Qualitative and Quantitative Data Notes
Identifying Quantitative and Qualitative Data
Two kinds of evaluation data:
Qualitative ("Qual")
Focuses on descriptions and experiences.
Examples:
"Participants reported being confused about how to login."
"Many users said that, when using their GPS system, they assumed that 'home' was set to their real home location."
"Most users reported finding Task 2 more difficult than Task 4."
Quantitative ("Quant")
Focuses on numerical data and measurements.
Examples:
"78% of users could not complete the login without requesting help."
"In 10,000 sampled GPS journeys, 0.1% involved the user having to correct the home location coordinates."
"On average, users took twice as long to complete Task 2 than to complete Task 4."
Surveys: Quantitative and Qualitative Data
Surveys can collect both quantitative and qualitative data.
Quantitative: e.g., rating scales (Excellent, Good, Poor)
Qualitative: e.g., open-ended questions about opinions
Example of quantitative data from a survey: Distribution of students in different courses.
Example of qualitative data from survey questions like "What have you liked most about the subject so far?" and "What would you like to change about this subject?"
Observations: Quantitative and Qualitative Data
Observations can yield both types of data.
Quantitative: direct measurements
Time to complete a task.
Number of errors made.
Number of users who can perform a task.
Number of times a task is successfully completed.
Offset from a target
Qualitative: descriptive insights into user behavior
Example: A user's thought process while trying to log in, revealing confusion about multiple login options.
Collecting Quantitative Data
Quantitative data involves:
Quantity, metrics, numbers.
Analysis through statistics.
Measurable aspects.
Not necessarily non-subjective, but not interpretative.
Common quantitative methods:
Surveys (with quantitative questions, including Likert scales).
Experiments.
Task-based evaluation.
What to Measure in Quantitative Data
Measures of usability should cover:
Effectiveness: Users' ability to complete tasks successfully.
Efficiency: Users' ability to perform tasks without expending excessive resources (e.g., time).
Satisfaction: Users' subjective reactions to using the system.
Measured by the SUS (System Usability Scale).
Five basic types of performance metrics:
Task success.
Time on task.
Errors.
Efficiency.
Learnability.
Types of Performance Metrics
Task Success:
Tasks must have a clear end state or goal.
Binary success (True/False).
Consider types of task failure:
User gives up.
Moderator gives up.
User takes too long.
User does something incorrectly.
Define success levels clearly before you start!
Time on Task:
Important for frequent and routine tasks.
Considerations:
When to start/stop the timer.
Whether to record only successful attempts or all attempts.
Issues:
Users may take longer if using a concurrent think-aloud protocol.
How to establish a target time.
Errors:
Errors lead to:
Loss of efficiency.
Costs to organization or end-user.
Task failure.
Consider how to collect, measure, analyze, and present errors.
Be careful not to “double count” errors.
Efficiency:
How much effort the user expends to achieve their goal (both cognitive and physical).
To consider:
What action(s) are to be measured?
When does the action start and end?
How many ‘meaningful’ actions are required to complete the task?
Measuring “backtracking” is common.
Learnability:
Whether/how users learn to use a new system.
Short term learnability:
Within the span of a single user session/a few hours or days.
Longer term learnability:
Users come back to the task after weeks, months, or years.
Types of Data
Categorical:
Nominal.
Ordinal.
Numerical:
Interval.
Ratio.
Categorical Data
Nominal Data:
Separate categories that do not have a hierarchy or structure.
No category is better or worse than another.
e.g. "Yes/No" responses; hair color.
Analysis: counts, frequencies, proportions.
Visualisation: bar charts, pie charts.
Example: Favorite animals - 60% cats, 25% dogs, 15% mice.
Ordinal Data:
Groups with a clear order or hierarchy.
The elements are not necessarily equally distanced from each other.
e.g. Educational experience (primary -> PhD); "low/med/high".
Analysis: (aggregated) counts and frequencies.
e.g. 40% of users rated the site “good” or “excellent” (“at least good”).
Visualisation: bar charts, pie charts.
Numerical Data
Interval Data:
Data that is measured along a scale.
Each value is of equal distance.
No 'true 0 point'; can be negative (e.g. -10 degrees Celsius).
e.g. Temperature (C or F); time of day.
Analysis: mean, mode, median, standard deviation.
Visualisation: Line graphs, histograms.
Ratio Data:
Data that is measured along a scale.
Each value is of equal distance.
Has a true 0 point.
Can compare, multiply and divide, calculate a ratio.
e.g. Height; age; distance.
Analysis: mean, mode, median, percentile, interquartile range, standard deviation.
Visualisation: Line graphs, histograms.
The importance of data types lies in how they can be mathematically manipulated.
Ratio data allows for statements such as "a participant can take twice as long on Task A as on Task B," while nominal/categorical data does not permit such proportional comparisons.
Analyzing Quantitative Data
Describe the overall tendency:
Mean (average), median (the middle value), mode (the most frequent value).
Describe the distribution:
Range (difference between highest and lowest).
Standard deviation (average distance of data point from the mean).
Identify relationships:
Correlation (how strongly two variables are correlated).
Interpreting the results:
Look into the results.
Propose theories & explanations.
Compare to other researchers’ results.
Correlation and Causality
Correlation: A and B both happen at the same place.
Causality: B happens because A happens. Much harder to demonstrate!
Correlation is relatively easy to demonstrate.
Example: Storks and Babies:
Birth-rate and number of storks correlate.
Explanation 1: storks cause children.
Explanation 2: children cause storks.
Explanation 3: a third unknown aspect causes both (Tertium Quid).
Presenting Quantitative Data
Describe:
Central tendency – the ‘middle’ of the data.
Dispersion or variability – how spread out are the data points?
Measures of central tendency:
Mean – the average for all the data points.
Median – the midpoint in the distribution (half the data are above this, half are below).
Mode – the most commonly occurring value.
Standard Deviation:
How different are the values? (i.e. what is the standard amount that these values deviate from the mean?)
A low standard deviation indicates that values are more similar; higher standard deviation shows that they are quite different.
How to report these statistics:
Where relevant, you will describe your data in the text in your findings, as well as depict them in visualisations.
e.g. "On average, users found this website to be easy to use (M=5.84, SD=0.7)".
Small Samples
Instead of percentages, present information in tables/sentences using raw numbers.
“Two of the seven users were unable to find their bin collection dates."
Outliers:
An outlier is a single instance of a finding, something you have observed in one user only.
Is this person unlike your target users? (disregard)
Is this a problem that is relevant to (some of) your target users? (discuss, but use number not %).
Do you need to know more? (note for future testing).
Reporting outliers example: discuss the specific case and its potential relevance rather than generalizing with percentages.
Quantitative Data Visualization
The Basics of Data Plotting:
Make sure there are clear labels for:
All variable names (with units) - often on graph axes.
Values.
Choose a sensible range of values for each variable.
Remove anything that is unnecessary.
Give each figure a title and a label (e.g. "Fig 1. The number of…").
Pie Charts:
Pie charts are tempting - but not always appropriate.
People aren't good at identifying percentages via slice sizes.
Lots of slices can look too busy.
If you do use pie charts, totals must add up to 100%.
Bar Charts
Line Charts
Nightingale Rose Charts
Collecting Qualitative Data
Qualitative data involves:
Words, transcripts.
Analysis through coding and sorting.
Focus on meaning and experience.
Often interpretative.
Typical methods:
Interviews.
Talk-aloud observations.
Surveys (open responses).
Analyzing Qualitative Data
The Basics:
Look for themes and patterns in the data.
What did you notice during the study?
What did participants say?
Consider critical incidents – helps to focus in on key events.
Remember the original motivating questions for your research – what were your objectives?
Methods for Qualitative Analysis:
Generally will involve coding and/or sorting the data.
Formal Coding (Codebook analysis).
Affinity Mapping.
Thematic Analysis.
Qualitative Coding:
Coding can be:
Top-down: using a pre-existing framework to code your data.
Bottom-up: coding your data based purely on what the data says (no pre-existing framework).
Formal Coding of Usability Problems:
Interface Problems:
Dissatisfaction about an aspect of the interface.
Confusion/uncertainty about an aspect of the interface.
Confusion/surprise at the outcome of an action.
Physical discomfort.
Fatigue.
Difficulty in seeing particular aspects of the interface.
Having problems achieving a goal.
The user has made an error.
Unable to recover from error without external help.
Suggestion for redesign of the interface.
Content Problems:
Dissatisfaction about aspects of the content.
Confusion/uncertainty about aspects of the content.
Misunderstanding of the content (the user may not have noticed this immediately).
Suggestion for re-writing the content.
Affinity Mapping:
Organising your codes across users into distinct clusters to help summarise your data.
You might use a codebook to help you generate clusters (top-down), or you might just let the data lead you (bottom-up).
Presenting Qualitative Data
First Step: Summarise Your Data:
What did you analyse?
How did you analyse it?
What are your key themes/clusters?
What does the reader need to know in order to understand your report?
Presenting Your Themes/Clusters:
Structure presentation by task or theme.
Group related findings together.
Present positive findings before problems.
Present the highest priority problems first.
Finish with “nice to have” suggestions.
Presenting Qual Data - Tips:
Only make claims that your data can support.
Graphical representations may be appropriate for presentation (e.g. visual groups, clusters, hierarchies).
Combine the analysis from the various sources (pre-test, post-test, observations, comments).
You can use tables & dot points to present information.
Use representative quotes!
Using Quotes:
Use quotes to provide evidence of your claims.
Be selective - use the most representative quotes.
Explain your quotes!
Ethics: Anonymisation:
Full disclosure – name participants, provide all details (not generally used).
Change or remove names and key information (most commonly used).
Change or remove names and key information, rephrase quotes, “rogetise” information (not often used – maximum anonymity).
Do not use this data (consent obtained inappropriately or withdrawn).