Behavioural Ecology and Sociobiology
Supplementary material for:
Escape ability and risk-taking behaviour in a Hungarian population of the collared flycatcher (Ficedulaalbicollis)
MónikaJablonszky, EszterSzász, GáborMarkó, JánosTörök, GáborHerczeg, LászlóZsoltGaramszegi
*address for correspondence:
MónikaJablonszky
Behavioural Ecology Group, Department of Systematic Zoology and Ecology, EötvösLoránd University, PázmányPétersétány 1/c, Budapest 1117, Hungary
This supplementary material contains additional information on the relevance of escape ability. We repeated the statistical analysis by using a binary-state variable for escape ability, with which we aimed to separate the effect of endurance from manoeuvrability. The methods, the results and a brief discussion of these analyses are presented.
Analysis with a binary escape ability variable
To make an effort to estimate escape ability unbiased by endurance, we binarized our focal variable based on a short-time window cut-off criterion and repeated the entire set of analyses. Given that the measured time to capture for birds that were difficult to catch within the first few seconds (i.e. the observer needed to make several approaches to exhaust them until a successful catch) more likely reflects endurance than manoeuvring capacity, we set the value of the binary-state variable to 0 if the time to capture was below 10s, and 1 otherwise. We selected this threshold criterion based on the following considerations. First, it resulted in relatively balanced data (0: 34, 1: 87). Second, it had to be short time period, thus we could rule out the effect of exhaustion. Accordingly, two birds, for which we derived different but generally high estimates for time to capture (20 and 120 sec for instance) would receive the same binary scores, so their apparent difference that is caused by exhaustion on the original scale becomes masked.
Similarly to the original analysis, there was a positive relationship between binarized time to capture and the first approach distance (N=39, β±SE = 1.280±0.637, t = 2.010, P = 0.044).
When we repeated the entire analysis with the binary state variable for time to capture variable that eliminates the confounding effect of exhaustion, we could generally derive similar results (supplementary tables 1-6, supplementary text 1). The only exception was that mass did not have significant relationship with time to capture in any of the models. Furthermore, in the model including only yearlings, the relationship of binary time to capture and FID did not reach significance. Focusing on effect sizes instead of statistical significance, we note that effect sizes derived for the binary state variable were approximately half of the effect sizes for the original time to capture variable in the case of the full model investigating the relationship between escape ability and proximal factors and that model investigating the connection of FID and escape ability in yearlings, but were greater differences in the effect sizes of the reduced model about the relationship of escape ability and mass, and of escape ability change and mass change.
These finding validates that the measured time to capture reflects an ability to react to a single predator attack in the wild, especially by ambush predators, such as A. nisus or A. gentilis(Kenward 1982), as the chase rarely last until the full exhaustion of the birds (Witter and Cuthill 1993; Veasey et al. 1998). The main differences in the results with the binary variable were that binary time to capture did not significantly relate to mass, neither to FID even in yearlings. Note that not only significance was altered but the effect sizes of these connection were also lowered indicating that the differences between the two sets of results are not due to the lower power associated with the binary state variable (as a consequence of information loss due to binarization). A possible explanation is that endurance in the original variable mediated these (but only these) relationships. Mass seemed to be influencing only the time to capture above 10s, thus it was probably connected to endurance.
Supplementarytable 1Repeatability of thebinarytimetocapturevariable
N / R / PWithinday / 45 / 0.112 / 0.131
Withinday (courtship) / 39 / 0.922 / 0.016
Withinday (chick-feeding) / 13 / 0 / 0.232
Withinyear / 52 / <0.001 / 0.658
Betweenyears / 31 / <0.001 / 0.976
Betweenyears (courtship) / 21 / 0 / 0.801
Betweenyears (chick-feeding) / 13 / 0 / 0.621
Supplementary table 2The results from the linear mixed model investigating the relationship between binary time to capture and the considered predictor variables in male collared flycatchers. The random factors were individual and experimenter identity and year. Period was excluded from the variables in order to investigate the effect of body mass that vary to a great extent between periods (in the model including also season, this variable was highly significant (N = 350, β±SE = 1.090±0.285, z = 3.822, P < 0.001), see main text for further details). In another model we included also experience with time to capture measurement, but it was not significant (N = 349, β±SE = -0.143±0.315, z = -0.456, P = 0.649). N=350
Predictorvariables / β (SE) / z / Pstandardisedmass / 0.440 (0.268) / 1.645 / 0.100
winglength / 0.065 (0.074) / 0.882 / 0.378
tarsuslength / -0.056 (0.252) / -0.222 / 0.824
age / 0.532 (0.325) / 1.637 / 0.102
standardisedwing patch size / 0.001 (0.002) / 0.856 / 0.392
forehead patch size / <0.001 (<0.001) / -0.319 / 0.750
date / 0.043 (0.027) / 1.623 / 0.105
time / 0.011 (0.091) / 0.119 / 0.906
Random effects / Variance / Standard
Deviation
year / <0.001 / <0.001
experimenter / <0.001 / <0.001
identity / 0.007 / 0.085
Supplementary text 1 Among the simplified models containing only mass and the considered random effects, with binary time to capture as the response variable, we found that a model that only included random intercepts offered the best fit to the data, when compared to the model considering random slopes (N = 173, LRT test: χ2 = 0, P = 1) or an interaction between random slopes and intercepts (LRT test: χ2 = 2.721, P = 0.257). The effect of mass was not significant in the final model (LRT test: χ2 = 0.421, P = 0.81).
Supplementary table 3 Results from the model investigating in male collared flycatchers the relationship between binary time to capture change and mass change, controlling for the original binary time to capture, with year as random factor. N=50
Predictorvariables / β (SE) / z / Pmass change / -0.880 (0.964) / -0.913 / 0.361
original binary time to capture / -37.156 (915.894) / -0.041 / 0.968
Random effects / Variance / Standard
Deviation
year / 0 / 0
Supplementary table 4The results of the model investigating the relationship between flight initiation distance (FID) and binary time to capture and the considered control variables in male collared flycatchers. The random factors taken into account in this model are identity of the observer, year and the identity of the male decoy. P-values were calculated with likelihood ratio test (LRT). N=192.In the model containing adult birds and experience with FID, this variable was not significant (N = 137, β±SE = 0.192±0.098, t = 1.958, P = 0.076).
Predictorvariables / β (SE) / t / LRT χ2 / Pbinarytimetocapture / -0.106 (0.100) / -1.063 / 1.119 / 0.290
age / -0.314 (0.088) / -3.554 / 11.747 / <0.001
standardisedmass / -0.131 (0.072) / -1.289 / 3.156 / 0.076
date / 0.012 (0.009) / -1.807 / 1.359 / 0.244
Random effects / Variance / Standard Deviation
year / 0.011 / 0.105
observer / 0.036 / 0.189
maledecoy / 0.014 / 0.119
residual / 0.276 / 0.526
Supplementary table 5 Results from the model investigating the relationship between flight initiation distance (FID) and binary time to capture with control variables in yearling male collared flycatchers. The random factor for this model is year (we left out the identity of the decoy bird in this model, because of the limited sample size and because it accounted for zero variance). P-values were calculated with likelihood ratio test (LRT). N=95
Predictorvariables / β (SE) / t / LRT χ2 / Pbinarytimetocapture / -0.215 (0.161) / -1.337 / 1.771 / 0.183
standardisedmass / -0.194 (0.115) / -1.686 / 2.747 / 0.097
date / <0.001 (0.014) / -0.048 / 0.002 / 0.965
Random effects / Variance / Standard Deviation
year / -0.007 / 0.082
residual / 0.378 / 0.615
Supplementary table 6 Result of the model examining the connection between standardized owl latency and binary time to capture with control variables in male collared flycatchers. N = 46. The significance level of the year factor was calculated with likelihood ratio test (LRT)
Predictorvariables / β (SE) / t / LRT χ2 / Pbinarytimetocapture / -0.054 (0.228) / -0.236 / 0.815
standardisedmass / -0.074 (0.216) / -0.343 / 0.733
date / -0.039 (0.027) / -1.449 / 0.155
age / 0.106 (0.335) / 0.316 / 0.754
year / 1.105 / 0.553
Supplementary references
Kenward RE (1982) Goshawk hunting behavior, and range size as a function of food and habitat availability. J Anim Ecol 51:69-80 doi:10.2307/4311
Veasey JS, Metcalfe NB, Houston DC (1998) A reassessment of the effect of body mass upon flight speed and predation risk in birds. Anim Behav 56:883-889 doi:10.1006/anbe.1998.0880
Witter MS, Cuthill IC (1993) The ecological cost of avian fat storage. Philos T Roy Soc B 340:73-92 doi:10.1098/rstb.1993.0050