ELECTRONIC Supplementary material
The effect of local dominance and reciprocal tolerance on feeding aggregations of ocellated antbirds
Johel Chaves-Campos1, Yimen Araya-Ajoy2, Claudia A. Lizana-Moreno2 and Kerry N. Rabenold1
1Department of Biological Sciences, PurdueUniversity, West Lafayette, IN, USA
2 Biology School, University of Costa Rica, San Jose, Costa Rica
Details on statistical analysis
Feeding location
Repeated measures analysis of variance: We used analysis of variance with binomial error distribution in Proc Glimmix, SAS 9.1, after checking for lack of overdispersion (Littell et al. 2006). We accounted for the repeated use of the same individuals in our analysis by including bird identity as a random effect and by specifying a first-order autoregressive covariance structure. As in subsequent models, pairwise comparisons of means were obtained using the DIFF command, and degrees of freedom were calculated using the between-within and the Kenward-Roger approach(Littell et al. 2006). All main effects and respective interactions were initially added, and non-significant terms were dropped, starting with the lowest F-value, until the remaining model included only significant (α 0.05) terms. Significant P-values correspond to final models.
Calculation of areas and the number of roosting areas expected for each distance: We determined the area of each distance category by calculating the area of a circle with radius equal to the distance category and subtracting it from the area of the concentric circle corresponding to the preceding category (the smallest category being a single roosting area; average radius = 125 m; Chaves-Campos & DeWoody 2008). We estimated the number of roosting areas per distance category by multiplying its area by the average density of roosting areas. This density was estimated by dividing the area of the adjacent-neighbour category by the average number of adjacent roosting areas that exist per roosting area (i.e. four; see above).
Prey intake
This and all subsequent analysis of covariance were estimated using proc Mixed in SAS 9.1 after confirming a normal distribution of residual errors.
Caveat on statistical analyses
Most data were analyzed with linear mixed models that accounted for several confounding factors, using small samples in some cases. A caveat in these cases is that small samples increase the risk of mistaken inference. We minimized this risk by using two different approaches to calculate degrees of freedom, including the Kenward-Roger procedure, which was developed to reduce small sample bias in complex models (Kenward and Roger 1997, Littell et al. 2006). Both methods yielded similar results, which gives us confidence in that results are robust and were interpreted as carefully as possible.
Figure S1. Map of the study area showing the roost centres of 17 individuals (each one representing a mated pair). Land use and forest type are shown. Roosting areas were rarely centred in modified forest despite intensive sampling in these areas with mistnets (to catch and radiotag individuals) and/or radio receivers(to detected tagged birds).
Figure S2. Response intensity of ocellated antbirds to experimental challenges from playbacks of male adjacent neighbours and non-neighbours summarized using principal components analysis. Stronger reactions (i.e. higher PC scores)indicate shorter latency to response, longer response duration, and more vocalizations per response (Table S3 for details). Least square means and s.e. are shown.
Figure S3. Spatialdistribution of swarms visited bynesting ocellated antbirds (grey bars) versus the expected distribution given the number of roosting areas estimated in each distance category (see main text). Swarms were found in the roosting area of: the focal individuals (own), adjacent neighbours, next-adjacent neighbours (1 range away), neighbours removed 2 ranges, or more distant locations (some of which did not include roosting areas).
Figure S4. Interaction between prey intake rate and weather according to the sex of the focal individual. Prey intake rates (least square means) were estimated using the model presented on Table S2, and are presented with standard errors. Weather conditions correspond to PC1 from Table S1. The interaction shows that females decreased prey intakes as PC1 scores increased, while males did not show this effect. Increased PC1 scores indicate lower temperatures and higher humidity and rainfall, conditions that likely make feeding harder due to decreased activity of the ants(Willis 1967, Teles da Silva 1977). This result suggests that both sexes obtain similar prey intake rates in good weather but females obtain lower intake rates during rainy conditions.
Figure S5. Estimated individual probability of answering acoustic challenges as a function of distance from the roost centre of focal individuals exposed to the playback and the presence of other adult males. The interaction between these two variables was not significant. Probabilities were estimated using a repeated measures logistic regression model with binary error structure and autoregressive covariance structure with the following significant parameters (standard errors in parentheses): Intercept: 3.187 (1.221), distance from roost centre: -0.004 (0.002), presence of other adult males: -1.564 (0.651).
Table S1. Eigenvectors from a principal components analysis on weather conditions when the swarm was observed (temperature, humidity and rainfall). The proportion of variance in the dataset explained by each principal component is shown within parentheses.
variable / PC1 (0.41) / PC2 (0.32) / PC3 (0.27)temperature / -0.67 / 0.05 / 0.74
humidity / 0.57 / -0.61 / 0.56
rainfall / 0.48 / 0.79 / 0.38
Table S2. Generalized linear mixed model (individual as random effect) used to analyze individual prey intake rates in relation to the distance from the roosting area of the focal individual (fixed effect). Other significant fixed effects in the model include PC1 from Table S1, swarm width, number of heterospecifics, sex of the focal individual, forest type, and prey size.
Effect / Num DF / Den DF / F / Proosting area / 2 / 13 / 4.44 / 0.0339
PC1 / 1 / 114 / 4.71 / 0.0321
swarm / 1 / 114 / 1.35 / 0.2484
heterospecifics / 1 / 114 / 0.67 / 0.4148
sex / 1 / 13 / 0.58 / 0.4582
forest type / 1 / 8 / 4.63 / 0.0635
prey size / 2 / 22 / 4.02 / 0.0324
PC1*sex / 1 / 114 / 7.86 / 0.0060
PC1*heterospecifics / 1 / 114 / 7.10 / 0.0088
PC1*forest / 1 / 114 / 8. 56 / 0.0042
PC1*swarm / 1 / 114 / 8.68 / 0.0039
Table S3. Eigenvectors from a principal component analysis on variables recorded from responses of ocellated antbirds to experimental acoustic challenges (playback experiments). The variables include the total number of loud songs (sensu Willis 1973), and the time spent singing (in seconds) during a standard 10 minute period after the playback. Time spent singing was not directly correlated with the number of songs due to variation in the number of final elements among songs. Final elements are discrete repeated units at the end of each loud song (Willis 1973) that are likely aggressive given their harsh structure (Morton 1977, Owings and Morton 1998). Latency was the time (in seconds) after the playback started until a respondent answered. The proportion of variance in the dataset explained by each principal component is shown within parentheses.
Variable / PC1 (0.54) / PC2 (0.27) / PC3 (0.19)# songs / 0.51 / 0.83 / -0.23
latency / -0.59 / 0.53 / 0.61
time / 0.63 / -0.17 / 0.76
The first component was used as explanatory variable in the analysis of the response according to stimulus (adjacentneighboursvs.non neighbours). We interpret this component as intensity of response to the stimulus, in which positive values refer to responses that were producedquickly and included more loud songs and more time singing.
The variables presented in Table S3were chosen because they could berecorded with little error given the simplicity of the measurement. We also measured the frequency and amplitude of the response to adjacentneighbours and to non-neighbours from spectrograms. Measurements of these variables likely include error caused by variation in climate, position of the bird while singing, and interfering vegetationthat reduced our power to detect differences. Those variables were also summarized using principal component analysis and analyzed using repeated measures ANCOVA. We found no significant differences between responses to adjacentneighbours and non-neighbours.
Table S4. Number of dyads and displacements observed at eight army ant swarms. Dyads were classified as adjacent-neighbour or non-adjacent neighbour. The total number of dyads (mate-mate dyads excluded) and non-adjacent neighbour dyads observed per swarm are provided, as well as the number of observed displacements between non-adjacent and between adjacent-neighbour dyads.
Observed dyads / Observed displacementsswarm / total / non-adjacent / non-adjacent / adjacent
1 / 35 / 32 / 1 / 0
2 / 45 / 39 / 1 / 0
3 / 10 / 9 / 2 / 0
4 / 13 / 11 / 1 / 0
5 / 5 / 3 / 2 / 0
6 / 18 / 11 / 2 / 0
7 / 2 / 2 / 1 / 0
8 / 6 / 5 / 1 / 0
TOTAL / 11 / 0
We used a resampling method to test the prediction that the total number of displacements between non-adjacent neighbours across swarms (i.e. 11) was higher than expected by chance (i.e. a one-tail test). The probability was obtained through Monte Carlo simulation by resampling dyads at random within each swarm and adding the number of non-adjacent neighbour dyads across swarmsin each iteration. The simulation was constructed using the program PopTools (Hood 2006).
A summary of the simulated distribution is provided below:
Mean / Variance / ≥11 / iterations / P ≥ 118.59 / 1.63 / 49 / 1000 / 0.049
The distribution shows that it is unlikely to obtain a total number of 11 displacements (or more) between non-adjacent neighbours by chance alone given the relative number of non-adjacent dyads per swarm. We interpret this result as evidence that displacements are biased towards non-adjacent neighbours.
References
Hood, G. M. 2006. PopTools version 2.7.5. Available on the internet. URL
Kenward, M. G., and J. H. Roger. 1997. Small sample inference for fixed effects from restricted maximum likelihood. Biometrics 53:983-997.
Littell, R. C., G. A. Milliken, W. W. Stroup, R. D. Wolfinger, and O. Schabenberger. 2006. SAS for mixed models, second edition. SAS Institute Inc., Cary, North Carolina, USA.
Morton, E. S. 1977. On the occurrence and significance of motivation-structural rules in some bird and mammal sounds. American Naturalist 111:855-869.
Owings, D. H., and E. S. Morton. 1998. The evolution of vocal communication: a new approach. Cambridge University Press, Cambridge.
Teles da Silva, M. 1977. Behavior of the army ant Eciton burchelli Westwood (Hymenoptera-Formicidae) in Belem region .1. Nomadic-stationary cycles. Animal Behaviour 25:910-923.
Willis, E. O. 1967. The behavior of bicolored antbirds. University of California Publications in Zoology 79:1-132.
Willis, E. O. 1973. The behavior of ocellated antbirds. Smithsonian Contributions to Zoology 144:1-57.
1