The cold surprise
Why is cold the surprise?
Mortality rises at both ends of the temperature distribution; that U shape is well documented, for Mexico by Cohen and Dechezleprêtre (2022). Emergency department visits do not follow it. They rise with temperature across the whole range: an additional day below 10 °C reduces them by 9.1%, a day above 30 °C raises them by 3.7%.
Two outcomes, one axis
Cumulative effects over 30 days for mortality and ED visits. Visits slope in one direction across the whole range; deaths rise at both ends.
estimate, with the 95% interval beneath
| Group | ≤10 °C | 10–15 °C | 15–20 °C | 20–25 °C | 25–30 °C | >30 °C |
|---|---|---|---|---|---|---|
| Emergency department visits | −9.1%−10.0% to −8.1% | −5.3%−5.8% to −4.9% | −2.7%−3.0% to −2.4% | reference | +2.5%+2.1% to +2.8% | +3.7%+3.1% to +4.4% |
| Deaths, all causes | −3.6%−4.5% to −2.6% | −3.4%−4.1% to −2.8% | −2.2%−2.7% to −1.7% | reference | +4.2%+3.7% to +4.7% | +9.2%+8.0% to +10.4% |
estimate, with the 95% interval beneath
| Group | ≤10 °C | 10–15 °C | 15–20 °C | 20–25 °C | 25–30 °C | >30 °C |
|---|---|---|---|---|---|---|
| Emergency department visits | −16.0%−19.7% to −12.4% | −11.8%−13.9% to −9.7% | −5.3%−6.6% to −3.9% | reference | +4.5%+2.8% to +6.1% | +2.7%+0.4% to +5.0% |
| Deaths, all causes | +9.4%+2.9% to +15.8% | +3.7%+0.5% to +6.9% | +0.8%−1.4% to +3.0% | reference | +2.5%+0.9% to +4.1% | +10.0%+5.3% to +14.7% |
What to read here
- Visits are monotonic. Each interval colder than the reference reduces them more than the last (2.7% at 15 to 20 °C, 9.1% below 10 °C) and each warmer interval raises them (2.5% at 25 to 30 °C, 3.7% above 30 °C). There is no turning point.
- Deaths behave differently. Once the following 30 days are accounted for, the mortality curve is higher at both ends than in the middle: cold days and hot days both raise deaths relative to a day at 20 to 25 °C.
- The disagreement is at the cold end. At the hot end both outcomes move up. At the cold end deaths move up while visits move down, and the reduction in visits grows rather than reverses: 9.1% on the day, 16.0% once the following 30 days are accounted for.
Why this matters
Emergency department volume measures demand for emergency care, not population health. At the cold end of the distribution fewer people come to the emergency room, on the day and in the weeks after. The paper's mechanism section discusses why visits fall: fewer infections spread by vectors and food, fewer injuries, and patients staying home.
The shape, not just the level
The paper's summary statement is that, unlike mortality, the demand for emergency care responds to temperature roughly linearly. That is what the visits series shows, and it is what the rest of this site unpacks: by age (the young respond most), by diagnosis (respiratory is the one chapter that rises after cold), and in the projections (a warming Mexico trades fewer cold days for more hot ones, with visits of different severity).
What is being estimated
Visits: the paper's main specification, a Poisson distributed lag model with 30 lags on the ED visit rate per 100,000 people by municipality and day, 2008 to 2021, with municipality-by-year-month and municipality-by-weekday fixed effects, weather controls, population weights, and standard errors clustered by municipality.
Deaths: all-cause deaths per 100,000 people by municipality and day over the same years, in the same Poisson distributed lag form, but with the model of Cohen and Dechezleprêtre (2022): municipality by day of year, municipality by year, and date.