Takeover warnings as information schedules: How pre-alert content, audiovisual integration, and lead time jointly shape takeover decisions and execution in conditionally automated driving

This dataset contains driving-performance and eye-tracking data from a driving-simulator experiment on takeover warning design in SAE Level 3 conditionally automated driving. It supports the article "Takeover warnings as information schedules: How pre-alert content, audiovisual integration, and lead time jointly shape takeover decisions and execution in conditionally automated driving." Eighty licensed drivers (mean age 23.2 years) each completed eight takeover events on a fixed-base simulator while engaged in a non-driving-related smartphone task. All warnings used a two-stage architecture: a preparatory pre-alert (alerting tone plus visual status display) issued 0, 3, 6, or 9 s before the takeover request, which was always issued 7 s before the critical event. Warning mode was manipulated between subjects (five levels forming an information-capacity gradient — status only; status plus an action recommendation; a pre-alert additionally enriched with the consequences of not taking over — crossed with complementary versus redundant audiovisual integration). Pre-alert lead time and scenario consequence (potential collision vs. potential delay) were manipulated within subjects. Each scenario defined a safer lane-change direction, which the takeover request recommended explicitly in the recommendation-bearing modes. Files. takeover_performance.csv (640 rows = 80 drivers × 8 trials; GB18030 encoding). eye_tracking_aoi.csv (2,345 rows; UTF-8): long format, one row per area of interest per trial, for the 61 drivers in the recommendation-bearing modes. Scripts (run in order; expect both CSVs in the working directory): 01_takeover_performance_gee.R — GEE models of takeover execution and of subjective workload and situation awareness across the five modes and lead times; 02_capacity_focused_2x2x4.R — factorial decomposition (capacity × integration × lead time) within the recommendation-bearing modes; 03_eye_movements_glmmTMB.R — zero-inflated Gamma mixed models of AOI fixation durations by stage, and logistic models linking mirror fixations to the recommended direction; 04_maneuver_choice.R — maneuver classification, compliance models, and attribution of collisions; 05_joint_gaze_performance.R — trial-level merge and models relating environmental sampling to safety margins. Analyses used R 4.6.0 with geepack 1.3.13, glmmTMB 1.1.14, and emmeans 2.0.3.