Abstract
The status signalling hypothesis aims to explain conspecific variation in ornamentation by suggesting that some ornaments signal dominance status. Here, we use multilevel meta-analytic models to challenge the textbook example of this hypothesis, the black bib of house sparrows (Passer domesticus). We conducted a systematic review, and obtained raw data from published and unpublished studies to test whether dominance rank is positively associated with bib size across studies. Contrary to previous studies, our meta-analysis did not support this prediction. Furthermore, we found several biases in the literature that further question the support available for the status signalling hypothesis. First, the overall effect size of unpublished studies was zero, compared to the medium effect size detected in published studies. Second, the effect sizes of published studies decreased over time, and recently published effects were, on average, no longer distinguishable from zero. We discuss several explanations including pleiotropic, population- and context-dependent effects. Our findings call for reconsidering this established textbook example in evolutionary and behavioural ecology, raise important concerns about the validity of the current scientific publishing culture, and should stimulate renewed interest in understanding within-species variation in ornamental traits.
Introduction
Plumage ornamentation is a very striking example of colour and pattern diversity in the animal kingdom, and has attracted a lot of research [1]. Most studies have focused on sexual selection as the key mechanism to explain this diversity in ornamentation [2,3]. The status signalling hypothesis explains conspecific variation in ornaments by suggesting that these traits signal individual dominance status or fighting ability [4]. Aggressive contests are costly in terms of energy use, and risk of injuries and predation [5–9]. These costs could be reduced if individuals can predict the outcome of such contests beforehand using so-called “badges of status” – individuals could base the decision of whether to avoid or engage in aggressive interactions based on the signal’s message [4].
Patches of ornamentation have been suggested to function as badges of status in a wide range of taxa, including insects [10], reptiles [11] and birds [12]. The status signalling hypothesis was originally proposed to explain variation in the size of mountain sheep horns [13,14], but the hypothesis has become increasingly important in the study of variability in plumage ornamentation in birds [4,12]. Among the many bird species studied [15], the house sparrow (Passer domesticus) has become the classic textbook example of status signalling [2,12,16,17]. The house sparrow is a sexually dimorphic passerine, in which the main difference between the sexes is a prominent black patch on the male’s throat and chest (hereafter “bib”). Many studies have suggested that bib size serves as a badge of status, but most studies are based on limited sample sizes, and have used inconsistent methodologies for measuring bib and dominance status [15,18].
Meta-analysis is a powerful tool to quantitatively test the overall (across-study) effect size (i.e. the “meta-analytic mean”) for a specific hypothesis. Meta-analyses are therefore able to provide more robust conclusions than single studies, and are increasingly used in evolutionary ecology [19–22]. Traditional meta-analyses combine summary data across different studies, where design and methodology are study-specific (e.g. effect sizes among studies are typically adjusted for different fixed effects). These differences among studies are expected to increase heterogeneity, and therefore, the uncertainty of the meta-analytic mean [23]. Meta-analysis of primary or raw data is a specific type of meta-analysis where studies can be analysed in a consistent manner [23]. This type of meta-analysis allows methodology to be standardized so that comparable effect sizes can be obtained across studies and is, therefore, considered the gold standard in disciplines such as medicine [24]. Unfortunately, meta-analysis of primary data is still rarely used in evolutionary ecology (but see [25–27]), perhaps due to the difficulty of obtaining the primary data of previously published studies [28].
An important feature of any meta-analysis is to identify the existence of bias in the literature [20,29]. For example, publication bias occurs whenever particular effect sizes (e.g. larger ones) are more likely found in the literature than others (e.g. smaller ones). This tends to be the case when statistical significance and/or direction of effect sizes determines whether results were submitted or accepted for publication [29]. Thus, publication bias can strongly affect the estimation of the meta-analytic mean, and distort the interpretation of the hypothesis [30]. Several methods have been developed to identify this and other biases [20,29]; however, such methods are imperfect, dependent on the number of effect sizes available, and therefore should be considered as types of sensitivity analysis [20,31].
Here we meta-analytically assessed the textbook example of the status signalling hypothesis in the house sparrow. Specifically, we combined summary and primary data from published and unpublished studies to test the prediction that dominance rank is positively associated with bib size across studies. We found that the meta-analytic mean was small, uncertain and overlapped zero. Hence, our results challenge the status signalling function of the male house sparrow’s bib. Also, we identified several biases in the published literature that call for substantial changes in scientific publication culture. Finally, we discuss potential biological explanations for our results, and provide advice for future studies testing the status signalling hypothesis.
Results
Overall, we obtained the primary data for seven of 13 (54%) published studies, and we provided data for six additional unpublished studies (Table 1, S1 and S2 Appendix).
Dominance hierarchies
Mean sampling effort was 36 interactions/individual (SD = 24), which highlights that, overall, dominance hierarchies were inferred reliably across groups [49]. The mean Elo-rating repeatability was 0.92 (SD = 0.07) and the mean triangle transitivity was 0.63 (SD = 0.28). Thus, the dominance hierarchies observed across groups of house sparrows were of medium both steepness and transitivity.
Meta-analytic mean
Our meta-analyses revealed a small overall effect size with large 95% credible intervals that overlapped zero (Table 2; Fig 1). Additionally, the overall heterogeneity (I2overall) was moderate (53%; Table 2). Thus, our results suggested that generally, bib size does not predict dominance status in male house sparrows.
Moderators of the relationship between dominance rank and bib size
None of the three biological moderators studied (season, group composition and type of interactions) explained differences among studies (Table 3). Sampling effort (i.e. the ratio of interactions to individuals recorded) was not an important moderator either (Table 3).
Detection of publication bias
There was no clear asymmetry in the funnel plots (Fig 2). Also, Egger’s regression tests did not show evidence of funnel plot asymmetry in any of the meta-analyses (Table 2 and S4 Table). However, published effect sizes were larger than unpublished ones, and the latter were not different from zero (Table 4; Fig 3). Additionally, we found evidence for a time-lag bias in the published literature as effect sizes decreased over time (Table 4; Fig 4).
Discussion
The male house sparrow’s bib is not the strong across-study predictor of dominance status once believed. In contrast to the medium-to-large effect found in the previous meta-analysis [18], our updated meta-analytic mean was uncertain and overlapped zero. Thus, the male house sparrows’ bib should not be unambiguously considered or called a badge of status. Furthermore, we found evidence for the existence of bias in the published literature that further undermines the validity of the available support for the status signalling hypothesis. First, the meta-analytic mean of unpublished studies was zero, compared to the medium effect size detected in published studies. Second, we found evidence for a time-lag bias. The effect size estimated in published studies has been decreasing over time, and recently published effects were on average no longer distinguishable from zero. Our findings call for reconsidering this textbook example in evolutionary and behavioural ecology, raise important concerns about the validity of the current scientific publishing culture, and should stimulate renewed attention to hypotheses explaining within-species variation in ornamentation.
The status signalling hypothesis [4] has been extensively tested to try and explain intraspecific trait variation (e.g. reptiles: [11]; insects: [10]; humans: [51]), particularly plumage variation [15]. Soon after the first empirical tests on birds, the black bib of house sparrows became a textbook example of the status signalling hypothesis [2,12,16,17], an idea that was later confirmed meta-analytically [18]. However, Nakagawa et al.’s [18] meta-analytic mean was over-estimated because only 9 low-powered studies were available (more in [52]). Here we updated that meta-analysis with newly published and unpublished data. Our results showed that the overall effect size is much smaller and much more uncertain than previously thought. The status signalling hypothesis is thus no longer a compelling explanation for the evolution of bib size across populations of house sparrows.
Similar contradicting conclusions have been reported for other model species. An exhaustive review and meta-analysis on plumage coloration of blue tits (Cyanistes caeruleus) revealed that, after dozens of publications studying the function of plumage ornamentation in this species, the only robust conclusion is that females’ plumage differs from that of males [53]. Another example is the long-believed effect of leg bands of particular colours on the perceived attractiveness of male zebra finches (Taeniopygia guttata), which has been also experimentally and meta-analytically refuted [54,55]. Finally, the existence of a badge of status in a non-bird model species, the paper wasp (Polistes dominulus; [10]) has also been challenged multiple times (e.g. [56–58]), generating doubts about its generality. Our findings corroborate studies showing that abundant replication is needed before any strong or general conclusion can be drawn [59], and highlight the existence of important impediments to scientific progress in ecology.
Indeed, our results showed that the published literature on status signalling in house sparrows is likely a biased subsample. The main evidence for this is that the mean effect size of unpublished studies was essentially zero and clearly different from the mean effect size based of published studies, which was of medium size. Furthermore, this moderator (i.e. unpublished vs. published) explained a large percentage of the model’s variance. In some of our own unpublished datasets, the relationship between dominance rank and bib size was never formally tested (Westneat & Bókony, personal communication), suggesting that these unpublished datasets are not strictly speaking examples of the “file drawer problem” (sensu [60]). Egger’s regressions failed to detect any funnel plot asymmetry, even in the meta-analyses based on published effect sizes only (S4 Table). However, because unpublished data indeed existed (i.e. those obtained for this study), the detection failure was likely the consequence of the limited number of available estimates [61].
An additional common bias in the published literature is the time-lag bias [62]. We detected evidence for such bias because the correlation between dominance rank and bib size in published studies has decreased over time. Year of publication explained a large percentage of variance, and accounting for year of publication resulted in a strong reduction of the mean effect size across published studies (Table 4 vs. S4 Table). Time-lag bias has been detected in other ecological studies [63,64], including a meta-analysis on status signalling across bird species [15]. In the latter study, a positive overall (across-species) effect size persisted regardless of the time-lag bias, and no strong evidence for other types of biases was found [15]. However, Santos et al. [15] did not attempt to analyse unpublished data, so additional evidence is needed to determine the effect that unpublished data have on the overall validity of the status signalling hypothesis across bird species. If effect sizes based on unpublished data for other species were of similar magnitude than those obtained for house sparrows, the validity of the status signalling hypothesis across species would need reconsideration. The existence of publication bias in ecology has long been recognized [65–67]. Publication bias leads to false conclusions if not accounted for [30], and is, thus, a serious impediment to scientific progress.
In addition to estimating the overall effect size for a hypothesis, meta-analyses can also assess differences among estimates or heterogeneity [68,69]. Understanding the sources of heterogeneity is an important step towards the correct interpretation of a meta-analytic mean, and it can be done using meta-regressions [20]. Here, we found that the percentage of variance that was not attributable to sampling error (i.e. heterogeneity) was moderate. This value is below what has been quantified in ecological and evolutionary meta-analyses [21], and indicates that we accounted for large differences among estimates. Our meta-regressions based on biological moderators explained 20-23% of the variance (Table 3). However, none of the biological moderators that we tested influenced the overall effect size, but this might be because of limited sample sizes.
The badge of status idea is more subtle than typically portrayed [70]. While the evolution of badges of status in New and Old World sparrows has been related to sociality during the non-breeding season [71], additional factors need to be involved if the signal is to function in reducing aggression but retaining honesty. A recent study on black-crested titmice (Baeolophus atricristatus) suggested that badges of status might function only in high-competition scenarios [72]. However, time of the year was not a strong predictor in our models, though most studies were conducted in captivity, where food is normally provided ad libitum. Badges of status are expected to function both within and between sexes [4,12]. Indeed, we found little evidence that the status signalling function of bib size differed between male-only and mixed-sex flocks. Interestingly, when competing for resources, possessing a badge of status would be beneficial for both males and females. However, males but not female house sparrows have a bib. This sexual dimorphism suggests that the bib’s function is likely more important when competing for resources other than essential, a priori non-sex-specific, resources such as food, water, sand baths and roosting sites. [73,74] reported that female house sparrows preferentially choose males with large bibs (but see [75]), and bib size has been positively correlated with sexual behaviour [76,77], which suggests that the bib may play a role in mate choice. Furthermore, the original status signalling hypothesis posits that the main benefit of using badges of status would be to avoid fights, which should be particularly important when interacting with non-familiar individuals [4,12]. Although we did not have data to test whether unfamiliarity between contestants is an important pre-requisite for the status signalling hypothesis, we found no change in mean effect size when only obviously aggressive interactions were studied. In practice, testing whether the bib is important in mediating aggression between unfamiliar individuals is difficult because the certainty of estimates of individual dominance increases over time as more contests are recorded, but so does familiarity between individuals.
Our analyses have several potential limitations. First, although the number of studies included in this meta-analysis is more than double that of the previous meta-analysis [18], it is still limited. Also, it is likely (see above) that additional unpublished data are stored in “file drawers” sensu [60]. Second, most tests included in this study were still low-powered in terms of group size (median = 6 individuals/estimate, range = 4-41), and the sample size is inflated because some of the published studies pooled individuals from different groups (Fig 4). Third, although our results showed little evidence of an effect of sampling effort on the overall effect size, the quality of the data on dominance and bib size may still be a potential factor explaining differences between studies. Additionally, data from studies not specifically designed to test the status signalling hypothesis (such as some of the unpublished data) may be less prone to confirmation bias [78], although other unknown factors might affect the adequacy of such data for testing the hypothesis. Fourth, experimental effect sizes will normally be larger, because effects of confounding factors can be reduced [67]. However, our systematic review only identified two studies where the status signalling hypothesis was tested experimentally in house sparrows [37,79], preventing us from estimating the meta-analytic mean for experimental studies. Note, however, that the results of those experiments were inconclusive, and could be partially explained by the phenomenon known as regression to the mean [78].
There are some additional explanations for the small and uncertain effect detected by our meta-analyses. First, different populations might be under different selective pressures regarding status signalling. Indeed, the population-specific heterogeneity (I2population ID) estimated in our meta-analyses was 15-16%, suggesting that population-dependent effects might exist. Second, although none of the moderators had a strong influence on the overall effect size, the study-specific heterogeneity estimated in our meta-analyses (I2study ID = 20-21%) suggests that the uncertainty observed could still be explained by the status signal being context-dependent. Although context-dependence is often invoked to explain variation between studies, there is little evidence for it. Last, most studies testing the status signalling hypothesis in house sparrows are observational (Table 1), and the only two experimental studies conducted so far were inconclusive [37,79]. Thus, it cannot be ruled out that the weak correlation observed between dominance status and bib size is driven by a third, unknown variable. In this respect, it has been proposed that the association between melanin-based coloration (such as the bib; e.g. [80,81]) and aggression is due to pleiotropic effects of the genes involved in regulating the synthesis of melanin (reviewed by [82]). Furthermore, bib size has been shown to correlate with testosterone, a hormone often involved in aggressive behaviour ([83]; but see [84]). Future studies should shift the focus towards understanding the function of bib size in wild populations and increase considerably the number of birds studied per group. The latter is essential because the statistical power of published tests of the status signalling hypothesis in house sparrows is alarmingly low (power = 8.5%, S4 Appendix) and lower than the average in behavioural ecology [85].
In conclusion, our results challenge an established textbook example of the “badge of status” hypothesis to explain variation in ornament size. In house sparrows, we find no evidence that bib size consistently acts as a badge of status across studies and populations, and thus, this can no longer be considered a textbook example of the status signalling hypothesis. Furthermore, our analyses highlight the existence of publication and time-lag biases in the published literature, further undermining the validity of past conclusions. Bias against the publication of small (“non-significant”) effects hinders scientific progress. We thus join the call for a change in incentives and scientific culture in ecology and evolution [78,86–88].
Materials and Methods
Systematic review
We used several approaches to maximize the identification of relevant studies. First, we included all studies reported in a previous meta-analysis that tested the relationship between dominance rank and bib size in house sparrows [18]. Second, we conducted a keyword search on the Web of Science, PubMed and Scopus from 2006 to June 2017 to find studies published after [18], using the combination of keywords [“bib/badge”, “sparrow”, “dominance/status/fighting”]. Third, we screened all studies on house sparrows used in a meta-analysis that tested the relationship between dominance and plumage ornamentation across species [15] to identify additional studies that we may have missed in our keyword search. We screened titles and abstracts of all articles and removed the irrelevant articles before examining the full texts. We followed the preferred reporting items for systematic reviews and meta-analyses (PRISMA: [89]; see S1 Appendix). We only included articles in which dominance was directly inferred from agonistic dyadic interactions over resources such as food, water, sand baths or roosting sites (S1 Table).
Summary data extraction
Some studies had more than one effect size estimate per group of birds studied. When the presence of multiple estimates was due to the use of different statistical analyses on the same data, we chose a single estimate based on the following order of preference: (1) direct reports of effect size per group of birds studied (e.g. correlation coefficient), (2) inferential statistics (e.g. t, F and χ2 statistics) from analyses where group ID was accounted for and no other fixed effects were included, (3) direct reports of effect size where individuals from different groups where pooled together, (4) inferential statistics from models including other fixed effects. When the presence of multiple estimates was due to the use of different methods to estimate bib size and dominance rank on the same data, we chose a single estimate per group of birds or study based on the order of preference shown in S2 Appendix.
Primary data acquisition
We requested primary data (i.e. agonistic dyadic interactions and bib size measures) of all relevant studies identified by our systematic review. Additionally, we asked authors to share, if available, any unpublished data that could be used to test the relationship between dominance rank and bib size in house sparrows. We emailed the corresponding author, but if no reply was received, we tried contacting all the other authors listed. One study [33] provided all primary data in the original publication and, therefore, its author was not contacted. Last, we included our own unpublished data (S2 Appendix).
Most studies recorded data from more than one group of birds (Table 1). For each primary dataset obtained, we inferred the dominance hierarchy of each group of birds from the observed agonistic dyadic interactions (wins and losses) between individuals using the randomized Elo-rating method, which estimates dominance hierarchies more precisely than other methods [49]. We then used the provided measures of individual bib size (e.g. area outlined from pictures) or, if possible, calculated bib area from length and width measures following [33]. Subsequently, we estimated the Spearman’s rho rank correlation (ρ) between individual rank and bib size for each group of birds. For one study [42], we received the already inferred dominance hierarchies for each group of birds, which we then correlated with bib size to obtain ρ.
Effect size coding
Regardless of their source (primary or summary data), we transformed all estimates (e.g. ρ, F statistics, etc) into Pearson’s correlation coefficients (r), and then into standardized effect sizes using Fisher’s transformation (Zr) for between-study comparison. We used the equations from [90] and [91]. Since log(0) is undefined, r values equal to 1.00 and −1.00 were transformed to 0.975 and −0.975, respectively, before calculating Zr. Zr values of 0.100, 0.310 and 0.549 were considered small, medium and large effect sizes, respectively (equivalent benchmarks from [50]). When not reported directly, the number of individuals (n) was estimated from the degrees of freedom. The variance in Zr was calculated as: VZr = 1/(n-3). Estimates (k) based on less than four individuals were discarded (k = 33 estimates discarded).
Meta-analyses
We ran two multilevel meta-analyses to test whether dominance rank and bib size were positively correlated across studies. The first meta-analysis, “meta 1”, included published and unpublished (re-)analysed effect sizes (i.e. effect sizes estimated from the studies we obtained primary data from), plus the remaining published effect sizes obtained from summary data (i.e. effect sizes for which primary data were unavailable).
Second, three studies reported “statistically non-significant” results without showing either the magnitude or the direction of the estimates (Table 1). Receipt of primary data allowed us to recover some but not all the originally non-reported estimates. Two “non-significant” estimates were still missing. To test the robustness of the results to those two cases of selective reporting, we ran an additional meta-analysis (see [92] for a similar approach). This second meta-analysis, “meta 2”, was like meta 1 but included the two non-significant non-reported estimates, which were assumed to be zero. Note that non-significant estimates can be either negative or positive, and thus, assuming that they were zero may have either underestimated or overestimated them, something we cannot know from non-reported estimates. Meta-analyses based on published studies only are shown in S3 Appendix.
We investigated inconsistency across studies by estimating the heterogeneity (I2) from our meta-analyses following [20]. I2 values around 25, 50 and 75% are considered as low, moderate and high levels of heterogeneity, respectively [69].
Meta-regressions
We tested if season, group composition and/or the type of interactions recorded had an effect on the meta-analytic mean. For that, we ran two multilevel meta-regressions that included the following moderators (hereafter “biological moderators”): (1) “season”, referring to whether the study was conducted during the non-breeding (September-February) or the breeding season (March-August); (2) “group composition”, referring to whether birds were kept in male-only or in mixed-sex groups; and, (3) “type of interactions”, referring to whether the dyadic interactions recorded were only aggressive (e.g. threats and pecks), or also included interactions that were not obviously aggressive (e.g. displacements). Because only three of 19 studies were conducted in the wild (k = 12 estimates; Table 1), we did not include a moderator testing for captive versus wild environments. The three biological moderators were mean-centred following [93] to aid interpretation.
The ratio of agonistic dyadic interactions recorded to the total number of interacting individuals observed (hereafter “sampling effort”) is a measure of sampling effort that correlates positively and logarithmically with the ability to infer the latent dominance hierarchy [49]. The higher this ratio, the more precisely the latent hierarchy can be inferred [49]. For the subset of studies for which the primary data of the agonistic dyadic interactions were available (12 out of 19 studies; Table 1), we ran a multilevel meta-regression including sampling effort and its squared term as z-transformed moderators [93]. The squared term was included because of the observed logarithmic relationship between sampling effort and the method’s performance [49]. This meta-regression tested whether sampling effort had an effect on the meta-analytic mean.
For all meta-regressions, we estimated the percentage of variance explained by the moderators (R2marginal) following [94].
Random effects
All meta-analyses and meta-regressions included the two random effects “population ID” and “study ID”. Population ID was related to the geographical location of the population of birds studied. We used Google maps to estimate the distance over land (i.e. avoiding large water bodies) between populations, and assumed the same population ID when the distance was below 50 km (13 populations; Table 1). Study ID encompassed those estimates obtained within each specific study (19 studies). Two studies tested the prediction twice for the same groups of birds (Table 1) and, within each population, some individuals may have been sampled more than once. However, we could not include “group ID” and/or “individual ID” as additional random effects due to either limited sample size or because the relevant data were not available.
Detection of publication bias
For the meta-analyses, we assessed publication bias using two methods that are based on the assumption that funnel plots should be symmetrical. First, we visually inspected asymmetry in funnel plots of meta-analytic residuals against the inverse of their precision (defined as the square root of the inverse of VZr). Funnel plots based on meta-analytic residuals (the sum of effect-size-level effects and sampling-variance effects) are more appropriate than those based on effect sizes when multilevel models are used [20]. Second, we ran Egger’s regressions using the meta-analytic residuals as the response variable, and the precision (see above) as the moderator [20]. If the intercept of such a regression does not overlap zero, estimates from the opposite direction to the meta-analytic mean might be missing and hence we consider this evidence of publication bias [20]. Further, we tested whether published estimates differed from unpublished estimates. For that, we ran a multilevel meta-regression that included population ID and study ID as random effects, and “unpublished” (two levels: yes (0), no (1)) as a moderator. This meta-regression was based on meta 1 (i.e. it did not include the two non-reported estimates). We did not use the trim-and-fill method [95,96] because this method has been advised against when significant heterogeneity is present [29,97], as it was the case in our meta-analyses (see below).
Finally, we analysed temporal trends in effect sizes that could indicate “time-lag bias”. Time-lag bias is common in the literature [63,64], and occurs when the effect sizes of a specific hypothesis are negatively correlated with publication date (i.e. effect sizes decrease over time; [62]). We ran a multilevel meta-regression based on published effect sizes only, where “year of publication” was included as a z-transformed moderator [20].
All analyses were run in R v. 3.4.0 [98]. We inferred individual dominance ranks from agonistic dyadic interactions using the randomized Elo-rating method from the R package “aniDom” v. 0.1.3 [49,99]. Additionally, we described the dominance hierarchies observed in the groups of house sparrows for which primary data was available. For that we estimated the uncertainty of the dominance hierarchies using the R package “aniDom” v. 0.1.3 [49,99] and the triangle transitivity [100] using the R package “compete” 3.1.0 [101]. We used the R package “MCMCglmm” v. 2.24 [102] to run the multilevel meta-analytic (meta-regression) models [103]. For each meta-analysis and meta-regression, we ran three independent MCMC chains for 2 million iterations (thinning = 1800, burn-in = 200,000) using inverse-Gamma priors (V = 1, nu = 0.002). Model chains were checked for convergence and mixing using the Gelman-Rubin statistic. The auto-correlation within the chains was < 0.1 in all cases. For each meta-analysis and meta-regression, we chose the model with the lowest DIC value to extract the posterior mean and its 95% highest posterior density intervals (hereafter 95% credible interval).
Data Accessibility
We provide all of the R code and data used for our analyses [104].
Supporting Information
The following supporting information is available for this article online:
- Supporting Information 1:
S1 Appendix: PRISMA flow diagram.
S2 Appendix: Information about data used.
S3 Appendix: Results based on published effect sizes only.
S4 Appendix: Power analysis.
- Supporting Information 2: decision spreadsheet of the systematic review.
Acknowledgments
AST and AG are members of and grateful for the support of the International Max Planck Research School (IMPRS) for Organismal Biology. We thank Katherine L. Buchanan, Fabrice Helfenstein, Anna Kulcsár, Ádám Z. Lendvai, Karin M. Lindström, Thor Harald Ringsby, Alfonso Rojas Mora, Bernt-Erik Sæther, Erling J. Solberg, Zoltán Tóth and Jarle Tufto for sharing the primary data of published studies, and Emmi Schlicht for maintenance of the house sparrow database at Seewiesen. We thank Wolfgang Forstmeier and Lucy Winder for constructive feedback on the manuscript. This work was supported by the Volkswagen Foundation (to JS), the People Programme (Marie Curie Actions) of the European Union’s Seventh Framework Programme FP7/2007-2013 (CIG PCIG12-GA-2012-333096; to JS) the U.S. National Science Foundation (to DFW) and the Natural Environment Research Council (NERC: NE/N013832/1; to TB). The authors declare that they have no conflict of interest.
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.
- 7.
- 8.
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.
- 33.↵
- 34.
- 35.
- 36.
- 37.↵
- 38.
- 39.
- 40.
- 41.
- 42.↵
- 43.
- 44.
- 45.
- 46.
- 47.
- 48.
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵