Twenty-eight studies (n=1,839) were included in the review.
In terms of quality, of the 28 included studies, 12 were randomised trials, 21 were controlled, and 16 had a sample size over 40.
The overall mean weighted effect size (d) of any therapy was 0.72 (standard error 0.02; range: -0.20, 2.08, p<0.001). There was significant heterogeneity between the studies (p<0.001). No significant effects of study quality, coder or coding time on overall effect size was found.
There was a significant difference in the mean weighted effect size between treatment groups (d=0.74) and no treatment groups (d=0.46; p<0.001). Significant heterogeneity (p<0.001) was found for both treatment and no treatment groups.
There were moderate mean effect sizes for social functioning (d=0.48) and large effects for behaviour, psychological distress, self-concept and 'other' outcomes (including academic functioning and risk assessment) (d=1.60, d=1.05, d=0.71 and d=1.49, respectively). For each of the outcome measures, significant heterogeneity remained (p<0.01 to 0.001).
Effect sizes (d) above 0.8 were achieved by cognitive-behavioural therapy, play therapy, supportive therapy, group therapy and abuse-specific therapy.
Effect sizes over 1 were reported for behaviour outcomes (supportive, group and abuse-specific therapies), psychological distress outcomes (cognitive-behavioural therapy, abuse-specific therapy, individual therapy and family therapy), self-concept outcomes (group therapy and no treatment) and other outcomes (play therapy, supportive therapy, group therapy and abuse-specific therapy). For social functioning, the largest effect size was achieved by play therapy (d=0.72).
No significant effects were found for age or gender of the participants or therapist training. Significant moderator effects were reported for ethnicity (effect sizes increased with an increasing proportion of non-Caucasians; p<0.001), intrafamilial abuse (effect sizes smaller with a greater proportion of intrafamilial abuse; p<0.002), and numbers of therapy sessions and duration of treatment (greater effects were seen with more sessions and longer duration; p<0.001).
No significant funnel plot asymmetry was found. The number of non significant studies needed to reduce the overall effect size to below statistical significance was 22.