> ## Documentation Index
> Fetch the complete documentation index at: https://meta.niceshare.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Illusory Truth Effect
> Illusory Truth Effect is a higher truth rating after a plausible statement is repeated. In 1977 the mean rose from 4.35 to 4.74. Later repeats added less.
**Category**: Effects
**Type**: A higher truth rating for a plausible statement that has been heard again, compared with a similar statement that has not
**Origin**: Lynn Hasher, David Goldstein, and Thomas Toppino, *Journal of Verbal Learning and Verbal Behavior* 16, 1977, pages 107–112. The paper's words are referential validity. It does not print Illusory Truth Effect
**Also known as**: Referential validity, the 1977 paper's own phrase for this rating
**Quick Answer** — The Illusory Truth Effect on this page is a higher truth rating for a plausible statement that has been heard again. Lynn Hasher, David Goldstein, and Thomas Toppino measured it in 1977. Across three sessions, two weeks apart, the mean for repeated statements rose from 4.35 to 4.67 to 4.74 on a 7-point scale. Four was "uncertain." A later online study found that the large step is the second hearing. Going from nine hearings to 27 did not add a significant further step.
## What is the Illusory Truth Effect?
The Illusory Truth Effect, as counted here, is a higher truth rating for a plausible statement that has been heard again than for a similar statement heard once.
> The repetition of a plausible statement increases a person's belief in the referential validity or truth of that statement.
That sentence is Hasher, Goldstein, and Toppino's summary of their own sessions. The rating stayed near the middle of the scale. The repeated mean moved from 4.35 to 4.74. It did not move to "definitely true." [Mere Exposure Effect](/effects/mere-exposure-effect) is a change in liking after repetition. [Priming Effect](/effects/priming-effect) is a faster word decision when the other word is a common associate. Neither one is this truth rating.
### The Illusory Truth Effect in 3 Depths
* **Beginner**: A plausible line, heard again two weeks later, was rated truer than it had been. On the 1977 scale, 4 means uncertain and 5 means possibly true. The repeated mean went from 4.35 to 4.67, then to 4.74.
* **Practitioner**: Write the true row and the false row before you quote one mean. False repeated statements went from 4.18 to 4.54 to 4.67. Statements that were not repeated did not rise.
* **Advanced**: A second hearing can move the rating more than the eighth extra hearing. In a 2021 study the new mean was 3.76 and the once-heard mean was 4.27. Nine hearings averaged 4.69. Twenty-seven was not a significant step past nine.
## Origin
The paper is Lynn Hasher, David Goldstein, and Thomas Toppino, "Frequency and the Conference of Referential Validity," *Journal of Verbal Learning and Verbal Behavior*, volume 16 (1977), pages 107–112. The scanned PDF read here does not print a month or a DOI. The publisher record is 10.1016/S0022-5371(77)80012-1, and the issue listing is February 1977. The note at the end says the paper was received on 2 June 1976. Hasher and Goldstein's byline is Temple University. Toppino's byline is Villanova University. Reprint requests went to Hasher at Weiss Hall, Temple University, Philadelphia.
The paper's own words are frequency and referential validity. It does not print illusory truth, illusion of truth, or Illusory Truth Effect. This page uses Illusory Truth Effect for the repetition contrast they measured. It does not treat that later name as a second result.
On three occasions, each separated by two weeks, students heard 60 plausible statements and rated each one. Twenty statements, 10 true and 10 false, occurred on every list. The other statements were new each time. The pool had 140 statements, written so that they were plausible and unlikely to be known for certain by most college students. The method names ten general topics. Table 1 prints ten rows of examples. Those two lists do not use the same ten labels. This page does not merge them. One true example in the table is "Lithium is the lightest of all metals." One false example is "The capybara is the largest of the marsupials."
The 20 repeated statements were one true and one false item from each category, chosen at random inside that constraint. The other 120 statements were split into three groups of 40, half true and half false. Each group was joined to the 20 repeated statements to make one list of 60. Inside a list, only new statements occupied the first 10 slots and the last 10. The repeated statements sat in the middle 40, mixed with other new statements. The comparisons use the middle 40. The authors' reason for the buffers is that a student who looked a statement up would be most likely to look up an item at either end. They wanted the repeated items left alone.
A speaker recorded the lists in standard English. Ten seconds separated the start of one statement from the start of the next. A statement took about five seconds, so about five seconds of quiet followed it. Two orders of the three lists were used. Half the students heard each order.
The scale ran from 1 to 7. Four meant "uncertain." Five, six, and seven meant "possibly true," "probably true," and "definitely true." The same adjectives, on the false side, were tied to three, two, and one. Students were told to rate each statement at once, before the next one. They were told the task was to help validate a new test of college students' general knowledge.
Forty students' ratings were analyzed. Ten other students were dropped because they missed a session. Twenty of the 40 were run in small groups of 1 to 10. The other 20 were one intact class. A check found no difference between those settings, and the analyses collapse them. Of 7,200 statement-occasions, 60 times 3 times 40, 29 were left blank. Twenty of those blanks were two nonrepeated statements in the second session, for the 20 students in the class. The authors call that a temporary disruption. The small groups left only two events blank.
## Key Points
Four readings keep each mean on the row that produced it. A rise is not a proof that the statement became true. A ninth hearing is not a second copy of the first rise.
The repeated statements and the comparison statements came from the same pool of plausible claims. In the first session the two sets did not differ, F(1, 39) = 1.69, mean square error .45, p > .05. The later gap is the gap after repetition. A comparison that brings in easier claims on one side has not repeated this control.
Repeated true statements averaged 4.52, then 4.80, then 4.80. Repeated false statements averaged 4.18, then 4.54, then 4.67. The repeated means, collapsing those rows, were 4.35, 4.67, and 4.74. Students still rated true statements higher than false ones overall, 4.52 against 4.28, F(1, 39) = 34.7, mean square error .20, p \< .01. Truth did not interact with repetition or with session. The rise is not a rise for false statements only, and it did not erase the true-versus-false gap.
Statements that were not repeated averaged 4.25, then 4.22, then 4.16. The authors say that line diminished slightly. The test of change for those statements is F below 1. The interaction of repetition and session is F(2, 78) = 11.80, mean square error .20, p \< .01. The rise across sessions for repeated statements is F(2, 78) = 16.34, mean square error .41, p \< .01. Quote the interaction. A note that only says the new statements fell has promoted a test the paper does not call significant.
Aumyo Hassan and Sarah J. Barber, *Cognitive Research: Principles and Implications* (2021), article 38, repeated trivia they treated as largely unknown, and collected a truth rating one week later. In Experiment 1 the new mean was 3.76 and the once-heard mean was 4.27, on a scale from 1 to 6. Nine hearings averaged 4.69. Among the repeated counts, the only significant pairwise step was one hearing against nine. [Availability Heuristic](/effects/availability-heuristic) treats an easy example as a common event. A higher mark on this scale is not that judgment.
## Applications
Write the session, the truth of the statement, and the number of hearings before a truth rating is used outside these tables.
The 4.74 is the repeated mean in session 3, averaged over true and false statements. The false repeated row in that session is 4.67. The true repeated row is 4.80, and it was already 4.80 in session 2. A slide that prints 4.74 as "people believed it" has left the scale and both rows behind.
In Hassan and Barber's Experiment 1, new statements averaged 3.76 and once-heard statements averaged 4.27. The means at three, five, and seven hearings were 4.51, 4.49, and 4.49. Nine was 4.69. The significant step among those repeated cells was one against nine. A plan that buys eight more repetitions in order to buy another 4.27-sized jump has not read Table 1.
The 4.35 and the 4.74 are 40 college students, taped trivia, and a cover story about a knowledge test. They are not a headline, a slogan, or a price. [Framing Effect](/effects/framing-effect) is the same fact in two wordings. A post can be that. It is not Table 2.
Hasher, Goldstein, and Toppino chose statements that were plausible and unlikely to be known for certain. Hassan and Barber used trivia they expected to be largely unknown, and every statement they showed was factually true. A line the listener can already check is a different test. These tables do not contain it.
## Case Study
Hasher, Goldstein, and Toppino asked college students to rate statements they were unlikely to know. The students heard a tape of 60 statements on three days, with two weeks between days. Each statement was scored at once, on a scale whose midpoint, 4, was labeled uncertain. The cover story was a new test of general knowledge. Forty students completed all three days. Ten others did not, and their ratings were not used.
Twenty statements returned every day. Ten of those were true and ten were false. The other statements on a given day were new. The first 10 and the last 10 slots were always new statements, so the comparison lives in the middle of the list. In that middle stretch a tape had between 8 and 12 true statements among the new items. The analyzed number is each student's mean for a category, not one shared statement.
Table 2 is those means. Repeated statements: 4.35 the first day, 4.67 the second, 4.74 the third. The standard deviations of the subject means were printed cell by cell. For the collapsed repeated row the paper prints the means only. Not-repeated statements: 4.25, 4.22, and 4.16. The repeated line rose, F(2, 78) = 16.34, p \< .01. The not-repeated line did not change, F below 1. On the first day the statements that would later be repeated were not rated differently from the statements that would not, p > .05.
The same table splits truth. Repeated false statements moved from 4.18 to 4.54 to 4.67. Repeated true statements moved from 4.52 to 4.80 and then stayed at 4.80. Across the whole design, true statements still outscored false ones, 4.52 against 4.28. That gap did not interact with repetition. The authors write that the increase with repetition was equivalent for true and for false statements, and that the students still discriminated the two.
The lesson is a rating inside the band from uncertain to possibly true. The highest cell mean in the table is 4.80. The lowest is 4.04. Forty students, three sessions, and a pool written to be plausible rather than familiar produced a rise for the lines that came back and no reliable change for the lines that did not. The rise is not a vote, and it is not a later week's news feed.
## Boundaries and Failure Modes
The Illusory Truth Effect is the wrong summary in three retellings. The 1977 rise is treated as any later message. A long string of repetitions is treated as a stack of equal jumps. A truth rating is treated as liking, or as proof the statement became true.
The first boundary is the statement. Both studies counted here used claims the listener was not expected to know. Hasher, Goldstein, and Toppino say there was no verifying information in the session about the actual truth of the statements. They also say the students may have had some reliable information, because true statements outscored false ones, or that the true statements may have been slightly more plausible. Hassan and Barber used 100 trivia statements, 66 adapted from items previously normed as relatively unknown but plausible, and 34 more that the research team judged the same way. Those 34 were not normed. Every statement in their sessions was factually true. They say they avoided false statements because a false belief can be hard to undo. A page that exports 4.67 as the rating of a known falsehood has left both designs.
A separate literature reports that repetition can still move a statement that contradicts what the person knows. The papers usually named for that claim were not re-read here. No cell from them is on this page. [Confirmation Bias](/effects/confirmation-bias) is a preference for the evidence that agrees with a belief already held. It is not Table 2.
The second boundary is the number of hearings. Hassan's byline is San Francisco State University. Barber's byline is Georgia State University, and Barber is the corresponding author. The PDF prints received 29 July 2020 and accepted 28 April 2021. The review board was Georgia State University, protocol H19217. Experiment 1 analyzed 51 adults in the United States, ages 20 to 55, mean 33.27. They had finished both sessions and passed the attention checks. Session 1 showed 50 statements, 10 at each of one, three, five, seven, and nine presentations, for 250 trials plus three attention checks. Each statement stayed up for 4 seconds. The rating in that session was interest, from 1 to 6, not truth. One week later the full list of 100 statements was rated for truth, from 1, not truthful, to 6, very truthful. Fifty were new. Collapsing the repetition counts, repeated statements averaged 4.49 (standard deviation 0.60) and new statements averaged 3.76 (standard deviation 0.67), t(50) = 7.16, p below 0.001, d = 1.00. The cell means were 3.76, 4.27, 4.51, 4.49, 4.49, and 4.69 for zero, one, three, five, seven, and nine hearings. Three against five is a difference of 0.02 points in the wrong direction for a "more is always higher" story, and it was not a significant pair. The correlation of a person's means with the logarithm of the counts, after adding 1 so that zero has a log, averaged r = 0.52. The correlation with the raw counts averaged r = 0.46. The log correlation was larger, t(50) = 4.83, p below 0.001, d = 0.67. Those correlations are across the six counts, including the new items. They are not a correlation of belief with a news diet.
Experiment 2 used the counts 1, 9, 18, and 27, with 57 adults ages 21 to 35, mean 29.46. Eight statements sat in each count. Session 1 was paced: 4 seconds to read, then 4 seconds to rate interest. Repeated statements, collapsed, averaged 4.66 (standard deviation 0.86) against 3.64 (standard deviation 0.65) for new statements, t(56) = 8.22, p below 0.001, d = 1.09. The cell means were 3.64, 4.26, 4.78, 4.72, and 4.87. The mean at 18 hearings is lower than the mean at nine. Pairwise tests mark one hearing as lower than nine, 18, and 27. They do not mark nine, 18, and 27 apart. A second t(56) = 8.22, with d = 0.63, is the comparison of the log correlation with the linear correlation. It is not a second copy of the 4.66-against-3.64 test. The paper's own summary is that the largest increase came from encountering a statement for the second time, and that further hearings added less. The discussion also says the increases past nine were not statistically significant.
The third boundary is the export. The 4.74, the 4.27, and the 4.87 are not a shop, a classroom, a household, or a campaign. Hasher and colleagues end on the idea that if people are told something often enough, they will believe it. Their table still tops out at 4.80, which is their label for a point between uncertain and possibly true. Hassan and Barber discuss politics and public health in the last section. The counts they cite from those outside papers are not re-read here and are not on this page. A team that treats every repeated sentence as a 4.74 has not read the false row, the unrepeated row, or the week-later cells that stop climbing.
## Common Misconceptions
Three shortcuts change what the tables measured. One keeps only the false row. One treats every extra hearing as a fresh copy of the first jump. One gives the 1977 paper a name it does not print.
False as a summary of this table. Repeated false statements did rise, from 4.18 to 4.67. Repeated true statements also rose, from 4.52 to 4.80, and then stayed. The truth gap, 4.52 against 4.28, remained, and it did not interact with repetition. Hassan and Barber's means are from true trivia only. A higher rating is not a conversion of a known falsehood, and it is not confined to false lines.
The 1977 steps are not equal. For repeated true statements the mean goes 4.52, 4.80, 4.80. The third session adds nothing on that row. In 2021 the step from new to once-heard is 3.76 to 4.27. The step from one hearing to nine is the only significant step among the repeated cells in Experiment 1. In Experiment 2 the mean at 27 hearings is 4.87 and the mean at nine is 4.78. That pair was not significant. Nine to 18 even prints a lower mean, 4.72.
The paper's title is about frequency and referential validity. The discussion says repetition increases belief in the referential validity or truth of a plausible statement. It does not print Illusory Truth Effect. This page uses that later name for the contrast they counted. [Illusory Correlation](/effects/illusory-correlation) is a relationship seen between two events that the counts do not tie together. A higher mark after a repetition is not that relationship.
## Related Concepts
The rating after a repetition sits next to liking after a repetition, a judgment that uses an easy example, a preference for agreeing evidence, a faster word decision, and a relationship the counts do not show.
Liking that rises after a stimulus is shown again. Hasher, Goldstein, and Toppino cite that work as a different use of frequency. Their own scale is truth, and 4 means uncertain.
An easy example treated as a common event. A statement rated 4.67 has not been counted in the world. The mark is a certainty rating inside one session.
A preference for evidence that agrees with a belief already held. The 1977 rise appears for statements the students were not expected to know, on both the true row and the false row.
A faster correct decision that two letter strings are words, when the words are common associates. The 1971 gap is milliseconds. It is not a 7-point truth rating two weeks later.
A relationship read into two events the counts do not connect. Repetition here changes the rating of one statement. It does not pair that statement with a second event.
## One-Line Takeaway
Before you treat a repeated line as believed, write the scale point, whether the line was true, and whether the hearing was the second one or the twenty-seventh.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.