Data Visualization

Data visualization is the practice of encoding data as visual marks—points, lines, bars, areas, colors, shapes—so that people can perceive patterns, compare quantities, and reason about evidence that would otherwise sit inert in a table.

11 frameworks, from 1786 to today · Data Science · Computer Science

still a working view ended · each leads to its entry below

Overview

Data visualization is the practice of encoding data as visual marks—points, lines, bars, areas, colors, shapes—so that people can perceive patterns, compare quantities, and reason about evidence that would otherwise sit inert in a table1. What is at stake is perception itself: the human visual system is extraordinarily good at judging relative length, position, and alignment, and surprisingly bad at mentally summing columns of numbers1. Every major advance in the field has been, at bottom, an argument about which visual encodings let people see what the data actually says—and, increasingly, about how to build systems that produce those encodings reliably, at scale, and interactively.

Maps, Measurement, and the Birth of Thematic Cartography

The oldest strand is cartographic. Maps showing where things are are ancient, but maps showing how much of something is where—thematic maps—are a modern invention tied to the rise of state statistics. In the 1820s and 1830s, French engineers and physicians began encoding rates and densities directly onto geography. Charles Dupin's shaded maps of illiteracy in France (1826) used graded shading to show a social variable across districts, and Louis-René Villermé used similar maps to argue that mortality in Paris tracked wealth rather than topography—an early case of a visualization doing analytical work, not just illustration.

The decisive technical problem was how to show a quantity that varies continuously across space. In 1837 the Irish engineer Henry Drury Harness drew a map of Ireland with graduated circles and flow lines for passenger traffic, and in the 1840s the physician John Snow's cholera map of Soho (1854) plotted deaths as marks clustered around the Broad Street pump—a case where the visual arrangement itself made a causal argument legible. The choropleth map (shading whole regions by a rate) and the isarithmic map (contours of equal value, like a topographic map of a statistical surface) were both formalized in the nineteenth century. The French cartographer Charles Joseph Minard's 1869 flow map of Napoleon's Russian campaign—combining a shrinking band for army size with a temperature line—became the canonical demonstration that a single graphic could carry a multivariate narrative123.

The lesson cartographers drew, and taught through the twentieth century, was that thematic mapping requires deliberate choices: classification schemes (how to bin continuous values into shades), normalization (rates rather than raw counts, so large regions don't dominate), and projection. These became the working doctrine of professional cartography.

Statistical Graphics and the Grammar of Charts

In parallel, statisticians developed a repertoire of abstract charts divorced from geography. William Playfair is usually credited with the first published line graph, bar chart, and pie chart in his Commercial and Political Atlas (1786) and Statistical Breviary (1801), designed to make economic time series comparable at a glance. The nineteenth century added the histogram, the scatterplot, and the box-like summaries that would later be formalized.

The twentieth century's contribution was methodological discipline. In the 1960s and 1970s John Tukey argued that graphics were not decoration appended to a finished analysis but a tool of discovery—that you should plot data before modeling it, precisely because plots reveal structure, outliers, and failures of assumption that summary statistics hide. This became exploratory data analysis (EDA): an attitude in which visualization is a first-class investigative method4. Tukey's Exploratory Data Analysis (1977) and his earlier work on stem-and-leaf displays and box plots gave analysts concrete instruments for looking at distributions.

A competing, more prescriptive tradition asked what makes a graphic good. Edward Tufte's The Visual Display of Quantitative Information (1983) argued for data-ink minimalism: maximize the proportion of ink devoted to data, strip away chart junk, and avoid distorting scales15. Tufte's "lie factor" and his critiques of deceptive axis choices gave the field a vocabulary for graphical integrity. His work was influential but also contested—later empirical work suggested that some "chart junk" and decoration can aid memory and engagement, and that minimalism is not universally optimal.

Semiology and the Systematic Study of Encodings

A more theoretical strand asked: what are the elements of a graphic, and how do they combine? Jacques Bertin's Sémiologie graphique (1967) proposed that visual variables—position, size, shape, value (lightness), texture, color hue, orientation—each have different capacities for encoding quantitative versus categorical information. Position, Bertin argued, is the most precise channel; color hue is good for categories but poor for magnitude. This graphical semiology gave designers a principled basis for choosing encodings rather than imitating convention.

Bertin's framework was later extended and tested empirically. In the 1980s William Cleveland and Robert McGill ran controlled experiments showing that people judge some encodings more accurately than others—position along a common scale beats length, which beats angle, which beats area and color—a ranking that became known as the "elementary perceptual tasks." This was part of a broader turn toward empirical visualization research: treating chart design as a hypothesis to be tested with human subjects rather than settled by taste. Researchers measured how quickly and accurately people read different encodings, how color palettes affect perception (including work on colorblind-safe and perceptually uniform palettes), and how animation and interaction affect comprehension.

The Grammar of Graphics and Computational Construction

By the 1990s, the proliferation of chart types and the need to generate them programmatically created a new problem: how to describe a graphic in a way a computer could execute, without hard-coding each chart type. Leland Wilkinson's The Grammar of Graphics (1999) proposed that any statistical graphic can be decomposed into a data set, aesthetic mappings (data variables to visual properties), geometric objects (points, lines, bars), statistical transformations (binning, smoothing), scales, and coordinate systems. A chart is then a composition of these components.

This idea was realized in software. The R package ggplot2 (Hadley Wickham, mid-2000s) implemented a layered grammar in which users add geoms, mappings, and facets incrementally; it became one of the most widely used visualization tools in statistics. Similar grammar-inspired systems appeared in other languages. The payoff was combinatorial: instead of a fixed menu of chart types, a small set of primitives could generate a vast space of graphics, and the same description could be rendered to screen, print, or web.

Interaction, the Web, and Business Dashboards

The web changed the medium. Early static charts gave way to interactive graphics where users could hover, filter, zoom, and brush—selecting a range of points in one view to highlight them in another. Ben Shneiderman's 1996 "eyes have it" formulation—overview first, zoom and filter, details on demand—became a design mantra for interactive information visualization. Toolkits such as D3.js (2011) let developers bind data to web documents and produce custom interactive graphics, and libraries like Vega and later Observable Plot carried the grammar-of-graphics idea into the browser.

In organizations, a parallel commercial tradition produced business intelligence dashboards: consolidated views of key metrics, refreshed from data warehouses, aimed at managers rather than analysts. Tools from the spreadsheet era through dedicated BI platforms made charts a routine artifact of reporting. Dashboards raised their own design questions—how many metrics, how much context, how to avoid misleading aggregation—and they made visualization a mass practice rather than a specialist one.

Visual Analytics: Visualization as Reasoning

Around the mid-2000s, a research community formed around visual analytics, defined roughly as the science of analytical reasoning facilitated by interactive visual interfaces. The motivating observation was that many analytical tasks—intelligence analysis, fraud detection, scientific discovery—involve large, heterogeneous data and open-ended questions where neither fully automated mining nor pure manual inspection suffices. The answer was tight coupling: computational methods propose patterns, the human sees and steers, and the loop repeats. Visual analytics contributed techniques for coordinated multiple views, provenance tracking (recording what the analyst did), and scalable rendering, and it drew explicitly on cognitive science and human-computer interaction.

Automated and Generative Visualization

The most recent turn asks how much of the design work can be done by the machine. Automated visualization systems recommend chart types and encodings based on the data's characteristics and the user's task—for example, choosing a bar chart for comparing categories and a scatterplot for two quantitative variables—drawing on the empirical rankings from earlier decades. Recommendation and "chart chooser" tools encode these rules. More recently, generative approaches use machine learning to produce charts, summaries, and natural-language descriptions of data, and large language models can draft visualizations from a prompt. These systems raise unresolved questions: whether automated choices match expert judgment, how to detect misleading outputs, and how to keep the human in the loop when the tool is opaque.

Where the Field Stands

Today data visualization is both a craft and a research discipline1. Practitioners work in a layered stack: data preparation, a grammar-based or library-based construction layer, a rendering layer (often the browser), and increasingly an interaction and automation layer. The historical strands remain live and sometimes in tension—cartographic conventions govern maps, statistical graphics govern exploratory plots, semiotic and perceptual research inform encoding choices, grammar-based systems provide the construction vocabulary, and visual analytics supplies the model of human-machine reasoning1. Open questions include how to evaluate visualizations rigorously outside controlled lab tasks, how to make automated and generative tools trustworthy and transparent, how to design for accessibility and diverse audiences, and how to scale interactive graphics to datasets and displays far beyond what the field's founders imagined.

References

  1. Data and information visualization (introduction, Overview, Visual perception and data visualization, Human perception/cognition and data visualization, Characteristics of effective graphical displays, Other perspectives), Wikipedia
  2. Scientific visualization (History), Wikipedia
  3. Flow map (History), Wikipedia
  4. Exploratory data analysis, Wikipedia
  5. Information design (History), Wikipedia

The frameworks of Data Visualization

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1786-Present

    Statistical Graphics

    How do we draw data so patterns become visible?

    Fly to its tower in Nousfarer

  2. 1826-Present

    Thematic map

    How do we show how much is where?

    Fly to its tower in Nousfarer

  3. 1960-Present

    Exploratory Data Analysis

    How can plots reveal structure before modeling?

    Fly to its tower in Nousfarer

  4. 1967-Present

    Graphical Semiology

    What are the elements of a graphic and how do they combine?

    Fly to its tower in Nousfarer

  5. 1983-Present

    Data-Ink Minimalism

    How much ink should a chart spend on data?

    Fly to its tower in Nousfarer

  6. 1984-Present

    Empirical Visualization Research

    Which encodings do people read most accurately?

    Fly to its tower in Nousfarer

  7. 1990-Present

    Business Intelligence Dashboards

    How do managers see key metrics at a glance?

    Fly to its tower in Nousfarer

  8. 1996-Present

    Interactive Information Visualization

    How can interaction help people explore data?

    Fly to its tower in Nousfarer

  9. 1999-Present

    Wilkinson's Grammar of Graphics

    How can a chart be described as composable components?

    Fly to its tower in Nousfarer

  10. 2005-Present

    Automated Visualization

    Can a machine choose the right chart for data?

    Fly to its tower in Nousfarer

  11. 2005-Present

    Visual Analytics

    How can people and computation reason together over data?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Statistical GraphicsThematic mapExploratory Data AnalysisGraphical SemiologyData-Ink MinimalismEmpirical Visualization ResearchBusiness Intelligence DashboardsInteractive Information VisualizationWilkinson's Grammar of GraphicsAutomated VisualizationVisual Analytics

Beyond the field

Where Data Visualization's ideas came from in other fields, and what they led to there: the strongest 12 of 41 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Data Visualization is a planet in Nousfarer

Its city has a tower for each of these 11 frameworks. Land, walk in, and climb one concept at a time.

Land on Data Visualization

Related fields

Fields its ideas are linked to

More in Data Science

See where Data Visualization sits on the Map of Human Knowledge