Against

The evidence against this topic's claims, by theme: fonts and folk rules, line length, polarity, layout, the limits of the evidence, and practice that contradicts a principle.

Dyslexia fonts and folk rules

  • Dyslexia-specific fonts: null effects in Rello 2013, Wery 2017 and Kuster 2018 (n = 170).
  • Italic, underline and justified text are widely discouraged. Only italic has direct support (worse in Rello 2013). No source of 10 or more was read on justification; Akhmadeeva 2016 is below the quality bar: an example, not evidence.
  • Sans serif is recommended for dyslexia; Bernard 2002 finds no speed difference for typical readers.
  • The letter-spacing gains come from one research group (Zorzi 2012); replications by other groups were not reviewed here.

Line length

  • Speed and comprehension peak at different lengths (Dyson 2004), and preference follows neither.
  • The 66-character ideal is typographic craft tradition, not a measured optimum.
  • CSS ch is not the same as characters per line.

Polarity

  • Piepenbrock 2013 favours dark-on-light; Dobres 2017 finds no daytime difference. At night white on black was clearly less legible (negative polarity disadvantage), so the night benefit belongs to dark text on light.

Layout and design

  • Tufte's data-ink rule (Tufte 1983): intuition; Bateman 2010 found embellishment harmless for accuracy.
  • F-pattern: a description of scanning on cluttered pages, not a law for structured text.
  • Chunking advice rests on Miller 1956 and anecdote.
  • Information mapping, a structured-writing method behind much chunking advice, was tested by its author. The test found no significant difference from the same content written as prose (20 and 21 students, adjusted means 19.56 and 17.65, t = 1.17, 38 df; Horn 1969).
  • Syntax highlighting: no comprehension effect (Hannebauer 2018).
  • Collapsed sections: reduce clutter but hide content (Wang 2023).

Limits of the evidence base

  • Many studies use short passages, small samples or screens of the 2000s.
  • Practitioner sources (Nielsen Norman Group, Bringhurst, BDA) give guidance, not controlled trials.

Practice that contradicts a principle

  • Readable measure says 45 to 75 characters. USWDS 2026 allows up to 90, and WCAG 1.4.8 (AAA) offers 80. Neither cites a study.
  • Sufficient contrast states one ratio. APCA 2026 says the threshold depends on size and weight; W3C has not adopted it.
  • The layer-cake claims of Nielsen Norman 2019 (below the quality bar: an example, not evidence) rest on what the page shows: one 9-participant heatmap and no measured effect; the F-pattern is descriptive (Nielsen 2006).