Who benefits from network analysis: ethics of social network research

Abstract

The success of social network research (SNR) has led to expectations that in addition to academic research, SNR can introduce people to one another, solve organizational problems, map the epidemiology of AIDS, and catch criminals and terrorists. Since SNR requires that names of both respondents and their contacts be collected and used in most analyses, Institutional Review Boards become very concerned. Experiences of the author, participants in the 2003 Sun Belt Conference and the Social Network List Serve illustrate ethical issues. Proper handling of the data and the analysis, including complete control by the investigator can virtually eliminate harm to respondents and those they nominate, though perhaps not to the satisfaction of IRBs. On the benefit side, academic researchers always benefit, organizations, society and science may benefit, but individual respondents rarely do.

Introduction

The social network field may have become a victim of its own successes. The mapping of social networks with names, dates and places has become a major industry. Barry Wellman reports that “Business 2.0 anointed ‘social network applications’ in 2003 as ’The Technology of the Year” (Wellman, 2003). The New York Times has celebrated social networks as one of the “new ideas” of the year (Gertner, 2003). PC Magazine reviewed five Internet sites that attempt to introduce people to one another.

Introduced by Stanley Milgram in 1967, the theory of six degrees of separation, which supposes that you’re just a half-dozen introductions away from anyone you want to meet, has found the Internet. Sites like LinkedIn, which take hold of the six-degrees concept and put it to practical use, let you take advantage of chains of acquaintances to contact people down the line. They’re known as social-networking services. Such services use the Internet to help users expand their networks of personal and business relationships.

The process is simple. After joining one or more of these sites, you send messages to people you know, asking them to join. They in turn invite people they know, and so on. In this way, you construct an enormous network of people to whom you have personal links.

… Friendster boasts three million users; none of the others has even approached 100,000 (Metz, 2004), p 131.

One of these sites has applied for a patent. The value of one's network of friends can be calculated.¹ At the other extreme there are maps of terrorist networks and firms that apparently make a living by providing crime fighting units with software to map criminal networks. Saddam Hussein was said to have been captured in part through the application of social network mapping (Fassihi, 2003). Many organizations attempt to improve their efficiency through sociometric analyses (Krebs, 2003). Research and development laboratories map major gatekeepers of critical information. There are maps of who works with whom in biotechnology. Epidemiology was founded on the tracing of agents who carried disease and modern network methods have been applied to the HIV positive field. Structures of national leaders and decision-makers have been studied, as well as the structure and function of corporate overlaps. One could go on and on and produce what is essentially a bibliography of important social network studies.

The ethical issues are both straightforward and complex. In standard practice social science research, anonymity and confidentiality are both routinely granted to respondents, informants, and subjects in experiments and observations. In large-scale survey research with at least several hundred respondents these guarantees are very easy to keep. The researcher has no interest in the particular names of respondents, except in the case of panel studies when prior respondents need to be contacted again. Looking them up serves no purpose whatsoever.² In smaller scale qualitative studies, often organization or small community studies, who are the respondents even when given promises of anonymity may be obvious to both potential readers and to the social scientists. The latter often cannot successfully analyze the data without knowing who the respondents are. Eventual publication usually involves changing the names of respondents as well as information such as their age and occupation that might give them away even though some of this background of subjects or respondents may be important to the narrative. When it comes to the comparative organization or community studies there are further difficulties because it may be impossible or even undesirable for analytic purposes to disguise the names of the organizations or communities. Organization consultants, whether academically based or not, further face the dilemma that the observations or surveys that they produce may have consequences for the individuals surveyed or observed that the subjects may not have been aware of (Borgatti and Molina, 2003).

An extreme, but clever new technique in organizational network studies utilizes real time data. An article in Technology Review offers the following by an innovative network researcher (Pentland, 2004):

“Who are the experts within your organization? Who has the most decision-making influence? Recently, managers have started mining data from e-mail, Web pages, and other digital media for clues that will help answer such questions. That's a start, says MIT Media Lab researcher Alex Pentland, but it misses the real action: studies of office interactions indicate that as much as 80 percent of work time is spent in spoken conversation, and that critical pieces of information are transmitted by word of mouth in a serendipitous fashion. Fortunately, the data infrastructure for mining real-world interactions is already in place. Most working professionals already carry microphones (cell phones), and many also carry PDAs with ample computational horsepower. This foundation of mobile communications and processing power will support an exciting new suite of business applications: reality mining.”

In addition to these issues that are well known but for which there may not necessarily be obvious solutions, social network data have one troublesome and distinctive attribute: the collection of names of either individuals or social units is not incidental to the research but its very point. Further, the network analyst in collecting information about who relates to whom is not confined to the names of respondents or informants within the study, for they may give the names of others who have no idea that they are being named (Borgatti and Molina, 2003). In studies of elites, for example, it is common practice for a connection to be considered as valid between two individuals not within the original list of respondents if they are both named by a respondent within the study (Alba and Kadushin, 1976, Alba and Moore, 1978, Higley and Moore, 1981, Moore, 1979, Moore, 1961). The relationship or tie between individuals may be obviously pertinent to the topic of the investigation, say influence on public policy, but may also be less obviously related and even more “private” such as who is friends with whom (Kadushin, 1995). In organization settings, while who works with whom might seem a legitimate topic, who is friends with whom may lie outside the purview of an employer (Borgatti and Molina, 2003).

Finally, one need not go to respondents and ask them about their relationships. Important work in social network analysis has been done with public databases, or archived material that could be turned into databases. For example, Valdis Krebs has published a map of relations between Internet companies based on public data http://www.orgnet.com/netindustry.html. More controversially, he has analyzed networks of 9/11 terrorists based on publicly available data (Krebs).³ Recently there has been concern about using network techniques in interrogations in Iraq.

There have been a number of studies of corporate overlap based on publicly available records, for example (Mintz and Schwartz, 1985).⁴ Network analysis always makes visible that which cannot be seen by the naked eye. In the case of analysis using public records, the data were already there for “all to see” but in fact, without first collecting data from various sources and putting them in a data base and only then analyzing and graphing them, the data would have remained invisible. If these data are public, at least in principle, is there an ethical issue? The ethics lie in making assertions about connections, centrality, and power when there may be some question about the reliability and accuracy of the sources of the data even when they are publicly available. Perhaps asserting that IBM has a central position in the network of computer firms in the Internet is one thing, but asserting that a given named person has a central role in a terrorist network (and thereby marking the person for arrest or even “elimination”) is another. Granting that this is an extreme example, though one that is now on the agenda, what about making visible the connection of a particular individual that may have negative (or positive) consequences for that individual? The person in question may never have expected that the available data be used in this way.

Epidemiology in some ways can be said to have “invented” network analysis in the 19th Century. Currently, it has wide use in tracing illnesses and more recently has been used in AIDS research, subject to all sorts of safeguards that Klovehdal discusses (Klovdahl, 2005). The fact that HIV status is often hidden for good reasons raises a number of problems (Shelley et al., 1995). Contact tracing had been the modus vivendi of work with TB and even venereal diseases. In the beginning, HIV status was tantamount to sentencing an individual to a short and nasty life and resulted in much stigma. As a result, in the early years of the discovery of the disease it was virtually impossible to introduce studies that rested on contact tracing. GMHC [Gay Men's Health Crisis], for example, was strongly opposed to such work. The organization currently notes in advice to on its Web site that “New York State's HIV Confidentiality law prohibits health care providers and social service providers from disclosing your HIV status without your written consent.”

There is another point of view, rooted in the history of the social network field. The inventor of “sociometry,” Jacob Moreno, insisted that sociometric data were not valid unless the subjects knew that their answers would have consequences. That was the principle behind his and Jennings famous study of the cottage system of delinquent women teenagers. One of the first applications of network analysis was in the early 30's when “sociometry” showed how artificially to construct primary groups of adolescent young women who were incarcerated in an institution that housed them in separate cottages. The aim was to place women who liked one another into the same cottage rather than having cottages populated by warring cliques (Moreno, 1953). When grouped into cottages with greater group cohesion, the women were less likely to be cantankerous. Moreno successfully constructed primary groups or natural cliques or what were described above as “strong ties” that substituted for the lack of functioning families in this situation. Since Moreno, the general tendency in the network field has been not to reveal individual identities. The famous “Bank Wiring Room” omitted details that would have revealed the identity of the subjects (Roethlisberger and Dickson, 1939). An immediate practical issue is the reaction of literate “natives” to the revelation of the details of their interrelations.

For academics, the abstract ethical issues are overshadowed by the bureaucratic need to submit all research before it proceeds to a Protection of Human Subjects Institutional Review Board (IRB).⁵

The fundamental issue is who benefits from the network analysis? This is another version of the cost benefit analysis so dear to the hearts of Human Subject Review Committees who often conclude in the case of network research that the hazards to the individual outweigh any benefits to “Science.” Further, many IRB's conclude that any way of identifying individual participants is illegitimate and unethical. When viewed in this way, with privacy as an absolute right, network data collection is harmful regardless of any possible benefits. When “Science” is not involved and IRB's are not consulted, as in the examples of interrogation of suspected terrorists, or in the discovery within an organization who are the “real” leaders, or when individuals stand to gain dates, business contacts or sales advantages from the linking of names, the issues become even more murky. The concern of “legitimate” network researchers is that concern over individual rights to privacy will ultimately torpedo the entire social network field. On the other hand, an IRB, as will be seen, can provide guarantees for respondents and an institutional structure that enforces these guarantees. This situation may actually aid in the willingness of respondents to participate. What follows is a more detailed discussion of these issues, based in part, by the contributions made by network researchers at the Sunbelt, 2003 and by various comments contributed to the Social Network List Serve.