Social Network Analysis in Animal Behavior: From Wild Societies to Biomedical Research

Abstract

This article provides a comprehensive overview of Social Network Analysis (SNA) in animal behavior, tailored for researchers and drug development professionals. It explores the foundational principles of animal social structures and their implications for health, disease transmission, and welfare. The content covers cutting-edge methodological approaches, including dynamic network modeling and AI-assisted data collection, while addressing key challenges in data definition and analysis. By comparing findings across species and validating network robustness, this review synthesizes how SNA can inform biomedical models, enhance experimental design, and contribute to the development of novel therapeutic strategies by leveraging naturally occurring animal social systems.

The Social Blueprint: Unraveling Animal Societies and Their Biological Significance

Core Conceptual Framework

Social network analysis (SNA) in animal behavior research provides a powerful quantitative framework for understanding the complex social structures of animal populations. This methodology translates observed behaviors into a mathematical graph, enabling researchers to move beyond dyadic interactions and analyze the broader social system. The foundational elements of any animal social network are its nodes (representing individual animals) and edges (representing the social interactions or associations between them). The entire structure is termed a social graph, which can be analyzed using a suite of metrics to quantify social structure, individual positions, and group dynamics [1] [2].

In practice, animal social networks are distinguished by three key levels of abstraction [2]:

• Theoretical Constructs: The latent social relationships of interest (e.g., dominance, affiliation, friendship), which are abstractions that cannot be observed directly.
• Interaction Networks: The level of quantifiable social interactions (e.g., grooming, aggression) used as a proxy for the theoretical construct.
• Measured Networks: The raw, observed data on social interactions, which are subject to sampling effort and error.

Defining Network Components: Nodes and Edges

Nodes (Vertices)

In animal social networks, nodes almost exclusively represent individual animals. The definition of an "individual" within a study population must be clearly delineated during the research design phase.

Operational Guidance:

• Nodes should be uniquely identifiable individuals.
• The total population (node set) should be defined by clear spatial or membership criteria relevant to the species and research question.
• In captive settings, the node set may include all individuals in an enclosure. In wild populations, it often includes all animals within a defined social group or geographical area.

Edges (Links or Ties)

Edges represent a measurable social interaction or association between two individuals. The definition of an edge is the most critical step in study design and must be precisely tailored to the biological question. Edges can be constructed in several fundamental ways [2] [3]:

• Interaction Edges: Defined by a specific, observed behavioral event (e.g., grooming, aggression, vocal exchange). These are typically directed (from actor to receiver) and may be weighted by the frequency or duration of interactions [1].
• Association Edges: Defined by spatial or temporal proximity, or shared use of a resource. Two individuals are connected if they are observed within a specified distance, or if they co-use a resource like a refuge or foraging site, potentially asynchronously [3]. These are typically undirected.

Table 1: Common Edge Definitions in Animal Social Network Research

Edge Type	Definition	Directionality	Common Weighting	Typical Research Application
Aggression	Observed agonistic interaction (e.g., chase, bite)	Directed	Frequency, intensity score	Dominance hierarchy analysis
Allogrooming	One individual grooms another	Directed	Duration	Affiliative relationships, social bonding
Spatial Proximity	Individuals within N body lengths	Undirected	Co-occurrence frequency	Group cohesion, social organization
Gambit of the Group	Individuals observed in the same subgroup	Undirected	Number of co-occurrences	Defining association networks in fission-fusion societies
Shared Resource Use	Individuals using the same resource (e.g., refuge)	Undirected	-	Disease transmission, environmental sociology [3]

Experimental Protocols for Network Construction

Constructing a robust social network requires a standardized protocol from data collection to graph assembly. The following workflow details the primary steps for building an association-based network, a common approach in behavioral ecology.

Protocol 1: Constructing an Association Network via the "Gambit of the Group"

Objective: To map the social association network of a population based on group co-membership.

Materials:

• Focal study population
• Data recording equipment (e.g., video cameras, GPS tags, dataloggers)
• Statistical computing environment (e.g., R, Python)

Procedure:

• Behavioral Sampling:
  • Conduct focal animal sampling or scan sampling across multiple sessions to record the group composition at each time point [2].
  • A "group" is typically defined as individuals within a predetermined spatial threshold (e.g., 5-10 meters for primates).
  • Record all unique individuals present in each observed group.

• Construct a Group-by-Individual Matrix:

  • Create a bipartite matrix B where rows represent sampling events/groups (g) and columns represent individuals (i).
  • B(g,i) = 1 if individual i was observed in group g, and 0 otherwise. This creates a bipartite graph linking individuals to the groups they were observed in [3].

• Calculate Association Strength:

  • Derive a symmetric association matrix A from matrix B. A common metric is the Simple Ratio Index (SRI):
    • SRI(x,y) = Nxy / (Nx + Ny - Nxy)
    • Where Nxy is the number of sampling events/groups in which both x and y were observed, Nx is the number of events where x was seen, and Ny is the number of events where y was seen.

• Generate the Social Graph:

  • Perform a one-mode projection of the bipartite graph B onto the set of individuals [3].
  • In the resulting social graph, an edge exists between two individuals if their association strength (SRI) meets or exceeds a pre-defined threshold, or the edge can be weighted by the SRI value itself.

Protocol 2: Constructing an Interaction Network from Focal Sampling

Objective: To map a directed social network based on observed behavioral interactions.

Procedure:

• Data Collection:
  • Conduct standardized focal follows on each individual in the population for equal durations.
  • Record all instances of a pre-defined set of social behaviors (e.g., grooming, aggression, food sharing), noting the actor and receiver for each event.

• Construct the Adjacency Matrix:

  • Create a matrix M where rows represent actors and columns represent receivers.
  • The value in each cell M(i,j) is the frequency or total duration of the behavior initiated by individual i and directed toward individual j.

• Generate the Social Graph:

  • Matrix M directly defines a weighted, directed social graph. The graph can be analyzed as-is, or can be dichotomized (e.g., an edge exists if interaction frequency is above a certain threshold).

Analytical Framework and Key Metrics

Once a social graph is constructed, its properties can be quantified at the individual, dyadic, and group levels. The table below summarizes key social network metrics and their biological interpretations.

Table 2: Key Social Network Metrics for Animal Systems

Metric	Level	Definition	Biological Interpretation
Degree Centrality	Individual	Number of direct connections a node has.	Measures an individual's gregariousness or social popularity [1].
Strength	Individual	Sum of weights of edges connected to a node (weighted degree).	Measures the total intensity or frequency of an individual's social interactions.
Betweenness Centrality	Individual	Number of shortest paths between other individuals that pass through the focal node.	Identifies individuals that act as bridges or brokers between different parts of the network, potentially controlling information flow [1].
Closeness Centrality	Individual	Average shortest path length from a node to all other nodes.	Measures how quickly an individual can interact with or access all others in the network [1].
Eigenvector Centrality/PageRank	Individual	Measure of a node's influence based on the influence of its connections.	Identifies individuals connected to other well-connected individuals; a recursive measure of influence [1].
Clustering Coefficient	Individual/Group	Measures the degree to which a node's neighbors are connected to each other.	Quantifies the cliquishness of local neighborhoods; high values indicate tight-knit friendship triangles [1].
Network Density	Group	Proportion of possible edges that are actually present.	Measures the overall level of social connectivity in the population [1].
Modularity	Group	Strength of division of a network into modules (communities).	Quantifies the presence of distinct social subgroups or communities within the larger population [1].

The relationships between a hypothesized driver, the collected data, and the inferred social network can be formally represented using a causal modeling framework, as shown in the following Directed Acyclic Graph (DAG) [2].

The Scientist's Toolkit: Research Reagent Solutions

This section details the essential analytical tools and conceptual frameworks required for modern animal social network research.

Table 3: Essential Tools for Animal Social Network Analysis

Tool Category	Specific Solution	Primary Function	Key Features
Data Collection	Focal/Scan Sampling Protocol	Standardized behavioral observation	Ensures reproducible and unbiased data collection for edge definition.
Statistical Computing	R Programming Environment	Data manipulation, analysis, and visualization	Supported by packages like asnipe, sna, igraph, and tnet for comprehensive SNA [1].
Graph Analytics	igraph Library (R, Python, C++)	Network construction and metric calculation	Efficient implementations of algorithms for centrality, community detection, etc. [1]
Causal Inference	Bayesian Multilevel Models (e.g., Social Relations Model)	Estimating causal drivers of network structure	Isolates effects of individual, dyadic, and group-level features while accounting for data dependencies [2].
Conceptual Framework	Directed Acyclic Graphs (DAGs)	Formalizing causal assumptions	A graphical model to encode hypotheses and identify confounding variables [2].
Graph Databases & Query	Neo4j, TigerGraph, PuppyGraph	Storing and querying complex network data	Useful for managing highly connected data and performing complex traversals [1].
Crovozalpon	Crovozalpon, CAS:2406205-67-0, MF:C20H19ClF2N2O3, MW:408.8 g/mol	Chemical Reagent	Bench Chemicals
WK692	WK692, MF:C26H28Br2N8O5, MW:692.4 g/mol	Chemical Reagent	Bench Chemicals

Application Notes

This document provides application notes and experimental protocols for investigating the links between social connectivity, health, and reproductive success within the context of animal behavior research. The content is designed to support a broader thesis in social network analysis, offering researchers standardized methods to quantify social structures and their fitness consequences.

A growing body of evidence across species indicates that social connection is a fundamental determinant of health and reproductive fitness. In humans, the World Health Organization reports that loneliness is linked to an estimated 871,000 deaths annually and increases the risk of conditions like stroke, heart disease, diabetes, and depression [4]. Conversely, strong social connections can reduce inflammation, lower the risk of serious health problems, and prevent early death [4]. Parallel findings in animal models reveal that different types of social bonds—such as pair bonds, territorial neighbors, and flockmates—can have contrasting, multi-faceted effects on components of reproductive success, including clutch size, laying date, and fledgling number [5]. These relationships are often mediated by mechanisms such as reduced territorial aggression, enhanced information sharing, and cooperative defense [6].

The following sections provide a structured framework for measuring these phenomena, including standardized data collection protocols, analytical models for dynamic network analysis, and key reagent solutions for field research.

Quantitative Data Synthesis

Table 1: Documented Impacts of Social Connectivity on Health and Fitness Metrics

Subject/Species	Social Connection Metric	Health/Fitness Outcome	Effect Size & Notes	Source
Human (Global Population)	Loneliness (Subjective feeling)	All-cause mortality	~871,000 deaths annually; risk comparable to smoking.	[4]
Human (Global Population)	Social Isolation (Objective lack of connections)	Mental Health (Depression)	Twice as likely to develop depression.	[4]
Great Tit (Parus major)	Strength of pairmate bond	Earlier egg laying	Stronger bonds correlated with earlier laying.	[5]
Great Tit (Parus major)	Number of spatial associates	Clutch size	More associates correlated with smaller clutches.	[5]
Great Tit (Parus major)	Overall social connectedness	Number of fledglings	More-social individuals had more fledglings.	[5]
Seasonal Territorial Animals	Neighbor familiarity (Dear-enemy effect)	Territory establishment costs & timing of breeding	Reduces costly aggression, facilitates earlier breeding.	[6]

Table 2: Core Components of Subjective Well-being for Measurement (OECD Guidelines)

Component	Description	Example Measure
Life Evaluation	Reflective, cognitive assessment of one's life.	Satisfaction with life.
Affect	Feelings or emotional states.	Experience of positive/negative emotions.
Eudaimonia	Sense of worth, meaning, and purpose in life.	Sense that activities are worthwhile.

Experimental Protocols

Protocol 1: Dynamic Social Network Analysis using Stochastic Actor-Oriented Models (SAOMs)

Application: For analyzing how social networks and individual traits co-evolve over time and influence fitness outcomes [7].

Workflow:

• Data Collection: Collect longitudinal, time-ordered data on individual associations or interactions across multiple discrete time points (e.g., weeks or seasons). Data should be binary (associated/not associated) for each dyad at each time point [7].
• Model Framework: Employ SAOMs, implemented via the RSiena package in R, to model transitions between network states. These models treat network change as a Markov process, where the future state depends on the current state [7].
• Parameter Inclusion: Include effects at multiple levels:
  • Network Structure: Endogenous effects like transitivity (friends of friends become friends) and reciprocity.
  • Covariates: Individual (e.g., dominance, health status), dyadic (e.g., kinship, spatial proximity), and environmental variables [7].
• Model Fitting & Interpretation: Fit the model to test specific hypotheses about how social networks drive trait changes (e.g., health decline influencing social position) or how traits drive network changes (e.g., dominance influencing connection formation). The output estimates parameters indicating the strength and direction of these effects [7].

Protocol 2: Dissecting Multi-Level Social Bonds and Reproductive Fitness

Application: For testing how different types of dyadic relationships within a multi-level society independently and jointly shape various reproductive traits [5].

Workflow:

• Identify Bond Types: Define and collect data on distinct social layers in the study population. In great tits, this includes pair mates, breeding neighbours, winter flockmates, and spatial associates [5].
• Fitness Data Collection: Record key reproductive metrics for each individual, such as clutch size, egg-laying date, and number of fledglings.
• Spatial Control: Account for the spatial environment (e.g., population density, habitat quality) as it can confound sociality-fitness links [5].
• Statistical Modeling: Use multivariate models (e.g., GLMM) with the multiple bond types as simultaneous predictors and reproductive metrics as separate response variables. This reveals which social layers most strongly affect different fitness components [5].

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for Social Connectivity Research

Item/Reagent	Function/Application	Example/Notes
Unique Animal Tags	Individual identification for constructing longitudinal social networks.	Passive Integrated Transponder (PIT) tags, colored leg bands, or GPS loggers.
Automated Data Loggers	Unobtrusive, continuous monitoring of individual associations at central points.	RFID readers at feeders or waterholes to record co-occurrences [5].
Social Connection Measurement Inventory	Repository of validated tools for quantifying social isolation, loneliness, and connection.	The Foundation for Social Connection's inventory lists 55+ measures with psychometric data [8].
R Package RSiena	Statistical analysis of longitudinal network data using Stochastic Actor-Oriented Models (SAOMs).	Allows modeling of complex co-evolution dynamics between networks and behavioral traits [7].
SILC Intervention Catalog	Resource for identifying existing interventions designed to foster social connection.	Catalog of solutions for addressing social isolation and loneliness; useful for designing experimental manipulations [8].
OECD Well-being Guidelines	Standardized modules for measuring subjective well-being (life evaluation, affect, eudaimonia) in a comparable way.	Provides question wording, answer scales, and survey design for robust data [9].
UF010	UF010, CAS:537672-41-6, MF:C11H15BrN2O, MW:271.15 g/mol	Chemical Reagent
ES9-17	ES9-17, MF:C10H8BrNO2S2, MW:318.2 g/mol	Chemical Reagent

Within the framework of social network analysis in animal behavior research, social environments are not merely backdrops but active evolutionary drivers. The structure of a population—the pattern of connections that dictate interactions and information flow—can fundamentally alter the trajectory of natural selection. This document provides application notes and detailed protocols for studying these selection pressures in networked populations, translating theoretical models from evolutionary graph theory and population genetics into practical experimental methodologies for biological research. The core principle is that network topology can accelerate or suppress the spread of beneficial traits, a effect empirically supported in microbial systems and predicted to influence social behaviors in animal groups [10].

Theoretical Foundation: Evolutionary Driving Forces in Networks

Evolution in a networked population is governed by the interplay of classic evolutionary forces and population structure.

• Evolutionary Driving Forces: The primary forces are mutation, natural selection, genetic drift, and gene flow (migration). Their effect on genetic diversity is modulated by population structure [11].
• Natural Selection in Networks:
  • Positive Selection increases the frequency of advantageous variants, potentially carrying linked, non-advantageous genes to high frequency through "genetic hitchhiking" [11].
  • Negative Selection removes deleterious mutations from the population [11].
  • Balancing Selection maintains multiple alleles, potentially explaining the persistence of certain behavioral polymorphisms in animal societies, akin to its role in immune-related diseases [11].
• The Topology Effect: Contrary to classic population genetics models that predict little effect of structure on fixation probability, Evolutionary Graph Theory (EGT) demonstrates that the specific arrangement of connections among individuals or subpopulations can amplify or suppress selection. This has been demonstrated experimentally, where a star network topology accelerated the spread of a beneficial mutation under low migration rates [10].

Key Experimental Findings and Data Synthesis

Empirical evidence, particularly from microbial systems, provides a controlled validation of theoretical predictions and a model for designing experiments in animal behavior.

Table 1: Quantitative Findings on Topology and Evolutionary Dynamics

Network Topology	Migration Rate	Effect on Spread of Beneficial Mutant	Key Experimental Evidence
Well-Mixed	High (e.g., 20-30%)	Faster spread than structured networks	Pseudomonas aeruginosa ciprofloxacin-resistant mutant reached higher final frequency in well-mixed vs. star topology [10].
Star Network	High (e.g., 20-30%)	Suppressed spread compared to well-mixed	The resistant mutant had a significantly lower final frequency at 30% migration (χ² = 15.348, p < 0.0001) [10].
Star Network	Very Low (e.g., <0.01%)	Amplified spread (transient effect)	Relative frequency of the mutant was significantly higher in the star network on Day 5 (χ² = 13.825, p = 0.0002) [10].
Bidirectional Star	Low	Amplification of selection	Consistent with EGT predictions that certain topologies can increase the fixation probability of beneficial mutations [10].

Detailed Experimental Protocols

The following protocol, adapted from empirical studies on microbial metapopulations, provides a template for investigating topology-driven selection.

Protocol 1: Tracking a Beneficial Trait in a Metapopulation Network

I. Research Objective: To quantify the effect of network topology (e.g., well-mixed vs. star) on the rate of spread and fixation probability of a beneficial mutant.

II. Experimental Workflow:

The following diagram illustrates the core procedural steps of this protocol.

III. Materials and Reagents. Table 2: Research Reagent Solutions

Item	Function/Description	Example/Specification
Model Organism	Subject of study; should be tractable and have a measurable beneficial trait.	Pseudomonas aeruginosa (for antibiotic resistance) [10]. Alternative: social animals with observable behaviors.
Selective Agent	Applies uniform selective pressure across the network, favoring the beneficial mutant.	Sub-inhibitory concentration of ciprofloxacin (e.g., providing ~20% fitness advantage) [10].
Growth Medium	Supports population growth during selection phases.	Standard liquid or solid growth medium (e.g., LB broth for bacteria).
Diluent / Migration Buffer	Medium for serially transferring and diluting populations during dispersal events.	Physiological saline or fresh growth medium without selective agent.

IV. Step-by-Step Procedure.

• Network Construction:

  • Design the physical implementation of the metapopulation network. For a 4-deme star network, use one central "hub" and three satellite "leaf" populations.
  • The well-mixed control should consist of demes with all-to-all connectivity at each dispersal event.

• Founder Population and Inoculation:

  • Grow cultures of the wild-type and isogenic beneficial mutant (e.g., a ciprofloxacin-resistant strain, cipR).
  • Inoculate all demes (hub and leaves) with the wild-type at a standard density.
  • Introduce the beneficial mutant at a low, defined initial frequency (e.g., 1:1000) into a single, predefined location (e.g., one leaf population) to mimic a novel mutation's origin.

• Selection Phase and Serial Transfer:

  • Incubate all populations under defined conditions (e.g., 37°C with shaking) for a fixed growth period (e.g., 1 day).
  • The growth medium in all demes contains the selective agent at a concentration calibrated to provide a known fitness advantage to the mutant.

• Controlled Dispersal:

  • Following the selection phase, implement dispersal according to the network topology and predefined migration rate (m).
  • For a star network with 10% migration:
    • Hub to Leaf: For each leaf, combine a volume of the hub culture representing 10% of the leaf's final volume with 90% of the leaf's own culture.
    • Leaf to Hub: Combine a volume from each leaf culture representing (10%/number of leaves) of the hub's final volume to create the new hub culture.
  • Mix the samples thoroughly before transferring to fresh medium for the next growth cycle.

• Monitoring and Data Collection:

  • Sample each population at every transfer time point.
  • Quantify the frequency of the beneficial mutant in each deme. This can be achieved by plating dilutions on selective and non-selective media or by using PCR-based genotyping techniques.
  • Track the frequency over time (e.g., for 5-6 days, ~35-40 generations) to capture the dynamics before the emergence of secondary mutations.

V. Data Analysis.

• Plot the frequency of the beneficial mutant over time for each network topology.
• Compare the rate of spread (the slope of the frequency curve) between topologies (e.g., star vs. well-mixed) at different migration rates using statistical tests like Chi-squared tests on final frequencies or generalized linear mixed models on the full time series data [10].
• The experiment provides a proxy for fixation probability; a faster rate of spread indicates a higher probability of fixation.

Protocol 2: Analyzing Structural Diversity in Observed Network Populations

For studies involving longitudinal observation of animal social networks, this protocol uses modal network analysis to identify major structural regimes that may impose different selection pressures.

I. Research Objective: To compress a temporal series of observed social networks (e.g., from animal tracking data) into a minimal set of representative "modal" network structures and identify shifts between them.

II. Methodology:

• Data Input: A population of S network samples (adjacency matrices) observed over time on the same fixed set of nodes (individual animals).
• Clustering and Mode Identification: Apply a nonparametric method based on the minimum description length (MDL) principle to automatically:
  • Determine the optimal number of representative networks, K.
  • Assign each observed network sample to one of K clusters.
  • Identify a single, sparse representative ("modal") network for each cluster that best explains the structures of the networks assigned to it [12].
• Interpretation: The transition points between clusters of modal networks represent potential shifts in the social environment. These shifts may correlate with ecological events (resource scarcity, predation pressure) or life-history events (mating seasons, dominance upheavals) and represent changes in the selective landscape.

The Scientist's Toolkit

Table 3: Essential Research Reagents and Computational Tools

Category	Item	Function/Application
Biological Models	Microbial Metapopulations (e.g., P. aeruginosa)	High-replication, controlled testing of evolutionary dynamics in structured populations [10].
	Social Animals (e.g., birds, mammals)	Studying the evolution of behaviors (cooperation, communication) in naturalistic, observable social networks.
Computational & Analytical Tools	Agent-Based Simulations (In silico)	Validating experimental results and exploring parameter spaces (e.g., migration rates, fitness advantages) [10].
	Modal Network Analysis (MDL Principle)	Identifying dominant social structures and regime shifts from longitudinal network data without pre-specifying the number of clusters [12].
	Evolutionary Graph Theory (EGT) Models	Providing a theoretical framework and generating testable predictions about fixation probabilities in different topologies [10].
G-5555	G-5555, MF:C25H25ClN6O3, MW:493.0 g/mol	Chemical Reagent
Maytansinoid B	Maytansinoid B, MF:C36H51ClN4O10, MW:735.3 g/mol	Chemical Reagent

Visualization of Key Concepts

The conceptual relationship between network topology and evolutionary outcome is foundational. The following diagram summarizes the core findings and their relation to theoretical frameworks.

In animal behavior research, the micro-level encompasses the immediate, dyadic interactions between individual animals, such as grooming, aggression, foraging, and communication exchanges [13] [14]. These individual behavioral events form the fundamental building blocks of social structure. In contrast, the macro-level represents the emergent, population-scale patterns that arise from these cumulative interactions, including social organization, information flow pathways, and collective behavior phenomena [14]. This hierarchical relationship—where micro-level processes generate macro-level structures—forms the core investigative focus of social network analysis (SNA) in behavioral ecology [13].

Social network analysis provides both the theoretical framework and methodological toolkit to quantify how repeated local interactions among individuals give rise to global population structures [13] [15]. By mapping these connection patterns, researchers can identify key individuals who disproportionately influence group dynamics, trace potential pathways for disease transmission, and understand how social innovations spread through animal populations [15]. The application of SNA in animal behavior research has revealed valuable insights into the relationship between individual behavior and emergent population-level patterns across diverse species, from primates to ungulates to social insects [13].

Theoretical Framework: Connecting Micro-Behaviors to Macro-Structures

Foundational Concepts in Social Network Analysis

Table 1: Core Social Network Analysis Concepts and Their Applications in Animal Behavior Research

Concept	Definition	Micro-Level Manifestation	Macro-Level Implication
Degree Centrality	Number of direct connections an individual maintains	Frequency of pairwise interactions	Identification of social hubs; potential super-spreaders in disease transmission
Network Density	Proportion of possible connections that actually exist	Rate and diversity of interactions across all possible dyads	Group cohesion and social resilience; higher density facilitates rapid information spread
Betweenness Centrality	Extent to which an individual connects otherwise disconnected groups	Individual movement between social subgroups	Structural bottlenecks or bridges for information/resource flow between communities
Modularity	Degree to which network divides into distinct subgroups	Preferential association within vs between subgroups	Population factionalization affecting cultural differentiation and genetic structuring
Path Length	Average number of steps between any two individuals	Efficiency of direct and indirect information transfer	Overall connectivity and integration of the social system

The theoretical underpinning of micro-macro integration in animal social networks draws from sociological foundations, where macro-level processes approach social life as it exists within broader systems and institutional structures, while micro-level processes focus on interpersonal and interactional exchanges [14]. In animal contexts, these "institutions" represent evolved social conventions and ecological constraints that shape individual decision-making. The network representation chosen by researchers—whether dyadic, bipartite, multigraph, or higher-order networks—fundamentally shapes which micro-macro relationships can be detected and how they are interpreted [13].

Analytical Approaches for Micro-Macro Integration

The methodological challenge in linking micro-interactions to macro-structures lies in developing analytical approaches that accommodate the non-independence of social data—where individual relationships constitute the fundamental unit of analysis rather than independent observations [15]. Advanced SNA methods address this through:

• Multilevel network modeling that simultaneously examines individual attributes, dyadic relationships, and group-level properties
• Temporal network analysis that tracks how micro-level interaction changes propagate to alter macro-structure over time
• Network diffusion models that simulate how behaviors, pathogens, or information spread through observed connection pathways
• Stochastic actor-oriented models that examine how network position influences individual behavior while individual behavior simultaneously shapes network structure [15]

These approaches allow researchers to move beyond simple network description toward causal inference about how specific interaction patterns at the micro-level generate observable macro-structures, and how those emergent structures subsequently constrain or enable future interactions [13] [15].

Application Notes: Practical Implementation in Behavioral Research

Data Collection Protocols for Interaction Mapping

Table 2: Data Collection Methods for Capturing Micro-Level Interactions in Animal Systems

Method	Protocol Description	Best For	Limitations
Focal Animal Sampling	Continuous recording of all interactions of a predetermined individual for a set period	Complete interaction profiles for centrality measures	Labor-intensive; may miss simultaneous interactions
All-Occurrence Sampling	Recording all instances of specific behaviors across entire group	Rare but important behaviors (aggression, copulation)	May overrepresent conspicuous individuals
Proximity Loggers	Automated recording of individuals within predetermined distance	Large groups; cryptic species; continuous data	Equipment cost; distance may not equal interaction
Video Tracking Systems	Automated extraction of movement and interaction from video	High-resolution temporal data; small organisms	Processing complexity; environmental constraints
Genetic Methods	Inferring relationships through kinship analysis	Historical interactions; cryptic species	Indirect measure; may not reflect current interactions

Standardized data collection protocols must be tailored to the research question and species. For whole-network studies—which capture all members of a defined group—researchers must establish clear observational boundaries and consistent operational definitions of interactions [15]. The protocol should specify:

• Definition of social connection: Determine whether interactions will be directed (A grooms B) or undirected (A and B are associated), and weighted (frequency/duration) or unweighted (present/absent) [15]
• Sampling intensity: Establish sufficient observation hours to capture representative interaction patterns while avoiding sampling artifacts
• Inter-observer reliability: Implement calibration sessions to ensure consistent coding across research team members
• Ethological relevance: Ensure defined interactions have biological significance for the study species

For longitudinal studies, maintain consistent protocols across time periods while allowing for necessary adaptations as animal groups change composition. Document all protocol modifications to ensure comparability across sampling periods.

Data Structuring for Network Analysis

Proper data structuring is fundamental for effective network analysis. Tabular data should be organized with rows representing observations and columns representing variables [16]. For network data, this typically involves two primary structures:

• Edge lists: Each row represents a relationship between two individuals, with columns for source, target, and relationship attributes (weight, type, context)
• Attribute data: Each row represents an individual, with columns for individual characteristics (sex, age, rank, personality measures) that may influence network position

Maintain clear data granularity where each row represents a single, non-aggregated observation [16]. This preserves the micro-level interaction data necessary for constructing accurate macro-level networks. Implement unique identifiers for each individual to ensure consistent tracking across observations and analyses [16].

Experimental Protocols: Methodologies for Key Research Questions

Protocol 1: Mapping Information Flow Networks

Objective: Quantify how novel information spreads from individuals to populations through existing social networks.

Materials:

• Focal species with established social groups
• Novel foraging task or puzzle box requiring innovation
• Automated tracking system or video recording equipment
• Social network analysis software (ORA, UCINET, or R packages)

Procedure:

• Baseline network mapping: Conduct pre-experiment observations to construct baseline social networks using association, grooming, or proximity data
• Seeding innovation: Introduce trained demonstrators (or allow spontaneous innovation) at specific network positions determined by pre-calculated centrality measures
• Diffusion tracking: Record the spread of behavioral innovation through the group with continuous monitoring
• Network comparison: Compare pre- and post-diffusion networks to identify structural changes resulting from information spread
• Pathway analysis: Use network metrics (degree, betweenness, closeness centrality) to predict adoption timing and spread patterns

Analysis:

• Calculate diffusion rate as number of new adopters per time unit
• Map transmission pathways using network connection data
• Model social influence using temporal autocorrelation methods
• Test for social learning using network-based diffusion analysis

Information Diffusion Network

Protocol 2: Quantifying Social Structure Stability

Objective: Measure how micro-level interaction patterns maintain or transform macro-level social structure across temporal scales.

Materials:

• Longitudinal observational data from multiple time periods
• Social network analysis software with temporal analysis capabilities
• Environmental monitoring equipment to record ecological covariates

Procedure:

• Temporal sampling: Collect network data across multiple time points (daily, seasonally, annually) using consistent protocols
• Network construction: Build separate networks for each time period using identical interaction criteria
• Stability metrics: Calculate network-level stability measures (Matrix Correspondence, Mantel tests) between consecutive time periods
• Node-level persistence: Measure individual consistency in network position across time
• Environmental correlation: Test associations between environmental variables and network stability metrics

Analysis:

• QAP regression to test environmental effects on network structure
• Temporal autocorrelation of individual network positions
• Stochastic actor-oriented models to identify micro-level rules driving structural change
• Network visualization to illustrate structural persistence or reorganization

The Scientist's Toolkit: Essential Research Reagents and Solutions

Table 3: Essential Research Materials for Animal Social Network Analysis

Category	Specific Tool/Solution	Function	Application Notes
Data Collection	Automated tracking system (e.g., RFID, GPS)	Continuous proximity/association data	Essential for large groups; provides high-resolution temporal data
Data Collection	Behavioral coding software (e.g., BORIS, Observer XT)	Standardized behavioral quantification	Enables reliable inter-observer reliability; facilitates complex ethograms
Network Analysis	SNA software packages (e.g., UCINET, ORA, R packages)	Network construction, visualization, and metric calculation	ORA preferred for dynamic networks; R provides greater analytical flexibility
Statistical Analysis	Specialized network tests (MRQAP, ERGM, SABM)	Hypothesis testing with non-independent data	Accounts for network autocorrelation; essential for valid inference
Visualization	Network graphing tools (e.g., NetDraw, Gephi, igraph)	Visual representation of social structure	Critical for pattern detection and result communication
Mps1-IN-4	Mps1-IN-4, MF:C26H31F3N6O2, MW:516.6 g/mol	Chemical Reagent	Bench Chemicals
CC-885	CC-885, CAS:1010100-07-8, MF:C22H21ClN4O4, MW:440.9 g/mol	Chemical Reagent	Bench Chemicals

Analytical Framework: From Raw Data to Ecological Interpretation

Data Processing Workflow

The transformation of raw behavioral observations into quantitative network metrics follows a structured pipeline:

• Data cleaning: Remove observational artifacts and ensure consistent individual identification
• Matrix construction: Convert interaction observations into adjacency matrices (individuals × individuals) with connection weights
• Network visualization: Create initial graphs to identify obvious patterns and potential data issues
• Metric calculation: Compute relevant network metrics at individual, subgroup, and whole-network levels
• Statistical testing: Implement appropriate null models to identify significantly non-random patterns
• Ecological interpretation: Relate network patterns to biological processes and ecological contexts

Analytical Workflow

Statistical Considerations for Network Data

The non-independent nature of network data requires specialized analytical approaches. Standard statistical tests that assume independence of observations produce inflated Type I errors when applied to network data [15]. Appropriate methods include:

• Multiple Regression Quadratic Assignment Procedure (MRQAP): Permutation-based method that maintains network structure while testing attribute relationships
• Exponential Random Graph Models (ERGMs): Probability models that test whether observed network structures occur more frequently than expected by chance
• Stochastic Actor-Oriented Models (SAOMs): Longitudinal models that examine co-evolution of networks and individual attributes
• Network Cross-Correlation Functions: Temporal methods that identify lead-lag relationships in dynamic networks

Model selection should be guided by research question, data structure, and underlying assumptions. Each method has specific requirements regarding network size, distribution, and missing data that must be addressed during study design and data collection phases.

Interpretation Guidelines: Bridging Micro-Macro Divides

Effective interpretation of animal social network analyses requires careful consideration of the relationship between micro-level mechanisms and macro-level patterns. Researchers should:

• Distinguish structure from function: Similar network structures can emerge from different interaction processes and serve different functions
• Consider multiple temporal scales: Micro-level interactions may have delayed effects on macro-structure that are not immediately apparent
• Account for ecological context: The same interaction pattern may have different consequences in different environments
• Acknowledge methodological constraints: Sampling intensity, operational definitions, and analytical choices all influence detected patterns

The explanatory power of social network analysis in animal behavior lies in its capacity to reveal how simple, local interaction rules generate complex population-level phenomena—and how those emergent structures subsequently feed back to shape individual behavioral opportunities and constraints [13] [14]. This iterative relationship between micro-level agency and macro-level structure represents the central theoretical contribution of network approaches to behavioral ecology.

Social network analysis (SNA) has become a fundamental tool in infectious disease ecology for quantifying contact patterns among individuals that influence pathogen spread [17]. In a network approach, the epidemiological units of infection (e.g., individuals, herds, farms) are defined as nodes, and these are inter-linked according to who is in contact with whom, where contact is assumed to represent transmission opportunities between two nodes [17]. Theoretical work has repeatedly demonstrated that incorporating contact pattern heterogeneity into epidemiological models can substantially alter model predictions, and empirical studies confirm that network connectivity influences an individual's risk of acquiring an infection [17].

A foundational question that often goes unaddressed is how to determine whether an observed pattern of cases is consistent with pathogen spread through an observed network presumed to represent potential transmission pathways [17]. Such validation is critical for developing accurate predictive models of pathogen spread [17]. These dynamic networks are fundamental aspects of an animal's environment, creating selection on behaviors and other traits, with applications ranging from disease epidemiology to the dynamics of group formation [7].

Key Analytical Frameworks and Statistical Tests

The Network k-Test for Epidemiologic Relevance

The network k-test is a novel permutation-based procedure designed to determine whether an observed contact network has epidemiologic relevance for a specific pathogen [17].

• Procedure: The k-statistic is defined as the mean number of cases observed to occur within one step (i.e., direct contacts) of an infected case in the network [17].
• Significance Testing: The observed k-statistic is compared to a permuted distribution of k-statistics generated by randomly re-allocating case locations within the network (node-label swapping) [17].
• Interpretation: A significantly high k-statistic (p < 0.05) suggests that the occurrence of cases results from pathogen propagation through network links, indicating the network's epidemiologic relevance [17].

This method is particularly powerful because it considers the global clustering pattern of cases within the network and is robust to missing data and a lack of temporal information [17].

Stochastic Actor-Oriented Models (SAOMs) for Dynamic Networks

While many ecological network analyses are static, SAOMs provide a framework for analyzing network dynamics [7].

• Purpose: SAOMs model gradual changes in networks and individual traits across discrete time points using hidden Markov models, allowing researchers to examine how social and non-social processes drive each other [7].
• Data Structure: These models typically use binary network data (associating or not) at each time point, with the duration of time periods determined by the study system and research questions [7].
• Application: SAOMs are individual-based models that can include effects and covariates for individuals, dyads, and populations (constant or variable), making them suitable for a wide range of ecological and evolutionary questions [7].

Table 1: Comparison of Statistical Methods for Analyzing Contact Networks in Disease Ecology

Method	Primary Function	Data Requirements	Key Strengths	Key Limitations
Network k-test [17]	Tests if case pattern is consistent with transmission through observed network	Static network data, infection status	Robust to missing data (up to 50%); does not require temporal data; accounts for global clustering.	Does not model the direction or rate of spread.
Stochastic Actor-Oriented Models (SAOMs) [7]	Models network and trait co-evolution over time	Longitudinal network data across discrete time points	Allows inference of causal drivers; models complex co-evolutionary processes.	Requires high-resolution longitudinal data; performance with uncertain relationships is unclear.
Degree Comparison (e.g., Kruskal-Wallis) [17]	Compares connectivity of infected vs. uninfected nodes	Static network data, infection status	Simple, intuitive, and commonly used.	Cannot account for global clustering; high centrality may correlate with other susceptibility factors.
Logistic Regression [17]	Tests if node degree predicts infection status	Static network data, infection status	Provides a familiar statistical framework and effect sizes.	Suffers from non-independence of network data; distills network into individual-level measures.

Application Notes and Experimental Protocols

Protocol 1: Implementing the Network k-Test

This protocol evaluates the epidemiologic relevance of an observed social contact network for a specific pathogen using the network k-test.

1. Research Question and Hypothesis Formulation

• Question: Is the observed pattern of infected cases clustered within the observed social contact network, consistent with pathogen transmission via these contacts?
• Hypothesis: The mean number of cases within one network step of a case (k-statistic) will be significantly greater than expected if cases were randomly distributed.

2. Data Requirements and Preparation

• Network Data: An undirected social contact matrix where nodes represent individuals and edges represent potential transmission contacts.
• Epidemiological Data: A binary vector indicating the infection status (case/non-case) for each node in the network.
• Data Integration: Align network nodes and epidemiological data so that each individual has a corresponding node ID, infection status, and list of contacts.

3. Computational Procedure

• Step 1: Calculate Observed k-statistic
  • For every infected node, count the number of its direct contacts that are also infected.
  • Sum these counts across all infected nodes.
  • Divide this sum by the total number of infected nodes to obtain the observed k-statistic [17].
• Step 2: Generate Permuted Null Distribution
  • Randomly reassign the "case" status among all nodes in the network, preserving the total number of cases.
  • For each permutation, re-calculate the k-statistic using the procedure in Step 1.
  • Repeat this process a large number of times (e.g., 1,000-10,000 permutations) to build a null distribution of k-statistics expected under random case distribution [17].
• Step 3: Calculate Statistical Significance
  • Compare the observed k-statistic to the permuted null distribution.
  • The p-value is the proportion of permutations that produced a k-statistic greater than or equal to the observed value [17].
  • A p-value < 0.05 indicates the network has epidemiologic relevance for the pathogen.

4. Interpretation and Reporting

• A significant result supports the hypothesis that the observed contact network represents viable pathogen transmission pathways.
• Report the observed k-statistic, the mean of the permuted distribution, the p-value, and the number of permutations performed.
• Visualize the network with cases highlighted to illustrate the clustering.

Protocol 2: Applying Stochastic Actor-Oriented Models (SAOMs)

This protocol outlines the steps for using SAOMs to analyze the co-evolution of social networks and disease status over time.

1. Research Question and Hypothesis Formulation

• Question: Do current social connections predict future infection status, or does current infection status predict changes in social connections?
• Hypothesis: Network structure and node attributes (like infection status) co-evolve, with each influencing the other over time.

2. Data Requirements and Preparation

• Longitudinal Network Data: A series of network observations (e.g., association data) collected at multiple discrete time points [7].
• Node Covariates: Time-varying or constant attributes of nodes (e.g., infection status, age, sex, dominance rank) [7].
• Data Structuring: Organize data into a panel format where the network and all node attributes are recorded at each observation time point.

3. Model Specification and Fitting with RSiena

• Step 1: Define the Dependent Variable
  • Specify the network that is changing over time as the primary dependent variable [7].
• Step 2: Include Network Structural Effects
  • Model parameters that capture endogenous network formation processes (e.g., density, reciprocity, transitivity) [7].
• Step 3: Include Covariate Effects
  • Infection Dynamics: Model infection status as a dependent behavior variable, influenced by network position and other covariates.
  • Network Dynamics: Include infection status as a node covariate affecting network formation (e.g., "infected actors change their socializing behavior").
• Step 4: Model Estimation
  • Use the RSiena package in R to estimate model parameters via Method of Moments [7].
  • Assess model convergence and goodness-of-fit.

4. Interpretation and Reporting

• Interpret significant parameters in the context of the research question.
  • A positive effect of infection status on network formation suggests the attribute influences social ties.
  • A positive effect of network position on infection dynamics suggests the network influences disease spread.
• Report parameter estimates, standard errors, and convergence t-ratios.

Table 2: Power and Robustness of the Network k-test Across Different Scenarios [17]

Scenario	Network Type	Pathogen Infectiousness	Prevalence	Power of k-test	Power of Degree Comparison
Baseline	Bernoulli	Moderate (β=0.04)	25%	High	Lower
Network Structure	Scale-free	Moderate (β=0.04)	25%	High	Lower
	Small-world	Moderate (β=0.04)	25%	High	Lower
	Modular	Moderate (β=0.04)	25%	High	Lower
Pathogen Transmissibility	Bernoulli	High (β=0.133)	25%	Very High	Lower
Epidemic Size	Bernoulli	Moderate (β=0.04)	5%	Moderate	Low
	Bernoulli	Moderate (β=0.04)	50%	High	Lower
Missing Data	Various	Moderate (β=0.04)	25%	Remains High (even with 50% missing)	Decreases

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Analytical Tools for Social Network Analysis in Disease Ecology

Tool / Reagent	Type	Primary Function	Application Example
R Statistical Software	Software Environment	Provides a comprehensive platform for statistical computing and graphics.	Core platform for implementing all analytical methods described.
igraph Package (R)	Software Library	Network analysis and visualization. Generation of theoretical network structures (Bernoulli, modular, etc.) [17].	Constructing and visualizing empirical contact networks; simulating network structures for power analyses [17].
RSiena Package (R) [7]	Software Library	Statistical analysis of longitudinal network data using Stochastic Actor-Oriented Models (SAOMs).	Modeling the co-evolution of animal contact networks and infection status over time [7].
Permutation Test Algorithm	Computational Method	Generates null distributions by randomizing data under a specific hypothesis.	Implementing the network k-test to assess the epidemiologic relevance of a contact network [17].
High-Resolution Tracking Data	Primary Data	Raw data on individual locations or interactions (e.g., GPS, proximity loggers).	Building the empirical contact networks used as input for k-tests or SAOMs.
Diagnostic Assays	Laboratory Reagent	Determines the infection status (case/non-case) of each individual in the network.	Providing the binary case status data required for the network k-test and as a covariate in SAOMs [17].
A2ti-2	A2ti-2, MF:C18H18N4O2S, MW:354.4 g/mol	Chemical Reagent	Bench Chemicals
BI-0474	BI-0474, MF:C30H37N9O2S, MW:587.7 g/mol	Chemical Reagent	Bench Chemicals

Visualization and Data Presentation Guidelines

Effective data presentation is critical for communicating complex network relationships. Adhere to the following guidelines for creating accessible visualizations.

• Color Contrast: All diagrams and figures must meet WCAG 2.1 AA minimum contrast ratios. For graphical objects and large text, a minimum contrast ratio of 3:1 is required. For standard text, a ratio of 4.5:1 is required [18] [19]. The provided color palette (#4285F4, #EA4335, #FBBC05, #34A853, #FFFFFF, #F1F3F4, #202124, #5F6368) is designed to facilitate this.
• Graphs vs. Tables: Use graphs when the message is in the shape or to reveal relationships among multiple values. Use tables for looking up individual values or when presenting precise values that involve multiple units of measure [20].
• Quantitative Data Encoding: Encode quantitative values using points, lines, or bars. For categorical subdivisions, use position or color, limiting the number of different hues to a maximum of eight for preattentive distinction [20].

From Data to Dynamics: Advanced Methodologies for Mapping Animal Social Interactions

The integration of advanced sensor technologies and artificial intelligence is revolutionizing the collection of behavioral data for animal social network analysis (SNA). These automated methods enable researchers to gather high-resolution, quantitative datasets on social interactions at scales and precision previously unattainable through manual observation [21] [22]. By capturing continuous, objective data on animal movements, identities, proximities, and interactions, these tools provide the foundational data required to construct robust and dynamic social networks, offering unprecedented insights into the ecological and evolutionary processes governing animal societies [13] [7].

The table below summarizes the primary sensor technologies employed in automated animal behavior monitoring, detailing their key applications in social network research.

Table 1: Sensor Technologies for Automated Animal Tracking and Social Behavior Analysis

Technology	Primary Data Collected	Key Applications in SNA	Considerations
Computer Vision (CV)	Animal position, pose, trajectory, and appearance from video [21] [22].	Proximity networks, interaction detection (e.g., grooming, aggression), group movement analysis [13] [23].	Requires clear line-of-sight; processing can be computationally intensive; performance depends on lighting and occlusion [21].
Wearable Inertial Sensors	Tri-axial acceleration, rotation, and movement dynamics via accelerometers and gyroscopes [24] [22].	Classifying specific behaviors (e.g., grazing, running), activity budgets, and inferring social states like estrus or lameness [22] [25].	Requires animal handling for attachment; potential for device loss; data from multiple individuals must be synchronized [24].
RFID	Unique animal identity at specific locations or feeders [24].	Constructing association networks based on co-occurrence at specific resources (e.g., feeders, watering holes) [7].	Provides data only at fixed reader points, offering sparse spatial tracks compared to CV.
Bioacoustic Sensors	Audible and infrasonic vocalizations [26].	Identifying call types for communication network analysis, tracking species presence, and detecting threats like gunshots [26].	Analysis complicated by background noise; requires sophisticated machine learning models for call discrimination [26].

Experimental Protocols for Social Network Data Collection

Protocol: Computer Vision-Based Proximity and Interaction Networks

Objective: To automatically construct dynamic social networks based on spatial proximity and specific interactions from video data.

Materials:

• JAX Animal Behavior System (JABS) or equivalent standardized hardware [23].
• High-resolution cameras (minimum 30 fps) with appropriate field of view.
• Computational hardware (GPU recommended) for video processing.
• Software: DeepLabCut, SLEAP, or SimBA for pose estimation [23].

Procedure:

• Video Acquisition: Record subjects in their enclosure using a calibrated camera setup. Ensure uniform lighting and minimize visual obstructions to optimize tracking accuracy [21] [23].
• Animal Detection and Pose Estimation: Process video frames using a pre-trained pose estimation model (e.g., DeepLabCut, SLEAP) to identify key body parts (e.g., nose, ears, tail base, limbs) for each individual [23].
• Tracking and Identity Maintenance: Link detected poses across frames to create continuous trajectories for each animal, maintaining unique identities throughout the recording session.
• Proximity and Interaction Annotation:
  • Proximity: Calculate the centroid position for each animal and compute pairwise distances between all individuals in each frame. Define a social association (an "edge") when the distance between two animals is below a pre-defined threshold (e.g., 2 body lengths) [13].
  • Interaction: Use a specialized classifier, trained in software like JABS-AL or SimBA, to identify specific social behaviors such as grooming, mounting, or nose-to-nose contact from the sequence of poses [23].
• Network Construction: For a given time window, create an adjacency matrix where nodes represent individuals and edges represent the frequency or duration of proximity/interactions. This matrix is the raw input for social network analysis [13] [7].

Protocol: Multi-Sensor Fusion for Holistic Behavioral Phenotyping

Objective: To integrate data from wearable accelerometers and RFID systems to link individual activity states with social association patterns.

Materials:

• Tri-axial accelerometer tags.
• RFID tags and stationary readers.
• Data synchronization unit.
• Machine learning platform for sensor fusion (e.g., Python with scikit-learn).

Procedure:

• Sensor Deployment: Fit each study animal with an accelerometer tag. Install RFID readers at strategic, resource-limited locations (e.g., feeding stations, water sources, nesting areas) [24].
• Data Collection and Synchronization: Collect continuous accelerometer data and timestamped RFID read events. Ensure all data streams are synchronized to a common clock.
• Behavioral Classification from Accelerometry: Extract features (e.g., mean amplitude, variance, signal entropy) from raw accelerometer data. Use a pre-trained machine learning model (e.g., Random Forest) to classify the data into discrete behaviors such as feeding, resting, walking, or ruminating [24] [22].
• Association Events from RFID: Define a co-occurrence event when two or more animals are detected by the same RFID reader within a short time window (e.g., 2 minutes).
• Sensor Fusion and Analysis: Fuse the classified behavior and association data to answer complex questions. For instance, analyze whether individuals exhibiting lethargic behavior (from accelerometer data) also show reduced social associations (from RFID data), which could indicate sickness. This fusion can occur at the feature level (combining data before classification) or decision level (combining results from separate classifiers) [24].

Protocol: Dynamic Social Network Analysis using Stochastic Actor-Oriented Models

Objective: To model and understand how social networks change over time and how individual traits influence network evolution.

Materials:

• Longitudinal social network data (multiple observation periods).
• R statistical software with RSiena package [7].
• Covariate data on individuals (e.g., sex, dominance rank, health status).

Procedure:

• Data Structuring: Organize your observed social networks (e.g., from Protocol 1 or 2) into a sequence of two or more discrete time points (e.g., networks observed on day 1, day 2, etc.) [7].
• Model Specification: In RSiena, specify a Stochastic Actor-Oriented Model (SAOM) that includes:
  • Structural Effects: Parameters that model how the network structure itself influences change (e.g., transitivity, the tendency to form ties with "friends of friends").
  • Covariate Effects: Parameters that test how individual attributes (e.g., dominance) influence the rate of change in ties or the tendency to form/retain ties.
  • Network-Behavior Co-evolution: If tracking a behavioral trait over time, the model can test how the network influences the behavior and vice versa [7].
• Model Estimation: Use the siena07 function in RSiena to estimate the parameters of the model, which represent the strength and direction of the various social forces driving network change.
• Model Assessment and Interpretation: Check the model for convergence and goodness-of-fit. Interpret the significant parameters to draw conclusions about the social processes that best explain the observed changes in the animal social network over time [7].

The Scientist's Toolkit: Key Research Reagents and Materials

Table 2: Essential Research Reagents and Solutions for Automated Behavioral Monitoring

Item	Function/Application
JAX Animal Behavior System	An open-source, integrated platform providing standardized hardware designs and software for data acquisition, behavior annotation, and classifier sharing, specifically validated on diverse mouse strains [23].
DeepLabCut/SLEAP	Open-source software toolkits for markerless pose estimation of animals based on deep learning. They allow researchers to train models to track user-defined body parts from video data [23].
SimBA	Open-source software for classifying defined behaviors from pose estimation data. It provides a graphical user interface for creating supervised machine learning classifiers [23].
Tri-axial Accelerometer Tag	A wearable sensor that measures acceleration in three spatial dimensions. The resulting data stream is used to infer body movement, posture, and specific behaviors [24] [22].
Passive Integrated Transponder System	A system comprising implanted or attached RFID tags and stationary readers. It is used for unambiguous individual identification and logging visits to specific locations [24].
RSiena Software Package	A statistical package for R used to analyze the evolution of social networks using Stochastic Actor-Oriented Models (SAOMs), allowing for the test of hypotheses about dynamic network processes [7].
Cleminorexton	Cleminorexton, CAS:2980518-93-0, MF:C24H24F4N2O4S, MW:512.5 g/mol
Cytosporin A	Cytosporin A, MF:C17H24O5, MW:308.4 g/mol

Workflow and Data Integration Visualizations

Computer Vision Workflow for Social Network Analysis

The diagram below illustrates the end-to-end pipeline for deriving social networks from video data using computer vision.

Multi-Level Sensor Fusion Architecture

This diagram outlines the three primary levels of sensor fusion for integrating data from multiple sources, such as accelerometers and RFID readers.

The protocols and technologies outlined provide a robust framework for integrating automated sensor data collection with social network analysis. This synergy allows researchers to move beyond static network snapshots to model the dynamic processes that shape animal societies [7]. As these technologies continue to mature, focusing on standardization, data sharing, and multidisciplinary collaboration will be key to unlocking their full potential in advancing our understanding of animal behavior, with applications spanning from fundamental ecology to conservation and precision livestock farming [21] [24] [22].

Stochastic Actor-Oriented Models (SAOMs) represent a class of individual-based models designed to analyze changes in social networks over discrete time points, making them particularly valuable for ecological and evolutionary studies [7]. In animal behavior research, where social relationships fundamentally influence fitness, disease transmission, information spread, and competition, SAOMs provide a dynamic framework that moves beyond static network analysis [7]. These models treat social networks as dynamically changing environments that create selection pressures on behaviors and other traits, allowing researchers to examine how social and non-social processes drive each other and what processes govern the development of network structure [7].

The fundamental limitation SAOMs address is the "snapshot problem" in social network analysis. Traditional static networks summarize relationships over a period, ignoring how individuals change interaction patterns over time and making causal inference difficult [7]. For instance, when an infected individual shows different social behavior, static analysis cannot determine whether the infection caused behavioral changes or whether pre-existing behavior patterns caused the infection [7]. SAOMs incorporate time-ordering into analyses, providing stronger evidence for potential causal pathways by observing how processes consistently precede and lead to changes in other processes [7].

Core Principles of Stochastic Actor-Oriented Models

Foundational Assumptions

SAOMs operate under several key assumptions that define their application scope and interpretation [7] [27]:

• Time as Continuous Between Observations: While data are collected at discrete panel waves (time points), the model assumes unobserved "mini-steps" occur between observations, explaining how network structures emerge gradually [27].

• Markov Process: The network state at time t depends only on the state at time t-1, with no direct influence from states further in the past (e.g., t-2) [27].

• Actor-Oriented Decisions: The model assumes that individuals (actors) control their outgoing ties and make changes to optimize their position based on network structure and covariates [27].

• One Tie Change at a Time: The network evolves through sequential micro-steps where only one outgoing tie changes at any moment, making the process computationally tractable [27].

• Homogeneous Objective Function: All actors share the same fundamental propensity for tie changes, with heterogeneity introduced through network effects and covariates [27].

Mathematical and Conceptual Framework

SAOMs model network evolution as a Markov chain, where individuals are presented with opportunities to change their outgoing ties based on rate functions and then make decisions based on objective functions and evaluation functions [7]. The model decomposes into two fundamental components:

• Rate Function: Determines how frequently each actor gets opportunities to change their outgoing ties
• Objective Function: Evaluates how attractive different network configurations are to each actor

The objective function incorporates effects representing social mechanisms (transitivity, reciprocity), actor attributes (sex, age, dominance), and environmental factors, weighted by parameters estimated from the data [7].

Table: Core Components of the SAOM Framework

Component	Mathematical Representation	Biological Interpretation
Rate Function	λ_i(ρ)	Determines how often individual i gets to change their ties; can depend on covariates
Objective Function	fi(β,x) = Σk βk s{ik}(x)	Evaluates network configuration attractiveness for individual i based on weighted effects
Evaluation Function	-	Determines probability of specific tie changes based on objective function
Endogenous Effects	s_{ik}(x)	Network structural effects (transitivity, reciprocity) influencing tie formation
Exogenous Effects	s_{ik}(x)	Individual, dyadic, or environmental covariates affecting network evolution

SAOM Application Protocol for Animal Social Networks

Data Requirements and Preparation

Implementing SAOMs requires specific data structures and preparation steps [7] [27]:

• Longitudinal Network Data: Networks must be recorded at multiple discrete time points, with the duration between observations determined by the biological process under study and data resolution [7].

• Binary Network Representation: Networks are typically represented as binary (association/no association) at each time point, though valued networks can be handled with modifications [7].

• Covariate Integration: The model can incorporate individual covariates (sex, dominance rank), dyadic covariates (spatial proximity, kinship), and environmental variables [7].

• Data Aggregation: For event-based data (specific interactions), researchers must aggregate events into states (e.g., "individuals A and B associated 4 out of 7 days this week") [7].

Table: Data Preparation Steps for SAOM Implementation

Step	Procedure	Considerations for Animal Systems
Time Scale Selection	Determine appropriate intervals between network observations	Balance behavioral relevance (e.g., daily cycles) with practical constraints
Network Construction	Create adjacency matrices for each time point	Define association criteria appropriate for species and research questions
Covariate Compilation	Organize individual, dyadic, and environmental variables	Ensure temporal alignment of covariates with network observations
Data Formatting	Structure data for RSiena input	Create network and covariate objects with consistent ordering of individuals

Model Specification and Implementation

The following diagram illustrates the complete SAOM workflow from data preparation to interpretation:

Step-by-Step Analytical Procedure

• Data Import and Formatting

  • Load network data for each time point into RSiena format
  • Specify and scale covariates appropriately
  • Create RSiena data objects using sienaDataCreate()

• Model Specification

  • Specify effects based on biological hypotheses
  • Include structural effects (transitivity, reciprocity)
  • Include covariate-based effects (homophily, sender, receiver)

• Parameter Estimation

  • Use method of moments estimation procedure
  • Run simulations to find parameter estimates
  • Check convergence statistics and overall maximum convergence ratio

• Model Assessment

  • Evaluate goodness of fit using simulation-based methods
  • Compare nested models using score-type tests or likelihood ratio tests
  • Check for potential degeneracy issues

• Interpretation

  • Interpret significant parameters in biological context
  • Calculate predicted probabilities for specific effects
  • Consider limitations and assumptions

Essential Effects and Biological Interpretations

Structural Effects

Structural effects capture how network topology itself influences its evolution, representing social preferences and constraints [7]:

• Density/Outdegree: Basic propensity to form ties, controlling for other effects
• Reciprocity: Tendency to form mutual connections
• Transitivity: Preference for forming triangles (friends of friends become friends)
• Popularity: Tendency for some individuals to receive more ties
• Activity: Tendency for some individuals to send more ties

Table: Key Structural Effects in SAOMs for Animal Behavior

Effect	RSiena Term	Biological Interpretation	Research Application
Reciprocity	recip	Mutualism, cooperative investment	Testing for reciprocal altruism in grooming networks
Transitive Triplets	transTrip	Closure of triads, alliance formation	Investigating hierarchical stability in primate groups
Three-Cycles	cycle3	Generalized exchange systems	Measuring cyclic dominance in conflict networks
Popularity (sqrt)	inPopSqrt	Preferential attachment, status	Modeling disease spread through highly-connected individuals
Activity (sqrt)	outActSqrt	Individual variation in sociality	Understanding information flow in animal collectives

Covariate Effects

Covariate effects examine how individual or dyadic characteristics influence network dynamics [7]:

• Ego, Alter, and Similarity Effects: How individual attributes affect tie formation, maintenance, and dissolution
• Dyadic Covariates: How external relationships (kinship, spatial proximity) affect social ties

The following diagram illustrates how different effects operate within the SAOM framework:

Essential Software and Analytical Tools

Table: Research Reagent Solutions for SAOM Implementation

Tool/Resource	Function	Implementation Notes
RSiena	Primary R package for SAOM estimation	Requires installation from R-Forge; comprehensive documentation available
Statnet Suite	Alternative ERGM/TERGM implementation	Useful for model comparison and specific extensions
RStudio	Development environment for analysis	Facilitates script management and visualization
asnipe	Animal Social Network Inference and Permutations	Useful for preliminary network construction and data preparation
UCINET	Social network analysis software	Alternative for basic network descriptive statistics

Data Collection and Management Solutions

• Automated Tracking Systems: GPS, proximity loggers, and computer vision systems for high-resolution behavioral data [7]
• Field Data Management: Standardized protocols for observational data collection across multiple time points
• Data Processing Pipelines: Custom scripts for aggregating raw interactions into network formats suitable for SAOMs

Advanced Applications and Future Directions

Coevolution Models

SAOMs can be extended to model coevolutionary processes where networks and individual characteristics mutually influence each other [7] [27]. This is particularly relevant for animal behavior research questions such as:

• How dominance hierarchies influence and are influenced by association patterns
• How behavioral syndromes (personalities) affect and are affected by social position
• How health status and social connectivity interact over time

Multi-Group and Comparative Analyses

The RSiena framework allows for multi-group analyses, enabling comparative studies across [27]:

• Different populations of the same species
• Different seasons or ecological conditions
• Experimental treatment and control groups
• Different age or sex classes within populations

Integration with Other Analytical Frameworks

SAOMs can be combined with complementary approaches to provide more comprehensive insights:

• Network-Based Diffusion Analysis: For studying social transmission of behaviors [7]
• Multi-Model Inference: Comparing SAOMs with alternative dynamic network models
• Hybrid Approaches: Integrating SAOMs with spatial or phylogenetic analyses

Critical Considerations and Limitations

Data Requirements and Quality

SAOMs require high-resolution longitudinal data, which may not be feasible in all study systems [7]. Missing data and uncertainty around social relationships remain challenging issues that require careful consideration during analysis [7].

Model Assumptions and Biological Realism

Researchers must critically evaluate whether SAOM assumptions align with their biological system, particularly regarding [7] [27]:

• The Markov assumption (lack of longer-term memory)
• Homogeneity in objective functions across individuals
• The actor-oriented decision framework
• Temporal scale of mini-steps between observations

Computational and Statistical Challenges

SAOM estimation can be computationally intensive, particularly for large networks or complex effect specifications. Convergence issues may arise with certain network structures or effect combinations, requiring careful model specification and diagnostic checking.

Social Network Analysis (SNA) has emerged as a powerful, quantitative framework for investigating the social structures of animal populations by characterizing relationships as networks of nodes (individual animals) connected by ties (social interactions or associations) [28]. This approach allows researchers to move beyond dyadic relationships to understand complex, population-level patterns. In behavioral ecology, SNA has proven invaluable for linking individual behavior to emergent social structures and quantifying the implications of these structures for critical outcomes such as disease transmission, information diffusion, and fitness [13] [29].

The application of SNA enables the quantification of social roles through precise metrics that capture an individual's position and importance within the broader social matrix. These metrics—including centrality, betweenness, and brokerage measures—provide objective tools for identifying keystone individuals, understanding social dynamics, and predicting behavioral transmission across animal populations [30] [28].

Core Metric Definitions and Biological Significance

Table of Key SNA Metrics and Their Interpretations in Animal Behavior

Metric Name	Definition	Biological Interpretation	Data Requirements
Degree Centrality	Number of direct connections an individual has [28].	Measures social connectedness or gregariousness; high values may indicate popular or socially active individuals [30].	Interaction data (grooming, aggression, proximity).
Betweenness Centrality	Number of shortest paths between all other individuals that pass through a given node [28].	Identifies potential information brokers or bridges between subgroups; individuals who connect otherwise separate parts of the network [30].	Complete network data on all possible connections.
Eigenvector Centrality	Measure of an individual's connection to well-connected others [28].	Reflects social influence; individuals connected to other central individuals have higher status or importance [29].	Network data with reciprocal or directional interactions.
Clustering Coefficient	Likelihood that two associates of a node are associates themselves [28].	Quantifies clique formation or subgroup cohesion; measures local transitivity in social relationships [28].	Triadic interaction data (A-B, B-C, A-C).
Bridge	An individual whose weak ties fill a structural hole between clusters [28].	Individuals providing the only connection between subgroups; critical for network cohesion and information flow [30].	Data on weak vs. strong ties across subgroups.

Advanced Brokerage Concepts in Animal Societies

Beyond basic centrality metrics, brokerage typologies provide nuanced understanding of how individuals mediate relationships within social networks. In animal societies, brokers occupy strategic positions that influence information flow, resource access, and social stability [30]. Five distinct brokerage roles have been identified:

• Coordinators mediate between individuals within their own subgroup, facilitating internal cohesion.
• Gatekeepers control flow from outside into their subgroup.
• Representatives speak for their subgroup to the outside.
• Consultants are outsiders who mediate between members of a subgroup they don't belong to.
• Liaisons connect two different subgroups without belonging to either [30].

Recent research in dynamic sow herds revealed that the most connected individuals predominantly engaged in coordinating behavior, demonstrating a clear relationship between overall connectedness and brokering type [30].

Experimental Protocols for SNA in Animal Behavior Research

Data Collection Methodologies

Protocol 1: Behavioral Sampling for Network Construction

Objective: To systematically collect behavioral interaction data for constructing social networks.

Materials:

• Video recording equipment (5 cameras recommended for full coverage) [30]
• Individual identification system (color-coded markings, ear tags) [30]
• Data logging software or standardized data sheets
• Timing device

Procedure:

• Individual Identification: Mark all subjects with unique, visible identifiers using non-toxic coloring or physical tags [30].
• Sampling Method Selection:
  • Focal Animal Sampling (FAS): Observe single individuals for predetermined periods, recording all social interactions [29]. Suitable for detailed data on individual behavior patterns.
  • All-Occurrences Behavior Sampling (ABS): Record all instances of specific behaviors within the entire group during observation periods [29]. Optimal for capturing rare behaviors or complete interaction networks.
• Behavioral Ethogram Definition: Clearly define and operationalize the social behaviors of interest (e.g., grooming, aggression, huddling, proximity) [29].
• Spatial Coverage: Position multiple cameras to cover all functional areas of the habitat or enclosure [30].
• Temporal Sampling: Conduct observations during peak activity periods (e.g., 08:00-09:00, 15:00-16:00, 20:00-21:00) across multiple days [30].
• Data Recording: For each interaction, record initiator, recipient, behavior type, duration, and intensity.

Validation Notes: Studies comparing FAS and ABS in macaques found correlations for degree centrality across grooming, huddling, and aggression networks, though correlations for eigenvector centrality varied by species and behavior type [29].

Protocol 2: Multiplex Network Analysis

Objective: To integrate multiple interaction types into a comprehensive multiplex centrality metric.

Materials:

• Data on multiple social behaviors (e.g., aggression, status signaling, conflict policing, grooming, huddling) [31]
• Statistical software with network analysis capabilities (R, UCINET, PARTNER CPRM)
• Computing resources for matrix operations

Procedure:

• Network Layer Construction: Create separate social networks for each distinct behavior type using the same individuals [31].
• Data Normalization: Standardize interaction frequencies across networks to account for different base rates of behaviors.
• Consensus Ranking: Apply consensus ranking algorithms to calculate multiplex centrality across all layers [31].
• Model Validation: Test whether consensus ranks detect known social patterns (e.g., association with dominance, sex, family size) [31].
• Statistical Analysis: Use generalized linear mixed models to analyze how individual attributes (sex, age, dominance rank, rearing history) predict multiplex centrality [31].

Validation Notes: In rhesus macaques, consensus ranking successfully detected known social patterns, showing greater multiplex centrality in high-ranking males with high certainty of rank and females from the largest families [31].

Table: Minimum Sampling Requirements for Reliable Network Metrics

Network Metric	Minimum Observations/Individual	Recommended Sampling Method	Species-Specific Considerations
Degree Centrality	10-20 samples [29]	FAS or ABS	Robust across methods for grooming, huddling, aggression [29]
Edge Weight Estimation	Minimum 20 samples [29]	FAS for precise measures	Critical for weighted centrality measures
Betweenness Centrality	15+ samples	ABS for complete network view	More reliable in tolerant species than despotic species [29]
Eigenvector Centrality	15+ samples	Method depends on social style	Correlated between FAS/ABS for grooming, not for huddling in despotic species [29]
Network Modularity	20+ samples	ABS preferred	Correlated between methods for affiliative but not aggression networks [29]

Visualization and Workflow Diagrams

Social Network Analysis Workflow for Animal Behavior

Brokerage Typologies in Animal Social Networks

The Scientist's Toolkit: Research Reagent Solutions

Essential Materials and Methodological Solutions for Animal SNA

Tool/Reagent	Function/Application	Implementation Notes
Color-Coded Marking System	Individual identification for behavioral tracking [30].	Use non-toxic paints or dyes; ensure visibility at distance; reapply as needed.
Multi-Camera Video System	Comprehensive behavioral sampling with spatial coverage [30].	Minimum 5 cameras positioned to cover functional areas; time-synchronized recording.
PARTNER CPRM Software	Network mapping and analysis with customizable color palettes [32].	Apply specialized color palettes (e.g., Dark2 for light backgrounds) for accessibility.
Focal Animal Sampling Protocol	Detailed individual-level behavioral data collection [29].	Optimal for capturing complete behavioral profiles of target individuals.
All-Occurrences Sampling Protocol	Group-wide data on specific behavior types [29].	Efficient for capturing rare behaviors or complete interaction networks.
Consensus Ranking Algorithm	Integrates multiple network layers into unified centrality metric [31].	Handles networks with different properties (sparse vs. dense) and biological meanings.
Propensity Score Matching	Creates comparable treatment/control groups in natural experiments [33].	Controls for confounding variables in observational studies of social dynamics.
Brokerage Typology Framework	Classifies five distinct brokerage roles in social networks [30].	Reveals how individuals mediate relationships within and between subgroups.
OXS007417	OXS007417, MF:C20H14F3N3O, MW:369.3 g/mol	Chemical Reagent
4-CPPC	4-CPPC, MF:C14H9NO6, MW:287.22 g/mol	Chemical Reagent

Applications and Case Studies in Animal Behavior Research

Case Study: Multiplex Centrality in Rhesus Macaques

A recent study validated consensus ranking as a method for quantifying multiplex centrality across five interaction layers (aggression, status signaling, conflict policing, grooming, and huddling) in seven social groups of rhesus macaques [31]. The analysis revealed that:

• Multiplex centrality was greater in high-ranking males with high certainty of rank
• Females from the largest families showed higher multiplex centrality
• Mother-reared individuals were more central than nursery-reared counterparts
• The method detected individuals whose social importance would have been missed in single-layer analyses [31]

Case Study: Natural Experiment of Hurricane Effects

Research leveraging Hurricane Ike as a natural experiment examined social network formation in college students, with implications for animal social dynamics research [33]. The study demonstrated:

• Affected individuals were more likely to strengthen interactions while maintaining similar numbers of friends
• Social relationships may serve as a coping mechanism in high-stress situations
• Methodology using propensity score matching created comparable treatment/control groups [33]
• This approach provides a template for studying natural disasters' effects on animal social networks

Case Study: Preferential Associations in Dynamic Sow Herds

Application of SNA to commercial pigs revealed complex social dynamics in unstable networks [30]:

• Social discrimination occurred in assortment by connectedness even without reciprocal ties
• The most connected sows were significantly more likely to be approached
• Brokerage typologies existed, with most connected sows predominantly engaging in coordinating behavior
• Reciprocity was not a primary motivator for establishing preferential associations [30]

These findings highlight how SNA can reveal social complexities in managed animal populations with practical implications for welfare and management practices.

Application Notes

The integration of artificial intelligence (AI) with social network analysis (SNA) is transforming the study of animal social structures. This approach enables the automated, high-resolution, and non-invasive collection of behavioral data on a continuous basis, revealing hidden social dynamics within commercial pig populations that were previously impractical to measure [34] [35]. These technological advances provide novel insights into how social hierarchies form and evolve, with direct applications for improving animal welfare, health, and productivity [34] [36].

Key Quantitative Findings from AI-SNA Studies

The following table synthesizes core quantitative findings from recent research applying AI and SNA to commercial pig populations:

Table 1: Key Quantitative Findings from AI-SNA Studies in Pigs

Metric Category	Specific Finding	Reported Value / Significance	Context and Implications
Temporal Dynamics	Increase in group-level centralization (degree, betweenness, closeness)	Significant increase (P < 0.02) from early to later growing period [34]	Indicates social structure becomes more defined and hierarchical as pigs mature [34].
Individual-Level Stability	Change in individual closeness centrality and clustering coefficient	Significant increase (P < 0.00001) over time [34]	Reflectates shifts in an individual's proximity to others and the tightness of its peer group as the hierarchy stabilizes.
Interaction Definition	Impact of proximity definition on network metrics	Degree centrality is less affected than eigenvector centrality and clustering coefficient [36]	Highlights the critical importance of standardizing proximity definitions for reproducible SNA.
Sampling Rate	Minimum sampling rate for robust networks	A rate beyond 1 frame every 6 minutes is recommended (r > 0.90 correlation with complete data) [36]	Provides a data-driven guideline for balancing computational efficiency and analytical accuracy.
Early-Life Agonistic Behavior	Weighted degree centrality in weaned pigs	Higher fighting intensity post-mixing compared to older age groups [37]	Quantifies the intense aggression during initial hierarchy formation, a key welfare concern.

The Researcher's Toolkit: Essential Reagents and Technologies

Table 2: Essential Research Reagents and Solutions for AI-SNA in Pigs

Item / Technology	Function / Application
2D Camera Systems with Multi-Object Tracking	Captures raw video data for individual identification and posture classification [34].
Deep Learning Algorithms (e.g., CNN, DeepCut)	Performs pose estimation, identifies key body points, and classifies postures/activities from video data [34].
Ear Tag Identification System	Provides unique identification for individual animals, enabling longitudinal tracking [34].
SNA Computational Pipeline (e.g., R packages: spatsoc, asnipe, igraph)	Constructs social networks from tracking data and calculates network metrics at group and individual levels [34].
GPS or Ultra-Wideband (UWB) Telemetry Tags	An alternative technology for collecting high-resolution spatial location data, especially in larger enclosures [38].
MC-Gly-Gly-Phe-Boc	MC-Gly-Gly-Phe-Boc, MF:C27H36N4O7, MW:528.6 g/mol
OncoFAP-GlyPro-MMAF	OncoFAP-GlyPro-MMAF, MF:C102H140F2N20O28S, MW:2164.4 g/mol

Experimental Protocols

Protocol 1: Constructing Social Networks from Automated Tracking Data

This protocol details the process of converting raw video data into quantifiable social networks, based on established methodologies [34] [36].

1.1 Data Collection and Preprocessing

• Equipment Setup: Install 2D cameras above pig pens to ensure a clear, overhead view of all animals.
• Individual Identification: Use a custom multi-object tracking algorithm that combines a deep learning-based pose estimation model (e.g., a customized DeepCut algorithm) with a reliable ear-tag reading system. This generates real-time data for each pig, including: unique ID, timestamp, posture (standing, lying, sitting), and 2D XY-coordinates of body points (e.g., shoulder, rump) [34].
• Data Validation: Implement a prolonged "learning period" (e.g., 9 months) to fine-tune equipment and algorithms. Data from this period should be excluded from final analysis to avoid bias [34].
• Activity Filtering: For analyzing active social dynamics, filter the data to include only frames where pigs are in a "standing" posture [34].

1.2 Defining Social Interactions via Proximity

• Spatial Threshold: Define a social interaction (edge) based on a sustained proximity threshold between standing animals. A common definition is a Euclidean distance of less than 0.5 meters between the shoulder points of two pigs, maintained for a duration longer than the average brief interaction time within the pen [34].
• Temporal Sampling: To reduce computational load while maintaining data integrity, downsample the video frames. Do not exceed a rate of 1 frame every 6 minutes to preserve high correlation (r > 0.90) with the complete dataset [36].

1.3 Network Construction and Metric Calculation

• Software: Use a computational pipeline in R, utilizing packages like spatsoc for group-based spatial analysis, asnipe for network generation, and igraph or sna for calculating network metrics [34].
• Generate Association Matrices: Create weighted, undirected networks where nodes represent individual pigs, and edges represent the frequency or total duration of proximity events between dyads.
• Calculate Key Metrics:
  • Group-Level: Degree centralization, betweenness centralization, density.
  • Individual-Level: Weighted degree centrality (connection strength), betweenness centrality (role as a bridge), closeness centrality (efficiency of access to the network), eigenvector centrality (influence) [34] [37] [36].

The following workflow diagram illustrates this multi-stage protocol:

Protocol 2: Assessing Robustness and Bias of Social Network Metrics

This protocol provides a framework for evaluating the reliability of SNA metrics, which is critical when making inferences from partially sampled populations or when comparing groups [38].

2.1 Testing for Non-Random Structure

• Objective: Determine if the observed network metrics reflect a true social structure rather than random associations.
• Method: Generate null networks by permuting the pre-network data stream (e.g., randomly shuffling individual identities). Compare the observed network metrics to the distribution of metrics from the null networks. If the observed value falls outside the 95% confidence interval of the null distribution, the structure is considered non-random [38].

2.2 Assessing Bias from Sub-Sampling

• Objective: Quantify how the sampling proportion of individuals affects global network metrics.
• Method: Repeatedly sub-sample your dataset (e.g., 1000 iterations) at progressively lower proportions of tagged/identified individuals (e.g., from 90% down to 10%). For each sub-sample, re-calculate the global metrics of interest (e.g., density, centralization). Plot the estimated metric values against the sampling proportion to visualize bias [38].

2.3 Quantifying Uncertainty with Bootstrapping

• Objective: Generate confidence intervals for global and node-level network metrics.
• Method for Global Metrics: Apply a bootstrapping technique (sampling with replacement) to create multiple simulated datasets from your original data. Calculate the global metric for each bootstrap sample. The 2.5th and 97.5th percentiles of the resulting distribution provide a 95% confidence interval for each metric [38].
• Method for Node-Level Metrics: Similarly, use bootstrapping to generate distributions for individual centrality metrics. This allows researchers to assess whether the centrality of one individual is significantly different from another, accounting for uncertainty [38].

The logical relationships and decision points in the robustness assessment protocol are shown below:

Application Notes: Integrating Social Network Analysis in Behavioral Research

Social Network Analysis (SNA) has become an indispensable toolkit for quantifying the complex social structures of animal societies, from primate grooming networks to avian flocking patterns. By representing individuals as nodes and their interactions as edges, SNA allows researchers to move beyond dyadic relationships and understand population-level phenomena, including information flow, disease transmission, and collective decision-making [13]. The cross-species application of these methods reveals both universal principles and unique adaptations in social organization.

In primate research, SNA has been pivotal in understanding how social bonds facilitate cooperation and manage conflict. A landmark study on chimpanzee communities at Ngogo, Kibale National Park, utilized grooming networks to document the precursors to a rare permanent community fission. Analysis of long-term data revealed that differentiation in male-male grooming networks between what would become the Ngogo Central and Ngogo West communities was detectable years before the fission was behaviorally obvious, highlighting the predictive power of SNA for major social upheavals [39]. Furthermore, research on rhesus macaques has demonstrated a direct link between social network dynamics and cognitive processes like social attention. Scientists quantified multi-dimensional social relationships—aggregating grooming, aggression, and proximity into a Social Engagement Index (SEI) and Individual Engagement Index (IEI)—and found that these indices significantly shaped patterns of social attention, a relationship that was subsequently modulated by oxytocin administration [40] [41].

Conversely, in avian ecology, SNA has transformed our understanding of mixed-species flocks. A large-scale study of 84 flock networks across the Andes used SNA to test the "open-membership hypothesis." This research examined how network connectivity and cohesion (e.g., modularity, connectance) vary across environmental gradients. The findings confirmed that in harsher, high-elevation environments, flocks function more as open-membership systems with numerous weak associations, while flocks in milder, low-elevation conditions are more structured and modular, reflecting higher costs of competition and activity matching [42]. The physics underlying these flocks is equally complex; research using robotic flapping wings has revealed that stable formations rely on "flonons"—coherent waves of aerodynamic interaction that can destabilize large, identical groups, suggesting that diversity in wingbeat phases is crucial for maintaining long-distance flock integrity [43].

The methodological framework for such studies is critical. A robust, multi-step protocol has been established to assess bias and robustness in social network metrics derived from partial population data, such as that from GPS telemetry. This protocol involves testing for non-random network structure, quantifying bias and uncertainty in global metrics through sub-sampling, and assessing the reliability of node-level metrics, ensuring that ecological inferences are based on statistically sound network representations [38].

Table 1: Key Findings from Cross-Species Social Network Studies

Study System	Key Social Network Metric	Primary Finding	Biological Implication
Chimpanzee Community Fission [39]	Grooming Network Differentiation	Grooming networks differentiated prior to observable behavioral signs of fission.	SNA can predict major social restructuring; fission driven more by male mating competition than grooming network constraints.
Rhesus Macaque Social Attention [40] [41]	Social Engagement Index (SEI), Individual Engagement Index (IEI)	Social attention patterns correlated with SEI and IEI; oxytocin altered these relationships.	Multidimensional social relationships shape cognitive processes; neuroendocrinology directly linked to network-influenced behavior.
Andean Mixed-Species Flocks [42]	Network Modularity, Connectance	Modularity decreased and connectance increased with elevation, indicating more open membership in harsher environments.	Flock structure is context-dependent, balancing costs and benefits of grouping across environmental stress gradients.
Robotic Flock Physics [43]	Spatial Formation Stability	Stable flocks require disruption of amplifying "flonon" waves via phase diversity or vacancy defects.	Individual variation is key to maintaining collective motion; challenges the idea of perfectly identical synchrony.

Experimental Protocols

Protocol for Assessing Robustness in Animal Social Networks

This protocol is designed to evaluate the reliability of social network metrics when only a subset of a population is observed, a common scenario in GPS-telemetry studies [38].

Step 1: Testing for Non-Random Structure

• Objective: To determine if the observed network structure is non-random.
• Method: Generate a null distribution of network metrics (e.g., density, centrality) by permuting the pre-network data stream (e.g., randomizing the timing or location of individual observations). If the observed metric falls outside the 95% confidence interval of the null distribution, it is considered non-random.

Step 2: Quantifying Bias in Global Network Metrics

• Objective: To assess how bias changes with the proportion of individuals sampled.
• Method: Iteratively re-calculate global network metrics (e.g., density, modularity) using random sub-samples of the tracked individuals (e.g., 90%, 80%, ... 20% of the sample). Plot the metric value against the sampling proportion to visualize bias and identify a robustness threshold.

Step 3: Bootstrapping for Uncertainty and Confidence Intervals

• Objective: To estimate the uncertainty of global network statistics.
• Method: Apply a bootstrapping technique (e.g., resampling individuals with replacement) to generate numerous simulated networks from the partial data. Calculate confidence intervals (e.g., 95% CI) for each global metric from the bootstrap distribution.

Step 4: Assessing Robustness of Node-Level Metrics

• Objective: To determine the reliability of individual-level metrics like centrality.
• Method: For each node-level metric, calculate the correlation (e.g., Pearson's r) between its value in the full sample network and its value in networks built from progressively smaller sub-samples. High correlations across sub-sampling levels indicate robustness.

Step 5: Generating Node-Level Confidence Intervals

• Objective: To provide estimates of uncertainty for individual network positions.
• Method: Use the bootstrapped networks from Step 3 to calculate confidence intervals for each node's network metrics (e.g., degree, betweenness centrality). This allows researchers to statistically compare the social connectivity of different individuals.

Protocol for Quantifying Multi-Dimensional Primate Social Relationships

This protocol outlines the method for integrating different social behaviors into a composite index to study their relationship with cognitive tasks or other ecological variables [40] [41].

Step 1: Data Collection on Core Social Behaviors

• Objective: To systematically record affiliative and agonistic interactions.
• Method: Conduct focal animal sampling or use automated video tracking (e.g., with YOLOv5 models) to collect data on:
  • Proximity: Frequency and duration of spatial closeness outside of aggressive contexts.
  • Grooming: Frequency, duration, and directionality (who grooms whom).
  • Aggression: Frequency, intensity, and directionality of aggressive acts (e.g., chases, threats).

Step 2: Calculating Dyadic Interaction Scores

• Objective: To quantify the directional relationship within each dyad for each behavior.
• Method: For each behavior (e.g., grooming), use a dichotomous framework. For a dyad (Monkey A and Monkey B), if A grooms B more than B grooms A, assign a score of +1 for A and -1 for B. This is repeated for all behaviors and all dyads.

Step 3: Computing the Social Engagement Index (SEI)

• Objective: To summarize an individual's overall social tendency within the group.
• Method: For each individual, sum their interaction scores across all partners and all three behavioral dimensions. Weights can be assigned to different behaviors based on the research question. SEI_A = (Proximity_Score_Total_A + Grooming_Score_Total_A + Aggression_Score_Total_A)

Step 4: Computing the Individual Engagement Index (IEI)

• Objective: To quantify the interaction tendency within a specific dyadic relationship.
• Method: For a given dyad (e.g., A-B), sum the interaction scores from all three behavioral dimensions for that specific pair. IEI_AB = (Proximity_Score_AB + Grooming_Score_AB + Aggression_Score_AB)

Step 5: Linking Indices to Dependent Variables

• Objective: To test how social relationships influence other phenomena.
• Method: Use regression models to correlate an individual's SEI or the IEI of a dyad with a dependent variable, such as performance in a social attention task (e.g., response times when distracted by images of groupmates) or physiological measures. Experimental manipulations like oxytocin administration can be introduced to probe causality.

Visualization of Workflows and Analytical Frameworks

Protocol for Robustness Assessment in Social Network Analysis

Multidimensional Social Relationship Analysis

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials and Tools for Social Network Analysis in Animal Behavior

Tool or Material	Function/Application	Example Use Case
GPS Telemetry Collars	High-resolution tracking of individual location and movement over time.	Core data source for constructing association networks based on spatio-temporal proximity in ungulates or other large animals [38].
Automated Video Tracking (YOLOv5)	Objective, high-throughput coding of specific social behaviors from video footage.	Automated detection of grooming, aggression, and proximity in captive primate groups [40] [41].
Network Analysis Software (UCINET, aniSNA R package)	Software for calculating and visualizing social network metrics and conducting statistical tests on networks.	Used across all studies to compute metrics like density, modularity, and centrality, and to implement the robustness assessment protocol [39] [38].
Oxytocin and Administration Kits	Experimental neuroendocrine manipulation to probe the biological mechanisms underlying social behavior.	Investigating the causal role of oxytocin in modulating the relationship between social network position and social attention in macaques [40] [41].
Robotic Model Systems (3D-printed flappers)	Controlled experimental study of the physical principles governing collective motion.	Isolating and testing the aerodynamic rules of flocking in birds, revealing the role of "flonons" and phase diversity [43].
[Orn5]-URP TFA	[Orn5]-URP TFA, MF:C48H62N10O10S2, MW:1003.2 g/mol	Chemical Reagent

Navigating Analytical Challenges: Robust Social Network Definition and Interpretation

Within the field of animal social network analysis, a fundamental challenge is the "Association Definition Problem"—the question of how to best define a social connection, or edge, between individual animals when direct behavioral interactions are difficult to observe. The method chosen to define these associations fundamentally shapes the constructed network and can influence subsequent ecological and evolutionary interpretations [44] [45]. Social network analysis provides a powerful framework for quantifying social structure, linking individual behavior to population-level patterns such as information spread and disease transmission [44]. However, the "real" social network is often inferred from observed patterns of co-occurrence, making the definition of an association a critical methodological decision [44].

This document details the application and protocols for three primary methods used to infer social associations from spatio-temporal data: the strict time-window, co-occurrence in a group (often using Gaussian Mixture Models), and arrival-time approaches. Framed within a broader thesis on animal social behavior, these notes provide researchers with the practical tools to select, implement, and validate these methods, ensuring that network edges possess biological relevance.

The table below summarizes the core characteristics, applications, and comparative findings of the three association definition methods based on empirical research from four avian study systems [46] [45] [47].

Table 1: Comparison of Association Definition Methods in Animal Social Network Analysis

Method	Core Definition of an Association	Typical Data Input	Key Advantages	Key Limitations	Comparative Findings (from Bird Studies)
Strict Time-Window [45]	Individuals detected at the same location within a fixed, pre-defined duration (e.g., Δt).	Time-stamped location data (e.g., from RFID feeders, acoustic telemetry).	Simple to implement and compute; highly transparent and reproducible.	Requires a priori justification for window length; may miss associations if Δt is too short or create false positives if too long.	Networks showed high similarity to those from other methods when Δt was ecologically relevant. Demonstrated robustness in network structure [46] [47].
Co-occurrence in a Group (GMM) [45]	Individuals belong to the same dynamically identified "group" based on bursts of activity at a resource (e.g., using Gaussian Mixture Models).	Dense time-stamped data from centralized resources like feeders.	Data-driven; identifies natural grouping events without a fixed time window; effective for fission-fusion dynamics.	Model complexity; may struggle in highly gregarious species with less clear group boundaries; designed for specific foraging contexts.	Effectively identified flocks in species like great tits. Subtle differences in network metrics emerged, influenced by species biology and feeder design [46] [45].
Arrival-Time [45]	Individuals arriving at a location in close temporal succession, indicating coordinated movement.	Precise arrival and departure times at a resource.	Captures fine-scale coordinated movement; may better reflect social attraction in highly gregarious species.	Highly dependent on precise timing data; may be sensitive to resource distribution and density.	Provided a finer-scale measure of association in gregarious species like house sparrows. Networks were largely comparable but sensitive to system ecology [45] [47].

Detailed Experimental Protocols

Protocol for the Strict Time-Window Method

This protocol is best suited for research questions where spatial and temporal proximity is the primary factor of interest, such as in disease transmission studies [45].

Workflow Overview:

• Data Collection: Gather time-stamped presence records (e.g., DateTime, Individual ID, Location ID) from RFID feeders, acoustic telemetry receivers, or GPS loggers [48].
• Define Association Time-Window (Δt): Critically, the choice of Δt must be justified biologically.
  • Example: For house sparrows foraging at an RFID feeder, a Δt of 3 seconds was used to define a co-occurrence [45].
  • Validation: Conduct sensitivity analyses by constructing networks over a range of plausible Δt values and comparing their stability and the repeatability of individual social traits (e.g., degree, strength) [46] [47].
• Construct the Association Matrix:
  • For every individual i and j, calculate an association index. A common metric is the Simple Ratio Index (SRI) [48]: SRI = x / (y_i + y_j - x) where x is the number of time windows where i and j were co-present, and y_i and y_j are the total number of time windows in which each individual was detected [44] [48].
  • The resulting SRI matrix is a weighted adjacency matrix representing the social network.

Protocol for the Co-occurrence in a Group (GMM) Method

This method is ideal for systems where animals interact in discrete, dynamic groups, such as fission-fusion flocks of great tits [45].

Workflow Overview:

• Data Preparation: Compile a data stream of all individual presences at a resource, sorted by timestamp.
• Group Detection: Use the gmmevents function in the asnipe R package (or equivalent) to algorithmically detect grouping events [45].
  • The Gaussian Mixture Model (GMM) identifies "bursts" of activity at a feeder, treating each burst as a distinct group. The model dynamically determines the start and end of each grouping event based on the intervals between arrivals [45].
• Generate Group-by-Individual Matrix: The output is a matrix where rows represent unique grouping events and columns represent individuals. A 1 indicates membership in that group.
• Construct the Association Matrix: Calculate the association matrix (e.g., using SRI) directly from the group-by-individual matrix using the get_network function in asnipe [45].

Protocol for the Arrival-Time Method

This protocol is particularly useful for gregarious species where coordinated arrivals at a resource may indicate stronger social bonds than simple co-presence [45].

Workflow Overview:

• Data Requirement: Obtain high-resolution data on arrival times at a resource. Precise timestamps are critical.
• Define Arrival Threshold: Set a maximum time difference (e.g., 30 seconds) between successive arrivals to consider them part of the same socially coordinated event [45].
• Identify Arrival Clusters:
  • Sort all arrivals by timestamp.
  • Group arrivals that occur within the defined threshold of each other into a single cluster.
  • Individuals within the same cluster are considered associated for that event.
• Construct the Association Matrix: Assemble a group-by-individual matrix based on these arrival clusters and then calculate the association matrix (e.g., SRI) as in the previous protocols.

Visualizing Method Workflows

The following diagram illustrates the logical flow and key decision points for each of the three association definition methods.

The Scientist's Toolkit: Research Reagent Solutions

Successful implementation of these protocols relies on specific technological and computational tools. The following table lists essential "research reagents" for animal social network analysis.

Table 2: Essential Materials and Tools for Social Association Research

Tool / Technology	Type	Primary Function in Research	Key Considerations
RFID Feeders [45]	Hardware	Automatically records visits of PIT-tagged individuals at a central resource, generating precise timestamps for association analysis.	Ideal for small birds and mammals; design affects data quality (e.g., limits simultaneous detections).
Acoustic Telemetry [48]	Hardware	Tracks movements and co-occurrences of large, free-ranging animals over wider areas using uniquely coded transmitters and receivers.	Receiver placement density is critical for accurately defining co-occurrences in aquatic and terrestrial environments.
Passive Integrated Transponder (PIT) Tag [45]	Hardware	A small, inert microchip implanted in or attached to an animal, providing a unique ID when read by a compatible scanner.	The standard for marking individuals in automated feeder systems.
R Package asnipe [45]	Software	Performs key social network analyses, including the GMM group detection method and network permutation tests.	Central to implementing the co-occurrence in a group method.
Simple Ratio Index (SRI) [48]	Analytical Metric	Calculates the strength of association between two individuals, correcting for individual variation in detection frequency.	A standard association index for binary (co-present/absent) data.
GPS Telemetry Tags [49]	Hardware	Provides high-resolution spatio-temporal movement data, enabling association definitions based on fine-scale proximity in space.	Decreasing size and cost is enabling tracking of smaller species and larger sample sizes.
Machine Learning Pose Estimation (e.g., DeepLabCut) [49]	Software	Tracks the position and orientation of multiple individuals and their body parts from video, enabling detailed interaction analysis.	Moves beyond simple co-occurrence to define edges based on directed behaviors and orientations.

Empirical evidence from comparative studies demonstrates that animal social networks are largely robust to the choice of association definition method, with networks constructed using different but ecologically justified methods showing similar overall characteristics [46] [45] [47]. However, subtle yet important differences in network structure and individual social metrics can arise, driven by the specific biology of the study species (e.g., great tits vs. house sparrows) and the design of data collection equipment [45].

Therefore, the central tenet for researchers is that methodological decisions must be guided by the biological context. The research question and the species' natural history should inform the a priori selection and parameterization of an association definition, rather than relying on a default method [46] [45]. Validation steps, such as testing the repeatability of social traits and conducting sensitivity analyses, are crucial for ensuring that the inferred network edges are meaningful representations of social relationships, thereby solidifying the foundation upon which all subsequent behavioral, ecological, and evolutionary inferences are built.

In animal social network analysis, missing data presents a fundamental challenge to inferring reliable conclusions about social structures and their drivers. Social networks are generally not observed directly but must be approximated from behavioural samples, creating significant potential for uncertainty and bias in estimated relationships [2]. The problem extends beyond simple data gaps to affect the core inferential tasks in behavioural ecology—understanding how phenotypic and ecological factors shape social relationships. When network data contain uncertainties, the observed edges become correlated through both biological and sampling processes, potentially leading to dysfunctional statistical procedures and incorrect results if not properly addressed [2]. This challenge is particularly acute in animal studies where researchers cannot simply ask subjects about their relationships and must instead infer connections from observed interactions.

The importance of properly handling missing data has been increasingly recognized as the field has matured. Early animal social network analyses often treated observed networks as complete representations of social systems, but methodological advances have revealed the profound consequences of missing data for estimating network properties and drawing biological inferences. As the field moves toward more sophisticated dynamic and causal inference approaches [7] [2], robust methods for addressing uncertainty in social relationships have become essential tools for behavioural ecologists.

Conceptual Framework: Understanding Uncertainty in Social Relationships

A sophisticated approach to missing data begins with recognizing three distinct levels of abstraction in any social network analysis [2]:

• Relationship Network: The theoretical construct representing latent social relationships (e.g., affiliation, dominance, friendship) that researchers aim to study but cannot observe directly.
• Interaction Network: The quantifiable social interactions used as proxies for theoretical constructs, representing the true interaction rates for specific behaviours.
• Measured Network: The actually observed data, subject to sampling limitations, technological constraints, and logistical challenges.

This conceptual separation clarifies that missing data occurs at the level of the measured network but propagates uncertainty to inferences about both interaction and relationship networks. The distinction is crucial because different types of missing data require different handling techniques, and the impact of missing data varies across these levels of abstraction.

Table: Common Sources of Missing Data in Animal Social Network Studies

Source Type	Specific Examples	Typical Impact
Sampling Limitations	Low resolution tracking, limited observation time, incomplete group monitoring	Underestimation of weak ties, biased centrality measures
Technical Constraints	Failed device deployment, sensor battery life, detection range limitations	Completely missing nodes or temporal gaps
Animal Behavior	Animals moving outside study area, cryptic behaviours, avoidance of observers	Systematic missingness related to behavioural traits
Logistical Challenges	Weather disruptions, resource limitations, field site inaccessibility	Irregular missing patterns across time and individuals

Missing data in animal social networks arises from multiple sources, each with different statistical properties and implications for analysis. Sampling limitations represent perhaps the most common challenge, where finite observation windows and limited tracking capabilities prevent researchers from observing the complete network [2]. The problem is particularly acute for wild populations where continuous monitoring is impossible, leading to networks constructed from sparse samples of behaviour.

Technical constraints associated with modern tracking technologies—including GPS tags, proximity loggers, and acoustic monitors—introduce another dimension of missing data. Deployments rarely achieve 100% coverage of study populations due to cost, capture challenges, or device failures, creating systematically missing nodes [7]. Additionally, these technologies have their own detection limitations that can miss certain interaction types or fail in specific environmental conditions.

Perhaps most challenging is missingness related to animal behaviour itself. Individuals may temporarily leave the study area, engage in unobservable behaviours, or modify their behaviour in response to observational methods. Each of these can create missing data patterns that are non-random and potentially correlated with traits of scientific interest, complicating statistical correction.

Quantitative Techniques for Missing Data Handling

Bayesian Approaches for Uncertainty Propagation

Bayesian methods provide a powerful framework for handling missing data in social networks by explicitly modeling the uncertainty in social relationships and propagating this uncertainty through subsequent analyses. These approaches treat missing network data as parameters to be estimated rather than as a problem to be eliminated through deletion [2]. The Bayesian Social Relations Model, a multilevel extension specifically adapted for network data, incorporates partial pooling to share information across individuals, improving estimates for sparsely observed animals [2].

In practice, Bayesian models for missing network data specify a joint probability distribution over both the observed data and the missing values, with the relationship between them explicitly defined by the model structure. This allows researchers to obtain posterior distributions not just for model parameters but for the missing data itself, properly representing the uncertainty in network structure when making biological inferences. The resulting analyses naturally incorporate this uncertainty, producing more accurate confidence intervals and reducing false positive findings.

Statistical Framework: Stochastic Actor-Oriented Models (SAOMs)

Stochastic Actor-Oriented Models represent a specialized class of individual-based models designed specifically for analyzing network dynamics between discrete time points [7]. SAOMs model gradual change in networks and individual traits using hidden Markov models, treating the network as a constantly evolving system rather than a series of static snapshots. These models are particularly valuable for addressing missing data in longitudinal studies because they can incorporate information from multiple time points to inform estimates at any single observation period.

The strength of SAOMs lies in their ability to model network change as the outcome of individual decisions, where animals probabilistically form, maintain, or dissolve ties based on network structure and covariates [7]. This individual-based approach naturally handles missing data by focusing on the processes that generate observable interactions rather than requiring complete network snapshots at every time point. When data are missing, SAOMs can leverage the Markov property, using information from available time points to inform likely network states during periods with missing observations.

Table: Comparison of Missing Data Handling Techniques

Technique	Appropriate Scenarios	Key Assumptions	Implementation Considerations
Bayesian Imputation	Small to moderate missingness, informative missingness patterns	Missing at random mechanism	Computationally intensive, requires expert knowledge
SAOMs	Longitudinal data with partial missing time points	Markov process, gradual network change	Requires multiple observation waves
Multiple Imputation	Various missing data patterns, auxiliary variables available	Correct model specification	Simpler implementation, combines frequentist framework with uncertainty propagation
Data Augmentation	MCMC frameworks, complex missing data patterns	Exchangeability of parameters	Technical implementation within Bayesian algorithms

Data Collection Protocols for Minimizing Missing Data

Table: Recommended Sampling Protocols for Different Study Systems

Study System	Minimum Recommended Sampling	Key Mitigation Strategies	Validation Approaches
Primate Groups	80%+ response rate, continuous monitoring ≥2hr sessions	Focal animal sampling with rotation, proximity logger deployment	Inter-observer reliability tests, comparison with genetic data
Ungulate Herds	GPS collars on ≥70% individuals, daily locations	Combined direct observations and automated tracking, aerial surveys	Sensor detection range testing, movement model validation
Social Birds	Color bands on ≥90% individuals, standardized observation protocols	Fixed observation posts, coordinated sampling schedules	Resampling methods to estimate detection probabilities
Marine Mammals	Photo-ID catalogs with high coverage, systematic surveys	Mark-recapture models, collaborative data sharing	Discovery curves to assess catalog completeness

Effective handling of missing data begins with study design and data collection protocols that minimize unnecessary gaps. For observational studies, achieving response rates of 80% or higher is considered the gold standard for obtaining reliable network data [50]. This involves systematic sampling designs that ensure all individuals and potential interactions have non-zero probability of being observed. For focal animal sampling, this means balanced observation schedules that avoid systematically overlooking particular individuals or time periods.

Technological approaches can significantly reduce missing data in modern studies. The strategic deployment of proximity loggers, GPS tags, and acoustic monitoring devices can fill observation gaps, particularly for cryptic behaviours or difficult-to-observe time periods. However, these technologies require careful validation to ensure they capture the social interactions of interest and don't introduce their own biases through differential detection probabilities.

Protocols should also include explicit documentation of sampling effort and conditions during data collection. This metadata is essential for diagnosing patterns of missingness and implementing appropriate statistical corrections. Recording factors like weather conditions, observer identity, time of day, and methodological variations creates the necessary auxiliary information to model missing data mechanisms.

Experimental Protocols for Validation and Sensitivity Analysis

Protocol 1: Sensitivity Analysis for Missing Data Mechanisms

Purpose: To evaluate how sensitive research conclusions are to different assumptions about missing data mechanisms.

Materials: Complete-case dataset, statistical software with multiple imputation capabilities (R preferred), high-performance computing resources for Bayesian methods.

Procedure:

• Begin with the original dataset containing missing values and document the pattern and extent of missingness.
• Implement three analysis approaches under different missingness assumptions:
  • Missing Completely at Random (MCAR): Use complete-case analysis or simple imputation
  • Missing at Random (MAR): Implement multiple imputation using observed covariates
  • Missing Not at Random (MNAR): Apply selection models or pattern-mixture models
• For each approach, estimate key network parameters of scientific interest (e.g., density, centrality measures, homophily coefficients)
• Compare the results across assumptions, noting particularly sensitive parameters
• Report the range of plausible values for each parameter given uncertainty about missingness mechanisms

Validation: Apply the protocol to a subset of data with artificially introduced missingness where the true values are known. Calculate recovery accuracy for key network metrics.

Protocol 2: Resampling Validation for Network Robustness

Purpose: To assess how robust network inferences are to missing nodes and edges through systematic resampling.

Materials: The most complete available network data, custom R or Python scripts for resampling, visualization tools.

Procedure:

• Start with the most complete network dataset available (aim for >90% node and edge coverage)
• Implement a jackknife procedure that systematically removes:
  • Random subsets of nodes (5%, 10%, 20%, 30%)
  • Random subsets of edges (5%, 10%, 20%, 30%)
  • Nodes with specific characteristics (e.g., low centrality, particular traits)
• For each resampled network, recalculate the network metrics of scientific interest
• Compare the distribution of metrics from resampled networks to the complete network
• Identify which metrics are most sensitive to which types of missing data
• Establish confidence intervals for key metrics that incorporate missing data uncertainty

Interpretation: Metrics showing large variation across resampling scenarios require particularly careful interpretation and should be accompanied by uncertainty estimates in final reports.

Visualization and Reporting Standards

Workflow for Handling Missing Data in Social Network Analysis

Table: Essential Computational Tools for Handling Missing Data in Social Networks

Tool/Resource	Primary Function	Application Context	Implementation Considerations
RStena Package	Fits Stochastic Actor-Oriented Models for longitudinal networks	Dynamic network analysis with missing observations	Requires multiple waves of data; computationally intensive for large networks
BRMS Package	Bayesian multilevel modeling with custom response distributions	Flexible imputation models for complex missing data patterns	Steep learning curve but extremely versatile for custom models
mice Package	Multiple Imputation by Chained Equations	General missing data handling for various variable types	User-friendly but requires careful specification of imputation models
Social Relations Model	Partitioning variance in social behavior into individual and partner effects	Estimating social differentiation with incomplete data	Particularly useful for round-robin designs with missing observations
Network Resampling Methods	Bootstrap and jackknife procedures for network data	Quantifying uncertainty in network metrics due to missing data	Computationally intensive but makes minimal assumptions

Addressing missing data is not merely a statistical technicality but a fundamental requirement for robust inference in animal social network analysis. By implementing the techniques outlined in these application notes—from careful study design through Bayesian modeling and comprehensive sensitivity analyses—researchers can substantially strengthen the validity of their conclusions about social relationships. The field is moving toward greater methodological sophistication, with emerging frameworks explicitly acknowledging and modeling the uncertainty inherent in measuring social relationships [51] [2]. As these approaches become standard practice, we can expect more reproducible and reliable insights into the ecological and evolutionary processes shaping animal social systems.

Distinguishing Social Associations from Non-Social Aggregations

In animal social network analysis, accurately distinguishing social associations from non-social aggregations is a fundamental methodological challenge. Social associations refer to spatio-temporal co-occurrences driven by social attraction and intentionality, where individuals choose to associate with specific others [45]. In contrast, non-social aggregations are groupings explained by proximate, non-social factors, such as individuals gathering independently at a localized resource like a water hole, feeder, or sleeping site [45]. The biological significance of an edge in a social network depends entirely on this distinction. Misclassifying an aggregation as an association can lead to severe inferential errors when interpreting the drivers of social structure, the transmission of information, or the dynamics of disease spread [45] [2]. This document outlines the conceptual framework and provides detailed protocols for researchers to define and extract biologically relevant social associations from observational data.

Theoretical Framework and Key Concepts

Defining Social Networks in Animal Systems

Social networks in behavioral ecology are abstractions used to represent social structures, and it is crucial to distinguish between three levels of abstraction [2]:

• Level 1: The Theoretical Construct: The latent social relationship (e.g., affiliation, dominance, friendship) that is the ultimate target of inference but cannot be directly observed.
• Level 2: The Interaction Network: The network of quantifiable social interactions, which serves as a proxy for the theoretical construct.
• Level 3: The Measured Network: The network that is actually observed and measured, which is a sample of the interaction network and is subject to noise and bias.

The process of moving from the measured network to an understanding of the theoretical construct requires careful causal and statistical modeling to account for uncertainty and potential confounding factors [2].

Association vs. Aggregation

The core distinction for building a meaningful social network is as follows:

• Social Association: An association is inferred when the spatio-temporal co-occurrence of individuals is likely due to mutual social attraction. The identity of the involved individuals matters, as these connections represent potential social relationships [45]. For example, two birds consistently arriving together at a feeder may indicate a social bond.
• Non-Social Aggregation: An aggregation is a grouping driven by external, non-social factors. The individuals are drawn to the same location independently, and their co-occurrence is incidental rather than social [45]. Examples include moths attracted to a light source or animals congregating at a scarce water resource during a drought. Here, the identity of individuals is less critical than the context that caused the grouping.

Table 1: Comparative Features of Social Associations and Non-Social Aggregations

Feature	Social Association	Non-Social Aggregation
Primary Driver	Mutual social attraction, social bonds	External environmental factors (e.g., resource location, predation risk)
Intentionality	High; individuals choose to associate	Low; individuals respond independently to external stimuli
Individual Identity	Critical; specific relationships are key	Less important; group composition may be random
Network Implication	Represents a potential social relationship	May represent a transmission or dilution risk, but not a social bond
Biological Question	Social structure, information flow, mate choice	Disease ecology, predator dilution, resource use

Methodological Approaches for Defining Associations

The following table summarizes three common methods used to define social associations from spatio-temporal data, such as records from Passive Integrated Transponder (PIT)-tagged animals at RFID feeders [45].

Table 2: Methods for Defining Social Associations from Spatio-Temporal Data

Method	Description	Best Suited For	Considerations
Strict Time-Window (Δt)	All individuals recorded at the same location within a predefined, fixed time window are considered associates [45].	Systems where spatial and temporal proximity is the primary research interest; simplest to implement.	The choice of time window is critical. Too short may miss associations; too long may include non-social aggregations [45].
Group Co-occurrence (GMM)	Uses Gaussian Mixture Models (GMM) to dynamically identify bursts of activity and define discrete grouping events [45].	Systems with clear fission-fusion dynamics and distinct foraging bursts (e.g., great tits) [45].	May struggle in highly gregarious species with loose group boundaries (e.g., house sparrows) [45].
Arrival-Time	Defines associations based on the time between successive arrivals of individuals at a resource [45].	Gregarious species where socially connected individuals are more likely to arrive in tight succession [45].	Captures fine-scale movement coordination; may be less sensitive to prolonged co-feeding in dense aggregations.

Experimental Protocols

Protocol: Implementing the Strict Time-Window Method

Application Note: This protocol is designed for processing raw timestamped data from RFID feeders or similar automated tracking systems.

Workflow Overview:

Detailed Methodology:

• Data Acquisition and Preprocessing:

  • Data Source: Collect data from an automated system (e.g., RFID feeders, GPS loggers). The raw data should, at a minimum, include: Animal_ID, Location_ID, and Timestamp.
  • Data Cleaning: Implement filters to remove physiologically impossible movements and erroneous detections (e.g., the same individual registered at two distant feeders within an impossibly short time).

• Defining the Association (Δt):

  • Parameter Selection: The choice of the time window (Δt) must be justified biologically. For example, a study on house sparrows used a Δt of 3 seconds to define an association at a feeder [45].
  • Ecological Validation: The selected Δt should reflect the typical spatio-temporal scale of social interactions for the study species. This can be informed by direct behavioral observations or prior literature.

• Generating the Association Matrix:

  • Algorithm: For every unique pair of individuals (i, j), count the number of times they were detected at the same location within the defined Δt window.
  • Implementation (Pseudocode):

  • Output: A symmetric association matrix (or GBI matrix) where cells represent the association strength (frequency or duration) between each dyad.

• Network Construction and Analysis:

  • Software: Use packages such as asnipe in R [45] or igraph to construct a network from the association matrix.
  • Network Metrics: Calculate individual-level social traits (e.g., degree, strength, betweenness) and global network properties.

Protocol: Comparative Analysis of Association Definitions

Application Note: This protocol guides researchers in testing the robustness of their social network results to different association definitions.

Workflow Overview:

Detailed Methodology:

• Data Processing: Start with a single, cleaned dataset of spatio-temporal detections, as described in Protocol 4.1.

• Parallel Network Construction:

  • Construct multiple social networks from the same underlying data using the three different methods outlined in Table 2.
  • Strict Time-Window: Follow Protocol 4.1.
  • Group Co-occurrence (GMM): Use the asnipe package in R to run the Gaussian Mixture Model, which infers group membership based on temporal clustering of detections [45].
  • Arrival-Time: Adapt the association definition to be based on the arrival intervals between individuals (e.g., all individuals arriving within t seconds of each other form an association) [45].

• Calculation of Social Traits:

  • For each resulting network, calculate a standard set of individual-based social traits for all individuals. Key traits include:
    • Degree: The number of unique social associates.
    • Strength: The sum of association weights (strengths) for an individual.
    • Betweenness: A measure of an individual's role as a connector in the network.

• Comparison and Validation:

  • Statistical Comparison: Use statistical tests like Mantel tests to assess the correlation between the association matrices generated by different methods.
  • Repeatability Analysis: Calculate the repeatability (e.g., using Intra-class Correlation Coefficient) of individual social traits across the different networks. High repeatability suggests network structure is robust to the definition used [45].
  • Biological Justification: The final choice of method should be driven by which network structure best predicts a biologically relevant outcome (e.g., pairing success, information spread) or aligns most closely with direct behavioral observations.

The Scientist's Toolkit: Research Reagents and Materials

Table 3: Essential Materials and Analytical Tools for Social Network Construction

Item	Function/Description	Example Use in Protocols
RFID Feeder System	Automated data collection system that records individual animal identities and precise timestamps at a resource [45].	Primary data source for Protocols 4.1 and 4.2. Provides the Animal_ID, Location_ID, Timestamp data stream.
PIT Tags	Passive Integrated Transponder tags uniquely identifying each individual animal.	Deployed on study subjects; detected by the RFID feeder system.
R Statistical Software	Open-source environment for statistical computing and graphics.	Primary platform for data analysis and network construction.
asnipe R package	A package for the analysis of social network data, including the GMM group detection method [45].	Used in Protocol 4.2 to implement the Group Co-occurrence (GMM) association definition.
igraph R package	A powerful and comprehensive network analysis library.	Used for network construction, visualization, and calculating network metrics (degree, betweenness, etc.) in all protocols.
Color Contrast Checker	A tool to ensure visualizations meet accessibility standards (e.g., WCAG 2.0).	Critical for creating accessible diagrams and figures for publications, ensuring sufficient contrast between foreground and background colors [52] [19].

In animal social network analysis, temporal resolution—the frequency and duration of behavioral observations—profoundly impacts the accuracy and ecological validity of network inferences. Social networks are dynamic constructs where interactions fluctuate across timescales, making the choice of observation period a critical methodological decision [2]. This protocol provides a structured framework for determining optimal observation periods tailored to specific research questions, enabling researchers to balance logistical constraints with scientific rigor. By integrating current methodologies from GPS-based telemetry and behavioral sampling, we address the challenge of deriving robust social metrics from data streams that are often autocorrelated and incomplete [38]. The guidelines presented are particularly critical for studies where only a subset of populations can be monitored, a common scenario in wildlife research due to financial and practical constraints [38].

To properly contextualize temporal resolution, one must first distinguish between three fundamental levels of network abstraction [2]:

• Relationship Network: The theoretical construct representing latent social relationships (e.g., dominance hierarchies, alliances, or friendships). This is the ultimate target of inference but is not directly observable.
• Interaction Network: The quantified rate of specific behavioral interactions (e.g., grooming, aggression) used as a proxy for the underlying relationship.
• Sampled Network: The actual observed data, which is a partial representation of the interaction network, subject to sampling error and temporal variation.

The core challenge is that the Sampled Network must adequately represent the Interaction Network to allow valid inferences about the Relationship Network. The chosen temporal resolution directly governs the fidelity of this representation [38].

Quantitative Guidelines: Temporal Resolution for Core Research Areas

Table 1: Optimal Observation Periods for Key Research Domains

Research Question	Behavioral Proxies	Minimum Sampling Frequency	Minimum Total Duration	Key Network Metrics	Evidence Base
Disease Transmission Dynamics	Physical contact, proximity	High (Minutes-Hours)	Multiple transmission cycles	Degree centrality, Betweenness	[53]
Information/Social Learning	Co-feeding, matched activity	Medium-High (Hours-Daily)	Weeks to Months	Clustering coefficient, Modularity	[53]
Mating Strategies & Reproductive Success	Courtship, consortships, agonistic interactions	Medium (Daily)	Full mating season(s)	Strength, Affinity	[53]
Dominance Hierarchy Stability	Agonistic interactions, submission	Context-Dependent (Event-based)	Until hierarchy stabilizes	Eigenvector centrality, David's score	[2]
Long-Term Social Structure	Association, group membership	Low (Weekly-Monthly)	Years / Multiple seasons	Density, Centralization, Community structure	[38]

Table 2: Effect of Sampling Proportion on Network Metric Reliability

Network Metric	Robustness to Low Sampling Frequency	Robustness to Low Sampling Proportion	Uncertainty Assessment Method
Global Network Density	High	High	Bootstrapping Confidence Intervals [38]
Clustering Coefficient	Medium	Medium	Pre-network data permutation [38]
Node-Level Degree	Low	Medium-High	Correlation analysis across subsamples [38]
Betweenness Centrality	Low	Low	Node-level bootstrapping [38]
Eigenvector Centrality	Low	Low (Unreliable for 4/5 species [38])	Regression analysis against sampling proportion [38]

Experimental Protocol: A Five-Step Workflow for Assessing Temporal Resolution Adequacy

This protocol allows researchers to determine if their observational sampling regime is sufficient for obtaining reliable network metrics. The workflow is adapted from a validated framework for GPS-based telemetry data [38].

Diagram 1: Five-step workflow for assessing data adequacy.

Step 1: Testing for Non-Random Network Structure

Objective: To determine if the observed network structure captures significant, non-random social associations.

Methodology:

• Calculate Observed Metric: Compute your target global network metric (e.g., global clustering coefficient) from the fully sampled dataset.
• Generate Null Networks: Create a null distribution of the metric by repeatedly permuting the pre-network data stream. This involves randomly shuffling the timestamps of associations between individuals while preserving the overall rate of interaction.
• Statistical Testing: Compare the observed metric value against the null distribution. A significant difference (e.g., p < 0.05) indicates the network possesses non-random structure.

Interpretation: If a metric does not significantly differ from its null distribution, it should not be used for further inference, as it may not reflect true social patterning [38].

Step 2: Assessing Bias with Decreasing Sampling

Objective: To quantify how the value of a global network metric changes as the proportion of sampled individuals or observations decreases.

Methodology:

• Systematic Subsampling: From your full dataset, create progressively smaller subsets by randomly removing 10%, 20%, up to 50% of the observed data points (or tracked individuals).
• Metric Recalculation: For each subset, recalculate the global network metrics of interest.
• Bias Calculation: Plot the metric values against the sampling proportion. The trend line reveals the direction and magnitude of bias introduced by incomplete sampling [38].

Step 3: Estimating Uncertainty via Bootstrapping

Objective: To generate confidence intervals for global network statistics, enabling statistical comparison between networks (e.g., across seasons or populations).

Methodology:

• Bootstrap Resampling: Generate a large number (e.g., 1000) of new datasets by randomly sampling your original data with replacement.
• Distribution Construction: For each bootstrap sample, recalculate the global network metric. This creates an empirical distribution of the metric.
• Confidence Interval Derivation: Calculate the 95% confidence interval from the 2.5th and 97.5th percentiles of this bootstrap distribution [38].

Step 4: Evaluating Node-Level Metric Robustness

Objective: To determine how reliably node-level metrics (e.g., an individual's centrality) represent the true population value.

Methodology:

• Subsample and Correlate: Create multiple subsets of your data at various sampling proportions.
• Rank Correlation: For each node-level metric, calculate the correlation (e.g., Spearman's rank correlation) between the node values from the full dataset and the node values from each subset.
• Regression Analysis: Model the correlation coefficient as a function of sampling proportion. A strong, positive correlation even at low sampling proportions indicates a robust metric [38].

Step 5: Generating Node-Level Confidence Intervals

Objective: To provide estimates of uncertainty for individual-level social metrics, which can then be used as predictors in ecological models (e.g., for survival or habitat selection).

Methodology:

• Node-Level Bootstrapping: Apply a bootstrapping procedure specifically for each node's metric.
• Interval Calculation: Generate a confidence interval for each individual's network metric value (e.g., degree centrality) [38].
• Downstream Analysis: Use these interval estimates in subsequent models to account for measurement error in social connectivity.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Key Reagents and Analytical Solutions for Social Network Research

Tool / Solution	Function	Application Note
GPS Telemetry Tags	High-resolution spatiotemporal data collection.	Enables construction of proximity-based networks. Minimum frequency should be aligned with the species' typical interaction rate [38].
R Package aniSNA	Implements the 5-step assessment protocol.	Provides a standardized workflow for assessing bias and robustness in network metrics derived from telemetry or observational data [38].
Pre-Network Data Permutation	Generates null models for hypothesis testing.	Crucial for Step 1 to confirm that observed network structure is non-random [38].
Bootstrapping Algorithms	Quantifies uncertainty in network metrics.	Essential for Steps 3 and 5 to generate confidence intervals, allowing for statistical comparison of networks [38].
Bipartite Network Projection	Models participation in grouped events.	Useful for analyzing interactions in contexts like shared space use or co-attendance at specific locations [54].
Dyadic Regression Models (e.g., DyNAM)	Analyzes sequences of relational events.	Suitable for modeling the dynamics of network tie formation and dissolution over time [55].

Integrated Workflow: From Data Collection to Robust Inference

Diagram 2: Integrated workflow from data to ecological inference.

Selecting an optimal temporal resolution is not a one-size-fits-all process but a strategic decision that must align with the specific research question and biological system. The protocols outlined here provide a rigorous, quantitative framework for making this decision, moving beyond anecdotal guidance. By systematically assessing bias and uncertainty, researchers can determine which network metrics are reliable given their sampling regime and use these metrics with greater confidence in ecological and evolutionary inferences. This approach is indispensable for building a more robust, reproducible, and predictive science of animal social networks.

In animal behavior research, social network analysis (SNA) has emerged as a powerful tool for quantifying the complex social interactions that define animal societies [13]. Constructing a social network from behavioral observation data is only the first step; the critical subsequent challenge is validating that the inferred edges (social connections) and calculated network metrics accurately reflect biologically meaningful relationships and not just statistical artifacts [56]. This document outlines application notes and protocols for ensuring the biological relevance of network edges and metrics, framed within the context of a broader thesis on SNA in animal behavior research. These strategies are designed to help researchers, scientists, and drug development professionals build more reliable models for understanding social behaviors, which can be crucial for assessing welfare, evaluating therapeutic interventions, and advancing behavioral neuroscience.

Foundational Concepts of Network Validation

The Edge Validation Problem

In animal SNA, a network edge typically represents a specific behavioral interaction (e.g., grooming, aggression, proximity) observed between two individuals (nodes) [13]. The core validation challenge lies in distinguishing a statistically co-occurring interaction from a biologically significant social bond. For instance, two pigs might be recorded in proximity due to shared environmental attraction (e.g., a common feeder) rather than a specific social affinity [56]. Without proper validation, network metrics (e.g., centrality, clustering coefficient) calculated from these edges may be misleading, potentially leading to incorrect conclusions about social structure, hierarchy, or the impact of a pharmacological agent on group behavior.

Key Terminology and Metrics

Node: An individual animal within the study population. Edge/Network Edge: A documented interaction or association between two nodes. Centrality: A class of metrics (e.g., degree, betweenness) identifying an individual's importance or influence within the network. Causal Strength: A quantitative measure of the direct causal influence of one node on another, extending beyond simple correlation [57].

Protocols for Edge Validation

Protocol: Multi-Modal Edge Corroboration

This protocol ensures that an observed interaction is a robust indicator of a social bond by requiring confirmation from multiple data modalities.

I. Purpose To validate hypothesized social edges by confirming behavioral interactions across independent data streams, thereby reducing the likelihood that observed associations are spurious or environmentally driven.

II. Experimental Workflow

• Primary Interaction Definition: Define the core interaction constituting a network edge (e.g., direct physical contact, sustained proximity within a specified distance).
• Secondary Data Collection: Simultaneously collect data from at least one secondary behavioral modality:
  • Behavioral Context: Note the context of interactions (e.g., during feeding, resting, play).
  • Affiliative/Aggressive Acts: Record supportive (allogrooming, sharing food) or antagonistic (aggression, displacement) behaviors that coincide with or occur between the same individuals outside the primary interaction window.
  • Vocalization Analysis: Document if specific contact calls or other vocalizations accompany the interaction.
• Data Integration and Threshold Setting: Integrate data streams. Set a pre-defined threshold for validation (e.g., an edge is considered biologically validated if the primary interaction is accompanied by a secondary affiliative behavior in >20% of observed instances).

III. Key Materials and Reagents

• Automated Tracking System: Cameras and movement sensors for unbiased, continuous data collection on animal location and posture [56].
• Ethogram Software: Digital tool for coding and time-stamping specific behavioral acts.
• Audio Recording Equipment: For capturing and analyzing vocalizations associated with interactions.

Protocol: Temporal Stability Assessment

A true social bond should demonstrate persistence over time. This protocol assesses the stability of edges across multiple observation sessions.

I. Purpose To distinguish transient, situational interactions from stable social relationships by testing the repeatability of edges across time.

II. Experimental Workflow

• Longitudinal Data Collection: Conduct structured observation sessions of the same animal group at multiple time points (e.g., daily over a two-week period).
• Network Slicing: Construct a separate social network for each observation session.
• Edge Persistence Calculation: For each potential dyad (pair of individuals), calculate the proportion of observation sessions in which the edge was present.
• Threshold Application: Apply a consistency threshold (e.g., edges persisting in >60% of sessions are considered validated stable bonds). The specific threshold should be justified based on the species and research question.

Quantitative Metrics and Their Validation

Causal Inference for Edge Directionality

Many interactions in animal societies are directional (e.g., one individual grooms another). Simply observing co-occurrence does not reveal this directionality or causality. The Cross-Validation Predictability (CVP) algorithm provides a statistical method to infer causal strength from observed data, which is highly applicable to non-time-series behavioral data [57].

CVP Algorithm Workflow:

• Hypothesis Formulation: For two observed variables (behaviors of two animals, X and Y), test if X causes Y.
• Model Construction:
  • Null Model (Hâ‚€): Y = f̂(Ẑ) + ε̂. Predict Y using all other relevant variables (ZÌ‚) except X.
  • Alternative Model (H₁): Y = f(X, Ẑ) + ε. Predict Y using X and all other relevant variables (ZÌ‚).
• Cross-Validation: Both models are trained and tested using k-fold cross-validation to obtain prediction errors: ê (from Hâ‚€) and e (from H₁).
• Causal Strength Calculation: CS_{X→Y} = ln(ê / e). If CS > 0 and is statistically significant, a causal relationship from X to Y is inferred [57].

Table of Key Validation Metrics

The following table summarizes core quantitative metrics used for validating network edges and their biological significance.

Table 1: Key Metrics for Validating Network Edges and Biological Relevance

Metric Name	Definition	Application in Validation	Interpretation
Causal Strength (CS) [57]	CS = ln(ê / e), where ê and e are prediction errors from null and causal models, respectively.	Quantifies the direction and magnitude of a causal influence between two nodes (e.g., does individual A's behavior cause a change in individual B's behavior?).	A positive CS value suggests a causal relationship. Higher values indicate a stronger, more predictable causal influence.
Edge Persistence Index	Proportion of observation sessions in which a specific edge between two nodes is observed.	Measures the stability and reliability of a social bond over time, distinguishing it from random encounters.	Values closer to 1 indicate a stable, persistent social bond. Low values suggest a situational or transient interaction.
Multi-Modal Correlation Coefficient	Statistical correlation (e.g., Pearson's r) between the frequency of a primary interaction and the frequency of a secondary, affiliative behavior.	Validates that the edge of interest is correlated with other positive indicators of a social relationship.	A significant positive correlation provides evidence that the primary interaction is part of a broader affiliative context.
Heritability of Behavior	The proportion of observed variance in a behavioral trait that can be attributed to genetic factors [56].	Assesses whether a specific interaction behavior or social role has a genetic basis, supporting its status as a robust biological trait.	Heritability of 20-40% (as found for pig behavior [56]) indicates the behavior can be shaped by evolution and selection.

The Scientist's Toolkit: Research Reagent Solutions

This section details essential materials and tools for conducting robust social network analysis and validation in animal behavior research.

Table 2: Essential Research Materials and Tools for SNA Validation

Item	Function/Application	Specific Example/Note
Automated Behavioral Monitoring System	Provides high-resolution, unbiased data on individual location, movement, and posture for defining edges [56].	Systems typically comprise overhead cameras and RFID or UWB sensors. Enables tracking of proximity and activity 24/7.
AI-Based Pose Estimation Software	Automates the identification and classification of specific behavioral acts from video footage, standardizing edge definition.	Tools like DeepLabCut can be trained to identify species-specific behaviors like grooming, play, or aggression.
Causal Inference Software Library	Implements algorithms like CVP [57] or Granger Causality to infer directionality and causal strength from observed data.	Custom scripts in R or Python can be developed based on the CVP mathematical framework.
Ethogram Coding Software	Allows researchers to systematically record and time-stamp behaviors of interest during live or video observations.	Software like BORIS or Observer XT facilitates the structured data collection needed to build the initial network.
Network Analysis Platform	A computational environment for constructing networks, calculating metrics (centrality, density), and performing statistical tests.	Platforms include R (igraph, statnet), Python (NetworkX), and specialized commercial software.

Integrated Workflow for Comprehensive Validation

The following diagram integrates the key protocols and analyses into a single, coherent workflow for ensuring the biological relevance of network edges and metrics.

Integrated Validation Workflow

Application Note: Case Study in Livestock Research

Background: A study on commercial pigs used automated monitoring, AI, and SNA to understand social structures and their link to welfare and productivity [56].

Application of Validation Strategies:

• Edge Definition & Multi-Modal Data: Edges were defined based on proximity data from automated sensors. This primary data was correlated with behavioral annotations from video (a secondary modality) to identify "positive" versus "antagonistic" interactions [56].
• Temporal Assessment: Researchers found that pig social interactions became more structured over time, with key individuals emerging as central figures. This temporal stability helped validate that the observed edges represented established social hierarchies, not random encounters [56].
• Metric Validation via Heritability: The researchers determined that the observed behavioral traits had a heritability of 20-40% [56]. This provided strong evidence that the network metrics were capturing biologically grounded, genetically influenced traits, making them suitable for further research into breeding or management strategies.

Outcome: This validated SNA approach provided a data-driven method to reduce stress-related behaviors like tail-biting and to make more informed decisions on breeding and welfare-friendly practices [56].

Validating Connections: Robustness, Heritability, and Cross-Species Comparisons

Application Notes: The Robustness of Animal Social Networks to Methodological Variation

Core Conceptual Framework

Social network analysis has become an indispensable tool in behavioral ecology for quantifying the social structures of animal populations. These networks—composed of nodes (individual animals) and edges (social connections between them)—provide critical insights into population-level processes including information transmission, disease dynamics, and cultural evolution [45] [13]. A fundamental challenge in this field concerns how social relationships, which are often latent theoretical constructs, should be operationalized from observable data [2].

Researchers must navigate three distinct levels of abstraction when working with animal social networks: (1) the theoretical construct representing social relationships (e.g., affiliation, dominance); (2) the true interaction rates for specific behaviors; and (3) the measured interaction rates obtained through observation [2]. This hierarchy highlights the inherent methodological gap between theoretical concepts and empirical measurements, creating a central tension in the field: how significantly do methodological decisions affect the resulting network structures and subsequent ecological inferences?

Empirical Evidence of Robustness

Recent evidence demonstrates that animal social networks exhibit remarkable robustness to variations in how social associations are defined. A comprehensive 2025 study analyzing automatically recorded feeder visit data from four avian systems compared networks constructed using three different association definitions: strict time-window, group co-occurrence (GMM), and arrival-time approaches [45]. The findings revealed that networks built using different methods but applying ecologically relevant parameters showed similar structural characteristics, suggesting an underlying stability in social representation despite methodological variations [45] [58].

Critical to this robustness is the application of ecologically informed parameters regardless of the specific method employed. When association definitions align with the biological context of the study system—considering factors such as species-specific social behavior, flocking dynamics, and resource distribution—the resulting networks capture consistent social patterns [45]. This robustness persists across network analysis levels, from individual social traits to overall network topology, though subtle differences emerge that reflect both species biology and experimental design [45].

Experimental Protocols

Protocol: Comparing Association Definition Methods

Objective

To empirically assess the impact of different association definitions on social network structure using temporally-stamped co-occurrence data.

Materials and Equipment

• Automated data collection system (e.g., RFID-enabled feeders, GPS trackers)
• Temporal proximity data (timestamped individual identities)
• Computing environment with social network analysis capabilities (R recommended)
• Specialized R packages: 'asnipe' for GMM method, 'sna' or 'igraph' for network analysis

Procedure

Step 1: Data Collection and Preparation

• Deploy automated monitoring systems to record individual identities with precise timestamps
• Collect data over sufficient duration to capture representative social dynamics
• Pre-process data to ensure accurate individual identification and timestamp alignment

Step 2: Implement Association Definition Methods Apply three distinct association definitions to the same dataset:

• Strict Time-Window Method

  • Define a fixed temporal threshold (Δt) based on ecological context
  • Assign individuals to the same group if their visits occur within Δt seconds
  • Vary Δt in sensitivity analyses (e.g., 2s, 5s, 10s)

• Group Co-occurrence Method (GMM)

  • Implement Gaussian Mixture Models via 'asnipe' R package
  • Identify natural grouping events based on activity bursts
  • Use model parameters to detect discrete arrival/departure clusters

• Arrival-Time Method

  • Calculate time intervals between consecutive individual arrivals
  • Apply biologically-informed threshold to define coordinated arrivals
  • Associate individuals arriving within this critical window

Step 3: Network Construction

• Create adjacency matrices for each method where edges represent association frequencies
• Use simple ratio index or half-weight index to normalize association rates
• Construct undirected, weighted networks for comparative analysis

Step 4: Network Comparison and Validation

• Calculate key network metrics at individual and population levels
• Assess metric correlations across methods using Mantel tests
• Evaluate individual position consistency across methods
• Conduct sensitivity analyses on parameter choices

Protocol: Causal Inference Framework for Social Drivers

Objective

To establish causal relationships between phenotypic/ecological factors and social network structure while accounting for methodological biases.

Procedure

Step 1: Define Causal Estimands

• Specify individual-level (age, sex, personality), dyadic-level (kinship, familiarity), and group-level (population density, resource distribution) factors of interest
• Formalize causal questions using Directed Acyclic Graphs (DAGs) to represent assumed relationships

Step 2: Implement Bayesian Multilevel Modeling

• Develop Bayesian Social Relations Model extensions
• Incorporate varying effects for individual identities as both senders and receivers of social interactions
• Specify prior distributions based on ecological knowledge
• Use Markov Chain Monte Carlo (MCMC) sampling for posterior estimation

Step 3: Validate and Interpret Models

• Conduct posterior predictive checks to assess model fit
• Compare alternative model specifications
• Compute causal contrasts from joint posterior distributions
• Translate statistical parameters to structural causal parameters

Data Presentation

Comparative Analysis of Association Definition Methods

Table 1: Comparison of social network metrics across three association definition methods applied to four avian systems

Study System	Association Definition Method	Mean Degree	Mean Strength	Global Clustering	Network Density
Great Tit	Strict Time-Window (Δt=2s)	8.3	12.5	0.45	0.21
	Group Co-occurrence (GMM)	7.9	11.8	0.42	0.19
	Arrival-Time (10s threshold)	8.1	12.1	0.43	0.20
House Sparrow	Strict Time-Window (Δt=5s)	15.2	25.7	0.38	0.31
	Group Co-occurrence (GMM)	14.6	24.3	0.35	0.29
	Arrival-Time (15s threshold)	14.9	25.1	0.36	0.30

Table 2: Repeatability of individual social traits across methodological variations

Social Trait	Species	Within-Method Repeatability	Cross-Method Consistency	Method Effect Size
Network Degree	Great Tit	0.72	0.68	0.04
	House Sparrow	0.65	0.61	0.04
Network Strength	Great Tit	0.69	0.65	0.04
	House Sparrow	0.71	0.67	0.04
Betweenness Centrality	Great Tit	0.58	0.52	0.06
	House Sparrow	0.49	0.43	0.06

Visualization

Methodological Workflow for Association Definition

Causal Inference Framework for Social Networks

The Scientist's Toolkit

Essential Research Reagents and Methodological Solutions

Table 3: Key methodological tools for robust social network construction

Tool Category	Specific Solution	Function & Application	Considerations
Data Collection Systems	RFID Feeder Arrays	Automated recording of individual visits with precise timestamps	Bridge et al. 2019 design enables continuous monitoring of small birds
	GPS Tracking Units	Spatial proximity data for wide-ranging species	Battery life and accuracy trade-offs must be considered
Association Definition Algorithms	Strict Time-Window (Δt)	Simple threshold-based association definition	Highly sensitive to threshold choice; requires ecological validation
	Gaussian Mixture Models (GMM)	'asnipe' R package identifies natural grouping events	Optimized for fission-fusion systems; may struggle with highly gregarious species
	Arrival-Time Method	Defines associations based on coordinated arrival patterns	Captures movement coordination; effective for gregarious species
Analytical Frameworks	Bayesian Social Relations Model	Multilevel modeling of social effects with uncertainty quantification	Accounts for network autocorrelation and sampling biases
	Causal Inference Framework	Directed Acyclic Graphs (DAGs) and Structural Causal Models	Distinguishes causal effects from spurious correlations
Validation Approaches	Mantel Tests	Matrix correlations for cross-method comparison	Assesses overall network structure similarity
	Individual Trait Repeatability	Quantifies consistency of social positions across methods	Reveals method-dependent variations in social trait estimation

Understanding the genetic architecture that underpins social behavior is a fundamental pursuit in behavioral ecology and neuroscience. Research consistently demonstrates that complex social phenotypes, including an individual's position within a social network and their overall sociability, are not merely products of environment and experience but are also significantly influenced by genetic variation [59] [60]. This application note situates itself within the broader context of social network analysis (SNA) in animal behavior research, providing a methodological framework for investigating the heritability of social behavioral traits. By quantifying social interactions as networks—where individuals are nodes and their interactions are edges—researchers can leverage powerful analytical tools to dissect the genetic contributions to sociality [44] [13]. The protocols herein are designed for researchers, scientists, and drug development professionals aiming to bridge the gap between observational behavioral data and genetic analysis, facilitating the discovery of genetic markers and biological pathways associated with social behavior.

Quantitative Foundations: Heritability Estimates and Genetic Effect Sizes

Compilation of data from quantitative genetic studies, particularly those using genome-wide complex trait analysis (GCTA) and quantitative trait locus (QTL) mapping, provides robust evidence for the heritability of social behaviors across species. The tables below summarize key quantitative findings essential for framing experimental hypotheses and power analyses.

Table 1: Heritability Estimates (h²) for Social Behaviors in Non-Human Primates

Species	Behavioral Phenotype	Heritability (h²) Estimate	Measurement Method	Citation
Rhesus Macaque	Spontaneous Social Behaviors (Composite)	0.17 - 0.53	GCTA/Ethogram	[61]
Rhesus Macaque	jmSRS Score (Atypical Sociality)	~0.29	Adapted Social Responsiveness Scale	[61]
Rhesus Macaque	Mutual Eye Gaze	~0.45	Focal Observation	[61]

Table 2: Effect Sizes of Behavioral QTLs Across Animal Taxa

Behavioral Category	Average Effect Size (% Variance Explained)	Notes	Citation
Courtship & Feeding	~30%	Significantly greater (approx. 3x) than other behaviors	[59]
Other Behaviors	~10%	Includes aggression, locomotion, etc.	[59]
General Conclusion	Most behavioral architectures fit an exponential distribution: a few loci of moderate-to-large effect and many with small effects.	[59]

Core Experimental Protocols

Protocol 1: Quantifying Social Networks in a Group-Living Species

This protocol outlines the steps for constructing a social network from raw behavioral observations, a prerequisite for generating the social phenotypes (e.g., network centrality) used in genetic analyses [44] [62].

1. Node and Edge Definition: - Define the Population: Identify all individuals in the study group. Each individual becomes a node in the network. - Define the Social Interaction: Choose a biologically relevant behavior that defines an edge. This could be grooming, spatial proximity (e.g., within 1 meter), aggression, or food sharing. Edges can be directed (e.g., who initiates grooming) or undirected (e.g., mere proximity) [44] [13].

2. Data Collection: - Method: Use focal animal sampling or group scans with a predefined sampling period. Automated tracking via RFID or computer vision is ideal for high-resolution data. - Duration: Sample for a sufficient period to capture representative social dynamics. The sampling period must be justified biologically [44]. - Data Recorded: For each interaction, record the identities of the two individuals, the time, and the duration or frequency.

3. Association Index Calculation: - Construct a socio-matrix where cells represent the strength of the relationship between each dyad. - Use an appropriate association index to control for observation bias. Common indices include the Simple Ratio Index (SRI) or Half-Weight Index (HWI), which calculate the proportion of samples two individuals were observed associating relative to the total number of sampling events they were both observed [44] [62].

4. Network Construction and Metric Extraction: - Input the socio-matrix into network analysis software (e.g., R igraph, asnipe; Python NetworkX). - Calculate node-level metrics for each individual. Key metrics for heritability studies include: - Strength: The sum of edge weights for a node; a measure of overall gregariousness. - Betweenness Centrality: The number of shortest paths that pass through a node; a measure of brokerage or connectedness across the network. - Clustering Coefficient: The probability that two of an individual's associates are themselves connected; a measure of clique formation [44] [63] [13].

Protocol 2: Assessing Heritability Using a Pedigreed Population

This protocol describes a standard quantitative genetic approach to estimate the proportion of phenotypic variance in a social trait attributable to genetic variance.

1. Study Population and Phenotyping: - Establish or utilize a population with a known pedigree. Free-ranging rhesus macaque troops or laboratory-bred lines of other species (e.g., mice, cockroaches) are common models [63] [61]. - Quantify social phenotypes for all individuals using the methods in Protocol 1. The jmSRS (juvenile macaque Social Responsiveness Scale) is an example of a validated composite score for atypical social behavior [61].

2. Statistical Modeling: - Use a linear mixed model (LMM) to partition the variance. The model can be structured as: Phenotype = µ + Fixed_Effects + a + e where a is the random additive genetic effect and e is the residual error. - Fixed effects like sex, age, and body mass must be included to control for confounding variables [63]. - The model is typically fitted using Bayesian (MCMCglmm) or Restricted Maximum Likelihood (REML) methods [59] [61].

3. Heritability Calculation: - Narrow-sense heritability (h²) is calculated as: h² = V<sub>A</sub> / V<sub>P</sub> where V<sub>A</sub> is the additive genetic variance and V<sub>P</sub> is the total phenotypic variance. - Significance is assessed by comparing the model to a null model without the genetic effect via likelihood-ratio tests or by examining the credibility intervals of the posterior distribution in a Bayesian framework [59] [61].

Protocol 3: Genotyping and Genome-Wide Association Study (GWAS)

For identifying specific genetic variants associated with social phenotypes.

1. DNA Collection and Genotyping: - Collect biological samples (e.g., blood, hair, saliva) from phenotyped individuals. - Perform high-density genotyping (e.g., SNP arrays) or whole-exome/whole-genome sequencing [59] [61].

2. Quality Control (QC): - Apply standard genomic QC filters: remove individuals and SNPs with high missing rates, exclude SNPs with low minor allele frequency (MAF), and check for Hardy-Weinberg equilibrium deviations.

3. Association Analysis: - For each SNP, test for association with the social network phenotype (e.g., strength, betweenness) using a linear regression model, typically including the genetic relationship matrix as a random effect to account for population structure (a Mixed Linear Model, MLM) [59]. - Causal inference techniques, such as those using Directed Acyclic Graphs (DAGs), are recommended to disentangle true causal effects from spurious correlations and confounders [2].

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Reagents and Resources for Social Behavior Genetics

Item/Tool	Function/Description	Example/Reference
Animal Social Network Repository (ASNR)	A multi-species repository of social networks for comparative analysis and hypothesis generation.	[62]
GraphML Format	A flexible XML-based file format for storing network structure and attributes; facilitates data sharing and reproducibility.	[62]
Proximity Loggers (RFID)	Automated data collection of spatial associations between individuals, minimizing observer bias.	[62]
jmSRS (juvenile macaque SRS)	A validated behavioral scale adapted from human research to quantify atypical social behavior in macaques.	[61]
Directed Acyclic Graph (DAG)	A causal modelling tool to formally represent and test assumptions about the drivers of social network structure.	[2]
MCMCglmm R Package	A software tool for fitting Bayesian generalized linear mixed models, ideal for estimating variance components and heritability.	[59]
DGRP (Drosophila Genetic Reference Panel)	A public resource of fully sequenced inbred D. melanogaster lines for powerful genotype-phenotype mapping.	[59]

Advanced Framework: Causal Inference in Social Networks

Observing a correlation between a genetic variant and a social phenotype is not sufficient to claim causation. Confounding factors, such as shared environment or other social processes, can create spurious associations [2]. A formal causal inference framework is essential for robust conclusions.

Key Steps:

• Define the Causal Estimand: Precisely define the effect you wish to measure (e.g., the total causal effect of a specific genotype on network betweenness).
• Construct a DAG: Build a Directed Acyclic Graph (DAG) that encodes your qualitative assumptions about the causal relationships between variables in the system (e.g., genotype, social phenotype, environment, sex).
• Identify the Estimand: Use graphical criteria (e.g., the backdoor criterion) to determine if the causal effect is identifiable from the observed data and what set of variables need to be adjusted for in the statistical model.
• Build a Generative Model: Implement a Bayesian multilevel model (e.g., an extension of the Social Relations Model) that incorporates the structure of the social network and the identified adjustment set to estimate the causal effect [2].

Application Notes: Core Concepts and Quantitative Comparisons

Social Network Analysis (SNA) provides a quantitative framework for comparing animal social structures across wild, captive, and domesticated contexts. Understanding these differences is critical for ecological research, conservation, and ensuring the welfare of animals in managed care [64] [65].

Table 1: Key Structural Differences in Meerkat Social Networks (Wild vs. Captive)

Network Metric	Wild Meerkats	Captive Meerkats	Interaction Type	Implications
Average Path Length	Longer	Shorter	Dominance, Foraging	Simplified connection pathways in captivity [64]
Network Density	Lower	Higher	Grooming	Reduced partner choice, forced associations in captivity [64]
Saturation	Higher	Lower	Dominance, Foraging	Altered intensity of competitive interactions [64]
Assortativity by Sex	Present	Differs/Wild Pattern Disrupted	Grooming	Increased intrasexual conflict in captivity due to inability to disperse [64]

Table 2: Contrasting Features of Wild, Captive, and Domesticated Social Systems

Feature	Wild Systems	Captive Systems	Domesticated Systems
Primary Selection Pressure	Natural & Sexual Selection [65]	Artificial (Housing/Husbandry) & Natural [64]	Strong Artificial Selection [65]
Key SNA Consideration	Dynamic, ecologically driven networks [7]	Static, human-managed networks; frequent group membership changes [64]	Genetically predisposed tolerance & human-directed interaction [65]
Typical Group Size	Larger (e.g., wild meerkats) [64]	Smaller (e.g., captive meerkats) [64]	Varies by human purpose & husbandry
Defining Characteristic	Heritable predisposition for human association absent [65]	Dependence on humans for food/shelter; genetic predisposition for human association may be present [65]	Permanent genetic modification for tameness & human association; human-controlled breeding [65]

Experimental Protocols

Protocol 1: Data Collection for Comparative SNA (Wild & Captive)

This protocol outlines a standardized method for collecting social interaction data, adaptable for both wild and captive settings, based on studies of meerkats [64].

I. Pre-Observation Planning

• Group Selection: Identify study groups with a mixture of sexes and ages. For a representative sample, study multiple groups (e.g., 8 wild and 15 captive groups) [64].
• Ethical Approval: Secure all necessary permits and IACUC approvals prior to data collection [66].
• Individual Identification: Ensure all animals in the study group are individually identifiable (e.g., natural markings, bands, tags).

II. Data Collection Procedure

• Observation Schedule: Conduct observations during active animal hours. In captive settings, observe during public hours (e.g., 8:00-17:00) to account for visitor effects [64].
• Sampling Method: Use focal animal sampling or all-occurrence sampling for specific behaviors.
• Recording Interaction Types: Document the following interactions for each dyad (pair of animals):
  • Agonistic/Dominance Interactions: Chasing, biting, forced retreats [64].
  • Affiliative Interactions: Allo-grooming, huddling, cooperative behaviors [64].
  • Foraging Interactions: Co-feeding, food-begging, or competitive displacements at food sources [64].
• Session Duration: Conduct multiple observation sessions per group, totaling several hours of data per group.

III. Post-Collection Data Structuring

• Compile data into an adjacency matrix for each interaction type, where rows and columns represent individuals, and cell values represent the frequency or duration of interactions.

Protocol 2: Social Defeat Stress Model for Investigating Depression-like Syndromes

This protocol details a standardized method for inducing a clear depressive-like syndrome in mice using repeated social defeat, useful for preclinical drug development [67].

I. Selection and Housing of Aggressor Mice

• Aggressor Mice: Select sexually experienced, aggressive male CD-1 mice.
• Screening: House CD-1 mice alone for several days. Introduce a C57BL/6J intruder for 5 minutes over 3 consecutive days. Retain only CD-1 mice that consistently display aggressive bouts within 1 minute of intrusion [67].
• Housing: House aggressor mice in large, transparent cages divided by a clear, perforated divider.

II. Social Defeat Sessions

• Subject Mice: Use C57BL/6J male mice (7-8 weeks old).
• Defeat Procedure: Once daily, introduce a C57BL/6J mouse into the home cage of an aggressor CD-1 mouse for 5-10 minutes, or until approximately 10 bouts of physical subjugation occur.
• Sensory Contact: After the physical confrontation, separate the mice using the clear, perforated divider, allowing for continuous sensory contact for the remainder of the 24-hour period.
• Duration: Repeat this process for 10 days, introducing the subject mouse to a different aggressor each day to prevent habituation [67].

III. Social Interaction Test

• Apparatus: Use an open arena with an empty, perforated enclosure at one end.
• Habituation: Place the experimental mouse in the arena for 2.5 minutes without a social target.
• Test Phase: Confine an unfamiliar, aggressive CD-1 mouse in the enclosure and place the experimental mouse back in the arena for 2.5 minutes.
• Automated Tracking: Use video tracking software (e.g., EthoVision) to quantify time spent in the "interaction zone" (area surrounding the enclosure).
• Classification: Mice that spend significantly less time in the interaction zone when the social target is present are classified as "susceptible," displaying social avoidance. Those that do not are classified as "resilient" [67].

Visualizations: Workflows and Logical Relationships

Social Defeat Stress Experimental Workflow

Dynamic Social Network Analysis (SNA) Logic

The Scientist's Toolkit

Table 3: Essential Research Reagents and Materials

Item	Function/Application
Video Tracking System (e.g., EthoVision)	Automated quantification of animal movement and behavior during social tests (e.g., time in interaction zone) [67].
R Statistical Environment with RSiena Package	Fits Stochastic Actor-Oriented Models (SAOMs) to analyze network dynamics and co-evolution with traits over time [7].
Standardized Social Defeat Arenas	Apparatus for social interaction testing, consisting of an open field with a removable, perforated enclosure for a social target [67].
Environmental Enrichment Items	Species-specific items (e.g., nesting material, huts, foraging treats) used in captive studies to promote natural behaviors and assess impact on social networks [66].
C57BL/6J Inbred Mouse Strain	Standardized subject strain for social defeat and other behavioral neuroscience models due to well-characterized responses [67].
Aggressor CD-1 Mice	Larger, aggressive mice screened for consistent offensive behavior, used as social stressors in the defeat model [67].

Social Network Analysis (SNA) provides a quantitative framework for quantifying and analyzing the social structures within animal groups. By treating individuals as "nodes" and their interactions as "edges," SNA maps the complex web of relationships, revealing key social roles, information flow pathways, and group dynamics [68] [69]. In both zoo and agricultural settings, understanding these social structures is crucial for making evidence-based management decisions that enhance animal welfare. This approach moves beyond simple counts of aggression or affiliation to a holistic analysis of the social environment, allowing caregivers to identify sources of social stress, predict the impacts of management changes, and promote positive welfare states [68] [70]. When integrated with modern sensing technologies, SNA transforms from a purely research-oriented tool into a practical component of daily animal care and welfare assessment [69].

Application Notes

The core application of SNA in managed animal settings is to move from a general, species-level understanding of behaviour to a precise, group-specific, and individual-oriented analysis. This "bottom-up" approach is fundamental for modern, individualised animal care [68].

Key Applications in Zoo Settings

In zoos, SNA helps manage often small, stable, and valuable groups of animals.

• Identifying Social Pressures and Roles: SNA can identify individuals who are frequent targets of aggression or who are socially isolated. A case study on black lemurs (Eulemur macaco) used SNA to reveal patterns of female dominance and identify one male as the primary recipient of aggressive interactions [68].
• Informing Social Management: Understanding social bonds is critical for decisions related to introductions, separations, or translocations. Removing a central, well-connected individual may destabilise a group more than removing a peripheral one [70].
• Guiding Environmental Improvements: Data on social dynamics can inform enclosure design, including the provision of visual barriers, multiple feeding stations, and complex terrain to reduce conflict over resources and allow subordinates to avoid aggressors [68].

Key Applications in Agricultural Settings

In agriculture, SNA is increasingly combined with sensor technology to monitor large groups and improve welfare in a production context.

• Monitoring Health and Welfare: Changes in social behaviour can be early indicators of disease or poor welfare. One study cited that cows with metritis altered their social competition strategies at the feed bunk, a change detectable through SNA [69].
• Understanding Subtle Behaviours: SNA helps decipher ambiguous behaviours. For instance, licking in cows can be either affiliative (e.g., a mother to her calf) or antagonistic, and prolonged sensor-based observation allows for accurate categorisation within the social network [69].
• Optimising Farm Management: Insights from SNA can guide management practices related to space allowance, group composition, and resource distribution to minimise aggression and stress, thereby improving overall welfare and potentially productivity [69].

Experimental Protocols

Implementing SNA requires a structured approach from data collection to analysis. The following protocols provide a standardised methodology.

Protocol 1: SNA for Zoo-Housed Primates

Objective: To map affiliative and agonistic networks to identify social roles and potential welfare concerns within a stable social group.

• Behavioural Ethogram Definition: Define and operationalise behaviours.
  • Affiliative Behaviours: Social grooming, social play, proximity within one body length, co-feeding.
  • Agonistic Behaviours: Chase, displace, threat display, physical contact aggression, submission (flee, avoid).
• Data Collection:
  • Method: Focal animal sampling with continuous recording.
  • Schedule: Observe each individual for a minimum of 30 minutes per day, distributed across at least five different days, including active and inactive periods.
  • Data Recorded: For each interaction, record the initiator, receiver, behaviour type, and time.
• Data Analysis:
  • Network Construction: Create separate adjacency matrices for affiliative and agonistic interactions. Rows and columns represent individuals, and cell values represent the frequency or duration of interactions.
  • SNA Metrics: Calculate key metrics using software like UCINET or R (igraph, sna packages).
• Interpretation: Identify central individuals in both networks. Correlate an individual's network position with independent welfare indicators (e.g., body condition, glucocorticoid levels). Use the results to guide management, such as providing additional escape routes for frequent targets of aggression [68].

Protocol 2: Sensor-Based SNA for Livestock

Objective: To automatically quantify social associations and interactions in a pen of dairy cows to identify changes indicative of health or welfare issues.

• Sensor Deployment: Fit all study animals with wearable sensors (e.g., ultra-wideband RFID or GPS collars) that record location at high temporal resolution (e.g., every second).
• Data Collection:
  • Duration: Collect data continuously over a period of at least one week to establish a baseline.
  • Validation: Conduct simultaneous video recording for a subset of 24 hours to validate that proximity data from sensors corresponds to observable social interactions.
• Data Processing and Network Construction:
  • Define an Association: Define a "social interaction" or "association" based on sensor data (e.g., individuals within 2 meters of each other).
  • Create Networks: Construct undirected, weighted networks where the edge weight between two cows is the total time they spent in proximity during the observation period.
• Analysis and Intervention:
  • Calculate network metrics for each cow.
  • Monitor these metrics over time. A significant change in an individual's social connectivity (e.g., a sudden drop) can trigger a health check [69].

Quantitative SNA Metrics for Welfare Assessment

The following table summarizes key SNA metrics and their relevance to animal welfare assessment.

Table 1: Key Social Network Analysis Metrics and Their Welfare Implications

Metric Name	Description	Interpretation for Animal Welfare
Degree Centrality	The number of direct connections an individual has.	High degree may indicate social integration; low degree may suggest isolation or ostracism [68].
Betweenness Centrality	The extent to which an individual lies on the shortest path between other individuals.	High betweenness individuals act as "brokers" in the network; their removal could fragment the group [70].
Eigenvector Centrality	A measure of an individual's influence based on the influence of its connections.	Identifies individuals well-connected to other well-connected individuals, potentially key for social stability and information spread.
Edge Weight/Strength	The frequency or duration of interactions between a pair of individuals.	Strong ties represent primary social bonds, which are critical for buffering stress and promoting positive welfare [70].

Visualization of SNA Workflows

The following diagrams illustrate the logical workflow for implementing SNA in animal welfare studies, integrating both traditional and sensor-based approaches.

Figure 1: A cyclical workflow for applying SNA to zoo animal management, from defining objectives to implementing and monitoring interventions.

Figure 2: A workflow for implementing automated, sensor-based SNA in a livestock setting to monitor welfare and detect issues.

The Scientist's Toolkit

Implementing a robust SNA study requires a combination of hardware and software tools.

Table 2: Essential Research Reagents and Tools for SNA in Animal Behaviour

Tool Name/Category	Function/Purpose	Specific Examples & Notes
Data Collection Hardware	To record the raw data on individual locations, identities, and/or interactions.	Wearable sensors (UWB, GPS, accelerometers), stationary cameras (for video tracking), RFID feeders. Essential for automated, high-resolution data [69].
Behavioural Coding Software	To facilitate the recording and organisation of observational data from video or live observation.	BORIS, Observer XT. Allows for the operationalisation of an ethogram and structured data entry for later matrix construction.
Social Network Analysis Software	To construct social networks from interaction matrices and calculate key SNA metrics.	R packages (igraph, sna, asnipe), UCINET with NetDraw. These are the core analytical engines for quantitative SNA [68] [13].
Ethogram	A predefined list of behaviours and their definitions that standardises data collection.	Must be species-specific and context-specific. Includes affiliative (e.g., allogrooming) and agonistic (e.g., chase) behaviours. The foundational "reagent" for any behavioural study [68].

Application Notes: Integrating Animal Social Network Analysis into Translational Research

Social network analysis (SNA) provides a powerful quantitative framework for understanding the structure and dynamics of animal societies. The core of this approach involves representing social systems as networks composed of nodes (individual animals) connected by edges (social interactions or associations) [44]. When properly applied, this methodology offers unparalleled insights into how social factors influence health, disease transmission, and aging across species—findings with significant translational potential for human health.

Key Translational Insights from Animal Social Research

Linking Social Structure to Health and Aging Mechanisms: Research using rhesus macaques has demonstrated that social network size directly correlates with the expansion of specific brain circuits, providing a potential neurological mechanism for the social determinants of health and aging [71]. Longitudinal studies in red deer have revealed that social connectivity decreases with age, as older individuals move to increasingly isolated areas, mirroring patterns of social isolation observed in aging human populations [71].

Methodological Innovations for Complex Social Systems: Modern animal SNA employs sophisticated statistical toolkits and software packages like the Animal Network Toolkit Software (ANTs), which enables researchers to compute global, polyadic, and nodal network measures; perform data randomization; and conduct statistical permutation tests for both static and temporal networks [72]. These tools allow for testing hypotheses about how individual attributes, sociodemographic characteristics, and ecological pressures shape social relationships and their health consequences [72].

Epistemological Considerations and Limitations

The translational value of animal social research must be considered within a broader epistemological framework. Animal studies traditionally occupy the lower tiers of the evidence hierarchy in biomedical research, with significant challenges in translating findings to human applications [73]. Systematic reviews have highlighted that fewer than 15% of clinical trials successfully progress beyond phase I in areas like cancer research, despite promising preclinical results in animal models [73]. These limitations underscore the importance of robust methodology and careful interpretation when bridging animal social findings to human health paradigms.

Table 1: Key Social Network Metrics and Their Potential Health Correlates

Network Metric	Definition	Potential Health Correlation
Degree Centrality	Number of direct connections an individual maintains	Associated with immune function, stress response, and disease susceptibility
Betweenness Centrality	Extent to which an individual connects otherwise disconnected groups	Potential indicator of social stress or information brokerage position
Network Density	Proportion of possible connections that actually exist	Group-level indicator of disease transmission potential or social support availability
Eigenvector Centrality	Influence of an individual based on their connections' influence	Potential correlate of social status and associated health benefits

Experimental Protocols for Animal Social Network Research

Protocol 1: Assessing Robustness of Social Network Metrics from Telemetry Data

Purpose: To establish a standardized protocol for evaluating bias and robustness of social network metrics derived from GPS-based telemetry data, particularly when monitoring limited individuals within a population [38].

Workflow:

• Determine Non-Random Network Structure: Generate null networks by permuting pre-network data streams to confirm that observed network metrics capture non-random aspects of social association.
• Assess Bias in Global Metrics: Systematically sub-sample the observed network to quantify how bias in global summary statistics varies with decreasing proportions of sampled individuals.
• Evaluate Alternative Sampling Scenarios: Apply bootstrapping techniques to subsamples to model how network properties would differ if an entirely different set of individuals had been tagged.
• Analyze Node-Level Metric Robustness: Use correlation and regression analyses to determine how node-level network metrics are affected by the proportion of individuals present in the sample.
• Generate Confidence Intervals: Employ bootstrapping to generate confidence intervals for each node's individual network metric values, enabling quantification of uncertainty [38].

Figure 1: Protocol for assessing the robustness of social network metrics from partial population data.

Validation: This protocol was validated using fallow deer populations with known population size where approximately 85% of individuals were directly monitored, demonstrating that global network metrics like density remain robust even with lowered sample sizes, while local metrics like eigenvector centrality show greater variability [38].

Protocol 2: Multilevel Analysis of Animal Social Networks Using ANTs R Package

Purpose: To provide a comprehensive workflow for analyzing animal social networks across multiple levels of organization using the specialized ANTs R package [72].

Data Input Specifications:

• Interaction/Association Data: Can be input as matrices or data frames in edge list format (for network permutations) or in specialized formats for data stream permutations (focal or group follow observations).
• Individual Attributes Data: Data frames with rows for each individual and columns for attributes (sex, age, dominance rank, hormone levels, etc.).

Analytical Workflow:

• Data Preprocessing: Format raw observational data according to ANTs requirements, ensuring proper coding of actors, receivers, weights, and control factors.
• Network Construction: Build networks from interaction data using appropriate association indices and sampling periods.
• Metric Computation: Calculate global, node-level, and polyadic network measures using ANTs' optimized functions.
• Statistical Testing: Perform permutation-based tests (correlation tests, t-tests, GLMs, GLMMs) to assess significance while accounting for non-independent data.
• Temporal Analysis: Conduct time-aggregated network analyses to track structural changes over time.
• Visualization: Generate network visualizations incorporating individual attributes and network metrics.

Figure 2: Multilevel analysis workflow using the ANTs R package for animal social networks.

Key Advantages: ANTs outperforms existing R packages in computation speed for network measures and permutations, provides specialized functions for animal behavior research, and integrates procedures that previously required switching between multiple software packages [72].

Table 2: Research Reagent Solutions for Animal Social Network Analysis

Tool/Category	Specific Examples	Function/Application
Data Collection Technologies	GPS telemetry collars, proximity loggers, automated tracking systems	Capture high-resolution movement and association data with minimal disturbance
Specialized Software	ANTs R package, SOCPROG, asnipe, igraph	Compute network metrics, perform permutations, and conduct statistical analyses
Statistical Frameworks	Data stream permutations, node label permutations, bootstrap methods	Account for non-independent data and sampling biases in network analysis
Validation Protocols	Sampling bias assessment, metric robustness evaluation, confidence interval estimation	Ensure reliability and interpretability of network metrics derived from partial sampling

Translational Framework: From Animal Social Systems to Human Health

Bridging the Translational Gap

The integration of animal social network analysis into human health research requires careful consideration of comparative validity and mechanistic pathways. The National Institutes of Health has recently prioritized human-based research technologies while acknowledging the continued importance of animal models for specific research questions [74]. This evolving research landscape emphasizes the need for robust translational frameworks that can effectively bridge findings across species.

Strategic Considerations for Translational Applications:

• Model Selection: Choose animal models with social systems relevant to the human health question under investigation. Research networks are actively exploring diverse models from rhesus macaques to ungulates and birds [71] [75].
• Mechanistic Focus: Prioritize research on conserved physiological pathways (e.g., stress response systems, neuroimmune pathways) that may mediate social determinants of health across species.
• Methodological Alignment: Standardize social network metrics and analytical approaches to enable direct comparison between animal and human studies.
• Ethical Framework: Implement ethical considerations that acknowledge both the instrumental value of animal research for human benefit and the intrinsic value of animal well-being [73].

The ongoing development of innovative animal models—such as naked mole-rats for studying longevity and cancer resistance, and birds for understanding neuroprotective mechanisms—continues to provide unique insights into fundamental biological processes with direct relevance to human health [75]. By applying rigorous social network methodologies to these models, researchers can build robust biomedical bridges that translate animal social findings into novel paradigms for human health research.

Conclusion

Social Network Analysis provides a powerful, quantitative framework for understanding the complex architecture of animal societies, revealing how individual interactions scale to population-level patterns with significant implications for health, disease transmission, and welfare. The integration of advanced technologies like AI and sensor systems with sophisticated analytical models such as SAOMs has revolutionized our capacity to capture dynamic social processes. While methodological challenges around association definition and data quality persist, studies demonstrate remarkable robustness in social structures across species and contexts. For biomedical research, animal SNA offers valuable models for understanding social determinants of health, disease spread dynamics, and the neurobiological underpinnings of social behavior. Future directions should focus on standardizing methodologies across systems, exploring the genetic architecture of sociality, and leveraging these insights to develop innovative approaches to managing social stress, enhancing welfare, and informing public health strategies.

References

• 1. Social Network Graphs: Concepts, Metrics & Tools [https://www.puppygraph.com/blog/social-network-graphs]
• 2. A causal framework for the drivers of animal social network ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12459825/]
• 3. Animal social networks as intersections graphs of random ... [https://arxiv.org/html/2511.11130v1]
• 4. Social connection linked to improved health and reduced ... [https://www.who.int/news/item/30-06-2025-social-connection-linked-to-improved-heath-and-reduced-risk-of-early-death]
• 5. Different types of social links contrastingly shape ... [https://link.springer.com/article/10.1007/s00265-025-03594-4]
• 6. Non‐breeding social behaviour as an investment in reducing ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12586285/]
• 7. Analysing animal social network dynamics - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC6849756/]
• 8. Research and Practice [https://www.social-connection.org/what-we-do/research-and-practice/]
• 9. OECD Guidelines on Measuring Subjective Well-being ... [https://www.oecd.org/en/publications/oecd-guidelines-on-measuring-subjective-well-being-2025-update_9203632a-en.html]
• 10. Experimental evidence that network topology can accelerate ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10693003/]
• 11. The Impact of Evolutionary Driving Forces on Human Complex ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4904122/]
• 12. Compressing network populations with modal ... [https://www.nature.com/articles/s42005-023-01270-5]
• 13. Conceptual representations of animal social networks [https://www.sciencedirect.com/science/article/pii/S0003347223001100]
• 14. Micro and Macro Level Processes | Research Starters [https://www.ebsco.com/research-starters/social-sciences-and-humanities/micro-and-macro-level-processes]
• 15. Use of social network analysis methods to study professional ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5651641/]
• 16. Structure Data for Analysis [https://help.tableau.com/current/pro/desktop/en-us/data_structure_for_analysis.htm]
• 17. Evaluating empirical contact networks as potential ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5014053/]
• 18. Elements must meet minimum color contrast ratio thresholds [https://dequeuniversity.com/rules/axe/4.8/color-contrast]
• 19. Color contrast - Accessibility - MDN Web Docs - Mozilla [https://developer.mozilla.org/en-US/docs/Web/Accessibility/Guides/Understanding_WCAG/Perceivable/Color_contrast]
• 20. Stephen Few: Show Me the Numbers; Designing Tables ... - Principus [https://principus.si/2021/03/22/stephen-few-show-me-the-numbers-designing-tables-and-graphs-to-enlighten/]
• 21. The quest to develop automated systems for monitoring ... [https://www.sciencedirect.com/science/article/pii/S0168159123001727]
• 22. Wearable Sensors-Based Intelligent Sensing and ... [https://www.mdpi.com/1424-8220/25/14/4515]
• 23. JAX Animal Behavior System (JABS): A genetics informed ... [https://elifesciences.org/reviewed-preprints/107259]
• 24. Machine Learning-Based Sensor Data Fusion for Animal ... [https://www.mdpi.com/1424-8220/23/12/5732]
• 25. Editorial: Advances in the application of technology for ... [https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2025.1715133/full]
• 26. A comparison of machine learning multiband sensor fusion ... [https://www.spiedigitallibrary.org/conference-proceedings-of-spie/13454/3051994/A-comparison-of-machine-learning-multiband-sensor-fusion-models-for/10.1117/12.3051994.full]
• 27. Introduction into SAOM and REM [http://rstudio-pubs-static.s3.amazonaws.com/362623_93d48d7f331a4c77908e9d35535187a3.html]
• 28. Social network analysis [https://en.wikipedia.org/wiki/Social_network_analysis]
• 29. Effect of behavioural sampling methods on local and global ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10698479/]
• 30. Preferential associations in an unstable social network [https://pmc.ncbi.nlm.nih.gov/articles/PMC10267343/]
• 31. A Multiplex Centrality Metric for Complex Social Networks [https://pubmed.ncbi.nlm.nih.gov/32211232/]
• 32. Make Your Network Maps Shine with 16 New Analyzer ... [https://visiblenetworklabs.com/2025/03/06/make-your-network-maps-shine-with-16-new-analyzer-color-palettes/]
• 33. A natural experiment of social network formation and dynamics [https://pmc.ncbi.nlm.nih.gov/articles/PMC4450406/]
• 34. Revealing the Hidden Social Structure of Pigs with AI ... [https://www.mdpi.com/2076-2615/15/7/996]
• 35. Editorial: Artificial intelligence in animal behaviour, ... [https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2025.1607623/full]
• 36. Impact of proximity definitions and sampling rates on social ... [https://www.nature.com/articles/s41598-025-93830-8]
• 37. Social network analysis - centrality parameters and individual ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4409614/]
• 38. A protocol for assessing bias and robustness of social network ... [https://movementecologyjournal.biomedcentral.com/articles/10.1186/s40462-024-00494-6]
• 39. Changes in Grooming Networks Among Male ... [https://pubmed.ncbi.nlm.nih.gov/41002186/]
• 40. Multi-dimensional social relationships shape ... [https://elifesciences.org/articles/104460]
• 41. Multi-dimensional social relationships shape ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11888598/]
• 42. Network structure of avian mixed-species flocks decays with ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10107241/]
• 43. How do birds flock together? Scientists reveal their ... [https://studyfinds.org/how-do-birds-flock-together/]
• 44. Constructing, conducting and interpreting animal social ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4973823/]
• 45. Animal social networks are robust to changing association ... [https://link.springer.com/article/10.1007/s00265-025-03559-7]
• 46. Animal social networks are robust to changing association ... [https://pubmed.ncbi.nlm.nih.gov/39927187/]
• 47. Animal social networks are robust to changing association ... [https://ecoevorxiv.org/repository/view/6524/]
• 48. Use of social network analysis to examine preferential co ... [https://animalbiotelemetry.biomedcentral.com/articles/10.1186/s40317-020-00201-3]
• 49. Behavioral ecology: New technology enables a more ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10449220/]
• 50. Social Network Analysis 101: Ultimate Guide [https://visiblenetworklabs.com/guides/social-network-analysis-101/]
• 51. Five misunderstandings in animal social network analysis [https://ecoevorxiv.org/repository/view/9817/]
• 52. Text has enhanced contrast | ACT Rule | WAI [https://www.w3.org/WAI/standards-guidelines/act/rules/09o5cg/]
• 53. Current directions in animal social networks [https://www.sciencedirect.com/science/article/abs/pii/S2352154616301619]
• 54. Analysing social network structures and thematic ... [https://link.springer.com/article/10.1007/s13278-025-01516-w]
• 55. ARS'25 Tenth International Workshop on Social Network ... [https://indico.unina.it/event/103/timetable/?view=standard]
• 56. When Pigs Click: Social Network Analysis Reveals New ... [https://www.porkbusiness.com/news/hog-production/when-pigs-click-social-network-analysis-reveals-new-findings-about-animal-beh]
• 57. Causal network inference based on cross-validation ... [https://www.nature.com/articles/s42005-025-02091-4]
• 58. Animal social networks are robust to changing association ... [https://research.rug.nl/en/publications/animal-social-networks-are-robust-to-changing-association-definit]
• 59. Assessing the Genetic Landscape of Animal Behavior - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC5937184/]
• 60. How genes and environment interact to shape behaviour [https://www.sciencedirect.com/science/article/pii/S0168159125000802]
• 61. Heritability of social behavioral phenotypes and preliminary ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8930433/]
• 62. A multi-species repository of social networks | Scientific Data [https://www.nature.com/articles/s41597-019-0056-z]
• 63. Cockroach social network position does not influence ... [https://link.springer.com/article/10.1007/s10164-025-00870-8]
• 64. Does the social network structure of wild animal ... [https://www.sciencedirect.com/science/article/abs/pii/S0376635721001303]
• 65. From Wild Animals to Domestic Pets, an Evolutionary ... - NCBI [https://www.ncbi.nlm.nih.gov/books/NBK219727/]
• 66. Environmental Enrichment and Behavioral and Social ... [https://research.wayne.edu/iacuc/envenrich]
• 67. A standardized protocol for repeated social defeat stress in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC3220278/]
• 68. Social Network Analysis as a Tool in the Care and Wellbeing ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10668822/]
• 69. Social Network Analysis in Farm Animals: Sensor-Based ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7914829/]
• 70. The potential of Social Network Analysis as a tool for ... [https://www.cambridge.org/core/journals/animal-welfare/article/potential-of-social-network-analysis-as-a-tool-for-the-management-of-zoo-animals/8B2445F5CF579F3F673EDBEED91660CF]
• 71. Animal Models for the Social Dimensions of Health and Aging ... [https://www.animalsocialaging-network.org/]
• 72. A multilevel statistical toolkit to study animal social networks [https://www.nature.com/articles/s41598-020-69265-8]
• 73. “Lost in translation?” Animal research in the era of precision ... [https://translational-medicine.biomedcentral.com/articles/10.1186/s12967-025-06084-3]
• 74. NIH to prioritize human-based research technologies [https://www.nih.gov/news-events/news-releases/nih-prioritize-human-based-research-technologies]
• 75. WEBINAR—Leveraging Novel Animal Models for ... [https://www.physiology.org/detail/event/2025/01/14/default-calendar/webinar-leveraging-novel-animal-models-for-translational-research]