Psychology
March 04, 2019
When our friends or family members face an episode of depression, we can often see signs and symptoms in their behavior and speech. A new study quantifies the link between writing and depression based on Facebook posts and medical records.
Major depression is one of the most common disorders in the United States. It also causes significant disability. According to the NIH, around 10 million American adults face at least one episode of major depression every year. 64% of of these Americans faced severe impairment or disability from their depressive event.
Researchers looked through Facebook status posts from 683 patients. Among the patients studied, 114 had a medical diagnosis of depression.
In order to see whether, and how, Facebook posts could predict depression, researchers looked through years of Facebook statuses for all patients. They then looked for the features that best predicted depression prior to the actual diagnosis.
In their results, they found that words pertaining to health were most predictive of depression. Users who used the words "hospital" or "pain" were significantly more likely to be diagnosed with depression within 6 months.
Words such as "better" and "hope" were also similarly predictive of depression, and had a higher significance than even more negative words such as "tears", "pain", and "crying."
These results suggest that while depression sometimes shows in an individual's writing, the words used may not be the ones we would naturally associate with a depressed mood. Words pertaining to health may reflect the physical symptoms of depression and other somewhat positive words may reflect a profoundly negative mood.
Source: Facebook language predicts depression in medical records
The words used in a post were just one of several factors studied by researchers. However, it was by far the most predictive. On a scale of 0.5-1, with 0.5 being random chance and 1 being perfect prediction, word choice scored a 0.68.
The language used in a post predicted depression better than the demographics of the person posting. This goes to show that depression can affect anyone. There is no specific profile of a depressed person.
Length and frequency of the posts were also predictive, albeit less so than word choice. This suggests that patients who are depressed write long posts and post more often on Facebook.
However, keep in mind that this is only a correlation. Not all long Facebook posts about the hospital are depressed patients. In fact, most users are probably just fine statistically speaking.
As expected, posts just prior to a diagnosis showed more signs of depression than earlier posts. A deeper dive into the data shows the algorithm developed by researchers could predict depression with some accuracy as far as 1 year prior to the actual diagnosis.
The data shows an impressive correlation between a user's choice of words and a medical diagnosis of major depression. If you suspect that a friend may be depressed based on their writing, these data provide statistical evidence for that intuition.
However, this correlation does not provide a diagnostic tool. Just Facebook posts alone do not provide enough "predictive power." The researchers behind the study speculate that one day, Facebook posts may be combined with other "signals" to provide a tool to identify potential depression patients. This tool would potentially save lives by preventing suicides.
These data also suggest one reason why depression often goes unnoticed. Even for users who write about their personal feelings in their Facebook posts, the words most predictive about depression often have nothing to do with depressed moods.