Data visualization is the practice of encoding data values as visual properties, such as position, length, angle, area and color, so that people can perceive patterns, comparisons, trends and outliers faster and more reliably than from tables of numbers. Often shortened to dataviz, it covers static charts, interactive dashboards, maps and specialized diagrams.
Within data work, visualization serves two different purposes: exploration, where an analyst looks for structure in unfamiliar data, and explanation, where a finished chart communicates a specific finding to an audience. The same dataset usually needs different charts for each purpose.
How data visualization encodes information
Every chart maps data variables to visual channels. A bar chart maps a category to position on one axis and a quantity to bar length; a scatter plot maps two quantities to horizontal and vertical position; a heatmap maps a quantity to color intensity. The designer chooses marks (points, lines, bars, areas) and channels (position, size, hue, lightness) for each variable.
Perception research shows that channels are not equally accurate. People judge position along a common scale most precisely, then length, then angle and area, while color saturation and volume are judged poorly. This ranking explains why bar charts usually outperform pie charts for precise comparison, and why 3D effects degrade readability.
The table below ranks the most common visual channels for quantitative data.
| Visual channel | Relative accuracy | Typical use |
|---|---|---|
| Position on a common scale | Highest | Bar, dot, line and scatter charts |
| Position on unaligned scales | High | Small multiples with separate axes |
| Length | High | Bars, bullet charts |
| Angle and slope | Medium | Pie charts, slope graphs |
| Area | Low to medium | Bubble charts, treemaps |
| Color lightness or saturation | Low | Heatmaps, choropleth maps |
| Volume | Lowest | 3D shapes, rarely justified |
Choosing a chart type for the question
Chart choice starts from the analytical question, not from the available chart gallery. Comparison, change over time, distribution, relationship, composition and flow each have a small set of chart types that answer them well.
| Question | Suitable charts | Avoid |
|---|---|---|
| Comparison across categories | Bar chart, dot plot | Pie chart with many slices |
| Change over time | Line chart, area chart, sparkline | Unordered bars for long series |
| Distribution of values | Histogram, box plot, density plot | Averages alone |
| Relationship between variables | Scatter plot, bubble chart | Dual-axis charts with arbitrary scales |
| Part-to-whole composition | Stacked bar, treemap, pie with two to five parts | Stacked areas with many layers |
| Flow between stages | Sankey diagram, funnel chart | Network diagrams for simple flows |
| Geographic variation | Choropleth map, symbol map | Raw counts on choropleths |
A frequent mistake is the choropleth of raw counts: populous regions dominate simply because they are large. Maps shaded by color should show normalized values such as rates per 100,000 people, while absolute counts suit proportional symbols.
Design principles that make charts readable
Effective visualization removes everything that does not carry information and emphasizes what does. The principle often described as maximizing the data-ink ratio means dropping heavy gridlines, borders, shadows and decorative backgrounds, and labeling data directly instead of relying on distant legends.
| Parameter | Recommended practice | Why it matters |
|---|---|---|
| Bar chart baseline | Start the value axis at zero | Bar length must be proportional to value |
| Line chart axis | May start above zero if labeled | Highlights change within a narrow range |
| Categorical colors | Roughly six to eight distinct hues at most | More hues cannot be told apart reliably |
| Sequential colors | Single hue varying in lightness | Preserves order from low to high |
| Diverging colors | Two hues around a meaningful midpoint | Shows deviation from a target or zero |
| Text contrast | At least 4.5:1 for normal text | Meets common accessibility guidelines |
| Sorting | Order bars by value unless order is inherent | Makes ranking visible at a glance |
Color accessibility deserves particular attention, because a meaningful share of people have some form of color vision deficiency, most often affecting red and green. Palettes should differ in lightness as well as hue, and critical distinctions should also be carried by labels, shapes or position.
Exploratory, explanatory and interactive visualization
The purpose of a chart determines how polished and how focused it must be. Exploratory charts are fast and numerous; explanatory charts are few, carefully annotated and built around one message; interactive dashboards let users filter and drill down within a fixed design.
| Mode | Audience | Characteristics |
|---|---|---|
| Exploratory | The analyst | Many quick charts, default styling, focus on discovery |
| Explanatory | Decision makers or the public | One message per chart, annotations, deliberate emphasis |
| Interactive dashboard | Recurring business users | Filters, drill-down, consistent layout, frequent refresh |
| Data storytelling | Broad audiences | Sequenced charts with narrative and context |
Interactivity is valuable when users genuinely ask different questions of the same data, but it shifts the burden of interpretation to the reader. For a single important finding, a well-annotated static chart is usually more effective than a dashboard full of controls.
Common pitfalls and misleading charts
Most misleading charts are not deliberate; they result from default settings and unexamined choices. Reviewers should check each chart against a short list of known failure modes before publication.
- Truncated bar axes that exaggerate small differences between categories.
- Dual y-axes whose independent scales can suggest any correlation the designer wants.
- Overplotting in scatter plots with many points, solved with transparency, binning or sampling.
- Rainbow color scales that create false boundaries and are hard to read in grayscale.
- Missing uncertainty, showing point estimates without intervals when samples are small.
- Too many series on one line chart, better split into small multiples.
Understanding it simply
Data visualization means turning numbers into pictures so the eye can do part of the thinking. A column of 50 monthly sales figures is hard to read. Drawn as a line, the same figures instantly show whether sales are rising or falling.
A good chart works like a well-drawn map. It leaves out unnecessary detail and makes the important route obvious. The best chart answers the reader’s question at a glance. It never uses tricks that make differences look bigger or smaller than they are.
Applied in practice
A hospital operations team tracks emergency department waiting times. Its old monthly report was a table of averages per weekday, and the averages looked acceptable. An analyst replaced it with a heatmap of median and 90th percentile waits by weekday and hour, using a sequential single-hue palette.
The heatmap immediately showed a dark band on Monday and Tuesday evenings between 18:00 and 22:00, where the 90th percentile wait was roughly twice the daily figure. The averages had hidden it because quiet night hours pulled them down. Management moved two clinician shifts to those evening hours, and the next quarter’s heatmap showed the band largely gone.