Main effect
To investigate whether anomalous nighttime temperatures harmed the sleep quality of individuals, we constructed a data set of individuals’ reported monthly nights of insufficient sleep linked with monthly historical nighttime temperature data. Our individual response data come from the Centers for Disease Control and Prevention Behavioral Risk Factor Surveillance Survey (BRFSS) pooled over the period 2002–2011. Randomly selected respondents answered the question: “During the past 30 days, for about how many days have you felt you did not get enough rest or sleep?”
Questions from the BRFSS have been assessed for validity (
22) and reliability (
23) and are largely consistent with other health-related activity measures, including our specific measure of perceived sleep insufficiency (
24). Further, this specific question is used in widely cited public health studies related to sleep (
25).
We combine these individual responses—marked by interview date and geolocated to the city level—with station-level daily temperature and precipitation and climate normals data from the National Centers for Environmental Information Global Historical Climatology Network–Daily (GHCN-D) (see Cities and Stations in the Supplementary Materials) (
26). Notably, our analysis is robust to the use of gridded daily weather data from the PRISM Climate Group instead (see PRISM Data in the Supplementary Materials) (
27). Our theoretical relationship of interest is the effect of nighttime temperature anomalies on insufficient sleep. We empirically model this relationship as
where β is our parameter of interest. In this pooled cross-sectional model,
i indexes individuals,
j indexes cities,
s indexes seasons, and
t indexes calendar days (our results are robust to the use of a negative binomial model instead; see Negative Binomial in the Supplementary Materials). Our dependent variable
Yijst represents respondents’ number of nights of insufficient sleep over the past 30 days (results are robust to dichotomizing this variable in a linear probability model; see Linear Probability Model in the Supplementary Materials). Our independent variable of interest,
Xjst, represents the 30-day average of daily minimum temperature deviations from their normal daily values (from 1981 to 2010) over the same 30-day window as respondents’ reported nights of insufficient sleep
where
r indexes the days before an individual’s survey response date.
The
Zη term in
Eq. 1 represents a set of climatic control variables that include average temperature range, precipitation anomalies, cloud cover, and humidity [with cloud cover and humidity data drawn from the National Centers for Environmental Prediction (NCEP) Reanalysis 2 project (
28)]. We included these other meteorological variables because their exclusion might bias our estimates of the effect of nighttime temperature anomalies (although the magnitude of β is largely unaffected by the exclusion of these variables; see Main Effect in the Supplementary Materials) (
29).
Further, unobserved characteristics may influence sleep. For example, people may sleep better in cities with lower noise pollution or higher prevalence of air conditioning, on days when they are more likely to have leisure time, or because of other city-specific seasonal factors. To be sure that geographic and temporal factors like these do not interfere with our estimate of the effect of nighttime temperature on human sleep, we included γ
t, and ν
js in
Eq. 1. These terms represent calendar date and city-by-season indicator variables that account for unobserved characteristics constant across cities and days as well as seasonal factors that might vary differentially by city (
30). Notably, our results are robust to varying the specification of these controls (see Time and Location Controls in the Supplementary Materials). Our empirical identifying assumption, consistent with the literature (
31–
33), is that temperature anomalies are as good as random after conditioning on these fixed effects. The estimated model coefficient β can thus be interpreted as the effect of temperature anomalies on reports of insufficient sleep (
34–
36).
Because our estimation procedure uses exogenous city-level variations in nightly temperature deviations to predict individual-level outcomes, we account for within-city and within-day clustering of standard errors by using heteroskedasticity robust errors clustered on both city and day. Finally, we omit nonclimatic control variables from
Eq. 1 because of their potential to generate bias—a phenomenon known as a “bad control” (
36)—in our parameter of interest (nonetheless, our results are robust to the inclusion of common demographic covariates; see Demographic Controls in the Supplementary Materials).
As can be seen in
Fig. 1, as temperature anomalies become more positive, the incidence of nights with insufficient sleep increases. The results of estimating
Eq. 1 indicate that a +1°C deviation in nighttime temperatures produces an increase of approximately three nights of insufficient sleep per 100 individuals per month (β = 0.028,
P = 0.014,
n = 766,761). Notably, nonlinear specifications of nighttime temperatures, precipitation, and daily temperature range return similar estimates of β, and a permutation test further supports our statistical inference (see Main Effect and Permutation Test in the Supplementary Materials) (
37). Putting scale to the magnitude of our estimated effect, a harmonized +1°C nighttime temperature anomaly, if extrapolated across the current population of the United States, would produce nearly 9 million additional nights of insufficient sleep per month or approximately 110 million extra nights of insufficient sleep annually.
Heterogeneous effects of nighttime temperature on sleep
The above β represents an estimate of the average effect of anomalous nighttime temperatures on sleep over the course of a full year. However, because excessive heat disrupts sleep by preventing normal decreases in core body temperature (
9,
21,
38), above-average summer nighttime temperatures might plausibly harm sleep more than above-average temperatures in cooler periods of the year. This leads us to our second question: Do the effects of nighttime temperatures on sleep vary by season?
To investigate this question, we stratify our sample and examine
Eq. 1 for each season of the year (excluding city-by-season indicator variables). We find that the effect size β during summer (β = 0.073,
P = 0.019,
n = 179,117) is almost three times the magnitude of the effects observed during any other season of the year, as can be seen in
Fig. 2A. The effects during spring, fall, and winter are all positive but are smaller in magnitude and fail to gain significance at the α = 0.05 level.
In addition to heterogeneous effects by season, we may expect that not all individuals will be similarly affected by anomalous increases in nighttime temperatures. This leads us to our third question: Are the observed effects most acute among those least able to cope with nighttime heat? For example, more wealthy individuals may be able to afford running the air conditioning at night, whereas those in lower-income brackets may not (
39). Further, it has been observed that older individuals can have deficient thermoregulation (
40), which may make their sleep cycles more vulnerable to anomalous temperatures.
To examine whether lower-income and elderly respondents are most acutely disturbed by above-average nighttime heat, we stratify our sample along relevant demographic covariates, again estimating
Eq. 1 for each subsample. Splitting the sample along its median income bracket ($50,000), we find that the effect of temperature anomalies on insufficient sleep is greatest for lower-income respondents (see
Fig. 2B). The effect for the lower-income group (β = 0.042,
P = 0.009,
n = 342,565) is over three times the magnitude of the higher-income group (β = 0.012,
P = 0.455,
n = 322,044). Next, splitting the sample along a common age dimension—over or under 65 years of age—we find that our effects in older adults (β = 0.041,
P = 0.043,
n = 223,211) are nearly twice the magnitude of those found in younger adults (β = 0.025,
P = 0.064,
n = 535,968) (see
Fig. 2C).
Combining these insights, the effect observed in a subsample of elderly, lower-income respondents during the summer (β = 0.175, P = 0.007, n = 30,532) is approximately 10 times the magnitude of the effect observed in the remainder of the sample excluding this group (β = 0.018, P = 0.089, n = 735,743). Thus, our data suggest that both lower-income and elderly individuals may be most susceptible to increasing nightly temperatures and that these individuals experience more severe sleep disruptions in response to atypically warm summer nights.
Potential implications of climate change for human sleep
Our historical data indicate that past nighttime temperature anomalies have likely altered historical sleep patterns in meaningful ways. In the future, climate change is likely to produce an increased frequency of above-average nighttime temperatures (see
Fig. 3A) (
41), and observed nighttime temperatures have been increasing more rapidly than daytime temperatures over the last century (
42). These observations lead us to our fourth question: Might warming nighttime temperatures due to climate change increase the incidence of insufficient sleep in the future?
To examine this question, we calculate projected nighttime temperature anomalies for 2050 and 2099 from NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) bias-corrected, statistically downscaled (to 25 km × 25 km), nightly temperature projections (
43) drawn from 21 of the Coupled Model Intercomparison Project Phase 5 (CMIP5) models (
44) run on the Representative Concentration Pathways “high emissions” scenario (RCP8.5) (
45). We extract the projections’ time series for the grid cells associated with each city and couple these predicted anomalies with our historical estimate of the relationship between nighttime temperature anomalies and insufficient sleep—β from
Eq. 1—to calculate a forecast of possible insufficient sleep due to future nighttime warming at the city level for each downscaled climate model. We define our forecast of the increase in insufficient nights of sleep due to climate change by 2050 (Δ
Ymj2050) as
and for the additional effect from 2050 to 2099 (Δ
Ymj2099) as
where
m indexes the 21 specific climate models,
j indexes the city, and
t indexes the day.
Xmjt—our measure of 30-day nighttime temperature anomalies—is calculated as in
Eq. 2, using 1981–2010 as the baseline. This forecasting procedure enables us to incorporate uncertainty regarding future downscaled climatic projections into our forecast (
29).
Figure 3B plots the estimation results. Each of the 219 cities in our analysis has a prediction for each of the 21 downscaled climate models, producing approximately 4600 estimates per period. On average, we project that climate change may cause approximately 6 additional nights of insufficient sleep per 100 individuals by 2050 and an excess of approximately 14 nights per 100 individuals by 2099 (see
Fig. 3C).
In addition, the effects of climate change on nighttime temperatures—and thus the potential impacts of climate change on sleep—are likely to vary geographically across the United States. To investigate the spatial distribution of potential reductions in sufficient sleep due to climate change, we take the ensemble average of the 21 NEX downscaled climate models for each of the 2010, 2050, and 2099 forecasts. We then take the yearly average of nighttime temperature anomalies for each grid cell in the continental United States in each year. For the 2050 forecast, we assign to each grid cell the difference in average nighttime temperature anomaly between 2010 and 2050. For the 2099 forecast, we assign to each grid cell the difference in average anomalies between 2010 and 2099. As can be seen in
Fig. 4—with effects scaled to be per 100 individuals—areas of the western and northern United States are likely to have the most significant changes in nighttime temperatures and, as a result, may see the greatest increase in climate change–induced nights of insufficient sleep.