• Aucun résultat trouvé

Air quality, infant mortality, and the Clean Air Act of 1970

N/A
N/A
Protected

Academic year: 2021

Partager "Air quality, infant mortality, and the Clean Air Act of 1970"

Copied!
74
0
0

Texte intégral

(1)

ion

MIT LIBRARIES DUPL

3 9080

02617

6872

' 'lI '1''> )i ;. i 1'

li

lit-i

1

fff 1 I

w

(2)
(3)

Digitized

by

the

Internet

Archive

in

2011

with

funding

from

Boston

Library

Consortium

IVIember

Libraries

(4)
(5)

^^^

^>/

5

Massachusetts

Institute

of

Technology

Department

of

Economics

Working

Paper

Series

Air

Quality, Infant Mortality,

and

the

Clean

Air

Act

of

1970

Kenneth

Y.

Chay

Michael Greenstone

Working

Paper

04-08

August

2003

Room

E52-251

50

Memorial

Drive

Cambridge,

MA

02142

This

paper can be

downloaded

without

charge from

the

Social

Science

Research Network Paper

Collection at

(6)

OFTECHWLOGY

'

I

MAR

n

2004

(7)

Air

Quality,

Infant

Mortality,

and

the

Clean Air

Act

of

1970^

Kenneth

Y.

Chay

and Michael Greenstone

August 2003

*

We

thank

David

Card,

Doug

Almond,

Josh

Angrist,

Gary

Becker,

David

Cutler,

John DiNardo,

Carlos

Dobkin,

Mark

Duggan,

Vernon

Henderson,

Charles Kolstad, JeffMilyo, Steve Pischke,

and

numerous

seminar

participants for helpful

comments.

Justin

McCrary,

Doug

Almond,

Fu-Wing

Lau,

and

Anand

Dash

provided

outstanding research assistance.

Funding

from

NSF

Grant No.

SBR-9730212

and

NIH

(NICHD)

Grants

No.

ROl

HD42176-01

and

R03 HD38302-01

isgratefully

acknowledged.

(8)
(9)

Air

Quality,

Infant

MortaUty,

and

the

Clean Air Act

of

1970

Abstract

We

examine

the effects

of

total

suspended

particulates

(TSPs)

airpollution

on

infanthealthusing theairquality

improvements induced

by

the

1970 Clean

Air

Act

Amendments (CAAA).

This legislation

imposed

strictregulations

on

industrialpolluters in

"nonattainment"

countieswith

TSPs

concentrations

exceeding

the federal ceiling.

We

use

nonattainment

status as

an

instrumentalvariablefor

TSPs

changes

toestimatetheir impact

on

infant

mortahty changes

inthe firstyearthatthe

1970

CAAA

was

inforce.

TSPs

nonattainment status isassociatedwith sharp reductions inboth

TSPs

pollution

and

infant

mortality

from

1971 to 1972.

The

greaterreductionsinnonattairmient counties nearthe federal ceiling relative to the "attainment" counties

narrowly

below

the ceilingsuggestthat theregulations are the cause.

We

estimatethata

one

percent declinein

TSPs

results ina 0.5 percent

dechne

inthe infantmortalityrate.

Most

of

these effects aredriven

by

a reduction indeaths occurringwithin

one

month

of

birth, suggesting

that fetal

exposure

isapotentialbiological

pathway.

The

results

imply

thatroughly 1,300 fewer infants

diedin

1972

than

would

have

inthe

absence of

the

Clean

AirAct.

Kenneth

Y.

Chay

Department of

Economics

University

of

California,

Berkeley

549

Evans

Hall Berkeley,

CA

94720

andNBER

kenchay@econ.berkeley.edu

Michael Greenstone

Department

of

Economics

MIT

E52

50

Memorial

Drive

Cambridge,

MA

02142-1347

American

Bar Foundation and

NBER

(10)
(11)

Introduction

The

1970 passage

ofthe

Clean

Air

Act

Amendments

(CAAA)

and

the establishment ofthe

Environmental

Protection

Agency

(EPA) marked

an unprecedented

attempt

by

theU.S.

government

to

mandate

lower

levels ofairpollution.

The

statedgoal

of

theseefforts

was

toachieveairqualitystandards

that "protect thepublic health." Inthe caseoftotal

suspended

particulates (TSPs),

widely

thoughtto

be

the

most

pernicious

form

ofairpollution, the

EPA

set

maximum

allowable concentrationsthatevery

county

was

requiredtomeet.' This study usesthereductions in

TSPs

concentrations

induced

by

the

1970

CAAA

to estimatethe effects

of

TSPs

on

infantmortahty.

Figure 1

shows

trends inaverage

TSPs

pollution

and

infantmortalityrates for the entire

United

States

from

1970-1990.^ It is apparentthatairquality

improved

dramatically inthe

1970s

and

1980s, withparticulateconcentrationsfalling

from

an average of 95 micrograms-per-cubic

meter

(|J.g/m^)to

about

60

|J.g/m^.

Almost

all

of

thereduction

occurred

in

two

punctuated

periods,

1971-1973

and

1980-1982.

The

figurealso

shows

thelargedecline inthe infantmortalityrate

over

the entire period.

Although

both

TSPs

pollution

and

infantmortality ratesfrended

downward,

theirtime-series

correspondence

does

notprovide conclusive evidence

of

a causalrelationship. First,the

punctuated

declines in

TSPs

arenot mirroredin the infantmortalityrateseries. Further, other potentialdeterminants

of

infant mortality(e.g.,

socialinsurance

programs,

medical

technology)

were

also

changing

duringthisperiod.

Chay

and Greenstone

(2003) establish that

most

of

the

1980-82 dechne

in

TSPs

was

atfributable

tothe differential

impacts

of

the

1981-82

recession acrosscounties.

The

study usesthe substantial differences inairpollutionreductions acrosssites toestimatethe

impact of

TSPs

on

infant mortality.

We

find that a

one

percent reductionin

TSPs

results ina 0.35 percent declineinthe infant mortality rateatthe

county

level.

Most

of

these effects aredriven

by

a declineindeaths occurring within

one

month

ofbirth,

'Inthis

period,the

EPA

defined

TSPs

toincludeallparticleswith diameterslessthan or equalto 100 micrometers

(|im).

The

focusoffederalregulationshiftedtoparticulateswithdiameterslessthan orequalto 10 |im(PM|o) and

2.5\im (PM2.5)in 1987 and 1997,respectively.

^

The

TSPs

pollution seriesisbasedonaverages acrossthe 1,000-1,300 countiesayear with

TSPs

monitors,while

infantmortalityratesare derivedfromthe entireU.S. andare forallcausesofdeath.

The

counties with

TSPs

data accountforupwards of

80%

of the U.S. population. Unlessotherwise noted, infantmortalityreferstodeath within

(12)

suggestingthatfetal

exposure

during

pregnancy

is a potentialpathophysiologic

mechanism.

Overall, the

estimates

imply

that

about

2,500

fewer

infants died

from

1980-82

than

would

have

inthe

absence of

the

dramatic reductionsin particulates pollution.

In

view

ofthe significance ofa potentialcausal link

between

TSPs

pollution

and

infant mortality,

we

believeitisimportanttovalidate the results

from

ouranalysis ofthe early 1980s. Inthispaper,

we

examine

data

on

infantmortality

and

airpollution

from

theother

major

episodeofsubstantial

changes

in

TSPs

levels in theearly 1970s.

We

begin

by

demonstratingthatthese

changes

are largely attributable to the

1970

CAAA.

This legislationdividedallU.S. countiesinto

TSPs

nonattainment

and

attainment

categories

based

on

whether

their

ambient

TSPs

concenfrations

were

above

or

below

the legislated

maximum.

Industrial emitters

of

TSPs

were

subject to significantlystricterregulations innonattainment countiesthaninattaiiunentones.

We

find that the entiredeclinein

TSPs

duringthe early

1970s

occurred

innonattainment counties

and

that two-thirds

of

the

1971-1974

declineinthese counties occurred

between 1971 and

1972,thefirstyearthatthe

1970

CAAA

was

in force. Consequently,this studyuses nonattaiimientstatusas

an

instrumental variablefor

1971-1972 changes

in

TSPs

toestimatetheir

impact

on

changes

ininfant mortality.

This researchdesign has severalattractivefeatures. First, itisplausible that federally-mandated regulatorypressure isorthogonaltocounty-level

changes

in infantmortalityrates, except

through

its

impact

on

airpollution

-

thus,nonattainmentstatus

may

be

a valid instrument. Second, nonattainment

status isa discretefunction

of

thepreviousyear's

TSPs

levels. Thisdiscontinuityinthe

assignment

of

regulations

can

be used

to

gauge

the credibilityoftheresearchdesign. Third, ouranalysisprovides direct

and

easily interpretedestimates

of

theairquality

and

healthbenefitsofthe

1970

Act

-

benefitsthat

can be

compared

with

the costs toindustryof

complying

withthe regulations.

Finally, thisstudyprovides a

unique

opportunitytocross-validate the results

from

our earher

work by

examining

adifferentdesign appliedto adifferentera. In contrast to therecession-driven

TSPs

reductions

of

theearly 1980s,

we

use regulation-induced

changes

thatoccurred during

an

economic

expansion

(1971-1972). Thus,

any

potentialbiases

due

to

economic

shocksarelikely to

be

mitigated.

Also, particulatespollution levels

were

much

higher inthe early

1970s

thaninthe early 1980s.

As

a

(13)

result,

we

can

examine

adifferent

range of

the pollution-infantmortality function

and

analyzethe health benefits atthe federally

mandated

concentrationceiling.

We

implement

thedesign usingthe

most

detailed

and comprehensive

data available

on

infant

births,deaths,

and

TSPs

levels for the

1969-1974

period.

We

find thatthe federally

imposed

county-level regulationsareassociatedwithlargereductionsinboththe

TSPs

pollutionlevels

and

infant mortality

ratesof

nonattainment

counties.

We

estimatethata

one

percentreductionin

TSPs

results ina 0.5 percent declineintheinfant mortalityrateatthe

county

level.

As

inour earUer study,

most of

these effectsare driven

by

a reductioninthe

number

of

deathsoccurring within

one

month

ofbirth, suggestingthat fetal

exposure

isapotentialbiological

pathway.

Our

attemptsto

probe

therobustness

of

theseconclusions suggestthattheinstrumentalvariables

estimates

of

the effect

of

TSPs

pollution

on

infant mortality arefar less sensitive tospecificationthan conventionalcross-sectional

and

fixed effects estimates. Further, thetiming

and

location

of

the

TSPs

and

infantmortality

improvements correspond remarkably

well.

As

atest

of

internalvalidity,

we

find thatthe

regulation-induced

TSPs

reductionsareuncorrelated withinfantdeaths

due

toaccidents

and

homicides.

Finally,

nonattainment

and

attainment countiesnearthe federalceiling exhibit discrete differences in

TSPs

and

infant

mortaUty

changes,but

do

notexhibitdifferences inotherobservablecharacteristics. This

provides our

most convincing evidence

thatthe regulations arecausallyrelatedtodeclines inboth

TSPs

and

infantdeath.

Overall, our findings supporttheconclusions that airpollution hasa causal effect

on

infant

mortality

and

that the

1970

CAAA

provided

significanthealth benefits

by

reducing

TSPs

concentrations.

The

estimates

imply

that

roughly

1,300

fewer

infantsdiedin

1972 than

would

have

inthe

absence

of

the

Clean

Air Act. Also, theresults

correspond

wellwith the findings in

Chay

and Greenstone

(2003).

The

paper proceeds

as follows. SectionI provides

background

on

the

1970 Clean

AirAct.

SectionsII

and

IIIdiscuss theresearchdesign

and

the

econometric

methods

used

to

implement

this

design. Section

IV summarizes

the data,

and

Section

V

presentstheempiricalfindings. Section

VI

(14)

I.

Background

on

the

1970 Clean

Air

Act

Amendments

Before

1970

thefederal

government

didnot playa significant roleintheregulation

of

air

pollution, a responsibilityleftprimarilytostategovernments.^ In the

absence of

federallegislation,

few

states actedto

impose

strictregulations

on

polluterswithintheirjurisdictions.

Concerned

withthe

detrimental health effects

of

persistently

high

concentrations

of suspended

particulates pollution,

and of

otherairpollutants,

Congress

passedthe

Clean

Air

Act

Amendments

of 1970.

The

centerpiece

of

the

1970

CAAA

was

theestablishment

of

separatefederalstandards,

known

as

theNational

Ambient

Air QualityStandards

(NAAQS),

for six pollutants.

The

CAAA's

goal

was

to

reducelocal airpollutionconcentrationssothatallU.S. counties

would

be

in

compliance

withthe

NAAQS

by 1975

(with thepossibility

of

an

extensionto 1977).

The

firststep inthisprocess

was

the

assignment

of

pollution-specific "attairunent-nonattainment" designationsto allU.S. countiesfor

each of

theregulatedpollutants.''

For

TSPs

pollution the

EPA

was

requiredtodesignatea

county

as

nonattainmentifits

TSPs

concentrations

exceeded

eitherof

two

thresholds: 1)theaimual geometric

mean

concentration

exceeded 75

flg/m^, or 2) the

second

highestdailyconcentration

exceeded

260

|J.g/m^ This standard prevailed

from

1971 until 1987,

when

the

EPA

shifteditsfocustotheregulationoffiner

particles.

To

achieve thesestandards, thefiftystates

were

requiredtoformulate

and

enforceState

Implementation

Plans (SIPs)thatspecifiedprecise

abatement

activities.

For

nonattainmentcounties,the

SIPsdetailedplant-specificregulationsfor

every

major

source

of

pollution.

These

local rules orderedthat

any

substantial investment

by

a

new

orexistingplant

must be accompanied

by

theinstallation

of

state-of-the-artpollution

abatement

equipment and

strictemissionsceilings.

The

SIPsalsosetemissionlimits for existing plants innonattairmientcounties.

^

Lave

and

Omenn

(1981)andLiroff

(1986) provide

more

detailsonthe

CAAAs.

Inaddition, seeGreenstone

(2002)and

Chay

and Greenstone(2000).

"

It isunclearwhetherthe

EPA

assignednonattainmentstatusatthecountylevelorata

more

aggregategeographic

unitin the early 1970s.

Our

analysisassumesthattheassignment

was

atthecountylevel.

The

rationale forthis

choiceisdiscussed inthe "NonattainmentData"subsectionoftheData Appendix. Further,below

we

find thatusing

(15)

Instark contrast tothe oversightin

nonattainment

counties, therestrictions

on

industrial polluters

inattainment counties

were

considerably less stringent. Large-scale investments, suchas

new

plants

and

largeexpansions atexisting plants, requiredthe installation

of

less

expensive

(andless effective)pollution

abatement

equipment.

Moreover,

existingplants

and

smallerinvestments

were

essentiallyunregulated.

Both

thestates

and

thefederal

EPA

were

givensubstantial

enforcement

powers

toensurethatthe

goals

of

the

CAAA

were

met.

To

limitvariation in the intensity

of

regulationacross states, thefederal

EPA

had

to

approve

allstateregulation

programs.

On

the

compliance

side, statesinitiated their

own

inspection

programs and

frequently fined non-compliers. Further, the

EPA

was

requiredto

impose

its

own

proceduresforattaining

compliance

in statesthatfailed toadequately enforcethe standards.

The

1970

CAAA

was

signedinto

law

by

Richard

Nixon

on

December

31, 1970.

Four

months

later

on

April30, 1971,the

EPA

announced

thefinal

publicadon of

the

NAAQS

thatspecifiedthe nationalstandards for

TSPs

concentrations.

On

August

14, 1971,the

EPA

pubhshed

"Requirements

for Preparation,

Adoption

and

Submittal

of Implementations

Plans"inthe

Code

of Federal Regulations,

which

setforth

how

states

were

towritetheir

SIPs

inordertoachieve

compliance

withthe

NAAQS

by

1975. Finally, the

SIPs

were

due

tothe

EPA

by

January of

1972, thefirstyearin

which

the

CAAA

was

enforced.

Appendix

Table

1

shows

the timeline

of

the

key

datesassociatedwiththe 1

970

CAAA,

and

the

Data

Appendix

gives additionaldetails.

Henderson

(1996) provides

evidence

that theregulations

were

successfully enforced.

He

finds that

ozone

concentrations declined

more

incountiesthat

were

nonattainment

for

ozone

than inattairunent counties.

Chay

and Greenstone

(2000)find that

TSPs

levelsfellsubstantially

more

incountiesthat

were

nonattainmentfor

TSPs

thaninattainment countiesafterthe

passage of

the 1

970

CAAA

and

throughout

the 1970s.^

^Greenstone

(2002) providesfurtherevidenceonthe effectivenessofthe regulations.

He

finds thatnonattainment

statusisassociatedwith modestreductionsintheemployment,investment,and shipments ofpolluting

manufacturers. Interestingly,the regulationof

TSPs

hashttleassociationwithchangesinemployment. Instead, the overall

employment

declinesaredrivenmostly

by

the regulationofotherairpollutants.

(16)

II.

Research Design

Previous researchhas

documented

astatisticalassociation

between

TSPs

concentrations

and

adult

mortality/

However,

thereliability

of

the evidence has

been

seriouslyquestionedforseveral reasons.

First, sinceairpollutionisnot

randomly

assigned acrosslocations,previouscross-sectional studies

may

not

be

adequatelycontrolling fora

number

of

potential

confounding

determinants

of

adultmortality

(Pope

and

Dockery

1996;

Fumento

1997).

For example,

areas with higherpollution levels alsotendto

have

higherpopulationdensities,different

economic

conditions,

and

higher

crime

rates, all

of

which

could

impactadult health. Second, the lifetime

exposure

of

adults toairpollutionis

unknown.

Many

studies implicitly

assume

thatthecurrentpollutionconcentration

observed

at asiteaccurately

measures each

resident's lifetimeexposure. Third, theexcessadultdeaths thatare attributed totemporarily elevatedair

pollution levels in time-series studies

may

be

occurring

among

the

aheady

sick

and

representlittlelossin

hfe

expectancy

(Spix, etal. 1994; Lipfert

and

Wyzga

1995).

In the

absence of

randomized

clinicaltrials,this study usestheair

quahty

improvements induced

by

the 1

970

Clean

Air

Act

Amendments

inthefirstyearthatthey

were

inforcetoestimatethe

impact

of

TSPs

on

infantmortality.

A

researchdesign

based

on

comparisons

between

nonattainment

and

attainment counties hasseveral attractive features.^ First,

TSPs

nonattairunent status isassociated

with

sharpdifferences in

changes

in

TSPs

acrosssites inthe early 1970s. Panel

A

of Figure 2

shows

the trends

in

TSPs

levels

from

1

969

to 1

974

separately for countiesthat

were

nonattainment

and

attainmentfor

TSPs

in

1972

-

thefirstyearof

CAAA

enforcement.^

*Studies ofthe adultmortalityeffectsofparticulatespollutionincludeLave andSeskin (1977),

Pope

and

Dockery

(1996),Dockery and

Pope

(1996),Dockery,etal.(1993), andPope, etal. (1995). Chay, Dobkin,and Greenstone

(2003)examinetheassociationbetweenadultandelderlymortalityandregulation-induced

TSPs

declines. 'Despite extensive

efforts,

we

were

unabletoobtain alistofthecountiesthatwereinitiallydesignated

TSPs

nonattaimnentbythe

EPA.

Giventhetimelinefor statesto

comply

withthe

CAAA,

states likelyascertainedthe identityofthecountieswith

TSPs

concentrationsabovethe

NAAQS

whilewritingtheirSlPs duringthesecondhalf

of1971.

We

assumethattheyreliedonthe available

TSPs

concentrations datairom 1970todeterminewhich

countieswould benonattainmentfor 1972. Thus,

we

assigncountieswith

TSPs

concentrationsexceedingthe

NAAQS

in 1970tothe 1972

TSPs

nonattaiimientcategory. Allothercountieswithnonmissing

TSPs

data are

designated attainmentfor thatyear.

The

measures of

TSPs

concentrationsarederivedfromthe

same

network of

TSPs

monitorsused

by

thestatesandthe

EPA

in thisperiod.

The

Data

Appendix

providesfurther detailsonour

assignmentrule.

*

The

sample

consistsofthe401 countieswithcontinuousmonitorreadings from 1969-1974

-

229

ofthese counties

(17)

The

figure

documents

thatbeforethe

1970

CAAA,

TSPs

concentrations

were

40-|ig/m^higher in

nonattainmentcounties.

While

thepollutiontrends are similarinthe

two

groups

from

1969

to 1971,there

isasharp

break

intrendinnonattainment countiesafter

implementation

ofthe

CAAA.

From

1

97

1-74

newly

regulated counties

had

astunning20-|J,g/m^ reductionin

TSPs,

while

TSPs

fell

by

only

3 p.g/m^in

attainmentcounties.

These comparisons

suggestthat virtually the entirenational

dechne

in

TSPs

from

1971-74

inFigure 1

was

attributabletothe regulations. In addition,two-thirds

of

the

1971-74

TSPs

declinein

nonattainment

counties occurred

from

1971-72,the firstyear

of

CAAA

enforcement. Consequently,

our

analysis focuses

on

thisabrupt, one-year

improvement

inairquahty.^

Second,

below

we

find

evidence

thatthisdesign

may

reduce

the role

of

omittedvariablesbias in

analyzingtheassociation

between

TSPs

and

infant mortality. Panel

B

of

Figure2 displaysthetrends in internalinfantmortality rates

(IMR)

inthe

same two

setsofcounties.'"

It

shows

thatbeforethe

CAAA

was

errforced, infantmortalityrates

were

120-150

deaths per

100,000

births higherinnonattaimnent

counties. Further, attainment

and nonattainment

counties

had

similar

IMR

trends

between

1969 and

1971, with attainment counties experiencing a slightly larger decline.

However,

thispattern isreversed

between 1971

and

1972,withtheattainment-nonattainment

IMR

gap narrowing

by

over

75

deathsper

100,000

births.

The

figurealsoreveals thatattainment counties

had

a largerdecline in infantmortality

between

1972

and

1974.

The

correspondence of

the trendbreaks in

TSPs

and

infantmortalitydifferences

between

nonattainment

and

attainment countiesin

1972

suggestsa causalrelationship. Since thepatterns after

1972

areless

compelhng,

we

probe

thecasefor causality

more

rigorouslyintheanalysis below.

To

1970.

'

We

probed

how

industrial

polluters

may

have significantlyreduced

TSPs

emissionsinthisnarrowtime frame.

Electrostatic precipitatorswere one oftheprimarycontroldevicesusedtoabateparticulatespollutionduringthis

period.

The

Industrial

Gas

CleaningInstitutereportsthatwhileannual sales ofprecipitatorswere approximately $50

millionfrom 1968 through 1970,thisfiguredoubledto

$100

millionin 1971 and 1972 (White 1984). Accordingto

Donald

Hug

oftheEnvironmental ElementsCorporation, asupplierofpollutionabatementequipment, an

electrostatic precipitatororderedin 1971 couldbeoperatingin 1972.

We

concludethatthe largedecreasein

TSPs

emissionsbetween 1971 and 1972is theplausible resultofpolluterresponsetothe 1970

CAAA.

'"

We

defineinternalinfantmortalitytobeinfantdeathduetohealth relatedcauses(i.e., 8"'International ClassificationofDiseases(ICD) codes 001 through799). This excludes

non

-health related"external" causesof

(18)

maximize

the signal-to-noiseratio in

TSPs

changes

and minimize

the potential biases

from confounding

changes

inothervariables,

we

focus

on

the strikingone-year

changes

thatoccurred

between

1971

and

1972.

The

analysisuses nonattainmentstatusas

an

instrumentalvariable for

1971-1972 changes

in

TSPs.

Since federally-mandated regulatory pressure isplausibly orthogonalto

changes

in infant mortalityrates,

except

through

its

impact

on

airpollution,nonattainmentstatus

may

be

avalidinstrument. Consistent withthis,

we

findlittleassociation

between

nonattainment status

and

otherobservablevariables,

includingparents' characteristics, prenatalcareutilization,

and

transfer

payments

from

socialprograms.

Further,nonattainmentstatusin

1972

was

adiscrete fiinction

of

the annualgeometric

mean

and

second

highestdailyconcentrationsof

TSPs

in theregulationselection year, 1970.

Thus,

the

assignment

of

regulatorystatushas thefeatureofaquasi-experimentalregression-discontinuitydesign

(Cook and

Campbell

1979). Ifother factors affecting infantmortalityare similar for countiesjust

above and

just

below

theregulatorythresholds, then

comparing

outcome

changes

innonattainment

and

attainment counties

with

pre-regulation

TSPs

levelsjust

around

thethresholdwill controlforallomittedfactors correlated

with

TSPs.

Under

this

assumption

discretedifferences in

mean

outcome

changes

between

nonattairmient

and

attairmientcountiesnear the federal ceilings

can be

attributed tothe regulations.

The

discontinuityinthe

assignment

ofregulationsprovides a valuableopportunityto

gauge

the credibilityof

theresearchdesign

and develop

convincingspecificationtests.

Third, our focus

on

infant, ratherthanadult, mortality

may

circumvent

the other important

criticisms of previous research.

The

problem of

unknown

lifetime

exposure

to pollutionis significantly mitigated, ifnotsolved,

by

the

low

migrationrates

of

pregnant

women

and

infants. Thus,

we

assign

TSPs

levels toinfants

based

on

the

exposure

ofthe

mother

during

pregnancy

and

the

exposure

of

the

newborn

during thefirst

few

months

afterbirth

(we

focus

on

deaths within

24

hours,

28

days,

and

1 yearofbirth).

Further, since the mortality rateishigher inthefirstyear

of

lifethaninthenext

20

years

combined

(NCHS

1993),itis likelythatinfantdeaths representa large loss inlife expectancy. In

contrast, time-series studies

of

adult

and

elderly

mortahty based

on

dailyairpollution fluctuations

may

be

detectingeffects

among

thealready veryill

and

dying. Inthiscase,the actual lossoflife

expectancy

may

be

much

smaller than thesestudies implicitlysuggest.

(19)

A

final

advantage

of

thedesignisthatitpermits a directanalysis ofthehealthbenefits of

TSPs

reductions atlevelsjust

above

the

EPA's

maximum

allowable concentration.

The

optimalityofthe

CAAAs

depends

crucially

on

theexistence

and

relative

magnitudes of

thehealtheffects

above and

near

the federal regulatorythreshold.

III.

Econometric

Methods

Our

analysis

compares changes

ininfantmortalityrates

and

TSPs

pollution

from

1971-1972

for

countiesthat

were

nonattainment

and

attainmentfor

TSPs

in 1972. Here,

we

discuss the

econometric

models used

to estimatethe infant

mortality-TSPs

association.

For

simplicity, it is

assumed

thatthe "true" effect

of exposure

toparticulatespollutionis

homogeneous

acrossinfants

and

overtime.

The

cross-sectional

model

predominantly used

inthe literature

on

airpollution

and

healthis:

(1)

yc

=

Xc'P +

eTc,

+

ec, Ect

=

ttc

+

Uc,

(2) Tct

=

Xct'n

+

Tic Tlet

=

K

+

Vet,

where

yet isthe infantmortalityratein

county

c inyeart,Xct isavector

of observed

characteristics,

T^

is

the

mean

of

TSPs

acrossall monitorsinthecounty,

and

8ct

and

Tictarethe

unobservable

determinants

of

infantmortalityrates

and

TSPs

levels,respectively.

The

coefficient

9

isthe "true" effect

of

TSPs

on

infantmortality.

For

consistent estimation, theleastsquares estimator

of 9

requiresE[ect'nct]=0. Ifthere areomitted

permanent

(ttc

and

Xc) ortransitory(Uct

and

Vd) factorsthat

covary with both

TSPs

and

infant mortality,then thecross-sectionalestimatorwill

be

biased.

With

repeated observations overtime, a"fixed-effects"

model

implies that first-differencing the

datawill

absorb

the

permanent county

effects, (Xc

and

^c- This leadsto:

(3) yc,- yct-I

=

(Xe,- Xct-O'P

+

9(Tct-Tc,-,)

+

(Uet -Uet-l)

(4) Tet-Tet-l

=

(Xct-

XcO'n

+

(Vct-V„.i).

For

identification, thefixedeffects estimator

of

requiresE[(uct-Uct-i)(Vct

-

Vct-i)]=0.

That

is, thereare

no

unobserved shocks

to

TSPs

levelsthat

covary

with

unobserved shocks

to infant mortalityrates.

While

first-differencing

removes

permanent

sources

of

bias, it

may

magnify any

attenuation bias

due

to

measurement

error. Since our

measure

of

TSPs

concentrations is

an average

across themonitorsin

(20)

a county, itimperfectly capturesindividual

exposure

to

TSPs."

This

mismeasurement

will attenuate the

estimated

TSPs

regressioncoefficientleadingtoabias thatisproportionaltothe fractionofthetotal

variation in

TSPs

that is attributable to

measurement

error.

To

theextent that theserialcorrelation in "true"

TSPs

exposure

is greaterthanthe autocorrelation inthe

TSPs

measurement

error, first-differencing

thedatawill increasethe

magnitude

oftheattenuationbias.'^

Now

suppose

thereexists

an

instrumentalvariable, Zc,that causes

changes

in

TSPs

without

having

adirect effect

on

infantmortality ratechanges.

Such

a variable

would

purge

theestimates

of

the biases

due

to

both

omittedvariables

and

measurement

error.

One

plausibleinstrumentisthe 1

970

C

AAA

regulatoryintervention for

TSPs,

measured

by

the

1972

attainment-nonattainment statusofacounty. Here,equation (4)

becomes:

(5) Tc72- Tcvi

=

(Xc72 -Xc7i)'nTx

+

Z^jiUtz

+

(Vc72" Vc7i),

and

(6) Zc72

=

1(Tc70

>

T

)

=

1(Ve70

>

T

-Xc7o'n - X,),

where

Zc72 istheregulatorystatusof

county

c in 1972, 1(»)is

an

indicatorfunction equalto

one

ifthe

enclosed statement istrue,

and

T

isthe

maximum

concentrationof

TSPs

allowed

by

the federal regulations.

Regulatory

status in

1972

isadiscretefunctionof

1970

pollution levels.'^

An

attractive feature

of

thisresearchdesign is thatthe

reduced-form

relations

between 1972

TSPs

nonattainmentstatus

and

the

two endogenous

variablesprovidedirect estimates

of

the benefits

of

the

regulations. In particular, FItz

from

equation(5)

measures

the

1971-1972

airquality

improvement

in

nonattainment countiesrelativetothe

change

inattairunentcounties. Intheother

reduced-form

equation, (7) yc72-yc7I

=

(Xc72-Xc7l)'nyx

+

Zc72nyz

+

(Uc72"Uc7l),

'

' Duringthe 1971-1972period, the

mean

and

median

numbersofmonitorsinacountyare4.2and2,respectively.

'^However,

inour contextitisunclearwhethertheserialcorrelationinthe

TSPs

measurement

errorfrom 1971to

1972issmalleror largerthantheserialcorrelation intrue

TSPs

exposure.

'^

Forsimphcity,

we

havewritten(6)asifregulatorystatusisa functionofasinglethresholdcrossing. If T^^yg and

T^"^^ are theannual geometric

mean

and2" highestdaily

TSPs

concentrations, respectively,thenthe actual

regulatoryinstrumentusedis 1(T^^q

>

75|ig/m^ or

T^^

>

260|ag/m^).

Only

sixcountieswerenonattairmientin

1972 forexceedingthe2"''

highestdailyconcentrationthreshold, butnottheannual geometric

mean

ceiling. See

footnote 7andtheData

Appendix

fordetailsonthetiming ofthe

CAAA

intervention.

(21)

riyzcapturesthe relative

improvement

inthe infant

mortahty

rateinnonattainment counties

between

1971

and

1972.

Since the instrumentalvariables (IV) estimator(9iv)

of

the effect

of

TSPs

on

infant mortalityis

exactlyidentified, itis equalto theratioofthe

two

reduced-form

relations

-

i.e.,Biv

=

Tlyz/TiTz-

Two

sufficientconditions for6iv toprovide a consistentestimate

of

the

TSPs

effect are

Htz

'^

and

E[Vc7o(Uc72

- Uc7i)]=0-

The

firstconditionrequires that theregulations

induced

airquality

improvements.

The

second

conditionrequires that transitory

shocks

to

TSPs

levels in

1970

are orthogonalto

unobserved

shocks to infantmortalityrates

from

1971-1972.'''

Even

ifE[Vc7o(Uc72 -Uc7i)]

^

0, causalinferences

on

9

may

be

possible

by

leveragingthe

regression discontinuity

(RD)

design implicit in the 1(•) functionthatdeterminesnonattainmentstatus.

For example,

ifE[vc7o(Uc72-Uc7i)]

=

inthe

neighborhood of

theregulatoryceiling(i.e.,

75

)J,g/m^), thena

comparison of changes

in

nonattainment

and

attainment counties inthis

neighborhood

willcontrolforall

omittedvariables.

Below,

we

present

IV

estimates

based

on

allcounties

and

RD

estimatesin

which

the

sample

is limitedtocounties with

TSPs

concentrations "near"the

75

|ig/m^thresholdintheregulation

selection year.'^ Inthenextsection,

we

show

thattheobservablecharacteristics

of

nonattairmient

and

attainment counties nearthe federal

TSPs

ceilingare

very

similar.

Another

potentialsource

of

bias

comes from

uruneasured

state-specificdeterminants ofinfant

mortality(e.g., generosity

of

state

Medicaid programs,

state

weather

conditions). Sincethe analysis is at

the

county

level, in

some

specifications

we

control for unrestricted state-time effects in infant mortality to

absorb

all

unobservables

that

vary

acrossstatesovertime. Here, the effect

of

TSPs

is identifiedusing"

only

comparisons

between

attainment

and nonattainment

countieswithinthe

same

state. Figure3

provides agraphical

overview of

the locationof attainment

and

nonattainment counties in 1972.

A

county's

shading

indicatesitsregulatorystatus

-

light

gray

for attainment,blackfor nonattairmient,

and

Inthesimplestcase,the

IV

estimatorwillbeconsistentif1972 nonattainmentstatus isorthogonaltounobserved

infantmortalityshocks

-

i.e.,E[Zc72(Uc72-Uc7i)]=0,

which

isa stronger condition thanE[Vc7o(Uc72-Uc7i)]=0. IftherelationshipbetweenVc7oand(Uc72-Uc7i)issufficiently"smooth"attheregulatoryceilings,then causal inference

may

alsobepossible

by

including smoothflmctionsoftheselection variable,Tc7o,ascontrols. Below,

we

alsoprovide

RD

estimatesthatare adjustedfora quadraticin Tc7o.

(22)

whiteforthe countieswithout

TSPs

pollutionmonitors.

The

pervasiveness

of

the

1972

TSPs

regulatory

program

is evident. Inadditiontothe traditional countiesinthe

Rust

Belt

and

inthe

South Coast

Air

Basin

around

Los

Angeles,scoresof other counties

were

designated nonattainment. Importantly,

most

statescontainboth attainment

and

nonattainment counties.

IV.

Data

Sources

and

Summary

Statistics

To

implement

our evaluation strategy,

we

use

county

leveldata

on

airpollution, infant health,

and

a

number

of

additional control variables for the

1969-1974

period. Here,

we

describe these data

and

provide

summary

statistics

showing

thatthenonattainment-attainment research design balancesthe

observablecharacteristics

of

counties.

More

details

on

thedatasources

and

variablesarecontainedin the

Data Appendix.

A.

Data

Sources

The

health

outcome

variables

and

a

number

of

the control variables

come

from

theNational MortalityDetail Files

and

National NatalityDetail Files,

which

arederived

from

censuses

of

death

and

birthcertificates.'^

We

merge

the individual birth

and

infantdeath recordsatthe

county

levelfor

each

yeartocreate

county

by

year cells.

For each

cell, the infantmortalityrate is calculated as theratio

of

the

total

number

of

infantdeaths

of

a certaintype (age

and

cause

of

death) to thetotal

number

ofbirths.

The

mortality rateswithin

24

hours,

28 days

(neonatal),

and one

yearofbirthare

computed

separatelyfor

each

year. Fatalitiesthatoccur

soon

afterbirtharethoughtto reflect

poor

fetal

development.

Thus,

we

use

the

estimatedeffects of

TSPs

on 24-hour and

neonatalmortality to

probe

whether

fetal

exposure

to

TSPs

may

have

adversehealtheffects.

We

also calculateseparate mortalityratesfordeaths

due

to internalhealthreasons

and

deaths

due

toexternal"non-health"relatedcauses

such

asaccidents

and

homicides.'' Internalinfant deaths arenot

'*

The

NatalityFilesprovidea

100%

censusofallbirthsin every yearfrom 1969-1974.

The

MortalityFilesprovide a

100%

censusofall deathsinevery year exceptfor 1972,inwhicha

50%

random

sample ofalldeathsisprovided.

"

"Internal"and"external"deathsspanallpossiblecausesofdeath. Seefootnote 10forthe definitionofthe internalandexternalcategoriesbased onthe8"'

ICD

codes.

(23)

disaggregatedfurther for

two

broad

reasons. First, coroners assignabout

70%

of

thesedeathsto

two

categories

"certaincauses

of

perinatalmorbidity

and mortahty" and

"congenitalanomalies." Second,

since thepathophysiological

pathways through

which

TSPs

may

affect fetalor infanthealth are largely

unknown,

thereis

no

aprioribasis fordividing these causes

of

deathinto ones thatare

and

arenot plausiblyrelatedtoTSPs.'^

However,

it

seems

reasonableto

assume

that thereis

no

causal

pathway

Unking

airpollution to externalcauses ofdeath.

So

we

usethe estimated association

between

external

mortalityrates

and

TSPs

pollution as a

check

on

the internal consistency

of

our

fmdings."

The

microdata

on

thecontrolvariables available in theNatalityFiles arealsoaggregated into

annual

county

cells.

The

detailedNatalityvariablesincludeinformationon:

socioeconomic

and

demographic

characteristicsofthe parents;

medical

system

utilization, including prenatalcare

usage

(Rosenzweig

and

Schultz 1983;Institute

of

Medicine

1985); maternal health

endowment,

such

as

mother's

age

and pregnancy

history

(Rosenzweig

and

Schultz 1983;

Rosenzweig

and

Wolpin

1991);

and

infantbirthweight. Since

TSPs

may

affectfetal

development,

we

useinfantbirth

weight

asanother

outcome

variable.

The

Data

Appendix

describes the Natalityvariables

used

intheanalysis indetail.

Annual,

monitor-level data

on

TSPs

concentrations

were

obtained

by

filinga

Freedom

of

Information

Act

requestwiththe

EPA.

This yielded

annual

summary

information

from

the

EPA's

nationwide

network of

pollutionmonitors, includingthe locationof

each

monitor

and

their readings.

The

dataalsocontaininformation

on

the

armual

geometric

mean TSPs

concentrationfor

each monitor

ina

county

and

the

number

ofdaily

monitor

readings

exceeding

the federal standardsin

each

year.

These

measures

are

used

to

determine

the

annual

nonattaimnent

status

of each

county

(see the

Data

Appendix

for

more

details).

From

the data,

we

alsocalculatedthe

mean

of

TSPs

concentrationsin

each

county

and

year,

which

we

use

as our

measure of

TSPs

exposure

inthe analysis.'"

' SeeUtelland

Samet

(1996)fora

summary

oftheevidenceonthepathophysiologicalrelationshipbetween

TSPs

and

human

health.

Thatis,the associationbetween

TSPs

andexternal infantdeaths

may

resultfrom unobservedsecularfactors that affectall typesofinfant death.

Specifically, thearmual

TSPs

concentrationusedistheweightedaverageoftheannualarithmetic

means

of each monitorin thecounty,withthe

number

of observations permonitor usedasweights.

Some

recent researchhas focused onthe effectsof

PM-IO

particles.

Most

ofthesestudiesmultiplied

TSPs

concentrationsbyaconstantfactor toobtain

PM-10

measures (e.g..Pope, Dockery,

and

Schwartz, 1995).

(24)

Inadditiontothe

key

infanthealth

and

pollution data,

we

usea variety

of

othercounty-leveldata

as controls. Per-capita

income

data

come

from

the

Bureau

of

Economic

Analysis' annual series,

which

provides the

most

comprehensive

measure of income

availableatthe

county

level. This filealsoprovides

annual, county-level data

on

per-capita net earnings, theratiooftotal

employment

tothetotal population,

and

theratio

of

total

manufacturing

employment

tothe totalpopulation.

The

Regional

Economic

Information

System

fileprovidesannual, county-level data

on

several different categoriesoftransfer

payments.

These

include separateseries fortotal transfer

payments;

total

medical

care

payments;

public expenditures

on medical

carefor

low-income

individuals(primarily

Medicaid

and

local assistance

programs);

income

maintenance

benefits;family assistance

payments,

including

AFDC;

Food

Stamps

payments;

and

Unemployment

Insurancebenefits.

Table

1 presents

summary

information

from 1969-1974

forthe

501

countieswithavailable

TSPs

and

infantmortality data in 1970, 1971,

and

1972. This is our

primary

sample

inthe

below

analysis. In

1969 and

1973-74,the

sample

is fiirtherresfrictedtocounties with

nonmissing

data

on

TSPs

and

infant

mortalityinthoseyears.

The

top

rows

of

the table

show

thatthesecounties accountfor over

60%

of

the 3.2-3.7 million

livebirthsthatoccur annuallyinthe

United

Statesin thisperiod.

The

next

rows

present infantfatalities

per

100,000

live births,

by

cause

of

death.

The

internalinfantmortality ratein 1971 is 1.8deaths within a yearofbirthper

100

births,whileinfantdeaths

due

to externalcausesare

much

rarer. In 1971, about

45

and 75

percentofall internalinfantdeathswithinayear

of

birthoccur withinthefirst

24

hours

and

28

days

of

birth,respectively.

The

tablealso

shows

national trends in infant mortality,average

TSPs

concentrations,

and

per-capita

income

across counties. Internalinfant mortalityratesdecreasedsteadily

from

1969-74.

TSPs, on

theotherhand,fell

by

over 18 percent

from

1971-74, but

were

relativelystable

from

1969-71. Per-capita

income

increased

by

8.5 percent

from

1971-73, beforefallingslightly

from

1973-74,

presumably

aresult

of

the recession.

The

remaining

rows of Table

1 present

1969-74

trends forseveral

of

the control variables

used

in

the analysis. Importantly, the

magnitude of changes

inthesevariablesissmall

when

compared

tothe

TSPs

pollution

changes

that occurred

from

1971-72.

Our

measures of

parental

background

characteristics 14

(25)

suggestthat theaverage

socioeconomic

status

(SES) of

parents

was

declining

from

1969-74, as the fractions

of

births to single,high school dropout,

and

black

mothers

increased.

On

theotherhand,the

Hkelihood

of

a

mother

receiving prenatal care or

having

aprenatalvisitinthe firsttrimesterincreased

slightly.

The

percentage

of

births atfributabletoteenagersrose,

and

theshare

among

women

aged 35

and

overfell."' Finally,the shareoffirst-timebirths increasedslightly

while

the fractionof

mothers

who

had

apriorfetaldeath

remained

stable.

B.

Balancing

of Observable

Characteristics

Before

proceeding,

we

examine

whether

the

nonattainment

instrumentalvariable is orthogonalto theobservablepredictors

of

infant mortality.

While

itisnotaformaltest

of

theexogeneity

of

the

instrument,it

seems

reasonableto

presume

thatresearchdesignsthat

meet

thiscriterion

may

suffer

from

smalleromittedvariablesbias. First, designs thatbalancetheobservable covariates

may

be

more

likely to also

balance

the

unobservables

(Altonji, Elder,

and Taber

2000)."'

Second,

iftheinstrumentbalancesthe

observables, thenconsistent inference

does

not

depend

on

fiinctional

form

assumptions

on

the relations

between

theobservable

confounders

and

infant mortality. Estimatorsthatmisspecify thesefiinctional

forms

(e.g.,

hnear

regressionadjustment

when

theconditional expectationsfianctionisnonlinear)will

be

biased.

Table

2

shows

the association

of

TSPs

levels,

TSPs

changes,

and

TSPs

nonattainmentstatuswith otherpotential correlates

of

infantmortalityusingthe

same

sample of

counties asin

Table

1.

Each

column

ofthe tablepresentsthedifferences inthe variable

means between two

sets

of

counties (with standarderrors inparentheses).

The

first

column

contains differences inthe 1971 levels ofthe covariates

between

counties with

high (above

the

median)

and

low

(below

the

median)

TSPs

concentrations in 1971.

If

TSPs

levels

were randomly

assigned acrosscounties,

one

would

expectvery

few

significant differences

between

the

two

groups.

However,

thereare significantdifferences for several

key

variables, including

"'

Rees,et. al. (1996)find that

much

ofthecorrelationbetween poorinfanthealthandmother'sageisduetothe higher incidenceoflowbirthweightbirths

among

teenagersand

women

35

and

older.

^^

For example,inexperimental designstestingfordifferencesintheobservablecharacteristicsofthetreatmentand

controlgroups isusedtoassesswhetherthe treatment

was

actually

randomly

assigned. 15

(26)

mother's

race,

immigrant

status, use

of

prenatalcare,

and

likelihood

of being

ateenager.

Though

not presentedinthetable, thereare also large

and

significantdifferences

between

the

county

groups in per-capita receipts of

income

maintenance

benefits, familyassistance

payments,

and

publicexpenditures

on

medical

carefor

low-income

individuals(e.g., see

Chay, Dobkin,

and Greenstone

2003). Overall, these

findings suggestthat"conventional"cross-sectionalestimates

may

be

biased

due

toomittedvariables.

The

next

two columns

of

Table

2

perform

a similar analysis for

1971-1972

TSPs

changes.

They

containdifferences in 1971 variable levels

and 1971-72

variable

changes

between

countiesthat

had

large

(above

the

median) and

small

(below

the

median)

reductionsin

TSPs

between

1971

and

1972.

These

two

county groups

exhibit significant differencesinthe 1971 levels of percapita

income,

mother's

immigrant

status,prenatal careuse, likelihood

of

being over

34

yearsold,

and pregnancy

history.

There

arealso significantdifferences in the

1971-72 changes

inmother's

and

father's educationlevels

and

marginally

significantdifferences inthe

changes

inthefi-actionsof

mothers

who

are teenagers

and

who

had

aprior

fetaldeath.

Thus,

fixed effects

models

may

alsoleadtobiasedinference.

The

next

two columns

ofthe table

show

the differencesin 1971 levels

and

1

97

1-72

changes

between nonattainment

and

attaiimientcounties.

While

nonattainmentstatusisassociated witha large

and

significantreductionin

TSPs

concentrations, it is

uncorrected

with nearlyall

of

the 1971 levels

and

1971-72 changes

inthe covariates, withtheonly exceptions

being

percapita

income and

mother's racein

1971.

The

differences inthe

changes

inmother's

and

father'seducation

and

inthe

changes

in the

incidences of

mothers

who

areteenagers

and

experienced aprevious fetaldeathare greatly

reduced

by

comparing nonattainment and

attainmentcounties. Also, incontrast totheprevious

two

sets

of county

group

comparisons, nonattainment

and

attainment counties exhibitlittle differencein theirreceipts

of

transfer

payments

firom social

programs

(resultsavailable fi^omtheauthors). Thus, itappearsthatthe

attainment-nonattainment researchdesigndoes a betterjob of balancingthe observables."^

^^

Itshould benotedthatthecompositionofcounties

who

arenonattainmentandattainmentin 1972isdifferent

fromthecompositionof countieswith "high"and"low"

mean

TSPs

in 1971 (i.e.,

column

1).

(27)

Inthelast

two

columns,

the

sample

isfurther restrictedtothe

176

countieswith

1970

geometric

mean TSPs

concentrations

between

60 and

90

|ag/m^

-

i.e., countieswith

TSPs

levels nearthe

TSPs

regulatoryceiling intheregulationselection year. Itis evidentthat focusing

on

nonattainment

and

attainment counties neartheregulatorythreshold eliminates all

of

the significant differencesinthe

covariatelevels

and

changes, including per capita income,

mother's

race,

and

the infantmortality ratein

theyear before

CAAA

enforcement.

While

nonattainment

status isorthogonaltoall ofthe other covariates, the

second

row

shows

thatitis still associated

with

a sharp reductionin

TSPs

levels

between

1971

and 1972

forcounties neartheregulatorythreshold."''

These

resultsprovide reassuring evidence

on

the quality

of

the instrumental variablesresearchdesign

used

in thisstudy.

We

probe

thisfurtherbelow.

V.

Empirical

Results A. Cross-Sectional

and Fixed

EffectsResults

First,

we

replicatethe conventionalcross-sectional

approach

toestimatingtheassociation

between

TSPs

pollution

and

infant mortalitythatis

common

intheliterature.

Table

3 presents the

regression estimates

of

the effect

of

TSPs

on

the

number

of

internal infantdeaths within a year ofbirth

per

100,000

livebirths for

each

cross-section

from

1969-1974.

The

final

row

containsthe results for the

pooled 1969-1974

data.

Column

1 presentstheunadjusted

TSPs

coefficient;

column

2 includesthebasic

control variables available inthe Natalitydata;

column

3 includesadditionalNatalityvariables

and

flexiblefunctional

forms

forthevariables;

and

column 4

further adjusts forper-capita

income,

earnings,

employment, and

transfer

payments by

source."^

Columns

5

and

6

add

unrestrictedstateeffects to the

column

2

and

column 4

specifications, respectively.

The

sample

sizes

and

R''s oftheregressions are

shown

in brackets.

There

is

wide

variabilityintheestimated effects

of

TSPs,

both across specifications foragiven cross-section

and

acrosscross-sections foragivenspecification.

While

the

raw

correlationsin

column

1

Giventhenotationinequations(6)and(7),itshouldbe notedthat 1972nonattainmentstatus isorthogonaltothe levelsofcovariatesin 1970aswell(resultsavailablefromthe authors).

^

The

controlvariablesincludedineachspecification arelistedintheData Appendix. 17

(28)

are allpositive,only those

from

the 1

969 and

1

974

cross-sections arestatistically significantat

conventional levels."^ Includingthebasic Natalitycontrolsin

column

2reduces thepointestimates substantially in

most

years,

even

astheprecision ofthe estimatesincreases

due

tothegreatly

improved

fit

of

the regressions. In addition, only

two

of

the estimatesin

columns

3

and 4

are significantat

conventionallevels,

and one of

thesehas a perverse negativesign. This isalso trueofthe

most

unrestricted specificationin

column

6thatalso adjusts forstatefixedeffects.

The

largestpositiveestimates

from

the cross-sectionalanalyses

imply

thata l-)J,g/m^reductionin

mean TSPs

resultsin

roughly

3 fewer internalinfantdeaths per

100,000

livebirths,

which

is

an

elasticity

of0.14. Overall,

however,

thereis littleevidence ofa systematic cross-sectionalassociation

between

particulatespollution

and

infantsurvivalrates.

While

the

1974

cross-section

produces

estimates that are positive, significant

and

slightlylesssensitive to specification, the 1

972

cross-sectionprovides estimates

thatare routinely negative.

The

sensitivity

of

the results to theyear analyzed

and

theset

of

variables

used

as controlssuggests that omittedvariables

may

play

an

importantroleincross-sectional analysis.

We

suspectthat thefragility

of

thecross-sectionalresultsto

changes

inspecification

and

sample

is likely to

applytotheestimation

of

the relationship

between

other

measures of

human

health(e.g., adult mortality)

and environmental

quality(e.g., otherairpollutants).

Table

4

presentsthefixed effects estimates

of

the association

between

mean TSPs

and

internal

infantmortalityrates

based

on

1969-1974 pooled

data(firstpanel)

and 1971-1972

first-differenceddata (secondpanel).

The columns

correspond

to similarregressionspecificationsas thosein

Table

3.

The

1971-1972

firstdifferences resultsprovide a baselinefor

comparison

tothe

subsequent

instrumental

variablesresultsthatusethe

same

samples

and

control variables.

The

inclusionof

county

fixedeffectsor the use

of

first-differenceddataeliminatesthe bias in the cross-sectional estimates attributable to

time-invariantomittedfactors that

vary

across counties.

However,

these

approaches

will

be

biasedifthereare

unobserved shocks

thatare correlated

with

changes

in

TSPs

and

infantmortalityrates.

^^

Itisworthnotingthatthecomparison ofcountieswith"high"and "low" 1971

TSPs

levelsinthefirst

column

of Table2 impliesanunadjusted

TSPs

effectof2.8 (129.8/45.8),whichis

75%

greaterthantheestimatedcorrelation for 1971 in Table3 (1.6). Thisisconsistentwithmeasurementerrorin

TSPs

leadingtosubstantive attenuationbias,

sincetheaggregationinTable2 "averages out"themeasurementerrorsover

many

counties.

(29)

The

1969-1974

panel

shows

thattheinclusion

of

county-specificintercepts greatly

improves

the

fit

of

theregressions

when

compared

tothe

pooled

cross-sectionalregressions in thelast

row

of Table

3.

The

raw

fixed effects correlation in

column

1 ispositive

and

highlysignificant.

However,

theestimated

TSPs

coefficientfalls significantly

when

the analysis controls for theNatalityvariables,

economic

conditions, transfer

payments,

yeareffects

and

state-yeareffects.

The

1971-1972

firstdifferences results alsosuggestlittleassociation

between

changes

in

TSPs

and changes

ininfantmortalityrates.

The

fixedeffects association

between

TSPs

and

infantmortality issmall

and

sensitive to specification,

which

is consistent

with

potentialbiases

due

toomittedvariables and/or

measurement

error.''

We

conclude

thatconventional

approaches

may

notprovidereliableestimates

of

the causal effect

of

TSPs

on

infantmortality

and

turnourattention totheinstrumentalvariables design outlined above.

B.

The

Impact

of the

1970 Clean Air Act

on

Air

Quality

and

InfantMortality

The

instrumentalvariablesestimate

of

theeffect

of

TSPs

isa fiinction

of

two

reduced-form

effects: the

impact of 1972

nonttainmentstatus

on

improvements

inairquahty,

and

its associationwith

declinesininfant mortality. Here,

we

provide

estimates

of

theairquality

and

infanthealthbenefits

of

the

1970 Clean

Air

Act

and

present

evidence

thatour design

may

plausibly identifythe causaleffects of

nonattainment

status

on both outcomes.

Figure

4

depicts the

1969-1974

time-series

of

the

raw

differencesin

TSPs

levels (Panel

A)

and

internalinfantmortalityrates(Panel

B) between

countiesthat

were

nonattainment

and

attainmentin 1972. Separate series are plottedforthreesamples: 1) allcountieswith

nonmissing

TSPs

data, 2) thesubset of counties

with

a

1970

geometric

mean TSPs

concentration

between

50-1 15 )ig/m^,

and

3) thesubset

of

counties

with

1970

TSPs

between 60-90

|ig/m\"^

The

last

two samples

arerestricted tothose counties with

TSPs

concentrationscloser tothe

75

|ig/m^regulationceiling intheregulation selection year.

^'

The

comparison ofcounties with"big"and"small" 1971-1972

TSPs

reductionsinthe third

column

of Table2 impliesanunadjusted

TSPs

effect of-0.73 (18.8/(-25.6)) for 1971-1972firstdifferences. Thissuggeststhatthe

small 1971-1972first-differencesestimatein

column

1 ofTable4 isnotthe resultof attenuationbias dueto

measurement

error. Seefootnotes 12and 26for

more

details. ^*

The

samplesconsistofthefixedsetofcountieswith continuousmonitorreadings

from

1969-1974.

The numbers

ofcountiesare401, 262,and 141, respectively.

(30)

Panel

A

shows

large trendbreaks inthenonattainment-attainmentdifference in

TSPs

that

correspond

with

thetiming

of

thefirstyearof

CAAA

enforcement

in 1972.

There

isa clear

convergence

in

TSPs

from

1971 to

1972 even

forthe subset

of

counties

with

1970

TSPs

levels closetotheregulatory

threshold.

Among

countieswith

1970

TSPs

between 60-90

[ig/m?, the 10.5 unit

nonattainment-attainment

gap

in

TSPs

in 1971 is

almost

eliminated

by

1972

before

rebounding

somewhat

in 1973.

For

countieswith

1970

TSPs

between

50-115 |J.g/m^,thereis also a significant relativereductionin

TSPs

from

1971-1972, though

the

break

intrendappearsto

have

starteda yearearlierthanfor the other

two

county

groups.

On

the whole,thetiming

and

location

of

the

TSPs

reductionssuggestthat

enforcement

of

the 1

970

CAAA

ledto

improved

airquality.

Panel

B

ofthe figure

shows

astriking

correspondence

between

thebreaksintrendininfant

mortalitydifferences

between

nonattainment

and

attainment counties

and

thosein

TSPs

differences.

For

thefullsample,thesharp

break

ininfantmortalitytrends

from

1971-1972 matches

the

break

in

TSPs

differences.

The

differentialtiming

of

the relativeshift

down

ininfant mortalityacrossthe three

samples

also

matches

the

TSPs

patterns.

While

thefull

sample and

the

sample

with 1

970

TSPs

between 60-90

)Lig/m^

show

relativemortality reductions

from

1971-1972

only, the

sample with 1970

TSPs

between

50-1 50-15|ag/m^ exhibits arelativemortalityreductiontheprevious yearalso.

Also

consistent

with

the patterns inPanel

A,

the

sample with 1970

TSPs

between 60-90

jig/m^ exhibitsthe largest

rebound

in relative

mortalityrates

from 1972

to 1973.

Below,

we show

that

some

ofthe

1972-1974 rebound

in infant

mortalitydifferences is explained

by

differential

changes

intheobservablecharacteristics

of

attainment

and

nonattaimnent

countiesafter 1972.

Taken

together, theseplots provideevidenceofa direct link

between

regulation-induced

improvements

inair

quahty

and

reductionsin infant mortalityrates. First, thetiming of the abrupt declines in

TSPs

and

infant mortality correlate

remarkably

wellacrossallthree

groups

ofcounties. Further,

none of

theobservablecovariates, including per-capita

income

and

transfer

payments,

exhibits

similartrend

break

patternsinnonattainment-attainmentdifferences. It

seems

unlikelythat

any

unobserved

factors

had

nonattainment-attainmentdifferencesthat

changed

as sharplyorabruptlyas the

differencesin

TSPs

and

infant mortality

from

1971

to 1972.

Based on

these

raw

numbers,

a l-]Xg/ny'

Figure

Table 1: Sample Statistics, 1969-1974
Table 2: Differences in County Characteristics by 1971 TSPs Level, 1971-1972 TSPs Change, and 1972 Nonatta Difference in Sample : Means between Groups of Counties
Table 3: Cross-Sectional Estimates of the Association between Mean TSPs and Infant Mortality Rates (estimated standard errors in parentheses)
Table 4: Fixed Effects Estimates of Association between Mean TSPs and Infant Mortality Rates (estimated standard errors in parentheses)
+7

Références

Documents relatifs

Jacques Turgeon, doyen de la Faculté de pharmacie de l’Université de Montréal, Mme Christine Landry et Mme Pascale Gervais, gagnantes du Prix Julien Braun (Université de

Comparison of the two time periods indicated that statistically significant areas of elevated infant and neonatal mortality shifted northwards for the city of Paris, are present only

What are the impacts of heat treatment on the structure and digestion kinetic of proteins of infant milk formulas mimicking the protein profile of human milk.

Here we use modeled concentrations from an ensemble of chemistry–climate models to estimate the global burden of anthropogenic outdoor air pollution on present-day premature

Estimates of global future premature mortality (IHD + STROKE + COPD + LC) for RCP8.5 in 2030 and 2100, for PM 2.5 reported by four models and PM 2.5 estimated as a sum of species

Model 2 estimated MOR values to vary between 1.1 and 1.14 for the risk of LBW and between 1.01 and 1.13 for the risk of TLBW. By comparison with the measure of association between

To our knowledge, no data are available on the relationship between temperature and mortality effect modified by chronic air pollution ex- posure and socio-economic factors

The main climate effect responsible for this increase in surface ozone pollution in Europe is the increase in temperature and solar radiation leading to an increase in