Overview
Social network analysis (SNA) applies to friendships in classrooms, collaboration among firms, information flow in online communities, co-authorship in science, and co-occurrence of words or codes in text. The first decisions are fundamental: what counts as a node, what counts as a tie, whether ties are directed or weighted, and where the network boundary lies. Those choices determine what every later measure means.
We help you design or clean the network data, compute descriptive measures (density, reciprocity, transitivity, path lengths), identify important actors with appropriate centrality measures, and find cohesive subgroups with community detection. For inferential questions — for example, whether ties form because of shared attributes or reciprocity — we can fit exponential random graph models (ERGMs) or related approaches.
Visualizations are powerful but easy to misread, because layout algorithms place nodes for readability, not meaning. We produce clear, labeled figures in Gephi or R and pair them with the numerical evidence that supports your claims.
Who it’s for and the problems it solves
This is for you if you are…
- Students and researchers in sociology, education, management, public health and political science collecting relational survey or archival data
- Researchers analyzing co-authorship, citation or collaboration networks
- Teams analyzing online interaction data such as reply or retweet networks
- Qualitative researchers who want to visualize code co-occurrence or stakeholder maps
Common problems we help with
- Your data are in a spreadsheet and you are not sure how to create an edge list or adjacency matrix
- You calculated several centrality measures and do not know which one answers your question
- Community detection gives different results each run
- Your network figure looks like a hairball
- Missing respondents have left gaps in a whole-network survey
What’s included — and what isn’t
Included
- Network design: node and tie definitions, directed or undirected, weighted or binary, one-mode or two-mode, whole or ego networks
- Data preparation: edge lists, adjacency matrices, node attribute tables and two-mode projections
- Descriptive analysis: density, degree distribution, reciprocity, transitivity, components, diameter and average path length
- Centrality: degree, betweenness, closeness, eigenvector and PageRank, chosen to match your theoretical question
- Community detection: Louvain, Leiden and other methods, with modularity and stability checks
- Ego-network analysis: size, composition, heterogeneity and brokerage measures
- Inferential models where needed: QAP regression, ERGMs, and advice on stochastic actor-oriented models for longitudinal data
- Publication-quality visualizations in Gephi or R (igraph, ggraph) and reproducible code
Not included
- Collecting network data from platforms in breach of their terms or your ethics approval
- Identifying or publishing named individuals in sensitive networks without consent and ethical clearance
- Treating standard regression p-values on network data as valid without addressing dependence
What you receive
- Network data files
- Clean edge list and node attribute files, plus Gephi (.gephi or .gexf) or R objects ready for reuse.
- Measures table
- Network-level and node-level statistics with definitions and a note on how each relates to your research questions.
- Community results
- Community assignments, modularity values, stability checks and a description of each group's composition.
- Figures
- Labeled network visualizations with a legend and methods note describing the layout and encodings.
- Code and walkthrough
- Annotated R or Python code and a session explaining the results.
How it works
Define the network
We agree nodes, ties, boundaries and the questions the network should answer.
Build and describe
We prepare the data and compute descriptive and centrality measures.
Find structure
We run community detection or inferential models as needed and check robustness.
Visualize and explain
We produce figures and walk you through interpretation.
Measures, models and tools
- Centrality
- Degree for activity or popularity, betweenness for brokerage, closeness for reach, eigenvector and PageRank for connection to well-connected others.
- Community detection
- Louvain and Leiden modularity optimization, Infomap and label propagation, compared across runs for stability.
- Structural holes and brokerage
- Burt's constraint and effective size, and Gould–Fernandez brokerage roles for ego networks.
- QAP and ERGMs
- Permutation-based regression for dyadic data, and exponential random graph models that test tie-formation mechanisms while accounting for dependence.
- Gephi
- Interactive exploration, ForceAtlas2 layouts and presentation-quality figures.
- R and Python
- igraph, tidygraph, ggraph and statnet in R; NetworkX and igraph in Python.
Which centrality measure answers your question?
Centrality measures capture different ideas of importance. Choosing one because it appears in a tutorial produces results that are hard to justify. We start from your theory: are you interested in who is most active, who controls information flow, who can reach others quickly, or who is connected to other influential actors?
- Who has the most direct contacts? → degree centrality
- Who sits between otherwise separate groups? → betweenness centrality
- Who can reach everyone in the fewest steps? → closeness centrality
- Who is connected to other well-connected actors? → eigenvector centrality or PageRank
Why network data need special statistics
Observations in a network are not independent: if A is tied to B, that affects B's ties too. Ordinary regression on ties or node measures can therefore understate uncertainty. For hypothesis tests we use methods built for dependence, such as QAP or ERGMs, and we explain their assumptions and limits.
Missing data and network boundaries
Whole-network studies are sensitive to non-response: a missing central actor can change the structure substantially. We help you assess the pattern of missingness, consider approaches such as symmetrizing reported ties where defensible, and describe limitations honestly.
What the work can look like
Illustrative example — fictional data
Illustrative collaboration network summary
A fictional study mapped advice-seeking among 64 teachers in one school using a roster survey with a 90% response rate. All numbers are invented.
Network description
The directed network had 312 ties (density 0.08) and a reciprocity of 0.41. Leiden community detection found four communities largely matching subject departments (modularity 0.47), stable across 100 runs.
Brokers
Three teachers had high betweenness but modest degree; all were members of cross-department committees, suggesting a brokerage role that the qualitative interviews could explore.
Illustrative only; the school, teachers and statistics are invented to show how findings are reported.
What you’ll need to provide
- Research questions and the theoretical ideas behind them (for example, brokerage, homophily, diffusion)
- Relational data in any format and a description of how it was collected
- Node attributes such as role, organization or group
- Ethics constraints regarding identification of individuals
You don’t need everything ready to ask for a quote — tell us what you have and we’ll say what’s missing.
What affects price and turnaround
Price depends on
- Network size and data preparation needs
- Number of networks or time points
- Descriptive analysis only versus inferential modeling (QAP, ERGM)
- Number and complexity of visualizations
Turnaround depends on
- Clarity of node and tie definitions at the start
- Data format and cleaning effort
- Model convergence time for ERGMs
Every project is quoted individually. How we price our work.
Questions people ask before starting
Do I need special survey questions for network analysis?
Usually yes. Whole-network studies often use a roster of names; ego-network studies use name generators and interpreters. We can review your instrument before data collection.
Gephi or R?
Gephi is excellent for exploring and visualizing. R or Python is better for reproducible analysis and statistical models. Many projects use both.
Can you analyze co-authorship or citation networks?
Yes. For bibliometric mapping with tools like VOSviewer or Bibliometrix, see our bibliometric analysis service; for deeper network statistics, this service applies.
How do you protect participants' identities?
We work with pseudonymized IDs and can produce figures without labels. Network data can be re-identifiable even without names, so we help you plan what to publish.