Naive realism
People take their own view of a situation to be an unmediated reading of it, and so treat anyone who reads it differently as biased, misinformed or lazy.
People take their own view of a situation to be an unmediated reading of it, and so treat anyone who reads it differently as biased, misinformed or lazy.
Ross and Ward set out three linked convictions, and they write them in the first person. The first is that one sees entities and events as they are in objective reality. The second is that other rational perceivers given the same information will share one's reactions. The third covers what to make of someone who does not, and it offers three explanations rather than one.
That person may have seen a different sample of information, may be lazy or irrational, or may be biased by ideology or self-interest. Ross and Ward note that only the first explanation leaves room for agreement, because pooling what each side knows can close it. Pronin, Gilovich and Ross place that third conviction at the centre of their account, because it turns a disagreement into a diagnosis.
Naive realism is a framework rather than a result, and it is not tested directly. What gets tested are its predictions, of which three have a substantial literature: that partisans see neutral coverage as hostile to them, that they overestimate how far apart the two sides really are, and that they attribute the other side's views to ideology and their own to evidence.
Replicated Two preregistered experiments reproduced the hostile media prediction on samples of 817 and 819 Americans, and it went missing in the one condition where the story was not contested.
Lo Iacono and Dores Cruz ran two preregistered experiments on Amazon Mechanical Turk, showing participants identical real news items labelled as coming from CNN or from Fox News. In the COVID-19 study, N = 819, a one-point rise on a seven-point conservatism scale raised perceived bias by 0.23 points on an index running from 0 to 4 when the source was CNN rather than Fox for a story about protests against distancing rules, and by 0.20 points for a story about compliance, both p < 0.001.
In the police study, N = 817, the same contrast gave 0.15 points for coverage of an officer killing a civilian, p < 0.001, and nothing for coverage of an officer saving a civilian's life. The authors report that null as beta = 0.05, p = 0.129, with a standard error of 0.06 in the text and 0.041 in the table.
Their design swaps the outlet label rather than holding one source neutral, so it tests the relative form of the claim, which entry 205 separates from the absolute one. The second prediction has preregistered support of its own. Lees and Cikara asked partisans how the other party would react to their own party's legislative moves, and compared the answers to what the other party actually reported.
Their nationally representative Experiment 4, N = 536, found meta-perceptions more negative than reality on all three measures, with the disliking contrast at b = -0.75, 95% CI [-0.95, -0.56]. In their Experiment 6, N = 1,122, the 358 participants told the true figures rated the out-group lower on obstructionism than the 396 controls, b = -4.08, p = 0.027, and the effect was largest among those who had been most wrong. These figures were read from the authors' accepted manuscript in Harvard's DASH repository, because the journal article is closed at Springer Nature.
The meeting everyone left differentlyillustration
Two colleagues leave the same meeting and each concludes the other was not paying attention, since the meeting plainly went one way. Neither treats his own account of it as an account rather than the thing itself.
The refereeing decision on the replayillustration
Supporters of both sides watch the same replay, see the obvious call, and conclude that anyone taking the other view is either blind or dishonest. The diagnosis of the dissenter is the part that does the damage.
The rota one flatmate calls fairillustration
A flatmate draws up a cleaning rota he considers evidently even, and reads the objection to it as laziness. The explanation for the disagreement comes from the objector's character rather than from the rota.
Cases marked as illustrations describe nobody in particular. They are there to make the pattern recognisable, not to report that a particular thing happened.
Vallone, Ross and Lepper showed 144 Stanford students 36 minutes of network news on the Sabra and Chatilla massacre, taken from six segments broadcast between 15 and 24 September 1982. Sixty-eight described themselves as pro-Israeli, 27 as pro-Arab and 49 as neutral. On a nine-point scale where 5 meant fair and impartial, pro-Arab viewers rated the same tapes at 6.7 and pro-Israeli viewers at 2.9, F(1, 141) = 98.70, p < .001.
The two groups also reported different content. Pro-Arab viewers said 42% of references to Israel were favourable and 26% unfavourable; pro-Israeli viewers said 16% and 57%, F(1, 141) = 54.84, p < .001. Each side expected neutrals to be swayed against it, with pro-Israelis putting the figure at 68% and pro-Arabs at 37%, F(1, 141) = 34.02, p < .001.
Controlling for perceived content left the bias difference intact, F(1, 92) = 30.89, p < .001. Within each partisan group the better-informed saw more bias, r = .31 among pro-Israelis and r = -.41 among pro-Arabs. Robinson and colleagues then measured the second prediction. Twenty-seven pro-choice and 25 pro-life students rated six abortion scenarios and estimated how each side would rate them.
Real differences existed, t(49) = 5.19, p < .001 on the composite, and both groups had presumed them larger, t(49) = 3.46 for pro-choice raters and t(49) = 3.10 for pro-life. A second questionnaire on the same topic went to an introductory psychology class, 66 of whom called themselves strongly pro-choice and 22 strongly pro-life. Their second study gave 23 liberals, 20 conservatives and 16 neutrals a synopsis of the Howard Beach case.
The real gap between liberal and conservative readings was 9 points on a 100-point construal scale, 63.2 against 54.2, t(41) = 3.14, p < .01, and all three groups put it far wider. Every group also thought both partisan camps had been driven more by ideology than the camps themselves claimed, in all but one case at p < .01.
Each group thought both camps had leaned less on evidence than they claimed, though that fell to p < .10 where conservatives judged conservatives. Liberals and conservatives differed little in what they said about their own reasoning.
2 primary research4 replication and reanalysis3 commentary and critique
Their title carries the term, and their paper is the empirical anchor of the account. Robinson and colleagues credit the name to Griffin and Ross, and to a Ross and Ward chapter then in press. That chapter appeared in 1996 as 'Naive realism in everyday life', and it is the one that sets out the three convictions. Pronin, Gilovich and Ross later cite Ross and Ward's 1995 and 1996 chapters together as the account's source. The hostile media study that the framework is best known for came ten years earlier, and entry 205 covers it.
The claim, and the replication this entry reads its figures from.
The strongest number in this literature could not be obtained. Hansen and Kim pooled 34 hostile media studies, and their abstract reports a moderate effect, larger as involvement rises but present at low involvement too, with no evidence that medium or study design matters. The article is closed at Taylor and Francis, which answers with a Cloudflare bot challenge and HTTP 403 rather than a paywall notice, to a browser-like agent and to a plain one alike.
Unpaywall, OpenAlex and Semantic Scholar all recorded no open copy on 21 September 2026, and Internet Archive Scholar and the Wayback Machine both answered with rate limits that day, so neither was a clean negative. A pooled correlation circulates in search-engine summaries of that paper, and it is not printed here, because a snippet is not a source.
The boundary condition is clearer than the size. Vallone and colleagues had already run pilot studies on Carter and Reagan coverage and found no tendency to perceive hostile bias, which they put down to their participants caring too little about a finished election. Lo Iacono's positive police story produced the same nothing. What the effect needs is a story the sides are already fighting over.
The false polarisation work has its own gaps. Robinson's first study rests on 52 recruited partisans, plus 88 more in a class who answered a second questionnaire, and his second on 59 students. Lees and Cikara found their negativity bias only in competitive scenarios: with cooperative ones, N = 499, meta-perceptions were accurate on all three measures, the largest coefficient being b = 0.12, 95% CI [-0.11, 0.35].
Naive realism itself is not falsifiable in the way its predictions are, since a framework that explains hostile media perception, false polarisation and the bias blind spot alike survives the failure of any one of them.
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 first misreading is that this is the bias blind spot under another name. The two are related and not the same. The blind spot is a self-other comparison on rated susceptibility to bias, and entry 133 covers its preregistered replications and their pooled estimate. Naive realism is the epistemic stance offered as one explanation for that comparison, and the evidence for it is indirect by construction.
The second is that partisans are wrong about their opponents. In Robinson's studies they overestimated the extremity of their own side as well, and the neutral participants in the Howard Beach study overestimated both camps. Robinson and colleagues conclude from that pair of findings that the rater's own partisanship was not the key, and that what everyone shared was an exaggerated notion of what partisanship does.
Lees and Cikara found the same thing in a sharper form: participants shown an unlabelled Party A and Party B were as pessimistic as those shown their real rivals, and in the nationally representative sample they overestimated how badly their own side would react too, b = -0.34, p < 0.001 for disliking. The third is that the Beirut tapes were established as neutral.
Vallone and colleagues say plainly that they had no measure of objective fairness, and that the neutral group's ratings sat closer to the pro-Israeli group's on every item. Their defence of the finding is not that the coverage was even-handed, but that the more knowledgeable neutrals landed midway between the partisans while both partisan groups saw hostility no neutral saw.
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.