Climate Change and Children's Health and Well-Being in the United States

I

Appendix B: Supplemental Information for
Analyses in the Extreme Heat Chapter

This appendix describes methods, data sources, and assumptions for the extreme heat
analyses presented in Chapter 3 of the main report. First is the information for the detailed analysis
of how heat experienced during the school year affects learning among children. Second is
information required for the discussion of emergency department visits to children's hospital
associated with high heat days.

Detailed Analysis of Heat and Learning
Losses

This section describes the detailed analysis of learning losses associated with heat
presented in Chapter 3 of the main report.

SUMMARY OF STUDIES USED IN THIS ANALYSIS

PARK ETAL. (2020)1

Park et al. investigated how heat inhibits learning among students in the U.S. and how air
conditioning (A/C) mitigates those effects. Learning was measured using test results among a
nationwide sample of 10 million PSAT takers, generally 10th and 11th grade students, which the
authors believe is a true indicator of cognitive performance as opposed to generalized intelligence.
Heat exposure was modeled using the heat experienced throughout the school year preceding the
test date, as measured by the National Oceanic and Atmospheric Administration's Daily Global
Historical Climatology Network. With these data, the authors developed a regression model to
estimate a causal impact of cumulative heat exposure on test scores by studying the variation among
individual students who take the PSAT at least twice. By employing "student fixed effects," the
authors controlled for many unobservable characteristics of individual students that may influence
test scores and therefore more readily isolate the effect of heat. The authors offered several
important findings. The detailed analysis presented in this report leverages results demonstrating
that school A/C reduces some of the effects of heat on learning losses. Using student and guidance
counselor responses to a survey about the availability of A/C in schools, the authors develop and
utilize these data to show that learning losses are significantly reduced in schools with A/C. When
considering the additional relief of having A/C at home, learning losses are mitigated almost entirely
on hot days. Table 1 describes the specific coefficients leveraged from the paper for this detailed
analysis (as well as the standard errors reported with the results).

CHETTY ETAL. (2011)2

Chetty et al. estimated the value of learning loss per standard deviation by studying a cohort of
11,571 students and their teachers from 1985 to 1989, as part of the Student/Teacher Achievement
Ratio (STAR) Project. The STAR Project empowered 79 schools in Tennessee to randomly assign

April 2023

1


-------
Climate Change and Children's Health and Well-Being in the United States

kindergarten students to different classrooms within their schools, some large in class size, some
small, where students remained until the 4th grade. The STAR Project oversampled lower-income
schools in Tennessee; thus, the socioeconomic characteristics of the STAR sample can be considered
as poorer (lower income) than U.S. averages. Chetty et al. were able to track data from tax returns of
95% of STAR Project participants 25-27 years old to observe socioeconomic outcomes in adulthood.
Chetty et al. found that randomly assigning students to a classroom that is one standard deviation
better than average in kindergarten generates an increase in earnings at ages 25-27 of $1,520
(+9.6%) per year for each student. This translates into lifetime earnings gains of approximately
$39,200(2009 dollars).

LEROY ETAL. (2021)3

LeRoy et al. estimated the cost of installing A/C in all public schools in the contiguous U.S. The
authors used a phone call campaign and news article research for a sample of schools/districts to
identify geographic regions with facilities already equipped with HVAC. Observing average
temperature data in those HVAC-equipped regions from 1955-1984 (baseline period), they
calculated the number of school-year days (32) above 80°F during the baseline period were needed
to justify current existence of HVAC in a school/district. The researchers assumed that all public
schools in the contiguous 48 states with 32 school-year days above 80°F during the baseline period
should already have HVAC, and thus would not be included in the "installations" cost estimate;
meanwhile, schools with fewer than 32 days above 80°F during the base period, but with at least 32
days during the study period - a 30-year average centered on 2025 - would be in the "installations"
cost estimate pool. Using RSMeans to find cost estimates for Department of Energy Standards-
compliant HVAC systems, the authors found the total cost of furnishing all contiguous U.S. public
schools with A/C to be about $42.4 billion, composed of $40.5 billion in new installations, $414.8
million in upgrades of existing HVAC technology, and $1.5 billion in annual operating and
maintenance costs.

ANALYSIS STEPS

The detailed analysis of the effects of heat on learning relies on the quantitative estimates presented
by Park et al., the most recent and nationally representative analysis of this relationship. The results
of the analysis describe learning losses experienced by a cohort of students graduating from high
school each year (age 17). The learning losses observed in late high school are interpreted as the
result of cumulative heat exposure across all preceding school years. Because Park et al. show the
importance of A/C in mitigating the impacts of heat, this analysis incorporates "baseline" A/C
investments consistent with the data used in Park et al. to present how heat will affect learning if A/C
access does not improve throughout the 21st century. In other words, the results are consistent with
a "no additional adaptation" approach.

Learning losses are valued in terms of lost future income using estimates from Chetty et al., as they
appear in Park et al. Applying the income losses from Chetty et al. allows us to consider total lost
income per child as well as the annual income losses across all graduating children in a given year.

April 2023

2


-------
Climate Change and Children's Health and Well-Being in the United States

Lost income associated with learning losses is then compared to the cost of installing A/C in schools
estimated by LeRoy et al.

Table 1 that follows details the analytic steps, data sources, and assumptions used to forecast and
value learning losses among children resulting from climate change. This analysis considers impacts
across all census tracts in the continental U.S. (over 72,000 tracts).

April 2023

3


-------
Climate Change and Children's Health and Well-Being in the United States

Table 1: Analytic Steps in Climate Change Impacts on Heat and Learning Among Students

Step

Data

Methods, Assumptions, and Notes

Baseline Risks

la. Determine baseline
school A/C and home A/C
levels

School A/C: extracted from
Figure 5, Panel A of the paper at
the county level (student
reports of hot classrooms)

Home A/C: provided bv authors
of Park et al. at the county/year
level for the years 1997-2011

School A/C: Given the proprietary nature of this data, the authors of Park et al.
were unable to share. Instead, the information from a county-level map displayed in
the paper (Figure 5, panel A, Park et al.) was extracted, which presents the fraction
of hot days when classrooms get too hot, as reported by students. The inverse of
this value (1-fraction of days when classrooms are too hot) is used as the percent of
schools with A/C access. Because the data is displayed as a range in the figure, the
midpoint of each range (e.g., 0-10 percent range=5 percent) is assumed. For
counties with missing data, state-level averages are applied.

Home A/C: Data are averaged across years. For counties with missing data, state-
level averages are applied.

Both school and home A/C are converted to values between 0 and 1, interpreted as
the portion of schools and homes with A/C.

lb. Identify baseline
"learning gains" each
academic year

Presented in Park et al.

Park et al. report the average gain in PSAT score performance between the 10th and
11th grades is about 0.3 standard deviations. This forms the basis of comparison for
future learning losses predicted in Step 3.

Future Climate Stressor

2. Calculate future average
daily maximum
temperatures during school
years

LOCA future climate data at the
census tract level

Start- and end-dates in school
year (provided by study authors
at state level)

Consistent with Park et al., future mean daily maximum temperature from all days
within the school year among the 365 days preceding the PSAT date (assumed
October 15 for all locations) are calculated. For example, for year 2021, and
assuming a school year that starts September 1 and ends May 31, the analysis takes
the average daily max temperature between October 15, 2020, and October 14,
2021, for all days not in June, July, August.

To remove the influence of time not spent in school, the school year start- and end-
dates provided by the study authors are used to align the date range with locally
appropriate school calendars; see Figure Al of the Supplemental Materials to Park
et al. for details. This approach assumes that the school calendar will not change in
response to future warming.

April 2023

4


-------
Climate Change and Children's Health and Well-Being in the United States

Step

Data

Methods, Assumptions, and Notes



3a. Estimate future learning
losses per student at the
census tract level under
baseline A/C conditions

Learning impact function from
Park et al. (Table 5, specification
2 of the study)

The following relationship is used to forecast learning losses under future heat and
baseline A/C levels*:

(-0.569*temp) + (0.235*temp*schoolAC) + (0.324*temp*homeAC)

Because A/C investments are held constant, only the temperature variable changes
in the future (see Step 2). Consistent with Park et al., the results are interpreted as
hundredths of a standard deviation in learning loss. In other words, the resulting
value is divided by 100 to calculate a measure of standard deviations.

c
0)

i_

2

!E
u
c
o





Future learning losses experience at each degree of global warming are compared
with both losses experienced in the baseline time-period as well as relative to
average learning "gains" each school year associated with learning progress in
general (see Step lb).

+¦»
u
OJ

£
LU

0)

3

3
u_

3b. Value learning losses
per student in terms of lost
future income

Valuation from Chetty et al., as
reported in Park et al.

See Chapter 2 of the main
report and Appendix A for
details on population methods
and data sources used
throughout the analysis.

As described in Park et al, Chetty et al. finds that having a teacher who raises test
scores by 0.1 standard deviations results in net present value of $7,000 (2009
dollars) in future increased earnings for current 12-year-olds. Park et al. apply a 5
percent discount rate to generate a net present value estimate of $8,500 (2009) for
the typical 16-year-old. This analysis uses Park et al.'s $8,500 values but adjusts to
2021 dollars using the GDP deflator to arrive at $11,000 in lost future income per
student per 0.1 standard deviations in learning losses observed in high school.



3c. Aggregate lost future
income across all
graduating students in a
given year

See Chapter 2 of the main
report and Appendix A for
details on population methods
and data sources used
throughout the analysis.

Total learning losses among a cohort of graduating students are aggregated using
the total number of graduating students in a given census tract. The total number of
children in the 14-17 age cohort are divided by 4 to approximate the number of
students graduating in each future year. (Note: In 2010, there were approximately
4.3 million children aged 17 in CONUS. By 2099, there are expected to be 6.7 million
children in the same age cohort, representing a 56 percent increase in population.)

Note: The coefficients leveraged for this analysis are all statistically significant at conventional levels. The standard errors are 0.104, 0.070, and 0.110 for the
coefficients on temp, temp*schoolAC, and temp*homeAC, respectively.

April 2023

5


-------
Climate Change and Children's Health and Well-Being in the United States

FUTURE HEAT AND BASELINE A/C

Table 2 presents baseline and future temperatures during school years by degree of global warming.
The spatial distribution of these temperature increases at 2°C and 4°C of global warming relative to
baseline levels of heat is showcased in Figure 1. Finally, Figure 2 describes the spatial distribution of
baseline school and home A/C coverage used in the analysis.

Table 2: Future National Average Maximum Daily Temperature During School Years (°F)





Increase in Projected

Degree of Global

Projected

Temperature

Warming (°C)

Temperature (°F)

Relative to Baseline
Temperature (°F)

Baseline

63.9

1°C	66.8	2.9

2°C

69.7

5.8

3°C

72.1

8.2

4°C

73.9

10.0

5°C	76.3	12.4

Notes: National averages incorporate state-specific school years and are
weighted based on populations. The baseline aligns with averages from

1986-2005.

April 2023

6


-------
Climate Change arid Children's Health and Well-Being in the United States

Figure 1: Future Average Maximum Daily Temperature During School Years (°F) at 2°C and 4°C of
Global Warming Relative to Baseline

2°C of Global Warming

4°C of Global Warming

0 1-4 5-6 7-8 9- 10 *11-15
Note: These maps describe the number of degrees (in Fahrenheit) above baseline levels during state-specific school
years projected for each degree of global warming listed. Darker shading describes larger increases, lighter shading
describes smaller increases. See Step 2 of Table 1 for additional details.

April 2023

7


-------
Climate Change arid Children's Health and Well-Being in the United States

Figure 2: Baseline School A/C (top) and Home A/C (bottom) Market Penetration (%)

0% - 20% 20% - 40% 40% - 60% ¦ 60% - 80% ¦ 80% - 100%

Note: These maps describe A/C market penetration identified in Park et al. and used in the detail analyses
presented in this report. See Step la of Table 1 for additional details.

April 2023

8


-------
Climate Change and Children's Health and Well-Being in the United States

EFFECTS ON CHILDREN RESULTS

Table 4 presents the national average learning losses per child in terms of standard deviations,
percent reduction relative to average gain in test scores per year, and how these results translate
into lost future income. Table 5 aggregates across graduating high school students to express lost
annual income at each degree of global warming under different population assumptions.

Table 4: Projected National Average Learning Losses Per Child

Degree of
Global

(1)

National Average Learning

(2)

National Average Learning

(3)

National Average Lost

Warming

Losses Per Child

Losses Per Child

Future Income Per Child

(°C)

(Standard Deviations)

(%)

Per Year





(2021 Dollars)

rc

0.0062

2.1%

$677

(-0.0020 to 0.0156)

(-0.7% to 5.2%)

(-$222 to $1710)

2°C

0.0119

4.0%

$1,310

(0.0034 to 0.0217)

(1.1% to 7.2%)

($378 to $2,390)

3°C

0.0173

5.8%

$1,900

(0.0090 to 0.0269)

(3.0% to 9.0%)

($988 to $2,960)

4°C

0.0213

7.1%

$2,340

(0.0140 to 0.0290)

(4.7% to 9.7%)

($1,550 to $3,190)

5°C

0.0265

8.8%

$2,920

Note: All estimates presented in columns (1) and (3) are incremental relative to baseline learning losses

associated with heat exposure. Averages per student are population weighted and assume population growth.
The learning losses in column (2) are presented in terms of a "percent" relative to 0.3 standard deviations
(average gain in PSATscore performance between 10th and 11th grades, see Table 1). The table displays the
average and range across climate models; a range for 5°C is not feasible because only one climate model reaches

this temperature threshold before 2100.

Table 5: Projected Total Lost Annual Income Related to Learning Losses Across Graduating High
School Students (Billion 2021 Dollars)

Degree of Global Warming (°C)

With Population Growth

Constant 2010 Population

rc

$3.2

$3.0



(-$1.0 to $8.2)

(-$0.9 to $7.4)

2°C

$6.9

$5.8



($1.9 to $12.7)

($1.7 to $10.6)

3°C

$10.6

$8.5



($5.6 to $16.6)

($4.5 to $13.2)

4°C

$13.4

$10.5



($8.9 to $18.3)

($7.0 to $14.2)

5°C

$17.1

$13.0

Note: All estimates presented are incremental relative to baseline learning losses associated with heat
exposure. See Table lfor additional details. The table displays the average and range across climate models; a
range for 5°C is not feasible because only one climate model reaches this temperature threshold before 2100.

April 2023

9


-------
Climate Change and Children's Health and Well-Being in the United States

Figure 3 presents the spatial distribution of learning losses per child at 2°C and 4°C of global
warming. The five states with largest impacts per child are outlined in black and listed below each
map.

Table 6 and Table 7 then follow with the impacts per child for each state at 2°C and 4°C of global
warming to provide perspective on the range of impacts across states, although there can be
considerable heterogeneity within states (see Figure 3).

Figure 4 shows the total lost income from learning losses across graduating students in each census
tract. Impacts are generally highest in areas with large children's populations. The five states with
largest total impacts are outlined in black and listed below each map. The relevant quantities or rates
presented in each figure are provided in parentheses after the state name in the lists of top 5 states.

April 2023

10


-------
Climate Change arid Children's Health and Well-Being in the United States

Figure 3: Projected Lost Future income Per Child Per Year Associated with Learning Losses from
Heat Exposure

2°C of Global Warming

Top five states: VT ($3,050), ME ($2,760), NH ($2,630), WY ($2,320), Ml ($2,230)

4°C of Global Warming

Top five states: VT ($5,520), ME ($4,980), NH ($4,620), WY ($4,280), MT ($4,060)

$0 $1 -31,450 $1,451-$2,350 $2,351 - $3,250 $3,251 - $4,300 ¦ $4,301 - $7,300

Note: These maps describe projected future lost income per child per year associated with learning losses from heat
exposure at 2°C and 4°C of global warming (expressed in 2021 dollars). The five states with the highest average
impacts are outlined in black (see Tables 6 and 7 for related details). See Table 1 for analytic details.

April 2023

11


-------
Climate Change and Children's Health and Well-Being in the United States

Table 6: Projected Lost Future Income Per Child Per Year Associated with Learning Losses by State
with 2°C Global Warming (with Population Growth)

State

Lost Future
Income Per
Child

State

Lost Future
Income Per Child

Vermont

$3,050

New Mexico

$1,330

Maine

$2,760

Kentucky

$1,300

New Hampshire

$2,630

Delaware

$1,290

Wyoming

$2,320

California

$1,270

Michigan

$2,230

North Carolina

$1,270

West Virginia

$2,230

Virginia

$1,250

Colorado

$2,150

Iowa

$1,240

Massachusetts

$2,140

Illinois

$1,240

Wisconsin

$2,100

Maryland

$1,160

Rhode Island

$2,090

Mississippi

$1,050

New York

$1,990

Georgia

$1,020

Montana

$1,970

Missouri

$1,010

Idaho

$1,960

South Carolina

$984

Pennsylvania

$1,890

Tennessee

$959

Utah

$1,860

Arkansas

$929

Ohio

$1,820

Nevada

$891

Washington

$1,800

Kansas

$885

Minnesota

$1,760

Arizona

$858

Connecticut

$1,740

Alabama

$856

North Dakota

$1,620

Oklahoma

$747

Oregon

$1,560

Nebraska

$733

Washington, DC

$1,550

Texas

$651

South Dakota

$1,500

Louisiana

$625

Indiana

$1,410

Florida

$538

New Jersey

$1,330



Notes: This table describes average future lost income per child per year at
2°C of global warming using the methods described in Table 1 averaged to
the state level (2021 dollars). States are listed from largest to smallest
impacts.

April 2023

12


-------
Climate Change and Children's Health and Well-Being in the United States

Table 7: Projected Lost Future Income Per Child Per Year Associated with Learning Losses by State
with 40C Global Warming (with Population Growth)

State

Lost Future
Income Per
Child

State

Lost Future
Income Per Child

Vermont

$5,520

New Jersey

$2,460

Maine

$4,980

New Mexico

$2,430

New Hampshire

$4,620

Delaware

$2,310

Wyoming

$4,280

Kentucky

$2,280

Montana

$4,060

Virginia

$2,180

Rhode Island

$4,040

North Carolina

$2,160

Michigan

$3,930

Illinois

$2,150

Colorado

$3,900

Maryland

$2,060

West Virginia

$3,900

Iowa

$2,050

Massachusetts

$3,890

Georgia

$1,750

Idaho

$3,880

Mississippi

$1,750

Washington

$3,790

South Carolina

$1,710

Wisconsin

$3,610

Tennessee

$1,660

New York

$3,560

Missouri

$1,650

Oregon

$3,420

Nevada

$1,600

Pennsylvania

$3,380

Arizona

$1,590

Utah

$3,370

Arkansas

$1,510

North Dakota

$3,250

Alabama

$1,480

Ohio

$3,190

Kansas

$1,400

Connecticut

$3,170

Nebraska

$1,270

Minnesota

$3,120

Oklahoma

$1,200

Washington, DC

$2,760

Texas

$1,070

South Dakota

$2,630

Louisiana

$1,030

California

$2,490

Florida

$924

Indiana

$2,480

--

Notes: This table describes average future lost income per child per year at
4°C of global warming using the methods described in Table 1 averaged to
the state level (2021 dollars). States are listed from largest to smallest
impacts.

April 2023

13


-------
Climate Change arid Children's Health and Well-Being in the United States

Figure 4: Total Lost Annual Income Across Graduating Students with 2°C and 4°C Global Warming
(with Population Growth)

2°C of Global Warming

Top five states: CA ($997 mil), NY ($679 mil), TX ($349 mil), Ml ($337 mil), PA ($325 mil)

4°C of Global Warming

Top five states: CA ($2.2 bil), NY ($1.3 bil), TX ($646 mil), Ml ($606 mil), PA ($575 mil)

$0 $1-$64,000 $64,001 -5140,000 $140,001 - $250,000 $250,001 - $500,000 H $500,001 - $7,200,000
Note: These maps describe projected lost annual income across each cohort of graduating high school students
associated with learning losses from heat exposure at 2°C and 4°C of global warming (expressed in 2021 dollars).
The five states with the highest impacts are outlined in black. See Table lfor analytic details.

April 2023

14


-------
Climate Change and Children's Health and Well-Being in the United States

Table 7 demonstrates hypothetical future learning losses under different scenarios featuring
increased installation of A/C in schools: a 10% increase in school A/C, 20% increase in school A/C, and
complete A/C coverage in schools.

Table 7: National Average Learning Losses Per Child Under Alternate Future School A/C Coverage
Relative to Baseline A/C Coverage and Temperatures

Degree of Global

10% Increase in School A/C

20% Increase in School A/C

Complete A/C Coverage in
Schools

Warming (°C)

Overall

Baseline
Corrected

Overall

Baseline
Corrected

Overall

Baseline
Corrected

Baseline

0.147



0.147



0.147



1°C

0.137

-0.010

0.122

-0.025

0.090

-0.057

2°C

0.142

-0.005

0.126

-0.021

0.093

-0.054

3°C

0.147

0.000

0.130

-0.017

0.096

-0.050

4°C

0.150

0.004

0.133

-0.014

0.099

-0.048

5°C

0.155

0.008

0.137

-0.010

0.102

-0.045

Note: This table provides hypothetical future learning losses under different scenarios featuring increased
installation of A/C in schools. The baseline considers both baseline temperature (relative to 1986-2005) as well as
baseline school A/C (see Table 1). The cells in gray demonstrate future learning "gains" relative to these baseline
levels.

LIMITATIONS

Below are several limitations of the detailed analysis of heat exposure on learning losses.

1.	PSAT scores may not be representative oflearning losses for students who do not plan to
attend college. Park et al. argue that PSAT scores are a suitable proxy for "learning" because
the test is not designed to measure overall intelligence of IQ. However, it may be a
subsample of high school students that elect to take the PSAT, especially college-bound
students. It is uncertain whether the learning losses measured by PSAT scores are necessarily
transferrable to students who do not take the PSAT.

2.	Valuation of learning losses relies on one study from Tennessee. I n o rde r to offe r a tota I
magnitude of learning losses across students, learning losses are valued using income losses
associated with standard deviations in learning losses using the same approach in Park et al.,
which transfers the value from Chetty et al. The income loss value in Chetty et al., however, is
drawn from a relatively small sample of students in Tennessee that over-samples among
poorer students. It is uncertain if the income losses observed in Chetty et al. and used by
Park et al. are necessarily transferrable to all students across the U.S.

3.	School and home A/C levels are observed at high level ofspatial granularity. The school A/C
data used in Park et al. are at the school level and reflects considerable variation across

April 2023

15


-------
Climate Change and Children's Health and Well-Being in the United States

space. The data are considered confidential and are not available for use in this study.

Instead, the analysis in this report relies on county-level information as presented in a map in
Park et al. which likely masks within-county variation in access to school A/C. Home A/C is
modeled in Park et al. and was provided by the authors at the county level. To the extent that
access to A/C significantly varies within counties, this analysis will not be able to decipher
these differences.

4.	Survey-based responses about school A/C penetration may be imperfect measures. Park et al.
rely on survey-based responses from students and staff at schools to predict A/C coverage.
While the authors determined this metric to be a good indicator on how well schools are air
conditioned, it relies on subjective information about whether classrooms were too hot and
at a single point in time.

5.	Cumulative learning losses maybe under-estimated. Pa rk et a I. demonstrate that heat
experienced in one year can affect PSAT scores four years later (see Table 4 of Park et al.),
although do not offer this relationship controlling for school and home A/C. The study
authors conclude that the cumulative effect of heat experienced in these four years is three
to five times higherthan the impact of the one school year prior. Given Park et al.'s findings,
the approach used in this report is likely to under-estimate total learning losses associated
with all heat exposure experienced during a child's time in grade school. Similarly, Roach and
Whitney provide evidence that learning losses associated with heat can start very early.

6.	Unable to differentiate between public and private schools. This study considers the impacts
on all students, although there may be important differences between students who attend
public vs. private schools. Moreover, the LeRoy et al. study that estimates the cost of
installing A/C is specifically focused on infrastructure gaps in public schools. Given less than
10% of school-aged students likely attend private school4 (U.S. Department of Education
2019) and the more considerable difference between the lost income from learning losses
and cost of installing A/C, the potential over-inclusion of private school students in our
valuation estimate is not concerning.

7.	A verage tempera tures during the school year may not be the best indica tor of hea t tha t
reduces learning: Park et al. find a statistically significant relationship between average
maximum temperature across the school year and learning. However, it is possible there is a
better indicator of school buildings being too hot for productive learning.

DATA SOURCES

Table 8: Summary of Data Sources Used in Heat and Learning Analysis

Data Type

Description

Data Documentation and
Availability

Historical

temperature

data

Daily temperature data from 3,000 weather stations
covered by the National Oceanic and Atmospheric
Administration's Daily Global Historical Climatology
Network between 1996 and 2014

Global Historical Climatology
Network dailv (GHCNd)

April 2023

16


-------
Climate Change and Children's Health and Well-Being in the United States

Data Type

Description

Data Documentation and
Availability

Baseline A/C
coverage in
schools

Generated by Park et al. using data from surveys
administered by the College Board to students and
guidance counselors at high schools across the U.S.
(provided to the study authors via a data use agreement
with the College Board, see Park et al. for details)

Data not publicly available

Baseline A/C
coverage in
homes

Generated by Park et al. using the 1980 decennial Census
(county-year level) with data on changes in penetration
overtime (census-region level) from the Energy
Information Agencies Residential Energy Consumption
surveys (see Park et al. for details)

1980 Census of Population and
Housing

Residential Enerev Consumption
Survev (RECS)

PSAT scores

Park et al. used test scores (math and reading) from the
universe of PSAT-takers from high school classes
between 2001 and 2014 (provided to the study authors
via a data use agreement with the College Board, see
Park et al. for details)

Data not publicly available

Income losses

Originally derived from Chetty et al. and adjusted in Park
et al. Converted to 2021 dollar-year using the U.S. Bureau
of Economic Analysis's "National Income and Product
Accounts" product

U.S. Bureau of Economic
Analvsis's "National Income and

Product Accounts" (Table 1.1.9
in the Section 1)

Future heat
during school
years

See Appendix A for data sources



Future

population of
children

See Appendix A for data sources



Note: See Park et al. for other data sources used in the underlying study, including demographic and other
control variables not leveraged for projection purposes.

,n B Heat and Emergency Department Visits
C3—I

1	1 Chapter 3 highlights research linking daily maximum temperature and the incidence of

ED visits among a sample of children's hospitals across the U.S. from Bernstein et al.
(2022).5 The study authors apply a two-stage analytic approach to estimate the association between
day-to-day variation in maximum temperature and the relative risk of emergency department visits
for the 47 children's hospitals in their sample, adjusting for temporal trends. Extreme heat was
associated with a relative risk (RR) of all-cause ED visits of 1.17 over the next 7 days relative to
hospital-specific minimum morbidity temperature (MMT) and the 95th percentile of the
temperature distribution. Given the geographic variation of the hospitals in their sample, especially
relative to existing literature studying similar relationships, the authors offer their results as
generalizable to other geographies and hospitals. The analysis in this report extrapolates the findings
from Bernstein et al. to other children's hospitals with emergency departments.

April 2023

17


-------
Climate Change and Children's Health and Well-Being in the United States

To determine the locations of all children's hospitals in the U.S. with emergency departments, and
characteristics similarto those in the study sample, the authors of the study were consulted to help
determine which hospitals among all 260 listed in a directory from the Children's Hospital
Association met these criteria.6 The authors identified 38 hospitals that are not pediatric acute care
facilities (e.g., burn hospitals, rehab hospitals, mental health facilities) and which do not have an ED.
The final list of relevant hospitals includes all other children's hospitals (222 total).

Healthcare Cost and Utilization Project's Kids' Impatient Database (HCUP KID), a database of hospital
admissions data for kids, was then leveraged for two recent years (2016 and 2019) to approximate
the baseline number of ED visit incidence among children. Following the below-listed steps revealed
that children's hospitals see approximately 22,000 children in their EDs each day in summer months:

1.	Identify all hospital admissions associated with children's hospital (KID_STRATUM variable)

2.	Identify all admissions with evidence of ED services in the discharge record (HCUP_ED
variable)

3.	Retain admissions for May through September, consistent with the months included in the
original study

4.	Apply the sample weighting approach described in HCUP (2005)7 to establish national counts

5.	Estimate the number of ED visits indirectly through a ratio of hospital admissions to all ED
visits (including those that do not result in hospital admissions) by dividing the total observed
hospital admissions in HCUP KID by 3.3% (McDermott et al., Table l8)

6.	Average the findings from 2016 and 2019 to produce an estimate of total annual number of
ED visits at children's hospitals between May and September

Information was not available to identify the MMT and 95th percentile for the locations of the 222
children's hospitals, which limited the application of Bernstein et al.'s findings to climate projections
used elsewhere in this chapter. Instead, the analysis assumes that the national average MMT is
approximately 62°F (drawn from Figure SI of Bernstein et al.) and the 95th percentile is
approximately 95°F. If the RR between those two temperature levels is 1.17, then that assumes a
0.5% increase in ED visits per degree between those two temperatures. Therefore, if temperatures
increase 1°F for each of the 153 days between May and September, this analysis estimates an
additional 113 ED visits at children's hospitals per day, or 17,340 total visits throughout the summer
months.

DATA SOURCES

Table 9: Summary of Data Sources Used in Heat and Emergency Department Visits Analysis

Data Type

Description

Data Documentation and
Availability

Children's
hospitals

Population and location of all children's hospitals in the
U.S. Discussions with the authors of Bernstein et al.
helped to narrow the dataset to only those hospitals
that included an ED.

Children's Hospital Association
directory

April 2023

18


-------
Climate Change and Children's Health and Well-Being in the United States

Data Type

Description

Data Documentation and
Availability

Baseline number
of ED visits at
children's
hospitals

Identified hospitalizations admitted through the ED.
These data were then "inflated" to correct for the
number of ED visits that did not result in hospitalization
using information presented in McDermott et al.

Healthcare Cost and Utilization
Project's Kids' Impatient
Database (2016 and 2019)

Note: This table only provides the data sources used in the analysis presented in this report. For the underlying
data sources used in Bernstein et al., see the paper.

April 2023

19


-------
Climate Change and Children's Health and Well-Being in the United States

References

1	Park, R.J., Goodman, J., Hurwitz, M. and Smith, J., 2020. Heat and learning. American Economic Journal:
Economic Policy, 12(2), pp.306-39.

2	Chetty, R., Friedman, J.N., Hilger, N., Saez, E., Schanzenbach, D.W. and Yagan, D., 2011. How does your
kindergarten classroom affect your earnings? Evidence from Project STAR. The Quarterly Journal of Economics,
126(A), pp. 1593-1660.

3	LeRoy, S., Matthews, M., Wiles R., 2021. Hotter Days, Higher Costs: The Cooling Crisis in America's Classrooms.
The Center for Climate Integrity. https://coolingcrisis.org/uploads/media/HotterDaysHigherCosts-CCI-
September2021.pdf

4	U.S. Department of Education, National Center for Education Statistics. 2019. "Public and private school
comparison." Available at: https://nces.ed.gov/fastfacts/display.asp?id=55. Accessed on September 16, 2022.

5	Bernstein, A.S., Sun, S., Weinberger, K.R., Spangler, K.R., Sheffield, P.E. and Wellenius, G.A., 2022. Warm Season
and Emergency Department Visits to US Children's Hospitals. Environmental Health Perspectives, 130(1),

p.017001.

6	Children's Hospital Association Directory, n.d. Available at: https://www.childrenshospitals.org/hospital-
directory#sort=%40z95xdisplayname%20ascending

7	Healthcare Cost and Utilization Project. 2005. "Calculating Kids' Inpatient Database (KID) Variances." Methods
Series Report #2005-5. Available at: https://www.hcup-
us.ahrq.gov/db/nation/kid/reports/CalculatingKIDVariances.pdf

8	McDermott, K.W., Stocks, C. and Freeman, W.J., 2020. "Statistical Brief #242: Overview of Pediatric Emergency
Department Visits. 2015." Available at: https://www.ncbi.nlm.nih.gov/books/NBK526418/

April 2023

20


-------