Microsoft Word - Volume37_final insna.org | Issues 1&2 | Volume 37 | 89 IkeNet: Email and Friendship Evolution Ian McCulloh Abstract Network evolution is an important problem for social scientists, management consultants, and social network scholars. Unfortunately, few empirical data sets exist that have sufficient data to fully explore evolution dynamics. Increasingly, more and more online data sets are used in lieu of offline, face-to-face data. The veracities of these findings are questionable, however, because there are few studies exploring the similarity of online-offline dynamics. The IkeNet project investigated online and offline network evolution. Empirical data was collected on a group of 22 mid-career military officers going through a one-year graduate program. Data collection included email communication collected from the Exchange server, as well as self- reported friendship, and time spent together, over a course of 20 weeks. Numerous attribute data on the individual actors was collected from their military personnel files. The data allows network scholars to conduct research into the dynamics of network evolution and allows edu- cators a real-world example data set for use in classroom instruction. Author Ian McCulloh holds joint appointments as a Parson’s Fellow in the Bloomberg School of Public health, a Senior Lecturer in the Whiting School of Engineering and a senior scientist at the Applied Physics Lab, at Johns Hopkins University. He is the Chief Technology Officer for Arrow Analytics, a management and neural marketing consulting firm. His current research is focused on strategic influence in online networks. His most recent papers have been focused on the neuroscience of persuasion and measuring influence in online social media firestorms. He is the author of “Social Network Analysis with Applications” (Wiley: 2013), “Networks Over Time” (Oxford: forthcoming) and has published 48 peer-reviewed papers, primarily in the area of social network analysis. He retired as a Lieutenant Colonel from the US Army after 20 years of service in special operations and improvised explosive device forensics. He founded the West Point Network Science Center and created the Army’s Advanced Network Analysis and Targeting (ANAT) program. Please send all correspondence to imccull4@jhu.edu Connections IkeNet: Email and Friendship Evolution 90 | Volume 37 | Issues 1&2 | insna.org 1. Overview The IkeNet data is from one year of a five- year strategic study of social networks among mid-career military officers admit- ted to a one-year graduate program run jointly by Columbia University and the U.S. Military Academy at West Point. Most of these students/officers serve for two to three years as tactical officers at West Point following completion of their master’s degree program. Tactical officers are responsible for overseeing the cadet chain of command at the U.S. Military Academy and for providing the primary leadership training for cadets. This data focuses on friendship formation and evolution during the first 20 weeks of their program. The stu- dents/officers were recruited for this study on the first morning of their graduate pro- gram, prior to meeting other participants for the first time, although some subjects may have had random interaction with each other in the past. The first network survey was collected during that morning, and then weekly for the following time pe- riods. One subject out of 22 chose not to participate in this study. The response rate on self-reported survey data among the participants was 97%, which creates a 92% overall response rate. Email data was cap- tured at the exchange server, therefore that response rate is 100%. This presents a unique data set for exploration of friend- ship formation, multi-level network analy- sis, longitudinal network analysis, and comparison between email and face-to- face networks. 2. Data Collection Social network data was collected on a group of 21 mid-career officers in the U.S. Military. The officers were all enrolled in the Eisenhower Leadership Development Program (ELDP) where they complete a one-year graduate program administered jointly by Columbia University and the U.S. Military Academy. Social networks included self-reported friendship, time spent together, and email communication collected from the central email exchange server (Ring et al., 2008). Data was col- lected beginning on the first day that the officers reported for duty and met each other for the first time. They were all given blackberries that allowed them to be con- nected to their military email accounts to facilitate email communication for the purpose of the project. The specific year of this data is in- tentionally omitted to protect the privacy of the respondents. The “IkeNet” project was conducted over five years from 2006 to 2011. In this particular year, 21 of 22 of- ficers in the program consented to be par- ticipants in this project. All data from the non-consenting officer were removed to protect their privacy. The U.S. Army and the West Point Institutional Review Boards approved this research. Surveys were conducted on a week- ly schedule, collecting network data on friendship and time spent together for 20 weeks. The boundary of the network was defined using a realist approach (Wasser- man and Faust, 1994; McCulloh et al., 2013). The boundaries in this experiment were the members of the Eisenhower Leadership Development Program (ELDP). This group consisted of 21 offic- ers. Two of the officers were women and 19 were men. One of the men was a U.S. Coast Guard officer, while the other 20 of- ficers were in the U.S. Army. There were 17 Caucasian officers to include the two women. Three of the officers were Black and one was Asian. The ages of the offic- ers ranged from 26 to 33, with most offic- ers aged 30. Attribute data was collected on the participants from their officer record brief (ORB). The ORB contains a wealth of in- formation on the officers. Six officers to include the two women were Roman Cath- olic. 12 officers were of various protestant Christian denominations. For the purpose of this study, they were coded as protestant. This decision was made, be- cause on military installations all protestant faiths meet in a non- IkeNet: Email and Friendship Evolution Connections insna.org | Issues 1&2 | Volume 37 | 91 denominational protestant service, while Roman Catholics meet in a different ser- vice. Therefore, it is likely that if protestant officers attend church, they are likely to attend the same church regardless of their specific denomination. One officer was Buddhist, one had no preference re- ported for religion, and no data was availa- ble for the Coast Guard officer. Several experiments have recently been run out of West Point concerning the formation of email networks. Two collec- tion methods identified include decentral- ized, client-side collection and a central- ized collection of messages from the mail server (Ring et al., 2008). The IkeNet stud- ies conducted at West Point have demon- strated findings that favor a centralized da- ta collection method over a client-side method (McCulloh and Ring, 2008). Addi- tionally, research has been conducted on how to unobtrusively collect these net- works to reduce respondent burden (McCulloh and Ring, 2008). The source and time of commission was also collected from the ORB. Nine of- ficers to include the two women were graduates of the U.S. Military Academy. The Coast Guard officer was a graduate of the Coast Guard Academy. Eight officers were commissioned through the Reserve Officer Training Corps (ROTC) and the remaining three were commissioned through Officer Candidate School (OCS). The military tracks officer promotions and career management by year groups. There- fore, each officer’s year group was also recorded. One officer was commissioned in 2000, 11 in 2001, seven in 2002, one in 2003, and the Coast Guard officer was commissioned in 2005. Another important attribute for so- cial interaction may be competence or ex- perience. The undergraduate grade point average (GPA) was collected for each of- ficer from his or her application to the graduate program. The GPA ranged from 2.3 to 4.0 with an average GPA of 3.05 and a standard deviation of 0.51. For experi- ence, the number of months an officer spent in command was recorded from their ORB. Command is the “key developmental job” (KD) for a captain and is an experien- tial requirement that an officer must com- plete prior to being admitted into the ELDP program. The only exception is for the Coast Guard officer who must meet different career milestones respective to his military branch of service. The maxi- mum time any officer spent in command was 32 months and the minimum time was 13 months. The average was 23.2 with a standard deviation of 6.1. Each week the officers completed two surveys: friendship and time spent to- gether. There was a 97% response rate among participants. With one out of 22 of- ficers choosing not to participate, the over- all response rate was 92%. In the cases of non-response, the dyadic rating between the preceding and proceeding weeks did not change, so it was simple to interpolate the missing data. Unfortunately, there was no indicator coded in the data for where this occurred. The missing data were ran- dom and believed to be due to oversight. Given that the omission was not discov- ered for about a week and that the ratings for those omissions did not change, there was no attempt to go back to respondents to correct their response. The friendship survey question asked respondents, “Please rate how well you like the other members of your ELDP cohort according to the following scale:” The scale was a seven-point Likert scale as follows: Connections IkeNet: Email and Friendship Evolution 92 | Volume 37 | Issues 1&2 | insna.org Coded Rating (this is what was shown to respondents) 1 “I strongly dislike this person” 2 “I do not care for this person” 3 “I am neutral toward this person” 4 “I like this person” 5 “This person is one of my better friends at this duty station” 6 “This person is one of my better friends overall” 7 “I consider this person one of my closest friends” The time spent together survey question asked respondents, “Please rate the time you spend with the other members of your ELDP cohort according to the fol- lowing scale:” The scale was a seven-point Likert scale as follows: Coded Rating (this is what was shown to respondents) 1 “I avoid this person” 2 “I associate with this person only for official business at work” 3 “I socialize with this person at work” 4 “I occasionally (monthly) get together with this person outside of work” 5 “I regularly (weekly) get together with this person outside of work” 6 “I regularly (weekly) spend time with this person at their/my home” 7 “I go on pass/leave [vacation] with this person” This project collected a rich group of networks surrounding 21 mid-career Army officers. Three networks were col- lected from the participants: friendship, email and the time spent together. The friendship and time spent together were collected through the use of questionnaires on a weekly basis. The email network was collected at the server level and only col- lected the header information of emails (Ring et al., 2008). 3. Data Files and Formats The data are provided as flat comma sepa- rated value files. The file “Attributes.csv” provides the attribute data. There are 34 columns in this file, corresponding to dif- ferent attributes and described by the table below. Column Attribute Description A long tour Number of overseas assignments the officer has had that exceed one year B short tour Number of overseas assignments the officer has had that are one year or less C commands Number of command assignments the officer has had D birthplace Officer’s state of birth E branch Two letter designation of the officers career field F Cbt Tour Number of combat deployments completed G Kids Number of children the officer lists as dependents H CmdMos Number of months spent in command assignments I comm Source of commission: United States Military Academy (US- MA), Reserve Officer Training Corps (ROTC), or Officer Candi- date School (OCS) J followon Follow-on assignment after officer finishes program IkeNet: Email and Friendship Evolution Connections insna.org | Issues 1&2 | Volume 37 | 93 Column Attribute Description K GPA Undergraduate GPA on program application L GRE1 Verbal score on graduate record exam from application M GRE2 Math score on graduate record exam from application N height Height, in inches O HOR State listed as home of record P lang1 Secondary language after English Q lang2 Tertiary language after English R marital Marital status: single, married, or divorced S OPNTour Number of operational deployments, not classified as combat T Race Caucasian (C), Black (B), and Asian (A) U Religion Religion as reported in personnel file V Rest Tour Unknown – Included in personnel file W sex Sex X skill1 Additional skill identifier to denote certain skills used in assign- ment decisions, such as paratrooper or ranger. Y skill2 As above Z skill3 As above AA skill4 As above AB skill5 As above AC spousebirth Spouse’s location of birth AD state State of residence on record for tax purposes AE Undergrad Major Academic major studied in undergraduate program AF Undergrad School Name of the undergraduate academic institution AG Weight Officer’s weight in pounds AH YG Officer’s year group, usually the year they commissioned The email network data is con- tained in files “email01.csv”, “email02.csv”, …, “email20.csv” corre- sponding to weeks one through 20. Each csv contains an adjacency matrix, where the elements of the matrix represent the number of email messages sent from the row element to the column element. The ordering of the rows and columns are con- sistent with the ordering of attributes in the “Attributes.csv” file and with other adja- cency matrices included. The friendship network data is con- tained in files “like01.csv”, “like02.csv”, …, “like20.csv” corresponding to weeks one through 20. Each csv contains an adja- cency matrix, where the elements of the matrix represent the friendship rating that the row element assigns to each column el- ement. The rating follows a seven-point Likert scale. Subjects do not rate them- selves, so “0” represents an N/A value. The ordering of rows and columns are con- sistent with the ordering of attributes in the “Attributes.csv” file and with other adja- cency matrices included. The time spent together network data is contained in files “time01.csv”, “time02.csv”, …, “time20.csv” corre- sponding to weeks one through 20. Each csv contains an adjacency matrix, where the elements of the matrix represent the friendship rating that the row element as- signs to each column element. The rating follows a seven-point Likert scale. Sub- jects do not rate themselves, so “0” repre- sents an N/A value. The ordering of rows and columns are consistent with the order- ing of attributes in the “Attributes.csv” file and with other adjacency matrices includ- ed. Connections IkeNet: Email and Friendship Evolution 94 | Volume 37 | Issues 1&2 | insna.org 4. Data Details Response Rate 100% for email networks, collected at the central server 92% for self-reported survey networks 100% for attribute data, collected from personnel files Non-Respondent Bias None. There was no dyadic change from the data collected in the pre- ceding and following time periods for the few cases of missing data, allowing easy interpolation of data Theoretical Grouping N/A Publication Using These Data None to date Data Context Friendship formation among a group of mid-career military officers in a one-year graduate school program Respondents Mid-career military officers in the Eisenhower Leadership Develop- ment Program of the US Army and Columbia University Longitudinal The first 14 weeks of the program. Temporality Multiple ties are recorded between actors on a weekly basis. Assign- ment cycles typically involved major assignments due every other week. Analytical or Pedagogical Utility  Longitudinal network data with 14 time periods  Multi-level networks with three relations: self-reported friend- ship, self-reported time spent together, email  Network evolution, capturing data as the actors meet for the first time  Node and edge covariates Known Issues References McCulloh, I., Armstrong, H. & Johnson, A.N. (2013). Social Network Analysis With Applications. Hobo- ken, NJ: Wiley. McCulloh, I. & Ring, B. (2008). Unobtrusive Social Network Data from E-mail. In Proceedings, 26th Army Science Conference. Orlando, FL. Ring, B., Henderson, S., & McCulloh, I. (2008). Gath- ering and Studying Email Traffic to Understand Social Networks. In H.R. Arabnia & R.R. Hashe- mi (Eds.), Proceedings of The 2008 International Conference On Information & Knowledge Engi- neering (pp. 338-343). Las Vegas: CSREA Press. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications (Vol. 8). Cambridge: Cambridge University Press.