Behavioral Ecology Advance Access originally published online on August 31, 2006
Behavioral Ecology 2006 17(6):1029-1040; doi:10.1093/beheco/arl041
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The blue tit's song is an inconsistent signal of male condition
a Edward Grey Institute of Field Ornithology, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK b Division of Biology, Ackert Hall, Kansas State University, Manhattan, KS 66506, USA c Department of Animal and Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK d School of Biological Sciences, University of East Anglia, Norwich NR4 7TJ, UK e School of Biological, Earth, and Environmental Sciences, University of New South Wales, Sydney, New South Wales 2052, Australia
Address correspondence to T.H. Parker, who is now at the Department of Biology, Whitman College, 345 Boyer Ave, Walla Walla, WA 99362, USA. E-mail: parkerth{at}whitman.edu.
Received 1 February 2006; revised 8 May 2006; accepted 3 August 2006.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
Sexually selected traits are often hypothesized to signal male condition or quality, though empirical evidence is mixed, and a number of alternative models of sexual selection do not require condition dependence. We examined the relationship between various measures of condition and dawn songs in male blue tits (Cyanistes caeruleus). We detected 6 largely independent measures of variation (i.e., variables) in these songs. None of these variables were related to blue tits' ultraviolet–blue plumage, a demonstrated sexual signal, thus failing to support the redundant signal hypothesis. We found some evidence that the song variables we measured signaled male quality. There were correlations between body size and certain song traits, though neither male age nor male recapture in the subsequent breeding season (apparent local survival) predicted any song variation. We combined our results with published effect sizes comparing blue tit song with male quality variables using meta-analysis and found that a few song measures are correlates of male quality, though as in our field data, neither male age nor survival appeared related to song. Our relatively large sample sizes (>60), combined with our meta-analytical integration of 89 effect sizes, make the results regarding the signaling value of our measured components of blue tit song robust. These results demonstrate that 1) only certain aspects of signal variation may be condition dependent and 2) even when components of a sexual signal appear correlated with condition in some studies, these signal components may be unrelated or inconsistently related to a variety of condition indices.
Key words: condition dependence, dawn song, meta-analysis, Parus caeruleus, survival, ultraviolet.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
Numerous studies have demonstrated that variation in the expression of sexual signals amongst individuals within a population can be related to differences in the condition of those individuals (Andersson 1994
; Ligon 1999). Condition-dependent ornamental traits convey information about male quality (Andersson 1986), and thus might facilitate female choice of high quality mates (Griffith and Pryke 2006), or male choices concerning competitive interactions (e.g., Parker and Ligon 2002). However, not all hypotheses regarding the evolution and maintenance of sexual signals predict strong condition dependence of ornamental traits. For instance, some sexual signals may induce a response in a receiver because of a sensory bias on the part of the receiver (Ryan 1998). Sexual signals may convey information about heritable fitness entirely due to sexual attractiveness, and sometimes males with reduced condition or survival may be most attractive (Kokko et al. 2002). Other sexual signals, especially those with multimodal expression, may help receivers identify individual signalers (Dale 2000). Alternatively, certain signal measurements may not vary with male condition or quality but the context in which the signal is used may (Mennill et al. 2002). Empirical generalizations about whether sexual signals are typically expressed with a high level of condition dependence remain difficult.
In many species, individuals display multiple sexual signals (e.g., Ligon et al. 1998; Pryke et al. 2001), and a number of hypotheses have been proposed to explain this phenomenon. The multiple message hypothesis proposes that different signals each convey different information about the signaler, although it could be that different signals convey redundant information instead (Møller and Pomiankowski 1993). It may also be that if one signal is condition dependent, selection will not favor condition dependence in other signals (Iwasa and Pomiankowski 1994) or multiple signals might evolve through a process of sensory drive, where new, more detectable signals spread rapidly because females notice them (Schluter and Price 1993).
Although many aspects of bird song have been extensively studied, the possibility that it can be a condition-dependent signal has only recently begun to receive extensive scrutiny, for instance with experimental manipulation (e.g., Spencer et al. 2003). Song is a sexual signal in birds, typically produced by males in the context of both territory defense and mate attraction (Catchpole and Slater 1995). At least 3 different mechanisms for condition dependence of bird song have been investigated. A component of the avian brain important to song learning, the high vocal center, has been shown to be sensitive to manipulations of condition in captive studies (Buchanan et al. 2004). This translates to an influence of condition, especially during brain development, on song learning (Nowicki et al. 1998), and both correlative and manipulative studies indicate that song complexity (e.g., repertoire size, the number of distinct song types an individual learns and sings) is condition dependent in some species (Nowicki et al. 2000; Spencer et al. 2003, 2004, but see Forstmeier and Leisler 2004). It has long been supposed that song production might be energetically demanding and that variables such as rate, length, or consistency of song production might therefore show condition dependence. There is support for this idea, but evidence varies among studies and species. For instance, evidence from the field suggests that fat reserves may be needed for long duration song output (e.g., Thomas 2002), but respirometry chamber results are mixed, with some suggesting singing is metabolically demanding (Eberhardt 1994; but see Gaunt et al. 1996) and others suggesting it is not (e.g., Ward et al. 2004). A third possibility is that constraints on song production are enforced by social interactions such that poor condition males of low status produce a substandard signal because of the risk of punishment from higher status males. For instance, dominant and subordinate individuals might alter aspects of their song production, such as the timing of the song phrase, in relation to other males singing in the vicinity (e.g., Mennill et al. 2002).
The blue tit (Cyanistes caeruleus) has multiple sexual ornaments. Particularly striking are the ultraviolet (UV)–blue plumage patches. The UV–blue crown plumage is sexually dichromatic (Andersson et al. 1998) and has been found to be related to both overwinter survival (Griffith et al. 2003) and the sex ratio of offspring (Sheldon et al. 1999). One measurement of this plumage trait, UV chroma (the proportion of the reflectance curve in the UV wavelengths), has consistently appeared important in sexual signaling (Andersson et al. 1998; Sheldon et al. 1999; Griffith et al. 2003). It has also been suggested that another plumage patch, the yellow breast, may be a signal of parenting ability (Senar et al. 2002). In addition to these ornamental colors, the male blue tit sings a moderately complex song (Bijnens and Dhondt 1984) that has been the focus of several studies (Bijnens 1988; Kempenaers et al. 1997; Doutrelant et al. 2000; Poesel et al. 2001; Foerster et al. 2002; Dreiss et al. 2006). Much of this work has focused on recordings made during the dawn chorus, the only extended period of uninterrupted singing in this species. As blue tits only participate in the dawn chorus after pairing and territory formation, dawn singing is thought to serve functions such as mate guarding (Mace 1989; Kempenaers et al. 1997), attraction of extrapair partners (Mace 1989; Kempenaers et al. 1997), or territory defense (Slagsvold et al. 1994). Components of dawn song may signal male quality or condition, but significant relationships between song production and variables such as body size, hormone level, and survival have not been replicated among studies (Appendix 1). For instance, tarsus length showed significantly positive relationships with song variables in 2 studies, but in one study, the song variable was the number of song types sung in a single chorus (Doutrelant et al. 2000), and in another it was performance time, the proportion of the song's duration that a male was actually vocalizing (Poesel et al. 2001). In a third study, performance time was significantly related to male mass and testosterone level, but a relationship with tarsus length was not reported (Foerster et al. 2002). Such results suggest a role for song in sexual selection, but the lack of consistency between studies is a concern, and clearly further work is necessary before strong, general conclusions can be drawn.
When attempting to draw general conclusions about a study system, meta-analytical synthesis of published data can be fruitful (e.g., Sheldon and West 2004; Parker et al. 2005) by combining published results into single statistical tests of hypotheses (Rosenthal 1984; Gurevitch and Hedges 1999). A meta-analytical approach is well suited to cases such as blue tit song condition dependence, where results differ among studies. This approach will allow identification of particular song variables that tend to be most strongly related to condition or vice versa.
Our objectives with this study of blue tits were to 1) define the major forms of variation (i.e., variables) in dawn song production and thus determine the extent to which different aspects of song were redundant signals, 2) test for redundancy between a well established plumage color signal and dawn song variables, 3) assess whether various potential condition or quality indices were predictors of dawn song variables, and 4) decide, based on a combination of published data and our own field data, whether current evidence supports the hypothesis that blue tit song production signals condition and, if so, determine which components of song appear involved.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
Study site
We studied blue tit song in 3 years (2002–2004) in Wytham Wood, a 380 ha woodland in Oxfordshire, UK (1°20'W, 51°47'N). Between 330 and 450 blue tit pairs bred in artificial nest boxes in each breeding season.
Song recording
During a female blue tit's fertile period (from several days before egg laying begins until the penultimate day of the laying period [Mace 1989]), each day before dawn her mate produces one bout of song typically lasting 15–60 min. This is the most predictable singing bout produced by male blue tits, and it is usually also their longest period of singing uninterrupted by other activities. Because of the link between this male signal and the female's fertile period and because earlier research suggested that dawn song may be sexually selected in blue tits (Doutrelant et al. 2000; e.g., Kempenaers et al. 1997; Poesel et al. 2001), we chose to focus on this aspect of male song.
As with some other studies of blue tit song (e.g., Doutrelant et al. 2000), birds on our study site were not individually color ringed. Therefore, we carefully observed behavior of individual birds in the weeks leading up to recording to determine territory boundaries associated with particular nest boxes. This involved confirming that males spent at least 5 min unchallenged in the vicinity of the nest box (e.g., Doutrelant et al. 2000) and following males to map locations of boundary conflicts and thus delineate territories. A male's dawn song was recorded from his first vocalizations of the day until the termination of his dawn song bout. The end of the dawn song is generally unambiguous and occurs when the pair copulates. After this, the song rate drops dramatically, and the male can be observed in other activities such as foraging. If we did not observe copulation, then the termination was defined as >5 min of either 1) no singing or 2) a switch to other activities, typically foraging, with long (>30 s) pauses between singing. Not all dawn recording efforts were successful. In some cases, we could not confirm unambiguous association with a nest box during or after the dawn chorus and such recordings were not used in our analyses. Because we were conservative in applying these standards, we are confident that few, if any, recordings were attributed to an incorrect male.
Although females are fertile before egg laying begins, we limited our assessment of songs to those recorded after the start of laying to increase standardization of the conditions the pairs were experiencing at the time of data gathering. Several males were recorded in more than one year. When comparing song variables with color and condition variables, we used only the first recording made for a given male to avoid pseudoreplication. After excluding from our data set the duplicate recordings, those that were recorded before or after the laying period, recordings from males we never subsequently captured and measured (see below), and recordings that could not be confidently attributed to particular males, we were left with 63 usable recordings over the 3 years.
Approximately half the songs were recorded using Marantz PMD680 digital recorders and Sennheiser short shotgun microphones (ME66). The other songs were recorded using a Sennheiser long gun microphone (MKH816T) and a Sony Pro-2 professional Walkman.
Song processing
Each recording was processed by one of 2 observers (IRB or THP) using programs Raven 1.0–1.2 (Cornell Lab of Ornithology; THP) or Avisoft SASLab 3.2 (Avisoft Bioacoustics, Berlin; IRB). Before beginning this process, we developed a standard library of note and strophe (group of notes sung together in a predictable pattern, typically lasting 0.5–2.0 s, Figure 1) types to facilitate repeatable classification of all songs.
|
Broad frequency, harmonic scold-type vocalizations (types B1-B7 in Bijnens and Dhondt 1984
) constituted varying portions of dawn song. When these notes were produced (with or without high frequency, pure tone introductory notes), they were excluded from our analyses. These calls are often much more variable in length and form than other dawn song strophe types (IR Barr and TH Parker, personal observation), are produced year round and at any time of day in response to threats (Bijnens and Dhondt 1984), and in blue tits and related species, the rate of scold note production is a function of the type and proximity of a threat (Bijnens and Dhondt 1984; Templeton et al. 2005). Thus, we decided not to consider these notes as part of the blue tit sexual signal. For each strophe in a male's entire dawn song, we classified its type, measured its length, and measured the length between adjacent strophes (Figure 1). When summing number of strophe types produced, we did not include types produced only once during the song. In the rare case when a strophe type was sung just once, we could not be confident that it was ever sung again and was actually a part of the individual's repertoire rather than an accidental combination of notes produced, for instance, by being interrupted or switching strophe types in mid-strophe.
Because a male's dawn song typically ceases when his mate emerges from the nest box, the duration of his song is largely dependent on his mate's decision rather than his own. Therefore, we did not use duration of the song as a variable in the analyses, and where relevant, all song variables were converted to rates. The first song rate variable, number of strophes per minute, may be likely to reflect condition if singing is metabolically demanding. The next 3 rate variables were designed to capture information about song complexity. For the number of strophe types per minute and the number of switches between strophe types per minute, the total number of strophe types or the total number of times the bird switched between strophe types for the entire dawn song was divided by the total number of minutes in the dawn song. For the third song rate variable designed to address song complexity, number of strophe types per strophe, the number of different strophe types produced over the entire dawn song was divided by the total number of strophes sung during the entire dawn song. One measure of song complexity we did not consider was absolute repertoire size. This is because we recorded each male on only one morning and so had insufficient data to determine asymptotic repertoire size. However, the song complexity variables described above provide information regarding the repertoire per unit time or song and thus may be influenced by the same factors hypothesizes to influence overall repertoire size.
We also calculated mean strophe length and mean pause length. Both of these might be influenced by metabolic demands. For mean pause length, we excluded the pauses between strophe types, as these pauses were often longer than pauses between strophes of the same type. We calculated performance time, another variable possibly influenced by metabolic demands, as the length of a strophe divided by the sum of the length of the strophe and the length of the preceding pause. These calculations excluded the initial strophe in a series of given strophe type because, as stated above, pause length between strophe types was often longer than pause lengths within strophes. We used the mean performance time value in subsequent analyses.
We calculated the coefficient of variation (CV) for strophe length and pause length because we hypothesized that consistency of strophe length or timing among strophes might reflect aspects of condition such as current metabolic fatigue or a legacy of developmental stress. These were calculated separately for all strophe types, though strophe types sung fewer than 10 times (and their associated pauses) were excluded. The strophe and pause CVs for each strophe type were then averaged within a song to produce the average CV of strophe length and the average CV of pause length.
It has been shown that while blue tits are singing a given strophe type, the pause length between individual strophes tends to increase between successive strophes, possibly because of fatigue (Poesel and Kempenaers 2000). Because strophe length does not tend to change, the increase in pause length leads to a decrease in performance time (Poesel et al. 2001). This is termed drift. Although it occurs within a series of strophes of the same type (strophe series), it is reset after each change in strophe type (Poesel and Kempenaers 2000). We estimated the strength of drift for individual birds as the slope from a general linear model (PROC GLM, SAS 8.2) in which performance time was the dependent variable, time was the predictor, and strophe series was a covariate. By controlling for strophe series, a separate intercept was allowed for each run of strophes of a given type, thus accounting for the tendency for drift not to carry over when song type is switched. The slope was negative for nearly all birds, demonstrating that drift was close to ubiquitous. Some authors have classified songs into those showing drift and those not showing drift based on a significance test for negative slope (Poesel and Kempenaers 2000); however, we considered drift to be a continuous variable. First, unlike a previously reported pattern (Poesel et al. 2001), we observed a normal rather than a bimodal distribution of slope. Second, sample size differed dramatically based on the length of the song in our data set, so a significantly negative slope would be more likely from longer songs if the strength of the relationship between time and performance time was held constant.
Other field methods
At each box containing a blue tit nest, we noted first egg date, clutch size, and hatching date. We attempted to capture each adult between day 6 and day 14 of the chick-feeding period. Any unringed adults were given a uniquely numbered BTO ring at this point. For each captured adult, we quantified the color of the crown and primary coverts. Spectral reflectance was measured using a USB2000 spectrometer (Ocean Optics, Dunedin, FL) with illumination from a xenon light source (Ocean Optics PX-2). A sheath was fixed to the fiber end to standardize measuring distance (7 mm) and exclude ambient light. The fiber-optic reflectance probe was held (not pressed) perpendicular to the plumage, and 3–5 scans were taken from the center of each patch, removing the probe between each. The reflectance was measured relative to a WS-2 white standard scanned prior to each individual. We chose to limit our index of color to the one measurement that has consistently appeared important to sexual signaling in this species, UV chroma (R320-400/R320-700). This measurement addresses the specific importance of UV (Andersson and Amundsen 1997; Andersson et al. 1998; Sheldon et al. 1999). Its relevance in signaling is indicated by previous observations of its sexual dimorphism, correlation with mate choice, prediction of offspring sex ratio, and prediction of over-winter survival (Andersson et al. 1998; Sheldon et al. 1999; Griffith et al. 2003). We also measured 5 potential condition indices. For each male, we measured tarsus and wing length (millimeters) and mass (grams) and determined age (second year vs. after second year) based on wing covert molt. We did not combine the 3 body size measurements prior to analyses because they were not strongly correlated with each other (r = 0.17–0.33) and because we wished to compare our results with those from other studies that have considered these body size measurements separately. Our fifth measure of male quality or condition was recapture in the subsequent year. Because most individuals bred in nest boxes, all boxes were monitored, and a major effort was made to capture the breeding male at every nest box; most individuals that survived to breed on the site in a subsequent year were detected, and thus, the variable "recapture" can be considered a good approximation of local survival. None of these variables alone are ideal condition or quality indices, but they have been used in this context before (e.g., Bijnens 1988; Doutrelant et al. 2000; Nowicki et al. 2000). Age often predicts expression of condition-dependent signals, presumably because yearling males are in poorer condition or of lower dominance rank on average (e.g., Greene et al. 2000; Parker et al. 2003; Griffith and Pryke 2006). Survival has been shown to be influenced by aspects of condition (e.g., Gosler 1996; e.g., Lambrechts and Dhondt 1986). Tarsus length has been used as an index of condition experienced during development (Nowicki et al. 2000) and mass as an index of current condition (but see, Gosler and Harper 2000). In blue tits, older males have longer wings than yearling males (TH Parker, IR Barr, and SC Griffith, unpublished data) and so wing length may be a condition indicator as well. Strong correlations between one of these condition indices and song production would be consistent with a role of song in condition signaling.
Data analyses
All our continuously distributed variables that were not normally distributed were transformed to approximate normality. Recapture and age were classified as binomial variables. Birds were either recaptured or not in the next breeding season and were either second year or after second year adults.
We next identified axes of independent variation among our original 10 song variables. The first step was to generate a correlation matrix to determine which, if any, song variables were strongly correlated with each other. We then included each group of correlated song variables in its own principal component analysis (PCA) to generate a single principal component that would capture the majority of the variation in this group of variables. In all such cases, we substituted this principal component for the group of correlated variables in subsequent analyses.
Before conducting further data analyses, we identified potential covariates that we hypothesized might influence song production or condition variable values. Any of our variables of interest could have varied by year or geographically among the 9 subareas of the study site. Song variables could have been influenced by the timing of the recording, and so we considered recording date and both the number of days after clutch initiation and the number of days before egg laying ended that the bird was recorded. We considered an effect of low temperature on the dawn of recording because a cold night might force birds to forage at the expense of singing. We also considered the possibility of an effect of identity of the scorer of the song recording (IRB vs. THP). For the morphological measurements, we considered a role of age (wing length and mass), identity of the measurer, and date of measurement (mass).
We then determined whether our song variables were providing the same information as the best-studied signal in this species, UV–blue plumage color. We compared each of our song variables with the UV chroma of the wing coverts and cap using general linear models, which allow inclusion of relevant covariates.
We determined which hypothesized condition correlates, if any, predicted our song production variables by conducting general linear model analyses in a 2-step process. First, for each song variable, we analyzed a global model containing all 5 condition variables as predictors. If any condition variables had at least a marginally significant (P < 0.1) utility in predicting the song variable in question, a second analysis excluding all nonsignificant condition variables was conducted for that song variable. We obtained effect sizes and parameter estimates for the nonsignificant predictor variables from the global model and for the significant predictor variables from the reduced model. Appropriate covariates were included in each analysis, as discussed below in Results.
Published data
We located each published study comparing blue tit song with potential indices of individual quality or condition (Appendix 1). We then extracted all statistical information from each comparison of a song variable with a condition variable. Our intention was to compare results based on sign and magnitude of correlation coefficients. If no correlation coefficient was presented for a given relationship, we estimated one based on statistics available in the published paper with the statistical calculator option in the meta-analysis program MetaWin 2.1 (Rosenberg et al. 1997). In the one paper (Dreiss et al. 2006) where most data reporting was insufficient for inclusion in meta-analysis, we obtained more detailed information from the authors.
We located 5 studies reporting 42 relationships between blue tit song variables and condition-related variables in sufficient detail for meta-analysis. We received unpublished information for an additional 11 relationships from one of these studies (Appendix 1). We plotted all these effect sizes (correlation coefficients, y axis) describing relationships between song variables and aspects of condition in blue tits against the sample size for the respective studies (x axis). This should produce a funnel-shaped plot with a large vertical spread of points to the left, where sample sizes are small and sampling variance is thus large, and a narrower distribution of points to the right, where sample sizes are larger and sampling variance is thus reduced (Palmer 1999). The points should converge on the true effect size(s) as the sample size increases on the right side of the plot. If, when plotted, the published effect sizes do not follow a funnel-shaped distribution, publication bias may be to blame (Palmer 1999). Typical publication bias is against negative or nonsignificant results, especially in studies with small sample sizes. If the expected effect size is positive, then negative and modestly positive effects are likely to be nonsignificant when samples are small, and these results may often go unpublished (Rosenthal 1984). This can lead to a linear, rather than a funnel-shaped, distribution, with published effects from small-sample studies mostly high and positive, but becoming lower with increasing sample sizes (Palmer 1999).
To supplement our visual inspection of the funnel plot, we conducted a regression analysis to test for the negative relationship between sample size and effect size predicted in the case of publication bias. Before making this comparison, we converted all correlation coefficients (r) to Fisher's z-transformation {z=1/2ln[(1+r)/(1–r)]} and log-transformed sample sizes. We included all published effects in this analysis, as well as the 11 effects for which we obtained necessary details from the authors. Because not all the details of these data were published, this is not strictly a test for publication bias, but rather a test for bias in the data we obtained from published and unpublished sources. We did not include our field data in the test for bias in previously available data.
Finally, we conducted meta-analyses to determine whether certain song variables appeared more likely to be related to condition or certain condition variables seemed more likely to be related to song. We combined our field results with those obtained from other published and nonpublished sources. For these analyses, we used mixed models and bootstrap-generated significance tests (based on 9999 iterations) in program MetaWin version 2.1. Mixed models in meta-analysis account for expected heterogeneity in true effect sizes among song and condition variables and bootstrapping accounts for violations of distributional assumptions (Rosenberg et al. 2000). These analyses were based on Fisher's z-transformation of correlation coefficients (r) and variance inversely proportional to the sample size (vz = 1/n–3, where n = sample size for the given study [Rosenberg et al. 2000]). With program MetaWin, it is only possible to model the effect of one class of variables at a time. Thus, we first tested for differences among effect sizes (z) related to song variable, and then we tested for difference among effect sizes related to male quality or condition variables. For each song and condition variable, we asked whether the bootstrap-generated 95% confidence interval (CI) included zero.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
Identifying independent variation in songs
Of the 10 song variables we compared with each other using a correlation matrix (PROC PRINCOMP, SAS 8.2), 4 were not strongly correlated with any others (all |r| < 0.42) and the remaining 6 fell into 2 groups with high intragroup correlation (each variable with at least one |r| > 0.65, Table 1). Thus, a minimum of 6 largely independent types of variation (i.e., 6 variables) were present in these blue tit songs. The 4 variables not strongly correlated with any others, mean strophe length, drift, CV of strophe length, and CV of pause length, were considered separately in further song analyses. The variation in the 2 groups of correlated variables was summarized in 2 PCAs (PROC PRINCOMP, SAS 8.2). In the first PCA, we included the variables strophes per minute, performance time, and mean pause length. Principal component 1 (PC1) explained 88% of the variation in these 3 variables and had strong loading from all 3 variables (negative for pause length), so we used PC1 in further analyses, and we termed this variable "singing intensity." In the second PCA, we included the number of strophe types per minute, the number of strophe types per strophe, and the number of strophe type switches per minute. PC1 from this analysis explained 74% of the variation in these 3 variables, and all 3 variables had strong loadings. Thus, we used this new variable, which we termed "strophe turnover," in all further analyses. We were thus left with 6 song variables for further analyses.
|
Identifying covariates of song and condition variables
We identified covariates of song and condition variables and included them in further analyses of these traits. Strophe length had no covariates, but drift was related to subarea of the study site; singing intensity was a function of the number of days after clutch initiation that the recording was made; and strophe turnover, CV strophe length, and CV pause length were all related to the scorer of the song recording (IRB vs. THP). The effect of identity of the song scorer may have resulted from the nonrandom geographic and temporal distributions of the sets of songs analyzed by the 2 scorers. We examined the scorer effect for 38 recordings made in one subarea of the study site (28 scored by IRB, 10 scored by THP), and the scorer effect disappeared for strophe turnover and CV pause length. We retained scorer as a covariate in these cases with the understanding that factors correlated with scorer identity were likely influencing the patterns we observed. Among male quality or condition variables, age had no covariates but wing length was a function of age, measurer, and subarea of the study site; tarsus length was related to measurer and year; and year also influenced both mass and recapture.
Assessing redundancy of song and color variables
According to our general linear model analyses (PROC GLM, SAS 8.2), the 6 largely independent song variables were not good predictors of the 2 color variables in the 61 individuals for whom we had sufficient color data (Table 2). No relationships were significant even before correction for multiple comparisons, and all correlations (|r|) were less than 0.24.
|
Condition indices as predictors of song output
We found evidence that some measured aspects of condition predicted song production in blue tits. Using general linear models with normal error (PROC MIXED, SAS 8.2), we identified one or more body size traits that predicted variation in 4 of 6 song variables (|r| < 0.41, Table 3). Birds with longer tarsi tended to sing longer strophes (Table 3). Tarsus length also explained the distribution of CV of strophe length, but counter to prediction, males with longer tarsi sang strophes that were less consistent in length within a strophe type (Table 3). One of the 2 strongest pattern was for heavier males to have a higher singing intensity (a greater proportion of time singing) (Table 3), and this was the only song–condition relationship to remain significant (P < 0.05) if we conducted a conservative Bonferroni adjustment for 30 comparisons (6 song variables x 5 condition variables). The relationship between body size and drift was inconsistent, with a strong trend for heavier males to show more drift, but a somewhat weaker pattern of males with longer tarsi showing less drift. Neither age nor recapture significantly predicted any song variables (Table 3).
|
Summarizing published relationships between song and condition
Published correlations between song variables and male condition variables ranged from r = –0.29 to 0.86 (Appendix 1). Both a visual assessment of the funnel plot (see especially open and filled circles, Figure 2) and a quantitative assessment of the same data indicate that there may have been bias against publishing of negative or nonsignificant relationships between condition variables and song production in blue tits (F1,51 = 11.8, P = 0.001, r = 0.43, slope = –0.49 ± 0.14 standard error [SE]). Studies with small sample sizes tended to report large positive effect sizes, and reported effect sizes declined with increasing sample size. Several of the particularly strong effects at small sample sizes describe relationships between testosterone level and song variables. Because effect sizes may tend to be larger for studies of physiological effects (Møller and Jennions 2002
), we conducted our test for publication bias a second time, excluding the testosterone effects. We still found a significant negative slope with this reduced data set (F1,48 = 4.0, P = 0.050, r = 0.28, slope = –0.31 ± 0.15 SE), again supporting the hypothesis of publication bias, though the effect seemed less marked.
|
When we combined our results with previously published results using meta-analysis, we found evidence that certain song traits more consistently correlated with condition than others (Table 4). Singing intensity and strophe length both had significantly positive effect sizes. Within-strophe drift, a variable assessed by only one study, also showed a significant positive effect size. A similar pattern was observed when the testosterone studies were excluded, but the strophe length effect was no longer significant (Table 4). Only one possible condition or quality indicator, testosterone level, had a significantly positive relationship with song variables, although the 95% CI for mass was only marginally nonsignificant (Table 4). These results are consistent with certain song variables signaling at least some aspects of condition in blue tits. However, these results must be interpreted with caution because the evidence for publication bias suggests that our rather low meta-analysis–generated effect size estimates may be biased upward.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
There are 5 major results of this study. We show that 1) a minimum of 6 largely independent variables exist in dawn blue tit song, 2) none of these variables correlated strongly with variation in an established measurement of color signal expression, 3) of male quality or condition indicators, aspects of body size, but not age or a survival index, predicted variation in certain song measurements, especially singing intensity, 4) a bias toward publication of strong positive results may have inflated the perceived importance of blue tit song as a condition-dependent signal, and 5) a subset of song variables, particularly strophe length and singing intensity, may be most likely to signal condition in this species. Our large sample sizes, combined with our thorough meta-analysis, make our conclusions particularly robust.
In blue tits, the dawn song appears to be a multicomponent signal, and we identified 6 largely independent forms of variation. There are any number of song variables that we did not quantify, such as amplitude (Forstmeier et al. 2002), within-strophe drift (Bijnens 1988), and the ratio of trill rate to frequency breadth (Podos et al. 2004), and with a complex signal such as song, the number of variables examined is limited only by the researcher's time and creativity. Thus, 6 independent song variables is a minimum estimate. However, just because one can measure a song component does not mean that it serves as a signal. Without demonstrated signal utility through study of conspecific responses to different song components, it will remain unknown what aspects of blue tit song serve as signals and therefore how many meaningful forms of variation exist.
We demonstrated that our measured song variables are not strongly correlated with UV chroma of cap and wing plumage in blue tits, a well established sexual signal. This is consistent with other findings (Dreiss et al. 2006), and thus, we can unambiguously reject the redundant signal hypothesis as an explanation of the evolution of the song and color signals we measured in this system (Møller and Pomiankowski 1993). To the extent that these different measures are signaling information about individual males, they are signaling different information. Other alternative hypotheses explaining the presence of multiple ornaments (e.g., Møller and Pomiankowski 1993; Schluter and Price 1993; Iwasa and Pomiankowski 1994) cannot be rejected or supported.
We found significant relationships between song variables and 2 of the 5 condition variables. These 2 condition variables were mass and tarsus length, and at least one of them was a significant predictor (before correction for multiple comparisons) of each of the following: strophe length, drift, singing intensity, and CV strophe. Although the relationship between strophe length and tarsus length in our field data was not robust to correction for multiple comparisons, when we combined our strophe length–male quality relationship effects with the 17 published relationships involving strophe length, we found a significant, albeit weaker, positive effect. Evidence for condition dependence of drift was mixed. Drift was positively related to one of our body size measurements but negatively related to another. Further, the effect of drift was very small and not significant in meta-analysis. Our results suggested the strongest role for the composite variable singing intensity in predicting male quality: heavier males tended to spend a larger proportion of their time singing, had shorter pauses, and sang more strophes per minute, and this result was highly significant. As with strophe length, meta-analysis results indicate that the proportion of time a male blue tit spends singing provides a signal of some aspect of his quality. Although the presence of publication bias means we must interpret the meta-analysis results with caution, it appears that both strophe length and singing intensity may be condition dependent. These 2 song variables could be linked to body size in multiple ways. Although it is possible that larger males have more energy reserves to devote to song production (Thomas 2002), another possibility is that song production is socially mediated (Mennill et al. 2002) and that only larger males can afford to risk aggression from other males (Parker and Ligon 2002) by singing longer or more frequent strophes.
Several other lines of evidence indicate that the components of blue tit song that we measured may not be general indicators of male condition. Although correlated with some body size traits, our measurements of blue tit song appeared completely unrelated to age or the probability of being recaptured after the subsequent winter. Further, in meta-analysis, neither of these condition variables was related to song. Many condition-dependent male sexual signals are age dependent (e.g., Greene et al. 2000; Parker et al. 2003; Griffith and Pryke 2006), presumably because older males have higher dominance rank or are more efficient at performing the tasks that determine the environmental effects on their signal expression. Thus, either age is unrelated to male condition in blue tits or, more likely, the song variables we examined are not particularly sensitive to male condition or the components of male condition that relate to age. In tits, it is well demonstrated that aspects of male quality and condition, including male dominance, influence overwinter survival (e.g., Gosler 1996; e.g., Lambrechts and Dhondt 1986), and so presumably our negative results with regard to recapture in the subsequent year are not due to a lack of correlation between male survival and condition but rather to song traits lacking dependence on the aspects of condition influencing survival.
Our effect sizes are within the wide range observed in other bird species (–0.6 < r < 0.6) where measures of condition similar to ours have been compared with aspects of song production (e.g., Lampe and Espmark 1994; Galeotti et al. 1997; Otter et al. 1997; Balsby 2000; Rinden et al. 2000; Gil et al. 2001; Forstmeier et al. 2006; Kipper et al. 2006). As with the blue tit data, results vary among and within studies. Certain combinations of condition and song variables appear important in some studies, and other combinations appear important in other studies. One fairly consistent pattern is that of an age effect on song complexity variables (Lampe and Espmark 1994; Balsby 2000; Rinden et al. 2000; Gil et al. 2001; Forstmeier et al. 2006), something we failed to detect in blue tits. Relationships between other condition and song variables are more varied, and a formal meta-analysis will be required to determine how the condition variables we studied tend to be related to song production across bird species.
Our meta-analysis demonstrates the importance of formal synthesis of published results rather than casual review of notable published relationships. Without meta-analysis of the blue tit song literature, any general conclusion we might have drawn would have been dubious because of the heterogeneity in results among studies. Meta-analysis does not eliminate uncertainty concerning interpretation of published results, but it provides a formal framework for assessing these results and thus minimizes biased interpretation. It also allows for identification of potential biases in the published literature and as such promotes caution in interpretation when such bias is identified. Because sexual signal content may vary among populations of a species (Badyaev et al. 2001; Forstmeier and Leisler 2004), an appropriate interpretation of our meta-analysis results is that they represent the distribution of, and average patterns for, song–condition relationships in this species. Thus, regardless of whether the distribution of data in our meta-analyses represents geographically and temporally divergent patterns or simply sampling error, we can still conclude that many aspects of blue tit song are either inconsistent or weak predictor of male condition or both. Even the most consistent blue tit song signal of condition, singing intensity, has a weaker meta-analysis–generated effect size than that detected in our field data and may not be condition dependent in all situations (e.g., Dreiss et al. 2006).
Although we found some evidence of blue tit song signal utility, we cannot eliminate the possibility that unmeasured components of blue tit song production are better signals of quality. For instance, although we quantified aspects of song complexity, we did not measure absolute repertoire size, a song variable found to be important in some case (Nowicki et al. 2000; Spencer et al. 2003, 2004, but see Forstmeier and Leisler 2004). We also did not consider the social context of song production. As with previous research on signal content in blue tit song, our song measures did account for neither the proximity of conspecifics such as competing males nor the ongoing vocalizations of other chorusing males. Male songbirds can adjust their singing behavior in response to that of their neighbors (Catchpole and Slater 1995), and patterns of counter singing have been shown in a closely related species to be an important source of information for females assessing male quality (Mennill et al. 2002). Future work with blue tits will need to address these possibilities if the signal content of male song is to be understood.
It could also be that we measured appropriate song variables but did not measure the best component of male quality to compare with these variables. One such unmeasured variable is male dominance rank, which can relate to song production and have meaningful fitness implications (Lambrechts and Dhondt 1986; Mennill et al. 2002). In any study with negative results, it always remains possible that important variable(s) went unmeasured.
Despite some unmeasured variables, we are still in a position to draw certain robust general conclusions. Many of the song variables we studied have previously been hypothesized to signal male condition (e.g., Eberhardt 1994; Nowicki et al. 1998; Poesel and Kempenaers 2000; Foerster et al. 2002), but for most of these song components, we found at best mixed evidence that they consistently signal male quality in blue tits. Thus, researchers need to continue to test hypotheses concerning condition dependence on a case by case basis and should remain hesitant to assume condition dependence of sexual signals without strong empirical support.
We are not yet in position to identify the evolutionary mechanism maintaining variation in blue tit song. However, one aspect of variation, the production of different song types, may be maintained by the necessity of neighbor recognition as seen in many other species (Catchpole and Slater 1995) and more generally for other types of multimodal signals (Dale 2000). Maintenance of variation in other traits is less clear. Although a lack of correlation between signal expression and condition is likely in some models of signal evolution that do not rely on a fitness payoff to female choice, such as sensory bias (Ryan 1998), theory predicts this relationship can be absent even in cases where genetic benefits to mate choice are important, for instance, if males trade-off condition against sexual attractiveness (Kokko et al. 2002). Our results lend support to the hypothesis that certain aspects of sexual traits signal condition but also lead us to reject the simple scenario of a sexual signal providing consistent information about male condition. Given the common perception of the importance of this sort of condition signaling, this is an important conclusion.
|
APPENDIX 1 |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX 1REFERENCES |
|---|
Comparisons of blue tit song variables to potential condition indices in other studies