User Research
Human-Computer Interaction
01 Oct 2026
Week FOUR
Part III of Hornbæk et al. (2025)
Interviews · Field Research · Surveys
Intro
For three weeks we’ve been building a general model of the human, which is powerful and a good starting point, but it’s not specific to your users.
This week’s question: you are not your user, so how do you find out who is, and what they do?
Case Study: Uber
What did Uber miss?
Uber got off to a rocky start with women. Because the all-male cofounders designed for themselves.
- They designed for convenience instead of safety. Uber assumed riders primarily valued speed, price, and ease of use over safety.
- They didn’t deeply understand women’s lived experiences. Women often approach transportation differently than men because of the risk of harassment or assault.
- They optimized for business metrics instead of trust, which is essential in a marketplace.
You are not the user
This is the founding commitment of user research. Three reasons it’s hard, and each one previews a method this week.
- The say-do gap: what people say they do and what they actually do diverge. Self-reported internet use correlates weakly with logged behavior (Scharkow).
- Tacit knowledge: much of what drives behavior cannot be put into words. Ask someone how they ride a bike.
- The future is unimaginable: no user of a batch system could have specified a GUI. People are poor predictors of their own future use.
Note
User research is the set of empirical methods for obtaining, analyzing, and representing knowledge about users, their activities, their contexts, and the systems they already use, in order to inform design.
Who is the user, then?
If you are not the user, your first job is to say who is. That is a three-step commitment, before any method.
- Target audience: the profiles of who the product is for, built from behavioral, technological, and demographic criteria.
- Other stakeholders: people affected by the system without using it, like the parents of a child playing a mobile game.
- Sampling: picking specific participants who represent that audience, balancing representativeness, variety, and cost.
Note
Identifying the user is a three-step commitment: specify the target audience, map the other stakeholders, then sample representatively. A failure at any step injects bias that is hard to detect and often only surfaces at deployment.
The stakeholders you forgot
The people who wreck your design are often the ones who never made it onto your participant list.
- Stakeholder analysis rank groups by power, legitimacy, and urgency so you don’t just serve the loudest.
- Latent and dormant stakeholders: some are affected only indirectly, or stay silent until a crisis, like legal experts after an accident.
- Managers sample badly: they attend to the stakeholders who matter to them, not the ones who are most urgent.
- Uber again: women riders were core to the audience but treated as an edge case, and safety was an urgent concern the all-male team never sampled for.
Concept check
Pause and think
You are designing a homework app for a school. Name three stakeholders who are not the students, and for one of them, argue why they might be more urgent than the students themselves.
No method is free
Every method trades off three things you cannot maximize at once (McGrath).
- Realism: how close to the naturally occurring situation. Field research wins here.
- Precision: how much control and detail. Lab and structured methods win here.
- Generalizability: how well it extends to other people and settings. Surveys win here.
Note
Because every method is biased in its own way, the fix is triangulation: combine methods with different weaknesses so they cover for each other. Hold this throughline: interviews, field research, and surveys each buy one corner of the triangle and pay for the others.
Judging the work: is it valid?
Choosing a method is only half the job. The other half is judging whether a study was done well. The textbook names four quality dimensions.
- Validity: whether the conclusions you draw are actually justified.
- Reliability: whether the method gives consistent results if you run it again.
- Transparency: whether your design, data, and analysis are open and inspectable.
- Ethics: whether collecting and reporting the data treats participants and society responsibly.
Note
Validity asks a different question than reliability: not “would I get the same answer again?” but “is this answer about the thing I think it is about?” A study can be perfectly reliable and completely invalid.
Reliable is not the same as valid
Reliability is a tight cluster of darts. Validity is whether they landed on the right board at all. You can measure the wrong thing very consistently.
- Four flavors of validity: internal (did your change cause the effect), construct (are you measuring what you claim), statistical conclusion (are the numbers trustworthy), external (does it generalize).
- The trap: a survey scale with a high Cronbach’s alpha is reliable, but if it is really measuring “wanting to please the researcher,” it is invalid.
- Look ahead: we meet the reliable-but-not-true version of this head-on in the surveys chapter.
Concept check
Pause and think
A team measures “user delight” with five questions and gets a very high Cronbach’s alpha, so they call the scale trustworthy. Which quality dimension have they shown, and which have they not? What would you check next?
Interviews
The neutral interview
- Interview: a planned, systematic conversation to elicit a user’s first-person view of a chosen theme, distinct from everyday talk by its preparation, focus, and recording.
- Open-ended (responsive): questions adapt to answers; structured: fixed script, closer to a survey; micro-phenomenological: tiny content-free questions unfolding one lived experience.
- Question types: main questions (tour, grounded, comparison), probes (“go on…”), and follow-ups that clarify, get concrete, and surface variation.
- Listen, do not talk: the interviewer spends more time listening than in any normal conversation.
Note
A leading question is one that implies a “right” answer and injects the interviewer’s prejudice into the interviewee’s response, for example “Don’t you think autocorrect should be on by default?”
Staying neutral is harder than it sounds
The rule is easy to state but brutal in practice, especially when you built the thing you are asking about.
- Kill the bias in wording: “How much did you like it?” becomes “Tell me about your experience using it.”
- Handle “why?” with care: it pushes people to rationalize and can make them defensive; use it sparingly.
- Do not ask people to forecast: “Is handwriting recognition useful?” invites speculation. Ask “How are you using it right now?”
- The hardest case is your own project: if you are personally invested, be honest about whether you can even be neutral on the subject.
In the wild