voluntary vs. involuntary changes in breeding site or status
A total of 102 changes were classified as voluntary, and a total of 362 as involuntary (Table 1). Birds of different categories differed significantly in the quality of their original breeding sites (Fig. 1; anovaF3,421 = 30·86, P < 0·001), and also in how much their site quality changed if they bred again (Fig. 2; anova: F3,344 = 10·72, P < 0·001). Birds that changed site voluntarily (VC) were breeders on low-quality sites before the change and obtained a site of much higher quality after the change (Figs 1 and 2). Birds that left voluntarily to float (VF) also obtained a better site when they bred again after a gap of 1 or more years. The original site of VF birds was of higher quality than that of VC birds (Fig. 1).
Figure 2. Means ± SE (box) and the 95% confidence interval for the mean (whiskers) for the change in site quality that a bird experienced when it changed its breeding site, measured as Q(new) –Q(previous). For floating birds (VF and IF), the new site is the site this bird bred next if it ever obtained a new site (the data set did not include birds who floated without ever breeding again). See text for significance tests.
Download figure to PowerPoint
Involuntary changes were recorded in birds that had been occupying very high-quality sites (IC, and especially IF in Fig. 1). When these birds bred again, they occupied lower-quality sites if they had been floaters during the intervening periods (IF, Fig. 2). If they changed site immediately (IC, Fig. 2), their new site was better than their old site (Fig. 2). This suggests that the IC category included birds that were evicted, but also birds that changed site voluntarily from a relatively good site to a very good site, and the relatively good vacancy left behind was immediately filled by another bird.
Although Figs 1 and 2 suggest adaptive site choice in guillemots, they could be artefacts because estimates of site quality are difficult to distinguish from the increase in individual breeding success over time (Harris et al. 1997). Therefore, we performed a randomization test on a null model (Appendix I), which assumed that all sites were equal in quality, all variation in breeding success arose from previous breeding experience of the bird and from demographic stochasticity, and that patterns of site use (probability of moving after n years of breeding; life span of birds; number of birds) equalled that of the field data. We assumed that birds keep their high experience score even if they move, making the test conservative as this tends to improve the breeding success scores after site changes. As the test statistic we used pinc, the proportion of increases (01) of all changes in breeding success following a site change. This proportion had the mean = 0·541 in the 10 000 simulation runs (range 0·448–0·634). The observed value in the field data was 0·766 (n01 = 317, n10 = 97), which is outside the range of the randomized null model, indicating P < 0·001. Birds thus changed sites to improve their breeding success.
For voluntarily leaving birds, we investigated whether they left low-quality sites regardless of their own breeding success, or whether the individual's own experience of breeding failure made it more likely to leave. In the null model, the proportion of site changes that were preceded by breeding failure (pdep; see Appendix I) had the mean = 0·241 and range 0·190–0·300 across all randomizations. In the field data, voluntarily leaving birds (VC and VF) had failed to rear a chick in 77·5% of cases (79 of 102). Thus, voluntary departures were associated strongly with breeding failure (10 000 simulations with range outside the observed value indicates P < 0·001).
Failure, however, often occurs disproportionately on poor quality sites. Did birds thus pay attention to site characteristics or to their own experience when leaving a site? Another randomization test took variable site quality into account. In the year prior to moving, the 64 birds of category VC reared a total of 15 fledglings. We calculated the expected distribution of the number of offspring produced prior to the VC birds moving, given the qualities of sites they owned. In a single run of such a randomization test, each site produced a fledgling with a probability equal to its estimated quality Q. The test statistic was the sum of all fledglings produced, calculated 10 000 times. The null hypothesis was that birds did not fail more often than expected based on the qualities of their sites. Using this null hypothesis, the birds produced 15 or fewer fledglings in 41 of 10 000 randomized cases. VC birds had thus been experiencing more breeding failures than predicted from their site occupancy (n = 64, two-tailed P = 0·0082). The same was true for VF birds (n = 38, P < 0·001). For victims of usurpation, the patterns were different depending on their category (IC and IF). IC birds had experienced more breeding failures than predicted by the null model (n = 202, P < 0·001), but the breeding success of IF birds did not differ significantly from expectation (n = 160, P = 0·29). Overall, the pattern suggests that individuals partly used their own breeding experience to determine whether they should attempt to move and potentially displace owners of higher quality sites. The group IF, that deviates from this pattern, may have consisted of true victims of usurpation (potentially particularly successful birds) and of birds that left voluntarily in search of a better breeding site after being unsuccessful. The opposite predictions for breeding success for these birds would thus cancel out in the data.
The response of birds to a breeding failure was not a simple rule of thumb but conditional on the quality of its site: among failed breeders, subsequent voluntary departure (VC, VF) was more likely if the failed bird bred on a poor quality site (logistic regression between breeding site quality Q and belonging to the VC or VF category, as opposed to staying as a breeder: β = −3·73, χ2 = 76·26, d.f. = 1, n = 732, P < 0·001). Birds on high-quality sites were thus more site-tenacious after a breeding failure than birds using low-quality sites.
Did usurpers simply target birds that owned sites with good physical characteristics, or did they use others’ breeding success as a cue for high-quality sites? If the latter, involuntarily leaving birds should have high breeding success for their sites, but none of the above results indicate this. For some birds of the category IC, this may simply indicate voluntary movement after a breeding failure. More conclusive evidence is provided by involuntarily floating (IF) birds, i.e. birds that are most likely to have been forced to leave. Because IF birds had had average breeding success for their sites, there was no suggestion that birds looking for new sites preferentially targeted neighbouring birds that had been successful in the previous year. In other words, usurpers appeared to vie for the generally high-quality sites occupied by IC and IF birds, whether or not the particular breeding attempt of these birds was successful.
changes in site occupancy and site change frequencies over time
Site quality was significantly positively related to its occupation status in all 19 years (Table 2). We therefore used the quality that predicted 50% site occupancy, Q50, as an index that reflected the quality of available sites at the current population size (Fig. 3). It is evident that poorer sites remained available as the study progressed (Fig. 3). As poorer sites were used for breeding in years with high population size, average breeding success decreased. This shows as a positive correlation between quality of available sites and average breeding success in all sites (marked and unmarked birds; Fig. 4). However, breeding success in the 316 sites that were occupied every year between 1982 and 2000 was unaffected by the quality of available sites (Fig. 4). The result supports the interpretation of pre-emptive use of best sites, leading to site-dependent density regulation (Rodenhouse et al. 1997).
Table 2. Logistic regression of site quality as an explanatory variable for site occupancy
Figure 3. Upper panels: two examples (years 1983 and 1999) of the logistic regression (Table 2) predicting site occupancy based on site quality. The arrow indicates the quality that relates to 50% occupancy probability. This forms the measure Q50 of site availability for each year. Lower panel: site availability measure Q50 for each year, with (a) and (b) denoting Q50 (the location of the 50% arrow) in years 1983 and 1999. The decreasing trend is significant (regression Q50 = α +β year: β = −0·0204, F = 63·7, P < 0·001).
Download figure to PowerPoint
Figure 4. The proportion of occupied sites that fledged an offspring was positively related to the measure of the quality of available sites (open dots). For the 316 popular sites that were occupied in every year between 1982 and 2000, fledging success was not significantly related to the quality of available sites (black dots).
Download figure to PowerPoint
Population regulation was evident in that the poorest available sites at the end of the study did not offer high enough breeding success to sustain a population. The mean survival from fledging to recruitment in 14 years of study was 33%. Available sites offered approximately 20% fledging success at the end of the study period (Q50 = 0·2, Fig. 4), which gives 1/2 × 0·2 × 0·33 = 0·033 same-sex recruits from one breeding attempt. More than 30 breeding attempts are required for such sites to yield a stable or growing population (1/0·033 = 30·3), clearly beyond the average life span of guillemots.
More inexperienced birds could have bred in later years, generating the observed decline in breeding success. Therefore, we checked if the pattern of declining breeding success could have arisen simply through the null model (Appendix I). Breeding success of marked birds followed over years improved with time (regression with mean success of marked birds = α × year +β: estimate of α = 0·0028), but the null model produced an even stronger positive trend, α falling below that of the true data set in 308 of 10 000 runs (P = 0·031). Thus the data contained more, rather than fewer, experienced birds over time, and the breeding success of birds declined relative to what was expected on the basis of their improving experience. Repeating the randomization by excluding those birds already breeding at the start of the study, whose true experience status was unknown, produced an even more significant result (α below observed in 34 of 10 000 runs, P = 0·0034).
It is possible that changes in breeding output are due to increasing local density, rather than birds using different site qualities than before. However, this alternative explanation appears unlikely. We had data for the mean number of close (touching) neighbours for individual nest-sites in the years 1985 and 2002. Despite a much higher population density in the latter year, we found no significant difference in the mean number of neighbours (1985: 1·33 ± 0·09; 2002: 1·32 ± 0·09; Student's t = 0·32, n1 = 552, n2 = 472, P = 0·75).
Finally, we checked if mortality changed with density. The probability of not returning (likely death) was significantly higher in years when available breeding sites were of low quality (Q50 and probability of not returning: Spearman's rS = −0·50, n = 18, one-tailed P < 0·05).