Unit 4: Data Plotting and Visualization

I. Foundations and Governing Principles

Data visualization is the systematic representation of data through graphical marks, spatial arrangements, and visual properties. Its governing principle is that a graphic should make patterns, comparisons, relationships, and uncertainty easier to perceive without distorting the underlying evidence.

• Purpose: Visualization converts numerical, categorical, spatial, or temporal data into forms that support exploration, explanation, and decision-making.
• Data-to-visual pipeline:
  • Acquire: Obtain data from surveys, databases, sensors, files, or APIs.
  • Prepare: Clean missing values, correct data types, and calculate required measures.
  • Encode: Map variables to position, length, color, size, shape, or other visual channels.
  • Render: Produce the chart using software or a programming library.
  • Interpret: Examine patterns while considering scale, context, and uncertainty.
• Core conventions: Titles identify purpose; axes state variables and units; legends explain encodings; annotations identify important values; sources establish provenance.
• Accuracy principle: Visual proportions must correspond to data proportions. For example, a bar representing 40 units should be twice as long as one representing 20 units.
• Audience principle: Chart complexity, terminology, interaction, and detail should match the viewers’ knowledge and intended task.
• Ethical principle: Designers should not conceal observations, manipulate scales, misuse color, or imply unsupported causal relationships.
• Accessibility principle: Meaning should not depend on color alone; readable labels, adequate contrast, and alternative descriptions should be provided.

II. Introduction to Data Visualization — Turning Data into Visible Evidence

A. Introduction to data visualization

Data visualization uses graphical representations to communicate the structure and meaning of data more efficiently than raw tables alone.

• Exploratory visualization: Analysts create charts while investigating data, such as a histogram used to discover skewness or unusual observations.
• Explanatory visualization: A refined figure communicates a selected finding, such as an annotated line chart showing when sales began to decline.
• Primary tasks:
  • Comparison: Determine which category has the highest value.
  • Trend detection: Observe how a measurement changes over time.
  • Distribution analysis: Examine center, spread, skewness, and outliers.
  • Relationship analysis: Identify association between variables.
  • Composition analysis: Show how parts contribute to a whole.
• Context: A value of 75 is interpretable only when its variable, unit, time period, population, and comparison point are known.
• Integrity: Three-dimensional decoration, truncated axes, and inconsistent intervals can exaggerate or obscure differences.

B. Applications and limitations

Visualization strengthens analysis, but a chart remains a model rather than the data itself.

• Applications: Dashboards monitor operations, maps reveal geographic patterns, and scientific figures communicate experimental results.
• Pattern advantage: A scatter plot can reveal clusters that are difficult to detect in hundreds of table rows.
• Aggregation limitation: A monthly average can hide daily volatility or differences among subgroups.
• Causality limitation: A visible correlation does not establish that one variable causes another.
• Data-quality limitation: Missing, biased, duplicated, or incorrectly measured data produce misleading graphics even when chart construction is technically correct.

III. Visual Encoding — Mapping Variables to Perceptual Channels

A. Visual encoding

Visual encoding assigns data fields to marks and visual channels so that values become perceptible graphical differences.

• Marks: Geometric objects represent observations or groups:
  • Points: Individual cases in a scatter plot.
  • Lines: Connections or ordered sequences in a time series.
  • Areas: Magnitudes in bars, maps, or filled regions.
• Channels: Properties of marks carry information, including position, length, angle, area, color, shape, and orientation.
• Effectiveness hierarchy: Position on a common scale usually enables more accurate comparison than length, angle, area, or color intensity.
• Variable matching:
  • Quantitative: Encode values such as income in position or length.
  • Ordinal: Encode ordered classes such as low, medium, and high with ordered lightness.
  • Nominal: Encode unordered groups such as departments with distinct hues or shapes.
• Redundant encoding: Representing a category by both color and shape improves accessibility and recognition.
• Scale mapping: For a linear position scale,

p = p_min + ((x - x_min) / (x_max - x_min)) × (p_max - p_min)

Here, x is a data value, x_min and x_max are the data bounds, and p, p_min, and p_max are the corresponding display positions.

B. Encoding accuracy and limitations

An encoding should maximize perceptual accuracy while preventing visual channels from implying the wrong relationship.

• Zero baseline: Bar length represents magnitude, so bar charts generally require a zero baseline; otherwise, small differences appear disproportionately large.
• Color choice: Sequential palettes suit increasing values, diverging palettes emphasize deviation from a midpoint, and qualitative palettes distinguish categories.
• Area limitation: Doubling a circle’s radius quadruples its area, so size encodings must scale area—not radius—proportionally to value.
• Overplotting: Thousands of overlapping points can hide density; transparency, sampling, jittering, binning, or aggregation can reduce the problem.
• Dual-axis risk: Two vertical scales can be adjusted to manufacture apparent agreement between unrelated series.

IV. Data Visualization Software — Platforms for Creating and Sharing Graphics

A. Data visualization software

Data visualization software provides interfaces for importing, transforming, plotting, and distributing data-driven graphics.

• Spreadsheet software: Microsoft Excel and LibreOffice Calc support tables, pivot summaries, conditional formatting, and standard charts.
• Business-intelligence software: Tableau and Microsoft Power BI connect multiple sources and create interactive dashboards with filters and drill-down operations.
• Statistical software: R environments, SAS, SPSS, and Stata combine statistical analysis with publication-oriented graphics.
• Notebook environments: Jupyter Notebook integrates executable code, output, equations, and narrative in one reproducible document.
• Selection criteria: Choice depends on data volume, chart requirements, collaboration, automation, cost, governance, and user expertise.
• Deployment: A static PNG suits a report, an SVG supports resolution-independent publishing, and an interactive web dashboard supports user-selected views.

B. Applications and limitations

Software should be evaluated by whether it supports a reliable workflow rather than merely producing attractive output.

• Advantages: Graphical interfaces enable rapid chart construction, while templates help maintain organizational consistency.
• Reproducibility: Scripted or recorded transformations are preferable when charts must be updated repeatedly.
• Vendor limitation: Proprietary formats, license costs, and platform-specific services can restrict portability.
• Automation limitation: Recommended-chart features cannot determine whether a visual matches the analytical question.
• Governance: Published dashboards require controlled data access, refresh schedules, versioning, and documented metric definitions.

V. Data Visualization Libraries — Programmable and Reproducible Graphics

A. Data visualization libraries

Visualization libraries are reusable code packages that translate data and plotting instructions into static or interactive graphics.

• Python ecosystem: Matplotlib provides low-level control; Seaborn supplies statistical defaults; Plotly and Bokeh support interaction; Altair uses a declarative grammar.
• R ecosystem: ggplot2 builds graphics through layers based on data, mappings, geometric marks, scales, and themes.
• JavaScript ecosystem: D3.js binds data to web-document elements, while Vega-Lite specifies charts through a declarative JSON grammar.
• Reproducibility: Code preserves transformations, filters, labels, and style settings, allowing the same figure to be regenerated with new data.
• Example:

import seaborn as sns
sns.scatterplot(data=df, x="height_cm", y="weight_kg", hue="group")

Here, df is a tabular dataset; height_cm and weight_kg are quantitative columns; group is a categorical column encoded by color.

B. Applications and limitations

Libraries are especially valuable when customization, automation, or integration with analysis is required.

• Applications: A script can generate hundreds of consistent regional charts or rebuild a report after each data refresh.
• Control: Programmers can define scales, annotations, facets, themes, and export formats precisely.
• Learning cost: Effective use requires knowledge of programming, data structures, and library conventions.
• Dependency risk: Version changes may alter output or break code, so environments and package versions should be recorded.
• Performance: Browser-based charts may become slow when they attempt to render millions of individual marks.

VI. Basic Data Visualization Tools — Accessible Methods for Routine Analysis

A. Basic data visualization tools

Basic tools provide straightforward chart creation for small or moderately sized datasets with limited programming.

• Tables: Sorting, filtering, and conditional formatting reveal exact values and simple ranking patterns.
• Spreadsheet charts: Bar, line, pie, scatter, and histogram tools support common comparisons without code.
• Pivot charts: Aggregation by category and time allows users to summarize measures such as total sales by region.
• Chart builders: Drag-and-drop interfaces map fields to axes, color, labels, and filters.
• Workflow: Select a clean rectangular data range, choose a chart suited to the analytical task, label variables and units, and remove unnecessary decoration.
• Best use: Basic tools are suitable for classroom analysis, administrative reporting, prototypes, and one-time presentations.

B. Applications and limitations

Basic tools prioritize speed and accessibility but provide less control over complex analysis and deployment.

• Strengths: Immediate visual feedback makes them effective for users without programming experience.
• Transparency: Visible formulas and pivot settings can make small analyses easy to inspect.
• Scale limitation: Large datasets, frequent updates, and many linked sources can make spreadsheet workflows fragile.
• Error risk: Manual range selection, copied formulas, and hidden cells can introduce unnoticed mistakes.
• Design limitation: Default colors, legends, and axis settings often require deliberate correction before publication.

VII. Advanced Data Visualization Tools — Interactive, Multidimensional, and Specialized Systems

A. Advanced data visualization tools

Advanced tools handle complex data, specialized visual forms, interactive exploration, and scalable delivery.

• Interactive dashboards: Filters, brushing, linked views, tooltips, and drill-down controls let users examine different levels of detail.
• Geospatial systems: QGIS, ArcGIS, and web-mapping frameworks combine coordinates, boundaries, projections, and thematic layers.
• Network visualization: Node-link diagrams and adjacency matrices represent entities and relationships, such as users connected by messages.
• Large-scale rendering: Aggregation, server-side queries, tiled maps, and GPU acceleration support datasets too large to draw mark by mark.
• Specialized outputs: Animated timelines, three-dimensional scientific views, and immersive displays can represent change or spatial structure.
• Analytical integration: Advanced platforms may incorporate forecasting, clustering, anomaly detection, or calculated metrics.

B. Applications and limitations

Advanced capabilities are justified only when interaction or specialized structure contributes directly to understanding.

• Applications: A public-health map can link geographic regions to time-series panels and demographic filters.
• Cognitive cost: Excessive controls, motion, dimensions, or simultaneous views can overwhelm users.
• Three-dimensional limitation: Perspective and occlusion make accurate value comparison difficult on ordinary two-dimensional screens.
• Operational cost: Advanced systems require data pipelines, hosting, access control, testing, maintenance, and performance monitoring.
• Validation: Every filter state and linked view must preserve consistent definitions and totals.

VIII. Data Visualization Types — Matching the Chart to the Analytical Task

A. Data visualization types

Visualization types should be selected according to the question, variable types, and relationships that need to be displayed.

• Comparison: Bar charts compare categorical magnitudes; dot plots offer a compact alternative when many categories are present.
• Temporal change: Line charts show ordered observations over time, while area charts emphasize cumulative magnitude.
• Distribution: Histograms group quantitative values into bins; box plots summarize median, quartiles, spread, and possible outliers; violin plots add density shape.
• Relationship: Scatter plots compare two quantitative variables; bubble charts add a third variable through area; heat maps encode values in a matrix using color.
• Composition: Stacked bars show part-to-whole relationships across categories; pie charts are most readable with few, clearly different proportions.
• Geospatial pattern: Choropleth maps shade regions by normalized rates, while proportional-symbol maps represent totals with symbol area.
• Hierarchy and flow: Treemaps display nested composition; Sankey diagrams encode flows by band width; network diagrams represent links among entities.
• Small multiples: Repeating the same axes across panels enables consistent comparison among groups.

B. Selection principles and limitations

A chart type is effective when its visual structure corresponds directly to the analytical task and data semantics.

• Counts versus rates: Mapping population counts by region may mainly reproduce population size; a rate per 100,000 people often supports fairer comparison.
• Bin sensitivity: Histogram shape depends on bin width, while choropleth interpretation depends on class boundaries.
• Ordering: Ranking bars by value improves comparison, whereas chronological and ordinal categories should retain meaningful order.
• Normalization: A 100% stacked bar emphasizes proportions, but an ordinary stacked bar preserves totals.
• Uncertainty: Confidence intervals, error bars, ranges, or probability bands should accompany estimates where sampling or measurement uncertainty matters.
• Final criterion: The simplest chart that accurately answers the intended question is generally preferable to a more decorative or technically elaborate alternative.