5D Library › Computational Aesthetics, Visual Saliency Maps and Cognitive Ergonomics in User Interface Design · 中文

Chapter 1 Two Popular Rules: the Golden Ratio and Three Clicks

Two rules turn up in almost every introduction to interface design. One says that dividing a page by the golden ratio φ=(1+5)/2≈1.618\varphi = (1+\sqrt5)/2 \approx 1.618 makes users more comfortable and conversion better; the other says that every piece of content must be reachable within three clicks, or users will leave. Both rules are simple, memorable and directly actionable, which is why they are popular. This chapter checks the sources and evidence for each, and then sets out the limits that can be defended mathematically.

The golden ratio: from Fechner to web pages

The history of the experiments

The tradition of testing “the golden rectangle is the most beautiful” by experiment began with Fechner. In his Vorschule der Aesthetik of 1876 he reported that when people chose the most pleasing of ten rectangles of equal area and different proportions, the one closest to the golden ratio was chosen most often (Fechner 1876). But Fechner himself did not conclude that only the golden ratio was beautiful: he thought a range of proportions was preferred, and he considered Zeising’s claims for its aesthetic value overstated (as later writers report from the first volume of his book).

More than a century of later experiments does not give a uniform picture. Green’s 1995 review in Perception, titled “All that glitters”, surveyed the field (Green 1995). McManus (1980), using paired comparisons, in which only two rectangles are compared at a time and which is regarded as a less biased method, found that the group preference for the golden ratio was weak, while individual preference functions were strong, stable and very different from one another (McManus 1980) (as reported in the introduction of a 2016 paper by Spehar and colleagues (Spehar et al. 2016)). Hekkert and colleagues (1994) found that merely changing the instructions (“which do you personally like” versus “which proportion is good”) and whether participants had art training could move the peak of the average preference between the golden ratio and the square (Hekkert et al. 1994). The mathematician Markowsky (1992) systematically criticised claims that the golden ratio is “everywhere”, including its “discovery” in architecture, painting and the human body (Markowsky 1992).

Evidence

Evidence on preference for the golden ratio (summary). Fechner 1876: among ten rectangles of equal area, the one nearest the golden ratio was most popular, but Fechner himself said only that preferences clustered in a range (Fechner 1876). McManus 1980: paired comparisons; weak group preference, stable but scattered individual preferences (McManus 1980). Hekkert and colleagues 1994: instructions and art training shift the peak of the average preference (untrained participants under subjective instructions near the golden ratio, art-trained participants near the square; under “objective” instructions both groups lean towards the square), with large individual differences (Hekkert et al. 1994). Green 1995, review: mixed evidence, conclusions depend on method (Green 1995). This round of checking found no peer-reviewed study showing that dividing a web page by 1.618 raises conversion, click-through or sales.

So the “universal myth” that the framework sets out to refute does not stand up to the evidence: even for simple rectangles the golden ratio is not a robust peak of preference, and every step from rectangles to web pages and from web pages to conversion lacks evidence.

Why the golden ratio can be “found everywhere”

If the evidence is weak, why are so many successful web pages, logos and products said to “contain” the golden ratio? One reason is that in any moderately complex layout, finding some ratio close to 1.618 by chance is almost certain.

The program below does three things. Part A lists common screen and paper ratios and their distance from φ\varphi; Part B generates random layouts of kk boxes (widths and heights uniform between 20 and 400 pixels) and computes the probability that at least one ratio between two lengths lies within φ±1%\varphi \pm 1\,\%; Part C simulates a designer who runs one A/B test on each of 20 layout variants when every true effect is zero.

Calculation

# How easy is it to "find" the golden ratio phi = 1.618... in an interface?
# A: common aspect ratios and their distance from phi (a tolerance decides the verdict).
# B: a random layout of k boxes (widths and heights uniform on 20-400 px); probability
#    that at least one of the ratios between its 2k lengths lies within +-1 % of phi.
# C: 20 layout variants, each tested against a control in a null A/B test (no true
#    effect): probability that at least one shows p < 0.05, without and with Bonferroni.
import math, random, statistics

PHI = (1 + math.sqrt(5)) / 2
RATIOS = [("16:9 screen", 16 / 9), ("16:10 screen", 16 / 10), ("4:3 screen", 4 / 3),
          ("3:2 (Song timber cai)", 1.5), ("A4 paper, sqrt 2", math.sqrt(2)),
          ("Fibonacci 8:5", 8 / 5), ("tall phone 19.5:9", 19.5 / 9)]

def near_phi(rng, k, tol):
    L = [rng.uniform(20, 400) for _ in range(2 * k)]
    return any(abs(a / b / PHI - 1) <= tol for a in L for b in L if a > b)

if __name__ == "__main__":
    print(f"A. phi = {PHI:.6f}")
    for name, r in RATIOS:
        d = r / PHI - 1
        ok = "yes" if abs(d) <= .05 else "no"
        print(f"   {name:24s} {r:6.3f}  {d:+7.1%}  within 5 %: {ok}")
    rng = random.Random(1618)
    print("B. random layouts, 2,000 each: P(some ratio within +-1 % of phi)")
    for k in (2, 4, 8, 12):
        p = statistics.fmean(near_phi(rng, k, 0.01) for _ in range(2000))
        print(f"   {k:2d} boxes ({k * (2 * k - 1):3d} ratios): {p:.3f}")
    m, sims = 20, 20000
    hit = hit_b = 0
    for _ in range(sims):
        z = [abs(rng.gauss(0, 1)) for _ in range(m)]
        hit += max(z) > 1.959964; hit_b += max(z) > statistics.NormalDist().inv_cdf(
            1 - 0.025 / m)
    print(f"C. {m} null A/B tests: P(at least one p < 0.05) = {hit / sims:.3f}"
          f" (exact {1 - 0.95 ** m:.3f}); Bonferroni {hit_b / sims:.3f}")
# A. phi = 1.618034
#    16:9 screen               1.778    +9.9%  within 5 %: no
#    16:10 screen              1.600    -1.1%  within 5 %: yes
#    4:3 screen                1.333   -17.6%  within 5 %: no
#    3:2 (Song timber cai)     1.500    -7.3%  within 5 %: no
#    A4 paper, sqrt 2          1.414   -12.6%  within 5 %: no
#    Fibonacci 8:5             1.600    -1.1%  within 5 %: yes
#    tall phone 19.5:9         2.167   +33.9%  within 5 %: no
# B. random layouts, 2,000 each: P(some ratio within +-1 % of phi)
#     2 boxes (  6 ratios): 0.071
#     4 boxes ( 28 ratios): 0.316
#     8 boxes (120 ratios): 0.795
#    12 boxes (276 ratios): 0.967
# C. 20 null A/B tests: P(at least one p < 0.05) = 0.641 (exact 0.642); Bonferroni 0.049

In Part A, a 16:10 screen and the Fibonacci ratio 8:5 are only 1.1 % from φ\varphi, whereas the most common 16:9 screen is 9.9 % away, 4:3 is 17.6 % away and the 3:2 section of Song-dynasty timber (cai, Chapter 7) is 7.3 % away. Part B is the main point: with only 2 boxes (6 ratios) the chance that some ratio falls within φ±1%\varphi \pm 1\,\% is 0.071; with 4 boxes it is 0.316; with 8 boxes 0.795; with 12 boxes 0.967. An ordinary product page has dozens of rectangular elements, so “finding the golden ratio in this page” after the fact carries almost no information. Figure 1 (left) plots this curve; with a tolerance of ±3 %, 6 boxes already exceed 0.9.

Part C shows another source: if each of 20 layout variants with no true effect is tested once, the chance that at least one gives p<0.05p < 0.05 is 0.641 (exactly 1−0.9520=0.6421 – 0.95^{20} = 0.642); with a Bonferroni correction it falls to 0.049. A case report saying “after we switched to the golden ratio conversion rose significantly” is not evidence unless it says how many variants were tried and whether the test was registered in advance. Book 33 treated multiple comparisons and the garden of forking paths; this simply applies them to layout tests.

Note

The mathematical boundary. The conclusion of this section is not that “the golden ratio is wrong” or “the golden ratio is ugly”. φ\varphi is a number with many elegant mathematical properties (such as φ2=φ+1\varphi^2 = \varphi + 1, and a continued fraction made entirely of ones), and those properties all hold. What fails are two inferences: (1) that people generally prefer rectangles in the ratio φ\varphi; and (2) that such a preference would turn into a difference in behaviour on a web page. A designer may use φ\varphi as a choice of proportion, but cannot claim that it therefore raises conversion. Book 3 discussed a similar question: the mathematical specialness of a length or ratio is not a physical or psychological effect.

The three-click rule

Origin and evidence

The origin of “every page should be reachable within three clicks” is hard to trace; it looks more like oral lore that formed in the web-design trade in the late 1990s. Few studies test it directly, and the one cited most is an article by Joshua Porter published on the website of User Interface Engineering (UIE) in 2003 (Porter 2003).

Evidence

Porter 2003 (industry research, not peer-reviewed). 44 users, 620 tasks, more than 8,000 clicks. The number of clicks was unrelated to whether a task succeeded: users were no more likely to give up after three clicks than after twelve, successful and failed tasks had similar distributions of clicks, and hardly anybody gave up after three clicks. Satisfaction did not vary with the number of clicks either: across click-path lengths the share of dissatisfied users ranged only from 46 % to 61 %. The author concluded that “fewer clicks do not make more satisfied users”; users complain about clicks because they cannot find what they want (Porter 2003). A 2019 Nielsen Norman Group article titled “The 3-click rule for navigation is false” takes the same view (Laubheimer 2019).

Note that this is an article published by a consultancy; the selection of samples, tasks and sites was not peer-reviewed, and the conclusion should be treated as an industry observation, not rigorous experimental evidence.

The closest peer-reviewed literature concerns menu depth and breadth. Miller (1981) compared four hierarchies holding 64 target items; according to secondary reviews, selection time increased with depth, and some accounts report the two-level 8 × 8 structure as fastest and most accurate (Miller 1981) (the original could not be checked for this book); Kiger (1984) also found that response time and errors increased with depth, with the two-level 4 × 16 structure fastest and least error-prone, while users subjectively preferred shallower structures (Kiger 1984). Larson and Czerwinski (1998) repeated the question on the web and found that depth did have a cost, but that a medium structure balancing depth and breadth performed best, and that “information scent” (how well link labels point towards the target) mattered as much as structure (Larson and Czerwinski 1998). Pirolli and Card’s information-foraging theory (1999) gives scent a theoretical framework (Pirolli and Card 1999).

So the peer-reviewed evidence supports neither “the shallower the better” nor “three is the limit”. It supports this: depth has a cost, breadth also has a cost (more labels to read at each level), and whether labels lead people to the target often matters more than the number of levels.

A menu-tree model

The program below is the author’s model, structured after Lee and MacGregor’s 1985 model of menu search time (Lee and MacGregor 1985) and the menu-performance model of Cockburn, Gutwin and Greenberg (2007) (Cockburn et al. 2007), with illustrative parameters. 4,096 items form a tree of breadth bb and depth d=log⁡b4096d = \log_b 4096; each level costs a click and page load (0.8 s), a search or decision step and a Fitts pointing movement. Novices read labels one by one (on average (b+1)/2(b+1)/2 labels, 0.3 s each); experts decide by Hyman’s entropy (0.2+0.15log⁡2b0.2 + 0.15\log_2 b s for bb equally likely links, as in Chapter 2). At each level the right link is chosen with probability ss (information scent); a wrong choice wastes one level and a step back.

Part A computes expected times. Part B simulates 620 tasks (depth 2 to 6, scent 0.60 to 0.98, “patience” of 1 to 5 wrong choices) and looks at how success relates to design depth, scent and the clicks actually made.

Calculation

# The "three-click rule" against a menu-hierarchy model (the author's model; structure
# after Lee & MacGregor 1985 and Cockburn, Gutwin & Greenberg 2007; parameters are
# illustrative assumptions). N = 4,096 items in a tree of breadth b and depth
# d = log_b N. Each level costs a click/load overhead, a search or decision step and a
# Fitts pointing movement. Novices read labels serially (self-terminating, (b + 1)/2
# labels on average); experts decide in a + b log2(b) (Hyman's entropy for b equally
# likely links; Chapter 2). At each level the
# user picks the right link with probability s ("information scent"); a wrong pick
# costs one wasted level plus a back step.
import math, random, statistics

N, LOAD, BACK, READ = 4096, 0.8, 1.0, 0.30       # s per click, s per back, s per label
HA, HB, FA, FB = 0.20, 0.15, 0.10, 0.15          # decision and Fitts constants

def level_time(b, novice):
    search = (b + 1) / 2 * READ if novice else HA + HB * math.log2(b)
    return LOAD + search + FA + FB * math.log2(b / 2 + 1)

def expected_time(b, s, novice):
    d = round(math.log(N, b)); t = level_time(b, novice)
    return d, d * (t + (1 / s - 1) * (t + BACK))   # geometric number of wrong picks

def corr(x, y):                                     # Pearson r (Python 3.8+)
    mx, my = statistics.fmean(x), statistics.fmean(y)
    sxy = sum((a - mx) * (b - my) for a, b in zip(x, y))
    return sxy / math.sqrt(sum((a - mx) ** 2 for a in x) * sum((b - my) ** 2 for b in y))

def task(rng, d, s, patience, b=8):
    clicks = wrong = level = 0
    while level < d:
        clicks += 1
        if rng.random() < s:
            level += 1
        else:
            wrong += 1; clicks += 1                 # wasted click and a back click
            if wrong > patience:
                return clicks, False
    return clicks, True

if __name__ == "__main__":
    print(f"A. expected time to reach one of {N:,} items (seconds)")
    print("   breadth depth  novice s=.95  s=.80   expert s=.95  s=.80")
    for b in (2, 4, 8, 16, 64, 4096):
        row = [expected_time(b, s, nv)[1] for nv in (True, False) for s in (.95, .80)]
        d = expected_time(b, .95, True)[0]
        print(f"   {b:7d} {d:5d}  {row[0]:12.1f} {row[1]:6.1f}  {row[2]:13.1f}"
              f" {row[3]:6.1f}")
    for nv in (True, False):
        best = min((expected_time(b, .95, nv)[1], b) for b in range(2, 4097)
                   if abs(round(math.log(N, b)) - math.log(N, b)) < 1e-9)
        print(f"   fastest exact tree for {'novices' if nv else 'experts'}:"
              f" breadth {best[1]}, {best[0]:.1f} s")
    rng = random.Random(2003); rows = []
    for _ in range(620):                            # 620 simulated tasks
        d = rng.randint(2, 6); s = rng.uniform(0.6, 0.98); pat = rng.randint(1, 5)
        c, ok = task(rng, d, s, pat); rows.append((d, s, c, ok))
    ok = [float(r[3]) for r in rows]
    print("B. 620 simulated tasks (depth 2-6, scent 0.60-0.98, patience 1-5 wrong picks)")
    for lab, i in (("design depth", 0), ("scent s", 1), ("clicks made", 2)):
        print(f"   correlation of success with {lab:12s}:"
              f" {corr([r[i] for r in rows], ok):+.2f}")
    print(f"   success rate {statistics.fmean(ok):.2f}; median clicks, successful"
          f" {statistics.median(r[2] for r in rows if r[3]):.0f},"
          f" failed {statistics.median(r[2] for r in rows if not r[3]):.0f}")
    print(f"   tasks done within 3 clicks: {sum(r[3] and r[2] <= 3 for r in rows)}/620;"
          f" successes after more than 3 clicks: {sum(r[3] and r[2] > 3 for r in rows)}")
# A. expected time to reach one of 4,096 items (seconds)
#    breadth depth  novice s=.95  s=.80   expert s=.95  s=.80
#          2    12          19.6   25.5           18.3   24.0
#          4     6          12.2   15.7           10.7   13.8
#          8     4          11.2   14.0            8.2   10.5
#         16     3          12.6   15.5            7.0    8.9
#         64     2          24.1   29.0            5.9    7.4
#       4096     1         649.6  771.6            4.8    5.9
#    fastest exact tree for novices: breadth 8, 11.2 s
#    fastest exact tree for experts: breadth 4096, 4.8 s
# B. 620 simulated tasks (depth 2-6, scent 0.60-0.98, patience 1-5 wrong picks)
#    correlation of success with design depth: -0.17
#    correlation of success with scent s     : +0.34
#    correlation of success with clicks made : -0.38
#    success rate 0.87; median clicks, successful 5, failed 8
#    tasks done within 3 clicks: 132/620; successes after more than 3 clicks: 407

In Part A, the fastest complete tree for novices is breadth 8, depth 4 (11.2 s at s=0.95s = 0.95); three levels of breadth 16 take 12.6 s, two levels of breadth 64 take 24.1 s, and putting all 4,096 items on one page takes 649.6 s, because novices read one by one. For experts, a decision time growing only as log⁡2b\log_2 b makes wide, shallow structures ever faster, and one page holding everything takes only 4.8 s. The “best number of levels” therefore depends on whether users already know the label they are looking for: the same tree ranks in opposite orders for novices and experts. Lowering scent from 0.95 to 0.80 raises the novice’s time at breadth 8 from 11.2 s to 14.0 s, almost the cost of an extra level. Figure 1 (right) plots both groups of curves.

In Part B, the success rate over 620 simulated tasks is 0.87. Success correlates only −0.17 with design depth, +0.34 with scent and −0.38 with the number of clicks actually made. The last correlation is the easiest to misread: failed tasks involve more clicks (median 8, against 5 for successes), but because users who are lost click more wrong links, not because clicking causes failure. Only 132 tasks were completed within three clicks, while 407 succeeded after more than three. Unlike Porter’s data, failed tasks in this model involve more clicks; but both say that the number of clicks is not itself the cause of failure—in the model, many clicks are a consequence of being lost.

Left: probability that a random layout contains at least one length ratio near the golden ratio; it approaches 1 quickly as the number of boxes grows. Right: in the author’s menu-tree model, the expected time to retrieve one of 4,096 items against menu breadth; novices (reading labels one by one) and experts (logarithmic decision time) have different optimal breadths, and lowering information scent from 0.95 to 0.80 costs nearly one extra level (illustrative parameters).

Hypothesis · not established · Scent matters more than levels

The author’s hypothesis (not established): in website navigation with a fixed number of items, rewriting link labels to point more clearly to their targets (raising information scent) affects task success more than compressing the hierarchy from four levels to three. What result would show it wrong: in a pre-registered 2 × 2 factorial experiment (labels: original / rewritten; depth: three / four levels), the main effect of depth is larger than that of labels, with a confidence interval for the difference excluding zero.

Chapter summary

The golden ratio 1.618 is a weak, method-dependent peak in rectangle-preference experiments, and no peer-reviewed evidence shows it raising web conversion; in any moderately complex layout a ratio close to 1.618 is almost certain to be found by chance, and chance “significance” across many layout tests is common. The main evidence against the three-click rule comes from an industry study (Porter 2003); peer-reviewed menu studies show that depth and breadth each have costs and that information scent matters as much. The author’s menu-tree model shows that the best number of levels depends on whether users are novices or experts, and that the correlation between clicks and failure comes from being lost, not from clicking. Both rules should be rewritten: proportion is a design choice, not a conversion lever; the number of clicks is a symptom to monitor, not a design constraint.

References

Cockburn, Andy, Carl Gutwin, and Saul Greenberg. 2007. “A Predictive Model of Menu Performance.” Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’07), 627–36. https://doi.org/10.1145/1240624.1240723.
Fechner, Gustav Theodor. 1876. Vorschule Der Aesthetik. Breitkopf & Härtel.
Green, Christopher D. 1995. “All That Glitters: A Review of Psychological Research on the Aesthetics of the Golden Section.” Perception 24: 937–68. https://doi.org/10.1068/p240937.
Hekkert, Paul, C. (Lieke) E. Peper, and Piet C. W. Van Wieringen. 1994. “The Effect of Verbal Instruction and Artistic Background on the Aesthetic Judgment of Rectangles.” Empirical Studies of the Arts 12 (2): 185–203. https://doi.org/10.2190/RMEM-K5KL-R2RN-7A2Y.
Kiger, John I. 1984. “The Depth/Breadth Trade-off in the Design of Menu-Driven User Interfaces.” International Journal of Man-Machine Studies 20: 201–13.
Larson, Kevin, and Mary Czerwinski. 1998. “Web Page Design: Implications of Memory, Structure and Scent for Information Retrieval.” Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’98), 25–32.
Laubheimer, Page. 2019. The 3-Click Rule for Navigation Is False. Nielsen Norman Group article.
Lee, Eric, and James MacGregor. 1985. “Minimizing User Search Time in Menu Retrieval Systems.” Human Factors 27: 157–62.
Markowsky, George. 1992. “Misconceptions about the Golden Ratio.” The College Mathematics Journal 23 (1): 2–19.
McManus, I. C. 1980. “The Aesthetics of Simple Figures.” British Journal of Psychology 71 (4): 505–24.
Miller, Dwight P. 1981. “The Depth/Breadth Tradeoff in Hierarchical Computer Menus.” Proceedings of the Human Factors Society 25th Annual Meeting, 296–300.
Pirolli, Peter, and Stuart Card. 1999. “Information Foraging.” Psychological Review 106 (4): 643–75.
Porter, Joshua. 2003. Testing the Three-Click Rule. User Interface Engineering (UIE) article, 16 April 2003.
Spehar, Branka, Nicholas Walker, and Richard P. Taylor. 2016. “Taxonomy of Individual Variations in Aesthetic Responses to Fractal Patterns.” Frontiers in Human Neuroscience 10: 350. https://doi.org/10.3389/fnhum.2016.00350.

Fan, Y. (2026). Computational Aesthetics, Visual Saliency Maps and Cognitive Ergonomics in User Interface Design: From the Hick–Hyman Law and Spatial Frequency Channels to Saccadic Trajectory Modelling, Read Beside the Kaogongji and Classical Treatise on Architectural Methods [计算美学、视觉显著性图谱与人机界面认知工程] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23836258 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).