Principles of Effective Data Visualisation

Introduction

Data visualisation is the disciplined transformation of data into visual representations that support exploration, explanation, comparison and decision-making. A chart is not merely a decorated table. It is an analytical interface between data, human perception and a particular audience. Effective visualisation enables viewers to identify patterns accurately and efficiently; ineffective visualisation introduces ambiguity, cognitive effort or even deliberate deception.

For students of artificial intelligence and machine learning, visualisation is important throughout the data life cycle. Analysts inspect distributions before modelling, engineers diagnose model errors, and decision-makers interpret predictions through dashboards and reports. A technically correct model can still lead to poor decisions if its results are represented misleadingly.

The central design question is therefore not “Which chart looks attractive?” but “What relationship in the data must the audience perceive, and what visual encoding will communicate it most accurately?”

Learning Objectives

By the end of this chapter, you should be able to:

• explain how visual perception influences interpretation;
• distinguish pre-attentive attributes from attention-demanding analysis;
• apply Gestalt principles to organise visual elements;
• select colour schemes that are meaningful, accessible and appropriate to the data;
• match chart types to analytical purposes and data structures;
• design visualisations with suitable scales, labels, annotations and hierarchy;
• identify clutter, chartjunk, distorted scales and deceptive representations;
• evaluate a visualisation for accessibility, ethics and professional use; and
• create a defensible visualisation specification for an analytical problem.

1. Visualisation as an Encoding and Decoding Process

A visualisation encodes data into graphical marks and attributes. A mark may be a point, line, bar, area, region, symbol or text label. Its attributes may include position, length, angle, size, colour, shape, orientation and texture. The viewer then decodes these attributes using perception, prior knowledge and context.

A useful design model contains five elements:

1. Data: observations, variables, categories, time points or geographical units.
2. Analytical task: comparison, ranking, distribution, relationship, composition, change or location.
3. Visual encoding: the mapping from data values to visual properties.
4. Context: titles, units, source, annotations, benchmarks and explanatory text.
5. Audience and decision: the knowledge level and action expected from the viewer.

The same dataset may require different visualisations for different tasks. A data scientist exploring model residuals may need a scatter plot and interactive filtering. A manager monitoring service quality may need a compact dashboard with reference lines and alert colours. Neither is universally superior; effectiveness depends on purpose.

Important terminology

• Variable: a measured or classified characteristic.
• Quantitative variable: a variable represented by numerical magnitude.
• Categorical variable: a variable whose values identify groups or labels.
• Ordinal variable: a categorical variable with meaningful order.
• Encoding: representing data using visual properties.
• Scale: the rule mapping data values to visual positions or sizes.
• Legend: an explanation of symbols, colours or patterns.
• Small multiples: repeated, consistently scaled charts used to compare subsets.
• Dashboard: a coordinated collection of visualisations designed for monitoring or decision support.

2. Visual Perception and Graphical Accuracy

Human vision processes some visual properties rapidly and accurately, while other properties require conscious comparison. Position along a common scale is generally easier to compare than angle, area or volume. Consequently, a bar chart is usually more precise for comparing values than a pie chart, and a dot plot often permits more accurate comparison than bubbles whose areas must be judged.

A practical hierarchy of quantitative encodings is:

1. position on a shared axis;
2. position on non-aligned scales;
3. length;
4. angle or slope;
5. area;
6. volume, shading or colour intensity.

This is not an absolute law. A carefully designed map, for example, may use colour to communicate a spatial pattern effectively. The hierarchy is a design guide: when exact comparison is important, prefer higher-accuracy encodings.

2.1 Preattentive attributes

Preattentive attributes are visual features detected rapidly and with little conscious effort. Examples include a unique colour, a sharp change in orientation, a distinctive shape, enclosure, position and considerable size difference. They are useful for directing attention to a relevant observation or exception.

For example, a scatter plot of 10,000 transactions may show all points in grey and highlight suspected fraud in orange. The orange points become a visual entry point for investigation. Preattentive emphasis should be limited: if every point is bright or every label is bold, nothing has priority.

Preattentive coding is not identical for all viewers. Colour perception, cultural conventions, visual impairment, display quality and surrounding colours influence interpretation. It should therefore support—not replace—labels, symbols and explanatory context.

2.2 Visual hierarchy and attention

A visual hierarchy establishes an order in which the viewer encounters information. A strong hierarchy commonly proceeds from:

• title and principal message;
• main pattern or comparison;
• labels, units and reference values;
• detailed annotations and source information.

Hierarchy can be created through scale, whitespace, contrast, position, typography and controlled colour. The most important visual element should not compete with a decorative logo, oversized title or unnecessary background.

3. Gestalt Principles for Organising Visualisations

Gestalt psychology describes tendencies through which people organise visual elements into meaningful groups. These principles are not mechanical rules, but they provide a useful foundation for layout and interaction design.

Proximity

Objects placed close together are perceived as related. In a dashboard, a metric and its explanatory sparkline should be spatially grouped. Excessive spacing between a chart and its legend can cause the viewer to search unnecessarily.

Similarity

Objects sharing colour, shape, size or orientation appear related. Consistent colours for product categories across several charts reduce interpretation time. Similarity must not imply a relationship that does not exist; identical colours should not be used for unrelated meanings.

Enclosure

A boundary or lightly shaded region can group elements. Enclosure is helpful for separating panels, but heavy boxes around every item create visual noise.

Continuity

The eye tends to follow smooth paths. A line chart uses continuity to communicate change over an ordered variable such as time. Lines should not connect categories when no meaningful order exists.

Closure

Viewers tend to complete incomplete shapes. Minimalist design can use this tendency, but excessive omission may make a chart ambiguous, particularly for inexperienced audiences.

Figure–ground

Viewers distinguish a focal figure from its background. A neutral background and strong contrast for data marks improve readability. Gridlines should usually remain subordinate to the data.

Common region and connectedness

Elements within the same region, or joined by a line, are perceived as associated. These principles are useful in network diagrams and process flows, but connections must represent real relationships rather than merely improve appearance.

4. Choosing a Chart for the Analytical Task

Chart selection should begin with the question being asked, not the software menu. The following table provides a practical guide.

Bars, dots and lines

Bars encode magnitude through length from a baseline and are effective for discrete category comparison. For standard bars, a baseline of zero is normally expected because truncated bars can exaggerate differences. A dot plot may use a non-zero axis more safely when the numerical positions are clearly labelled, although the choice must still be disclosed.

Line charts are appropriate when the horizontal variable has an ordered progression, especially time. Connecting unrelated categories with a line can falsely suggest continuity. Multiple lines require restrained use of colour and direct labelling where possible.

Distributions

A histogram groups numerical observations into bins. Its interpretation depends strongly on bin width and boundary placement. A box plot summarises a distribution using the median, quartiles and possible outliers, but it can conceal multimodality and sample size. A strip plot or jittered dot plot may be preferable for small datasets because it displays individual observations.

Relationships

A scatter plot is useful for examining association between two quantitative variables. A trend line can summarise an association, but it does not establish causality. Overplotting can hide dense regions; transparency, sampling, hexbinning or aggregation can help. Axes, units and missing values must be documented.

Composition

A stacked bar chart shows how parts contribute to a total. Only the baseline segment is easy to compare precisely across categories; interior segments are harder to compare. A pie chart may be acceptable for a small number of clearly different parts of a single total, but it is generally weak for precise comparisons or many categories. Composition charts must distinguish between parts that sum to a meaningful whole and unrelated metrics that merely appear together.

5. Scales, Axes and Numerical Integrity

A scale is an argument about how numerical differences should appear visually. It must be truthful, interpretable and appropriate to the data.

Baselines and truncation

A bar chart whose vertical axis begins at 95 rather than 0 may make values of 96 and 100 appear dramatically different. This is potentially deceptive because bar length conventionally represents magnitude from a baseline. If a truncated axis is necessary, a dot or line chart may be more appropriate, or the truncation should be prominent and justified.

Truncation is not always wrong. In a line chart showing temperature changes between 19 °C and 21 °C, a focused axis can reveal small variations. The chart should make the scale visible, avoid heavy visual emphasis on the difference, and provide sufficient context.

Unequal intervals and logarithmic scales

Tick marks must represent equal numerical intervals unless the scale explicitly communicates otherwise. A logarithmic scale is legitimate for data spanning several orders of magnitude, but it must be labelled clearly. On a base-10 logarithmic scale, equal distances represent equal ratios rather than equal differences. Thus, the distance from 1 to 10 is the same as from 10 to 100.

Dual axes

Dual-axis charts can compare variables with different units, but they are easy to manipulate. Changing either axis range can create or eliminate an apparent relationship. Use separate aligned panels when possible. If dual axes are necessary, clearly label both axes, use visually distinct encodings and explain the analytical reason.

Area and volume distortion

If a circle’s radius is doubled, its area becomes four times as large because:

where A is area and r is radius. A three-dimensional sphere’s volume becomes eight times as large when its radius doubles:

where V is volume. Therefore, scaling a pictograph by width and height can exaggerate differences. Symbols should generally be scaled by area when area is intended to represent magnitude, and three-dimensional effects should be avoided unless volume itself is analytically meaningful.

Percentages and denominators

Percentages require a stated denominator. A rise from 2 cases to 3 cases is a 50% increase but only one additional case. A chart should show counts, rates or both when the denominator affects interpretation. Comparisons across populations require consistent definitions and, where relevant, normalisation such as cases per 1,000 people. Rates must not be compared without considering exposure, time period and population composition.

6. Colour Theory and Accessibility

Colour has three commonly discussed dimensions:

• Hue: the colour family, such as blue or orange;
• Lightness: perceived brightness from dark to light;
• Chroma or saturation: intensity or vividness.

Colour should have a declared semantic role. It may identify categories, show ordered magnitude, distinguish positive and negative values, or highlight an exception. A single visualisation should not use the same colour both for category identity and for quantitative intensity unless the meanings are unambiguous.

Colour schemes

• Categorical schemes use distinct hues for nominal groups. They should contain a limited number of clearly distinguishable colours.
• Sequential schemes move from light to dark to represent ordered values from low to high.
• Diverging schemes use two contrasting hues with a meaningful central value, such as zero or a target.
• Qualitative emphasis schemes use a neutral majority and one accent colour for the focal item.

Rainbow palettes are often poor for quantitative data because lightness is not ordered consistently and some colour differences are difficult to perceive. A diverging palette should not be used when the midpoint has no substantive meaning.

Accessibility

Approximately 8% of men and 0.5% of women with Northern European ancestry have some form of red–green colour-vision deficiency; prevalence varies by population and condition. Regardless of prevalence, accessible design is good professional practice.

Use colour together with direct labels, line styles, shapes or patterns. Check contrast between text and background, test charts in grayscale, and use colour-vision simulation tools where available. Do not communicate “good” and “bad” solely through green and red. Provide text alternatives and meaningful descriptions for digital charts, and ensure interactive elements can be operated with keyboard navigation where applicable.

Accessibility also includes language, font size, cognitive load, screen size and data literacy. A chart that is technically compliant but unreadable on a mobile device is not effective communication.

7. Clarity, Composition and Information Design

Reduce chartjunk

Chartjunk consists of non-data elements that consume attention without improving interpretation. Common examples include decorative illustrations, excessive borders, gradients, 3-D perspective, ornamental shadows, dense backgrounds and redundant labels. The aim is not to remove every non-data element, but to ensure that every element has a communicative function.

Use titles and annotations analytically

A descriptive title identifies the subject; an explanatory title can state the principal finding. Subtitles should define population, period and measure. Annotations can identify an event, threshold or unusual observation. They should explain why a feature matters rather than merely repeat its value.

Labels and legends

Direct labels reduce the need to move between chart and legend, particularly for a small number of series. Units should appear in axis titles or subtitles. Abbreviations must be familiar or defined. A source and data date are essential for professional reports.

Small multiples and consistency

Small multiples can replace a cluttered multi-series chart by repeating the same design for groups. Shared scales support comparison; independent scales reveal within-group variation but hinder cross-group comparison. The choice should be stated or made obvious.

Interaction and dashboards

Interaction should answer a real analytical need, such as filtering, drilling down or viewing exact values on demand. It should not conceal essential information. A dashboard should provide an overview first, allow focused exploration second, and preserve context during filtering. Avoid excessive animation because motion attracts attention and can make comparison difficult.

8. Avoiding Deceptive Visualisation

Deception may be intentional or accidental. Common sources include:

• truncated or irregular axes;
• omitted data periods or categories;
• inconsistent denominators;
• cherry-picked time windows;
• inappropriate aggregation;
• misleading area or volume scaling;
• dual axes that imply correlation;
• hidden uncertainty or missing observations;
• colour schemes that exaggerate differences; and
• titles that claim causation from observational association.

A trustworthy visualisation makes design choices inspectable. It states the source, population, time period, units, transformations and exclusions. If smoothing, indexing or normalisation has been applied, the method should be disclosed.

Uncertainty should be represented when it affects a decision. Possible methods include confidence intervals, prediction intervals, error bars, shaded bands and distributional displays. Error bars must be defined: they may represent standard deviation, standard error, a confidence interval or another quantity. These are not interchangeable.

9. Visualisation in Artificial Intelligence and Machine Learning

Visualisation supports both model development and responsible communication.

Data understanding

Histograms, box plots and missingness matrices reveal skew, outliers, imbalance and data-quality problems. Scatter plots can expose nonlinear relationships and possible leakage. Grouped visualisations may reveal that a pattern differs across demographic or operational groups.

Model evaluation

A confusion matrix shows counts or rates of predicted versus actual classes. Precision–recall curves are often informative for imbalanced classification, while receiver operating characteristic curves show ranking performance across thresholds. Calibration plots compare predicted probabilities with observed frequencies. These charts answer different questions and must not be treated as interchangeable measures of model quality.

Residual plots can reveal systematic error, changing variance or neglected structure. Feature-importance charts should distinguish association from causation and should state the importance method, because permutation importance, tree-based importance and local explanation methods have different interpretations.

Explainability and risk

Explanatory charts should communicate uncertainty and limitations rather than presenting a model as objective or infallible. A fairness dashboard may compare selection rates, false-positive rates or calibration across groups, but group comparisons require appropriate definitions, sample sizes and privacy safeguards. Suppressing small cells can protect confidentiality, while excessive aggregation may conceal harm.

10. A Professional Design Workflow

A repeatable workflow improves quality:

1. Define the decision: What should the audience understand or do?
2. Identify the analytical task: comparison, trend, distribution, relationship, composition or location.
3. Audit the data: types, units, missing values, outliers, denominators and uncertainty.
4. Select an encoding: prefer position and length for precise comparison.
5. Choose a chart form: select the simplest chart that supports the task.
6. Construct honest scales: use appropriate baselines, intervals and transformations.
7. Establish hierarchy: make the principal message visually prominent.
8. Add context: title, labels, units, source, date, definitions and annotations.
9. Test accessibility: contrast, colour alternatives, text size, keyboard access and mobile layout.
10. Review critically: ask whether a reasonable viewer could reach a false conclusion.
11. Validate with users: observe whether representative viewers can answer the intended question.
12. Document the specification: preserve data preparation, calculations and design decisions.

Worked Example: Comparing Service Delays

A university service centre records the following median response times for four request types:

The analytical task is to compare response times and identify where improvement may be needed. An ordered horizontal bar chart is appropriate because there are four discrete categories and the audience must compare magnitudes. The bars should begin at zero, use a single neutral colour, and include data labels in hours. If the centre’s target is 4 hours, a reference line at 4 hours should be added. A highlight colour can identify the two request types exceeding the target.

A pie chart would be less suitable because the values are not parts of a common total. A line chart would imply that request types form a continuous sequence. A three-dimensional bar chart could distort apparent lengths. The final design should state whether the median was calculated over a particular week or month, how many requests were included and whether unresolved requests were excluded.

Applied Case: Illustrative Hospital Readmission Dashboard

Background

This fictional case concerns a regional hospital seeking to reduce 30-day readmissions. Its analytics team prepares a dashboard for clinical managers. The data include monthly readmission counts, admissions, patient age groups, diagnosis categories and model-generated risk scores.

Problem

An initial dashboard displays a large red percentage labelled “readmissions,” but the denominator is not visible. Managers interpret an increase as a deterioration in care. In fact, the number of readmissions increased while total admissions increased more rapidly, so the readmission rate fell. The dashboard also uses red and green alone to distinguish performance and presents a model’s risk score as if it were a probability of readmission.

Analysis

The team first defines the measures. A rate can be expressed as:

The numerator is the number of eligible patients readmitted within 30 days, and the denominator is the number of eligible discharges in the same period. The calculation is valid only when eligibility rules, time window and population definition are consistent. The team checks exclusions, delayed reporting and changes in coding practice.

The dashboard is redesigned with a line chart of monthly rate and a separate column showing the count. A subtitle identifies the denominator and eligibility rule. A target reference line is included, and a shaded band represents the uncertainty or monitoring threshold agreed with clinical governance. A small-multiple display compares diagnosis groups using the same vertical scale. The model panel labels its output as a risk score and provides calibration information rather than calling it a guaranteed probability.

For accessibility, the team replaces red–green status colours with blue–orange accents and adds symbols and text labels. Patient groups with very small counts are suppressed or combined according to privacy policy. A footnote records data refresh date, exclusions and responsible data owner.

Outcome

Managers can distinguish count from rate, identify the diagnosis groups requiring investigation and understand the model’s limitations. The dashboard does not claim that visual patterns prove a cause. Instead, it supports questions for clinical review and quality-improvement analysis.

Learning points

• Every rate requires a visible and appropriate denominator.
• Counts and rates answer different questions and may move in opposite directions.
• A model score must not be presented as a calibrated probability without evidence.
• Accessibility and privacy are part of visualisation quality, not optional additions.
• Governance metadata increases trust and reproducibility.

Challenges and Limitations

Visualisation cannot repair biased, incomplete or poorly defined data. Aggregation may conceal variation between groups, while disaggregation may create privacy risks or unstable estimates. Interactive charts can improve exploration but may make results difficult to reproduce if filters and state are not recorded. Automated chart generation can produce visually polished but semantically inappropriate outputs, especially when software chooses defaults without understanding the analytical task.

Human perception also varies. Viewers differ in experience, language, colour vision and expectations. User testing is therefore preferable to relying solely on designer intuition. Finally, a clear visualisation can still be misinterpreted when circulated without its title, source or explanatory text. Charts should be designed for their actual medium and context of use.

Contemporary Developments and Professional Practice

Modern visualisation systems increasingly combine statistical graphics, natural-language interfaces, geospatial analysis and machine-learning assistance. These tools can recommend chart types, detect anomalies or generate narrative summaries. However, automated recommendations require human review for scale integrity, accessibility, uncertainty, privacy and semantic validity.

Reproducible visualisation is becoming standard practice. Analysts should preserve source data or a governed extract, transformation code, chart specification, version information and publication date. Declarative systems such as grammar-based visualisation tools encourage explicit separation of data, marks, encodings and transformations. This supports review and reuse, but it does not remove the need for domain knowledge.

Responsible practice also requires attention to data protection, confidentiality, intellectual property, accessibility legislation and organisational governance. In sensitive domains such as health, education, employment and public safety, visualisations can influence allocation of resources and perceptions of groups. Ethical review should consider who benefits, who may be misrepresented and whether uncertainty is being hidden.

Best-Practice Checklist

Before publishing a visualisation, ask:

• Is the analytical purpose explicit?
• Does the chart type match the data and task?
• Are the scale, baseline, units and denominators clear?
• Are transformations, exclusions and uncertainty disclosed?
• Does the visual hierarchy direct attention appropriately?
• Are colour, contrast, labels and symbols accessible?
• Have chartjunk, unnecessary decoration and redundant legends been removed?
• Could a reasonable viewer infer a false comparison or causal claim?
• Are privacy, fairness and governance requirements satisfied?
• Can another analyst reproduce the result?

Chapter Summary

Effective data visualisation combines data understanding, perceptual principles, graphical integrity and communication purpose. Position and length generally support more accurate comparison than area, angle or volume. Preattentive attributes can direct attention, while Gestalt principles such as proximity, similarity, continuity, enclosure and figure–ground help organise information.

Chart selection should follow the analytical task: bars and dots support comparison, lines show ordered change, distribution plots reveal variability, scatter plots examine relationships, and composition charts show meaningful parts of a whole. Honest scales, visible denominators, appropriate transformations and explicit uncertainty are essential. Colour should have a clear semantic role and must be supplemented for accessibility. Clutter, chartjunk, excessive interaction, decorative perspective and deceptive axes reduce trust and comprehension.

In AI and machine-learning practice, visualisation supports data quality assessment, model evaluation, explainability and fairness analysis. Professional visualisation requires documentation, reproducibility, privacy protection and critical review. The best visualisation is not the most elaborate one; it is the one that enables the intended audience to perceive the relevant evidence accurately and act responsibly.

Key Terms

Visual encoding: Mapping data values to graphical properties.

Preattentive attribute: A visual feature detected rapidly with limited conscious effort.

Gestalt principle: A perceptual tendency used to organise visual elements.

Visual hierarchy: The ordering of attention established by layout and emphasis.

Scale: A rule mapping data values to visual positions or sizes.

Baseline: The reference level from which a graphical magnitude is displayed.

Chartjunk: Non-data decoration that distracts from interpretation.

Categorical colour scheme: Distinct colours used to identify nominal groups.

Sequential colour scheme: Ordered lightness or intensity used for ordered values.

Diverging colour scheme: Two-sided colour progression around a meaningful midpoint.

Accessibility: Design that permits people with varied abilities and conditions to use and interpret information.

Small multiples: Repeated charts using a consistent design to compare subsets.

Overplotting: Visual overlap that hides individual observations in a chart.

Dual-axis chart: A chart using two value axes, often with different units or ranges.

Denominator: The reference population or quantity used to calculate a rate or percentage.

Uncertainty: The range or lack of precision associated with a measurement or estimate.

Calibration: Agreement between predicted probabilities and observed frequencies.

Dashboard: A coordinated set of visual displays for monitoring or decision support.

Data provenance: Information about a dataset’s origin, preparation and transformation.

Deceptive visualisation: A representation likely to produce a materially false or misleading interpretation.

Review Questions

Short Answer Questions

1. What is a preattentive attribute, and how can it be used responsibly?
2. Why is position on a common scale usually more accurate than area for comparison?
3. Distinguish a sequential colour scheme from a diverging colour scheme.
4. Why should a line chart generally be used only when the horizontal variable has meaningful order?
5. State two ways in which a visualisation can support accessibility.

Long Answer Questions

1. Explain the major Gestalt principles and show how they can improve the design of a dashboard.
2. Compare bar charts, line charts, histograms, scatter plots and stacked bars in terms of analytical purpose and limitations.
3. Discuss how truncated axes, dual axes, area scaling and omitted denominators can produce misleading visualisations.
4. Develop a professional workflow for designing and validating a visualisation for a non-technical audience.
5. Explain the role of visualisation in machine-learning data preparation, model evaluation and responsible communication.

Analytical / Application Questions

1. A report shows sales for four regions using large 3-D cylinders. The vertical axis begins at 80, and the chart uses a red–green legend. Identify at least four design problems and propose a revised visualisation.
2. A hospital’s readmissions rise from 120 to 150, while eligible discharges rise from 2,000 to 3,000. Calculate both readmission rates and explain why counts alone could lead to an incorrect conclusion. Use the rate definition given in the case.
3. Design a visualisation specification for examining whether a fraud-detection model performs differently across customer groups. State the chart types, measures, encodings, accessibility provisions, uncertainty considerations and governance safeguards you would use.

Multiple Choice Questions with Answers

Question 1

Which visual encoding generally supports the most accurate comparison of quantitative values?

A. Position on a common scale

B. Three-dimensional volume

C. Colour hue alone

D. Decorative texture

Correct Answer: A. Position on a common scale

Question 2

Which chart is most appropriate for showing the distribution of a numerical variable and its quartiles?

A. Pie chart

B. Box plot

C. Network diagram

D. Choropleth map

Correct Answer: B. Box plot

Question 3

What is the main purpose of a preattentive attribute?

A. To hide missing values

B. To direct attention rapidly to a relevant feature

C. To replace all labels

D. To prove causation

Correct Answer: B. To direct attention rapidly to a relevant feature

Question 4

Which colour scheme is most suitable for values ordered from low to high without a meaningful midpoint?

A. Sequential

B. Diverging

C. Random categorical

D. Red–green status coding

Correct Answer: A. Sequential

Question 5

Why can a truncated baseline be misleading in a bar chart?

A. It prevents categorical sorting

B. It can exaggerate apparent differences in bar lengths

C. It makes all values equal

D. It removes the need for units

Correct Answer: B. It can exaggerate apparent differences in bar lengths

Question 6

Which chart is generally best for examining the relationship between two quantitative variables?

A. Scatter plot

B. Pie chart

C. Stacked area chart of unrelated categories

D. Single-value KPI card

Correct Answer: A. Scatter plot

Question 7

In the hospital case, why must the readmission denominator be stated?

A. Because percentages have no meaning without a reference quantity

B. Because counts cannot be displayed on dashboards

C. Because all rates must use population age as denominator

D. Because a denominator determines the chart colour

Correct Answer: A. Because percentages have no meaning without a reference quantity

Question 8

What is chartjunk?

A. A missing data record

B. A statistical confidence interval

C. Non-data decoration that distracts from interpretation

D. A logarithmic transformation

Correct Answer: C. Non-data decoration that distracts from interpretation

Question 9

A logarithmic base-10 axis means that equal visual distances represent equal:

A. Absolute differences

B. Ratios or multiplicative changes

C. Category counts only

D. Percentages of zero

Correct Answer: B. Ratios or multiplicative changes

Question 10

Which practice best improves accessibility when colour carries meaning?

A. Use more saturated colours only

B. Add labels, shapes or line styles as redundant cues

C. Remove all text from the chart

D. Use red and green for every status

Correct Answer: B. Add labels, shapes or line styles as redundant cues