The study was carried out on Lundy Island, a small island off the coast of south-west England (51°10′N, 4°40′W). Natural migration of house sparrows to, and presumably from, Lundy Island is rare (the minimum distance to the mainland is 19 km; Griffith et al., 1999a) because of their sedentary nature and their flight ability not being suited for a long, continuous distance (Summers-Smith, 1963). Over 95% of breeding occurs in nestboxes. Since 2000, all breeding birds and fledglings have been marked with unique colour combinations in addition to a metal ring supplied by the British Trust for Ornithology. The parental care of house sparrows at nestboxes was monitored during the breeding seasons of 2004 and 2005 (from late April to late August) for this study.
Parental care data
Recordings of incubation and nestling provisioning were carried out using four Sony digital video camera recorders with 90-min tapes. A video camera was set at 2- to 5-m distance from nestboxes; the field of view included a 30-cm radius around the nestbox. All recordings were conducted between 05:00 and 12:00 hours GMT. Nestboxes were checked for the number of eggs before videotaping and for the number of chicks before and after taping. Incubation activity was recorded once for each clutch in 2004 on day 11 (where the first egg was laid on day 1), whereas in 2005 recordings were obtained twice for a clutch (on days 9 and 12). Food provisioning was recorded twice for each brood: on days 7 and 11 (with day 1 being the day on which the first chick hatched; in most cases, all chicks hatched within 24 h). If there was heavy rain on an observation morning, recordings were conducted on the subsequent day, although such recordings were not common (< 5%).
Only the segment of recordings starting with the first appearance of either of the breeders and finishing at the end of a tape were used (named effective observational time), to exclude the time when birds might have been getting accustomed to our visit to the nest-box and the presence of the camera. The mean video recording length was 93.4 min (±1.58 SD, n = 1051) and the mean effective observational time was 89.4 min (±4.49 SD, n = 1051). For incubation recordings, we considered that a breeder staying in a nestbox for more than 1 min represented an incubation bout; three types of observations were possible during the incubation period: male incubation time, female incubation time and nonincubation time. For provisioning recordings, a visit in which a breeder entered a nestbox or a visit in which a breeder fed chicks from outside were considered to be feeding visits.
All statistical analyses were conducted in the R environment (version 2.3.1; R Development Core Team, 2006). Incubation and provisioning data were not always normally distributed. Therefore, prior to statistical tests, appropriate transformations were identified and applied using the Box–Cox transformation to achieve normality.
The within-year and between-year repeatabilities (2004 and 2005) for incubation time (min h−1) and feeding visits (visits/chick h) were calculated using a procedure using within- and between-variance components in a linear mixed effects model (LMM) with the restricted maximum-likelihood method (REML, nlme package in R; Pinheiro & Bates, 2000) with bird identity as a grouping random factor in the model (Lessells & Boag, 1987; Falconer & Mackay, 1996; Díaz-Uriarte, 2002). Standard errors for repeatability were estimated by the method described in Becker (1984). The incubation time of both sexes tended to decline with increasing clutch size and both male and female feeding visits per chick also tended to decline with increasing brood size (Nakagawa et al., 2007a), so these effects were controlled for. Therefore, another set of repeatability estimates included clutch and brood size as a fixed factor to control for the effect of brood and clutch size, and bird identity as a grouping random factor. Statistical significance of repeatability was estimated from a likelihood ratio test comparing deviances between an LMM without a grouping random factor (i.e. individual identity) and an LMM with this factor. The use of the parametric bootstrap for likelihood ratios, rather than the chi-squared distribution, is recommended because likelihood ratios obtained in LMM often do not follow chi-squared distributions and the bootstrap approach provides more accurate approximation of P values (Faraway, 2006). Therefore, we used parametric bootstraps to create the distributions of likelihood ratios (5000 times) and the observed likelihood ratios are compared against these distributions to obtain P values (for more details of this method, see Faraway, 2005, 2006)
It should be noted that the observation periods for each data point differed between the analyses for incubation and feeding and also between years: incubation watches were made for ca. 90 min in 2004 and ca. 180 min in 2005, whereas provisioning watches were made over ca.180 min in both years. All the data used in the analyses were from broods where both parents were socially monogamous; data from broods belonging to polygamous males (i.e. social fathers observed at more than one nest at the same period of the breeding season) were excluded. For the within-year repeatability analyses (separate analyses for 2004 and 2005), data on individuals for which we had recordings from more than one brood (for incubation, two to four broods per year; for feeding, two to four broods per year) were used. For between-year repeatability, mean incubation or feeding visit values obtained for each individual within a year were averaged (i.e. brood average for each year), to compare one value from 2004 and another from 2005 for both incubation and feeding. Thus, a data point for the between-year repeatability analysis consisted of approximately 90- to 900-min observation for incubation (mean 354.8 ± 205.9 SD, n = 184) and approximately 180- to 720-min observation for feeding (mean 323.1 ± 133.3 SD, n = 138). These differences in data points should be borne in mind because they change the degrees of measurement errors in data (Falconer & Mackay, 1996)
Because individuals may maintain a similar amount of effort when they are with the same partner, the between-year repeatability for individuals that changed their partners during the 2004–2005 breeding season were also calculated. There were a small number of birds that changed partners within a breeding season. However, these individuals were in the minority (Table 1), so that separate calculations of within-year repeatabilities were not conducted. To compare repeatability estimates between the sexes and also to compare repeatability estimates between all birds and for birds that changed partners (i.e. MC in Table 1), we used a type of paired t-test which accounted for different sample sizes by employing LMMs (‘sex’ or ‘mate change’ as a fixed factor along with an appropriate random grouping factor; for a similar procedure, see Garamszegi, 2006).
Table 1. Within-year and between-year repeatabilities (R) of incubation time and feeding rate per chick; in the types of repeatability below, MC stands for mate change (including only individuals that changed their breeding partner during the 2004–2005 breeding seasons) and CSA and BSA stand for clutch- and brood size adjusted, respectively.
|Parental care||Year||Sex (type)|| R (SE)|| n[i] (n[c]), n[b]|| P ||Transformation|
|Incubation time (min h−1)||Within 2004||Male||0.345 (0.109)||49 (4), 116||0.0002|| 1|
|Male (CSA)||0.342 (0.109)||49 (4), 116||0.0008|| 1|
|Female||0.108 (0.104)||50 (7), 117||0.1616|| 1|
|Female (CSA)||0.086 (0.116)||50 (7), 117||0.2326|| 1|
|Within 2005||Male||0.342 (0.086)||61 (3), 170||< 0.0001|| 1|
|Male (CSA)||0.344 (0.086)||61 (3), 170||< 0.0001|| 1|
|Female||0.338 (0.087)||60 (6), 167||< 0.0001|| 1|
|Female (CSA)||0.350 (0.087)||60 (6), 167||< 0.0001|| 1|
|Between 2004 and 2005||Male||0.242 (0.140)||46 (24), 92||0.0468|| 1|
|Male (CSA)||0.218 (0.141)||46 (24), 92||0.0834|| 1|
|Male (MC & CSA)||0.213 (0.197)||24 (24), 48||0.1670|| 1|
|Female||0.349 (0.130)||46 (18), 92||0.0074|| 1|
|Female (CSA)||0.340 (0.131)||46 (18), 92||0.0090|| 1|
|Female (MC & CSA)||0.251 (0.224)||18 (18), 36||0.1540|| 1|
|Feeding rate (visits/chick h)||Within 2004||Males||0.462 (0.151)||25 (1), 53||0.0022|| 0.8|
|Male (BSA)||0.458 (0.152)||25 (1), 53||0.0034|| 0.8|
|Female||0.348 (0.165)||26 (2), 55||0.0280||−0.2|
|Female (BSA)||0.304 (0.170)||26 (2), 55||0.0512||−0.2|
|Within 2005||Male||0.442 (0.121)||37 (2), 83||0.0004|| 0.8|
|Male (BSA)||0.488 (0.115)||37 (2), 83||0.0002|| 0.8|
|Female||0.186 (0.141)||37 (1), 83||0.0450||−0.2|
|Female (BSA)||0.296 (0.135)||37 (1), 83||0.0060||−0.2|
|Between 2004 and 2005||Male||0.580 (0.110)||37 (17), 74||< 0.0001|| 0.65|
|Male (BSA)||0.625 (0.101)||37 (17), 74||< 0.0001|| 0.65|
|Male (MC & BSA)||0.682 (0.132)||17 (17), 34||0.0008|| 0.65|
|Female||0.284 (0.164)||32 (10), 64||0.0488||loge|
|Female (BSA)||0.271 (0.165)||32 (10), 64||0.0670||loge|
|Female (MC & BSA)||0.307 (0.294)||10 (10), 20||0.2126||loge|
We used LMMs to investigate whether incubation behaviour was related to subsequent food provisioning effort. First, to test simply for an association between these two parental behaviours, which was found to be positive in a previous study (Kopisch et al., 2005), we constructed two models, with either male or female incubation time as the response variable and fitting the corresponding feeding visit rate and brood size as fixed factors and year and bird identity as random factors. We then constructed LMMs using a global model approach. For these models, three fixed factors were fitted: brood size and any two of male incubation time, female incubation time or nonincubation time (these three variables always add up to 60 min so that we could not fit all of the three simultaneously, due to a singularity problem) and their second-order interactions. Thus, three different models for both male and female feeding visits were examined. For these latter models, the Akaike information criterion (AIC) was used for model minimization (the minimal adequate model had the smallest AIC of all of the alternatives; cf. Burnham & Anderson, 2002). It should be noted that LMMs using the maximum-likelihood method (ML) were used for model selection and then the final model was constructed using REML to obtain accurate parameter estimates. This is because the AIC values obtained from REML with different fixed factors are not comparable (Pinheiro & Bates, 2000). The effect sizes (correlation coefficients) and their 95% confidence intervals (CI) were also presented instead of adjusted P (i.e. Bonferroni-corrected) values, following a recent recommendation by Nakagawa (2004). The CIs for effect size were estimated according to Nakagawa & Cuthill (2007). Cohen's (1988) benchmark values are r = 0.1, 0.3 and 0.5, categorized as small, medium and large effects, respectively.