Researchers at IBM found that engagement levels of employees may be inferred from their word choices in social media posts (1). Meanwhile, in his article in Wired, Tim Simonite sheds light on the current debate surrounding the efficacy and ethics of academic institutions monitoring student social media posts to prevent tragedies such as school shootings (2). As social media monitoring continues and the applicability of such monitoring is explored, a likely complement will be the heightened attention to mental health exhibited today by employers and academic institutions alike.
Social media data has been repeatedly used to predict mental illnesses like depression (3, 4). I find it likely that these findings will trickle into what employers and schools are looking at in their existing use of social media data. I predict we will soon see employers and academic institutions using social media data to predict the prevalence of mental illness among their employees and students, just as they are doing so to measure engagement levels and preempt dangerous behavior. I would like to express a view of caution against this notion. While I celebrate the increased care that society is broadly taking toward mental wellness, I am concerned for the implications of using analytics surrounding social media to take such care.
First, my concerns regarding social media analytics to monitor and measure mental health stem from the fact that social media use has been linked to diminished levels of psychological well-being (5). If employers and academic institutions introduce the monitoring of their employees’ and students’ social media as part of their measures to care for their mental health (whether or not the subjects of the monitoring are made aware of it), will this result in an encouragement by these organizations for increased use of social media for the sake of robust data sets? If social media data were used to predict mental illness, it would likely be used in tandem with mental health initiatives aimed at improving mental well-being as a way to measure the efficacy of these initiatives. However, any encouragement or prompting of social media use on behalf of these organizations could very well exacerbate the issues they would be aiming to abate.
Second, I worry for what I predict could be an over-reliance on social media analytics to monitor concerns as nuanced and important as mental health. Consider the nature of social media as a manifestation of an individual’s mental state as compared to data sources such as surveys or network data. There’s an argument to be made that social media data, given its informal tone and fostering of free-form self expression, might offer more accurate insights into an individual’s mental health than do formal surveys or measurements of the frequency of an individual’s interactions with others. However, I argue that social media data in fact trends in the opposite direction, away from accuracy. Social media introduces another layer that stands between an individual’s true mental state and the way it is captured in data. This layer consists of the careful curation, presentation, and social comparison that are all at play when a person posts on social media. These elements are also what an individual, whether deliberately or subconsciously, may use to present what is specifically an inaccurate depiction of their own well-being. It is an unfortunate cliche of our time that an unhappy person will use social media to feign happiness. Even if a sophisticated algorithm were able to capture such deceit, what about the element of curation? If individuals choose only to publicize truly positive moments or sentiments on social media and refrain from posting anything negative, would an algorithm detect such absent content? We can again give an algorithm the benefit of the doubt in terms of sophistication, and trust that it would be able to predict mental illness using word choice in any kind of social media posting, be it explicitly positive, negative, or neutral in tone. Still, there is further nuance to acknowledge in that a person might change their levels of engagement with social media depending on their mental state, preventing any data collection from capturing a full and accurate picture. All of this is to say that, with a concern like mental health, it will not be enough for organizations like employers or schools to monitor social media and trust that they have the full picture. Social media analytics must not become a box that organizations can check off and move on from. With mental health and mental illness, employers and academic institutions owe it to their employees and students to be many-model thinkers.
(1) Inferring Employee Engagement from Social Media, https://dl.acm.org/doi/abs/10.1145/2702123.2702445
(2) Schools Are Mining Students’ Social Media Posts for Signs of Trouble, https://www.wired.com/story/algorithms-monitor-student-social-media-posts/
(3) Multi-Kernel SVM Based Depression Recognition Using Social Media Data, https://link.springer.com/article/10.1007/s13042-017-0697-1
(4) Detecting Depression and Mental Illness on Social Media: An Integrative Review, https://wwbp.org/papers/detecting_depression_twitter_2017.pdf
(5) Media Use Is Linked to Lower Psychological Well-Being: Evidence from Three Datasets, https://link.springer.com/article/10.1007/s11126-019-09630-7