Methods

Field work

Selvagem Grande (Lat 30.140556, Long -015.868889) is an island nature reserve that forms part of the Madeiran Archipelago, Portugal. Around 30 000 pairs of Cory’s Shearwaters breed on the island, arriving at their nest burrows between February and March to undertake a long period of nest defence and mating (the pre-laying period) before laying a single egg in early June, which they incubate for around two months. Over 450 regularly occupied nest cavities within rubble walls are the subject of long-term research, with annual monitoring of nest occupancy and breeding output resulting in hundreds of ringed individuals (Granadeiroet al. 2006). In 2016 and 2017 we deployed a large number of geolocation loggers (GLS, Migrate Technology model C330; 122 in 2016, 104 in 2017) to track individuals’ migratory movements over winter, revealing the migratory strategy of 110 males (98 overwinter movements 2016-2017 and 83 overwinter movements 2017-2018). GLS were leg mounted and weighed less than 1% of the birds’ body weight, which, given the attachment method, can be assumed to bear minimal or no effects on survival and fitness (Bodeyet al. 2017). The resighting rate of birds fitted with GLS is consistent with the rate of return of breeding birds from one year to the next, at ca. 85% (Mouginet al. 2008).

Behavioural tests

We carried out behavioural tests over three breeding seasons between 2016 and 2018.
We assessed exploration in an emergence/novel environment test. Birds were extracted from their nest, transported in a closed box to a small room which served as the novel environment, and transferred into a box with a removable flap door situated in a corner of the room. Over a 10 minute time bracket, birds could leave the open box into the empty room while being filmed. The whole process lasted 20 minutes and had no implication on the breeding success of the individuals tested. Thirty two birds were tested during the incubation period of 2016, 69 during the pre-laying period of 2017, and 43 during the incubation period of 2017. Twenty seven individuals were tested twice over the two years. From this test we extracted two scores; whether or not the bird left the box within the stipulated time (binary “emergence” score) and, if it did, the latency (in seconds) for the bird to leave it.
In 2017 and 2018 we tested the reaction on extraction from the nest in male Cory’s Shearwaters on first encounter, either when retrieving GLS in the pre-laying period or as part of annual nest monitoring during incubation. A binary “extraction score” was assigned depending on the birds’ reaction to an approaching hand, reaction on being caught, and resistance to being pulled out of the nest (Table 1), classifying birds as either “reactive” (1) or “unreactive” (0). Behaviours that could not be classified into either of the two classes according to this classification were not considered. Around 60% of all assessments resulted in a classification of either “reactive” or “unreactive”. All assessments were performed by the same person to minimise subjectivity, and were only carried out at nests where the target individuals were the sole occupants of the nest at the time of testing, were facing towards the nest entrance, and could be reached easily without the use of a noose – this is the scenario for the vast majority of nests in the study site. Extraction scores were obtained during pre-laying (2017: 56 individuals; 2018: 147 individuals) and incubation (2017: 168 individuals; 2018: 107 individuals). 17 individuals were assessed twice during pre-laying in 2018 to determine short-term repeatability. In total, 153 males were scored; 43 males twice, 17 three times, and 11 more than three times.

Statistical analysis

Statistical analyses were performed on R Statistical Package, version 3.5.1 (R Core Team 2018).
In order to identify whether age and annual cycle stage (pre-laying vs incubation) could influence the results of the behavioural traits, we ran generalised linear (mixed-effect) models (function glm, package stats or function glmer, package lme4). Age was treated a factor of six 5-year age classes. The latency to leave the box was fitted with a quasi-poisson error structure, while the emergence and extraction scores were fitted with a binomial error structure. Individual ID was added as a random intercept in models of emergence and extraction due to repeated assessments on the same individuals. To control for the effect of migratory strategy on extraction score (see Results), we  fitted this model only on migratory individuals. Previous breeding success does not influence extraction score (see Appendix 1, Supporting Information).
We estimated individual repeatability in temperament over the entire period of study to ensure that individuals were consistent in their behaviour in the short term, and that therefore temperament could be related to migratory strategy at least over this time period. Repeatability was calculated as the intraclass correlation coefficient with a binomial error distribution (package rptR, Nakagawa & Schielzeth 2010) using bootstrapping without randomisation (1,000 iterations) to estimate confidence intervals. For assessing exploration behaviour, this was calculated on the emergence score using a binomial error distribution.
To investigate links between individual temperament and migratory strategy (migratory/resident) we ran generalised linear mixed effect models for the binary emergence and extraction scores as dependent variables in separate models, with individual as a random intercept. Age and annual cycle stage were added as fixed effects in the model explaining emergence.
Linear regression estimates (“Est”) are presented with their standard error.