Directed Modularity Analysis
Directed Modularity Analysis (Leicht-Newman Directed Community Detection) · Also known as: directed community detection via modularity, directed Q-modularity, digraph modularity optimization, Leicht-Newman modularity
Directed modularity analysis extends the classic Newman-Girvan modularity framework to directed graphs, where edges carry a source and a destination. Formalized by Leicht and Newman in 2008, it partitions nodes into communities by maximizing a modularity score that accounts for each node's separate in-degree and out-degree in the null model, making it the standard approach for community detection in citation networks, information flows, and other asymmetric relational data.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use directed modularity analysis when your network has asymmetric edges — citations, follows, information flows, trade links, or hyperlinks — and you want to discover groups of nodes that form coherent communities given the direction of ties. It is the appropriate generalization of modularity whenever undirected modularity would ignore meaningful asymmetry in the data. Minimum network size is roughly 30-50 nodes; very small networks rarely show stable community structure. Do not use it when edges are genuinely undirected (co-authorship, co-occurrence) — standard undirected modularity is simpler and more interpretable. Also avoid it when the network is very sparse or nearly acyclic (e.g., strict temporal DAGs), where the directed null model may be poorly calibrated.
Strengths & limitations
- Correctly handles asymmetric edge structure by separating in-degree and out-degree in the null model.
- Produces an interpretable scalar Q score that quantifies how strongly communities exceed random expectation.
- Compatible with fast heuristics (Louvain, Leiden) that scale to networks with millions of edges.
- Well-grounded in the physical-review literature with a clear mathematical derivation (Leicht & Newman 2008).
- Identifies overlapping sender/receiver roles within communities that undirected methods miss.
- Modularity maximization is NP-hard; all practical algorithms are heuristics and can return different partitions across runs.
- Suffers from the resolution limit: sub-communities smaller than a scale set by network size may be merged into larger groups.
- The modularity score Q alone does not determine the correct number of communities; additional validation is needed.
- Performance degrades on very sparse directed networks where the null model P_ij becomes a poor baseline.
Frequently asked
Why can I not just symmetrize my directed network and run standard modularity?
Symmetrizing (treating each directed edge as undirected) loses the distinction between nodes that predominantly send edges and those that predominantly receive them. In citation networks, for example, a highly cited paper and a prolific citing paper play very different structural roles; directed modularity preserves this distinction while undirected modularity conflates them.
What Q value indicates a good partition?
There is no universal threshold, but in practice Q above 0.3 is often taken as evidence of meaningful community structure. However, Q depends on network size and density, so always compare against a null distribution or randomized baseline rather than relying on an absolute cutoff.
Which algorithm should I use to maximize directed modularity?
The Louvain algorithm with directed-edge support (available in igraph and NetworkX) is the most common choice for moderate to large networks due to its speed. The Leiden algorithm is a more recent refinement that avoids some instabilities in Louvain partitions. For small networks, spectral optimization of the directed modularity matrix B is exact enough to be practical.
How do I choose the number of communities?
Directed modularity optimization selects the number of communities implicitly as part of maximizing Q. If you need a specific number, use the resolution parameter (gamma) in Louvain/Leiden: increasing gamma yields more, smaller communities; decreasing it yields fewer, larger ones. Always validate the result qualitatively by inspecting who belongs to each community.
Can directed modularity handle weighted edges?
Yes. Replace the binary adjacency matrix A with a weighted adjacency matrix W, replace counts with total weight, and redefine k_i^out and k_j^in as the total out-strength and in-strength respectively. The formula for Q and the null model P_ij adapt directly; most graph libraries support this natively.
Sources
- Leicht, E. A., & Newman, M. E. J. (2008). Community structure in directed networks. Physical Review Letters, 100(11), 118703. DOI: 10.1103/PhysRevLett.100.118703 ↗
- Newman, M. E. J., & Girvan, M. (2004). Finding and evaluating community structure in networks. Physical Review E, 69(2), 026113. DOI: 10.1103/PhysRevE.69.026113 ↗
How to cite this page
ScholarGate. (2026, June 3). Directed Modularity Analysis (Leicht-Newman Directed Community Detection). ScholarGate. https://scholargate.app/en/network-analysis/directed-modularity-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Betweenness CentralityNetwork analysis↔ compare
- Directed Community DetectionNetwork analysis↔ compare
- Directed Social Network AnalysisNetwork analysis↔ compare
- Modularity AnalysisNetwork analysis↔ compare
- Stochastic Block ModelNetwork analysis↔ compare