Structural equation modelling in developmental psychiatry

The paper Green & Dunn ¹ may prove to be of merit in the interpretation of causal relationships between interventions and outcomes. In particular, the recommendation that randomised controlled trial (RCT) methodology should be embedded within statistical methods from observation studies is long overdue. Such an approach would greatly assist in the interpretation of results which seem completely counterintuitive to those in everyday clinical practice. One such result is the finding of Byford et al ² that cognitive–behavioural therapy provides no added or separate advantage to selective serotonin reuptake inhibitors in the treatment of adolescent depression.

I have a quibble with the length of time it has taken for basic concepts on causality introduced by Green & Dunn to appear in psychiatric research. These concepts have been commonplace in social science research for more than 20 years and their section on causal inference in analysis is little more than a primer. For a more complete coverage of principles of causality, I can recommend Judea Pearl's book, Causality: Models, Reasoning and Inference. ³

Is there any particular reason why Green & Dunn, having put their toes in the water by introducing basic concepts on causality, have not taken their paper further or are we to await a follow-up? In particular, why is there no mention of structural equation modelling, otherwise known as covariance structure analysis? Structural equation modelling has been extensively used in social science research for the past 20 years and adaptations of the method such as multiple-indicator, multiple-cause (MIMIC) seem to address the issues on confounding variables adequately without the need to revert to RCT methodology. It would be interesting to hear from Green & Dunn their thoughts as to how necessary would RCT methods be in developmental psychiatry research whenever a structural equation model is being employed.