Species selection
We used data for 38,245 threatened and Near Threatened species from Red List version 2020-1 and coded species to assign them to a category pertaining to threat from international trade based on available information in Red List assessments (hereafter “assessments”). Background on the Red List and limitations to using these data for this purpose are provided in Supplementary Methods 2.1–2.3.
To identify species that may be threatened by international trade we queried the Red List and constructed a MS Excel database of candidate species. A PostgreSQL database, which contains a copy of all data from current published assessments, was used for data extraction; we ran five SQL queries on this database using pgAdminIII (database querying software). We used the combined results to assign species to a category using automated and manual coding (see below).
Query 1
The first query extracted the threat category and all data from the rationale, threats, and use and trade sections (text fields) of assessments, for species selected based on the following criteria: (1) species categorised as Critically Endangered (CR), Endangered (EN), Vulnerable (VU), Near Threatened (NT), Low Risk/near threatened (LR/nt) or Low Risk/conservation dependent (LR/cd); and either (2) assessments contained one or more of 53 text strings (e.g., “commercial use”, full list in Supplementary Methods 2.2) within the rationale, threats, and/or use and trade sections; or (3) assessments included one or more of 11 threat codes relating to BRU (5.1.1, 5.1.4, 5.2.1, 5.2.4, 5.3.1, 5.3.2, 5.3.5, 5.4.1, 5.4.2, 5.4.4, 5.4.6; Supplementary Methods 2.2). Species classified as LR/nt and LR/cd were treated as NT, as per Red List guidance. We excluded Least Concern (LC) species on the basis that they are likely to be at lower risk from overexploitation. We also excluded Data Deficient (DD) species. This resulted in a database of 21,714 species.
The 53 text strings were chosen as those most likely to return species that may be threatened by international trade. We searched assessments using these text strings because for species listed as Extinct (EX), Extinct in the Wild (EW), CR, EN, VU, NT, LC and DD, it is a requirement when completing assessments that supporting information is provided in the threats text field in the form of a narrative on threats. For utilized species, it is recommended, though not mandatory, that supporting information be included in the use and trade text field in the form of a narrative on use and trade.
Regarding threat codes, it is a requirement when completing assessments for species listed as EX, EW, CR, EN, VU and NT (but not LC or DD) that major threats to the species be coded according to the IUCN standardized Threats Classification Scheme38. We selected species where the threats included one or more of the 11 aforementioned threat codes on the basis that these species may be threatened by international trade. We included threat codes where motivation is unknown because while the coding suggests that it is not known if the species is the target (of harvest), assessors are known to use this code when use is intentional, but the scale is not known34. We included threat code 5.4.4 (Biological Resource Use [BRU] à Fishing & harvesting aquatic resources à Unintentional effects: large scale) because such species could theoretically be threatened by international trade, despite harvest being unintentional.
Query 2
The second query enabled us to add information from the IUCN Use and Trade Classification scheme39 to our database, specifically the end uses for which species were coded in the end use table in assessments. On completing assessments for species that are utilized, it is recommended, though not mandatory, that supporting information on trade and/or use be included by means of indicating whether use is one or more of “subsistence”, “national” and/or “international”. Assessors are also asked to indicate the purpose of use from a list of 18 different purposes (e.g. “Food – human”, full list in Supplementary Methods 2.2). We used these data rather than the scale of use (e.g., “Local livelihood – Subsistence”) because doing so enabled us to distinguish between uses (at subsistence, national and/or international levels) comprising a threat to species and those that are not when combined with other information and applying our criteria to species.
We cross-referenced the results of our first two queries to identify any species that had any “international” uses coded but were not captured by our first query. This resulted in the addition of one species, Cynanchum itremense, to our database, and 21,715 candidate species See Species Verification for detail on the process meaning our final dataset had 21,745 candidate species.
Query 3
The third query enabled us to add information to our database on whether international trade is recorded as a significant driver of threat to species. For a subset of threat codes (5.1.1, 5.2.1, 5.3.1, 5.3.2, 5.4.1, and 5.4.2) relating to intentional use, assessors are asked to code whether international trade is a significant driver of that threat to species, or not, or whether it is unknown. However, this code was only recently added to the data system, is not consistently applied, and has only been used in a subset of assessments, and therefore it is not yet a reliable indicator of the number of species threatened by international trade on the Red List.
Query 4
The fourth query extracted data on coded threats to all species on the Red List, including whether threats were current, past or future; temporal data were added to our database for corresponding species. This enabled evaluation of coded threats to species relating to BRU.
Query 5
The fifth query extracted data from the IUCN Use and Trade Classification scheme39 for candidate species, specifically from the field “no use/trade information for this species”. This field is intended to be used to indicate that it is known or highly likely that the species is used and/or traded, but further information is not available (Supplementary Methods 2.3).
Species categorisation
We developed criteria in order to assign species to a category - Likely or Unlikely to be threatened by international trade, or Insufficient information, adapting an approach developed by IUCN in 201540. We applied the criteria to the 21,715 species that were selected using the process outlined above and using a combination of automated and manual coding (i.e., coding by a human being). Our criteria are:
Likely to be threatened by international trade:
- Intentional use is coded as a threat and “is international trade a significant driver of threat” is coded as yes; or
- There is evidence to suggest that use and/or trade is a (probable or certain) threat to one or more populations/subpopulations (from threat code or description in rationale, threats or use and trade sections) and that form of use and/or trade is to some extent international (from international use being coded as yes and/or a relevant international end use is coded and/or from description in rationale, threats or use and trade sections).
Insufficient information to determine if species is threatened by international trade:
- There is evidence to suggest that use and/or trade takes place (from threat codes or description in rationale, threats or use and trade sections, or “no use/trade information for this species” is coded as yes) and is a (probable or certain) threat to one or more populations/subpopulations (from threat codes or description in rationale, threats or use and trade sections), but there is no evidence it is international and also no evidence that it is not international (from description in rationale, threats or use and trade sections and international/national/subsistence uses not coded); or
- There is evidence to suggest that use and/or trade takes place (from threat codes or description in rationale, threats or use and trade sections, or end uses, or “no use/trade information for this species” is coded as yes), there is no evidence that it is not international (from description in rationale, threats or use and trade sections or international use is coded as yes), and either (i) there is no evidence that it is a threat and also no evidence that it is not a threat (from description in rationale, threats or use and trade sections) or (ii) it is described to be a past, future, potential, possible (or similar) threat; or
- There is no evidence that use or trade takes place (from threat codes or description in rationale, threats or use and trade sections, and no uses are coded, and “no use/trade information for this species” is blank), but it is described as a potential future (or similar) threat.
Unlikely to be threatened by international trade:
- There is no evidence that use or trade takes place (from threat codes or description in rationale, threats or use and trade sections, no end uses are coded, and “no use/trade information for this species” is blank), and it is not described as a potential future (or similar) threat; or
- There is evidence to suggest that use and/or trade takes place (from threat codes or description in rationale, threats or use and trade sections, end uses, and “no use/trade information for this species” is coded as yes) but that it is subsistence and/or national level and not international (from description in rationale, threats or use and trade sections, or subsistence and/or national use coded as yes and international as no); or
- There is evidence to suggest that use and/or trade takes place (from threat codes or description in rationale, threats or use and trade sections, end uses, or no use/trade information for this species is coded as yes) but that it is not a threat (from description in rationale, threats or use and trade sections).
We took an evidentiary but precautionary approach (i.e., assumed greater rather than lesser risk to species) to reasonably deduce from available information in each assessment whether international trade constitutes a threat to species, or not. We focused on determining categorically whether there was evidence that international trade was a threat to species, regardless of the level of threat (Supplementary Methods 2.4). If we were unable to deduce from available information in each assessment that a species was threatened in any way by international trade, even if it is a species known to be impacted by international trade from other information sources, then it was categorised as Insufficient information or Unlikely based on the information available. We used data on “international trade is a significant driver of threat” (Query 3) to categorise species but did not use other responses (“no” and “unknown”) because the aim was to determine whether international trade posed any level of threat to species rather than being a significant driver of threat necessarily.
Automated coding
We coded 9,320 species to assign them to one of the three aforementioned categories using automated coding where it was feasible to do so based on the “use and trade” and “is international trade a significant driver of threat” fields, and the relevance of use-related threat codes using R version 4.0.341 (Extended Data Fig. 4; Supplementary Methods 2.4). Species that were coded “yes” for whether international trade is a significant driver of threat were coded Likely. Where the use and trade text field of assessments contained phrases such as “information regarding the trade and use of this species is not known” or similar, the species was coded Insufficient information. Where the use and trade text field included phrases such as “there is no known use and trade in this species” or similar, the species was coded Unlikely. Where it was evident that species of fauna and funga had been included in our database based only on the presence of flora-related text strings (e.g., “timber”) in assessments, they were categorised as Unlikely.
Manual coding
We manually coded the remaining 12,395 species to assign them a category because the available information needed to be interpreted by a human coder. This is because there is no direct link on the Red List between end uses and threats or scale of use beyond information in the text fields. Manual coding entailed reading the information and data for each assessment—text fields, threat codes, scale of use codes, purpose of use codes, “no use/trade information on this species” field, and “is international trade a significant driver of threat” field—and categorizing species aided by a decision tree (Extended Data Fig. 5). For instance, a species with a relevant threat code may be used at the subsistence, national and/or international level and interpretation of the text fields was necessary to determine whether trade at the international level, rather than the subsistence and/or national level, comprised any level of threat (Supplementary Methods 2.4). Prior to coding, all coders trained on six batches of 100 randomly chosen species from our dataset. Before coding the full dataset, we measured our inter-rater reliability to ensure coders were categorising species in a standardised way using 100 randomly selected amphibian species. We used Fleiss’ Kappa in SPSS v.28 to test if agreement between all four coders was higher than would have been expected by chance. κ = 0.85 (95% CI, 0.79–0.91), p < .0005, indicating almost perfect agreement42. Remaining uncertainties were clarified among coders prior to coding the full dataset.
If a species could have been placed in one of two categories, we chose the most precautionary option i.e., assumed greater rather than lesser risk to the species. For example, coded a species as “Likely” rather than “Insufficient information”. However, we respected the qualification of coded BRU threats (e.g., as “possible”) (Extended Data Fig. 5). This also applied if there were contradictions between different pieces of information and data. We considered information in assessments to be current, recognizing that some assessments are older than 10 years (Supplementary Methods 2.3). Where threat codes were qualified (e.g., “past (unlikely to return)”) we interpreted them as past or current accordingly (Supplementary Methods 2.4). Regarding flora, we treated species as threatened by use even if only 5.3.5 (Biological resource use [BRU] à Logging & wood harvesting à Motivation Unknown/Unrecorded) was coded as a threat unless it was evident in the text fields that the species was not a tree, it was stated that code 5.3.5 applies to the species' habitat (i.e., not to the species), or other information meant it was not relevant (e.g., past threat). Following coding, species categorised as Likely and Insufficient information were checked for accuracy of coding.
Taxonomy alignment
Following the categorisation of species, we determined which species are, and are not, included in the CITES Appendices to determine those taxa currently subject to CITES trade measures. The full list of official species names from the CITES Appendices were downloaded from the Checklist of CITES Species43 and cross-checked with all 21,715 species to determine if the names corresponded to CITES-listed species. We considered species to be the same when the scientific name matched, even though we acknowledge that the species concept may differ, as taxonomies differ between the Red List and CITES (Supplementary Methods 2.7). Where no match was found, synonyms were considered to ensure that species treated as synonyms by either IUCN or CITES, and which were accepted names in the other taxonomy, were not overlooked. For potential matches involving synonyms, particularly cases involving two synonyms, additional verification was carried out by manually checking the Red List assessment to ensure that the match was logical; species too distantly related or clearly referring to a separate species were discounted. Higher taxonomic listings in CITES (e.g., primates), were cross-checked to ensure that even where there was not an exact match in nomenclature, species on the Red List within the corresponding genus, family or order of relevance received the corresponding CITES listing. For example, if there was a newly described primate on the Red List not yet recognised in the CITES nomenclature, the species was assumed to be covered by the Appendix II listing for primates, or the Appendix I listing for the relevant genus or family.
For species with CITES listings that only cover certain populations (e.g., Diospyros, populations of Madagascar) or involve other exclusions (e.g., the Euphorbia listing only applies to succulents), the distribution or other attributes were checked, where feasible, to ensure that the CITES listing or characterisation as “non-CITES” were correct. Where uncertain, we consulted the CITES nomenclature specialists for fauna and flora, respectively.
Species verification
As the Red List had been updated on completion of coding, we verified whether each species in our dataset remained distinct. We did this by cross-referencing the unique identifier number for all species on the 2020-1 version of the Red List and the species in our dataset to identify those species no longer on the Red List (e.g., because their taxonomy had changed). The Species Information Service database (SIS), which is used to store all current and historic Red List assessment data, was used for this purpose. Following the removal of 25 and addition of 74 species, but removing 19 LC and DD species, resulted in 21,745 candidate species. We coded the additional species and cross-referenced them with the CITES listing information as described. 3,815 of these species mapped to CITES-listed species.
Species calculations
We calculated the number of species in each category (e.g., Likely), those included and excluded from CITES, and the proportion of species with BRU as a threat that are Likely threatened by international trade, or not, overall and by class. To account for the uncertainty of species categorised as having Insufficient information we followed previous studies44,45 to estimate the proportion of these species that would be expected to be categorised as Likely and Unlikely if there was sufficient information (Supplementary Methods 2.5).
Repeatability
We assessed if the process could be fully automated using an advanced automated coding method and used Fleiss’ Kappa to test for agreement between approaches (Supplementary Methods 2.8). We retrospectively re-coded all Actinopterygii (n=1,187) and Amphibia (n=329) species that were manually coded and compared the advanced and manual coding results. We also tested the advanced coding against the initial coding of all Actinopterygii and Amphibia species (i.e., including taxa that were coded using the simpler automated coding) and tested whether it could correctly categorize species in these classes with new or updated assessments. We then tested the approach on all animals (kingdom Animalia) in our dataset. The advanced coding achieved 83% accuracy for Actinopterygii and Amphibia species compared to manual coding (κ = 0.72, 95% CI: 0.68–0.75, p .000) and 92% across all species initially coded in these classes (κ = 0.82, 95% CI: 0.80–0.85, p .000) respectively. It achieved 88% accuracy for new or updated assessments. For all animals it achieved 77% accuracy (κ = 0.6, 95% CI: 0.58–0.62, p .000). These results demonstrate that the advanced coding performs well (Supplementary Tables 3–6, Supplementary Results 4.3) and this process can be used to generate data to inform decision-making in CITES (Supplementary Discussion 5.2).
Data availability
Data generated in this study are included in the supplementary data files.
Code availability
Source code for advanced automated coding of species is available on GitHub (link).
Method references
38. IUCN. Threats Classification Scheme (Version 3.2). https://www.iucnredlist.org/resources/threat-classification-scheme (2012).
39. IUCN. General Use and Trade Classification Scheme https://www.iucnredlist.org/resources/general-use-trade-classification-scheme (2020).
40. CITES. IUCN Red List Assessments of Asian Snake Species (Decision 16.104), CITES AC28 Doc. 14.3. https://cites.org/sites/default/files/eng/com/ac/28/E-AC28-14-03.pdf (2015).
41. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/ (2020).
42. Landis, J.T. & Koch, G.G. The Measurement of Observer Agreement for Categorical Data. Biometrics 33(1), 159–174. https://doi.org/10.2307/2529310 (1977).
43. UNEP-WCMC (Comps.). The Checklist of CITES Species Website. CITES Secretariat, Geneva, Switzerland. Compiled by UNEP-WCMC, Cambridge, UK. http://checklist.cites.org (Accessed 19 February 2021).
44. Schipper, J. et al. The Status of the World’s Land and Marine Mammals: Diversity, Threat, and Knowledge. Science 322(5899), 225–230. https://doi.org/10.1126/science.1165115 (2008).
45. Hoffmann, M. et al. The Impact of Conservation on the Status of the World’s Vertebrates. Science 330(6010), 1503–1509. https://doi.org/10.1126/science.1194442 (2010).