2015).
2. UNEP-WCMC & IUCN. Protected Planet Report 2016 (United Nations
Environment Programme (UNEP) World Conservation Monitoring Centre
(UNEP-WCMC) and International Union for Conservation of Nature (IUCN),
2016).
3. Secretariat of the CBD. Aichi Target 11. Decision X/2. Convention on Biological
Diversity (2011).
4. UN. United Nations Sustainable Development Goal 14: Conserve and
sustainably use the oceans, seas and marine resources. http://www.un.org/
sustainabledevelopment/oceans/ (2016).
5. Lester, S. E. et al. Biological eects within no-take marine reserves: a global
synthesis. Mar. Ecol. Prog. Ser. 384, 33–46 (2009).
6. Mascia, M. B., Claus, C. A. & Naidoo, R. Impacts of marine protected areas on
shing communities. Conserv. Biol. 24, 1424–1429 (2010).
7. Edgar, G. J. et al. Global conservation outcomes depend on marine protected
areas with ve key features. Nature 506, 216–220 (2014).
8. Pollnac, R. B., Crawford, B. R. & Gorospe, M. L. G. Discovering factors that
inuence the success of community-based marine protected areas in the
Visayas, Philippines. Ocean Coast. Manage. 44, 683–710 (2001).
9. Basurto, X., Blanco, E., Nenadovic, M. & Vollan, B. Integrating simultaneous
prosocial and antisocial behavior into theories of collective action. Sci. Adv. 2,
e1501220 (2016).
10. Mascia, M. B. in Marine Reserves: A Guide to Science, Design, and Use
(eds Sobel, J. & Dahlgren, C.) 164–186 (Island Press, 2004).
11. Pollnac, R. et al. Marine reserves as linked social-ecological systems. Proc. Natl
Acad. Sci. USA 107, 18262–18265 (2010).
12. Fox, H. E. et al. How Are Our MPAs Doing? Challenges in assessing global
patterns in marine protected area performance. Coast. Manage. 42, 207–226
(2014).
13. Ferraro, P. J. Counterfactual thinking and impact evaluation in environmental
policy. New Dir. Eval. 2009, 75–84 (2009).
14. Ahmadia, G. N. et al. Integrating impact evaluation in the design and
implementation of monitoring marine protected areas. Philos. Trans. R. Soc. B
Biol. Sci. 370, 20140275 (2015).
15. Ostrom, E. Governing the Commons: The Evolution of Institutions for Collective
Action (Cambridge Univ. Press, 1990).
16. Scianna, C., Niccolini, F., Gaines, S. D. & Guidetti, P. ‘Organization Science’:
A new prospective to assess marine protected areas eectiveness. Ocean Coast.
Manage. 116, 443–448 (2015).
17. Hockings, M. et al. Evaluating Eectiveness: A Framework for Assessing
Management Eectiveness of Protected Areas 2nd edn (International Union for
Conservation of Nature (IUCN), 2006).
Supplementary Information is available in the online version of the paper.
Acknowledgements This research was supported by the National Socio-
Environmental Synthesis Center (SESYNC) under funding received from the
National Science Foundation DBI-1052875, as part of the working group:
Solving the Mystery of Marine Protected Area (MPA) Performance: Linking
Governance, Conservation, Ecosystem Services and Human Well Being. D.A.G.
was jointly supported by postdoctoral fellowships from the Luc Hoffmann
Institute and SESYNC. We thank the following data providers: Atlantic Gulf
Rapid Reef Assessment (AGRRA) contributors and data managers, Conservation
International, Healthy Reefs Initiative, I. Williams (NOAA Coral Reef Ecosystem
Program), NOAA Coral Reef Conservation Program, K. Knights (Global Database
for Protected Area Management Effectiveness), G. Edgar and R. Stuart-Smith
(Reef Life Surveys), The Nature Conservancy, Wildlife Conservation Society, and
the World Conservation Monitoring Centre. We also thank other members of
the SESYNC MPA Pursuit team: A. Agrawal, G. Cid, A. Henshaw, I. Nur Hidayat,
W. Liang, P. McConney, M. Nenadovic, J. E. Parks, B. Pomeroy, C. Strasser and
M. Webster, and P. Marchand of SESYNC for scientific support. We acknowledge
GEF, USAID, and the many other funders who supported authors’ time and data
collection. This is contribution no. 9 of the research initiative Solving the Mystery
of MPA Performance.
Author Contributions H.E.F. and M.B.M. conceived the study. D.A.G. led the
analysis and data compilation with the assistance of H.E.F., M.B.M., G.N.A., L.G.,
S.E.L., M.B., I.C., E.S.D., C.M.F., J.G., S.H., O.P.J., L.C., G.G., P.J.M, H.T., S.W. and S.W.
C.M.F. prepared the maps. D.A.G., H.E.F., M.B.M., G.N.A., L.G. and S.E.L. wrote the
manuscript with the input of all the other authors.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the paper.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations. Correspondence and
requests for materials should be addressed to D.A.G. (dgill@conservation.org).
Reviewer Information Nature thanks A. Rosenberg, B. Worm and the other
anonymous reviewer(s) for their contribution to the peer review of this work.
18. Coad, L. et al. Measuring impact of protected area management interventions:
current and future use of the Global Database of Protected Area Management
Eectiveness. Philos. Trans. R. Soc. B Biol. Sci. 370, 20140281 (2015).
19. Claudet, J. et al. Marine reserves: size and age do matter. Ecol. Lett. 11,
481–489 (2008).
20. Lester, S. & Halpern, B. Biological responses in marine no-take reserves versus
partially protected areas. Mar. Ecol. Prog. Ser. 367, 49–56 (2008).
21. McClanahan, T. R., Marnane, M. J., Cinner, J. E. & Kiene, W. E. A comparison of
marine protected areas and alternative approaches to coral-reef management.
Curr. Biol. 16, 1408–1413 (2006).
22. Cinner, J. E. et al. Comanagement of coral reef social-ecological systems. Proc.
Natl Acad. Sci. USA 109, 5219–5222 (2012).
23. Nolte, C. & Agrawal, A. Linking management eectiveness indicators to
observed eects of protected areas on re occurrence in the Amazon
rainforest. Conserv. Biol. 27, 155–165 (2013).
24. Williams, I. D. et al. Human, oceanographic and habitat drivers of central
and western Pacic coral reef sh assemblages. PLoS One 10, e0120516
(2015).
25. Wenger, A. S. et al. Eects of reduced water quality on coral reefs in and out of
no-take marine reserves. Conserv. Biol. 30, 142–153 (2016).
26. Advani, S., Rix, L. N., Aherne, D. M., Alwany, M. A. & Bailey, D. M. Distance from a
shing community explains sh abundance in a no-take zone with weak
compliance. PLoS One 10, e0126098 (2015).
27. Ferraro, P. J. & Pressey, R. L. Measuring the dierence made by conservation
initiatives: protected areas and their environmental and social impacts. Philos.
Trans. R. Soc. B Biol. Sci. 370, 20140270 (2015).
28. Watson, J. E. M., Dudley, N., Segan, D. B. & Hockings, M. The performance and
potential of protected areas. Nature 515, 67–73 (2014).
29. Sandvik, B. World Borders Dataset. http://thematicmapping.org/downloads/
world_borders.php (2016).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
METHODS
Data reporting. Sample size was not based on power analysis but on availa-
ble global, regional and national datasets of management and fish survey data
(Supplementary Table 2). The sample meets the requirements for the selected mod-
elling approaches used in the study. As the study was based on observational data,
the experiments were not randomized, and quasi-experimental procedures were
used in order to replicate the conditions of a randomized experiment.
MPA attribute and zone information. MPA geospatial and attribute data (that is,
location, shape/boundaries, age, area, fishing regulations) were sourced from the
October 2015 version of the World Database on Protected Areas (WDPA)30 as well
as other regional and international MPA datasets (see Supplementary Information).
Where possible, these data were supplemented and/or validated using scientific
publications, reports, other official gover nment and non-government sources, the
ecological data providers, and local expert knowledge (Supplementary Table 4).
For the purpose of this study, ‘fishing prohibited’ refers to an MPA or zone within
an MPA that prohibits any type of fishing activity, including subsistence and
recreational fishing.
MPA management data. Data on MPA management processes were sourced
from three management assessment tools: Management Effectiveness Tracking
Tool (METT)
31
, the World Bank MPA Score Card
32
, and the NOAA Coral Reef
Conservation Program’s (CRCP) MPA Management Assessment Checklist33
(Supplementary Table 2).
Management indicator scores were rescaled to ensure construct validity between
the assessments (Supplementary Table 3). To assist with the interpretation of the
different scoring levels and criteria, we defined binary thresholds for each indicator
based on the description of the scoring levels and social theory (Supplementary
Tables 1 and 3). These thresholds were for descriptive purposes only; we used
the rescaled indicator scores (as described in Supplementary Table 3) in the sta-
tistical models. For MPAs that had multiple management assessments, we used
the most recent assessments available for describing the status of management
processes in MPAs worldwide (for example, for results in Fig. 1). For the models
testing relationships with ecological outcomes, we used the assessment that was
closest in time to when the ecological surveys were done, preferably before the
ecological data were collected. If no assessment was available before the ecological
surveys, we chose the one closest in time after the survey. When there was more
than one assessment in the same year we used the median score. There were a few
cases of survey respondents reporting non-integer scores (for example, 2.5) or
cases when such scores arose from calculating the median value for a specific year
(see Extended Data Fig. 8). No rounding was carried out on non-integer scores,
however; MPAs with thes e non-integer values were excluded from maps and graph-
ics (Figs 3b and 4) to simplify interpretation.
Ecological impact data. We derived ecological data on marine fish populations
from seven independent global and regional datasets, with the majority compris-
ing species-level data from underwater visual census (UVC) surveys on coral or
rocky nearshore reefs (Supplementary Table 2), and the remainder coming from
meta-analyses5,20. For the UVC data (15,978 survey sites), biomass represents the
total biomass of all recorded fish species, averaged across all transects at each
site (grams per 100 m
2
). Variations in sample methods meant that the choice of
recorded species varied between datasets (Supplementary Table 2); therefore
response ratios were never calculated among surveys from different datasets.
Biomass values were calculated by the data providers or the authors using the
individual body lengths and allometric length–weight data obtained either from
the data provider or from FishBase (http://www.fishbase.org).
Isolating MPA causal effects. We identify MPA causal effects by comparing MPA
survey sites to comparable non-MPA sites (outside MPA boundaries and/or before
establishment) and calculating lnRR values. Here we use statistical matching and
other procedures (described below) to account for: i) selection biases in MPA
placement; ii) spatiotemporal dynamics of f ish response to protection (for example,
spill-over, recovery time); and iii) other biological, social and physical factors that
can affect fish populations14.
Effective assessment of MPA impact necessitates the isolation of response to
protection (MPA treatment) from other confounding factors34. Statistical matching
allows us to develop a functional counterfactual by using the same factors that
determine where MPAs are placed (for example, opportunity costs for fishing)
to select control sites
13,14
. Other factors that explain variation in fish populations
(for example, habitat, depth, wave energy) can also be used as covariates in the
matching process. This assumes t hat, conditional on confounding covariates (both
observed and unobserved), the control and treatment sites are interchangeable,
that is, from the same population35. Thus, with appropriate metrics or proxies
of potentially confounding variables, control (non-MPA) and treatment (MPA)
survey sites can be appropriately matched, with the majority of the remaining
variation in the differences between the two groups attributable to the treatment
(MPA protection) effect36.
Controlling for spill-over and response time lags. Before matching, we removed
survey sites that might confound the measurement of effects. To account for (spatial)
spill-over effects, only control survey sites greater than one kilometre away from an
established MPA boundar y were use d in the ana lysis (1,116 control sites removed).
Despite many individual species having larger home ranges37,38, a review of studies
examining spill-over effects of marine reserves39 indicates that one kilometre is a
sufficient distance beyond which most population-level MPA effects can no longer
be detected. Any spill-over effects present in sites beyond this range will result in
a more conservative estimate of MPA effects as it will reduce the inside–outside
differences.
To account for time lags in fish response to protection, we assigned a survey site
to an MPA only if the MPA was established for at least three years. Initial detectable
responses to protection can be quite rapid (for example, 1.5–2 years
40
, 1–3 years
41
,
2–5 years
42
) and three years appeared to be sufficient time for MPA impacts to
become detectable. All sites within an MPA less than three years old were not
used as MPA (treatment) sites (n = 579 sites). All survey sites located within the
boundaries of an MPA before the first (complete) year of MPA establishment were
treated as ‘before’ (control) sites given that a protection response is unlikely to
occur within so short a period of time (n = 123 sites or 3.0% of 4,125 control sites).
After removing the above mentioned sites and sites with ambiguous loca-
tions (n = 1,882 sites total), we proceeded with matching on 14,096 survey sites,
comprising 9,971 treatment (MPA) and 4,125 (non-MPA) control sites.
Matching to control for observable bias. On the basis of existing literature on
MPA site-selection biases and factors affecting variation in fish populations,
Supplementary Table 5 describes the variables compiled for each survey site and
used in the matching process. We performed multivariate matching using the
Matching package 4.9-0 (ref. 36) in the statistical software R v3.2.3 (ref. 43). We
assessed the performance of various matching iterations using the post-matching
covariate match balance outputs (Supplementary Table 6) and quantile–quantile
plots. Here we attempted to reduce the standardized mean differences between
covariates for control (non-MPA) and treatment (MPA) to below 5%, which is
considered appropr iate for studies assessing casual inference
44
. We chose ne arest-
neighbour multivariate matching algorithms (based on Mahalanobis distances),
as they performed better than propensity score algorithms for our data. As there
were fewer control than treatment sites, we matched with replacement, and
allowed multiple control sites to be matched to each treatment site. Matching with
replacement prevents ordering effects and allows the algorithm to choose the best
available match from the entire population of control sites. Allowing multiple treat-
ment–control matches reduces the influence of outliers by increasing the number
of matched pairs. For our data, matching two controls to each treatment site
(2:1 ratio) resulted in lower standardized mean differences in treatment–control
covariates than 1:1 matching, or using higher ratios (for example, 3:1,4:1). All
covariates carried equal weight, however covariate ‘callipers’ were used to ensure
lower differences between the treatment and control sites for select covariates14
(see Supplementary Table 5). To help determine appropriate callipers, we used random
forest models and partial dependency plots to explore the relationship between each
covariate and fish biomass (using no-take sites to control for fishing impact). These
were useful in determining both the strength of the relationship between the covari-
ate and fish biomass, and to identif y asymptotic peaks be yond which the covariate
has no effect (for example, shore distance appeared to have litt le effect on fish bio-
mass beyond 20 km). Callipers improved the quality of the matching, but reduced
the overall number of possible matches; 2,335 (23%) treatment (MPA) sites were
dropped owing to failure to find appropriate controls to match the treatment sites.
Some of these drops were due to failure to find an appropriate control site within
the same country or close in time to match with the treatment site. This resulted in
15,821 matched observations for 7,636 treatment sites in 178 MPAs. These matched
pairs were used to derive response ratios (and their natural logarithm) for total fish
biomass, which were averaged to the MPA level (Extended Data Fig. 8k).
We used Rosenbaum’s bounds sensitivity analysis to assess the vulnerability of
our MPA treatment effects to unobserved biases (that is, factors not included in our
list of matching covariates t hat could confound our estimates of MPA impact
35,45
).
Rosenbaum’s sensitivity bounds do not indicate whether or not such biases exist,
but merely the potential for such a bias to influence our findings. When assessing
the sensitivity of our estimates of MPA effects on fish biomass to an unobserved
variable, we find that if such a variable were able to change the odds of a site being
protected by a factor (Γ) of 1.35, it would confound our estimate of effect. While
Γ = 1.35 suggests some sensitivity in our findings to potential unobserved bias,
there is no evidence to suggest such a bias exists. Our extensive list of observed
covariates (Supplementar y Table 5) were identified through expert knowledge, the
scientific literature, and available primary and secondary data as key factors that
affect both MPA participation and outcomes. Further, covariates that remained
significant after matching (for example, shore distance, chlorophyll) were con-
trolled for in subsequent models (Supplementary Table 9).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
We supplemented the matched UVC data (n = 178 MPAs) with MPA-level fish
biomass ratios from existing datasets
5,20
(n = 40 MPAs), w hich comprise response
ratios derived from 149 peer-reviewed publications that examine the ecological
effects of areas where fishing is prohibited (marine reser ves or no-take areas) and
areas where fishing is allowed but restricted (multi-use). Where data were avail-
able for an MPA in both the existing and matched datasets (n = 11 MPAs), we
chose the latter. No matching was required for the existing data as response ratios
were already formulated by the authors in their meta-analysis. The final ecological
dataset totalled 218 MPAs (see Extended Data Fig. 2 for data compilation steps).
Management and ecological data analysis. We used random forests with con-
ditional inference trees
46
to identify the management processes (Supplementary
Table 4) that best explained the variation in ecological impact (n = 62 MPAs).
Random forests account for higher-order interactions and nonlinear relationships
between predictors, and do not require many of the strict assumptions of linear
parametric models that are difficult to meet47. These qualities make random forests
an ideal approach for our analysis, where many interacting and nonlinear relation-
ships among management processes, MPA attributes, and ecological outcomes are
expected11. Random forests are also able to effectively estimate variable importance
in ‘small n, large p’ models and models with missing data47,48.
In this study, we used the R party package v1.0-25 (ref. 49) to estimate the
relative variable importance of the ten management indicators using the fish
biomass lnRR values as the response variable and the metric for e cological impact.
In addition to the management indicators, we also included other non-management
variables as predictors in the model. Many of these were identified in the
literature as being important in explaining variability in fish populations and MPA
ecological outcomes (MPA age, MPA size, fishing regulations)
7,19,20
, and include
many of the variables used in the matching process (mean MPA depth, shore
distance, market distance, human population density, chlorophyll, wave exposure,
sea surface temperature, ecoregion, countr y; Supplementar y Table 5). This allowed
us to assess the relative importance of the management indicators as predictors,
while accounting for (and allowing interactions with) these potentially important
non-management factors.
Given that we were investigating the MPA-level impact of management, the
MPA was considered as the unit of analysis. Therefore all variables, including
response ratios, were averaged to the MPA level. All non-management predic-
tors represent the MPA-level average of the conditions at each fish survey site
(for example, mean depth represents the mean depth of the fish survey sites in
that MPA). All continuous predictors were transformed to the natural log scale
to reduce the effect of extreme outliers, with the exception of depth, which did
not need to be transformed. Proportion no-fishing represents the proportion of
survey sites for an MPA sampled from within a prohibited-fishing (no-take) zone
(0: all multi-use, 1: all prohibited fishing). See Supplementary Information and
Extended Data Fig. 9 for more details on the procedures and variables used in the
random forest modelling.
We also ran a series of general linear mixed-effects models (Supplementary
Table 9) to examine the direction and strength of the relationships between each
of the management indicators and ecological impact. The linear mixed effects
models allowed us to examine the predictor–response relationships in a hierar-
chical model structure, while controlling for other important non-management
factors. These non-management variables were those identified as important in
the random forest models (mean chlorophyll, mean shore distance, mean MPA
age, MPA size) and those found to be important in the literature (that is, fishing
regulations: ‘proportion no fishing’). For the hierarchical structure, we included
a random intercept for country to account for potential non-independence in the
fish response to protection between MPAs in the same country (for example, MPAs
managed by the same national agency). Including country as a random intercept
performed similarly to other random effect structures that account for spatial
hierarchy (see Supplementary Table 8). We used the R nlme package v3.1-128
(ref. 50) to implement the linear mixed models and only included one management
predictor in each model owing to strong correlation (Extended Data Fig. 6) and
missing data amongst some of the predictor variables. The results are shown in
Supplementary Table 9.
Data availability. The authors declare that the source data supporting the findings
of this study are available within the paper and its Supplementary Information,
including source data for Figs 1–3 and Extended Data Figs 3–7 and 9. All other
data and R code are available from the corresponding author upon reasonable
request.
30. IUCN and UNEP-WCMC. The World Database on Protected Areas (WDPA).
(United Nations Environment Programme (UNEP) World Conservation
Monitoring Centre (UNEP-WCMC) and International Union for Conservation of
Nature (IUCN) http://www.protectedplanet.net (2015).
31. Stolton, S. et al. Management Eectiveness Tracking Tool: Reporting progress in
Protected areas sites; second edition (World Bank/WWF Forest Alliance and
WWF, 2007).
32. Staub, F. & Hatziolos, M. E. Score Card to Assess Progress in Achieving
Management eectiveness goals for Marine Protected Areas (Prepared for the
World Bank, 2004).
33. NOAA. NOAA Coral Reef Conservation Program MPA Management Assessment
Checklist (National Oceanic and Atmospheric Administration (NOAA), 2010).
34. Mora, C. & Sale, P. Ongoing global biodiversity loss and the need to move
beyond protected areas: a review of the technical and practical shortcomings
of protected areas on land and sea. Mar. Ecol. Prog. Ser. 434, 251–266
(2011).
35. Rosenbaum, P. R. Design sensitivity and eciency in observational studies.
J. Am. Stat. Assoc. 105, 692–702 (2010).
36. Sekhon, J. S. Multivariate and propensity score matching. J. Stat. Softw. 42,
1–52 (2011).
37. Alerstam, T., Hedenstrom, A. & Akesson, S. Long-distance migration: evolution
and determinants. Oikos 103, 247–260 (2003).
38. Green, A. L. et al. Larval dispersal and movement patterns of coral reef shes,
and implications for marine reserve network design. Biol. Rev. Camb. Philos.
Soc. 90, 1215–1247 (2015).
39. Halpern, B. S., Lester, S. E. & Kellner, J. B. Spillover from marine reserves and
the replenishment of shed stocks. Environ. Conserv. 36, 268–276 (2010).
40. Russ, G. R. et al. Rapid increase in sh numbers follows creation of World’ s
largest marine reserve network. Curr. Biol. 18, 514–515 (2006).
41. Halpern, B. S. & Warner, R. R. Marine reserves have rapid and lasting eects.
Ecol. Lett. 5, 361–366 (2002).
42. Gell, F. R. & Roberts, C. M. Benets beyond boundaries: the shery eects of
marine reserves. Trends Ecol. Evol. 18, 448–455 (2003).
43. R Development Core Team. R: a language and environment for statistical
computing, version 3.2.3. (2015).
44. Caliendo, M. & Kopeinig, S. Some practical guidance for the implementation of
propensity score matching. J. Econ. Surv. 22, 31–72 (2008).
45. Keele, L. An overview of rbounds: an R package for Rosenbaum bounds
sensitivity analysis with matched data. (2010).
46. Hothorn, T., Hornik, K. & Zeileis, A. Unbiased recursive partitioning: A conditional
inference framework. J. Comput. Graph. Stat. 15, 651–674 (2006).
47. Strobl, C., Malley, J. & Tutz, G. An introduction to recursive partitioning:
rationale, application, and characteristics of classication and regression trees,
bagging, and random forests. Psychol. Methods 14, 323–348 (2009).
48. Hapfelmeier, A., Hothorn, T., Ulm, K. & Strobl, C. A new variable importance
measure for random forests with missing data. Stat. Comput. 24, 21–34
(2012).
49. Hothorn, T., Hornik, K., Strobl, C. & Zeileis, A. Package ‘Party’: A Laboratory for
Recursive Partytioning. R package version 3.1-128. (2015).
50. Pinheiro, J. C. & Bates, D. M. Linear and nonlinear mixed-eects models, R
package version 3.1-128. http://cran.r-project.org/web/packages/nlme/index.
html (2016).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
Extended Data Figure 1 | Key domains and illustrative indicators for assessing management efficacy and equity. Indicators with asterisks are those
that were used in this study. Details on indicator descriptions, sources and citations are located in Supplementary Table 1.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 2 | Sources and major steps in the data compilation and analysis. See Supplementary Table 2 for more details on data sources.
CRCP, Coral Reef Conservation Program.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
Extended Data Figure 3 | Per cent of MPAs by managing authority
exceeding or falling below threshold values for indicators of effective
and equitable management processes. Details on indicators, scores and
threshold values in Supplementary Tables 1 and 3. Dark blue bars (right)
indicate the proportion of MPAs with scores at or above the threshold
value, light blue bars (left) indicate the proportion below the threshold.
Scores are from the latest assessment year where data were available from
433 MPAs.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 4 | Mean fish biomass response ratios (lnRR) by fishing regulations. Mean (dot) and 95% confidence intervals (error bars) for
areas where fishing is prohibited (dark blue) and multi-use MPA areas (light blue) in 254 zones in 218 MPAs.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
Extended Data Figure 5 | Relationship between mean fish biomass
response ratios (lnRR) and key predictor variables used in the
analysis of the relationship between MPA management processes and
ecological impact (n ≤ 62 MPAs). a–j, Mean (dot) and 95% confidence
intervals (error bars) of the response ratios for each management score
and indicator. Details on threshold levels and score descriptions in
Supplementary Table 3. k–t, Smoothed LOESS lines (blue line) along with
the standard error regions (shaded area) for relationships with continuous
variables. Number of MPAs in parentheses.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 6 | Spearman rank correlations amongst
management indicators, national variables and other key variables
(n = 433 MPAs). Variables ordered using hierarchical clustering,
displaying values for significant correlations only (P < 0.05). Circle size
and colour indicate the correlative strength and direction, respectively
(blue, positive; red, negative). Most of the management indicators
for procedural efficacy were significantly correlated with each other
(for example, correlation coefficient for monitoring and management
plan = 0.49). National level variables (GDP, HDI) were poorly correlated
with management indicators and were not included in this study.
ENF, acceptable enforcement capacity; BGT, acceptable budget capacity;
REG, appropriate MPA regulations; MON, monitoring informing
management activities; MPL, implementing existing management plan;
BND, clearly defined boundaries; LEG, legally gazetted; STF, adequate staff
capacity/presence; DEV, non-state/shared management; IDM, inclusive
decision-making; SIZ, MPA size (ln[km2]); AGE, MPA age (ln[years]);
HDI, Human Development Index 2010; GDP, gross domestic product per
capita (ln[US$ PPP]) 2013.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
Extended Data Figure 7 | Spearman rank correlations amongst fish
metrics, management indicators, and other key variables for the
62 MPAs used in the management and ecological data analysis. Circle
size and colour indicate the correlative strength and direction, respectively
(blue, positive; red, negative). Variables ordered by type (that is, ecological,
management, and so on) and not hierarchical clusters, displaying values
for significant correlations only (P < 0.05). BIO, lnRR; DEN, natural
logarithm of fish density response ratio; FSZ, natural logarithm of
fish mean size response ratio; RCH, natural logarithm of fish species
richness response ratio; DEV, non-state/shared management;
IDM, inclusive decision-making; LEG, legally gazetted; REG, appropriate
MPA regulations; BND, clearly defined boundaries; ENF, acceptable
enforcement capacity; MON, monitoring informing management
activities; MPL, implementing existing management plan; STF, adequate
staff capacity/presence; BGT, acceptable budget capacity; NTZ, proportion
of survey sites for an MPA sampled from within a prohibited-fishing
(no-take) zone; SIZ, MPA size (ln[km2]); AGE, MPA age (ln[years]); CHO,
chlorophyll a concentration (ln[mgm−3]); SHR, distance from shore
(ln[km]).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 8 | Frequency distribution of MPA management,
ecological and other key variables. a–n, White bars indicate the
distribution of scores from the latest available management assessments
in433 MPAs (a–j); MPAs where fish biomass data were available (n ≤ 218
MPAs) (k–n). Grey bars indicate MPAs used in the analysis modelling
the relationship between management processes and ecological impact
(n ≤ 62 MPAs). Indicators for inclusive decision-making (b) and
enforcement (g) have a maximum score of 2. Non-integer values were
reported scores by few managers, or represent the median value of
multiple assessments in the latest year. k, Mean (MPA-level) response
ratios (natural log scale) for fish biomass. l, Proportion of survey sites
for an MPA sampled from within a prohibited-fishing (no-take) zone
(0, all multi-use area; 1, all no-take/prohibited fishing area). m, MPA age
(years between establishment and fish survey). n, MPA size (thousand
km2). MPA age and size were transformed to the log scale for the analysis.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE
RESEARCH
Extended Data Figure 9 | Random forest variable importance plots.
Random forest variable importance measures for management (blue
bars) and other (non-management; grey bars) variables as they relate
to ecological impact in 62 MPAs. a, b, Results from models with all
management indicators (as shown in Fig. 3a in the main text) (a) and
management indicators with few missing data and not highly correlated
with other predictors (that is, excluding legal status, acceptable budget,
management plan, country and ecoregion) (b). Only values greater than
the red dashed line are considered to have non-random importance scores.
c, d, Predicted and observed response ratio values from the random forest
models in a and b respectively, along with the linear fitted line (dashed
blue line) and a smoothed LOESS line along with the standard error region
(grey line and shaded area). R2 values for the linear fit are also shown.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.