How does the accuracy of fisher knowledge affect seahorse conservation status?

Authors


Correspondence
Kerrie P. O'Donnell, Project Seahorse, Fisheries Centre, University of British Columbia, 2202 Main Mall, Vancouver, BC V6T1Z4, Canada. Tel: +604 827 5137
Email: odonnell@zoology.ubc.ca

Abstract

Despite a growing interest in incorporating fisher knowledge into quantitative conservation assessments, there remain practical impediments to its use. In particular, there is some debate about the accuracy of fisher knowledge. In this study, we report an attempt to quantify assumptions about how accurately fishers report past events (retrospective bias). Then we examine how the assumption we make about retrospective bias affects the characterization of changes in the fishery and extinction risk. We link fisher interviews and fisher logbooks to establish a catch rate (catch per unit of effort) trend for the history of a data-poor, small-scale seahorse fishery in the Philippines. We find that fishers perceive historic declines in fishing rate that are not apparent in more recent logbook trends, and the extent of the decline (and therefore extinction risk) hinges on assumptions we make about the accuracy of fisher recall. Scenarios that ignore retrospective bias result in the most severe declines and the most worrying extinction risk classifications. Furthermore, the historic baseline set by interviews suggests that relying on recent decades of data alone may underestimate extinction risk for our study species, and others that have been historically exploited. Attempting to link interviews with logbooks also illustrates differences between fisher-derived datasets: retrospective interviews may exaggerate early fishing rates and capture less variability than logbooks. In addition to being the first seahorse fishery reconstruction, our work contributes to the emerging interest in how fisher knowledge can guide conservation assessment. Future studies that incorporate fisher knowledge into quantitative assessments require (1) clearly stated assumptions about fisher knowledge bias; (2) clear criteria to compare fisher knowledge collected with different methods; (3) evaluation of the impact of assumptions on assessments.

Introduction

Historic baselines for the world's ocean fisheries are largely unknown (Pauly, 1995) and it is generally impractical to gather basic information on the many unstudied regions (Johannes, 1998). Formal scientific data required for managing many marine resources have not been gathered. Moreover, the large scale of most exploited areas means that controlled and replicated experiments are impossible (Ludwig, Hilborn & Walters, 1993). Catch rates [catch per unit of effort (CPUE)] have been used in formal fishery assessments as proxies of changing fish abundance, but have only been systematically collected for commercially important industrial fisheries (FAO, 2009). Small-scale fisheries that employ 98% of the world's fishers (Berkes et al., 2001) and land more than half the world's annual marine catch (Davy, 2000) are poorly understood because there is little formal research or monitoring of exploitation.

Careful accumulation of fishers' knowledge is beginning to fill the gaps in our quantitative understanding of the threatened status of marine species and ecosystems. We know from government reports, local knowledge and fisher logs that the Chinese bahaba Bahaba taipingensis and giant humphead parrot fish Bolbometopon muricatum have been fished to the brink of extinction (Sadovy et al., 2003; Dulvy & Polunin, 2004) and current cod stocks Gadus morhua on Canada's Scotian Shelf are a small fraction of their historic numbers (Rosenberg et al., 2005). In Indonesia, analysis of local knowledge suggests an ecosystem-wide decline of exploited coral reef species (Ainsworth, Pitcher & Rotinsulu, 2008).

Despite a growing interest in incorporating fisher knowledge into quantitative fisheries and conservation assessments, there remain practical impediments to its use. When compared directly, fisher and scientific knowledge have been found to both agree (e.g. Neis et al., 1999; Begossi, 2008; Lozano-Montes, Pitcher & Haggan, 2008) and disagree (e.g. Otero et al., 2005; Daw, 2008), raising concerns about the accuracy of fisher knowledge. Work reviewed by Daw (2008) illustrates that it is easier to recall events that are unusual or rare, especially if they were positive (e.g. a particularly lucrative fishing excursion), than other less striking circumstances (Bradburn, Rips & Shevell, 1987; Tourangeau, 2000). Rare, positive events may thus be overemphasized in retrospective interview datasets. ‘Retrospective bias’ has been recognized in recording historic fisher perceptions (Neis et al., 1999; Ainsworth et al., 2008), but to our knowledge, there are no reports on how different assumptions of bias affect quantitative assessments.

We report our efforts to address retrospective bias in CPUE data for a small-scale, data-poor seahorse fishery in the Philippines. Our work had three objectives: (1) reconstruct CPUE using fisher interviews and daily logbooks; (2) consider how assumptions about the accuracy of fisher recall affect the assessment of conservation status for that species; (3) compare current and past conservation status. We do not test the reliability of interviews by directly comparing them with logbooks because the datasets did not overlap in time; instead, we explore simple calculations of CPUE for logbooks that might account for recall bias in interviews, allowing us to link interview and logbook data into one time series.

Seahorses are collected as a cash commodity for export as non-food items, either when dried for use in traditional medicines and as curiosities or live for the aquarium trade. They have been traded internationally for decades, but we lack understanding of the status of most populations; hence, we turn to fisher knowledge. Almost all seahorses are caught in small-scale fisheries that are not formally monitored or managed, but trade analyses led to estimates of at least 20 million dried animals per year in 1995 (Vincent, 1996). Furthermore, all qualitative indicators of overfishing reveal that seahorses are overfished (Vincent, 1996; Martin-Smith et al., 2004). The IUCN Red List (IUCN, 2008a) classifies extinction risk for 33 seahorse species: seven as Vulnerable (including our study species Hippocampus comes), one as Endangered and 25 as Data Deficient (i.e. having insufficient data to be classified).

The Philippines has many unmanaged and unrecorded fisheries and has historically been a dominant source of seahorses in international trade. Almost all of the millions of seahorses exported (Vincent, 1996) are obtained in the small-scale fisheries that engage 90% of Filipino fishers (Luna et al. 2004). Exports have been recently curtailed by domestic legislation that has inadvertently (as a consequence of international trade restrictions) prohibited all seahorse exploitation since 2004. Nonetheless, domestically illegal fishing continues because there is little enforcement, seahorses command a high price on the internationally legal market and fishers have few alternative sources of income. Quantifying the extent of declining seahorse catch rates is one way to estimate key parameters needed to evaluate effects of possible recovery options.

Methods

Study area, fishery and target species

Our study takes place in the Danajon Bank region of the Central Philippines (Fig. 1). Destructive fishing has degraded this once-rich coral reef system to the point where marine life is grossly depleted. The people living in the region's coastal communities are among the poorest in the Philippines. In 2002, 80% of fishing households in Northern Bohol fell below the poverty threshold (Green et al., 2003). Our focal community, Handumon, has the largest concentration of lantern fishers who catch the most seahorses in the region (Meeuwig et al., 2003). Project Seahorse has been working with the Handumon community since 1996 as part of a larger initiative to empower them to manage their marine resources sustainably.

Figure 1.

 Danajon Bank study region of the central Philippines and our focal lantern fishing community of Handumon.

Although the lantern fishery is the most extensively studied seahorse fishery, it is understood solely through fisher-derived knowledge and trade records. Lantern fishers swim at night under a kerosene lantern, and free-dive for seahorses, other finfish and invertebrates. This open-access, small-scale, multi-species fishery has not been formally monitored or managed, but fishers and traders have reported declines in seahorse availability (Vincent, 1996; Pajaro et al., 1997), increasing demand, price and effort, suggesting overexploitation (Martin-Smith et al., 2004). Analysis of the recent fishery suggests that current CPUE is seasonal and much lower than the anecdotal fisher reports of historic CPUE (Vincent et al., 2007). The historic anecdotes and magnitude of the decline have not yet been assessed thoroughly.

The tiger-tail seahorse H. comes Cantor, 1850, is the most frequently caught seahorse species in lantern fishery (Vincent, 2007). Hippocampus comes is found at night on coral reefs, sponge and seaweed habitats (Perante et al., 2002). It is listed as Vulnerable on the IUCN Red List (IUCN, 2008a) and studies of its life history, abundance and distribution suggest that it fares poorly under high levels of fishing pressure (Morgan & Lourie, 2006; Morgan & Vincent, 2007).

Analysis

To reconstruct the history of H. comes exploitation by lantern fishers, we searched Project Seahorse unpublished data and reports. The only record of the earliest years of the fishery was contained in retrospective fisher interviews, whereas catch rates for more recent years were recorded in daily fisher logbooks. Because these two datasets do not overlap in time, we cannot be sure as to how comparable they are. Interview data may be biased toward maximum catch nights, and so we tried a number of calculations that could correct for retrospective bias linking interviews to logbooks as a single trend.

Although CPUE data may not be a reliable measure of changes in exploited populations (e.g. Harley, Myers & Dunn, 2001), we used it because (1) it is the only metric of historic change available; (2) previous work suggested that it may be a robust indicator of H. comes population size (Vincent et al., 2007); (3) CPUE is an accepted proxy for abundance for IUCN Red Listing (IUCN, 2008b).

Interview CPUE (1970s to 1994) data collection and analysis

From March to July 1995, Project Seahorse conducted extensive semi-structured interviews to document fisher understanding of the biology, fishery, trade, conservation and management of local seahorses. The second author, a Filipina, conducted interviews with 21 of the approximately 35 seahorse fishers in Handumon in the local Cebuano or Filipino language. All fishers were invited to be interviewed; all who were willing and available were included. Their individual fishing experience ranged from 6 months to 24 years.

From these interviews, we extracted only respondent biographical information and answers to three structured questions focused on understanding fishers' perceptions of historic change in the seahorse fishery between 1969 and 1993 (Table 1). Fishers were asked if CPUE had increased or decreased in recent years (Q1) and completed trend diagrams (Deguit et al., 2004) representing relative changes in CPUE over their career (Q2). Fishers were also asked to recall a typical catch night−1 for a few specific years (Q3). Years were chosen to correspond with events that might aid fishers' recall, for example change of presidents, large typhoons or something personal like the birth of a child (Means & Loftus, 1991). The number of responses to Q3 varied by time period (mean±1 sd: 12±7), with fewer fishers available and able to recall earliest catches than for more recent years. Our best estimate of the total number of seahorse fishers in the first decade of the fishery (roughly the 1970s) was between five and 10 individuals; hence. the two fishers interviewed comprise between 20 and 40% of all those who were fishing.

Table 1.   Interview schedule
Biographical information: name, age, number years fishing, number years fishing seahorses.
Q1. Has the number of seahorses you catch per night increased or decreased over the past 2 years? Over the past 5 years?
Q2. On this timeline (where we have established the beginning, middle and most recent year of your seahorse fishing career) please mark a number of X's at each time period to represent relative change in CPUE.
Q3. What was the usual number of seahorses you caught per night in: 1994, 1990, 1980s and 1970s?

To convert each fisher's trend diagram (Q2) into a relative percent change in career CPUE, we totaled the number of X's for the initial year of a career (CPUEinitial) and the most recent year of career (CPUEfinal). We then calculated the change in career CPUE as (CPUEfinal−CPUEinitial)/CPUEinitial. We analyzed typical catch night−1 (Q3) raw interview transcripts by standardizing responses, compressing years into bins and then calculating two CPUE metrics corresponding to different assumptions about retrospective bias.

Interview median

With this metric, we assumed that when fishers were asked about ‘typical’ CPUE, they reported a measure of center, with some error. We choose a median, rather than a mean, because distributions of counts (number of seahorses) tend to be skewed, making the median a more suitable measure of central tendency. Fishers differed in how they provided ‘typical’ CPUE: some provided a range while others provided a single value. To standardize responses, when fishers responded with a range, we took the median value; when they provided a single value, we used that. We then grouped years into bins that were larger for earlier time periods. Because our goal was to look at long-term trends and we assumed fishers' ability to recall precisely which year a CPUE occurred would be worse for dates farther back in time (Bradburn, 2000), we grouped responses into four categories: 1994, 1990, during the 1980s and during the 1970s. We then calculated the median of all fishers' CPUEs in a given time period to produce a single CPUE value for each time period.

Interview max

With this metric, we assume that fishers consistently over-estimated CPUE. We chose only the highest CPUE for each time period as the single yearly maximum CPUE.

Logbook CPUE (1996–2003) data collection and analysis

From 1996 to 2003, fishers worked with Project Seahorse to record daily seahorse catch in logbooks. Logbooks were completed by 43 fishers between 1996 and 2003 with a mean participation rate in the logbook program of 16±5 (1 sd) fishers year−1 (Vincent et al., 2007). To minimize bias, we present results only for years that had equal sampling during peak and lean seasons. We then carried out three calculations to explore how we might ‘correct for’ biases in interview data allowing us to link interview and logbook data into a single trend:

Logbook median

For each fisher, we calculated the median catch night−1 for each year from the logbook data. We then took the median of these catches per fisher to produce a single CPUE value for each year.

Logbook median no zeros

We repeated the calculations used for logbook median, but removed nights on which zero seahorses had been caught.

Logbook max

We took the highest catch night−1 for each year as the single yearly maximum.

Evaluating extinction risk

Drawing on CPUE trends calculated from interview and logbook data, we evaluated the extinction threat for H. comes using the IUCN Red List (http://http://www.redlist.org) criteria (IUCN, 2001). We calculated the percent decline in interview and logbook CPUE from the start to the end of three time periods, the most recent of which (1990–2001) spanned interview and logbook datasets. We compared results with IUCN Criteria to assign extinction risk. We focused on Criterion A, which measures the ‘observed, estimated, inferred or suspected’ decline in the number of mature individuals over a 10-year period or three generations, whichever is longer. Information needed to assess the exact generation time as prescribed by the IUCN (2008b) is unavailable for H. comes. Therefore, we chose the default 10-year time period because it is longer than three times the best estimate of maximum lifespan (Morgan, 2008) for this species.

We were particularly concerned with Criterion A2, in which the causes of decline may not have ceased, may not be understood or may not be reversible. In these contexts, a decline of >30% would see a species listed as Vulnerable, >50% as Endangered and >80% as Critically Endangered. Because we were unable to ascertain the number of ‘mature individuals’ over the entire history of the fishery, we used the tally of all fish. Limited to the years with data available for this calculation, our time periods were either slightly less or more than 10 years, and the final year of one time period became the starting point of the subsequent time period.

Results

Interview CPUE (1970s to 1994)

Fishers consistently reported declines in CPUE. All fishers said that their CPUE in the past 2 years and past five years had declined. All changes in career CPUE were also declines, ranging from 25 to 80% (mean 67%), with the maximum decline reported by a fisher who had been fishing since the 1970s. When asked specifically to recall numbers, fishers reported collecting a median CPUE of 150  seahorses fisher−1 night−1 at the outset of the market for seahorses in 1969. This had declined to 10 seahorses fisher−1 night−1 by 1994, a 93% reduction. Max CPUE followed a similar trend, declining from 200 to 50 seahorses fisher−1 night−1 (75% decline) over the same time period (Fig. 2). One fisher mentioned that he was motivated to start catching seahorses because two other fishers could catch more than 100 seahorses night−1 in the 1970s and early 1980s. Only five fishers reported having any zero seahorse catch nights.

Figure 2.

 All CPUE trend calculations for interview data (maximum: ▴, median: ▪) and logbook data (maximum: ▵, median no zeros: □, median: ⋄). Arrow indicates break between interview (1970–1994) and logbook (1996–2003) datasets. All calculations are in units of seahorses fisher−1 night−1. Error bars are 25th and 75th percentiles for median values calculated for all medians, but large enough only to be visible on the graph for interview medians.

Logbook CPUE (1996–2003)

From fisher logbooks, we recorded a total of 31 381 H. comes caught over 11 179 nights of fishing from 1996 to 2003, with catches ranging from 0–80 seahorses night−1. We observed no clear trend in CPUE over the course of the logbook program (Fig. 2), although there was variability among years. This was true of all metrics, with or without zero catches, although absolute values varied according to the metric. Logbook data showed that catch night−1 was highly skewed, with no seahorses on nearly 70% of fishing nights (Fig. 3a). Medians were reduced as a result of the high number of low catch nights, but there were also nights with large catches, in all years and for all fishers (Fig. 3b).

Figure 3.

 Frequency (a) and distribution (b) of catch night−1 logbook data. Box plot shows 25th, 75th percentile and median; whiskers are 90th percentiles; hollow circles are outliers.

Entire fishery CPUE trend (1970s to 2003)

Overall reconstructed CPUE showed a fishery that experienced a severe historic decline with more recently low and variable CPUE. Our assumption about fisher recall bias, and therefore our choice of CPUE calculation, affected the inferred magnitude of decline in catch rate. Even the most optimistic calculation, using maxima, yielded a 77% decline in CPUE from 1970 to 2003. When we assumed that interviews overestimated historic CPUE, we chose logbook calculations based on maxima or medians that excluded zero catch nights (Fig. 2). This choice revealed a fishery that was in decline from 1970 to 1993, but has been stabilizing at lower levels since 1996. When we assumed that interviews accurately portrayed historic CPUE, we chose the logbook median CPUE calculation that included zero catch nights. This choice of CPUE calculation suggested that the fishery had collapsed rather comprehensively.

Evaluating extinction risk

Extinction risk classification depended on the assumed reliability of fisher recall and the time period of data used. Different assumptions about recall bias, calculated as a means of linking interview and logbook data, translated into a wide range of extinction risk classifications for the most recent decade (1990–2001: Table 2). For example, when we assumed that interviews overestimated historic CPUE and then calculated a metric that might make logbook comparisons valid (logbook max), we concluded that CPUE was increasing and, therefore, this species should be classified as Least Concern. On the other hand, if we assumed that interviews and logbooks were comparable without ‘correction’, and calculated a median to summarize their central tendencies (interview median plus Logbook median), we concluded that CPUE had declined by 97% during the 1990s and that H. comes should be listed as Critically Endangered.

Table 2.   Extinction risk categorization for IUCN Red List criteria by time period and CPUE calculation: CR=Critically Endangered, EN=Endangered, VU=Vulnerable, LC=Least Concern
CPUE calculationTime period
InterviewLogbook1970–19801980–19901990–2001
  1. *VU indicates current Red List classification.

Max ENEN 
Median ENEN 
MaxMax  LC
MedianMedian no zeros  VU*
MedianMedian with zeros  CR

When we compared historic with current extinction risk for H. comes, we found that for earlier decades, extinction risk was consistently more worrying. Declines between 50 and 80% justified the classification of Endangered for both 1970–1980 and 1980–1990 (Table 2), regardless of how accurate we assumed interviews to be. In contrast, since 2002, H. comes has been listed only as Vulnerable on the IUCN Red List (2008a).

Discussion

Handumon lantern fisher interviews tell the story of a seahorse fishery that underwent severe historic decline. The severity of that decline (and therefore extinction risk) hinges on assumptions about the accuracy of fisher recall. Fisher interviews set a historic baseline, which suggests that H. comes, and therefore other historically exploited species, may be more at risk than current Red Listing reflects.

Fishers perceived historic declines in CPUE that were not apparent in more recent logbook trends. Fisher interviews provided the only record of historic catches and suggested that CPUE had declined from 75 to 93% by 1994. Without fisher interviews, we might have concluded that recent low CPUE from logbooks was also typical in the past. It is clear from the consistent declines reported by all fishers and across triangulated questions that all fishers perceived a decline. Declining catch rates revealed by interviews are similar for other marine resources in Danajon Bank (Green et al., 2004) and other fishery declines in the Philippines (Stobutzki et al., 2006). Fishers' concerns over dwindling seahorse CPUE catalyzed the establishment of a community-managed no-fishing zone in the hopes of recovering seahorses and other target species (Pajaro et al., 1997). Certainly, any management of seahorses should consider declines perceived by fishers to avoid the shifting baseline syndrome (Pauly, 1995).

How perceived historic declines relate to recent logbook trends depended on assumptions we made about fisher recall bias. Nights on which fishers caught zero seahorses dominated logbook data (Fig. 3a), but were mostly absent from interviews, making us wonder why fishers did not report zero catch nights. It is possible that if seahorses were more abundant historically, fishers caught them every night, but it is also likely that fisher reports were biased toward positive rare events (Bradburn et al., 1987; Tourangeau, 2000). Combined with the possibility that fishers may err toward pessimism (Ainsworth & Pitcher, 2005), fisher interviews likely tend to exaggerate the perceived decline. Although there are no standard methods for quantifying retrospective bias, our calculations show that if we assumed that fishers overestimated historic CPUE, the decline to more recent CPUE in logbooks was less severe (interview max with logbook max: Fig. 2). On the other hand, if we assumed fisher historic interviews were accurate, the decline to present was more drastic (interview median with logbook median: Fig. 2).

In addition to recall bias, interviews and logbooks generated data with different characteristics, requiring careful consideration during analysis. Logbooks captured more variability than interviews, which influenced conservation assessments. Logbook data had more samples per time than interview data: more nights in a year and more years in a row. It is then not surprising that logbooks captured a wider spectrum of events from common to rare, zero to many seahorses caught, respectively. By extension, extinction risk calculated from logbooks (1990–2001) spanned a wider range than when calculated only from interview data (1970–1980 and 1980–1990 Table 1).

Despite uncertainties about how best to link interviews and logbooks, our analysis suggests that our focal population of H. comes may be more at risk of extinction than its current listing of Vulnerable. Our uncertainty about the compatibility of interviews and logbooks placed our Red List assessment for the period 1990–2001 between Least Concern and Critically Endangered (Table 2). Least Concern is an unrealistic classification because it is based on a single max CPUE value per year, probably a rare value and possibly an outlier. Hippocampus comes should therefore be classified at least as Vulnerable, but possibly as at risk as Critically Endangered (CR). In situations where the spread of plausible values qualifies a taxon for two or more categories of threat, the IUCN recommends adopting a precautionary, but realistic, approach and choosing a more threatened category (IUCN, 2008b). In our study, the more threatened choice of CR is based on data, which assumes that historic CPUE from interviews was not overestimated relative to logbooks. Clarification of this classification requires interviews that span the entire history of the fishery. The current listing of Vulnerable (IUCN, 2008a) is supported by our calculations that account for retrospective bias (Median interview and Median no zeros: Table 2), but may underestimate the actual threat level, especially if the biggest declines occurred before 1990.

For species like H. comes, that have undergone historic exploitation, our results suggest that relying on recent decades of data alone may underestimate extinction risk. Ignoring historic data, the current IUCN Red List assessment is Vulnerable (IUCN, 2008a). In contrast, assessment of the earliest data qualifies this species as Endangered. While a similar pattern has been found for Gulf grouper (Mycteroperca jordani; Sáenz-Arroyo et al., 2005), the opposite pattern has been found for amphibians (Stuart et al., 2004). For marine species that have been impacted by fishing for at least hundreds of years (Jackson et al., 2001), assessment of contemporary conservation and fishery status must be conducted relative to historic baselines (Pauly, 1995). Our analysis highlights a potential limitation to the Red Listing process. Because the default time frame for assessment is only a 10-year window, there is little capacity for incorporating long-term changes for relatively short-lived species. Capturing historic changes is especially important for species that have undergone historic declines. This type of retrospective threat classification is uncommon, but is becoming possible by incorporating fisher knowledge.

We have focused on examining retrospective bias to improve methods for incorporating fisher knowledge into quantitative assessments, but other questions about the magnitude of decline have not yet been answered. It is perhaps surprising that despite severe perceived declines in CPUE, there are still seahorses to be caught. Certainly, the flexible, multi-species nature of the fishery allows lantern fishers to continue taking seahorses whenever they see them (Vincent et al., 2007) and establishment of no-fishing zones may allow for un-fished individuals to seed fishing grounds. In addition to retrospective bias, another way by which declines may be overestimated is that some deep-water-dwelling seahorses are less vulnerable to fishing because lantern fishers fish only in relatively shallow water. If deeper seahorses were not being fished, fishers would perceive their reachable stocks to be declining at a greater rate than the overall population. This bias applies to scientifically gathered CPUE data as well and is known as hyperdepletion. We cannot assess this problem as yet due to lack of data.

Our study that links two historic fisher-derived datasets can inform management, prioritize future research and provide insight into how we might tackle challenges of integrating fisher knowledge into quantitative assessments. Fishers perceive severe historic declines in seahorse catch rates, which means seahorses may be more endangered than we think because current Red listing does not incorporate historic status. The success of future management and avoidance of the shifting baseline syndrome requires a better understanding of fisher perceptions of the decline. Before our CPUE reconstruction can be used to evaluate effects of possible management options, we will need to clarify the magnitude of interview bias and improve our understanding of how CPUE relates to H. comes abundance. Future studies that incorporate fisher knowledge into quantitative assessments require (1) clearly stated assumptions about fisher knowledge bias; (2) clear criteria to compare fisher knowledge collected with different methods; (3) evaluation of the impact of assumptions on assessments.

Acknowledgments

This is a contribution from Project Seahorse. We sincerely thank the fishers of Handumon, the village captain, council and citizens for their generous and kind support. We also thank the current and past teams of researchers at Project Seahorse Foundation (Philippines) for their assistance. An earlier version of the paper was substantially improved by comments from Philip Molloy and Iain Taylor and two anonymous reviewers. This work was generously supported by the People's Trust for Endangered Species (UK). K.P.O. was funded by a National Science Foundation Graduate Research Fellowship. Further support came from Guylian Chocolates Belgium and the John G. Shedd Aquarium through their partnerships for marine conservation with Project Seahorse.

Ancillary