The Case of the Manus Green Tree Snail Papustyla Pulcherrima

Harnessing local ecological knowledge for conservation decision making via Wisdom of Crowds: the case of the Manus green tree snail Papustyla pulcherrima

N ATHAN W HITMORE

Abstract The shell of the Manus green tree snail Papustyla conservation intervention (e.g. Weitzman, ; Bottrill pulcherrima is renowned for its beauty and is subject to et al., ; Wilson et al., ). Implicit in such methods international protection under CITES, having been har- is an understanding of the level of crisis and the related dri- vested intensively in the past. To determine its threat status, vers affecting the species in question. However, such threat and whether further conservation action is justified, an in- assessments incur costs, and in the case of poorly under- expensive Wisdom of Crowds approach was used to esti- stood or cryptic species, such as invertebrates, the cost of mate the change in relative density of the snail between even a superficial in situ assessment may exceed the funding  and . Local men and women were approached available. around the main market on Manus Island, Papua New Species-specific conservation programmes predominantly Guinea, and asked to map the relative abundance of the target vertebrates, with only a small fraction of the available snail on an ordinal scale, based on their personal observa- funding being assigned to invertebrate conservation (Lydeard tions in  and  (a year of cultural significance). et al., ; Cardoso et al., ). Whereas most vertebrates The spatial abundance data from  surveys were analysed have been categorized on the IUCN Red List (IUCN, ), using an information-theoretic approach. A suite of cumu- , .% of all described arthropods and , % of all described lative link models incorporating geographical factors was molluscs have been categorized (Cardoso et al., ). Given used to determine the magnitude of the change and to inves- the scarcity of funding for invertebrate conservation, there is a tigate possible biological influences underpinning the re- need to allocate available funds in a prudent manner. ported pattern. High abundance of the snail was Researchers, especially in fisheries, are recognizing the associated with intact forested areas, high elevation and potential of local ecological knowledge to address knowl- low population density. A slow decline was evident, with edge gaps regarding the status of natural resources and the median percentage of map cells where the snail was ca- their management (e.g. Dulvy & Polunin, ; Sáenz– tegorized as plentiful decreasing by c. % between the  Arroyo et al., ; Arlinghaus & Krause, ; Cano & years. On this basis a categorization of Near Threatened Tellería, ). Assessing such knowledge using a Wisdom was advocated for the species. Although it is arguable that of Crowds methodology is one potential way to produce Wisdom of Crowds methods cannot be substituted for in cost-efficient estimates for threat status assignment and tri- situ quantification, the approach appears to have utility as age assessments. Wisdom of Crowds is a mechanism by a preliminary assessment for further conservation expend- which the aggregated knowledge of a large number of indi- iture, and as a tool for determining threat status. viduals is used to generate quantitative estimates. It was de- monstrated over a century ago by Francis Galton (Galton, Keywords Cumulative link model, Data Deficient, decision ), and popularized as a named concept by Surowiecki making, local ecological knowledge, Papustyla pulcherrima, (). By aggregating information from a large sample threat status, triage, Wisdom of Crowds population the idiosyncratic noise of individuals is dimin- ished (Welsh, ). Often the resulting estimate, including those of quantity estimation and spatial reasoning, has been shown to be surprisingly accurate (Galton, ; Surowiecki, Introduction ; Arlinghaus & Krause, ). uring the past  decades increasing attention has been Surowiecki () outlined four basic criteria for Wisdom Dgiven to triage methods to prioritize species for of Crowds, which can be summarized as () diversity of opinion (i.e. individual interpretations are not censored), () independence (i.e. the opinion of an individual is not influenced NATHAN WHITMORE Wildlife Conservation Society, Papua New Guinea by others), () decentralization (i.e. individuals are able to Programme, P.O. Box 277, Goroka, Eastern Highlands Province, Papua New Guinea. E-mail [email protected] draw conclusions based on their local knowledge), and () ag- Received  November . Revision requested  March . gregation (i.e. there is a mechanism for translating private Accepted  April . judgements into a collective estimate).

Although the application of the methodology is relatively Methods new, Wisdom of Crowds is being investigated in a variety of fields, including politics (e.g. Bosler & Schneider, ), eco- NW and four community facilitators (local Manusian staff nomics (Nofer & Hinz, ), computer science (e.g. trained in community engagement and conservation) sur- Welinder et al., ) and sociology (Lorenz et al., ). veyed adult market-goers, stall holders and local residents Although it has been proposed as a mechanism to assist at random during – September  (in a manner in the management of fisheries in data-poor situations accordance with the ethical standards of the British (Arlinghaus & Krause, ), it has yet to become an estab- Sociological Association). The survey began  days after lished research method underpinning conservation and Papua New Guinea Independence Day, and market activity natural resource management initiatives. appeared to be typical over the survey period. The Manus green tree snail Papustyla pulcherrima is a Each respondent was given two maps of Manus Island forest-dwelling arboreal snail found only on Manus and adjacent islands, overlain with an indexed grid of  Island, a , km densely forested island to the north of , -km map cells marked with village aid posts, town- Papua New Guinea. Its conical, emerald green shell, with ships and roads. Respondents were asked to mark map cells a fine yellow border to the whorl, and a white lip, attains a in which they had observed and were confident of the abun- maximum length of  cm and is renowned for its beauty dance of Manus green tree snails, using a five-point ordinal (Pitman, ). After the snail’s existence was popularized scale: () not found, () previously found but now absent, () internationally in the s demand for its shell grew, par- scarce, () moderate numbers, () plentiful. Community fa- ticularly for jewellery, which resulted in the export of large cilitators assisted with interpretation of the map and com- numbers of shells, and concerns about possible extinction of pletion of a form. On one map respondents recorded their the species (Pitman, ). This resulted in the snail being observations of snail distribution and abundance in , the first invertebrate to be listed on the Endangered and on the other they recorded the same information Species Act of  (U.S. Government, ). International based on their recollections from  (the World Sea trade has been controlled by export permit since  under Kayaking Championships were held at Manus Island in CITES Appendix II (UNEP–WCMC, ) but the snail still , and therefore this year had special significance in faces potential threats from overexploitation for legal do- local memory). Only respondents that were  years or mestic trade (with the shells used for adornment), and habi- older in  were asked to complete the  map. Sites tat loss (Shearman et al., , estimated that c. % of the of known extinction were differentiated from sites of natural forest on Manus Island had been cleared or degraded during absence (e.g. because of habitat preference), but for pur- –). As of May  the snail was still categorized as poses of modelling between years a single category repre- Data Deficient on the IUCN Red List (Mollusc Specialist senting absence was used. Unless the surveyor understood Group, ), with its extinction risk unclear. the local language of the respondent (there are  indigen- The majority of resident Manusians are indigenous and ous languages spoken in Manus Province) interviews were retain a subsistence lifestyle (Hide et al., ), and thus re- conducted in Tok Pisin. present a repository of local ecological knowledge. The ob- Name, age, sex and home village were recorded for each jective of this study was to utilize a Wisdom of Crowds respondent, as well as the name of the snail in their local lan- approach to local ecological knowledge to ascertain tem- guage. All maps were checked for mistakes associated with poral changes in the distribution and relative abundance duplication (for people with the same name), age threshold, of the Manus green tree snail, assess the need for further and form-filling, and indications of influence or interfer- conservation efforts, and explore the utility of the Wisdom ence by a third party. There is no recommended threshold of Crowds technique. for the sample size required for Wisdom of Crowds methods but basic principles suggest that the sample should be suffi- ciently large to diminish idiosyncratic noise. I estimated we would need a minimum sample size of  respondents, and Study area aimed to maximize the sample within the constraints of our The study took place in Lorengau, the provincial capital of budget. We conducted a total of  surveys, of which  Manus Province, and the only major town on the island of were deemed valid following exclusions for third-party Manus, Papua New Guinea (Fig. ). Surveys were carried out interference, duplication, and incomplete forms. in and around the central covered market, which operated In addition to the mapping exercise  structured inter- daily. This location was chosen because of the opportunity views with open-ended questions were used to gather quali- to survey people from diverse locations across Manus Island tative information on aspects of the snail’s life history, and the neighbouring offshore islands. As such, it was commercial value and temporal population trends from a deemed to be broadly cross-sectional and representative of random subsample of the respondents (pertinent metadata the wider community. are reported here). To estimate the sales turnover from

http://journals.cambridge.org Downloaded: 26 Jan 2016 IP address: 69.74.254.25 Manus green tree snail 3

FIG. 1 Manus Island, Papua New Guinea, where surveys were carried out in the town of Lorengau.

snail shells, a community facilitator recorded the number of times a grid cell was reported on; it was considered to be a shells on sale and sold each day during – September . proxy of human population density. Biogeography (biog) Data analysis was carried out in R v. . (R Development was a categorical variable representing mainland or offshore Core Team, ), with additional functionality provided by islands (for the purposes of this study the island of Los various packages. Geographical forest cover data (percent- Negros, which is separated from Manus Island by a narrow age tree cover in ; losses and gains during –) body of water, Lonui Passage, was considered to be part of were sourced from Global Forest Change tagged image files the mainland). Surveyor (surveyor) represented the individ- (TIFs) derived from Landsat data (Hansen et al., ), and ual effect of each interviewer. Period (period) represented a altitude data for Manus Province were sourced from the time effect distinguishing the year under assessment ( Papua New Guinea Resource Information System as or ). Nineteen candidate cumulative link models were MAPInfo files (Bryan & Shearman, ). Both datasets created using combinations of these explanatory variables, were converted into raster maps, in the Universal and tested using the package ordinal (Christensen, ). A Transverse Mercator projection, in R using the package ras- deliberately poor model (based on a singular intercept) was ter (Hijmans, ) in conjunction with rgdal (Bivand et al., run within the suite of candidate models to represent a base- ), and aggregated to the same resolution as the snail sur- line for poor performance. vey grid. Forest change data were converted into two layers, This suite of  candidate models was assessed in an infor- representing  and  coverage. The raster values cor- mation–theoretic model testing framework (sensu Burnham responding to the non-empty survey map cells were collated & Anderson, ) using the package AICcmodavg to form covariates within a single data frame. (Mazerolle, ). Predicted values for the top model (lowest The ordinal data were analysed using cumulative link Akaike information criterion with a small sample correction, models (also known as ordered logit regression). The re- AICc) were calculated using the package MASS (Venables & sponse variable from such models represents the accumu- Ripley, ). Based on a priori knowledge I expected that the lated log-odds of occurrence in or below a particular model coefficients would be positive for forestcov and alt, and ordinal abundance category (Agresti, ), and can be negative for period (i.e. declining) and sample (i.e. inversely used to calculate the predicted probabilities for each ordinal proportional to reporting density); I had no particular expec- abundance category in each of the , map cells. tations for biog and surveyor effects. Six explanatory variables were considered in the models: altitude, forest cover, sample, biogeography, period and surveyor. Altitude (alt) was a continuous variable, measured in Results metres. Forest cover (forestcov) was the canopy cover in  and , used as proxies for cover in  and , respect- The cells reported on by respondents broadly reflected the ively (which was not available from the Hansen et al., , known pattern of human population distribution (Fig. ): dataset), and was assigned a value between  (no cover) cells in the east of the island were heavily reported on, and  (% cover). Sample (sample) was the number of whereas elsewhere reports were primarily from areas around

relative abundance could be seen as a difference in the probability of assignment between ordinal abundance categories (Fig. ). It was apparent that the probability of a cell being assigned the highest ordinal abundance category (i.e. plentiful) declined between  and  and the probability that a cell would be assigned a lower category increased (Fig. ). A comparison of medians of the plentiful category suggested a decline from % of map cells in  to % in . Among the  respondents to the structured interviews there was a high rate of concurrence regarding where and when the snails could be found feeding (Table ). There was marginally more support for the perception that the snail population was increasing rather than decreasing. FIG. 2 Visualization of Manus green tree snail Papustyla pulcherrima abundance on Manus Island (Fig. ) as reported for Thirty-seven percent of respondents claimed they could  and . Cell colouration depicts abundance as a identify the sex of the snails, although members of the fam- continuous variable, under the assumption of equal distance ily Camaenidae are known to be hermaphroditic (Jordaens between ordinal abundance categories. et al., ). Over half of the interviewees reported selling shells at some time. coastal villages, and were most sparse in the vicinity of the Of  shells for sale at the Lorengau market over a -day central forest. Thus, unreported and less frequently reported period,  were sold, by seven vendors. Of these,  were grids probably coincide with areas less affected by human bought by a local hotel, presumably to replenish their stock activities. on sale for tourists. The asking price was determined by The top model was the most highly parameterized model size: PGK  for small shells and PGK . for large shells (period + alt + forestcov + biog + sample + surveyor;  para- (equivalent to USD . and ., respectively). One vendor meters) and had considerable support (model weight = %; specialized in snail shell jewellery and sold a variety of neck- Table ). Only one of the other candidate models had any laces and earrings for PGK – (equivalent to USD  and ). support: period + alt + forestcov + biog + surveyor (model He claimed to sell – items per week. weight = %). The most important parameters in the top model, in terms of magnitude, given the range of covariates were alt and forestcov (Table ). Discussion The parameter period indicated that relative abundance was consistently lower in  than in  (Table ). The The results of the Wisdom of Crowds method show con- support for the surveyor parameter in the top model gruity with my a priori expectations of model coefficient (Table ) and the relatively high surveyor coefficients direction: positive for forest cover and altitude, and nega- (Table ) indicate that respondents’ replies were influenced tive for period and sample reporting (as a proxy for by the surveyor. A comparison of the top model with and human population distribution). The support for the in- without the surveyor parameter, however, resulted in a com- clusion of multiple environmental parameters, developed paratively minor change in the estimation of the period par- a priori through preliminary knowledge of the snail’s biol- ameter (with: .; without: .), which suggests ogy and immediate threats, raises confidence that the re- − − that regardless of variation between surveyors, the effect of spondents’ reports have a reasonable ecological basis. period was real. When converted into an odds ratio the per- Within the suite of models there was preferential support iod effect coefficient estimated that the map cells were % for the inclusion of the parameter period (which appears more likely to be assessed in a lower ordinal category in  in the top two models and accounts for % of model than in . weight support). This indicates there was an additional ef- The predicted probability for any given map cell Y occur- fect (a decline) in the response that was not accounted for ring in an ordinal abundance category j (or below) could be by my choice of environmental variables. The cause of the calculated from the coefficients of the top model (Table ): underlying decline is unclear; likely candidates include harvesting pressure and environmental effects (e.g. climat- P Y j ( ≤ ) ic change) but psychological effects (e.g. people remem- exp a b period b alt b forestcov b biog b sample b surveyor ( j − 1 − 2 − 3 − 4 − 5 − 6 ) bering the past being better than it was) cannot be = 1 exp aj b1period b2 alt b3 forestcov b4 biog b5 sample b6 surveyor + ( − − − − − − ) ruled out. When the predicted values for the top model were examined Cumulative link models can show the magnitude of between time periods the magnitude of the change in change on a relative scale but describing an absolute change

http://journals.cambridge.org Downloaded: 26 Jan 2016 IP address: 69.74.254.25 Manus green tree snail 5

TABLE 1 Candidate models for the prediction of ordinal categories, with number of parameters, Akaike’s information criterion with a small sample correction (AICc), ΔAICc, model weight, cumulative model weight, and maximized value of the log-likelihood function (LL). The models are ranked by AICc.

Model Cumulative Rank Model No. of parameters AICc ΔAICc* weight weight LL 1 period + alt + forestcov + biog + sample + surveyor 12 2,182.05 0 0.93 0.93 1,078.92 − 2 period + alt + forestcov + biog + surveyor 11 2,187.15 5.1 0.07 1 1,082.49 − 3 alt + forestcov + biog + sample + surveyor 11 2,200.48 18.43 0 1 1,089.15 − 4 alt + forestcov + biog + surveyor 10 2,211.31 29.25 0 1 1,095.58 − 5 period + alt + forestcov + biog + sample 8 2,220.21 38.16 0 1 1,102.06 − 6 period + alt + forestcov + biog 7 2,224.11 42.05 0 1 1,105.02 − 7 period + alt + forestcov 6 2,224.21 42.16 0 1 1,106.08 − 8 alt + forestcov + biog + sample 7 2,237.02 54.97 0 1 1,111.47 − 9 alt + forestcov + biog 6 2,245.56 63.51 0 1 1,116.75 − 10 alt + forestcov 5 2,245.99 63.94 0 1 1,117.98 − 11 alt + biog + surveyor 9 2,309.84 127.79 0 1 1,145.86 − 12 period + alt + biog 6 2,313.21 131.16 0 1 1,150.58 − 13 period + alt 5 2,324.78 142.72 0 1 1,157.37 − 14 alt + biog 5 2,339.19 157.14 0 1 1,164.58 − 15 alt 4 2,350.55 168.5 0 1 1,171.26 − 16 forestcov 4 2,536.46 354.4 0 1 1,264.21 − 17 period + sample 5 2,831.03 648.97 0 1 1,410.49 − 18 sample 4 2,875.53 693.48 0 1 1,433.75 − 19 period 4 2,904.46 722.41 0 1 1,448.22 − 20 Singular intercept 3 2,935.74 753.69 0 1 1,464.86 − *Difference between the AICc of the current model and the top model cannot be undertaken unless equality of distance between the traded internationally, despite the restrictions on inter- ordinal categories is assumed. Although this assumption may national trade. One interviewee reported giving  shells be helpful for visualization (especially mapping; e.g. Fig. ) to a relative for sale in Australia; another market vendor re- there is no logical basis for its application. However, model ported an expatriate buying her entire stock of  shells for predictions of the probability of assignment to each ordinal PGK  per shell (five times the standard price). Based on abundance category can clarify the magnitude of the change. these anecdotes it is conceivable that the number of shells These predictions show that the probability of the snail being being supplied internationally may approach or eclipse the assigned to the plentiful category declined from  to , domestic trade. with a compensatory increase in the probability of all the Online searches revealed the shells were being sold, often other categories (Fig. ). For both years the majority of as antiques, in open forums and internet market places responses fell in the plentiful category, with the predicted based in New Zealand, Australia and the USA. The majority probability declining from  to % between  and . of listings avoided using the common name or a variant, in- The modelling suggested that respondents were influ- stead referring to the snails simply as land snails, and using enced by the surveyor. It is likely that each surveyor’s indi- the antiquated scientific pseudonym Papuina pulcherrima vidual manner and phrasing influenced respondents in their rather than Papustyla pulcherrima. The avoidance of con- interpretations of where the transition points between rela- ventional nomenclature may be an attempt to avoid detec- tive abundance categories lay. Although the inclusion of a tion by authorities. In some cases sellers in internet market surveyor effect in the top model caused little change to the places were based in CITES member countries (including period coefficient estimate, it caused the estimated qualita- Australia, Italy, New Zealand, Singapore and USA). The tive decline to increase from c.  to c. % over the -year quantity of shells for sale in such online market places ap- period. I used the higher figure in a precautionary approach pears to be low, suggesting that the existing controls on but am aware that the difference is probably an artefact and international trade may be adequate. However, given that indicates greater measures to ensure surveyor consistency the prices of shells for sale online were often orders of mag- should be undertaken in future. nitude greater than market prices on Manus Island, vigi- Despite limited sales data collected from the Lorengau lance will be required to ensure that illegal international market, a crude extrapolation of shell sales over a -day per- demand does not fuel a resurgence in snail collection. iod suggests that annual sales may approach ,. From the As of May  the Manus green tree snail was categor- interviews it was apparent that some shells were being ized as Data Deficient (Mollusc Specialist Group, ).

TABLE 2Coefficientsforthepredictionofordinalcategories,basedon the top model (period + alt + forestcov + biog + sample + surveyor), with estimates on the logit scale, and range of covariates.

Estimate ± SE Covariate range

αj threshold coefficients Absent/scarce 1.0612 ± 0.3349 Few/moderate 2.1308 ± 0.3445 Moderate/plentiful 2.7233 ± 0.3483 βn coefficients Period (2013) 0.6198 ± 0.1386 0 or 1 Alt 0.0201 ± 0.0016 0–435 Forestcov 3.5588 ± 0.3788 0.009–0.970 Biog (offshore island) 0.4023 ± 0.3595 0 or 1 Sample 0.0423 ± 0.0158 1–22 − Surveyor Person 1 0.0267 ± 0.1998 Person 2 0.6301 ± 0.1697 − Person 3 0.0082 ± 0.5269 Person 4 0.5805 ± 0.1855

FIG. 3 A comparison of the predicted probability of assignment IUCN guidelines specify that assessors should, where pos- to each ordinal abundance category for  and  on the basis of the top model sible, refrain from assigning this category where habitat de- (period + alt + forestcov + biog + sample + surveyor). terioration and other causal factors are known (IUCN, ). This study provided two pieces of information to fa- cilitate categorization of the snail: () a qualitative decline of protection of the remaining primary forest, which covers c. % over  years, and () the positive relationship be- much of central Manus. Specific interventions for the Manus tween forest cover and snail abundance. green tree snail are unlikely to be necessary at present but we If the qualitative rate of decline is assumed to be equiva- are trying to raise awareness among the public, tourists, provin- lent to an absolute change (a worst case scenario) it is still less cial government, and the relevant customs agencies of its than the threshold of $ % decline over  years or three CITES II status and the repercussions of illegal international generations required by IUCN for a categorization of trade. Vulnerable under criterion A (IUCN, ). Of the remain- It is important to note that this work does not advocate ing assessment criteria, the snail appears to come closest to the use of Wisdom of Crowds methods to supplant quanti- attaining a Vulnerable categorization under criterion B (i.e. tative field monitoring. Well-structured quantitative moni- continuing decline) but fails to meet the threshold because toring (sensu Lindenmayer & Likens, ) remains the the population is neither highly fragmented nor subject to ideal. However, Wisdom of Crowds methods are probably extreme fluctuations. On this basis a petition was made to sufficient for basic conservation decision making, such as the IUCN, and our recommended categorization of Near determining a species’ threat status or undertaking a triage Threatened was accepted and was conferred in version assessment in circumstances where there is local knowledge . of the Red List (Whitmore et al., ). We informed of the resource or species in question and a large sample size the Manus provincial government of this development. can be acquired but a quantitative survey is not possible be- However, forest clearance is an ongoing threat on Manus cause of financial constraints. For these reasons Wisdom of Island, with proposed logging, mining, road building and Crowds methods may be of particular utility in rural areas in agroforestry developments (Wildlife Conservation Society, the developing world. unpubl. data). If large-scale deforestation were to occur, a Describing resource change based on personal or collect- review of the snail’s threat status would be advisable. ive memory unavoidably raises questions about how people The results suggest that the Manus green tree snail does not perceive and recall environmental change. Papworth et al. require immediate in situ conservation intervention, despite () proposed two categories of possible memory artefacts: the threats outlined. According to the Red List (IUCN, ) () personal amnesia, in which the respondent remembers there are other species on Manus Island facing a greater risk past conditions inaccurately, and () generational amnesia, of extinction. In the short term, therefore, it is proposed there in which environmental baselines are reset with each passing should be no further conservation expenditure on this snail generation because of a lack of intergenerational memory. In species per se but that conservation of this and other species the case of personal amnesia, Wisdom of Crowds methods, as on the island is likely to be served best by focusing on the a consequence of their requirement for a large sample size,

http://journals.cambridge.org Downloaded: 26 Jan 2016 IP address: 69.74.254.25 Manus green tree snail 7

TABLE 3 Concurrence rate of information provided by  respondents in open structured interviews.

Information % interviewees concurring Snails feed on leaves or leaf biofilms. 97 The best time to search for snails is during early morning. 73 Snails are encountered most commonly after primary forest has been cleared for the establishment of gardens. 54 The respondent sells snail shells. 52 The snail population is stable or increasing. 38 The respondent could identify the sex of snails. 37 The respondent could not identify the sex of snails. 36 The snail population is stable or decreasing. 32 Snails are found on the leaves of the sago palm. 29

should be capable of dampening idiosyncratic noise and dis- screening of expert panellists, nor information feedback sys- cerning the underlying pattern if the errors in memory are tems, and the conclusion is reached statistically using a large highly individualized and random rather than skewed in na- sample size rather than by the consensus of a small group. ture. Although there was no clear evidence of generational Given the need for cheap and rapid species assessment amnesia in this study, c. % of all respondents had no knowl- tools and the impetus for judicious use of conservation edge of the snail’s name in their native language, which sug- funds, Wisdom of Crowds methods are an overlooked but gests some loss of cultural knowledge had taken place. potentially valuable tool for conservation. For widespread I suggest that prior to the application of Wisdom of acceptance of the Wisdom of Crowds methodology con- Crowds an examination of what Palmer & Wadley () de- firmatory field surveys will be required to validate the tech- scribed as local ecological talk (i.e. what people say they know) nique. Pragmatically, high-profile species, which have the is required. There is potential for confusion between local eco- potential to elicit reasonable levels of funding for compara- logical talk and local ecological knowledge, which for the pur- tive field surveys, will be the focus of such validation tests. In poses of Wisdom of Crowds must be experiential in nature. the long term, however, it is likely that Wisdom of Crowds Examples of this dichotomy can be seen in some of the an- methods will be used primarily in countries where conser- swers to open interview questions. Questions that could be vation budgets are meagre, and focused on species for which clarified by simple observation elicited a narrow range of opportunities to elicit funding are limited and for which near unanimous answers (e.g. % mentioned that the snails conventional field survey methods are unaffordable. fed on leaves and/or leaf biofilms), which could be considered local ecological knowledge. In contrast, the answers to the question of whether respondents could identify the sex of Acknowledgements snails, and how, were equivocal and variable, suggesting this This survey would not have been possible without the assist- was an example of local ecological talk rather than knowledge. ance of the Wildlife Conservation Society (WCS) Manus It is therefore advisable to use a Wisdom of Crowds approach community facilitators: Heoa Itariri, Jenny Wak, Julien only for questions that can be resolved by superficial observa- Benjamin and Pondrilei Sori. WCS staff members June tion (e.g. changes in spatial patterns and abundance), and not Poloman, Daniel Charles, Lilly Ugi, Junior Walker, John for matters of biology, which by their hidden nature are sub- Ole and Marygrace Wanamp-Puri provided logistical sup- ject to conjecture and local ecological talk. port for the project. I am grateful to Arison Arihafa, John Although other researchers have advocated the use of Lamaris, Wallace Takendu, Mazzella Maniwavie, Sven local ecological knowledge to address questions regarding Frijlink, Richard Cuthbert and two anonymous reviewers natural resources and their management (e.g. Huntington, for their insightful feedback. This project was funded by ; Fazey et al., ; Raymond et al., ; Sutherland The Christensen Foundation as part of a WCS project on et al., ), much of the emphasis has been on the screening nature and culture. I am grateful to the Manus Provincial of local experts (e.g. Davis & Wagner, ; Martin et al., Government for allowing this project to take place. I ) and the development of group consensus, typically thank all the survey respondents in Lorengau for their pa- using a Delphi technique framework (e.g. van der Hoeven tience and involvement in this study. et al., ; Leite & Gasalla, ). This essentially operates by allowing the opinion of independent experts to evolve through a series of iterative information feedback loops References until a consensus is reached (Hsu & Sandford, ).

Wisdom of Crowds methods, by comparison, represent a AGRESTI, A. () Analysis of Ordinal Categorical Data. nd edition. substantial departure from this paradigm as there is no John Wiley & Sons Hoboken, USA.

ARLINGHAUS, R. & KRAUSE, J. () Wisdom of the crowd and JORDAENS, K., DILLEN, L. & BACKELJAU,T.() Effects of mating, natural resource management. Trends in Ecology & Evolution, , breeding system and parasites on reproduction in hermaphrodites: –. pulmonate gastropods (Mollusca). Animal Biology, , –. BIVAND, R., KEITT,T.&ROWLINGSON, B. () rgdal: Bindings for LEITE, M.C. & GASALLA, M.A. () A method for assessing fishers’ the Geospatial Data Abstraction Library. R v. .–. Http://CRAN. ecological knowledge as a practical tool for ecosystem-based R-project.org/package=rgdal [accessed  June ]. fisheries management: seeking consensus in south-eastern Brazil. BOSLER, N. & SCHNEIDER, G. () The oracle or the crowd? Experts Fisheries Research, , –. versus the stock market in forecasting ceasefire success in the LINDENMAYER, D.B. & LIKENS, G.E. () Effective Ecological Levant. In Global Governance: Political Authority in Transition. Monitoring. CSIRO Publishing, Collingwood, Australia. International Studies Association Annual Conference. Montreal, LORENZ, J., RAUHUT, H., SCHWEITZER,F.&HELBING, D. () How Canada. social influence can undermine the wisdom of crowd effect. BOTTRILL, M.C., JOSEPH, L.N., CARWARDINE, J., BODE, M., COOK, C., Proceedings of the National Academy of Sciences, , –. GAME, E.T. et al. () Is conservation triage just smart decision LYDEARD, C., COWIE, R.H., PONDER, W.F., BOGAN, A.E., BOUCHET, making? Trends in Ecology & Evolution, , –. P., CLARK, S.A. et al. () The global decline of nonmarine BRYAN, J.E. & SHEARMAN, P.L. (eds) () Papua New Guinea mollusks. BioScience, , –. Resource Information System Disks. PNGRIS Publication MARTIN, T.G., BURGMAN, M.A., FIDLER, F., KUHNERT, P.M., No. . University of Papua New Guinea, Port Moresby, Papua New LOW-CHOY, S., MCBRIDE, M. & MENGERSEN, K. () Eliciting Guinea. expert knowledge in conservation science. Conservation Biology, , BURNHAM, K.P. & ANDERSON, D.R. () Model Selection and –. Multimodel Inference: A Practical Information–Theoretic Approach. MAZEROLLE, M.J. () AICcmodavg: Model Selection and nd edition. Springer-Verlag, New York, USA. Multimodel Inference Based on (Q)AIC(c). R v. .. Http://CRAN. CANO, L.S. & TELLERÍA, J.L. () Local ecological knowledge as a R-project.org/package=AICcmodavg [accessed  June ]. tool for assessing the status of threatened vertebrates: a case study in MOLLUSC SPECIALIST GROUP () Papustyla pulcherrima. In The Vietnam. Oryx, , –. IUCN Red List of Threatened Species v... Http://www. CARDOSO, P., ERWIN, T.L., BORGES, P.A.V. & NEW, T.R. () The iucnredlist.org [accessed  August ]. seven impediments in invertebrate conservation and how to NOFER, M. & HINZ, O. () Are crowds on the internet wiser than overcome them. Biological Conservation, , –. experts? The case of a stock prediction community. Journal of CHRISTENSEN, R.H.B. () ordinal: Regression Models for Ordinal Business Economics, , –. Data. R v. .–. Http://www.cran.r-project.org/package= PALMER, C.T. & WADLEY, R.L. () Local environmental ordinal [accessed  June ]. knowledge, talk, and skepticism: using ‘LES’ to distinguish ‘LEK’ DAVIS, A. & WAGNER, J.R. () Who knows? On the importance of from ‘LET’ in Newfoundland. Human Ecology, , –. identifying ‘experts’ when researching local ecological knowledge. PAPWORTH, S.K., RIST, J., COAD, L. & MILNER-GULLAND, E.J. () Human Ecology, , –. Evidence for shifting baseline syndrome in conservation. DULVY, N.K. & POLUNIN, N.V.C. () Using informal knowledge to Conservation Letters, , –. infer human-induced rarity of a conspicuous reef fish. Animal PITMAN, R.W. () Manus Island’s green tree snails at home. Conservation, , –. Hawaiian Shell News, , –. FAZEY, I., FAZEY, J.A., SALISBURY, J.G., LINDENMAYER, D.B. & RAYMOND, C.M., FAZEY, I., REED, M.S., STRINGER, L.C., ROBINSON, DOVERS, S. () The nature and role of experiential knowledge G.M. & EVELY, A.C. () Integrating local and scientific for environmental conservation. Environmental Conservation, , knowledge for environmental management. Journal of –. Environmental Management, , –. GALTON,F.() Vox populi. Nature, , –. RDEVELOPMENT CORE TEAM () R: A Language and Environment HANSEN, M.C., POTAPOV, P.V., MOORE, R., HANCHER, M., for Statistical Computing. R Foundation for Statistical Computing, TURUBANOVA, S.A., TYUKAVINA, A. et al. () High-resolution Vienna, Austria. global maps of st-century forest cover change. Science, , SÁENZ-ARROYO, A., ROBERTS, C.M., TORRE, J. & CARIÑO-OLVERA, –. M. () Using fishers’ anecdotes, naturalists’ observations and HIDE, R.L., ALLEN, B.J., BOURKE, R.M., FRITSCH, D., GRAU, R., grey literature to reassess marine species at risk: the case of the Gulf HELEPET, J.L. et al. () Manus Province: Text Summaries, Maps, grouper in the Gulf of California, Mexico. Fish and Fisheries, , Code Lists and Village Identification. Agricultural Systems of Papua –. New Guinea. Working Paper No . The Australian National SHEARMAN, P.L., BRYAN, J.E., ASH, J., HUNNAM, P., MACKEY, B. & University, Canberra, Australia. LOKES, B. () The State of the Forests of Papua New Guinea. HIJMANS, R.J. () raster: Geographic Data Analysis and Modeling. R Mapping the Extent and Condition of Forest Cover and Measuring package v. .–. Http://CRAN.R-project.org/package=raster the Drivers of Forest Change in the Period –. University of [accessed  June ]. Papua New Guinea, Port Moresby, Papua New Guinea. HSU, C.C. & SANDFORD, B.A. () The Delphi technique: making SUROWIECKI, J. () The Wisdom of Crowds: Why the Many Are sense of consensus. Practical Assessment, Research & Evaluation, , Smarter Than the Few. Little Brown, London, UK. –. SUTHERLAND, W.J., GARDNER, T.A., HAIDER, L.J. & DICKS, L.V. HUNTINGTON, H.P. () Using traditional ecological knowledge in () How can local and traditional knowledge be effectively science: methods and applications. Ecological Applications, , – incorporated into international assessments? Oryx, , –. . UNEP–WCMC () Checklist of CITES Species (CD-ROM). CITES IUCN () Guidelines for Using the IUCN Red List Categories and Secretariat, Geneva, Switzerland, and UNEP World Conservation Criteria. Version . Standards and Petitions Subcommittee, IUCN, Monitoring Centre, Cambridge, UK. Gland, Switzerland. U.S. GOVERNMENT () Endangered Species Act of , As IUCN () IUCN Red List of Threatened Species v. .. Http:// Amended Through the th Congress. Http://www.fws.gov/ www.iucnredlist.org [accessed  August ]. endangered/esa-library/pdf/ESAall.pdf [accessed  June ].

http://journals.cambridge.org Downloaded: 26 Jan 2016 IP address: 69.74.254.25 Manus green tree snail 9

VAN DER HOEVEN, C.A., DE BOER, W.F. & PRINS, H.H. () Pooling of the Cognitive Science Society, Sapporo, Japan, August –,  local expert opinions for estimating mammal densities in tropical (eds N. Miyake, D. Peeble & R.P. Cooper), pp. –. Curran rainforests. Journal for Nature Conservation, , –. Associates Inc., Red Hook, USA. VENABLES, W.N. & RIPLEY, B.D. () Modern Applied Statistics with WHITMORE, N., ARIHAFA, A., TAKENDU,W.&LAMARIS, J. () S. th edition. Springer, New York, USA. Papustyla pulcherrima. In The IUCN Red List of Threatened Species WEITZMAN, M.L. () The Noah’s ark problem. Econometrica, , v. .. Http://www.iucnredlist.org [accessed  June ]. –. WILSON, H.B., JOSEPH, L.N., MOORE, A.L. & POSSINGHAM, H.P. WELINDER, P., BRANSON, S., PERONA,P.&BELONGIE, S.J. () The () When should we save the most endangered species? Ecology multidimensional wisdom of crowds. In Advances in Neural Letters, , –. Information Processing Systems : th Annual Conference on Neural Information Processing Systems  (eds J.D. Lafferty & C.K.I. Williams) pp. –. Curran Associates Inc., Red Hook, Biographical sketch USA. WELSH, M.B. () Expertise and the wisdom of crowds: whose NATHAN WHITMORE’s research interests are sustainable harvesting, judgments to trust and when. In Building Bridges Across Cognitive population modelling, and incorporation of local knowledge in conser- Sciences Around the World: Proceedings of the th Annual Meeting vation planning.