Hacker News new | past | comments | ask | show | jobs | submit login
Experimental evidence of massive-scale emotional contagion thru social networks (2014) (pnas.org)
166 points by pseudolus on July 14, 2020 | hide | past | favorite | 63 comments



Here's a simple mechanism for emotional contagion:

1. Mirroring is the behavior in which one person unconsciously imitates the gesture, speech pattern, or attitude of another. -- https://en.wikipedia.org/wiki/Mirroring

2. The facial feedback hypotheisis is that one's facial expression directly effects their emotional experience -- https://en.wikipedia.org/wiki/Facial_feedback_hypothesis

So you subconsciously imitate expressions and gestures from characters in media, even text, which activates corresponding emotional states. The states persist long enough to propagate to people you interact with, like ripples.

It's just that the faster mirroring, feedback and propagation of bits over atoms are generating waves rather than ripples. Beware of tsunamis.



Thank you for this link! I haven't heard about this particular effect before, but it shows the importance of context very well I think


What's old is new again.

https://en.wikipedia.org/wiki/Extraordinary_Popular_Delusion...

Extraordinary Popular Delusions and the Madness of Crowds is an early study of crowd psychology by Scottish journalist Charles Mackay, first published in 1841 under the title Memoirs of Extraordinary Popular Delusions.[1] The book was published in three volumes: "National Delusions", "Peculiar Follies", and "Philosophical Delusions".[2


> When positive expressions were reduced, people produced fewer positive posts and more negative posts; when negative expressions were reduced, the opposite pattern occurred.

More precisely, this research appears to show how emotions expressed in Facebook posts get mimicked by connections. Should this be taken as literally reflecting people’s underlying emotional states[0]? I’d accept that they correlate to a degree, but not completely, especially not on a platform that mandates real names.

[0] Anecdotally, I could be on balance in a happy state but express annoyance at minor aspects of life in my posts. In fact, the existence of things to complain about is a sign things are going relatively well (when things are looking down on a fundamental level, going on record with that under a real name is risking to alienate IRL connections, and minor nuisances of life seem too insignificant to be worth mentioning).

Conversely, seeing more negative posts would make me feel compelled to post more of the same, but on balance feel better about my own situation after seeing what emotions others have to deal with.


Yes if I'm remembering correctly this study was published by Facebook data science after some psych lab published a study that went semi-viral suggesting people were more unhappy when their network made mostly happy posts (which was explained by jealousy/comparing self to others).

I remember thinking both studies had limitations, the main one here being that the interpretation is a stretch and to me was obviously trying to push back on the original popular narrative.

Point being, I agree it is definitely plausible that seeing positive posts encourages more showing off rather than actual happiness. I would be interested if anyone is aware of any more recent follow-up studies?


I wish we didn't have to rely on tweets that are possibly lies and certainly have bias, but I found this pretty interesting.

It's a travesty the scientific comment can't and won't comment on it. So random tweets are what we have.

"7 out of 9 teenage girls at this school have detransitioned since lockdown and one is considering it." <snip bias>

https://twitter.com/transgendertrd/status/127768841560624742...


I think AI recommendations algorithms are too unregulated. Say a person is having or had for example a depression. Recommendation algorithms for advertising tech wants customer to stay on site and view content to sell advertising. So potentially recommendations algorithms could recommend videos or content that transfers depressed or other emotional states to viewers as long as it generates ad revenue. Depending on what the algorithms senses what others in the same emotional state are viewing. What is the legal liability For transfer these emotions for companies providing these platforms?

On the other side the platform could transfer positive emotions. The issue I see is that we have unregulated emotional and potentially addiction transfers directly to consumers brains. What I mean for cigarettes there is warning here there is nothing but the product is probably as addictive minus health effects.


AI recommendations is something for bored people in my opinion. Sure, you may stumble upon content you would not have considered, but the examples people tend to bring up aren't too convincing.

Some might think that showing depressed people some cute dog pics would lighten them up, but that isn't how it works. On the contrary, the expectations towards themselves to be happy can be counterproductive.

I believe this is the crux of the matter, since social media tends to load people with expectations, depression or not and advertisers try to use this handle to promoted their products.

If a site would try to "transfer positive emotions" I would be seriously concerned. Humans don't even know what content would be best for an individual. How would you train an AI? From resonance of other users? No, that leaves out a lot of context.

But I still don't think AI curators should be regulated, because the quality of regulation would certainly be horrible.


In some places though (see: algorithmic views of the primary content e.e. facebook or twitter, youtube recs), the recommendations can't be disabled.

You cannot avoid the machine entirely, and all it might take is showing a thumbnail or a preview of something to you.

This sort of thing has led me to getting off those services.


Are studies of this sort still being done, perhaps privately at Fb or other social networks?

This study was unethical, and revealed something very scary about the potential of social networking, but I haven't heard much since then. Did they stop all research into social behavior, find ethical ways to do it, or just stop publishing publicly?


>Are studies of this sort still being done, perhaps privately at Fb or other social networks?

According to my facebook feed they said no, they don't do anything unethical. To be serious though this is pretty scary. The power these companies have to manipulate society is not understood by much of society.


This is insanity. How did this study pass an ethics council? They admit to exposing people to negative emotional cues at random?

Their N is 689K. At this point, they are certainly responsible for some number of suicides.

The authors should be ashamed of themselves. Their salary is being paid by manipulating people's emotions on purpose. Something is wrong with that.


Unless I am mistaken, this study reports on Facebook’s own test of News Feed algorithms. Facebook ran the test in 2012, supposedly to determine whether posts with emotional content are more engaging. This study (published in 2014, by the way—could someone edit the title?) appears to be examining the data from that test.


FWIW there was some backlash at the time: https://www.wsj.com/articles/furor-erupts-over-facebook-expe...


There is a link to the answer right on the page under "Editorial Expression of Concern". This answers how it bypassed an ethics council: https://www.pnas.org/content/111/29/10779.1


>They admit to exposing people to negative emotional cues at random?

I'm not sure the "negative" part matters. Manipulating someone's information in any way is irresponsible. A message intended to be uplifting can be harmful, say if someone is in a manic state or is unable to relate to the positive message. In politics, I think we all know the dangers of burying bad news.

Making people sad is more obviously evil than making people happy, but Facebook has no business trying to do either.

>At this point, they are certainly responsible for some number of suicides.

I doubt that. The effect size was small (~2% change in frequency of emotional words), and it seems dubious to me that the frequency of positive and negative words actually reflects someone's emotional state.

But failure isn't an excuse. Trying to hurt people isn't OK just because you aren't very good at it.


> Manipulating someone's information in any way is irresponsible.

I agree. But what about advertising, filter bubbles, curated feeds, aggregators? Is it OK if done algorithmically to make more money?



The issue is that they published it in a journal? Otherwise an ethics board isnt required for running an A/B test


Seems like there are several concerns: the journal, a question of whether the FB usage agreement permitted such a report from being published, and generally whether it is ethical (separate from the review board question). Some more discussion:

http://ulan.mede.uic.edu/alansz/2014/07/an-open-letter-to-pn...


The ethical question is whether to do A/B tests privately vs to do A/B tests and publish scientific papers about them.

I would be interested to understand the ethical basis for removing the ability of companies to run A/B tests.


And further, whether to do A/B tests or not, and whether the subject consents, since the test might have negative effect


How should a user consent to a Google A/B test other than agreeing to use the software that has A/B tests in it? They run 10s of thousands of tests every day


I think that the vast majority of A/B tests are unethical.


Sorry but you can't hold someone else responsible for suicide unless they were specifically taunting and influencing someone to commit that act.


But they were specifically taunting them and influencing them. They put up posts with negative emotions. If just one post caused someone to commit suicide in the negative study group, and facebook showed it to them one time more than normal, they are culpable. Given the large N value, I find it difficult to believe this didn't happen.


Negative emotions is already a significant chunk of social media in general and news media as well. So...very slippery slope definition for culpability you've got there.


That's why social media feeds must use a dumb algorithm.


"Normal" is already determined by their AB testing which is already massive scale human experimentation.


I wonder at what point people will be able to say "It's not my fault, the Algorithm made me do it" and have that accepted as a valid excuse by a majority of society.


Isn't that just the generalisation of 'it is only business' which the majority of society already accepts as a valid excuse for some pretty unethical behaviour.


There's a much wider discussion (educational process?) we need to have as a society. The idea that your or my opinions are our own, that we came to them independently or that they're based on facts and logic is basically disproven at this point.

But people insist on pretending they are and mistakenly believe that facts change minds.

This is leading to disastrous public policy and significant damage to people's mental health. These are the issues (directly or indirectly) of our time. Solving them is the big challenge on the 2000s...


This blindness afflicts people who self-identify as "rational" the most, in my opinion. They see themselves as strong, independent thinkers who are immune from efforts to manipulate. No one is immune, the same way no human is independent in the way that individualists would hope they are.




Needs a (2014)


The article appeared in a footnote to another post that offers something of a historical overview of the use of predictive data [0].

[0] https://news.ycombinator.com/item?id=23836500


Recommendation algorithms have dramatic effects on a user's emotions, beliefs, vocabulary, and choices.

This applies not only to social networks but also to search engines, content recommendation, and other applications built on metrics like "relevance" and "engagement".


I think calling it contagion misrepresents that this quality of emotional transfer is the original internet


> "It's just that the faster mirroring, feedback and propagation of digits over atoms are generating waves rather than ripples. Beware of tsunamis."

that's how the hysteria concerning corona has crashed all over our economy and social infrastructure.

as in, lockdowns won't stop the virus and they're immaterially better than distancing (with masks for when you can't), but the economic and social costs (including the politically motivated backlash) are massive, and yet we keep imposing that enormous cost on ourselves to make us feel like we're doing something about it, no matter how trivial the effect or how great the costs, to assuage our irrational fears.


We detached this subthread from https://news.ycombinator.com/item?id=23839314.


ICU filled up fast in Houston after the lockdown ended. How does that fit with your idea here?


correlation not causation. the end of lockdown unleashed pent-up frustration and disregard for even simple yet effective distancing precautions--a case of throwing the baby out with the bathwater behavior as backlash.


Arguing this at the level of "What should individuals do?" makes no sense, because we have little-to-no data on how effective masks or social distancing are at that level. We're really at the infancy stages of understanding this virus--keep in mind up until a few weeks ago we thought spread from high traffic surfaces was a major transmission vector.

At the level of "What should government recommend?"--that's a fundamentally different question from "What should individuals do?" because governments have secondary effects such as people throwing the baby out with the bathwater. Policy decisions have to take into account this effect, so saying that this effect is the real cause of a sudden spike in cases after lockdown ended, is irrelevant. At a policy level, lockdown is more effective than social distancing and masks, and that's because when you lift the lockdown, people don't social distance or wear masks.

I live in a town where businesses that attract tourism have reopened. People are wearing masks as chin straps and congregating, and that wasn't happening during lockdown.

And to be clear, there are other factors: a country which creates numerous hurdles to access for even the miniscule aid it provides to the poor cannot be surprised when the poor don't want to give up what little income they have to comply with a lockdown. But we have to admit, lockdowns work. That's a fact. The fact that our economy is so fragile that it can't handle the only effective countermeasure to this virus is a separate fact.


no, the point is that lockdowns, being basically a superset of distancing, don't add enough protection/reduction on top of distancing for it's enormous cost.


I understood that as your point the first time.

And I'm saying, if you don't lock down, distancing doesn't happen.

It's more complex than just that, but that's the gist.


> And I'm saying, if you don't lock down, distancing doesn't happen.

First - our lockdowns were mostly unenforced (except against businesses).

Bars were closed but house parties raged.

Second - Sweden didn’t do lockdowns and yet people did practice social distancing.


First: I don't think there's much data to make an argument here, but I think we can agree that house parties alone don't create as much social contact as house parties AND restaurants, salons, gyms, nonessential retailers, hotels, etc.

Second: If your goal is to demonstrate that lockdowns aren't appreciably better than social distancing, Sweden is a great example of why you're wrong.

"As time has gone on the numbers don’t seem to have borne out this strategy, as Sweden is seeing more deaths from the virus than it might have otherwise."[1]

"The death rate in Sweden has now risen significantly higher than many other countries in Europe, reaching more than 22 per 100,000 people, according to figures from Johns Hopkins University, controlled for population. By contrast, Denmark has recorded just over seven deaths per 100,000 people, and both Norway and Finland less than four."[2]

Sweden is, in fact, an example where relying on social distancing and masks rather than lockdowns has resulted in more than three times the number of deaths per-capita that were experienced by countries with similar demographics.

[1] https://www.forbes.com/sites/joshuacohen/2020/05/23/swedens-...

[2] https://www.cnn.com/2020/04/28/europe/sweden-coronavirus-loc...


> First: I think we can agree that house parties alone don't create as much social contact as house parties AND restaurants, salons, gyms, nonessential retailers, hotels, etc.

In places with much lighter lockdowns (or no lockdowns at all) we still had significant social distancing. It turns out most people don’t want to catch a deadly disease.

So we are comparing strict lockdowns compared with sensible social distancing - not with doing nothing.

There are strong arguments to be made, and we will have to see how they play out over the next 12-18 months, that strict lockdowns actually resulted in more deaths because the fear they induced stopped people from doing regular health checkups etc.

> The death rate in Sweden has now risen significantly higher than many other countries in Europe, reaching more than 22 per 100,000 people

There is endless misinformation being put out over COVID.

Comparing the other Nordic countries to Sweden is misleading because it was on a different trajectory before any lockdowns were implemented.

Note if you compare Sweden to the US it’s around half of the US rate per 100,000 despite lack of lockdowns.

Now the nice thing about Sweden is they have open data and accurate reporting of data - and so you can dive in and see that their total deaths haven’t changed significantly compared to prior years.

They have had ~100 excess deaths/week - and if you remove the older population (>65) then it’s basically 0 excess deaths/week.

It will be interesting next year to see if (as typically happens with the flu) we see significantly fewer total deaths - which would indicate for those 100/week excess deaths this year the total life lost per death was 1 year. But that’s just my speculation for now.

Edit: A clear trend is emerging that a countries deaths come down to two factors:

1) Ability to properly protect the old

See the failures in Italy and NYC as examples.

2) The overall health of your young population

See for example Singapore where the majority of infections were in young healthy men - and their corresponding tiny 26 deaths for ~45000 positive cases.


> So we are comparing strict lockdowns compared with sensible social distancing - not with doing nothing.

No, we are not comparing lockdowns with sensible social distancing. We aren't comparing lockdowns with doing nothing, either. We're comparing lockdowns with what people actually do when you don't have a lockdown, which is in some cases sensible social distancing, but in other cases traveling 150 miles to wear your mask as a decorative chin strap and cough in a waiter's face.

Yes, some people sensibly social distance. But given the opportunity, many people don't.

> There are strong arguments to be made, and we will have to see how they play out over the next 12-18 months, that strict lockdowns actually resulted in more deaths because the fear they induced stopped people from doing regular health checkups etc.

Didn't you say, just one paragraph back, that "It turns out most people don't want to catch a deadly disease."? Wouldn't this cause them to not do regular health checkups, etc. as well?

I suppose it's possible that Covid itself causes just enough fear to cause people to social distance but not skip preventive care, and lockdowns just happen to tip people over into too much fear so they skip preventive care. But I don't think you have the evidence to claim that's what's happening.

That's a long way of saying, [citation needed].

> Comparing the other Nordic countries to Sweden is misleading because it was on a different trajectory before any lockdowns were implemented.

[citation needed]

> Note if you compare Sweden to the US it’s around half of the US rate per 100,000 despite lack of lockdowns.

Comparing Sweden to the US is misleading because the US doesn't have socialized medicine and is much more densely populated in the major cities where the disease spread. You can [citation needed] me on either one of those claims if you want and I'll provide links, but I think they're well known enough to not need citation.

> They have had ~100 excess deaths/week - and if you remove the older population (>65) then it’s basically 0 excess deaths/week.

Well, there it is. The age of the people dying is only relevant if you don't care about the elderly. You should be ashamed of yourself. Grow a conscience.

> 1) Ability to properly protect the old

Lockdowns protect the old.


> We're comparing lockdowns with what people actually do when you don't have a lockdown

As long as we are comparing with what people actually do when there is a lockdown.

This is what makes the data analysis so difficult - perhaps in some areas people actually followed lockdown rules - where I live basically nobody did.

Yet on paper we had similar lockdowns.

> Wouldn't this cause them to not do regular health checkups, etc. as well?

It doesn’t seem so - but it’s hard to know for sure because you are comparing different cultures and demographics.

Swedes kept going to routine checkups whereas Californians didn’t - but was that caused by the lockdowns or differences in culture?

Like I said in 12-18 months the data will be a lot clearer.

> [citation needed]

Follow Michael Levitt on Twitter - https://mobile.twitter.com/MLevitt_NP2013

> Comparing Sweden to the US is misleading

Yes, and so is comparing Sweden and Denmark - but you were happy to do that.

Latest excess deaths show Sweden is doing better than the European average - https://mobile.twitter.com/MLevitt_NP2013/status/12838142549...

> Well, there it is. The age of the people dying is only relevant if you don't care about the elderly. You should be ashamed of yourself. Grow a conscience.

Nonsense - a 5 year old dying is completely different from an 80 year old dying.

That doesn’t mean I don’t care about the elderly.

> Lockdowns protect the old.

That’s not clear.

NYC for example has had a ruinous number of deaths in nursing homes.

Maybe you meant “Lockdowns are intended to protect the old”?


> As long as we are comparing with what people actually do when there is a lockdown.

We are.

> This is what makes the data analysis so difficult - perhaps in some areas people actually followed lockdown rules - where I live basically nobody did.

Really? People in your area showed up to closed businesses and did what, congregated outside pointlessly? I'm gonna have to say, I don't believe that.

> Swedes kept going to routine checkups whereas Californians didn’t - but was that caused by the lockdowns or differences in culture?

[citation needed]

> Follow Michael Levitt on Twitter - https://mobile.twitter.com/MLevitt_NP2013

[citation needed] still, because I have no idea where you're seeing anything there which you're claiming, and more to the point, a biophysicist is not the same as an epidemiologist, and even more to the point, even an epidemiologist needs evidence to back up their point.

In short, link a study, or you are just making crap up.

> > Comparing Sweden to the US is misleading

> Yes, and so is comparing Sweden and Denmark - but you were happy to do that.

No, it's not.

The difference is that I gave reasons why the US and Sweden are different, whereas you don't seem to feel that a lack of evidence should prevent you from saying things.

Again, provide evidence for your claims, or you're just making crap up.

> NYC for example has had a ruinous number of deaths in nursing homes.

NYC didn't lock down until the disease had already been widespread.


I guess Americans aren't as responsible as Swedes.


>correlation not causation.

>the end of lockdown unleashed pent-up frustration and disregard*

The debate totally set aside, realize that you proposed a causal relationship one sentence after saying "correlation not causation."


yes, but "undistancing->more cases" is a simpler if subtler explanation than "unlockdown->more cases", and thereby invalidating the implied causation of the question via occam's razor.


> as in, lockdowns won't stop the virus

I wish we could be done with this absolutely false misinformation. Case in point, New Zealand is COVID free.


Not completely, we still have a few active cases as far as I'm aware.

We also basically shut down our borders, which I don't think is that feasible for a lot of other countries.


There have been weeks-long stretches within the past month where New Zealand has had no active cases.

Currently, there are 2, apparently new, according to this:

https://www.health.govt.nz/our-work/diseases-and-conditions/...

This is about as close to "eliminated" as you can get; the new cases are all from incoming international travel.

This directly refutes the OP's statement that lockdowns "won't stop the virus."


Lucky for them to be a tiny island nation with totally controlled borders. Long Island alone is almost twice as big.

Early on when New York was surging and Trump suggested limiting inter-state travel Cuomo suggested that would be akin to a federal declaration of war. [1]

It can be true that strategies that might work for New Zealand could never possibly work for the continental US.

Hawaii has a similar situation, got new case count down to zero for a few days, but are dealing with a few cases per day recently.

[1] - https://www.theguardian.com/us-news/2020/mar/28/donald-trump...


The good thing about the scientific method is you can decide what you are doing isn't working and try something else.


> Case in point, New Zealand is COVID free

Unlike many countries, they can easily close their borders, many people live in isolated communities and they aren't a major transport hub.


> Case in point, New Zealand is COVID free.

Unlike NZ, the US can’t control its borders - as decades of uncontrolled illegal immigration has shown us.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: