Impact of an IUCN national Red List of threatened flora on scientific attention

: Red Lists are thought to attract attention to the conservation of threatened species. Determining the impact of these lists on the attention of scientists is a matter of consequence for biodiversity conservation. We evaluated trends in mentions of Brazilian angiosperm plants in the biodiversity conservation literature and tested the effect of the Red List of Brazilian Flora (RLBF) publication on these mentions. We collected mentions in the literature available in Google Scholar from the years 1990−2020, for 2449 Brazilian angiosperm species assessed in different IUCN categories. We used a Bayesian structural time-series method to test the effect of the RLBF publication on the number of mentions for the set of species in the IUCN categories, angiosperm families, and plants of commercial interest. The results showed a gap in mentions for many threatened and Data Deficient species in the scientific literature. We also found that the mentions were biased toward species of commercial interest and were unrelated to their threat status. Publication of the RLBF positively affected the number of mentions for IUCN threat categories and for more than half of the angiosperm families. These results were obtained after a few species of commercial interest were excluded from each treated group. This study suggests that the Red List assessments are essential to determine priorities for resource allocation to scientific activities. However, this effect was not sufficient to reduce the bias in scientific attention. Our findings support the need to stimulate more effective programs to fund research on threatened plant species.


INTRODUCTION
Research conducted by the scientific community is essential to increase our knowledge of threatened species. The available information on taxonomy, biology, population status, threats, and protection measures undergirds action and management plans, which are needed to change the threat category of conservation targets to the IUCN status of 'Least Concern' (Baillie et al. 2004, Sutherland et al. 2004). However, research bias, i.e. the disproportionate investigation of certain species more than others, is a well-known problem in the study of conservation (Clark & May 2002). Several authors have suggested that research efforts may be geographically, taxonomically, and phylogenetically biased (Moustakas & Karakassis 2005, Lawler et al. 2006, Brodie 2009, Zhang et al. 2015, Donaldson et al. 2016, and unrelated to the threat status of the studied species (Murray et al. 2015). In particular, species that are threatened or facing a high probability of extinction may attract proportionately less research effort than nonthreatened species, as has been shown for sturgeon and marine mammal species (Jarić & Gessner 2012, Jarić et al. 2015. Conversely, charismatic threatened species in habitats in the developed world and those that have socioeconomic or socio-cultural value dominate the scientific literature (Amori & Gippoliti 2000, Sitas et al. 2009).
The Red List of Threatened Species (RLTS) assembled by the IUCN is a robust system for compilation, synthesis, and dissemination of species data (https:// www.iucnredlist.org/en). In summary, the list includes 9 categories, ranging from 'Not Evaluated' to 'Extinct', with intermediate categories reflecting both the state of knowledge and the level of threat (Baillie et al. 2004). Among the numerous contributions to conservation strategies is the capacity of the RLTS to influence the allocation of resources to species in the most critical categories (Rodrigues et al. 2006), including research efforts aimed at increasing knowledge of the species of concern. Thus, the determination of the impact of RLTS publication on research efforts is a matter of consequence for biodiversity conservation (Betts et al. 2020). Recent studies have tested this possible contribution; for example, the IUCN global Red List assessment was pivotal in attracting research efforts to several groups of animals, although this effect was most pronounced for species classified as Data Deficient (Jarić et al. 2017). Similarly, the creation of the list of the 'World's 25 Most Endangered Primates' in creased the research efforts mentioning the listed primates (Acerbi et al. 2020).
Brazil, a globally recognized center of biodiversity, houses approximately 37 000 vascular plant species, of which slightly more than half are considered endemic (Flora do Brasil 2020). This number represents 9−11% of all known vascular plant species (Nic Lughadha et al. 2016, WCVP 2020. The country is also considered one of the highest priorities for flora conservation (Myers et al. 2000 Brazil harbors much of the global biodiversity that is of interest to research groups dedicated to conservation biology; however, few attempts have been made to analyze the global research effort on threatened plants. Among the biodiversity conservation studies assessing the research effort directed toward Brazilian taxa (Frehse et al. 2016, de Barros et al. 2020, Guerra et al. 2020, only a few dealt with threatened animals (Gomes 2016, Tourinho et al. 2020, and only exceptionally were threatened plants addressed (Ribeiro et al. 2016). Additionally, the propensity of biodiversity scientists to undertake studies following the publication of a national Red List has not been tested, indeed comprehensive studies to determine this effect are recent and deal only with groups of animals (Jarić et al. 2017, Acerbi et al.2020. Here we used a bibliometric approach to evaluate trends in mentions of Brazilian angiosperm plants in the biodiversity conservation literature and to test the effect of the RLBF publication on these mentions. Specifically, these investigations were conducted for the sets of species in different IUCN categories (Critically Endangered, Endangered, Vulnerable, Near Threatened, and Data Deficient), species of commercial interest, and angiosperm families.

Selection of species
The RLBF comprised an assessment of 4582 vascular plants species (Martins et al. 2018). However, we included only angiosperms for our study list. In addition, for each angiosperm species, we excluded those that changed IUCN categories or their taxonomic status, i.e. the merging of several species into a single species ('lumping'), or the division of a species into 2 or more species ('splitting') during the period from 2013 to January 2021. Thus, we ensured that the scientific attention of the species studied was not affected by these types of recategorizations (Jarić et al. 2017, Tessarolo et al. 2017. Ultimately, 2449 angiosperm species were included in the dataset. Of these, 358 angiosperm species were assessed as Critically Endangered (CR), 912 as Endangered (EN), 422 as Vulnerable (VU), 322 as Near Threatened (NT), and 435 as Data Deficient (DD). We built a list with the relevant synonyms of the study species, following the Brazilian Flora 2020 database (Flora do Brasil 2020), as this combination of accepted names and synonyms (Table S1 in the Supplement at www. int-res. com/ articles/ suppl/ n046 p175 _ supp. pdf) can significantly increase the accuracy of the data recovered (Correia et al. 2018).
We also selected Brazilian angiosperm species categorized as Least Concern (LC) as the control group for the analyses, because we considered the baseline trend of frequency of mentions over time (see Sec-tion 2.3). For this category, we selected 1276 species that were submitted to the same exclusion criteria as the species in the dataset. These species were evaluated from the IUCN criteria, although they are not mentioned in the RLBF. We used information from the CNCFlora-JBRJ database (CNCFLORA-JBRJ 2020) to compile this list.
We conducted a systematic search for each scientific name in papers (Shanley & Medina 2005, Mendonça 2006, Coradin et al. 2011, Vieira et al. 2016) and a database (CNCFLORA-JBRJ 2020) that compiled information on species with current or potential economic value, for instance, species used for medicinal, timber or handicraft purposes or those which are edible. As a result, 96% (2353 of the 2449) of the study species were classified as 'unusable' for commercial purposes, and 4% (96 species) as 'usable'. For the control group, 91.7% (1170 of 1276) were classified as 'unusable' and 8.3% (106 species) as 'usable' (Table S2).

Scientific attention
We chose the Google Scholar database as a bibliometric indicator of the species. This database also includes comprehensive gray literature (e.g. systematic reviews, meta-analyses, or synopses re sulting from unpublished dissertations and theses), and this kind of content has provided evidence to inform many conservation decisions (Haddaway & Bayliss 2015, Calver et al. 2017. Furthermore, Google Scholar retrieves a wider range of literature when the content involves endemic species, compared to Scopus and Web of Science (Calver et al. 2017). Last, Brazil has the fifth largest number of university domains indexed in this database (Aguillo 2012), which increases the range of publications in Portuguese and reduces the effect of the massive number of North American, European, and Asian pub lications found in other databases (Holmgren & Schnitzer 2004, Li et al. 2018. We conducted a sequence of searches in Google Scholar to quantify scientific attention given to a species. In the search terms, we associated each individual scientific name with the set of terms 'Conservation' AND 'Status' AND 'Threats' AND 'Action'. We used these terms because they are key words in biodiversity conservation literature (Salafsky et al. 2008). In addition, they recorded more data from the scientific literature on conservation of threatened species reported by Google Scholar than other search terms (see our previous controlled experiment in Text S1 for more detail). We also conducted the same searches with the Portuguese language translations of the terms. The searches were also repeated for each synonym. The search results (hereafter, number of mentions) from Google Scholar were used as a measure of scientific attention. The period analyzed was 1990 through 2020, i.e. a total of 31 yr of observations. The search was last updated in March 2021.

Method for detecting effects
We applied a Bayesian structural time-series (BSTS) model (Brodersen et al. 2015) to test for effects of the RLBF publication on the number of mentions of the listed species in the biodiversity conservation literature. The BSTS model quantifies the impact of an event on a response metric of interest. This method combines concepts from time-series models and synthetic control methods to construct a synthetic counterfactual time series from a donor pool of control cases. In summary, the model requires one or more time series from a set of candidates that are classified by a matching algorithm as similar enough to the treated time series in the pre-treatment period; here, before the RLBF publication. Thus, using BSTS, the relationship between the matched series (hereafter, the control group) and the treated series is modeled on the pre-treatment period and used to predict the post-treatment series of the treated group. This post-treatment prediction is counterfactual to the treated series, under the scenario where no intervention was applied. The difference between the predicted counterfactual time series and the actual data is a measure of the impact of the intervention (Schmitt et al. 2018).
We used the 'Causal Impact' package, version 1.2.4 in R software version 3.6.2 (R Core Team 2019) to calculate the BSTS model. We selected 3 measures provided by the package to estimate the difference between the treated series and the counterfactual series: (1) absolute average effect (i.e. the average yearly difference); (2) absolute cumulative effect (i.e. the average difference of the entire post-intervention period); and (3) relative effect (i.e. the percentage of the difference of the entire post-intervention period). To quantify the error imposed on an effect size, the package provides the confidence interval for each estimation. In general, we consider there to be an effect (increase or decrease) if the confidence interval of the difference between the treated and counterfactual series does not include zero, which means we consider there to be no effect if the confidence interval is centered around zero. Thus, the tests were adjusted to a 99% confidence interval. To test the probability of obtaining any causal effect, the package also provides posterior tail-area probabilities, which can be interpreted as classical p-values, i.e. p = 0.05 means a 95% chance of obtaining this effect by chance (see Brodersen et al. 2015 for more detail).
The analysis period of our study was subdivided into pre-intervention (1990 through 2012) and postintervention (2014 through 2020), considering the year of RLBF publication (2013) as the axis of our causal analysis. We followed the recommendation of a pre-intervention period of approximately 2 or 3 times the length of the post-intervention (Brodersen et al. 2015). The post-intervention interval covers a reasonable time for scientists to respond to the Red List publication (e.g. Jarić et al. 2017).
We used 1276 LC angiosperm species as a control group. This decision was based on 2 fundamental premises of the BSTS model (Brodersen et al. 2015). The first assumption is that the time series values of the control group will not be affected after the intervention. Thus, it is reasonable to assume that the number of mentions in the literature for LC species will be indifferent to the publication of the Red List.
Research attention directed at this group is related to other factors, such as their charisma, economic value, suitability for use as model species, and accessibility (Jarić et al. 2019). The second assumption is the existence of a statistical relationship between the treated and the control group during the pre-intervention period.

Data analysis
First, the BSTS analyses were designed for the number of mentions of species in the different IUCN categories (CR, EN, VU, DD, and NT). For each dataset, we built a time series using an average number of mentions per species for each year from 1990 through 2020. For example, to construct the time series for the CR dataset, we computed the average number of mentions of the 358 species in this category for each year from 1990 through 2020. Second, we also built a time series for 39 angiosperm families in the RLBF (32.5% of the 120 families). The families were selected because they recorded at least 10 species listed as threatened and were therefore considered the most important. Third, new time series were created for the same IUCN categories and angiosperm families, using only the 'unusable' spe-cies. This approach avoids a confounding factor caused by the possible bias in mentions toward a few commercial species. For all tests, which totaled 88, we used the average number of mentions of the LC category as the control group.
We used Wilcoxon-Mann-Whitney U-tests for pairwise comparisons between the average number of mentions of species in different IUCN categories in the post-intervention period 2014−2020. The tests were also done using the sets of unusable and usable species within each category. Non-parametric tests were used because some variables did not show a normal distribution based on a Shapiro-Wilk test.

RESULTS
We recorded 33 898 mentions in the biodiversity conservation literature for the species studied from 1990 through 2020. However, 40.4% (684 of 1692) of threatened angiosperm species (CR, EN, and VU) had only 1 or no mention during the entire study period. We found these parameters for 38.8% (125 of 322) of species for NT. This was especially pronounced for DD with 62.7% of species (273 of 435). In contrast, 27.7% (354 of 1276) of the LC species had only 1 or no mentions.
The BSTS indicated no effects on mentions after RLBF publication for IUCN threatened categories (CR, EN, and VU), DD, and NT species during 2014−2020 (Table 1). On the other hand, positive effects were observed in the threatened categories when the few species of commercial interest were excluded from the analysis (unusable datasets, Table 1). For the CR (unusable) species, the mentions showed an increase of 0.407 (99% CI: 0.122, 0.664) compared to their counterfactual series throughout the entire post-intervention period (absolute cumulative effect). This increase was equivalent for EN (unusable) species with 0.400 (99% CI: 0.0542, 0.7) more mentions. The positive impact for the VU (unusable) species was most prominent as the mentions increased by 0.87 (99% CI: 0.527, 1.23); in relative terms, this represents an increase of 57% (99% CI: 35%, 80%). There was no effect on mentions for the DD (unusable) species (0.217; [99% CI: −0.0098, 0.427]). The test for NT (unusable) species was not significant (p > 0.05) and is not subject to interpretation.
We reported positive effects for 17 angiosperm families, no effects for 14, and negative effects for 8 (left panel in Fig. 1). The tests using only unusable species indicated positive effects for 20 families, while no effects were detected for 14 and negative effects for 5 families (right panel in Fig. 1). Table S4 summarizes the BSTS results.
The number of mentions per year for the LC, NT, VU, and DD species had peaks of average mentions concentrated in 2017, with 1.3, 1, 1.4, and 0.3 mentions per species, respectively (Fig. 2). EN species also had their second-highest mentions (0.4) in 2017. CR and EN species had peak mentions in 2014, immediately after the publication of the RLBF (0.4 and 0.5, respectively), although this number of mentions was not sustained in the following years. The VU group also experienced this rapid growth in 2014, with 1.4 mentions per species, although in the following year (2015) it decreased. Also starting in 2014, DD species showed an increase in mentions, which was sustained until 2018. Conversely, LC and NT species had no increase in mentions in 2014. The mean ± SD number of mentions per species between 2014 and 2020 was 4.86 ± 19.65. Average mentions for species in less-threatened categories (4.58 ± 16.30 for NT and 7.78 ± 21.73 for LC) and VU (7.97 ± 22.86) were higher than for species belonging to DD (1.43 ± 5.02) and to the most threatened categories (1.48 ± 2.37 for CR and 2.24 ± 6.38 for EN; Fig. 3). Table S5 summarizes the statistical tests for both categories.

DISCUSSION
This is the first study to evaluate the impact of a national Red List on scientific attention for the listed species reported in the biodiversity conservation literature. We found that the publication of the RLBF had a clear, significant effect on the allocation of scientific attention to species in the most concerning IUCN categories and for most angiosperm families, when the few species of commercial interest were excluded from the analysis. Species of commercial interest may have already had a high number of mentions in the pre-intervention period, which perhaps made any further increase in mentions unlikely. In any case, the development and implemen- Although the RLBF did not have the expected impact on the DD species (Jarić et al. 2017), the positive effect detected here is in line with a similar pattern found for threatened animals (Acerbi et al. 2020). This means that the Red List assessments and the inclusion of threatened species in official national lists is likely an important component in catalyzing actions by the scientific community toward threatened species (Rodrigues et al. 2006). This is good news regarding the globally high investments in Red List assessments of species. Surprisingly, the Red List publication seemed to have an impact on the scientific community already in the year after the publication of the RLBF, as mentions of species in the threat categories showed a clear increase in 2014 but those of LC species did not. However, this scientific attention was not sustained the following year. A similar rapid positive response pattern was observed in media penetration in searches for the term 'Red List' following the IUCN World Conservation Congress (Betts et al. 2020), and for threatened primate species after the publication of the 'World's 25 Most Endangered Primates' list (Acerbi et al. 2020). In view of the short period of just 1 yr, the published responses of scientists may have been limited to the inclusion of a plant name in a species list. Some examples are mentions in local floristic inventories (Kortz et al. 2014), or in non-peer-reviewed documents such as an ex situ survey of threatened flora in botanical gardens (Costa 2014), action plans aimed at priority areas that harbor the listed species (Loyola et al. 2014), or aspects of environmental legislation . It was not possible to determine the exact content of the papers from the year 2014 that mention the species. This gap inherent to the bibliometric approach is a recognized limitation for conservation biology studies (Proulx et al. 2014) and indicates the need for further research.
We found a strong bias arising from socio-economic preferences for certain plants. The absolute number of mentions of usable species was up to 33 times higher than for species of no commercial interest in some threat categories. At the same time, several threatened and DD species with no commercial interest remain understudied. Indeed, the socio-economic factors associated with species may be consistent drivers of scientific attention, while also helping to perpetuate research biases (Jarić et al. 2019). This asymmetry imposes challenges to infer general principles and to put in place effective strategies for biodiversity conservation (Troudet et al. 2017). For example, it is difficult to develop action plans that incorporate interactions between ecological, evolutionary, and environmental processes for species that are poorly known (Sitas et al. 2009). For this reason, it is urgent to allocate a similar or greater effort (Trimble & Aarde 2010) that is focused on nationally or regionally endemic threatened plants compared to  Fig. 2). Error bars represent standard deviations; number of species per category is shown in parentheses wild plants that contribute to human livelihoods. Although our analysis of scientific activity after the publication of the RLBF suggests that some researchers have transferred their scientific interest to these historically neglected species, this transfer is still insufficient.
Remarkably, research output is disproportionately directed toward less threatened species. The only exception is the Vulnerable category, which may be due to the large number of species of commercial interest in this group. This scientific attention unrelated to threat status is not unique to Brazil (Brito 2008, Brito & Oprea 2009, Jarić et al. 2015, Roberts et al. 2016 or to plants (Jarić & Gessner 2012, Gessner et al. 2013. Several factors other than threat status can increase scientific attention towards a particular taxonomic group. Those factors most closely related to plants are abundance, range size, range proximity to or overlap with research facilities, and habitat accessibility (Jarić et al. 2019). That is, the prevalence of rarity and endemism of the threatened and DD species (Kunin & Gaston 1997, Pimm et al. 2014 may create additional difficulties for scientific research. First, the locations of rare plants may be very remote and difficult to reach, affecting the likelihood of locating populations in the wild (Royle & Nichols 2003, Gu & Swihart 2004, Chen et al. 2009); this difficulty is aggravated when dealing with life forms that are less conspicuous or have short life cycles (Kéry & Gregg 2003, Moore et al. 2011. This set of constraints may pose greater challenges in tropical environments (Banks-Leite et al. 2014), which are typically highly biodiverse (Wright 2002). Second, these difficulties also require more practical training of researchers in plant identification (Ahrends et al. 2011). Finally, research groups may require more time and consequently more financial resources and effort. As a result, these problems may lead specialists to select species that are more abundant, widely distributed, and close to research facilities (Moerman & Estabrook 2006).
The degree of scientific attention is not only a matter of the choice of specific scientific questions. Scientists' activities are strongly affected by the possibility of obtaining research funding to support particular lines of research, especially in competitive funding environments (Himanen et al. 2009). In addition to the scarce funding for species conservation programs in developing countries (Lawler et al. 2006, Waldron et al. 2013, the degree of investment in conservation-oriented research is often controlled by the preferences and interests of sectors of society outside the academic world (Wil-son et al. 2007). In particular, government funding allocated to species research programs is directed primarily toward vertebrates; similarly, NGOs usually prioritize species with high public appeal (i.e. charismatic species or species with economic value) to concentrate their conservation efforts (Martín-López et al. 2009). As funding follows these societal preferences, scientific attention would follow that preference as well (Jarić et al. 2019). To our knowledge, no studies have examined the influence of sociological factors on project funding for the conservation of Brazilian threatened species. However, the patterns found in our study, such as the strong bias toward commercial species and the lack of relationship to the threat status, suggest that these factors possibly influence how research goals have been set and funded.
Although we recognize that our study was not exhaustive, the database that we constructed is the most comprehensive assessment of scientific attention to threatened plants in the field of biodiversity conservation to date. This allowed us to show that biases and the lack of scientific attention affect research on plants even after an IUCN listing. On the other hand, the development and implementation of the IUCN Red List has again proved to foster positive results for conservation efforts. Surprisingly, its impact on scientific attention can occur within a short period and benefit historically neglected species. However, its impact has proven to be insufficient to reduce biases in research on threatened plants. Our findings indicate the need to design studies that incorporate the information available in national Red Lists and to devote attention to gaps in knowledge of threatened flora. To accomplish this, there is a need for better integration among stakeholders in the biodiversity conservation arena with the goals of the Global Strategy for Plant Conservation and the Aichi Biodiversity Targets regarding scientific knowledge of threatened species.