Abolishing Daylight Saving Time and the Clock-Hour Distribution of Road Crashes: Evidence from Brazil (2019) and Mexico (2022)

Most evidence on daylight saving time (DST) and road safety comes from the twice-yearly clock transitions or from comparisons of periods and places with and without DST. Abolition is different: it removes the evening light shift for a whole season without a sleep-loss night. This paper studies the end of DST in Brazil (2019) and most of Mexico (2022). The registered design compares the ratio of evening (17:00-20:59) to morning (05:00-08:59) crashes, in local clock time, between states that used to observe DST and states that never did, before and after abolition. In Brazilian federal-highway data (15 November to 10 February, 2014/15 to 2025/26, excluding 2020/21), the registered difference-in-differences estimate of ln(evening/morning) is -0.120 (95% cluster-bootstrap interval -0.194 to -0.052; state-label permutation p = 0.007), the opposite sign to the light hypothesis. The registered falsification test in June to August, when no state changed its clocks, gives an estimate of the same size (-0.126, interval -0.205 to -0.041), so the primary estimate cannot be distinguished from a regional divergence present in months without DST. An exploratory triple difference is 0.006 (interval -0.111 to 0.125). With an exploratory pre period from the recording system used since 2017 only, the primary estimate is -0.053 (interval -0.167 to 0.050) and the falsification estimate -0.070 (interval -0.197 to 0.057). In Mexican urban crash data (May to September, 2016-2019 vs 2023-2025), the registered year rule applied to the current release gives 0.059 (interval -0.045 to 0.164); on the older file at the registered address the estimate is 0.095 (interval -0.022 to 0.209). The exploratory Mexican triple difference is -0.066 (interval -0.113 to -0.011), against the prediction. Neither country shows a detectable abolition-specific shift of crashes toward the evening.