Digital Hermits

Abstract

When users share multi-dimensional data about themselves with a firm, the firm learns about
the correlations of different dimensions of user data. We incorporate this type of learning into
a model of a data market in which a firm acquires data from users with privacy concerns. Each
user can share no data, only non-sensitive data, or their full data with the firm. As the firm
collects more data and becomes better at drawing inferences about a user’s privacy-sensitive
data from their non-sensitive data, the share of new users who share no data (”digital hermits”)
grows. This growth of digital hermits occurs although the firm offers higher compensation for a
user’s non-sensitive data and a user’s full data as its ability to draw inferences improves. At the
same time, the share of new users who share their full data also grows. The model thus predicts
a polarization of users’ data sharing choices away from non-sensitive data sharing to no sharing
and full sharing. Our model suggests that recent privacy policies, which are focused on control
of data rather than inferences, may be misplaced.

Keywords: data, privacy, informational externalities, prediction, digital markets

JEL Classification: D42, D82, D83, L20