Phase 1: Collection of factors

The goal of this phase is an exhaustive, but structured set of factors which influence diffusions [the process of spreading ideas, concepts, technology, or products.]. I am using three sources of data: informal, ad-hoc interviews with Debian contributors, project forums and mailing lists, and the literature. If viable, I will consider the use of axial coding [SC1998] to normalise and structure the factors.

Phase 2: Delphi study

The set of factors from phase 1 will be inevitably biased. To eliminate this bias, I plan a Delphi study with around 30 participants. To guard against influencing the participants right at the start, I will only use these factors in the Delphi study to fuel the discussion, should this be necessary.

A Delphi study is an instance of moderated communication: as moderator, I serve a series of questions to the participants, who are also known as experts or panelists, and who return their answers to me. I then anonymise, summarise, and structure the responses and make them available to the group, along with a new set of questions, and the second round beings. The questions may be either unchanged (in which case I would be looking for convergence among the panelists on an answer), or modified (if I seek more broad input). The process continues for as many iterations as it takes for the group to converge on an answer (which may well include disagreements). Usually, Delphi studies consist of two or three iterations.

The approach is based on the theory that crowds have more wisdom than the individual because, given enough diversity in the members of the crowd, their averaged statements converge closer to the "right answer" than an individual's response [Sur2004]. To ensure diversity, I do not limit the Delphi panel to Debian contributors, but will also consider developers from other distributions and other complex FLOSS projects, as well as social science researchers familiar with the Debian Project, and possibly scholars of management science. Nevertheless, I need to select participants on a clear rationale or convergence may not be possible at all.

I am considering starting the study with questions yielding quite personal, subjective responses in the first round, e.g. "what are your experiences?" or "what would you do to…?". In subsequent rounds, I would then ask similar questions, but in the context of "the stereotypical Debian contributor." My theory is that this leads people to approach the second set of questions (whose answers are the ones I seek) with the chance of having first considered one's own opinion, and the opportunity of reading each other's thoughts thereafter.

As part of the selection process, I might present each potential participant with a small questionaire to help me see their opinions in the right light.

I don't see the outcome of this study to be an agreement on a set of factors (the intersection of everyone's opinion), but rather as a cumulation of factors (the union of...). Obviously, I expect some of these to stick out more than others, and these nuances are what I look for in the Delphi discussion.

Phase 3: Community survey

The goal of this phase is to get an idea of the community members' perception of the relevance of each of the factors. I would thus like to administer a community survey using a novel survey strategy I previously introduced in my blog. I have to admit that the community survey does not really blend in nicely with the rest of the approach, and my main motivation for it is to try the survey strategy I devices. I may end up administering this survey separate from this research.

As part of this survey, I also intend to probe participants about their individual adoption behaviour and their perceptions of specific diffusions. The resulting data are used during the fourth phase of the research.

A possible approach to combining a community survey with the set of factors from the first phase is conjoint analysis [Baj2000].

Phase 4: Application and verification

In the final phase of my research, I seek to test the framework developed thus far. Even though a prescriptive application would be ideal, I leave that for further research as it is outside the scope of my current research. Instead, I intend to descriptively investigate the diffusion of two classes of methods in the Debian Project: package build helpers (debhelper, cdbs, and yada), and patch management and version control systems for package maintenance (dpatch, quilt, SVN, and Git).