Can Emotions Be Contagious Online?

Just how exactly and how deeply do emotions transfer across social networks?

With the ever-increasing adoption of social media, one can only wonder whether there are any impacts on users apart from just the ability to communicate and sharing content.

Let’s go down a rabbit-hole of the current research literature surrounding the phenomenon to see if we can get to the bottom of things.

To start things off, one 2014 article published by PNAS makes an interesting claim that in fact social media can become a medium for a massive-scale emotion contagion. To elaborate, the article claims that emotional states can be transferred to others through social media so that people can experience the same emotions without their awareness.

Specifically, the authors provide experimental evidence that the dispersion of emotions can occur without direct interaction between people and without any nonverbal cues (non-verbal communication such as body movements, nuances of the voice, and facial expressions).

One important argument concerns the trend of a growing importance of social media and how its increasing popularity is affecting people in the offline world; the argument made suggests that negative news feed can generate more negative emotions within social media users for up to several days and vice versa.

There is an evidence-based interpretation of how emotions spread based on three key areas of data: how social media platforms mediate the transfer of emotions, the differences in susceptibility to emotional transfers among different individuals, and if positive and negative emotions spread at distinct rates.

Briefly, the research attempts to form predictive models around how user behaviors are impacted and influenced by emotion through statistical modeling.

Primarily, establishing the central ideas of my argument requires a specific basis, there must be research that proves that although indirect, online transmission of information contributes to tangible differences in the emotional states of users.

Understanding the process by how this indirect transmission of emotional information occurs is also very significant. For this basis, consider the work by Coviello et al. (2014), which develops a framework for actually measuring how a “contagion of emotional expression” spreads across social networks and helps to clarify the extent to which emotions are mediated by online, indirect transfers of information.

Coviello et al. (2014) makes one of their key claims with regards to how the perception of rainfall by users affects users in cities with no rain, surmising that:

“For every one person affected directly, rainfall alters the emotional expression of about one to two other people, suggesting that online social networks may magnify the intensity of global emotional synchrony” (1).

The authors of this study make their observations and measurements with regards to one of the most used and commonplace social media networks: Facebook.

They note that there is a wealth of existing research that establishes that emotions are directly transferable or contagious, and that emotions can even further be transferred online or by indirect means:

“Experiments have demonstrated that people can ‘catch’ emotional states they observe in others over time frames ranging from seconds to months…” (Coviello et al., 2014, 1).

Before moving onto to their methods, the authors establish the fact that in many of these observational studies, it is not possible to conclude whether this similarity in emotions is due to social contact termed “contagion” or choosing social contacts with similar emotions, termed “homophily”.

Moreover, the uncertainty inherently associated with the design of many large-scale sociological studies should occupy a salient point in the conclusiveness and confidence in the results published by these authors: this is a limitation.

The incorporation and modeling of a large amount of social media information does not prove causation even if correlation is found, and this point is especially relevant when it relates to something as fluid and ever changing as social interactions themselves.

The authors themselves are keenly aware of such a limitation, and attempt to account for it by changing the context of their study:

“Here, we propose an alternative method for detecting emotional contagion in massive social networks… However, since this is infeasible in our massive-scale setting, we identify a source of variation that directly affects the users’ emotional expression… rainfall” (Coviello et al. 2014, 1).

In other words, the authors of the paper are aware of the fact that the scale of their study requires them to choose variables that are more universal in how they impact one’s mood, such as rainfall.

This is an important consideration on the part of the study’s design because it delineates between an approach that would possess more confounding factors and be more explicit, such as studying an event such as a friend’s marriage or a personal loss.

Keeping this point about the limited nature of the study’s model in mind, it is possible to explore the robustness and rationale of the model more earnestly and thoroughly. Coviello et al. (2014) describes their model concerning social interaction in the following way:

“Instead of changing the user’s emotion directly with an experimental treatment, we let rainfall do the work for us by measuring how much the rain-induced change in a user’s expression predicts the change in the user’s friend’s expression” (2).

This is the unique aspect of the alternative model that Coviello et al. (2014) has developed: pairing the user’s change in emotional state with their friends in a way that fits the user’s changes to predict the changes in the friend’s of the user’s emotional states.

This solves two issues.

• First, by using rainfall as a conductor of emotional change, the authors are more likely to get consistent, wide-reaching results in comparison to something like relationship breakups, which do produce large emotional impacts but are less homogenous in how different groups cope and therefore harder to measure in terms of how much online transference played a role in mediating the emotional changes in comparison to a life-changing event.
• Second, matching the user’s emotional change to the group of their friends emotional changes helps to control for the issue of homophily mentioned before. Even if the users preferentially choose groups of friends that are more likely to express a certain set of emotions, those idiosyncrasies will be captured when measuring a unique individual against a unique group of friends that they have chosen. This improves the reliability of the study’s results when making claims regarding large and massive scales of social media information or trends.

A number of other studies corroborate the results of Coviello et al.’s (2014) points regarding how social media platforms appear to be mediators of a process that increases the global emotional synchrony of a particular emotion.

For example, Kramer, Guillory, and Hancock’s (2013) work is seminal in how it identifies the possibility that social media platforms themselves are mediators for emotional contagions that can have positive or negative qualities and that can persist in the long term.

Kramer et al. (2013) use a massive sample (around 700,000 Facebook users) to draw their conclusion that “emotions expressed by others on Facebook influence our own emotions constituting experimental evidence for massive-scale contagion via social networks…” and that in-person interaction and nonverbal cues are not strictly necessary for emotional contagion.

The authors of both draw from large samples of user data gathered from Facebook and they both contextualize this data within a unique interpretative framework.

The three major areas of evidence in our thesis derive from the conclusions made in Kramer et al.’s (2013) study and the work that Coviello et al (2014) have put forth into identifying the mechanism and extent of indirect emotional contagion.

Recall that the first area involves establishing how social media platforms mediate the transfer of emotions despite a lack of physical interaction and cues between individuals.

The next section deals additionally with this first area in explaining how certain posting functions of Twitter mediate the spread and reinforcement of political opinions.

In a similar framework to Coviello et al. (2014) utilized rain as a litmus test for the change in a emotional states for a large number of individuals on Facebook, authors Stefan Stieglitz and Linh Dang-Xuan note that Twitter functions as an ideal social media platform that creates total emotional synchrony over political opinions due to how the feature of “retweeting” reinforces certain political tendencies through allowing fast, informal gestures of public approval or disapproval.

The authors breakdown the nature of a tweet as existing within the context of a continuous social environment where individuals can easily access and judge the feelings or sentiments of other users.

The authors define the unit or mode of transfer emotions as grounded in the site’s format of a short Twitter message or “tweet” and concludes that:

“Based on a data set of 64,431 political tweets, we find a positive relationship between the quantity of words indicating affective dimensions including positive and negative emotions associated with certain political parties or politicians, in a tweet and its retweet rate”.

Again, in terms of rhetorical techniques the use of numbers and qualification is incredibly important in signaling to the reader what conclusions the authors are trying to draw.

First, the sample size is large: over 60,000 tweets are gathered for analysis of their content.

Second, the framework applied to the interpretation qualifies the mediator of emotional contagion as occurring through tweets with political content.

By defining a statistical standard as well as framework for what constitutes a political tweet, Steiglitz and Dang-Xuan (2012) are able to point the discussion towards the causal link between political tweets and an online emotional contagion transmitted through large social media platforms under specific conditions.

Further, Stieglitz and Dang-Xuan (2012) structure their paper in a logical manner in order to enhance the persuasive impact of their proposed evidence-based models based on their observations.

This structuring is best shown through how these authors offer a robust background regarding the relationship between Twitter and political communication or between retweeting practice and information diffusion in a background literature review.

On the former, Stieglitz and Dang-Xuan again bolster their argument through rational means by citing pertinent statistical information surrounding previous results:

“A number of studies focusing on different parliamentary uses of Twitter have been published… For example, Golbeck et al. focused on the U.S. Congress and analyzed the contexts of over 6,000 tweets from members of Congress” (p. 3502).

They similarly cite a number of studies, such as from Lerman and Ghosh on the empirical analysis those authors had previously performed on user activity on social media platforms such as Digg or Twitter in order the further establish the precedent for how retweeting links to information diffusion in a persuasive argument that relies on a mixture of the credibility of the authors (ethos) and the rational conclusions they have drawn (logos).

After establishing this background, Stieglitz and Dang-Xuan confidently begin to build their own framework as an extension of the previous work done by other authors on these data and trends concerning how political tweets function affect global emotional synchrony.

They find trends such as the conclusion that:

“…leftists seemed to stimulate the discussion by being actors who are highly retweeted. This was also in line with the election results. As an implication, it is important for politicians and political parties to identify the most influential users and follow these discussions…” (Stieglitz and Dang-Xuan 2012, p. 3507).

Finally, Stieglitz and Dang-Xuan finish rounding off their appeal to logic by acknowledging the flaws of their study in the lack of generalizability in their sample because of how their sample had been “restricted to regional political events”, legitimizing the academic intent of their study through this acknowledgement (p. 3507).

Regarding the second area of data, the relationship between positive and negative emotions in spreading emotions through social media platforms, Stieglitz and Dang-Xuan (2012) repeat these rhetorical methods in drawing a relationship between the fact that certain emotionally charged tweets may have greater rates of information diffusion or be retweeted more often in comparison to those that are neutral.

In a separate paper, Stieglitz and Dang-Xuan (2013) again begin with the rhetorical technique of providing background information to establish a level of credibility among their readers as well as to introduce the reader to the conclusions made in literature. For instance, consider how at the outset of the paper, Stieglitz’s is stated to be a faculty member with a Ph.D. while Dang-Xuan is stated to be a Ph.D. candidate.

The signaling of these authors credentials and their university affiliations is a clear appeal to authority, though passive. Like before, a number of a background studies are offered to contextualize the problem before Stieglitz or Dang-Xuan even begin their arguments: they contextualize the relevance of Twitter in transmitting information and the categorical nature of some of these communications:

“Recent studies have shed light on the user of Twitter in various contexts. Kwak et al. conducted a large-scale study to analyze the topological characteristics of Twitter and reveal its power as a new medium of information sharing” (p. 220).

It is apparent how Stieglitz and Dang-Xuan (2013) are framing their argument through this background and priming their audience to their idea concerning the relationship between the emotional quality of a tweet and its impact on political information dissemination: this is a rhetorical move.

After a wide presentation of statistical data and tabulated information sorting various tweets (n=160,000, a large sample size again) according to their emotional quality, influence of the tweeter, and overall impact, Stieglitz and Dang-Xuan (2013) make their conclusion:

“…we found that the affective dimensions (positive or negative sentiment) of political Twitter messages are indeed significantly associated with retweet behavior in terms of retweet quantity, in the way that emotionally charged tweets are more likely to be disseminated compared to neutral ones” (p. 241).

Significantly, we can see these rhetorical techniques applied to another study in the second area of evidence regarding how the emotional content or charged nature of messages among social media platforms relates to how fast they spread.

In their study on the relationship between sentiment and information diffusion in social media (Facebook and Twitter), Emilio Ferrara and Zeyao Yang (2015) claim that their findings show:

“…negative messages spread faster than positive ones, but positive ones reach larger audiences, suggesting that people are more inclined to share and favorite positive contents, the so-called positive bias”.

They also remark that positive conversation is usually associated with anticipated events while negative conversations usually center around unexpected events in their conclusions regarding the temporal dynamics of entire conversations (Ferrara & Yang 2015).

Following the above rhetorical model that we have established previously, it is apparent that Ferrara and Yang (2015) also subscribe to this method of persuasion as they too begin their research by presenting background information and citing the studies of other authors to contextualize their own findings as an extension of ongoing research.

The first paragraph contains a massive amount of previous research regarding the field of “computational social science on studying the characteristics of techno-social systems to understand the effects of technologically-mediated communication on our society” (Ferrara and Yang 2015).

Each piece of background information that leads up to Ferrara and Yang’s (2015) foray into how positive and negative emotions play a role in the rate of information dissemination on social media platforms is supported by three to five other peer-reviewed journal articles.

Further Reading and Sources

Coviello, Lorenzo, Yunkyu Sohn, Adam D. I. Kramer, Cameron Marlow, Massimo Franceschetti, Nicholas A. Christakis, and James H. Fowler. “Detecting Emotional Contagion in Massive Social Networks.” PLOS ONE. Public Library of Science, 12 Mar. 2014. Web. 28 Feb. 2017.

Ferrara, Emilio, and Zeyao Yang. “Measuring Emotional Contagion in Social Media.” PLoS ONE. Public Library of Science, 2015. Web. 22 Feb. 2017.

Ferrara, Emilio, and Zeyao Yang. “Quantifying the Effect of Sentiment on Information Diffusion in Social Media.” PeerJ Computer Science. PeerJ Inc., 30 Sept. 2015. Web. 28 Feb. 2017.

Kramera, Adam D. I., and Jamie E. Guillory. “Adam D. I. Kramer.” Proceedings of the National Academy of Sciences. PNAS, 25 Mar. 2015. Web. 22 Feb. 2017.

Stieglitz, Stefan, and Linh Dang-Xuan. “Emotions and Information Diffusion in Social Media — Sentiment of Microblogs and Sharing Behavior.” ResearchGate. Journal of Management Information Systems, Apr. 2013. Web. 28 Feb. 2017.

Stieglitz, Stefan, and Linh Dang-Xuan. “Political Communication and Influence through Microblogging — An Empirical Analysis of Sentiment in Twitter Messages and Retweet Behavior — IEEE Xplore Document.” Political Communication and Influence through Microblogging — An Empirical Analysis of Sentiment in Twitter Messages and Retweet Behavior. IEEE Xplore Digital Library, 9 Feb. 2009. Web. 28 Feb. 2017.