Selective perception
People given the same complex event pick out the parts that suit their allegiance or their job, and report back a different event from the one their neighbour reports.
People given the same complex event pick out the parts that suit their allegiance or their job, and report back a different event from the one their neighbour reports.
The name is a label over three literatures that share a premise and little else. The first is the New Look work of the 1940s and 1950s on perceptual defence, which measured how long a word had to be shown before a viewer could read it. The second is social perception, where partisans watch one event and describe two.
The third is organisational, where a manager reads a business case and names the problem belonging to his own department. Dearborn and Simon give the shared premise in one line: the more complex or ambiguous the stimulus, the more the perception is determined by what is already in the subject and the less by what is in the stimulus.
They cite Bruner's paper on perceptual readiness for it, at pages 132 and 133. Those pages carry Bruner's seven propositions about how accessible a category is, and the word ambiguous appears nowhere in his article. The sentence is theirs, not his. Nothing in the label fixes where the selection happens. It might be in what the eye goes to, in what gets stored, or in what a person says a week later, and the classic studies were not built to tell those apart.
Repeats disagree, or the effect holds in some conditions and not others. The other 248.
Mixed Rival fans still report different games, d = 0.83 on a judgement made ten days later, while the gaze and the recall of those same fans showed nothing.
Huff and colleagues invited the 58 keenest of 132 applicants, 33 Borussia Dortmund fans and 25 Bayern Munich fans, to watch the live broadcast in separate cubicles. Ten to twelve days later they rated how much each team had contributed in possession, tackles won and scoring chances. Bayern fans rated Bayern's contribution higher than Dortmund fans did, t(49) = 2.89, p = 0.006, d = 0.83.
The earlier stages gave nothing. Gaze coherence with one's own fan group minus coherence with the other group came to M = -0.03, SD = 0.27, t(14) = -0.46, p = 0.655, with a Bayes factor of 5.15 favouring the null. Event segmentation agreement gave t(33) = -1.49, p = 0.144, and a Bayes factor of 12.50 for the null.
Cued recall showed no effect of fandom immediately or at the delayed test, F(1,54) = 0.13, p = 0.725. Confidence in those same recall answers did move. Each group was surer of itself on the clips showing its own team with the ball, F(1,54) = 9.52, p = 0.003. The authors' own summary is that fandom biases subjective judgements and not basic perceptual and cognitive processes.
A second modern study finds a difference earlier than that. Working with 21 conservatives and 22 liberals in a scanner, de Bruin and colleagues measured how each viewer's brain broke political video into states. On a Vice-Presidential Debate clip about immigration, people of like ideology switched states together in the dorsomedial prefrontal cortex, beta = 0.119 with a standard error of 0.036, t(805.3) = 3.326, P = 0.002.
The same test gave nothing on a policing clip, P = 0.658, or on a neutrally worded video about abortion, P = 0.294. Boon and Davies had run a modernised version of the football design in 1996. Their abstract reports that perceptions were still distorted towards the preferred team, that participants did not know it, and that they justified the distortion afterwards.
Their article sits behind a bot challenge at the publisher rather than a paywall, and the abstract quoted here is the one in the Crossref record. The management strand went the other way. Walsh gave a 1,000-word case to 121 midcareer managers and measured their belief structures with a 50-card sort. His first hypothesis failed, in that work history did not predict what a belief structure contained.
His third failed too, in that information use did not vary by group. The second and fourth drew support he calls marginal and limited, from a marketing group that saw more external problems and an accounting group that asked for more accounting figures. A narrower test of Dearborn and Simon's own domains of human relations, accounting and marketing was not significant.
Managers named 7.12 problems each on average and used 25.09 pieces of case information. Beyer and five co-authors then ran a systematic replication of both studies. Their abstract reports that the two sets of conclusions differ because the procedures differ, and that managers perceive more broadly when encouraged to name more problems. It adds that belief structure did not mediate between functional experience and selective perception, against Walsh's account, and that functional experience can produce selective imperception as well as selective perception. Only that abstract was read, and it was read at OpenAlex.
The report three departments readillustration
One long account of a struggling firm goes to every manager. The one from sales names a sales problem and the one from accounts names a costs problem. Each picked out the part his own day is made of.
The match both sets of supporters watchedillustration
Two people watch the same game and come away with different counts of fouls by the other side. Neither invented an incident; the ambiguous ones were resolved in opposite directions and then remembered as seen.
The row overheard by two friendsillustration
Two friends hear the same argument between a couple they both know, and each returns with an account favouring the one they like better. Both kept accurate parts of it, and they kept different parts.
Cases marked as illustrations describe nobody in particular. They are there to make the pattern recognisable, not to report that a particular thing happened.
McGinnies ran the perceptual-defence experiment at the University of Alabama on eight male and eight female undergraduates. Eleven neutral and seven taboo words were flashed at .01 second, then at longer steps, until the viewer read each one aloud. Mean thresholds were higher for the taboo words by .045 second, t = 3.96 for 15 degrees of freedom, P < .01.
Skin-resistance readings taken before recognition differed by 1.98 microamperes, t = 5.10 for 15 degrees of freedom, P < .01. The guesses viewers made before recognising a word also fell differently across four categories, chi-square 31.26, P < .01. Hastorf and Cantril studied the Dartmouth against Princeton game of 23 November 1951 at Palmer Stadium. Princeton won, and the official statistics recorded 70 penalty yards against Dartmouth and 25 against Princeton.
A week later they gave a questionnaire to psychology students at both colleges, 163 at Dartmouth and 161 at Princeton. Of the Princeton students 93 percent called the game rough and dirty and none called it clean and fair. Among Dartmouth students 42 percent said rough and dirty and 13 percent clean and fair, while 39 percent wrote in a category of their own, rough and fair.
Asked who started the rough play, 86 percent of Princeton students blamed Dartmouth, against 36 percent of Dartmouth students. Then a film of the game was shown to two Dartmouth fraternities on 7 December and to two Princeton undergraduate clubs early in January. Watching it, 48 Dartmouth students marked a mean of 4.3 infractions by Dartmouth and 4.4 by Princeton.
The 49 Princeton students marked 9.8 by Dartmouth and 4.2 by Princeton. The paper reports one test, a footnote to the table saying the Dartmouth-team counts differ at the .01 level. Dearborn and Simon gave 23 middle managers from one manufacturing firm a 10,000-word teaching case about the Castengo Steel Company and asked each to write down the most important problem a new president should tackle first.
Five of the six sales executives named sales, against five of the other 17, which the authors report as 83 percent against 29 percent and significant at the 5 percent level. Four of the five production executives named organisation, 80 percent against 22 percent, again at the 5 percent level. Three of the four accounting executives named sales too, and the authors fold them in with sales to reach eight of ten against two of the remaining 13. No test statistic is printed for either comparison.
3 primary research9 replication and reanalysis3 commentary and critique
A five-page note that put the term on the phenomenon in the form management and popular writing still use. The demonstration everyone cites is four years older and does not use the term: Hastorf and Cantril's study of a football game appeared in 1954.
The claim, and the replication this entry reads its figures from.
Both famous studies are small and neither reports an effect size. Hastorf and Cantril tested the film on 48 and 49 students drawn from two fraternities and two undergraduate clubs, shown a month apart. Their own footnote warns that a film of a football game follows the ball and omits much of the action. Dearborn and Simon had 23 people from one company, and only 10 of the 23 named sales at all.
The perceptual-defence strand was contested within a year, and the whole exchange has now been read. Howes and Solomon of Harvard answered McGinnies in 1950 with two objections. Duration thresholds fall as the printed frequency of a word rises, correlating -0.79 across sixty words in their own data, and taboo words are rare in print. And a student may see the word penis and wait before saying it in front of his professor and a woman assistant.
McGinnies replied in the next six pages. Fitting a line within each set of words gave a slope near zero. The taboo words gave Y = -0.003X + 0.119 and the neutral ones Y = 0.001X + 0.061, against Y = -0.018X + 0.116 for the two sets pooled. He read the pooled slope as an artefact of joining two distributions whose means were set by meaning rather than by frequency.
He then went back to the laboratory against the withholding objection, and that repeat is the part nobody cites. With Sherman in 1952 he showed 20 male undergraduates a taboo word at full exposure, then a neutral five-letter word matched to it for frequency. Neutral words that followed a taboo word took more exposures, a mean difference of 9.85 exposures, t = 2.87, P < .01, and no taboo word ever had to be spoken.
Cable repeated that design in 1969 on 30 male students, adding nonsense paralogs as a third kind of prime. Neutral primes gave a mean of 5.50 exposures, taboo primes 6.67 and paralogs 8.33. Only the paralogs differed from the neutral primes, at p < .01, and Cable read his result as set rather than defence. York, Mandour and Jex repeated the 1949 experiment in 1984, and their abstract reports taboo words still above neutral words on threshold while the skin-conductance difference did not come out.
York returned in 1996 with four predictors of the effect, one of them skin conductance, and reports a multiple correlation of .19 accounting for 3 percent of the variance. The modern experimental base is thin. Crossref, OpenAlex, PubMed, Semantic Scholar and the works citing Hastorf and Cantril since 2010, searched on 21 September 2026, turned up two studies that separate an earlier stage from a judgement, and the two disagree.
Huff and colleagues ran on 58 people with the eye-tracking arm down to 15, and de Bruin and colleagues on 21 conservatives and 22 liberals. Beyer and colleagues' replication and Boon and Davies' replication were both unreachable on that date, each behind a bot challenge at the publisher rather than a paywall, so this entry quotes their abstracts and no figures from either.
Each of these is named in this entry's own prose as something it gets taken for. Where the two came out differently, that is worth knowing before you quote either.
The football result is almost always told as both sides seeing their own team as cleaner, and that is not what the table says. Dartmouth students saw 4.3 infractions by Dartmouth and 4.4 by Princeton, which is no gap at all, and Princeton students saw 4.2 by their own team, close to the 4.4 the Dartmouth students gave it.
One number moved: the 9.8 that Princeton students charged to Dartmouth. Nor did Hastorf and Cantril claim anyone was wrong. They footnote that they are not concerned with guilt, and the referees that day penalised Dartmouth 70 yards to Princeton's 25. Boon and Davies add that the study is widely cited as evidence about eyewitness testimony, which its original purpose and method do not support.
The second misreading is that Dearborn and Simon established something. Walsh counted 60 citations to it in the management literature over 1978 to 1987. He then read their published appendix of raw answers and concluded that it is not at all clear their data support their conclusion. Their sales managers named organisational problems, he points out, and their accounting managers named marketing ones.
The third is treating the name as a synonym for confirmation bias, entry 5, or motivated reasoning, entry 59. Confirmation bias is about the search for and reading of evidence bearing on a hypothesis. Motivated reasoning is about wanting a conclusion. The hostile media effect, entry 205, is the partisan strand carried into news coverage and keeps its own record.
Selective perception as its founders meant it was a claim about the perceptual act itself, and that is the version Huff and colleagues tested and did not find. The fourth misreading is taking McGinnies as settled evidence that people see a taboo word late. Howes and Solomon's answer, that the viewer saw it on time and held the word back, is as old as the finding and sits in the same issue as McGinnies' reply to it.
Every claim on this page was held against these sources on 2026-09-21. Nothing here is written from memory. Written and checked by Krishna Chagti; how entries are written and corrected. Found a mistake? Report it, with the sentence and a source.