Networks of action in social science research - Routines - Michigan ...

SBE 2020 White Paper Networks of action in social science research 1,2 Brian T. Pentland Eli Broad College of Business Michigan State University E. Lansing, MI 48824-­‐1122 ABSTRACT Social and economic research has traditionally focused on familiar units of analysis, such as individuals, groups and firms. In this white paper, I suggest that networks of action might provide a fruitful addition to our repertoire as a basic unit of analysis in social and economic research. A network of action can be used to represent recognizable, repetitive patterns of interdependent actions, typically carried out by multiple actors. Such patterns are the foundation of economic organization (e.g., organizational routines) and social institutions in general. Two current trends make research on action networks feasible: (a) on-­‐going improvement in our tools for pattern recognition and network analysis; and (b) increased availability of data that results from digitization of economic and social processes. Reversing figure and ground, and focusing on actions, rather than actors, provides an opportunity to address social and economic research from a new perspective. Networks of action as a unit of analysis The social world is filled with patterns of action. Within organizations, patterns of action make up the basic work processes that transform inputs into outputs. In society in general, patterns of action produce and reproduce all kinds of social and economic outcomes. While this seems obvious, social science research generally focuses on the actors (individuals, groups, organizations, etc.) and their properties (attitudes, demographics, network position, etc). Actions are treated as secondary. In this white paper, I suggest that NSF should promote research that makes patterns of action the direct focus of inquiry. This research would involve a number of challenges, such as: • How to gather meaningful data on action patterns? • How to scale up from analyzing a few patterns, over a small span of time, to analyzing ecologies of patterns as they change and evolve over time? This line of research is promising because it matches our growing capabilities in network analysis and pattern recognition to the growing availability of data on events and actions that is the natural by-­‐product of increased digitization in organizations and society in general (e.g., workflow systems, 1 This work is licensed under the Creative Commons Attribution-­‐NoDerivs 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-­‐nd/3.0/ or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA. 2 I want to thank Martha Feldman for her comments on a draft of this document. Many of the ideas expressed here flow from our research and writing on organizational routines.

Networks of action cloud computing, cell networks, etc.). The ability to track patterns as they change over time, and explain those changes, seems particularly promising. This idea builds on the work of Andrew Abbott, who has been advocating the analysis of sequence and event data in sociological research since the early 1990s. With recent progress in the available tools and sources of data, we are poised to make significant advances and apply these methods to a broad range of social phenomena. This line of research offers transformative potential because, in effect, it introduces a new unit of analysis into social science research. Traditional units of analysis include individuals, groups, organizations and so on. This approach focuses on patterns of action as the unit of analysis. In some respects, this is a simple reversal of figure and ground; the scene is the same, but we focus on the actions rather than the actors. But the shift is transformational because it is allows us to study certain kinds of social and economic phenomena directly, rather than studying their outlines or shadows. An Example: Invoice processing To help visualize the concept of an action network, consider the example of invoice processing. In this research, data were collected from a computerized workflow system for processing invoices (Pentland, Haerem and Hillison, 2010). In addition to helping participants perform the work, the workflow system provides a digitized, archival record of the sequence of actions required to process each invoice. We collected data on two routines that are required to process each invoice: a data entry routine and an approval routine. Figure 1 shows these two routines represented as directed graphs. Each node in the graph represents a type of action, while the arcs represent the sequential relationship between actions. These graphs are like first order Markov models, except that the nodes are events rather than states. The data presented here were tested for higher order effects, and these processes are first order. Statistically, the next action depends only on the current action. Invoice EntryInvoice ApprovalFigure 1: Patterns of action as directed graphs (adapted from Pentland, Haerem and Hillison 2010) Modelling the pattern of action as a network provides a rich variety of tools to visualize, analyze and compare the structure of the patterns. For example, Figure 1 shows that the approval routine (on the right) is more densely connected than the entry routine (on the left). This indicates greater variety or complexity in the pattern of action. 2

Networks of action The conceptual shift is from networks of people to networks of action. Even though the images in Figure 1 were generated by traditional social network software, the nodes in Figure 1 are not actors; they are actions. Also, these patterns were not stationary; they were changing over time. This conceptual shift is potentially important because it provides a novel way to approach sociomaterial systems – systems that involve heterogeneous networks of human and non-­‐human actants (e.g., information and communication technologies). Approximately 65 people were involved in carrying out the patterns of action shown in Figure 1, but only about half of the actions were performed by those people. The rest of the actions were performed by the workflow system itself, which can take actions based on rules that have been programmed into it. For example, when all approvals are complete, each invoice is automatically queued for transfer to the accounts payable system. Thus, the patterns of action shown in Figure 1 were generated by a combination of human and non-­‐human actants. Capability to be created/Scientific strategy Many of the core capabilities required for this line of research already exist. These include tools for network analysis developed by sociologists (such as UCINET and SIENNA), and tools for pattern recognition developed by computer scientists. With NSF support, (NSF-­‐0924786, Innovation and Organization Science), we are working on applying and extending these tools to explore path dependence in repetitive patterns of action in the context of organizational routines. To use action patterns as the unit of analysis in a broader range of phenomena, some additional definitional, theoretical and methodological work is needed. Some basic issues include: Identification. How can we define the boundaries on a pattern of action, to create an identifiable unit of analysis? In other words, when can we say that two patterns of action are of the same kind or category? Comparison. How can patterns of action be compared (cross-­‐sectionally and over time)? A fairly extensive set of tools already exists for this, but further definition and codification would be useful. Hierarchical nesting. Actions can usually be reduced into finer-­‐grained sub-­‐actions, resulting in a potential infinite regress in granularity. Thus, reasonable principles need to be established for defining relationships between patterns. Time and space. Patterns of action can be widely distributed in time and space, and they may operate at different time scales, which complicates the problems of identification and comparison mentioned above. Visualization. Many excellent visualizations exist; the network graphs shown in Figure 1 are among the simplest. With this kind of complex phenomenon, visualization can be very useful. Since methods are relatively well developed, the key issue is availability of meaningful data on actions and events. Traditionally, this kind of data has been collected through qualitative fieldwork. Researchers like Stephen Barley and Martha Feldman have conducted influential studies using traditional field research methods. I expect that qualitative field research would be necessary to continued progress in this domain. 3

Networks of action The new opportunity, however, comes from the increased potential of archival data that is generated as a by-­‐product of network-­‐based digital communications and workflows. As with the use of any archival electronic data source, privacy is likely to be a concern. I would think that the ideal research project would contain both qualitative fieldwork (to ensure validity and provide conceptual clarity), and the analysis of archival sources (to foster generality and to detect patterns that individual fieldworkers would not be able to see). Intellectual Context Since the 1950s, Herbert Simon, James March, Sidney Winter, Michael Cohen and others have placed routines (repetitive patterns of action) at the center of economic organization. Over the last several years, Martha Feldman and I have emphasized the practice-­‐based view of routines, which emphasizes that routines can be understood as repetitive patterns of interdependent action carried out by multiple actors. I selected Pentland and Feldman (2007) as recommended as background reading because it explains the network perspective, and makes connections to the role of information and communication technologies. Recent advances in methodology and the availability of data have made it possible to analyze patterns of action based on large samples of data. This work provides a “proof-­‐of-­‐concept” for a broader line of research on patterns of action (e.g., Pentland, Haerem and Hillison 2010). Scholars interested in large scale social network analysis provide another important point of reference. Studies based on Internet traffic and cell phone records have broken new ground in the analysis of social structures. However, it is important to note that their work is still focused on traditional networks of actors, not networks of action. In some respects, action patterns can be seen as related to “memes”, an idea introduced by Richard Dawkins. However, the “meme” is primarily a cognitive concept, connected to individual level psychology; memes spread “from mind to mind.” The focus here is behavioral. Further, patterns of action are often distributed across multiple actors, and those actors may or may not be human. Standing questions Andrew Abbott has shown that many traditional social science questions about objects (e.g., people or organizations) can be restated and analyzed in terms of actions. Here are some examples of questions that might be addressed in this research program. They are phrased here in terms of “organizational routines,” since that has been my own disciplinary interest, but similar phenomena could be investigated in any context where patterns of action occur. 1) Stability and change. • How do patterns of action react to environmental change? In organization theory, routines are famous for being resistant to change (a source of inertia). At the same time, routines can also drift and change over time. • Can change within a low level routine propagate to higher level routines? To the extent that routines exist in nested hierarchies, it seems important to explore how hierarchical structures enable or constrain change over time. 2) Diffusion of innovation (and other phenomena involving propagation). • How does change propagate among routines? In social networks, complex contagion is generally stronger than simple contagion (peer pressure is stronger when it comes from 4

Networks of action multiple peers). But do the same kinds of effects occur in networks of routines? Remember that in a routine, different actions may be carried out by different actors; there is not necessarily a one-­‐to-­‐one mapping. 3) How can we account for the role of artifacts? • How does the mix of human and non-­‐human actors affect patterns of action? In real routines, many actors are not human (consider workflow systems, for example). The properties of non-­‐human actors may influence variation, selection and retention of the overall system. The role of artifacts is important because social and economic phenomena are mediated by technology. Recent work by Stephen Barley, Wanda Orlikowski, and others has emphasized the role of sociomateriality. Orlikowski (2007) is recommended background reading because it makes a forceful and eloquent case for the inclusion of material artifacts in social science research. For years, social scientists have struggled with ways to incorporate technology (especially information and communication technology), with limited success. The approach advocated here provides a common denominator – patterns of action – that can be used to investigate sociomaterial systems regardless of the mix of actants. Range of disciplines that could contribute There are several disciplines that could fruitfully contribute to this line of research, including: • Computer science (pattern recognition and process mining); • Sociology (network analysis and qualitative field studies); • Organization theory (repetitive patterns of action are definitional of the phenomena we recognize as “organization”). Application areas include healthcare, public health, energy and environment, technological innovation, and so on. Implications for future research within and across disciplines It is difficult to say what the implications this research might have. However, the transformative potential is considerable because we are talking about a new unit of analysis in social science research. Analysis of action patterns would provide a new approach to some truly foundational questions in social theory, as well as potential insights into practical problems that hinge on our ability to detect and influence patterns of action in society. Three background readings Orlikowski, W. J. (2007). “Sociomaterial practices: Exploring technology at work,” Organization Studies, 28, 1435-­‐48. Pentland, B. T. and Feldman, M. S. (2007) “Narrative Networks: Patterns of technology and organization,” Organization Science, 18(5), 781-­‐795. Pentland , B, T. Haerem and D. Hillison (2010) “Comparing Organizational Routines as Recurrent Patterns of Action”, Organization Studies, 31(7), 917-940. 5