Geographical gradients of sexual selection

To visualise global patterns, we mapped sexual selection scores for 9,836 bird species with geographical range data using standard GIS approaches (see Methods). This procedure generates mean sexual selection scores for all species occurring in each grid cell, revealing lower average levels of sexual selection in tropical regions (Fig 3A). To account for spatial autocorrelation among neighbouring grid cells, we modelled the relationship between latitude and average sexual selection for each cell using spatial simultaneous autoregressive (SAR) models. The results confirmed that sexual selection follows a latitudinal gradient, increasing significantly towards higher latitudes (Fig 3B). However, when we used the same procedure to assess geographic biases in knowledge, we found that average data certainty also increased in the temperate zone, with sexual selection scores being particularly robust in North America and Europe, where many bird species have been studied intensively for decades (Fig 3C and 3D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Geographical distribution of sexual selection in birds.

(a) Worldwide variation in sexual selection scores for 9,836 bird species included in a global phylogeny (www.birdtree.org [50]), averaged for all species occurring in 5-km grid cells based on breeding range maps. Sexual selection was scored based on mating behaviour from monogamy (scored 0) to extreme polygamy (scored 4; Table 1). (b) Relationship between avian sexual selection and breeding latitude. (c) Geographical variation in data certainty, ranging from no evidence (scored 1) to strong evidence (scored 4; see Table A in S2 Text). (d) Relationship between data certainty and latitude. In (b) and (d), points denote mean sexual selection per 200-km grid cell (Behrmann projection); dashed lines were generated from spatial simultaneous autoregression (SAR) models predicting mean sexual selection strength. Additional SAR models on a subset of species with higher-quality data (scored 3–4 for data certainty) showed similar patterns (S1 Text and Table C in S2 Text). To reduce noise, cells with <10 species were excluded from all plots and models. To aid visualisation, plots were coloured using discrete intervals with an equal number of cells. Results are plotted using geographical range polygons provided by BirdLife International (www.datazone.birdlife.org) cropped to Earth’s land-surface using the BIO1 climate layer (www.chelsa-climate.org). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

https://doi.org/10.1371/journal.pbio.3002856.g003

These patterns reflect a clear correlation between improved data quality and higher sexual selection scores (S2 Fig), suggesting that geographical variation in ornithological knowledge could potentially explain the overall latitudinal gradient in sexual selection. This does not appear to be the case, however, because the gradient in sexual selection remains strongly significant when we restrict the analysis to 7,592 species with moderate to high data certainty (scored 3–4; S3A and S3B Fig). Even when we restricted the analysis to the highest category of certainty (n = 2,851 species), the positive relationship between sexual selection and latitude remained, albeit weaker because highly polygamous tropical species such as cotingas and manakins are more easily classified with certainty than monogamous species, explaining the accumulation of high-certainty species in the tropics (S4 Fig) and the relatively high average sexual selection score among the best-known Amazonian birds (S3C and S3D Fig). The positive latitudinal gradient in sexual selection scores identified by our analyses may seem counter-intuitive given the occurrence of largely monogamous clades such as seabirds and raptors at higher latitudes, and reflects intense sexual selection reported in other temperate and polar species, including both passerines [62] and non-passerines [63–65].

Univariate models predicting sexual selection

To understand the mechanisms driving latitudinal gradients in sexual selection, we re-examined geographical patterns through the lens of ecological traits. This approach revealed contrasting gradients when comparing primary consumers (herbivores, frugivores, granivores, and nectarivores) with secondary consumers (omnivores, carnivores, and scavengers). The positive latitudinal gradient in sexual selection is removed (nonsignificant) in primary consumers (Fig 4A and 4B) and strongly reversed (negative) in an ecologically important subset: frugivores (Fig 4C and 4D). Conversely, we found a strong positive gradient in secondary consumers (Fig 4E and 4F), particularly the invertivores, which comprise approximately 40% of the world’s birds (Fig 4G and 4H). Thus, the overall positive gradient across all birds is predominantly driven by the species-rich invertivores, concealing an opposite (negative) gradient in fruit-eating species (frugivores), which constitute a smaller proportion (approximately 10%) of global bird diversity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Global distribution of sexual selection partitioned by trophic level and diet.

Upper panels show strength of sexual selection in primary (1^ry) consumers mapped globally (a) and plotted against latitude (b), as well as for frugivores only (c, d). Lower panels show strength of sexual selection in secondary (2^ry) consumers mapped globally (e) and plotted against latitude (f), as well as for invertivores only (g, h). We treated omnivores, carnivores and scavengers as secondary consumers. Sexual selection was scored from monogamy (0) to extreme polygamy (4; see Methods). For mapping, we averaged sexual selection scores in each 5-km grid cell based on all species with breeding ranges overlapping the cell. To aid visualisation, maps were coloured using discrete intervals with an equal number of cells. In scatterplots, points represent mean sexual selection per 200-km grid cell; dashed lines were generated from spatial simultaneous autoregression (SAR) models predicting mean sexual selection strength. Sensitivity analyses on a subset of species with higher-quality data (scored 3–4 for data certainty) showed similar patterns (Table D in S2 Text). To reduce noise, cells with <10 were excluded from all plots and models. Results are plotted using geographical range polygons provided by BirdLife International (www.datazone.birdlife.org) cropped to Earth’s land-surface using the BIO1 climate layer (www.chelsa-climate.org). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

https://doi.org/10.1371/journal.pbio.3002856.g004

Switching focus to another key ecological trait, migration, we found that both migratory and non-migratory species showed a significant positive gradient in sexual selection across latitude (S5 Fig), aligning with the general trend observed in all birds. A similar positive trend was also found in territorial species, with sexual selection again increasing towards high latitudes from a low baseline in the tropics (S6 Fig). The gradient for non-territorial species was different, either flat when we accounted for spatial autocorrelation in SAR models (S6 Fig) or reversed when we modelled sexual selection at the species level using Bayesian regression models (S7 Fig). In all other cases, the results of our SAR models were replicated by species-level models (S7 Fig) and remained unchanged when we restricted analyses to well-known species (Table C in S2 Text), indicating that our results are not biased by modelling approach or data certainty. These highly consistent results suggest that a combination of ecological strategies relating to diet, migration, and resource defence shape global trends in avian sexual selection. However, the relative importance of each trait is difficult to determine based on raw patterns given our finding that sexual selection is phylogenetically conserved in birds (Fig 5A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. The role of climate in regulating avian sexual selection.

(a) A family-level consensus phylogenetic tree showing the distribution of sexual selection scores across 194 families (n = 9,988 species). Terminal bars show average strength of sexual selection estimated for each family; longer bars indicate higher scores (see Methods). Coloured segments demarcate major clades; icons depict representative families. Panels (b–d) show average sexual selection scores for each family, partitioned by trophic level (b), migratory behaviour (c), and territoriality (d); black points show predicted mean values for each group from species-level Bayesian phylogenetic models; whiskers show 95% credible intervals. In (b), 1^ry = primary; 2^ry = secondary. Species traits shown in (b–d) are all hypothetically linked to temperature seasonality, which is also related to sexual selection scores (e); dashed line shows model fit from a species-level Bayesian phylogenetic model; shaded area denotes 95% credible intervals from model predictions. In (b–e), data points for each family are coloured by major clades (see phylogeny), with size and transparency of each data point scaled to within-family species richness. Statistics show effect size and p-value comparing main effects with reference groups: secondary consumer, no migration, and no territoriality, respectively. For ease of visualisation, data shown in panels are family-level averages, which do not fully reflect the stronger species-level correlations reported in statistics. Running the same models on high-certainty data produced similar results (Table F in S2 Text). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609. Icons were generated using R (rphylopic package) and are under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

https://doi.org/10.1371/journal.pbio.3002856.g005

To evaluate the relative role of species traits while accounting for phylogenetic non-independence, we re-examined the correlation between traits and sexual selection using Bayesian phylogenetic models. As a first step, we ran a species-level model including trophic level as a binary predictor, with phylogenetic co-variance as a random effect (see Methods). We then repeated the same univariate model for migration and territoriality. All 3 ecological traits emerged as significant predictors (Tables E and F in S2 Text), with the intensity of sexual selection increasing in primary consumers (Fig 5B), migratory species (Fig 5C), and non-territorial species (Fig 5D). These ecological traits are all strongly tied to climatic seasonality [20,66], which we also found to be a powerful predictor of global trends in sexual selection score (Fig 5E and Tables E and F in S2 Text). This raises 2 important questions: first, whether any particular ecological trait is driving the overall patterns, and second, whether climatic factors play an important role, either directly or via their effect on species ecology.

Multivariate models

To examine the relative roles of ecological and climatic factors, we included all predictors in a multivariate Bayesian phylogenetic model (Fig 6 and Table E in S2 Text) along with 2 key interactions accounting for complex relationships between diet and territoriality [20] and between diet and seasonality [67]. We found that the strongest constraint (negative effect) on sexual selection was territoriality, with significantly lower levels of sexual selection in species defending seasonal or year-round territories. Conversely, the strongest driver (positive effect) was climatic seasonality, with significantly higher levels of sexual selection in species breeding in the most variable climates. While accounting for both these effects, we found that trophic level and migration also retained their independent significant positive effects on avian sexual selection (Fig 6 and Table E in S2 Text).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Relative roles of ecology and climate as drivers of sexual selection in birds.

Results shown are from Bayesian phylogenetic models testing drivers of sexual selection in 9,836 species. Predictors include 3 life history variables (green), 1 climatic variable (pink), and 2 key interactions between diet and the dominant effects (territoriality and seasonality; blue). 1ry consumer = primary consumer. The reference groups for the 3 categorical predictors are as follows: secondary consumer; no migration; and no territoriality, respectively (see Methods for definitions). Models were run on a sample of 50 phylogenetic trees extracted from www.birdtree.org [50], grafted to the Prum and colleagues [61] genomic backbone. Dots show mean effect size estimates from 12,500 posterior draws. For each effect, broad bases of whiskers show 66% credible intervals; narrow tips of whiskers show 95% credible intervals. Coloured distributions indicate the spread of effect size estimates, generated from a sample of 1,000 posterior draws. Full statistical results are presented in Table F in S2 Text. The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

https://doi.org/10.1371/journal.pbio.3002856.g006

Although the independent effect of trophic level on sexual selection was relatively small, this ecological trait appears to have a key role in shaping variation in sexual selection among species through significant interactions with both territoriality (β = 0.80, 95% CI = 0.08, 1.51) and climatic seasonality (β = −0.75, 95% CI = −1.33, −0.18; Fig 6 and Table E in S2 Text). In other words, dietary factors regulate the effects of territoriality and seasonality, which in turn have different implications for species at different trophic levels. We found that a phylogenetic model incorporating all ecological traits, along with the effects of seasonality and the key interactions with diet, explained a very high proportion of variation in sexual selection (R² = 0.90; 95% CI = 0.88, 0.92). The proportion of variance explained was equally strong (R² = 0.92; 95% CI = 0.91, 0.94) when we ran a more conservative analysis focusing exclusively on species with high-certainty data (S8 Fig and Table F in S2 Text).

Sexual selection in the context of climate change

Our findings indicate that geographical variation in sexual selection may be sensitive to shifts in seasonality projected to occur worldwide and particularly in mid- to high latitudes [81]. In mid-latitudes, reduced climatic fluctuations and milder winters may lead to increased survival of resident birds, in parallel with year-round availability of insects, nectar, and fruit. This can reduce selection for migration [74], causing migratory bird populations to become sedentary [82], while also increasing the duration of breeding seasons for resident species [83]. Our results suggest that these behavioural changes will dampen sexual selection, in line with previous studies showing that higher temperatures [77] and aseasonal climates [84,85] can modify competition for mates and mating opportunities. Nonetheless, the precise impacts of climate change on mechanisms of sexual selection and their adaptive implications [4,86–88] are hard to predict because lengthening warm seasons have variable and species-specific effects on reproductive phenology in both plants [89] and animals [83,90].

Sexual selection scores for all birds: Validity, robustness, and research priorities

Our analyses offer the most comprehensive synthesis of geographical variation in sexual selection, estimated for 10,671 bird species, including all 9,988 valid species aligned with a global phylogeny [50]. We provide this information on sexual selection, data sources, and data certainty for all species in S1 Data, representing a major step change in data availability from the most recent compilations of sexual selection scores, covering 3,250 passerine species [32,48]. We extended these earlier data sets, not only in terms of species coverage, but also by scoring sexual selection based on a wider accumulation of evidence, including molecular evidence of polygamy and extra-pair paternity.

We found that these revised scores correlated relatively weakly with a widely used metric of sexual selection in birds: residual testes mass (Fig 2A). One reason for this weak relationship is that residual testes mass appears to decline in species with the highest levels of sexual selection. This makes sense because females tend not to remate in lekking systems [91] and males consequently invest heavily in pre-copulatory sexual competition [92]. Thus, testes mass reflects sperm competition but is a relatively crude metric for sexual selection. In contrast, our scores correlated more strongly with 2 other metrics—Bateman gradient and I_S (Fig 2B and 2C)—which provide a more complete estimate of the strength of sexual selection. Overall, these correlations suggest that our scores provide a robust metric of variance in mating success, at least in cases where sufficient evidence is available.

Our results also reveal a strong latitudinal gradient in data certainty (Fig 3C and 3D), reflecting a long-standing bias towards ornithological research in the temperate zone [70] as well as the difficulty of tracking down relevant literature published in languages other than English. However, we note that most high-certainty species in our data set are found in the tropics (S4 Fig), suggesting that biases in knowledge to some extent reflect the sheer number of tropical species compared with higher latitudes. The fact that high-certainty scores are distributed across all latitudes (S4 Fig) is crucial, allowing us to re-run our analyses with a more conservative data set, confirming that all our main results are robust to biases in knowledge (Tables C–F in S2 Text).

We hope that the open release of comprehensive estimates of sexual selection for over 10,000 bird species provides a template for further macro-scale research. General hypotheses linking sexual selection to evolutionary mechanisms and ecological contexts [93,94] can now be examined at unprecedented scale. We also hope our global data set can re-energise efforts to investigate avian mating systems in data-poor regions. In particular, the distribution of bird species with low-certainty data (Figs 3C and S4) offers a detailed map of research priorities, highlighting the urgent need for observational and molecular studies, especially in the tropics, to quantify polygamy and EPP in species with low to moderate levels of sexual selection.

Conclusions

Our analyses show that global-scale gradients in avian sexual selection are driven largely by temperature seasonality, with the strength of selection also strongly constrained by territoriality, promoted by migration, and modified by interactions with dietary niche. This complex but consistent interplay of environmental factors and species traits helps to clarify some of the fundamental yet elusive eco-evolutionary processes driving global variation in mating systems and sexual traits in animals [12]. On one hand, we identify context-dependent associations that provide a clear explanation for the conflicting latitudinal patterns reported in previous studies, thus helping to resolve a long-standing debate about the validity of latitudinal gradients in sexual selection and their underlying mechanisms [13,15]. More broadly, these analyses establish a standardised metric for addressing major research objectives, including the role of sexual selection in speciation [3] and the evolutionary impacts of environmental change [82,95,96].

Quantifying sexual selection

We scored sexual selection for all bird species using information published in primary and secondary literature, building on previous estimates of sexual selection [32,48,58,97]. Most information at family- and species-level was extracted from regional or global handbooks, including the Handbook of the Birds of the World series [98] with recent updates [99]. When published information was inconclusive, we used expert knowledge to guide scoring decisions.

To assign scores, we adapted a well-established system based on the estimated proportion of polygamous individuals of either sex in the population [32,39–41]. By our definition, monogamy involves both sexes maintaining the same breeding partner for the duration of a reproductive event or season, and thus includes both sequential and perennial (lifelong) monogamy. In data-poor cases, we used observations of breeding behaviour, including pair-territoriality and offspring provisioning roles, to infer the degree of social monogamy and polygamy. In addition, we further developed previous approaches by integrating information on EPP and display behaviour where possible. We assumed that high rates of EPP are associated with larger variance in male reproductive success [21], and thus reflect elevated levels of sexual selection even when species are socially monogamous. Assignments were made based on quantitative estimates or textual descriptions of observed polygamy, extra-pair behaviour, or behavioural displays, with the addition of genetic estimates of EPP data, now available for a growing number of species [49].

Using the full set of criteria described in Table 1, we assigned sexual selection scores of 0 (strict monogamy), 1 (frequent monogamy), 2 (regular polygamy), 3 (frequent polygamy), or 4 (extreme polygamy) to all species. We assumed that sexual selection is lowest in strict monogamy (0) and highest for extreme polygamy (4). Allocation to the first 4 categories was based on thresholds for polygamy suggested by previous studies: <0.1%, 0.1%–5%, 5%–20%, and >20% for scores of 0–3, respectively [32,39–41]. We used similar but higher quantitative thresholds for EPP data for 2 reasons. First, species may be strictly monogamous yet still have reported EPP rates above zero because of factors such as mate switching or brood parasitism [100]. Consequently, following previous authors [101], we define monogamy as <5% EPP. Second, many socially monogamous species have relatively high levels of EPP, so we adjusted thresholds upwards (assigning populations with 5%–25%, 25%–50%, and >50% EPP to sexual selection scores 1–3, respectively). This avoids inflating sexual selection scores for species with EPP data, a biased subset of well-studied birds mainly inhabiting the temperate zone [49]. Finally, species with highly elaborate mating display behaviours, including social lekking (e.g., Birds-of-Paradise, Paradisaeidae) or permanent display locations (e.g., bellbirds, Procnias), were given the highest score (4) because variance in the reproductive success of males is likely to reach its peak in these systems [102–104]. Note that our scoring system does not assume correlation between different components of our score because, for example, polygamous species may have low EPP while species with high levels of EPP may be socially monogamous. Instead, we assigned the highest sexual selection scores triggered by any criterion.

Unlike previous bidirectional scoring systems [32], we treated sexual selection as a unidirectional variable. That is, we make no distinction between male or female-biased mating systems, instead grouping polyandry and polygyny under the general definition of polygamy. To facilitate any future research that relies on discriminating between female-biased and male-biased sexual selection, we provide an updated list of sex-role-reversed bird species (S1 Data) so that a bidirectional score can be generated by reversing the sign on scores for sex-role-reversed species.

To enable phylogenetic analyses, we aligned sexual selection scores and associated geographical range data to the most comprehensive phylogeny available [50]. The taxonomy used in this phylogeny requires updating, so we also provide an additional data set of sexual selection scores aligned with 10,671 species included for a more recent taxonomy [57]. Although we do not use this data set in our own analyses, we hope that it will enable future studies using forthcoming phylogenetic data sets [105], as well as several rich sources of information, including eBird citizen-science data [106], Birds of the World [99], and AVONET [52], all of which use the same updated taxonomic format [57]. Sexual selection scores under both taxonomic treatments, along with the sources of information supporting these assignments, are presented in S1 Data.

Testing sexual selection metrics

To validate the accuracy of our proposed scoring system in estimating sexual selection, we conducted a comparative analysis against 3 widely used measures of sexual selection from existing published data sets: residual testes mass [97,107], Bateman gradients (β_SS) [108], and the opportunity for sexual selection (I_S) [45] (for a full explanation, sources and derivation of these metrics see S1 Text). Given the scarcity of studies estimating sexual selection, we also derived additional estimates of zero I_S from studies that recorded 0% EPP rates, as summarised by Brouwer and Griffith [49]. We assessed the strength of association between our scores and each metric using Bayesian phylogenetic models (S1 Text), with each metric as a single predictor. To test the accuracy of each metric, we extracted standard p-values for the strength of association, and calculated marginal R² values to determine how closely scores and metrics aligned. To confirm a significant relationship with I_s was due to estimates from species with 0% EPP, we performed a sensitivity analysis excluding the 51 species with estimated I_s scores, which produced results consistent with the full models, as presented in Table B in S2 Text.

Accounting for biases in data certainty

Assembling data on sexual selection is challenging because published information for many species is sparse or based on few observations. Basic details of breeding behaviour are now available for almost all birds (e.g., [99]), but the format, quality, and relevance of information varies greatly across species. To address these biases, we scored data certainty using a system adapted from Tobias and colleagues [20], from 1 (lowest certainty) to 4 (highest certainty; see Table A in S2 Text for full definitions). Species lacking direct observations receive higher certainty scores if they belong to taxonomic groups with consistent and well-documented behaviour (e.g., a poorly known species of sandgrouse would be scored 3 because all known Pteroclidae are monogamous with biparental care). Species in groups with some variation in breeding systems would instead be scored 2 because the degree of confidence in inferred data is reduced.

Ecological variables

To assess the role of evolutionary drivers, we extracted data on species-specific traits from published data sets [20,52]. We selected 3 fundamental ecological traits previously proposed to influence levels of sexual selection: trophic level (linked to both spacing and constraints on parental care [22,53,109]), migration (linked to breeding synchrony or constraints on the duration of breeding seasons [54]), and territoriality (linked to movement and spacing patterns [55]).

To facilitate the comparison of categorical and continuous data, we converted trophic level, migration, and territoriality to binary variables [110]. For trophic level, we group species as either obligate primary consumers, defined as those with more than 60% of their energetic requirements provided by plant material, or secondary consumers, including omnivores, carnivores, and scavengers from higher trophic levels. This was because primary consumers generally experience a more heterogeneous landscape than higher trophic levels, often with scarce resources of high nutritional value. Furthermore, previous research has shown a potential relationship between herbivory and polygamy in mammals [111] that may also be present within birds.

Migration was scored in 3 categories: sedentary, partially migratory (minority of population migrates long distance or most individuals migrate short distances), and migratory (majority of population undertakes long-distance migration). We dichotomized migration by grouping sedentary and partial migrants together, because obligate long-distance migrants tend to maintain consistent departure times, distances, and directions [112], which should all lead to higher breeding synchronicity. Territorial behaviour was classified as either: “none” (never territorial or at most defending very small areas around nest sites), “weak” (weak or seasonal territoriality, including species with broadly overlapping home ranges or habitually joining mixed species flocks), and “strong” (territories maintained throughout year). We dichotomized territoriality by grouping weak and strong territorial behaviour together, as both potentially create strong spatial constraints that in turn may influence the mating opportunities [55,113].

Climatic seasonality

To account for the effects of climate seasonality in models of sexual selection, we included species-specific data on average temperature seasonality extracted from the CHELSA bioclim data set v2.1 [114]. We selected temperature seasonality (bio4) as our primary measure of climate variability because temperature has been shown to predict mating systems and sexual selection [77,115]. The key metric (bio4) provides an index of local intra-annual temperature variation, with grid cells equal to the standard deviation of monthly mean temperatures across each year, using global data from 1981 to 2010 at 30 arcsecond resolution. To extract an average seasonality estimate for each species, we overlaid climate data with expert-drawn breeding ranges provided by BirdLife International [116] (S1 Text). Grid cells that fell within each species’ breeding range were averaged using a Behrmann equal area projection, disregarding cells with less than 50% overlap. Removing species without range maps, this left 9,836 species for subsequent statistical analysis.

Statistical analysis

To identify latitudinal gradients in sexual selection, we first quantified mean sexual selection score for each grid cell using a 200 km resolution Berhmann equal area projection (S1 Text). Grid cell values were calculated from all bird species with available geographical range data (n = 9,836). We then modelled mean sexual selection using spatial SAR models from the R package spatialreg [117], with absolute latitude (distance from the equator) as the sole predictor. Next, to identify any latitudinal bias in data certainty, we ran similar models on the same sample of species with their mean data certainty score as the response variable. Given a strong correlation between latitude and data certainty (Fig 3C and 3D), we ran subsequent latitudinal models predicting sexual selection for higher quality data only. In this case, we limited the sample to species with moderate to high data certainty (scored 3–4; n = 7,592 species) and also ran a maximally conservative model restricted to high-certainty data only (scored 4; n = 2,851 species).

To determine how species ecology shapes global trends, we ran subsequent latitudinal models under the following partitions: primary consumers (n = 2,753), frugivores (n = 1,025), secondary consumers (n = 7,083), invertivores (n = 4,694), long-distance migrants (n = 901), resident and short-distance migrants (n = 8,935), territorial species (n = 7,261), and non-territorial species (n = 2,575). To account for spatial differences in species richness, we also ran Bayesian regression models (species-level) from the R package brms [118], using the centroid latitude of each species’ range as the predictor (S1 Text). Given the ordinal nature of our sexual selection scores, we used a cumulative family distribution, which can be interpreted similarly to standard generalised linear models [119]. To account for variability in data quality, we also repeated each set of models on the subset of species with moderate to high data certainty (certainty scores 3 and 4). It was not possible to restrict these models to species scored 4 for data certainty because of low sample size in many categories.

After identifying raw latitudinal gradients in sexual selection, we used univariate and multivariate Bayesian phylogenetic models to link potential evolutionary drivers with the strength of sexual selection across species. Preliminary analyses show that the strength of sexual selection is highly conserved in birds (see S1 Text). Consequently, we included a phylogenetic co-variance matrix as a random effect, using the Jetz and colleagues [50] tree topology grafted to the Prum and colleagues [61] genomic backbone. To ensure our results are robust to uncertainty in tree topology, we repeated each model over 50 randomly selected trees, combining draws into a single posterior distribution.

To determine the role of species ecology in directly predicting sexual selection, we ran phylogenetic models incorporating information on trophic level, migration, territoriality, and seasonality. As the only continuous variable, seasonality was log-transformed to approximate normality and standardised to 2 standard deviations to facilitate comparison with categorical predictors [110]. All categorical predictors were centred on zero to reduce the collinearity with interaction terms [120]. To address the hypothesis that spatial and temporal heterogeneity of resource abundance drive sexual selection, we also included trophic level interactions with seasonality (temporal heterogeneity) and territoriality (spatial heterogeneity). All predictors and interactions had variance inflation factor (VIF) values below 3 (Table G in S2 Text), indicating that collinearity would not affect model interpretation. To account for biases in data quality, we re-ran univariate and multivariate analyses on higher quality data (certainty 3 or 4; n = 7,592). All analyses were conducted in R version 4.2.1 [121].