To maximise fitness, parents should optimise their investment in each breeding attempt. When there are multiple carers, the fitness of each individual may also depend on the relative timing of their investment, with coordination of care hypothesised to maximise its efficiency and reduce predation risk. The aim of this study was to test the hypothesis that carers coordinate provisioning as an antipredator measure that reduces the time that a brood’s location is advertised to predators (‘predation hypothesis’). We presented predatory and non-predatory model birds to provisioning long-tailed tit Aegithalos caudatus parents and helpers, predicting that coordination would increase, and carer activity near the nest would decrease following predator presentation, relative to controls. First, carers reduced provisioning rates and took longer to resume provisioning following the predator presentation. Second, contrary to predictions, we found no significant change in any metric of coordination following predator presentations, relative to controls. Moreover, following predator presentation carers spent more time near the nest, resulting in greater near-nest activity compared to controls. In conclusion, although provisioning long-tailed tits are sensitive to perceived predation risk, our findings do not support the prediction of the predation hypothesis that carers adjust coordination behaviour in response to that threat.

Study system and general field protocol

Data were collected during the breeding seasons (March–June) of 2021 and 2022 from a population of long-tailed tits in a ~3km² field site in the Rivelin Valley, Sheffield, UK (53°23′N, 1°34′W). Nests were located by following adult birds during nest building, and once located, were monitored at 1–3 day intervals to record lay dates and the start of incubation, with daily checks to record hatch date within 24 h. Brood size was recorded on day 11 (d11) after hatching (d0) when chicks were ringed. Long-tailed tits typically build their nests in low vegetation (<3 m) or tree forks (>3 m). Our experimental protocol required model presentation within 2 m of the nest, so we conducted experiments only on nests within reach of an observer (<3 m) (N = 22). All birds in our study were individually identifiable after being ringed with a British Trust for Ornithology ring and a unique combination of two colour rings either as nestlings in our field site or during nest building if immigrants. In total, 21 breeding females, 20 breeding males and 11 helpers (10 males, 1 female) were observed at 22 nests. It was not possible to record data blind because our study involved focal animals in the field.

Experimental protocol

Long-tailed tits provision their brood in the nest from hatching (d0) until failure or fledging (d16–18). Breeding females brood the chicks until around d5, during which time males provide most food, but from d6 onwards brooding ceases and carers provision chicks directly. We conducted experiments on d8–10 at 22 nests (d8: N = 12; d9: N = 6; d10: N = 4) in April–May 2021 and 2022. At each nest, we performed an experiment comprising a series of provisioning watches with intermittent periods of model presentation (Fig. 1). First, an observer (CH) set up a wooden pole (1.2 m tall) for model presentations and a video camera on a tripod ~2 m from the nest to record the time, to the nearest second, that each carer fed the brood, before retreating to an observation position >20 m from the nest. The first watch was a control provisioning period of 1h which commenced upon the first feed observed after an initial 10-minute acclimation period. The observer recorded the time, to the nearest second, that a carer arrived within 15 m of the nest, so that the time between arrival and feeding, termed the ‘loitering period’, could be calculated. If two or more carers arrived simultaneously, they were recorded as arriving at the same time, but the observer noted the order in which they were identified. Some arrivals did not result in feeds (4.69%, N = 1,561), so were omitted from our final analysis. In addition, some arrivals (4.03%, N = 1,491) were missed, in which case the time of arrival was assumed to be the time first seen on camera; the longer a carer loitered the lower the chance that their arrival was missed, so we reasoned that missed arrivals would typically occur with short loitering periods. Watches were performed for one hour after the first observed feed, and watch duration was calculated as the time between first arrival and final feed.

Following the control period, the observer fixed the first taxidermic model atop the pole before retreating to the same observation position. The model was either a non-predatory (rock dove, Columba livia) or predatory (Eurasian jay) bird; jays commonly depredate long-tailed tit eggs and chicks, but not adults. The first model was presented for ~15 minutes (dove mean (±SE): 15 mins 19s ± 6s, N = 22; jay mean (±SE): 15 mins 22s ± 8s, N = 22), while the observer and video camera recorded arrival and feed times to the nearest second. Although we did not systematically quantify other behaviours during this period, carers usually responded to the jay by intensely alarm-calling and mobbing the model before retreating from the nest area. In contrast, during the dove presentation, carers alarm-called less and rarely mobbed the model, often resuming provisioning within 2 minutes of the presentation. After 15 minutes the observer retrieved and concealed the model before returning to the same observation position to perform another 1h provisioning watch. The observer then repeated the presentation procedure and subsequent observation period for the other model. All watches at a focal nest were performed on the same day, back-to-back, with minimal breaks between watch periods and the next model presentation. Each experiment typically lasted 4–5 h. The order of model presentations was stratified to minimise confounding order effects with model treatment (dove first: N = 10 trials; jay first: N = 12 trials). Nevertheless, we tested the effect of presentation order on all measures of coordination, finding that it was not statistically significant in any case and no other result qualitatively changed with its inclusion (Table S1). Therefore, we did not include presentation order in our final analyses.

Calculating coordination

Effect of experimental treatments on provisioning behaviour

To establish whether model presentations disturbed regular provisioning behaviour, we first calculated the time between the removal of each model (dove or jay) and the first feed by any carer during the subsequent watch, termed the ‘Lag time’ (Fig. 1). Secondly, we calculated the ‘Number of feeds’ by all carers during control watches, during model presentation periods (dove or jay present) and in watches following model removal (post-dove and post-jay). This was analysed as the total number of feeds per watch, which functioned as a measure of provisioning rate when watch duration was included in the model.

Does predation threat increase coordination?

To test the prediction that carers increase their coordination in response to elevated predation risk, we calculated provisioning and coordination metrics by all carers during control watches and post-dove and post-jay watches, as follows. ‘Alternation’ – the number of alternated feeds, defined as the number of feeds that occurred following the feed of another carer, i.e. non-consecutive feeds (e.g. A-B-A-C-B) (median = 16, range = 3–37, N = 66 watches at 22 nests). ‘Arrival synchrony’ – the number of synchronised arrivals, defined as the number of arrivals that occurred within a time window of 2 minutes following an arrival by another carer (as in: Bebbington and Hatchwell 2016; Halliwell et al 2022, 2023) (median = 10, range = 1–30, N = 66). ‘Feed synchrony’ – the number of synchronised feeds, defined as the number of feeds that occurred within a 2 minute time window of a feed by another carer (median = 10, range = 1–30, N = 66). ‘Present upon arrival’ – the number of feeds where the focal carer arrived back within 15 m of the nest with another carer loitering nearby (median = 6, range = 0–24, N = 66). ‘Present upon feed’ – the number of feeds where the focal carer fed whilst another carer loitered nearby (median = 5, range = 0–24, N = 66). ‘Loitering time’ – the mean loitering period duration by all carers who provisioned the brood during a watch (mean (±SE) = 45.8s ± 3.8s, N = 66) and ‘Duration of time with carer(s) nearby’ – the total time during each watch where one or more carers loitered within 15 m of a nest (mean (±SE) = 11 mins 54s ± 36s, N = 66); when watch duration was included as a covariate this measure functioned as an analysis of the proportion of a watch where at least one carer was loitering nearby.

To further test the dynamics of any potential response to predation threats, we compared several metrics of carer behaviour between sub-sections of control watches and post-dove and post-jay watches, each split into thirds by watch duration (mean third duration (±SE) = 18 mins 47s ± 8s, N = 198); response variables were: ‘Number of feeds’ (provisioning rate), ‘Loitering time’, and the levels of each coordination metric (‘Arrival synchrony’, ‘Present upon arrival’, ‘Present upon feed’, ‘Feed synchrony’, ‘Alternation’). This analysis attempted to determine the time frame on which carers adjusted their provisioning behaviour in response to elevated predation risk.

Statistical analysis

All statistical analyses were performed in R version 4.2.1 (R Core Team 2022). All models were built using lme4 (Bates et al 2015) and P values were extracted using lmerTest (Kuznetsova et al 2017). Where appropriate, we used the package emmeans (Lenth et al 2019) to perform post hoc testing. Figures were produced using the packages ggplot2 (Wickham 2016), cowplot (Wilke 2020) and ggsignif (Ahlmann-Eltze and Patil 2021). Our general approach to analyses was to use full mixed effects models with all biologically relevant fixed and random effects. To confirm that our findings were not influenced by overparameterisation of these models, we also conducted analyses of the same dataset using stepwise elimination of non-significant terms and an information theoretic (AIC) approach, but results were very consistent between these methods (see online resource details, table S3). Therefore, here we present only the results of full mixed-effect models.

Effect of experimental treatments on provisioning behaviour

To determine whether model presentation treatments affected provisioning behaviour, we first built a normally distributed linear mixed effects model (LMM) to compare the lag time between model removal and the first subsequent feed for dove and jay treatments. The response variable was log-transformed ‘Lag time’ because the assumption of normality was met only when the response variable was appropriately transformed. This model’s term of interest was ‘Treatment’ (post-dove or post-jay), with the fixed effects covariates as follows. ‘Provisioning rate’ – the total number of feeds by all carers per hour in the watch following model removal (in this case: mean (±SE) = 25.09 feeds/h ± 1.33, range = 7.94–46.42, N = 44) ‘Carer number’ – the number of unique carers which provisioned during each watch following model removal (in this case: 2: 70.5%, 3: 20.5%, 4: 2.3% and 5: 6.8%; N = 44). ‘Brood size’ – the number of live chicks recorded in the nest on d11 (median = 9, range = 2–11, N = 22). ‘Watch start time’ – the time of day each experiment started (mean = 08:30 BST, range = 07:20–12:40, N = 22). ‘Brood age’ – the number of days since hatching upon which a watch was performed (d8: N = 12; d9: N = 6; d10: N = 4). Finally, ‘Hatch date’ – the number of days since March 1 of each year on which each brood hatched (median = April 30, range = April 19 – May 29, N = 22). Random effects were ‘Year’ and ‘Nest ID’ – a unique identifier for each nest. Detailed descriptions of all fixed and random effects are available in Table S2 (online resource). Secondly, we built a Poisson-distributed generalised linear mixed effects model (GLMM) to compare provisioning rates during control watches, model presentations, and watches following model removal (post-dove and post-jay). The response variable was ‘Number of feeds’ and the term of interest was ‘Treatment’ (control, dove present, jay present, post-dove or post-jay) with fixed effect covariates as follows: ‘Carer number’ – for provisioning watches, this was the number of carers that provisioned during each watch (2: 69.7%, 3: 21.2%, 4: 3.0% and 5: 6.1%; N = 66); for presentation periods, this was the maximum number of carers observed provisioning the nest at any point during the experiment (2: 68.2%, 3: 22.7%, 4: 0.0%, 5: 9.1%; N = 44), ‘Watch duration’ – the time, in minutes, between first arrival and last feed of each watch (including display periods: mean (±SE) = 41 mins 18s ± 125s, N = 110), ‘Brood size’, ‘Watch start time’, ‘Brood age’ and ‘Hatch date’, as above. Random effects were ‘Year’, ‘Nest ID’ and ‘Rowref’, which was an observation level random effect providing a unique identifier for each provisioning watch; included throughout to account for overdispersion in Poisson-distributed models.

Does predation threat increase coordination?

To test the prediction that carers increased coordination in response to a perceived predation threat, we first produced a series of Poisson-distributed GLMMs to compare coordination metrics between control watches and watches following model presentations. The response variables were as follows: ‘Arrival synchrony’, ‘Present upon arrival’, ‘Present upon feed’, ‘Feed synchrony’ and ‘Alternation’, as described above, with ‘Treatment’ (control, post-dove or post-jay) being the term of interest in each model. Fixed effect covariates for these models were: ‘Provisioning rate’ (in this case: mean (±SE) = 22.94 feeds/h ± 1.06, range = 7.94–46.42, N = 66), ‘Carer number’, ‘Watch duration’ (in this case: mean (±SE) = 58 mins 36s ± 48s, N = 66), ‘Brood size’, ‘Watch start time’, ‘Brood age’, ‘Hatch date’ and ‘Maximum possible alternation’ – a variable that reflects the disparity in provisioning rate within groups of carers, e.g. if two carers feed at the same rate, all feeds (except the first) are potentially alternated, but if one carer feeds more than all others combined there exist several feeds which cannot be alternated (mean (±SE) = 87.80% ± 1.34%). Therefore, this variable represents the percentage of visits performed during a watch which could theoretically be alternated (or synchronised) given the relative number of feeds by all carers during that watch (see Table S2 for further details). Random effects were ‘Year’, ‘Nest ID’ and ‘Rowref’.

Secondly, to test whether treatment influenced loitering times and consequently, the total time that carers were nearby the nest, we built two normally distributed LMMs. The response variables were log-transformed ‘Loitering time’ and ‘Duration of time with carer(s) nearby’, with the term of interest being ‘Treatment’ (control, post-dove or post-jay). Fixed effect covariates were: ‘Provisioning rate’, ‘Carer number’, ‘Watch duration’, ‘Brood size’, ‘Watch start time’, ‘Brood age’ and ‘Hatch date’. Additionally, to further investigate the time during a watch with carer(s) nearby we re-ran the original model with mean loitering time included as a covariate to account for the effect of potentially different loitering times during different watches; we report both model outputs. Random effects were ‘Year’, ‘Nest ID’ and ‘Rowref’.

As the final test of this hypothesis, we used a series of Poisson-distributed GLMMs to compare ‘Provisioning rate’ and coordination metrics between successive sections (1^st, 2^nd and 3^rd) of control watches and watches following model removal. The response variables were: ‘Number of feeds’, ‘Arrival synchrony’, ‘Present upon arrival’, ‘Present upon feed’, ‘Feed synchrony’, ‘Alternation’ and ‘Loitering time’, with the terms of interest being ‘Section’ and its interaction term with ‘Treatment’, which measures whether watches exposed to different treatments varied in their subsequent coordination through time. Fixed effect covariates were: ‘Provisioning rate’, ‘Carer number’, ‘Watch duration’, ‘Brood size’, ‘Watch start time’, ‘Brood age’, ‘Hatch date’ and ‘Maximum possible alternation’ (except the provisioning rate and loitering time models), which was calculated for each individual section of the watch, i.e. ‘Maximum possible alternation’ for 1^st, 2^nd and 3^rd sections separately. Random effects were ‘Year’, ‘Nest ID’, ‘Rowref’ and ‘Watch ID’ – a unique identifier for each watch from which a section was sampled.

Microsoft Excel

R

RStudio