Downloaded by guest on September 23, 2021 h otvleal oteeavreevrnetlconditions environmental adverse these among to are vulnerable status most socioeconomic the lower increase and visits conditions men- health hospital preexisting tal with psychiatric Those (31–33). and temperatures hotter (28–30), the during amplify suicide drought of and heat risk with dis- both stress associated Furthermore, posttraumatic (22–27). as is well order floods as For depression and acute extremes. of hurricanes symptoms weather to and exposure events example, climatic discrete on 21). focused (2, compelling is between risk relationship health theoretical mental and the (20). adversi- change estimate, climate losses climate-induced to ecological these difficult of of is ties face magnitude the precise (19). the in migration While grief liveli- associated forced of and disruption produce hood experiencing and already are (18), communities Some conflict (17), activity economic social harm (16), spur amplify health temperatures physical human Rising to ways. risks costly to in expected systems also human are alter climate in in changes Gradual disruption regions. societal affected psycho- and of injury, damage, intensity infrastructure physical trauma, cause and logical often frequency which the (13, (15), amplify disasters health to natural mental likely to is poses Warming change 14). climate that threats indirect adversity. ability with the cope reducing functioning, to psychological (12) impair immunological can and distress life 11) of (10, of levels quality subclinical reduce Even overall and 9). (8), worsen (3, productivity disorders anxi- diminish 7), psychological with (6, majority These health (4), the (5). comprising lifetime diagnoses disorders their are of mood Americans in and difficulties of stress-related, illness one-half health ety, mental Nearly Mental experience (3). costly fac- (2). will and risk disorders common known exacerbate mental already to for likely is tors change climate systems, S climate to threats pose health. change mental climate envi- analyses human by that Our produced conclusion metric. the stressors this for ronmental in support Katrina increase quantitative Hurricane added point to provide prevalence 4% exposure a the that with in and associates increase issues, point 1 health 2% mental that of a points, with 0.5% associates by warming difficulties health mental temperatures of monthly between from residents shifting 25 that US between find sampled We 2012. randomly and drawn million 2002 difficulties 2 health mental nearly reported from with meteorological data couple climatic we weather, so, and do extreme To associate health. each more mental exposure worsened to with cyclone Here, tropical exposure and hypothesis. large- warming, short-term this multiyear Few that examined change. show empirically climate we by have undermined studies wellbeing—has be 2018) human scale 26, to of January review potential facet for (received critical the 2018 21, health—a August mental approved and Sound AZ, Tempe, University, State Arizona Turner, L. B. by Edited 02139 MA Cambridge, Technology, of Institute Massachusetts www.pnas.org/cgi/doi/10.1073/pnas.1801528115 74136; OK Tulsa, Research, 02138; MA Cambridge, a Obradovich Nick by posed change risks climate health mental of evidence Empirical ei a,MsahstsIsiueo ehooy abig,M 02139; MA Cambridge, Technology, of Institute Massachusetts Lab, Media miia netgtoso hsrltosi aeprimarily have relationship this of investigations Empirical and direct the highlighted have scholars decade, past the Over at fpyhlgclwlbig() ydsutn these disrupting By (1). determi- wellbeing critical psychological are systems of physical nants and economic, ocial, | etlhealth mental ◦ n 30 and C a,b,1 c scooyDprmn,EihNus oesMmra eeasHsia,Bdod A01730; MA Bedford, Hospital, Veterans Memorial Rogers Nourse Edith Department, Psychology | oy Migliorini Robyn , aua disasters natural e scityDprmn,Uiest fClfri,SnDeo aJla A903 and 92093; CA Jolla, La Diego, San California, of University Department, Psychiatry ◦ to C >30 ◦ | nrae h probability the increases C psychology c atnP Paulus P. Martin , | weather ◦ f5-year of C d,e n ydRahwan Iyad and , b efrCne o cec n nentoa far,HradUniversity, Harvard Affairs, International and Science for Center Belfer te tnadmaue fmna elhsau 3,4) While 40). (39, status health mental of measures with standard validity convergent other and (38) reliability test–retest chometric otherwise zero and period (see the over difficulties health mental cate d 30 good?” past not the health during mental days your many was how and for depression, (37). emotions, stress, thinking with includes 2012 problems which “Now health, and question: mental your following 2002 about Factor the between Risk answered Center (BRFSS) Behavioral Respondents the Prevention’s System from and Surveillance drawn Data- Control is Disease Harvard’s health 2002 for on mental between available reported residents individuals’ are US materials health sampled verse: (study mental randomly 2012 the million and and 2 conditions of climatic historical by between posed (36), risks sleep health (2). lacking mental and is change the (35), climate of expression quantification in emotional large-scale ways (34), such out- the phenomena, cognition health understand psychological as mental to other severe influences beginning most weather are which the we only Although to local comes. limited qualitative, or been to scale, have difficult in investigations been most have as impacts characterize, population-level However, (21). ulse nieOtbr8 2018. 8, October online Published 1073/pnas.1801528115/-/DCSupplemental. y at online information supporting contains article This 1 aadpsto:Tesuymtrasaeaalbeo avr’ Dataverse: Harvard’s on available are materials org/10.7910/DVN/OVQY76 study The deposition: Data the under Published Submission.y Direct PNAS a is article This interest.y of conflict no the declare wrote authors I.R. The and M.P.P., R.M., N.O., research; and designed data; N.O. paper. analyzed y research; N.O. the research; of performed conceived N.O. R.M. and N.O. contributions: Author owo orsodnesol eadesd mi:[email protected] Email: addressed. be should correspondence whom To ec otegoigltrtr ikn lmt hneand change climate linking evi- literature large-scale health. growing mental added worsened the provide to to results in linked dence is increase These future, to health. the likely mental in cyclones, intensity issues, tropical and health to frequency asso- mental exposure warming of that prevalence multiyear and increased that that an find with health, ciates we mental collection, worsen precipitation added data each and temperatures of hotter decade with experience US a sampled randomly across million residents 2 and nearly from direct coupled information of data with variety meteorological daily a Using through mechanisms. indirect health mental likely are undermine change climate to of impacts the have that Scholars indicated health. recently mental sound without falters Wellbeing Significance ecd epne oti usina n frsodnsindi- respondents if one as question this to responses code We relationship the on report we gap, this address to begin To .Ti esr a ensont oss ohpsy- both possess to shown been has measure This Data). .Ormaueof measure Our https://doi.org/10.7910/DVN/OVQY76). PNAS NSlicense.y PNAS a,f | coe 3 2018 23, October . y f nttt o aa ytm,adSociety, and Systems, Data, for Institute | d araeIsiuefrBrain for Institute Laureate o.115 vol. www.pnas.org/lookup/suppl/doi:10. | o 43 no. | 10953–10958 https://doi.

SUSTAINABILITY PSYCHOLOGICAL AND SCIENCE COGNITIVE SCIENCES not a direct measure of psychiatric disorders, this item has a number of strengths. First, it is to our knowledge the best large- scale, randomly sampled measure of individual mental health status in the United States. Second, the measure is spatially and temporally referenced in a manner that enables precise pair- ing with meteorological data. Third, it is likely able to capture both clinical and subclinical distress across a wide set of possible symptoms (5) and unlike measures of health care utilization, can account for the substantial portion of US adults who fail to seek treatment (41). We combine our mental health data with meteorological data and empirical tools drawn from the climate econometrics liter- ature to investigate the historical relationship between climatic variations and human mental health. These historical relation- ships can aid in estimating the magnitude of the risks that climate change poses to mental health. Following the theoretical frame- work of Bourque and Cunsolo Willox (14), we examine three selected types of environmental stressors likely to be produced by climate change: short-run meteorological exposure, multiyear warming, and acute exposure to natural disasters. We examine three questions. First, anthropogenic warming is likely to present humans with increasingly extreme meteorological conditions in any given year Fig. 1. Higher temperatures and precipitation increase risk of mental (42). Do recently experienced meteorological stressors affect health issues. A presents projections of warming in the United States along individuals’ reported mental health? Additionally, are those who the RCP8.5 high-emissions scenario. The annual number of days with max- ◦ are most vulnerable to mental health challenges more affected imum temperatures > 30 C is projected to markedly increase over this by such meteorological stress (14)? In our sample, more vulner- century. B plots the projected number of annual days with measurable pre- > able populations include those with lower incomes, those who cipitation ( 1 mm). C draws from nearly 2 million respondents’ reports of monthly mental health issues between 2002 and 2012. It plots the predicted experience a higher average burden of mental health problems probability of reporting any mental health issues for each 30-d average (43), and those who may be less able to smooth adverse temper- maximum temperature bin. The probability of mental health issues steadily atures (44) as well as women, who are more susceptible to the increases past 10◦ C to 15◦ C. D depicts that precipitation days increase the mental health difficulties captured by our measure (4). To exam- probability of mental health issues. Dotted lines represent 95% confidence ine this set of questions, we use pooled cross-sectional analyses, intervals. leveraging exogenous meteorological variation to examine the effect of short-run (past 30 d) weather exposure on individuals’ mental health outcomes (Pooled Cross-Section and SI Appendix, days (coefficient: 2, P < 0.001, n = 1,961,743). Putting scale Fig. S2) (45). to the magnitude of this estimated relationship, a uniform shift Second, climate change is also likely to increase the rates from monthly temperatures between 25 ◦C and 30 ◦C to averages of year-over-year and decade-over-decade warming of local cli- greater than 30 ◦C, if extrapolated across the current popula- mates and the chronic stressors that such warming produces tion of the United States over a 30-d period, would produce (14, 46). Does longer-term warming have detrimental impacts almost 2 million additional individuals reporting mental health on individual mental health over time? To examine this ques- difficulties. tion, we use the long-differences approach (47), examining To examine if poorer respondents are more sensitive to tem- the association between spatial variation in multiyear warm- perature, we stratify our sample along income quartiles and ing and longer-run changes in mental health (Long Differences estimate regressions for the lowest and highest quartiles (SI and Fig. 3B). Appendix, Table S8). Fig. 2A shows that the negative effect of Third, climate change is also likely to amplify the fre- temperatures greater than 30 ◦C on the probability of men- quency and intensity of acute climatic events, like tropical tal health issues is larger for low-income respondents (coeffi- cyclones (14, 15). Does direct exposure to costly tropical cient: 2.309, P = 0.002, n = 438,518). This is 1.6 times the cyclones worsen individual mental health outcomes? To exam- effect observed among the highest-income adults in the sample ine this question, we use a difference-in-differences approach, (coefficient: 1.458, P = 0.005, n = 509,608). leveraging the landfall of Hurricane Katrina to examine Fig. 2B shows that the negative effect of temperatures greater the association between tropical cyclone exposure and our than 30 ◦C on the probability of mental health difficulties is measure of reported mental health outcomes (Difference-in- largest for women (coefficient: 1.41, P = 0.005, n = 1,211,220). Differences) (48). This is 1.6 times the effect observed among men in our sample (coefficient: 0.879, P = 0.031, n = 750,523). Combining these Results insights, the effect observed in the subsample of low-income Short-Run Weather Exposure. The results of estimating our pooled women (coefficient: 2.742, P = 0.002, n = 300,570) is approx- cross-sectional regression indicate that exogenous increases in imately two times the magnitude of the effect observed in the monthly temperature (Fig. 1C and SI Appendix, Fig. S2) and high-income men in the sample (coefficient: 1.373, p = 0.03, n = added precipitation days (Fig. 1D) each amplify the monthly 235,098). probability of experiencing mental health issues (Pooled Cross- Looking forward, we observe that days with temperatures that Section and SI Appendix, Table S1). Average maximum tempera- exceed 30 ◦C are likely to become more common in the future tures greater than 30 ◦C amplify the probability of mental health (Fig. 1A), particularly in the US South. Changes in precipi- issues by over 1% point compared with 10 ◦C to 15 ◦C (coeffi- tation are projected with less certainty (Fig. 1B). Our exami- cient: 1.275, P < 0.001, n = 1,961,743). Months with greater than nation of these projections indicates that any climate change- 25 d of precipitation increase the probability of mental health induced alteration in US mental health outcomes would be issues by 2% points compared with zero monthly precipitation more likely due to changes in future temperature distributions

10954 | www.pnas.org/cgi/doi/10.1073/pnas.1801528115 Obradovich et al. Downloaded by guest on September 23, 2021 Downloaded by guest on September 23, 2021 nrprso etlhat susi h eidatrKatrina after period the (Difference-in-Differences in period and issues pre-Katrina health the with mental compared of reports Gulf in US the US along in individuals disasters of (49). worst millions the Coast the affected of of 2005. and one use August history in make late resulted in we Katrina Katrina health, Hurricane Hurricane mental of landfall and catastrophic exposure cyclone cal Disaster. Natural S13). Table Appendix, (SI 4 from Fig. effects in estimated models the by-season display these of dur- We months. occurred prevalence winter that and the warming fall on ing with effects compared issues largest health the mental has summer and spring issues. health mental of prevalence in esti- increase multiyear median a The that right. suggests the 1 procedure on permutation this histogram 3D from the mate Fig. in not We estimates in do all that estimates. points of intervals unique as confidence 278 zero 95% include in with Appendix, estimates (SI results postperiod the where procedure the depict sample before This y the S8). 5 in least Fig. years at is of preperiod regression combinations the our possible estimate we all postperiod, across and preperiod of choice 1.293, (coefficient: (48). error estimate measurement smaller a produce P sample years the in some year ing each across subjects 2.055, have (coefficient: that cities the only the at significant are are issues effects health marginal mental All on with women. temperatures among higher Those larger of income. effects temperatures. The men- higher of gender. reporting to by response of quartiles in probabilities issues amplified fourth health have tal quartile and first the first in incomes the between sample women. among 2. Fig. baoihe al. et Obradovich 1.913, (coefficient: preva- difficulties in health increase mental p point reported 2% of approximately lence an with periods, 1 associates both a in that representation find have that we sample our in cities the Differences (Long equation 3A). (Fig. in outcomes changes health longer-term mental rela- and the warming identify multiyear to between 3B) tionship (Fig. 2007–2011 between and changes 2002–2006 (28, city-level years in approach the variation long-differences spatial the leveraging use 47), we 45, health, mental on ing Warming. Multiyear days precipitation in Data). shifts (Climatic projected magnitude smaller the than SEM. are bars Error level. 0.007, = ◦ i.5A Fig. the during warming multiyear that observe we Furthermore, specific our to sensitive not are estimates our that ensure To i.3C Fig. 0.134, = nraei aiu eprtrspoue .%point 0.8% a produces temperatures maximum in increase C .Cte ihdcae disasters declared with Cities S14). Table Appendix, SI etefcsaemr ct mn o-noersodnsand respondents low-income among acute more are effects Heat rsnsterslso siaigorlong-differences our estimating of results the presents n hw htnniatraesosre decrease a observed areas nondisaster that shows 5) hsetmt ssihl agri einclude we if larger slightly is estimate This 156). = n A 8,psil u oatnainfo added from attenuation to due possibly 78), = ipastemria fet rdcdb pitn the splitting by produced effects marginal the displays ◦ oeaieterltosi ewe tropi- between relationship the examine To nraei vrg aiu temperatures maximum average in increase C p oivsiaeteefcso utya warm- multiyear of effects the investigate To 0.014, = and n 8,wiecte htaemiss- are that cities while 78), = .Across S9). Table Appendix, SI n ipa h distribution the display and B taie h sample the stratifies α = 0.05 eut a eprilyatiual oslcinoto affected of where out selection analysis, to Katrina attributable Hurricane limitation areas. partially our This be of may time. case results the over in mental factors notable monitoring climatic is to precisely responses individual- more for health and controlling the enable characteristics track to would specific time data of over source individuals ideal same An respondents. of section needs. treatment and risks unique confer chronic psy- each and psychiatric and stress distress, specific use subclinical psychopathology, substance to Varied as change chosis. such climate symptoms, by neurobehavioral posed and risks inves- should the studies contain tigate Future would symptoms. individual’s health each measure of mental optimal severity particular and an nature the First, on information results. added our of tation and points). Katrina size (+4% Hurricane difficulties the health to one-half mental exposure of turn, occurrence between in association is, the which of points), (+2% association issues the health of size 1 the between one-quarter approximately is points) monthly average from 30 shifting mental of 25 worsened between effect to temperatures relate The warming, each outcomes. multiyear exposure health weather, cyclone short-term tropical that and show expe- they we that rienced, exposures environmental mental the of to reports difficulties individuals’ health linking temporally and spatially By with Discussion regressions placebo assignments. 100,000 group on permuted randomly estimated difference-in- distribution coefficients our the outside of from substantially estimate falls regression coefficient differences the that indicates 4.12, nondis- (coefficient: with areas compared mental points aster of 4% over occurrence approximately indi- by issues the issues increased estimate health health exposure difference-in-differences Katrina mental that The of cates period. reports same in this increase an observed l ofcetetmtsfo hspruainpoesi h itga to intervals. histogram confidence the 95% in in process points in permutation right are 95% this the with zero from process contain estimates this coefficient not from all do estimates coefficient that between The intervals difference y. in confidence 5 the years least where at more data is or periods our 1 in missing combinations cities any long-differences the missing only respectively. cities including sample, excluding our model model the the and observations, years, all including model, city. each period. from same respondents and of 2002–2006 number the between B by issues scaled health are mental Points of 2007–2011. prevalence in change the 3. Fig. lt h hnei rdcl vrg aiu eprtrsoe this over temperatures maximum average cell grid in change the plots eod u aacnito admysmld oldcross- pooled sampled, randomly a of consist data our Second, interpre- the to important are considerations of number A ◦ ntepoaiiyo etlhat ifiute (+0.5% difficulties health mental of probability the on C ogrtr amn n ae fmna elhissues. health mental of rates and warming Longer-term rneerrlnsrpeetoeS,adbu ie represent lines blue and SE, one represent lines error Orange D. ◦ C f5ywrigadtepeaec fmental of prevalence the and warming 5-y of C lt h ofcetetmtsfo h long-differences the from estimates coefficient the plots PNAS D | lt h eut fapruaino l possible all of permutation a of results the plots coe 3 2018 23, October ◦ n 30 and C p 0.007, = ◦ oaeae rae than greater averages to C | n o.115 vol. ,6,4) i.5B Fig. 1,961,743). = | o 43 no. edisplay We D. A presents | 10955

SUSTAINABILITY PSYCHOLOGICAL AND SCIENCE COGNITIVE SCIENCES ABCD health burden in the face of the acute environmental threats produced by warming in natural systems. Given the vital role that sound mental health plays in personal, social, and economic wellbeing—as well as in the ability to address pressing personal and social challenges—our findings provide added evidence that climatic changes pose substantial risks to human systems. Methods Data. We dichotomize our BRFSS measure of mental health difficulties, as recall of mental health difficulties may be susceptible to memory limitations given the interaction of psychological distress and cognitive functioning Fig. 4. Multiyear warming and prevalence of mental health problems (56), adding error to the 30-d measure (48). Supporting this, the distribu- across the seasons of the year. A plots the results of estimating the long- tion of the answers to our mental health variable presents a clustering at differences model for spring, B plots the results of estimating the long- weeks of the month as well as at 5-d intervals, although the true underly- differences model for summer, C plots the results of estimating the ing distribution is likely continuous (SI Appendix, Fig. S9). Whether or not a long-differences model for fall, and D pots the results of estimating the person experienced any mental health difficulties over the past 30 d (our long-differences model for winter. Orange error lines represent one SE, and dichotomized measure) is likely more reliable. SI Appendix, Fig. S1 plots blue lines represent 95% confidence intervals. respondents over time. We use gridded (at ∼4 km) daily temperature and precipitation from the PRISM Climate Group (57) and average daily cloud cover, relative humid- Third, we do not uncover the causally mediating factors under- ity, and wind speed from the National Centers for Environmental Prediction lying our results. Exposure to more extreme meteorological Reanalysis II project (58). Global circulation model projections use NASA conditions may produce physiological stressors that precipitate Earth Exchange’s maximum temperature and precipitation projections for 2010, 2050, and 2099 (59) from 21 of the Coupled Model Intercomparison poor mental health, such extremes may initiate inflammatory Project Phase 5 (CMIP5) models (60) run on Representative Concentration processes that worsen mental health, or the effects may run Pathway 8.5 (RCP8.5) (high-emission scenario) (61). SI Appendix, Figs. S10 entirely through reductions in health maintenance behaviors, and S11 presents the spatial and temporal distributions of these projected like exercise (50) and sleep (36). Future studies and advances changes. in climate econometric methods (45) are needed to investigate causal mediation in this setting. Pooled Cross-Section. Our relationship of interest is the marginal effect of Fourth, measurement error exists between the temperatures recent meteorological conditions on the probability of experiencing mental that we observe and the temperatures that respondents actually health issues. We model this as experienced, attenuating the magnitude of our estimates (48). Yijkts = f(AVG.TMAXijkts) + g(PRECIP.DAYSijkts) This added measurement error suggests that our temperature- [1] related estimates may represent a lower bound of the effects of + Zη + αj + µt + νks + ijkts. temperature on mental health. Fifth, we observe that the mental health of low-income indi- In this pooled cross-sectional linear probability model, i indexes individuals, viduals may be most harmed by a changing climate. However, j indexes cities, k indexes states, t indexes calendar days, and s indexes cal- our data are from a wealthy country with a temperate climate. endar years. Our dependent variable Yijkts is binary and represents whether respondents in city j in state k on calendar day t within calendar year s Regions with less-temperate climates, insufficient resources (51), reported experiencing any days of poor mental health over the 30 d before and greater reliance on ecological systems may see more severe their interview date.

effects of climate change on mental health (28). AVG.TMAXijkst and PRECIP.DAYSijkts represent the 30-d average of daily Sixth, we examine three selected types of environmental adver- maximum temperatures and number of precipitation days over the same sity likely to be produced by climate change. As a result, our esti- 30-d window as respondents’ reported mental health, respectively. We con- mates only represent a sampling of the possible risks that climate trol for average diurnal temperature range, percentage of cloud cover, change poses to mental health. Unfortunately, clear historical relative humidity, and wind speed, which are represented via Zη, as fail- analogs for other climate-induced environmental stressors—like ure to do so may bias our primary estimates (45). Bold terms in equations inundation from sea-level rise—are more difficult to measure. denote matrices. Nonetheless, it is vital for future large-scale empirical studies to investigate the many additional ways that climate change might harm mental health. AB Seventh, while robust to many tests (Methods), we cannot definitively rule out unobserved heterogeneity with respect to our long-differences and difference-in-differences estimation procedures. As a result, care is warranted with respect to the causal interpretation of these estimates. Eighth, we measure the direct effect of exposure to environ- mental stressors on self-reported mental health. Worry about cli- mate change itself may exacerbate these environmental impacts on mental wellbeing (52, 53). Moreover, interactive effects Fig. 5. Worsened mental health in areas affected by Hurricane Katrina. A between exposure to climatic stressors and other social stressors depicts the mean prevalence of mental health difficulties for the Katrina- may modify these effects. exposed and non—Katrina-exposed groups before and after Hurricane Ninth, our observed effects may not persist into the future. Katrina landfall. Locales not exposed to Katrina observed a drop in rates Humans may adapt technologically and physiologically to of mental health issues, while locales exposed to Katrina experienced an warmer climates to minimize the impact of warming on mental increase. The difference-in-differences estimate indicates that Hurricane Katrina exposure associates with an increase in prevalence of mental health health (44). Individuals may also adapt via psychological coping issues. All differences are significant at α = 0.05. Error bars are SEM. B plots mechanisms, such as avoidance, seeking social support (54), or the results of randomly permuting 100,000 treatment status assignments fostering mental preparedness (55). and conducting our difference-in-differences regression for each permuta- Ultimately, if observed relationships from the recent past per- tion. The true difference-in-differences estimate falls substantially outside sist, added climate change may amplify the society-wide mental the distribution of permutation coefficients.

10956 | www.pnas.org/cgi/doi/10.1073/pnas.1801528115 Obradovich et al. Downloaded by guest on September 23, 2021 Downloaded by guest on September 23, 2021 ii h xenlvldt forfindings. that our ways of in survey validity BRFSS external the fac- to the meteorological response respondents that limit of concerned of rates be differential number produce might the tors one by Finally, city-day. regressions each weighting on and analysis and of division, unit census one any S6 by Table driven entirely not are results respondents, state’s sam- each particular one answered our any respondents note, all Of not con- as estimation. varies, demographic the regressions and into question. of these bins effects in nonparametric inclusion size fixed with ple varying as controls con- to demographic them it our introduce robust model as are We specification, trols. variation). results effects confounding our fixed most that effects primary the fixed for of our sets trols prefer stringent less we choosing when (although even persist forms tional controls. meteorological ping temperature maximum specified Appendix additional the 1 in having falling of effect window the issues 30-d bin. represent health the model mental of this reporting over day in of temperatures estimates maximum probability daily Coefficient on of (63). distribution period full precipitation the the and using quadratic to a robust bins. with linearly. precipitation temperatures days and maximum temperature average our ing of size the Appendix varying to robust variables. meteorological main 45). (18, interest of parameters our in bias non- posttreatment Eq. omit We from in (48). variables correlation level control health serial state climatic the and mental at spatial of clustered for occurrence SEs adjust tem- heteroskedasticity-robust 30-d We maximum the 62). of (45, on effects issues days causal the precipitation as and interpreted perature be thus can estimation n similarly, and measure, outcome issues health tal period to (2002–2006) period before porally from coefficients mated aewt h ete.T oto o hs oeta ofud,w include we confounds, potential these for corre- µ control spuriously To weather. could the that with outcomes late include time health region-specific mental we or our considerations outcomes, influencing daily trends health unobserved be mental may there on more, weather of effect α the of to mates relative range day associ- precipitation issues or health temperature baselines. mental these particular recent a reporting with of ated as probability estimates in our interpret change and the variables indicator days precipitation 10 0–5 the and omit We (50). outcomes men- health and rep- tal variables meteorological are main which our between days, relationship precipitation nonlinear of bin 5-d by each resented for and bin temperature IAppendix after (SI random 62) as (45, good effects as fixed these is on precipitation conditioning and temperature maximum 30-d respectively. effects, baoihe al. et Obradovich independent and dependent periods, our separate of two mean over the mental variables take average we in outcomes, changes multiyear health and temperatures fixed maximum average full our of Differences. Long size the estimate model). binomial to negative R our a in (to in matrix tractability exists effects computational package for no effects day- knowledge, year-month, fixed substitute city-level we and model, Appendix binomial of-week, (SI negative model the of binomial In context negative the a S19). Table and in squares variable robust least outcome are ordinary days meteorological an health results both mental the cross-sectional of by pooled count the driven our using Finally, bias to use. indi- selection we collection, sample that sample of variables on threat meteorology low of a effect cating the of evidence minimal t j rpeetn oprmti iyfie fet—nEq. effects—in fixed city nonparametric —representing Furthermore, inclusion. variable control varying to robust also are estimates Our our of specification form functional varying to robust are estimates Our euesprt niao aibe o ah5 each for variables indicator separate use We oesr httm-nain iyseicfcosd o isoresti- our bias not do factors city-specific time-invariant that ensure To u miia dniyn supini httermiigvrainin variation remaining the that is assumption identifying empirical Our and ν hw htorrslsaerbs oageaigt h city-day the to aggregating to robust are results our that shows ks al S1 Table , S2 Table and S4 Fig. , nEq. in f )and () TMAX IAppendix SI IAppendix SI omdlterltosi ewe utya hne in changes multiyear between relationship the model To ersnigclna aeadsaeb-erfixed state-by-year and date calendar representing 1, ) epciey hsalw o eil siaino a of estimation flexible for allows This respectively. g(), jka hw htrslsaerbs oporsieydrop- progressively to robust are results that shows b b o the for 20–01 o ahct.Freape o u men- our for example, For city. each for (2007–2011) 4,4) u anmdlue aafo period from data uses model main Our 47). (45, f (AVG.TMAX i.S n al S3 Table and S5 Fig. , i.S6 Fig. , IAppendix SI TMAX IAppendix SI hw htrslsaerbs omodel- to robust are results that shows 1 Y hw htorrslsaentdie by driven not are results our that shows eas fterptnilt generate to potential their of because jkts ijkts ijkst IAppendix SI a and and ◦ ecalculate we , and ) o15 to C al S4 Table , IAppendix SI i.S3 Fig. , Zη IAppendix SI b ihperiod with , g(PRECIP vrtesm eid This period. same the over ◦ ◦ 0daeaemaximum average 30-d C i.S7 Fig. , aiu temperature maximum C hw htrslsare results that shows hw htrslsare results that shows hw htorfunc- our that shows Y .DAYS al S7 Table , jka i.S2 Fig. , al S5 Table , niae htour that indicates 1 as a ijkts 4) Further- (48).  IAppendix SI ijkts n 1 aln tem- falling .Teesti- The ). P rmthis from ) yusing by presents t ∈a shows Y ijkts SI SI a , , nEq. In health mental and (48): exposure design difference-in-differences disaster a using between outcomes relationship the estimate (SI variable Difference-in-Differences. outcome days health mental of that count specifications the are Appendix the using results to long-differences prefer robust our we Finally, also controls,” (45). vari- “bad controls meteorological demographic as in exclude serve changes thus, longer-term and by variables ables demographic driven these be as However, themselves spec- well. varied may as and inclusion controls demographic the these to in of robust ification changes are results measure Our externally cities. controls to of composition Bureau demographic Census US of the from results set these to varied Adding questions. is a demographics BRFSS across the using calculated persist results differences oaeu hcsadta iatrafce ra eetedn similarly trending (48). landfall were hurricane areas before affected areas nondisaster disaster to that and shocks poraneous mental of subsequent Unbiasedness residents’ outcomes. and health Katrina Hurricane with experience city’s interest, a of estimate Our terms. two otherwise. zero and otherwise. zero and city 2012) 1, February level. to state 2005 24, the at errors our cluster POST again and equations earlier from indices estimation, our found temperature of level absolute than change. rather baseline from temperature mum period either S10 Table in observations missing with cities (47). interest alter two of that outcomes our variables health between time-varying mental our unobserved temperature with maximum uncorrelated mean are periods in changes city-level that inb h ubro epnet nact o ttsia fcec and efficiency statistical for city a in errors respondents our of cluster number again, the regres- the by Eq. weight In We sion variables. issues. meteorological health other mental in changes reporting represents individuals of rates in change β with left are We regressions. cross-sectional model: long-differences pooled explicitly our the we in that way them the for in control factors city approach time-invariant long-differences for the controls Thus, implicitly confounds). potential not thus are (and nEq. In Eqs. differencing Then, the Calculating factors. period city-specific for time-invariant metrics same for control that effects Eq. In eut nteblweuto o period for equation below the in results ocnrlfrpsil isidcdb nbevdhtrgniy(48). attempt heterogeneity to unobserved by controls induced demographic bias respondent-level possible of for set control varied to Furthermore, fixed city- a factors. time-invariant date temporal of idiosyncratic for and any Appendix controls as city SI well flexibly as with more factors coupled specific specification variable This indicator effects. exposure Katrina IAppendix SI rmEq. from IAppendix SI con- could variables time-varying unobserved that given Furthermore, Appendix SI j eevdafdrldsse elrto nrsos oHriaeKatrina Hurricane to response in declaration disaster federal a received ijkts 6, 2, h iylvl ieivratfie fet (α effects fixed time-invariant city-level, the 4, Y seult n o h otKtiapro nordt fo August (from data our in period post-Katrina the for one to equal is j al S19). Table , hw htrslsaerbs ouigpretg hnei maxi- in change percentage using to robust are results that shows ijkts nee cities, indexes al S15 Table , 5 Y ersnsa niiulsmna elha nEq. in as health mental individual’s an represents jkb al S9 Table , hr euedmgahcvralsdrawn variables demographic use we where S12, Table , rvdsa siaeo utya amn nmultiyear on warming multiyear of estimate an provides al S14 Table , − Y PNAS ijkts Y POST Y Y ∆Y jka jkb jka = 2 = jk b = = ∆ k γ IAppendix SI and ijkts + hw htrslsaerbs oecuigthose excluding to robust are results that shows hw htrslsaerbs oteinclusion the to robust are results that shows β = | elvrg h adalo urcn arn to Katrina Hurricane of landfall the leverage We eut nteblwequation: below the in results POST β β nee tts and states, indexes ( ( Z jk TMAX β ψ TMAX TMAX coe 3 2018 23, October niae htrslsaerbs ouiga using to robust are results that indicates × b ∆TMAX 3 ttesaelvl nisdesof Unbiasedness level. state the at POST − TREAT eut in results ijkts Z a jkb jkb ψ jka )η ijkts + ψ eursteasmto fn contem- no of assumption the requires − + + κTREAT ijkts + jk ersnsterltosi between relationship the represents , al S11 Table , × TMAX Z Z ∆Zη + (α a: b a TREAT ersnsteitrcino these of interaction the represents η η j − + + ijkts jka α | α α α j ∆ + ijkts j j )+ ) o.115 vol. j a + + + TREAT ersnsct-ee fixed city-level represents rperiod or +   ( hw htorlong- our that shows jka  jkb jk jkb  . ijkts . . ijkts − j . | em rpout drop terms )  seult n if one to equal is jka o 43 no. b. ). IAppendix SI eretain We 1. β requires | 5, 10957 ∆Zη [6] [5] [4] [3] [2] ,

SUSTAINABILITY PSYCHOLOGICAL AND SCIENCE COGNITIVE SCIENCES However, as these demographic variables may be affected by Katrina expo- declarations as the control group. Furthermore, not all cities in the sample sure, we prefer the specifications that exclude demographic controls (45). are represented in both pre- and post-Katrina periods. SI Appendix, Table SI Appendix, Table S16 displays that our results are robust to aggregating S18 shows that results are robust to selecting respondents from only those to the city-day and weighting the regression by number of respondents cities with representation in the sample both before and after landfall. per city-day. Finally, our difference-in-differences results are robust to using the count We also explore alternative selection rules for Katrina-exposed and of mental health days outcome variable (SI Appendix, Table S19). non–Katrina-exposed groups. SI Appendix, Table S17 shows that results are robust to using only neighboring nondisaster declaration cities from ACKNOWLEDGMENTS. We thank our anonymous reviewers for their states that geographically border the states that had cities with disaster constructive feedback on this manuscript.

1. Allen J, Balfour R, Bell R, Marmot M (2014) Social determinants of mental health. Int 33. Chan EY, et al. (2018) Association between ambient temperatures and mental disor- Rev Psychiatry 26:392–407. der hospitalizations in a subtropical city: A time-series study of Hong Kong special 2. Berry HL, Waite TD, Dear KB, Capon AG, Murray V (2018) The case for systems thinking administrative region. Int J Environ Res Public Health 15:754. about climate change and mental health. Nat Clim Change 8:282–290. 34. Zivin JG, Hsiang SM, Neidell M (2018) Temperature and human capital in the short 3. Whiteford HA, et al. (2013) Global burden of disease attributable to mental and sub- and long run. J Assoc Environ Resource Economists 5:77–105. stance use disorders: Findings from the global burden of disease study 2010. Lancet 35. Baylis P, et al. (2018) Weather impacts expressed sentiment. PloS One 13:e0195750. 382:1575–1586. 36. Obradovich N, Migliorini R, Mednick SC, Fowler JH (2017) Nighttime temperature and 4. Kessler RC, et al. (2005) Lifetime prevalence and age-of-onset distributions of DSM-IV human sleep loss in a changing climate. Sci Adv 3:e1601555. disorders in the national comorbidity survey replication. Arch Gen Psychiatry 62:593– 37. Centers for Disease Control and Prevention (2017) Behavioral Risk Factor Surveillance 602. System (US Department of Health and Human Services, Centers for Disease Control 5. Kessler RC, Chiu WT, Demler O, Walters EE (2005) Prevalence, severity, and comorbid- and Prevention, Atlanta). ity of 12-month DSM-IV disorders in the national comorbidity survey replication. Arch 38. Andresen EM, Catlin TK, Wyrwich KW, Jackson-Thompson J (2003) Retest reliability Gen Psychiatry 62:617–627. of surveillance questions on health related quality of life. J Epidemiol Community 6. Moussavi S, et al. (2007) Depression, chronic diseases, and decrements in health: Health 57:339–343. Results from the world health surveys. Lancet 370:851–858. 39. Moriarty DG, Zack MM, Kobau R (2003) The centers for disease control and preven- 7. Prince M, et al. (2007) No health without mental health. Lancet 370:859–877. tion’s healthy days measures–Population tracking of perceived physical and mental 8. Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D (2003) Cost of lost productive health over time. Health Qual Life Outcomes 1:37. work time among US workers with depression. JAMA 289:3135–3144. 40. Kroenke K, et al. (2009) The phq-8 as a measure of current depression in the general 9. Rapaport MH, Clary C, Fayyad R, Endicott J (2005) Quality-of-life impairment in population. J Affective Disord 114:163–173. depressive and anxiety disorders. Am J Psychiatry 162:1171–1178. 41. Substance Abuse and Mental Health Services Administration (2016) Key Substance 10. Lewinsohn PM, Solomon A, Seeley JR, Zeiss A (2000) Clinical implications of Use and Mental Health Indicators in the United States: Results from the 2016 National “subthreshold” depressive symptoms. J Abnorm Psychol 109:345–351. Survey on Drug Use and Health, HHS Publication No. SMA 17-5044, NSDUH Series H- 11. Marshall RD, et al. (2001) Comorbidity, impairment, and suicidality in subthreshold 52 (Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental PTSD. Am J Psychiatry 158:1467–1473. Health Services Administration, Rockville, MD). 12. Glaser R, Kiecolt-Glaser JK (2005) Stress-induced immune dysfunction: Implications for 42. Fischer E, Knutti R (2015) Anthropogenic contribution to global occurrence of heavy- health. Nat Rev Immunol 5:243–251. precipitation and high-temperature extremes. Nat Clim Change 5:560–564. 13. Berry HL, Bowen K, Kjellstrom T (2010) Climate change and mental health: A causal 43. Dohrenwend BP, et al. (1992) Socioeconomic status and psychiatric disorders: The pathways framework. Int J Public Health 55:123–132. causation-selection issue. Science 255:946–952. 14. Bourque F, Cunsolo Willox A (2014) Climate change: The next challenge for public 44. Barreca A, Clay K, Deschenes O, Greenstone M, Shapiro JS (2016) Adapting to climate mental health?. Int Rev Psychiatry 26:415–422. change: The remarkable decline in the US temperature-mortality relationship over 15. Webster PJ, Holland GJ, Curry JA, Chang H-R (2005) Changes in tropical cyclone the twentieth century. J Polit Economy 124:105–159. number, duration, and intensity in a warming environment. Science 309:1844–1846. 45. Hsiang S (2016) Climate econometrics. Annu Rev Resource Econ 8:43–75. 16. Watts N, et al. (2017) The Lancet countdown: Tracking progress on health and climate 46. Smith SJ, Edmonds J, Hartin CA, Mundra A, Calvin K (2015) Near-term acceleration in change. Lancet 389:1151. the rate of temperature change. Nat Clim Change 5:333–336. 17. Burke M, Hsiang S, Miguel E (2015) Global non-linear effect of temperature on 47. Burke M, Emerick K (2016) Adaptation to climate change: Evidence from US economic production. Nature 527:235–235. agriculture. Am Econ J Econ Policy 8:106–140. 18. Hsiang SM, Burke M, Miguel E (2013) Quantifying the influence of climate on human 48. Wooldridge JM (2010) Econometric Analysis of Cross Section and Panel Data (MIT conflict. Science 341:1235367. Press, Cambridge, MA). 19. Bohra-Mishra P, Oppenheimer M, Hsiang SM (2014) Nonlinear permanent migration 49. Lambrew JM, Shalala DE (2006) Federal health policy response to Hurricane Katrina: response to climatic variations but minimal response to disasters. Proc Natl Acad Sci What it was and what it could have been. JAMA 296:1394–1397. USA 111:9780–9785. 50. Obradovich N, Fowler JH (2017) Climate change may alter human physical activity 20. Cunsolo A, Ellis NR (2018) Ecological grief as a mental health response to climate patterns. Nat Hum Behav 1:0097. change-related loss. Nat Clim Change 8:275–281. 51. Bhugra D, et al. (2017) The WPA-Lancet psychiatry commission on the future of 21. Clayton S, Manning C, Hodge C (2014) Beyond Storms & Droughts: The Psychologi- psychiatry. Lancet Psychiatry 4:775–818. cal Impacts of Climate Change (American Psychological Association and ecoAmerica, 52. Searle K, Gow K (2010) Do concerns about climate change lead to distress? Int J Clim Washington, DC). Change Strateg Management 2:362–379. 22. Kessler RC, et al. (2008) Trends in mental illness and suicidality after Hurricane Katrina. 53. Clayton S, Manning C, Krygsman K, Speiser M (2017) Mental Health and Our Chang- Mol Psychiatry 13:374–384. ing Climate: Impacts, Implications, and Guidance (American Psychological Association 23. La Greca AM, Silverman WK, Lai B, Jaccard J (2010) Hurricane-related exposure experi- and ecoAmerica, Washington, DC). ences and stressors, other life events, and social support: Concurrent and prospective 54. Kaniasty K (2012) Predicting social psychological well-being following trauma: The impact on children’s persistent posttraumatic stress symptoms. J Consulting Clin role of postdisaster social support. Psychol Trauma Theor Res Pract Policy 4:22–33. Psychol 78:794–805. 55. Reser JP, Swim JK (2011) Adapting to and coping with the threat and impacts of 24. Pietrzak RH, et al. (2012) Resilience in the face of disaster: Prevalence and longitudinal climate change. Am Psychol 66:277. course of mental disorders following Hurricane Ike. PLoS One 7:e38964. 56. Kizilbash AH, Vanderploeg RD, Curtiss G (2002) The effects of depression and anxiety 25. Bei B, et al. (2013) A prospective study of the impact of floods on the mental and on memory performance. Arch Clin Neuropsychol 17:57–67. physical health of older adults. Aging Ment Health 17:992–1002. 57. Di Luzio M, Johnson GL, Daly C, Eischeid JK, Arnold JG (2008) Constructing retro- 26. Boscarino JA, et al. (2013) Mental health outcomes at the Jersey shore after hurricane spective gridded daily precipitation and temperature datasets for the conterminous sandy. Int J Emerg Ment Health 15:147–158. United States. J Appl Meteorol Climatology 47:475–497. 27. Jermacane D, et al. (2018) The English national cohort study of flooding and health: 58. Kanamitsu M, et al. (2002) NCEP–DOE AMIP-II reanalysis (R-2). Bull Am Meteorol Soc The change in the prevalence of psychological morbidity at year two. BMC Public 83:1631–1644. Health 18:330. 59. Thrasher B, Maurer EP, McKellar C, Duffy P (2012) Technical note: Bias correcting 28. Carleton TA (2017) Crop-damaging temperatures increase suicide rates in India. Proc climate model simulated daily temperature extremes with quantile mapping. Hydrol Natl Acad Sci USA 114:8746–8751. Earth Syst Sci 16:3309–3314. 29. Hanigan IC, Butler CD, Kokic PN, Hutchinson MF (2012) Suicide and drought in New 60. Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment South Wales, Australia, 1970–2007. Proc Natl Acad Sci USA 109:13950–13955. design. Bull Am Meteorol Soc 93:485–498. 30. Burke M, et al. (2018) Higher temperatures increase suicide rates in the United States 61. Riahi K, et al. (2011) RCP 8.5—a scenario of comparatively high greenhouse gas and Mexico. Nat Clim Change 8:723–729. emissions. Clim Change 109:33–57. 31. Hansen A, et al. (2008) The effect of heat waves on mental health in a temperate 62. Carleton TA, Hsiang SM (2016) Social and economic impacts of climate. Science Australian city. Environ Health Perspect 116:1369–1375. 353:aad9837. 32. Wang X, Lavigne E, Ouellette-kuntz H, Chen BE (2014) Acute impacts of extreme tem- 63. Deschenesˆ O, Greenstone M (2011) Climate change, mortality, and adaptation: perature exposure on emergency room admissions related to mental and behavior Evidence from annual fluctuations in weather in the US. Am Econ J Appl Econ disorders in Toronto, Canada. J Affective Disord 155:154–161. 3:152–185.

10958 | www.pnas.org/cgi/doi/10.1073/pnas.1801528115 Obradovich et al. Downloaded by guest on September 23, 2021