Human perception and cognition for designers

Every interface is processed by a human perceptual and cognitive system with well-studied strengths and limits. Designers who know those limits can predict where people will look, what they will notice, what they will forget and how long decisions will take. This lesson surveys the parts of perception and cognition that matter most for interface design: vision and colour, the organization of what we see (Gestalt principles), visual search and attention, memory, decision time and reading. It stays at the level a designer needs; PSYC-355 (sensation and perception) and PSYC-258 (cognitive psychology) cover the science in depth, and ROB-430 applies attention and workload to robot operators.

A useful frame is the model human processor of Card, Moran and Newell (1983), which describes a person as three interacting processors (perceptual, cognitive and motor) with memories, each with approximate cycle times: around 100 milliseconds for the perceptual processor (with a range of roughly 50 to 200), around 70 milliseconds for a cognitive cycle and around 70 milliseconds for a motor movement. The model is a deliberate simplification, but it explains why two flashes closer than about a tenth of a second blur together, why animation needs at least ten or so frames per second to look continuous, and why responses within about 100 milliseconds feel instantaneous, a figure that also appears in classic guidance on response times alongside about 1 second (the user's flow of thought is uninterrupted) and about 10 seconds (the limit of attention on a waiting task).

Vision

The eye resolves fine detail only in the fovea, a small central region covering about 2 degrees of visual angle, roughly the width of a thumbnail at arm's length. Outside it, acuity falls off steeply, though peripheral vision remains sensitive to motion and changes in brightness. People therefore read and inspect by moving their eyes in rapid jumps called saccades, with fixations of about 200 to 300 milliseconds in between; vision is effectively suppressed during the saccades themselves. Two consequences for design follow. A message placed far from where the user is looking, such as an error at the top of a long form when the user is at the bottom, may never be seen. And motion in the periphery is very good at capturing attention, which is why blinking banners and animated advertisements are distracting.

Visual angle is the right way to think about size. A target that subtends 1 degree is easy to see; text needs to be large enough that its x-height subtends at least about 0.2 degrees for comfortable reading, and larger for people with low vision. Because phones are held closer than monitors, the same physical size subtends a larger angle on a phone, which is one reason platform guidelines specify sizes in device-independent units rather than pixels.

Colour vision depends on three types of cone cells, sensitive to long (L), medium (M) and short (S) wavelengths. About 8 per cent of men and 0.5 per cent of women of northern European ancestry have some form of colour vision deficiency, most commonly a reduced or absent response from M cones (deuteranomaly or deuteranopia) or L cones (protanomaly or protanopia); rates are lower in many other populations. The practical result is difficulty distinguishing reds, greens, oranges and browns of similar lightness. Blue–yellow deficiencies (tritan) are rare. Colour vision also weakens with age as the lens yellows, and contrast sensitivity declines.

The design rule that follows is simple and appears in WCAG 2.2 as success criterion 1.4.1 (use of colour): never use colour as the only means of conveying information. Pair it with text, icons, shape, position or pattern. RideLine's old app showed trip status with a green badge for on time and a red badge for late, with no text. The program below uses published simulation matrices to estimate how those two badges look to people with the two common dichromacies.

# Simulating dichromatic vision with the Machado, Oliveira and Fernandes (2009) matrices
# (severity 1.0). This is an approximation for design checks, not a clinical model.
def to_lin(c):                                   # sRGB 0-255 to linear light 0-1
    c = c / 255
    return c / 12.92 if c <= 0.04045 else ((c + 0.055) / 1.055) ** 2.4
def to_srgb(v):
    v = min(max(v, 0.0), 1.0)
    v = 12.92 * v if v <= 0.0031308 else 1.055 * v ** (1 / 2.4) - 0.055
    return round(v * 255)
M = {
    "protanopia":   [[0.152286, 1.052583, -0.204868], [0.114503, 0.786281, 0.099216], [-0.003882, -0.048116, 1.051998]],
    "deuteranopia": [[0.367322, 0.860646, -0.227968], [0.280085, 0.672501, 0.047413], [-0.011820, 0.042940, 0.968881]],
}
def simulate(hexcode, kind):
    rgb = [int(hexcode[i:i + 2], 16) for i in (1, 3, 5)]
    lin = [to_lin(c) for c in rgb]
    out = [sum(M[kind][r][k] * lin[k] for k in range(3)) for r in range(3)]
    return "#" + "".join(f"{to_srgb(v):02x}" for v in out)
def dist(h1, h2):                                # plain RGB distance, enough to flag near-identical pairs
    a = [int(h1[i:i + 2], 16) for i in (1, 3, 5)]; b = [int(h2[i:i + 2], 16) for i in (1, 3, 5)]
    return sum((x - y) ** 2 for x, y in zip(a, b)) ** 0.5
on_time, late = "#2e9e44", "#d23b2f"             # RideLine's old status badges: green and red
for kind in M:
    s1, s2 = simulate(on_time, kind), simulate(late, kind)
    print(f"{kind:13} on time {on_time} -> {s1}, late {late} -> {s2}, distance {dist(on_time, late):.0f} -> {dist(s1, s2):.0f}")

protanopia    on time #2e9e44 -> #a0903b, late #d23b2f -> #685e2c, distance 193 -> 77
deuteranopia  on time #2e9e44 -> #93864b, late #d23b2f -> #8d7f29, distance 193 -> 35

For simulated protanopia the distance between the two badges falls to about 40 per cent of its original value, and for deuteranopia to under a fifth: the badges become two similar olive or brownish tones, differing mainly in lightness under protanopia and barely at all under deuteranopia. The fix is not to pick different hues alone but to add redundancy: the redesigned RideLine badge reads 'On time' or 'Late 6 min', uses a check mark or a clock icon, and differs in lightness as well as hue. Lightness contrast matters for everyone and is the basis of the contrast ratio calculation in lesson 10.

Organizing what we see: Gestalt principles

Early twentieth-century Gestalt psychologists, including Max Wertheimer, Wolfgang Köhler and Kurt Koffka, described how the visual system groups elements into wholes. Several of their principles are everyday tools of interface layout. Proximity: elements close together are seen as a group, so a label should be closer to its own field than to the next one. Similarity: elements that share colour, shape or size are seen as related, so all clickable links should look alike and unlike plain text. Common region: elements inside a shared boundary, such as a card or panel, form a group. Continuity: the eye follows smooth lines and curves, which helps with alignment. Closure: we complete incomplete shapes, which is why simple, partly drawn icons still read clearly. Figure–ground: we separate an object from its background, which modal dialog boxes exploit by dimming the page behind them. Common fate: elements that move together are grouped, as when selected items move as one.

Gestalt principles often explain layout problems that users cannot articulate. If the label 'Destination' sits equidistant between two text fields, users hesitate over which field it labels. If a 'Delete' button sits inside the same card as the save controls, it is grouped with them and pressed by mistake. Grouping by proximity and region is free, needs no extra words, and works for sighted users at a glance; the same structure must also be exposed in the markup (headings, lists, field labels) so that screen reader users get the grouping too.

Visual search and pre-attentive features

Some visual features are processed in parallel across the whole visual field, before focused attention: a single red item among grey ones, a tilted line among upright ones, a large item among small ones, a moving item among still ones. Searching for such a pop-out target takes about the same time no matter how many distractors there are. Searching for a conjunction of features, such as a red circle among red squares and blue circles, requires serial inspection, and time grows roughly linearly with the number of items. Anne Treisman's feature integration theory (1980) explains this difference. Designers use pre-attentive features to make the one thing that matters, such as a service disruption or a validation error, stand out; using them for many things at once destroys the effect.

Eye-tracking studies of web pages have found common scanning patterns, such as the F-shaped pattern in which users read the first lines of text, then a shorter line further down, then scan the left edge. Such patterns describe how people skim text-heavy pages when they have little reason to read, not a law to design for. Good information design (front-loaded headings, meaningful link text, short paragraphs) reduces the need to scan at all.

Attention

Attention is limited and selective. Selective attention lets people focus on one source while filtering others, as in following one conversation at a noisy party. Divided attention, doing two tasks at once, works only when the tasks use different resources and are well practised; Christopher Wickens's multiple resource theory describes how visual and auditory channels, spatial and verbal codes, and perception and response compete or coexist, and is the basis of guidance on in-vehicle interfaces (lesson 11). Sustained attention, or vigilance, declines over time on monotonous monitoring tasks.

Two striking phenomena show how much people miss. In change blindness, people fail to notice large changes in a scene when the change coincides with a brief disruption, such as a page reload or a saccade; a total that silently updates while the user looks elsewhere may go unnoticed. In inattentional blindness, people focused on one task fail to see an unexpected object in plain view, as in the famous experiment by Daniel Simons and Christopher Chabris (1999) in which many observers counting basketball passes missed a person in a gorilla suit. Designs that rely on users noticing a subtle change, an unobtrusive toast message or a banner far from their focus are fragile. Important state changes should be placed where attention already is, signalled through more than one channel, and, for screen reader users, announced through live regions.

Memory

Working memory holds the small amount of information a person is actively using. George Miller's 1956 paper famously suggested 'seven, plus or minus two' items, but later work, notably by Nelson Cowan (2001), puts the capacity at about four chunks when rehearsal and grouping are controlled. A chunk is a meaningful unit: the sequence 1 9 6 8 is four digits to one person and a single chunk, a year, to another. Working memory contents also decay within seconds without rehearsal and are easily displaced by interruptions.

Long-term memory is vast but retrieval is the bottleneck, and recognition is much easier than recall. This is the basis of one of Nielsen's heuristics (lesson 4), recognition rather than recall: menus, autocomplete, recently used lists and visible options let users recognize what they want instead of remembering commands. Command lines demand recall and reward experts; graphical interfaces trade some speed for recognition. Designs should never require users to carry information from one screen to another in their heads, such as a confirmation code shown on one page and needed on the next; WCAG 2.2 success criterion 3.3.7 (redundant entry) similarly requires that information already entered be auto-filled or selectable rather than re-typed.

Mental effort is often discussed as cognitive load. John Sweller's cognitive load theory distinguishes intrinsic load (the inherent complexity of the task), extraneous load (load imposed by poor presentation) and germane load (effort that builds understanding). Interface design mostly aims to cut extraneous load: removing irrelevant choices, keeping related information together, using consistent layouts and defaults, and showing the user's current position and options.

Decision time: the Hick–Hyman law

How long does it take to choose among options? William Hick (1952) and Ray Hyman (1953) found that choice reaction time increases with the information content of the choice. For n equally likely alternatives, the Hick–Hyman law is

where a and b are fitted constants and the + 1 accounts for the possibility of not responding. For unequal probabilities, the logarithm is replaced by the information (entropy) of the choice, so frequently chosen options are found faster. The program below fits the law to invented, seeded reaction time data and shows the effect of skewed usage.

# Hick-Hyman law: choice reaction time grows with the information in the choice,
# T = a + b * log2(n + 1) for n equally likely options.
# Invented, seeded data stand in for a lab study; we recover a and b by least squares.
import math, random
random.seed(421)
true_a, true_b = 0.20, 0.155                  # seconds and seconds per bit (illustrative)
rows = []
for n in (1, 2, 4, 8, 16, 32):
    for trial in range(40):
        bits = math.log2(n + 1)
        rows.append((bits, true_a + true_b * bits + random.gauss(0, 0.06)))
xs, ys = [r[0] for r in rows], [r[1] for r in rows]
mx, my = sum(xs) / len(xs), sum(ys) / len(ys)
b = sum((x - mx) * (y - my) for x, y in rows) / sum((x - mx) ** 2 for x in xs)
a = my - b * mx
ss_res = sum((y - (a + b * x)) ** 2 for x, y in rows); ss_tot = sum((y - my) ** 2 for y in ys)
print(f"fitted: T = {a:.3f} + {b:.3f} * log2(n + 1)   (R^2 = {1 - ss_res / ss_tot:.2f})")
for n in (4, 8, 16, 64):
    print(f"  predicted time to choose among {n:2} equally likely items: {a + b * math.log2(n + 1):.2f} s")
# unequal probabilities: information is the entropy H = sum p * log2(1/p + 1) in Hyman's form
p = [0.6, 0.2, 0.1, 0.05, 0.05]               # one menu item is chosen most of the time
H = sum(pi * math.log2(1 / pi + 1) for pi in p)
print(f"five items, skewed use: effective bits {H:.2f} vs {math.log2(6):.2f} for five equal items")

fitted: T = 0.204 + 0.153 * log2(n + 1)   (R^2 = 0.93)
  predicted time to choose among  4 equally likely items: 0.56 s
  predicted time to choose among  8 equally likely items: 0.69 s
  predicted time to choose among 16 equally likely items: 0.83 s
  predicted time to choose among 64 equally likely items: 1.13 s
five items, skewed use: effective bits 2.15 vs 2.58 for five equal items

The fitted constants are close to the values used to generate the data, and the predictions show the law's most important property: doubling the number of options adds a constant increment, not a doubling, of decision time. Two cautions keep the law in its place. It applies to choices where the user knows what they want and must find and select it among familiar options; it does not apply to unfamiliar menus that must be read item by item, where search is closer to linear, nor to genuinely hard decisions such as choosing a fare product without understanding the zones. And it says nothing about errors: more options can also mean more wrong choices. Designers sometimes cite Hick's law to argue for fewer options, but the more defensible lesson is to organize options so that users can narrow them quickly (grouping, ordering by frequency, progressive disclosure) and to put the most common choice first.

Reading

Reading on screens benefits from the same typographic practices as print: adequate size, high contrast, comfortable line length (often cited as about 45 to 75 characters for body text), generous line spacing and left-aligned text in left-to-right languages, since justified text produces uneven word spacing. People read words largely by recognizing their shapes and letter combinations within a fixation; all-capital text is slower to read in long passages because it removes the variation of ascenders and descenders. Readability formulas such as Flesch–Kincaid are crude but can flag dense prose, and plain language standards, such as ISO 24495-1 (2023), give organizations a tested process for writing clear content. Lesson 7 returns to typography as part of visual design.

Worked examples