The Combinatorics of Sibling Conflict

How many ways can n children take sides against each other? The answer is a Stirling number, and it grows exponentially.

Mathematics · A pet project

The Combinatorics of Sibling Conflict

I have four kids, each two years apart, and I grew up with only one sibling. The number of ways they could get into a fight seemed never-ending, so I counted it. The answer for four is 25.

Read the Paper (PDF) Repository

Preprint — queried to The Mathematical Intelligencer on 2026-08-12, reply pending. Not published, not peer-reviewed. Revised since the query, so the PDF here is newer than the one the editors have; the as-queried version is kept in the repository.


The question

A conflict configuration is an unordered pair of disjoint, nonempty subsets of the n children — the two sides — with everyone else uninvolved. That third category is the interesting part: a child can be on side A, on side B, or out of it entirely.

The result

Each child goes to side A, side B, or neither, giving 3n assignments. Drop the ones where a side is empty (2·2n − 1, by inclusion–exclusion), halve it because the labels A and B are interchangeable, and you get:

\[C(n) = \frac{3^n - 2 \cdot 2^n + 1}{2}\]

Which is not an ad hoc expression — it’s S(n+1, 3), a Stirling number of the second kind, OEIS A000392.

The obvious thing to compare it against is Bossard’s 1945 count of family relationships, n(n−1)/2. I had those two side by side as rival quantities until a reader pointed out that they aren’t: a sibling pair is precisely a configuration in which both sides are single children. The pairs don’t run alongside C(n), they sit inside it. Everything else in C(n) is a split where at least one side is a bloc.

Diagram titled 'Three children can take sides in 6 ways, and 3 of those are one against one.' Six small groups of three lettered circles A, B and C. A filled green circle is a child on one side, a ringed circle a child on the other, a grey circle a child staying out. The first row, headed 'One against one — the sibling pairs (3 of 6)', shows A vs B with C staying out, A vs C with B staying out, and B vs C with A staying out. The second row, headed 'One against two (3)', shows A vs B+C, B vs A+C and C vs A+B, each with nobody left out. A note states that A vs B and B vs A are the same split and are counted once. Source strip: all 6 configurations for three children, enumerated, computed 2026-08-13, OEIS A000392, S(n+1, 3), exact counts, nothing estimated.

All six ways three children can take sides. The three one-against-one cases in the top row are exactly the three sibling pairs — the subset drawn rather than asserted.
Conflict configurations C(n) = S(n+1, 3), against the one-against-one configurations they contain, n(n−1)/2 — Bossard’s sibling pairs. All values exact.
Children n Conflict configurations C(n) One against one, n(n−1)/2
1 0 0
2 1 1
3 6 3
4 25 6
5 90 10
6 301 15
7 966 21
8 3,025 28

Of the 25 ways four children can take sides, 6 are one against one. Of the 3,025 ways eight children can, 28 are. The containing space grows exponentially; the thin slice of it where nobody sides with anybody grows quadratically.

Line chart titled 'Of the 25 ways four children can take sides, 6 are one against one.' Horizontal axis, number of children, 1 to 4; vertical axis 0 to about 25. A solid green line, labelled 'All ways to take sides: 25' at its right end, runs 0, 1, 6, 25. A dashed grey line, labelled 'One against one: 6', runs 0, 1, 3, 6. The two lines coincide to two children and part from three onward, where the green line is labelled 6 and the grey line 3. Subtitles read 'Every possible split into two opposing sides, with anyone left over staying out of it' and 'These are possibilities, not arguments that happened.' Source strip: OEIS A000392, S(n+1, 3); one against one is n(n-1)/2; computed 2026-08-12; exact counts, nothing estimated.

The range most families are actually in, n = 1 to 4, with every value readable.

Line chart titled 'Of the 3,025 ways eight children can take sides, just 28 are one against one.' Horizontal axis, number of children, 1 to 8; vertical axis 0 to 3,000. A solid green line, labelled 'All ways to take sides: 3,025' at its right end, runs along the axis to four children then climbs steeply through labelled values 90, 301 and 966. A dashed grey line, labelled 'One against one: 28', stays flat against the axis for its whole length. An italic note reads 'Each child multiplies the total: 6x at first, nearer 3x by eight.' Subtitles read 'Every possible split into two opposing sides, with anyone left over staying out of it' and 'These are possibilities, not arguments that happened.' Source strip: OEIS A000392, S(n+1, 3); one against one is n(n-1)/2; computed 2026-08-12; exact counts, nothing estimated.

The same two counts out to n = 8, where the shape of the growth takes over.

What it doesn’t claim

It counts a possibility space — not how often siblings actually fight, how badly, or how any of it turns out. Plenty of big families are close and plenty of small ones aren’t.

It’s also not a novelty claim over the field. Bossard (1945) counted family relationships and Kephart (1950) extended that to subgroupings; the paper claims only to extend that line to opposing-sides configurations, and to say plainly that Bossard’s count sits inside the new one rather than beside it.

The counts were verified two ways — brute-force enumeration of all 3n assignments for n = 1–8, and the closed form against OEIS. The figures went through four rounds of blind reading panels; the third is what caught the subset error above. The records are in the repository.