Blog
Random Is Not the Same Thing as Fair
Equal chances and fair outcomes are different goals, and most disputes are about the second one.
Equal chances and a fair outcome are different things, and most disputes about a draw are about the second one while being conducted in the language of the first.
Here are six situations where a perfectly unbiased draw produces a result nobody involved would call fair.
Teams that are equal in expectation and lopsided in fact
Split ten players of genuinely different ability into two fives at random. Every player had an equal chance of every team, which is exactly what fair means procedurally.
More than half the time — 52%, by direct enumeration of the 252 possible splits — one team gets three or four of the strongest four, and the match is decided before it starts. Nobody enjoys that game, including the team winning it.
The procedure was fair and the outcome was not, and pretending those are the same thing is how a group ends up with an argument that has no available resolution.
A rota that is unbiased every week and unfair every quarter
Draw the bathroom cleaning at random each week among four housemates. Each week is a clean one-in-four for everyone.
Over ten weeks, the expected count per person is 2.5 with a standard deviation of about 1.4 — so somebody getting five is within the ordinary range. That person has done the worst job twice as often as their share, from a process that was never biased against them.
They will describe the system as broken and they will be describing the outcome accurately.
Bill-splitting that converges too slowly to matter
Four friends take turns paying for dinner by random draw. Fair in expectation, and the expectation takes a very long time to arrive.
After ten dinners the typical gap between the most and least frequent payer is about three occasions. After a hundred it has grown to about ten, while the proportional imbalance has fallen — the group experiences the growing absolute gap, not the shrinking ratio.
The method is defensible and it needs the group to still be eating together in three years for the defence to cash out.
Slots that are not interchangeable
An on-call rota drawn at random over fifty-two weeks treats the week containing Christmas as one shift and a quiet February week as one shift. They are not the same shift.
A draw that is uniform over undifferentiated weeks will reliably hand one person two premium slots and another none, and no amount of procedural fairness makes that acceptable to the person with two.
The same problem appears in presentation order, judging order, and anywhere a position carries a real cost that the draw does not model.
Coverage failures
Thirty independent draws from a class of thirty leave an expected eleven students never picked. Every draw was fair; a third of the class was not reached.
This is the purest version of the distinction, because the goal — reaching everyone — is not a fairness goal at all, and a fair procedure is the wrong instrument for it.
The fix is to give up equal chances deliberately in exchange for a coverage guarantee, which is what a no-repeat cycle does.
What to do about it
The general move is to identify which property you actually want before choosing a mechanism. Equal chances, balanced outcomes, guaranteed coverage and even totals over time are four different goals, and no single draw delivers all of them.
The second move is to say which one you chose. A team split described as fair when it was balanced, or a rota described as random when it was rotated, invites a specific objection later from someone who assumed the other property.
And the third is to accept that some of these have no clean answer. A one-off presentation order has a genuinely worse last slot, and distributing that fairly is the most anyone can do.
The vocabulary problem underneath all six
Part of why these arguments are so hard to resolve is that both parties are using the word fair correctly and meaning different things by it. One is describing a property of the procedure and the other a property of the outcome, and neither is being sloppy.
English does not give us separate words, which is unusual — most fields that care about this distinction invent some. Procedural and distributive justice is the pair philosophers use, and it maps directly onto the disagreements on this page.
The practical move is to name which one you mean before proposing a mechanism. "Everyone has an equal chance" and "everyone ends up with a similar amount" are both reasonable goals, they are rarely compatible, and a group that has not chosen between them will re-litigate every result.
Frequently asked questions
What is the difference between random and fair?
Random means equal chances at the moment of drawing. Fair usually means something about the outcome or its accumulation, which a single unbiased draw does not guarantee.
How often does a random team split go wrong?
For ten players split 5v5, about 52% of splits put three or four of the top four on one side. Only 47.6% divide them evenly.
Why does a rota feel unfair when each week is unbiased?
Because people judge the totals. Over ten weeks with four people, somebody getting five of a job against an expected 2.5 is entirely ordinary.
What are the four goals?
Equal chances, balanced outcomes, guaranteed coverage, and even totals over time. No single mechanism delivers all four.
Which should I choose?
Whichever matches the complaint you are trying to prevent — then say which one you chose, so nobody assumes a different one.
Do any of these have no answer?
Yes. A one-off presentation order has a genuinely worse last slot, and distributing that cost fairly is the most anyone can do.