Tufte Mapped Pictures

Mapped Pictures: Images as Evidence and Explanation

Overview

A representational image shows what something looks like; a mapped picture lets a viewer measure it, place it, compare it, and test the claim being made about it. Tufte's argument in Beautiful Evidence is that explanatory, journalistic, and scientific images should nearly always be mapped — fitted with scales, a coordinate grid, labels, and overlays — and that a mapping is itself a claim, so it must be held to the same standards of credibility as any explanation. The chapter's two halves are equally important: how to map an image (scale, grid, labels), and how to tell a credible mapping from a crank one (explanatory tightness, falsifiability, flatland-vs-spaceland). The current skill title drops "and Explanation"; do not — the second half is where most real failures live.

"Mapped pictures combine representational images with scales, diagrams, overlays, numbers, words, images." — Tufte, Beautiful Evidence, p. 13

Chapter and citation note. "Mapped Pictures: Images as Evidence and Explanation" runs printed pp. 12–45. Page citations throughout are printed page numbers. Four items marked (pp. 112–121) come from the adjacent "Words, Numbers, Images — Together" chapter; they are flagged where they occur because they serve this thesis more directly than any other.

Provenance across the canon. The idea does not originate here. Visual Explanations ch. 1 supplies the direct ancestor — direct labels, encodings, and self-representing scales (p. 13), the NCSA tripod-of-scales redesign (pp. 20–23), Repton's pole-people as human-scale referents (pp. 16–17). Envisioning Information pp. 108–109 gives the earliest worked case: an aerial photograph of Hoboken overlaid with a route and place labels.

§1. What a Mapped Picture Is: Image Qualities + Diagram Qualities

A mapped picture is not an illustration with decoration added. It is the fusion of two complementary things: the evidential richness of a representational image and the focusing power of a diagram. Each contributes qualities the other lacks; the map is what binds them.

The test for whether a picture has been mapped (not merely captioned): could a stranger, with no prior knowledge, extract a measured value and locate any feature in shared coordinates without leaving the image? If the scale, grid, or labels live only in a separate caption or legend that travels separately, the image is still unmapped.

DO: Decide the single number or comparison the viewer should be able to read off the image, then build the scale/grid/labels to deliver it.

DON'T: Treat annotation as a finishing coat applied after the image is "done" — added late, it decorates instead of measures.

§2. The Scale of Measurement: Put Every Image on the Universal Ruler

The first and most-skipped obligation. Without a scale of measurement, sizes and distances are guesses, and two images cannot be compared at all. Tufte's worked examples are a catalog of how to attach a ruler to an image.

Forms of scale reference (worked from the chapter)

• Native unit overlaid on the drawing. Galileo recorded Jupiter's satellites with hundreds of annotated, scaled sketches — observation time noted, the satellites labeled, distances measured in Jovian-radii, the only relevant unit available. The annotation is what turned telescopic sketches into credible quantitative evidence of orbital motion.
• An explicit scale bar with a stated conversion. The Crick–Watson DNA demonstration model is photographed with a scale reading 0 to 10 ångströms. The payoff is the conversion the scale forces into view: 1 cm = 100,000,000 ångströms (10⁸), 1 inch = 254,000,000 ångströms. The bar both quantifies the model and signals the image is explanatory, not commercial art.
• A comparison object of known size. Placing Earth beside Saturn gives millions of casual viewers a sense of scale that a lone "celebratory photograph" never does. Scale is part of the news whatever the marketing department believes.
• A ruler drawn into the plate. Brisson/Martinet's 1760 cockatoo carries a "scale of 6 inches" (échelle de 6 pouces) at lower right; objects of roughly known size (plant leaves) reinforce it.
• Rescaling instructions when the drawing is reduced (pp. 34–35). Lilienthal's plate calculates the wing area of a 4 kg stork with rectangles and measurements overlaid on the drawing. Lilienthal drew it at ⅛ natural size; because Beautiful Evidence reproduces it at a further reduction, Tufte adds the instruction to multiply by 9.4 to obtain real stork size. The two figures are not in conflict — ⅛ is the original plate's scale, 9.4 is the factor for the reproduction in front of you, which is exactly why a rescaled image must state its own factor rather than inheriting the original's. The flapping-wing space-time tracings carry their own separate scales (1/20 head-on, 1/50 bird-side view). Tufte's complementary complaint on the same page: this plate and 7 others were folded up and stuffed in the back of the book, convenient for printing and binding and inconvenient for seeing and reading.

The anti-example: the "different and unknown scale"

Bloch's Ichthyologie (1795) draws 216 fishes — seahorses to sharks — each forced to fill an identical framing box. The beauty hides a defect: every fish therefore has a different and unknown scale, so no two can be compared. Tufte's fix is the redraw — gentle redesigns that add 0/5/10 cm measurement scales and mark cross-section locations, "mapping the images on the universal ruler."

Density is not a substitute for scale, but it is the reward

Dense annotated images repay the scale apparatus. Dürer's 1525 construction packs ~48 numbers mapping 3 line-traces (one of 100 such plates in a book literally titled a course in the art of measurement); Mersenne's 1636 lute engraving carries 80 numbers (312 digits) over a 19-string lute and 21-string theorbo. The lesson: when the image is information-rich, the ruler is what makes that information legible rather than ornamental.

DO: State the conversion (1 cm = N units) so the viewer can compute any measurement directly.

DON'T: Force heterogeneous subjects into identical frames — it silently rescales each one and destroys comparison.

§3. The Map Metaphor: Labels Belong on Images, Grids Scale Them, Context Makes Data Credible

Bayer's Uranometria (1603, engravings by Alexander Mair; pp. 20–21) is Tufte's cleanest statement of the map metaphor. The star atlas locates stars on a measured, labeled two-dimensional grid, encodes brightness by star-size (like city-size on a road map) and by Greek letter (α brightest, β next), and so yields a dual context: the universal sky grid plus the local neighborhood of nearby stars.

"For showing evidence, the map metaphor suggests that labels belong on images, that external grids help to scale images, and that data are more credible when contextualized." — Tufte, Beautiful Evidence, pp. 20–21

Three operational rules fall out of the metaphor:

1. Labels go on the image, next to what they name (see §5).
2. An external grid scales the image without competing with it (keep grid weight well below the data marks).
3. Context confers credibility — an isolated mark is a rumor; the same mark on a shared grid is evidence.

Note Bayer's honest weakness, which Tufte flags rather than hides: where real stars run out, the myth (Leo the lion) fills the gap — a reminder that the decorative layer must never be confused with the measured one. Galileo's telescope soon replaced the myth with real stars.

§4. The Universal Grid: x, y, z, t

The chapter's governing principle, stated at the close:

Every image — whether for explanation or for exploration — should reside on the universal measurement grid of 3-space and time: x, y, z, t (chapter close, p. 45). The grid must travel with the image through every transformation: it should accompany rescaling and zooming in and out, so a viewer never loses the absolute frame while inspecting a detail (as in the 3-image photosynthesis sequence, where every magnification still carries its scale bar and z-range).

Relative-only scaling is insufficient. Peterson's bird field guide scales birds relative to one another, which is not universal — Tufte's remedy is to add a constant external referent (a common plant, a few clams for shorebirds) or, far better, a real measurement scale. The principle generalizes the special cases of §2 and §3: a scale bar fixes one axis, a grid fixes two, the universal grid fixes all four and binds every image in a document into one comparable space.

DO: Keep the universal grid (or at minimum a scale + orientation) attached through every zoom, crop, and rescale.

DON'T: Rely on within-image relative scaling ("this bird is bigger than that bird") as if it were measurement — it breaks the moment the image is compared with anything else.

§5. Labels on the Image vs. the Dreaded Letter Code

Where the label sits decides whether the viewer reads or decodes.

• Names directly on the image tie the word to the thing. The dog group portrait labels each animal (Anna, Zerlina, Alex, Abby) in place; Bella's nine-week growth series puts the measurement labels in the photograph where they belong — Tufte's standard for honest scientific imaging ("Science should use Bella reporting standards"), set against astronomical images that are extensively color-processed yet "notoriously unscaled and dequantified."
• The dreaded letter code (A, B, C keyed to a separate list) should be avoided whenever possible. It forces the eye to ping-pong between image and legend and loses the tie between sign and referent.
• The diagram-as-caption / profile outline is the legitimate workaround when the image is too dense or too precious to letter directly. A traced outline placed beside the image (Marey's 1902 reunion: hat-and-head silhouettes carrying each scientist's name and affiliation; the dogs' profile-outline key) can hold richer captions than fit on the picture, and it can be drawn at a reduced scale to act as a compact legend. Loran's Cézanne mappings work the same way — placed adjacent to the painting, not overlaid on it.

DO: Put the name on the thing; if the image is too dense, trace a parallel outline as the legend.

DON'T: Default to A/B/C codes — Lichtenstein could parody Loran's earnest "DIAGRAM/A-B-C" precisely because the code-and-legend habit is so mechanical.

§6. Explanatory vs. Exploratory Mapping

Whether to annotate aggressively depends on the job.

• Explanatory presentation (you already know what matters and are showing it): point out the important comparisons with annotations, arrows, labels, highlighting. Peterson's field guide uses pointer-line "keys" to call out the distinctive marks that separate similar birds — exactly right for teaching identification.
• Exploratory analysis (you are still discovering what is there): premature mapping diverts the viewer, forecloses alternative readings, and masks subtle detail. The analyst should look with fresh eyes and deliberately ignore the directive hints. Premature closure and "viewing-to-please" turn open analysis into advocacy and marketing.

The both-views method (underused): show the unmapped image and the mapped image — a clean look first, then what the mapping brought to it. This lets the viewer audit the mapping instead of swallowing it. Tufte says this should be done much more frequently.

DO: In exploration, present the clean image first; add the mapping only after the viewer (or you) has looked freshly.

DON'T: Pre-annotate an image you are still trying to understand — you will see only what the arrows told you to see.

Where the mapping apparatus comes from (p. 115, adjacent chapter)

Having decided to map, how do you choose what apparatus to add? Tufte's rule: metaphors and analogies can genuinely assist reasoning about analytical design, but the metaphors should derive from the content and from the reasoning associated with that content — not from computer interfaces.

His worked positive case is the Golden Field Guide's jay spread (Robbins, Bruun & Zim, Birds of North America, 1966, illustrated by Arthur Singer). The task is visual recognition of physical objects, joined to new verbal, geographic, and graphical information, and every element of the apparatus is derived from that task:

The location maps are the sharpest illustration. They earn their space by preventing a specific, nameable error class: confidently identifying a bird in Montreal that in fact lives only in Acapulco. That is what content-derived apparatus looks like — each addition eliminates a known failure of the actual task.

The rule for §6, stated operationally: derive the mapping apparatus from the identification or reasoning task, not from a UI pattern library. "Add a tooltip / a legend / a zoom control" are interface metaphors and answer no question the content posed. Ask instead: what mistake will a competent viewer make with this image, and what annotation makes that mistake impossible?

DON'T: Base the design of an evidence display on a computer-interface metaphor — Tufte's explicit exclusion. The apparatus should look like the subject's own reasoning, not like software.

§7. Explanatory Tightness: Credible Mapping vs. Crank Mapping

The chapter's hardest and most original argument: when a mapping makes an explanatory claim about an image, it inherits the standards of credible explanation. A mapping can be drawn beautifully and still be worthless evidence.

The dividing line

What separates Loran's Cézanne analyses and Hockney's optical-instrument hypothesis (perspective lines traced on Holbein's Ambassadors to argue divergent vanishing points imply a mirror lens) from crank work is explanatory tightness: their mappings are specific, coherent, credible, and testable — they try to explain a particular thing and could in principle be wrong. Mössel's 300+ geometric overlays (in a 534-page book claiming Ur-Forms beneath nearly all art; p. 29) are the opposite: flexible enough to fit anything, refutable by nothing — Tufte's verdict at p. 31 is that all 300 are not replicated research but the drawings of a monomaniac, and that different maps fit the same artwork because many symmetries and proportions can be claimed of any image.

"...no possible empirical evidence can ever refute a Mössel, who explains everything and therefore nothing." — Tufte, Beautiful Evidence, p. 31

Why crank mappings always "work"

• Too many degrees of freedom. Different grids fit the same artwork; dots could plausibly sit elsewhere; even fixed dots can be connected many ways. A scheme that fits every case discriminates none. (Feller demolished the same move in statistics: a "law of logistic growth" was fitted to populations, bacteria, railroads, plant and animal size — but normal, Cauchy, and other distributions fit the same data as well or better, so the logistic explained nothing distinctive.)
• Alignment-to-please / viewing-to-please. A registration will "lock right into place if the desired answer is already known." Tufte aligning a sculpture plate to a north marker could gain ±3° just by tilting his head; land surveyors mapping the same parcel get divergent results depending on the client's interest. Construct the mapping independently of the result you want.
• The flatland–spaceland mismatch (catastrophic; p. 31). Mössel maps a single one-eyed flatland photograph of a real 3-dimensional object and treats the result as a fact about the object. A photograph is not the artwork; a flatland mapping cannot capture spaceland reality. No amount of re-aligning a grid-net over one projection recovers the lost dimension.

Test before you trust a mapping

DO: Demand a mapping be specific, coherent, credible, and testable, and build it blind to the result you hope for.

DON'T: Accept a mapping because it "fits" — a theory that fits everything has told you nothing; check it could have failed.

§8. Mapping Motion, and Images That Map Other Images

Motion in still-land ("movie-land")

Showing a moving thing in a static frame is inherently hard. The toolkit: directional arrows, floor plans, trace lines, linked images, and words. The 1762 contredanse pamphlet maps a dance as small multiples — 8 movements across a consistent grid, each cell combining figures, two floor-plans, trace-tracks, and music. Read down a column for the sequence within a movement, across for the sequence of movements; the steady repeated format is "a canvas upon which to paint, preserve, and reason about the changing information." Tufte flags its honest gap: the floor-tracks lack an explanatory legend. How to Ski by the French Method does the same with 200 motion-stopped photos, overlaying 2-color diagrams so the analytical variables (time, slope grade) read left to right. Good still-land strategies usually translate to actual motion graphics — design for both.

Images mapping images

An image can map another image: a before/after history, a painting beside its forgery, similar species side by side. The photosynthesis AFM triptych maps molecular dynamics with x-y scale bars per panel and a color scale for z (depth). Tufte's own Rock/Braque/Picasso trio (each carrying a 1 cm bar) maps a fractured rock against two Cubist paintings to argue that Cubism pictures real fractured 3-D planes. The rule that makes these work is §4: even comparison images must share the universal grid, or the comparison is impressionistic.

DO: Use a constant repeated frame (small multiples) for sequence and motion so only the data changes cell to cell.

DON'T: Compare two images that lack a shared scale/grid — "looks bigger" is not a measurement.

§9. Worked Example: The Vigilante (the chapter's own mapped picture)

The 1823 anti-slavery plan and cross-section of the slave ship Vigilante (pp. 22–23; London Religious Society of Friends, Case of The Vigilante, A Ship Employed in the Slave-Trade; folding plate engraved by J. Hawkesworth) is the chapter's central demonstration that a moral document gains force from engineering-diagram rigor.

• Plan, elevation, and cross-section of the ship captured off Africa in April 1822, with a "Scale of Feet" along the bottom and a "Water Line" on the section — the full apparatus of a measured drawing.
• Nearly identical human figures count out the numbers: 227 men and 120 women, 347 people; the elevation's double layer shows the packing; the labels mark Captain's cabin and wine lockers against the human cargo.
• The facticity — individuals cumulating into an overview, hidden lines, plan views, elevations, labels, a measurement scale — gives it "the straightforward quality and credible precision of an engineering diagram," and that precision is exactly what makes it more damning than any still-life or photograph. The detail signals the ship was examined carefully, which adds to the credibility of the image — and of the surrounding claim about the 12,000,000 to 20,000,000 people eventually carried into America and the New World by the trade (p. 22). Tufte's own comparison on the page is to Minard's 1812 graphic: both earn their force from the straightforward quality and credible precision of an engineering diagram.

The lesson: mapping does not cool an image; the measured grid is what makes the evidence undeniable.

§9b. The Designer-as-Researcher Standard (pp. 116–117, adjacent chapter)

The Vigilante shows what a well-mapped image achieves. This shows what production arrangement makes one possible.

Tufte's model of mapped motion is Megan Jaegerman's "Spotting a Hidden Handgun," parts of which appeared in The New York Times, 26 May 1992. It choreographs movement in 3-space using sequences, call-outs, motion arrows, mappings, and multiple viewpoints — silhouette, 3-dimensional figure, and flatland footprints — with words and images working together to describe an extended causal sequence: asymmetrical gait, upper-body shift, the quick adjustment when a gun's weight shifts, running from the rain, conspicuous clothing.

The claim that matters is not about the drawing. Jaegerman did both the research and the design, breaking what Tufte calls their common alienation. The design amplifies the content because the designer created the content — she knew which asymmetry mattered because she had interviewed the detective, watched him act out the motions, and handled the weapons herself.

He then reproduces her backup documentation as evidence of the care the work required: confirmation of Detective Robert T. Gallagher's identity and record against a public database, notes on his 18 years with the NYPD and his years in the anti-robbery tactical unit, the 1,200 gun arrests and near-flawless conviction rate, and — decisively for the graphic's logic — the legal sequence from suspicious demeanor to probable cause to frisk to arrest, which is why the tell-tale signs shown are the ones that matter.

The rule for this skill: mapping quality is downstream of whether the mapper did the reporting.

§10. Failure Modes

THE DIFFERENT-AND-UNKNOWN SCALE

What happens: Heterogeneous subjects are forced into identical frames (Bloch's 216 fishes in equal boxes), so each is silently rescaled and none can be compared. Fix: Put every image on the universal ruler — a real scale bar per image (the redraw), never uniform framing as a stand-in for scale.

THE DEQUANTIFIED CELEBRATORY IMAGE

What happens: A scientific or journalistic image is published as a beauty shot — color-processed, dramatic, and unscaled ("the marketing department"); the most extensively manipulated images (deep-space photos) are the least quantified. Fix: Bella reporting standards — scale bar plus measurement labels embedded in the image; a comparison object of known size at minimum.

THE DREADED LETTER CODE

What happens: Features are tagged A/B/C and keyed to a separate legend, breaking the tie between sign and referent and taxing the eye. Fix: Name things directly on the image; when density forbids it, use an adjacent profile-outline diagram as the legend.

THE FLATLAND–SPACELAND MISMATCH

What happens: A single flat photograph of a 3-D object is mapped and the result is claimed as a truth about the object (Mössel's grids over one-eyed photos of sculptures). Fix: Map the object in 3-space — multiple viewpoints, cross-sections, x/y/z scales; never over-read one projection.

THE UNFALSIFIABLE MAPPING

What happens: A scheme (Ur-Forms, golden-section-everywhere) is flexible enough to fit any image and refutable by no evidence — it explains everything and therefore nothing. Fix: Require explanatory tightness — specific, coherent, credible, testable; confirm the mapping could have failed and survives connecting the marks differently.

ALIGNMENT-TO-PLEASE / PREMATURE CLOSURE

What happens: The mapper unconsciously tilts the registration until it confirms the wanted answer (±3° free just by moving the head); or premature annotation in exploration forecloses alternative readings. Fix: Construct the mapping independently of the desired result; in exploration show the unmapped image first; state alignment tolerance honestly.

§11. Evaluation Checklist

Run before publishing any explanatory or evidential image.

Scale and grid

• A measurement scale is embedded in the image (not only the caption), with its conversion stated
• The scale survives reproduction/rescaling, with an explicit rescale factor if reduced
• The image sits on a universal frame (x, y, z, t as applicable), and the grid stays attached through zoom/crop
• Comparison images share one scale/grid, not relative-only sizing

Labels and context

• Names sit on the things they name; no A/B/C code unless density truly forbids in-place labels
• If a code/legend is unavoidable, it is a parallel outline diagram, not a bare list
• Each mark is contextualized (neighborhood + universal frame), not floating

Explanatory integrity

• The mapping makes a specific, falsifiable claim — name what evidence would refute it
• It was built independently of the desired answer (no alignment-to-please)
• It maps the real object, not a single flat photo treated as the object
• It still holds if the marks are connected a different way

Mode and motion

• Exploratory images shown clean first; aggressive annotation reserved for explanation
• Both unmapped and mapped versions offered where the mapping is contestable
• Motion/sequence uses a constant small-multiple frame; any track/legend gap is acknowledged

Quick Reference: Mapping Decision Table