Background. Quality improvement collaboratives are often labeled as black boxes because effect studies usually do not describe exactly how the results were obtained. In this article we propose a way of opening such a black box, by taking up a dynamic perspective based on Actor-Network Theory. We thereby analyze how the problematisation process and the measurement practices are constructed. Findings from this analysis may have consequences for future evaluation studies of collaboratives. Methods. In an ethnographic design we probed two projects within a larger quality improvement collaborative on long term mental health care and care for the intellectually disabled. Ethnographic observations were made at nine national conferences. Furthermore we conducted six case studies involving participating teams. Additionally, we interviewed the two program leaders of the overall projects. Results. In one project the problematisation seemed to undergo a shift of focus away from the one suggested by the project leaders. In the other we observed multiple roles of the measurement instrument used. The instrument did not only measure effects of the improvement actions but also changed these actions and affected the actors involved. Conclusions. Effectiveness statistics ideally should be complemented with an analysis of the construction of the collaborative and the improvement practices. Effect studies of collaboratives could benefit from a mixed methods research design that combines quantitative and qualitative methods.