Pat Monaghan, Division of Environmental and Evolutionary Biology, Institute of Biomedical and Life Sciences, Graham Kerr Building, University of Glasgow, Glasgow G12 8QQ, UK. E-mail: P.Monaghan@bio.gla.ac.uk
1Using data from a 20-year study of individually marked red-billed choughs, we examine how reproductive performance varies with age in male and female breeders, and investigate whether population-level trends result from changes in individual performance and/or the phenotypic composition of the breeding population.
2Across the population, mean clutch size, the probability of breeding successfully and the number of offspring fledged during successful attempts increased and then declined with female age. Male age did not explain a significant proportion of the residual variation.
3All three measures of reproductive performance improved and then declined with age within individual females.
4Females that died young laid relatively small clutches and fledged few offspring before death. Thus mean performance improved across young age classes partly because some poor breeders were absent from older age classes.
5Females that ultimately reached the greatest ages had laid small clutches and fledged few offspring during their first few breeding attempts. Females that were more productive when they were young had relatively shorter lives. These data indicate a trade-off between early reproduction and future survival in choughs, and suggest that individuals that reach old age are phenotypically distinct from an early stage in their breeding lives.
6We emphasize that age-specific changes in mean reproductive performance observed across wild populations are due to a complex interplay between improvement and senescence at the individual level, as well as changes in the phenotypic composition of the breeding population.
An individual's life-history comprises its age-specific pattern of growth, reproduction and survival (Stearns 1992). While theories predicting broad patterns of life-history variation are reasonably well developed (for example the occurrence of iteroparity vs. semelparity, Schaffer 1974; Orzack & Tuljapurkar 1989), the way in which traits should vary over an iteroparous lifetime is less well understood (Stearns 1992). Because age-specific patterns of performance can influence social structures and population dynamics (Pärt, Gustafsson & Moreno 1992; Coulson et al. 2001), understanding how reproduction and survival vary with age is an important area for development in life-history theory.
Reproductive performance could, initially, improve with age if older individuals forage or breed more efficiently, or acquire higher quality mates or territories (constraint hypotheses, Curio 1983; Nol & Smith 1987; Komdeur 1996; Pärt 2001). Performance could also improve because an individual's optimal level of reproductive effort increases with age due to changes in reproductive costs or residual reproductive value (restraint hypotheses, Williams 1966; Pianka & Parker 1975; Clutton-Brock 1988). Similarly, changes in performance in old age may be due to constraint as a consequence of somatic deterioration, or to further optimization of an individual's life-history. With respect to the latter, the optimal reproductive effort in old age will depend on the complex interplay between reproduction and survival; both a decrease (to maximize remaining life span) and an increase (if survival prospects decline in old age) are theoretically possible. The overall pattern of age-specific variation in reproductive performance is therefore difficult to predict, and empirical data from natural populations are essential to advance our overall understanding of life-history strategies.
Measuring how reproductive performance varies with age in wild populations is challenging (Partridge & Barton 1996). As ages can often be assigned only by marking individuals at birth and extrinsic hazards mean that most individuals die before old age, even long-term studies may not amass sufficient data to describe performance in the oldest age classes. Furthermore, individuals within populations differ in their ability to acquire and utilize resources and thus in their optimal life-history allocation (van Noordwijk & de Jong 1986). The resulting phenotypic heterogeneity can mask or exacerbate individual allocation patterns when trends are averaged across a population (Vaupel & Yashin 1985; McDonald, Fitzpatrick & Woolfenden 1996; Cam & Monnat 2000). While reproductive performance could improve across young age classes because component individuals perform better, mean performance could also improve because individuals that breed poorly also die young, or those that recruit later breed more successfully (selection and recruitment hypotheses, Curio 1983; Nol & Smith 1987; Forslund & Pärt 1995). Less well recognized, however, is the possibility that variation in individual life-history optimization could also cause apparent senescent declines in mean reproductive performance measured across a population. For example, if individuals with high reproductive effort consequently have poorer survival, mean reproductive performance may decline across older age classes because only individuals that invest relatively little in each attempt survive to old age. While estimates of mean age-specific performance are valuable for population modelling, cross-sectional averages cannot be used to test predictions and assumptions of individual-based life-history theory. Thus it is crucial to clarify the extent to which population-level trends reflect changes in individual performance rather than the phenotypic composition of the breeding population (Forslund & Pärt 1995).
Table 1. Studies that have investigated whether mean reproductive performance improves across young breeders because individuals that initially breed poorly also die young. Studies tested for positive correlations between an individual's early reproductive performance and an index of subsequent survival: an individual's final lifespan (analysis A, a quantitative index) or its survival to some future breeding season (analysis B, a binary index). Some analyses additionally stratified data by breeder age or environmental conditions (analysis C). Studies reported a statistically significant positive correlation (+), a statistically significant negative correlation (–) or no significant correlation (0) between each component of current reproduction and subsequent survival. Overall, seven of 24 (29%) studies reported a positive correlation between a component of reproductive performance and survival, and thus supported the differential mortality hypothesis
Using data from a 20-year study of individually marked red-billed choughs Pyrrhocorax pyrrhocorax Linnaeus inhabiting the Scottish island of Islay, we investigate whether three measures of reproductive performance, clutch size, the probability of breeding failure and the number of offspring fledged during successful attempts vary with the ages of the breeding male and female. Focusing on relationships between female age and performance, we use longitudinal and cross-sectional data to investigate whether overall patterns reflect age-specific changes in individual performance, or population-level consequences of covariation between reproduction, recruitment age and longevity. Thus we investigate the role of variation in individual life-histories in creating population-level patterns in this system, and discuss the implications of our findings for age-specific variation in survival in choughs.
The small Scottish population of red-billed choughs is confined to the western seaboard (Monaghan et al. 1989). The inner Hebridean island of Islay, which lies 25 km west of the Argyll coast in south-west Scotland (55°N, 6°W), has long been the main stronghold of the Scottish population (Monaghan et al. 1989) and currently holds approximately 60 breeding pairs of choughs (Finney & Jardine 2003). Nineteen of the remaining 20 pairs in Scotland breed on the adjacent island of Colonsay (Finney & Jardine 2003). These birds are resident, breed once each year and rarely miss breeding years, and nest in traditional cavity sites that have been accurately mapped (Bignal, Bignal & McCracken 1997). Each year from 1981 to the present, the Scottish Chough Study Group (SCSG) checked 50–60% of occupied nest sites on Islay towards the end of chick-rearing and recorded the number of offspring fledging during each breeding attempt (Bignal et al. 1997). During 1981–96, nests were additionally visited early during the breeding season to record clutch size. To provide a sample of known-age adults, 992 fledgling choughs were individually colour-ringed between 1981 and 2000, as part of an ongoing colour-ringing project. Breeding adults were checked for colour-rings during subsequent seasons, and sexes were distinguished by behaviour and relative size (Tella & Torre 1993). Thus, while not all colour-ringed breeders were monitored in all years, considerable data describing the reproductive performance of known-age parents have been accumulated. As choughs on Islay show high site fidelity and rarely leave the island (Monaghan 1989; Bignal et al. 1997; SCSG, unpublished data), the disappearance of a colour-ringed breeder from its territory generally reflects mortality rather than dispersal (see also Reid et al. 2003). Repeated nest checks on both Islay and Colonsay, plus observations made outwith the breeding season, allowed the years in which colour-ringed choughs recruited to and disappeared from the population to be recognized with confidence. Thus, age at first breeding and breeding lifespan (the total number of breeding years) were recorded for most colour-ringed breeders whose reproductive performance was regularly monitored (Reid et al. 2003).
Backward elimination linear regression models were used to test whether breeder age explained a significant proportion of variation in clutch size. Due to the frequency of breeding failure, the number of offspring fledging per nest was bimodally distributed, and was not normalized by transformation. Consequently, predictors of breeding failure (whether or not any offspring fledged, a binary variable) and fledging success (the number of offspring fledged during successful attempts) were investigated separately, using logistic and linear regression models, respectively. Thus, we examined relationships between breeder age and three different measures of reproductive performance. Where inspection suggested that relationships were non-linear, squared terms were included in models. Because a significant quadratic regression does not prove that a dependent variable declines after the peak rather than simply levelling off, we additionally tested whether independent and dependent variables were correlated negatively beyond empirical peaks in performance. To distinguish influences of male and female age on reproductive performance, initial analyses used data from breeding attempts where the ages of both pair members were known. As these analyses indicated that performance varied primarily with female age (see Results), to maximize sample sizes all data from known-age females, including attempts where male age was unknown, were used in further analyses.
On Islay, mean reproductive performance varied among years (across all observed attempts, clutch size: anovaF15,372 = 1·7, P = 0·05, frequency of breeding failure: = 34·5, P = 0·02, fledging success: anovaF19,444= 1·7, P = 0·04). To minimize confounding effects of observation year on age-specific patterns of performance, we therefore needed to incorporate year-effects into our analyses. The most rigorous approach would be to model year as a random factor (Bennington & Thayne 1994). However, relatively few known-age choughs were monitored in each year, and in some years these are likely to have been a non-representative subset of the age-structured population (for example, no old breeders were identifiable during early study years). Estimates of random year-effects based solely on data from known-age individuals may therefore be biased. Accordingly, we standardized clutch size and fledging success data by subtracting the annual population-wide mean performance (calculated from all observed breeding pairs, not solely known-age breeders) from each observed value, and dividing by the population-wide standard deviation (Zar 1999). To allow for annual variation in the probability of breeding failure (the binary variable), the proportion of all observed attempts that failed in each year was included as a covariate in logistic models. Similar approaches have been widely used in previous analyses (e.g. Perdeck & Cavé 1992; Robertson & Rendell 2001; Laaksonen et al. 2002). Although between-year variation in reproductive performance is linked to Islay's weather (Reid et al. 2003), mean performance may also vary with the age-structure of the small breeding population. However, if year-specific performance estimates were biased by unequal inclusion of age classes, our standardization procedures would tend to under-rather than over-estimate age-effects on performance.
As many individuals bred in multiple years, the data included observations that were not independent. While repeat observations of known individuals are essential for investigating age-specific variation in individual performance, correlations within dependent variables can deflate variance estimates, clouding interpretation of F-ratios. Although the degree of non-independence in our data was relatively minor, we ran further analyses to validate our conclusions. While modelling relationships between female age and performance, we included female identity as a random factor in generalized linear mixed models (Bennington & Thayne 1994; Steele & Hogg 2003). However, as the ages of paired males and females are correlated across the duration of the pair-bond, the independent effects of male and female age could not be separated while specifying pair identity as a random factor. Analyses where both breeder ages were included were therefore repeated using a single randomly selected observation per pair. Analyses were carried out in SPSS (SPSS Inc, Version 10·0). Mean standardized age-specific performance is presented ± 1 SE.
individual change and differential recruitment and mortality
For most individuals, reproductive performance was not recorded at every age. To investigate whether performance varied with age within individual females, lifespan was divided into three age classes (young, middle-aged and old). Paired tests were used to examine whether performance changed as individuals moved between classes. This approach required that discrete age categories be defined. However, it was not always clear where divisions should be drawn: for example, whether 4-year-olds were young or middle-aged (see Results). To minimize ambiguity, we compared a female's performance aged 2–3 (young) to that aged 6–9 (middle-aged) and 12 years or older (old). While few individuals were monitored within all three categories, data from two categories were available for a reasonable number, although inevitably few reached old age. Where individuals were monitored more than once within each category we averaged performance across observed attempts.
If mean reproductive performance were to improve across young age classes because poor breeders died young, a positive correlation between a female's early performance and her breeding lifespan would be predicted (Thomas & Coulson 1988; Wooller et al. 1990; Cam & Monnat 2000). However, if females that consistently invested less in reproduction were in fact more likely to survive, individuals comprising the oldest age-classes within the population-level analyses may equally reproduce at a relatively low rate. Thus, to investigate whether differential mortality of reproductive phenotypes contributed to trends in mean performance, we used further linear and logistic regression models to test for linear or quadratic relationships between age-specific reproductive performance and a female's breeding lifespan.
Overall, 87% of females that recruited on Islay did so aged 2 or 3 years (Reid et al. 2003). To test whether mean reproductive performance improved with age because individuals that recruited later performed better, we compared reproductive performance between 3-year-old females that were making their first and second breeding attempts. Females whose recruitment age or lifespan were uncertain were excluded from these analyses.
Capture–mark–recapture models (program mark, White & Burnham 1999) were used to estimate age-specific probabilities of apparent survival from resightings of colour-ringed choughs (see also Reid et al. 2003). As male and female choughs are not reliably distinguishable in the field other than by breeding behaviour, the sexes of some colour-ringed birds remained unknown. Thus we initially used encounter histories of known-sex individuals to test whether resighting or apparent survival probabilities differed between males and females. General models with full sex- and age-dependence in survival and sex- and year-dependence in resighting probability were fitted initially. Parametric bootstrap goodness-of-fit simulations were used to test whether starting models fitted the data, and the variance inflation factor (č) was calculated to quantify the degree of data overdispersion (Cooch & White 1998). Although bootstrapping may underestimate overdispersion (White 2002), more robust estimators are not yet fully developed for models other than the fully time-dependent Cormack–Jolly–Seber formulation (Cooch & White 1998; White 2002). General models were then constrained with respect to sex and age, and Akaike's Information Criterion (AIC), corrected for small sample sizes and overdispersion (qAICc), was used to select the most parsimonious model (Anderson & Burnham 1999). As we could not predict the relationship between age and mean survival probability clearly, we merged consecutive age-classes sequentially and used qAICc values to identify the most parsimonious groupings (as Orell & Belda 2002).
age and mean reproductive performance
In species where partners remain together over multiple seasons, male and female ages will inevitably be correlated across all breeding attempts. In choughs, male and female ages were not correlated at pair formation (r39 = 0·06, n = 41, P = 0·72) and remained moderately correlated across all observed attempts (r = 0·4).
Across breeding attempts where both male and female ages were known, clutch size, the probability of breeding failure and fledging success (offspring fledged during successful breeding attempts) varied significantly with female age, while male age did not explain a significant proportion of the residual variation (Table 2). Analyses based on single randomly selected observations per pair gave similar results. As there was no evidence for an independent effect of male age, we verified relationships between female age and reproductive performance using all observations of known-age females. Mean clutch size, the probability of breeding successfully and fledging success increased and then declined significantly with female age (Fig. 1). Figure 1 presents standardized parameter values, correcting for year effects. To indicate the absolute magnitude of age-specific variation in mean performance, mean observed clutch size increased from 3·8 ± 0·2 in young females (ages 2–3) to 5·2 ± 0·1 in middle-aged females (ages 6–9) and declined to 4·4 ± 0·4 in old females (age 12 or over). Mean probabilities of breeding failure were 0·60 ± 0·01, 0·24 ± 0·01 and 0·36 ± 0·03 for these same categories, while mean fledging success was 1·4 ± 0·2, 2·4 ± 0·2 and 1·6 ± 0·3. Thus mean reproductive performance varied substantially with female age. We examine below whether overall patterns resulted from changes in individual performance and/or the differential recruitment or survival of reproductive phenotypes.
Table 2. Relationships between standardized clutch size (CS), the probability of breeding failure (BF) and standardized fledging success (FS) and male and female breeder ages. Analyses included Nobs total observations of Nfemales individual females, Nmales individual males and Npairs different pairings. Partial F or χ2 statistics are presented for each independent variable. For variables that were not retained in the final model, statistics reflect their addition to the minimum acceptable model. Mean clutch size, the probability of breeding failure and fledging success varied significantly with female age and female age2, but not male age or male age2. Interaction terms were not significant (all P > 0·3)
F = 26·3
F = 21·9
F = 0·3
F = 1·3
F2,81 = 14·1
P < 0·001
P < 0·001
P = 0·61
P = 0·25
P < 0·001
R2 = 0·27
χ2 = 5·2
χ2 = 6·1
χ2 = 0·1
χ2 = 0·2
χ2 = 5·9
P = 0·023
P = 0·014
P = 0·76
P = 0·63
P = 0·02
P = 0·015
R2 = 0·12
F = 6·3
F = 7·5
F = 1·2
F = 1·2
F2,94 = 4·0
P = 0·010
P = 0·007
P = 0·38
P = 0·28
P = 0·02
R2 = 0·08
Individual females laid significantly larger clutches in middle-age than when young (year-standardized means of 0·50 ± 0·15 and −0·47 ± 0·19, respectively, paired t-test t14 = 4·1, n = 15, P = 0·001). Individuals laid smaller clutches when old than in middle-age (standardized means for these individuals of −0·28 ± 0·24 and 0·60 ± 0·22, respectively, paired t-test t5 = 2·6, n = 6, P = 0·05), but similar clutches to those laid when young (paired t-test t4 = 1·2, n = 5, P = 0·32).
Individual females were less likely to fail in middle-age than when young (median residual frequencies of failure of −0·48 and 0·05, respectively, Wilcoxon matched pairs test, Z20 = 2·1, P = 0·04), and when middle-aged than when old (medians of −0·13 and 0·33, respectively, for these individuals, Wilcoxon matched pairs test, Z9 = 2·0, P = 0·05). Females that lived to old age were no more susceptible to breeding failure when old than when young (Wilcoxon matched-pairs test, Z7 = 0·3, P = 0·80).
Females that bred successfully when both young and middle-aged fledged more offspring in middle-age (standardized means of −0·14 ± 0·20 and 0·36 ± 0·15, respectively, paired t-test t14 = 2·3, n = 15, P = 0·04). Females that bred successfully when old fledged fewer offspring than they did when middle aged (standardized means of –0·88 ± 0·27 and 0·23 ± 0·22 for these females, respectively, paired t-test t6 = 2·9, n = 7, P = 0·03). Fledging success when old did not differ from that when young (paired t-test t4 = 0·7, n = 5, P = 0·54). Thus clutch size, the probability of breeding successfully and fledging success all improved and then declined with age within individual females.
phenotypic composition of age classes
To test whether mean performance improved across young age-classes because better females recruited slightly older, we compared the performance of 3-year-old first-time breeders with that of 3-year-olds that had recruited at age 2 years. These groups did not differ with respect to any of the three measures of reproductive performance (Table 3).
Table 3. Mean (± 1 SE) standardized clutch size (CS), residual frequency of breeding failure (BF) and standardized fledging success (FS) of 3-year-old females that recruited aged 2 years (second-time breeders) and aged 3 years (first-time breeders)
Recruitment age (years)
−0·48 ± 0·52
−0·65 ± 0·25
t23 = 0·50, n = 25, P = 0·63
−0·31 ± 0·28
0·12 ± 0·19
Z = −1·2, n = 34, P = 0·25
−0·13 ± 0·29
−0·28 ± 0·24
t23 = 0·42, n = 25, P = 0·68
To investigate whether differential mortality of reproductive phenotypes contributed to changes in mean age-specific performance, we examined relationships between a female's performance at each age up to 8 years and her final breeding lifespan. Beyond 8 years sample sizes were small and the distribution of possible lifespans was inevitably truncated; 9-year-old choughs could not have short lifespans. In 3- and 4-year-old females, standardized clutch size varied non-linearly with lifespan: females that ultimately had either short or long breeding lives had laid smaller clutches when aged 3 and 4 than females that ultimately bred an intermediate number of times (Fig. 2). Clutch sizes were largest in 3- and 4-year-old females that ultimately bred approximately five to seven times, and declined in females that ultimately survived longer (F3,37 = 5·6, P = 0·003, age P = 0·018, lifespan P = 0·009). However, this relationship between age-specific clutch size and lifespan was no longer evident in middle age: clutch sizes of 5–8-year-old females did not vary with their final lifespan (P > 0·6 for ages 5–7, P = 0·13 for performance at age 8).
The probability of breeding failure did not vary significantly with a female's breeding lifespan within any age class (all P > 0·25).
Across successful breeding attempts where a female's age and final breeding lifespan were known, females that ultimately had short or long breeding lives had fledged fewer offspring when aged 3, 4, 5 and 6 than females that ultimately bred an intermediate number of times (Fig. 3). Fledging success peaked in young females that ultimately bred six to eight times, and declined with increasing lifespan (F4,21 = 10·0, P < 0·001, age P = 0·07, lifespan P = 0·001). The number of offspring fledged by 7- and 8-year-old females did not vary significantly with the female's final breeding lifespan (P = 0·88 and P = 0·13, respectively).
In summary, a female's breeding lifespan varied with her productivity during her first few breeding attempts; both particularly short- and particularly long-lived females produced relatively few offspring early in life (Figs 2 and 3).
age and survival
The general model (ϕ(a,s) p(y,s), Table 4) fitted the data from choughs of known sex (P = 0·56, č = 1·08). Constrained models provided no clear evidence that, across these data, resighting or apparent survival probabilities differed between males and females (Table 4a). Thus sexes were pooled, and the encounter histories of all colour-ringed choughs were used to investigate age-specific variation in apparent survival. The general model ϕ(a) p(y) fitted the data (P = 0·32, č = 1·10), and survival and resighting probabilities varied with age and year, respectively (Table 4b). Resighting probability averaged 0·73 ± 0·04 across years (range 0·42–0·95). The most parsimonious age-structured model comprised seven age-classes, covering ages 1, 2, 3–5, 6, 7–9, 10–12 and 13 + (Table 4b). Apparent survival probability declined in choughs aged 13 and older (Fig. 4, Table 4b).
Table 4. Capture–mark–recapture models of sex- and age-dependence in apparent survival probability (ϕ) and sex- and year-dependence in resighting probability (p), using resightings of (a) colour-ringed choughs of known sex (n = 300) and (b) all colour-ringed choughs (n = 923). a, y and s indicate full age, year and sex dependence in probabilities. Adjusted AIC scores (qAICc), the AIC difference between the current and the best model (ΔqAICc), AIC weights, the number of parameters and the deviance are presented for each model. The most parsimonious model in each section is indicated in bold. The number of survival parameters differed between (a) and (b) because sexes were only distinguished when choughs recruited. In (a), ϕ for years 1 and 2 therefore equalled one for both sexes
(a) Choughs of known sex
1 ϕ(a,s) p(y,s)
2 ϕ(a,s) p(y)
3 ϕ(a) p(y)
(b) All colour-ringed choughs
4 ϕ(a) p(y)
5 ϕ(a) p
7 ϕ(a1,2,3–5,6,7–9,10–12,13 +) p(y)
8 ϕ(a1,2,3−5,6,7−9,10+) p(y)
State variables other than age, including breeding experience, the time since territory occupation and pair-bond duration, can influence reproductive performance, reflecting very different biological phenomena (Pärt 1996, 2001; Cam & Monnat 2000; Black 2001; Pyle, Sydeman & Hester 2001). These factors may both cause apparent age effects, and explain residual variation in performance after age is accounted for. However, in choughs, breeding experience, residence time and pair-bond duration were highly correlated with age (r = 0·98, r = 0·75 and r = 0·55, respectively). Their effects could not be separated statistically as attempted in systems where recruitment age varies or established breeders miss seasons or frequently change sites or mates (e.g. Cam & Monnat 2000; Pyle et al. 2001). We therefore focused solely on investigating age-specific patterns of reproductive performance.
Mean reproductive performance improved from young to middle-aged females partly because individual females laid larger clutches, were less likely to fail and fledged more offspring in middle-age than when young. Similar improvements in individual performance have been demonstrated in other species (Clutton-Brock 1988; Saether 1990; Forslund & Pärt 1995). While we cannot distinguish conclusively the importance of constraint vs. reproductive restraint in limiting early performance in choughs, there was no evidence that adult survival probability dropped markedly during the period of improvement (Fig. 4). In relatively long-lived birds such as choughs, residual reproductive value may not decline fast enough for early improvements to be due solely to adaptive increases in reproductive effort (Forslund & Pärt 1995). As choughs showed high mate and site fidelity, performance could not in general have improved because middle-aged females acquired better males or territories. Some constraint, such as the need to acquire foraging skills (Forslund & Larsson 1992; Desrochers & Magrath 1993; Wheelwright & Schultz 1994) or to allocate more resources to mate or territory acquisition, may therefore limit performance in young choughs.
As recruitment age varied little in choughs, mean performance is unlikely to have improved across young females because better breeders recruited later (Lessells & Krebs 1989; Wiktander et al. 2001). Further, 3-year-old females breeding for the first time did not lay larger clutches, breed more successfully or fledge more offspring than 3-year-olds that had recruited at age 2. However, females that had short breeding lives (4 years or less) tended to lay small clutches and fledge few offspring before death. Thus mean clutch size and fledging success, measured across the population, increased across young age-classes partly because some poor breeders died young. While some previous studies investigating early improvements in performance have demonstrated similar covariation between early reproduction and survival, most analyses have found no evidence of such effects (Table 1). Differential mortality has been suggested consequently to be an unlikely cause of age-specific improvements in mean performance, particularly in relatively long-lived species where mortality rates are low (Forslund & Pärt 1995; although see Cam & Monnat 2000). On Islay, chough life-history traits are notably correlated positively across cohorts, possibly reflecting persistent consequences of early environmental conditions for individual investment ability (Reid et al. 2003). However, positive covariation among life-history traits is a common general phenomenon (e.g. Smith 1981; Sandercock et al. 2000), for example, underlying the accepted need for experimentation in the study of physiological trade-offs (van Noordwijk & de Jong 1986; Stearns 1992). It is consequently surprising that so few studies dissecting early improvements in mean reproductive performance have detected correlations between early productivity and survival. Many analyses have simply compared performance between discrete classes of breeders that did and did not survive to subsequent years (analysis B, Table 1). This approach assumes that ‘surviving’ individuals are phenotypically homogeneous with respect to early reproduction. If, as in choughs, early reproduction varies in fact with an individual's long-term as well as short-term survival, these analyses may underestimate the role of differential mortality in causing early improvements in mean reproductive performance. Indeed, analyses which correlated an individual's reproductive performance with its final lifespan (a quantitative index of future survival), rather than survival to an arbitrary subsequent season (a binary index) have shown differential mortality more frequently (Table 1, analysis A: four of four studies, analysis B: three of 20 studies, = 7·9, P = 0·005). Furthermore, two of the three binary studies that detected differential mortality used more finely structured analyses (analysis C).
declining performance in old age
Across the population, mean clutch size, the probability of breeding successfully and fledging success declined significantly across old females. Similar declines have been reported in other species, suggesting reproductive senescence (e.g. Clutton-Brock 1988; Wooller et al. 1990; Komdeur 1996; Newton & Rothery 1997; Robertson & Rendell 2001; Saino et al. 2002). However, studies have not generally distinguished whether declines in mean performance result from changes in individual performance rather than increased mortality of individuals that invest heavily in reproduction. In choughs, individual females laid smaller clutches, were more likely to fail and fledged fewer offspring when old than they did during middle-age; these declines in individual performance were clear despite the small number of females reaching old age. However, females that ultimately lived longest laid relatively small clutches and fledged few offspring when they were young. Particularly long-lived females were therefore a non-random subset of the population with respect to reproductive performance, at least during their early years. However, we could not detect such clear relationships between lifespan and breeding performance during middle age. As our ability to detect these relationships was reduced by small sample sizes and data truncation at older ages, it is difficult to evaluate fully the extent to which mean reproductive performance declined across old age classes because individuals whose reproductive outputs were consistently at the lower end of the age-specific performance distributions survived for longer. More data describing the life-long reproduction of long-lived females are needed to investigate this in more detail.
early reproduction and survival
Correlations between early fecundity and lifespan have been observed repeatedly in Drosophila (Rose 1984; Zwaan, Bijlsma & Hoekstra 1995; Partridge, Prowse & Pignatelli 1999) and a causal relationship between the two has been demonstrated (Sgró & Partridge 1999). However, there is relatively little evidence that similar long-term trade-offs apply in natural populations, and in the absence of experiments negative relationships may often be masked by qualitative differences amongst individuals. Increased reproductive effort reduced future fecundity in Ficedula albicollis Temminck (Gustafsson & Pärt 1990) and future survival in Parus montanus Baldenstein (Orell & Belda 2002), Parus major L. (McCleery et al. 1996) and Homo sapiens L. (Westendorp & Kirkwood 1998). Bérubéet al. (1999) and Cam & Monnat (2000) found no evidence of these relationships in Ovis canadensis Shaw or Rissa tridactyla L. In female choughs that survived for multiple breeding seasons, final lifespan was correlated negatively with both clutch size and fledging success during the female's first few breeding attempts, suggesting a trade-off between early reproduction and future survival in this species. The occurrence of such a long-term physiological trade-off is central to the disposable soma theory of ageing, and a senescent decline in survival might consequently be predicted (Kirkwood & Rose 1991; Partridge & Barton 1996; Orell & Belda 2002). As with reproductive performance, patterns of mean age-specific survival measured across a population can reflect complex combinations of individual trajectories, potentially obscuring evidence of senescence in individuals (Vaupel & Yashin 1985; McDonald et al. 1996; Nisbet 2001). While age-specific changes in individual reproduction can be described by monitoring performance across multiple seasons, as individuals can die only once, measuring age-specific variation in an individual's survival probability is considerably less tractable (Partridge & Mangel 1999; Service 2000; Cam et al. 2002). The Islay chough data are not yet sufficient to permit the multistate survival analyses that may shed the most light on this problem (e.g. Vaillefont, Cooch & Cooke 1995; Yoccoz et al. 2002). However, while heterogeneity could potentially obscure evidence of senescence (e.g. McDonald et al. 1996), it is not clear that artefactual declines could be created: this would require individuals with the lowest survival probability to survive the longest. Our unstratified models therefore provide some evidence that survival does decline in old age in choughs.
We thank Earthwatch, Merial Animal Health, the Nature Conservancy Council, the Royal Society for the Protection of Birds (RSPB), the Scottish Executive Environment and Rural Affairs Department, Scottish Natural Heritage (SNH) and WWF–UK for financial support during the study, and the RSPB and SNH for funding data compilation and analysis. Islay farmers kindly allowed access to nest sites. Numerous people assisted with data collection, including Martin and Robin Bignal, John and Pamela Clarke, David Jardine, Clive McKay, Neil Metcalfe, Allen Moore, Malcolm Ogilvie, Elizabeth Still, Judy Stroud, Paul Thomson and RSPB, SNH and Islay Natural History Trust staff. Lukas Keller advised on analyses, and Peter Arcese, Neil Metcalfe and Arie van Noordwick commented helpfully on manuscript drafts. Nest visits were licensed by NCC and SNH. J.M.R. is supported by Killam and Green College Postdoctoral Fellowships.