Chart Selection and Effective Visual Encoding

Introduction

Data visualization is not simply the production of attractive graphics. It is the disciplined transformation of data into visual representations that help people detect patterns, compare alternatives, identify exceptions and make decisions. In business analytics, a chart is successful when its visual structure matches the analytical question and communicates the intended evidence with minimal interpretation effort.

Selecting a chart therefore begins with the purpose of the analysis rather than with a software menu. A manager may ask: Which region has the highest sales? How have costs changed over time? What is the distribution of delivery times? Is advertising expenditure associated with revenue? What proportion of customers belongs to each segment? Each question implies a different visual task. A chart designed for one task may be ineffective or misleading for another.

This chapter presents a practical framework for selecting and constructing charts in Tableau and Power BI. It explains visual encoding, chart families, scale design, clutter reduction, interaction and accessibility. The emphasis is on comparison, distribution, composition, relationship, change over time and ranking—the common analytical purposes encountered in organisational reporting.

Learning Objectives

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

1. Explain how data attributes are represented through visual channels.
2. Select an appropriate chart for comparison, distribution, composition, relationship, time change and ranking.
3. Construct effective charts in Tableau and Power BI using suitable dimensions, measures and aggregation levels.
4. Detect and correct misleading scales, inappropriate baselines, excessive decoration and visual clutter.
5. Design charts that support accessibility, ethical communication and responsible decision-making.
6. Evaluate a dashboard or report for analytical clarity and visual integrity.

1. The Analytical Purpose Comes First

A useful chart-selection process begins with four questions:

1. What is the analytical task? For example, comparison, trend detection or relationship assessment.
2. What are the data types? Categorical, ordinal, quantitative, temporal, geographic or textual.
3. What is the intended audience and decision? An executive summary, operational investigation and statistical exploration require different levels of detail.
4. What level of aggregation is appropriate? Daily observations, monthly totals and customer-level records answer different questions.

A chart can be technically correct but analytically unsuitable. A pie chart may accurately display percentages while making close comparisons difficult. A line chart may show a smooth trend even though the horizontal values are irregularly spaced. A map may attract attention without revealing a meaningful geographic pattern. Accuracy is necessary, but usefulness also depends on perceptual effectiveness and contextual honesty.

Important terminology

• Dimension: A field used to group, categorise or describe data, such as Region or Product Category.
• Measure: A quantitative field that can be aggregated, such as Sales, Profit or Quantity.
• Mark: A visual object representing a record, category or aggregated value.
• Encoding: The mapping of a data attribute to a visual property.
• Aggregation: The process of summarising observations using totals, averages, counts or other functions.
• Granularity: The level of detail represented in a dataset or view.
• Scale: The mapping between data values and positions, lengths, colours or other visual properties.
• Dashboard: A coordinated collection of views designed to support monitoring or analysis.

2. Visual Encoding: Turning Data into Perception

Visual encoding is the process of representing data through position, length, angle, area, colour, shape, size or other visual channels. The effectiveness of a channel depends on how accurately people can perceive and compare it.

For quantitative comparison, position on a common scale is generally the most precise channel. Length, as used in bars, is also effective. Angles and areas are less precise, which is why a bar chart usually supports more accurate comparisons than a pie chart or bubble chart. Colour hue is useful for distinguishing categories, but is weak for showing ordered magnitude. Colour intensity can suggest magnitude, provided that the scale is perceptually ordered and explained.

A practical approximate hierarchy for comparing numerical values is:

1. Position along a common axis.
2. Position along non-aligned axes.
3. Length.
4. Angle or slope.
5. Area.
6. Colour intensity or saturation.
7. Shape or colour hue for categorical distinction.

This is not an absolute law. Context, chart construction and audience experience matter. However, it is a useful design principle: assign the most important quantitative message to the most accurately perceived channel.

Preattentive attributes

Some visual properties are detected rapidly before conscious inspection. These include colour contrast, position, enclosure, orientation and size. They can direct attention to an exception or target value. Used carefully, they improve a chart; used excessively, they create competing focal points. For example, highlighting one region in a muted bar chart can focus attention on a selected business unit. Highlighting every category removes the distinction.

Data types and visual implications

3. A Chart-Selection Framework

The following framework connects a business question to a chart family.

The table is a starting point, not a mechanical rule. A chart must also consider the number of categories, whether values are positive or negative, the need for precise reading and the possibility of overlapping marks.

4. Charts for Comparison and Ranking

Bar charts

Bar charts are appropriate when comparing a quantitative measure across discrete categories. Bars should normally share a common baseline, especially when their lengths imply magnitude. Horizontal bars are particularly effective for long category names and ranked displays.

A professional bar chart should:

• use a meaningful category order, usually descending for ranking;
• label the measure and units;
• avoid unnecessary three-dimensional effects;
• use one neutral colour unless colour encodes a meaningful variable;
• display data labels when exact values are important and the number of marks is manageable;
• distinguish zero, missing and negative values clearly.

A grouped bar chart compares subcategories within each major category. It works when the number of groups and series is small. If there are many series, the chart becomes difficult to scan. A small-multiple design or a line of dots may be clearer.

A stacked bar chart supports part-to-whole comparison, but only the segments sharing a baseline are easy to compare. Middle segments are difficult to compare across bars. If comparing each component is the main purpose, use grouped bars or separate panels instead.

Dot plots and lollipop charts

Dot plots place values along a common scale and often provide more efficient comparison than bars when many categories are present. Lollipop charts can reduce visual weight, but they should not be used merely for decoration. A simple dot plot is often the clearer choice.

Ranking

Ranking requires ordering categories explicitly. Alphabetical order is rarely suitable when the question concerns performance. A ranked chart should show the ranking variable, period and direction of “better” or “worse.” If a ranking is volatile, include the time period and avoid presenting a temporary rank as a permanent status.

In Tableau, ranking can be created by sorting a dimension by a measure or by applying a rank table calculation. In Power BI, a bar chart can be sorted by the selected measure, and a rank measure can be created with DAX when ranking must respond to filters. The analyst should verify that the ranking is recalculated at the intended filter context.

5. Charts for Distribution

A distribution describes how values are spread, including centre, variability, skewness, gaps and unusual observations.

Histograms

A histogram groups a quantitative variable into contiguous intervals called bins. The horizontal axis represents value intervals and the vertical axis represents frequency or, in some designs, density. A histogram is suitable for examining the shape of one numerical variable, such as order value or delivery time.

Bin width affects interpretation. Very wide bins conceal structure; very narrow bins produce noise. There is no universally correct bin width. It should reflect sample size, measurement precision and the decision context. The number of observations and the bin definition should be available to the reader.

A histogram is not the same as a bar chart. Histogram bars represent adjacent numerical intervals, so gaps between bars usually have a specific meaning. Bar-chart categories are discrete, and their order may be changed.

Box plots

A box plot summarises a distribution using the median, quartiles and possible outliers. The interquartile range is:

where Q₁ is the first quartile and Q₃ is the third quartile. Under a common convention, observations below Q₁ − 1.5 × IQR or above Q₃ + 1.5 × IQR are displayed as potential outliers. This rule is a visual convention, not proof that an observation is erroneous. Outliers may represent genuine business events, data-quality problems or a different population.

Box plots are useful for comparing distributions across regions, products or departments. They conceal some detail, particularly when sample sizes are small. Combining a box plot with jittered individual points can reveal both summary and observation-level information.

Density and strip plots

Density plots estimate the relative concentration of observations. They can reveal multiple peaks but may be difficult for unfamiliar audiences and depend on a smoothing parameter. Strip plots display individual values and are useful for small or moderate datasets. The choice should balance statistical detail with interpretability.

6. Charts for Composition

Composition concerns how a total is divided among components. The analyst must decide whether the important question is the total, the component values or the percentage share.

Stacked bars

Stacked bars show components within totals. They are effective when the number of components is limited and the overall total matters. Use consistent segment colours and place the most important or largest segment at a stable baseline where possible.

A 100% stacked bar normalises each bar to the same total and shows percentage composition rather than absolute size. It is appropriate for comparing shares across categories, but it hides differences in total volume. A report should not use it when both total size and composition are essential unless the totals are shown separately.

Pie and doughnut charts

Pie charts can communicate a small number of clearly different shares, especially when one category dominates. They are poor for precise comparison among similar shares because people judge angles and areas less accurately than aligned lengths. Avoid pie charts with many slices, long labels or small differences. Doughnut charts do not solve this perceptual limitation; the central hole may reduce available area without improving comparison.

Treemaps

Treemaps use nested rectangles to represent hierarchical composition. They can display many categories compactly, but area comparisons are difficult and small items may be unreadable. A treemap is best for exploration or high-level pattern recognition, not for precise comparison.

Composition can also be represented with a table, a 100% stacked bar or a set of aligned bars. The best choice depends on whether the audience needs exact values, shares, hierarchy or a compact overview.

7. Charts for Relationships

Scatter plots

A scatter plot places one quantitative variable on each axis. Each mark represents an observation or an aggregated group. Scatter plots are appropriate for examining association, clusters, non-linearity and potential outliers.

A visible association is not proof of causation. Confounding variables, selection effects, time trends and reverse causality may explain the pattern. Analysts should show sample size where relevant and avoid fitting a trend line without explaining its purpose.

Correlation is commonly summarised by Pearson’s correlation coefficient:

where cov(X,Y) is the covariance between variables X and Y, and σₓ and σᵧ are their standard deviations. The coefficient is dimensionless and ranges from −1 to +1. Its interpretation assumes that a linear association is meaningful; it can be strongly affected by outliers and does not establish causation.

For a sample of n paired observations, an equivalent computational form is:

The formula applies when the paired observations are appropriately measured and the denominators are non-zero. The units cancel because covariance and the product of standard deviations have the same units.

A scatter plot with colour, size and shape may encode additional variables, but adding channels increases cognitive load. Use small multiples or filters when the plot becomes crowded.

Heat maps

Heat maps encode values in a matrix using colour intensity. They are useful for identifying patterns across two categorical dimensions, such as product category by month. Because colour is less precise than position, include values or supporting labels when exact comparison matters. Use a sequential palette for ordered non-negative values and a diverging palette when a meaningful central value separates positive and negative results.

8. Charts for Change over Time

Time-series charts require a genuine temporal order and an appropriate time interval. Line charts are effective when the continuity of time matters and observations are sufficiently frequent. Columns may be preferable for discrete periods such as annual totals or monthly counts when individual periods must be emphasised.

Best practices include:

• use a chronological axis rather than alphabetical month names;
• distinguish calendar year, financial year and rolling period;
• avoid connecting unrelated categories with lines;
• show missing periods rather than silently interpolating them;
• identify whether values are totals, averages, rates or index values;
• use annotations for significant events, policy changes or data breaks.

An indexed line chart can compare growth from different starting values. If a series X begins at X₀ and has later value Xₜ, its index is:

The index is unit-free and equals 100 at the base period. It compares relative change, not absolute level. Two series with equal index values may have very different actual values.

Dual axes should be used sparingly. They can create an apparent relationship by allowing independent scales to be adjusted visually. Prefer separate panels, normalisation or a clearly explained common unit. If dual axes are necessary, use distinct visual treatments, label both axes prominently and explain the scales.

9. Scale, Axis and Baseline Integrity

Scales determine how visual differences are perceived. A truncated bar-chart axis can exaggerate differences because bar length no longer corresponds naturally to the full magnitude from zero. For example, values of 98 and 100 differ by only about 2% in magnitude, but a bar chart beginning at 95 may make the difference look much larger.

For bars representing magnitude, begin the quantitative axis at zero unless there is a clear and disclosed reason not to. Line charts may use a non-zero range when the purpose is to show variation over time, but the axis should not conceal important context. Always display units, tick marks and sufficient scale information.

Common scale problems include:

• unequal intervals presented as equally spaced;
• inconsistent axes across small multiples;
• percentages displayed without a denominator;
• logarithmic scales used without explanation;
• automatic aggregation that changes when filters are applied;
• mixing nominal and real monetary values without stating the price basis.

A logarithmic scale can be appropriate for multiplicative growth or values spanning several orders of magnitude. It should be labelled clearly because equal distances represent equal ratios rather than equal absolute differences.

10. Reducing Clutter and Improving Reading

Clutter is visual content that does not support the analytical task. It includes unnecessary borders, heavy gridlines, redundant legends, excessive decimals, decorative images and too many colours.

A useful editing sequence is:

1. Remove marks and labels that do not support the question.
2. Sort categories by analytical meaning.
3. Reduce the number of colours and use a consistent semantic palette.
4. Replace legends with direct labels where feasible.
5. Use concise titles that state the finding or question.
6. Add annotations only where they explain a meaningful event.
7. Check readability at the size and device on which the report will be used.

The principle of data-ink efficiency is helpful: visual emphasis should primarily represent data rather than decoration. It does not mean removing all white space or every gridline. A faint gridline can support accurate reading; a dark background and ornamental gradients may not.

11. Constructing Charts in Tableau and Power BI

The software should implement a sound analytical design, not determine it accidentally.

Tableau workflow

In Tableau, place dimensions and measures on shelves or the Marks card. Select the mark type that matches the task: bars for category comparison, lines for time, circles or shapes for scatter plots, and square marks for heat maps. Confirm whether a measure is being summed, averaged, counted or treated as a continuous field. Use filters deliberately and inspect the generated view for unexpected aggregation.

Useful Tableau practices include:

• use discrete fields for categories and continuous fields for numerical axes when appropriate;
• apply sorting based on the displayed measure;
• use reference lines, bands or box plots to provide context;
• use parameters for controlled user choices rather than exposing every field;
• use dashboard actions to connect views without overwhelming the user.

Power BI workflow

In Power BI, choose a visual from the Visualizations pane and assign fields to its wells. Check the summarisation setting for each measure. Prefer explicit measures when business logic is important, because an implicit sum or average may not reflect the intended definition. Use slicers and cross-filtering to support exploration, but test their interaction carefully.

Useful Power BI practices include:

• create a proper date table for reliable time intelligence;
• format measures with correct currency, percentage and decimal conventions;
• use conditional formatting only with an explicit semantic purpose;
• use small multiples when comparing repeated category patterns;
• provide accessible alternative text and logical tab order;
• test bookmarks, drill-through and filter context before publication.

In both tools, verify the underlying data and aggregation before interpreting the visual. A polished chart based on duplicated rows or an incorrect join remains incorrect.

12. Detailed Applied Case: Regional Retail Performance

This is an illustrative case. A retail company operates in four regions and wants to decide where to focus a quarterly improvement programme. The dataset contains transaction date, region, product category, sales revenue, profit, order value and delivery time.

Background and problem

The executive team initially receives a dashboard containing a pie chart of sales by region, a three-dimensional column chart of monthly sales and a map coloured by profit. Managers report that the dashboard looks impressive but does not answer three questions reliably:

1. Which regions generate the highest sales and profit?
2. Are delivery times consistently poor or are a few extreme orders responsible?
3. Is higher sales associated with higher profit at the category level?

Analysis and approach

The analyst first defines the grain of each view. Regional performance is aggregated to region and quarter. Delivery-time distribution uses order-level observations. Category relationship analysis uses category-quarter observations. This prevents an order-level distribution from being confused with a regional summary.

For the first question, the analyst replaces the pie chart with a sorted horizontal bar chart for profit, accompanied by a compact sales label. Profit is selected as the primary measure because the decision concerns improvement impact. Sales remains available through a tooltip and a secondary table. The bar axis begins at zero, and currency units are stated.

For the second question, the analyst uses side-by-side box plots of delivery time by region, with sample size shown in the subtitle. The median and spread become visible, and unusually late orders can be investigated rather than mistaken for the typical experience. A histogram is provided in a drill-through view for operations staff who need the detailed distribution.

For the third question, the analyst constructs a scatter plot in which each mark is a product category for a selected quarter. The horizontal axis is sales, the vertical axis is profit, and colour indicates region only when the number of categories remains manageable. A reference line at zero profit distinguishes profitable from loss-making categories. A trend line is shown as a descriptive aid, not as evidence that sales causes profit.

Outcome

The revised dashboard shows that the region with the highest sales is not the region with the highest profit margin. Delivery times in one region have a similar median to other regions but a much wider upper tail, indicating a reliability problem rather than a general shift in typical performance. The scatter plot shows that two high-sales categories have low or negative profit, prompting investigation of discounting and fulfilment costs.

Learning points

• The chart should reflect the decision variable, not merely the most available measure.
• Aggregation level must be stated and controlled.
• Distribution charts prevent averages from hiding operational risk.
• A relationship chart supports investigation but does not establish causation.
• Multiple coordinated views are more useful than one overloaded chart.
• Definitions of sales, profit, margin and delivery time must be documented in the report metadata.

13. Accessibility, Ethics and Governance

Visualisation can mislead without containing a numerical error. Ethical practice requires the analyst to present relevant context, avoid selective omission and disclose transformations.

Accessibility includes more than colour choice. Do not rely on red and green alone to communicate status; use labels, symbols or patterns where appropriate. Maintain adequate contrast, readable text and logical reading order. Provide descriptive titles, alternative text and tables or downloadable data for users who cannot perceive the visual. Avoid dense dashboards that require precise hover interaction on small screens.

Governance requires a traceable definition of measures, data sources, refresh dates, filters and exclusions. A dashboard should identify whether a figure is a count of rows, count of distinct customers, sum of transactions or an estimate. Personal or sensitive information must be protected through access controls, aggregation and appropriate suppression. Small-cell displays may enable re-identification even when names are removed.

AI-assisted chart recommendation is increasingly available in Tableau and Power BI. Such tools can accelerate exploration, but their recommendations may reflect default assumptions, hidden aggregation or correlations that are not decision-relevant. Analysts remain responsible for validating the data, chart type, scale, narrative and accessibility of AI-generated outputs.

14. Best-Practice Checklist

Before publishing a chart, ask:

• What specific question does it answer?
• Is the chart type appropriate for the data and task?
• Are categories ordered meaningfully?
• Are units, time period and aggregation stated?
• Does the scale preserve honest visual comparison?
• Are missing values and outliers handled transparently?
• Is colour used consistently and accessibly?
• Can the intended audience read the chart quickly?
• Does interaction clarify the analysis rather than hide it?
• Can another analyst reproduce the measure and filters?
• Does the chart imply causation, certainty or precision beyond the evidence?

Chapter Summary

Effective visualisation begins with the analytical purpose. Comparison and ranking are usually served by aligned bars or dots; distribution by histograms, box plots or strip plots; composition by carefully designed stacked bars or other part-to-whole views; relationships by scatter plots and heat maps; and time change by chronological line or column charts. Each chart has perceptual strengths and limitations.

Visual encoding determines how easily values can be compared. Position and length are generally more precise than angle, area or colour intensity. Scale design, baseline choice, aggregation and labelling strongly affect interpretation. Truncated bar axes, unexplained dual axes, excessive decoration, inconsistent small-multiple scales and hidden denominators can produce misleading conclusions.

Tableau and Power BI provide powerful construction and interaction features, but they do not replace analytical judgement. Professional practice requires validating data grain, measure definitions, filter context and accessibility. Ethical visual storytelling communicates uncertainty and limitations, protects sensitive information and avoids implying causation where only association has been shown.

Key Terms

Visual encoding: Mapping data attributes to visual properties such as position, length, colour or size.

Analytical task: The purpose of a visual analysis, such as comparison, ranking or distribution assessment.

Dimension: A categorical or descriptive field used to group data.

Measure: A quantitative field that can be aggregated.

Aggregation: Summarising records using operations such as sum, average or count.

Granularity: The level of detail represented by a data record or visual view.

Bar chart: A chart using lengths from a common baseline to compare categories.

Histogram: A chart showing frequencies across contiguous numerical intervals.

Box plot: A distribution summary based on quartiles, median and possible outliers.

Composition: The parts that make up a total.

Scatter plot: A chart showing paired quantitative observations on two axes.

Heat map: A matrix in which colour encodes values across two dimensions.

Time series: Observations arranged in temporal order.

Baseline: The reference position from which a visual length begins.

Dual axis: A chart containing two independently scaled quantitative axes.

Outlier: An observation unusually distant from other observations under a stated criterion.

Small multiples: Repeated, consistently scaled charts used to compare groups.

Preattentive attribute: A visual property detected rapidly, such as contrast or position.

Dashboard: A coordinated set of visual views supporting monitoring or analysis.

Review Questions

5 Short Answer Questions

1. Why is a bar chart generally more effective than a pie chart for comparing similar category values?
2. What is the difference between a histogram and a bar chart?
3. When is a 100% stacked bar chart appropriate?
4. Why should dual-axis charts be used cautiously?
5. Name two accessibility practices for business dashboards.

5 Long Answer Questions

1. Explain the relationship between analytical purpose, data type and chart selection. Use at least three examples.
2. Discuss how aggregation and granularity can change the meaning of a visualisation.
3. Compare histograms, box plots and strip plots for analysing delivery-time data.
4. Explain how misleading scales and decorative design can distort business interpretation.
5. Describe a professional workflow for constructing and validating a chart in Tableau or Power BI.

3 Analytical / Application Questions

1. A manager wants to compare monthly sales for eight products over two years. Recommend a chart design and justify your choices concerning time, category overload, colour and interaction.
2. A dashboard shows that one region has the highest average order value, but its box plot reveals a long upper tail and a low median. Explain how you would communicate this finding and what additional analysis you would request.
3. Design a coordinated dashboard for deciding whether to expand a product category. Specify the charts, measures, aggregation levels, scales and governance information you would include.

Multiple Choice Questions with Answers

Question 1

Which chart is generally most suitable for ranking many categories by a quantitative measure?

A. Sorted horizontal bar chart

B. Pie chart with many slices

C. Unordered doughnut chart

D. Decorative 3-D area chart

Correct Answer: A. Sorted horizontal bar chart

Question 2

What is the main purpose of a histogram?

A. To show parts of a whole

B. To show the distribution of a quantitative variable

C. To show geographic locations

D. To display a hierarchy of categories

Correct Answer: B. To show the distribution of a quantitative variable

Question 3

A 100% stacked bar chart primarily compares:

A. Absolute totals only

B. Geographic distance

C. Percentage composition across categories

D. Individual observations over time

Correct Answer: C. Percentage composition across categories

Question 4

Which visual channel is generally most precise for comparing quantitative values?

A. Position on a common scale

B. Colour hue

C. Area of a bubble

D. Three-dimensional perspective

Correct Answer: A. Position on a common scale

Question 5

What does the interquartile range represent?

A. The maximum value minus the minimum value

B. The mean divided by the median

C. The third quartile minus the first quartile

D. The number of observations above the mean

Correct Answer: C. The third quartile minus the first quartile

Question 6

Why can a truncated bar-chart axis be misleading?

A. It always removes missing values

B. It can exaggerate apparent differences in bar lengths

C. It prevents categories from being sorted

D. It changes categorical data into temporal data

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

Question 7

Which chart is most appropriate for examining the association between advertising expenditure and revenue?

A. Scatter plot

B. Pie chart

C. 100% stacked bar chart

D. Histogram of product names

Correct Answer: A. Scatter plot

Question 8

An index with a base-period value of 100 is mainly used to compare:

A. Relative change from a common starting point

B. The exact number of transactions

C. The geographic location of observations

D. The number of categories in a dataset

Correct Answer: A. Relative change from a common starting point

Question 9

Which practice improves colour accessibility?

A. Using red and green as the only status signals

B. Using many unrelated saturated colours

C. Combining colour with labels, symbols or patterns

D. Removing all category labels

Correct Answer: C. Combining colour with labels, symbols or patterns

Question 10

What is a major limitation of a dual-axis chart?

A. It cannot display any numerical data

B. Independent scales can create an apparent relationship

C. It always requires a histogram

D. It prevents the use of chronological data

Correct Answer: B. Independent scales can create an apparent relationship