Evaluating bias in data visualisations: HSC Enterprise Computing Data Visualisation
“Evaluate bias in data collection, storage and analysis when developing visualisations, including accuracy, audience, data source and unconscious bias”
Bias can enter a visualisation during collection, storage and analysis, and through presentation choices such as truncated axes and cherry-picked ranges. Evaluate it through accuracy, audience, data source and unconscious bias, judge how much it distorts the conclusion, and reduce it with representative data, full context and diverse review.
Jump to a section
What this dot point is asking
"Evaluate" means making a judgement about how much bias is present and how it affects the visualisation's conclusions. You need to trace bias through three stages (collection, storage, analysis) and use the four lenses NESA names: accuracy, audience, data source and unconscious bias.
The answer
Bias across the data life cycle
- Collection: unrepresentative samples, leading questions, sensors that work poorly for some groups, and data only from people who use a service.
- Storage: forms that force limited categories, dropped records with missing values, inconsistent coding, and old data that reflects past discrimination.
- Analysis: choosing time ranges, excluding "inconvenient" outliers, choosing averages that hide spread, and inferring causation from correlation.
The four lenses
- Accuracy: do the values and the visual impression match reality? A correct number can still be shown misleadingly.
- Audience: who will view it, what is their data literacy, and is the design exploiting or supporting them?
- Data source: who collected the data, why, and how? Is it disclosed and checkable?
- Unconscious bias: what assumptions did the people collecting, analysing or designing hold without realising?
Presentation techniques that introduce bias
- Truncated axes exaggerate differences (bar charts should start at zero).
- Cherry-picked time ranges create false trends.
- Dual axes with chosen scales make unrelated series look linked.
- 3D and area effects distort size perception.
- Emotive colours and titles push an interpretation ("Crisis!").
- Aggregation can hide subgroups (an overall average hides that one group is worse off).
Reducing bias
- Use representative data and disclose sources and methods.
- Show full context: long time ranges, consistent scales, uncertainty.
- Test the visual with diverse reviewers and the real audience.
- Separate observation (what the data shows) from interpretation.
A bank's dashboard shows loan approval rates by suburb.
- Collection: only people who applied are included; people who never applied because they expected rejection are invisible.
- Storage: postcodes are stored, which can act as a proxy for ethnicity or income.
- Analysis: an overall approval rate of 70% hides that two suburbs are at 40%.
- Presentation: a map colours those suburbs grey (neutral) instead of highlighting the gap.
- Evaluation: the dashboard risks hiding unfair lending. It should break down results by suburb and applicant group, highlight gaps and prompt a fairness review.
- Saying a chart is unbiased because the data is accurate
- Presentation choices can still mislead.
- Only discussing sampling bias
- Use all four lenses.
- Stopping at identification
- Evaluate the effect and suggest improvements.
Practice questions
Original practice questions graded from foundation to exam level, each with a full worked solution. Try them before revealing the solution.
foundation2 marksA bar chart compares two phone plans' customer satisfaction: 82% and 85%. The y-axis starts at 80%. Explain why this is biased.Show worked solution →
Starting the axis at 80% makes the 85% bar look more than twice as tall as the 82% bar, exaggerating a 3-percentage-point difference. Bar length should be proportional to the value, so bar charts should start at zero.
Marking guide: 1 mark for identifying the truncated axis, 1 mark for explaining the exaggeration.
core4 marksA company's recruitment dashboard shows that most successful hires come from three universities, so it recommends only advertising there. Evaluate the bias in this analysis.Show worked solution →
- Data source and collection
- The data only includes people the company already hired. If it historically advertised at those three universities, candidates from elsewhere were never in the pool, so the pattern reflects past practice, not ability.
- Unconscious bias
- Recruiters may have favoured familiar universities, and that bias is now built into the data.
- Accuracy
- "Successful hire" may be defined by past manager ratings, which can also carry bias.
- Judgement
- The recommendation would reinforce the original bias and narrow diversity. The company should widen advertising and compare performance of hires by capability measures rather than university.
Marking guide: 1 mark each for data source, unconscious bias and accuracy issues, 1 mark for a justified judgement.
exam6 marksA political party publishes a line chart showing unemployment falling during its term. The chart starts at the year's peak unemployment month and uses a y-axis from 5.0% to 5.8%. Evaluate the chart for bias with reference to accuracy, audience, data source and unconscious bias, and suggest improvements.Show worked solution →
- Accuracy
- Starting at the peak month (cherry-picking the range) guarantees a fall, and the narrow axis makes a small change look dramatic. The values may be accurate but the impression is not.
- Audience
- Voters scrolling social media are unlikely to check axis labels or the start date, so the design exploits limited attention and data literacy.
- Data source
- If the source is the ABS labour force series, it is reliable, but the chart should name it, state whether it is seasonally adjusted and show the full series so people can check it.
- Unconscious bias
- The designers may genuinely believe the policy worked, so they notice data that supports it and ignore longer-term trends or national factors.
- Improvements
- Show a longer period (for example five years before and during the term), use a clearly labelled axis, cite the ABS series and include context such as the national trend.
- Judgement
- The chart is biased by presentation choices rather than false data, and is misleading for its intended audience.
Marking guide: 1 mark each for accuracy, audience, data source and unconscious bias (4 marks), 1 mark for improvements, 1 mark for a judgement.