Email reciprocity and personal communication bias Bernie Hogan Dept of Sociology, University of Toronto 725 Spadina Avenue Toronto, ON MS5 2J4, Canada bernie.hogan@utoronto.ca Danyel Fisher Microsoft Research 1 Microsoft Way Redmond, Washington USA 98052 danyelf@microsoft.com ABSTRACT In this paper, we discuss differences in reciprocity in core networks in email inboxes using data from 538 cases at a high- tech organization. We highlight the fact that reciprocity in email behavior is different between multi-recipient and dyadic mail and that a consistent number of alters send disproportionate volumes of mail. Keywords E-mail; Social Network Analysis; Multilevel analysis; Egocentered networks 1. INTRODUCTION Email use has transcended mere tool to become a habitat in which work gets done [2]. It is asserted that one of the advantages of this space is its ability to level the conversational playing field. Studies have shown that compared to face-to-face conversations, email communication is more egalitarian [7] and is a source of ‘status equalization’. While there has been much attention to the type of conversations taking place in email, whether they are aggressive or more collaborative in nature, little attention has been paid to the relative volumes of email being distributed, and how (and why) individuals vary in the amount of email sent to their alters. No doubt there is some expected variation in the amount of email sent and received, but we have little knowledge of who sends more mail, and whether this imbalance is because individuals intensively send more mail to a small number of alters, or extensively send more mail to large swaths of their networks. Knowledge of distributions of reciprocity can build on communications research to give a clear profile of the extent to which certain individuals in a company are dominating inboxes and perhaps contributing to email overload. It will also help give individuals a normative expectation of how much email is too much and under what circumstances. We assert that present the results of an analysis of 538 email inboxes at a large High-Tech company in the U.S. to illustrate our findings. In doing so, we hope to contribute to the literature on exchange theory and to the CSCW literature on email use in the workplace. 2. RELEVANT LITERATURE Studies highlight the benefits to network research of using recorded communication data as it enables precise metrics on network patterns when recall data is not appropriate or entirely reliable. To this end researchers have looked at turn-taking and exchange in online forums [4] and at the structure of the networks produced by email [5,6,7]. These studies however, have given little attention to the differences in propensity to reciprocate via email. We focus mainly on this propensity by examining the proportion of ties who have sent more messages than ego (alter-biased ties), the proportion who have sent less messages than ego (ego- biased ties) and the proportion of ties who sent roughly the same amount. Here we present the relative distributions of these ties. Expanded versions of this work examine these proportions in relation to organizational role and look more intensively at possible reasons for the different biases. 3. METHODS Data for this project came from the deployment of an experimental technology designed to facilitate email checking. The program was released internally on July 13, 2005. During the 4-week adoption window 538 people used the program long enough to upload an uncorrupted snapshot. To enable certain functionality, this application created pseudo-anonymized inbox databases (or ‘snapshots’) that constitute the raw data for this project. The snapshots do not contain message bodies nor do they contain identifiable email addresses. Moreover, the same email address was represented by a different identifier in a different respondent’s snapshot. Email networks: To build the email egonets we selected messages 0-60 days prior to the date of snapshot submission. The networks are directed with weights representing the number of messages sent. One can conceptually think of three nested graphs of interest for each respondent (figure 1). The largest graph includes all addresses and arcs in the respondent’s inbox. The second graph is the neighborhood of ego. In the third there must be an arc to and from ego and the sum of the weights of these two arcs have to be greater than 3. This criteria eliminates distribution lists, bots, spam and individuals with whom ego has not had much correspondence. We consider reciprocity between ego and the alters in graph 3. Figure 1. Selection criteria for three nested email egonets. Copyright is held by author/owner(s). NetSci2006, May 22–25, 2006, Bloomington, IN, USA.