Cocktail Party Effect

​

What is Cocktail Party Effect?

Cocktail Party Effect is a later report that one heard one’s own name in a spoken channel one had been told to ignore, while repeating another channel word by word.

How do we recognize what one person is saying when others are speaking at the same time (the “cocktail party problem”)?

That question is the one on page 976 of Cherry’s 1953 paper. The words in parentheses are his name for the job of separating one speaker from another. This page does not use that name for the name report. Wood and Cowan call the name report the cocktail party phenomenon. Conway uses that phrase and, in quotation marks, cocktail party effect. This page keeps Cocktail Party Effect for the name report. It does not rotate the 1953 problem, or the word phenomenon, into the page name after this section. The counted report is narrow. The listener wears headphones. One ear gets a list of unrelated one-syllable words and must be repeated aloud. The other ear gets a different list and is to be ignored. At one marked time the listener’s own first name replaces a word on the ignored list. After the lists stop, the listener is asked what, if anything, was noticed. A report of one’s own name is the effect. A report of the sentence around that name is not what the studies measured. Frequency Illusion is the belief that a thing, once noticed, is happening a lot. Catching a name once is not that belief. Priming Effect is a faster correct decision that two strings are words, when the words are associates. A name in an ignored ear is not that decision.

​

Cocktail Party Effect in 3 Depths

• Beginner: In Wood and Cowan, 5 people in the 4-minute placement and 4 people in the 5-minute placement later said they had heard their own name. The paper prints that pair as 34.6% of the people whose names were presented. Eight other people heard lists with no name in them. None of those eight reported a name. Nobody reported the other person’s name that had been inserted as a control.
• Practitioner: Before you treat a clean repetition as proof the name never registered, look at the two attended words that came after the name. People who later recalled the name made more errors on those two words than people who did not. The attended word that arrived with the name was not that row. The two words before the name were not that row either.
• Advanced: Conway, Cowan, and Bunting found the report in 65% of the low-span group and 20% of the high-span group, in a sample of 40. The tests print t(32) and t(15). A split of 40 does not produce 32 degrees of freedom, and 20% of a group of 20 against the rest of that group does not produce 15. This page does not repair the degrees of freedom, and it does not move 65% into a meeting.

​

Origin

The problem and the name report are two publications. Cherry poses the separation job. Wood and Cowan count the name. Conway splits the name report by a memory span. Neville Moray’s 1959 paper is the study they are replicating. That paper was not re-read. The counts below that name Moray are Wood and Cowan’s report of his table, not a second reading. E. Colin Cherry was at Imperial College, University of London, and at the Research Laboratory of Electronics, Massachusetts Institute of Technology, as a visiting professor under a Fulbright grant. The paper is Journal of the Acoustical Society of America, volume 25, number 5, September 1953, pages 975–979. It was received on 5 May 1953. The copy read here does not print a DOI. A footnote says the work was supported in part by the Signal Corps, the Air Materiel Command, and the U.S. Office of Naval Research, and was presented with the permission of Professor J. B. Wiesner. The page-976 question quoted above is the cocktail party problem. Cherry’s rejected-ear tests do not insert the listener’s name. In the tests where normal speech went to the ignored ear, he writes that in no case did listeners fail to identify it as speech, and that in every such instance they could not identify any word or phrase, or identify the language as English. A change from a male voice to a female voice was nearly always identified. A steady 400-cycle tone was always observed. Reversed speech was called something queer by a few listeners and was taken for normal speech by others. When the ignored ear was switched into German, spoken by an Englishman, the listener “did not know at all, but assumed it was English.” He says the result was similar with different, unprepared listeners. No head count is printed for those listeners. He also says it is unfair to run that German test more than once on the same person, and that no listener was used for more than one test. Those sentences are not a name result. The name counts are Noelle Wood and Nelson Cowan, “The Cocktail Party Phenomenon Revisited: How Frequent Are Attention Shifts to One’s Name in an Irrelevant Auditory Channel?,” Journal of Experimental Psychology: Learning, Memory, and Cognition, volume 21, number 1, 1995, pages 255–260. The copy read here does not print a DOI. It was received on 16 December 1993, the revision was received on 28 February 1994, and it was accepted on 10 March 1994. Both authors were at the University of Missouri, Columbia. Correspondence went to Cowan at 210 McAlester Hall, Columbia, Missouri 65211. National Institutes of Health grant HD-21338 supported the work. They write that Moray’s detection rate was only 33%, a point often overlooked, and that his Table IV, page 58, showed 4 of 12, or 33%, heard their name on both the first and the second presentation. A third presentation, with warning, was reported as detected by 80%, and they say that third presentation should probably be viewed as divided attention. This page does not treat 80% as the unannounced rate, and it does not re-read Moray. The span split is Conway, Cowan, and Bunting, “The cocktail party phenomenon revisited: The importance of working memory capacity,” Psychonomic Bulletin & Review, volume 8, number 2, 2001, pages 331–335. The copy read here does not print a DOI. The manuscript was received on 22 July 1999, and the revision was accepted on 14 June 2000. Conway and Bunting were at the University of Illinois at Chicago, Department of Psychology, 1007 West Harrison Street, Chicago, Illinois 60607-7137. Cowan was at the University of Missouri, Columbia. NIH grant R01 HD-21338 supported the work. Conway writes that, contrary to popular belief, not all subjects show the effect, and he summarizes Moray as 33% and Wood and Cowan as 34.6%. His own counted result is the span split in the case below. Donald Broadbent’s 1958 book, which Conway cites for an early filter, was not re-read. Anne Treisman’s 1960 and 1964 papers, which Wood and Cowan cite for an attenuating filter, were not re-read. No result from those three sources is on this page.

​

Key Points

Four cuts keep a rejected word, a recalled name, and a span percentage from inheriting one another. The headphones, the questionnaire, and the degrees of freedom each have their own limit.

​

Applications

Use the contrast when a noticed name is treated as proof that the ignored sentence arrived, or when a laboratory share is treated as a property of every room. Keep each count on the question that produced it.

​

Case Study

In 1995 Wood and Cowan asked how often a name in an ignored auditory channel is later recalled, and whether that recall lines up with a slip in the repeating task at the moment of the name. The listeners were not at a party. They were undergraduates at the University of Missouri, Columbia, tested one at a time in a sound-attenuated chamber. Thirty-four remained in the sample, 25 men and 9 women. Nine further people were excluded, for the reasons above. Everyone repeated a list of 330 one-syllable words in a monotone female voice, 60 words a minute, played to the right ear. The left ear carried 300 one-syllable words in a monotone male voice at the same rate. The right-ear list began 30 seconds before the left-ear list. Both ended together after 5.5 minutes. Levels in both ears were set in a range of 65 to 70 dB(A). The listener’s own first name, and one other participant’s first name, were recorded in that same male voice and inserted by computer, replacing a word. One’s own name came at 4 minutes or at 5 minutes. The other name came at the other mark. Eight people received neither name. The questionnaire came after the lists stopped. For a name at 5 minutes, that was 30 seconds after the name. For a name at 4 minutes, it was 90 seconds after the name. Listeners were asked what they could remember from the left ear, whether anything unusual was heard, and whether they had heard their own name. Five of the 13 people with a name at 4 minutes, and 4 of the 13 with a name at 5 minutes, recalled their own name. The paper prints 34.6%, and a Fisher’s exact test between those two conditions at p = .50. Nobody in the group of 8 reported a name. Nobody reported the yoked name, the other participant’s name inserted as a control. The paper also says 14.7% of “our participants” recalled a specific word such as left or praise, and 38.2% volunteered that the ignored channel was a male voice. It does not print the head count behind either percent. This page does not turn 14.7 or 38.2 into a number of people. The online record agrees with the questionnaire only in a short window. Across the session, recallers averaged 17.0 shadowing errors and non-recallers 20.5, a nonsignificant difference. On the attended word after the name, and on the word after that, recallers made more errors. The F tests are F(1, 128) = 12.60 and F(1, 128) = 15.74, both p < .001. The other positions in the six-word window were not significant on the Fisher’s exact tests listed above. In the error-free subset, all 4 who later recalled the name, and 2 of 12 who did not, showed at least one latency among the three following words that was at least 50 milliseconds above any of the three words up through the name, p < .01. The lesson is a name, inserted once into an ignored word list, later reported by 9 of the 26 people who received one, with a short disruption in the repeated list among those who reported it. It is not a count of sentences understood at a party, and the 8 people who received no name reported none.

​

Boundaries and Failure Modes

Cocktail Party Effect is the wrong summary when 34.6% is applied to every guest, when Cherry’s unnoticed words are retold as a name result, or when 65% is applied to anyone described as forgetful. The first boundary is the denominator, and the forewarned presentation that does not belong beside it. The 34.6% is 5 plus 4 among the people whose names were presented. The two name groups were 13 and 13. The 8 who heard no name are inside the 34 and outside the 34.6%. Wood and Cowan’s report of Moray’s Table IV is a different study: 4 of 12 heard their name on both the first and the second presentation, which Wood and Cowan also summarize as 33% and, in the discussion, as 33.3%. They say the match with 34.6% could be coincidental, because Moray’s sample was small. The 80% belongs to a third presentation that the listeners had been warned about. Wood and Cowan say that one should probably be viewed as divided attention. A sentence that says “about a third, or 80% if you are listening for it” has glued the unannounced table to the warned presentation. Moray 1959 was not re-read, so this page does not add any cell that Wood and Cowan do not print. The second boundary is Conway’s span split, and it is where the printed tests do not line up with the head count a reader can form. The sample is 40, half high-span and half low-span, so 20 in each group is half of 40. The reports are 20% and 65%. Fisher’s exact test is p = .005. Overall shadowing errors differ by span at t(32) = 2.18. Errors on the two words before the name do not, t(32) = 1.0 in each of those two tests. Errors concurrent with the name do, t(32) = 2.59, p = .01. On the two words after the name, high-span listeners who reported the name differed from high-span listeners who did not, t(15) = 3.04, p = .01, and the low-span pair differed at t(15) = 2.50, p = .03. Three words after the name, both contrasts are above .10: t(15) = 0.75 and t(15) = 1.74. A two-group test on 40 people would ordinarily carry 38 degrees of freedom, not 32. Twenty percent of 20 is 4, and 65 percent of 20 is 13. Those head counts are arithmetic from the printed percents and from half of 40. The paper does not print 4, 16, 13, or 7. A contrast of 4 against 16 would ordinarily carry 18 degrees of freedom, not 15. This page does not invent the people who would make t(32) or t(15) match, and it does not drop the t tests. The span scores themselves stay in their groups: 24.85 against 8.22, with the ranges above. They are not percentages of names noticed. The third boundary is the common misuse, including the hope that a party is the experiment and the hope that a noticed name means the ignored message was understood. Cherry’s listeners could reject an ear and still miss its words, its language, and a change into German. They could notice a voice and a tone. He prints no name and no party tally. Wood and Cowan’s listeners who recalled the name were not worse across the whole session. The extra errors sat on two words. Fourteen point seven percent recalling some specific ignored word, with no printed head count, is not a transcript of the left ear. Processing Fluency is a feeling of ease used as a cue about the thing itself. A name that breaks into an ignored list is not a hard font, and none of these shares is a minute estimate from that page.

​

Common Misconceptions

Three retellings turn a word-list questionnaire into a party, Cherry’s problem into a name, or a low span score into a better ear.

​

Related Concepts

The name report sits next to a faster word decision, a belief that a noticed thing is suddenly everywhere, a visual item that stands out, a feeling of ease used as a cue, and a memory difference when a statement is stored outside the head.

​

One-Line Takeaway