This is a brief discussion on the need for a new contrast method to guide designers of web content.
The old WCAG 2.x method and 1.4.3 contrast is being replaced in WCAG 3.0, and as I am the author of the WCAG 3.0 contrast, and leading the ongoing research in this area, I wanted to discuss how the new contrast method fits into the conformance model in a non binary way.
First of all, WCAG 2.x contrast and 1.4.3 and the related understandings and guidelines are the single "most wrong" part of WCAG 2.x, and the part that was most in need of updates due to the massive related changes in web content and CSS functionality since 2008, when that standard was first introduced. I am on record with this since April 2019 in WCAG issue 695.
There are a number of reasons that 1.4.3 and related are faulty, one of which is the binary nature of the SC for a property that does not apply in a binary way across perception nor impairments. It is a prime example of why "binary" is a fail for conformance models for properties relating to human perception. Humans are not binary computers, and it is important to understand the non-linear and non-binary aspects of perception, and to set guidelines that correctly model perception as opposed to "brute forcing" arbitrary values that ultimately do more harm than good.
Let's begin with a general understanding of what color and contrast actually is outlined below. See more in my whitepaper "Introduction to Color and Accessibility" and I also invite you to see the more in-depth discussions of Visual Contrast conformance at the W3 Wiki for Visual Contrast
First, like color, contrast is not real. It is a perception. It is not a measure of the distance between two colors.
Like all perceptions it is context sensitive. It is also substantially affected by spatial frequency, and is primarily a function of the luminance channel, while hue/chroma are far distant in effect, particularly as related to readability. High luminance contrast is required for fluent readability such that the text's words and letter-pairs are properly directed to and filtered by the Visual Word Form Area (VFWA) of the brain, leading into lexical processing.
Above a certain amount, contrast constancy may come into effect wherein further increases in contrast as far as color distances does not have an effect on perception for readability. However, as spatial frequency increases (meaning smaller and thinner letters) perceived contrast drops rapidly, such that lightness distance must be increased to compensate for small thin fonts.
For fluent readability, Dr. Lovie-Kitchin demonstrated, and Dr. Legge also showed, that contrast must be at least ten times the contrast sensitivity threshold (CS) of just noticeable differences (JND), and twenty times is preferred for adequate contrast reserve above the critical contrast.
This defines the area of "suprathreshold critical contrast for readability." I refer to this as "readability contrast" for simplicity. A similar suprathreshold exists for acuity in terms of font size, which is separate and in addition to the font sizing issues related to contrast spatial frequency. Contributing factors are white space, tracker/leading (letter and line spacing) aspect ratio, weight, etc etc...
Spot reading contrast has a lower requirement than fluent readability contrast. Non-lexical object contrast also has a lower luminance contrast requirement. And there are some differences in terms of the importance of color (hue/chroma) for some categories of stimuli, and certainly for information coding.
No single figure such as 4.5:1 is even remotely useful as a blanket statement, and I have demonstrated that it can even be harmful. Some gists with colorful examples of this are:
And finally, the problems of 4.5:1 are not only in accessibility, but fails for normal vision, and further causes such problems for designers that it is often just ignored. 1.4.3 is in fact sometimes ignored even by those who claim to be A11y supporters.
APCA, the Advanced Perceptual Contrast Algorithm is a product of the larger SLUV Advanced Perceptual Color (SAPC) appearance model, which is specifically related to appearance on self illuminated RGB computer displays, and for modeling impairments.
The APCA generates a contrast value based on a color pair, but this value is perceptually based: that is, regardless of how light or dark the colors are, a contrast value of 60 represents the same perceived readability contrast. This is absolutely not the case with WCAG 2.x, which far overstates contrast for dark colors to the point that 4.5:1 can be functionally unreadable.
The APCA contrast value is perceptually uniform, and based on the point where the CS curve flattens due to contrast constancy. Halving or doubling the APCA value relates to a halving or doubling of the perceived contrast.
The APCA then uses a lookup table to associate font size and weight to the readability contrast. The lookup tables shown thus far are for latin fonts. Different languages may want to have their own modified lookup tables specific to their needs.
A principal point here is that a strict pass/fail is not instructive, is arbitrary, and not necessarily solving a given user need. In fact user needs when it comes to contrast are conflicting in that what is good for one can be harmful to another. This is even true of font size.
More than anything this points to the importance of real user personalization, an area where the technology is literally missing (and a work in progress). For the guidelines though we can set ranges for targets and expectations toward making the web readable for all. Developing personalization capabilities is ongoing.
Because of the wide and variable nature of user needs vs use cases in visual contrast, we've divided the scoring into ranges.
- Preferred: this is the stealth level 5 — it is not a 5 in terms of score, that would still be 4 — but it is a target to encourage designers to work toward when possible.
- Score 4 is the level for fluent readability best practices, and intended to accommodate sighted users including those with impairments that do not otherwise need assistive technology.
- Score 3 is a "safety catch" for well designed sites that in testing had a flaw that should not fail the site, but provide guidance for future improvement.
- Score 2 and 1 are "deficient" and "poor", intended to help sites that currently pass WCAG 2.x but have readability problems still "pass" while providing guidance for improvement.
- Score 0 is a fail, and fails the site. If a site scores 0 it was likely not passing WCAG 2.x either, though there may be edge cases there.
The method for predicting readability contrast is not binary, and the scoring of results is also not binary. It is range based, and dependent on multiple factors, yet is simple enough to be fully automated. So I don't think using the term "binary" is descriptive of this scoring methodology. It is clearly not "pass/fail" except at the extreme edge cases, and demonstrates the value of a range-based approach.
In reference to the example tool for demonstrations, the contrast figure is associated with live updates of minimum font size and weight for score level 4: https://www.myndex.com/APCA/simple
I hope this clarifies the useful differences of a range based or multi-level score from a simplistic pass/fail approach.
Thank you,
Andy
Andrew Somers
W3 AGWG Invited Expert
Myndex Color Science Researcher
Inventor, SAPC and APCA
NOTICE: Personal opinions expressed are mine and may or may not reflect those of the W3 or AGWG unless specifically stated so.