Translated from Japanese

Before Choosing Standard Time, We Should Also Read the Correction to the Health Model

The idea of using circadian rhythm models to compare health impacts is understandable. The PNAS paper by Weed and Zeitzer compares year-round standard time, daylight saving time, and twice-yearly switching using light conditions across the continental U.S. and county-level health data. However, these are not measured health outcomes following a policy change, but predictions based on assumptions about light exposure.

PNAS Paper (September 15, 2025)

If you are using that prediction to support Standard Time, I think we should also read the April 27, 2026, correction. According to the authors, the calculation erroneously included light after sunset due to missing constraints on solar angle. They explain that the original implementation affected all models and time policies, and that both the code and public data have been updated.

PNAS Correction Notice (April 27, 2026)

Changes remain in the public code as well. An April 2 update added logic to set illuminance to zero if the solar altitude is zero or less. A July 17 "Sign correction" changed the divisor in the formula used for longitudinal time adjustments from -15 to 15. It would be best not to treat the April correction notice and subsequent code changes as the same single version.

Public Code: April 2 Change

Public Code: July 17 Change

Simply having the correction published doesn't mean the health impact figures have been re-verified. I haven't run the code myself, and I didn't see a table comparing the pre- and post-correction effect sizes in the notice. At the same time, I want to avoid dismissing the entire study as meaningless just because there was an error.

Personally, I place more weight on brightness in winter mornings, so I am leaning toward year-round Standard Time. I also understand the hesitation in the original post regarding a preference for evening brightness. If we are to use health outcomes as a basis for choosing Standard Time, how much of a comparison regarding how the predictions changed per policy in the corrected version has actually been confirmed?

Loading comments…