Why the season, not the clock, is the hard part to rule out
Daylight saving time arrives twice a year, and both times sit inside a much larger seasonal change. Dawn and dusk are already shifting, and at latitudes away from the equator those changes in light timing are tangled up with the clock change itself [5]. So if crashes or heart events rise in the week after a spring transition, the cause could be the lost hour of sleep, the earlier sunrise, or something else about that time of year.
This is what researchers mean by seasonal confounding: a seasonal factor that moves alongside the thing you want to study and can produce the same pattern. The practical consequence is that a simple before-and-after comparison around a transition date cannot, on its own, tell you whether the clock shift did anything. The design has to give the season somewhere else to go.
Designs that separate the clock change from the season
The strongest approach in this set compares jurisdictions that change their clocks with jurisdictions that do not, using a difference-in-differences design. That method asks whether the change around the transition is larger in the places that shifted than in the places that stayed put, so a shared seasonal trend cancels out. One traffic-crash study used exactly this setup with non-DST control jurisdictions, and then added further controls and fixed effects because the basic version might still leave residual seasonal confounding or day-of-week-specific differences in the treatment [1].
A second approach handles the problem statistically rather than through a comparison group. A study of vehicular crash fatalities used a multilevel model to account for temporal and spatial heterogeneity, meaning it let the baseline risk vary by place and time instead of pooling everything together [4]. That matters because crash risk is not uniform across a country or a calendar, and a model that ignores this can attribute ordinary local variation to the clock change.
A third approach uses large administrative datasets with models built for event rates. A nationwide US study of adverse cardiovascular events applied Bayesian hierarchical Poisson regression to estimate how event rates related to DST transitions [3]. The hierarchical structure lets the model borrow strength across groups and absorb some of the natural variation between them, which is a way of keeping the estimate from being driven by a handful of unusual places or periods.
What these designs can and cannot settle
Even a well-controlled design answers a narrower question than it first appears. The difference-in-differences crash study is framed around whether DST increases traffic crashes, and its added controls are there to protect that specific estimate from residual seasonality [1]. The cardiovascular study estimates associations between transitions and event rates across a large population [3]. Neither design, as described, establishes a biological mechanism; they estimate whether the timing of events shifts with the clock.
Sleep research faces a related limit. A longitudinal Fitbit study tracked sleep patterns across seasonal transitions with a comparison group, which is a sensible way to see whether sleep timing shifts when clocks do [2]. But the broader picture of downstream health effects remains mixed across outcomes and study designs, as that work itself notes [2]. Mixed results across designs are a signal to look at how each study handled season and comparison groups, not proof that the effect is absent.
How future studies should handle the season
The clearest lesson from these designs is to build the season into the comparison rather than correcting for it afterward. Using non-DST control jurisdictions, as the crash study did, gives the seasonal trend a natural place to show up without being mistaken for a clock effect [1]. Where such controls are unavailable, multilevel or hierarchical models that let risk vary by place and time serve a similar purpose by absorbing heterogeneity that would otherwise contaminate the estimate [3][4].
A second lesson is to check whether the basic design is doing enough. The crash study's authors treated their baseline difference-in-differences estimate as potentially insufficient and added controls for residual seasonality and day-of-week-specific treatment differences [1]. That is a useful habit: state what the simple design assumes, then test whether those assumptions hold. Finally, because results differ across outcomes and study designs [2], future work is most informative when it reports how the season was handled and what comparison group was used, so readers can judge whether a null or positive result reflects the clock or the calendar.
Sources used in this answer
Does daylight saving time increase traffic crashes? Seasonal confounding and a morning fatality effect
A difference-in-differences design using non-DST control jurisdictions, augmented with controls and fixed effects, is presented as a way to isolate clock-shift effects on traffic crashes from residual seasonal confounding.
Daylight Saving Time, Acute Sleep Disruption, and Return-to-Baseline Patterns in All of Us Fitbit Data
A longitudinal Fitbit study with a comparison group tracked sleep patterns across seasonal transitions, while noting that downstream health effects remain mixed across outcomes and study designs.
Daylight saving time practice and the rate of adverse cardiovascular events in the United States: a probabilistic assessment in a large nationwide study
A large nationwide US study used Bayesian hierarchical Poisson regression to assess associations between DST transitions and rates of adverse cardiovascular events.
Spatiotemporal Context for Vehicular Crash Fatalities Days Before and Days After Daylight Saving Time Transitions
A multilevel statistical approach was used to examine vehicular crash fatalities around DST transitions while accounting for temporal and spatial heterogeneity.
The human circadian clock's seasonal adjustment is disrupted by daylight saving time
Survey and longitudinal data indicate that the human circadian clock's seasonal adjustment is disrupted by the twice-yearly clock change, a setting where light timing is confounded with seasonal change.