Research conducted by a team of psychologists and computer scientists has found that women tend to be warmer and friendlier than men on Facebook.
The study, published in the open-access journal PLOS One, analyzed the language of more than 68,000 Facebook users. Women were more likely to make references to social relationships and emotions, while men were more likely to make references to sports and occupations. Men were also more likely to swear.
But the effect sizes — a measurement of the average difference between men and women — was relatively small, suggesting there is a large overlap in how men and women use language.
The research team also found a “surprising degree” of similarity between men and women’s assertiveness on the social networking website.
PsyPost interviewed the study’s corresponding author, David Bryce Yaden of the University of Pennsylvania. Read his responses below:
PsyPost: Why were you interested in this topic?
Yaden: The first thing to understand is that self-identified gender is a complex, multidimensional, and fluid concept. It was not our intention to attempt a definition of gender or to explain the causes of gender differences in this study, but rather to let the data speak for itself.
The study began with this question: While the vast majority of language used by women and men is the same, researchers have found that algorithms can be used to predict users’ self-identified gender correctly over 90% of the time. What words, phrases, and linguistic themes account for the differences that these algorithms are detecting to make such accurate gender predictions?
A team of psychology and computer science researchers in the World Well-being Project at the University of Pennsylvania approached this question using a very large sample of participants who agreed to allow us to analyze their Facebook posts using computational linguistic analyses.
Greg Park, Peggy Kern, Johannes Eichstaedt, Lyle Ungar, and Andrew Schwartz contributed most prominently to this study. Please visit www.WWBP.org to learn more about this research lab.
What should the average person take away from your study?
First, we invite and encourage people to try the age and gender prediction demo here: https://wwbp.org/agegender.html
Our hope is that using the demo will be interesting to readers – and using it helps us to improve our prediction models.
People might be interested to learn that researchers can now use algorithms to predict a number of personal traits by analyzing written language. Besides gender, language analysis can predict users’ age, relationship status, drug use, ethnicity, religion, sexual orientation, political orientation, personality, intelligence, popularity, and well-being. In other words, all of these personal qualities can be predicted from what someone has written on social media.
Importantly, at least from the standpoint of psychology, researchers can look at the words and phrases that most differentiate people across these various traits. These words and phrases often contain clues about key, previously undetected features of various traits. For example, Andrew Schwartz found language that differentiates personality traits (such as extraversion versus introversion), which can be viewed here: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0073791
The method can also be used on physical health outcomes. In Johannes Eichstaedt’s study on the language of heart disease, he found the words and phrases indicative of “hostility” and “disengagement” were important predictors of communities with high incidence of heart disease. You can read Johannes Eichstaedt’s Heart Disease article here: https://pss.sagepub.com/content/early/2015/01/20/0956797614557867.abstract
In this study, we looked particularly at the dimensions of assertiveness and affiliativeness across gender. Past research has tended to characterize language differences between males and females by describing male language as more assertive (e.g. more imperative statements) and female language as more affiliative (e.g. more agreement). However, we found male and female language differed more in terms of interpersonal warmth (e.g. friendliness), with female participants showing higher interpersonal warmth. We also found, contrary to previous research, that self-identified female participants were slightly more assertive than self-identified male participants.
Are there any major caveats? What questions still need to be addressed?
Yes, there a number of important caveats and future directions.
Our sample comes from users who signed up to use an app that would ask them a number of personality-related questions and would grant researchers access to their social media posts. The kinds of people who might sign up for such an app may differ from the normal population in important ways. As with all psychological research, future studies should replicate these findings in other, more diverse samples.
Also, we looked at the dimensions of assertiveness and affiliativeness because of how often these concepts have been used in previous gender research, but there are a number of other dimensions that could be explored in future research (like how much people write, how often people reference themselves, etc…). Also, we hope to have access to data in the future that transcends the gender binary in order to include others who fit between or beyond these categories, and who have received less representation in this research topic.
Finally, again, readers should remember that self-identified gender is a complex, multidimensional, and fluid construct, and that the vast majority of language used by men and women is the same. Our findings should be interpreted in this context.
The study, “Women are Warmer but No Less Assertive than Men: Gender and Language on Facebook,” was co-authored by Gregory Park, David Bryce Yaden, H. Andrew Schwartz, Margaret L. Kern, Johannes C. Eichstaedt, Michael Kosinski, David Stillwell, Lyle H. Ungar, and Martin E. P. Seligman.
