Attentional Breadth Measurement: A Latent Variable Approach

Abstract

Attentional breadth refers to the scope of information individuals attend to, which can be biased toward the overall global picture (e.g., the forest) or the local elements within it (e.g., the trees). Laboratory tasks measuring global/local bias often show good reliability over time, but researchers have struggled to identify meaningful relationships between tasks, raising concerns about whether attentional breadth reflects a unified construct. This study addressed this issue using structural equation modeling (SEM) to model shared variance across tasks and sessions, while accounting for measurement error, stimulus-specific processes, and task-specific variance. A two-session design was used, incorporating three hierarchical stimulus tasks: (1) target detection, measuring RTs to find targets at global or local levels; (2) forced-choice, where participants chose whether an item that matched the target at a local level or an item that matched the target at a global level better represented the hierarchical target (3) Navon interference, assessing interference from the irrelevant global or local level. An embedded figures task, where participants located a smaller shape within a larger display, was also included. Each task was presented in two stimulus variants, with participants completing both versions in each session. Two conceptual breadth tasks (Remote Associates Test and Object Categorization Task) were also administered. Data were analyzed from 105 Brock University undergraduates who had complete task data across testing sessions. Results showed moderate to strong test-retest reliability for all tasks except both Navon interference versions, which lacked stability and within-session convergence. Navon Tasks were subsequently excluded from SEM analyses. Conversely, latent factors for detection, forced choice, and dis-embedding (response time and accuracy) showed acceptable fit and were retained in the final SEM. Notably, a significant positive relationship was observed between the detection and forced choice latent factors, demonstrating – for the first time – a shared attentional breadth construct across distinct paradigms. A second significant positive relationship emerged between forced choice and dis-embedding accuracy, but in a counterintuitive direction. Beyond these associations, traditional bivariate correlations failed to detect relationships among attentional breadth tasks, underscoring the value of latent variable modeling and multi-session designs in isolating stable, trait-like constructs from task-specific noise.