data-visualization

# Data Visualization Principles and best practices for effective data visualization. ## Core Principles ### Tufte's Foundations **Data-Ink Ratio**: Maximize ink used for actual data - Remove unnecessary gridlines, borders, backgrounds - Eliminate 3D effects, shadows, decorative elements - Every visual element must justify its existence **Lie Factor**: Graphical representation must match data - Lie Factor = Size of Effect in Graphic / Size of Effect in Data - Ideal = 1. Substantial distortion when >1.05 or <0.95 - Avoid: non-zero baselines, inconsistent scales, 3D distortion **Chart Junk**: Remove non-data ink and redundant data-ink ### Graphical Integrity | Practice | Rule | |----------|------| | Bar charts | Must start at zero | | Proportions | Size encodements reflect actual ratios | | 3D effects | Never use—distorts perception | | Pie charts | Maximum 3-4 slices | ## Visual Perception ### Gestalt Principles | Principle | Application | |-----------|-------------| | **Proximity** | Cluster related data; space different categories | | **Similarity** | Consistent color/shape for categories | | **Continuity** | Connected line charts for trends | | **Closure** | Complete shapes, avoid unnecessary borders | | **Figure/Ground** | Data stands out against background | | **Connection** | Lines/links show relationships | ### Preattentive Attributes Processed in <200ms before conscious attention: **Hierarchy:** Position > Color > Size > Shape > Orientation | Attribute | Use Case | |-----------|----------| | Position (spatial) | Ranking, trends | | Color (hue) | Categorical distinction | | Size | Quantitative comparison | | Shape | Category distinction | | Intensity | Highlighting differences | ## Color ### Palette Types **Sequential**: Ordered data (low → high), single hue light to dark **Diverging**: Data with meaningful midpoint, two hues meeting at neutral **Categorical**: Nominal data, distinct equally-spaced hues (max 6-8) ### Color-Blind Safe Palettes ~8% of men and 0.5% of women have color vision deficiency. Design for them by default. **Okabe-Ito Palette** (recommended default): - Black `#000000`, Orange `#E69F00`, Sky Blue `#56B4E9` - Bluish Green `#009E73`, Yellow `#F0E442`, Blue `#0072B2` - Vermillion `#D55E00`, Reddish Purple `#CC79A7` **Other tested palettes**: Viridis, Cividis, Paul Tol, ColorBrewer (colorblind filter) ### Color Blindness Types | Type | Prevalence | Confusion | |------|-----------|-----------| | Deuteranomaly | 5% of men | Red-green (most common) | | Protanopia | 1% of men | Red-green | | Deuteranopia | 1.3% of men | Red-green | | Tritanopia | 0.0001% | Blue-yellow (rare) | | Achromatopsia | 0.003% | No color (greyscale) | ### Cultural Considerations | Color | Western | China | Other | |-------|---------|-------|-------| | Red | Danger | Good luck | Death (some African) | | White | Purity | Death | | | Green | Environment | Infidelity | | ## Accessibility ### Fundamental Rule **Never use color alone to encode data.** Always combine with at least one other channel: - Color + pattern/texture (bar charts) - Color + line style: solid, dashed, dotted (line charts) - Color + shape (scatter plots) - Color + direct labels (all charts) ### Contrast Requirements (WCAG) | Element | AA (minimum) | AAA (enhanced) | |---------|-------------|----------------| | Body text | 4.5:1 | 7:1 | | Large text (18pt+) | 3:1 | 4.5:1 | | Non-text (lines, bars, icons) | 3:1 | — | | Focus indicators | 3:1 | 3:1 + 2px space | ### Low Vision Design ~253 million people globally have visual impairment. **Typography in charts:** - Minimum 12px for labels (14px preferred) - Sans-serif fonts for readability - Line height ≥1.5× - High contrast text (4.5:1 minimum) **Layout:** - Support browser zoom to 200% without content loss - No horizontal scrolling at zoomed levels - Direct data labeling reduces magnification needs - Avoid cluttered, dense layouts **Dark mode:** - Avoid pure white (#fff) on pure black (#000) — causes halation - Use slightly heavier font weights on dark backgrounds - Maintain all contrast ratios ### Screen Reader Accessibility **Alt text for charts** (follow the four-level model by Lundgard & Satyanarayan, 2021): 1. **What**: Chart type and data ("Bar chart showing 2024 sales by region") 2. **How**: Encoding method ("Bars represent revenue in USD") 3. **Readout**: Key values ("North America leads at $450K") 4. **Insight**: Patterns/trends ("45% year-over-year growth") **SVG accessibility:** ```html <svg role="img" aria-labelledby="title desc"> <title id="title">2024 Sales by Region</title> <desc id="desc">Bar chart. North America leads with 45% of revenue.</desc> </svg> ``` **Always provide**: data table alternative or detailed description for complex charts. ### Multi-Modal Access Beyond visual, consider: - **Sonification**: Map data values to pitch/rhythm (Highcharts Sonification Studio, TwoTone) - **Tactile graphics**: Raised-surface charts for blind users (swell-touch paper, 3D print) - **Haptic feedback**: Vibration intensity proportional to data values - **MAIDR**: Multi-Access Interactive Data Representation (text + audio + tactile) ### Accessibility Legislation | Jurisdiction | Law/Standard | Requirement | |-------------|-------------|-------------| | **US** | Section 508, ADA | WCAG 2.0/2.1 AA (federal); courts use WCAG for ADA | | **EU** | European Accessibility Act (2025), EN 301 549 | WCAG 2.1 AA for public + private sectors | | **UK** | Equality Act 2010, PSBAR 2018 | WCAG 2.1 AA for public sector | | **Japan** | JIS X 8341-3:2016 | Aligned with WCAG, AA recommended | | **Australia** | Disability Discrimination Act 1992 | WCAG 2.1 AA benchmark | | **Canada** | Accessible Canada Act, AODA (Ontario) | WCAG 2.0 AA minimum, moving to 2.1 | | **Singapore** | Digital Service Standards | WCAG 2.1 AA for government | | **China** | GB/T 37668-2019 | Aligned with WCAG 2.0 | | **South Korea** | KWCAG 2.1 | Aligned with WCAG 2.1 | ### Accessibility Testing Checklist - [ ] Color not sole encoding method (WCAG 1.4.1) - [ ] Text contrast ≥4.5:1 (WCAG 1.4.3) - [ ] Non-text contrast ≥3:1 (WCAG 1.4.11) - [ ] Supports 200% zoom (WCAG 1.4.4) - [ ] Text spacing adjustable (WCAG 1.4.12) - [ ] Alt text or data table provided (WCAG 1.1.1) - [ ] Keyboard navigable (WCAG 2.1.1) - [ ] Test with color blindness simulator (Coblis, ColorOracle) - [ ] Test with screen reader (NVDA, VoiceOver) - [ ] Grayscale test: still understandable? ## Chart Selection ``` Data type: ├─ Categorical comparison → Bar chart ├─ Part-to-whole → Treemap/stacked bar (avoid pie >4 slices) ├─ Time series → Line chart ├─ Distribution → Histogram, box plot, violin ├─ Correlation → Scatter plot ├─ Geographic → Choropleth, proportional symbol └─ Network/flow → Network graph, Sankey ``` ## Common Mistakes ### Avoid - Truncated Y-axis in bar charts - Dual Y-axes (false correlations) - >4 pie chart slices - 3D charts - Rainbow palettes without meaning - Over-plotting (too many points) - Color-only encoding (accessibility failure) - Insufficient contrast on chart elements ### Fixes - **Clutter** → Small multiples, sparklines - **No context** → Add baseline, benchmarks - **Hard to compare** → Consistent scales, aligned axes - **Data overload** → Filter, aggregate, progressive disclosure - **Inaccessible** → Redundant encoding, alt text, data tables ## Domain Guidance ### Financial - Candlestick charts for prices - Treemaps for portfolio allocation - Log vs linear scale for long timeframes - Annotate key events (earnings, mergers) ### Security/SOC - Heatmaps for activity over time - Network graphs for connection analysis - Sankey for traffic flow - Red/amber/green severity (**with icons** for color-blind users) - Dark theme preferred (reduce eye strain) ### Scientific - Vector graphics (SVG, PDF) - Field-specific conventions - Follow Nature 2025 checklist: clarity, accessibility - 300+ DPI, clear labeling - Color-blind safe palettes mandatory for publications ## Tools | Use Case | Tool | |----------|------| | Custom web viz | D3.js, Plotly, Olli (accessible) | | BI dashboards | Tableau, Power BI, Apache ECharts | | Static reports | Matplotlib, Seaborn, ggplot2 | | Rapid prototyping | Flourish, Google Data Studio | | AI/ML integration | Python (Matplotlib, Plotly, Altair) | ### Accessibility Tools | Purpose | Tool | |---------|------| | Color blind simulation | Coblis, ColorOracle, Chrome "Let Me Color" | | Contrast checking | WebAIM Contrast Checker, axe DevTools | | Screen reader testing | NVDA (free), JAWS, VoiceOver | | Accessibility audit | WAVE, Lighthouse, Accessibility Insights | | Accessible charts | Olli (MIT), Highcharts Sonification | | Color palette design | ColorBrewer, Paul Tol, Accessible Colors | ## Quick Reference 1. **Start with grayscale** — add color only to encode data 2. **Redundant encoding** — color + pattern/shape/label always 3. **Okabe-Ito palette** — default color-blind safe choice 4. **4.5:1 / 3:1** — text contrast / non-text contrast minimums 5. **Small multiples** — same chart across subsets solves clutter 6. **Tufte's test** — can you remove this element and still understand? 7. **Alt text + data table** — provide text alternatives for every chart 8. **Test accessibility** — simulator + screen reader + grayscale ## Resources - Tufte's 4 books (foundational) - Wilke, *Fundamentals of Data Visualization* (2019) - Lundgard & Satyanarayan, "Accessible Visualization via Natural Language Descriptions" (2021) - Hajas et al., "Rich Screen Reader Experiences" (2022, MIT) - Nature 2025 scientific visualization checklist - W3C WAI: w3.org/WAI - ColorBrewer: colorbrewer2.org - WebAIM: webaim.org - DIAGRAM Center: diagramcenter.org