Are Moods & Emotions ‘Contagious’ on Facebook?

Some news outlets are blaring that a new study recently published demonstrates moods are “contagious” on online social networks, like Facebook. Parroting the tone and talking points from the news release on the study, it appears nobody bothered to read the actual study before doing their reporting on it.

However, it doesn’t take an empirical study to understand that our moods impact one another. If you’re depressed and you live with your family, your depressed mood is going to affect your family. If you’re manic and hang out with your friends, chances are some of that manic energy is going to rub off on them.

We would expect that same thing to occur online, wouldn’t we?

The study was conducted on data collected from people living in the top 100 most populous cities over 3 years on Facebook from January 2009 to March 2012. It’s unclear whose data was collected, as the researchers do not say (which is an odd thing to leave out, since one would assume whose data was being collected is important to note).

However, since two of the authors were working at Facebook at the time, we can assume they collected all U.S. users of data of people living in the most populous cities. You did know you agreed to allow research on everything you upload to Facebook, right?

But the primary problem is the use of the analysis tool that has become a favorite amongst researchers analyzing online text — the LIWC. The Linguistic Inquiry Word Count (LIWC) is a rudimentary, somewhat crude automated analysis tool for language. Those aren’t my words — those are the words of one of the LIWC’s creators (Tausczik & Pennebaker, 2010):

Despite the appeal of computerized language measures, they are still quite crude. Programs
such as LIWC ignore context, irony, sarcasm, and idioms. (Emphasis added.)

Ummm… those are pretty big things to leave out of an analysis of the nuances and complexities of social, informal language, don’t you think? In fact, the LIWC’s accuracy rate has been called into question by other researchers in at least one analysis of a set of tweets from Twitter (Gonzalez-Ibanez et al, 2011) ((“We found that automatic classification can be as good as human classification; however, the accuracy is still low. Our results demonstrate the difficulty of sarcasm classification for both humans and machine learning methods.”))

But let’s ignore the fact that the current researchers are using a crude analysis tool generally unsuited for the purpose they’re using it for. ((The researchers justify its use by saying it’s “widely used” for this sort of text analysis. It’s an odd thing to read in a scientific paper, just because something’s popular doesn’t make it the right tool to use.))

Let’s look at a hypothetical example of a Facebook status update interaction to understand why some of the assumptions the researchers made were probably not ideal:

You: I’m having a bad day… just wish this day would end already!

Friend A: Oh wow, sorry to hear that. Some days do indeed suck.

Friend B: Bummer, that sucks.

The LIWC would code this exchange as negative, with two negative responses.

But did the first post actually do anything to impact the mood of the two respondents?

We simply don’t know. The LIWC can’t tell us, because it doesn’t really understand social context. All it understands is the rudimentary of negative and positive words.

Is this an Effect That Actually Matters?

Even if we say the effect the researchers found is a robust one as they claim (because they controlled for one variable out of hundreds — the weather), it doesn’t appear to be a very important one. How big was this effect of a mood “contagion?”

If you post positively on Facebook, amongst all of your hundreds of friends, your post will generate an additional 1.75 positive posts. That’s not almost 2 posts per friend — that’s just 2 posts amongst all of your friends. If all of your friends post a combined total of 50-100 status updates a day (not an unreasonable amount, since the average number of friends a person has on Facebook is 338), that’s probably less than a 4 percent change.

If you post negatively on Facebook, your post will generate a mere 1.29 additional negative posts — again, total, from all your friends. ((It seems many mainstream media outlets are reporting this data incorrectly, saying that a negative post “spreads through” 1.29 percent of one’s friends.))

These effects don’t seem all that big when put into any sort of real-life context. It’s like finding statistical significance in your data, but nothing that would make a clinical (or real-world) difference.

What the researchers may have shown — if you throw out the limitations of the LIWC as a data analysis tool — is that sharing begets sharing on online social networks. If you share you like popcorn, others are going to chime in that they like popcorn too. If you share your cat is the cutest thing since Barnie, well, your cat lover friends will respond in kind.

And if you share a mood state on Facebook, surprise, surprise, others will be ever-so slightly more likely to share theirs as well. Does this make sharing a “contagion?” Not likely.

 

CBS’s news release-based reporting: Emotions spread through Facebook are contagious, study says

The Guardian’s regurgitation of other news stories on the topic: Facebook Transfers Contagious Emotions

 

References

Corviello, L. et al. (2014). Detecting Emotional Contagion in Massive Social Networks. PLOS One.

Gonzalez-Ibanez, R. Muresan, S., & Wacholder, N. (2011). Identifying Sarcasm in Twitter: A Closer Look.
Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, 581-586.

Tausczik YR, Pennebaker JW (2010) The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology 29(1): 24 — 54.