One Construct, Many Channels: Interpreting Convergence and Discordance in Digital Psychiatric Measurement

Digital phenotyping now measures the same psychiatric construct through many channels at once: speech, voice, face, keystrokes, mobility, movement, self-report and clinician ratings. When these channels disagree, the field usually treats the disagreement as noise to be averaged away or as a signal to be fused into a classifier. Neither approach asks what the disagreement means. We argue that cross-channel discordance is clinically informative, but only if three things are specified. First, an a priori statement of when channels measuring one construct should converge, and when partial convergence is the expected, theory-consistent result. Second, a differential diagnosis of discordance: impaired insight, psychotic content, affective bias, apathy, concealment, alexithymia, construct mismatch, resolution or context mismatch, measurement artifact and medication effects, each with a testable signature. Third, a measurement approach that captures both the magnitude and the direction of discordance within persons, at matched temporal resolution, without the known flaws of raw difference scores. The framework extends two established lines of work, the Operations Triad Model for informant discrepancies and response-coherence theory in affective science, to digital channels, which differ from human informants in having no perspective but many technical failure modes. To our knowledge, this is the first such extension. We illustrate the framework with published findings on physical activity in schizophrenia and mood in bipolar disorder, and close with testable predictions, including whether discordance itself predicts relapse.