In more complicated biological problems, it often becomes difficult to express all of the selective forces in terms of relative variances among groups. The problem is that patterns of interaction may differ with respect to different processes, such as mating, competition between certain individuals such as males, and competition between other individuals such as females. In that sort of realistic scenario, it is far easier to trace pathways of causation through a series of partial correlations that can be interpreted as an extended form of kin selection analysis (Frank, 1986, 1998). In practice, it is rarely sensible to express such multiple pathways of causation by expressions of relative amounts of variance among groups, although such expressions may be possible mathematically. For that reason, kin selection often becomes a more natural form of analysis for realistic biological problems, leading to a generalized path analysis framework.
I think this is part of it. The idea of a "trait group" represents an attempt to overcome this issue. However there are other reasons too:
- Kin selection is more quantification-friendly - with its r, b and c;
- In practice, use of group selection often seems to lead to naive group selection;
- Group selection avoids problems where kin selection is obviously responsible;