Three people is not a study#
SLOP2805 · Week 11 · Lecture
Somebody who has never seen your interface is about to use it,
hit a stall you did not warn them about,
and say out loud what they think is happening.
Where we were, and where we are going#
Week 10 produced a register of every claim your interface makes, all of them now traced to an event.
Today is the only evidence in the course that comes from outside your own head — and it is much weaker than you will want it to be.
Existence proofs and rates#
“At least one person read the stall as a failure.” Established by one observation. Real, reportable, often enough to act on.
“Most people read the stall as a failure.” Not established by three observations, or thirty, or any number you will reach.
Nearly every overreach in a write-up is the first sentence rewritten as the second.
Do not write proportions. “Two of three” counts this room. It is not a rate.
Records, not readings#
✗ “The user was confused.” A conclusion with its evidence deleted.
✓ “They moved the pointer to Cancel, stopped, and said is it stuck or is it me.” A record.
Only the second survives next week, when you disagree with yourself about what it meant.
Name your sample precisely#
Not “a small sample”. These people:
- they are in this course
- they know a stall is coming, because stalls were week 8
- they have built the same interface
- they are simultaneously the most forgiving and the most suspicious users your prototype will ever have
Generic humility is not a limitation statement. Specific bias is.
Worked example — one observation, three legitimate answers#
What was recorded (P2, simulated stall at 45%):
pointer to Cancel at 0:38 · did not click · “is it frozen or is it just slow” · pressed Cancel at 0:52
| Answer | When it is right |
|---|---|
| Change it. Show elapsed-since-last-progress after 10 s | you saw it twice, you understand the cause, the fix is smaller than the problem |
| Leave it, say why. One person, first use, cannot separate novelty | you saw it once |
| Leave it, name the test. “Show 5 people a 60 s stall; count who cancels before 45 s” | you think it is real and one observation cannot settle it |
All three score. Only the unexamined change loses marks.
Your turn — Thursday#
- Rotation, groups of four. Three rounds of 15 minutes each way.
- Consent read aloud before each round; stop if asked.
- Fill your sheets before discussing anything with anyone.
- Three handling decisions, each naming the observation behind it.
- At most three changes. A ceiling, not a target.
Not compulsory — in either role, for any reason, with no explanation. Tell a tutor and you get the structured walkthrough, graded identically.
The mistake to expect#
Manufacturing a change to look responsive.
Three changes on one thin observation apiece is a worse outcome than one change and two honest deferrals, and it is marked as one.
The other mistake belongs to the alternative task: a walkthrough written up as though somebody had been watched. A walkthrough is evidence about internal consistency, not about people. That is the one way to lose this criterion outright.
What this week cannot tell you#
How often anything happens. Whether your interface is good. Whether a stranger under real pressure, with a file they needed, would do what P2 did.
You are marked on whether your conclusions are proportionate to your evidence — never on how many people you watched, whether you watched anybody, or how many things you changed.
A confident claim from three observations loses marks. A careful claim from one does not.
What leaves the room#
- three observation records, in participants’ words
- three handling decisions, each traced to a line
- at most three changes, deployed, register still matching
Where it goes: the Evaluation record is deliverable 3 of Assignment 3, and Evaluation and revision is 20 marks.