Interpersonal & Intercultural Communication

A 1948 paper about wires

Claude Shannon worked at Bell Telephone Laboratories, and in July and October of 1948 he published a two-part paper in the Bell System Technical Journal called A Mathematical Theory of Communication. It contains a diagram you have almost certainly seen: an information source, a transmitter, a channel with noise entering it, a receiver, and a destination.

The paper was about telephone lines. Shannon was solving the problem of how to reproduce at one point a signal selected at another point, and he said explicitly that the meaning of the message was irrelevant to the engineering question he was asking. His concern was fidelity: did the bits that went in come out.

A year later Warren Weaver wrote an accompanying essay suggesting the model might apply more broadly, and the diagram escaped into psychology, education, management training and communication textbooks. It has been drawn on whiteboards ever since as though it described people talking. It does not, and the gap between what it describes and what happens between two people is the subject of this lesson.

Key idea: A model is a tool for noticing. Every model makes some features of an interaction visible and hides others, so choosing one is choosing what you will fail to see.

What the linear model gets right

It would be easy to dismiss the transmission model, and that would be a mistake, because three of its ideas are genuinely useful.

Noise is real. Shannon meant electrical interference. Communication scholars extended the term to physical noise, such as a loud room; physiological noise, such as being ill or exhausted; psychological noise, such as preoccupation or anxiety; and semantic noise, where a word or jargon term blocks understanding. Naming the type of noise is often the fastest route to fixing a breakdown.

Encoding and decoding are distinct steps. What you intend, what you say, and what is understood are three different things, and they can each fail separately. That distinction is the reason misunderstanding is not automatically anyone's fault.

Channel matters. The same words land differently in a text message, a phone call and a face-to-face conversation, because each channel carries different amounts of additional information. Module 6 returns to this in detail.

So the model earns its place as a vocabulary. The trouble begins when it is treated as a picture of what people do.

Four things the transmission model cannot show

It has no feedback. The arrow runs one way. In a real conversation the listener is responding continuously, with nods, frowns, small sounds and shifts of posture, and the speaker is adjusting to those responses while still speaking. Take a moment to notice this the next time you explain something to someone whose face goes blank. You will change what you are saying before you finish the sentence.

It treats meaning as cargo. The diagram implies that a meaning is loaded up, shipped, and unloaded intact. But words do not carry meanings; they prompt meanings in a receiver who already has a history. The same sentence, spoken to two different people, produces two different meanings, and neither of them is the one that was sent.

It has no relationship in it. The boxes are interchangeable. In reality, who is speaking to whom is often the most important information in the exchange. Your manager saying that a report needs work is not the same message as a colleague saying it, even with identical words.

It has no history and no future. Each transmission is independent. Real conversations are episodes in ongoing relationships, and what was said last week constrains what can be said today.

The point: The transmission model is fine for describing how a signal survives a wire. It is a poor description of two people who are simultaneously interpreting and being interpreted.

The interactive model: adding the loop

Wilbur Schramm's 1954 model repaired the most obvious defect by making the process circular. Both parties encode and decode; feedback runs back to the source; and the process continues in turns.

Schramm added a second idea that is more important than the loop, and it is the one most often skipped. Each participant brings a field of experience, meaning everything they have learned, lived and been socialised into. Communication succeeds to the extent that those fields overlap. Where they do not overlap, the same word activates different meanings and neither party has any way to notice.

Think about what that predicts. Two software engineers who trained at the same company can communicate very efficiently in a shorthand nobody else follows. The same two people talking to a client have to work much harder, not because the client is slower but because the overlap is smaller. When you find yourself thinking that someone is being obtuse, the fields-of-experience model asks a more productive question: what do they know that I do not, and what am I assuming that they never learned?

The remaining limitation is that the interactive model is still a turn-taking model. Person A sends, person B receives and sends back. Real conversation is not that tidy.

The transactional model: everything at once

Dean Barnlund's transactional model, published in 1970, makes the decisive move. There is no sender and no receiver. Both people are sending and receiving simultaneously and continuously, and each is a participant rather than occupying a role that alternates.

Three consequences follow, and they are the reason this is the model most interpersonal research now assumes.

Communication is simultaneous. While you are speaking, you are reading the other person's face and adjusting. While they are listening, they are communicating to you, whether or not they intend to. The exchange has no clean turns.

Communication constitutes relationships rather than merely occurring inside them. A friendship is not a container that conversations happen in. It is made out of the conversations. Change the pattern of communication and you have changed the relationship, because there is nothing else it is made of.

Communication constitutes selves. Who you are in a given interaction is produced partly by that interaction. You are a slightly different person with your grandmother than with your closest friend, and not because you are being false with either.

	Transmission	Interactive	Transactional
Shape	One direction	Circular, in turns	Simultaneous and continuous
Roles	Sender and receiver, fixed	Alternating	Both at once, no fixed roles
Where meaning is	In the message	In the overlap of experience	Created between participants
Relationship	Absent	Implied	Central and constituted by the process
Best for	Signal fidelity, mass messaging	Explaining misunderstanding	Ongoing personal relationships
Blind spot	Almost everything human	Simultaneity, identity work	Hard to diagram, harder to measure

Two levels in every message

Paul Watzlawick, Janet Beavin and Don Jackson, working with families at the Mental Research Institute in Palo Alto, published Pragmatics of Human Communication in 1967 and made a distinction that is worth more than most of the models.

Every message carries a content level, which is what is literally said, and a relationship level, which is a statement about how the speaker sees the relationship and how the message should be taken. The relationship level is usually carried nonverbally and is usually not discussed.

Consider the sentence: could you close the window. The content is a request about a window. The relationship level, depending on tone, timing and who is speaking, could be a courtesy, an order, a complaint about the other person's inattention, or a test of whether the other person will comply. Most persistent conflicts in close relationships are fought on the relationship level while both parties argue about the content, which is why resolving the content question so often fails to end the argument.

The same authors introduced punctuation: participants divide an ongoing sequence into cause and effect differently, and each version feels obviously correct from inside. One partner withdraws because the other nags; the other nags because the first withdraws. Neither is describing the sequence wrongly. They are starting the story at different points, and the sequence has no natural beginning.

In short: When an argument keeps returning after the content is settled, the dispute is almost certainly at the relationship level, and it will not be resolved by relitigating the content.

The axiom that is not quite true

The best known line from that 1967 book is that one cannot not communicate. The argument is that all behaviour in the presence of another has message value, so silence, refusal to engage and leaving the room all communicate.

It is a genuinely useful corrective, and it is also overstated, which is worth saying plainly because it is repeated as settled doctrine in a great many textbooks. Michael Motley argued in 1990 that the axiom collapses a real distinction: behaviour that a receiver interprets is not the same thing as communication, which normally implies some encoding on the sender's side. If you yawn because you are tired and I decide you are bored with me, I have made an inference from your behaviour. Calling that communication makes the term cover everything and therefore explain nothing.

The defensible version is narrower and more useful: in the presence of another person, your behaviour is available for interpretation whether you intend it to be or not, and you do not control what is inferred. That claim survives, and it still supports the practical advice people usually draw from the axiom.

Working a case with each model

Here is a small scene. Ana sends her flatmate Ben a message: the dishes are still in the sink. Ben reads it, feels criticised, and replies: I said I would do them.

Through the transmission model you look for noise and encoding failure. The message was terse and text has no tone, so semantic and channel noise are candidates. The advice that follows is to be clearer.

Through the interactive model you look at fields of experience. Ana grew up in a household where stating a fact was neutral. Ben grew up in one where stating a fact about an undone chore was always a rebuke. The overlap is small on exactly this point. The advice is to make the intent explicit, because you cannot assume shared coding.

Through the transactional model you notice that this exchange is not an isolated event. It is the fourteenth time, it is happening in a relationship in which who does the chores is an unresolved question, and the message is doing relationship-level work whether Ana meant it to or not. The advice is different: the content problem is dishes, but the exchange is about fairness and standing, and no amount of clearer wording about dishes will settle it.

Three models, three diagnoses, three different pieces of advice. Only the third is likely to help, and you would not have reached it from the first.

Common misconceptions

"Shannon's model was a bad model of human communication." It was an excellent model of the problem it addressed, which was signal transmission over a noisy channel. The error was made by the people who applied it to conversation, not by Shannon, who said the semantic question was outside his scope.

"Better communication means transmitting more clearly." Clarity helps with content problems. It does nothing for relationship-level disputes, and increasing the volume and precision of the content message in a relationship-level conflict usually makes things worse.

"Meaning is in words." Meanings are in people. Words are prompts that trigger meanings assembled from the receiver's history, which is why identical wording produces different understandings and why dictionaries do not settle arguments.

"You cannot not communicate." Stated flatly, this is contested rather than established. Behaviour is always available for interpretation, which is the part that holds up; whether unintended behaviour counts as communication depends on a definition that scholars genuinely disagree about.

Where this leaves us

Shannon's 1948 diagram was engineering, and its migration into the social sciences carried assumptions that do not survive contact with two people talking: no feedback, meaning treated as cargo, no relationship, no history. Its useful residue is real, though, in the vocabulary of noise, of encoding and decoding as separable steps, and of channel effects. Schramm added feedback and the fields of experience whose overlap determines whether understanding is possible. Barnlund's transactional model added simultaneity and the recognition that communication constitutes relationships and selves rather than passing between pre-existing ones. Watzlawick and colleagues added the content and relationship levels of every message, and punctuation, which explains why two people describing the same recurring argument each sound entirely reasonable. The axiom that one cannot not communicate is worth holding in its narrower form: your behaviour is available for interpretation and you do not control the inference. What controls that inference is the subject of the next lesson.

Sources

1. University of Minnesota Libraries Publishing. (2016). Communication: History and forms. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. Encyclopaedia Britannica. (n.d.). Communication. britannica.com
3. Shannon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27(3), 379-423 and 27(4), 623-656.
4. Barnlund, D. C. (1970). A transactional model of communication. In K. K. Sereno & C. D. Mortensen (Eds.), Foundations of communication theory. Harper & Row.
5. Watzlawick, P., Beavin, J. H., & Jackson, D. D. (1967). Pragmatics of human communication. W. W. Norton.
6. Motley, M. T. (1990). On whether one can(not) not communicate: An examination via traditional communication postulates. Western Journal of Speech Communication, 54(1), 1-20.

An essay written under orders

In 1967 Edward Jones and Victor Harris ran an experiment at Duke University that has been repeated in some form in almost every social psychology course since. Participants read a short essay either supporting or opposing Fidel Castro's regime in Cuba and were asked to estimate the writer's real opinion.

Some participants were told the writer had chosen the position freely. Others were told, explicitly, that the writer had been assigned the position by the experimenter and had no choice at all. That second group had every reason to conclude nothing about the writer's actual views.

They concluded something anyway. Participants who read a pro-Castro essay rated its author as more pro-Castro even when they knew the author had been ordered to write it. The situational information was available, unambiguous and simply not used.

That result is the anchor for this lesson, and it is also, as you will see, more limited and more argued over than the confident version you may have been taught. Both halves matter.

Why this matters: Everything you conclude about what another person meant is an inference, built from a fraction of the available information, using rules that are systematically skewed in directions you can learn to recognise.

Perception has three stages, and each one loses information

Selection. Your senses take in far more than attention can process, so most of it is discarded before you are aware of it. What survives is predictable. Intense stimuli survive, as do repetitive ones and novel ones. So does anything that matches what you expect or want. If you have decided a colleague is unreliable, you will notice the two times she was late this month and not register the eighteen times she was not. Nothing dishonest happens; the non-events were never encoded.

Organisation. Whatever survives selection gets structured. You group things by proximity, by similarity, and by difference from a background. More consequentially, you fit people and situations into schemata, which are organised mental structures built from experience, and into scripts, which are schemata for sequences of events. A restaurant script tells you what happens after you sit down. A job interview script tells you what kind of question comes third. Scripts make interaction efficient and they make surprises hard to process, which is why an interaction that violates a script feels disproportionately unsettling.

Interpretation. Finally you assign meaning, which for social perception means assigning a cause. That step is called attribution, and it has its own research literature.

Kelley's covariation model, worked

Fritz Heider, in The Psychology of Interpersonal Relations in 1958, described ordinary people as naive scientists who explain behaviour by locating its cause either in the person or in the situation. Harold Kelley formalised how they do it in 1967, proposing that observers use three kinds of information.

Consensus: do other people behave this way toward this target? Distinctiveness: does this person behave this way only toward this target, or toward everyone? Consistency: does this person behave this way toward this target across time and settings?

Take a case. Your colleague Maya spoke sharply to you in a meeting on Tuesday. Two patterns of information lead to opposite conclusions.

Information	Pattern A	Pattern B
Consensus: do others also speak sharply to you?	High. Two other colleagues did this week.	Low. Nobody else does.
Distinctiveness: does Maya speak sharply to others?	High. Only to you.	Low. To everyone.
Consistency: does she do it repeatedly?	High	High
Kelley's conclusion	The cause lies with the target, which is you or your behaviour	The cause lies with Maya

Notice what pattern A demands. High consensus and high distinctiveness point away from Maya's personality and toward something about you or about the situation you create. That is not a comfortable conclusion, and it is exactly the conclusion people are least likely to reach unaided, which is why having the model written down is useful.

Bernard Weiner later added a second scheme with three dimensions that matter for how people react emotionally: locus, internal or external; stability, whether the cause is durable; and controllability. Whether you forgive a late colleague depends far less on whether the cause was internal than on whether you judge it controllable. Sleeping through an alarm and being stuck behind a road closure are both external in one sense and are treated completely differently.

The error, stated precisely

Lee Ross named the pattern in 1977: the fundamental attribution error, the tendency of observers to overweight dispositional explanations and underweight situational ones when explaining another person's behaviour. A more careful term, preferred by many researchers, is correspondence bias, the tendency to infer that a behaviour corresponds to a stable underlying trait even when the situation adequately explains it.

The classic companion claim is the actor-observer asymmetry: you explain your own behaviour situationally and other people's dispositionally. I was short with you because I had been awake since four; you were short with me because you are difficult.

Here is where the honest account diverges from the textbook one. Bertram Malle published a meta-analysis in 2006 covering 173 studies of the actor-observer asymmetry and found that the overall effect was close to zero. It appeared under some conditions, reversed under others, and did not hold as a general law of social perception. That is a large and awkward result for a claim taught as established in a great many introductory courses, and you should know it before you repeat the claim.

The correspondence bias itself has held up better than the asymmetry, but it too is conditional. It is stronger when observers are cognitively busy, when the situational constraint is not salient, and when the observer has little information. It weakens when people are given time, motivation and explicit reason to consider the situation.

Bottom line: Correspondence bias is real and well replicated. The neat symmetric story about actors and observers is not, and Malle's 2006 meta-analysis is the reason to stop repeating it.

Culture changes the pattern

The second qualification is that these tendencies are not uniform across societies, which matters for the intercultural half of this course.

Joan Miller published a study in 1984 comparing explanations given by Americans and by Hindu Indians for everyday behaviours. Americans made more dispositional attributions; Indians made more contextual and situational ones. The most telling detail was developmental: young children in both groups explained events similarly, and the divergence widened with age. Whatever produces the difference is learned rather than built in.

Michael Morris and Kaiping Peng found a related pattern in 1994 by examining newspaper coverage of two comparable mass shootings, one by a Chinese graduate student and one by an American postal worker. English-language coverage emphasised the killers' personalities and dispositions. Chinese-language coverage emphasised situational and relational context, including pressures and circumstances. Same events, systematically different causal framing.

The lesson is not that one culture is right. It is that the attribution style you find natural is a habit acquired somewhere, and when you interact with someone whose habit differs, you will each find the other's explanations subtly unreasonable without being able to say why.

The bias that runs the other way

People are not uniformly harsh in their attributions. For their own outcomes, they are systematically generous. Self-serving bias is the tendency to attribute one's successes to internal factors and one's failures to external ones. You got the job because you interviewed well; you did not get the other one because the process was political.

Thomas Pettigrew described the group version in 1979 and called it the ultimate attribution error. Applied to an out-group, a negative act is attributed to the group's character while a positive act is explained away as an exception, luck, or special circumstances. Applied to one's own group, the pattern reverses. This is a mechanism by which stereotypes survive disconfirming evidence, and it returns in Module 6.

What actually helps, and what does not

The standard advice is to take the other person's perspective. It sounds obviously right and the evidence is not kind to it.

Tal Eyal, Mary Steffel and Nicholas Epley reported a series of twenty-five experiments in 2018 testing whether perspective taking improves the accuracy of judgments about what another person is thinking or feeling. It did not. In most of their studies it produced no improvement, and in several it made accuracy slightly worse, apparently because imagining another mind produces confident guesses rather than correct ones. What did improve accuracy was straightforward: asking the person and listening to the answer. They called this getting perspective rather than taking it.

That finding is worth sitting with, because it inverts a great deal of communication advice. Imagining your way into someone's head increases your confidence without increasing your accuracy. The correction is boring and effective: ask, and treat the answer as data rather than as a prompt for further inference.

Two other habits have better support than imagination. Perception checking is a specific three-part move: describe the behaviour you observed, offer at least two possible interpretations, and request clarification. It works because offering two interpretations forces you to generate an alternative to the first one that occurred to you. And slowing down helps, because correspondence bias is strongest when the observer is cognitively loaded.

Worth holding on to: Getting perspective beats taking perspective. Ask, do not simulate.

Common misconceptions

"People explain their own behaviour situationally and others' dispositionally." This is the actor-observer asymmetry, and Malle's 2006 meta-analysis of 173 studies found the overall effect near zero. The correspondence bias in judging others is well supported; the neat mirror-image version is not.

"Perception is a recording of what happened." Selection discards most of the input before awareness, organisation forces the remainder into existing schemata, and interpretation assigns a cause. By the time you have a memory of an interaction, it has been through three lossy transformations.

"The fundamental attribution error means people ignore situations entirely." They underweight them, conditionally. The bias shrinks when situational constraints are made salient and when observers have time and motivation to think.

"Empathy means imagining how the other person feels." Imagining produces confidence, not accuracy, according to Eyal and colleagues. Asking produces accuracy. The two are often confused, and only one of them works.

"Attribution styles are universal." Miller's 1984 developmental comparison and Morris and Peng's 1994 newspaper analysis both show systematic cross-cultural differences that grow with age, indicating a learned habit rather than a fixed human tendency.

The takeaway

Jones and Harris showed in 1967 that people infer an author's attitude from an essay even when told the position was assigned, which is the founding demonstration of correspondence bias. Perception reaches that point through three lossy stages: selection, which keeps what is intense, repetitive, novel or expected; organisation, which forces the remainder into schemata and scripts; and interpretation, which assigns a cause. Kelley's covariation model gives you a usable procedure for that last step, using consensus, distinctiveness and consistency, and it will sometimes point the finger back at you, which is precisely why it is worth applying deliberately. The correspondence bias is well replicated but conditional, stronger when the observer is busy and the situation is not salient. The actor-observer asymmetry, by contrast, did not survive meta-analysis in 2006. Attribution habits vary across cultures in ways that are learned rather than innate, and self-serving and ultimate attribution errors run in the opposite direction for oneself and one's own group. The most effective correction is not imaginative perspective taking, which twenty-five experiments found does not improve accuracy, but asking the person and taking the answer seriously.

Sources

1. University of Minnesota Libraries Publishing. (2016). Perception. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. Encyclopaedia Britannica. (n.d.). Social psychology. britannica.com
3. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
4. Jones, E. E., & Harris, V. A. (1967). The attribution of attitudes. Journal of Experimental Social Psychology, 3(1), 1-24.
5. Malle, B. F. (2006). The actor-observer asymmetry in attribution: A (surprising) meta-analysis. Psychological Bulletin, 132(6), 895-919.
6. Miller, J. G. (1984). Culture and the development of everyday social explanation. Journal of Personality and Social Psychology, 46(5), 961-978.
7. Eyal, T., Steffel, M., & Epley, N. (2018). Perspective mistaking: Accurately understanding the mind of another requires getting perspective, not taking perspective. Journal of Personality and Social Psychology, 114(4), 547-571.

A door between the kitchen and the dining room

In the early 1950s Erving Goffman spent about a year doing fieldwork on Unst, in the Shetland Islands, for his Chicago doctorate. One of the things he watched most closely was a hotel. Staff moved back and forth between a dining room, where they were deferential and composed and used a particular accent, and a kitchen, where the same people were loud, profane, mocked the guests and dropped the accent entirely. The transformation happened at the door, in about a second, in both directions.

Goffman published The Presentation of Self in Everyday Life in 1956, in a small run from the University of Edinburgh, and in a wider edition in 1959. His argument was that this is not a special feature of hotel work. It is what everyone does everywhere, and the theatrical vocabulary he used for it, front stage and back stage, performance and audience, has been in the field ever since.

The claim that unsettles people is not that we perform. It is Goffman's further point that there is no unperformed self sitting behind the performances waiting to be revealed. What you find backstage is another region with its own norms and its own audience, not the truth.

The core of it: Identity is not a thing you have and then express. It is something produced, maintained and repaired in interaction, which is why communication research treats it as a process rather than a property.

Where a self-concept comes from

Charles Horton Cooley described the looking-glass self in 1902 in three steps: we imagine how we appear to another person, we imagine their judgment of that appearance, and we experience a feeling about ourselves as a result. Note the second step. What shapes you is not what others actually think, which you cannot access, but your guess about what they think, which can be badly wrong and is rarely checked.

George Herbert Mead, whose lectures were published as Mind, Self, and Society in 1934, added structure. He distinguished the I, the spontaneous acting part, from the me, the self as seen through others' eyes. Children learn the me first from significant others, specific people whose views matter, and later from the generalised other, an internalised sense of what people in general expect. That progression is why a five year old behaves for their parent and a fifteen year old behaves for an imagined audience that is not in the room.

Leon Festinger added the comparison mechanism in 1954. Lacking objective standards, people evaluate themselves against others. Upward comparison, against people doing better, can motivate or can deflate. Downward comparison, against people doing worse, tends to protect self-esteem. Which comparison you make is largely a matter of who is available, which is why the composition of the group you spend time in shapes how you feel about yourself more than your actual standing does.

Front stage, back stage, and the definition of the situation

Goffman's dramaturgical vocabulary is precise enough to use as an analytic tool rather than a metaphor.

Front stage is where a performance is given to an audience whose presence shapes it. It has a setting, the physical scene, and a personal front, comprising appearance, which signals status, and manner, which signals the role the performer will play.

Back stage is where the performance is prepared and where the performer can drop it. Access is restricted, and a great deal of interactional work goes into keeping audiences out of it.

Performance teams cooperate to sustain a single definition of the situation. Two parents presenting a united decision to a child are a team, and their private disagreement is backstage information that would spoil the performance if it leaked.

The definition of the situation is the shared working agreement about what is going on and who everyone is in it. It is fragile, and most of the small courtesies of interaction exist to protect it. When someone breaks it badly, by weeping in a business meeting or telling a joke at a funeral, the disruption is out of all proportion to the act, because everyone present must now improvise a new definition in real time.

Face, and the four ways to ask for something

In a 1955 paper Goffman defined face as the positive social value a person effectively claims through the line others assume they are taking in an interaction. Face is not private self-esteem. It is a claim made publicly and sustained cooperatively, and one of the quieter findings in this area is that people work hard to protect each other's face, including the face of people they dislike, because a collapse in anyone's face threatens the whole interaction.

Penelope Brown and Stephen Levinson refined this in 1987 by splitting face in two. Positive face is the desire to be approved of and included. Negative face is the desire to be unimpeded and not imposed upon. Many ordinary acts threaten one or both, and they called these face-threatening acts. Asking a favour threatens the other's negative face. Criticising threatens their positive face. Apologising threatens your own.

Their politeness strategies form a scale you can hear in any workplace. Suppose you need a colleague to send you a file today.

Strategy	What it sounds like	When it fits
Bald on record	Send me the file today.	Emergencies, very close relationships, or clear authority with an accepted right to direct
Positive politeness	You are always so quick with these. Any chance of the file today?	Where closeness and inclusion are the relevant currency
Negative politeness	I am sorry to add to your workload, but would it be possible to have the file today?	Where the imposition is real and the other's autonomy needs acknowledging
Off record	I still cannot get into that folder.	Where you want to leave the other an exit and avoid a recorded request

Brown and Levinson argued that the strategy chosen depends on three variables: the social distance between the parties, the relative power, and the size of the imposition in that culture. That last clause is important, because what counts as a large imposition is not constant, and misjudging it is one of the most common sources of intercultural friction. Stella Ting-Toomey's face-negotiation theory develops this, arguing that patterns of face concern and conflict style vary across cultural groups. Module 5 gives that idea the treatment it needs, including a warning about applying group averages to individuals.

Remember: Most of what looks like politeness ritual is face management, and most requests fail not because the content was unclear but because the strategy misjudged distance, power or imposition.

Self-disclosure, and what it actually predicts

Irwin Altman and Dalmas Taylor proposed social penetration theory in 1973, with the durable image of a personality as an onion. Disclosure has breadth, the number of topics, and depth, how central and risky those topics are. Relationships develop by increasing both, usually breadth first, and de-escalate by reversing the process.

Disclosure is governed by a strong reciprocity norm. If someone tells you something moderately personal, you are under real pressure to match it, and failing to match reads as rejection. This is why an interaction with someone who discloses far more than you can comfortably reciprocate is uncomfortable rather than flattering.

The relationship between disclosure and liking is better established than most claims in this course, and it runs three ways. Nancy Collins and Lynn Miller's 1994 meta-analysis found that people who disclose more are liked more; that people disclose more to those they already like; and that people come to like those to whom they have disclosed. The third of those is the least intuitive and the most useful: revealing something appropriate to someone increases your liking for them, not only theirs for you.

The heavy qualifier is appropriateness. Disclosure that outruns the stage of the relationship is penalised rather than rewarded, and the same sentence that deepens a friendship of two years is alarming on a first meeting. There is no general rule that more openness is better, and advice that says so is ignoring most of the literature.

Joseph Luft and Harrington Ingham's Johari window, from 1955, gives a simple four-cell map: the open area known to self and others, the hidden area known only to self, the blind area known to others but not to self, and the unknown area. Disclosure enlarges the open area by shrinking the hidden one. Feedback from others shrinks the blind one, which is the half people usually neglect.

Many identities, and the gaps between them

You do not have one identity. Henri Tajfel and John Turner's social identity theory distinguishes personal identity, based on individual attributes, from social identity, based on group memberships, and shows that people shift between them depending on which is made salient by the situation. That shift has consequences for intergroup behaviour that Module 6 takes up.

Michael Hecht's communication theory of identity is useful here because it names where the friction sits. It describes four frames: the personal frame, how you see yourself; the enacted frame, what you express in interaction; the relational frame, who you are in a particular relationship; and the communal frame, the identity a group holds collectively. Identity gaps occur when the frames do not line up, and gap size predicts distress reasonably well. Someone whose enacted identity at work is far from their personal identity is not merely uncomfortable; they are doing continuous, effortful maintenance, which is one of the mechanisms behind the emotional labour research in Module 3.

The self-esteem claim that did not hold

Through the 1980s and 1990s, raising self-esteem was treated as a general-purpose intervention, on the assumption that people with higher self-esteem perform better and behave better.

In 2003 Roy Baumeister, Jennifer Campbell, Joachim Krueger and Kathleen Vohs published a review in Psychological Science in the Public Interest examining the actual evidence. Their conclusions were unwelcome. Self-esteem correlates modestly with academic achievement, but the causal arrow appears to run mostly from achievement to self-esteem rather than the reverse. It does not predict better job performance in any strong sense. It does not reduce aggression; if anything, some forms of high but unstable self-esteem are associated with more aggression when threatened. What high self-esteem reliably does is make people feel better and judge themselves more favourably, including in domains where objective measures do not support the judgment.

The point for this course is not that self-esteem is worthless. It is that a widely promoted intervention rested on a correlation whose direction had not been established, and the correction came only when someone examined the whole literature rather than the studies that supported the programme.

So what?: When a self-improvement claim is repeated everywhere, the useful question is not whether the correlation exists but which way the causal arrow was shown to run, and by what design.

Common misconceptions

"Backstage is the real self." Goffman's argument is that backstage is another region with its own audience and norms. The kitchen performance is a performance too. There is no final unperformed layer.

"Impression management is dishonesty." It is the ordinary work of sustaining a definition of the situation, and everyone does it constantly. Dressing appropriately for a funeral is impression management. The interesting question is not whether someone manages impressions but which impression and to what end.

"More self-disclosure always improves a relationship." Only when it fits the stage of the relationship. Disclosure that outruns the relationship is penalised, and the reciprocity norm means mismatched disclosure creates pressure rather than intimacy.

"Face means saving embarrassment, mainly in Asian cultures." Goffman developed the concept studying Americans and Britons. Face concerns are universal; what varies is which acts threaten which kind of face and how much, which is a difference of content rather than of presence.

"Raising self-esteem improves performance." Baumeister and colleagues' 2003 review found the causal arrow runs mainly the other way. Self-esteem is largely an outcome of doing well, not a cause of it.

Summing up

Goffman's Shetland hotel staff, transformed by a doorway, is the founding image for treating identity as something produced in interaction rather than expressed from within. The self-concept assembles from reflected appraisals, which are guesses about others' judgments rather than the judgments themselves, and from social comparisons whose direction depends mostly on who happens to be around. Dramaturgy gives a working vocabulary: front and back stage, setting and personal front, performance teams, and the fragile shared definition of the situation that ordinary courtesy exists to protect. Face is the public claim a person makes and others cooperatively sustain, split by Brown and Levinson into positive face, wanting approval, and negative face, wanting not to be imposed upon, with politeness strategies chosen according to distance, power and the size of the imposition. Social penetration describes relational development through breadth and depth of disclosure under a strong reciprocity norm, and Collins and Miller's meta-analysis established that disclosure and liking are linked in three directions, subject to a heavy appropriateness condition. Identity is multiple, and gaps between personal, enacted, relational and communal frames predict strain. Finally, the widely promoted claim that raising self-esteem improves performance did not survive a full review of the evidence in 2003, which is a useful reminder to check the direction of a causal arrow before acting on a correlation.

Sources

1. University of Minnesota Libraries Publishing. (2016). Communication and the self. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. OpenStax. (2020). Self-presentation and the social self. In Psychology 2e. Rice University. openstax.org
3. Goffman, E. (1959). The presentation of self in everyday life. Anchor Books.
4. Brown, P., & Levinson, S. C. (1987). Politeness: Some universals in language usage. Cambridge University Press.
5. Collins, N. L., & Miller, L. C. (1994). Self-disclosure and liking: A meta-analytic review. Psychological Bulletin, 116(3), 457-475.
6. Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2003). Does high self-esteem cause better performance, interpersonal success, happiness, or healthier lifestyles? Psychological Science in the Public Interest, 4(1), 1-44.

The reference letter that says nothing bad

H. P. Grice gave a set of lectures at Harvard in 1967, published as Logic and Conversation in 1975, and used an example that has been quoted ever since. A professor is asked to write a reference for a student applying for a philosophy job. The entire letter reads, in substance, that the student's command of English is excellent and his attendance at tutorials has been regular.

Nothing negative has been said. No claim in the letter is false. And every reader understands it as a devastating assessment of the student's philosophical ability.

Where did that meaning come from? Not from the words. It came from the gap between what was written and what a cooperative writer would have written if there had been anything better to say. The reader reasons: he must know more, he is expected to say more, so his silence on the central question is itself informative.

That reasoning is the engine of everyday conversation, and this lesson is about it. Most of what people communicate is not stated.

The point: Meaning is generated as much by what a speaker could have said and did not as by what they did say, which is why a transcript of a conversation reliably misrepresents it.

The cooperative principle

Grice proposed that conversation works because participants assume each other to be cooperating toward a shared purpose. From that assumption he derived four maxims, which are not rules of etiquette but expectations that make inference possible.

Maxim	The expectation	What flouting it communicates
Quantity	Be as informative as required, and no more	Saying less than expected implies there is a reason, as in Grice's letter
Quality	Do not say what you believe false or lack evidence for	Obvious falsehood signals irony, sarcasm or hyperbole
Relation	Be relevant	An apparent non sequitur invites the hearer to find the connection, which is how topic changes signal discomfort
Manner	Be clear, brief and orderly; avoid ambiguity	Unusual wordiness or vagueness signals evasion or delicacy

The key move is implicature: a hearer who believes the speaker is cooperating, but who observes an apparent violation, infers an additional meaning that restores cooperation. Ask a friend how the party was and hear that the food was good. You have been told something about the party, and nobody said it.

Two practical consequences follow. First, adding information can change the meaning of what you already said. Telling a colleague that their draft is clear is a compliment; telling them it is clear and grammatically correct is not, because the second item was too obviously worth mentioning. Second, in written and asynchronous channels the maxims are harder to apply, because you cannot check whether an apparent violation was deliberate. Module 6 develops that point.

Words do not contain meanings

C. K. Ogden and I. A. Richards published The Meaning of Meaning in 1923 with a diagram called the triangle of meaning. Its three corners are the symbol, which is the word; the referent, which is the thing in the world; and the reference, which is the thought in a person's head. The crucial feature of the diagram is that the line between symbol and referent is drawn as broken. There is no direct connection between a word and a thing. The link runs only through a mind.

This is why arguments that consist of asserting what a word really means so rarely settle anything. Words have conventional uses, and those conventions vary between communities and drift over time.

Two distinctions do useful work here. Denotation is the conventional, dictionary sense of a word; connotation is the freight of association and evaluation it carries. Thrifty and stingy denote roughly the same behaviour and connote opposite judgments. Most disputes about wording are disputes about connotation, and pointing to a dictionary is beside the point.

The second is the abstraction ladder, popularised by S. I. Hayakawa. At the bottom sit specific observable particulars; at the top sit general categories. She was unprofessional sits high; she arrived eleven minutes after the meeting started and did not have the figures sits at the bottom. Ascending the ladder is efficient and it is also where most unnecessary conflict lives, because a high-abstraction statement is unfalsifiable and feels like a verdict on the person. Descending the ladder is the single most reliable technique for making a difficult conversation tractable.

Does language shape thought?

Edward Sapir and Benjamin Lee Whorf are associated with the idea that the language you speak affects how you think. The claim comes in two very different strengths, and conflating them has caused a century of confusion.

Linguistic determinism, the strong version, holds that language determines thought and that a concept without a word for it cannot be thought. This is rejected. People routinely think things they have no word for, then coin a word for it, which would be impossible if the strong version were true. Speakers of languages without a numeral system can still perceive quantity, and bilingual people do not become different thinkers when they switch.

Linguistic relativity, the weak version, holds that habitual patterns in a language make certain distinctions easier and more automatic. This has real support.

Russian has two basic colour terms where English has one blue, distinguishing lighter goluboy from darker siniy. Jonathan Winawer and colleagues reported in 2007 that Russian speakers were faster than English speakers at discriminating between two blues when the pair straddled that lexical boundary, and that the advantage disappeared when participants performed a verbal interference task. The language was doing work in the moment of perception.

Stephen Levinson and colleagues have documented languages, including Guugu Yimithirr in Queensland, that use absolute cardinal directions rather than relative left and right for everyday spatial description. Speakers of such languages maintain an accurate sense of orientation in circumstances where speakers of relative-frame languages become disoriented, and they perform differently on non-linguistic spatial reasoning tasks.

Now the correction. Claims about grammatical gender changing how people describe objects have had a mixed replication record, and you should treat individual striking results in this area cautiously. And the single most repeated example in the popular literature is simply false.

The claim that Inuit or Eskimo languages have dozens or hundreds of words for snow traces back through a garbled chain. Franz Boas mentioned four distinct roots in 1911. Whorf inflated the figure. Later writers inflated it further, with no one checking. Laura Martin documented the chain in a 1986 paper, and Geoffrey Pullum popularised the correction in a 1991 essay whose title calls it a hoax. English has plenty of snow terms too, and the counting method that generates large numbers for polysynthetic languages would generate large numbers for English as well. When you see this example used to support linguistic relativity, the person using it has not read the source literature.

In short: Language does not determine what you can think. It does make some distinctions faster and more automatic, and that effect is measurable, but the famous snow example is not evidence of anything.

Saying is doing

J. L. Austin's How to Do Things with Words, delivered as lectures in 1955 and published in 1962, made a point that now seems obvious and was not. Some utterances do not describe the world; they change it. I promise, I apologise, I resign, I now pronounce you married. Saying the words performs the act.

Austin distinguished three layers in any utterance. The locutionary act is the saying of the words with their sense. The illocutionary act is what is done in saying them, which is the intent: promising, warning, requesting. The perlocutionary act is the effect produced on the hearer, which may be quite different from the intent.

That three-way split is diagnostically useful, because interpersonal failures land in different layers. If your friend did not know the word you used, the locutionary layer failed. If they heard a request as a criticism, the illocutionary layer failed. If they understood you perfectly and were hurt anyway, both earlier layers succeeded and the effect was still bad, which is why saying that you did not mean it that way is a claim about the illocutionary layer and no answer at all about the perlocutionary one.

John Searle sorted illocutionary acts into categories, including assertives, which commit the speaker to a truth; directives, which attempt to get the hearer to do something; commissives, which commit the speaker to a future action; expressives, which convey a psychological state; and declarations, which change reality by being uttered. He also specified felicity conditions, the circumstances that must hold for an act to work. A promise from someone with no ability to deliver is not a promise, and an order from someone with no authority is not an order.

Most requests in English are indirect directives. Can you pass the salt is not a question about capability. It is a directive dressed as a question, and the disguise is negative politeness from the previous lesson, protecting the hearer's autonomy by leaving them a formal exit they will not take.

Speech that sounds powerless

Some speech features are consistently read as tentative: hedges such as sort of and I think; tag questions appended to statements; disclaimers such as this may be a stupid question; hesitations; and rising intonation on declaratives.

Robin Lakoff described these in the 1970s as characteristic of women's speech. William O'Barr and Bowman Atkins tested it directly by analysing courtroom testimony, and published a result in 1980 that reframed the whole question. The features clustered not by gender but by social status and courtroom experience. Male witnesses of low status used them; female expert witnesses did not. They proposed renaming the cluster powerless language, because that is what it tracks.

That correction matters twice over. It shows how easily a status effect gets misread as a gender effect. And it complicates the advice usually drawn from the finding, which is to strip hedges from your speech. Hedges do face work. Prefacing a challenge with I might be wrong about this is not weakness; it is protecting the other person's positive face while leaving room for them to disagree without a confrontation. Removing every hedge produces speech that is read as blunt rather than confident.

A related popular claim deserves the same scrutiny. Meta-analytic work on gender and interruption, including a 1998 review by Kristin Anderson and Campbell Leaper, finds that the difference in intrusive interruptions is real but small, on the order of a fifth of a standard deviation, and depends heavily on setting and group composition. That is a long way from the confident assertion that men interrupt women constantly, and a long way from nothing.

Confirming and disconfirming responses

One final framework earns its place because it is directly actionable. Communication scholars distinguish messages by whether they confirm the other person's existence and worth.

Confirming responses come in ascending strength: recognition, which is simply acknowledging the person exists; acknowledgment, which engages with what they actually said; and endorsement, which affirms the value of what they said or felt.

Disconfirming responses include the impervious response, which ignores the person entirely; the interrupting response; the irrelevant response, which changes topic with no connection; the tangential response, which acknowledges and then redirects to the responder's own interest; and the incongruous response, where verbal and nonverbal channels contradict each other.

The category worth watching for is the tangential response, because it feels polite from inside. Someone describes a problem at work, and you reply that something similar happened to you, then describe your version. You have acknowledged them and taken the floor. Done occasionally it is normal reciprocity; done consistently it reads as never being heard.

What matters here: Disconfirmation is rarely delivered as insult. It is usually delivered as a topic change.

Common misconceptions

"Eskimo languages have hundreds of words for snow." The claim is a documented chain of inflation from a modest observation by Boas, traced by Martin in 1986 and Pullum in 1991. It supports nothing, and its persistence is a better illustration of how claims spread than of how language works.

"The Sapir-Whorf hypothesis has been disproved." The strong determinist version is rejected; the weak relativity version has genuine experimental support, including the Russian blues result and the spatial frame work. Treating them as one hypothesis loses the whole point.

"Being clear means saying exactly what you mean." Grice showed that hearers infer from what you omit relative to expectation, so adding an obviously unnecessary compliment can create an implication you did not intend. Clarity includes managing what your silences imply.

"Hedges and tag questions are women's speech." O'Barr and Atkins found the cluster tracks social status and situational power, not gender. The original label was a misreading of a status effect.

"You should eliminate hedges to sound confident." Hedges perform face work and invite disagreement without confrontation. Removing them all produces speech read as blunt, and the research does not support a blanket rule.

Pulling it together

Grice's reference letter communicates a devastating judgment while stating nothing negative, because hearers assume cooperation and infer meaning from departures. The four maxims of quantity, quality, relation and manner make that inference possible, and flouting them deliberately generates implicature. Ogden and Richards's triangle of meaning explains why words never connect directly to things, only through minds, which is why denotation rarely settles disputes that are really about connotation, and why descending the abstraction ladder from unprofessional to eleven minutes late without the figures makes a conflict tractable. Linguistic determinism is rejected, but linguistic relativity has real support from the Russian blues discrimination result and from absolute spatial frame languages, while the famous snow-words example is a documented chain of inflation. Austin and Searle showed that utterances perform acts, and separating locutionary, illocutionary and perlocutionary layers tells you which part of an exchange failed. Powerless language tracks status rather than gender, as O'Barr and Atkins established in 1980, and hedges do genuine face work rather than merely signalling weakness. Finally, the most common form of disconfirmation is not insult but the tangential response, which acknowledges you and then takes the floor.

Sources

1. University of Minnesota Libraries Publishing. (2016). Language and meaning. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. Encyclopaedia Britannica. (n.d.). Language. britannica.com
3. Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics 3: Speech acts (pp. 41-58). Academic Press.
4. Austin, J. L. (1962). How to do things with words. Oxford University Press.
5. Pullum, G. K. (1991). The great Eskimo vocabulary hoax and other irreverent essays on the study of language. University of Chicago Press.
6. O'Barr, W. M., & Atkins, B. K. (1980). Women's language or powerless language? In S. McConnell-Ginet, R. Borker, & N. Furman (Eds.), Women and language in literature and society. Praeger.
7. Winawer, J., Witthoft, N., Frank, M. C., Wu, L., Wade, A. R., & Boroditsky, L. (2007). Russian blues reveal effects of language on color discrimination. Proceedings of the National Academy of Sciences, 104(19), 7780-7785.

Two small experiments and a number that got loose

In 1967 Albert Mehrabian published two studies. In the first, with Morton Wiener, listeners heard single words such as maybe and really, recorded in tones that were positive, neutral or negative, and judged the speaker's attitude. When the word's meaning and the tone conflicted, listeners went with the tone. In the second, with Susan Ferris, listeners judged attitude from a single spoken word paired with a photograph of a face, and the face dominated.

Mehrabian combined the regression weights from the two studies and got a rough split: about 7 per cent of the judgment attributable to the words, 38 per cent to vocal tone, and 55 per cent to facial expression.

Those three numbers left the laboratory and never came back. You have almost certainly been told that 93 per cent of communication is nonverbal, and it is one of the most confidently repeated false statements in the field.

Look at what the studies actually involved. Single words, not sentences. No context of any kind. Judgments only of whether the speaker liked the listener, not of content, information or intent. Conditions deliberately constructed so that the channels contradicted each other. And a small number of speakers, all women in at least one of the studies. Mehrabian himself has said repeatedly that the equation was never meant to describe communication in general and does not.

The defensible finding is narrow and genuinely useful: when verbal and nonverbal channels conflict on a question about feelings or attitude, listeners weight the nonverbal channels heavily. That is worth knowing. It says nothing about how information is conveyed, and it does not imply that words carry seven per cent of anything.

Bottom line: The 7-38-55 rule is a real finding about conflicting cues in a judgment of liking, generalised beyond recognition. Treat any speaker who quotes it as a general law as someone who has not read the papers.

The channels, named properly

Nonverbal communication is not one thing. It is a set of channels that operate simultaneously and can be studied separately.

Kinesics covers body movement. Paul Ekman and Wallace Friesen sorted it in 1969 into five categories that are still standard. Emblems have direct verbal translations within a culture, such as a thumbs up, and their meanings vary sharply between cultures, which makes them a reliable source of accidental offence. Illustrators accompany speech and depict what is being said, such as gesturing the size of a fish. Affect displays convey emotion, mostly through the face. Regulators manage the flow of interaction, such as the small nods and gaze shifts that hand over a turn. Adaptors are self-directed behaviours such as touching your face or fidgeting, associated with discomfort but, as you will see, not reliably diagnostic of anything specific.

Proxemics covers use of space. Edward Hall proposed four zones in 1966: intimate distance out to about 45 centimetres, personal distance to about 1.2 metres, social distance to about 3.7 metres, and public distance beyond that. These numbers are useful and they are not universal; Hall derived them from observations of middle-class Americans, and preferred distances vary by culture, gender composition, relationship and setting. Quoting the numbers as though they applied everywhere is exactly the error Module 5 is about.

Haptics covers touch, which is the most regulated nonverbal channel and the one where the rules vary most between relationships and cultures. Vocalics, or paralanguage, covers everything about the voice except the words: pitch, volume, rate, pauses, and vocal fillers. Chronemics covers the use of time, including who is kept waiting and how quickly a message is answered, which is a status signal in almost every workplace. Appearance and artifacts cover clothing, grooming and objects, and environment covers the arrangement of the space itself.

Ekman and Friesen also described what nonverbal behaviour does relative to speech: it can repeat the verbal message, substitute for it, complement it, accent part of it, regulate the interaction, or contradict the words. That last function is the one Mehrabian's experiments engineered, and it is comparatively rare in ordinary conversation.

Do faces read out emotions?

Ekman's cross-cultural work from the late 1960s, including studies with the Fore people of Papua New Guinea who had little contact with Western media, supported a strong claim: a small set of basic emotions have universal facial expressions that can be recognised across cultures. That claim became foundational, and it is now embedded in commercial products for hiring, security screening and so-called emotion recognition software.

In 2019 Lisa Feldman Barrett, Ralph Adolphs, Stacy Marsella, Aleix Martinez and Seth Pollak published a review of more than a thousand studies in Psychological Science in the Public Interest, and their conclusion was blunt. The evidence does not support the claim that a person's emotional state can be reliably inferred from their facial movements.

Their argument has three legs. First, reliability: people scowl when angry only a minority of the time, and they scowl for many reasons unrelated to anger, so the mapping between configuration and state is loose in both directions. Second, specificity: the same facial configuration accompanies different states in different contexts. Third, method: much of the classic evidence used forced-choice tasks in which participants matched a posed photograph to one of six supplied emotion words, and when Maria Gendron, Barrett and colleagues ran free-sorting tasks with Himba participants in Namibia, the neat cross-cultural agreement largely disappeared.

None of this means faces carry nothing. Facial movements are informative, especially in context and combined with voice and situation. What is not supported is the inference from a configuration to an internal state with the confidence that a hiring algorithm or a border screening system requires. The practical stakes here are high, and they are the reason this particular dispute is worth knowing in a communication course rather than only in a psychology one.

Why this matters: The gap between what a research literature supports and what a product built on it claims is the gap where real harm happens, and here the product claims are running well ahead of the evidence.

Detecting lies from behaviour: what the numbers say

Ask people how to tell if someone is lying and most will say they avoid eye contact. The Global Deception Research Team surveyed more than two thousand people across 58 countries in 2006 and found gaze aversion to be the most commonly named cue almost everywhere. It is a nearly universal belief.

It is also wrong. Bella DePaulo and colleagues published a meta-analysis in 2003 examining 158 candidate cues to deception across the accumulated literature. Almost none showed a substantial relationship with lying. Gaze aversion, in particular, does not distinguish liars from truth-tellers. A handful of cues showed small effects, such as slightly reduced detail and slightly more tension, but nothing approaching a usable individual diagnostic.

Charles Bond and DePaulo followed in 2006 with a meta-analysis of deception judgments drawing on more than 200 studies and roughly 24,000 judges. Mean accuracy at distinguishing truths from lies was about 54 per cent, against 50 per cent for a coin. That is a real effect and a practically negligible one.

Two further results in that literature are worth carrying. Professional groups who believe themselves expert, including police officers and customs officials, generally do not outperform students, though they are frequently more confident. And people are better at detecting deception when they hear a story and evaluate its content than when they watch for behavioural tells, which points the practical advice in a direction opposite to most training.

That evidence has had policy consequences. The United States Government Accountability Office reviewed the Transportation Security Administration's behaviour detection programme in 2013 and concluded that the available evidence did not support the ability of officers to identify threats reliably from behavioural indicators, recommending that funding be limited until effectiveness was demonstrated.

The useful thing to take from this is not that lying is undetectable. It is that behavioural tells are close to worthless, while inconsistencies in an account, checkable facts and the sequence of what someone knew when remain informative. Investigate the story, not the fidgeting.

Power posing, and how a replication failure looks from inside

In 2010 Dana Carney, Amy Cuddy and Andy Yap published a study with 42 participants reporting that holding an expansive posture for two minutes raised testosterone, lowered cortisol, increased risk tolerance and increased feelings of power. The idea became one of the most widely viewed pieces of psychology in the world.

In 2015 Eva Ranehill and colleagues ran a much larger replication with 200 participants. The feeling of power replicated. The hormonal changes did not. The behavioural change in risk taking did not.

What followed is instructive. In 2016 Carney, the first author of the original paper, published a statement saying she no longer believed the effects were real and would not teach the finding. Cuddy and colleagues, using different analytic approaches, have continued to argue that the self-reported felt-power effect is robust, and on that narrower point the evidence is reasonably supportive.

So where does that leave the practical claim? Adopting an expansive posture plausibly makes you feel somewhat more powerful for a short time. There is no good evidence it changes your hormones or your behaviour, and the version of the claim that spread is not the version that survives. Reporting both halves accurately is the point of including it here.

What nonverbal behaviour does reliably do

After that much subtraction it is worth being clear about what is well supported, because the answer is not nothing.

Immediacy behaviours are the most robust finding in the area, and, in an irony worth noting, they are Mehrabian's own most durable contribution. Eye contact, forward lean, reduced distance, open posture, direct body orientation, appropriate touch and a warm vocal tone reliably increase perceived warmth, liking and approachability. The effect appears in classrooms, in clinical settings and in ordinary conversation, and it is large enough to act on.

Behavioural mimicry increases liking. Tanya Chartrand and John Bargh's 1999 studies, in which a confederate subtly copied a participant's posture or mannerisms, found that participants liked the mimicking confederate more and reported smoother interactions, generally without noticing the mimicry. The effect has been extended and qualified in later work, and the core finding has held up better than most of the priming literature from that era.

Voice carries more than people expect. Michael Kraus reported in 2017 that participants judging another person's emotions were more accurate when they only heard the voice than when they had both voice and face, apparently because visual information invited over-interpretation. That result sits comfortably alongside the Barrett review: faces are noisier readouts than intuition suggests.

Brief observations carry some signal. Nalini Ambady and Robert Rosenthal's 1992 meta-analysis on thin slices found that judgments made from very short silent clips predicted outcomes such as teacher evaluations above chance. The effects are modest and real, and the popular reading of them as evidence for reliable snap judgment about individuals overstates what an above-chance aggregate correlation means.

The upshot: Nonverbal behaviour is highly informative about warmth, engagement and the state of a relationship, and poorly informative about specific internal states such as which emotion someone is feeling or whether they are lying.

Common misconceptions

"93 per cent of communication is nonverbal." The figure comes from two 1967 studies of single words in engineered conflict, judging liking only. It describes cue weighting when channels contradict, not the informational content of communication.

"You can read someone like a book if you know body language." Individual cues are ambiguous. Crossed arms may mean defensiveness, cold, or comfort. Interpretation requires context, a baseline for that person, and clusters of behaviour, and even then it is inference rather than reading.

"Liars avoid eye contact." The most widely believed cue in the world, across 58 countries, and among the least diagnostic. Meta-analysis of 158 cues found nothing that supports it.

"Trained professionals can spot deception." Meta-analytic accuracy sits near 54 per cent overall, and professional groups typically do not exceed it, though they report more confidence. The 2013 review of the TSA behaviour detection programme reached a comparable conclusion about operational practice.

"Power posing changes your hormones." The 2015 replication with 200 participants found the felt-power effect but not the hormonal or behavioural effects, and the original study's first author publicly withdrew support for those claims in 2016.

What to remember

Mehrabian's 1967 experiments used single words, no context and engineered conflict between channels, and support only the narrow claim that people weight nonverbal cues heavily when judging attitude from contradictory signals. Nonverbal communication runs through many channels at once, including kinesics with Ekman and Friesen's five categories, Hall's proxemic zones which are culturally specific rather than universal, haptics, vocalics, chronemics, appearance and environment, and it can repeat, substitute for, complement, accent, regulate or contradict speech. The strong claim that facial configurations are reliable readouts of emotion did not survive the 2019 review by Barrett and colleagues, which found the mapping loose in both directions and much of the classic evidence dependent on forced-choice methods. Deception detection from behaviour runs at about 54 per cent accuracy across roughly 24,000 judges, the near-universal belief in gaze aversion is unsupported by a meta-analysis of 158 cues, and professionals are more confident rather than more accurate. Power posing produces a felt-power effect and not the hormonal changes the original 42-person study reported. What does hold up is immediacy behaviour, which reliably raises perceived warmth; mimicry, which raises liking; the surprising informativeness of voice alone; and small but real thin-slice effects. Nonverbal behaviour tells you a great deal about a relationship and very little about a specific internal state.

Sources

1. University of Minnesota Libraries Publishing. (2016). Nonverbal communication. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
3. American Psychological Association. (n.d.). Monitor on Psychology. apa.org
4. Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements. Psychological Science in the Public Interest, 20(1), 1-68.
5. Bond, C. F., & DePaulo, B. M. (2006). Accuracy of deception judgments. Personality and Social Psychology Review, 10(3), 214-234.
6. DePaulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., & Cooper, H. (2003). Cues to deception. Psychological Bulletin, 129(1), 74-118.
7. Ranehill, E., Dreber, A., Johannesson, M., Leiberg, S., Sul, S., & Weber, R. A. (2015). Assessing the robustness of power posing: No effect on hormones and risk tolerance in a large sample of men and women. Psychological Science, 26(5), 653-656.

Eleven seconds

In 1984 Howard Beckman and Richard Frankel recorded a set of primary care consultations to find out what happened after a doctor asked what brought the patient in. The answer was that the doctor usually stopped listening quickly. Patients' opening statements were interrupted after a median of about eighteen seconds, and only a small minority were allowed to finish.

The finding was widely cited, and it prompted replications. A 1999 study in JAMA found a similar figure, around twenty-three seconds. A 2019 analysis of recorded clinical encounters found that clinicians solicited the patient's agenda in only about a third of visits, and that when patients did begin to set it out, they were interrupted after a median of eleven seconds.

Nobody in those rooms was being rude. Doctors interrupt because they are pattern-matching to a diagnosis and because they are under time pressure, which are both defensible. The cost is that the interruption usually arrives before the patient has said the thing they came to say. Patients often lead with the acceptable complaint and reach the real concern third or fourth.

That is a medical example of a general problem, and it is the reason this lesson exists. Public Speaking, COMM 101 on this site, covers the stages of the listening process and the standard listening types, and this lesson does not repeat that material. Here the question is narrower and harder: what does listening do to the person being listened to, and which specific behaviours produce that effect?

The point: Listening is not the absence of talking. It is a set of behaviours with measurable effects on the speaker, and the default in most conversations is to take the floor well before the other person has finished constructing their point.

Preference is not ability

People differ in what they attend to. Graham Bodie, Debra Worthington and Christopher Gearhart's revised Listening Styles Profile describes four habitual orientations. Relational listeners attend to feelings and to the connection. Analytical listeners withhold judgment and try to take in the full picture before deciding. Task-oriented listeners focus on efficiency and on what needs to happen. Critical listeners evaluate accuracy and consistency as they go.

Two cautions come with this framework and are usually omitted. First, a style is a preference, not a competence, so a strong critical listener is not a better listener, only a differently deployed one. Second, and more importantly, these are self-report measures, and self-reported listening correlates only modestly with what an observer would score as listening behaviour. Almost everyone rates themselves as an above-average listener, which is arithmetically impossible.

The useful application is mismatch diagnosis. A relational listener describing a hard week to a task-oriented listener will receive solutions and will experience that as not being heard, while the task-oriented listener will experience the frustration as ingratitude. Neither has done anything wrong. They are running different programmes on the same input.

What listening does to the person speaking

The most interesting recent work does not ask whether listeners understood correctly. It asks what changes in the speaker when the listening is good.

Guy Itzchakov and Avraham Kluger have run a programme of experiments in which speakers talk about an attitude they hold while a trained listener either listens well, using attention, questions and no evaluation, or listens in a distracted or interrupting way. Several effects show up repeatedly.

Speakers who are listened to well report lower social anxiety during and after the conversation. They express less extreme attitudes, not because they have been argued with but because they have had room to notice the qualifications in their own position. They show greater attitude complexity, acknowledging arguments on more than one side. And they report greater self-insight, having heard themselves think out loud with somebody paying attention.

Kluger and Itzchakov's 2022 review of listening at work gathers this work and makes an argument that goes against most persuasion advice. If you want someone to hold a position less rigidly, arguing tends to entrench it, while being genuinely listened to tends to loosen it. The mechanism is not that good listening is persuasive. It is that defensiveness maintains extremity, and being listened to lowers defensiveness.

Remember: Good listening changes the speaker, not just the listener's understanding. It lowers anxiety and moderates extremity, which is a stronger claim than anything in the usual advice about eye contact.

Support: what makes a comforting message work

Brant Burleson spent a career on a question that sounds simple and is not: when someone is upset, what should you say? His answer, developed over decades of message-evaluation studies, centres on person-centredness, meaning the degree to which a message acknowledges, elaborates and legitimises the other person's feelings and perspective.

Level	What it does	What it sounds like
Low person-centred	Denies, criticises or challenges the feeling, or tells the person how to feel	You should not let it get to you. It is not worth being upset about.
Moderate person-centred	Acknowledges the feeling implicitly, often by distracting, offering sympathy, or explaining the situation away	That is rough. Look at it this way, at least you learned something.
High person-centred	Explicitly recognises and legitimises the feeling, invites elaboration, and helps the person make sense of it in their own terms	That sounds genuinely painful. What was the worst part of it for you?

The finding, replicated across many studies and in several cultural contexts, is that high person-centred messages are rated as more sensitive and more helpful, and produce greater improvement in the distressed person's emotional state. The effect is large enough to be practically useful and it is not a matter of taste; low person-centred comfort is worse comfort.

Notice which everyday move sits in the low category. Telling someone to look on the bright side, that it could be worse, or that they should not let it bother them are all attempts at kindness, and all of them communicate that the feeling is unwarranted. So does moving straight to solutions.

That last point needs qualifying, because advice is not forbidden. Erina MacGeorge's research on advice finds that advice is well received when three conditions hold: the content is actually feasible and useful, the source is seen as having relevant expertise and good intentions, and it arrives after emotional support rather than instead of it. Advice given first reads as dismissal; the same advice given after the feeling has been acknowledged is often welcome. Sequence does most of the work.

The other half: responding to good news

Nearly all advice about listening concerns bad news. Shelly Gable, Harry Reis, Emily Impett and Evan Asher pointed out in 2004 that people bring good news to their partners far more often than bad, and that what happens next predicts relationship quality.

They described four response styles along two dimensions, active or passive and constructive or destructive.

	Constructive	Destructive
Active	Enthusiastic engagement, questions, reliving the event with them	Pointing out problems, deflating, listing the downsides
Passive	Quiet, understated pleasure, a brief acknowledgment	Ignoring it, changing the subject, turning to your own news

Their finding, which has held up in subsequent work, is that active-constructive responding predicts relationship well-being and commitment, and that it does so over and above how partners respond to each other's problems. The passive-constructive quadrant is the surprising one: a mild, pleasant, brief response to good news does not feel supportive to the person who brought it, even though nothing negative has happened.

This is called capitalisation, and the practical implication is direct. If you want to strengthen a relationship and can only change one behaviour, changing how you respond to good news may be a better lever than changing how you respond to complaints, and it is a great deal easier.

Does paraphrasing actually help?

Active listening technique, in most trainings, means paraphrasing what the other person said before responding. It is worth asking whether that works, since it is easy to do badly and can sound like a script.

Harry Weger and colleagues tested it directly in 2014 in initial interactions, comparing paraphrasing with simple acknowledgment such as nodding and brief verbal signals, and with advice giving. Paraphrasing produced significantly greater feelings of being understood than the alternatives, though the difference was moderate rather than dramatic.

Two things follow. The technique has genuine support, so it is not merely a training-room ritual. And the size of the effect suggests that what matters most is not the specific verbal formula but whether the listener is actually attending, which paraphrasing forces but does not guarantee. A paraphrase produced from a script while the listener plans their reply is not the thing that was measured.

Worth holding on to: Paraphrasing works because it makes attention verifiable, both to the speaker and to yourself. It stops working the moment it becomes a formula you can produce without attending.

Common misconceptions

"Good listening means never giving advice." Advice is well received when it is feasible, comes from a credible source, and follows emotional support rather than replacing it. The problem is almost always sequence, not advice itself.

"Telling someone to look on the bright side is supportive." It is a low person-centred message. It communicates that the feeling is unwarranted, and message-evaluation research finds it consistently rated less helpful than acknowledgment.

"People know whether they are good listeners." Self-reported listening correlates only modestly with observed listening behaviour, and nearly everyone rates themselves above average, which cannot be true.

"Listening is about accuracy." Accuracy matters, and the experimental work shows that good listening also lowers speakers' anxiety and moderates their attitudes. Those effects occur even when the listener says almost nothing and offers no interpretation.

"Supporting someone means being there when things go wrong." Gable and colleagues found that responses to good news predict relationship well-being over and above responses to bad news, and that a mild, pleasant response to good news does not register as support.

Putting it together

Physicians interrupt patients' opening statements after a median that studies have put at eighteen seconds in 1984 and eleven seconds in a 2019 analysis, usually before the real concern has been reached, which is a specific instance of a general default: listeners take the floor early. Listening styles describe habitual preferences rather than abilities, and the self-report measures correlate only modestly with behaviour, so their best use is diagnosing mismatch rather than ranking people. Experimental work by Itzchakov and Kluger shows that high-quality listening lowers speakers' social anxiety, reduces attitude extremity, increases attitude complexity and increases self-insight, which means listening changes the speaker rather than only informing the listener. Burleson's person-centredness research establishes that messages which explicitly acknowledge and legitimise a feeling outperform those that minimise it or move straight to solutions, and MacGeorge's work shows advice is welcome when it is feasible, credible and sequenced after support. Gable and colleagues' capitalisation research adds the neglected half: active-constructive responses to good news predict relationship well-being over and above responses to bad news. And Weger and colleagues found that paraphrasing genuinely does increase the sense of being understood, moderately, because it makes attention verifiable rather than because the words themselves are magic.

Sources

1. University of Minnesota Libraries Publishing. (2016). Listening. In Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. American Psychological Association. (n.d.). Monitor on Psychology. apa.org
3. Beckman, H. B., & Frankel, R. M. (1984). The effect of physician behavior on the collection of data. Annals of Internal Medicine, 101(5), 692-696.
4. Burleson, B. R. (2003). Emotional support skills. In J. O. Greene & B. R. Burleson (Eds.), Handbook of communication and social interaction skills. Lawrence Erlbaum.
5. Gable, S. L., Reis, H. T., Impett, E. A., & Asher, E. R. (2004). What do you do when things go right? The intrapersonal and interpersonal benefits of sharing positive events. Journal of Personality and Social Psychology, 87(2), 228-245.
6. Kluger, A. N., & Itzchakov, G. (2022). The power of listening at work. Annual Review of Organizational Psychology and Organizational Behavior, 9, 121-146.
7. Weger, H., Castle Bell, G., Minei, E. M., & Robinson, M. C. (2014). The relative effectiveness of active listening in initial interactions. International Journal of Listening, 28(1), 13-31.

Smile training in Atlanta

Arlie Russell Hochschild spent time in the early 1980s watching recruits at Delta Air Lines being trained in Atlanta. What she recorded was not instruction in how to behave. It was instruction in how to feel. Trainees were told to think of the cabin as their own living room and the passengers as personal guests, and to produce a smile that came from inside rather than one pasted on, because passengers can tell.

Her 1983 book The Managed Heart named what she was seeing: emotional labour, the work of managing your own feelings to produce the emotional state an employer requires. She estimated that a large share of jobs involved it, and the proportion has only risen since.

Hochschild distinguished two ways of doing it. Surface acting means displaying an emotion you do not feel, managing the outward expression only. Deep acting means working on the feeling itself, using memory, imagination and reframing until the required emotion is genuinely present. The Delta training was explicitly aimed at deep acting, on the theory that surface acting is detectable and exhausting.

Three decades of research have tested that intuition. A 2011 meta-analysis by Ute Hulsheger and Anna Schewe, covering three decades of studies, found that surface acting is consistently associated with emotional exhaustion and reduced job satisfaction, while deep acting shows weaker or negligible associations with strain and is sometimes linked to better performance and customer ratings. Hochschild's distinction, drawn from observation, held up against the numbers.

Why this matters: Feeling is not a private state that communication reports on. In a great many settings, producing a particular feeling is the work, and doing it one way rather than another has measurable costs.

What an emotion is, for our purposes

Emotion research is contested territory, but for communication purposes a working decomposition is enough. An emotional episode typically involves an appraisal of an event, some physiological change, an expressive component, an action tendency, and a subjective feeling.

The appraisal component does most of the explanatory work. In Richard Lazarus's account, emotion follows from an evaluation of an event relative to one's goals and resources, not from the event itself. Two people receive identical feedback on a draft. One appraises it as a threat to their standing and feels shame; the other appraises it as useful information from someone who bothered to read carefully, and feels gratitude. Nothing about the feedback differs.

This is the single most useful idea in the lesson, because appraisals are contestable in a way that raw feelings are not. Telling someone they should not feel angry is useless. Asking what they took the event to mean sometimes changes the appraisal, and the feeling follows.

It also matters for how you frame your own reactions. When you find yourself saying that someone made you feel a certain way, you are treating the emotion as caused directly by their behaviour. The appraisal step sits in between, is yours, and is often the part that can move.

Display rules

What is felt and what is shown are different, and the gap is governed by learned conventions. Ekman and Friesen described four operations people perform on expression: intensify, showing more than is felt; deintensify, showing less; neutralise, showing nothing; and mask, showing a different emotion entirely.

Display rules vary by culture, by occupation, by gender expectation, by relationship and by setting, and they are learned early. A child who has learned to look pleased at a disappointing gift has learned a display rule and can usually articulate it.

The intercultural point, which Module 5 develops, is that display rule differences are frequently misread as differences in feeling. A person from a context where deintensifying negative emotion in public is strongly expected may be read as cold or evasive by someone from a context that expects intensification, and vice versa. Neither is feeling less. They are following different rules about the relationship between feeling and showing.

Emotional contagion, and a famous experiment about it

Emotions spread. Elaine Hatfield, John Cacioppo and Richard Rapson proposed that people automatically mimic the expressions, postures and vocal patterns of those around them, and that this mimicry feeds back into their own emotional state. The effect is well documented in face-to-face interaction and is small per exchange, which is what you would expect from a mechanism operating below awareness.

In 2014 Adam Kramer, Jamie Guillory and Jeffrey Hancock published a study testing whether contagion operates through text alone. Working with Facebook, they manipulated the news feeds of 689,003 users for a week, reducing either positive or negative posts, and measured the emotional content of what those users subsequently wrote.

Two findings came out of it. The statistical one: contagion appeared to operate through text without face-to-face contact, and the effect size was extremely small, on the order of a fraction of a per cent change in word use. With 689,003 participants, an effect that small is statistically overwhelming and practically almost invisible, which is a useful illustration of the difference between the two.

The other finding was ethical, and it dominated the response. Participants had not consented to an experimental manipulation of their emotional environment, and the argument that the platform's terms of service constituted consent was widely rejected. The journal issued an expression of concern. The episode changed how large platforms describe their research and remains a standard case in research ethics teaching.

In short: Emotional contagion is real and small. The 2014 Facebook study demonstrated it at scale and demonstrated, more memorably, that scale does not confer permission.

Emotional intelligence: the construct and the marketing

Peter Salovey and John Mayer proposed emotional intelligence in 1990 as a specific ability: perceiving emotions accurately, using them to facilitate thought, understanding emotional information, and managing emotions. It was framed as an intelligence, testable with right and wrong answers.

Daniel Goleman's 1995 book made it famous, and made much larger claims, including that emotional intelligence could matter more than cognitive ability for life success. The idea reached corporate training, schools and hiring within a few years.

The evidence supports a narrower claim. Meta-analytic work distinguishes ability-based measures, which use performance tests, from mixed or trait measures, which use self-report and bundle in traits like optimism and assertiveness. Ability measures show modest incremental validity for job performance over cognitive ability and personality, and that validity is larger in jobs with high emotional labour demands, which makes sense. Mixed-model measures predict outcomes somewhat better but overlap so heavily with established personality dimensions that it is unclear what is being added beyond a new name for conscientiousness, extraversion and emotional stability.

So the honest position is that emotional intelligence names something real, that it is measurable with effort, that its predictive power is modest and context-dependent, and that the claim about outweighing cognitive ability is not supported by the research literature that the claim is usually attributed to.

Expressing and regulating: two things that do not work as advertised

Two pieces of folk wisdom about emotion have been tested directly and failed.

Catharsis. The idea that expressing anger discharges it is very old and very durable. Brad Bushman tested it in 2002 by having angered participants either hit a punching bag while thinking about the person who provoked them, hit it while thinking about fitness, or sit quietly. Those who vented were subsequently more aggressive, not less. Sitting quietly produced the least aggression. Rehearsing anger practises anger. The general finding has been replicated, and it means that advice to let it out is close to exactly backwards.

Suppression. James Gross's process model of emotion regulation distinguishes strategies by when they intervene: choosing or changing the situation, redirecting attention, reappraising the meaning of the event, or modulating the response after the emotion is underway. Reappraisal, which works early, generally produces better outcomes than expressive suppression, which works late.

The interpersonal evidence is sharper than the individual evidence. Emily Butler and colleagues reported in 2003 that when one partner in a conversation was instructed to suppress their emotional expression, the other partner, who did not know, showed elevated blood pressure during the interaction. Suppression is not a private act. It degrades the interaction for the person who was not doing it, and it reduces rapport and the other person's willingness to affiliate.

What does help is more boring. Naming the emotion specifically, in your own words and as your own state, tends to reduce its intensity, an effect sometimes called affect labelling. Lisa Feldman Barrett's work on emotional granularity finds that people who habitually distinguish finely between emotional states, using words like frustrated, resentful, embarrassed and disappointed rather than a general bad, regulate more effectively and use fewer maladaptive strategies. The vocabulary is not decoration; it is part of the mechanism.

In conversation this translates into a specific form. Saying I felt dismissed when the decision was made without me describes your appraisal and its trigger. Saying you made me feel dismissed asserts that the other person caused the feeling directly, skips the appraisal step, and invites a dispute about causation rather than a conversation about the event.

The core of it: Venting rehearses; suppressing leaks into the other person's physiology; naming precisely helps. The advice that survives contact with evidence is unglamorous.

Common misconceptions

"Venting anger gets it out of your system." Bushman's 2002 experiments found that venting increased subsequent aggression relative to doing nothing. Expressing anger rehearses it rather than discharging it.

"Hiding your feelings protects the other person." Butler and colleagues found that a suppressing partner's uninformed conversational partner showed elevated blood pressure. Suppression is transmitted even when it is not detected consciously.

"Emotional intelligence matters more than IQ." This popular claim is not supported by the meta-analytic literature. Ability-based emotional intelligence shows modest incremental validity, largest in emotionally demanding jobs, and mixed-model measures overlap heavily with ordinary personality traits.

"Events cause emotions." Appraisals do. That is why identical feedback produces shame in one person and gratitude in another, and why arguing with a feeling is useless while asking what someone took the event to mean sometimes works.

"Someone who shows little emotion is feeling little." Display rules intervene between feeling and expression, with four documented operations, and they vary by culture, occupation and setting. Reading expression as a direct measure of feeling is the same error as the facial expression overclaim in the previous lesson.

The short version

Hochschild's observation of Delta training in the early 1980s named emotional labour and split it into surface acting, which manages display, and deep acting, which manages the feeling itself; meta-analytic work three decades later confirmed that surface acting predicts exhaustion while deep acting largely does not. Emotion is best understood as following an appraisal of an event relative to goals, which is why identical events produce different emotions and why appraisals rather than feelings are the negotiable part. Display rules govern the gap between feeling and showing through intensifying, deintensifying, neutralising and masking, and they vary enough across cultures to be routinely misread as differences in feeling. Emotional contagion is real and small, and the 2014 Facebook experiment with 689,003 users demonstrated it at a scale that also made it a landmark case in research ethics. Emotional intelligence names a real construct whose validity is modest and context-dependent, well short of the claim that made it famous. And the two most popular pieces of advice about handling emotion both fail: venting increases later aggression, and suppression raises the blood pressure of the person you are talking to. What works is precise naming, owned as your own appraisal.

Sources

1. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
2. OpenStax. (2020). Emotion and motivation. In Psychology 2e. Rice University. openstax.org
3. Hochschild, A. R. (1983). The managed heart: Commercialization of human feeling. University of California Press.
4. Hulsheger, U. R., & Schewe, A. F. (2011). On the costs and benefits of emotional labor: A meta-analysis of three decades of research. Journal of Occupational Health Psychology, 16(3), 361-389.
5. Bushman, B. J. (2002). Does venting anger feed or extinguish the flame? Catharsis, rumination, distraction, anger, and aggressive responding. Personality and Social Psychology Bulletin, 28(6), 724-731.
6. Butler, E. A., Egloff, B., Wilhelm, F. H., Smith, N. C., Erickson, E. A., & Gross, J. J. (2003). The social consequences of expressive suppression. Emotion, 3(1), 48-67.
7. Kramer, A. D. I., Guillory, J. E., & Hancock, J. T. (2014). Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences, 111(24), 8788-8790.

Doors, stairwells, and friendship at Westgate West

In 1950 Leon Festinger, Stanley Schachter and Kurt Back published a study of the housing built for married veterans studying at MIT. Apartments were handed out as they fell vacant rather than chosen, which made the arrangement close to random, and the researchers asked every resident to name their closest friends in the project.

The answers tracked the floor plans. Roughly four in ten residents named the person living next door. Roughly two in ten named someone two doors along. About one in ten named someone at the far end of the same short corridor, a walk of a few seconds.

Festinger's team named the thing they had found functional distance: not how far apart two doors are, but how likely the layout makes it that two people will cross paths. Residents in the apartments beside a stairwell had more friends on the other floor than anyone else, because everyone going up passed their door. A corridor with the mailboxes at one end produced more acquaintances at that end than at the other.

Why this matters: Opportunity comes first. Selection then operates on the pool that proximity delivers, and that pool is far smaller, and far less chosen, than it feels from inside.

The woman who attended lectures and never spoke

Robert Zajonc's experiments in the 1960s established the mere exposure effect: repeated exposure to something, with no reward attached, tends to increase liking for it. He showed it with nonsense syllables, with Chinese characters shown to people who could not read them, and with photographs of faces.

Richard Moreland and Scott Beach ran the version that matters here. In their 1992 study, four women of broadly similar appearance were assigned to attend a large university lecture course. One attended none of the classes. The others attended five, ten or fifteen. They sat near the front, spoke to nobody, and left. At the end of term the students were shown slides of all four and asked to rate them.

Ratings of attractiveness, and ratings of how similar the woman seemed to the rater, rose with the number of classes attended. The women had done nothing at all except be seen.

Now stack the two findings. Proximity produces exposure. Exposure produces mild liking. Mild liking makes an approach more likely. The approach is the part you remember as the beginning of the friendship, and it is the last link in a chain whose first three links were invisible to you.

Similarity, and the point where it stops working

Donn Byrne spent the 1960s and 1970s running a procedure so tidy it was called the attraction paradigm. A participant fills in an attitude questionnaire. Later they are shown a questionnaire supposedly completed by a stranger, which the experimenter has actually written to share a set proportion of the participant's answers, and asked how much they think they would like this person. Liking rose with the proportion of shared attitudes in a relationship so regular that Byrne wrote it as a law.

Which is where the popular version of the finding stops, and where the interesting part starts. Ricardo Montoya, Robert Horton and Jeffrey Kirchner published a meta-analysis in 2008 separating two things the paradigm had run together: actual similarity, measured by comparing two people's independently reported attitudes, and perceived similarity, meaning how alike one person believes the two of them to be.

In no-interaction studies of the Byrne type, actual similarity predicted attraction, as it always had. In short-interaction studies the effect shrank. In studies of people already in an ongoing relationship, actual similarity did essentially nothing. Perceived similarity predicted attraction at every stage, and predicted it strongly.

Read that carefully, because it inverts the usual story. Couples are not, on the whole, unusually well matched on attitudes. They are unusually convinced that they are. Liking appears to generate the perception at least as much as the perception generates liking.

The mirror claim, that opposites attract, has fared worse. Complementarity of needs was proposed in the 1950s, tested repeatedly, and has never accumulated general support. Specific complementarity in specific domains, such as one person preferring to lead a conversation and the other preferring not to, is a different and much narrower claim.

What happens when you try to predict this in advance

In 2012 Eli Finkel, Paul Eastwick, Benjamin Karney, Harry Reis and Susan Sprecher published a long review of online dating in Psychological Science in the Public Interest. They were careful about what the technology does well: it enormously widens the pool of available partners, and it allows communication before meeting. On matching algorithms, their conclusion was blunt. The published evidence does not support the claim that any algorithm can pair two people who have never met better than chance, and the sites that make the claim have not released the data that would let anyone check it.

Samantha Joel, Eastwick and Finkel then tested the underlying assumption directly in 2017. They ran speed-dating sessions, collected over a hundred self-report measures from each participant beforehand, covering personality, values, preferences, attachment and stated ideals, and trained machine learning models to predict who would want to see whom again.

The models worked for two of the three things they were asked to predict. They could predict how much a given person tended to like their partners in general, and how much a given person tended to be liked by their partners in general. On the third quantity, the specific chemistry between this person and that one, which is the entire promise of matching, prediction was indistinguishable from zero.

The upshot: Individual tendencies to like and be liked are predictable from data collected in advance. The compatibility of a particular pair, so far, is not. Anything sold to you as an algorithmic match is selling the widened pool and calling it science.

Three accounts of why a relationship continues

Attraction gets two people talking. It does not explain why they are still talking in four years. Three theories do most of the work here, and they are not rivals so much as different levels of zoom.

Account	Core claim	What it explains well	Where it runs out
Uncertainty reduction (Berger and Calabrese, 1975)	Early interaction is driven by the need to reduce uncertainty about the other person, using passive observation, active third-party enquiry, or interactive questioning	The shape of first conversations: the question and answer trade, the biographical inventory, the search for common ground	Michael Sunnafrank argued in 1986 that people are not chasing certainty for its own sake. They forecast outcome value, and will end a promising-to-be-dull conversation while still deeply uncertain
Social exchange (Thibaut and Kelley, 1959)	People track rewards against costs, and judge the result against two standards: the comparison level, meaning what they believe they deserve, and the comparison level for alternatives, meaning what they believe is available elsewhere	The gap between satisfaction and dependence. Satisfaction comes from the first comparison; staying comes from the second	It treats the accounting as conscious and stable, which it is not, and it says little about how the standards themselves are formed
Investment model (Rusbult, 1980)	Commitment is built from satisfaction, from poor alternatives, and from investments that would be lost on leaving: time, shared property, a joint social network, a version of yourself that only exists in this relationship	Persistence. Le and Agnew's 2003 meta-analysis of fifty-two studies found the three components together accounted for roughly two thirds of the variance in commitment, and commitment predicted whether relationships lasted	It predicts staying, not flourishing. Caryl Rusbult applied it directly to abusive relationships, where high investment and low alternatives produce commitment without satisfaction

The third row is the one worth sitting with. A model that explains why people stay in relationships that harm them is not a cynical model; it is a model that stopped assuming staying means happy. If you have ever watched a friend remain somewhere you could not understand, the investment model is usually a better guide than any account built on satisfaction alone.

Knapp's staircase, read as a description

Mark Knapp's staircase model, first set out in the 1970s, describes ten stages, five going up and five coming down.

Coming together	What it sounds like	Coming apart	What it sounds like
Initiating	Scripted openings, a few seconds long	Differentiating	We turns back into I; differences get emphasised
Experimenting	Small talk as a search for common ground	Circumscribing	Topics get fenced off; the safe list shrinks
Intensifying	More self-disclosure, first use of we, private nicknames	Stagnating	Conversations are rehearsed in advance because the outcome is known
Integrating	Social identities merge; others treat you as a unit	Avoiding	Physical and conversational distance is arranged deliberately
Bonding	A public, often formal, commitment	Terminating	Distance summaries, the account of what happened, the dividing of a network

Used as a description this is a useful vocabulary. Circumscribing in particular is worth being able to name: the list of things you cannot discuss grows quietly, nobody announces it, and the relationship gets smaller without anyone deciding it should.

Used as a script it produces two errors. The first is thinking the stages must be climbed in order and without skipping, which they are not. The second is reading the descending stages as failure. Most relationships stabilise somewhere partway up and stay there for years, which is what a friendship is; and movement is not one way, since differentiating can be followed by intensifying and often is.

The tensions that never resolve

Leslie Baxter and Barbara Montgomery's relational dialectics starts from a different premise: that relationships are held together by contradictions which are managed, not solved. Three recur.

• Connection and autonomy. Wanting to be close and wanting to be a separate person, at the same time, permanently.
• Openness and closedness. Intimacy requires disclosure; a self requires some things not disclosed.
• Predictability and novelty. Reliability is most of what a relationship is for, and enough of it is deadening.

People manage these in identifiable ways. Cyclic alternation swings between the poles by season or by week. Segmentation assigns different domains different rules, so that finances are wholly open and past relationships are not. Balance splits the difference and usually satisfies nobody completely. Reframing redefines the terms so the contradiction dissolves, as when time apart is understood as a component of closeness rather than a subtraction from it.

Worth holding on to: If you are waiting for the moment when you have got the closeness question settled, you will wait indefinitely. The question is a permanent feature. The skill is in noticing which strategy you are using and whether it is working now.

Maintenance: the ordinary behaviours that were actually measured

Laura Stafford and Daniel Canary set out to find what people actually do to keep relationships going, rather than what advice books tell them to do. Their factor analyses produced five behaviours, which have held up across many samples.

1. Positivity. Being cheerful and uncritical in ordinary interaction. Not a technique, just the general temperature.
2. Openness. Direct talk about the relationship itself, including how it is going.
3. Assurances. Statements and acts implying a future together. This one is usually the strongest correlate of commitment in the data.
4. Social networks. Spending time with shared friends and family, which builds external structure around the relationship.
5. Sharing tasks. Doing your share of the joint work, which is far less romantic and far more predictive than it sounds.

Marianne Dainton's work added a distinction that matters. Maintenance can be strategic, done deliberately with the relationship in mind, or routine, done without thinking about maintenance at all. Routine behaviour is far more frequent, and it predicts satisfaction at least as well. The relationship is mostly kept alive by things nobody is counting.

Attachment, handled carefully

In 1987 Cindy Hazan and Phillip Shaver printed a short questionnaire in a Denver newspaper asking readers to pick which of three paragraphs best described how they felt in close relationships. The three paragraphs corresponded to the secure, anxious and avoidant patterns Mary Ainsworth had described in infants. The distribution of adult answers roughly matched the distribution found in infancy, which was the point of the exercise.

Three things have changed since. The categories were replaced by two continuous dimensions, attachment anxiety and attachment avoidance, because people do not fall into three boxes; Kelly Brennan, Catherine Clark and Shaver's 1998 measure is built that way. Stability over time is moderate rather than fixed. And attachment turns out to be partly relationship-specific: the same person can be measurably more secure with one partner than another.

So the honest form of the idea is that people carry expectations about availability and responsiveness, formed early and revisable, which shape how they read ambiguous behaviour. The dishonest form, now common online, treats attachment style as a fixed personality type that explains a partner's every action and licenses a diagnosis. The research does not support that, and the two-dimensional measures were designed partly to stop it.

Common misconceptions

"We became close because we had so much in common." Montoya and colleagues found that among people already in relationships, actual similarity predicts almost nothing while perceived similarity predicts a great deal. The commonality is substantially a product of the closeness, not only its cause.

"Opposites attract." Complementarity of needs has been tested since the 1950s without accumulating support as a general principle.

"A good algorithm can find your match." Joel and colleagues could predict how much someone would like others in general and be liked in general, and could not predict the pair-specific component at all from more than a hundred pre-meeting measures.

"If they stay, they must be reasonably happy." Social exchange separates satisfaction, which comes from comparing outcomes to what you expect, from dependence, which comes from comparing them to your alternatives. Low alternatives and high investment produce staying without satisfaction.

"A relationship that comes apart failed." Knapp described a sequence, not a verdict. Most relationships settle partway along and remain there, and movement runs in both directions.

What to carry forward

Festinger's residents named the neighbours whose doors they passed, which makes proximity and functional distance the first filter on everyone you know. Mere exposure supplies the liking that makes an approach likely, as Moreland and Beach showed with four women who attended lectures and said nothing. Similarity works powerfully in the laboratory and mostly through perception in real relationships. Prediction in advance fails at exactly the point the industry sells: Finkel's review found no support for matching claims, and Joel's models could forecast general liking but not pair-specific chemistry. Continuation is best explained by commitment built from satisfaction, alternatives and investments, which is why the model also explains staying in relationships that are bad. Knapp's stages describe how coming together and coming apart sound, without prescribing either. Baxter and Montgomery's dialectics say the central tensions are managed and never solved. And what actually maintains a relationship, in the measured data, is assurances and shared tasks and ordinary uncritical warmth, mostly performed routinely by people who are not thinking about maintenance at all.

Sources

1. University of Minnesota Libraries Publishing. (2016). Communication in the real world: An introduction to communication studies. open.lib.umn.edu
2. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
3. Festinger, L., Schachter, S., & Back, K. (1950). Social pressures in informal groups: A study of human factors in housing. Harper.
4. Moreland, R. L., & Beach, S. R. (1992). Exposure effects in the classroom: The development of affinity among students. Journal of Experimental Social Psychology, 28(3), 255-276.
5. Montoya, R. M., Horton, R. S., & Kirchner, J. (2008). Is actual similarity necessary for attraction? A meta-analytic synthesis of actual and perceived similarity. Journal of Social and Personal Relationships, 25(6), 889-922.
6. Finkel, E. J., Eastwick, P. W., Karney, B. R., Reis, H. T., & Sprecher, S. (2012). Online dating: A critical analysis from the perspective of psychological science. Psychological Science in the Public Interest, 13(1), 3-66.
7. Le, B., & Agnew, C. R. (2003). Commitment and its theorized determinants: A meta-analysis of the investment model. Personal Relationships, 10(1), 37-57.

Fifteen minutes, wired to instruments

In a laboratory in the early 1980s, Robert Levenson and John Gottman attached sensors to married couples, recording heart rate, skin conductance and gross bodily movement, and asked each pair to spend fifteen minutes discussing a continuing disagreement in their marriage. Two cameras ran. Afterwards each partner watched the tape and turned a dial to rate, second by second, how positive or negative they had felt at the time.

The result that made the method famous was not about the words. It was that the degree to which the two people's physiological arousal moved together during the argument predicted the decline in their marital satisfaction over the following years. Bodies that tracked each other closely in conflict belonged to marriages that got worse.

Bottom line: Conflict is a physiological event as much as a verbal one, and the body's state constrains what the conversation can possibly achieve.

What a conflict actually is

Communication scholars usually define interpersonal conflict as an expressed struggle between interdependent parties who perceive incompatible goals, scarce resources, and interference from the other in achieving what they want. Four parts of that definition do work.

Expressed. An unspoken irritation is not yet a conflict. It becomes one when it is communicated, including by pointed silence, which is expression.

Interdependent. You only fight with people whose behaviour affects your outcomes. This is why conflict is a sign of connection rather than its absence, and why it tends to increase, not decrease, as a relationship deepens.

Perceived. The incompatibility does not have to be real. A large proportion of conflicts are about goals that turn out, on inspection, to be compatible, and were fought because neither party stated theirs.

Interference. Someone has to be in the way. Two people wanting different things in separate rooms are not in conflict.

Add the content and relationship levels from Module 1 and you can see why arguments about the dishwasher are not about the dishwasher. The content is the loading of the dishwasher. The relationship claim underneath is about who gets to set standards, whose time counts, and whether the other person notices what you do.

The pattern with a name: demand and withdraw

Andrew Christensen and Christopher Heavey described the best-documented conflict structure in the field. One partner pursues, criticises and demands discussion. The other avoids, defends and withdraws. The pursuit intensifies because the withdrawal is read as indifference; the withdrawal intensifies because the pursuit is read as attack. Neither person is doing anything unreasonable given how they have read the other, and the loop is self-feeding.

In heterosexual couples the demanding role is more often taken by the woman and the withdrawing role by the man, which for a while was explained as a difference in emotional style. Christensen and Heavey's 1990 experiment undercut that. They varied whose issue was being discussed. When the topic was a change the woman wanted, the female-demand pattern appeared strongly. When the topic was a change the man wanted, the asymmetry shrank considerably.

The structural reading is that the person seeking change demands, and the person who benefits from the status quo withdraws, because withdrawal wins by default. Whoever is happy with things as they are can end the conversation and keep the outcome. Where household arrangements assign more of the unwanted status quo to women, the roles fall out accordingly.

Paul Schrodt and colleagues meta-analysed the literature in 2014 across dozens of studies and found demand-withdraw reliably associated with lower relational satisfaction, at a correlation in the region of minus point three. That is a solid, replicated association and it is not enormous, which is the honest way to hold it: a real pattern with real costs, not a diagnosis.

So what?: If you are the withdrawer, the useful move is to name a time you will return to it and keep that appointment. Withdrawal is read as a verdict on the relationship; a scheduled return is not.

The four horsemen

From coding thousands of minutes of conflict tape, Gottman identified four behaviours that distinguished couples whose marriages later deteriorated.

Behaviour	What it does	Sounds like	The stated antidote
Criticism	Attacks the person rather than the behaviour, using global and permanent terms	You never think about anyone but yourself	A complaint about a specific act, plus a stated need
Contempt	Communicates superiority and disgust, through mockery, name calling, sneering or eye rolling	Oh, well done, genius	Deliberate description of what you appreciate, built as a habit outside conflict
Defensiveness	Refuses any part of the responsibility, usually by counter-attacking or playing the victim	I only did that because you	Accepting the fraction that is yours, however small
Stonewalling	Shuts down the listening signals entirely: no eye contact, no back-channel, no response	Silence, a turned body, a phone picked up	Stopping the conversation explicitly and physiologically settling before returning

Contempt is the one Gottman reports as the strongest single discriminator, and it is different in kind from the others. Criticism and defensiveness are escalations. Contempt is a statement about standing: it says the other person is beneath the speaker. That is very hard to take back.

Stonewalling has a physiological story attached that is worth knowing. Gottman uses the term flooding for the state in which heart rate climbs past roughly a hundred beats per minute and the person's access to their own reasoning, humour and listening degrades sharply. Stonewalling is often what flooding looks like from the outside. The practical implication is that continuing to talk at a flooded person accomplishes nothing, because the equipment required to process what you are saying is temporarily offline. Twenty minutes of genuine break, not twenty minutes of rehearsing your argument, is the recommendation, and the rehearsal caveat matters given what Module 3 said about venting.

Those prediction accuracies, and why they are not predictions

You have probably encountered the claim that Gottman can predict divorce from a few minutes of conversation with accuracy in the region of ninety per cent. Figures between eighty-three and ninety-four per cent circulate, attached to different studies.

Richard Heyman and Amy Slep examined what those numbers rest on in a 2001 paper with a title that gives away the answer: the hazards of predicting divorce without cross-validation. Their point is technical and completely decisive.

Here is the problem, in the plainest form. Suppose you have observed a hundred couples and know which ones later divorced. You look for the combination of behaviours that best separates the two groups in that sample, and you fit your model afterwards, choosing the variables and the cut-off points that work best on the data in front of you. Naturally the model then classifies that sample very accurately. It was built to. What you have produced is a description of a hundred couples, not a prediction rule, and the only way to find out whether it predicts anything is to apply the fixed model, unchanged, to a fresh sample nobody used in building it.

Heyman and Slep did that with independent data. Accuracy dropped substantially. The models that had separated the original samples so cleanly did much less well on couples they had not been fitted to, which is the ordinary fate of models that have not been cross-validated.

Notice exactly what this does and does not undermine. It does not show that the four horsemen are unreal, that contempt does not matter, or that the observational coding was sloppy. Those findings have held up in other hands. What it shows is that a specific set of accuracy percentages, repeated for two decades in books, talks and articles, describes how well a model fitted a sample it was built on. Applied to your marriage, or to any couple not in the original data, it does not mean what it sounds like it means.

Remember: Post hoc classification of a known sample is not prediction. When you meet an impressive accuracy figure, the question to ask is whether the model was fixed before it met the data it is being scored on.

What longitudinal research does support

Benjamin Karney and Thomas Bradbury reviewed more than a hundred longitudinal studies of marriage in 1995 and proposed the vulnerability-stress-adaptation model, which is still the standard frame. Three things interact. Enduring vulnerabilities are what each person brings, including personality and family history. Stressful events are what happens to the couple, including illness, money and children. Adaptive processes are how the pair handles both. The model's contribution is that it stops treating conflict behaviour as the whole story: the same couple, handled the same way, does better under less stress.

Samantha Joel led a study published in 2020 that pooled forty-three longitudinal datasets covering more than eleven thousand couples and used machine learning to find which of many self-report measures best predicted relationship quality. Three results are worth carrying.

• Relationship-specific perceptions were the strongest predictors: how committed you believe your partner is, how appreciated you feel, how satisfying the sexual relationship is, how much conflict there is.
• Individual characteristics, including personality traits, added very little once those perceptions were in the model. Who you are matters less than how this particular relationship feels from inside it.
• Predicting how satisfaction would change over time was near hopeless for every model tested, which is a striking result given the size of the data.

Startups, repairs, and the ratio

The first three minutes. How a conflict conversation opens predicts a great deal about how it goes, and Gottman's labs report being able to forecast the tenor of a fifteen minute discussion from its opening. A harsh startup leads with criticism or contempt. A softened one names a specific behaviour, states the speaker's own feeling, and makes a positive request: not you never help, but the kitchen was left for me again last night and I felt taken for granted; can we settle who does it on Tuesdays.

Repair attempts. A repair attempt is any move that tries to interrupt escalation: a joke, an apology, a question, an admission, a physical touch, a change of tone. Gottman's claim is that repair attempts are made in almost all couples, and that the difference between stable and unstable pairs lies in whether they are received. A joke that lands stops a fight. The identical joke, refused, becomes evidence that the other person is not taking this seriously.

The often-quoted ratio comes from coding those discussions: stable couples produced roughly five positive behaviours for every negative one during conflict, while couples heading for separation were closer to parity. Two cautions. It is a ratio observed in laboratory conflict discussions, not a rule for daily life, and it is a correlation rather than a demonstration that manufacturing positives fixes anything.

Styles, and why collaborating is not always the right answer

The standard five-style scheme sorts approaches to conflict along two axes: concern for your own goals and concern for the other's.

Style	Own goals	Other's goals	When it is the right choice
Avoiding	Low	Low	The issue is trivial, or one of you is flooded, or the problem is perpetual and unsolvable
Obliging	Low	High	The issue matters far more to them, or you were wrong
Dominating	High	Low	Speed matters and the cost of a bad outcome falls on you, or a safety line is being crossed
Compromising	Medium	Medium	Time is short, goals are genuinely opposed, and a partial win each is available
Integrating	High	High	The relationship is ongoing, the issue is important to both, and there is time to find the underlying interests

Training courses tend to present the last row as the answer and the rest as deficiencies. That is not what the research supports. Gottman's group reports that around two thirds of the recurring conflicts in long-term couples are perpetual: they arise from durable differences in personality, family background or need, and they will still be there in twenty years. For a perpetual problem, the sensible goal is not resolution but dialogue that keeps the difference from becoming contempt. Avoidance, chosen deliberately and by both, is a reasonable strategy for a difference that cannot be settled.

Nickola Overall and James McNulty have pushed further against the be-positive default. Their reviews find that direct and even negative conflict behaviour, including blunt statements of dissatisfaction, is associated with improvement when the problem is severe and requires the partner to change, while soft, positive communication is more useful for minor problems. Warmth is not universally optimal. It depends on what has to change.

Accommodation, apology and the limits of forgiveness

When a partner does something inconsiderate, four responses are available, and the labels come from Rusbult's adaptation of Albert Hirschman's scheme: exit, voice, loyalty and neglect. Exit and neglect are destructive, actively and passively. Voice and loyalty are constructive, actively and passively. The accommodation finding is that committed partners suppress the impulse to retaliate and choose a constructive response instead, and that this suppression, unlike the emotional suppression discussed in the previous lesson, is associated with better outcomes because it is about action rather than concealed feeling.

Apologies that work tend to contain identifiable components: a specific statement of what you did, an acknowledgement of its effect on the other person, no conditional clause beginning with the word if, and a statement of what will be different. The conditional is the usual failure. I am sorry if you were upset relocates the problem into the other person's reaction.

Forgiveness is where the honest account gets uncomfortable. It is generally good for the forgiver's own wellbeing. But McNulty's longitudinal work found that in marriages where a partner behaved badly, more forgiving spouses saw that behaviour continue or increase over time, while less forgiving spouses saw it decline. Forgiveness removes a consequence. Where the behaviour was mild and rare, that is generous and costless; where it was serious and repeated, it can maintain the thing being forgiven.

Common misconceptions

"Gottman can predict divorce with ninety per cent accuracy." Heyman and Slep showed those figures came from models fitted to the same samples they were scored on. Cross-validated on fresh data, accuracy falls substantially. The behaviours are real; the percentages are not predictions.

"Happy couples do not fight." Conflict requires interdependence, so it increases with closeness. The observational data separate couples by how they fight, not whether they do.

"Women want to talk about problems and men do not." Christensen and Heavey found the asymmetry largely tracked whose desired change was on the table. The person seeking change demands; the person served by the status quo withdraws, because withdrawal preserves the outcome.

"Every conflict should be resolved." Around two thirds of recurring conflicts in long-term couples are perpetual, arising from stable differences. Dialogue that prevents contempt is the realistic goal for those.

"Being positive and gentle is always the best approach." Overall and McNulty find directness, including negative directness, associated with improvement on severe problems, and gentleness better suited to minor ones.

What you now know

Levenson and Gottman's wired couples established that conflict is a bodily event whose physiology predicts later satisfaction, and flooding explains why talking at a shut-down partner cannot work. Demand-withdraw is the best-documented destructive structure, its gender asymmetry largely follows whose change is being demanded, and it correlates with dissatisfaction at around minus point three. The four horsemen name behaviours that discriminate deteriorating marriages, with contempt the strongest, and repair attempts matter chiefly according to whether they are received. The famous accuracy percentages fail on cross-validation, which is a lesson about statistics rather than about marriage. What longitudinal work supports is Karney and Bradbury's three-way interaction of vulnerabilities, stress and adaptation, and Joel's finding that relationship-specific perceptions beat personality while nothing predicts change well. Two thirds of recurring conflicts are perpetual, which makes deliberate avoidance a legitimate strategy rather than a failure, and directness beats gentleness when the problem is severe. Forgiveness, finally, is good for you and is not always good for the behaviour.

Sources

1. University of Minnesota Libraries Publishing. (2016). Conflict and interpersonal communication. In Communication in the real world. open.lib.umn.edu
2. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
3. The Gottman Institute. (n.d.). The four horsemen and their antidotes. gottman.com (the research group's own account of its model)
4. Levenson, R. W., & Gottman, J. M. (1983). Marital interaction: Physiological linkage and affective exchange. Journal of Personality and Social Psychology, 45(3), 587-597.
5. Christensen, A., & Heavey, C. L. (1990). Gender and social structure in the demand/withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59(1), 73-81.
6. Heyman, R. E., & Slep, A. M. S. (2001). The hazards of predicting divorce without crossvalidation. Journal of Marriage and Family, 63(2), 473-479.
7. Karney, B. R., & Bradbury, T. N. (1995). The longitudinal course of marital quality and stability: A review of theory, method, and research. Psychological Bulletin, 118(1), 3-34.
8. Joel, S., Eastwick, P. W., Allison, C. J., et al. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32), 19061-19071.

The questions at the dinner table

Jack McLeod and Steven Chaffee were not studying families. In the early 1970s they were studying how children learn to process news and form political opinions, and to get at it they asked parents what was allowed to be said at home. Could a child argue with a parent about politics? Was a child told that there are things you do not discuss outside the family? Was a child encouraged to look at both sides of an issue, or to agree with the parents?

Two clusters came out of the answers, and they turned out to describe far more than news habits. Ascan Koerner and Mary Anne Fitzpatrick later relabelled them in the terms now standard, and the pair of dimensions has organised family communication research ever since.

Conversation orientation is the degree to which a family creates a climate in which everyone is encouraged to talk freely about a wide range of topics. High-conversation families discuss decisions openly, tolerate disagreement about facts and preferences, and treat talk as the ordinary way things get worked out.

Conformity orientation is the degree to which a family emphasises a uniformity of attitudes and values, usually organised around the authority of parents. High-conformity families expect agreement, treat challenges to a parent's position as a challenge to the parent, and prioritise harmony and the family's presentation to outsiders.

The dimensions are independent, which is the part people get wrong. A family can be high on both.

Four family types, and what each one is good at

Type	Conversation	Conformity	What it looks like	Associated with
Consensual	High	High	Everything is discussed at length, and the discussion is expected to arrive where the parents already were. Children are heard, then persuaded	High closeness, and children who take parental values on board but may struggle with unresolved disagreement
Pluralistic	High	Low	Open discussion with no requirement to agree. A child's argument can win on its merits	The most consistently favourable pattern in the research: better conflict skills, higher communication confidence
Protective	Low	High	Obedience expected, explanation not offered. Decisions are announced	Children who are easily persuaded by outside authority and less confident in their own judgement
Laissez-faire	Low	Low	Few conversations, little guidance, members largely independent of one another	Low emotional involvement; children rely more on outside sources for values

Read the table against your own upbringing and one thing usually becomes visible immediately: the difference between a family where you could disagree and a family where you could talk. Those are not the same thing, and the consensual cell is where people most often confuse them. A great deal of discussion can take place in a family where the conclusion was never actually available.

The point: Two independent dimensions, not a single scale from closed to open. The useful diagnostic question is not how much your family talked, but what happened when someone reached a different conclusion.

Families that have to explain themselves

Define a family structurally, by blood and marriage, and you immediately misdescribe a large share of the households in any country. Define it functionally, by who performs the roles, and you capture more, at the cost of a fuzzy boundary.

Kathleen Galvin's concept of discourse dependence cuts through this usefully. All families are constituted partly through talk. But families formed through adoption, remarriage, fostering, donor conception, single parenthood by choice, or the chosen families common in queer communities depend on discourse more heavily, because the relationship is not legible to outsiders at a glance and cannot be asserted by resemblance.

Galvin separates the internal work from the external. Internally, families do naming, which includes what a stepparent is called and by whom; discussing, meaning the explicit conversations about how this family works; ritualising, which is the invention of practices that mark the unit as a unit; and storytelling. Externally, they do labelling, as when a parent introduces a child in a way that forecloses the question; explaining, which is offering an account before one is demanded; defending, when the account is challenged; and naming again, in the different register used with strangers.

Consider what this predicts. A stepfamily at a school registration desk performs more identity work in ten minutes than a birth family performs in a month, because the birth family's structure does the work silently. That labour is invisible to the people who do not have to do it, and it is exhausting for the people who do.

Rituals, stories, and a scale that got ahead of its evidence

Family rituals are repeated practices carrying meaning beyond their function: the specific way a birthday runs, the Sunday call, the seat that is a particular person's seat. Rituals differ from routines in that a routine can be skipped and a ritual cannot, without someone noticing that something has happened.

Stories do related work. Families tell origin stories about how the couple met, survival stories about the year everything went wrong, and character stories that fix what each member is like. Those stories teach values without ever stating them, which is why they persist.

In the early 2000s Marshall Duke, Robyn Fivush and Amber Lazarus at Emory University built a twenty-item scale called Do You Know, asking children questions such as whether they know how their parents met, where their grandparents grew up, and a story about a serious illness in the family. Scores correlated with measures of self-esteem, sense of control and family functioning.

The finding travelled, in a much stronger form than the researchers stated. The popular version says that telling children family stories makes them resilient. Three cautions are needed. The study was correlational and cross-sectional, so it cannot separate the story-knowledge from the kind of family that talks a lot in general, which is exactly the more likely cause. The sample was modest. And Fivush has been explicit in later writing that the intergenerational narrative, not the quiz score, is what the work is about.

Worth holding on to: A twenty-item scale that correlates with wellbeing is not a twenty-item intervention. The claim that survives is that families who talk about their past tend to be families that talk, and that talking is the plausible active ingredient.

Whose information is it?

Sandra Petronio's communication privacy management theory starts from a claim that sounds obvious and turns out to be productive: private information is owned, and disclosure makes it co-owned rather than transferred.

When you tell your sister that you have been made redundant, she does not simply now know it. She becomes a co-owner of the information, subject to rules about what she may do with it, and those rules are often assumed rather than stated. Petronio's term for what happens when the rules were never negotiated, or were negotiated differently by each party, is boundary turbulence.

Families run several boundaries at once. There is a boundary around the household, which is what makes we do not discuss that outside this house a coherent instruction. There are dyadic boundaries inside it, so that two siblings hold something the parents do not. And there are individual boundaries, which adolescence largely consists of establishing.

The theory makes a practical suggestion that costs nothing. When you disclose something you care about, state the rule at the same time. Not just here is what happened, but here is what happened and I have not told our mother. The alternative is that the other person invents a rule and is then blamed for following it.

Thirty million words: the anatomy of an overreach

In 1995 Betty Hart and Todd Risley published Meaningful Differences in the Everyday Experience of Young American Children. Their method was serious and laborious. Forty-two families in Kansas City were visited monthly for roughly an hour, from when the child was about nine months old until age three, and everything said in the room was recorded and transcribed. The families spanned professional households, working-class households, and households receiving welfare.

They found substantial differences in how much speech the children heard, and they extrapolated from the recorded hours to a lifetime figure. By age four, they estimated, a child in a professional family would have heard about thirty million more words than a child in a family on welfare. The number entered education policy, early-intervention programmes, city campaigns and countless keynote talks.

Now look at what the number is made of.

• It is an extrapolation. One hour a month, multiplied out to every waking hour of four years. Any systematic difference in how families behave while being recorded is multiplied by the same factor.
• The sample is small and unrepresentative. Forty-two families in one city, with only six in the welfare group, is a thin base for a claim about a population.
• It counted a narrow slice of the language environment. The recordings centred on speech in the room with the primary caregiver present.

Douglas Sperry, Linda Sperry and Peggy Miller published a direct challenge in 2019 in Child Development. Working across several communities and counting speech from everyone in the child's environment rather than the primary caregiver alone, they did not find the pattern in the form the original claim requires. In some working-class and rural communities, children heard as much speech as, or more than, children in the professional comparison, because talk arrived from grandparents, cousins, neighbours and older siblings. A method that counts only the mother will systematically undercount the language environment of any culture that does not organise childrearing around a single adult.

The response, led by Roberta Golinkoff and colleagues in the same journal, conceded that the thirty million figure is an extrapolation that should not be quoted as a measurement, while maintaining that socioeconomic differences in some measures of language input are real and replicated in larger samples. Separate work using automated recorders has found gaps in the same direction but considerably smaller.

Meanwhile the more interesting result came from a different direction. Rachel Romeo and colleagues reported in 2018 that what predicted children's language scores, and activation in language-related brain regions during listening, was the number of conversational turns the child took part in, not the number of words the child heard. The effect held after adjusting for socioeconomic status and for total adult word count. Being talked with beat being talked at.

Why this matters: A real difference was found, packaged into a memorable number by extrapolation, repeated until it became a fact about children rather than an estimate from forty-two families, and then partly dismantled. The direction survives. The number does not. And the actionable finding turned out to be about turns rather than volume, which is a different instruction entirely.

What families avoid

Topic avoidance in families is not the same as secrecy, and the research separates them. Avoidance is the steering away from a subject in conversation; a secret is information deliberately withheld by some members from others. Both are normal, both increase around adolescence, and both are usually justified by the avoider on relational grounds: to protect the relationship, to protect the other person, or to avoid a conflict that will change nothing.

The distinctive feature of family conflict is exit costs. In a friendship, sustained conflict can be resolved by drifting apart. Family members generally cannot exit, or can only exit at a price that includes the whole network. That constraint is why the destructive patterns from the previous lesson do more damage here: a demand-withdraw loop between two people who will be at the same funerals for forty years does not simply run its course.

Common misconceptions

"An open family is one where everyone talks a lot." Conversation and conformity are independent dimensions. Consensual families talk a great deal and expect the discussion to land where the parents already are, which is not the same as open.

"Poor children hear thirty million fewer words." That figure is an extrapolation from about one recorded hour a month with forty-two families, counting a narrow slice of the language environment. Studies counting all speakers in the child's world have not reproduced it, and even sympathetic researchers now treat the number as an estimate that should not be quoted as a measurement.

"Telling children family stories makes them resilient." The Do You Know work is correlational. Families whose children know the stories are families that talk, and talking is the more plausible active ingredient than the knowledge itself.

"Once I have told someone something, it is theirs to handle." Communication privacy management treats disclosed information as co-owned under rules. Where the rules were never stated, turbulence follows and the recipient is blamed for a rule they never agreed to.

"A family is defined by blood and marriage." Structural definitions misdescribe a large share of households. Galvin's discourse dependence explains what families formed by other routes actually do: naming, ritualising, explaining and defending, which is real labour that structurally legible families never have to perform.

Recap

McLeod and Chaffee's questions about dinner-table arguments produced two independent dimensions, conversation and conformity, which sort families into consensual, pluralistic, protective and laissez-faire types, with the pluralistic pattern showing the most consistently favourable outcomes. Galvin's discourse dependence explains why families formed by adoption, remarriage or choice do more identity work in talk, internally through naming and ritual and externally through labelling, explaining and defending. Rituals and stories transmit values without stating them, though the Do You Know scale supports a weaker claim than its popular version. Petronio's privacy management recasts disclosure as co-ownership under rules, and boundary turbulence as what happens when the rules were assumed. The thirty million word gap is the module's case study in a finding that outgrew its evidence: a real direction, a number produced by extrapolating from one hour a month with forty-two families, a replication attempt that counted the whole speech community and did not find it, and a more useful successor finding about conversational turns. And family conflict differs from all other conflict in one structural respect: you cannot leave.

Sources

1. University of Minnesota Libraries Publishing. (2016). Communication in relationships. In Communication in the real world. open.lib.umn.edu
2. OpenStax. (2021). Marriage and family. In Introduction to sociology 3e. Rice University. openstax.org
3. American Psychological Association. (n.d.). APA dictionary of psychology. dictionary.apa.org
4. Koerner, A. F., & Fitzpatrick, M. A. (2002). Toward a theory of family communication. Communication Theory, 12(1), 70-91.
5. Galvin, K. M. (2006). Diversity's impact on defining the family: Discourse dependence and identity. In L. H. Turner & R. West (Eds.), The family communication sourcebook (pp. 3-19). Sage.
6. Petronio, S. (2002). Boundaries of privacy: Dialectics of disclosure. State University of New York Press.
7. Hart, B., & Risley, T. R. (1995). Meaningful differences in the everyday experience of young American children. Paul H. Brookes.
8. Sperry, D. E., Sperry, L. L., & Miller, P. J. (2019). Reexamining the verbal environments of children from different socioeconomic backgrounds. Child Development, 90(4), 1303-1318.
9. Romeo, R. R., Leonard, J. A., Robinson, S. T., West, M. R., Mackey, A. P., Rowe, M. L., & Gabrieli, J. D. E. (2018). Beyond the 30-million-word gap: Children's conversational exposure is associated with language-related brain function. Psychological Science, 29(5), 700-710.

The nursing units where the better teams reported more errors

In the early 1990s Amy Edmondson, then a doctoral student, was collecting data in the nursing units of two teaching hospitals. Her hypothesis was straightforward and, she assumed, safe: units with better leadership and a stronger team climate would make fewer medication errors.

The correlation came out the wrong way round. Units rated as better led and better functioning reported more errors, not fewer, and the relationship was strong enough that it could not be waved away.

Rather than discard the result, Edmondson went back and looked at how errors came to be recorded. The units differed enormously in whether a nurse who had given the wrong dose would say so. In some, an error was an occasion for learning and was written down. In others, it was an occasion for blame and, where possible, was not. The measure she thought was counting mistakes was substantially counting willingness to speak about mistakes.

That accident produced the construct she named psychological safety, and it is the single most useful idea in this lesson.

Key idea: The safest-looking data can be a report on who feels able to speak. Before believing any organisational measure, ask what a person had to say out loud for the number to move.

What psychological safety is, and the four things it is not

Edmondson's 1999 definition is narrow and worth memorising: a shared belief, held by members of a team, that the team is safe for interpersonal risk-taking. The risks in question are specific and small. Asking a question that might reveal ignorance. Admitting an error. Proposing an idea that might be poor. Disagreeing with a person senior to you. Saying I do not think this will work.

Four confusions are common enough to be worth naming.

• It is not niceness. A team where everyone is pleasant and nobody raises a problem is not psychologically safe; it is polite.
• It is not comfort. Edmondson pairs safety with accountability deliberately. High safety with low standards produces a comfort zone in which nothing much happens. High standards with low safety produces anxiety, and anxiety produces silence and concealment. Both together produce the learning zone she is actually arguing for.
• It is not a personality trait. It is a property of a team, and the same person can be measurably safe on one team and silent on another in the same building.
• It is not permission to say anything. The construct concerns the interpersonal risks listed above, not freedom from any professional consequence.

Project Aristotle, and how to read an internal study

Psychological safety reached general awareness in 2016, when Google published an account of an internal research effort called Project Aristotle. Having examined a large number of its own teams and failed to find that the mix of individual talent explained performance, the company reported that psychological safety came out as the strongest differentiator of effective teams, followed by dependability, structure and clarity, meaning, and impact.

The finding is consistent with the academic literature, and that consistency is the reason to take it seriously rather than the study's own authority. Be clear about what it is. It was internal research, not independently peer reviewed; the sample was one company's teams; the performance criteria were the company's own; and the public account is a summary rather than a method section anyone can check. Google finding something is evidence. It is not stronger evidence than a meta-analysis merely because the company is famous.

The independent picture is reasonably good. Pamela Frazier and colleagues published a meta-analysis in 2017 pooling the psychological safety literature and found consistent positive associations with information sharing, learning behaviour, task performance and job attitudes, with supportive leadership and work design among the reliable antecedents. The construct survives outside Google.

Your manager has an in-group, and you are in it or you are not

Most leadership theory before the 1970s assumed a manager had a style, applied more or less uniformly to subordinates. Leader-member exchange theory started from the observation that this is plainly false. Managers develop different relationships with different reports, and the differences are large.

High-quality exchanges involve trust, mutual obligation and influence running in both directions. The subordinate gets more interesting assignments, more information, more access, and more latitude when something goes wrong. Low-quality exchanges run on the employment contract: defined duties, formal authority, limited information. The theory's blunt term for the two groups is in-group and out-group.

Meta-analytic work, notably by Charlotte Gerstner and David Day, finds exchange quality associated with job performance, satisfaction, organisational commitment and reduced turnover intentions. Two cautions belong with those results.

First, most of this is measured by asking the subordinate to rate the relationship and also to rate their own attitudes, so a good deal of the correlation may be one person's general positivity showing up twice. Where leader and member ratings of the same relationship are compared, agreement is only moderate, which is itself informative: the two parties frequently do not think they are in the same relationship.

Second, the theory describes rather than endorses. Differentiation happens; it is also, from the out-group's side, indistinguishable from favouritism, and the perception of unfairness carries its own costs across the team.

So what?: If you manage people, the useful question is not whether you have differentiated. You have. It is whether the people in the out-group can name the criteria that put them there, and whether those criteria are defensible.

Feedback: the intervention that sometimes makes performance worse

Avraham Kluger and Angelo DeNisi published a meta-analysis in 1996 that ought to be better known than it is. They gathered over six hundred effect sizes from studies of feedback interventions and found that on average feedback improved performance, by a moderate amount.

Then they looked at the distribution rather than the average. In more than a third of the cases, feedback made performance worse than no feedback at all. Not neutral. Worse.

Their explanation is about attention. Feedback works by directing attention somewhere, and where it lands determines what happens next. Feedback that directs attention to the task, and to the specific gap between what was done and what is required, tends to help. Feedback that directs attention to the self, meaning to the recipient's worth, ability or standing, tends to hurt, because attention spent on defending the self is attention not spent on the task. Praise directed at the person can do this as readily as criticism can.

This has consequences for practices that are taught as obviously correct.

• The feedback sandwich, in which criticism is inserted between two compliments, has no good evidence behind it. It reliably teaches recipients that compliments precede bad news, which degrades the compliments, and it obscures the message often enough that the recipient leaves unsure what was being asked.
• Ratings of the person, as against descriptions of the work, are close to the definition of self-directed attention.
• Asking first is unglamorous and well supported by the same logic: a recipient who has stated their own view of the gap is oriented to the task before you say anything.

The usable form is descriptive and specific: here is what I observed, here is the effect it had, here is what I would like to be different, what do you see. None of those clauses is about the person.

Friendship inside an institution

The American Time Use Survey, run by the Bureau of Labor Statistics, finds that employed Americans spend close to eight hours working on days that they work, which for most people is more waking time than is spent with anyone outside the household. Proximity and repeated exposure, from the previous module, do their usual work. Workplace friendships are not a bonus feature of employment; they are the predictable output of putting the same people in the same room for a decade.

They also sit under constraints no other friendship has. The relationship is embedded in an authority structure that can change without warning, when one friend is promoted over the other. It carries information asymmetries, since knowing something you may not share is a routine feature of many roles. And it is involuntary in origin, which means the exit costs discussed in the family lesson apply in a weaker form: falling out with a colleague does not end the requirement to work with them.

You will encounter the claim, drawn from Gallup's employee engagement survey, that having a best friend at work predicts engagement and performance. Treat it carefully. The item comes from a proprietary instrument, the analyses are the vendor's own, and the wording has been questioned for years, including by people sympathetic to the finding. The general association between workplace friendship and satisfaction is supported by independent research. The specific best-friend claim rests on data that outsiders cannot inspect.

What the remote work trials actually found

Opinions about remote work vastly outnumber experiments about it, so the experiments are worth knowing precisely.

Nicholas Bloom and colleagues ran a randomised trial at the Chinese travel agency Ctrip, published in 2015. Call-centre employees who volunteered were randomly assigned to work from home four days a week or to remain in the office. Home workers' performance rose by around thirteen per cent, most of it from working more minutes per shift because they started on time and took shorter breaks, and attrition fell by about half. Promotion rates for home workers, however, declined relative to office workers, which is the finding people quoting the study usually leave out.

Bloom, Ruobing Han and James Liang published a second randomised trial in 2024, this time of hybrid work at Trip.com, with over sixteen hundred employees assigned to two days a week at home or five days in the office. Over two years, hybrid work had no detectable effect on performance ratings or promotions, and it cut quit rates by about a third, with the largest reductions among women and among non-managers.

Put together, those two trials support a narrow and useful conclusion: for the roles studied, hybrid arrangements retained people without measurable performance cost, and fully remote arrangements in the earlier study carried a visible career cost. Survey work from the Pew Research Center adds the demand side, consistently finding that a large majority of workers whose jobs can be done from home would prefer to do so at least some of the time. What none of this settles is the effect on the things hardest to measure: mentoring, weak ties, and the transmission of how things are done here.

Common misconceptions

"Psychological safety means being nice to each other." Edmondson pairs it with accountability precisely because safety without standards produces comfort rather than learning. It is about the safety of specific interpersonal risks: questions, admissions, disagreement.

"Google proved psychological safety matters." Project Aristotle was internal, unreviewed research on one company's teams using its own criteria. It agrees with an independent literature, and that agreement, not Google's name, is the reason to believe it.

"A good manager treats everyone the same." Leader-member exchange research finds differentiation is near universal. The defensible version is not equal treatment but criteria the out-group can name and challenge.

"Feedback is always useful if it is delivered kindly." Kluger and DeNisi found that more than a third of feedback interventions reduced performance. What predicts harm is attention directed at the self rather than the task, and warm praise of the person can do that as effectively as criticism.

"Remote work is settled either way." The two randomised trials available point at different things: fully remote raised output and lowered promotion in one call centre, and hybrid was performance-neutral while sharply improving retention in another. Neither measures mentoring or informal learning.

Looking back

Edmondson's inverted correlation in hospital nursing units produced psychological safety, defined narrowly as a shared belief that a team is safe for interpersonal risk-taking, and paired with accountability rather than offered as comfort. Project Aristotle popularised it and should be read as internal research that agrees with an independent meta-analytic literature. Leader-member exchange describes the in-groups and out-groups every manager actually creates, with the caution that most of the evidence comes from subordinates rating both the relationship and their own attitudes, and that leaders and members agree with each other only moderately. Kluger and DeNisi's meta-analysis is the finding to carry out of this lesson: feedback helps on average and harms in more than a third of cases, depending on whether attention lands on the task or on the self. Workplace friendships are the predictable product of proximity operating inside an authority structure, and the best-friend claim rests on data nobody outside the vendor can check. And the two randomised trials of remote and hybrid work support a narrower conclusion than either side of the argument usually wants.

Sources

1. U.S. Bureau of Labor Statistics. (n.d.). American Time Use Survey. bls.gov
2. OpenStax. (2019). Work motivation and group dynamics. In Organizational behavior. Rice University. openstax.org
3. Pew Research Center. (n.d.). Research on work and employment. pewresearch.org
4. Edmondson, A. C. (1996). Learning from mistakes is easier said than done: Group and organizational influences on the detection and correction of human error. Journal of Applied Behavioral Science, 32(1), 5-28.
5. Edmondson, A. C. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44(2), 350-383.
6. Frazier, M. L., Fainshmidt, S., Klinger, R. L., Pezeshkan, A., & Vracheva, V. (2017). Psychological safety: A meta-analytic review and extension. Personnel Psychology, 70(1), 113-165.
7. Gerstner, C. R., & Day, D. V. (1997). Meta-analytic review of leader-member exchange theory: Correlates and construct issues. Journal of Applied Psychology, 82(6), 827-844.
8. Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119(2), 254-284.
9. Bloom, N., Han, R., & Liang, J. (2024). Hybrid working from home improves retention without damaging performance. Nature, 630, 920-925.

More than a hundred thousand questionnaires from one company

Between 1967 and 1973 IBM surveyed its own employees about their work values. Geert Hofstede, a Dutch psychologist running the company's personnel research department, ended up with well over a hundred thousand completed questionnaires from more than fifty countries, from people doing broadly the same jobs inside broadly the same corporate culture.

That last detail is what made the data interesting and what made the eventual claims contestable. Hofstede argued that because occupation, employer and job level were held roughly constant, differences in the country averages could be attributed to nationality. He factor-analysed the country means and found dimensions. He published Culture's Consequences in 1980, and the resulting six-dimension model became the most cited framework in cross-cultural research and the most misused framework in intercultural training.

Key idea: Every score in this lesson is a property of a country average. None of them is a property of a person, and the difference between those two sentences is the whole point of the lesson.

Hall first: context and time

Edward T. Hall was an anthropologist who trained American diplomats at the Foreign Service Institute in the 1950s, which is where his interest in the concrete mechanics of cross-cultural misunderstanding came from. His proxemic distances were covered in Module 2. Two further distinctions belong here.

High and low context. In a high-context communication style, most of the meaning sits in the situation, the relationship, the shared history and what is not said. The words carry a smaller share of the load, and a direct refusal may be conveyed by a pause and a change of subject. In a low-context style, meaning is expected to be in the message: say what you mean, put it in writing, and treat the explicit text as the agreement. Hall placed Japan, China and Arab countries towards the high-context end and Germany, Switzerland and the United States towards the low-context end.

Monochronic and polychronic time. Monochronic time treats time as a line divisible into segments that can be allocated, saved, wasted and spent, with one thing done at a time and schedules taken as commitments. Polychronic time treats several things and several people as simultaneously legitimate claims, with relationships taking precedence over the schedule. A meeting that starts twenty minutes late because the previous conversation had not finished is not a failure under the second arrangement; it is the arrangement.

These are genuinely useful categories for noticing your own assumptions. They are also weaker as science than their popularity suggests. Hall arrived at the country placements through observation, interviews and long experience rather than through measurement, and he never published data behind the rankings. Later reviews of the empirical literature have found that the high and low context classification is cited far more often than it is tested, and that the studies attempting to measure context empirically do not place countries consistently. Hold the distinction as a lens, not as a finding.

The six dimensions

Dimension	What a high score on the country average means	Where it came from
Power distance	Less powerful members of institutions expect and accept that power is distributed unequally; hierarchy is treated as natural rather than as a convenience	Original IBM factor analysis
Individualism versus collectivism	People are expected to look after themselves and their immediate family; identity is grounded in the individual rather than in the in-group	Original IBM factor analysis
Masculinity versus femininity	A preference for achievement, assertiveness and material reward over cooperation, modesty and quality of life; often relabelled in recent writing because the original name obscured the content	Original IBM factor analysis
Uncertainty avoidance	Discomfort with ambiguity and unstructured situations, handled through rules, codes and a preference for the known	Original IBM factor analysis
Long-term orientation	Perseverance and thrift oriented towards future reward, against respect for tradition and fulfilling social obligations now	Added later from the Chinese Value Survey developed with Michael Harris Bond, originally called Confucian dynamism
Indulgence versus restraint	Relatively free gratification of natural drives related to enjoying life, against suppression of gratification by strict social norms	Added in 2010 with Michael Minkov, using World Values Survey data

Notice the last two rows. The framework people describe as Hofstede's model of culture was assembled in stages over thirty years, using different instruments and different populations, and two of the six dimensions do not come from the IBM data at all.

The ecological fallacy, stated plainly

In 1950 the sociologist William Robinson published a short paper that ought to be required reading for anyone using country scores. Using the 1930 United States census, he calculated the correlation between the proportion of a state's population that was foreign-born and the proportion that was illiterate. Across the forty-eight states the correlation was strongly positive, in the region of point five: states with more immigrants had more illiteracy.

He then calculated the same relationship at the individual level, using individuals rather than states as the unit. It was slightly negative. Foreign-born individuals were, if anything, marginally less likely to be illiterate than native-born individuals.

Both numbers are correct. The state-level correlation exists because immigrants had settled in states whose native-born populations happened to be more literate, so aggregating hid the individual relationship and manufactured a different one. Robinson called the error of reading the first number as if it were the second an ecological correlation problem, and the mistake is now called the ecological fallacy.

Now apply it. Hofstede's dimensions were derived by factor-analysing country means. The factors emerged from correlations among country averages, not among people. This means the dimensions are properties of the aggregate, and their structure at the individual level may be different, weaker, or absent. Hofstede said so himself, repeatedly, and warned that his scores should never be used to describe individuals.

The practical consequence is that the sentence "Japan scores high on uncertainty avoidance" is a statement about a national average from a particular dataset, and the sentence "Yuki avoids uncertainty" does not follow from it in any degree that would justify acting on it. Within-country variation on these measures is enormous, far larger than the differences between country means. If you sample two people at random, one from each of two countries whose scores differ substantially, you will frequently find the individual difference runs the other way.

Why this matters: Applied to a person, a country score is a stereotype with a citation attached. The citation makes it more persuasive and no more accurate.

How much does culture predict?

Vas Taras, Bradley Kirkman and Piers Steel meta-analysed three decades of research using Hofstede's dimensions, covering several hundred studies. Their findings are the honest summary of what the framework buys you.

• Cultural values do predict organisational outcomes, at an average correlation in the region of point two. That is a real effect and a modest one.
• Predictions were considerably stronger when the values were measured and applied at the group or country level than at the individual level, exactly as the ecological argument implies.
• For many outcomes, personality traits and demographic variables predicted at least as well as cultural values did.
• The effects were stronger for managers and for older respondents, and weaker in more recent studies, which is what you would expect if the country scores are ageing.

A second meta-analysis, by Daphna Oyserman, Heather Coon and Markus Kemmelmeier, examined individualism and collectivism directly across a large body of studies and found the standard contrasts frequently failed. European Americans did score higher on individualism than most comparison groups. They did not consistently score lower on collectivism, and on several comparisons Japanese and Korean samples did not differ from American samples in the predicted direction at all. The reliable finding was smaller and more specific than the textbook version.

McSweeney's objection

Brendan McSweeney published a sustained critique in 2002 with a title that tells you its temperature: a triumph of faith and a failure of analysis. His main charges are worth knowing because they are not all equally strong.

1. The sample. Employees of a single multinational, in a narrow band of occupations, in a particular decade. In several countries the response numbers were small enough to make a stable national estimate implausible.
2. The inference. Answers to a workplace attitude questionnaire were treated as evidence of deep values held outside work.
3. The assumption of uniformity. The model requires a national culture that is shared, central and consistent within borders. Nations contain regions, classes, languages, religions and generations that differ from one another more than some national means differ.
4. The unit. Nation states are political and administrative units, drawn by history and war. There is no reason to expect culture to align with them.

Hofstede replied in the same journal, arguing that dimensions are constructs rather than things, that country scores are relative positions rather than absolute measurements, and that the framework's validation lies in its correlations with independent country-level data on wealth, health, education and legal arrangements, which are substantial.

Where does that leave a reader? The critique lands hard on the claim that these scores describe coherent national cultures. It does not show that the country scores are noise: they correlate with a range of independently measured national indicators, which random numbers do not do. The defensible position is that the dimensions capture something real about aggregate national tendencies, in a dataset now half a century old, and that everything beyond that has been oversold.

The successors

Framework	Method	Distinctive contribution	Main limitation
Hofstede (1980 onwards)	Factor analysis of country means from IBM employee surveys, extended later with other instruments	Six dimensions; the first systematic large-scale mapping; enormous downstream literature	Ageing data, one employer, national uniformity assumed, chronically misapplied to individuals
GLOBE (House and colleagues, 2004)	Around seventeen thousand middle managers in three industries across sixty-two societies	Nine dimensions, and the separation of practices, meaning how things are, from values, meaning how respondents think things should be	Also managerial samples; the practices and values scores correlate negatively on several dimensions, which is difficult to interpret
Schwartz (1990s onwards)	Value surveys of teachers and students across seventy-plus countries, with a theoretically derived value structure	Three bipolar cultural dimensions: embeddedness against autonomy, hierarchy against egalitarianism, mastery against harmony; explicit separation of individual-level and culture-level structures	Less used in applied settings; samples of teachers and students are their own narrow slice

The GLOBE practices-and-values split produced the most interesting single result in this area. On several dimensions, societies that scored high on the practice scored low on the corresponding value, and the other way round. Societies with high power distance in practice reported wanting less of it. One reading is that people want what they do not have. Another is that the two scales are measuring different things and should not be compared. The disagreement between Hofstede and the GLOBE authors, conducted in print in 2006, is a useful thing to read if you want to see how much of cross-cultural measurement is contested by the people who built it.

Using a dimension without committing the fallacy

Here is a concrete case. You are in a meeting with a colleague who works in an office in a country whose scores on power distance are considerably higher than your own country's. You ask the group whether anyone sees a problem with the plan. Your colleague says nothing. Later, privately, they raise three serious objections.

The fallacy version says: high power distance culture, therefore this person defers to authority, therefore ask them privately from now on. That reasoning has skipped the individual entirely and locked in a conclusion.

The defensible version treats the dimension as a source of hypotheses you would not otherwise have generated. It might be about hierarchy. It might be that the objections were only half formed at the time. It might be a low-context expectation that unsolicited criticism in front of a client is rude. It might be that this person is quiet in meetings and always has been, which is a fact about them and not about their country. The dimension has done its work if it widened your list of candidate explanations from one to four.

Then you do the only thing that resolves it, which is to ask. Not asking whether the culture is hierarchical, which invites a generalisation, but asking a specific question about the specific situation: what would make it easier to raise that kind of objection in the meeting itself, or would you rather I checked with you beforehand.

Common misconceptions

"Hofstede's dimensions tell you how someone from that country will behave." They are country averages produced by factor-analysing country means. Robinson's 1950 census example shows why a relationship at the aggregate level can be absent or reversed at the individual level. Hofstede issued this warning himself.

"The six dimensions all come from the IBM study." Four do. Long-term orientation came later from the Chinese Value Survey and indulgence versus restraint was added in 2010 using World Values Survey data.

"High-context and low-context cultures are an established empirical classification." Hall derived the placements from observation and never published supporting data, and later attempts to measure context empirically have not placed countries consistently.

"Culture is the main thing shaping how a colleague behaves at work." Taras and colleagues found average correlations around point two, stronger at country level than individual level, and comparable to or weaker than personality and demographics for many outcomes.

"Collectivist cultures are simply the opposite of individualist ones." Oyserman and colleagues found the two are not a single scale in the data, and several of the most repeated national contrasts did not appear.

Where this leaves us

Hall gave the field its most usable vocabulary, context and time, on the strength of observation rather than data, and the categories should be held as lenses. Hofstede assembled six dimensions over thirty years from a corporate survey and two later instruments, and produced country scores that correlate meaningfully with independent national indicators. Robinson's 1950 paper explains exactly why those scores cannot be transferred to a person: an aggregate correlation can be absent, or reversed, at the individual level, and within-country variation on these measures dwarfs the differences between country means. The meta-analytic picture is a modest real effect, stronger at the level the data were collected at, and rivalled by personality and demographics. McSweeney's critique defeats the strong claim about coherent national cultures without reducing the scores to noise. GLOBE and Schwartz improved the measurement and did not settle the argument, and GLOBE's practices-and-values split shows how unresolved this area still is. The competent use of any of it is to generate hypotheses you would not otherwise have had, and then ask the person in front of you.

Sources

1. University of Minnesota Libraries Publishing. (2016). Culture and communication. In Communication in the real world. open.lib.umn.edu
2. Encyclopaedia Britannica. (n.d.). Culture. britannica.com
3. Hofstede Insights. (n.d.). The six dimensions of national culture. hofstede-insights.com (the model's own commercial home; useful for the scores, not for their limits)
4. Hall, E. T. (1976). Beyond culture. Anchor Press.
5. Hofstede, G. (2001). Culture's consequences: Comparing values, behaviors, institutions and organizations across nations (2nd ed.). Sage.
6. Robinson, W. S. (1950). Ecological correlations and the behavior of individuals. American Sociological Review, 15(3), 351-357.
7. McSweeney, B. (2002). Hofstede's model of national cultural differences and their consequences: A triumph of faith, a failure of analysis. Human Relations, 55(1), 89-118.
8. Taras, V., Kirkman, B. L., & Steel, P. (2010). Examining the impact of Culture's Consequences: A three-decade, multilevel, meta-analytic review of Hofstede's cultural value dimensions. Journal of Applied Psychology, 95(3), 405-439.
9. Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and collectivism: Evaluation of theoretical assumptions and meta-analyses. Psychological Bulletin, 128(1), 3-72.
10. House, R. J., Hanges, P. J., Javidan, M., Dorfman, P. W., & Gupta, V. (Eds.). (2004). Culture, leadership, and organizations: The GLOBE study of 62 societies. Sage.

A talk to the Women's Club of Rio de Janeiro

In 1954 Kalervo Oberg, an anthropologist working in Brazil, addressed the Women's Club of Rio de Janeiro on a condition he had watched afflict foreign residents. He called it culture shock, and his description was concrete to the point of bluntness: excessive hand-washing, excessive worry about the drinking water and the dishes, fear of being cheated, irritation over delays, an absent-minded faraway stare, and a strong desire to spend all available time with other people from home.

His diagnosis was better than his list of symptoms. Culture shock, he said, is precipitated by the loss of familiar signs and symbols of social intercourse. Not by the food or the heat, but by the removal of the thousand small cues that tell you when to speak, how close to stand, what a laugh means, whether an agreement has been reached. Those cues are learned so early and used so automatically that their absence is felt as something wrong with the world rather than as a gap in your knowledge.

Oberg also proposed four stages, and that is where the trouble starts.

The point: The insight about lost cues has held up and is worth carrying. The stage model attached to it has not.

The curve that kept being drawn

The year after Oberg's talk, Sverre Lysgaard published a study of Norwegians who had spent time in the United States on Fulbright grants. Those who had been in the country between roughly six and eighteen months reported more difficulty than those who had arrived more recently or stayed longer. Plot adjustment against time and you get a U: an initial high, a trough, a recovery.

The shape was intuitive, memorable and quickly extended. John and Jeanne Gullahorn proposed a W-curve in 1963, adding a second dip for the return home, which sojourners often report as harder than the departure. The U-curve and W-curve entered orientation programmes, expatriate handbooks and study abroad briefings, where they remain.

Then people checked. Timothy Church reviewed the sojourner adjustment literature in Psychological Bulletin in 1982 and concluded that support for the U-curve was weak, overgeneralised and based on inadequate designs. Stewart Black and Mark Mendenhall revisited the question in 1991 and found that most of the studies commonly cited in its support did not actually test it properly: many were cross-sectional, comparing different people at different stages rather than following the same people over time; the measures of adjustment varied so much that studies were not comparable; and few established a baseline before departure.

What the better longitudinal work finds is that adjustment trajectories are heterogeneous. Some people do dip and recover. Some decline steadily. Some improve from the start. Averaging across a group with different shapes produces a curve that describes nobody.

Two things are worth taking from this. First, if you are abroad and having a hard time in month eight, the U-curve does not entitle you to expect month fourteen to be better. Second, and more generally, a model that survives because it is easy to draw on a whiteboard is exactly the kind of model to be suspicious of.

Two different things called adjustment

The replacement for the curve is better than the curve, and it comes from Wendy Searle and Colleen Ward. They separated two outcomes that had been run together.

	Psychological adjustment	Sociocultural adaptation
What it is	Emotional wellbeing and satisfaction in the new setting	The ability to fit in and manage daily life: getting things done, reading situations, being understood
Predicted by	Personality, social support, life changes, coping style	Length of residence, language ability, cultural distance, amount of contact with hosts
Time course	Variable, often fluctuating with events	A learning curve: fastest early, then flattening
What helps	Support, both from co-nationals and hosts, and stability in other parts of life	Practice, instruction, feedback, and exposure with correction

The distinction has immediate practical use, because the two problems have different solutions. Someone who is miserable but functioning has a psychological adjustment problem, and more language lessons will not help. Someone who is cheerful but keeps causing offence has a sociocultural adaptation problem, and reassurance will not help. Confusing the two is the standard failure of well-meaning support programmes.

Berry's four strategies, and who actually controls them

John Berry framed acculturation as the answer to two questions a person or group faces on entering a new society. Is it valuable to maintain my heritage culture and identity? Is it valuable to maintain relationships with the larger society? Cross the two and four strategies appear.

• Integration: yes to both. Maintain the heritage culture and participate in the larger society.
• Assimilation: no to heritage, yes to participation. Absorb into the receiving society.
• Separation: yes to heritage, no to participation. Maintain the original culture and avoid the wider society.
• Marginalisation: no to both. Little interest in or possibility of either.

Meta-analytic work on biculturalism finds integration associated with better psychological and sociocultural outcomes than the alternatives, which is the finding usually quoted. Two qualifications keep it honest.

The first is Berry's own, and it is routinely dropped. These are not free individual choices. Integration is only available when the receiving society permits it. Berry pairs each individual strategy with a societal one: multiculturalism makes integration possible, a melting pot ideology pushes assimilation, segregation enforces separation, and exclusion produces marginalisation. A person who would choose integration and lives in a society that will not have them has not chosen separation; it has been chosen for them. Any use of this framework that treats the four cells as personal attitudes has removed the part that carries the politics.

The second is Floyd Rudmin's methodological critique. The way the two dimensions are measured and scored strongly affects which cell people land in, different scoring approaches produce substantially different distributions, and marginalisation is endorsed so rarely that it may be partly an artefact of the measurement rather than a real position. The integration advantage is probably real; its size depends on decisions made in the coding.

Competence, and who gets to judge it

Intercultural competence is usually built from three components: motivation, meaning whether you want to engage and how anxious you are; knowledge, meaning what you know about the other culture and about culture in general; and skill, meaning what you can actually do. The definition that matters adds a fourth element that changes everything: competence is judged in terms of appropriateness and effectiveness, and appropriateness is assessed by the other person, not by you.

Darla Deardorff produced the most-used process model by a route worth knowing. Rather than proposing a model herself, she ran a Delphi study: successive rounds of questions to a panel of intercultural scholars, feeding each round's results back until agreement emerged about what the field actually accepted. The panel converged on a sequence. Attitudes come first, specifically respect, openness and curiosity. These enable knowledge and skills, particularly listening, observing and analysing. Those produce an internal outcome, an adaptable and empathic frame of reference. That produces the external outcome, which is behaving and communicating effectively and appropriately as judged by others.

The panel also agreed on something that most training programmes ignore: competence should be assessed by the other party rather than through self-report, because a confident sense of one's own cultural sensitivity is exactly what incompetence often feels like from inside.

Milton Bennett's developmental model describes six positions, moving from ethnocentric to ethnorelative: denial that difference exists, defence against it as a threat, minimisation which acknowledges surface difference while insisting everyone is fundamentally the same, then acceptance, adaptation and integration. Minimisation is the one to watch, because it is the position most learners mistake for the destination. We are all human underneath sounds generous and functions to make the other person's actual difference unmentionable. The model's developmental claim, that people move through these in order, rests mostly on cross-sectional data and should be held loosely; the descriptions of the positions are useful regardless.

Cultural intelligence, developed by Christopher Earley and Soon Ang, breaks capability into four factors: metacognitive, meaning awareness of and control over your own cultural thinking; cognitive, meaning knowledge; motivational, meaning drive and confidence; and behavioural, meaning flexibility of action. It correlates with adjustment and performance. Be aware that most of that evidence comes from self-report measures answered by the same people who report the outcomes, and that the incremental prediction over personality and general mental ability is modest.

Does going there work?

Here is the question a study abroad brochure never asks. Does living in another country actually increase intercultural competence?

The Georgetown Consortium Project, reported by Michael Vande Berg, Jeffrey Connor-Linton and R. Michael Paige in 2009, is the largest attempt to find out. More than a thousand American students in over sixty study abroad programmes were assessed before and after using an established measure of intercultural development, with a control group who stayed home.

On average, students abroad gained very little. Immersion by itself did not produce measurable development, and some programmes produced none at all.

What did predict gains was structure. Programmes with cultural mentoring, meaning regular guided reflection with someone trained to do it, produced substantially larger gains. So did taking courses alongside local students rather than in a group of visiting students, and living arrangements that required interaction with hosts. The students who improved most were those whose experience was deliberately designed and facilitated. The students who improved least were those who went abroad with their compatriots and stayed with them.

Bottom line: Exposure is not learning. Contact produces development when it is structured, reflected on and supported, and this is the same conclusion the contact literature reaches in the next lesson from a completely different direction.

A tool worth the five minutes it takes to learn

Describe, interpret, evaluate is a discipline for separating three things people do simultaneously and unconsciously.

A colleague on a video call looks away from the camera for most of a performance review. Written out properly:

• Describe. Only what a camera would record. She looked down or to the side for most of the twenty minutes and made brief eye contact when she spoke.
• Interpret. Generate several, not one. She was uncomfortable. She was showing respect to a senior colleague. She was taking notes. She finds sustained eye contact on video draining. She was reading something else. In her workplace, looking directly at someone while being appraised would be read as challenge.
• Evaluate. Only after checking, and only if a judgement is genuinely needed.

The value is not that the tool produces the right answer. It is that it makes visible how quickly you had produced a wrong one, and how the interpretation you reached first was the one your own culture supplied.

Common misconceptions

"Culture shock follows a U-curve." Church's 1982 review and Black and Mendenhall's 1991 reassessment found the evidence weak and the supporting studies mostly cross-sectional. Individual trajectories vary enough that the average curve describes nobody in particular.

"Adjustment is one thing." Searle and Ward showed psychological adjustment and sociocultural adaptation have different predictors and different time courses. Language classes do not fix loneliness, and reassurance does not fix causing offence.

"Acculturation strategy is a personal choice." Berry pairs each individual strategy with a societal one. Integration requires a receiving society that permits it, and where it does not, the resulting separation or marginalisation was not chosen.

"Recognising that we are all fundamentally the same is the goal." That is Bennett's minimisation position, still on the ethnocentric side of his model. It functions to make the other person's actual difference unmentionable.

"Studying abroad makes you interculturally competent." The Georgetown Consortium Project found little average gain from immersion alone. Cultural mentoring, courses with local students and structured reflection predicted the gains.

The takeaway

Oberg named the real mechanism in 1954: the loss of the small cues that make social life automatic. The stages he attached to it, and Lysgaard's U-curve, did not survive review, because adjustment trajectories differ enough that the average shape belongs to no individual. Searle and Ward's split between psychological adjustment and sociocultural adaptation is the better frame and points at different remedies. Berry's four strategies are only intelligible alongside the societal strategies that permit or forbid them, and their measurement is contested even where the integration advantage holds. Competence is motivation, knowledge and skill judged for appropriateness by the other party, which is why Deardorff's panel put attitudes first and self-assessment last, and why Bennett's minimisation stage is the trap that feels like arrival. Cultural intelligence is a useful construct measured mostly by self-report. And the largest study of study abroad found that immersion on its own does very little, while structured mentoring and genuine contact with hosts do a great deal.

Sources

1. University of Minnesota Libraries Publishing. (2016). Intercultural communication. In Communication in the real world. open.lib.umn.edu
2. American Psychological Association. (n.d.). APA dictionary of psychology: acculturation, culture shock. dictionary.apa.org
3. OpenStax. (2021). Culture. In Introduction to sociology 3e. Rice University. openstax.org
4. Oberg, K. (1960). Cultural shock: Adjustment to new cultural environments. Practical Anthropology, 7(4), 177-182.
5. Church, A. T. (1982). Sojourner adjustment. Psychological Bulletin, 91(3), 540-572.
6. Black, J. S., & Mendenhall, M. (1991). The U-curve adjustment hypothesis revisited: A review and theoretical framework. Journal of International Business Studies, 22(2), 225-247.
7. Searle, W., & Ward, C. (1990). The prediction of psychological and sociocultural adjustment during cross-cultural transitions. International Journal of Intercultural Relations, 14(4), 449-464.
8. Berry, J. W. (1997). Immigration, acculturation, and adaptation. Applied Psychology: An International Review, 46(1), 5-34.
9. Deardorff, D. K. (2006). Identification and assessment of intercultural competence as a student outcome of internationalization. Journal of Studies in International Education, 10(3), 241-266.
10. Vande Berg, M., Connor-Linton, J., & Paige, R. M. (2009). The Georgetown Consortium Project: Interventions for student learning abroad. Frontiers: The Interdisciplinary Journal of Study Abroad, 18, 1-75.

Two ways to build a housing project

In the late 1940s, public housing authorities in New York City and in Newark took different approaches to race. In the New York projects Morton Deutsch and Mary Ellen Collins studied, Black and white families were assigned to apartments as they came available, so the buildings were integrated floor by floor. In the Newark projects, families were assigned to separate buildings within the same development, so residents lived in the same project and different worlds.

Deutsch and Collins interviewed white women residents in both. The differences were large. Women in the integrated projects reported far more contact with their Black neighbours, described those relationships in more favourable terms, and frequently volunteered that their own views had changed since moving in. Women in the segregated projects, living a few hundred metres from the same neighbours, reported little contact and held to their previous views.

The design matters as much as the result. Residents did not choose which project or which building they were assigned to, so the usual objection, that tolerant people seek out mixed neighbourhoods, has much less purchase here than in an ordinary survey. Published in 1951, this study is one of the reasons the idea that followed it was taken seriously.

Worth holding on to: The variable was not attitude, education or exposure to argument. It was the floor plan. Structural arrangements decide who has the contact that changes anything.

Allport's four conditions

Gordon Allport published The Nature of Prejudice in 1954, and its most influential passage does not say that contact reduces prejudice. It says contact reduces prejudice under conditions, and names four.

1. Equal status within the situation. Not equal status in society at large, which is rarely available, but equal standing in the contact itself. A manager and a cleaner meeting as manager and cleaner does not qualify.
2. Common goals. Something both parties are trying to achieve, which orients attention outward.
3. Intergroup cooperation. The goal must require working together rather than competing. This is the condition that Muzafer Sherif's summer camp experiments had already shown to matter, where competition between two groups of boys generated hostility fast and only superordinate goals dismantled it.
4. Support of authorities, law or custom. The institution must visibly sanction the contact. Contact that participants believe their employer, school or government disapproves of does not work the same way.

Thomas Pettigrew later argued for a fifth: the situation must offer the potential for friendship, meaning repeated contact over time with the possibility of getting to know individuals rather than encountering representatives.

Notice that all five are properties of situations. None is a property of an attitude. Whatever else the contact literature has established, it has consistently found that the way to change intergroup attitudes is to change intergroup situations.

What stereotypes are doing, and why they are not simply errors

Module 1 established that perception works by selecting, organising and interpreting, and that categorisation is not optional. Stereotypes are what categorisation looks like when the objects being categorised are people. That is why they cannot be abolished by good intentions, and why the useful questions are about content and application rather than existence.

Susan Fiske, Amy Cuddy, Peter Glick and Jun Xu proposed the stereotype content model, which organises group stereotypes on two dimensions: warmth, meaning whether the group is seen as intending good or harm, and competence, meaning whether it is seen as able to act on those intentions. Cross them and four quadrants appear, each with a characteristic emotion and a characteristic behaviour.

	High competence	Low competence
High warmth	Admiration; active and passive help. Usually the in-group and close allies	Pity and sympathy; passive help alongside neglect. Often the elderly and disabled people
Low warmth	Envy and resentment; passive association with active harm under stress. Often applied to prosperous minorities and professionals	Contempt and disgust; active harm and active neglect. Applied to the most marginalised groups

The model's contribution is the ambivalent cells. A stereotype that grants competence while denying warmth licenses resentment, and one that grants warmth while denying competence licenses condescension, which is why being liked is not the same as being respected and why some prejudice is expressed entirely through kindness.

Pettigrew added a bridge to Module 1's attribution material. The ultimate attribution error is the group-level version of the fundamental attribution error: a negative act by an out-group member is attributed to something dispositional about their group, while the same act by an in-group member is attributed to circumstances. Positive acts reverse the pattern and are explained away as exceptions, luck, or special advantage. The result is that the stereotype cannot be disconfirmed by evidence, because the attribution machinery processes disconfirming evidence into exceptions.

The implicit bias story, told straight

In 1998 Anthony Greenwald, Debbie McGhee and Jordan Schwartz published the Implicit Association Test, which measures how quickly people sort words and images when categories are paired in different ways. It became one of the most widely taken psychological instruments in history, and its results, which showed that most people respond faster to some pairings than others, were widely interpreted as revealing hidden individual prejudice.

Here is what the subsequent evidence supports, and it is narrower.

• Prediction of behaviour is weak. Frederick Oswald and colleagues meta-analysed studies relating IAT scores to discriminatory behaviour and found correlations small enough that individual scores carry very little information about what a person will do.
• Individual scores are unstable. Test-retest reliability for the race IAT is modest, which means a person can score quite differently on two occasions. That is fatal for any use of the score as a personal diagnosis.
• Changing the score does not change behaviour. Patrick Forscher and colleagues meta-analysed nearly five hundred studies of procedures designed to shift implicit measures. Many procedures did shift the measures, at least briefly. Those shifts did not produce corresponding changes in explicit attitudes or in behaviour.

None of this shows that automatic associations do not exist, or that discrimination is not real and measurable by other means. What it shows is that the instrument tells you about associations available in a culture, aggregated across many people, and does not tell you what any particular person is going to do. Using an IAT score to select, screen or assign blame is not supported by the research, and Greenwald and Banaji have themselves cautioned against individual diagnostic use.

The upshot: A measure can be scientifically productive at the group level and worthless as an individual test. That is not a contradiction; it is the ecological point from the previous module arriving from a different direction.

What contact actually does

Pettigrew and Linda Tropp published the definitive quantitative summary in 2006: a meta-analysis of over five hundred studies, comprising more than seven hundred independent samples and a quarter of a million participants across dozens of countries.

Two results stand out. The average association between intergroup contact and reduced prejudice was around minus point two, negative meaning more contact with less prejudice, and it held across outgroup types, age groups and countries. And Allport's conditions turned out to be facilitating rather than necessary: studies meeting the conditions showed larger effects, but studies without them still showed effects in the same direction.

The effect also generalised. Contact with particular members of an outgroup was associated with more favourable attitudes towards the outgroup as a whole, and sometimes towards other outgroups not involved in the contact at all.

Then, in 2019, Elizabeth Levy Paluck, Seth Green and Donald Green audited what that literature is made of, and the audit is as important as the meta-analysis. Their finding was about design. Very few studies in the contact literature randomly assign people to contact and then measure outcomes at least a day later, which is the minimum for a claim about durable causal effect. Most are correlational, and correlational contact studies have an obvious confound: people who are less prejudiced seek more contact. Among the studies that do meet the stronger design standard, a large share concern outgroups other than racial and ethnic ones, and the evidence about the settings people most want to apply this to is correspondingly thinner.

The honest summary is that contact is one of the better supported interventions in social psychology and that the strength of that support is frequently overstated. Both halves of that sentence are load-bearing.

A fraud, a retraction, and a real result

In December 2014 Science published a paper by Michael LaCour and Donald Green reporting that a brief conversation with a gay canvasser produced large and lasting changes in attitudes to same-sex marriage. It was covered everywhere, and it was too good.

David Broockman and Joshua Kalla, graduate students planning an extension, tried to replicate the survey methodology and found the response rates impossible to reproduce. Working with Peter Aronow, they examined the dataset itself and concluded the survey data had not been collected as described. Green, who had not collected the data, requested a retraction. Science retracted the paper in May 2015.

The important part came next. Broockman and Kalla ran their own field experiment properly and published it in Science in 2016. Canvassers went door to door in Miami and had roughly ten-minute conversations with voters about transgender rights, built around actively taking the perspective of someone unlike themselves and inviting the voter to recall a time they had been judged for being different. Around five hundred voters were randomly assigned. The intervention reduced measured prejudice against transgender people, and the effect was still detectable three months later, and survived exposure to attack advertising.

So the field lost a spectacular fabricated result and gained a smaller genuine one, obtained by the people who caught the fraud. If you want a single case study in how error correction is supposed to work, this is it, and the sequence matters more than either paper.

Why diversity training mostly does not do what it is bought to do

Alexandra Kalev, Frank Dobbin and Erin Kelly examined more than seven hundred United States firms across three decades, using federal employment data to ask which diversity programmes were followed by actual increases in the representation of women and minorities in management. Mandatory diversity training for managers was not associated with the increases, and in some analyses was associated with declines. What was associated with change was structural: establishing responsibility through diversity taskforces and dedicated staff, and formal mentoring programmes.

Katerina Bezrukova and colleagues later meta-analysed four decades of diversity training evaluations. Training reliably produces gains in what people know, meaning cognitive learning, and those gains persist reasonably well. Effects on attitudes and on behaviour are smaller, and they decay. Training run as a one-off session, and training that is mandatory and framed as a legal requirement, does worst.

Read alongside the contact literature, the pattern is consistent and unsurprising. Structures that change who works with whom, on what terms, produce changes. Sessions that change what people can say produce changes in what people can say.

Speaking across the line: accommodation and its failures

Howard Giles's communication accommodation theory describes what speakers do with their speech when they interact with someone they perceive as different. Convergence adjusts towards the other person's style, in accent, pace, vocabulary or gesture, and generally signals liking and reduces social distance. Divergence accentuates difference, sometimes to assert identity, sometimes to mark distance deliberately.

The failure mode has a name. Overaccommodation adjusts too far, on the basis of a group stereotype rather than the person in front of you. Speaking loudly and slowly to someone with a foreign accent, using simplified vocabulary with a wheelchair user, or addressing an older person in the high-pitched, sing-song register sometimes called elderspeak are all overaccommodation. Each communicates an assessment of the listener's capacity that the listener did not authorise.

The communication predicament of ageing model traces where this goes. A speaker sees age cues, applies a stereotype, and speaks in a patronising register. The older person, addressed as incapable, participates less and offers less. The reduced participation confirms the speaker's original assessment, and the next conversation begins further along. Everyone involved is behaving reasonably given what they perceive, and the loop degrades the older person's actual opportunities to use the abilities they have.

Common misconceptions

"Allport showed that contact reduces prejudice." He argued it does so under four conditions, all of which are properties of situations rather than of attitudes. Pettigrew and Tropp later found the conditions facilitate rather than gate the effect.

"The IAT reveals your hidden prejudice." Individual scores predict behaviour weakly, are only moderately stable across occasions, and shifting them does not shift behaviour. The measure is informative about cultural associations in aggregate and is not a personal diagnostic.

"Five hundred studies prove contact works." Pettigrew and Tropp found a consistent association. Paluck, Green and Green's audit found that few of those studies randomly assign contact and measure outcomes later, and that the strongest designs cluster on outgroups other than racial and ethnic ones.

"Diversity training reduces bias." Kalev and colleagues found mandatory manager training unassociated with, and sometimes inversely related to, actual increases in managerial diversity, while structural accountability measures were associated with gains. Meta-analysis finds knowledge gains that last and attitude and behaviour effects that decay.

"Adjusting how you speak to someone is always courteous." Overaccommodation adjusts on the basis of a group stereotype and communicates an assessment of capacity, which is why elderspeak and simplified foreigner talk are experienced as insults by their recipients.

Putting it together

Deutsch and Collins found the attitudes of white residents differed sharply between integrated and segregated public housing that they had not chosen, which is why Allport's four conditions, all situational, were taken seriously. Stereotypes are categorisation applied to people, organised by warmth and competence in ways that make condescension and resentment distinct forms of prejudice, and protected from disconfirmation by the ultimate attribution error. The IAT measures something real at the level of cultures and does not work as an individual diagnosis: weak behavioural prediction, modest test-retest stability, and no transfer from changed scores to changed behaviour. Contact is genuinely one of the better supported interventions in the field, with an average association around minus point two across a quarter of a million participants, and Paluck's audit shows the causal evidence is thinner than the headline count implies. The LaCour retraction and the Broockman and Kalla replication show error correction working, and leave a real ten-minute intervention with effects lasting three months. Diversity training produces knowledge; structures that assign responsibility produce diversity. And accommodation, applied from a stereotype rather than to a person, becomes the mechanism that manufactures the incapacity it assumed.

Sources

1. Encyclopaedia Britannica. (n.d.). Prejudice. britannica.com
2. American Psychological Association. (n.d.). APA dictionary of psychology: prejudice, stereotype, contact hypothesis. dictionary.apa.org
3. University of Minnesota Libraries Publishing. (2016). Intercultural communication and social identity. In Communication in the real world. open.lib.umn.edu
4. Deutsch, M., & Collins, M. E. (1951). Interracial housing: A psychological evaluation of a social experiment. University of Minnesota Press.
5. Allport, G. W. (1954). The nature of prejudice. Addison-Wesley.
6. Fiske, S. T., Cuddy, A. J. C., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82(6), 878-902.
7. Pettigrew, T. F., & Tropp, L. R. (2006). A meta-analytic test of intergroup contact theory. Journal of Personality and Social Psychology, 90(5), 751-783.
8. Paluck, E. L., Green, S. A., & Green, D. P. (2019). The contact hypothesis re-evaluated. Behavioural Public Policy, 3(2), 129-158.
9. Forscher, P. S., Lai, C. K., Axt, J. R., Ebersole, C. R., Herman, M., Devine, P. G., & Nosek, B. A. (2019). A meta-analysis of procedures to change implicit measures. Journal of Personality and Social Psychology, 117(3), 522-559.
10. Broockman, D., & Kalla, J. (2016). Durably reducing transphobia: A field experiment on door-to-door canvassing. Science, 352(6282), 220-224.
11. Kalev, A., Dobbin, F., & Kelly, E. (2006). Best practices or best guesses? Assessing the efficacy of corporate affirmative action and diversity policies. American Sociological Review, 71(4), 589-617.

1984, and a set of terminals

In 1984 Sara Kiesler, Jane Siegel and Timothy McGuire published a paper in American Psychologist reporting what happened when small groups made decisions by typing to each other rather than sitting in a room. Three things differed from face-to-face groups. Participation was more equal, with the usual dominance of high-status members reduced. Decisions took longer and reached more extreme positions. And uninhibited behaviour was more common: rude, impulsive and hostile remarks that participants would not have made in person. Kiesler's group gave that behaviour its lasting name.

They were describing text terminals on a university network, before the web, before mobile phones, before any of the tools you use. Every result has been argued about since. But the paper set the terms of the field, and its central assumption, that removing cues removes something essential, dominated thinking for the next decade.

Key idea: The founding question of this literature was what mediated communication lacks. That framing built in an answer, and the field's biggest correction came from someone who asked a different question.

The cues-filtered-out era

Two theories, both developed for organisational settings, formalised the assumption.

Social presence theory, from John Short, Ederyn Williams and Bruce Christie in 1976, held that media differ in the degree to which they convey the sense that another person is really there, and that lower social presence produces more impersonal, task-focused interaction.

Media richness theory, from Richard Daft and Robert Lengel in 1986, ranked channels by their capacity to carry information: the availability of immediate feedback, the number of cues, the naturalness of the language, and the ability to be personally focused.

Channel	Richness	Recommended for
Face to face	Highest: immediate feedback, full cue set, personal focus	Ambiguous, emotionally loaded or conflictual matters
Video call	High, minus touch, shared space and reliable eye contact	Discussion needing visible reactions across distance
Telephone	Moderate: immediate feedback and vocal cues, no visual channel	Negotiation and sensitive news when distance rules out meeting
Personal written message	Lower, but addressed to a specific person	Detail that must be retained, reasoning that must be reread
Bulk or formal document	Lowest: no feedback, no personal address	Unambiguous information going to many people

The theory's practical advice is still sound. Delivering bad news, resolving a conflict or discussing anything with high potential for misreading belongs in a richer channel, and the impulse to handle a difficult conversation by message is usually the impulse to avoid the conversation. The prediction the theories made about relationships, though, was wrong.

Walther's correction: rate, not capacity

Joseph Walther pointed out in 1992 that almost all of the studies underlying the cues-filtered-out conclusion were one-shot laboratory tasks with zero-history groups, typically lasting under an hour and with no expectation of future interaction. If people using text need more messages to convey the same relational information, then a study that gives them thirty minutes will find them behind, and will attribute to the medium what is actually a fact about time.

Social information processing theory makes two claims. First, people are motivated to form impressions and develop relationships regardless of channel, and will use whatever cues the channel provides. Second, text-based communication does carry relational information, in word choice, message length, response timing, punctuation, emoji, and the willingness to disclose. What differs is the rate at which relational information accumulates, not the ceiling.

The prediction is testable and has been tested. Over extended interaction, differences between computer-mediated and face-to-face groups in relational tone and impression development shrink and often disappear. Anyone who has maintained a close friendship largely by text over years has run the experiment personally.

Note what the two positions each get right. Media richness is a good theory of a single conversation. Social information processing is a good theory of a relationship. Choosing a channel for tonight is a richness question; the developing closeness over two years is a rate question.

Hyperpersonal: when mediated beats face to face

Walther went further. Under some conditions, mediated communication produces relationships that are more intimate and more idealised than the equivalent face-to-face relationship. He called this the hyperpersonal model, and it has four moving parts.

1. The sender selects. Writing lets you present a curated version: your best thoughts, without the pauses, the bad hair day, or the sentence you would have started badly.
2. The receiver idealises. Given limited cues, people do not suspend judgement. They fill the gaps, and they tend to fill them favourably, exaggerating the similarity and desirability of the person they are constructing.
3. The channel permits it. Asynchrony gives you time to compose. Editability lets you revise. Neither is available in speech, where the sentence leaves before you have finished evaluating it.
4. Feedback confirms it. This is the part people miss. If someone treats you as witty and thoughtful, you behave more wittily and thoughtfully, which supplies evidence for their impression. The loop is self-reinforcing, and neither party is deceiving anyone.

This explains the specific disappointment of meeting an online correspondent in person. You have not discovered that they were lying. You have lost the conditions that produced the idealisation on both sides, and you are both now unable to edit.

Warranting theory follows from the same logic. Information that the target cannot easily control carries more weight in impression formation than information they can. A person's own description of themselves is cheap; a comment left by a friend, a photograph taken by someone else, a record of past behaviour, or an interaction visible to third parties is expensive to fake and is weighted accordingly. This is why people read the comments and not the biography.

Deception online is real and smaller than expected, and warranting is why. Catalina Toma, Jeffrey Hancock and Nicole Ellison compared online dating profiles with the physical measurements of the people who wrote them, and found that a large majority contained at least one inaccuracy about height, weight or age. The inaccuracies were mostly small: a couple of centimetres, a few kilograms, a year or two. Daters were lying within the range they expected to be able to explain when the other person met them. The anticipation of a face-to-face meeting acts as a warrant and constrains the lie.

What people actually do with these tools

It is worth separating what surveys measure from what commentary asserts. Pew Research Center's tracking finds that around nine in ten American adults own a smartphone, and that roughly half of American teenagers report being online almost constantly. Those numbers describe access and frequency, and they are frequently used to support claims about isolation that they do not contain.

What the survey work consistently finds about purpose is more interesting: most personal use of these tools is directed at maintaining relationships that already exist, rather than at forming new ones. The dominant activity is contact with people you already know. That is a substantially different picture from the one implied by the phrase online life, and it means much mediated communication is doing the maintenance work described in Module 3, in a channel that makes assurances cheap and frequent.

Why video calls are tiring

Jeremy Bailenson set out four mechanisms in 2021, and the value of his account is that each is specific and each suggests a remedy.

• Excessive close-up eye gaze. On a grid of faces, everyone appears to be looking at you, at a size and distance that in person would signal either intimacy or threat. Your nervous system is receiving a lot of very close faces.
• Cognitive load. Nonverbal behaviour that is automatic in person becomes effortful. You exaggerate nods to signal listening, you hold your head in frame, you monitor whether your expression reads correctly through a compressed video stream.
• Constant self-view. You spend hours looking at a live image of your own face, an arrangement that has no precedent in human interaction and that reliably increases self-focused attention.
• Reduced mobility. The camera pins you to a small physical box. In a room or on a phone call you move, and the movement is not incidental to thinking.

The remedies follow directly: hide self-view, shrink the window so faces are not life-sized, use audio only for some meetings, and stand up. Work developing fatigue scales has found reliable individual differences too, with women reporting higher fatigue on average, which is consistent with the self-view and mirror-anxiety mechanisms.

The screen time claim, and what a specification curve is

You will have heard that screen use is damaging a generation. Amy Orben and Andrew Przybylski tested the claim in 2019 in a way that deserves explaining, because the method is more important than the result.

The problem with large survey datasets is that a researcher makes dozens of defensible choices: which measure of technology use, which measure of wellbeing, which control variables, which subgroup. Each combination is a specification, and the number of defensible combinations runs into the thousands or millions. Report one, and you have reported a choice. A specification curve analysis computes them all and shows the whole distribution of results.

Orben and Przybylski did this on three large datasets covering more than three hundred thousand adolescents. Across the full space of specifications, the association between digital technology use and adolescent wellbeing was negative on average and extremely small: on the order of a few tenths of one per cent of the variation in wellbeing. To make the size concrete, they compared it with associations in the same datasets for other variables, and found it comparable to associations with things like wearing glasses and eating potatoes.

Be careful what you take from this. It does not show that nothing harms anyone: an average this small is consistent with real harm to some young people from some uses, and with benefits to others, and the field has since moved towards asking which uses, at what times, for whom. It does show that the total-hours measure, which is what almost every alarming headline rests on, does not support the size of the claims made from it. That is the same failure as the 7-38-55 rule from Module 2: a real but tiny finding scaled up into a slogan.

Disinhibition, both kinds

John Suler catalogued six factors that loosen behaviour online, and the useful part of his account is that it explains the good version and the bad version with the same mechanism. Anonymity separates online actions from an offline identity. Invisibility removes the sight of the other person's reaction. Asynchronicity means you do not have to sit through the consequence in real time. Solipsistic introjection is the way you supply an imagined voice for the person you are reading. Dissociative imagination treats the online space as a game with different rules. And status cues are minimised, so authority carries less weight.

Run those through a hostile person and you get abuse. Run them through a frightened one and you get the disclosure they could never have made face to face, in a support forum at two in the morning. Kiesler's flaming and the striking intimacy of some online support communities are the same set of mechanisms with different inputs.

Common misconceptions

"Text-based communication cannot support real intimacy." Social information processing theory showed the early evidence came from short zero-history laboratory tasks. Given time and enough messages, relational development in mediated channels reaches comparable levels; what differs is the rate.

"If you idealise someone you met online, they must have deceived you." The hyperpersonal model explains idealisation without deception: selective self-presentation, favourable gap-filling, a channel that allows editing, and a feedback loop where each person becomes more like the impression they are being given.

"Online dating profiles are full of lies." Most contain at least one inaccuracy about height, weight or age, and the magnitudes are small. Anticipating a face-to-face meeting acts as a warrant and constrains the size of the lie.

"Video calls are tiring because of the technology's lag." Bailenson's account locates the fatigue in close-up gaze, the effort of producing and reading nonverbal cues, hours of self-view, and physical confinement, each of which has a specific remedy.

"Screen time is destroying adolescent wellbeing." Across the full space of defensible analyses on more than three hundred thousand adolescents, the association is negative and tiny, comparable in magnitude to eating potatoes. Real harms to particular young people from particular uses are compatible with that; the total-hours claim is not supported.

What to remember

Kiesler, Siegel and McGuire's 1984 study framed mediated communication as a subtraction, and produced findings about equal participation and uninhibited behaviour that still hold up. Social presence and media richness theories formalised the subtraction view, and remain good advice about choosing a channel for one difficult conversation. Walther's social information processing theory corrected the error at the level of relationships: the cues are different, not absent, and the ceiling is the same while the rate is slower. The hyperpersonal model explains how mediated relationships can exceed face-to-face ones through selective presentation, idealisation, an editable channel and a confirming feedback loop, and warranting explains why information the target cannot control counts for more, which is also why online dating lies are small. Video fatigue has four identified mechanisms and four corresponding fixes. And the largest test of the screen time claim found an association so small that the honest question is not whether screens are harmful but which uses, for whom, and when.

Sources

1. Pew Research Center. (n.d.). Mobile fact sheet. pewresearch.org
2. Pew Research Center. (n.d.). Social media fact sheet. pewresearch.org
3. University of Minnesota Libraries Publishing. (2016). Communication and technology. In Communication in the real world. open.lib.umn.edu
4. Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer-mediated communication. American Psychologist, 39(10), 1123-1134.
5. Daft, R. L., & Lengel, R. H. (1986). Organizational information requirements, media richness and structural design. Management Science, 32(5), 554-571.
6. Walther, J. B. (1992). Interpersonal effects in computer-mediated interaction: A relational perspective. Communication Research, 19(1), 52-90.
7. Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1), 3-43.
8. Toma, C. L., Hancock, J. T., & Ellison, N. B. (2008). Separating fact from fiction: An examination of deceptive self-presentation in online dating profiles. Personality and Social Psychology Bulletin, 34(8), 1023-1036.
9. Bailenson, J. N. (2021). Nonverbal overload: A theoretical argument for the causes of Zoom fatigue. Technology, Mind, and Behavior, 2(1).
10. Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3, 173-182.