Biological science practicesSix-decade research bias towards fancy and familiar bird species
Abstract
Human implicit biases towards visually appealing and familiar stimuli are well documented and rooted in our brains’ reward systems. For example, humans are drawn to charismatic, familiar organisms, but less is known about whether such biases permeate research choices among biologists, who strive for objectivity. The factors driving research effort, such as aesthetics, logistics and species’ names, are poorly understood. We report that, from 1965 to 2020, nearly half of the variation in publication trends among 293 North American male passerine and near-passerine birds was explained by three factors subject to human bias: aesthetic salience (visual appeal), range size (familiarity) and the number of universities within ranges (accessibility). We also demonstrate that endangered birds and birds featured on journal covers had higher aesthetic salience, and birds with eponymous names were studied about half as much as those not named after humans. Thus, ornithological knowledge, and decisions based thereon, is heavily skewed towards fancy, familiar species. This knowledge disparity feeds a cycle of public interest, environmental policy, conservation, funding opportunities and scientific narratives, shrouding potentially important information in the proverbial plumage of drab, distant, disregarded species. The unintended consequences of biologists’ choices may exacerbate organismal inequalities amid biodiversity declines and limit opportunities for scientific inquiry.
1. Introduction
Human aesthetic values and preferences have been shaped by evolution, culture and experience. Our colour preferences are thought to be adaptive because humans are typically drawn to colours associated with favourable objects or conditions that confer fitness benefits [1]. For example, humans are drawn to blues, likely because blues are associated with clean water and clear sky [1], and we tend to associate greens with healthy nature and calmness and yellows with happiness [2]. On average, humans are drawn to familiar [3,4] and emotionally positive stimuli [5]. Brain reward centres activate when an object is perceived as aesthetically pleasing [6] and/or familiar [7], resulting in subconscious biases towards ‘beautiful’ and familiar objects that elicit pleasure [5]. Notably, humans have been described as the ‘world’s greatest evolutionary force’ [8] and thus, given our biases, what are the consequences for non-human organisms?
Human value judgements about non-human organisms have contributed to global ‘biological annihilation’ [9] and hence necessitated the concept of conservation as a ‘crisis’ discipline [10]. For example, overexploitation and perceived species value in the global wildlife trade [11–13] have manufactured a species salience hierarchy, effectively resulting in ‘taxonomic chauvinism’ [14]. One of conservation biology’s primary postulates is that all species have intrinsic value [10]; however, value may not be equally distributed among species, leaving practitioners with evidence gaps that compromise conservation efficacy [15]. Our value-based judgements drive species’ conservation designations and resource allocation [16–21], decisions often influenced by species’ aesthetic salience (i.e. human interest in/attention towards species based on visual appeal [22–24]) and familiarity. In general, humans favour species perceived as aesthetically pleasing, such as larger animals [20,25], those with bright [26] or multiple colours [18], and those that are geographically familiar [27,28], consistent with the mere-exposure effect [4] and the familiarity-attraction hypothesis [3]. These subjective preferences, and even preferences suggested by organisms’ names [29], can have tangible consequences for conservation, policy, public interest and scientific research [16,19,30,31], and thus for our broader understanding of biology [32,33].
Birds are the most colourful land vertebrates [34], although few birds have colourful plumage on the human-visible spectrum (i.e. vivid colours such as blue and green are rarer than ‘drab’ colours like brown and grey [24,35]). Birds are also globally ubiquitous and culturally prominent [36] and are a relatively well-studied taxon [14,37]. We leverage the great variation in aesthetic salience and geographic distribution among North American passerine and near-passerine birds to investigate whether research effort, and therefore biological knowledge, is skewed by multiple human biases. On average, humans prefer larger [27,36; but see 24,38], more colourful [24,39–41] and boldly patterned [38,42–44] birds with crests [24,39; but see 36], as well as species with large populations adjacent to humans [24,45,46]. Furthermore, functional magnetic resonance imaging studies show that birdwatchers and ornithologists perceive, recognize and remember birds similarly to human faces [47], especially in locally familiar species [48]. This aspect of human psychology may enhance bias potential, especially given familiarity [4] and/or visual biases [1,2]. However, it is unknown whether and how these biases inherent to the public impact study species selection by scientists, who strive for objectivity.
Here, we systematically quantified parameters empirically demonstrated to vary in human visual preferences of birds to test whether variation in 293 North American passerine and near-passerine species’ aesthetic salience, familiarity and accessibility is associated with variation in research effort. We focused our analyses on male birds because our objective was to assess potential bias in North American ornithological literature, and that literature is known to focus primarily on male birds [49–53]. We calculated breeding range sizes and the number of institutions of higher learning (hereafter, ‘universities’) within each breeding range as proxies for species’ familiarity and accessibility, respectively, to researchers in the continental USA and Canada. We quantified research effort from 1965 to 2020 for each species using an intensive bibliometric approach evaluating >27 000 publications. We hypothesized that bird species with greater aesthetic salience, larger breeding ranges and ranges encompassing more universities are more well studied than bird species that are less aesthetically salient, less widespread and have fewer universities within their ranges. We also investigated differences in aesthetic salience of species listed as federally endangered and those featured on covers of scientific journals, as well as whether species’ common names influenced research effort or were associated with aesthetic salience, familiarity or accessibility. We hypothesized that aesthetically salient species are more likely to be listed as endangered (because species status decisions, like scientific knowledge, could also be influenced by human biases) and are featured more often on journal covers. Further, we hypothesized that species with aesthetically salient-suggestive English common names have more publications than those with drab-suggestive names. We did not have a priori hypotheses regarding the relationships between eponymous English common names and research effort, aesthetic salience, familiarity or accessibility.
2. Methods
(a) Species selection
We studied 293 passerine and near-passerine species in the continental USA and Canada and accounted for changes in taxonomy and nomenclature (for the species list, see dataset [54]; see electronic supplementary material, S1, for species inclusion criteria and nomenclature methodology).
(b) Bibliometrics
We used the number of publications (i.e. a count of the number of peer-reviewed book chapters, journal articles, etc.) indexed by Web of Science (https://www.webofscience.com) as a proxy for research effort (response variable in models). We obtained bibliometric data from Web of Science in February 2022. We searched all publications that were published between 1 January 1965 and 31 December 2020 using species’ English common names and scientific names (nomenclature information in electronic supplementary material, S1). We constructed Boolean search strings to search Web of Science and retrieved pertinent publication data for each species (e.g. titles, abstracts, etc.; aggregated in .csv files; see electronic supplementary material, S2, for search details).
We limited searches to only species’ names that were in the title, abstract or keywords to constrain our initial bibliometric data to those publications in which the target species was included in the study. We subsequently conducted a qualitative assessment of each species’ search output to further constrain our dataset to include only publications in which the species was a primary focal species. This assessment involved reading each title and abstract and, when necessary, reviewing the full publication to exclude those in which the species was not the primary focus. We limited inclusion of publications in our dataset to those that were clearly focused on any aspect of the biology or ecology of ≤3 bird species, also including vagrant records outside of a species’ expected distribution, in addition to non-ecology studies (e.g. kinematics of hummingbirds compared with vehicles; publications in journals such as Bioinspiration & Biomimetics or Nutrition Journal). We excluded search results that were not full publications (e.g. published conference abstracts); we excluded irrelevant publications that had nothing to do with birds; we excluded publications that mentioned a species in the abstract as part of a bird community or for comparative purposes but for which the species was not a focus of the study; and we excluded publications focused on a different species elsewhere in the world with similar or identical common names (e.g. ovenbirds (Furnariidae) of Central and South America and yellow warblers (Acrocephalidae) of Africa; see electronic supplementary material, S2, for filtering details). We acknowledge that our publication filtering did not include determination of whether a publication focused on male birds, female birds or both, and thus we likely included some publications on female birds in our dataset. However, it is unlikely that inclusion of any such publications in our dataset would impact our results, especially due to the traditional literature focus on male birds [49–53].
(c) Scoring species’ aesthetic salience
We created a system to score each species’ aesthetic salience, a cumulative value incorporating documented human visual preferences towards birds. Our aesthetic salience system is based upon human visual perception rather than bird perception because human perception drives the aesthetic salience of birds to people. We obtained a colour-plate illustration (hereafter ‘Sibley image’) of the adult male for each species from the Sibley Birds mobile phone application (https://www.sibleyguides.com/product/sibley-birds-v2-app/) to standardize visual comparisons among images and limit potential biases in image colour, lighting, etc. (issues with comparing photographs). For nearly all species (n = 288), we used profile-view Sibley images of adult males in breeding plumage (see electronic supplementary material, S3, for five exceptions and details). We intentionally focused on males due to documented biases in North American ornithology towards studying [49–53] and collecting [55] male birds, making their plumage more representative of potential factors driving study species choices.
We quantified aesthetic salience for each species’ Sibley image using a systematic, repeatable process we created for assigning cumulative values among seven categories associated with average human visual preference towards birds (table 1). We weighted scores for each category such that they contributed differently to aesthetic salience based on relative importance inferred from the literature (table 1). These features, in order of estimated importance to human perception, included colour score, contrast/pattern, body mass (proxy for size), plumage lightness, iridescence, presence and size of a crest, and presence of other extraordinary features (e.g. tail streamers). See electronic supplementary material, S3, for details.
characteristic | values | reference(s) | significant predictor of no. of publications |
|---|---|---|---|
colour score | |||
(those distinctly present on a bird’s plumage, scored by consensus among authors) | 1.5 each: blue, green, yellow | Lišková & Frynta [40], Echeverri et al. [39], Senior et al. [12] | yes |
1.0 each: red, orange, purple | |||
0.5 each: black, brown, grey, white | |||
contrast/pattern | |||
(coefficient of variation of pixel tonal values measured with contrast CoVAnalysis macro in the PAT-GEOM v.1.0.1 plugin [56] in ImageJ [57]; value divided by 70 to scale contrast value to contribute approximately the same as colour score to aesthetic salience) | mean 3.2; range 1.0−7.5 | Frynta et al. [43], Vall-llosera & Cassey [13], Garnett et al. [38], | yes |
body mass | |||
(mean mass in grams from [58]; ln transformed, then divided by 2 to scale body mass value such that it contributed half as much as colour score to aesthetic salience) | mean 1.6; range 0.5−3.4 | Correia et al. [27], Schuetz & Johnston [36] | no |
lightness | |||
(median tonal value from histogram in Adobe Photoshop Elements 2021; multiplied by 2000 to scale lightness value such that it contributed half as much as colour score to aesthetic salience) | mean 1.7; range 0.4−3.3 | Lišková & Frynta [40] | no |
iridescence | |||
(absence, presence) | 0, 1 | Meert et al. [59] | no |
crest | |||
(absence, presence small, presence sizeable; scored by consensus among the authors) | 0, 0.5, 1 | Echeverri et al. [39] | no |
extraordinary features | |||
(presence of at least one; e.g. tail streamers, waxy tips of feathers, striking eye colour) | 0, 1 | Frynta et al. [43] | noa |
total aesthetic salience | mean 9.1; range 5.5−13.2 |
(d) Estimating species familiarity and accessibility
To test for familiarity and accessibility bias in research effort, we determined each species’ breeding range size and the number of institutions of higher learning (hereafter, ‘universities’) within each breeding range. We calculated 292 species’ breeding range sizes (km2) in ArcMap (v.10.7.1; Environmental Systems Research Initiative) using shapefiles obtained from BirdLife International [60] and clipped ranges that extended beyond mainland USA and Canada to include only the portion of those ranges within this area of interest. In one case, BirdLife International did not have range map shapefiles (i.e. hoary redpoll (Acanthis hornemanni)); thus, we referenced a current range map (https://www.audubon.org/field-guide/bird/hoary-redpoll) to manually draw a breeding range polygon in ArcMap.
To determine USA universities, we started with the US Geological Survey college and university shapefile, composed of all US Post-Secondary Education facilities as defined by the Integrated Post-Secondary Education System (https://www.sciencebase.gov/catalog/item/4f4e4acee4b07f02db67fb39). To determine Canadian universities, we used Google Maps (https://www.google.com/maps) as a reference, searched ‘college’ and ‘university’, scrolled across the map of Canada and manually added points to our Geographic Information Systems (GIS) layer for all entities confirmed as 4 year universities. Points were accurate to approximately 1 km and not to exact address locations, which was adequate for the scale of our analysis. Across both countries, we included only doctoral/research universities, master’s colleges and universities, and baccalaureate colleges (i.e. we excluded community colleges, technical/vocational schools, etc., because they typically have less active research programmes). We did not screen universities to determine whether ornithology research has been conducted there within the period of interest. Rather, within the US university shapefile, we manually screened the attributes table and removed all universities that were unlikely to host ornithology or ecology research. For example, we excluded seminaries; independent medical schools; law schools; art, music, and culinary institutes; business schools; and all 73 University of Phoenix locations (an online university). We also excluded duplicate locations within a city for a university (e.g. multiple campuses across a city or off-campus buildings identified independently in the shapefile), such that each university was represented by one point. We acknowledge that we did not further classify universities or constrain our university dataset to those with biology departments, and thus our proxy assumes similar potential accessibility or research output potential among universities.
(e) Phylogeny and statistical analysis
We obtained a phylogenetic tree from birdtree.org [61]. We pruned the ultrametric Hackett tree (https://birdtree.org/subsets/) to include the 293 species in our study and downloaded 1000 trees to account for phylogenetic uncertainty. We used TreeAnnotator (v.2.6.2; https://beast.community/treeannotator) to generate a single maximum clade credibility tree. Final node heights were estimated using common ancestor heights. We used a phylogenetic comparative approach to quantify the variables that predicted the number of publications for each of the 293 species. Because the response variable (number of publications) consisted of counts, we performed Poisson regression. All continuous traits were standardized using the ‘standardize’ function in the ‘effectsize’ package [62] in R [63], which avoids problems of traits that differ in scale and enhances interpretation.
We applied Markov Chain, Monte Carlo generalized linear mixed models (MCMCglmm) to assess the contribution of each predictor variable to the number of publications using the ‘MCMCglmm’ R package [64]. Prior to modelling research effort, we evaluated correlations among the continuous predictor variables to determine whether there was evidence of multicollinearity. We used the ‘vif’ function in the ‘car’ R package to identify predictor variables with high correlations by calculating variance inflation factors. All variance inflation factor values were <5, suggesting no evidence of multicollinearity. Thus, no highly correlated predictor variables were included in models of research effort. We ran the MCMCglmms including phylogeny as a random effect for 4 00 000 iterations with a burn-in of 10 000 and a thinning interval of 100. As suggested by Hadfield [64], we used uninformative priors. We ran each model with four chains and checked for convergence with the Gelman–Rubin diagnostic [65]. We also calculated phylogenetic heritability, which may be used as an estimate of phylogenetic signal.
We first constructed a model (model 1) that included the number of publications as the response variable with overall aesthetic salience, breeding range size (km2) and number of universities within species’ breeding ranges as predictor variables; the model structure (in R notation) was as follows:
We next assessed the contribution of individual aesthetic components (table 1), as well as main species colour, as predictors of overall aesthetic salience (response) using an MCMCglmm phylogenetic model (model 2):
Based on the results from model 2, we constructed an MCMCglmm (model 3) with the significant individual component predictors of overall aesthetic salience (from the previous model, model 2) to assess their contributions to the number of publications (response). The formula for model 3 was the same as model 2, except it excluded the main colour. Lastly, in our overall final MCMCglmm (model 4), we included the significant-to-research-effort aesthetic predictors (from model 3), breeding range size and number of universities as predictors of the number of publications (response), expressed as:
We estimated marginal means of predictor variables in all MCMCglmms using the ‘ggeffects’ R package [66]. We considered α = 0.05 to indicate statistical significance.
(f) Endangered species
To compare aesthetic salience of species listed as federally endangered with those not listed as endangered, we retrieved US and Canada federal conservation status for each species following US Fish and Wildlife Service’s Environmental Conservation Online System (https://www.fws.gov/southeast/conservation-tools/environmental-conservation-onlinesystem/) list under the Endangered Species Act (ESA; USA) and the Environment and Climate Change Canada list under the Species at Risk Act (Canada). We included a species as endangered if the entire species, one or more subspecies, or a distinct population segment was listed as endangered in either or both countries during our period of interest (i.e. even if the species was recently delisted, e.g. Kirtland’s warbler (Setophaga kirtlandii) in the USA). We compared aesthetic salience of species listed as endangered in each country (nUSA = 9; nCanada = 13) to aesthetic salience of species not listed as endangered in either country (n = 268) using t-tests. Note the total species counts do not match the total birds studied because two species were listed as endangered in both countries.
(g) Journal covers
We reviewed cover images from 29 relevant peer-reviewed journals that publish cover images (mainly photographs but some illustrations)—or did so during some portion of our period of interest—associated with publications in each issue to compare the aesthetic salience of ‘cover birds’ to birds that were not featured on covers using t-tests. We counted the number of instances each of the 293 species was represented in a cover image of a peer-reviewed journal (list in electronic supplementary material, S4), excluding covers that displayed collages of >3 bird species.
(h) Species’ common names
We were interested in whether species’ English common names influenced research effort, a topic not often investigated [29] but warrants consideration because names carry weight and potential consequences [67]. We were initially interested in comparing research effort between species with drab-sounding names and those with aesthetically salient-sounding names, and following the American Ornithological Society’s November 2023 announcement (https://americanornithology.org/american-ornithological-society-will-change-the-english-names-of-bird-species-named-after-people/) to replace English bird names that are directly attributed to humans (i.e. eponyms or honorifics [68]), we also investigated research effort between species with eponymous names versus those with non-eponymous names. We noted which of the 293 species had current common names that included words/colours suggesting drab appearance (e.g., ‘brown’, ‘dusky’, ‘gray’, ‘plain’) and those with words/colours in their names suggesting aesthetic salience (e.g. ‘blue’, ‘gold’, ‘green’; see dataset [54]). For species whose common names included both a drab-suggestive and a salience-suggestive word, we considered them in the aesthetically salient-sounding group. We noted which of the species in our dataset had current common names that were ‘primary eponyms’; 46 species had ‘primary eponyms’ (n = 45 with genitive possessive construction; n = 1 without: Blackburnian warbler (Setophaga fusca)). All other species, including those with ‘secondary eponyms’, were marked as not having eponymous names (n = 247). We compared research effort between species with drab-suggesting (n = 41) versus salience-suggesting names (n = 64) using a one-tailed t‐test because we expected any difference to be in the predictable direction. We compared aesthetic salience, familiarity, accessibility and research effort between species with eponymous names and those without eponymous names using two-tailed t-tests. For details, see electronic supplementary material, S5.
3. Results
We downloaded publication information from 27 254 publications, 50% (13 620/27 254) of which we retained in our dataset following filtering. The median number of publications per species was 19 ± 79 s.d., and research effort among species was heavily skewed, ranging from 0 to 597 publications (figure 1b). The three species with the most publications were the tree swallow (Tachycineta bicolor; n = 597; figure 1), the red-winged blackbird (Agelaius phoeniceus; n = 499) and the song sparrow (Melospiza melodia; n = 469). Ten species (3%) were the primary focus of zero publications (e.g. black-chinned sparrow (Spizella atrogularis), crissal thrasher (Toxostoma crissale), Philadelphia vireo (Vireo philadelphicus)). Values for aesthetic salience ranged from 5.5 to 13.2 (median and x̄ = 9.1 ± 0.10 s.e.; figures 1b and 2, electronic supplementary material, figure S1). The species with the highest aesthetic salience (13.2) were the bohemian waxwing (Bombycilla garrulus) and the tree swallow, whereas the black swift (Cypseloides niger) and the chimney swift (Chaetura pelagica) had the lowest aesthetic salience (5.5).
Figure 1. Dataset overview indicating skewed research effort among 293 bird species from 1965 to 2020. (a) Examples of species with high (tree swallow; Tachycineta bicolor), medium (Baltimore oriole; Icterus galbula) and low (gray vireo; Vireo vicinior) aesthetic salience (visual appeal to humans; see §2). Species’ breeding range maps (grey fill) were accessed from BirdLife International [60]. Black dots depict universities within species ranges. (b) Histograms (right) indicate data spread of the number of universities within species’ breeding ranges, number of publications per species and aesthetic salience. Black dashed lines in histograms denote dataset medians. Grey, orange and blue dashed lines denote where the three species occur in the data distribution. Note that the Baltimore oriole’s aesthetic salience was equal to the mean and median aesthetic salience; thus, the orange line in the aesthetic salience histogram also represents the dataset mean and median. Species illustrations by Silas E. Fischer.
Figure 2. Ancestral state estimation of aesthetic salience (visual appeal to humans; see §2) for 293 bird species on a time-calibrated phylogeny. Aesthetic salience values range from 5.5 (brown branches) to 13.2 (blue branches).
(a) Predictors of research effort
An initial MCMCglmm (model 1) revealed significant effects of aesthetic salience, breeding range size and number of universities on research effort (pMCMC < 0.01; electronic supplementary material, table S1). Thus, species with higher aesthetic salience, larger breeding ranges and breeding ranges containing more universities had more publications (figure 3). Specifically, species in the top 10% of aesthetic salience were studied 3.0× more than birds in the bottom 10%, on average (figure 3a). Species in the top 10% of breeding range size and university abundance were studied on average 3.8× and 3.5× more often than species in the bottom 10% of those categories, respectively (figure 3b,c; see also electronic supplementary material, figure S2, depicting the relationship between range size and universities). We analysed the relative influence of seven individual aesthetic characteristics (table 1), as well as main colour, on overall aesthetic salience (model 2), and found a non-significant effect of main colour on aesthetic salience; all remaining variables were significant contributors to aesthetic salience (pMCMC < 0.0001; electronic supplementary material, table S2). A subsequent MCMCglmm (model 3) assessing specific contributions of the significant aesthetic characteristics (from model 2) to the number of publications revealed that three were significant predictors, i.e. colour score, contrast/pattern and extraordinary features (pMCMC < 0.05; electronic supplementary material, table S3). A final MCMCglmm (model 4) analysing the influence of colour score, pattern, extraordinary features, breeding range size and number of universities revealed that all variables (pMCMC < 0.05) except extraordinary features predicted research effort (table 2). Breeding range size and the number of universities had the largest effect sizes on research effort, followed by colour score and contrast/pattern (pMCMC < 0.001; table 2).
(b) Endangered species
Of 293 focal species, 9 were listed as endangered in the USA and 13 were listed as endangered in Canada (2 were listed in both countries). Aesthetic salience of endangered species in the USA (x̄ = 9.98 ± 0.30 s.e.) was higher than that for species not listed as endangered in either country (x̄ = 9.04 ± 0.10 s.e.; t‐test, p = 0.01). Aesthetic salience of endangered species in Canada (x̄ = 9.45 ± 0.42 s.e.) was relatively similar to aesthetic salience of species not listed as endangered in either country (x̄ = 9.04 ± 0.10 s.e.; t‐test, p = 0.36). The difference between countries appeared to be driven by Canada recognizing the species with the lowest aesthetic salience in our dataset (black swift; aesthetic salience = 5.5) as endangered. Excluding the black swift, the other 12 species listed as endangered in Canada had higher aesthetic salience than species not listed as endangered in either country (t‐test, p = 0.03).
(c) Species on journal covers
Of 293 focal species, 94 (32%) were featured on journal covers. Species featured on journal covers had higher aesthetic salience (x̄COVER = 9.48 ± 0.15 s.e.) than those not featured on covers (x̄NON-COVER = 8.90 ± 0.12 s.e.; t‐test, p = 0.003).
(d) Consequences of species’ common names
Within our dataset, birds with eponymous names (n = 46) did not differ from other birds (n = 247) in aesthetic salience (t‐test, p = 0.22) but did have 2.2× smaller breeding ranges (t‐test, p < 0.0001), 4.5× fewer universities in their breeding ranges (t‐test, p < 0.0001) and received less than half of the research effort (x̄ = 20 publications; t‐test, p < 0.0001) compared with birds without eponymous names (x̄ = 51 publications). In addition, birds with names including words suggesting they might be visually drab (n = 41) were studied less than half as often (x̄ = 30 publications) as those with words suggesting they might be aesthetically salient (n = 64; x̄ = 63 publications; t-test, p = 0.02).
Figure 3. Predictors of research effort in 293 North American bird species from 1965 to 2020. Marginal mean number of publications (i.e. research effort) for the three significant predictor variables: (a) aesthetic salience (visual appeal to humans), (b) number of universities within species’ breeding ranges and (c) breeding range size (km2). Species with higher aesthetic salience, breeding ranges containing more universities and larger breeding ranges had more publications. Note that predictor variables are scaled to a mean of 0 and s.d. = 1.
source | posterior mean | LCI | UCI | effective sample size | pMCMC |
|---|---|---|---|---|---|
intercept | −0.073 | −0.421 | 0.268 | 4662 | 0.681 |
colour score | 0.142 | 0.021 | 0.253 | 3386 | 0.013 |
pattern | 0.151 | 0.034 | 0.270 | 4145 | 0.011 |
extra feature | −0.362 | −0.754 | 0.051 | 4145 | 0.073 |
number of universities | 0.192 | 0.056 | 0.330 | 4900 | 0.007 |
range size (km2) | 0.281 | 0.150 | 0.425 | 4419 | <0.001 |
4. Discussion
Our analyses revealed heavily skewed research effort towards aesthetically pleasing birds that occur in larger breeding ranges encompassing more universities, indicating a substantial influence of human implicit bias on the focus of ornithological study in the USA and Canada over nearly six decades. Combined, our variables representing visual appeal, familiarity and accessibility accounted for 45% of the variation in published research, adding to a growing body of evidence that signals pervasive, non-random knowledge gaps in biological research foci [14,22,25,32,37,69–71], including ornithology [45,72–76]. Science plays a pivotal role in a complex positive feedback loop dictating public interest and awareness, policy, conservation status and decisions, and funding disparities [30], all subject to bias. We do not intend to imply that all ornithological research is biased. However, any subjectivity in study species choice may feed a snowball of bias accumulation by iteratively producing greater research effort, attention and citations for a select few species because of their appearance and geographic convenience rather than critical research gaps or conservation needs [77], similar to circularity in funding and perceived biodiversity [78]. This bias accumulation, by which some species are over-represented in science and the public spotlight [22,71,79], can lead to wasted conservation efforts by diverting resources away from under-studied species lacking basic information [80]. In a highly competitive research environment where success is often measured by money coming in and publications going out, scientists may have little choice but to perpetuate this problem—even if they are aware of it—by studying popular, well-funded species rather than risk career advancement. Such bias-driven feedback loops [30] emulate the ‘Matthew effect’ (i.e. the rich get richer and the poor get poorer [81]) and further divide the organismal ‘haves’ and ‘have-nots’ in a world with limited research and conservation resources and accelerating biodiversity loss. Furthermore, because scientific and societal interests are enmeshed [30,31,77], lack of research on visually unremarkable and unfamiliar birds may ultimately result in their ‘societal extinction’ [82].
It was not our objective to explain all variation in ornithological research effort, which could also be associated with charismatic behaviours, vocalizations, cognition, habitat associations, value as game species, model species, nesting substrate and other factors. For example, migratory birds are studied 10× more than resident species [74], a factor we did not consider here. However, we note most of these other factors are not free from entanglement with human biases (e.g. migratory species also tend to be more colourful and have lighter-coloured feathers than non-migrants [83]). Unexpectedly, we found that body mass was not associated with research effort. However, our study was constrained to passerines and near-passerines (i.e. a narrow range of body masses) and therefore does not necessarily contradict studies detecting mass-related biases when comparing a wider range of taxa (e.g. [70,74,84]). There is not broad consensus in the literature regarding human preferences for avian body mass; some studies suggest that small birds are most aesthetically salient [24,38], whereas others suggest the opposite [27,36]. We found no association between research effort and presence or size of crests, consistent with Schuetz & Johnston [36] but not Echeverri et al. [39] or Santangeli et al. [24]. Further, we did not find a meaningful association between research effort and plumage lightness, iridescent plumage or extraordinary features, partially contrasting with Santangeli et al. [24]. We note that those studies included a broader range of species and investigated public preferences [24] and/or birdwatchers’ interests [36,39], who may exhibit even stronger aesthetic biases as a group compared with scientists [25,85]. Further, some preferences may be taxon and/or region specific (e.g. [86]). Future work could investigate how colour elaboration [24], colour uniqueness [12] or morphology (e.g. body shape, eye-to-body ratio [40]) influence research effort.
In our dataset, species in the top 10% of aesthetic salience were studied 3.0× more than birds in the bottom 10%, on average. We purposefully only assessed male plumage because the North American ornithological literature has traditionally focused on males. Female birds have been historically overlooked in most, if not all, aspects of North American ornithology [49–53] for many reasons, one of which we suspect is their typically drabber plumage [24,87]. Cooper et al. [55] showed that natural history museums harbour disproportionately more male bird specimens, a key suspected reason being ‘…deliberate selection for large, ‘impressive’ male specimens, especially where males are larger or more colourful than females…’ Strikingly, this skew was most evident in sexually dimorphic songbirds [55], suggesting that males’ on-average more aesthetically salient plumage [24] may be an important driver of the observed collection bias. Our results also suggest that this may be true—and arguably could contribute to the male-centric paradigm in North American ornithology—but further assessment is required.
Our finding that both breeding range size (familiarity) and the number of universities within a species’ range (accessibility) were independently significant predictors of ornithological research effort in the USA and Canada corroborates previous studies assessing geographic [22,69,72,74–76] and urban research bias [88] across many taxa. This disparity suggests that many ornithologists may be selecting study species based on logistical convenience, likely resulting from limited funding among other factors [16]. Much of this bias may stem not from researchers themselves, but from the interconnected structures of funding, policy and public interest [30,31,77] that can shape research agendas [89]. Funding agencies often dictate study species choices by identifying priority species and/or topics [90], sometimes influenced by politics, public interest and their intersection [91]. Inequitable access to research resources, for instance at smaller institutions (i.e. prestige bias [92]) or those with lesser research capacity [15,93], may amplify research unevenness among species. Thus, our aim is not to assign blame to any single link in the web but to question the broader framework enabling non-random omissions in the first place.
Even some species with relatively high aesthetic salience, such as Scott’s oriole (Icterus parisorum), have received scant research attention, likely due to occurring in a small breeding range containing few universities (aesthetic salience = 9.7, publications = 0, range size = 23rd percentile, universities = 20th percentile). Thus, drab species occurring in small breeding ranges encompassing few universities, such as many birds in Southwestern USA drylands, are rarely represented in scientific literature (e.g. gray vireo (Vireo vicinior; aesthetic salience = 5.7, publications = 13, range size = 17th percentile, universities = 10th percentile; figure 1a), Bendire’s thrasher (Toxostoma bendirei; aesthetic salience = 6.2, publications = 1, range size = 21st percentile, universities = 18th percentile), Lucy’s warbler (Leiothlypis luciae; aesthetic salience = 6.2, publications = 4, range size = 12th percentile, universities = 12th percentile)). We recognize that range size may relate to other factors in addition to familiarity; e.g. larger ranges may provide ample heterogeneity in abiotic or biotic factors of interest that could render them more appropriate systems within which to test particular hypotheses. Likewise, other factors may influence human familiarity with certain species—such as the portrayal of some species as sports team mascots [36] or their presence in visual media [94]—which we did not incorporate in our familiarity proxy. We emphasize that the disparities we report are among birds—a well-studied taxon [14,37]—specifically those that occur within the USA and Canada, a well-studied region [15,70,74] of the Global North with substantial research capacity [93]. Research choices in other regions may be driven by other context-specific factors not explored here. While we are explicitly not advocating for parachute science (see [93]), we suspect that the field of ornithology has omitted exciting and important knowledge by excluding drab and geographically distant birds from scientific inquiry.
We showed that birds featured on journal covers had higher aesthetic salience compared with non-cover species, which is unsurprising regardless of any research effort bias because journal editors and publishers likely select cover images that they believe will catch readers’ attention (e.g. [94]), thus exploiting human biases. Papers featured in journal cover images often garner greater Altmetric scores (https://www.altmetric.com; [95,96]) and more citations relative to non-cover publications in the same journal [95,97], and thus can perpetuate the bias feedback loop. This bias recognition is similar to selecting aesthetically pleasing species as ‘poster children’ or ‘flagships’ for conservation and fundraising campaigns [38], but these terms are not biological concepts [98], and care should be taken that they are not exploited to justify erroneous conservation or research activities. We did not explore whether aesthetic salience was associated with species’ h-indices (i.e. a measure of publication impact [84]), but previous work indicates that species’ h-indices and research effort are typically positively correlated [22,45,85].
The higher aesthetic salience of endangered birds may reflect aesthetic bias in the process and support for petitioning, listing, prioritizing and expending resources for species recovery [16,17]. Decision-making under the ESA relies on the ‘best available science’ [99], but evaluation processes stall when ‘candidates’ for listing lack sufficient scientific data [21]: it is difficult to accurately assess threat status, much less implement meaningful, evidence-based conservation actions [100], for overlooked species [15,70,73], many of which were of low aesthetic salience in our dataset. Alarmingly, well-researched species are often assessed as facing a greater number of threats [101,102] than less-studied species (e.g. those listed as ‘data deficient’) under the International Union for Conservation of Nature. Thus, it is plausible that aesthetically salient species may be more often categorized as endangered, rendering it possible that species assessment, prioritization and listing processes may be directly influenced by visual biases in addition to other perception-related factors (e.g. detectability or familiarity [98,103]).
Human visual perception, and perhaps geographic bias, can even permeate the names we attribute to organisms [34]—as assigning names is an inherently creative [67] and oftentimes arbitrary act [29] but importantly is not bias free. Species names can evoke emotions, including fear, disgust and apathy [104], which may hamper research effort [29] and/or conservation potential [105]. English common names, particularly in birds compared with other taxa, are standardized and widely used by both scientists and the public [68], and 57% of birds’ English common names worldwide are based upon physical traits (often plumage colour [106]). Taken together, these aspects of avian naming conventions could propagate opportunities for inequity among species via multiple feedback pathways [30]. We found that species with drab-suggestive names and species with eponymous names were both studied less than half as often as other birds in our dataset; eponymously named species also tended to be more geographically remote and near fewer universities. Given these results, we encourage committees tasked with renaming species to reflect upon how names may influence research effort and to consider potential consequences of naming species based on human visual perception.
When Charles Darwin embarked on his famous expedition aboard the HMS Beagle, he was enthralled by, for example, ‘beautifully coloured’ Brazilian planarians [107, p. 27] and ‘brilliantly coloured’ birds of Maldonado [107, p. 40] but was unimpressed by the drab organisms inhabiting the Galápagos Islands (e.g. ‘…The insects…are small-sized and dull-coloured…All the plants have a wretched, weedy appearance, and I did not see one beautiful flower…none of the birds are brilliantly coloured’ [107, p. 381]). Hence, Darwin may have initially thought little of the small, drab-plumaged Galápagos ‘finches’ (Thraupidae), first described by a previous explorer as ‘…not remarkable for their novelty or beauty’ [108, pp. 8−9]. Indeed, Darwin did not even correctly record the localities of the representative ‘finch’ specimens he collected [109]. Thus, his interest in the ‘finches’ was largely retrospective [108,109]. Had Darwin not studied this system, now a textbook example of natural selection and adaptive radiation, due to his initial biases against the ‘…drab sameness of their plumage’ [110, p. 434], where would the biological sciences be today? What would we have missed out on without Darwin’s insights and all subsequent work informed and inspired by it? With a growing understanding that awareness is the first step in overcoming implicit biases in many areas of human life, it is our sincere hope that a concerted effort will emerge in biology to consider what we may be missing by neglecting visually unremarkable and geographically distant species across the tree of life.
Ethics
This work did not require ethical approval from a human subject or animal welfare committee.
Data accessibility
Data and code are available on Dryad [54].
Supplementary material is available online [111].
Declaration of AI use
We have not used AI-assisted technologies in creating this article.
Authors’ contributions
S.E.F.: conceptualization, formal analysis, investigation, methodology, visualization, writing—original draft, writing— review and editing; J.G.O.: conceptualization, investigation, methodology, writing—original draft, writing—review and editing; A.M.L.: conceptualization, investigation, methodology, writing—original draft, writing—review and editing; D.M.: formal analysis, visualization, writing—original draft, writing—review and editing; H.S.: conceptualization, formal analysis, investigation, methodology, supervision, writing —original draft, writing—review and editing.
All authors gave final approval for publication and agreed to be held accountable for the work performed therein.
Conflict of interest declaration
We declare we have no competing interests.
Funding
D.B.M. was supported by National Science Foundation grant DEB 1950636. S.E.F. and A.M.L. were supported by The University of Toledo University Fellowship. S.E.F. was also supported by the Out to Innovate LGBTQ+ Career Development Fellowship.
Acknowledgements
We are grateful for initial input on study conceptualization and design from B.E. Carpenter, E. Zeigler, M. Rizk and T.L. Spanbauer. We thank J. Refsnider, J. Zubcevic, C.E. Nemes, G.R. Kramer, M. Neiman and four anonymous reviewers for comments on the manuscript.
