Transcript Document
Experiments and statistics
Classic “lab” study
Studying cause and effect
– “novel navigation for faster task times”
But it might be something else…
Ensure causal chain
QNT, Paul Cairns, University of York
2
Big problem
People vary
– Unlike say gravity or water
Context influences people
Need to isolate causal chain
Need to account for natural variation
QNT, Paul Cairns, University of York
3
Experimental design
Independent variable (IV)
– Causes change in
Dependent variable (DV)
– Which is numeric
Confounding variables
– Are eliminated
QNT, Paul Cairns, University of York
4
Statistics
Accounts for natural variation
Assume IV has no effect (null)
Prediction is crucial
Calculate probability of DV data
Unlikely data suggests real effects
QNT, Paul Cairns, University of York
5
Challenges
Hard concepts: involvement, aesthetics
– Construct validity
Context
– External/ecological validity
Very small questions
QNT, Paul Cairns, University of York
6
Surveys and interviews
How to ask questions
Questionnaires
Interviews
Focus groups
QNT, Paul Cairns, University of York
8
Outcomes
Demographics
Reveal thoughts/opinions/experiences
Formulate new hypotheses
Measurement
– Questionnaire as instrument
QNT, Paul Cairns, University of York
9
Asking questions
Closed
– Likert scales
– Options
Open
Wording
QNT, Paul Cairns, University of York
10
Bad questions
“Why don’t you like this interface?”
“Isn’t this good?”
“We made this. What do you think of
it?”
“I don’t like the colours and would
prefer a simpler interaction: Agree?”
“How old are you?”
QNT, Paul Cairns, University of York
11
Analysis
Closed questions: counting and stats
Open questions: coding
– Patterns, themes
– Dimensions and attributes
– Theoretical sampling
QNT, Paul Cairns, University of York
12
Challenges
Learning to ask questions
Qualitative analysis
Informing design
QNT, Paul Cairns, University of York
13
Case studies
Misconceptions
Sample of one
“Go in and learn”
Purely descriptive
QNT, Paul Cairns, University of York
15
Characterising Case Studies
Current
Context and object blur
Multiple sources of evidence
Theoretically informed
QNT, Paul Cairns, University of York
16
Outline of process
Theoretical prediction
Gather data
– Obs, interviews, documents, resources
Analyse
– Maintain causal chain
Theoretical development
QNT, Paul Cairns, University of York
17
Methodological development
Devise method to address problems
Use method in a project
Analyse impact of method
Develop method
QNT, Paul Cairns, University of York
18
Challenges
No formal analysis methods
Often need intervention
Action Research for HCI?
– Notion of participation
– Joint solutions
QNT, Paul Cairns, University of York
19