The Time-Based Resource Sharing (TBRS) model, developed by Pierre Barrouillet, Sophie Bernardin, and Valerie Camos (2004), provides a computational account of working memory that emphasizes the temporal dynamics of attention. The central claim is that a single, domain-general attentional bottleneck must be rapidly switched between two functions: processing (performing the concurrent task) and refreshing (reactivating decaying memory traces).
The Attentional Bottleneck
TBRS assumes that attention can serve only one function at a time: either process information or refresh memory traces. During complex span tasks, where participants alternate between encoding items and performing a distracting task, attention must be time-shared between these competing demands. The critical variable is cognitive load (CL), defined as the proportion of time during which attention is captured by processing:
where Σ tₐ is the total time attention is captured by processing activities and T_total is the total time available. When CL is high, little time remains for refreshing, and memory traces decay. When CL is low, frequent refreshing maintains traces at high activation levels.
Decay and Refreshing
Memory traces in TBRS undergo temporal decay whenever attention is diverted to processing. The decay is assumed to be time-based: the longer attention is unavailable, the weaker the trace becomes. Refreshing operates by briefly directing attention to each memory trace in sequence, boosting its activation back toward the encoding level. The interplay between decay and refreshing produces a dynamic equilibrium:
where δ is the decay rate and t_unattended is the time since the trace was last refreshed. The model predicts that WM span should be a linear, decreasing function of cognitive load, a prediction that has been confirmed across numerous experiments with different processing tasks.
Predictions and Evidence
The TBRS model makes several specific predictions: (1) Span should decrease linearly with cognitive load, regardless of the nature of the processing task. (2) The pace of processing matters more than the number of processing steps: fast-paced distractors leave less time for refreshing. (3) Free time between processing steps should be used for refreshing, improving recall. All three predictions have received substantial empirical support across studies using reading span, operation span, and custom-designed computer-paced tasks.
TBRS* Computational Implementation
Oberauer and Lewandowsky (2011) implemented a computational version called TBRS* that formalizes the model's assumptions in sufficient detail for quantitative simulation. TBRS* was tested against the SOB-CS model in a series of benchmark experiments. While TBRS* successfully accounts for cognitive load effects, it had difficulty with some interference-based phenomena, leading to a productive debate about the relative contributions of decay and interference to working memory forgetting.
TBRS's reliance on temporal decay has been challenged by interference-based models (e.g., SOB-CS) that explain the same data without assuming that traces decay over time. This debate remains one of the most active controversies in working memory research, with TBRS proponents arguing that time-based effects persist even after controlling for interference, and opponents arguing that all apparently time-based effects can be reduced to interference.