Delayed cost of reproduction and senescence in the willow tit Parus montanus

Authors

  • Markku Orell,

    Corresponding author
    1. Department of Biology, University of Oulu, PO Box 3000, FIN-90014 University of Oulu, Finland
      Markku Orell, Department of Biology, University of Oulu, PO Box 3000, FIN-90014, University of Oulu, Finland (fax + 358 8553 1061; e-mail Markku.Orell@Oulu.Fi).
    Search for more papers by this author
  • Eduardo J. Belda

    1. Department of Biology, University of Oulu, PO Box 3000, FIN-90014 University of Oulu, Finland
    Search for more papers by this author

Markku Orell, Department of Biology, University of Oulu, PO Box 3000, FIN-90014, University of Oulu, Finland (fax + 358 8553 1061; e-mail Markku.Orell@Oulu.Fi).

Summary

  • 1We studied age-specific survival rates in the willow tit Parus montanus in northern Finland using 15 years of capture–recapture data obtained from birds during the breeding seasons 1986–2000. In addition, short- and long-term costs of reproduction were investigated by comparing survival probabilities of breeding and non-breeding individuals.
  • 2We did not find evidence supporting age-specific survival probabilities in males. However, in females there was a significant decline in survival after the age of 5 years.
  • 3Reproduction did not impair individuals’ chances of being alive in the subsequent year (short-term cost) because breeding males and females had similar survival rates as non-breeders.
  • 4Demographic costs of breeding appeared later in life. Females skipping breeding earlier in life showed a higher probability of survival after the age of 5 years than females that bred every year until that age. This effect was non-significant in males.
  • 5The observed decline in survival probability late in life is likely to result from an increased cost of reproduction due to higher allocation of resources to breeding earlier in life, i.e. increased effort early in life is traded with survival late in life. The results also suggest that income breeders, such as small passerines, may pay long-term costs of reproduction. This is in agreement with the disposable soma theory of ageing.

Introduction

Natality, mortality, immigration and emigration are the four key parameters that affect population density and abundance. Early studies on avian population dynamics assumed linear survivorship after the attainment of maturity, which would imply a constant annual rate of mortality independent of age (Lack 1954). This idealized constant mortality rate has been presented in many ecological textbooks (Begon, Harper & Townsend 1996; Pianka 2000), although we have to allow for species-specific and perhaps also population-specific variation in age-related survival rate (Newton 1989). In many studies on small and large birds it has been claimed that there is age-dependent variation in survival of adult birds (Hildén 1978; Perrins 1979; Scott 1988; Thomas & Coulson 1988; McCleery & Perrins 1989; Newton 1989; Sternberg 1989; Wooler et al. 1989; Møller & de Lope 1999). In some other papers increasing mortality with age has not been demonstrated (McCleery & Perrins 1988; Ollason & Dunnet 1988). However, none of these interpretations was based on proper analyses, i.e. modelling both survival and recapture probabilities (Nichols, Hines & Blums 1997).

When studying survival probabilities in the wild the most reliable data come from successive samples of individually marked animals (Lebreton et al. 1992). If the data include, as they usually do, incomplete registration of survivors, then statistical methods to estimate separately survival and capture rates are needed. The Cormac–Jolly Seber (CJS) model considers time-dependent capture and survival probabilities, and modifications of it allow us to study age-dependent variation in survival and recapture rates (Lebreton et al. 1992). In some papers age-related survival rates for long-lived non-passerines (Francis et al. 1992; Newton & Rothery 1997; Nichols et al. 1997) have been investigated by applying these capture–recapture models for open populations, but the specific tests for the effects of senescence produced contrasting results. By using cohort-based Jolly Seber modelling, Loery et al. (1987) suggested a decline in survival probability of a short-lived passerine, the black-capped chickadee Parus atricapillus L., after the age of 3 years, based on birds banded and recaptured during the non-breeding season. However, because the birds were not aged at capture, their inferences about age-specificity applied to time since the initial capture not to real bird age. McCleery et al. (1996) analysed capture–recapture data of the breeding great tit Parus major L., specifically addressing the question of senescent decline in survival. Both sexes showed increased mortality rates after the age of 5 years during the period when nest predation was reduced to negligible with special nest boxes. Sanz & Moreno (2000) did not find any evidence for declining survival rate in the latter half of life in the pied flycatcher Ficedula hypoleuca Pallas. To our knowledge these three are the only papers where age-related survival probabilities instead of return-rates have been investigated on small birds with a relatively short life span.

Thus, we may interpret that patterns of age-specific survival rates are not well known for birds and further studies on the occurrence of senescence, especially in small passerines, are needed. The problem for studies of senescence is the unreliability of survival estimates of ‘older’ individuals in natural populations (Carey et al. 1992). Among the cohorts marked as juveniles, mortality decreases the number of individuals of known age in old age classes. Thus, large numbers of birds of known age must be marked during long-term studies in order to achieve sufficient sample sizes for estimating age-related survival probabilities (Nichols et al. 1997).

There are several hypotheses about how senescence may arise. Medawar (1952) suggested that deleterious mutations accumulate during the life span of an individual and result in an increase in age-dependent mortality. A second hypothesis is the ‘disposable soma’ theory, which is an optimality account of ageing, in which allocation of resources to reproduction jeopardizes somatic repair mechanisms and hence longevity (Kirkwood 1977; Partridge & Barton 1993 and references cited therein). This hypothesis can be regarded as a special case of Williams’ pleiotropy model (Williams 1957).

The willow tit Parus montanus Conrad, a small insectivorous passerine, is viewed as a highly sedentary species after settling during the first autumn or early spring (Ekman 1979; Koivula & Orell 1988; Orell, Rytkönen & Koivula 1994b). Because of this, capture–recapture data achieved by monitoring colour-ringed birds proved to give reliable estimates of survival rates between breeding and during non-breeding seasons (Orell et al. 1994a, 1999; Lahti et al. 1998). In this study we used a 15-year (14 cohorts) capture–recapture data set from willow tits monitored during breeding seasons at Oulu, northern Finland, to investigate the existence of the age-related annual survival probability. Furthermore, the presence of non-breeding individuals of both sexes in the population (Orell et al. 1994a) made it possible to investigate senescence of individuals differing in their allocation of resources to reproduction.

Materials and methods

STUDY AREA AND DATA COLLECTION

The willow tit population breeding in the Oulu area (c. 65° N, 25°30′ E), northern Finland, has been studied since 1975 (Orell & Ojanen 1983; Rytkönen et al. 1995; Orell et al. 1999). The study area has increased gradually from c. 8 km2 in 1986 to 24 km2 in 1996 and later. Coniferous and deciduous forests of varying ages, including stands of young trees, bogs and clear cuttings, are the main habitat types, forming a highly fragmented and mosaic structure in the study area. The dominating tree species consist of Norway spruce Picea abies L., Scots pine Pinus sylvestris L. and birch Betula spp. L. For detailed information about the habitats see Orell & Ojanen (1983) and Orell & Koivula (1988). Similar habitat types continue outside the study plot.

The majority (c. 90–95%) of the breeding population was individually marked with unique combinations of numbered aluminium and coloured plastic rings during the preceding winter or breeding seasons (Orell & Koivula 1988). We have no evidence suggesting that losses of coloured rings might have biased our data. In sexing the birds, we followed the criteria described in Koivula & Orell (1988). In this study we analysed survival of the cohorts born in the years 1985–99 until the year 2000. Ageing (yearling vs. older individual) of immigrants from outside the area, and without rings, was done according to Laaksonen & Lehikoinen (1976).

In our study area, willow tits nearly exclusively breed in natural holes that the parents excavate in decaying stumps (Orell & Koivula 1988). By intensive monitoring of the population before and during the breeding season (early April until late June) we were able to locate the territories and the nests and also identify the non-breeding individuals (Orell et al. 1994a). Taped territorial song of the willow tit, played back in breeding habitats, was used to locate the territory owners and unpaired individuals of both sexes. To confirm the status of an individual or a pair suspected not to be breeding, we used the following criteria (Orell et al. 1994a).

First, based on our knowledge of location of stumps suitable for nest holes in the territories, we checked whether or not the birds were excavating a hole(s). If a hole with a completed nest was not found, or the excavation had stopped, and the pair was seen in the territory throughout the breeding season performing non-breeding behaviour (e.g. continuous hoarding), this was recorded as a non-breeding pair. This judgement was confirmed when we did not find such a pair with young (without rings) during the time other parents attended fledged offspring. Secondly, if an individual lost its mate at the beginning of the season (before egg laying started) and was seen alone thereafter, it was recorded as non-breeding. Thirdly, individuals, usually males, remaining alone in the area without a mate throughout the breeding season were considered non-breeders. Fourthly, intensive singing of a male, including an eager response to play-back at the time of season when territorial singing of breeding males had decreased, confirmed that he was a non-breeder.

So, by definition, non-reproducing birds were those that did not own a nest in a given year. The breeding status of some of the pairs close to the border of the study area could not always be confirmed and possibly they bred outside the area. Therefore, such pairs were not considered as non-breeders although we did not find their nest that year. We think our approach is conservative because we may have scored some non-breeders as breeders.

Local survival of willow tits from year n refers here to birds detected in the study area during the breeding season in year n + 1 (the sampling period being from 1 May until early July; Orell et al. 1994a, 1999). Thus, only the birds present in the population during the breeding time were included in the data set. The analyses were based on the capture histories of 546 males and 560 females aged 1 year old at first capture.

The study area has increased during the course of the investigation. This, together with possible permanent emigration, might have biased our results, because the survival estimates of capture–recapture data include both death and permanent emigration. We do not consider this a serious bias in our data because of the high site-fidelity of the willow tit after establishment (average breeding dispersal distance for males and females ranged from 209 to 253 m; Orell et al. 1999). If permanent emigration were a serious problem in our data, we should have found individuals with incomplete capture histories (i.e. an individual not resighted in a given year but recorded afterwards) appearing, especially after the increase of the study area. This was not the case. We modelled capture and survival rates considering the size of the study area but this factor was not significant. Therefore, we excluded this variable from the final analyses presented in this paper.

SURVIVAL ANALYSES

Survival analyses were carried out with the CJS model according to the statistical framework reviewed in Lebreton et al. (1992). We used the program mark (White & Burnham 1999). The validity of the CJS model to the data was assessed by the goodness-of-fit test of the program release (Burnham et al. 1987). The goodness-of-fit tests considering age effects were conducted using the parametric bootstrap approach implemented in mark. The parameter estimates (survival and recapture rates) of the model were used to simulate data. These simulated data exactly met the assumptions of the model, i.e. no over-dispersion was included, probability of being recaptured was the same for all individuals, and no violations of model assumptions were included. Then we checked whether the observed deviance for the general model fell within the distribution of all the deviances from the simulated data. The number of simulations with deviance larger than the one we observed for our general model, divided by the total number of simulations, finally gave the probability of obtaining by chance a deviance value as large or larger than the one we observed. We used the significance level P < 0·05 for rejecting the null hypothesis. Model notations follow those suggested in Lebreton et al. (1992).

MODEL SELECTION

Estimation of survival rates depends heavily on selection of an appropriate statistical model. We used the procedure described in Lebreton et al. (1992). Model selection starts from a general model, which fits the data, and uses likelihood ratio tests (LRT) to test specific hypothesis and the Akaike information criterion (AIC) to assist in selecting the most parsimonious model (Lebreton et al. 1992). When LRT are used in model selection, there is a substantial risk of both type I and II errors. We used an α level of 0·15 for rejecting a model and an α level of 0·05 to assess significance strictly (Lebreton et al. 1992; Bauchau & van Noordwijk 1995). When the P-value was between 0·05 and 0·15 the effect was kept in the model selection process and tested in later steps. The decision whether to include them in the final model selected was based on AIC.

CAPTURE RATES

Out of the 1106 individual histories, there were only 27 incomplete capture histories of individuals still alive (15 males and 12 females). This could be explained by the intensive capture effort (see above). Therefore, we decided only to test time effects on capture probabilities, not considering age effects on capture rates. There was no significant effect of time on capture rates for the females (LRT = 15·99, d.f. = 12, P = 0·2), whereas we detected a significant effect of year for the males (LRT = 36·97, d.f. = 8, P < 0·01). Most of the incomplete histories came from the years 1998 and 1999 (13 out of 27), when the number of breeding pairs was more abundant than in any other study year. This may have contributed to the recapture success. The estimate of capture rate for the females was 0·95 (SE = ±0·01) and it ranged from 0·84 to 1·0 in the males.

AGE EFFECTS AND SENESCENCE

We modelled survival and capture rates separately for males and females. We started from a model considering age and time dependence, and proceeded by simplifying it to a model considering three age classes (McCleery et al. 1996). Once a model of three age groups was achieved we tested for time effects on survival in each age class using LRT. For testing the effect of senescence we compared, using the LRT test, the model incorporating the three age classes with the model where the second and third age classes were lumped together (Lebreton et al. 1992; McCleery et al. 1996). Similarly, grouping the yearlings and the second age class and comparing this model with the model considering survival differences between these two age classes allowed us to test for the differences in survival of yearlings and the second age class.

COSTS OF REPRODUCTION: THE EFFECT OF BREEDING STATUS

Recent developments in the analyses of capture–recapture data allowed us to study the possible existence of a cost of reproduction in survival by using multi-strata models (Brownie et al. 1993; Nichols et al. 1993; Clobert 1995). Transition probabilities (Ψ; the probability of making the transition from non-breeder to breeder or vice versa) can be estimated together with survival probabilities (S) in each stratum (i.e. survival of breeders and non-breeders) and capture probabilities (P) using a first-order Markov process (Nichols et al. 1994). Such a process makes the assumption that survival probability from stage i to stage i + 1 depends only on state at time i. This model can be used to study the existence of a ‘direct cost’ of breeding on survival, i.e. if non-breeders have higher survival rates than breeders. In this model survival is estimated separately from the transitional probability Ψ, i.e. the survival estimate controls for heterogeneity due to breeding status. We used the symbol ∅ when referring to CJS and the symbol S when estimates were obtained using multi-strata models, following Sandercock et al. (2000). Analyses were conducted using mark and model selection according to AIC values. For each capture occasion, the breeding status of each willow tit was determined as breeder or non-breeder. The approach is similar to the one described in Nichols et al. (1994).

Some costs or benefits will not show up directly in the following year, but will only be manifested over an individual’s life. Therefore, the assumption that probability from stage i to stage i + 1 depends only on state at time i does not necessarily hold (see above). A simple approach to study the possible existence of a delayed cost of reproduction is to group the birds along trait lines (breeders and non-breeders) and look for differences among the traits in these two groups (McCleery et al. 1996).

Results

GOODNESS-OF-FIT TEST

Capture probabilities were fairly high and therefore the 3.Sm component of TEST 3 and the TEST 2 are not informative (Burnham et al. 1987; Lebreton et al. 1992). Thus, in this case the information relating to the goodness-of-fit test was contained in the 3.SR component. The results of the tests for the CJS model considered the sexes separately; model ∅(t), P(t) fit the data (∅, survival probability; P, capture rate; t, time effect; TEST 3.SR, males χ2 = 14·25, d.f. = 13, NS, females χ2 = 11·16, d.f. = 13, NS). A goodness-of-fit test for the model ∅(a*t), P(a*t) (a= age) can be performed by testing the fit of the CJS model for each age- and sex-determined cohort class separately (Lebreton et al. 1992). There were not sufficient data for testing time-dependence models for birds aged 6 years or older, so we pooled these age classes with the age of 5 years. For the 5-year or older birds data were still too sparse in 6 out of 10 possible years. About 95% of the males and females seemed to die before the age of 6 years, albeit more males than females reached the oldest ages (Table 1). We started with a model considering only five age classes (1-, 2-, 3-, 4- and 5-year or older birds) with time-dependent survival probabilities (Table 2, model 1), although we were aware that there may not have been enough data for properly analysing the time effects on the oldest age class. We let capture rate vary with time in the males whereas we kept it constant in the females (see the Materials and methods). The goodness-of-fit parametric bootstrap approach suggested that our starting model fitted the data (males P = 0·17; females P = 0·23; 1000 simulations).

Table 1.  Number of different male and female willow tits captured at different ages and age-specific survival probabilities based on the CJS model ∅(a*t)P(constant)
AgeNo. of malesNo. of females% males surviving% females survivingMale survival (CJS)Female survival (CJS)
15465601001000·610·58
2322320 58·97 57·140·650·63
3184180 33·69 32·140·660·59
4104 94 19·05 16·780·630·59
5 63 50 11·53  8·930·640·51
6 33 20  6·04  3·570·630·43
7 19  9  3·48  1·610·790·21
8 15  2  2·74  0·360·500·49
9  7  1  1·28  0·180·331
10  2  1  0·36  0·180·510
11  1  0  0·18  0  
Table 2.  Model selection for age effects on survival rates of male (a) and female (b) willow tits breeding in Oulu, Finland. The values for the deviance (DEV), maximum number of parameters (np) that can be estimated in each model, Akaike’s information criterion (AIC) and the likelihood ratio test (LRT), with degrees of freedom (d.f.) and P-values, are shown. In the column ‘Models tested’ the digits refer to the models compared with the LRT. Model notation is as follows: t, time dependence; *, interaction between factors. ai, means survival from age i to age i + 1; aij, means survival probability of age i = age j; ai+, means survival probability of age i = age j… over any older age class. Capture rates for males were time dependent and for females they were constant. The selected models are in bold letters
ModelDEVnpAICModels testedLRTd.f.PEffect tested
(a) Males
1(a1,a2,a3,a4,a5+)*t1675·5731821·5     
2(a1,a2,a3+)*t1706·9521810·9     
3(a1,a2–3,a4+)*t1702·8511804·8     
4(a1,a2–4,a5+)*t1700·8501800·8     
5a1*t, a2–4*t, a5+1712·9421799·9     
6a1*t, a2–4, a5+1724·9301784·9     
7a1, a2–4, a5+1746·8171788·8 7–621·93130·06Time on a1
8a1*t, a2+1725·1291783·1 8–6 0·15 10·70Senescence
9a1, a2+1747·0161779·0 9–7 0·18 10·67Senescence
10(constant)1748·7151778·710–9 2·23 10·14Age
11(a1, a2+)*t1718·7401789·7     
12t1735·6271789·612–1116·89130·20Age
(b) Females
1(a1,a2,a3,a4,a5+)*t1662·7601782·7     
2(a1,a2,a3+)*t1676·2401756·2     
3(a1,a2–3,a4+)*t1674·0391752·0     
4(a1,a2–4,a5+)*t1675·5381751·5     
5a1*t, a2–4*t, a5+1681·0291739·0     
6a1*t, a2–4, a5+1691·7171725·1     
7a1, a2–4, a5+1713·2 41721·2 7–621·4130·07Time on a1
8a1*t, a2+1697·9161729·9 8–6 6·21 10·013Senescence
9a1, a2+1719·4 31725·4 9–7 6·23 10·013Senescence
10a1–4*t, a5+1694·5161726·510–513·52130·41a1a2–4
11a1–4, a5+1714·0 31720·011–7 0·81 10·37a1a2–4

AGE EFFECTS AND SENESCENCE

The oldest age observed for males was 11 years, 10 for females. We simplified our starting model (Table 2, model 1) to a model considering three age classes: age1 (yearlings), age2 (2 years) and age3+ (3 years or older) and allowed time-dependent variation for all these groups (Table 2, model 2). Then we simplified this age structure by adding years to the second age category [e.g. age1 (yearlings), age2–3 (2 or 3 years old) and age4+ (4 years or older); Table 2, model 3]. In both sexes a model with three age classes, age1, age2–4 and age5+, had lower AIC values than other more complex models (Table 2, model 4). Then we tested for time effects in survival probabilities. We did not find time effects on age2–4 and age5+ (Table 2, models 5 and 6). This made it easier to proceed with model selection targeting our principal aim, the possible decline in survival at older ages. However, a model without considering time effect on yearling survival did not seem to fit the data as well as a model with time effect (Table 2, model 7 vs. model 6). The LRT for time effects on yearling survival had P-values of 0·057 and 0·065 for the males and females, respectively (Table 2). Therefore, we continued model selection considering models with and without time dependence in age1.

We did not find any evidence for senescence in the males. The survival estimate of 0·65 (SE = ±0·02) for age2–4, and 0·63 (± 0·04) for age5+, suggested no decline in male survival probability in older ages (Fig. 1). This was confirmed by comparing models 8 and 6 (Table 2a; LRT = 0·15, d.f. = 1, NS). Thus, model 8 implied constant survival probability after the age of 2 years. In females, survival from 1 to 5 years averaged 0·61, while survival after the age of 5 was 0·46 (± 0·06). The logistic regression coefficient for age5+ was negative (b =−0·08 ± 0·01), suggesting declining survival probability after the age of 5 (Fig. 1). This difference in survival between the two age groups was significant (Table 2b, model 8 vs. model 6; LRT = 6·21, d.f. = 1, P < 0·01). Similar results were obtained for both sexes when we considered models with constant survival for yearlings (Table 2, model 9 vs. model 7).

Figure 1.

Survival probabilities (± SE) for male and female willow tits at different ages: a1= yearlings; a2–4; a5+. Estimates from model ∅(a1,a2–4,a5+)P(t) for the males and ∅(a1,a2–4,a5+)P(constant) for the females.

The suggested time-dependent variation in yearling survival probability and the constant rate in older ages (Table 2, model 6) made it difficult to test the differences of survival between these two age groups. One possible approach was to study whether a model with no time effects on juvenile survival fit the data. For the males, comparing model 9 with model 10, where the first and second age classes were lumped together (Table 2a), would test for the difference in survival probability between these two age groups. The difference was non-significant (Table 2a). In another approach, we considered models with time effects on survival in both age groups, model 11, and compared it with model 12 with only a time effect on survival (Table 2a). The difference between these models was not significant, suggesting no age effect in the males. For females, the comparison between model 7, no time effect on age, with model 11, where the first two age classes were pooled together (Table 2b), was not significant. Considering time effects on survival of these two age groups, model 5, and comparing it with model 10, lumping the first two age groups together, also suggested no difference in survival between yearling and age2–4 females (Table 2b).

Sauer & Williams (1989) described a general chi-square statistic for the comparison of several survival estimates. This statistic can be implemented using the program contrast (Hines & Sauer 1989). We used that method to address the hypothesis of no differences in survival probability between the mean survival of first year birds and the survival probability of older ages. We also tested the hypothesis that yearling survival is not time dependent. For the males we compared the time-specific estimates of yearling survival with the survival of older birds, but the differences were not significant (χ2 = 1·86, d.f. = 1, NS). For the females we compared the constant survival estimates of age2–4 with the time-dependent survival estimates of age1. The differences were not significant (χ2 = 0·63, d.f. = 1, NS). We did not find support for the hypothesis of time-dependent variation in survival at age12 = 0·386, d.f. = 1, NS; χ2 = 2·99, d.f. = 1, NS, for males and females, respectively). These results suggest that there was no age or time effect in male survival. For the females there was no difference in survival between yearlings and birds aged 2–4 years. This was in agreement with our previous results.

The final model selected for the males included constant survival and time-dependence in recapture rate, ∅(c), P(t) (Table 2a, model 10). The final survival estimate for males was 0·63 ± 0·015. For the females, the selected model included two age classes, with constant survival from 1 to 4 years and declining survival rates for older birds [∅(a1–4,a5+), P(c); Table 2b, model 11]. Year to year survival estimate for females until the age of 5 years was 0·60 ± 0·015, and after the age of 5 years 0·46 ± 0·06.

There were two potential biases in our analyses. First, the method of data collection automatically led to older ages not being present in all the study years. If there were systematic temporal changes in the survival probability during the course of the study, this could affect the age-related survival pattern detected. To detect possible trends in survival we fitted a model in which survival was constrained to be a linear function of time. To analyse the significance of this covariate we used F-tests based on analysis of deviance (anodev; Skalski, Hofmann & Smith 1993). We did not find significant trends in survival estimates in either sex (anodev; F1,12 = 0·07, P = 0·80; F1,13 = 1·17, P = 0·30, for males and females, respectively). Secondly, a possible bias is related to non-random sampling of ages through cohorts (old age classes come only from early years’ cohorts). To analyse this possibility we limited our modelling to the cohorts from 1986 to 1995, given that all these birds could have attained the age of 5 years by 2000. The results obtained from these analyses did not differ from the results reported above (data not shown).

COST OF REPRODUCTION

There were 149 cases of non-breeding males and 85 females from 1986 to 2000. Most of the non-breeding birds were yearlings (67% in males and 65% in females). We addressed the question of whether there is a difference in survival among breeders and non-breeders. This question focuses on the existence of a ‘direct cost of reproduction’. We started from a model considering three age classes (age1, age2–4 and age5+) for survival (S) and transition probabilities (Ψ), and with breeding status effect on all parameters [bs= breeding status, i.e. breeders and non-breeders; model S{(a1, a2–4, a5+)*bs}, P(bs), Ψ{(a1, a2–4, a5+)*bs}]. We ran separate analyses for males and females.

We only found a significant effect of breeding status on transition probabilities (Table 3, model 4 vs. model 3). In males the probability of transition from non-breeder to breeder was 0·86 ± 0·04 (conditional on being alive), while the probability of finding a breeder as a non-breeder in the next year was 0·06 ± 0·01. In females the values were 0·91 ± 0·04 and 0·035 ± 0·01, respectively. No evidence for differences with age effects in transition probabilities was found (Table 3, model 3 vs. model 2). We did not find a significant effect of breeding status on survival probabilities (Table 3b, model 5 vs. model 4; model 6 vs. model 5; model 7 vs. model 6). Considering breeding status did not change the pattern of age-specific survival rates, and the final model selected included constant survival in males (Table 3a, model 9), while in females senescence in survival after the age of 5 years was found (Table 3b, model 8). In summary, non-breeders did not tend to survive better to the next year than breeders did.

Table 3.  Akaike’s Information Criterion (AIC) values and likelihood ratio tests (LRT) to the hypotheses that (1) survival probability from stage i to stage i + 1 did not differ by breeding status (breeding or non-breeding, s) and that (2) the transition probabilities (from breeder to non-breeder or vice-versa) differ. Model notation is as follows: S survival probability; P capture rate;? transition probability; ai means survival from age i to age i + 1; ai–j means survival probability of age i = age j; ai+ means survival probability of age i = age j… over any older age class; bs means breeding status; *interaction between factors; c constant effect. Model selected in bold letters
 ModelAICModels testedLRTd.f.PEffect tested
(a) Males
1S{(a1,a2–4,a5+)*bs}, P(bs), Ψ{(a1,a2–4,a5+)*bs}2184·6     
2S{(a1,a2–4,a5+)*bs}, P(c), Ψ{(a1,a2–4,a5+)*bs}2182·82–1  0·211  0·65Status in P
3S{(a1,a2–4,a5+)*bs}, P(c), Ψ(bs)2180·43–2  5·774  0·22Age in Ψ
4S{(a1,a2–4,a5+)*bs}, P(c), Ψ(c)2456·64–3278·31< 0·01Status in Ψ
5S{(a1,a2–4)*bs(a5+)}, P(c)(bs)2178·65–3  0·201  0·66Status in a5+
6S{(a1*bs(a2–4)(a5+)}, P(c), Ψ(bs)2176·76–5  0·121  0·73Status in a2–4
7S(a1,a2–4,a5+), P(c), Ψ(bs)2174·87–6  0·191  0·66Status in a1
8S(a1,a2+) P(c), Ψ(bs)2173·18–7  0·251  0·62a2–4, a5+
9S(c), P(c)(bs)2172·69–8  1·581  0·21Age
(b) Females
1S{(a1,a2–4,a5+)*bs}, P(bs), Ψ{(a1,a2–4,a5+)*bs}1950·1     
2S{(a1,a2–4,a5+)*bs}, P(c), Ψ{(a1,a2–4,a5+)*bs}1948·22–1  0·191  0·66Status in P
3S{(a1,a2–4,a5+)*bs}, P(c), Ψ(bs)1945·83–2  3·652  0·16Age in Ψ
4S{(a1,a2–4,a5+)*bs},P(c), Ψ(c)2137·44–3195·71< 0·01Status in Ψ
5S{(a1,a2–4)*bs(a5+)}, P(c), Ψ(bs)1946·25–3  2·421  0·12Status in a5+
6S{(a1*bs(a2–4)(a5+)}, P(c), Ψ(bs)1944·26–5  0·061  0·81Status in a2–4
7S(a1,a2–4,a5+), P(c), Ψ(bs)1942·27–6  0·071  0·84Status in a1
8S(a1–4, a5+), P(c), Ψ(bs)1941·08–7  0·811  0·32a1, a2–4
9S(c), P(c), Ψ(bs)1944·59–8  5·481  0·02Senescence

The observed senescence in females may be due to a delayed cost of reproduction. Therefore, we compared survival from age of 5 years between the females that were non-breeders at least once before that age with the females that bred (successfully or non-successfully) every year. It was after the age of 5 when the effect of senescence appeared for the females (Table 2b, model 11). The LRT showed a significant effect of breeding status on survival probability after the age of 5 (model ∅(a5+*bs), P(constant) vs. ∅(a5+), P(constant); LRT = 4·07, d.f. = 1, P = 0·04). The female breeders showed a lower survival probability than the non-breeders (Fig. 2). For the males the effect of breeding status was non-significant (LRT = 0·03, d.f = 1, P = 0·86; Fig. 2).

Figure 2.

Survival probabilities (± SE) for willow tits after the age of 5 years or older (a5+) in relation to their breeding status. Non-breeders are referred to as individuals that skipped breeding at least once before the age of 5 years. Breeders are those that bred (successfully or non-successfully) every year until that age.

Discussion

Senescence is the drop in survival probability and/or fertility later in the life of individuals. The present work presents evidence for a decline in survival probability in female willow tits that are 5 years or older. This result contrasts with the widespread view that, in birds after a certain age, survival rate remains constant and independent of age until the very end of the potential life span (Lack 1954).

Evidence for senescence effects in birds is scarce, especially for short-lived passerines. We are only aware of one other study demonstrating senescence effects on survival using capture–recapture techniques in short-lived passerines (McCleery et al. 1996). They also found a decline in survival probabilities for females of 5 years or older. We did not find evidence for senescence in the males, although we cannot exclude that ageing may appear in older ages, but small sample sizes for older ages hampered us from detecting it. Ageing effects on survival have also been claimed for the black-capped chickadee (Loery et al. 1987) and the barn swallow (Hirundo rustica L.; Møller & de Lope 1999). Different approaches of these studies make comparisons difficult.

There are even fewer works on birds dealing with how ageing might arise. We are only aware of two where the authors not only show senescence effects but also suggest why they may appear (McCleery et al. 1996; Møller & de Lope 1999). If the senescence decline is due to age-specific accumulation of deleterious mutations, then a similar survival probability at the age when senescence effects appear will be expected between breeder and non-breeder females. Alternatively, if trade-offs are important in the senescence effect then a drop in survival will be expected in those individuals investing more in earlier reproduction (Partridge & Barton 1993), i.e. breeding females have lower survival probabilities than non-breeding females. We found evidence for a delayed cost of reproduction in females. Thus our results support the idea that the senescence effects observed in females were, at least partly, due to trade-offs between present reproduction and future survival. This is in line with the disposable soma hypothesis (Partridge & Barton 1993). The lack of a significant effect on survival of breeding status in males may be due to the fact that males expend energy setting up and defending a territory and thus they pay similar costs whether they breed or not (McCleery et al. 1996).

Various authors have suggested that non-breeding could be a part of an adaptive strategy to preserve residual reproductive value (Coulson 1984; Wooler et al. 1989; Orell et al. 1994a). Life-history theory predicts that it pays to invest in offspring when these face high, and parents low, survival prospects, and to invest in survival when the expectations are reversed (Schaffer 1974; Stearns 1992). Orell et al. (1994a) investigated direct consequences of skipping breeding in the willow tit. Lack of a mate was an important cause for non-breeding in males, but not in females. They showed that the proportion of non-breeding females in the population showed a negative correlation with clutch size and subsequent juvenile survival, and suggested that non-breeding in females was a phenotypic response to low offspring value. In this work we have used appropriate statistical tools (Lebreton et al. 1992; Nichols et al. 1994) and a longer data set to study the consequences of non-breeding in the willow tit. We did not find evidence of a direct survival cost of reproduction (but see Ekman & Askenmo 1986). Breeder and non-breeder willow tits had similar probabilities of surviving to the next year, independent of age and sex. None the less, this result differs from recent works that reported non-breeding birds to have lower survival than breeders (Cam et al. 1998; Sandercock et al. 2000). It is unlikely that our result would be due to problems in the identification of the breeding status of the birds (see the Materials and methods). Another possible source of bias is that non-breeders were more likely to emigrate from the study area permanently. Although we cannot completely rule it out, the willow tit is considered to spend its entire life in the home area after settling during the first autumn or early spring (Ekman 1979; Koivula & Orell 1988; Orell et al. 1994a,b). It could also be that non-breeding individuals were of lower quality, as suggested by Orell et al. (1994a). This may result in lower survival probabilities in non-breeders (Cam et al. 1998; Sandercock et al. 2000) or that, by not breeding, individuals in poor condition increase their survival perspectives (Orell et al. 1994a). This last explanation fits our results.

The existence of a trade-off between current reproduction and survival incorporates the idea of a cost of reproduction. Stearns (1989, 1992) suggested that income breeders were likely to show only ‘short-term’ costs of reproduction. That will be especially true in organisms with high metabolic rates. Some studies conducted on small passerines failed to find evidence for a trade-off between present reproduction and adult survival, whereas others did so (reviewed in Stearns 1992; Tinbergen & Verhulst 2000). However, Jönsson (1997) suggested that an income breeder might expose itself to a higher risk of parasites or diseases from increased metabolic activity in connection with compensatory foraging. The demographic costs of breeding may then well be expressed only after current breeding. Our data support the idea that income breeders such as small passerines may pay long-term costs of reproduction. The negative effects of current reproductive effort can even persist for many years. McCleery et al. (1996) found that those females that were not observed breeding at least once before the age of 5 years had higher survival rates after that age. They assumed that those females undetected did not reproduce successfully. In the present work, we could identify the breeding status of birds. Thus this is the first time, to our knowledge, that a delayed cost of reproduction due to the effect of breeding earlier in life has been clearly documented in birds. None the less, the result must be taken cautiously because of the small sample size. Orell et al. (1996) did not find a survival cost when the amount of breeding effort was altered in a brood size manipulation experiment in the same willow tit population. Because of these contradictory results, Orell et al. (1994a) suggested that, as far as survival costs is concerned, in the willow tit the effect of whether to breed or not is much more important than the decision on how many offspring to produce. This is in agreement with a recent model about the evolution of ageing that suggests that organisms need to pay an energetic ‘overhead’ before any offspring can be produced (Shanley & Kirkwood 2000). The present work is able to show that, by non-breeding, willow tits improve future survival perspectives, giving support to the above prediction and confirming suggestions of Orell et al. (1994a) about why females skip breeding for one year.

It has been argued that senescence effects on survival would be of little importance in natural populations, as it would be expected to influence a relatively small proportion of the individuals in the population. This is supported by our data as less than 4% of the females arrived to the age when senescence effects in survival probabilities were detected. None the less, senescence could have a large effect on net reproductive rate. It can be especially important for individual reproductive success. Several studies have shown that life span is a major factor affecting lifetime reproductive success in birds (Newton 1989). On the other hand, the observed decline in survival with ageing has an effect on reproductive value (Fisher 1930). Thus, a question relevant to the evolution of life histories open to future research is whether or not breeder and non-breeder individuals differ in their reproductive value.

Acknowledgements

Ari-Pekka Auvinen, Lluis Brotons, Kirsi Ilomäki, Petri Kärkkäinen, Kari Koivula, Markus Keskitalo, Kimmo Kumpulainen, Kimmo Lahti, Minna Ronkainen, Seppo Rytkönen, Claudia Siffczyk, Mika Soppela and Petteri Welling helped in the field. Comments of two anonymous referees improved the text. The study was supported by the Research Council for Biosciences and Environment of the Academy of Finland, the University of Oulu and Thule Institute of the University of Oulu.

Ancillary