Statistics from Altmetric.com
Systematic review with meta-analysis.
The reviewers searched PsychInfo, ERIC, Medline and Dissertation Abstracts. The Journal of Abnormal Child Psychology, Journal of Clinical Child Psychology, Journal of Consulting and Clinical Psychology, Journal of the Academy of Child and Adolescent Psychiatry, and reference lists of 2 previous reviews were hand searched.
Clinical studies presenting a quantifiable measure of the association between therapeutic relationship variables and treatment outcomes in children or adolescents were eligible. Studies had to include at least 5 participants, and be presented in English. Treatments meeting inclusion criteria were parent management training, individual therapy and family therapy. Studies examining the association of relationship variables with session quality or impact were excluded.
Data on study design, treatment characteristics and participants’ age, gender and diagnosis, were extracted by two independent reviewers and coded for methodological quality. Most studies reported results as correlations. The product-moment correlation coefficient r was used to estimate effect size. Results were converted to Fisher’s Z equivalents and weighted in proportion to the number of participants in the study. These results were re-converted to r to aid comparison.
23 studies met the inclusion criteria. Median sample size was 47, median length of treatment was 19 sessions. The product-moment correlation coefficient r for the association between therapeutic relationship variables and treatment outcomes was 0.24, giving a weighted effect size of 0.21 (a modest effect). This is similar to the effect size for the equivalent association in adult populations. This effect was not altered by the age of participants, the type of therapy (behavioural versus non-behavioural, or individual versus family versus parent), but was more marked among children and adolescents with externalising than internalising disorders (p<0.05).
The therapeutic relationship has a fairly consistent, small effect as a non-specific contributing factor to the ultimate outcome of child and adolescent psychotherapy across age groups and type of therapy. However, the studies reviewed did not compare the size of this effect with the contribution of specific items of therapy, such as completion of therapy tasks, so the relative importance of each item is unknown.
The quality of the therapeutic alliance is an important potential mediator variable in psychological therapy. Regrettably, the quality of the therapeutic alliance is sometimes trivialized as a “non-specific” component of treatment, thus devaluing the clinical skill that is required to attain it. As with meta-analyses of studies involving adult patients, the correlation between therapeutic alliance and treatment outcome in the present systematic review was modest, but nevertheless an effect that cannot be ignored. The association was stronger for patients with externalising disorders than it was for patients with internalising disorders. A good therapeutic alliance is often hard to win with a conduct disordered child and his or her family, so one might anticipate a greater variability in scores on measures of therapeutic alliance. This variability in scores leads, in turn, to a greater chance that an association will be found.
Systematic reviews are often more helpful for what they can tell us about improvements that are needed in future research than what they can tell us about existing research. This is true of the present review. Most of the primary studies were uncontrolled, and it is unlikely that the outcome ratings were blind. Only two studies involved observational ratings of the therapeutic interaction, yet this should probably be considered the “gold standard”. There was a lack of consistency across studies in choice of informant. Ideally one would like to see both therapist and patient ratings of the alliance, and blind ratings of the outcome. It is particularly problematic when the therapist is the informant about both the alliance and the outcome, as perceptions about one are likely to contaminate perceptions of the other. Many different rating scales were used to measure the therapeutic alliance. Standardization of research to one or two well-validated instruments is desirable.
Source of funding: Not specified.
For correspondence: S Shirk. Department of Psychology, University of Denver, Denver, Colorado 80208. E-mail:
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.