Aesthetic-usability effect

The founding result is a correlation between two rating scales, not a usability measurement. Kurosu and Kashimura had people rate 26 ATM layouts — same components, different arrangements — for how easy each looked and how beautiful it was: apparent usability correlated .589 with aesthetics, while the seven layout variables the designers held to determine inherent usability, such as key grouping and the position of the confirm and cancel keys, barely mattered, keypad type aside. Tractinsky replicated it three times in Hebrew, expecting a weaker correlation in a culture he describes as action-oriented rather than aesthetic, and got .832 to .921; the lower figure comes from the version that rated beauty and ease in separate randomised rounds to rule out response dependency. Two caveats travel with those numbers and are usually dropped: each is computed across the 26 designs on ratings averaged over participants, so none is the correlation between one person's two judgements, and nobody involved had used an ATM. Tractinsky himself wrote that it was “still premature to rule out the possibility of method variance as a major cause of the obtained results.” What the founding studies license is a claim about static screens: a layout that looks beautiful gets called easy before anyone touches it.

The study that put people in front of a working system is also the only one that found beauty masking a defect. Tractinsky, Katz and Ikar had 132 students run 11 tasks on a simulated ATM, with nine of the original layouts at three levels of rated beauty crossed with two levels of real usability. Perceived aesthetics and perceived ease of use correlated .66 before use and .71 after it. The usability manipulation worked on behaviour — 23.5 seconds per task in the high-usability condition against 37.5 in the low one — and never reached the ratings: post-use perceived usability moved with the aesthetics factor and not the usability factor, and in the high-aesthetics cells ratings were indifferent to whether the system had been crippled. Aesthetics did nothing for behaviour in return: no effect on completion time, none on the 1.26 errors participants averaged. The hole in it is that 9 of the 14 seconds separating the usability conditions were deliberate system delays, so much of what the beautiful interface absorbed was waiting rather than confusion.

Two later experiments that manipulated both variables found the arrow pointing the other way. Tuch, Roth, Hornbæk, Opwis and Bargas-Avila built four versions of an online shop and found no effect of aesthetics on perceived usability, while degraded usability lowered post-use ratings of beauty, with frustration as the proposed mediator. Hamborg, Hülsmann and Kaspar repeated that design on a mobile phone prototype with 88 participants: aesthetics did not affect perceived usability “either directly or indirectly,” and usability again moved perceived beauty. Hassenzahl's studies on MP3-player skins suggest why the literatures disagree: beauty rested largely on identification and stayed stable across use, while perceived usability became a strong determinant of overall goodness once people had used the skins. Which judgement leads depends on whether the interaction gave the user anything to be annoyed about, so a flow with a visible failure point has its looks marked down rather than its failure covered up.

The halo also has a shelf life. Sonderegger, Zbinden, Uebelbacher and Sauer followed 60 mobile-phone users across three sessions of a two-week field experiment, and the boost an attractive product gave to perceived usability “began to wane with increasing exposure time.” Their own recommendation is multiple-session usability tests, to keep that boost out of the ratings. The effect is largest in a single first session, which is the condition most design research is run under.

Whether beauty improves actual performance is a separate and weaker claim. Sonderegger and Sauer gave 60 adolescents two functionally identical simulated phones differing only in appearance: the appealing one was rated more usable, 6.13 against 5.19 on the PSSUQ, and was also operated better, 147.7 seconds per task against 198.5 and 2.0 errors per trial against 4.8 — though the authors concede the unattractive prototype may simply have looked unfamiliar. Moshagen, Musch and Göritz ran 257 volunteers through search tasks on four versions of a health website and found an interaction rather than a main effect: high aesthetics sped up completion only where usability was poor. Thielsch, Haines and Flacke gave 331 participants an aesthetic or unaesthetic version of one site (VisAWI 4.46 against 3.13, g = 1.27) and found no effect on accuracy or response time in search, creative or transfer tasks — while the usability ratings, which they never manipulated, still came out higher on the attractive version (5.25 against 4.59, g = 0.53). Schlamann, Nestler and Thielsch's preregistered meta-analysis of 31 studies and 18,794 participants pools the performance effect at g = 0.29, with high unexplained heterogeneity, interaction device and typography manipulation as the moderators, and a sign that studies defining their aesthetic conditions after the fact were confounded by prior user experience.

What the surface reliably does is get judged fast, and carry credibility while it does. Lindgaard, Fernandes, Dudek and Brown showed homepages for 500 ms and then for 50 ms, and appeal ratings were highly correlated across the two exposures: half a glance is enough. Reinecke and colleagues had 548 volunteers rate 450 websites, and their models of perceived colourfulness and visual complexity, plus age and education, account for about half the variance in appeal after a 500 ms viewing. Fogg and colleagues had 2,684 people comment on the credibility of live sites and found “design look” raised in 46.1% of comments, ahead of information structure and information focus. Beauty is the faster of the two judgements too: in Tractinsky's timings it took 7.58 seconds against 8.68 for ease of use. So the defensible version is narrow. Visual quality decides the first impression and inflates what people say about ease of use at first contact; that inflation shrinks with exposure, vanishes in some well-powered experiments, and has never shown up in completion times or error rates. Polish earns you the first session; the ratings it produces are not evidence that the flow works.