Predicting gender from electronic discourse

There is substantial evidence of gender differences in face-to-face communi cation, and we suspect that similar differences are present in electronic c ommunication. We designed three studies to examine gender-preferential lang uage style in electronic discourse. In Expt 1, participants sent electronic messages to a designated 'netpal'. A discriminant analysis showed that it was possible to successfully classify the participants by gender with 91.4% accuracy. In Expts 2 and 3, we wanted to determine whether readers of e-ma ils could accurately identify author gender. We gave participants a selecti on of messages from Expt 1 and asked them to predict the author's gender. I t was found that for 14 of the 16 messages used, the gender of author was c orrectly predicted. In the third experiment, six messages about gender-neut ral topics were composed. Using a subset of the variables identified in Exp t 1, female and male versions of each message were created. When participan ts were asked to rate whether a female or a male wrote these messages, thei r ratings differed as a function of the message version. These findings est ablish that people use gender-preferential language in informal electronic discourse. Furthermore, readers of these messages can use these gender-link ed language differences to identify the author's gender.