Colorful traits in females are suggested to have evolved and be maintained by sexual selection. Although several studies have evaluated this idea, support is still equivocal. Evidence has been compiled in reviews, and a handful of quantitative synthesis have explored evidence of the link between condition and specific color traits in males and females. However, understanding the potential function of females’ colorful traits in sexual communication has not been the primary focus of any of those previous studies. Here, using a meta-analytic approach, we find that evidence from empirical studies in birds supports the idea that colorful female ornaments are positively associated with residual mass and immune response, clutch size, and male-mate preferences. Hence, colorful traits in female birds likely evolved and are maintained by sexual selection.

A systematic search of literature was performed using the PRISMA method (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [51]. We looked for literature using the Google Scholar and Web of Science search engines, using the keywords: “female bird ornament”, “female bird ornamentation”, “female bird traits”, “female bird quality”, “female bird condition”, “female bird attractiveness”, “female bird color”, “female bird colour”, “female bird coloration”, “male bird choice”, “male bird preference”. References within papers were also checked. The last search was conducted on October 15^th 2020.

We excluded studies that were missing data on statistical values, did not provide independent estimates for each sex, or did not specify the type of coloration analyzed. We detected 83 studies in our search, 59 of which satisfied the inclusion criteria (Figure S1). Of the 47 total species (8 species were used in more than one study), 42 had a socially monogamous mating system, four were facultatively polygamous, and one was polygamous. Carotenoid-dependent ornaments were present in 23 species, melanin-based in 18, and structural colorations in 11 (Table S1). The intensity of coloration was estimated using the number and/or size of colored structures/patches, spectrophotometry, digital image analysis (RGB or LBA), and visual rank scales (color charts and visual rank score).

Using independent meta-analyses, we evaluated the evidence of an association of colorful female ornaments with condition, reproductive performance, and male preferences (see below). Studies assessing more than one of these associations were included in each meta-analysis that applied. Effect sizes for quality and reproductive success were analyzed as Pearson’s correlation coefficients (r). When this coefficient was not directly available from papers (i.e., different statistics were reported, such as F, t, χ²), reported values were transformed to correlation coefficients following Rosenberg’s formulas [52]. Pearson’s coefficients were subsequently transformed to Fisher’s z-values for statistical analyzes [53, see equations in Supplementary material]. All effect sizes were weighted using the inverse of the sampling variance [54, 55]. Effect sizes for male preferences were reported as Hedges’ g, calculated from Pearson’s coefficients (reported statistics were transformed when necessary, see above) [53], and weighted using the g’ variance [53, see equations in Supplementary material].  

Three moderators were considered in the model evaluating the association between female color and condition: Condition proxy (residual body mass, immune response -humoral or cellular -, and parasite load), ornament type (feathers or integuments), and coloration type (carotenoid-dependent, melanin-based, or structural). In the model evaluating the association between female color and reproductive performance, we used the same ornament type and coloration type moderator variables but replaced the condition proxy with a reproductive performance proxy (laying date, clutch size, or fledging success). When clutch size and fledging success were assessed in the same study, only fledging success was considered. In the analyses exploring the relationship between female color and male mating preferences, only four effect sizes were available for melanin-based and structural colors (one and three, respectively), so we fit this model including effect sizes only from studies evaluating carotenoid-dependent colorations (n = 11). The ornament type (feathers or integuments) was included as a moderator.

In the case that two or more effect sizes were available from a single paper testing the same hypothesis and using the same ornament type and coloration (e.g., two carotenoid-dependent feather patches), those effect sizes were averaged to avoid overrepresentation. In studies that contained multiple effect sizes from different ornament or coloration types (e.g., one carotenoid-dependent feather patch and one melanin-based integument patch), we included each of the effect sizes separately and study identity as a random factor.  We used Cochran's Q as a measure of effect size heterogeneity and the Q_E-test, and Q_M-test to estimate whether moderators were associated with estimates of effect size. Publication bias was illustrated using funnel plots [56] and assessed by the trim and fill method [57, 58]. All analyses were conducted using the R package metafor and the function rma.mv [59] in R 4.0.2 software [60].

Ornaments exhibited by different species may share evolutionary history, generating phylogenetic non-independence. To account for this, a phylogeny for each analysis was obtained using a maximum clade credibility consensus tree and a sample of 100 phylogenies downloaded from BirdTree (www.birdtree.org) [61] based on the Ericson et al. [62] backbone. The influence of phylogenetic signal was assessed using two independent approaches (Table S2). First, we determined the phylogenetic signal in model residuals as Pagel’s λ [63] and Blomberg’s Kappa (K) [64] using the phylosig function from the phytools package [65]. Second, we used phylogenetic generalized least squares models (PGLS) to estimate Pagel’s λ through maximum likelihood, by fitting the model as effect sizes ~ 1, using the R package caper [66]. Results obtained suggest that closely related species show similar relationships between female coloration and reproductive performance (Table S2) Here, we present the results of the analyses including phylogeny, following Nakagawa and Santos [67].