Background: There is a well-rehearsed debate within and outside the Cochrane Collaboration about the place of non-randomised intervention studies (NRS) in systematic reviews. A spectrum of opinion is reflected in the varying policies of different Cochrane review groups. The debate has focused on the evaluation of complex interventions, often at health care system or population level. In the field of domestic violence research the resistance to randomised controlled trials is also based on a perception that they are not feasible (even for interventions with individual patients) and that they are possibly unethical.

Objectives: To identify the consequences of including or excluding NRS in the area of domestic violence research by comparing results from a systematic review of advocacy interventions to help women experiencing intimate partner abuse.
To disentangle ethical concerns arising from any controlled study with abused women from specific problems with randomisation.

Method: Comparison of conclusions of a systematic review of advocacy interventions for women experiencing domestic violence, including or excluding non-randomised studies. Analysis of correspondence with the Cochrane Developmental, Psychosocial and Learning Problems Group about inclusion of NRS. Articulating the ethical implications of different study designs.

Results: Nine studies fulfilled the wider review inclusion criteria: four RCTs, two parallel group studies, and three `before and after' studies. NRS were conducted in similar settings to RCTs (healthcare, community, legal services), and variations in effect size (when this could be calculated) were independent of study design. Within the field of domestic violence research the resistance to randomised studies seems to be based on conflation of the implications of randomisation with the ethical challenge of `no treatment' groups. The latter needs to be considered for any controlled study. The implications of randomisation include restriction of studies to settings that allow cluster randomisation as well as diminished external validity, which strengthens the case for including high quality NRS.

Conclusions: Randomised controlled trials are feasible and ethical for testing interventions for domestic violence. However, inclusion of high quality NRS can be justified for reviews of interventions to improve outcomes for women experiencing domestic violence.