Abstract
A number of quantitative studies of pair programming (the practice of two programmers working together using just one computer) have partially conflicting results. Qualitative studies are needed to explain what is really going on. We support such studies by taking a grounded theory (GT) approach for deriving a coding scheme for the objective conceptual description of specific pair programming sessions independent of a particular research goal. The present article explains why our initial attempts at using GT failed and describes how to avoid these difficulties by a predetermined perspective on the data, concept naming rules, an analysis results metamodel, and pair coding. These practices may be helpful in all GT situations, particularly those involving very rich data such as video data. We illustrate the operation and usefulness of these practices by real examples derived from our coding work and present a few preliminary hypotheses regarding pair programming that have surfaced.