No excess in fatal traffic crashes around the 8 April 2024 total solar eclipse: a replication of the 2017 design in FARS 2024 against a pre-specified benchmark, with an analysis plan committed after partial outcome inspection

A widely reported study of the 21 August 2017 total solar eclipse found 31% more individuals involved in fatal US traffic crashes over the three days around the eclipse than on matched control days, and warned of a possible similar surge at the 8 April 2024 eclipse. This paper applies the published design without change to the Fatality Analysis Reporting System (FARS) 2024 file: individuals involved in fatal crashes on 7-9 April 2024 against the same weekdays one week before and one week after, using negative binomial regression of daily national counts. The same code first reproduced the published 2017 estimate exactly (741 versus 1,137 individuals, rate ratio (RR) 1.31, 95% CI 1.15-1.49). The decision rule used an interval calibrated on 52 holiday-free non-eclipse weeks from 2022 and 2023, which overlap and are not independent, with the 2017 RR of 1.31 as the pre-specified benchmark. The analysis plan was committed after the eclipse and after national 2024 death and crash counts had been inspected, but, by the author's account, before the primary outcome or any model estimate was computed; its timing cannot be independently verified. In 2024, 695 individuals were involved in fatal crashes on 7-9 April against 1,442 on the six control days: RR 0.96 (model-based 95% CI 0.86-1.07; placebo-calibrated 95% interval 0.80-1.15), which excludes the 2017 effect of 1.31. Deaths (RR 0.98) and fatal crashes (RR 1.00) showed no excess, no gradient by distance to the path of totality was found, and against 28 control days with Easter excluded the RR was 1.06 (0.91-1.24). Afternoon and evening counts on eclipse day were above the control-Monday means (ratios 1.31 and 1.44) but within the range of the same ratio among the control Mondays. Because the calibration rests on overlapping windows, robustness of the exclusion to calibration uncertainty is suggested but not established.