Design reading notes

Edward Tufte — legal full-text readings log

Edward Tufte — legal full-text readings log

Project: Meaning of Work Mixed Methods Study

Output dir: Manuscript/LEC_poster/digital_goody_bag/design_readings/

Date of loop: 2026-09-26

Method: WebSearch + WebFetch on edwardtufte.com notebooks and Graphics Press uploads; university-hosted author papers; local PDF download only from edwardtufte.com or .edu course archives that host Tufte coauthored journal articles. No Sci-Hub / LibGen / Anna’s Archive / Z-Library / Scribd scraped books. No reconstruction of the four Graphics Press monographs from memory.

Saved legal PDFs in this folder:

---

Documents read in full (legal sources)

1. Sparkline theory and practice Edward Tufte

1. Definition: a sparkline is a “small intense, simple, word-sized graphic with typographic resolution” (opening).

2. Graphics need not be special occasions in boxes; they can sit wherever words/numbers sit (opening).

3. General design rule stated: max[data], min[design] (Twitter analytics section).

4. Prefer design minimization, not data minimization (Apple Watch critique).

5. Glucose example: noun + number gains meaning when placed in a path of prior readings (Beautiful Evidence excerpt in ET comment 27 May 2004).

6. Add normal-range band so deviations are visual, not only numeric (same excerpt).

7. Stack many variables as small multiples for parallel comparison of hundreds of series (same).

8. Sparklines are “datawords”: data-intense, design-simple, word-sized (same).

9. Be “approximately right rather than exactly wrong” (Tukey cited; financial table discussion).

10. Sparklines reduce recency bias by showing recent change against long history (euro / mutual-fund tables).

11. Aspect ratio: prefer “lumpy” profiles (banking toward ~45°) over spiky/flat (Cleveland cited).

12. Avoid unintentional optical clutter (moiré, heavy frames, activated negative space); ask whether prominent visual effects carry content.

13. Paper resolution >> screens for serious sparkline work; screens ~10% of paper.

14. Layout metaphor: sparklines like words; page of sparklines like sentences/paragraphs; prefer adjacency in space over stacking in time (500 sparklines on A3 ≈ many screens/slides).

15. Avoid data frames; location of words/numbers/graphics enforces the grid (ET reply to Bissantz, 1 Jun 2004).

16. “Importance flags” (modest color fields) for notable segments; theory points to layering/separation and smallest effective difference (ET, 19 Aug 2005).

17. Contextual zooms preferred over erasing prior scale (micro/macro) (ET, 16 Oct 2005).

18. Interocular Trauma Test: compare observed sparkline to randomized replays of same data to dilute streak narratives (opening + later posts).

---

2. Sparklines History by Tufte: 1324 to now

1. Sparklines: data-ink / data-pixel ratio = 1.0; no frames, ticks, non-data paraphernalia.

2. Scaling context often comes from nearby words/numbers, not chart furniture.

3. Fundamental principle recalled from VDQI: “Above all else show the data.”

4. Three overlapping drivers: maximize data density; minimize/zero non-data; shrink principle (“graphics can be shrunk way down”).

5. Deliberate reduction of large examples (Bertin page; ECG) increases density.

6. Yale students reduced Excel charts by factors of ~3–5 to sentence/cell size.

7. Prior art across centuries (illuminated manuscripts, Halley contours, ECGs, Kubrick 2001 grid graphics).

8. Contours as three-dimensional sparklines.

9. 1989 HP consulting: “postage stamp graphics” / mini-graphics beside spreadsheet rows.

10. Critique of Microsoft “sparklines in the grid” patent relative to prior art.

---

3. Making better inferences from statistical graphics Edward Tufte

1. Combat the “rage to conclude” that sees patterns where none exist.

2. Statistical jargon (“explained variance,” “significant”) can itself feed premature conclusion.

3. Everyday integrity failures: ridiculous alternatives, millions of fitted models, “little tilts.”

4. First QC mechanism: a documentation box with data links, location in full dataset, analyst names, and whether the graphic was pre-specified or a found object.

5. Documentation boxes make cherry-picking more obvious.

6. Software should emit documentation boxes by default (as it does scales/labels).

7. Dilute cluster/streak bias via inline randomization / Interocular Trauma Tests.

8. Sparklines as tools against recency bias and for approximate high-res context (Beautiful Evidence pp. 50–51 excerpted).

9. Graphical resolution far exceeds tabular character density (cartographic 0.1 mm resolving power cited).

10. Mutual-fund sparkline table: ~5,000 more numbers in only ~21% larger data area.

11. Cite Ioannidis and oncology replication failures as motivation for better visual inference hygiene.

12. Edmond Murphy on bimodal distributions and dubious two-mechanism stories (VDQI p. 169 referenced).

---

4. Analytical design and human factors

1. Purpose of analytical displays: assist thinking about evidence.

2. First design question: What evidence-thinking tasks must this display serve?

3. Tasks: describe data, compare, understand causality, assess credibility.

4. “Logic of design replicates the logic of analysis.”

5. Deep design principles come from cognitive tasks of analytical reasoning, not from marketing/ISO/focus groups.

6. Great designs usually come from people “possessed by the content.”

7. Work should be self-exemplifying (show amazing evidence displays).

8. Outside-in interface design: content experts specify screens first; software is by-product.

9. Short form: “Simple designs, complex information.”

10. Maximize content-reasoning time; minimize design figuring-out time.

11. Theme across books named here: data-ink ratio (VDQI); micro/macro + layering and separation (Envisioning Information); smallest effective difference (Visual Explanations).

12. Models: good maps, aerial photographs, scientific flood-of-data practice.

13. Admiration for Cleveland’s analytically aware graphical perception work; skepticism of dust-bowl human-factors experiments.

---

5. Maps moving in time: a standard of excellence for data displays

1. Swiss mountain maps as comparison standard for high-end displays.

2. Checklist: content-first; high resolution; third dimension; local detail in larger context.

3. Light/realistic colors to avoid optical clutter; content-driven color.

4. Typographic excellence; type size proportional to labeled object (quantified type).

5. Contours = intense quantitative data at sparkline resolution; contours as 3D sparklines.

6. Natural integration of words, numbers, depictions; no annoying pop-ups.

7. Zero chartjunk; all pixels carrying content; avoid dequantification.

8. Pairwise adjacent comparison: proposed viz vs Swiss map (write into contracts).

9. Extend with motion/panning for gentle 3D reading without contraptions.

10. Google Maps as workaday comparison set: content-rich, design-straightforward, scale bar present, layered micro/macro, free of drop-shadow label boxes.

11. Marketplace ethics / open formats over focus-group blandness.

---

6. Slopegraphs for comparing gradients: Slopegraph theory and practice

1. Slopegraphs compare changes over time for nouns on an ordinal/interval scale.

2. Classic tax/GNP example from VDQI pp. 158–159: readable vertically (ranks), across (change), and down the slopes (unusual gradients).

3. Information both integrated (content) and separated (uncluttered reading paths).

4. Cancer survival slopegraph (Beautiful Evidence pp. 174, 176): every visual element shows data.

5. Table and slopegraph are colleagues, not competitors (ET reply on SEs).

6. Prefer independent replication for credibility over decorating slopes with SE bars when assumptions fail (ET reply).

7. Option: replace slope line with year-by-year sparkline of survival.

8. Bumps charts as slopegraphs (Beautiful Evidence pp. 56–57; Envisioning Information p. 111).

9. Implementation subtleties: thin gray lines, refined type, avoid line/label crashes, optional quiet colors for selected lines.

10. Require a separate data-documentation box (source, link, responsible person) in implementations.

11. Slopegraphs focus on slopes/deltas with high data and straightforward reading.

---

7. PowerPoint Does Rocket Science–and Better Techniques for Technical Reports

1. CAIB and Return-to-Flight boards: PP inappropriate for engineering reports; technical report superior.

2. Prefer intense summary matrices on 11×17 paper over “chippy and twiddly” slides for PRA/risk.

3. Risk models that omit on-ground analytic failure miss much of observed shuttle loss risk (ET PRA comments).

4. Critique of STS-121 FRR deck: dequantification, branding consuming ~20% of slide space, empty hierarchy, decorative 3D arrows, pitch tone.

5. In real science, photographs carry scales; low-res slides prevent that.

6. Bullet lists as “base-touching grunts”: effects without causes, verbs without subjects.

7. Bad presentational style can place truth in disrepute, not only enable sloppiness.

8. Prefer reading/exploring rich tables (sports/weather/finance metaphor) over stoplight PP.

9. Sparklines proposed for weekly indicators to fight recency bias (Iraq I&W slide notes).

10. Over-reaching methodologies (A3 fad example): when a narrow technical idea sprawls metaphorically, credibility must be re-earned locally (Beautiful Evidence p. 151 quoted).

11. Teaching/scientific papers better metaphors for presentations than marketing.

12. Amazon/Bezos and later Jobs anecdotes aligned with ET meeting method (materials read in advance; don’t present the deck).

---

8. New edition of “The Cognitive Style of PowerPoint”

1. First printing sold out; new edition added pages (Columbia board material, Feynman on bullet lists, FAQ).

2. FAQ themes named: Is PP just a tool? Must I use PP because colleagues do? PP-free talk methods? NASA Earth-centered slide.

3. 2nd edition (2006) = 32 pages; also a chapter in Beautiful Evidence.

4. Confirms Columbia/Boeing slide analysis as central evidence case.

5. Essay remains commercial product for full continuous reading (see Not obtained).

---

9. Corrupt Techniques in Evidence Presentations

1. Chapter focus: what audiences should look for when assessing credibility.

2. Published chapter length noted as ~16 pages in Beautiful Evidence.

3. Dequantification of categorical data (means-only bars) parallel to Galenson critique; insist averages come with variability (ET, 5 Feb 2005).

4. Tukey box-and-whiskers as responsible summary; redesign referenced VDQI pp. 123–125.

5. FDA cox-2 hearings: omitting inconvenient published data is a credibility failure (ET, 19 Feb 2005).

6. Cherry-picking metaphor (Spence Dixie cups) endorsed as classroom-clear.

7. Self-grading of systems one is selling is systematically associated with glowing assessments (JAMA EHR editorial cited).

8. Ioannidis “Why Most Published Research Findings Are False” posted as support for selection/bias themes.

9. Image-fingerprint / duplicated-graph fraud detection (Hwang; Schön) as evidence consumers’ tools.

10. Extraordinary findings deserve extraordinary review (possibly on-site) before publication.

11. Over-reaching section (cross-referenced from Rocket Science thread): precise technical ideas become metaphors only with fresh local evidence (BE p. 151).

---

10. Links, Causal Arrows, Networks

1. Draft Beautiful Evidence material on linking lines and causal arrows.

2. Links themselves treated as explanatory evidence; assess credibility of links.

3. Typographic org chart replacing bureaucrats-in-boxes.

4. Related prior threads: Barr Art Chart; Feynman diagrams.

5. Network viz collections often lack scales and tend toward “amazing visualization” over quantified explanation (Visual Explanations ch. 1 point recalled).

6. Useful question for each image: What did I learn beyond elegant architecture?

7. Screen light good for anti-aliased images; pixelation weaker for fine typography vs paper.

---

11. Overlapping data graphics to make comparisons

1. Mutual-fund sparkline-table then overlapped/stacked piles for comparison (Beautiful Evidence p. 51).

2. Practical comparison technique: physical alignment (ECG folded to light) as low-tech overlapping.

3. Overlap/stacking as a method to reveal common pattern vs divergent series (e.g., PIMCO vs stock funds in related sparkline material).

---

12. Pie Charts (forum thread with ET replies)

1. Spatially located nouns + area-coded circles + pies = stacked difficulties (map already uses 2D).

2. Prefer ordered table or table-graph with little bars; map nearby if needed.

3. Log-normal geographic distributions complicate area encodings.

4. First pie chart attributed to Playfair Statistical Breviary (1801), discussed VDQI pp. 44–45.

5. Later: direct pie vs table comparison (“Coalition of the Willing”).

6. Forum orthodoxy against pies is treated as design reasoning, not mere fashion.

---

13. Cleaning up Excel’s poshlust graphics

1. Default Excel graphs are chartjunky; students historically hacked Excel toward sparklines.

2. Community add-ins/custom chart types aimed at reducing non-data decoration.

3. Prefer pipelines that preserve SVG/vector quality (R/gnuplot → Inkscape) over style-heavy Excel defaults.

4. “Styles/effects” proliferation framed as making communication harder, not easier.

---

14. Chartjunk (notebook index post)

1. Page is a pointer to book loci, not a standalone essay: VDQI 106–121; Envisioning Information 52–65; Visual Explanations 88–89, 146–150; Beautiful Evidence 158–159, 170–179.

2. Confirms “chartjunk” as a book-term with multi-volume treatment, not fully restated here.

---

15. Computing Lie Factor by Dividing Percentages

1. Forum cites VDQI p. 57 definition and fuel-economy example (data 18→27.5; graphic 0.6→5.3).

2. Debates whether “size of effect” = percent change (Tufte LF = 14.8) vs simple factor ratios.

3. No full original definitional chapter text is posted; only secondary discussion of the book formula.

4. Related integrity theme: area encodings of one-dimensional quantities (banking market-cap circles comment).

---

16. Feynman–Tufte principle

1. ET prefers the general form: “simple design, intense content” over van metaphors.

2. Anchors the same spare-design / rich-content stance as Analytical Design thread.

---

17. Improving Conventional Graphics for Statistical Data (1977 sketches)

1. Contents map proto-VDQI topics: lying graphics; word-worth of one picture; less is more; blot maps; area and effect; text–graphic integration; small multiples; friendly graphics; parallel plots; consciously 3D graphics; aesthetics; case studies.

2. Opens with warning that charts can make falsehood look true (Rockefeller/charts anecdote).

3. “Word-worth of one picture”: long news prose vs one clear time-series of franked mail volume.

4. Critique of lousy component bar charts and overbuilt annual-report imagery (light bulbs for profits).

5. Small multiples and parallel plots as early organizing strategies.

6. Friendly graphics: strip unused digits / noise (e.g., delete trailing “.0”).

7. Integration of written text and graphics as production and cognitive problem.

8. Scatterplot redesign examples (midterm elections; registration rates) as iterative clarity work.

9. Population age–sex pyramids as information-dense multivariate displays.

10. Overall: excellence via showing data clearly and comparatively, failure often in news/PR graphics—trajectory later formalized in VDQI.

---

18. Citation Classic — The Visual Display of Quantitative Information (Tufte, 1992)

1. VDQI analyzes excellence (science/cartography) and failure (news graphics) since Playfair.

2. Practical theory: maximize data density and multivariate/comparative quality.

3. Book should be self-exemplifying; author controlled design (Gralla collaboration).

4. Integrate graphics into text, sometimes mid-sentence—eliminate usual text/image separation.

5. Self-publishing chosen to protect design integrity and accessibility of price.

6. Making the book is part of the scholarship (resolution, color, production as intellectual expression).

7. Origin story: teaching journalists; Tukey seminars concentrating the mind.

8. Thin prior literature too often about ruling pens, not quantitative reasoning.

---

19. ET Truth Matters: Evidence and Conclusions (open source chapter)

1. Take anonymous statistical lives as seriously as named individual lives.

2. Universal standards of statistical reasoning about truth.

3. Core questions: How do I know that? How do we know what we don’t know?

4. Feynman’s law: do not fool yourself—you are the easiest to fool.

5. Fundamental analytical principles: causality/mechanism, explain, compare, assess measurement/inference credibility, validate, enforce integrity.

6. These principles cannot be repealed by sponsors, fashions, or monetizing.

7. Observe data at the moment of measurement, not only at publication (Cuthbert Daniel).

8. Track provenance and life history of measurements; follow the money in medical measures.

9. Prefer forensic spreadsheet audits over mere “data cleaning.”

10. Avoid surrogate endpoints that replace meaningful outcomes (e.g., PFS theater vs overall survival/QoL).

11. Show the data; glib summaries / double-binning / imaginary thresholds signal falsification risk.

12. Model multiplicity: same data, many models, divergent messages (soccer red-card crowdsourcing study).

13. Good models explain; they do not merely memorize (overfit).

14. Classical “error/confidence/significance” puns are poor substitutes for empirical uncertainty detective work.

15. Exemplary credible reporting: show competing models, acknowledge unquantifiable unknown unknowns (exomoon paper praised).

16. Credible covid dexamethasone trial traits: RCT, all-cause mortality, absolute+relative risks, independent replication.

17. Meta-research / Chalmers: weaker designs → more enthusiasm for favored treatments; Ioannidis on false findings.

18. Quick credibility scoring checklist (RCT +, conflicts −, binning −, show data +, etc.).

19. Excellent live viz model: TEE during heart surgery—close to data, real-time QC, documented artifacts.

20. Graphics of vaccination natural experiments and life expectancy vs health expenditure as high-integrity displays of statistical lives.

---

20. Powsner & Tufte — Summarizing Clinical Psychiatric Data (1997)

1. One-page summary designed to serve patient care, not only admin/legal paperwork.

2. Retain diversity of data types: numeric, narrative, diagrammatic, temporal, relational.

3. Elements: IDs; visit/hospitalization timeline; background narrative; genogram; recent notes (newest on top); measurement/dosage small-graph matrix.

4. Nonlinear timescale: past year large; prior years compressed as context.

5. Columns group problems (psychosis, mood, diabetes) with findings above interventions.

6. Vertical position encodes clinical significance (normal / elevated / critical), not only raw units.

7. Current value printed and plotted as rightmost point; stopped measures omit “current.”

8. Horizontal bars for continuous dosage over time.

9. Goal: assess relations between interventions and outcomes; invite alternative strategies.

10. High-resolution display of coherent chronologies across generations → lifetime → clinic → weeks.

11. Explicitly cites VDQI, Envisioning Information, Visual Explanations page ranges for small-graph method.

12. Computer systems should not cast institutional habit into silicon before redesign.

---

21. Powsner & Tufte — Graphical Summary of Patient Status (1994)

1. One-page graphical summary of patient status for clinical use.

2. Matrix of small plots for labs/treatments over time (foundation for sparkline medical stacks).

3. Integrates quantitative tracks with clinical context on a single eyespan.

4. ET later note: prefer sparklines with normal limits over boxed normal-limit frames (graphical-summaries-for-medical-patients notebook).

---

Not obtained (no legal full copy found / not free)

Four Graphics Press books (full)

BookStatus
The Visual Display of Quantitative Information (1983/2001)Not obtained. Sold commercially; only sketches (1977), Citation Classic commentary, and scattered page citations/excerpts on edwardtufte.com.
Envisioning Information (1990)Not obtained. No legal full PDF; principles named in notebooks (micro/macro, layering and separation) without full chapter text.
Visual Explanations (1997)Not obtained. No legal full PDF; “smallest effective difference” named in notebooks; Challenger material referenced but not fully posted as book text.
Beautiful Evidence (2006)Not obtained as full book. Substantial legal excerpts appear as notebook drafts/images (sparklines, corruption, links/arrows, overlapping, slopegraphs) and as the open-source Truth Matters chapter from the later book Seeing with Fresh Eyes—not a substitute for BE entire.

The Cognitive Style of PowerPoint: Pitching Out Corrupts Within

Other

---

Sourced in the essays read vs commonly attributed without a source in hand

Clearly present in legal texts read here

Rule / ideaWhere grounded
Above all else, show the dataSparklines History; Citation Classic trajectory
Maximize data density; minimize non-dataSparklines History; Citation Classic; Analytical Design
Shrink principle / word-sized graphics / sparklinesSparkline theory; Sparklines History; medical papers
max[data], min[design] / simple design, intense contentSparkline theory; Feynman–Tufte; Analytical Design
Recency bias; approximately right not exactly wrongSparkline theory; Making better inferences
Documentation box / anti-cherry-pickingMaking better inferences; Slopegraphs contest notes
PP weak for technical reports; prefer paper reports/matricesRocket Science thread; New edition PP thread
Swiss maps / Google Maps as excellence standardsMaps moving in time
Slopegraphs for gradientsSlopegraphs notebook
Observe measurement at the source; show the data; statistical livesTruth Matters open-source chapter
Small multiples / parallel comparison stacksSparkline theory; 1977 sketches contents; medical summaries
Chartjunk as named enemy of content pixelsMaps moving in time; Chartjunk page pointers; Excel thread

Named as book-located but not fully readable in legal free text here

Commonly cited ruleStatus in this legal haul
Data-ink ratio (formal definition/formula)Named and endorsed (Analytical Design; Sparklines History ratio=1.0 for sparklines) but full VDQI chapter not obtained
Lie factor (formula and worked chapter)Discussed second-hand on forum citing VDQI p. 57; no author-posted full chapter
Chartjunk taxonomy/examplesIndex to book pages only; no full chapter body
Graphical integrity six principles as a listNot retrieved as continuous legal text
Smallest effective differenceNamed (Visual Explanations); not full chapter
Layering and separation; micro/macroNamed (Envisioning Information); not full chapters
PP cognitive style full monograph argumentPartial via free notebooks; full 32-page essay is paid
“The representation of numbers… should be directly proportional…” etc.Classic VDQI integrity language—not verified from a legal full file in this loop

Common attributions—treat as unverified here unless bought/borrowed

Do not treat the following as “read from source” for this project until a legal copy is in hand:

---

Search loop closure

Repeated queries (data-ink, chartjunk, sparklines, small multiples, lie factor, PowerPoint, Beautiful Evidence excerpt, Truth Matters PDF, sparklines Tufte PDF, site:edwardtufte.com/wp-content/uploads filetype:pdf, Powsner Tufte site:.edu) eventually returned duplicates of the notebooks and PDFs above. No additional free legal full texts of the four Graphics Press books appeared. Stopping rule met.

Practical next legal step if needed: purchase Graphics Press editions / PowerPoint ebook, or use library print copies; continue using the open-source Truth Matters PDF and the notebooks already archived here.