Eric Gilbert and Karrie Karahalios of the University of Illinois at Urbana-Champaign have begun to study how social media can be used to predict tie strength in a sample of 2,184 Facebook relationships:

They asked 35 people to rate a number of their Facebook friends that were randomly selected. Then they looked at the Facebook behavior of these people to see if could accurate predict the strength of their ties as reported by the participants. Dimensions such as the use of words associated with intimacy and intensity, duration of communication, social distance, and other factors were looked at for their predictive power. The complete set of variables was combined to form the predictive model that worked in 85% of the instances.

The authors used 74 predictive variables, spanning connection intensity, intimacy, duration, reciprocal services, structure, emotional support, social distance, demographics, and usage:

We identified 74 Facebook variables as potential predictors of tie strength. Table 1 presents 32 of these variables along with their distributions. In choosing these predictive variables, we tried to take advantage of Facebook’s breadth while simultaneously selecting variables that could carry over to other social media.