Reflective or formative?
The silent choice behind every diagnostic model — and why, for treatment, it makes all the difference.
Suppose: a patient scores high on various symptoms — sleep problems, fatigue, low mood, loss of interest, rumination. We perform a factor analysis on the scales, and a common factor emerges, which we call depression. It accounts for the coherence between the symptoms.
What, then, is depression actually? Two fundamentally different answers are possible — and they lead to very different treatments.
Answer 1: The reflective model
In the reflective model, depression is an underlying entity — an illness, a state, a disposition — that causes the symptoms. The symptoms are reflections or indicators of the underlying state. Just as fever, headache and muscle pain are together reflections of an underlying flu.
The implication for treatment is clear: interventions on separate symptoms are palliative, not curative. A paracetamol fights the headache but not the flu. True improvement comes from intervention on the underlying cause.
Schematically: one box in the centre (depression), with arrows pointing outward (sleep problems, fatigue, low mood, et cetera). The common cause.
Answer 2: The formative model
In the formative model, depression is not an underlying entity, but an index. The symptoms interact and amplify one another, and what we call depression is the summary of that mutually amplifying constellation. The factor explains no causal working — it describes the configuration.
The implication for treatment is quite different: interventions on separate symptoms may well change the whole system. For if symptoms amplify one another, a breakthrough in one may change the whole constellation. Restoring sleep may break the spiral — not because sleep was "the cause", but because sleep is a node in the network.
Schematically: no centre, but a network of symptoms pointing at each other with arrows, with a box over the whole that we call "depression" — more as label than as causal entity.
Statistically indistinguishable
Here lies a dilemma. On the basis of correlational data — a questionnaire, one measurement, a factor analysis — these two models are statistically equivalent. Both fit the data equally well. The choice between them is theoretical, not empirical.
"Underlying common causes are unsatisfactory if they cannot be identified independently of the observed relations they are supposed to explain." — after Van der Maas et al. (2006)
This citation is the core of Denny Borsboom's argument against the reflective model in psychopathology. In medicine the reflective model works because we can establish underlying entities independently. Flu can be shown by a test, apart from the symptoms. A heart valve can be seen on an echocardiogram. The underlying cause has its own ontological existence.
In psychiatry that is rarely possible. We have no depression test that is independent of the symptoms. We have no scan that tells us: yes, here is a depression. The "underlying factor" is inferred from precisely the symptoms that it is supposed to explain. That is, strictly speaking, no longer an explanation — it is a renaming.
What this means in practice
Suppose we take the network model seriously. What then changes?
In diagnostics: we look not only at the "severity score" on a total scale, but at the configuration of symptoms. Which symptoms are most central in this patient? Which symptoms feed which others? Which symptoms appear, surprisingly, to be absent?
In the case-conceptualisation conversation: we draw no "diagnosis" but a network. With the patient present. It quickly becomes an instrument by which she recognises her own pattern — and identifies the points of intervention for change.
In treatment: we need not first cure "the cause" before we are allowed to do anything with the symptoms. Working on a central symptom — often sleep, or activation, or social contact — is no surrogate treatment, it is a legitimate way to disturb a network. It does not make evidence-based interventions superfluous; it reinterprets them as interventions on the network.
In the conversation with the patient: the reflective model carries a moral charge that the formative model lacks. One who says "you have depression" implicitly says that something is wrong with you, a thing-in-you that requires treatment. One who says "your sleep-mood-activation-rumination amplify one another and now sustain themselves" says something more technical, and possibly something more liberating. It is not that you are depressed; you are caught in a pattern that feeds itself.
An honest caveat
The formative model is not automatically better. For some disorders — for instance classical psychotic disorders with clear neurobiological correlates — the reflective model still offers much. And even within depression or anxiety there may be parts that do have an underlying cause (a trauma, a hormonal dysregulation, a social situation structurally amiss).
What network thinking offers is no replacement but a complement. It is a second language, alongside the first — and for much of what we encounter in mental health care it seems the more useful of the two.
One who delves into it discovers quickly that there are by now solid technical tools: psychometric network analysis (qgraph, mgm, bootnet in R), longitudinal networks for within-person dynamics (mlVAR, lmerlite), Ising models for binary symptom data. The methodology follows the theory, and in some respects now runs ahead of it.
The silent choice between reflective and formative is now being stated aloud. That is a good development. For every choice we make aloud we can better justify — and, where needed, revise.