One symptom, two broken computations: why effort fails differently in psychosis and mood disorders
- In 920 participants spanning schizophrenia, first-episode psychosis, bipolar disorder, depression, clinical high-risk states, and healthy controls, computational modeling of an effort-for-reward task revealed that negative symptoms do not arise from one shared mechanism but from distinct, diagnosis-linked failures in how the brain converts reward into action.
- People with schizophrenia and first-episode psychosis were best described by a "bias" model: they largely ignored reward magnitude and probability when deciding whether to exert effort, choosing as if the payoff barely registered.
- Those with clinical high-risk states, depression, other clinical presentations, and healthy controls were best described by a "full subjective value" model, using both reward size and the odds of receiving it; the bipolar group was mixed.
- Regardless of diagnosis, participants who fit the bias model carried the heaviest burden of negative symptoms and the deepest cognitive impairment, which means the computation a person uses predicted their clinical state better than their diagnostic label.
Negative symptoms – blunted motivation, diminished pleasure-seeking, social withdrawal, poverty of speech – are among the most treatment-resistant features we encounter, and they cut across diagnoses. The intuitive assumption has been that the same broken motivational machinery sits underneath them wherever they appear. This study, led from the University of Georgia with collaborators at Emory, Yale, Maryland and Northwestern, tests that assumption directly and finds it wanting.
What the data shows
The team gave the Effort Expenditure for Rewards task to 920 participants and then did something the field rarely does at this scale: instead of comparing raw choice counts, they fit competing computational models to each person's decisions. Three accounts were pitted against each other. A full subjective-value model assumes the person weighs both how large a reward is and how likely they are to get it. A partial model uses only one of those. A bias model assumes the person is not really tracking reward information at all, defaulting to a fixed tendency to work hard or not.
The winning model differed by group. Schizophrenia and first-episode psychosis were best captured by the bias model: their effort choices were largely decoupled from what was on offer. The clinical high-risk, depression, mixed-clinical and control groups were best captured by the full subjective-value model, integrating magnitude and probability as one would expect of intact cost-benefit reasoning. Bipolar participants were heterogeneous, fitting no single account cleanly.
The decisive result concerns equifinality – the idea that different routes converge on the same endpoint. It was not supported. Across every group, the people whose behavior fit the bias model had the most severe negative symptoms and the greatest cognitive deficits. The mechanism, not the diagnostic category, tracked the clinical severity.
Why the computation matters more than the label
For clinicians this reframes negative symptoms as a problem of information use, not simply of drive. A patient with schizophrenia who shows avolition may not have a depleted reward system so much as a system that has stopped reading the value of available rewards. That is a different therapeutic target. Behavioral activation, contingency management, and skills work all implicitly assume the patient can register and act on reward information; if that registration is what has failed, interventions may need to make value explicit and external rather than relying on the patient to compute it internally.
The transdiagnostic finding also cautions against treating negative symptoms as interchangeable across conditions. The same observable apathy may demand different strategies in psychosis versus mood disorder, because the underlying computation differs. In practice, this argues for assessing how a given patient is weighing effort against reward – through task-based or careful behavioral observation – rather than inferring it from their diagnosis, and for tailoring motivational and cognitive remediation work to the specific point where the cost-benefit calculation breaks down.
The computation a patient used to decide whether a reward was worth the effort predicted the severity of their negative symptoms better than their diagnosis did.
This is a cross-sectional, task-based study; it captures how people decide in a laboratory paradigm, not how motivation operates in daily life, and it cannot establish whether the computational profile causes negative symptoms or merely accompanies them. Diagnostic subgroups varied widely in size, with depression and bipolar groups small, and medication status was not the focus, leaving treatment effects on effort decisions unresolved.