We thank Dr MacFarlane for his favourable comments on our views. The development of research designs that can rigorously test the complexities of mental health intervention and also have face validity to clinicians is at the centre of our concern. In a brief editorial we could do no more than whet the readers' appetites. There was no mention of structural equation modelling because of lack of space, and not because we do not have sympathies with the technique. In fact, one of us (G.D.) has taught structural equation modelling for nearly 20 years. ^{Reference Dunn, Everitt and Pickles1} When used wisely and with correctly specified models, structural equation modelling approaches can be very powerful – but they do not obviate the need for good design (including the randomisation in an RCT). In particular, MacFarlane is mistaken when he suggests that the use of structural equation modelling (MIMIC) models can successfully address issues of hidden confounding in the absence of appropriate design. Although enthusiasts in the social and behavioural sciences have used structural equation models and ‘causal models’ interchangeably for many years, their naïvety has frequently brought structural equation modelling into disrepute. Pearl's book covers structural modelling in the appropriate way, but many readers of this journal will find it a bit heavy going. We do indeed plan to publish on these issues in much greater detail in the near future.