A researcher detects 12 communities in a corporate email network and immediately reports them to management as the organization's actual informal work groups. What important caveat should she have applied?
AThe algorithm can only detect communities in networks with fewer than 1,000 nodes, so the results may be unreliable
BDetected communities are structural patterns — dense connectivity signatures — not self-identified groups; they require validation against substantive knowledge before being treated as real social units
CCommunity detection always finds exactly the number of communities corresponding to formal organizational divisions
DModularity optimization guarantees that detected communities match self-reported group memberships
A detected community is a structural signature: a set of nodes connected more densely to each other than to the rest of the network. This pattern may correspond to real work groups, or it may reflect shared project timelines, shared managers, or artifacts of the data collection process. Treating structural communities as real social units without validation is a category error — the network shows who emailed whom, not who identifies with whom.
Question 2 Multiple Choice
What does a high modularity score (Q) indicate about a partition of a network into communities?
AEvery node in each community has the same number of connections as every other node in that community
BThe communities were each formed through a distinct social process
CThe communities contain substantially more internal edges than would be expected in a random network with the same degree sequence — the detected structure is non-random
DThe network has been partitioned into the maximum possible number of communities
Modularity is defined relative to a null model: a random network with the same degree sequence. High Q means the actual edge distribution within communities substantially exceeds what chance would predict given each node's degree. This distinguishes genuine community structure from random clustering. Option A describes a regular graph; option D is wrong because modularity does not increase monotonically with number of communities.
Question 3 True / False
The Louvain algorithm efficiently detects communities in large networks but may merge small genuine communities into larger ones — a limitation known as the resolution limit.
TTrue
FFalse
Answer: True
The resolution limit is a mathematically demonstrated limitation of greedy modularity optimization: below a certain community size (which depends on the number of edges in the full network), the algorithm cannot distinguish a genuine small community from random noise and may merge it with neighboring communities. Researchers working with networks that plausibly contain small communities should consider spectral or probabilistic methods instead.
Question 4 True / False
In a social network, structurally detected communities and self-identified social groups typically correspond — dense connectivity reliably mirrors how people identify their own group memberships.
TTrue
FFalse
Answer: False
Structural communities and self-identified groups often diverge. Dense connectivity can reflect shared organizational role, common event attendance, or transient collaboration rather than group identity. Conversely, people may identify strongly with a community whose members communicate infrequently. Validating detected communities against member attributes, roles, or self-reports is a necessary step, not an optional one.
Question 5 Short Answer
Why is comparing actual within-community edge density to a random null model — rather than simply maximizing raw density within groups — a better criterion for detecting meaningful communities?
Think about your answer, then reveal below.
Model answer: Raw density fails because high-degree nodes form dense clusters by chance — a node with 500 connections will appear at the center of a 'community' simply because it is well-connected, not because it is embedded in a cohesive group. The null model controls for degree: it asks how many within-community edges would be expected if edges were distributed randomly among nodes with the same degrees. Modularity measures how much the observed clustering exceeds this baseline, identifying groups where internal connectivity is genuinely elevated relative to what node degree alone would produce.
The null model is the key conceptual innovation of modularity. Without it, community detection degenerates into finding well-connected hubs and their neighborhoods. With it, the algorithm identifies groups where internal connectivity is a genuine structural signal — a much better proxy for real social cohesion than raw density, and one that allows meaningful comparisons across networks of different sizes and densities.