Why the average person has fewer friends than their friends
In almost every network, most members are less connected than their own contacts are on average. The comparison is rigged by how contacts get sampled.
Intuition test — answer before you read on
In most measured networks the average member has fewer connections than their connections do. Why?
Survey a school, a company or a platform and ask each member how many connections they have. Then ask how many connections their connections have. The second number is larger, in nearly every population ever measured, and it is larger for most individuals at once. Nobody is mistaken and nothing is unusual about the population.
What everyone sees
The result reads as a report about self-esteem or social hierarchy: most people really are on the edges while others sit at the centre. It is taken as evidence that popularity is concentrated and that the respondent is falling behind. The interpretation treats the statistic as a fact about people rather than a fact about sampling.
What is actually happening
Feld’s friendship paradox is a consequence of weighted sampling. When you list your contacts, a person with a hundred connections appears on a hundred lists while a person with two appears on two. The set of other people’s friends is therefore biased towards the highly connected, and the mean degree of that sampled set exceeds the mean degree of the population. The mathematics holds for any network in which connection counts vary.
Why it stays hidden
The distortion hides because the sample feels exhaustive. A person sees all of their own friends, so the group appears complete rather than selected — yet it was selected, by the friends themselves, in proportion to how many lists they appear on. Social platforms amplify the effect by displaying follower counts, making the biased sample the most visible statistic available.
Your contacts are a sample weighted by connection count. The comparison is against the well-connected by construction.
Your contacts are a sample weighted by connection count. The comparison is against the well-connected by construction.
Collect this card
Your contacts are a sample weighted by connection count. The comparison is against the well-connected by construction.
0 / 10,000 collected
Sources & further reading 3
- Feld, "Why Your Friends Have More Friends Than You Do", American Journal of Sociology, 1991
- Newman, "Networks: An Introduction", 2010
- Hodas, Kooti & Lerman, "Friendship Paradox Redux: Your Friends Are More Interesting Than You", ICWSM, 2013
Cross-references
Related files
-
Hidden Logic Entry #0010Why “free” costs more than one cent
Dropping a price from one cent to zero does not shave one cent off the decision. It changes which part of the brain makes it.
MasterThe hidden part #0010Zero is not the smallest price. It is a different category of thing entirely.
Logic Open file -
No visual on fileParadox & Proof Entry #0160
The reason a prediction can cause its own failure
The forecast changed the system it was forecasting. The prediction was accurate about the world before it was announced, and inaccurate about the world after.
MasterThe hidden part #0160A prediction that enters the system it describes is no longer a prediction. It is a participant, and participants change outcomes.
Paradox Open file -
No visual on fileDecision Architecture Entry #0135
The reason irreversible choices feel better afterwards
An option that cannot be returned is rated higher than the same option with a return window. The closed door is doing the work.
MasterThe hidden part #0135A closed door stops the comparison loop. The mind then makes the best of what it has, and it is good at this.
Decisions Open file
Annotations are reserved for archive members.
Sign in to annotate