Picture superiority effect

The asymmetry is old and robust. Across Standing's four single-trial learning experiments on recognition, recall and choice reaction time, memory capacity followed a power law of the amount presented, pictures beat words throughout, and adding alternatives to the recognition test left the picture advantage intact. One result from the same paper is rarely quoted — verbal memory showed a higher retrieval speed than picture memory. Brady and colleagues put a ceiling on picture capacity rather than on the comparison: after viewing 2,500 objects over 5.5 hours, participants picked the seen image out of a pair at 92% against a foil from a novel category, 88% against another exemplar of the same category and 87% against the same object in a different state. That study had no word condition, so it shows how much detail an image holds, not that an image beats a label.

Hockley pushed the effect past single-item recognition. Students studied random pairs of concrete words or of line drawings, then separated intact pairs from rearranged ones; pictures won here too, entirely on a higher hit rate for intact picture pairs, with no difference in false alarms. The advantage persisted when the test pairs were the verbal labels of the pictures studied — the gain is made at encoding, so it is the drawing you show rather than how you later ask about it that does the work.

The standard explanation has since lost, and that changes what you draw. Dual coding attributed the advantage to a picture being spontaneously named as well as seen. Ensor, Surprenant and Neath instead varied physical distinctiveness: half the words in plain black type, half in varying fonts, sizes, colours and capitalisation; half the pictures in colour, half in black and white. Against distinctive words, black-and-white pictures lost their advantage entirely. Higdon and colleagues reproduced the pattern in associative recognition and free recall and concluded that dual coding is no longer viable. A screen of fifteen illustrations in one house style, one palette and one size has bought pictures and no distinctiveness.

It also reverses on the retrieval cue. Weldon, Roediger and Challis found words remembered better than pictures on word-fragment and word-stem completion — finishing -l-ph-nt or ele----- — under both implicit and explicit instructions, while pictures won when the cue was a semantically related word from outside the list, and words won again when both cue types were combined under implicit instructions. The cue decided the outcome, not the instruction. If what the user does later is type a word, scan a text label or spot a term in an email, the picture advantage is not in play, and verbal memory is the faster one — which is the case against an icon-only control.

In applied settings the effect is real and smaller than its reputation. Guo, Zhang, Landau Wright and McTigue meta-analysed 39 experiments and found a moderate effect of adding graphics to text on reading comprehension, Hedges's g = 0.39, holding across grade levels, with pictures, pictorial diagrams and flow diagrams not differing significantly from one another. Lindner tested the distinction interfaces usually ignore: across 404 students, a representational picture beside a science or maths item raised performance, perceived ease and test-taking pleasure, while a decorative one moved neither performance nor perceived ease and cut time on task in maths items. Houts and colleagues' health communication review names the conditions: pictures closely linked to the words beside them raise attention and recall, help most when they show relationships among ideas or spatial relationships, help low-literacy readers most, and want plain captions without distracting detail.

The field trials mark the limits. Noar and colleagues pooled 37 experiments, 48 samples, N = 33,613, on pictorial against text-only cigarette pack warnings: more attention (d = 0.79), far more cognitive elaboration (d = 1.70), stronger negative affect and higher intention to quit (both d = 0.54), more psychological reactance (d = −0.50) — and no better recall or recognition of the warning text at all (d = −0.03, not significant), no change in perceived likelihood of harm (d = 0.02) or self-efficacy (d = 0.01). Nine of the 17 pooled effects were heterogeneous, and only one study measured smoking behaviour. In Durand and colleagues' surgeon-randomised trial of breast cancer surgery conversation aids, the picture aid beat usual care on knowledge where the text-only aid did not, with more impact among women of lower socioeconomic status and health literacy — but the two aid arms did not differ significantly from each other. Scalia and colleagues randomised 407 people to an icon array or a bar graph for the same stroke-and-death risks and found no difference in understanding at any level of numeracy or graph literacy, though 34% preferred the icon array. Dowse and colleagues tested 10 pharmaceutical pictograms with 90 isiXhosa-speaking adults: mean comprehension 7.9 out of 10, 8 of the 10 meeting the ISO criterion of 66.7% correct, comprehension associated with health literacy (p = 0.002), age, education and English proficiency. A symbol is not self-explanatory; it is something your audience has to already read.

Memory for pictures is confident rather than accurate. Weinstein and Shanks showed participants object pictures, then a word list quietly including names of objects never shown; participants later claimed with high confidence and a feeling of recollection that they had seen pictures of those objects, while still telling studied pictures from new ones accurately. So the usable claim is narrow: a picture that carries the content, looks unlike the pictures around it, sits next to the words it explains and is retrieved by something visual is remembered better than the words alone. A decorative image on the same text buys attention and feeling, and in the one large meta-analysis that measured it the text was remembered no better at all.