Simple: Only change one variable at a time.

Thanks a lot Lawrence, 

But, does this mean that the aim is, for instance, to enhance the generalizability? I mean I have got to be more specific 

Replication as a positivist concept is not particularly meaningful in an interpretive approach.  Because every situation we study has its own characteristics, and our inquiry is shaped in part by the particulars of the situation, it doesn't make sense to think the study could be "replicated."  What this means for generalizability is addressed nicely in Frederick Erickson's article "Qualitative Methods in Research on Teaching" (including his reference to Hamilton's work on generalization).  In interpretive investigation we are not aiming for statistical generalization (extrapolating from a sample to the population), but for logical generalization, which is done by the reader.   The researcher must provide enough depth and detail of information about the context of her study that a reader in a different setting can look at the lessons drawn, see the connection to the particulars of the study setting, and then think about how that setting is similar to or different from her own setting, to be able to extrapolate lessons (generalize).  When I studied race and gender in the Oakland Fire Department, I didn't argue that all fire departments would have the same kind of experience.  But I tried to give enough context that readers working in different departments could see how the study results were or were not useful in their own settings.  Based on the feedback I've received from fire personnel around the country, they were able to do that, and found the work useful even if there were differences between OFD and their own departments.

Thank you very much Carol, 

very useful contribution. 

Indeed, my concern is that I adopt critical realist view which allows me to adopt interpretive approach but at the same time supports the logic of replication. This mix confuses me!