Degree Distribution Analysis¶

Why It Matters¶

Human intuition defaults to assuming rough equality - we expect people to have similar numbers of friends, websites to have similar numbers of links, proteins to have similar numbers of interactions. This assumption fails catastrophically in many real networks. Degree distribution reveals whether the network has vulnerable concentration points (hubs) or is robustly distributed. It determines whether random removal affects the network much (hub networks: little effect; homogeneous networks: proportional effect) and whether targeted attacks devastate quickly (hub networks: yes; homogeneous: no).

The Key Move¶

When examining any connected system, ask: how unequal is the connectivity? Plot or estimate the distribution of connection counts. Check if removing the top 10% most-connected nodes would fragment the system. If yes, you have hub structure with specific failure modes. If no, you have distributed structure with different vulnerabilities.

Classic Application¶

The internet's structure. Early studies revealed that the web has heavy-tailed degree distribution - most pages have few links, but rare hubs like Google or Wikipedia have millions. This explains both robustness to random node failures (most nodes are peripheral) and vulnerability to targeted hub attacks (removing key sites fragments access).

Surprising Application¶

Career capital accumulation. Skills can be mapped as a network where nodes are competencies and edges are "commonly combined" relationships. Most people accumulate skills in homogeneous clusters (all skills have 3-5 related skills). High performers often have power-law distributions - a few hub skills that connect to many domains (public speaking, writing, data analysis) plus many specialized peripheral skills. This explains why generic advice to "diversify skills" fails - the distribution shape matters more than the count.

Failure Modes¶

Confusing degree with importance - high-degree nodes aren't always critical (they might be redundantly connected). Assuming static distributions - degree distributions evolve, especially in growing networks. Ignoring weighted vs. unweighted - having 100 weak connections differs from 10 strong ones, but raw degree counts don't distinguish. Treating heavy tails as power laws - many distributions are heavy-tailed without being true power laws, and the distinction matters for prediction.

Go Deeper¶

Barabási, "Network Science" Chapter 4 (Degree Distribution); Newman, "Networks: An Introduction" Chapter 8 (Random Graphs)