Wide-Field Ethnography

As I and my research colleagues work with the BeamCoffer dataset, and others similar datasets, we have realized that such datasets are qualitatively different from most other datasets. The amount of data and variety of data in the BeamCoffer dataset is allowing us to do things that would have been difficult before. It is changing how we view our research, and what types of questions we can address.

Given this, we have coined a new term to describe this way of gathering and analyzing such datasets. We call this work wide-field ethnography (WFE).

Ethnography. WFE has its roots in the well proven technique of using video to augment ethnographically-informed research. These techniques have been used for decades to study situated work in a variety of fields, including sociology, anthropology, and software engineering. Gathering videos from these settings allows researchers to analyze the fine-grain details of how humans use their bodies (e.g., speech, gesture, gaze, body orientation), artifacts (e.g., tools, whiteboards, paper), and environment (e.g., office configuration) to do work. Not only does video make these aspects visible in all of their particulars, they can be viewed again and again to identify and document patterns. And videos can be shared with other researchers. These benefits motivated us to use video cameras to gather audio-video streams of data of software professionals collaborating in their normal workplace.

Wide-field. When we gathered the BeamCoffer dataset, however, we did not just use video cameras. Instead of deploying one, two, or perhaps four video cameras, we deployed nine wide-angle GoPro cameras, six high-quality Zoom H2n audio recorders, and screen capture software all running concurrently to gather a much “wider” set of data. Wider in that it covers a more extensive physical area of the place of work. Wider in that each camera has a wide-angle lens, showing more of the context of activity. Wider in that we ran the data recorders for 11 days. This is part of the reason for the ‘wide-field’ part of the wide-field ethnography term.

Physical-cyber-social systems (PCSSs). As we analyze the BeamCoffer dataset we are coming to appreciate the importance of recording the physical, cyber, and social aspects of the systems we are studying. Teams of software developers use physical tools, artifacts, people, and their environment to mediate their work in physical spaces. Many of these tools are digital. And software developers work in teams within a social system whose values, norms, and processes have profound influences on what the software developers do or even can conceive of doing.  All three of these aspects (physical, digital, social) are critical to successful software development.  Furthermore, the products of software development contain, connect to, and require a variety of physical, cyber, and social systems.

In other words, software development is done within a network of physical-cyber-social systems (PCSSs), and software development results in PCSSs (products and services). PCSSs are the engine and product of software development. To study software development, therefore, is to study networks of PCSSs. A software development organization is a PCSS that contains a set of interacting PCSSs. A team is a PCSS that interacts with many other PCSSs. And so on. Much of knowledge work today is done by PCSSs and creates PCSSs. To paraphrase a popular expression “its PCSSs all the way down”.

WFE datasets. Now that we see studying software development as studying PCSSs, we see that one benefit of WFE is that it allows to “see” more of each of these aspects of the PCSS under study. WFE datasets like BeamCoffer help make visible the physical, cyber, and social aspects of the PCSS representing the software development organization. WFE datasets are multi-modal; the BeamCoffer dataset, for instance, includes video, audio, screen capture, time-lapse photography, photos from a hand-held camera, field notes, and interviews. WFE datasets are multi-stream; BeamCoffer contains dozens of recordings (e.g., audio, video) that were streamed into thousands of files. WFE datasets are large; BeamCoffer, for instance, contains six terabytes of data spread across thousands of files.

Analysis tools. WFE datasets also produce problems. The BeamCoffer dataset, for instance, has too much data (thousands of files) for a research to easily understand, navigate, search, filter, and analyze it. Data from different recorders are not synchronized, making it hard to synchronously play a subset of the streams of collected data, such as playing the higher quality audio from the Zoom recorder while viewing the GoPro video. And in the future, we expect to gather much larger datasets, which will exacerbate these issues. So, we have started to create an ecosystem of tools.

Wide-field ethnography. Our conjecture is that WFE is a new thing. That dealing with this amount of different types of unstructured and structured data is qualitatively different from what has been possible before very recently. That WFE is providing a new way of studying PCSSs. That WFE will become increasingly important as sensor technology allows us to gather yet more types of streams and as we build the tools to help deal with the large, multi-modal, multi-stream datasets WFE produces. Our conjecture is that WFE may even change how some types of research is done. That WFE may be a new way of doing research.

For more information, see the following publications:

  • Socha, D., Adams, R., Franznick, K., Roth, W.M., Sullivan, K., Tenenberg, J., Walter, S. (2016). Wide-Field Ethnography: Studying Software Engineering in 2025 and Beyond. Proceedings of the 38th International Conference on Software Engineering (ICSE).
  • Socha, D., Jornet, A., Adams, R.. (2016). Wide Field Ethnography and Exploratory Analysis of Large Ethnographic Datasets. CSCW 2016 Workshop: Developing a Research Agenda for Human-Centered Data Science.