and the investigation of relational factors. In meta-analyses some variation will always exist, whether due to chance or to differences in trial context or content [5,6]. On its own, variance in a meta-analysis does not de-legitimize that meta-analysis, nor does it necessarily mask key causes of variation. Indeed, once these relational factors are specified , measured and tested in the same way in which specific factors are currently dealt with, meta-analyses of the trials of these new tests would still be subject to variation in outcome due to remaining unspecified factors. Miller & Moyers are right to point out that, in focusing exclusively on specific factors related to treatment content, research into addiction treatments may be overlooking important relational factors and their associated effects. However , in acknowledging genuine and measurable causes of variation-more trees, arguably-it is important that a true forest view is not obscured. Using existing methodology , findings can be aggregated without masking the impact of underlying mechanisms, as long as these potential mechanisms are identified in advance. Relational factors that can be measured empirically and in which therapists can be trained can be tested in randomized controlled trials, as has recently been done with empathy in the context of physician training [7,8]. Such trials could then be aggregated in meta-analyses. Even where not tested directly , the contribution of these relational factors could be examined in systematic reviews through meta-regression, as has already been performed with specific factors [9,10].