Forward Collision Warning Effects Across Driver Attention State Clusters Under Comparable-Risk Naturalistic Driving Conditions

Using Naturalistic Driving Study data, this study examined how the association between Forward Collision Warning (FCW) activation and perception-reaction time (PRT) varied across driver attention states during car-following. Head-orientation, gaze-direction, and upper-body posture features were used in a Hidden Markov Model to estimate unstable-attention probability over time, and clustering identified three attention-state profiles: stable, fluctuating, and persistently unstable. The stable profile reflected consistently low unstable-state probability, the fluctuating profile reflected frequent shifts in unstable-state probability, and the persistently unstable profile reflected sustained high unstable-state probability. A Gamma generalized estimating equation model was applied to comparable-risk events. The association between FCW activation and PRT showed evidence of variation across attention states, with the largest estimated warning-related reduction observed in the persistently unstable attention state. These findings suggest that considering pre-event attention state may enable a more informative assessment of the support FCW can provide in situations with a greater risk of delayed driver response.