The eighty-eighth article was about a metric that discards information. This one is about a statistic that was never about what it is used for, and about the weakest sourcing this series has had to work with.
A Warning About Our Sourcing
We put this first rather than in the method note, because it should change how a reader weighs everything below.
We obtained no abstract from either 1967 paper. Both predate the digital record and neither publisher record we found carries one.
Consequently four of our six sources are non-academic: an encyclopedia, two specialist websites and a personal essay. Every description of what the studies did reaches this article through one of them.
What partly compensates is that they agree with each other in detail, were written independently, and two of them quote the same sentence from the paper identically. But agreement among secondary sources is not the same as reading the paper.
We would have preferred not to write this article on this evidence. We are writing it because the claim it examines is taught to business owners as fact, and because one part of the argument requires no sources at all, which the arithmetic section demonstrates.
So we have separated the article into two parts, and a reader should treat them differently. The historical account of what the studies did is secondhand throughout, and if our sources have it wrong then those sections are wrong.
The arithmetic sections depend on nothing but the coefficients themselves, which are quoted identically by independent sources and are in any case the numbers everybody uses. That part stands whatever the history turns out to be.
The Verdict, Stated First
Five claims, in descending order of confidence.
One. The popular version fails a test every reader has already run. If words carry seven percent, you would understand ninety-three percent of a conversation in a language you do not speak. This requires no source and settles the matter.
Two. The coefficients sum to exactly one, which tells you the formula decomposes a single inferred quantity into shares rather than describing how much meaning each channel carries.
Three. The figure attached to body language came from still photographs, on every description we obtained, and a headshot contains no body language at all.
Four. Neither study examined all three channels together, and the combining sentence appears in a discussion section without showing how it was done.
Five. The author has stated the equations do not apply unless a communicator is discussing feelings or attitudes, which excludes nearly every business conversation.
Our Grades For These Claims
Applying the scheme from the first article in this series, and the grades here are lower than usual.
Grade A for our own arithmetic, which needs no source: the coefficients are what they are and their implications follow.
Grade B for the quoted sentence, which two independent non-academic sources reproduce identically and which we did not verify against the paper.
Grade C for the study descriptions, which reach us only through secondary accounts, though they agree in detail.
Grade C for the author's stated limitation, from an encyclopedia rather than from his own writing, which we did not obtain.
Grade D for the critical literature, which reaches us as titles in a reference list on a non-academic page.
Our position: the arithmetic is decisive and the historical account is secondhand, and we have kept those separate throughout.
A Note On Method
Everything here is verified to August 2026.
We did not obtain either 1967 paper or any abstract of one.
The quoted sentence reaches us from two non-academic sources that reproduce it identically[2][3].
The study descriptions come from an encyclopedia[1] and two specialist websites[2][3], all flagged at every use.
We obtained citation confirmations from a peer-reviewed book chapter[4], which is our only academic source and confirms only that the papers exist.
The critical literature reaches us as titles in a reference list on a non-academic page[5], and we obtained none of those works.
All arithmetic is ours. This article discusses research on communication and is not communications, sales or training advice.
The Claim As Taught
What business audiences are told.
The version in circulation is that communication is seven percent words, thirty-eight percent tone of voice, and fifty-five percent body language, often compressed to the claim that communication is ninety-three percent nonverbal.
A paper in the critical literature is titled "Communication is 93% nonverbal: An urban legend proliferates"[5], which establishes at least that the claim circulates in that form.
Four observations, ours.
The claim is quantitative and precise, which is most of its appeal. A vague statement that delivery matters would not be memorable and would not be repeatable in a slide.
It is also actionable in a specific direction, telling a listener that effort spent on wording is nearly wasted relative to effort spent on presence.
And it is attributed, which is what distinguishes it from folk wisdom and gives it standing in a room of professionals.
Which is why it is worth the trouble of tracing. An unattributed rule of thumb invites scepticism; a cited one does not, and that is precisely the failure mode the last several articles have documented.
Two further reasons this one is worth a full article, ours. It is older than any other claim in this series, having circulated since 1967, which makes it a good test of how long a misreading can last.
And it is unusually easy to check. Most of the claims in these articles required obtaining a paper; this one can be dismantled by asking what it predicts about a phone call, which is the part we recommend readers keep.
The Two Papers
Where the figures come from.
Mehrabian, A., and Wiener, M. (1967), Decoding of Inconsistent Communications, Journal of Personality and Social Psychology, 6(1), 109–114, DOI 10.1037/h0024532[2][4].
Mehrabian, A., and Ferris, S. R. (1967), Inference of Attitudes from Nonverbal Communication in Two Channels, Journal of Consulting Psychology, 31(3), 248–252, DOI 10.1037/h0024648[2][4].
We obtained neither.
Four observations, ours.
There are two papers, not one, published in different journals in the same year, and the rule requires combining them.
The titles are informative and neither mentions communication in general. One is about decoding inconsistent communications and the other about inferring attitudes.
Neither title contains the words meaning, understanding, persuasion or effectiveness, which are the things the popular rule is used to make claims about.
And the second title specifies two channels, which is the detail the whole three-channel rule depends on and contradicts.
The Sentence
The origin of the numbers, quoted by two independent non-academic sources in identical terms.
The paper is reported as stating: "It is suggested that the combined effect of simultaneous verbal, vocal, and facial attitude communications is a weighted sum of their independent effects, with the coefficients of .07, .38, and .55, respectively."[2][3]
Both sources are non-academic, flagged here and at every use, though they reproduce the sentence identically and were written years apart.
Four observations, ours.
"It is suggested that" is the hedge a researcher uses for a conjecture rather than a result. It is not the language of a finding.
The subject is "attitude communications", not communication. The sentence is about inferring how somebody feels, which is the thing both studies measured.
The third channel is "facial", not bodily. The word body does not appear.
And "combined effect" tells you two separate results are being joined, which is the next section.
Where That Sentence Sits
The detail that determines how much weight it can carry.
One source records that the statement "is included in the discussion section of the paper and includes no details about how the results of the two studies were combined."[2]
A second, independently, records that the authors "don't show how they combined the results of the two experiments and arrived at this formula."[3]
Both non-academic, and we did not verify either against the paper.
Four observations, ours.
A discussion section is where authors speculate beyond their data, and readers of research treat it accordingly. It is the least binding part of a paper.
The absence of a stated method for combining is the substantive problem. Two experiments with different subjects, different materials and different designs cannot simply be added, and if there is a principled way to do it, it was not shown.
Two independent sources making the same observation is the strongest thing we have here, since they would have had to both miss the same passage to be wrong.
And this is the whole provenance of the most-repeated statistic in business communication. A hedged suggestion, in a discussion section, without a stated derivation.
Two things that does not establish, ours, because it would be easy to overread. It does not mean the suggestion is wrong, since researchers frequently propose things in discussions that later turn out well.
And it does not mean the underlying studies were poor. The papers appeared in reputable journals and their narrow findings may be entirely sound, which is a separate question from whether one speculative sentence should carry a training industry.
Neither Study Tested All Three
The structural problem, which follows from the titles.
One source states it directly: "neither experiment looked at all three communication channels together."[3] Non-academic, flagged.
On the descriptions we obtained, the first study compared words against vocal tone[1], and the second compared facial expression against vocal tone[1].
Four observations, ours.
Two channels each, with tone appearing in both, which is presumably the hinge on which the combination was attempted.
But a three-way weighting cannot be established from two pairwise comparisons without assumptions, and the assumptions were not stated.
The everyday version of the problem is easy to see. Knowing that A beats B and that C beats B does not tell you how A and C compare, still less what share of a total each carries.
And the encyclopedia records the second study's result as a ratio of 3 to 2 between its two channels[1], which is 60 against 40, not 55 against 38. The published coefficients are not simply the study's own ratio, and how they were adjusted we do not know.
What Was Actually Presented
The materials, on the descriptions we obtained.
The first study used single spoken words. Our sources give examples including "dear," "terrible," "honey," "brute" and "maybe"[1], read in a tone that contradicted the word's meaning.
The second used black and white photographs of facial expressions together with tape-recorded vocal tone[1][2].
All descriptions are from non-academic sources, flagged.
Four observations, ours.
Single words is the detail that most changes the meaning of the finding. The verbal channel was given the smallest possible amount to work with: one word, with no argument, no evidence and no structure.
Of course tone dominated. A single word carries almost no information and a tone of voice carries a good deal, and the result is close to a definition rather than a discovery.
Which means the seven percent figure is not a measurement of what words can do. It is a measurement of what one word did, against a tone contradicting it.
And a business conversation is not one word. A quote, a proposal or an explanation is hundreds of words carrying an argument, which is the condition the study specifically did not test.
One way to see how much the design constrains the result, ours. Ask what the seven percent would have been if the verbal channel had been a paragraph rather than a single word, holding everything else fixed.
Nobody knows, because nobody ran that study on our sources' account. But the direction is not in doubt, and a figure obtained under the least favourable possible condition for one channel is not a general constant.
Headshots Are Not Body Language
The observation that should end the popular version on its own.
One source notes: "Because the study used static images to assess facial expressions and the photographs were headshots, conclusions can fairly speak to static facial expressions, but not body language."[2] Non-academic, flagged.
Four observations, ours.
A headshot photograph contains no posture, no gesture, no stance, no proximity, no movement and no hands. Those are what people mean by body language.
So the fifty-five percent figure, whatever it measures, was never about body language, and the entire content of a great many training courses rests on a substitution of one word for another.
The substitution is not innocent, because it changes what a listener is told to do. Facial expression is largely involuntary; posture and gesture are coachable, and only the second supports a training product.
And the images were static and monochrome, which removes even the dynamic component of facial expression that a real conversation carries.
One consequence for how the finding should be read, ours. A still photograph is closer to a symbol than to a face, in that it presents one fixed configuration to be identified rather than a stream to be interpreted.
Which may make the visual channel look either stronger or weaker than life, and we cannot say which. A posed expression is unusually legible; a frozen one loses the timing that carries much of the meaning, and those pull in opposite directions.
The Word In The Title
The scope condition, which is stated in the first paper's title and dropped everywhere else.
The title is "Decoding of Inconsistent Communications." One source describes the finding as applying to "judgements of attitude from 'inconsistent' messages involving single words spoken with intonation."[3]
Four observations, ours.
Inconsistent means the channels disagreed: a positive word in a hostile tone, or the reverse. That is the condition the study created deliberately.
Which makes the finding sensible and much narrower than advertised. When the words and the tone contradict each other, people believe the tone, and most readers would have guessed that.
It says nothing whatever about the ordinary case where the channels agree, which is what happens in almost every conversation anybody has.
And that scope condition sits in the title of the paper, which is the same failure this series documented in the eighty-third article, where the qualification was in the abstract. Here it is one word further forward and still did not survive.
The Author's Own Limitation
What the researcher has said about the use of his figures.
An encyclopedia records the position as: "unless a communicator is discussing their feelings or attitudes, these equations are not applicable."[1]
A further source records more generally that he "has always been clear that the results of these studies are limited in application."[3]
Both non-academic, flagged, and we did not obtain his own statement of this.
Four observations, ours.
The stated condition is discussing feelings or attitudes, which is what both studies measured and is a narrow slice of what people talk about.
Nearly every business conversation falls outside it. A price, a specification, a deadline, a contract term or a diagnosis is not a communication about feelings, and on the author's own limitation the equations do not apply.
Which puts the sales-training use in an odd position. The one setting where the rule is most taught is a setting its author has excluded, on the account we obtained.
Two readings of that are available and we prefer the charitable one, ours. A researcher publishing a narrow finding is not responsible for what an industry does with it, and issuing clarifications is the most any author can do.
The failure sits with everyone in between. A claim passes through textbooks, courses and slide decks, and each pass drops a qualification, which is the mechanism this series has now documented in five consecutive articles.
And we would rather have this from him directly. An encyclopedia paraphrasing an author's clarification is two steps from the source, and this claim deserves better sourcing than we could give it.
The Coefficients Sum To Exactly One
Our own arithmetic, and it requires no source because the coefficients are stated.
0.07 plus 0.38 plus 0.55 equals 1.00 exactly.
Four observations.
That is not a coincidence and it is not a rounding accident. A set of weights summing to one is a decomposition, dividing a single quantity into shares.
Which tells you what the formula is for. It apportions one inferred judgement among its inputs, in the way a variance decomposition apportions variation among sources.
And a decomposition of an inference about attitude is not a statement about how much meaning each channel carries, because those are different quantities and only the first was measured.
The distinction is the whole article in one line. The formula answers: when someone judges how I feel, what are they using? The popular version answers: how much of what I say gets through? Nothing measured the second.
What Kind Of Formula That Is
Drawing the consequence. Ours.
Four observations.
Because the weights are shares of a fixed total, they must trade off against each other by construction. If one rises another falls, whatever happens in the world.
So the formula cannot tell you that improving your words has little effect, which is the use it is put to. A share is not an effect size.
An analogy makes it plain, and it is ours. If a firm's costs are 7 percent rent, 38 percent materials and 55 percent labour, that is a true decomposition and it does not mean cutting rent is worthless. It means rent is a small share of costs.
And the eighty-third article's lesson applies exactly. A share of a total and an effect of a change are different questions, and a number answering the first gets read as answering the second.
Two ways to notice this error in the wild, ours, since it is not confined to this rule. Watch for weights that sum to one hundred percent, which almost always means a decomposition and almost never means a set of effect sizes.
And ask what would happen if you improved every component at once. Under a decomposition the shares are unchanged and the total may move a great deal, which is the tell that the shares were never about magnitude.
What The Popular Version Implies
Our own arithmetic, taking the popular claim literally and asking what share of communication each medium delivers.
Face to face, all three channels: 100 percent.
A telephone call, words and tone only: 45 percent. A voice note: 45 percent.
Video with no sound: 55 percent.
A written contract: 7 percent. An email: 7 percent. A text message: 7 percent.
And standing in a room where you understand none of the words spoken: 93 percent.
Four observations.
A signed contract conveys seven percent of its meaning, on this claim, which is a conclusion no reader of this publication will accept.
A silent video conveys more than a phone call, at 55 against 45, which anyone who has done both knows is false.
These are not strawmen. They are the claim applied to cases, and a rule that cannot be applied to cases is not a rule.
And a defender cannot rescue it by narrowing after the fact, ours. Restricting the rule to face-to-face conversations about feelings leaves it true and useless, since it then says nothing about the meetings, proposals and negotiations it is invoked for.
And the last line is the one that settles it.
Before that, one anticipated objection, ours. A defender might say the rule was never meant to be applied to written media at all, which is fair and is precisely our point: the scope conditions were dropped, and without them the rule cannot say where it stops applying.
A rule with no stated boundary gets applied everywhere. If seven percent is a fact about communication, it is a fact about email, and if it is not a fact about email then it was never a fact about communication.
The Test Everyone Has Already Run
The argument that needs no sources at all. Ours.
If words carry seven percent, then a person watching a broadcast in a language they do not speak receives the tone and the visual channel, which is 93 percent of the communication.
Four observations.
Everyone reading this has run that experiment, in a hotel room, on a flight, or in a country whose language they do not have.
What you get is tone, mood, and roughly nothing else. You cannot tell whether the news is about an election, a flood or a football result.
So the claim is refuted by universal experience, and this does not depend on anything in the preceding sections being right. Even if our secondhand account of the 1967 studies were wrong in every particular, the arithmetic would stand.
Which is why we were willing to write this article on weak sourcing. The historical part is secondhand and the decisive part is not sourced at all, because it does not need to be.
Two properties of that kind of argument make it worth reaching for generally, ours. It cannot be undermined by finding a better source, since it rests on the claim's own terms rather than on evidence about the world.
And it is available to any reader immediately, which matters for a business audience that will not be chasing a 1967 journal. Working out what a statistic implies in a case you know well costs a minute and settles most of them.
Why This Costs Money
The reason a business publication covers it. Our own arithmetic, applying the claim's own weights.
If the rule held, a ten point improvement in each channel would move the total by: words, 0.7 points; tone, 3.8 points; body language, 5.5 points.
Four observations.
The claim implies that improving what you say is worth about one eighth of improving how you look, the exact ratio being 7.9 to 1.
That ratio is what the training industry is selling. A course on posture, eye contact and vocal delivery is justified by it, and a course on structuring an argument is implicitly devalued by the same number.
For a small firm the cost is not only the course fee. It is the proposal that was polished in delivery and left vague in substance, because the rule said the substance was seven percent.
And the arithmetic gives no support to the opposite error either. Nothing here says delivery is unimportant, which the next sections address.
One specific cost worth naming for a professional firm, ours. An accountant, a lawyer or an adviser is selling the content of what they say, and a rule valuing that at seven percent is telling them their actual product is a rounding error.
Which is obviously false and is nonetheless taught in rooms full of professionals. The advice a client pays for is the words, and the manner is how they decide whether to trust them, which is a different function rather than a larger share.
The Critical Literature
What has been published about this, though we obtained none of it.
A reference list on a non-academic page names: Lapakko, D. (1997), Three cheers for language: A closer examination of a widely cited study of nonverbal communication, Communication Education; Lapakko, D. (2015), Communication is 93% nonverbal: An urban legend proliferates, Communication and Theater Association of Minnesota Journal; Patterson, M. L., Fridlund, A. J., and Crivelli, C. (2023), Four misconceptions about nonverbal communication, Perspectives on Psychological Science; and Hartwig, M., and Bond, C. F., Why do lie-catchers fail? A lens model meta-analysis of human lie judgments, Psychological Bulletin[5].
These are titles in a reference list on a non-academic page, and we obtained no part of any of them.
Three observations, ours.
The titles establish that the objection is not ours and is not new, with the earliest dating to 1997 and the most recent to 2023 in a leading psychology review journal.
The 2015 title uses the phrase "urban legend" and the 2023 title uses "misconceptions", in peer-reviewed venues, which is unusually direct language for academic titles.
And we grade this at the bottom of our scale because a title tells you the question and not the finding, which is the discipline the eighty-first article set and we are keeping to.
What Actually Survives
Our reading, stated directly.
Five statements.
The popular version is false, on arithmetic that needs no sources: it predicts you understand 93 percent of a language you do not speak.
The coefficients decompose an inference about attitude, and their summing to exactly one identifies them as shares rather than effects.
The visual figure came from still headshot photographs, on every description we obtained, which contain no body language.
The verbal channel was tested with single words, against a contradicting tone, which is close to the least favourable condition possible.
And the author's own stated limitation excludes communications that are not about feelings or attitudes, which is most of business.
Taken together those five say something narrower than a debunking and more useful. There is a real finding here about inconsistent messages, obtained under artificial conditions, which was extended by one speculative sentence into a claim about communication in general.
Delivery Still Matters
Because this article could be read as licence to ignore how you present, which would be wrong. Ours.
Four observations.
Nothing above shows that tone and manner are unimportant. It shows that a particular number does not measure what it is used to measure.
The narrow finding is real and useful on its own terms. When your words and your manner contradict each other, people believe your manner, which is worth knowing and is what the studies actually examined.
That has a direct business application which the rule obscures. Saying you are happy to help while visibly resenting it communicates the resentment, and no amount of wording fixes it.
And the honest version is more useful than the false one. Consistency between what you say and how you say it is the finding, rather than a ratio telling you to neglect one of them.
What Is Actually Known
The boundary of our own knowledge, stated plainly. Ours.
Four observations.
We do not know what share of meaning any channel carries, and we doubt the question is well posed, since it would depend entirely on the message.
A phone call and a site visit differ, and so do a contract and a conversation. Any real answer would be a function of what is being communicated, not a constant.
The 2023 review title suggests the field has views on this[5], and we did not obtain it, which is the largest gap in this article.
And that is the honest ending. We can show the rule is wrong and we cannot tell you what is right, which is a less satisfying article and a more accurate one.
Two things we would say in place of a number, ours and untested. The channel that matters most is the one carrying the information the other party lacks, which varies by conversation and is knowable in each case.
And mismatch is more informative than any channel alone. When manner and content disagree, something is worth asking about, which is the narrow finding turned into a practice rather than a percentage.
One place that practice earns its keep, ours and untested. A client saying a price is fine in a tone that says otherwise is telling you the deal is at risk, and the useful response is to ask rather than to proceed on the words.
Against Overcorrecting
A caution we take from one of our own sources, and which applies to this article. Ours, prompted by theirs.
A personal essay re-examining the model argues that "swinging to the extremes, either wholly accepting the 7-38-55 rule as absolute or completely debunking Mehrabian's model as a myth, overlooks the subtleties of communication research", and that the right response is to go back to the original papers rather than reading the opinions of others[6].
A personal essay, not an academic source, flagged.
Four observations, ours.
The point stands regardless of where it comes from, and it is uncomfortable for us specifically. We could not go back to the original papers, which is exactly what the essay says a reader should do.
So we have tried to be precise about which claim we are refuting. The popular three-channel rule applied to communication in general is false on arithmetic alone. The underlying studies may be perfectly sound work on a narrow question.
Those are different verdicts and a reader is entitled to both. A finding can be modest, well conducted and correct, and still be the origin of a false claim, without anyone having done anything wrong except summarise.
And the essay's prescription is the one this series has followed for eighty-eight articles. Read the paper rather than the account of it, which we recommend here while admitting we could not do it.
Which is worth one further word of honesty, ours. A reader with access to a university library could settle in an afternoon what we could not settle at all, and if anyone does so and finds our secondhand account wrong, the historical sections here should be discarded and the arithmetic kept.
Why This One Spread So Far
Our reading of the mechanism, offered as reasoning rather than evidence.
Four observations.
The numbers are memorable and oddly specific, and seven, thirty-eight and fifty-five sound like measurements rather than estimates, which is most of their authority.
The claim also flatters the person repeating it. Telling a room that most of communication is invisible positions the speaker as someone who sees what others miss.
And it is commercially load-bearing. An industry exists to teach the fifty-five percent, and no industry exists to teach that the figure came from headshots.
One further reason, ours, and it is the least cynical. The claim contains something true. Manner does matter, people do read more than words, and a listener who has experienced that will find the number confirms what they already believe.
Which makes it very hard to dislodge, because refuting the number appears to deny the experience, and it does not. The experience is real and the number is not a measurement of it.
Which is the general condition under which a claim outlives its evidence, and this series has now documented it repeatedly. A number is repeated because somebody benefits from repeating it, and checked only by people who do not.
Assessing A Training Course
The practical use. Ours, and not training advice.
Four points.
If a course opens with the 7-38-55 figures, ask where they come from. A trainer who can answer knows their material and one who cannot has read a slide.
That is not a reason to dismiss the course. Good delivery training may be perfectly worthwhile and justified badly, which is a common and forgivable situation.
It is a reason to ask what the course claims to improve and how it would know. A programme promising better outcomes should be able to say what it measures, and the eighty-sixth article's requirements apply.
And the substitution to watch for is the specific one this article documents. A course teaching posture and gesture on the authority of a study about still photographs of faces has changed the subject, and it is worth noticing whether anyone in the room knows.
One thing we would not do, ours, and it is worth saying because the temptation is real. We would not raise this in the room to embarrass a trainer, who is almost certainly repeating in good faith what they were themselves taught.
The useful version is private and afterwards. Ask whether the programme's claims can be stated without the statistic, and if they can, the statistic was decoration and the course may still be worth its fee.
Bibliographic Note
The series keeps a count, and this article produced two.
The second paper's issue number is given as 31(3) by most of our sources[2] and as 31, no. 6 by a peer-reviewed book chapter[4].
And its page range is given as 248–252 almost everywhere and as 248–452 in one reference list[3], which is a transposed digit.
Three observations, ours.
The issue-number disagreement is the more consequential, since a reader searching by issue would look in the wrong place, and the two cannot both be right.
The page range with 452 in it is self-evidently wrong, because a five-page article cannot run to page 452 from 248.
That brings the running count of bibliographic variants across this series to thirty-four.
Two at once is the most any single article has produced, and the reason is worth noting. These are 1967 papers with no reliable digital record, so every citation of them was copied from another citation rather than from the source.
Which is the same mechanism as the article's main subject, operating on the bibliography instead of the finding. A chain of copies drifts, and nothing in the chain has an incentive to check.
What To Do
Do not use the 7-38-55 figures. Taken literally they predict a written contract conveys seven percent of its meaning and that you understand 93 percent of a language you do not speak.
Notice that the coefficients sum to exactly one. That makes them shares of a single inferred judgement, and a share is not an effect of a change.
Keep the narrow finding, which is real and useful. When your words and your manner contradict each other, people believe your manner.
Do not conclude that delivery is unimportant. This article refutes a number, not a practice.
Do not neglect substance on the authority of the rule. On the claim's own weights it says words are worth about one eighth of appearance, at a ratio of 7.9 to 1, and that ratio is what devalues the work of getting a proposal right.
Ask a trainer where the figures come from, and treat the answer as information about the trainer rather than about the topic.
Watch for the word substitution. The visual figure came from still headshot photographs on every description we obtained, and headshots contain no posture, gesture or stance.
And apply the test to other numbers. Working out what a statistic implies in a case you know well is the cheapest check available, and it settled this one without any sources at all.
The Limits Of This Analysis
Several caveats matter, and the sourcing caveat is the largest this series has published. This article discusses research on communication and is not communications, sales or training advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain either 1967 paper, or any abstract of either, and consequently every description of what those studies did, presented, or found reaches this article through secondary sources, four of which are non-academic. The quoted sentence containing the coefficients is reproduced identically by two independent non-academic sources and was not verified by us against the paper; if their reproduction is inaccurate, the sections resting on its exact wording would need revision. The author's stated limitation on the equations reaches us from an encyclopedia paraphrasing him rather than from his own writing, which is two steps from the source and which we would not normally accept for a claim carrying this much weight. The critical literature reaches us as four titles in a reference list on a non-academic page, and we obtained no part of any of those works, so we report that objections exist in peer-reviewed venues and nothing about what they establish; the 2023 review is the paper we would most want and did not get. All arithmetic is ours: the coefficients are as reported, and every application of them, every medium in the implications table, and the ratio of 7.9 to 1 are our own constructions demonstrating what the claim entails. That arithmetic does not depend on any of the historical account being correct, which is why we were willing to publish on this evidence, and readers should weigh the two parts of this article separately. And we cannot tell you what the correct figures are, because we doubt the question has a constant answer and we obtained nothing that would settle it.
Frequently Asked Questions
Where does the 7-38-55 rule come from?
What is the simplest reason it cannot be right?
Was the study really about body language?
What does it mean that the coefficients sum to one?
Did the author dispute the popular use?
So delivery does not matter?
What are the correct percentages then?
References
- Encyclopedia entry on the researcher, recording that the 7 percent, 38 percent, 55 percent rule is based on two studies reported in 1967 papers titled Decoding of Inconsistent Communications and Inference of Attitudes from Nonverbal Communication in Two Channels; that both studies focused on the communication of positive or negative emotions using single spoken words such as "dear" or "terrible"; that the first compared the relative importance of the semantic meaning of the word with the tone of voice and found the latter much more influential; that the second dealt with facial expressions shown in black-and-white photographs and vocal tone heard in a tape recording, and found the relative contributions of the two channels had a ratio of 3 to 2; and that unless a communicator is discussing their feelings or attitudes, these equations are not applicable. Note: an encyclopedia, NOT an academic source, flagged at every use. Our source for the study materials and for the author's stated limitation, which reaches us as an encyclopedia's paraphrase rather than from his own writing. en.wikipedia.org
- Specialist website essay on the misrepresentation of this research, reproducing the sentence reported to appear in the 1967 paper: that it is suggested the combined effect of simultaneous verbal, vocal and facial attitude communications is a weighted sum of their independent effects, with the coefficients of .07, .38 and .55 respectively; recording that this statement is included in the discussion section of the paper and includes no details about how the results of the two studies were combined; recording that the speakers were instructed to vary tone of voice to communicate liking, neutrality and dislike, and that female models displayed facial expressions to the same end; and observing that because the study used static images to assess facial expressions and the photographs were headshots, conclusions can fairly speak to static facial expressions but not to body language. It gives the second paper as Journal of Consulting Psychology, 31(3), 248. Note: a specialist website, NOT an academic source, flagged at every use. One of two independent sources reproducing the key sentence identically. We did not verify the quotation against the paper. andrewquagliata.com
- Specialist website account of the same studies, reproducing the same coefficients sentence; recording that the authors do not show how they combined the results of the two experiments to arrive at the formula; stating that neither experiment looked at all three communication channels together; describing the first study's finding as applying to judgements of attitude from inconsistent messages involving single words spoken with intonation, based primarily on the tonal component; and stating that the author has always been clear that the results of these studies are limited in application. Its reference list gives the second paper as Journal of Consulting Psychology, 31, 248–452. Note: a specialist website, NOT an academic source, flagged at every use. The second independent reproduction of the key sentence. Recorded also as a bibliographic variant: its page range transposes a digit, since a paper beginning at 248 in a five-page span cannot end at 452. speakingaboutpresenting.com
- Peer-reviewed book chapter on business pitching, DOI 10.1007/978-1-137-51233-8_4, whose notes confirm Albert Mehrabian and Morton Wiener, Decoding of Inconsistent Communications, Journal of Personality and Social Psychology, 6, no. 1 (May 1967), 109–114, and Albert Mehrabian and Susan R. Ferris, Inference of Attitudes from Nonverbal Communication in Two Channels, Journal of Consulting Psychology, 31, no. 6 (June 1967), 248–252. Note: our only academic source, and it confirms only that the papers exist and where. Recorded also as a bibliographic variant: it gives the second paper's issue as number 6 where our other sources give number 3. link.springer.com
- Non-academic page on the rule, whose reference list names Lapakko, D. (1997), Three cheers for language: A closer examination of a widely cited study of nonverbal communication, Communication Education; Lapakko, D. (2015), Communication is 93% nonverbal: An urban legend proliferates, Communication and Theater Association of Minnesota Journal; Patterson, M. L., Fridlund, A. J., and Crivelli, C. (2023), Four misconceptions about nonverbal communication, Perspectives on Psychological Science; and Hartwig, M., and Bond, C. F., Why do lie-catchers fail? A lens model meta-analysis of human lie judgments, Psychological Bulletin; and recording that the rule was constructed by combining findings from two separate 1967 studies. Note: a non-academic page, flagged at every use; titles only. We obtained no part of any work named here, and report that objections exist in peer-reviewed venues and nothing about what they establish. The 2023 review is the paper we would most want and did not obtain. counseloraid.com
- Personal essay in two parts re-examining the model, confirming both 1967 citations with their DOIs, 10.1037/h0024532 and 10.1037/h0024648; describing the second study as investigating the decoding of both consistent and inconsistent communications of attitude conveyed through facial and vocal channels; and arguing that swinging to either extreme, wholly accepting the rule as absolute or completely debunking the model as a myth, overlooks the subtleties of communication research, and that it is important to go back to the original papers rather than reading the opinions of others. Note: a personal essay, NOT an academic source, flagged at every use. Recorded principally for its argument against overcorrection, which we have tried to honour, and because it makes the point about returning to originals that our own sourcing did not permit us to follow. medium.com
This article discusses research on communication and is not communications, sales or training advice. Neither 1967 paper was obtained, nor any abstract of either; every description of what those studies did reaches this article through secondary sources, four of which are non-academic and are flagged at every use. The quoted coefficients sentence was not verified against the paper. All arithmetic is the authors' own and does not depend on the historical account being correct.