Data visualization: chart types, design and best practices

Data visualization represents information through elements such as position, length, shape and color to support exploration and communication. A useful visualization makes a comparison or pattern easier to understand while preserving the meaning, context and limitations of the data.

Choosing a chart means deciding what the audience needs to compare, which properties of the data matter and what context makes the result interpretable. The same dataset can support several representations, but those representations may answer different questions or create different impressions.

This guide explains chart selection, scales, color, accessibility, data storytelling and interaction. It also covers how visualizations can be tested and how tools and custom visuals should be evaluated.

What makes a data visualization useful?

A visualization turns variables into perceptible properties. The length of a bar can represent a quantity, the position of a point can relate two numerical variables and a color can distinguish a category or highlight an exception.

Usefulness depends on the task. Comparing two similar values requires clear references. Following a trend requires a coherent time axis. Locating an exception requires enough emphasis to distinguish it without concealing what is happening elsewhere.

The article on the value of visualization in business decision-making discusses this communication role. The business dashboards guide covers the wider management instrument, including KPI selection and implementation. This guide focuses on how information is represented and understood.

Exploration and explanation serve different needs

An exploratory view supports questions that may not yet have an answer. It can provide distributions, filters and access to detail. An explanatory view communicates a finding that has already been investigated, together with the evidence and limitations needed to interpret it.

Audience knowledge also matters. An analyst working with a dataset every day and a manager seeing a chart once in a presentation need different levels of explanation. A simple composition can be sufficient when it answers the relevant question accurately.

Visualization can support both investigation and communication throughout an analytical project. It is not restricted to a final presentation stage. The data analytics guide places these activities within a broader process of framing, analysis, validation and action.

How should a chart be chosen?

The starting point is the comparison the audience needs to make. Data type, aggregation, number of observations and viewing format then help determine a suitable representation. The following chart families provide starting points rather than universal rules.

A starting point for choosing a data visualization
QuestionInitial optionWhat to check
Which category has the higher value?Bar chartCommon baseline, ordering and units.
How does a value change over time?Line chartIntervals, incomplete periods and definition changes.
Are two numerical variables related?Scatter plotMeaning of each point and level of aggregation.
How do observations vary?Histogram or quantile-based viewBins, sample size and extreme values.
What is the exact value?TableLabels, precision and ease of retrieval.

Comparing categories and rankings

Bar charts support comparisons of quantities across categories. Horizontal bars can make long category names easier to read. Ordering by value can help identify a ranking, while a natural order may be more appropriate for stages in a process.

The Tableau documentation on bar charts illustrates category comparisons. A shared reference and clear labels help readers compare lengths. Perspective effects and decorative shapes can interfere with that comparison.

A chart showing revenue by region needs to specify whether the values represent total sales, net sales or another measure. Visual clarity does not resolve an ambiguous business definition.

Showing change over time

A line chart can show how a quantity changes across an ordered time axis. Dates and intervals need to be coherent. A missing observation is different from a recorded zero, and a change in coverage or calculation may interrupt comparability even when the line looks continuous.

With many series, small multiples can give each category its own view while retaining a shared structure. Common scales can support comparisons between panels. When independent scales are necessary, their difference needs to be clear because similar-looking slopes may represent very different changes.

Examining relationships and distributions

A scatter plot positions observations according to two numerical variables. The meaning of each point and the level of aggregation need to be explicit. The Tableau scatter plot documentation develops this representation. A visible relationship can motivate further investigation without demonstrating causation.

A distribution shows how values vary, rather than reducing the dataset to a single average. Histograms group observations into intervals; changing those intervals can change the visible pattern. Quantiles, ranges and individual observations can provide additional context where appropriate.

For example, an average handling time can conceal a small group of very slow cases. A distribution can make that variation visible, provided the observation period and population are clearly defined.

Showing composition and location

A part-to-whole chart requires a meaningful total and compatible components. Missing categories, overlapping groups or a change in the denominator can make the displayed shares misleading. Aligned bars can make small differences easier to compare than segments that lack a common starting position.

A map is useful when location or a spatial pattern is part of the question. Comparing regions may require a rate rather than an absolute count. A large geographic area should not appear more important simply because it occupies more space on the map.

When is a table the better choice?

A table can be more suitable when the task is to retrieve precise values or reconcile figures. A chart can be more suitable for recognising a trend or comparison. Both can appear together when they serve different tasks without creating unnecessary duplication.

The article on Power BI visuals and the visual vocabulary provides further chart-selection examples. A visual vocabulary supports design choices; it is not a requirement to use every chart family or a guarantee that one representation works for every audience.

Scales, context and uncertainty

A chart needs enough context to identify the measure, unit, population, period and comparison. Source and update information matter where they affect interpretation. Observed results, targets and estimates should be distinguishable.

Why do axis choices matter?

In a bar chart, length represents magnitude. Starting the numerical axis above zero can distort the apparent proportions between values. A line chart may use a narrower range to show variation, provided the scale is clear and the presentation does not exaggerate the change. The Government Analysis Function guidance on charts explains these scale considerations.

Rounding also follows the task. A management overview may use values in millions, while reconciliation requires more precision. Percentage change and percentage-point change need different labels because they express different comparisons.

Making evidence limitations visible

A forecast needs to be distinguishable from historical observations. An interval or shaded area needs an explanation of what it represents and the assumptions behind it. A prediction interval, a confidence interval and a range of scenarios have different meanings.

Provisional figures, incomplete periods, excluded observations and small groups need appropriate disclosure. A continuous line should not imply that a comparison is valid when the underlying definition or coverage has changed.

The data quality guide and the article on data quality use cases address checks on the information behind a visualization. A readable chart can still contain an incorrect total if preparation, relationships or aggregation are wrong.

Color and accessibility in data visualization

Color should have a recognisable purpose and should not be the only way to communicate an important distinction. Labels, position, patterns or symbols can provide additional cues. Equivalent concepts need consistent treatment across views.

W3C's guidance on the use of color explains this principle. A small color palette can reduce visual noise, but reducing the number of colors is not sufficient to make a chart accessible.

Matching the palette to the information

Categories need distinctions that do not imply a nonexistent order. Ordered magnitudes can use a progression that makes that order perceptible. A scale with two directions needs a meaningful reference point and an explanation of each side.

Emphasis is most useful when it has a limited, clear role. If every mark is highlighted differently, the reader has more signals to decode before answering the question. Brand styling should preserve legibility and avoid confusing a management status with a decorative choice.

The 10 tips for using color in data visualization present alternatives such as direct labels, selective emphasis and small multiples. Different shades can suggest order, so their use for unrelated categories still needs care.

Text, graphical contrast and alternatives

Legibility needs to be assessed at the size and on the device actually used. W3C's non-text contrast guidance addresses meaningful graphical objects and interface cues. Its application depends on the elements needed to understand or operate the specific visualization.

For a complex chart image, a short identification alone may not communicate its essential information. The W3C tutorial on complex images explains the combination of a short description and a more detailed text alternative. An accessible data table can complement that explanation where useful; a list of values may not convey the pattern on its own.

The Bismart guide to color in data visualization provides practical material for reviewing emphasis and presentation. Essential definitions should also remain available without depending exclusively on a mouse hover.

What is data storytelling?

Data storytelling combines evidence, visual representation and explanation to communicate a finding to an audience. It can organise a presentation or a sequence of views while keeping observations, interpretations and recommendations distinguishable.

A useful sequence establishes the question, presents the finding and shows the comparisons supporting it. It then identifies what remains uncertain and what action could be evaluated. A narrative should not turn an untested explanation into a fact simply to provide a more satisfying ending.

The article on data storytelling in business dashboards develops narrative structure and context. Exploratory environments may support several competing hypotheses rather than one predetermined conclusion.

Titles and annotations

A finding-led title can highlight a supported result. A descriptive title may be more appropriate when the audience will explore different selections. Titles containing figures or claims need to remain consistent with active filters.

An annotation can locate a campaign, an incident or a definition change in time. The coincidence of an event and a movement in a chart does not establish that the event caused the movement. An explanatory note needs the same evidential discipline as the chart itself.

In financial communication, scenarios and periods need explicit definitions before a variance is interpreted. The historical article on the CFO's role and financial communication in 2022 provides business context from that period. Visual presentation supports the discussion; the figures and accounting definitions still require appropriate review.

Interaction and different viewing formats

Interaction adds value when it helps answer a question. A filter changes the population, a selection may highlight or filter related views, and navigation may reveal more detail. The user needs to recognise what changed and how to return to a previous state.

Periods and active selections should remain understandable throughout the exploration. If two views respond to different filters, the distinction needs to be visible. Totals and percentages also need a clear interpretation after the selection changes.

From an interactive report to a static document

A screenshot or export loses some of the original experience. Supporting definitions, filter states and source information need to remain available in the format that will actually circulate. A label available only in a tooltip cannot explain a printed page.

On a small screen, reorganising views around priority tasks can be more useful than shrinking the complete layout until it fits. Keyboard navigation and the intended assistive technologies also need testing. The dashboard design and usability tips provide related functional guidance.

The historical example of visual and interactive Reporting Services reports illustrates the relationship between audience and delivery format. Its specific implementation details belong to its publication context.

Data visualization tools and custom visuals

Tool choice depends on the work to be produced and maintained: interactive reports, reproducible analytical charts, web experiences or static documents. Data connections, editing skills, accessibility, distribution, collaboration and maintenance can matter more than the total number of chart types.

The comparison of data visualization tools provides a starting point. The older overview of visualization in a business intelligence environment provides historical context. Product names, capabilities and commercial conditions need verification against current vendor documentation before selection.

What needs checking in a custom visual?

  • Task fit: the representation answers a question the audience can understand.
  • Calculation: aggregations and displayed values correspond to the intended data model.
  • Behavior: filters, missing values, new categories, exports and planned screen sizes work as expected.
  • Access: labels, alternatives and navigation support the intended users.
  • Operation: licensing, maintenance, updates and dependencies are suitable for the project.

Microsoft's documentation on certified Power BI visuals describes code requirements and testing. Certification does not establish that a chart is the right choice for a message or that the calculations in a report are correct.

Power BI, Zebra BI and product history

The introduction to data visualization with Power BI provides a platform-specific route. The Power BI guide covers the wider product environment. This guide remains focused on visual communication across tools.

The article on data visualization with Zebra BI introduces its reporting applications. The Zebra BI 5.4 update overview and the Zebra BI Office launch for Excel and PowerPoint are historical product references. They do not replace current vendor documentation or demonstrate that a visual prevents all interpretation errors.

Charticulator and other implementation contexts

The article on Charticulator and custom chart creation without code describes a Microsoft Research project. The original Microsoft Charticulator repository was archived on April 26, 2024 and is read-only. Any implementation under consideration needs a separate maintenance review; the article is a historical reference rather than a current product recommendation.

The introduction to Elasticsearch and visualization in its ecosystem describes another implementation context. Search, storage and visual presentation perform different functions. Their integration does not remove the need to check what is calculated and how it is represented.

How can a visualization be tested?

Testing covers both fidelity to the data and audience comprehension. Displayed values and aggregations can be checked against known reference cases. Representative users can then perform realistic tasks without being told the intended answer in advance.

Useful observations include whether they identify the period, make the intended comparison and recognise the limits of the evidence. Incorrect interpretations provide a basis for revising the design and testing whether the change resolves the problem.

Technical checks should include empty selections, extreme values, long labels and new categories. A chart that works with a small sample can become unreadable after an update. Review is needed when the meaning of the measure, the audience or the delivery format changes.

The downloadable 21 Best Practices for Power BI Reporting provides additional platform-specific material. Design guidance complements calculation checks and user testing.

Worked example: a change from 90% to 92%

Illustrative scenario: a report compares an on-time delivery rate of 90% with a later rate of 92%. The increase is two percentage points. Relative to the original value, the increase is approximately 2.2%. Those statements describe different comparisons and need different labels.

Bars truncated between 89% and 93% would make the difference look disproportionate to the represented rates. A suitable reference and explicit labels support interpretation. A line chart could show the progression across several periods with its scale clearly indicated.

Before interpreting the difference as improved service, the reporting definitions and populations need to be comparable. An incomplete second period needs a visible qualification. Changes in order volume or mix may also deserve investigation.

A title can describe the observed increase without attributing it automatically to a new policy. Evaluating the policy's effect requires additional analysis. The figures illustrate a design decision and are not Bismart client results.

Common data visualization mistakes

  • Novelty before task: an unfamiliar chart adds complexity without improving the relevant comparison.
  • Missing context: the period, unit, population or reference is unclear.
  • Distorted magnitude: the scale, aggregation or graphical proportions create a misleading impression.
  • Missing values shown as zero: absence of an observation becomes an invented result.
  • Color-only or hover-only meaning: essential information is unavailable through other cues or access methods.
  • Narrative presented as certainty: a hypothesis is described as an established explanation.
  • Testing only the initial view: filtering, updates and exports change the meaning or readability unnoticed.

Frequently asked questions about data visualization

What is data visualization?

Data visualization represents information using visual elements such as bars, lines, positions, shapes and colors. It supports exploration and communication of comparisons, patterns and exceptions while preserving the meaning and limitations of the underlying data.

What is the best chart for business data?

The answer depends on the task, data and audience. Category comparisons, distributions and time trends require different design choices. A suitable chart allows the intended question to be answered accurately, with enough context to avoid a misleading interpretation.

Does every visualization need to be interactive?

No. Interaction is useful when it supports exploration or retrieval. A well-contextualized static chart can be more suitable for a document, presentation or a single message.

Does using fewer colors make a report accessible?

It can reduce visual noise, but it is insufficient on its own. Contrast, labels, text alternatives, navigation and access to essential information also need review.

What is the difference between data visualization and data storytelling?

Data visualization represents information visually. Data storytelling combines representation, evidence and explanation to communicate a finding. A visualization can support an exploratory task without presenting one narrative conclusion.

Can a chart prove why a result changed?

A chart can show patterns, comparisons and associations. A causal explanation needs an appropriate analytical basis. A title, annotation or coincidence between two curves does not establish that one variable caused the other.

Resources for improving data visualization

For an implementation project, Bismart's self-service BI consulting for usable, governed business reporting connects visual communication with shared measures, information access and adoption by business teams.

Review the use of color in business reporting

The guide explores presentation choices and ways to emphasize information without overloading visualizations with competing colors.