Tweet the words “love,” “OMG” and “cute,” and people are more likely to think you are female. Tweet the words “ebola,” “sports” and “war,” and people are more likely to assume you are male.
However, both of these assumptions would be wrong, according to a recent Penn study.
Penn researchers analyzed how people interpret the language of tweets in a study about stereotypes. In their study, entitled “Real Men Don’t Say Cute,” 3,000 participants read 120,000 different tweets and were able to predict the gender, education level, age and political orientation of the tweet’s author 68 percent of the time.
The study was published in the journal Social Psychological and Personality Science in November.
Just by looking at the text of a tweet, participants could identify the gender of the tweeter 76 percent of the time, whether the person was above or below 24 years old 69 percent of the time and the person’s political orientation 82 percent of the time.
Only in guessing education level did stereotypes lead people astray — participants correctly guessed whether the tweeter had no degree, a bachelor’s degree or an advanced degree only 46 percent of the time.
“The main message is that people are mainly correct but that stereotypes are inaccurate in the sense that they are always exaggerated,” said Daniel Preoţiuc-Pietro, a postdoctoral student at Penn’s Positive Psychology Center and one of the lead researchers of the study.
Both Preoţiuc-Pietro and fellow lead researcher Jordan Carpenter, now a postdoctoral student at Duke University, were involved in the World Well-Being Project of the Positive Psychology Center. The project is a three-year-old interdisciplinary collaboration between statisticians, psychologists like Carpenter and computer scientists like Preoţiuc-Pietro, all using big data to analyze language.
Preoţiuc-Pietro said the study represents a novel approach — rather than asking people to describe the stereotypes they hold, the researchers asked people to use their stereotypes in judging isolated tweets. This method reduces bias in the way people represent themselves and reveals implicit biases, reversing the way researchers had been looking at the problem, he said.
“In face-to-face interactions, people simultaneously use information from multiple channels to categorize others, which makes it ambiguous what cues were most important,” the study said. “Using social media language lets us isolate a single channel within the context of everyday life, allowing us more certainty that the identified stereotypes are real.”
In the study, participants largely assumed that feminine-sounding language belonged to liberals, that tweets related to technology or sports were written by men, that tweets about personal matters were written by women and that people who hold doctoral degrees did not use foul language.
Many of these stereotypes were supported by the data, but participants always over-blew them, assuming they held true far more than they actually did.
“The most useful aspect is making people aware of their stereotypes in order to intervene,” Preoţiuc-Pietro said. “People are usually not aware of these things, but they should be, because that is the first step of correcting these incorrect stereotypes.”
Preoţiuc-Pietro said the research will also be helpful in the field of computer science and artificial intelligence. For example, a computer scientist could program educational software for children to use language that they would perceive as more youthful, making it more relatable.
“Our studies indicate the power of big data methods to quantitatively compare actual and perceived behavioral tendencies across groups,” the study said. “Using social media text to unobtrusively measure both behaviors and perceptions of those behaviors can reveal surprising, important features of people’s stereotypical beliefs and their levels of correctness.”
All comments eligible for publication in Daily Pennsylvanian, Inc. publications.