In a series of repeated trials, schizophrenic patients often fluctuate in p
erformance. Our data suggest that it may be useful, not just to report an i
ncreased variance relative to nonschizophrenics, but to model these fluctua
tions concretely as transitions between a relatively normal and an abnormal
cognitive state - an intermittent degradation in performance that may be r
elated to transient abnormalities in CNS functioning. We define 'dialipsis'
as a temporary substitution of a less efficient process of task performanc
e. This phenomenon is mentioned in the literature, but the descriptions of
dialipsis are heuristic rather than based on a statistical model. We presen
t a mixture model in which the ordinary and degraded states are described b
y distinct ANOVA structures, each with its own task, subject and interactio
n effects, with transitions between them occurring at random times. We disc
uss ways of detecting dialipsis and comparing the mixture model statistical
ly with alternative models. (C) 1999 Elsevier Science B.V. All rights reser
ved.