Tufte Space Time Graphics

Space-Time Graphics: Narratives of Space and Time

Overview

Space-time graphics place a spatial dimension (route position, stations, distance along a path) on one axis and time on the other, so a moving entity traces a line whose angle encodes its speed, whose crossings reveal encounters, and whose density reveals congestion. The form was developed by the Paris engineer Charles Ybry in his 1846 British patent for railway scheduling, and it extends to any phenomenon that moves or develops through both space and time at once. (The famous Paris–Lyon chart is VDQI material — EI never connects Ybry to it, and VDQI p. 31 notes Marey himself credited that schedule to the engineer "Ibry.") The payoff: one flat graphic narrates what Tufte calls a "four or five variable story" — places, times, identities, speed — without a table of numbers (EI p. 101).

Source note (read this first). The substance here lives in Envisioning Information (EI), the chapter "Narratives of Space and Time," pp. 97–119 — Galilean satellites pp. 97–101, itineraries and timetables pp. 101–113, dance notation pp. 114–119 — with the grid-weight progression stated in EI p. 59 and The Visual Display of Quantitative Information (VDQI) pp. 115–116. Visual Explanations (VE) itself redirects this material: its footnote 12 (p. 93) points the reader to EI pp. 32, 45, 97–113. VE's key related display is the Salyut-6 cyclogram (pp. 92–95), a time-by-time grid — see §13, which bridges this skill to VE.

§1. The Canonical Axis Setup

The standard graphical timetable for a linear route:

Station placement: stations sit as horizontal bands or tick lines at their true proportional distance from the terminus. Unequal real spacing produces unequal vertical gaps — that is correct. Forcing equal vertical spacing onto unequal distances corrupts every slope on the chart.

Time scale: uniform clock time across the horizontal. To cover a full 24 hours without a hard cut at midnight, Tufte gives two devices (EI p. 107; see §5b for the source context): glue the schedule end-to-end onto a cylinder so midnight joins the next day seamlessly, or simply prolong the grid a few extra hours (as in his Atlanta-to-Chicago airplane schedule) to expose the complete cycle.

Dimensional compression: the graphic measures distance along the rail line itself, collapsing three-space geography into a one-dimensional "lineland" path — the same trick as an itinerary or strip map (EI pp. 112–113; developed in tufte-data-maps). Curvature and compass heading of the real route are discarded; only along-track position survives on the spatial axis.

Which axis gets space vs. time: time is normally horizontal, space vertical. Two named departures:

• Reversed (time vertical): the Jupiter/Saturn satellite charts run successive nights down the page and satellite east-west position across it — time vertical, space horizontal. This holds for Galileo's (1613) and Cassini's (1668) dotted observation tables and for the modern corkscrew redrawings alike (§7).
• Doubled (two variables on one axis): the Japanese-beetle life cycle (§7) carries both time and a horizontal ground-position on the horizontal axis while a second spatial dimension (depth, underground→surface) runs vertical.

Note — common mislabeling: the esophageal transit study is NOT a reversal. It uses the standard orientation — time horizontal (12 s), space (mouth→stomach) vertical. The reversed/doubled cases above are the only departures; §7 has the full set.

§2. Reading Speed from Line Angle

Every moving entity traces a diagonal. Its angle against the horizontal is its speed.

vertical axis   = distance (km or stations)
horizontal axis = time (hours)
slope of line   = Δdistance / Δtime = speed

• Direction: opposite travel directions produce opposite slopes (one leans \, the other /); the two line populations lean against each other across the grid.
• Comparison without arithmetic: an express reads as one steep clean diagonal; a local reads as a shallower line broken by many short horizontal dwell steps. Speed comparison is pure angle comparison.

Ybry's 1846 patent already states the design use: with the table in hand, drivers and guards regulate train speeds exactly, and for a special train its speed line can be ruled straight onto the chart to keep it clear of every preceding train. (Charles Ybry, British Patent No. 11,868, 1846, paraphrased; quoted in EI p. 108.)

§3. Crossings, Gaps, and Density

Line crossings — trains meeting

Two lines intersecting means two trains share one location at one moment. Unremarkable on double track; a hard constraint on single track.

• x-coordinate of the crossing = the time they meet.
• y-coordinate = the station or milepost where they meet.
• Crossing between stations on a single-track line = a physically impossible schedule — a fatal conflict requiring redesign.

Gaps — dwell time

A horizontal segment = a stopped train; its horizontal length = minutes stopped. Instantly visible here; in a numeric timetable dwell time is invisible unless explicitly listed, and it rarely is.

Density — frequency and congestion

Tight clustering of lines in a time window = high frequency = rush hour, packing toward spaghetti. The density is itself the message: when lines crowd that tightly, service is frequent enough that the rider just shows up. Sparse regions (few lines, wide gaps) = low-frequency midday or late night. The whole day's service pattern is legible in one glance — something a column of numbers cannot deliver.

§4. Single-Track Planning: Crossings Only at Sidings

On single track, opposite-direction trains can pass only where a siding exists. Tufte's example is a Swiss Federal Railroad chart of a few daily trains from La Chaux-de-Fonds (1932): the diagonals cross only at stations, revealing that the line is single-track and trains pass only at sidings within those stations.

• Visual signature: every crossing lands on a horizontal station band. A crossing between bands = a collision designed into the schedule.
• Correctly planned: up-trains and down-trains interleave at sidings with strict rhythmic regularity.
• Planning use: railways used the graphical form as the primary design tool for negotiating passing sequences across systems of immense complexity (thousands of station stops) long before computational optimization. Doing the same task nonvisually is clumsy and error-prone.

§5. Grid Treatment

The space-time graphic needs a grid more than most charts, because readers interpolate precise times and locations from it — but the grid must never compete with the data lines.

The weight progression Tufte demonstrates (VDQI pp. 115–116):

• Subordination rule (EI p. 59): only when a graphic serves as a genuine look-up table — "rare indeed" — does a grid earn any presence at all, and even then it must stay subordinate: a delicate gray grid supports more accurate data reconstruction than a heavy dark one.
• Interval (concrete): for transit, a gray grid at ten-minute intervals — fine enough to interpolate arrivals, coarse enough to stay quiet.

  "The gray grid is set at ten-minute intervals in order to ease visual interpolation of the times of arrival." — Tufte, Envisioning Information, p. 108 (the Hoboken bus-schedule passage)

• Paper trick (EI p. 59; also VDQI): ordinary graph paper prints its grid too dark for data; work on the reverse imprinted side so the rules show through faintly — and if the paper is heavily gridded on both sides, throw it out.

§5b. Cyclic Closure — Playfair's Torus and the Ratchet Effect (EI p. 107)

The §1 cylinder/prolong-the-grid devices have their source context in William Playfair's chart of Ordnance expenses. Playfair handled outlying wartime spending peaks in two steps: first temporarily extending the grid upward — Tufte's analogy is ledger lines in musical notation — and then topologically gluing the data surface from top around to bottom, the graphical counterpart of octave displacement. The glued chart is a torus-graphic, with thick horizontal bars flagging the war periods.

• The topology produces a finding, not just a tidier chart: a visible ratchet effect — postwar expense never shrinks back to its prewar level. (Tufte's footnote 6 adds a caution: Playfair's failure to deflate monetary units weakens his own argument.)
• The same gluing serves timetables: a 24-hour schedule joined end-to-end on a cylinder shows a fully connected daily cycle with nothing running off the edge at midnight; prolonging the flat grid a few extra hours does the same job on paper (the Tufte/Druckrey Atlanta–Chicago airplane schedule).
• Recycled globes complete the family: repeat the world map until every country and ocean appears whole somewhere in the picture — repetition spent to remove an arbitrary cut, and to avoid ethnocentrism. (Mechanics in tufte-redundant-encoding §6.)

§6. Spatial Detail vs. Temporal Precision: The Tradeoff

Two finite axes must be split between two continuous variables. Every choice shifts emphasis:

Worked resolutions:

• Czechoslovak Air Transport (1933): a route network is the spatial structure, with flight times and flight numbers annotated on the edges. Sacrifices minute-by-minute reading; reveals topology.
• China Railway index (200-page): route map carries page numbers along each line, pointing to where the detailed schedule lives. Sacrifices temporal data on the map entirely; provides a spatial index into a large document and avoids a witless alphabetical index.
• Hoboken-NYC bus schedule (Tufte & Druckrey): layers two full-fidelity representations — an aerial photograph at house resolution for space, a complete graphical timetable for time — neither compressing the other. Residents personalize the photo (finding their own street), then read the schedule above it.

There is no single correct ratio. Decide by reader task: planning routes → spatial detail dominant; planning departures → temporal precision dominant.

§7. The Space-Time Grid's Natural Universality

"The space-time grid has a natural universality, with nearly boundless subtleties and extensions." — Tufte, Envisioning Information, p. 110

One spatial dimension on one axis, time on the other — the same structure spans unrelated domains. Note the orientation column; most are standard, two are not.

Esophageal timetable (medical)

Eight consecutive video frames of 0.2 s each, 64×64 pixels, are compressed: each frame is summed along its horizontal rows into a single 1-pixel-wide column. The columns are assembled side by side — 60 in the full study — to build one condensed dynamic image of a complete swallow.

• Horizontal axis: 12 seconds total. Vertical axis: mouth (top) → stomach (bottom).
• Reading: a clean downward bolus trajectory is normal; descent rate reads from slope; reflux would appear as upward motion. (Standard orientation — time horizontal.)

Japanese-beetle life cycle (biological)

Popillia japonica Newman's full year is shown by doubling variables on the horizontal axis: it carries both the months (Jan–Dec) and the beetle's horizontal ground-position, while the vertical axis is depth (deep underground at bottom, surface at top). The illustrated organism sits at its true spatial-temporal position at each stage — a smooth escape from flatland.

Jupiter/Saturn corkscrew (astronomy)

The corkscrew diagrams are modern, 20th-century treatments — EI's versions are redrawn from Sky & Telescope (1988) and Bureau des Longitudes ephemerides (1987–88) — plotting continuous computed trajectories of Io, Europa, Ganymede, and Callisto. The historical precursors plotted dots, not curves: Galileo (1613) and Cassini (1668) recorded longer time-series "flickering with discontinuities" — separate nightly positions. Linking those dots into continuous curves was a design advance that arrived only in the 20th century, roughly 300 years after Galileo's discovery, despite the intellectual caliber of the astronomers involved. The result is a true micro/macro space-time grid — one spatial dimension stretched by time — but reversed: time runs vertically, east-west position horizontally, in the historical tables and modern redrawings alike. Tufte mutes the horizontal "prison-bar" gridlines to kill the 1+1=3 clutter (see §5); the smooth curves report every position of the moons, fitting data even for a few hours of viewing.

The satellites as a longitude clock (pp. 98–99): once satellite motion could be forecast precisely, it served as a worldwide signaling clock for determining longitude (24 hours corresponds to 360°). The predictions were entabled — "even gridlocked" — in ephemerides such as the 1766 Connaissance des Temps. Jovilabes, nomogram-like mechanical computation devices, recounted the orbits; solar-system machines showed why the earthbound view flickers — a shifting vantage point, plus Jovian shadows masking satellites from Earth view.

Bumps chart (competitive rowing)

The example is the Oxford University Torpids (Oxford colleges; redrawn from The Times, 3 March 1987) — not Oxford-and-Cambridge. The rivers' narrowness precludes more than two crews from rowing side-by-side — on bends there may be room for only one boat — so crews start spaced apart at intervals and chase the boat ahead; catching it ("bumping," historically a literal touch) advances a rank. Two abreast is possible; that is exactly why overtaking, rather than mass racing, structures the event.

• Vertical axis: crews by starting rank. Horizontal axis: race days.
• Crossed lines record each overtake. The form depends on the physical no-passing constraint of the narrow river.

§8. Conventional Typographic Timetable vs. Graphical Timetable

Tufte's analysis of the New York–New Haven Metro-North table (1983), Envisioning Information pp. 104–105:

"Only 21 percent of the timetable's area is devoted to display of times that trains run." — Tufte, Envisioning Information, p. 104

The graphical form resolves all of these at once: direction = slope direction; rush-hour load = visible crowding; speed comparison = angle comparison; dwell = horizontal segment; whole-day structure = one glance. It gives both the precise reading of one train and the macro overview of the day's structure — micro and macro together. (Tufte's redesign also reset the numbers in Carter's Bell Centennial, a typeface built for legibility in tight space.)

The deeper defects (p. 104):

• The most troubling flaw is the content: for several trains, the same journey ran no faster than it did 70 years earlier — design critique bottoms out in data critique.
• The antique October 1913 schedule shown alongside put the names of the officers responsible for operations on its cover: a sign of pride, and an effective force for quality control.
• Well-designed schedules separate hours from minutes with a visually quieter dot rather than a colon.
• The table's elaborate but false appearance of systematic order is Tufte's Vicksburg-courthouse failure: a grand portico of inflated Ionic columns, with the real work done in back rooms.

§9. Do / Don't Pairs

§10. Failure Modes

1. Equally spaced stations (false geography). Equal vertical spacing on unequal real distances distorts every slope, misrepresents speed, and puts crossings at wrong locations. Slope is meaningful only when vertical position is proportional to true distance.
2. Dark grid destroying data. A full black grid over the lines vibrates (moiré / 1+1=3) at intersections and buries individual trains. Fix: light gray grid, or the reverse side of graph paper (VDQI).
3. Ignoring single-track constraints. Crossings drawn between stations on single track are physical scheduling errors — invisible in a numeric table, immediately exposed graphically. Use the graphical form as the planning tool.
4. Collapsing space and time onto one axis. Listing stations as table rows and times as columns destroys both proportional spacing and slope-as-speed — that is the conventional timetable, with all its defects.
5. Truncating time at midnight. A 6am–midnight cut conceals overnight service and creates an artificial discontinuity. Wrap (cylinder) or prolong the grid to show the full cycle.
6. Graphical form for the public without spatial anchoring. Ybry's hope for public use went unrealized: the form is native to engineers, alien to passengers. Anchor it — the Hoboken-NYC bus schedule grounds the graphical strip on an aerial photo at house resolution so riders locate their own street first.
7. Omitting dwell time. Horizontal segments at stations are load-bearing. Treating arrival and departure as one point falsifies the schedule and erases which stops are operationally expensive.
8. Bumps-chart logic without a passing constraint. The bumps chart works because a narrow river physically prevents mass side-by-side racing (no more than two crews abreast; on bends only one). Apply the ranked-crossing form to a contest that allows simultaneous racing and you lose the constraint that gives it meaning.
9. Serpentined data (EI p. 106). When data collide with a rigid frame, they serpentine. The named family: the New Haven table's folded time sequence forcing the eye onto a snaking path; rivers meandering "boustrophedonically" around a tight frame, weakening comparison of their lengths; and the very bent bar chart in the 1935 Graphic Illustration of Tokyo City. Tufte allows that authentic graphical treasures can result, especially for lognormally distributed data — but as schedule design it is a failure: the frame, not the data, is dictating the path. Fix the frame, not the data.

§11. Implementation Formulas

Speed from line angle

speed = (distance between two stations) / (time to travel between them)
      = Δy / Δx        [in the axes' own units]
slope (rise/run) = speed     →    steeper = faster

Single-track feasibility check — for each pair of opposite-direction trains on a shared single-track segment:

1. find x (time) where their lines would cross
2. find y (location) of that crossing
3. if y falls between station bands → CONFLICT
   → shift departure times or add a siding at y

Pixel-compression (esophageal / any one-spatial-axis video)

n frames, each H pixels tall → n columns, each 1 px wide, assembled left→right
column[i] = Σ over rows j of pixel intensity in frame[i]   (sum each frame to one column)
result: (n px wide) × (H px tall) condensed dynamic image
horizontal axis = time, vertical axis = the single spatial dimension
worked values: 8 shown / 60 total frames, 0.2 s each, 64×64 source, → 12 s span

Rush-hour density signal — no formula. When lines are too dense to trace individually, let density itself say "service so frequent no schedule is needed." Do not annotate; do not color-code individual lines.

§12. When Space-Time Graphics Beat the Alternatives

Schedules rank among the most widely reproduced information displays — comparable in printed volume to road maps, weather charts, and telephone books — and 150 years of worldwide design effort have produced a rich range of strategies (EI p. 101). The graphical timetable is the form that turns that fussy numeric array into a legible multivariate narrative.

The live timetable — Seibu Railways (EI p. 45 n. 6): on a small Japanese line serving baseball spectators, railway workers monitor the game on television and dynamically adjust the train graph — irregular game length and early runaway-crowd departures feed straight into the graphical schedule (Hideo Ohki, Japanese Railway Engineering, 1979). Tufte's point: the logic is so visual that the graph entirely replaces entabled times — the graphical timetable as a live re-planning instrument, not a printed artifact.

§12b. Dance Notation — The Hardest Space-Time Case (EI pp. 114–119)

Roughly a third of the chapter treats dance notation: systems that must translate live human movement into signs transcribed onto flatland, permanently preserving the visual instant. Because the problem is so hard, notation deploys nearly every technique in the book at once — small multiples, close text-figure integration, parallel sequences, details with panorama, a polyphony of layering and separation, data compression into content-focused dimensions, and avoidance of redundancy (p. 114). Some systems even carry a visual elegance of their own, independent of the motions described.

Small multiples are the default (p. 115). Notation texts inevitably begin taxonomically, laying out fundamental movements as a visual dictionary of dance elements. Since the analytical task is unswervingly comparative, the design of choice is the small multiple; variations, sequences, and combinations follow.

Repetition beats page-flipping (p. 116). Dufort's 1728 Trattato del Ballo Nobile repeats identical small-multiple step diagrams on every double-page spread, so coordinating text with figure never requires turning a page — Descartes did the same in his Principia, repeating one diagram 11 times. The counter-lesson sits on the same spreads: separating text from graphic, even on one page, forces encoding to link the parts — Dufort's readers must jump back and forth via 22 alphabetical letters that impose a sequence on the illustrations but are otherwise arbitrary codes, saying nothing about the movements they mark.

Zorn's Cachucha: grid-prison → implicit grid (p. 117). Zorn's score for the ballerina Fanny Elssler runs stick-figure notation beneath the castanet music, but heavy gridlines disrupt the free play of dance and music — active stripes visually disproportionate to the tiny sum of content conveyed. The revision makes the grid implicit or nearly transparent and adds two devices: per-dancer floor plans showing each dancer's two-space path, and parallel motion profiles running with the music. Read down a column for the simultaneous actions of four dancers; read across a row for sequence. De-gridding restores the depiction of continuous movement, avoiding the arrested, disembodied quality of some notation compilations — and the representations can shrink to tiny stick-figures while still preserving fine detail.

Rameau's swirling words (p. 118). The drawings from Rameau's 1725 Dancing Master bend words along the path of the movement itself — an uncommon demonstration of the informational, rather than merely ornamental, use of swirling word-image integration.

The symbolic dead end (pp. 118–119). Modern systems make symbolic abstraction the prevailing doctrine, assembling meticulous codes for thousands of movements into huge dictionaries (Labanotation and its kin). Lincoln Kirstein's choreographic verdict on the whole lineage: every such system is logically conceived and invitingly rendered, yet for reconstructing old dances with any objective authority, practically worthless — systems so difficult to decipher that students feel triumphant deciphering even a single short solo. The practitioner counterweight (EI p. 27 n. 18): Ann Hutchinson Guest's Dance Notation makes the surprising demonstration that abstract symbolic notation is preferable to film and stick-figure portrayals, at least from a dancer's viewpoint — the verdict on notation depends on who is reading it.

The chapter's closing thesis (p. 119). That informed frustration is the essential dilemma of all narrative design: reducing the magnificent four-dimensional reality of time and three-space into little marks on paper flatlands. Tufte's forecast — high-resolution, slightly abstracted, dynamic and animated computer visualization as a partial lightening of the encoding burden — is the chapter's capstone, and the outer boundary of what any flat space-time graphic can do.

§13. Cross-Reference: The Salyut-6 Cyclogram (the bridge into Visual Explanations)

The space-time grid is not confined to EI. Inside Visual Explanations the key instance is the Soviet Salyut-6 cyclogram (VE pp. 92–95): a time-by-time grid — orbit-minutes on the vertical axis against trip-days on the horizontal — which VE says resembles a graphical timetable because time in orbit is interchangeable with distance (fn. 12, p. 93). The cosmonauts' repeating daily activity cycle stacks across the duration of the flight.

• Why it matters here: VE's footnote 12 (p. 93) explicitly sends the reader back to EI pp. 32, 45, 97–113 for the timetable / space-by-time material — VE points out to EI rather than developing it. The cyclogram is the natural hinge between the two books.
• VE's own space-time-equivalence anchors: "on paper, time and space are as one" (VE p. 57); and for still-land images that depict movement, space replaces time as the sequencing dimension (VE p. 109).
• Grid weight has a VE citation too: VE p. 74 demonstrates grid-calming on a graphical train timetable among its examples — so the §5 lesson can cite VE p. 74 alongside EI p. 59 and VDQI pp. 115–116.
• Use: when working from VE, treat the cyclogram as the worked space-time example and reach for EI's "Narratives of Space and Time" for the full design vocabulary (slope = speed, crossings, dwell, density, grid weight, single-track planning) laid out above.

Sources & scope

• Primary: Tufte, Envisioning Information, "Narratives of Space and Time," pp. 97–119 — Galilean satellites and longitude clock (pp. 97–101), Ybry/graphical timetable (p. 108), Czech 1933 (p. 102), China 200-page index (p. 103), Metro-North critique (pp. 104–105: 21% / 41 in / 24 AM-PM / 410 chars / dot-not-colon / Vicksburg courthouse), serpentined data (p. 106), Playfair torus + cylinder/midnight gluing + recycled globes (p. 107), Hoboken-NYC bus schedule (p. 108: ten-minute gray grid, house-resolution aerial photo), Chaux-de-Fonds 1932 single track (p. 110), Japanese beetle (p. 110), esophageal timetable and Oxford Torpids bumps chart (p. 111), strip maps (pp. 112–113), Wagner operas (p. 111), dance notation (pp. 114–119).
• Grid weight: EI p. 59 (look-up-table rule; reverse-side graph-paper trick); VDQI pp. 115–116 (dark→thinned→gray progression); VE p. 74 (grid-calming on a timetable).
• Bridge: Tufte, Visual Explanations, pp. 92–95 (Salyut-6 cyclogram) and footnote 12, p. 93 (redirect to EI pp. 32, 45, 97–113).
• Cross-book: VDQI p. 31 — Marey's schedule, credited by Marey to the engineer "Ibry," plus the TGV comparison (under 3 h vs. 9 h); VDQI pp. 176–177 — Minard's other carte figurative, the elephant campaign of Hannibal, drawn in the same flow-map grammar as the Napoleon march and worth citing as evidence that Minard had a reusable method for space-time narrative rather than one inspired one-off; Seeing With Fresh Eyes pp. 31, 118 — the measles/Project Tycho quilt: 88 million cases located in space and time 1888–2011, each state its own control, against ~103 million U.S. cases prevented since 1924.