by Dr. Judith A. Effken, Senior Editor and
Effken , J. & Benham-Hutchins, M. (February, 2011). Technology-Enhanced Social Network Analysis: An Old Idea Whose Time Has Come—Again. Issues, Impacts and Insights Column. Online Journal of Nursing Informatics (OJNI),15 (1). Available at http://ojni.org/issues/?p=326
Mention social networks today and we immediately think of Facebook, Twitter or Linked In. These modern tools allow people to connect and network. However, social networks and the analysis of social networks were around long before these technologically enhanced methods of connecting individuals.
Remember sociograms? If not, one is shown in the figure below. Sociograms depict the connections or interactions between members of groups (networks!) pictorially. For example, friendship networks show individuals such as Sally, Ted, and Jack and who in their social network are friends with whom. Conceived by sociologists, sociograms and their analysis have recently gotten a facelift of sorts—and it’s all due to improved technology.
A social network, as shown in a sociogram, is comprised of a set of actors (individuals in the network) and the relations (ties or links) between them, which may be more or less permanent (i.e., some links are very short-lived; others, such as those in a family, may last a lifetime). Social network analysis (SNA) emphasizes the structural properties of the network, which are determined by the relationship between the individuals (relations), rather than by individual characteristics (Wasserman & Faust, 1994) The links between people can be valued in terms of some characteristic (e.g., strength of friendship, frequency of communication, or quality of communication). Relations may or may not be reciprocated (for example, I may like you, but you may not like me!); and actors can be connected to more than one other actor or group. Researchers define the structure of the network based on the pattern of relationships among the individuals that make up the network (Wasserman & Faust, 1994). For example, a network may be highly centralized (i.e., the network acts as if it were a a single group) or decentralized into several small groups with unique patterns of behavior (Rivera, Soderstrom, & Uzzi, 2010).
Sociograms provided static pictures of the communication, friendship, or collaboration among a relatively small group of people. They could be hand-drawn or software might be used to create the nodes and links of the diagram. Statistical analysis of the networks was complex and not incorporated into the software used to draw the sociograms. Just as SPSS and SAS put statistical analysis within the reach of more researchers, computer applications that incorporate methodologically sound network analysis metrics have done the same for network analysis. Researchers are no longer limited to the few measures of network characteristics that could be carried out manually. Instead, they have an extensive arsenal of statistical measures at their fingertips that more completely characterize the network. SNA used to be limited to relatively small groups and usually a single network. However, today’s tools are nothing less than sociograms on steroids! These new dynamic network analysis tools can be used to examine organizations and other large groups–even those comprised of multiple networks.
Interest in SNA is growing proportionately as availability of network analysis software increases. The number of papers that list “networks” as a keyword increased from 1.2% in 1980 to 11.6% in 2005 (Rivera, Soderstrom, & Uzzi, 2010). Interest in SNA is also growing among healthcare researchers (O’Malley & Marsden, 2008). SNA has been used to examine the communication structure in private practices (Scott & Tallia, 2005), emergency departments (Creswick, Westbrook & Braithwaite, 2009), and neonatal ICU teams (Gray, Davis, Pursley, Smallcomb, Geva, & Chawla, 2010); and to describe medication advice-seeking in a renal unit (Creswick & Westbrook, 2010). SNA has also been used to explore the impact of information technology on healthcare organizations and teams (Aydin & Rice, 1990; Anderson & Jay, 1985; Anderson & Aydin, 1997; Anderson, 2005; Burkhardt, 1994); and to study how mutual understanding developed in multidisciplinary primary health care teams (Quinlan & Robertson, 2010. Some researchers have examined intrinsic (internal) constraints that contribute to network structure, such as competition (Burt, 1992), power, and influence (Ibarra & Andrews, 1993); others have focused on extrinsic (external) constraints that alter network structure, such as organizational change and workforce reductions (Feeley & Barnett, 1996; Kwon, Oh & Jeon, 2007).
ORA (Organization Risk Analyzer), which was developed by Carley and colleagues at Carnegie Mellon University, is one of those dynamic network analysis tools. ORA allows researchers to model communication networks, visualize them while highlighting particular features, then analyze them using more than 80 different metrics. ORA is not restricted to modeling a single network, but can integrate multiple networks into a single “metamatrix” structure or merge two or more networks into a virtual network that represents the overlap in the original networks (much like overlapping Venn diagrams where the overlapping section is what is being examined).
Given its original name (Organization Risk Analyzer), it is not surprising that ORA has been used heavily by the military, law enforcement and intelligence agencies to understand criminal networks and identify interventions (Hutchins & Benham-Hutchins, 2010). It can help these professionals understand who in these networks has what knowledge and who has or shares resources. Data from web based social networking sites, web usage patterns and even phone records can be imported into ORA to identify communication patterns and commonalities. Metrics are available to identify key individuals; and “what if” analysis allows examination of the influence of removing specific individuals would have on the network.
Nursing is full of communication issues that ORA might help us understand better. While law enforcement may be looking for ways to break up efficient communication patterns nursing is interested in how to strengthen this type of tie. For example, let’s say that we’re interested in communication around change of shift handoffs because it has been implicated in errors. We collect data about the frequency with which staff on the day shift “give” information related to patient care to the oncoming shift. We also collect data about the frequency with which staff on the night shift “get” information related to patient care from the oncoming shift. Now we have two networks—one for the day shift and one for the night shift. Using ORA, we can merge the networks in such a way that we have a new network which is comprised of just the common connections between the two networks. In fact, Effken and colleagues did just such an experiment (Effken, Gephart, Bianchi & Verran, 2010).
The beauty of ORA lies in its visualizations and graphics depicting the networks, which are not only visually appealing, but also highly informative. Node shape, size, and color can be mapped onto various attributes (e.g., education or centrality) and link color or size can be mapped onto measures of communication frequency or quality, for example. From a good visualization, the observer can begin to understand communication patterns, including who on the nursing unit may influence the flow of information (e.g., act as a gatekeeper) or who has more or authority due to their coordination role in the network.
The intuitions generated by the visualizations can be validated through metrics that specifically identify who the key players are in a particular network. Effken et al. (2010) found that on the nursing unit, the key players were not always nurses; nearly as often they were PCTs or Unit Clerks. We’ve all heard about the importance of the informal network; and these results confirmed it. Benham-Hutchins and Effken (2010) obtained a similar result when examining the communication networks that emerged during a patient handoff from the Emergency Department to a Medical-Surgical unit. The key roles (authority, etc.) varied across handoffs, even though the handoffs were in the same hospital. In this interdisciplinary study, the person with the most information was often the pharmacist, rather than a physician or nurse.
Despite the beauty of the visualizations, the real power of ORA lies in the metrics that describe the overall network or particular nodes (individuals). As noted earlier, initially social network analysis was a rather low-technology technique, with only a few metrics available. ORA changes that. It allows the user to analyze not only density, but also the speed with which information diffuses across the network, the strength of the various links between nodes (connections between people); the individuals who are most central to the communication on the unit, who is connected to other highly connected people, the extent of hierarchical (top-down) communication, etc. Merrill used ORA’s metrics to understand the communication of a public health organization and how it might be affected by an upcoming merger (Merrill, Caldwell, Rockoff, Gebbie, Carley, & Bakken, 2008). The results were shared with managers to help them successfully manage the change. Once you have network metrics, then new possibilities emerge. Suppose you’re interested in whether differences in the communication patterns on different nursing units, different shifts, or different hospitals are associated with different safety and quality outcomes. In study of 7 medical/surgical nursing units in 3 hospitals, Effken examined the correlations among a set of 14 network metrics with 7 commonly measured patient outcomes (patient fall rate, ADEs, change in patients’ ability to perform simple or complex self care from admission to discharge; change in patients’ ability to manage their symptoms from admission to discharge, and patient satisfaction) (Effken et al., 2010). Not only were statistically significant correlations found between specific metrics and outcomes, but the patterns of those correlations were different for the different types of outcomes. For example, the communication patterns that were associated with fewer patient falls were different than those associated with fewer ADEs, or improved satisfaction.
Using ORA, researchers can identify high and low performers and then examine the simulated network without those people to see what the effect would be. For example, let’s say that we have a number of staff on a particular nursing unit who are isolates (that is, they don’t appear to communicate with anyone about patient care). With them as part of the network, we find that diffusion of information is fairly slow and negatively associated with increased ADEs. So we remove them (from the data, not their jobs!), and run the simulation again to see whether the relationship changes. If increasing information diffusion by removing the unconnected individuals changes the relationship with ADEs in a positive way, then the Nurse Manager might want to consider how to get those individuals more involved in unit communication.
Several challenges remain in the use of these tools. The first is the need to collect data about communication patterns. Most often this is done through survey instruments. Because surveys are self report tools, they can be subject to biases related to faulty memory, fatigue, etc. Ideally, one would like to be able to observe everyone’s actual communication on the unit; but that is not really practical. Benham- Hutchins and Effken (2010) observed the handoffs from the ED to medical surgical units described earlier; but then gave surveys to the people who were observed participating in the handoffs, as well as to those participants identified who were not actually observed.
Second, we know almost nothing about how valid these snapshot views of networks are over time. Effken et al. (2010) collected data from nursing staff on two separate days. The days were selected intentionally to have the least possible overlap among staff members. Of the seven units in the study, four showed no statistically significant difference across the two days, suggesting that communication patterns were stable—at least over this short period of time—despite different staff. However, three of the units were significantly different across the two days, pointing to possible instability of communication patterns over time. More research is needed to understand these phenomena.
Third, a very high response rate is needed to assure that the survey accurately reflects the unit. Researchers have recommended individual response rates between 40 and 50% if data are to be successfully aggregated to the unit level (Kramer, Schmalenberg, Brewer, Verran, & Keller, 2009). But for social network analysis, we recommend a response rate of over 90% to avoid having significant holes in the network.
Still, despite these limitations, ORA and other similar analysis tools make it possible to study social networks in healthcare in ways that we never dreamed possible. Initial results of several studies begin to suggest the importance of communication networks and the value of these complex analysis tools. Given the importance of communication to achieving quality and safety outcomes, the value of being able to analyze complex healthcare networks and their relationship to patient outcomes cannot be overstated. Ultimately, they may become as ubiquitous in our organizations as spreadsheets.
Dr. Effken’s current research, as summarized above, is funded by the National Library of Medicine (NIH) 1R01LM009516-01A1.
Anderson, J. G. (2005), Evaluation in health informatics: Social Network Analysis, in J. G. Anderson & J. C. Aydin (Eds.), Evaluating the organizational impact of healthcare information systems (2nd ed., pp. 189-220), New York, NY: Springer.
Anderson, J. G. & Aydin, C. E. (1997). Evaluating the impact of health care information systems. International Journal of Technology Assessment in Health Care,13, 380-393.
Anderson, J. G., & Jay, S.J. (1985). Physician utilization of computers: A network analysis of the diffusion process. In L. Frederikcsen & A. Riley (Eds.), Computers, people and productivity (pp. 21-36). New York: Haworth Press.
Aydin, C. E., & Rice, R. E. (1990). Bringing social worlds together: Computers as catalysts for new interactions in health-care organizations. Journal of Health and Social Behavior, 33L,168-185.
Benham-Hutchins, M. M. & Effken, J. A. (2010). The influence of health information technology on multiprofessional communication during a patient handoff. International Journal of Medical Informatics, 79, 252-267.
Burkhardt, M. E. (1994). Social interaction effects following a technological change: a longitudinal investigation, Academy of Management Journal, 37, 869-898.
Burt, R. S. (1992). Structural holes: The social structure of competition, Cambridge, MA: Harvard University Press.
Creswick, N., & Westbrook, J. I. (2010). Social network analysis of medication advice-seeking interactions among staff in an an Australian hospital. International Journal of Medical Informatics, 79, e116-e125.
Creswick, N., Westbrook, J. I., & Braithwaite, J. (2009). Understanding communication networks in the emergency department. BMC Health Services Research, 9, 247. doi:10.1186/1472-6963-9-247 Retrieved from http://www.biomedcentral.com/I472-6963-9-247
Effken, J, Gephart, S., Bianchi, D., & Verran, J. (2010). Correlations of nursing communication network metrics with patient outcomes. In Proceedings AMIA Fall Symposium 2010, 13-17, Washington D. C.
Feeley, T. H., & Barnett, G. A. (1996). Predicting employee turnover from communication networks, Human Communication Research, 23, 370-387.
Gray, J. E., Davis, D. A., Pursley, D. M., Smallcomb, J. E., Geva, A., & Chawla, N. V. (2010). Network analysis of team structure in the neonatal intensive care unit. Pediatrics, 125, e1460-e1467.
Hutchins, C., & Benham-Hutchins, M. (2010). Hiding in plain sight: Criminal network analysis. Computational Mathematical Organizational Theory, 16(1), 90-111. doi: 10.1007/s10588-009-9060-8
Ibarra, H., & Andrews, S. B. (1993). Power, social influence, and sense making: Effects of network centrality and proximity on employee perceptions. Administrative Science Quarterly, 38, 277-303.
Kramer, M., Schmalenberg, C., Brewer, B. B., Verran, J. A., & Keller, J. (2009). Accurate assessment of nurse work environment: Response rate needed. Research in Nursing and Health, 32, 229-240.
Kwon, D., Oh, W., & Jeon, S. (2007). Broken ties: The impact of organizational restructuring on the stability of information-processing networks. Journal of Management Information Systems, 24, 201-231.
Merrill, J., Caldwell, M., Rockoff, M. L, Gebbie, K., Carley, K. M., & Bakken, S. (2008). Findings from an organizational network analysis to support local public health management. Journal of Urban Health: Bulletin New York Academy of Medicine, 85(4), 572-584.
O’Malley, J., & Marsden, P. V. (2008). The analysis of social networks. Health Services Outcomes Research Methods, 8, 222-269.
Quinlan, E., & Robertson, S. (2010). Mutual understanding in multi-disciplinary primary health care teams. Journal of Interprofessional Care, 24, 565-578.
Rivera, M. T., Soderstrom, S. B., & Uzzi, B. (2010). Dynamics of dyads in social networks: Assortative, relational, and proximity mechanisms. Annual Review of Sociology, 36, 91-115.
Scott, J., Tallia, A., Crosson, J. C., Orzano, A. J., Stroebel, C., DiCicco-Bloom, B., ……Crabtree, B. (2005). Social network analysis as an analytic tool for interaction patterns in primary care practices. Annals of Family Medicine, 3, 443-448.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. New York, NY: Cambridge University Press.
Dr. Effken is a Professor in the College of Nursing at the University of Arizona. She earned her BA in Psychology from the University of Hartford. her Masters of Science in Nursing Management and PhD in Psychology from the University of Connecticut. She was awarded the Ada Sue Hinshaw Research Award for Significant Work in Improving Healthcare in 2008 and was elected to be a Fellow in the American Academy of Nursing and in the American College of Medical Informatics in 2005.
Dr. Effken’s research interests focus around Design and evaluation of clinical information displays, Human-computer interaction, the Impact of organizational and unit characteristics on staff and patient outcomes, and E-learning. She has worked on DyNADS: A Dynamic Network Analysis Decision Support Tool for Nurse Managers.