Papers:
[1] Experimental evidence of massive-scale emotional contagion through social networks – Kramer et. al.
[2] A 61-million-person experiment in social influence and political mobilization – Bond et. al.
Summary:
[1] talks about emotional contagion in social networks; how emotions expressed by others influence our own emotions. It hints that these effects may even be long term, and are often subconsciously experienced. It tries to quantify whether seeing positive posts on their feed makes a user themselves post more positively, and vice versa.
[2] also talks about the importance of social influence for effecting behavior change, but it focuses on social contagion in actions rather than emotions. It randomly showed users either an informational message (i.e. statistics about how many people voted, and encouragement to vote), a social message (i.e. profile pictures of friends who voted along with the informational message) or no message, and analyzed the effects of the messages on users, and also analyzed the effects of close friends getting the message on users.
Reflection:
The authors in [1] state that they determined a post to be positive or negative if it contained at least one word which was positive or negative according to LIWC respectively, but this seems to be a fairly weak way of measuring the emotional content of the post, and the data might contain lots of “weakly positive” or “weakly negative” posts.
[1] was also an interesting insight into statistical analysis. It used a weighted linear regression model to test the hypothesis. The authors took care to try and reduce the effects of other factors, by ensuring that the subjects had no significant deviations in their emotional expression during the week before the experiment, and conducted t-tests to check the correlation between variables.
However, since emotional state is a complex thing to quantify, and since the sample size was so large, there is a lot of scope for external unforeseen variables which might muddy the results. Since the sample size is large, small effects can be found to have “significant” p-values. Moreover the effects of social contagion were not very pronounced. The percentage of change in number of words of a particular emotion used by a user after changing the news feeds were less than 0.1%.
As the study in [1] was only conducted over a week, studies can also be conducted to see the effects of emotional contagion for shorter or longer periods of time, or measure the rate of change of emotional word usage over time.
In [2], since a major factor in deciding whether users would vote or become interested in voting were the close friends of the user, the results become hard to isolate from effects beyond social media. This is because user’s have plenty of real world interactions with close friends, and this might be what influenced them to vote, not the messages on social media.
Interestingly, in [2], the users who saw the social message were 0.39% likelier to actually vote, but 2.08% likelier to say they voted. This hints that users are aware of the social stigma that comes with not voting, and often the validation from their peers for having voted is a stronger driving force than the inherent desire to vote. Maybe a system can be designed so that only verified voters can click the I Voted button, and the effects that has on voting patterns can be analyzed.
The probability of encouraging a user to vote through online social contagion is found to be quite small, although the authors do (rightly) state that even the small effects can have large consequences, since voting races are narrowly decided.
