Friendship Paradox, explained.
The friendship paradox shows how sampling people through friendship links can yield a higher average connection count than sampling people equally.
Why it happens
A person with many connections appears at the end of many links. Their inclusion chance is therefore proportional to degree when sampling edge endpoints. The link-based mean is a degree-weighted average, which emphasizes well-connected people.
The friendship paradox arises because people with more friends are represented more often when you sample friendship links than when you sample people equally.
Read the result
Increase the star's leaves and compare the ordinary mean degree with the edge-endpoint mean. The latter is not a claim about every person's friends or each person's separately averaged neighbor count.
A worked example
A six-leaf star
Six people each know one central person, who knows all six.
The ordinary average degree is twelve divided by seven, about 1.71. The mean degree at a randomly sampled edge endpoint is 3.5.
The center appears repeatedly in link-based sampling, giving it more influence on that second average.
OPTIONAL DEEPER DETAILGo deeper: inside the model
Inside this model
A star network contains one center with n links and n leaves with one link each. Mean degree is 2n/(n+1). Following a uniformly sampled edge endpoint gives mean degree sum(k²)/sum(k), equal to (n+1)/2 here. This is an edge-weighted average, not every individual's neighbor average.
Where this idea is useful
A practical use
A contact recruited through referrals may be better connected than a randomly selected participant. Account for the sampling method when estimating a community's typical connectivity.
A common misconception
“Everyone's friends must have more friends than they do.”
The central person in the example has fewer-connected friends. The paradox is an aggregate sampling result, not a universal individual experience.
What this explanation leaves out
- The comparison is about averages in an undirected graph. It does not mean every person's friends have more friends, and real ties need not form a star.
When are the two means equal?
In a regular undirected network where every person has the same degree, link weighting cannot favor a more connected group, so the means agree.
Are your participants sampled directly or found through someone else's connections?
Associated thinkers
Further reading
Explore the original research or the teaching reference behind this experiment.