Dataset DOI: 10.5061/dryad.wdbrv162g

Description of the data and file structure

This study investigates how wild rock hyrax (Procavia capensis) social networks respond to short-term changes in perceived predation risk and food availability. Between 2017 and 2020, we conducted 15 experimental trials in two neighboring canyons at the Ein Gedi Nature Reserve, Israel—nine “fear-related” experiments using wolf call playbacks or simulated aerial attacks by drones, and six “food-related” experiments using artificial food patches. Each trial included pre-treatment, treatment, and post-treatment stages. Social data were collected using proximity biologgers, direct behavioural observations, and trail cameras, from which we derived proximity-based social networks. To account for differences in data types, resolution, and sampling effort, we applied a double permutation framework and a meta-analytic approach to quantify treatment effects, explore technical sources of variation, and test ecological moderators of social responses.

Files and variables

File: Experiments.xlsx

Description: This table lists all experimental replications conducted on wild rock hyraxes at the Ein Gedi Nature Reserve between 2017 and 2020. Each row represents one experimental stage (pretreatment, treatment, posttreatment, or control). The table includes information about the experimental setting, timing, and data collection methods used.

Variables

• number – Identifier for each experimental replication. Numbers separated by commas indicate that the same experimental stage applied to multiple replications.
• canyon – Name of the canyon where the experiment took place (Arugot or David).
• target – Name(s) of the specific social groups targeted in the experiment.
• phase – Experimental stage: Pretreatment (baseline data collection), Treatment (experimental manipulation), Posttreatment (post-manipulation data collection), or other control phases.
• type – Method used for the experimental manipulation (playbacks or drone).
• start – Start date of the phase (DD/MM/YYYY).
• end – End date of the phase (DD/MM/YYYY).
• duration – Duration of the phase in days.
• observations – x if direct behavioural observations were conducted during this phase; blank if not.
• loggers – x if proximity biologgers were deployed and retrieved during this phase; blank if not.
• cameras – x if trail cameras were used during this phase; blank if not.

File: Life_History.xlsx

Description: This table contains the complete life-history records of all rock hyraxes ever identified in the study population. Each individual has a unique key ID, and records are organized by year, providing demographic, biological, and field observation details. The dataset includes information on age, sex, maturity stage, body measurements, field presence, and trapping history, allowing reconstruction of individual life trajectories and group membership over time. 

Variables

• Key – Unique identifier assigned to each individual hyrax.
• Chip – Microchip number implanted in the individual.
• Chip_in_files – Microchip number as recorded in related datasets for cross-referencing.
• ID – Field identification code used by observers (if any).
• Site – Canyon where the individual’s group resides (Arugot or David).
• Group – Demographic social group to which the individual belonged in that year.
• Year – Year of observation.
• Year.of.Birth – Known birth year.
• Maturity.stage – Age class (pup, juvenile, adult).
• Age_General.(months) – Estimated age in months based on general records.
• Camille.predicted.age.(months) – Model-predicted age in months (if available).
• Minimal.age.(years) – Minimum possible age in years based on field records.
• Year.of.Birth.(inferred) – Estimated year of birth when exact birth year is unknown.
• Age_Adi.(months) – Adjusted age in months from alternative estimation methods.
• Sex – Biological sex (M or F).
• male.status – Social status of males (dominant, subordinate); blank for females.
• Observed – TRUE if the individual was directly observed that year, FALSE if not.
• Trapped – TRUE if the individual was trapped that year, FALSE if not.
• Present_on_field – TRUE if the individual was confirmed present in the population that year, FALSE if absent.
• weight – Weight in kilograms.
• body_length – Body length in centimeters.
• body_girth – Body girth in centimeters.
• head_size – Head size in millimeters.
• death.recovery – Notes on recovery of the body after death (if applicable).

Empty cells throughout the sheet mean the data was either not available or could not be inferred by our models.

File: gbi.list.rds

Description: This RDS file contains a list of data objects for each experimental replication. Each element includes:

• gbi – A data frame of group-by-individual (GBI) data representing co-occurrences in space between members of a social group for that experiment. The columns of the GBIs correspond to individuals in the social network and each line correspond to behavioural observation. A 1 in line t, column j means individual in column j was observed during the time point t.  If individual j was interacting with any other group member (e.g., individual a, b, and f) during this time point, 1s will be present in the columns a, b, and f of line t, 0s otherwise.
• times – A data frame containing the timestamps of the recorded events corresponding to the GBI matrix. It contains has many rows as gbi, and two or three columns:
  • start:  featuring the starting time stamp when the corresponding behavioural observation in the gbi data frame starts. (numerical or DD:MM:YYYY HH:MM:SS format)
  • end:  featuring the ending time stamp when the corresponding behavioural observation in the gbi data frame ends. (numerical or DD:MM:YYYY HH:MM:SS format)
  • location (optional): certain datasets relate to trail camera data and feature the location when the camera recording the interaction was placed. It is a coded location name set as a character string.
• Other metadata – Additional information for each experiment, including the experiment ID, data collection type, and other relevant contextual details, presented as vectors of length 1.

This structure allows linking spatial co-occurrence patterns with their exact timing, while preserving experiment-level context for downstream analyses.

File: Subgroup_stimulus_mean_fear.csv

Description: This CSV file contains subgroup-level meta-analytical estimates derived from subgroup meta-analyses. Each row represents results for a given subgroup, including effect size estimates, statistical significance, and associated metadata.

Variables

• Coef: Estimated coefficient from the model.

• beta: Effect size estimate.

• SE: Standard error of the estimate.

• tstat: t-statistic value.

• df_Satt: Satterthwaite-approximated degrees of freedom.

• p_Satt: p-value based on Satterthwaite approximation.

• df: Model degrees of freedom.

• CI_L: Lower bound of the confidence interval.

• CI_U: Upper bound of the confidence interval.

• meta: Identifier for the meta-analysis.

• dcol: Data collection type or grouping variable.

• measure: The network or behavioral measure analyzed.

• transition: Experimental transition or treatment stage.

File: Stimul_compare_mean_fear.csv

Description: This CSV file contains statistical comparisons between subgroup-level estimates obtained from the subgroup meta-analysis. Each row represents a pairwise comparison of effect sizes, including test statistics, confidence intervals, and adjusted p-values.

Variables

• estimate: Difference in effect size estimates between subgroups.

• stderror: Standard error of the estimate.

• zval: z-statistic value for the comparison.

• pval: Raw p-value for the comparison.

• ci.lb: Lower bound of the confidence interval.

• ci.ub: Upper bound of the confidence interval.

• meta: Identifier for the meta-analysis.

• measure: The network or behavioral measure analyzed.

• transition: Experimental transition or treatment stage.

• padj: Adjusted p-value (multiple testing correction).

File: meta_regressions.csv

Description: This CSV file contains the results of meta-regressions examining how network traits respond to experimental treatments and other predictors. Each row corresponds to a meta-regression model output, including coefficients, test statistics, confidence intervals, and model descriptors.

Variables

• Coef: Name of the predictor variable in the meta-regression.
• beta: Estimated effect size (regression coefficient) for the predictor. 
• SE: Standard error of the coefficient estimate. 
• tstat: t-statistic value for the predictor. 
• df_Satt: Satterthwaite degrees of freedom for the test. 
• p_Satt: p-value from the Satterthwaite test. 
• df: Model degrees of freedom. 
• CI_L: Lower bound of the 95% confidence interval. 
• CI_U: Upper bound of the 95% confidence interval. 
• centrality: Type of network centrality or trait analyzed. 
• transition: Experimental transition or treatment stage. 
• p_perm: p-value obtained from permutation tests (if applicable). 
• ctrl: Indicator of whether control variables were included in the model. 
• es_corr: Effect size corrected for potential biases. 
• heterogeneity: Measure of heterogeneity (e.g., I²) from the meta-regression. 
• exp_type: Type of experimental manipulation. 
• p: Model p-value or overall significance level.

File: Groups_compare_mean_food_extra.csv

Description: This CSV file contains the subgroup estimates from the subgroup meta-analysis for food-related experiments. Each row reports the effect size estimates and associated statistics for a given subgroup comparison, including confidence intervals and meta-analysis details.

Variables

• estimate: Estimated effect size for the subgroup.
• stderror: Standard error of the estimate. 
• zval: z-statistic value for the estimate. 
• pval: p-value for the subgroup estimate. 
• ci.lb: Lower bound of the 95% confidence interval. 
• ci.ub: Upper bound of the 95% confidence interval.
• meta: Meta-analysis identifier or grouping factor.
• measure: Network measure or trait analyzed.
• transition: Experimental transition or treatment stage.

File: Subgroup_tests_mean_food_extra.csv

Description: This CSV file contains the statistical comparisons between subgroup estimates from the subgroup meta-analysis for food-related experiments. Each row reports the comparison effect size, associated test statistics, and confidence intervals for differences between subgroups.

Variables

• estimate: Estimated difference in effect size between subgroups.

• stderror: Standard error of the estimated difference.

• zval: z-statistic for the comparison.

• pval: p-value for the comparison.

• ci.lb: Lower bound of the 95% confidence interval for the difference.

• ci.ub: Upper bound of the 95% confidence interval for the difference.

• meta: Meta-analysis identifier or grouping factor.

• measure: Network measure or trait analyzed.

• transition: Experimental transition or treatment stage.

File: average_permuted_traits.rds

Description: This RDS file contains the average permuted network traits calculated from 1000 data-stream permutations, providing a null distribution for assessing whether observed changes differ from random expectations. NaN means the data could not be computed. All measures are unitless.

Variables:

• name: Unique individual identifier.
• year: Year of data collection.
• site: Study site where the data were collected.
• exp: Experiment identifier.
• type: Type of experiment or treatment applied (e.g., playbacks, drone, or food)
• phase: Experimental phase (e.g., Pretreatment, Treatment, Posttreatment).
• group: Social group assignment of the individual (historical names used in the project).
• sex: Sex of the individual, F or M.
• weight: Body weight of the individual at the time of data collection (in kg).
• stage: Life-history stage (adult, juvenile, or pup).
• data_collection: Method used to record social interaction data (e.g., observations).
• id_num_loc: Cluster the individual belongs to during that experiment, when social bonds are aggregated across all experimental stages (Pretreatment, Treatment, Posttreatment). The cluster was determined using a walktrap community detection algorithm.
• id_num_loc_pha: Cluster the individual belongs to during that experiment, when social bonds are aggregated during one of the experimental stages only (Pretreatment, Treatment, Posttreatment). The cluster was determined using a walktrap community detection algorithm.
• diff_bonds_glo: Social differentiation score measured at the level of the study population, including all social connections in the network (multiple social groups).
• diff_bonds_loc: Social differentiation score measured at the level of the group including only the social connections in the social unit cluster, extended to its ego network (n = 1).
• diff_bonds_ind: Social differentiation score measured at the level of the individual including all the social connections in its cluster, extended to its ego network (n = 1).
• size_tot: Total network size (number of individuals in the study population that year).
• size_glo: Global network size (number of individuals detected in the study population that year, during that experimental replication).
• size_loc: Local network size (number of individuals in an animal's cluster).
• transitivity_glo: Global clustering coefficient measured at the level of the study population (multiple social groups).
• transitivity_loc: Local clustering coefficient measured at the level of a cluster, , extended to its ego network (n = 1).
• transitivity_ind: Individual-specific clustering coefficient, , extended to its ego network (n = 1).
• density_glo: Global network density, measured at the level of the study population (multiple social groups).
• density_loc: Local network density measured at the level of a cluster, extended to its ego network (n = 1).
• net_centrality_glo: Global network centralization measured at the level of the study population (multiple social groups).
• net_centrality_loc: Local centralization computed within a cluster, extended to its ego network (n = 1).
• net_modularity_glo: Global network modularity measured at the level of the study population (multiple social groups).
• degree: Degree centrality (number of associates).
• strength: Strength centrality (sum of weighted associations).
• eigen: Eigenvector centrality (overall influence within the network).
• assoc_stab_w: association stability (weighted). Within-group association stability, defined as the Pearson correlation between the weighted associations of an individual at two consecutive experimental stages.
• assoc_stab_b: association stability (binary). Within-group association stability, defined as the Jaccard index between the identity of an individual's social affiliates at two consecutive experimental stages.
• sd.degree: Standard deviation of individual degree centrality within a cluster.
• sd.strength: Standard deviation of individual strength centrality within a cluster.
• sd.eigen: Standard deviation of individual eigenvector centrality across permutations within a cluster.
• sd.diff_bonds_ind: Standard deviation of individual-level social differentiation scores within a cluster.

File: average_observed_traits.rds

Description: This RDS file contains the average observed network traits across individuals and experimental stages, serving as the empirical benchmark against which permuted changes can be compared. NaN means the data could not be computed. All measures are unitless.

Variables:

• name: Unique individual identifier.
• year: Year of data collection.
• site: Study site where the data were collected.
• exp: Experiment identifier.
• type: Type of experiment or treatment applied (e.g., playbacks, drone, or food)
• phase: Experimental phase (e.g., Pretreatment, Treatment, Posttreatment).
• group: Social group assignment of the individual (historical names used in the project).
• sex: Sex of the individual, F or M.
• weight: Body weight of the individual at the time of data collection (in kg).
• stage: Life-history stage (adult, juvenile, or pup).
• data_collection: Method used to record social interaction data (e.g., observations).
• id_num_loc: Cluster the individual belongs to during that experiment, when social bonds are aggregated across all experimental stages (Pretreatment, Treatment, Posttreatment). The cluster was determined using a walktrap community detection algorithm.
• id_num_loc_pha: Cluster the individual belongs to during that experiment, when social bonds are aggregated during one of the experimental stages only (Pretreatment, Treatment, Posttreatment). The cluster was determined using a walktrap community detection algorithm.
• diff_bonds_glo: Social differentiation score measured at the level of the study population, including all social connections in the network (multiple social groups).
• diff_bonds_loc: Social differentiation score measured at the level of the group including only the social connections in the social unit cluster, extended to its ego network (n = 1).
• diff_bonds_ind: Social differentiation score measured at the level of the individual including all the social connections in its cluster, extended to its ego network (n = 1).
• size_tot: Total network size (number of individuals in the study population that year).
• size_glo: Global network size (number of individuals detected in the study population that year, during that experimental replication).
• size_loc: Local network size (number of individuals in an animal's cluster).
• transitivity_glo: Global clustering coefficient measured at the level of the study population (multiple social groups).
• transitivity_loc: Local clustering coefficient measured at the level of a cluster, , extended to its ego network (n = 1).
• transitivity_ind: Individual-specific clustering coefficient, , extended to its ego network (n = 1).
• density_glo: Global network density, measured at the level of the study population (multiple social groups).
• density_loc: Local network density measured at the level of a cluster, extended to its ego network (n = 1).
• net_centrality_glo: Global network centralization measured at the level of the study population (multiple social groups).
• net_centrality_loc: Local centralization computed within a cluster, extended to its ego network (n = 1).
• net_modularity_glo: Global network modularity measured at the level of the study population (multiple social groups).
• degree: Degree centrality (number of associates).
• strength: Strength centrality (sum of weighted associations).
• eigen: Eigenvector centrality (overall influence within the network).
• assoc_stab_w: association stability (weighted). Within-group association stability, defined as the Pearson correlation between the weighted associations of an individual at two consecutive experimental stages.
• assoc_stab_b: association stability (binary). Within-group association stability, defined as the Jaccard index between the identity of an individual's social affiliates at two consecutive experimental stages.
• sd.degree: Standard deviation of individual degree centrality within a cluster.
• sd.strength: Standard deviation of individual strength centrality within a cluster.
• sd.eigen: Standard deviation of individual eigenvector centrality across permutations within a cluster.
• sd.diff_bonds_ind: Standard deviation of individual-level social differentiation scores within a cluster.

File: Groups_compare_mean_food.csv

Description: This CSV file contains the subgroup estimates from the subgroup meta-analysis for food-related experiments. Each row reports the effect size estimates and associated statistics for a given subgroup comparison, including confidence intervals and meta-analysis details.

Variables

• estimate: Estimated effect size for the subgroup.

• stderror: Standard error of the estimate.

• zval: z-statistic value for the estimate.

• pval: p-value for the subgroup estimate.

• ci.lb: Lower bound of the 95% confidence interval.

• ci.ub: Upper bound of the 95% confidence interval.

• meta: Meta-analysis identifier or grouping factor.

• measure: Network measure or trait analyzed.

• transition: Experimental transition or treatment stage.

File: Subgroup_tests_mean_food.csv

Description: This CSV file contains the statistical comparisons between subgroup estimates from the subgroup meta-analysis for food-related experiments. Each row reports the comparison effect size, associated test statistics, and confidence intervals for differences between subgroups.

Variables

• estimate: Estimated difference in effect size between subgroups.

• stderror: Standard error of the estimated difference.

• zval: z-statistic for the comparison.

• pval: p-value for the comparison.

• ci.lb: Lower bound of the 95% confidence interval for the difference.

• ci.ub: Upper bound of the 95% confidence interval for the difference.

• meta: Meta-analysis identifier or grouping factor.

• measure: Network measure or trait analyzed.

• transition: Experimental transition or treatment stage.

File: Subgroup_tests_mean_fear.csv

Description: This CSV file contains the statistical comparisons between subgroup estimates from the subgroup meta-analysis for fear-related experiments. Each row reports the comparison effect size, associated test statistics, and confidence intervals for differences between subgroups.

Variables

• beta: Estimated effect size for the comparison between subgroups.
• SE: Standard error of the estimated effect size.
• tstat: t-statistic for the comparison.
• df_Satt: Satterthwaite-adjusted degrees of freedom for the t-test.
• p_Satt: p-value associated with the Satterthwaite-adjusted t-test.
• df`: Degrees of freedom used in the comparison (alternative or unadjusted).
• CI_L: Lower bound of the 95% confidence interval for beta.
• CI_U: Upper bound of the 95% confidence interval for beta.
• meta: Identifier for the meta-analysis or subgroup comparison.
• measure: Name or type of the outcome measure analyzed.
• transition: Labels or description of the specific experimental transition being compared.

File: extracted_Z_fear_food_*.rds

Description: This series of 50 RDS files contains Z-scores computed by comparing observed network traits to network traits generated from 1,000 data-stream permutations. The entire permutation procedure was repeated 500 times, Z-scores were recalculated for each iteration, and stored as 50 lists of 10 data frames. All data frames have the same structure. Measures are unitless. Missing, or unavailable data are replaced with NA.

Variables

• name: Individual identifier.
• year: Year of data collection.
• site: Study site (David or Arugot).
• exp: Experiment identifier.

sex: Sex of the individual (F or M).

• group: Social group assignment (historical names used in the project).

• weight: Body weight of the individual (in kg).

• stage: Experimental stage (e.g. Pretreatment, Treatment, Posttreatment, Wolf).

• type: Type of treatment (e.g. playbacks, drone, food) .

• data_collection: Method of data collection (e.g. observations, loggers, cameras).

• measure: Network trait or behavior being measured.

• diff.1, diff.2, diff.3, diff.4, diff.5: Observed differences in network traits between pairs of experimental stages. Some experiments present a Wolf stage hence, up to 5 differences can be computed (1: Treatment - Pretreatment, 2: Posttreatment - Treatment, 3: Wolf - Treatment, 4: Posttreatment - Wolf, 5: Postreatment - Pretreatment).

• ctrl.1, ctrl.2, ctrl.3, ctrl.4, ctrl.5: Average of 1000 permutation-based (i.e. model predicted) differences in network traits between pairs of experimental stages. Number of permutations is 1000.

• pval.1, pval.2, pval.3, pval.4, pval.5: Permutation-based p-values assessing wether the observed differences (diff) are statistically difference from their control values (ctrl). Number of permutations is 1000.

• wm.pre, wm.tre, wm.pos, wm.wol: observed centrality values per stage.

• z_dev1, z_dev2, z_dev3, z_dev4, z_dev5: z-scores quantifying the deviance of observed differences between stages from the permutation-based distribution. Number of permutations is 1000.

File: z.fear.food.csv

Description: This CSV file contains the changes in network traits for individual hyraxes across various pairs of experimental stages, integrating both fear- and food-related manipulations. It centralizes individual-level data, experimental context, and summary statistics (mean, median, differences, standard errors, proportions, and p-values) to facilitate downstream analyses and plotting. The file contains both observed values and the average and median values extracted from the 500 data frames stored in the extracted_Z_fear_food_*.rds files. Measures are unitless. Missing or unavailable data are left blank or contain NA.

Variables

• name: Individual identifier.

• sex: Sex of the individual (F or M).

• group: Social group assignment (historical names used in the project).

• weight`: Body weight of the individual (in kg).

• stage: Experimental stage (Pretreatment, Treatment, Posttreatment`).

• exp`: Experiment identifier.

• site: Study site (DavidorArugot`).

• year`: Year of data collection.

• type: Type of treatment (e.g. playbacks, drone, food`) .

• data_collection: Method of data collection (e.g. observations, loggers, cameras`).

• measure`: Network trait or behavior being measured.

• age`: Age of the individual (months).

• status: Social or reproductive status for males (subordinateordominant`). NA for females.

• cluster`: Assigned cluster within network analysis.

• social.status: Social rank or dominance category (MS: subordinate males, MR: dominant/resident males, MB: bachelor/solitary males, F`: females).

• wei_ind: Individual-level weighted network measure to use in meta-analyses.

• season: Season during data collection (summer or winter).

• gp.size: Group size as determined by a walktrap algorithm on the global network.

• diff.1, diff.2, diff.5: Average differences in network traits between pairs of stages across permutations (1: Treatment - Pretreatment, 2: Posttreatment - Treatment, 5: Posttreatment - Pretreatment). Note that the real (observed) changes in network traits are stored in the first data frame of the extracted_Z_fear_food_1.rds (i.e., permutations = 0).

• diff.se1, diff.se2, diff.se5`: Standard errors of differences across permutations.

• pval.1, pval.2, pval.5`: p-values of average differences, computed as the bonferroni p-value of 500 p-values. Number of permutations is 1000 per chain, repeated in 500 chains.

• prop.1, prop.2, prop.5`: Number of times the permutation-based p-value was significant out of 500. Number of permutations is 1000 per chain, repeated in 500 chains.

mean.z1, mean.z2, mean.z5`: Mean z-scores for each pair of stages computed across 500 chains. Number of permutations is 1000 per chain, repeated in 500 chains.

• mean.se1, mean.se2, mean.se5: Standard errors of 500 mean z-scores. Number of permutations is 1000 per chain, repeated in 500 chains.
• prop.z1, prop.z2, prop.z5: Number of times the Zscores were larger than the arbitrary value of 2 out of 500 Z-scores. Number of permutations is 1000 per chain, repeated in 500 chains.
• median.z1, median.z2, median.z5: Median z-scores for each stage computed across 500 chains.
• median.se1, median.se2, median.se5: Standard errors of 500 median z-scores.

Code/software

All data files can be viewed and analyzed using R (version 4.3.2, “Eye Holes”) on macOS (darwin20, aarch64). The workflow relies on R scripts that manage data, perform network analyses, compute meta-analyses, and generate plots. The following R packages were used:

• Data management: openxlsx, dplyr, plyr, lubridate
• Plotting and social network analysis: ggplot2, ggpubr, ggthemes, cba, ggdendro, dendextend, tidygraph, asnipe, igraph
• Meta-analysis: dmetar, meta, metafor, clubSandwich
• Parallelization: largeList, parallel, doParallel, foreach, doSNOW

The workflow begins with reading and pre-processing raw and summarized data, followed by computation of individual and group-level network metrics, permutation-based significance testing, and meta-analytic comparisons across experimental stages. Scripts generate intermediate files (e.g., Z-scores, permuted trait averages, etc.).

R session settings were standardized using a setup script to ensure reproducibility, including locale (en_GB.UTF-8), language (English), time zone (Asia/Jerusalem), and warning suppression.

Included scripts:

• Script0_Session_setup.R – Sets up the R session, defines custom functions, and standardizes environment settings.
• HighstatLibV10.R – Contains custom functions ported from the deprecated ADE package.
• Script1_parallelizedRandomTraits.R – Performs 1,000 data-stream permutations on raw co-occurrence data, repeated 500 times, measuring 19 network traits per iteration.
• Script2_parallelizedZscoreExtraction.R – Measures changes in network traits between experimental stages and extracts Z-scores comparing observed traits to permuted distributions, repeated 500 times.
• Script3_metaAnalyses.R – Conducts meta-analysis of Z-scores and performs subgroup meta-analyses.
• Script4_metaRegressions.R – Performs meta-regression analyses and node-level permutation tests.
• Script5_SI2_forestPlot.R – Generates forest plots for meta-analyses in the Supplementary Information.
• Script6_SI2_Subgroup_estimates.R – Performs subgroup meta-analyses for the Supplementary Information.

This structured workflow ensures reproducibility and facilitates downstream analyses, plotting, and interpretation of social network dynamics under experimental manipulations of predation risk and food availability in wild rock hyraxes.