Abstract: Visual working memory (VWM) underpins higher cognitive activities such as reasoning and learning. State-based models propose that task-relevant representations are maintained in an active state, whereas representations that are temporarily irrelevant but still needed are transferred to a passive state. Previous studies have shown that passive representations resist irrelevant distractors and time decay, yet memory performance consistently declines once representations enter the passive state. However, the source of this loss was still unclear; in particular, it remained unknown whether subsequent active encoding would impair passive representations that had already been formed.
Thus, this study aimed to examine whether subsequent active encoding impaired passive representations and whether this impairment was modulated by encoding load. We adopted a modified sequential change- detection paradigm, in which the first memory array (M1) was transferred to the passive state before later arrays (M2 and M3) were encoded into the active state; M1 was always tested last, so that its accuracy reflected passive memory. A total of 77 university students participated in three experiments. Experiment 1 (n = 23) manipulated the number of active encoding operations (one vs. two) during passive maintenance while holding the retention interval constant. Experiment 2 (n = 30) fixed two encoding operations and manipulated the load of the second encoding (M3: two vs. three items). Experiment 3 (n = 24) manipulated the load of the first encoding (M2: two vs. three items) to test whether M2 entered the passive state and competed with M1 for storage resources. We predicted that, if subsequent active encoding impaired passive representations, passive memory would be poorer after two encoding operations than after one; if this impairment depended on encoding load, a higher load would further reduce passive memory performance.
The results of Experiment 1 showed that passive memory performance was significantly poorer after two encoding operations than after one, even with the retention interval matched, whereas extending the retention interval alone did not impair passive memory. Experiment 2 replicated this effect; although increasing M3 load significantly reduced memory performance for M3, it did not affect passive memory for M1. Experiment 3 again revealed a significant decline in passive memory after two encoding operations; increasing M2 load reduced memory performance for M2 itself but left passive memory for M1 unaffected, ruling out resource competition between M2 and M1 in the passive state. Response times revealed no speed-accuracy trade-off in passive memory. These findings indicated that subsequent active encoding impaired passive representations that had already been formed. Specifically, this impairment was sensitive to the number of encoding operations but insensitive to encoding load, suggesting that it arose from the encoding event itself rather than from competition for storage resources.
In summary, during the maintenance of passive representations, subsequent active encoding is an important source of passive memory loss, and the extent of this loss depends on how often encoding occurs rather than on how much is encoded. These findings suggest that the passive state is not an unconditionally safe store, and extend state-based models by identifying an external source of passive memory loss beyond state-transition costs.

Key words: visual working memory, state-based models, passive storage, active encoding