Effects of marine reserve age on fish populations: a global meta-analysis


*Correspondence author. E-mail: pmolloy@sfu.ca


1.  Marine reserves are widely used for conservation and fisheries management. However, there is debate surrounding the speed of population recovery inside reserves and how recovery differs among species. Here, we determine how reserve effectiveness in enhancing fish density changes with reserve age. We also examine how the effects of protection vary between fished and non-fished species and among species of different body sizes, which we use as a proxy for life history and ecology.

2.  We meta-analysed over 1000 ratios of fish densities (inside : outside reserves) taken from reserves of 1–26 years old from around the world.

3.  Overall, older reserves were more effective than younger reserves, with fish densities increasing within reserves by ∼5% per annum relative to unprotected areas. Reserves older than 15 years consistently harboured more fish compared with unprotected areas; younger reserves were less reliably effective.

4.  Large, fished species responded strongly and positively to protection in old (>15 years) and, unexpectedly, in new and young (≤10 years) reserves. Small, fished species and non-fished species of all sizes showed weaker responses to protection that did not vary predictably with reserve age.

5.  We expected large fish to respond more slowly to protection than smaller species. We also expected small species to decline after large fish had recovered (i.e. trophic cascades). Neither prediction was supported.

6. Synthesis and applications. Our meta-analyses demonstrate that, globally, old reserves are more effective than young reserves at increasing fish densities. Our results imply that reserves should be maintained for up to 15 years following establishment, even if they initially appear ineffective. If protection is maintained for long enough, fish densities within reserves will recover and such benefits will be particularly pronounced for large, locally fished species.


Of the numerous threats to marine biodiversity, fishing is arguably the most pervasive (Halpern et al. 2008) and damaging (Watling & Norse 1998; Roberts & Hawkins 1999; Worm et al. 2006). The removal of fishing pressure, through the establishment of marine reserves in which all extraction is prohibited, results in a broad range of positive effects within reserves, such as increased biomass, abundance, average size and diversity of fish and invertebrates (Russ & Alcala 1996; Mosqueira et al. 2000; Roberts et al. 2001; Halpern 2003; Micheli et al. 2004; Guidetti et al. 2005; Micheli & Halpern 2005; Claudet et al. 2008). There is also evidence that these benefits can lead to similar improvements outside reserves through spillover of adults, eggs and larvae (Russ et al. 2003; Goñi et al. 2006; Kellner et al. 2007; Harmelin-Vivien et al. 2008), resulting in economic benefits from enhanced fisheries and tourism (White et al. 2008).

Reserves vary greatly in their effectiveness (Côtéet al. 2001) due to differences in reserve design, such as sizes of no-take and buffer zones (Halpern 2003; Parnell et al. 2006; Claudet et al. 2008) and reserve shape (Kramer & Chapman 1999). Social factors such as level of enforcement (Walmsey & White 2003; Guidetti et al. 2008) and community support (Walmsey & White 2003) also affect effectiveness.

Reserve age is expected to influence the benefits provided by marine reserves as recovery of some populations may take decades (Myers et al. 1997; Jennings 2001). More specifically, old reserves should have higher fish densities (relative to unprotected areas) than young reserves. Despite the intuitive appeal of this prediction, broad-scale tests of the importance of reserve age on fish densities have yielded conflicting results. Syntheses of data from marine reserves around the world show that increases in fish densities resulting from protection are not greater in older than in younger reserves (Côtéet al. 2001; Halpern & Warner 2002). By contrast, a meta-analysis of European reserves found more large individuals of fished species in older reserves (Claudet et al. 2008). Claudet et al. (2008) postulated that this effect is not seen in global meta-analyses due to the diversity of reserves and species considered combined with low statistical power. As the time required for reserves to ‘start producing’ is critical to the development of management plans and community support, resolving this age effect on a global scale is an important task.

Reserve effectiveness can also vary as a result of species differences in response to protection (Mosqueira et al. 2000; Micheli et al. 2004; Molloy et al. 2008). While spectacular increases in the abundance of some large-bodied, commercially important species following reserve establishment have been noted (e.g. Russ & Alcala 1996; Guidetti et al. 2005; Claudet et al. 2008), most species respond less predictably to protection (Chapman & Kramer 1999; Mosqueira et al. 2000; Micheli et al. 2004). The magnitude and speed of responses to protection vary with the extent of fishing pressure nearby (Mosqueira et al. 2000; Tetreault & Ambrose 2007) as well as with various species attributes. For example, species with fast life histories should rebound quickly when protected (Jennings 2001); species with small home ranges may gain complete protection within reserves (Kramer & Chapman 1999), but may be less likely to colonize a new reserve (Denny et al. 2004). For highly fecund species with pelagic phases, recovery via larval recruitment from neighbouring populations may be important (Domeier 2004). Predatory species, which are commonly fished (Pauly et al. 1998), are likely to benefit from protection, while prey species may suffer through ensuing trophic cascades (Dulvy et al. 2004; Micheli et al. 2004; Díaz et al. 2005; Baskett et al. 2007).

Many life-history and ecological characteristics correlate strongly with body size (Blueweiss et al. 1978; Charnov 1993; Calder 1996; Denney et al. 2002; Jetz et al. 2004). For this reason, body size has been advocated as a useful surrogate for life-history and ecological drivers of population recovery in data-poor situations (Jennings 2001). However, as fishing disproportionately kills large species (Heino & Godo 2002), care is required to disentangle the effects of body size (and its associated correlates) and local fishing pressure on responses to protection.

Reserve age probably interacts with species behaviour, ecology and life history to explain variation in reserve effectiveness. Arguably, small species, with rapid life histories, will have greater potential to increase quickly when fishing stops (Denney et al. 2002). Small species should therefore exhibit larger increases in new reserves than large species (Baskett et al. 2007). Larger species should increase in abundance slowly but steadily, relative to unprotected areas. Such processes have been invoked to explain the time-lag between reserve implementation and spillover by large, slow-growing, late-maturing fish (Russ & Alcala 1996; Roberts et al. 2001). Moreover, in older reserves in which (large) species at high trophic levels have recovered, trophic cascades could lead to an apparent reduction in the effectiveness of older reserves at protecting small (prey) species (Díaz et al. 2005; Baskett et al. 2007).

The aim of this study was to examine the interaction between duration of protection and fish body size in determining the effectiveness of marine reserves at enhancing fish populations. We meta-analyse 1086 paired inside–outside estimates of fish densities from 33 studies of reserves of different ages from around the world. To disentangle the covariation between body size and fishing pressure, we examine the effects of fish size on response to protection for locally fished and non-fished species separately, in reserves of different ages.

Material and methods

Fish density estimates

We used the database of fish densities estimated in and out of marine protected areas compiled by Molloy et al. (2008), updated with data from two studies (Tuya et al. 2006; Barrett et al. 2007) (Supporting Information Table S1; Fig. S1). Studies were included if: (1) density estimates inside and outside a reserve were available at the species level, (2) the number of transects or point counts used to generate density estimates was reported and (3) the reserve was either no-take or had minimal, well-regulated fishing, as described by the authors of the source papers (see Table S1 for fishing activities permitted). When studies reported data from several years, we used the most recent data. We excluded studies that reported data aggregated over several reserves if the reserve ages differed by more than 5 years (N = 1 study: McClanahan et al. 1999).

Densities were recorded as the mean number of individuals per unit area inside and, separately, outside each reserve (or group of reserves in the two cases where data were aggregated across reserves McClanahan 1994; Edgar & Barrett 1999). Some studies reported multiple density values for a species in different parts of a reserve or non-reserve area. These values were aggregated into a weighted mean for each area (Mosqueira et al. 2000). When studies presented paired inside–outside density values for several reserves, data from each reserve were included separately in the database. García-Charton et al. (2004) reported data from several reserves that did not have paired non-reserve sites. Following Molloy et al. (2008), these reserves were separated in the dataset and each compared with the same averaged non-reserve density. When densities were only presented graphically (e.g. Barrett et al. 2007), we estimated them from the figures using data-extraction software (Tummers 2006).

Predictors of reserve effectiveness

We examine three predictors of reserve effectiveness: reserve age, exploitation status of the species and maximum reported (total) length of the species. Reserve age was the number of years between enforcement of protection and survey year. We examined reserve age as both a continuous and categorical variable. For the latter, we initially divided reserves into young (1–9 years) and old (10+) (Micheli et al. 2004; Molloy et al. 2008); however, our large sample size permitted a more sensitive gauge of when reserves become effective. Thus, reserves were divided into new (1–5 years), young (6–10 years), established (11–15 years) and old (16+ years).

Exploitation status for each species, where available, was obtained from the source studies to ensure that species were assigned a locally relevant status. Four categories were used: fished, non-fished, bycatch and unknown. Bycatch and ‘unknown’ species were excluded from analyses involving exploitation status as the fishing pressure on these species was unclear.

Maximum reported (total) length (referred to as body size) was used as a proxy for life history and ecology. Body size for each species was obtained from FishBase (Froese & Pauly 2008). Standard or fork lengths were converted to total length using known relationships when necessary (Froese & Pauly 2008). To remove the potential confounding effect of body-size variation among reserves, median sizes were calculated for each reserve, and species in each reserve were classified as above (large) or below (small) the reserve-specific median. Species with body sizes equal to the median were excluded.


To conduct meta-analyses, we generated a standardized effect size for each species in each reserve. We used the log-response ratio ln RR, calculated as ln(XI/XO), where XI and XO are mean densities of a species inside and outside the reserve respectively. Use of the natural logarithm is advocated because of its preferred statistical properties (Rosenberg et al. 2000). We added 0.0001 to all raw abundance values to allow calculation of ln RR for species that were absent inside or outside a reserve (Molloy et al. 2008).

Effect sizes are commonly weighted to ensure a greater contribution of the most robust studies. Robustness is usually based on (inversed) sample variance (Rosenberg et al. 2000); to include studies that did not provide variance estimates, we assumed that studies that had surveyed a greater area were more robust (sensuMosqueira et al. 2000; Côtéet al. 2001; Molloy et al. 2008). We therefore weighted each effect size by the natural logarithm of the total area surveyed.

We calculated a grand overall effect size as:


where ln RRi and Wi are, respectively, the effect size and weighting for species i. A 95% bias-corrected bootstrapped confidence interval (CI) around inline image was generated from 5000 iterations. Effect sizes (back-transformed for ease of interpretation) were considered significant when their 95% CI did not include 1: CI > 1 indicates a positive effect of protection, CI < 1 indicates a negative effect. All meta-analyses were performed using metawin2.0 (Rosenberg et al. 2000).

Reserve-level analyses

We calculated reserve-specific effect sizes, inline image, as the average ln RR of the species recorded within reserve j. CIs were calculated as above.

We first performed a meta-regression to examine the effect of reserve age on protection effectiveness and assessed the model fit using a Q-regression (QReg) test (Rosenberg et al. 2000). We then carried out a categorical mixed model meta-analysis comparing each of the four reserve-age classes. Differences in overall effect sizes among classes were tested using QM statistics (Rosenberg et al. 2000). Some studies only considered a subsample of the local fish community. To avoid biases that could be introduced by subsampling, we repeated analyses of effects of reserve age using only studies that considered all species. These results were similar to those including all studies; hence, we do not present them.

Species-level analyses: exploitation status and size

We sought to determine the effect of protection on fished and non-fished species over time, and changes in the effect of body size on response to protection over time. Fished species in our data set were larger than non-fished species (Kruskall–Wallis test, inline image = 345·51, < 0·001). We therefore categorized species according to exploitation status (fished or not) and body size (large or small) in each of the four reserve-age classes. As metawin currently does not support multivariate meta-analyses, we conducted categorical meta-analyses of reserve-age effects within each of the four fishing-status × body-size classes. These within-group categorical analyses allowed us to test: (1) whether species groups (e.g. large, fished species in old reserves) respond significantly to protection by examining whether CIs overlapped 1, (2) the magnitude (via the QT statistic) and significance of heterogeneity in species responses to protection within each category and (3) whether there were significant differences among reserve-age classes using the QM statistic. As QM statistics do not identify specifically which groups differ, we performed categorical meta-analyses for all body size and reserve-age pairs (e.g. large vs. small, fished species in old reserves; large, fished species in old vs. new reserves). We controlled for multiple testing using false discovery rate (FDR, Benjamini & Hochberg 1995). The specific data used varied across analyses; therefore, sample sizes reported differ from the total number of response ratios in our database.

Finally, we examined the effects of several confounding variables. We used general linear models to test differences in inline image across latitudes (three categories: tropical, subtropical and temperate), habitat (five categories: coral, kelp, rocky, rocky and kelp, and rocky and seagrass), geographical regions (five categories: Pacific, Indo-Pacific, Atlantic, Caribbean and Mediterranean), size of no-take area (log10 x + 1) and size of entire reserve (log10) (both continuous). To ensure that variable species composition did not drive differences among reserve-age categories, we performed an analysis of similarity (anosim) based on species presence/absence in or out of reserves using primer (Plymouth Routines in Multivariate Ecological Research v. 5.2.4; PRIMER-E Ltd, Plymouth Marine Laboratory, Plymouth, UK). For this analysis, each study was considered a sample, and Bray–Curtis similarity coefficients between pairs of reserves were computed (Clarke & Warwick 2001). anosim generates an R-statistic, which varies between 0 (similarity within reserve = similarity between reserves) and 1 (reserves within age classes are more similar to each other than to reserves across groups), and we tested for difference from zero with a permutation test (= 999). Species composition similarity was tested for each of the four body-size × fishing-status groups, and the significance threshold adjusted for FDR (see above).

Publication bias and fail-safe number

A common problem with meta-analyses is publication bias, which arises when significant results are disproportionately published (Begg 1994). We tested for publication bias using the normal quantile plot (Wang & Bushman 1998). Publication bias is suspected when <95% of the data points lie inside the confidence lines. We also calculated the fail-safe number for our grand overall effect size, i.e. the number of non-significant, unpublished or missing studies that would be required to overturn a significant result. If the fail-safe number is greater than 5n + 10 (= number of studies in the meta-analysis), the result can be considered robust (Rosenthal 1979).


Overall effect of protection

Overall, fish were 66% more abundant inside than outside reserves (inline image = 1·66, CI 1·29–2·11, = 1086). This difference is significant as the CI does not span 1. The normal quantile plot revealed that >95% of the data points were within the confidence limits, providing no evidence of publication bias. The fail-safe number for the overall effect of protection was 5517, which exceeds Rosenthal’s suggested threshold (here, 5440) for confidence in the results.

Effect of reserve age on effectiveness

When all studies were considered, reserve effectiveness increased significantly with reserve age (slope = 0.050, SE = 0·027; QReg = 3·54, = 0·049, = 33; Fig. 1). This slope translates into an average increase in fish density, relative to unprotected areas, of ∼5% per year.

Figure 1.

 The effect of marine reserve age on the ratio of fish densities in and out of reserves (response ratio, inline image). The black line shows the best-fit meta-regression model: inline image = −0.0727 + 0.05 × reserve age. = 33 studies of marine reserves. The grey line at RR = 1 represents equal fish densities in and out of reserves; inline image > 1 means more fish in reserves, inline image < 1 means fewer fish in reserves. Y-axis shown on logarithmic scale for ease of interpretation.

In line with our continuous result, only old reserves had significantly more fish inside than outside their boundaries (Fig. 2). Overall, there were no differences among reserve-age classes (QM 3,29 = 4·75, = 0·18).

Figure 2.

 Ratio of fish densities in and out of marine reserves (response ratio, inline image) of four age classes. Overall reserve-level response ratios are shown with 95% bias-corrected CIs. The number of reserves in each age class is given above each error bar. The dashed line at inline image = 1 represents equal fish densities in and out of reserves; inline image > 1 means more fish in reserves, inline image < 1 means fewer fish in reserves. Y-axis shown on logarithmic scale for ease of interpretation.

None of the potential confounding factors considered influenced reserve effectiveness (GLMs, latitude: F2,30 = 1·34, = 0·28; region: F4,28 = 0·91, = 0·47; habitat: F4,28 = 0·26, = 0·90; log-size no-take zone: F1,31 = 1·93, = 0·17; log-size full marine reserve: F1,31 = 1·04, = 0·32). Species composition was not significantly different across reserve-age categories for any of the body-size × fishing-status combinations (anosim tests, < 0·12, 0.06 < Padj < 0·35), which reflects the relatively even distribution of tropical, subtropical and temperate reserves across age categories.

Exploitation status and body size

Fished species

In vs. out of reserves. Small fished did not respond significantly to protection in any reserve-age class (Fig. 3a). Large, fished species were significantly more abundant inside new, young and old reserves, but not inside established reserves (Fig. 3a).

Figure 3.

 Ratio of densities of (a) fished and (b) non-fished species in and out of reserves (response ratio, inline image) of four different age classes. Mean inline images are shown with 95% bias-corrected CIs. Black: large-bodied species (relative to the median of all fish in the same exploitation × reserve-age class); grey: small-bodied species. Sample sizes (i.e. number of species) are given above each error bar. The dashed line at inline image = 1 represents equal fish densities in and out of reserves; inline image > 1 means more fish in reserves, inline image < 1 means fewer fish in reserves. Y-axis shown on logarithmic scale for ease of interpretation.

Differences among reserve-age classes. After controlling for FDR, responses to protection of large and small, fished species did not differ significantly among reserve-age classes (Table 1a; Fig. 3a).

Table 1.   Significant pairwise comparisons of responses to protection, measured as the ratio of fish density in vs. out of reserves, among reserve-age categories for fished and non-fished species
ComparisonData subgroupQMPPadj
  1. QM values were obtained from mixed model meta-analyses and indicate the magnitude of the difference between groups. Padj represents P-values after adjustment for false discovery rate.

Fished species
 Young vs. oldLarge fish3·590·060·35
 Established vs. oldLarge fish4·940·040·35
Non-fished species
 Young vs. establishedSmall fish4·100·070·41
 Established vs. oldSmall fish6·970·0140·16

Differences between large and small species. Large and small species did not differ significantly in their responses to protection in all reserve-age classes (Table 1b; Fig. 3a). Large and small, fished species showed non-significant heterogeneity in effect sizes (both QT < 158·61, d.f. = 147, > 0·24), suggesting consistent responses to protection within each size category.

Non-fished species

In vs. out of reserves. Small, non-fished species were unexpectedly variable in their response to protection. They were significantly less abundant inside established reserves, and significantly more abundant inside old reserves (Fig. 3b). Large, non-fished species did not respond significantly to protection in any reserve-age categories (Fig. 3b).

Differences among reserve-age classes. After controlling for FDR, small, non-fished species were relatively more abundant in old reserves than in new and established reserves (Table 1a; Fig. 3b).

Differences between large and small species. Small and large, non-fished species responded similarly to protection in all reserve-age classes (Table 1b; Fig. 3b). Moreover, both large and small, non-fished species showed non-significant heterogeneity in within-group effect sizes (large: QT = 101.80, d.f. = 102, = 0.49; small: QT = 109.97, d.f. = 99, = 0.21).


How quickly fish populations recover inside marine reserves and how recovery rates differ among species are important questions for reserve stakeholders. Our analyses of reserves distributed globally show that older reserves harbour higher relative densities (i.e. more fish in than out of a reserve) than younger reserves. Closer inspection revealed that this benefit was generated consistently by reserves older than 15 years. However, different species responded to protection differently. Large, locally fished species had significantly higher population densities in new and young reserves (≤10 years). This apparent protection benefit was also observed in reserves older than 15 years but not in reserves of intermediate ages. Small, fished species and large, non-fished species showed no response to protection, while the responses of small, non-fished species varied unpredictably with reserve age.

Previous attempts to link reserve age and overall reserve effectiveness have yielded conflicting results. Halpern & Warner (2002) compiled data on relative densities of fish and invertebrates from reserves around the world and found density gains in young reserves that persisted but did not increase over time. Similarly, Côtéet al. (2001) meta-analysed drivers of overall effectiveness of 12 reserves and found that reserve age did not predict relative fish densities. It is possible that low sample sizes, combined with amalgamating data across taxa, sampling methods and locations, may have veiled age effects in these analyses (Claudet et al. 2008). In a global, species-level meta-analysis focusing on fish, Micheli et al. (2004) found a weak, positive relationship between relative density of individual species and reserve age, but did not examine the link between reserve effectiveness and age. Another meta-analysis of 12 European reserves showed that the relative density of large individuals of commercially important fish increased with reserve age (Claudet et al. 2008). However, these European reserves are homogeneous in many respects (e.g. species composition, habitat, socio-economy, reserve design, management techniques and objectives), suggesting that age effects may be relatively weak (hence detectable only when variation among reserves is controlled), limited to the Mediterranean Sea or to the most heavily exploited segment of fish populations. By calculating overall effect size for reserves around the world and analysing them in a true meta-analytical framework, we show that a positive relationship between relative fish densities and reserve age is a global phenomenon.

Overall fish densities within reserves increased linearly by ∼5% per annum relative to unprotected areas nearby. This estimate is comparable with that reported by Micheli et al. (2004; 4% per annum), who used species as independent data, and Claudet et al. (2008; 8% per annum), who considered only large individuals of locally targeted species. The similarity between these three estimates is striking given the differences in data.

The result from our categorical analysis of reserve effectiveness showed that only reserves older than 15 years reliably increased overall relative fish densities. This result does not indicate that younger reserves will not be effective; indeed, benefits were observed in many younger reserves (Fig. 1). Rather, it implies that reserve stakeholders may see a measurable improvement in overall fish densities in less than 15 years but can be confident of such benefits given longer durations of protection. Alternatively, it is possible that only reserves that performed well soon after establishment are maintained to become ‘old’. If so, our analyses would be biased by a lack of data on ineffective old reserves. If only old reserves have stood the test of time because they have always been effective, relative densities in reserves at different times should be positively related, i.e. good reserves always have the highest relative densities and poor reserves always have the lowest. Data for multiple years were available for 11 reserves. The time between the first and last surveys ranged 2–15 years. There was no relationship between the first and last estimates of reserve effects (Pearson’s = 0.10, P = 0.76, = 11; without outlying reserve: Pearson’s = −0.49, = 0.15, = 10; Fig. 4).

Figure 4.

 Relationship between the earliest and most recent estimates of ratios of fish density in and out of various marine reserves (response ratio, inline image). Inner and outer colours of each data point represent ages at the time of the earliest and most recent surveys respectively. White: new reserve (≤5 years), light grey: young (6–10 years), dark grey: established (11–15 years), black: old (15+ years). The square point (Ninepine marine reserve, New Zealand; Barrett et al. 2007) is a statistical outlier. The dashed horizontal and vertical lines at inline image = 1 represents equal fish densities in and out of reserves; inline image > 1 means more fish in reserves, inline image < 1 means fewer fish in reserves. Both axes shown on logarithmic scale for ease of interpretation.

We had predicted that large species would recover slowly. However, we observed significantly higher relative densities of large fish in new and young reserves. Rapid responses to protection, particularly by large species, could result from immigration and changes in distribution of fishing. Large species tend to have large home ranges (Calder 1996; Kramer & Chapman 1999). While large ranges are likely to encompass unprotected waters, they may also facilitate initial encounter and colonization of reserves. This mechanism was invoked to explain the rapid recovery of snapper Pagrus auratus within the Poor Knights Islands reserve, New Zealand (Denny et al. 2004). In addition, immigration could occur from deeper areas where fish have sought refuge from depth-limited fishing (e.g. spearfishing). Their movement into shallower protected waters (P. Guidetti, personal communication) could also make them more likely to be observed in surveys.

The effect of immigration, however, is likely to be short lived. As densities within reserves grow and exceed those of surrounding areas, emigration due to density-dependent processes should eventually exceed immigration; indeed, an important function of reserves for fisheries management is to provide spillover of adults to neighbouring fishing grounds (McClanahan & Kaunda-Arara 1996; Roberts et al. 2001; Kaunda-Arara & Rose 2004; Ashworth & Ormond 2005). In concert with short-term immigration effects, the temporary displacement of fishers to neighbouring areas (‘fishing the line’, McClanahan & Kaunda-Arara 1996; Goñi et al. 2006; Kellner et al. 2007) could lower fish densities outside reserves, leading to an apparent protection benefit when measured with the response ratio inline image.

The mechanisms leading to long-term benefits of reserves are likely to differ from those driving short-term responses. In our study, most fish species were relatively more abundant in old reserves. Recruitment is more likely to explain the long-term benefit of protection. The dynamics of larval recruitment into reserves are poorly understood despite their potential importance in population recovery (Watson & Munro 2004; Barrett et al. 2007). Recruitment rates of many fishes are highly variable (Sponaugle & Cowen 1996) and characterized by pulses (Warner & Chesson 1985). Long durations of protection may therefore be required for reserves to experience one or more recruitment episodes that effectively translate into strong year classes and drive recovery of populations (Secor 2000).

We had predicted that small species should exhibit larger increases in new reserves than large species, owing to their small home ranges (Kramer & Chapman 1999) and potential for rapid population increase (Jennings 2001). This was not the case but small species did benefit from protection in old reserves. The fact that small species do not recover faster than large species implies that life-history and ecological correlates of population recovery are less important than the known effects of fishing status (Mosqueira et al. 2000; Micheli et al. 2004; Claudet et al. 2008). It is possible that the subtle effects of ecology and life history may be detectable with more data or in analyses of more homogeneous sets of reserves (e.g. Claudet et al. 2008), both of which would overcome the noise generated by variation in, for example, management regimes and socio-political context.

Our last prediction concerned trophic cascades. We expected a low effectiveness of older marine reserves at protecting small (prey) species, if (large) species at high trophic levels had recovered in these reserves. Contrary to this prediction, smaller species benefited in old reserves despite increases in the densities of their potential predators (i.e. large, fished species). Furthermore, small, non-fished species exhibited variable responses across reserve-age classes. Responses by low trophic-level species to protection may be complex and involve interactions among community dynamics (Díaz et al. 2005; Barrett et al. 2007), habitat (McClanahan & Arthur 2001) and metapopulation dynamics (Man et al. 1995). They are therefore difficult to predict.

While our predictions focused on species size-related effects of protection, another important potential effect of marine reserves is alteration of within-species size structure, usually by increasing the abundance of large individuals. We did not have data on population size structures within reserves. However, it is likely that increases in the abundance of large-bodied species occurred in concert with increases in within-species body size (e.g. Halpern 2003). While the former can re-establish lost trophic interactions and restore the ecological integrity of communities (Guidetti 2006), the latter may have more direct consequences for population recovery within and beyond reserves. For example, the association between body size and fecundity may increase recruitment within the reserve itself and to neighbouring fishing grounds (e.g. Evans et al. 2008). Alternatively, the association between body size and home range size may promote spillover of adults into neighbouring sites (Kramer & Chapman 1999).

In summary, three key insights into the management of marine reserves emerge from our study. First, the overall effectiveness of marine reserves at enhancing fish densities improves over time. Although some reserves may be effective quickly, most reserves require more than a decade to yield significant benefits. Secondly, reserves may provide benefits in the long run, irrespective of their early performance. Finally, given enough time fish species that are locally exploited do recover in reserves. Reserve stakeholders therefore need to have realistic expectations and management plans need to incorporate short-term uncertainty and a long-term perspective.


S. Jennings, G. Edgar, N. Barrett, E. Macpherson, A. Gordoa, A. García-Rubies, C. Denny, T. Willis, R. Babcock, J. Claudet, D. Pelletier, J.-Y. Jouvenel, F. Bachet and R. Galzin for contributing raw data to earlier versions of our data base; M. Fish for assistance producing Figure S1. Funding was provided by the Leverhulme Trust and Canadian Bureau for International Education to PPM and the Natural Sciences and Engineering Research Council (NSERC) of Canada to IMC. IBM was supported by an NSERC Undergraduate Summer Research Assistantship.