The seventy-sixth article was about dividing money. This one is about something an owner of a small firm controls more directly and thinks about less: what walking into a room in a bad temper does to the people already in it.
Key Takeaway
The founding study reports that positive contagion "led to improved cooperation, decreased conflict, and increased perceptions of task performance."[1] The word perceptions is the author's own. On our own arithmetic, the massive-scale follow-up's sample of 689,003 makes an effect of one hundredth of a standard deviation statistically detectable, which moves the median person four tenths of a percentile.
The Verdict, Stated First
Five claims, in descending order of confidence.
One. Emotional contagion in work groups is established. Two managerial simulations with converging measures, published in a leading organisational journal, and the direction of the effects is not in dispute.
Two. What was measured is narrower than what is claimed for it. The abstract reports improved cooperation, decreased conflict, and increased perceptions of task performance, all of them ratings.
Three. The famous large-scale study is famous for the wrong reason. It attracted an ethical controversy severe enough to draw a formal expression of concern from the journal that published it.
Four. Its sample size makes statistical significance nearly automatic. On our own arithmetic, 689,003 participants make a hundredth of a standard deviation detectable, and we did not obtain the effect size the authors reported.
Five. The commercial application is real and modest. The mechanism is established, the magnitude in a workplace is not, and anyone selling you a mood programme on the strength of this literature is going further than it goes.
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A for the 2002 findings as stated, from an abstract obtained verbatim from an economics database, with the published paper's front matter confirmed from a copy hosted by a university laboratory.
Grade A for the 2014 study's design and the journal's response, both obtained verbatim from the publisher.
Grade F, in effect, for every magnitude in this article. We obtained no effect size from either study, which for a topic this widely quoted is the largest gap we have reported.
Grade A for our own arithmetic, which is standard.
Our position: the direction is well established and the size is unknown to us, and we would rather say that than borrow a number from a secondary source.
A Note On Method
Everything here is verified to August 2026.
We obtained the 2002 study's abstract verbatim from an economics database, in the form attached to its earlier working paper[1], and its published front matter from a copy hosted by a university media laboratory[2]. We did not obtain the paper's results.
We obtained the 2014 study's significance statement and abstract verbatim from the publisher[3], together with the full text of the journal's Editorial Expression of Concern and its separate Correction[4]. We did not obtain the paper's results.
We obtained no effect size for either study, and we do not report one.
We did not obtain the 1993 foundational work or the 2001 paper on negativity, and report both from citation records[5].
All arithmetic is ours, and the interaction counts in one section are invented.
This article discusses research on group emotion. It is not management, human resources or wellbeing advice, and nothing here bears on the lawfulness of any employment practice.
The 2002 Study
The source.
Barsade, S. G. (2002), The Ripple Effect: Emotional Contagion and Its Influence on Group Behavior, Administrative Science Quarterly, 47(4), 644–675, December[2].
Three observations, ours.
The journal is one of the most senior in organisational research, and the paper is published on behalf of a business school rather than a psychology department, which matters for how it should be read: it is a management finding, not a laboratory curiosity.
The title's metaphor has travelled better than its contents. "The ripple effect" is the part everyone remembers, and the qualifications are in the abstract.
And we did not obtain the paper, which for a study this central to the article is a real limitation and the reason every magnitude below is missing.
One further note on what we did obtain, because the sourcing here has an unusual shape. The abstract we quote throughout is attached to the earlier working paper version, retrieved from an economics database, and the published paper's own abstract is not something we obtained separately.
We treat the two as the same claims because working paper and published abstracts normally match closely, and we flag that we are relying on it. Where the two versions demonstrably differ, in title and stated affiliation, we record it in a section below rather than passing over it.
Two Managerial Simulations
The design, from the abstract.
"Emotional contagion, the transfer of emotion between individuals, and its influence on work group dynamics was examined in two managerial simulations using multiple, convergent measures of emotions and group dynamics."[1]
Four observations, ours.
Two simulations rather than one, which is a stronger base than most single papers in this series report.
"Multiple, convergent measures" is the phrase that matters methodologically. The same conclusion reached by different instruments is much harder to explain away than one measure moving.
A managerial simulation is a structured exercise with participants playing roles, which sits between a laboratory task and a real workplace. Closer to work than most psychology experiments and not work.
And the phrase "was examined" conceals the mechanism the study is known for. The standard account is that a trained confederate was placed in each group to introduce a mood, and we did not obtain the method section to confirm the detail.
What It Found
The result, verbatim.
"As predicted, experiencing positive emotional contagion led to improved cooperation, decreased conflict, and increased perceptions of task performance, all as rated by self, other group members, and outside video-coders. The opposite was the case when experiencing negative emotional contagion."[1]
Four observations, ours.
Three outcomes moved together and in the expected direction, which is a coherent result rather than a scattered one.
The three rating sources are the strength of the design. Self, other group members, and outside video-coders means the effect was not simply people reporting their own good mood.
"The opposite was the case" for negative contagion establishes both directions in the same study, which is more than most such findings offer.
And the phrase "as predicted" is worth registering. This was a confirmatory test of stated hypotheses rather than an exploratory result, which is the stronger form.
Two things that phrase does not settle, and this series has learned to check both. Predictions can be registered in advance or written after the fact, and pre-registration was not standard practice in 2001, so "as predicted" here means the hypotheses appear in the paper before the results rather than in a public record before the study.
And a confirmed prediction is only as good as the alternative it beat. The hypothesis that a cheerful group member improves cooperation is not a bold one, and confirming it is weaker evidence than confirming something the field would have bet against.
The Word In The Abstract
The thing we would most want a reader to take from this article. Ours.
Four observations.
The abstract says "increased perceptions of task performance." Not increased task performance. The word is the author's own and it is doing careful work.
Every popular retelling we encountered while researching this drops it. The claim in general circulation is that good moods make teams perform better, and the claim in the abstract is that they make performance be rated higher.
Note also that the outside video-coders do not rescue the distinction. An independent rater watching a video is producing a perception too, and a more disinterested one, but still a judgment rather than a measurement of output.
And the other two outcomes are not subject to the same caveat in the same way. Cooperation and conflict are behaviours a coder can observe, so the finding on those is on firmer ground than the one on performance.
Why That Word Matters
Because there is a specific reason to expect ratings to move without output moving. Ours.
Four observations.
A pleasant group is pleasant to be in and pleasant to watch, and both raters and participants may transfer that pleasantness to their judgment of how well the work went.
That is the halo effect this publication has covered separately, and it predicts exactly the observed pattern. A mood manipulation that changes perceived performance without changing performance is what a halo would look like.
We are not claiming that is what happened. The study may well have measured output too, and we did not obtain the results, so we cannot tell you.
What we can tell you is that the abstract, which is the author's own summary of her own findings, chose the word perceptions, and researchers do not add qualifications to abstracts for fun.
The Second Dimension
A finding in the same abstract that gets lost entirely.
"The studies tested hypotheses on differential contagion effects due to the degree of pleasantness of the emotion, and the energy with which this pleasantness was conveyed."[1]
Four observations, ours.
The design has two dimensions, not one. How pleasant the emotion is, and how energetically it is conveyed, and those are separable.
That matters practically because it distinguishes four states rather than two. Cheerful and energetic, cheerful and calm, negative and energetic, negative and flat, and there is no reason to assume they behave alike.
The popular version collapses this to good mood versus bad mood, which discards half the design. An owner who is quietly content and one who is loudly enthusiastic are in different cells, and this study was built to tell them apart.
And we did not obtain the results on that dimension, which is the second most frustrating gap in this article. The abstract says the hypotheses were tested and does not say what came back.
The omission is itself mildly informative, and we would not read much into it. Abstracts report what the authors considered the headline, and a tested dimension that does not appear in the results summary is more often a null or a complication than a clean finding.
The 2014 Study
The study most people have heard of, for reasons unrelated to its findings.
Kramer, A. D. I., Guillory, J. E., and 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, June 17, DOI 10.1073/pnas.1320040111[3].
Its significance statement: "We show, via a massive (N = 689,003) experiment on Facebook, that emotional states can be transferred to others via emotional contagion, leading people to experience the same emotions without their awareness. We provide experimental evidence that emotional contagion occurs without direct interaction between people (exposure to a friend expressing an emotion is sufficient), and in the complete absence of nonverbal cues."[3]
Three observations, ours.
The scientific contribution is specific and genuine: contagion without direct interaction and without nonverbal cues, which the 2002 design could not test.
The first author's affiliation is the company whose platform was manipulated, which is stated in the paper and is relevant to how the study was possible at all.
And the paper carries a note that it was edited by a named psychologist at Princeton and approved in March 2014, having been received the previous October, which is a normal review record and worth stating because of what happened next.
One more thing the abstract records, which is worth keeping because it shows the authors were careful about their own field. They note that data from a large real-world social network collected over twenty years suggests longer-lasting moods can be transferred through networks, "although the results are controversial"[3].
That is the authors flagging a disputed prior result in their own opening rather than leaning on it, which is the behaviour this series usually finds absent.
Massive Scale
What the design actually was, and why it could only be done by one kind of organisation. Ours.
Four observations.
The study altered what 689,003 people saw in their feeds, reducing the proportion of emotionally positive or negative posts, and then measured the emotional content of what those people subsequently wrote.
That design has an advantage no laboratory can match. It is a randomised experiment on real people going about their ordinary lives, which is the gold standard that field research almost never achieves.
It has a corresponding property that is inseparable from the advantage. The experiment was only possible because the experimenter controlled the product, and the participants were unaware throughout, which the significance statement itself states as a finding.
And the scale is the reason the study is famous and the reason it is misread. A very large sample buys precision, not importance, and the arithmetic below sets out what that precision amounts to here.
The Expression Of Concern
What the journal did, quoted at length because it is unusual.
"PNAS is publishing an Editorial Expression of Concern regarding the following article... This paper represents an important and emerging area of social science research that needs to be approached with sensitivity and with vigilance regarding personal privacy issues. Questions have been raised about the principles of informed consent and opportunity to opt out in connection with the research in this paper."[4]
The journal then quotes the authors' own position: "[The work] was consistent with Facebook's Data Use Policy, to which all users agree prior to creating an account on Facebook, constituting informed consent for this research."[4]
And states its own: "It is nevertheless a matter of concern that the collection of the data by Facebook may have involved practices that were not fully consistent with the principles of obtaining informed consent and allowing participants to opt out."[4] It is signed by the editor-in-chief.
What A Journal Saying That Means
Reading it, since the form is unfamiliar to most readers. Ours.
Four observations.
An expression of concern is not a retraction. The paper stands, the findings are not disputed by the journal, and the notice sits alongside it permanently.
The specific objection is to consent, and the authors' defence is that agreeing to a data use policy on signing up constitutes it. The journal declines to accept that and says so in print.
The phrase "opportunity to opt out" is the practical heart of it. Nobody in the study knew they were in it, and that is what a mood manipulation on a live product requires.
And the journal published a separate Correction the same day, on an unrelated matter of author affiliation[4], so this paper carries two formal notices, which is rare.
What A Sample Of 689,003 Can Detect
The question we would ask of any very large study. Our own arithmetic, standard power analysis.
The smallest effect detectable at 80 percent power and a five percent level, given that sample:
Split across 2 conditions, roughly 344,502 each: a Cohen's d of 0.0068.
Split across 4 conditions, roughly 172,251 each: a d of 0.0095.
Four observations.
So an effect of about one hundredth of a standard deviation would be reported as statistically significant at this sample size.
That is not a criticism of the study. Establishing that an effect exists at all was the paper's stated contribution, and for that purpose a large sample is exactly right.
It is a warning about how the result travels. Statistical significance at this N carries almost no information about magnitude, and the headline that emotions spread through networks does not distinguish an effect that matters from one that does not.
And we must be explicit about what we do not have. We did not obtain the effect size the authors reported, so we cannot tell you whether their result was near this floor or well above it. The arithmetic above establishes what the design could detect, not what it found.
What A Hundredth Of A Standard Deviation Is
Putting the floor in terms anyone can picture. Ours.
Taking a person at the median and applying an effect of the stated size, they move to the:
At d equals 0.01: the 50.4th percentile. At 0.05: 52.0th. At 0.10: 54.0th. At 0.20: 57.9th. At 0.50: 69.1st. At 0.80: 78.8th.
Four observations.
A detectable effect at that sample size moves the median person by four tenths of one percentile, which no individual would notice in themselves or anyone else.
For comparison, this series' own median effect size across its first forty-nine articles was 0.41, which on this table is worth about sixteen percentiles.
The gap between the two is a factor of forty. An effect can be real, replicable, statistically overwhelming and practically irrelevant, and all four at once.
And this is why we decline to quote the study's effect size from a secondary source. In a case where magnitude is the whole question, a number obtained at second hand is not good enough, and we would rather report the gap.
We should be fair to the study about what it was for, since the point above can read as a criticism of it and is not. Its stated contribution was that contagion occurs without direct interaction and without nonverbal cues, and for establishing that a mechanism operates at all, a tiny effect is perfectly adequate evidence.
The problem enters entirely at the point of transfer. A finding sized to answer "does this happen" gets quoted to answer "does this matter", and those need different evidence. That pattern is now the most frequent single failure this series has recorded across seventy-seven articles.
The General Problem With Large Samples
The lesson beyond this literature. Ours.
Four observations.
A very large sample makes significance nearly automatic and magnitude the only remaining question. The test answers "is it exactly zero", which is rarely what anyone wanted to know.
The commercial version arrives constantly. A platform reporting a statistically significant improvement from millions of observations has told you almost nothing, and the number to ask for is the size.
This connects directly to the seventy-fourth article, from the opposite end. Small business decisions have too few observations to detect anything; large platforms have so many that everything is detectable, and neither situation lets significance stand in for importance.
And the practical instrument is one question. Ask what the effect would do to a typical case, not whether it was significant, which converts a statistical claim into one a person can weigh.
Two ways that question gets deflected, both worth recognising. The first is a percentage change with no base: a forty percent improvement in something that was almost never happening is still almost never happening.
The second is the sample size offered as the reassurance. "Across two million users" is a statement about precision, not about size, and a very large N is a reason to demand the effect size rather than to accept its absence.
Two Studies Doing Different Jobs
Setting them side by side, because they are usually cited together and they answer different questions. Ours.
Four observations.
The 2002 study asks whether one person's mood changes how a group works. Small samples, rich measurement, a workplace-shaped setting, and outcomes that matter to a manager.
The 2014 study asks whether emotion transfers at all without direct contact. Enormous samples, one thin measure, no workplace, and an outcome that matters to a theorist.
Neither substitutes for the other and citing them jointly obscures that. The first has the relevant outcomes and cannot establish generality; the second has the generality and no relevant outcome, which is a familiar division of labour and rarely stated.
And it explains why this article can be confident about direction and silent about size. The study with workplace outcomes is small and old; the study with a precise estimate measured word usage in social media posts, and neither supplies a figure an owner could act on.
What Actually Survives
Our reading, stated directly.
Five statements.
Emotional contagion in work groups is established in both directions, on two managerial simulations with converging measures published in a leading organisational journal.
What moved were cooperation, conflict, and perceptions of task performance, and the third of those is a rating rather than an output measure, by the author's own wording.
Contagion works without direct interaction or nonverbal cues, which is the genuine contribution of the 2014 study.
That study carries a formal expression of concern from its journal over informed consent and the opportunity to opt out.
And we obtained no effect size for either study, so this article can tell you the direction of an effect and not its size, which for a practical decision is the more important half.
Exposure In A Small Firm
Why a small effect might still matter here, and might not. Our own arithmetic, with invented interaction counts.
A firm with an owner and four staff, over a 48-week year:
At 5 interactions per person per week: 960 exposures a year. At 10: 1,920. At 20: 3,840. With nine staff at 10: 4,320.
Three observations.
The volume is substantial, and it is the reason the question is worth asking at all. A small firm delivers thousands of these a year, from one person, to the same few people.
That is a structurally different situation from the 2014 study, where hundreds of thousands of people each received a small number of exposures from many sources.
And a large firm is different again. An owner of two hundred people is a rumour to most of them, and whatever this effect is, the delivery mechanism does not exist at that scale.
Which produces an inversion worth noticing, since the advice usually runs the other way. Leadership writing about setting the emotional tone is aimed at large organisations and applies best to tiny ones, because the mechanism it describes is face-to-face contact and only a small firm has enough of it per person.
And the corollary is a caution for a growing business. An owner whose presence genuinely set the tone at five people cannot do it at fifty by the same means, and continuing to try substitutes symbolism for a mechanism that has quietly stopped working.
Not An Argument That It Accumulates
The section this article needs, because the previous one invites an inference we cannot support. Ours.
Four observations.
Nothing above shows that these effects add up. Mood is not a running total, and the obvious null hypothesis is that each exposure has a transient effect which resets.
We found no study on the cumulative effect of repeated exposure to the same person's mood over a year, and would not assert one from the studies we have.
So the exposure arithmetic establishes something weaker and still worth knowing. It shows the conditions under which a small per-instance effect could matter, not that it does.
And the honest reading cuts both ways, which is why we would not use it as an argument in either direction. High exposure means a small effect gets many chances to act, and it equally means a small effect gets diluted across a thousand ordinary interactions, and we cannot tell you which.
Bad Is Stronger Than Good
An asymmetry we can name and not evaluate.
Reference lists identify Baumeister, R. F., Bratslavsky, E., Finkenauer, C., and Vohs, K. D. (2001), Bad is stronger than good, Review of General Psychology, 5(4), 323–370[5].
We did not obtain it and report the title and citation, which state a thesis.
Four observations, ours.
If the general asymmetry holds here, the two directions in the 2002 study are not equal and opposite, and negative contagion would be the larger effect.
The 2002 abstract does not say. It reports that "the opposite was the case" for negative contagion, which asserts direction and not magnitude.
That is the single most practically important open question in this article, and it decides what an owner should do. If the asymmetry holds, avoiding bad days matters more than manufacturing good ones, and the effort goes to a different place.
And we did not obtain the evidence either way, so we flag it as the question rather than answering it.
Two reasons we think it is likely to hold, offered as reasoning rather than evidence. The 2002 abstract's own summary of negative contagion is "the opposite was the case", which is symmetric language, and researchers who had found an asymmetry would usually say so.
Against that, a practitioner organisation's article describes the negative confederate as having "seemed to disrupt the groups and reduce efficacy" while the positive one was associated with increased cooperation, fewer conflicts and heightened task performance[6], which reads as a stronger characterisation of the negative case. That is a practitioner source, not an academic one, flagged here and at its other use, and it is a difference in wording rather than in data. We would not build on it.
The Owner's Mood
The first application. Ours, untested, and not management advice.
Four points.
The finding says a mood introduced into a group changes cooperation and conflict as observed by outsiders, which is the part on firmest ground and the part that matters operationally.
In a firm of five that transmission runs from one person, repeatedly, to the same few people. Nobody else has the standing to set the tone, which is a structural fact rather than a claim about leadership.
The modest and defensible version of the advice is about timing rather than temperament. Where a bad hour is unavoidable, spending it alone costs nothing and removes the transmission entirely.
And we would resist the stronger version. Nothing here supports requiring cheerfulness of anyone, including yourself, and the section below explains why we think that reading is a mistake.
Two further things about that recommendation are worth making explicit, because it is deliberately narrow. It does not require you to change how you feel, which is mostly not available on demand, and asks only about where you are when you feel it.
And it has a cost we should name rather than pretend away. An owner who withdraws when things are bad is also unavailable when things are bad, which is often exactly when people need a decision, so the rule cannot be applied absolutely.
Hiring For Tone
The second application, and it is where we would apply the most caution. Ours, and nothing here bears on the lawfulness of any employment practice.
Four points.
The finding is sometimes used to justify screening candidates on disposition. The study establishes that a mood introduced to a group spreads, not that some people are hire-worthy and others are not.
The seventy-fourth article's conditions apply with full force here. A small firm hires too rarely, with feedback too delayed, to learn whether its judgments about disposition predict anything.
And the construct is a poor one for selection in any case. A mood is a state and a hiring decision is about a person, and the study manipulated the first rather than measuring the second.
So we would take the finding as being about how a team is run rather than who is in it, which is where the evidence actually points and is also the part an owner controls.
One legitimate use does survive that objection, and it is worth separating from the illegitimate ones. Where a role puts one person in sustained contact with a whole team, being a supervisor, a receptionist, a dispatcher, the transmission mechanism this literature describes applies with more force than to an isolated role.
That is a statement about role design rather than candidate screening: it says where the effect can operate, not who to hire. The remedy it suggests is structural, being how much contact a difficult day gets to have, rather than a judgment about a person's character.
The Manipulation Problem
The uncomfortable part, which the 2014 study makes unavoidable. Ours.
Four observations.
The 2014 study demonstrates that mood can be shifted without the person's awareness, and its own significance statement says so in those words.
That is precisely what drew the expression of concern. An effect that works without awareness cannot be consented to in the moment, which is why consent has to be obtained beforehand or not at all.
The commercial version arrives immediately and is rarely examined. Deliberately managing your visible mood to change how staff feel is an influence attempt they cannot see, and it sits on a spectrum whose far end a journal formally objected to.
We would draw the line at a familiar place and not pretend it is sharp. Not doing harm you would be ashamed to explain is different from engineering a feeling you would not disclose, and the second is what this literature enables.
Two tests we would apply, ours, and neither is original. Would you be willing to tell the person what you were doing? An owner choosing to take a bad hour alone can explain that without embarrassment. A programme designed to induce a mood without disclosure cannot.
And does the person benefit, or only the firm? Most of what this literature supports genuinely helps both, which is why the honest applications are dull ones. The interesting applications are the ones where the two come apart, and those are the ones to decline.
What Not To Do With This
Ours, and this is the practical warning.
Four points.
Do not buy a programme on the strength of this literature. The direction is established and no magnitude reaches this article, so nobody can tell you what an intervention would return.
Do not quote the performance claim without the qualifier. The abstract says perceptions of task performance, and dropping the word converts a rating into an output.
Do not treat mandated positivity as an implication. The study introduced a mood; it did not test requiring people to display one, and the crowding-out literature this publication covered separately gives reason to expect a mandate to behave differently.
And do not use the 2014 study as support for anything commercial without knowing what it carries. A paper with a formal expression of concern about consent is a poor foundation for an argument about influencing people who have not consented.
One thing we would encourage rather than warn against, since this article has been mostly cautionary. The cooperation and conflict findings are the solid part and they are also the useful part. Those are observable behaviours, coded by outsiders, and they are what actually determines whether a small team gets through a difficult week.
An owner who takes only that from this literature has taken the best-supported thing in it. How a group treats each other is affected by the mood introduced into it, and that is worth knowing even without a number attached.
Three Variants In One Citation
The bibliographic entry, since this series keeps a running count.
The 2002 paper's earlier working paper version is titled "The Ripple Effect: Emotional Contagion In Groups" where the published version reads "...and Its Influence on Group Behavior"[1][2].
The working paper record is filed under one university's series and gives the author's affiliation as a different university's business school, while the published paper states a third[1][2].
And the working paper is dated 2001 against the published 2002[1].
Three observations, ours.
None of these is an error. Working papers change titles, authors move institutions, and publication lags, and all three are ordinary.
They are worth recording because a reader searching the working paper title will not find the published paper, and because the abstract we quote throughout is the working paper's, which we flag here and in the method note.
That brings the running count of bibliographic variants across this series to twenty-four.
The count is worth a word of explanation at this point in the series, since it now runs to two dozen. Almost none of them changes a finding. They are wrong page ranges, shifted volumes, altered author orders and titles that changed between draft and print.
We record them because the cumulative picture is the point rather than any individual case. Reference lists in published, peer-reviewed papers contain errors at a rate most readers would not guess, and a reader who cites from one rather than from the source is inheriting them.
What To Do
Keep the qualifier. The founding abstract reports improved cooperation, decreased conflict, and increased perceptions of task performance, and the third is a rating.
Ask for the effect size, not the significance. On our own arithmetic a sample of 689,003 makes a hundredth of a standard deviation detectable, which moves the median person four tenths of a percentile.
Treat the two directions as possibly unequal. A separate literature argues bad is stronger than good, and if it holds here, avoiding bad days matters more than manufacturing good ones.
Spend an unavoidable bad hour alone. It costs nothing and removes the transmission, which is the most defensible thing this literature supports.
Read the finding as being about how a team is run, not who is in it. The study manipulated a mood; it did not establish anything about screening people.
Notice the scale mismatch before applying anything. A firm of five delivers thousands of exposures a year from one person; a firm of two hundred has no such mechanism.
Do not assume the effects accumulate. We found no evidence they do rather than reset, and high exposure dilutes a small effect as readily as it multiplies it.
Check what a study carries before citing it. The most famous paper here has a formal expression of concern from its journal about informed consent.
The Limits Of This Analysis
Several caveats matter, and the central one is unusually large. This article discusses research on group emotion and is not management, human resources or wellbeing advice; nothing here bears on the lawfulness of any employment practice, and the applications are our own reasoning and untested. Everything is verified to August 2026. We obtained no effect size from either study, which means this article establishes the direction of an effect and says nothing about its magnitude; for a practical decision the magnitude is the more important half, and a reader should treat every application below as conditional on a number we do not have. We did not obtain the 2002 paper, only the abstract attached to its earlier working paper version, which carries a different title and a different stated affiliation, plus the published front matter from a copy hosted by a university laboratory; we therefore report nothing about its method beyond what the abstract states, and the widely repeated account of a trained confederate is not something we could confirm. We did not obtain its results on the energy dimension, which the abstract says was tested and does not report, nor any comparison of the sizes of the positive and negative effects. We did not obtain the 2014 paper's results, only its significance statement and abstract, so the arithmetic on detectability establishes what that design could detect and not what it found. We did not obtain the 1993 foundational work or the 2001 paper on negativity and report both from citation records; a title is not a finding. All arithmetic is ours, and the exposure counts are entirely invented; that section shows conditions under which a small effect could matter and does not establish that effects accumulate, which we found no evidence for either way.
Frequently Asked Questions
Is emotional contagion at work real?
What is the word everyone drops?
How big is the effect?
What about the famous Facebook study?
Does a sample of 689,003 prove the effect is important?
Should I be relentlessly positive at work?
Can I hire for this?
References
- Economics database record for the working paper version of the 2002 study, listed as Sigal G. Barsade, 2001, The Ripple Effect: Emotional Contagion In Groups, Yale School of Management Working Papers ysm167, with the author's affiliation given as the University of Pennsylvania, The Wharton School. It reproduces the abstract in full: on emotional contagion, the transfer of emotion between individuals, and its influence on work group dynamics being examined in two managerial simulations using multiple, convergent measures of emotions and group dynamics; on the studies testing hypotheses on differential contagion effects due to the degree of pleasantness of the emotion, and the energy with which this pleasantness was conveyed; on the author, after determining that emotional contagion existed in groups, examining its influence on individual-level attitudes and group processes; on positive emotional contagion leading, as predicted, to improved cooperation, decreased conflict, and increased perceptions of task performance, all as rated by self, other group members, and outside video-coders; on the opposite being the case for negative emotional contagion; and on theoretical implications and practical ramifications being discussed. Note: an economics database record for the working paper version. The abstract quoted throughout this article is the working paper's, not the published paper's; the two carry different titles, different stated affiliations and different years, and we did not obtain the published paper's own abstract or any of its results. ideas.repec.org
- Copy of the published paper hosted by a university media laboratory, reproducing its front matter: Barsade, S. G., The Ripple Effect: Emotional Contagion and Its Influence on Group Behavior, Administrative Science Quarterly, Vol. 47, No. 4, December 2002, pages 644–675, published by Sage Publications on behalf of the Johnson Graduate School of Management, Cornell University, stable archive reference 3094912, with the author's affiliation given as Yale University and acknowledgements to three anonymous reviewers among others. Note: a copy hosted by a university media laboratory. We obtained the front matter, confirming the published citation and affiliation, and did not obtain the paper's method or results. web.media.mit.edu
- 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, June 17, DOI 10.1073/pnas.1320040111. Publisher record reproducing the significance statement and abstract: on the authors showing, via a massive experiment on Facebook with N equal to 689,003, that emotional states can be transferred to others via emotional contagion, leading people to experience the same emotions without their awareness; on emotional contagion being well established in laboratory experiments with people transferring positive and negative emotions to others; on data from a large real-world social network collected over a twenty-year period suggesting that longer-lasting moods such as depression and happiness can be transferred through networks, citing Fowler and Christakis (2008), although the results are controversial; and on the authors providing experimental evidence that emotional contagion occurs without direct interaction between people, exposure to a friend expressing an emotion being sufficient, and in the complete absence of nonverbal cues. The record gives the first author's affiliation as the Core Data Science Team at Facebook and the remaining authors' as Cornell University departments, and notes the paper was edited by a named psychologist at Princeton University and approved in March 2014 having been received in October 2013. Note: the publisher's record. We obtained the significance statement and abstract only; we did not obtain the paper's results and therefore report no effect size from it. pnas.org
- Two formal notices published by the same journal on the same paper. The first, an Editorial Expression of Concern, records that the journal is publishing it regarding the article; that the paper represents an important and emerging area of social science research that needs to be approached with sensitivity and with vigilance regarding personal privacy issues; that questions have been raised about the principles of informed consent and opportunity to opt out in connection with the research; that the authors noted in their paper that the work was consistent with the platform's Data Use Policy, to which all users agree prior to creating an account, constituting informed consent for this research; that the journal considered it appropriate to publish the paper; and that it is nevertheless a matter of concern that the collection of the data may have involved practices not fully consistent with the principles of obtaining informed consent and allowing participants to opt out. It is signed by the journal's editor-in-chief. The second, a Correction published in the same issue, records a change to the author and affiliation lines concerning the middle author's position at the time of the study and her present address. Note: the publisher's own notices, obtained in full. Recorded because a paper carrying two formal notices, one of them an expression of concern about consent, is unusual and material to how it should be cited. pnas.org
- Academic preprint reference lists confirming Barsade, S. G., The ripple effect: Emotional contagion and its influence on group behavior, Administrative Science Quarterly, 47(4), 644–675, 2002; Baumeister, R. F., Bratslavsky, E., Finkenauer, C., and Vohs, K. D., Bad is stronger than good, Review of General Psychology, 5(4), 323–370, 2001; Hatfield, E., and colleagues (1993), on emotional contagion; and Van Kleef, G. A., How emotions regulate social life: The emotions as social information model, Current Directions in Psychological Science, 18(3), 184–188, 2009. Note: reference lists carried on academic preprints; citations only. We obtained none of the works named beyond those cited elsewhere here, and a title is not a finding. arxiv.org
- Article on a practitioner organisation's website about emotional contagion, describing the 2002 study and recording that a negative group member seemed to disrupt the groups and reduce efficacy, while having a positive confederate was associated with increased cooperation, fewer group conflicts and heightened task performance; and carrying a reference list confirming Barsade, S. G. (2002), The ripple effect: Emotional contagion and its influence on group behavior, Administrative Science Quarterly, 47(4), 644–675; Barsade, S. G., and Gibson, D. E. (2007), Why does affect matter in organizations?, Academy of Management Perspectives, 36–59; and Kramer, A. D. I., Guillory, J. E., and Hancock, J. T. (2014), PNAS, 111(24). Note: a practitioner organisation's website, not an academic source, flagged at every use. Recorded only because its characterisation of the negative condition is stronger than the study abstract's own symmetric wording, which is a difference in description rather than in data; we did not obtain the results either account is describing. 6seconds.org
This article discusses research on group emotion and is not management, human resources or wellbeing advice; nothing here bears on the lawfulness of any employment practice. No effect size was obtained from either underlying study, so this article reports direction without magnitude. The 2002 abstract quoted throughout is the working paper's, which carries a different title, affiliation and year from the published version. All arithmetic is the authors' own; the exposure counts are invented, and that section does not establish that effects accumulate.