Since the submission of my essay on the coherence theory of truth as a basis for developing a ‘softer’ and less ambitious realism model for psychiatric illness (Kendler, Reference Kendler2014), I have read in more detail about theories of truth and come to agree with key points made by Eronen in his thoughtful and helpful letter (Eronen, Reference Eronen2015). I was, in retrospect, over-ambitious in wading into the complex metaphysical issues that arise in the consideration of the nature of truth. This is a hazard for those of us who try to contribute to the psychiatry–philosophy interface without long training in the subtleties of the relevant analytic philosophical tradition – in this case, the nature of truth. So this response gives me an opportunity to scale-back some of my initial claims.

The essential problem with my initial claim – which is termed the ‘Specification Objection’ to the coherence theory of truth in Young's useful and accessible review (Young, Reference Young, Zalta, Nodelman and Anderson2015) – is the problem of regress. One major advantage of a coherentist theory of truth for psychiatry is that it provides a conceptual rubric for what we have done in our nosology for decades – that is the systematic use of validators to help decide whether to change our diagnostic criteria (Robins & Guze, Reference Robins and Guze1970). I wrote:

The problem is what I meant by our ‘knowledge base’. In common sense terms, I was referring to the kinds of studies that populate the psychiatric literature showing that a given diagnostic formulation is associated with a particular outcome, imaging findings, treatment response or set of molecular genetic markers. I wanted to claim that the validity of a given psychiatric disorder could be assessed by examining the degree to which it cohered with what else we knew about the syndrome – the more coherence the more validity. I did not, however, stop to consider how we then knew that the results of those validating studies were ‘true’. If I advocated a coherentist theory of truth for disorders, shouldn't I assume – for consistency – the same approach to truth for the entire psychiatric literature? But, of course, that will get us into trouble. If a good double blind study shows that syndrome X responds well to drug Y, or syndrome A has a particular set of genetic markers which distinguish it from controls, the natural approach of all biomedical researchers would be to say that these results might be true because they reflect actual processes in the external world (that is the correspondence theory) rather than because they are consistent with prior things we believe (the coherence theory). So, as pointed out by Eronen, my proposal, taken literally, to adopt a coherence theory of truth – tout court – was unsustainable.

So I am left with a somewhat odd but I think defensible position which might be termed a ‘restricted’ coherentist view that applies specifically to psychiatric nosology – wherein we are trying to evaluate diagnostic criteria. The ‘truth’ of a new set of criteria for major depression, we could argue, can be best understood as the degree to which it coheres with other things we know from our relevant sciences – that is our ‘knowledge base’ (epidemiology, drug trials, imaging studies, molecular genetics). Those studies, however, we interpret in a more traditional correspondence theory of truth.

This approach will take care of Eronen's concerns that the study of psychopathology would become disconnected from the rest of science if we widely adopted the coherence theory of truth. I am less concerned about his argument about changing truth with historical time. The concerns of nosologists are largely in the present – how well does criteria set A compare to criteria set B? What happened 50 years ago is not of major concern.

There is not here adequate space to discuss his idea of robustness. It is for me, one primary criterion for causal inference and indeed for the value of any scientific finding. I suspect it overlaps substantially with the concept of diagnostic validity as I and others use it in nosology. For example, the guidelines we adopted for the DSM-5 scientific review committee required for major changes support from a range of different validator studies by different research groups (Kendler, Reference Kendler2013). This is pretty close, I think, to Eronen's idea of robustness.

Is it too odd to postulate a different idea of truth for our nosology than for the rest of psychiatric science? While I am not sure, I think it merits consideration. Through over 30 years of working on the DSMs, it has become quite clear that nosology is not ‘science’ as we narrowly understand it. It is a hybrid between science, policy and values. A system of validators cannot in and of itself decide a nosologic question because we need to decide on a hierarchy. What do we want our diagnoses to do? (Kendler, Reference Kendler1990). There are never definitive experiments that decide nosologic issues. So the idea that our guiding principles (here instantiated as theories of truth) would differ in nosology and traditional psychiatric science is not an outlandish claim.

In closing, I want to re-affirm my commitment to a program of ‘graduated’ realism for psychiatric disorders – what Eronen calls a ‘middle way’ – trying to find a comfortable and defensible place in-between the broad claims of realism (which, especially in its physics/chemistry form, is far too ambitious for us) and instrumentalism which, although of relevance, is ultimately inaccurate for our disorders and unfair both to our discipline and to our patients. I have recently outlined my position on these issues at some length (Kendler, Reference Kendler2016).