Fertility is an unusually difficult outcome to study in relation to weather. Unlike hospital admissions or electricity demand, births are observed only after a long and partly private sequence of biological events and personal decisions. A single hot day can influence how couples spend their time, whether conception occurs, whether an early pregnancy survives, how long gestation lasts, and when a delivery finally happens. Monthly birth counts therefore bundle together several margins that need not move in the same direction or at the same time. A dip in births nine months after a hot month might mean fewer conceptions, unrecorded early pregnancy losses, or simply shifts in delivery timing—and aggregate statistics alone cannot tell these stories apart.

The exposure measure creates a second, subtler problem, and it is here that the study makes its central contribution. Daily maximum temperature records only the hottest moment of the day, typically a brief afternoon peak. A short spike and a relentlessly hot day can land in the same statistical category, even though they impose very different burdens on the body. Daily minimum temperature tells a different story: a high minimum usually means the night offered little relief, disrupting sleep and physiological recovery. Daily mean temperature summarizes conditions across the whole day. Because these three metrics are correlated but not equivalent, an analysis that relies on just one of them may attribute to a fleeting peak a relationship that actually belongs to sustained heat.

To disentangle these possibilities, Fu assembled an unusually rich dataset. Monthly vital statistics for all 47 Japanese prefectures from January 2012 through December 2019—a balanced panel of 4,512 prefecture-months containing more than eight million live births—were linked to daily weather reconstructed directly from raw Japan Meteorological Agency station records. For each prefecture, the three stations nearest its geographic center were selected and combined using inverse-distance-squared weights, with strict quality screening to exclude unreliable observations. The outcome variable was the monthly general fertility rate: live births per 1,000 women aged 15 to 49, a denominator that clearly defines the population at risk.

The statistical design is deliberately demanding. Each regression includes a full vector of temperature bins, lagged precipitation, and counts of missing weather days, alongside three layers of fixed effects: year-month effects that absorb nationwide shocks, prefecture-by-calendar-month effects that let each prefecture follow its own seasonal rhythm, and prefecture-by-year effects that soak up slow-moving local conditions. This means the temperature coefficients are identified not from comparing hot prefectures with cool ones, but from within-prefecture departures from a location’s usual month-specific exposure. Standard errors were clustered by prefecture, with additional checks using two-way clustering, spatial-temporal inference, and Benjamini-Hochberg adjustments across the full dynamic path of 28 lead and lag estimates per metric.

The results draw a sharp line between exposure measures. Days with daily maximum temperatures of at least 30 degrees Celsius showed only a fragile association with fertility: the lag-nine estimate of roughly 0.23 percent lower fertility was borderline under standard clustering, not significant under two-way clustering, and essentially vanished once complete daily-mean temperature bins entered the model jointly. By contrast, a day with a minimum temperature of at least 25 degrees Celsius was associated with 0.39 percent lower fertility nine months later and 0.25 percent lower fertility ten months later, with the lag-nine estimate surviving both two-way clustering and false-discovery adjustment. A day with a mean temperature of at least 30 degrees Celsius was associated with about 0.50 percent lower fertility at lag ten, and the cumulative mean-temperature association remained negative even when maximum-temperature bins were controlled simultaneously.

The dynamic analysis reinforces this pattern. Plotting the full path from placebo leads through 24-month lags, the maximum-temperature series shows a negative point at lag nine but no concentration of precise negative estimates across the gestational window. The minimum-temperature path reaches its most negative and most precisely estimated value at lag nine, while the mean-temperature path is negative at both lags nine and ten. Future-temperature placebo leads, which test whether the design is merely picking up residual seasonality, were centered around zero and none survived adjustment. Under the within-metric false-discovery corrections, only the minimum-temperature lag-nine estimate cleared the strictest threshold.

What can these estimates actually mean? The study is careful on this point. A decline in births nine or ten months after hot weather is compatible with several mechanisms: couples conceiving less often, early pregnancies being lost before clinical recognition, or gestational lengths shifting. The recorded fetal-death check did not resolve the question, partly because Japan’s statistical definition begins at 12 completed weeks of gestation and combines spontaneous and induced fetal deaths, omitting the earliest losses entirely. Nor did the data identify a behavioral pathway; prior time-use evidence from Hungary found no significant effect of hot days on sexual activity, weighing against treating reduced intercourse as an established general mechanism. The estimates are reduced-form relationships between outdoor measures and monthly fertility, not laboratory parameters ranking the physiological potency of different temperature metrics.

The study also translates its estimates into concrete scale. Using the jointly estimated lag-nine and lag-ten coefficients, the model implies that observed mean-temperature exposure was associated with roughly 5,418 fewer births relative to the reference-bin counterfactual over 2013 to 2019. An illustrative scenario of one additional day per prefecture-year with mean temperature at or above 30 degrees implies about 533 fewer births per year. Crucially, these are gross short-run accounting differences around the gestational horizon, not forecasts of lifetime births. The analysis cannot follow individual households across time, and the absence of a clear rebound through 24 months does not prove that conceptions never occur later—compensation may happen beyond the observed window or through timing shifts that aggregate imperfectly.

The broader lesson extends well beyond Japan. Temperature metrics are not interchangeable labels for a single treatment: maximum temperature emphasizes the daytime peak, minimum temperature the overnight trough and recovery environment, and mean temperature the cumulative daily burden. If buildings, air conditioning, and behavior mediate exposure, the mapping from ambient conditions to personal heat load varies across people and hours, and studies that compare coefficients without reconstructing all metrics consistently risk conflating data construction with biology. For climate-resilient reproductive-health planning, the findings suggest that heat-health surveillance should not rely on daytime maxima alone, since warm nights and sustained heat may flag the exposures that matter most. What remains needed, the author notes, are individual conception and pregnancy histories, gestational timing, early-loss measurement, indoor exposure data, and direct measures of adaptation—tools that would finally allow researchers to say not just whether heat shapes fertility, but how.

Subject of Research: The relationship between ambient temperature measures and short-run fertility rates in Japan

Article Title: Too hot to give birth? Temperature shocks and general fertility rates in Japan

Article References: Too hot to give birth? Temperature shocks and general fertility rates in Japan. (n.d.). https://doi.org/10.3389/fhumd.2026.1899647

Image Credits: AI Generated

DOI: 10.3389/fhumd.2026.1899647

Keywords: heat, fertility, Japan, birth rates, temperature, climate change, general fertility rate, nighttime temperature, conception, pregnancy loss, environmental demography, fixed effects