A selection is biased when it deviates from uniformity — not because anyone intended to favour anyone, but because human judgement entered the process somewhere.
That distinction matters. Nobody in the scenarios below decided to be biased. The bias is in the method of choosing, and it is invisible precisely because it feels like judgement rather than selection.
1. Recency
What it is. Weighting whatever happened most recently. The student who volunteered last is the one you pick; the task mentioned five minutes ago feels urgent and gets done.
What it looks like. “Who wants to go next?” — and you take the first hand, every time. The student who raised it thirty seconds ago has better claim on your attention than the one who raised it in minute one.
The fix. Draw instead of taking the first hand. There is no ordering to exploit.
2. Similar to me
What it is. Favouring people who remind you of yourself — same interests, same communication style, same background.
What it looks like. You pick the student who gives an answer like the one you would have given. Not deliberately; it registers as “they had a good answer” rather than “they are like me”.
The fix. Commit to the result before the draw. The selection is made and you accept it, which is the entire point — a draw you would have overridden is not a selection.
3. Position
What it is. Systematic preference for entries at the start, end, or middle of a list.
What it looks like. Scanning an attendance register and stopping when you find someone suitable. The top of the list gets more attention. With a wheel, entries in visually distinctive positions read as more important.
The fix. Use a shuffled draw. If the order is visible and fixed, position is a real input regardless of your intentions.
4. Confirmation
What it is. Interpreting outcomes as proof of what you already believed.
What it looks like. A student who has a reputation produces results that get read as confirmation, while the same result for a student you think highly gets read as a good surprise. The draw was fair; the interpretation was not.
The fix. Write down the reason for a decision before you make it. If you cannot state why before seeing the result, the decision was not being made on those grounds.
5. Availability
What it is. Treating who is present as a proxy for who should be included.
What it looks like. Two students are absent; the draw includes them; you skip them when they come up. Now the pool was not the published pool, and the deviation was not random.
The fix. Remove absent participants before the draw and say you have. If the count of remaining names is published, the result is defensible.
6. Momentum
What it is. Treating a pattern as if it were a mechanism. Skipping someone because they have been picked a lot recently is a legitimate fairness instinct, and it produces a result that is not random.
What it looks like. “They’ve had three turns, I’ll leave them.” The next draw includes them again, and the compensation has shifted the distribution without anyone deciding it should be.
The fix. If balance over a term matters, it needs a rotation, not a draw with manual adjustments. Those are two different processes and mixing them produces neither.
The Underlying Issue
Most of these share one cause: the person choosing wants to be seen to have chosen well, and cannot tolerate a result that looks bad. The bias is not in the preference for a good outcome — that is normal — but in the willingness to override the procedure to get it.
This is why a public draw works better than a private one even when the mechanism is identical. A draw someone else can see cannot be overridden as easily, and a draw that has been watched is not something you quietly modify afterwards.
It is also why an external tool is not automatically unbiased. The tool applies the same probabilities; the person operating it can still re-run, re-order, or skip. The mechanism being fair and the process being fair are separate properties, and only the second one prevents the argument.
What Actually Prevents It
Draw from a published list. Write it down, show it, do not change it after the result.
Let someone else run the draw. The strongest single measure available. A student holding the device, or a team member, is a witness who cannot later be suspected of favouring anyone.
Record it. A screenshot with the list and the result together settles the question in advance. It is worth doing for any draw that will matter.
Do not re-run unless a rule said you could. Re-rolling after a disappointing result is the clearest possible signal that the process is not actually random — and it is the mistake most likely to be observed.
Do not hand-correct. Removing an absent name afterwards, skipping an inconvenient entry, or compensating for a past run are all the same error in different clothes.
A Note on Fairness
Worth being precise about one thing: random selection gives everyone an equal chance, which is not the same as equal outcomes. A class of thirty will not have every student chosen equally often over a term, and someone will notice.
Uniform is not balanced. If the requirement is balance rather than impartiality, a rotation is the correct tool and a draw is not a substitute for one — even a well-run draw will not deliver what a rotation delivers.
The fairness page covers the mechanism, including why Math.random() is unsuitable and what rejection sampling fixes. Worth knowing if you are choosing a tool for a decision that will be questioned.
The Short Version
Every bias above comes from the same place: not being willing to accept a result you did not choose.
That is why the useful practice is not “be less biased” — it is to make the decision before you want it, in front of people, from a list you have already published. The biases stop having anywhere to act.
And if you find yourself re-running the draw, you have the answer. Stop, take the result, and move on.