Chartjunk

Overview

Chartjunk is any graphical element that does not communicate data — it burns ink, area, and viewer attention while adding nothing to meaning. Tufte sorts it into three kinds: vibrations (optical interference from busy patterns), grids (reference lines that overpower the data they support), and ducks (decoration that becomes, or buries, the data). The governing law is that a graphic succeeds or fails on its content, gracefully displayed: ornament can amplify a dull graphic's failure but has never rescued a thin data set from its own emptiness. The practical test is therefore subtractive — strip every mark that does not encode data or serve as a minimal navigational aid, then see what is lost.

§1. Vibrations — Optical Interference from Busy Patterns

A vibration is a perceptual artifact: dense, regular, high-contrast patterns applied to a graphic make the eye see shimmer, motion, or depth that is not in the data. The dominant source is moiré — two or more fine regular patterns (crosshatching, diagonal hatching, dot screens) overlaid or printed close together produce an unstable, eye-straining flicker. Moiré is unintentional optical art: the same effect 1960s Op artists pursued on purpose, smuggled into statistical graphics by default fill patterns.

Mechanism

The visual system reads closely spaced alternating lines as movement or relief — a low-level artifact, not information. Adjacent active lines also imply forms that were never drawn (the perceptual "1 + 1 = 3 or more" effect Tufte names in Envisioning Information): the more active marks crowd together, the more spurious structure the eye invents. The result is a raised visual noise floor that contaminates the whole graphic and makes the underlying quantities harder to read.

Tufte's moiré surveys (pp. 109–112)

Two samples, one finding. Random samples (1980–82) from the ten most-cited scientific journals put moiré vibration at 2% of graphics in Biochemistry, rising to 17% in Science and 21% in the New England Journal of Medicine (p. 110). A parallel survey of graphics manuals and software documentation runs worse: from Brinton's 12% up through Spear's 46%, Tell-A-Graf's 53%, and SAS/GRAPH's 68% (p. 112). The takeaway is structural, not anecdotal: when the manual or the tool ships vibration as the default, vibration is produced at scale by people who never chose it — the software-documentation peak marks exactly that. Even a statistics journal institutionalized it: JASA's own style sheet displayed a figure "prepared in the proper form" that needed 131 line-strokes plus 15 digits to convey trivial information — vibrating, unevenly spaced, misaligned (p. 110). Tufte also notes vibration may well peak for equally spaced bars (p. 109) — precisely where default fills land.

The Bertin rebuttal

Jacques Bertin argued a designer could court controlled moiré — flirt with ambiguity without surrendering to it. Tufte rejects this for data graphics: moiré is an undisciplined ambiguity with an eye-straining, illusive quality that pollutes the whole image, and there are no good examples of statistical graphics that gain anything from it. For data, vibration is always cost without benefit.