ion
MIT LIBRARIES DUPL
3 9080
02617
6872
' 'lI '1''> )i ;. i 1'li
lit-i
1
fff 1 Iw
Digitized
by
the
Internet
Archive
in
2011
with
funding
from
Boston
Library
Consortium
IVIember
Libraries
^^^
^>/
5Massachusetts
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 atOFTECHWLOGY
'I
MAR
n
2004
Air
Quality,
Infant
Mortality,
and
the
Clean Air
Act
of
1970^
Kenneth
Y.Chay
and Michael Greenstone
August 2003
*
We
thankDavid
Card,Doug
Almond,
Josh
Angrist,Gary
Becker,David
Cutler,John DiNardo,
CarlosDobkin,
Mark
Duggan,
Vernon
Henderson,
Charles Kolstad, JeffMilyo, Steve Pischke,and
numerous
seminar
participants for helpfulcomments.
JustinMcCrary,
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
isgratefullyacknowledged.
Air
Quality,
Infant
MortaUty,
and
the
Clean Air Act
of
1970
Abstract
We
examine
the effectsof
totalsuspended
particulates(TSPs)
airpollutionon
infanthealthusing theairqualityimprovements induced
by
the1970 Clean
AirAct
Amendments (CAAA).
This legislationimposed
strictregulationson
industrialpolluters in"nonattainment"
countieswithTSPs
concentrationsexceeding
the federal ceiling.We
usenonattainment
status asan
instrumentalvariableforTSPs
changes
toestimatetheir impact
on
infantmortahty changes
inthe firstyearthatthe1970
CAAA
was
inforce.TSPs
nonattainment status isassociatedwith sharp reductions inbothTSPs
pollutionand
infantmortality
from
1971 to 1972.The
greaterreductionsinnonattairmient counties nearthe federal ceiling relative to the "attainment" countiesnarrowly
below
the ceilingsuggestthat theregulations are the cause.We
estimatethataone
percent declineinTSPs
results ina 0.5 percentdechne
inthe infantmortalityrate.Most
of
these effects aredrivenby
a reduction indeaths occurringwithinone
month
of
birth, suggestingthat fetal
exposure
isapotentialbiologicalpathway.
The
resultsimply
thatroughly 1,300 fewer infantsdiedin
1972
thanwould
have
intheabsence of
theClean
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
ofEconomics
MIT
E52
50
Memorial
Drive
Cambridge,
MA
02142-1347
American
Bar Foundation and
NBER
Introduction
The
1970 passage
oftheClean
AirAct
Amendments
(CAAA)
and
the establishment oftheEnvironmental
ProtectionAgency
(EPA) marked
an unprecedented
attemptby
theU.S.government
tomandate
lower
levels ofairpollution.The
statedgoalof
theseeffortswas
toachieveairqualitystandardsthat "protect thepublic health." Inthe caseoftotal
suspended
particulates (TSPs),widely
thoughttobe
the
most
perniciousform
ofairpollution, theEPA
setmaximum
allowable concentrationsthateverycounty
was
requiredtomeet.' This study usesthereductions inTSPs
concentrationsinduced
by
the1970
CAAA
to estimatethe effectsof
TSPs
on
infantmortahty.Figure 1
shows
trends inaverageTSPs
pollutionand
infantmortalityrates for the entireUnited
States
from
1970-1990.^ It is apparentthatairqualityimproved
dramatically inthe1970s
and
1980s, withparticulateconcentrationsfallingfrom
an average of 95 micrograms-per-cubic
meter
(|J.g/m^)toabout
60
|J.g/m^.Almost
allof
thereductionoccurred
intwo
punctuated
periods,1971-1973
and
1980-1982.
The
figurealsoshows
thelargedecline inthe infantmortalityrateover
the entire period.Although
both
TSPs
pollutionand
infantmortality ratesfrendeddownward,
theirtime-seriescorrespondence
does
notprovide conclusive evidence
of
a causalrelationship. First,thepunctuated
declines inTSPs
arenot mirroredin the infantmortalityrateseries. Further, other potentialdeterminantsof
infant mortality(e.g.,socialinsurance
programs,
medical
technology)were
alsochanging
duringthisperiod.Chay
and Greenstone
(2003) establish thatmost
of
the1980-82 dechne
inTSPs
was
atfributabletothe differential
impacts
of
the1981-82
recession acrosscounties.The
study usesthe substantial differences inairpollutionreductions acrosssites toestimatetheimpact of
TSPs
on
infant mortality.We
find that aone
percent reductioninTSPs
results ina 0.35 percent declineinthe infant mortality rateatthecounty
level.Most
of
these effects aredrivenby
a declineindeaths occurring withinone
month
ofbirth,'Inthis
period,the
EPA
definedTSPs
toincludeallparticleswith diameterslessthan or equalto 100 micrometers(|im).
The
focusoffederalregulationshiftedtoparticulateswithdiameterslessthan orequalto 10 |im(PM|o) and2.5\im (PM2.5)in 1987 and 1997,respectively.
^
The
TSPs
pollution seriesisbasedonaverages acrossthe 1,000-1,300 countiesayear withTSPs
monitors,whileinfantmortalityratesare derivedfromthe entireU.S. andare forallcausesofdeath.
The
counties withTSPs
data accountforupwards of80%
of the U.S. population. Unlessotherwise noted, infantmortalityreferstodeath withinsuggestingthatfetal
exposure
duringpregnancy
is a potentialpathophysiologicmechanism.
Overall, theestimates
imply
thatabout
2,500fewer
infants diedfrom
1980-82
thanwould
have
intheabsence of
thedramatic reductionsin particulates pollution.
In
view
ofthe significance ofa potentialcausal linkbetween
TSPs
pollutionand
infant mortality,we
believeitisimportanttovalidate the resultsfrom
ouranalysis ofthe early 1980s. Inthispaper,we
examine
dataon
infantmortalityand
airpollutionfrom
theothermajor
episodeofsubstantialchanges
inTSPs
levels in theearly 1970s.We
beginby
demonstratingthatthesechanges
are largely attributable to the1970
CAAA.
This legislationdividedallU.S. countiesintoTSPs
nonattainmentand
attainmentcategories
based
on
whether
theirambient
TSPs
concenfrationswere
above
orbelow
the legislatedmaximum.
Industrial emittersof
TSPs
were
subject to significantlystricterregulations innonattainment countiesthaninattaiiunentones.We
find that the entiredeclineinTSPs
duringthe early1970s
occurredinnonattainment counties
and
that two-thirdsof
the1971-1974
declineinthese counties occurredbetween 1971 and
1972,thefirstyearthatthe1970
CAAA
was
in force. Consequently,this studyuses nonattaiimientstatusasan
instrumental variablefor1971-1972 changes
inTSPs
toestimatetheirimpact
on
changes
ininfant mortality.This researchdesign has severalattractivefeatures. First, itisplausible that federally-mandated regulatorypressure isorthogonaltocounty-level
changes
in infantmortalityrates, exceptthrough
itsimpact
on
airpollution-
thus,nonattainmentstatusmay
be
a valid instrument. Second, nonattainmentstatus isa discretefunction
of
thepreviousyear'sTSPs
levels. Thisdiscontinuityintheassignment
ofregulations
can
be used
togauge
the credibilityoftheresearchdesign. Third, ouranalysisprovides directand
easily interpretedestimatesof
theairqualityand
healthbenefitsofthe1970
Act
-
benefitsthatcan be
compared
with
the costs toindustryofcomplying
withthe regulations.Finally, thisstudyprovides a
unique
opportunitytocross-validate the resultsfrom
our earherwork by
examining
adifferentdesign appliedto adifferentera. In contrast to therecession-drivenTSPs
reductions
of
theearly 1980s,we
use regulation-inducedchanges
thatoccurred duringan
economic
expansion
(1971-1972). Thus,any
potentialbiasesdue
toeconomic
shocksarelikely tobe
mitigated.Also, particulatespollution levels
were
much
higher inthe early1970s
thaninthe early 1980s.As
aresult,
we
can
examine
adifferentrange of
the pollution-infantmortality functionand
analyzethe health benefits atthe federallymandated
concentrationceiling.We
implement
thedesign usingthemost
detailedand comprehensive
data availableon
infantbirths,deaths,
and
TSPs
levels for the1969-1974
period.We
find thatthe federallyimposed
county-level regulationsareassociatedwithlargereductionsinboththeTSPs
pollutionlevelsand
infant mortalityratesof
nonattainment
counties.We
estimatethataone
percentreductioninTSPs
results ina 0.5 percent declineintheinfant mortalityrateatthecounty
level.As
inour earUer study,most of
these effectsare drivenby
a reductioninthenumber
of
deathsoccurring withinone
month
ofbirth, suggestingthat fetalexposure
isapotentialbiologicalpathway.
Our
attemptstoprobe
therobustnessof
theseconclusions suggestthattheinstrumentalvariablesestimates
of
the effectof
TSPs
pollutionon
infant mortality arefar less sensitive tospecificationthan conventionalcross-sectionaland
fixed effects estimates. Further, thetimingand
locationof
theTSPs
and
infantmortality
improvements correspond remarkably
well.As
atestof
internalvalidity,we
find thattheregulation-induced
TSPs
reductionsareuncorrelated withinfantdeathsdue
toaccidentsand
homicides.Finally,
nonattainment
and
attainment countiesnearthe federalceiling exhibit discrete differences inTSPs
and
infantmortaUty
changes,butdo
notexhibitdifferences inotherobservablecharacteristics. Thisprovides our
most convincing evidence
thatthe regulations arecausallyrelatedtodeclines inbothTSPs
and
infantdeath.Overall, our findings supporttheconclusions that airpollution hasa causal effect
on
infantmortality
and
that the1970
CAAA
provided
significanthealth benefitsby
reducingTSPs
concentrations.The
estimatesimply
thatroughly
1,300fewer
infantsdiedin1972 than
would
have
intheabsence
of
theClean
Air Act. Also, theresultscorrespond
wellwith the findings inChay
and Greenstone
(2003).The
paper proceeds
as follows. SectionI providesbackground
on
the1970 Clean
AirAct.SectionsII
and
IIIdiscuss theresearchdesignand
theeconometric
methods
used
toimplement
thisdesign. Section
IV summarizes
the data,and
SectionV
presentstheempiricalfindings. SectionVI
I.
Background
on
the1970 Clean
Air
Act
Amendments
Before
1970
thefederalgovernment
didnot playa significant roleintheregulationof
airpollution, a responsibilityleftprimarilytostategovernments.^ In the
absence of
federallegislation,few
states actedtoimpose
strictregulationson
polluterswithintheirjurisdictions.Concerned
withthedetrimental health effects
of
persistentlyhigh
concentrationsof suspended
particulates pollution,and of
otherairpollutants,
Congress
passedtheClean
AirAct
Amendments
of 1970.The
centerpieceof
the1970
CAAA
was
theestablishmentof
separatefederalstandards,known
astheNational
Ambient
Air QualityStandards(NAAQS),
for six pollutants.The
CAAA's
goalwas
toreducelocal airpollutionconcentrationssothatallU.S. counties
would
be
incompliance
withtheNAAQS
by 1975
(with thepossibilityof
an
extensionto 1977).The
firststep inthisprocesswas
theassignment
of
pollution-specific "attairunent-nonattainment" designationsto allU.S. countiesforeach of
theregulatedpollutants.''
For
TSPs
pollution theEPA
was
requiredtodesignateacounty
asnonattainmentifits
TSPs
concentrationsexceeded
eitheroftwo
thresholds: 1)theaimual geometricmean
concentration
exceeded 75
flg/m^, or 2) thesecond
highestdailyconcentrationexceeded
260
|J.g/m^ This standard prevailedfrom
1971 until 1987,when
theEPA
shifteditsfocustotheregulationoffinerparticles.
To
achieve thesestandards, thefiftystateswere
requiredtoformulateand
enforceStateImplementation
Plans (SIPs)thatspecifiedpreciseabatement
activities.For
nonattainmentcounties,theSIPsdetailedplant-specificregulationsfor
every
major
sourceof
pollution.These
local rules orderedthatany
substantial investmentby
anew
orexistingplantmust be accompanied
by
theinstallationof
state-of-the-artpollution
abatement
equipment and
strictemissionsceilings.The
SIPsalsosetemissionlimits for existing plants innonattairmientcounties.^
Lave
andOmenn
(1981)andLiroff(1986) provide
more
detailsontheCAAAs.
Inaddition, seeGreenstone(2002)and
Chay
and Greenstone(2000)."
It isunclearwhetherthe
EPA
assignednonattainmentstatusatthecountyleveloratamore
aggregategeographicunitin the early 1970s.
Our
analysisassumesthattheassignmentwas
atthecountylevel.The
rationale forthischoiceisdiscussed inthe "NonattainmentData"subsectionoftheData Appendix. Further,below
we
find thatusingInstark contrast tothe oversightin
nonattainment
counties, therestrictionson
industrial pollutersinattainment counties
were
considerably less stringent. Large-scale investments, suchasnew
plantsand
largeexpansions atexisting plants, requiredthe installation
of
lessexpensive
(andless effective)pollutionabatement
equipment.Moreover,
existingplantsand
smallerinvestmentswere
essentiallyunregulated.Both
thestatesand
thefederalEPA
were
givensubstantialenforcement
powers
toensurethatthegoals
of
theCAAA
were
met.To
limitvariation in the intensityof
regulationacross states, thefederalEPA
had
toapprove
allstateregulationprograms.
On
thecompliance
side, statesinitiated theirown
inspection
programs and
frequently fined non-compliers. Further, theEPA
was
requiredtoimpose
itsown
proceduresforattainingcompliance
in statesthatfailed toadequately enforcethe standards.The
1970
CAAA
was
signedintolaw
by
Richard
Nixon
on
December
31, 1970.Four
months
later
on
April30, 1971,theEPA
announced
thefinalpublicadon of
theNAAQS
thatspecifiedthe nationalstandards forTSPs
concentrations.On
August
14, 1971,theEPA
pubhshed
"Requirements
for Preparation,Adoption
and
Submittalof Implementations
Plans"intheCode
of Federal Regulations,which
setforthhow
stateswere
towritetheirSIPs
inordertoachievecompliance
withtheNAAQS
by
1975. Finally, the
SIPs
were
due
totheEPA
by
January of
1972, thefirstyearinwhich
theCAAA
was
enforced.Appendix
Table
1shows
the timelineof
thekey
datesassociatedwiththe 1970
CAAA,
and
theData
Appendix
gives additionaldetails.Henderson
(1996) providesevidence
that theregulationswere
successfully enforced.He
finds thatozone
concentrations declinedmore
incountiesthatwere
nonattainment
forozone
than inattairunent counties.Chay
and Greenstone
(2000)find thatTSPs
levelsfellsubstantiallymore
incountiesthatwere
nonattainmentfor
TSPs
thaninattainment countiesafterthepassage of
the 1970
CAAA
and
throughoutthe 1970s.^
^Greenstone
(2002) providesfurtherevidenceonthe effectivenessofthe regulations.
He
finds thatnonattainmentstatusisassociatedwith modestreductionsintheemployment,investment,and shipments ofpolluting
manufacturers. Interestingly,the regulationof
TSPs
hashttleassociationwithchangesinemployment. Instead, the overallemployment
declinesaredrivenmostlyby
the regulationofotherairpollutants.II.
Research Design
Previous researchhas
documented
astatisticalassociationbetween
TSPs
concentrationsand
adultmortality/
However,
thereliabilityof
the evidence hasbeen
seriouslyquestionedforseveral reasons.First, sinceairpollutionisnot
randomly
assigned acrosslocations,previouscross-sectional studiesmay
not
be
adequatelycontrolling foranumber
of
potentialconfounding
determinantsof
adultmortality(Pope
and
Dockery
1996;Fumento
1997).For example,
areas with higherpollution levels alsotendtohave
higherpopulationdensities,different
economic
conditions,and
highercrime
rates, allof
which
could
impactadult health. Second, the lifetime
exposure
of
adults toairpollutionisunknown.
Many
studies implicitlyassume
thatthecurrentpollutionconcentrationobserved
at asiteaccuratelymeasures each
resident's lifetimeexposure. Third, theexcessadultdeaths thatare attributed totemporarily elevatedair
pollution levels in time-series studies
may
be
occurringamong
theaheady
sickand
representlittlelossinhfe
expectancy
(Spix, etal. 1994; Lipfertand
Wyzga
1995).In the
absence of
randomized
clinicaltrials,this study usestheairquahty
improvements induced
by
the 1970
Clean
AirAct
Amendments
inthefirstyearthattheywere
inforcetoestimatetheimpact
of
TSPs
on
infantmortality.A
researchdesignbased
on
comparisons
between
nonattainmentand
attainment counties hasseveral attractive features.^ First,
TSPs
nonattairunent status isassociatedwith
sharpdifferences in
changes
inTSPs
acrosssites inthe early 1970s. PanelA
of Figure 2shows
the trendsin
TSPs
levelsfrom
1969
to 1974
separately for countiesthatwere
nonattainmentand
attainmentforTSPs
in1972
-
thefirstyearofCAAA
enforcement.^*Studies ofthe adultmortalityeffectsofparticulatespollutionincludeLave andSeskin (1977),
Pope
andDockery
(1996),Dockery and
Pope
(1996),Dockery,etal.(1993), andPope, etal. (1995). Chay, Dobkin,and Greenstone(2003)examinetheassociationbetweenadultandelderlymortalityandregulation-induced
TSPs
declines. 'Despite extensiveefforts,
we
were
unabletoobtain alistofthecountiesthatwereinitiallydesignatedTSPs
nonattaimnentbythe
EPA.
Giventhetimelinefor statestocomply
withtheCAAA,
states likelyascertainedthe identityofthecountieswithTSPs
concentrationsabovetheNAAQS
whilewritingtheirSlPs duringthesecondhalfof1971.
We
assumethattheyreliedonthe availableTSPs
concentrations datairom 1970todeterminewhichcountieswould benonattainmentfor 1972. Thus,
we
assigncountieswithTSPs
concentrationsexceedingtheNAAQS
in 1970tothe 1972TSPs
nonattaiimientcategory. AllothercountieswithnonmissingTSPs
data aredesignated attainmentfor thatyear.
The
measures ofTSPs
concentrationsarederivedfromthesame
network ofTSPs
monitorsusedby
thestatesandtheEPA
in thisperiod.The
DataAppendix
providesfurther detailsonourassignmentrule.
*
The
sampleconsistsofthe401 countieswithcontinuousmonitorreadings from 1969-1974
-
229
ofthese countiesThe
figuredocuments
thatbeforethe1970
CAAA,
TSPs
concentrationswere
40-|ig/m^higher innonattainmentcounties.
While
thepollutiontrends are similarinthetwo
groups
from
1969
to 1971,thereisasharp
break
intrendinnonattainment countiesafterimplementation
oftheCAAA.
From
197
1-74newly
regulated countieshad
astunning20-|J,g/m^ reductioninTSPs,
whileTSPs
fellby
only
3 p.g/m^inattainmentcounties.
These comparisons
suggestthat virtually the entirenationaldechne
inTSPs
from
1971-74
inFigure 1was
attributabletothe regulations. In addition,two-thirdsof
the1971-74
TSPs
declinein
nonattainment
counties occurredfrom
1971-72,the firstyearof
CAAA
enforcement. Consequently,our
analysis focuseson
thisabrupt, one-yearimprovement
inairquahty.^Second,
below
we
findevidence
thatthisdesignmay
reduce
the roleof
omittedvariablesbias inanalyzingtheassociation
between
TSPs
and
infant mortality. PanelB
of
Figure2 displaysthetrends in internalinfantmortality rates(IMR)
inthesame two
setsofcounties.'"It
shows
thatbeforetheCAAA
was
errforced, infantmortalityrateswere
120-150
deaths per100,000
births higherinnonattaimnentcounties. Further, attainment
and nonattainment
countieshad
similarIMR
trendsbetween
1969 and
1971, with attainment counties experiencing a slightly larger decline.
However,
thispattern isreversedbetween 1971
and
1972,withtheattainment-nonattainmentIMR
gap narrowing
by
over75
deathsper100,000
births.The
figurealsoreveals thatattainment countieshad
a largerdecline in infantmortalitybetween
1972
and
1974.The
correspondence of
the trendbreaks inTSPs
and
infantmortalitydifferencesbetween
nonattainment
and
attainment countiesin1972
suggestsa causalrelationship. Since thepatterns after1972
arelesscompelhng,
we
probe
thecasefor causalitymore
rigorouslyintheanalysis below.To
1970.
'
We
probedhow
industrialpolluters
may
have significantlyreducedTSPs
emissionsinthisnarrowtime frame.Electrostatic precipitatorswere one oftheprimarycontroldevicesusedtoabateparticulatespollutionduringthis
period.
The
IndustrialGas
CleaningInstitutereportsthatwhileannual sales ofprecipitatorswere approximately $50millionfrom 1968 through 1970,thisfiguredoubledto
$100
millionin 1971 and 1972 (White 1984). AccordingtoDonald
Hug
oftheEnvironmental ElementsCorporation, asupplierofpollutionabatementequipment, anelectrostatic precipitatororderedin 1971 couldbeoperatingin 1972.
We
concludethatthe largedecreaseinTSPs
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" causesofmaximize
the signal-to-noiseratio inTSPs
changes
and minimize
the potential biasesfrom confounding
changes
inothervariables,we
focuson
the strikingone-yearchanges
thatoccurredbetween
1971and
1972.
The
analysisuses nonattainmentstatusasan
instrumentalvariable for1971-1972 changes
inTSPs.
Since federally-mandated regulatory pressure isplausibly orthogonalto
changes
in infant mortalityrates,except
through
itsimpact
on
airpollution,nonattainmentstatusmay
be
avalidinstrument. Consistent withthis,we
findlittleassociationbetween
nonattainment statusand
otherobservablevariables,includingparents' characteristics, prenatalcareutilization,
and
transferpayments
from
socialprograms.Further,nonattainmentstatusin
1972
was
adiscrete fiinctionof
the annualgeometricmean
and
second
highestdailyconcentrationsofTSPs
in theregulationselection year, 1970.Thus,
theassignment
of
regulatorystatushas thefeatureofaquasi-experimentalregression-discontinuitydesign(Cook and
Campbell
1979). Ifother factors affecting infantmortalityare similar for countiesjustabove and
justbelow
theregulatorythresholds, thencomparing
outcome
changes
innonattainmentand
attainment countieswith
pre-regulationTSPs
levelsjustaround
thethresholdwill controlforallomittedfactors correlatedwith
TSPs.
Under
thisassumption
discretedifferences inmean
outcome
changes
between
nonattairmient
and
attairmientcountiesnear the federal ceilingscan be
attributed tothe regulations.The
discontinuityinthe
assignment
ofregulationsprovides a valuableopportunitytogauge
the credibilityoftheresearchdesign
and develop
convincingspecificationtests.Third, our focus
on
infant, ratherthanadult, mortalitymay
circumvent
the other importantcriticisms of previous research.
The
problem of
unknown
lifetimeexposure
to pollutionis significantly mitigated, ifnotsolved,by
thelow
migrationratesof
pregnantwomen
and
infants. Thus,we
assignTSPs
levels toinfants
based
on
theexposure
ofthemother
duringpregnancy
and
theexposure
of
thenewborn
during thefirstfew
months
afterbirth(we
focuson
deaths within24
hours,28
days,and
1 yearofbirth).Further, since the mortality rateishigher inthefirstyear
of
lifethaninthenext20
yearscombined
(NCHS
1993),itis likelythatinfantdeaths representa large loss inlife expectancy. Incontrast, time-series studies
of
adultand
elderlymortahty based
on
dailyairpollution fluctuationsmay
be
detectingeffects
among
thealready veryilland
dying. Inthiscase,the actual lossoflifeexpectancy
may
be
much
smaller than thesestudies implicitlysuggest.A
finaladvantage
of
thedesignisthatitpermits a directanalysis ofthehealthbenefits ofTSPs
reductions atlevelsjust
above
theEPA's
maximum
allowable concentration.The
optimalityoftheCAAAs
depends
cruciallyon
theexistenceand
relativemagnitudes of
thehealtheffectsabove and
nearthe federal regulatorythreshold.
III.
Econometric
Methods
Our
analysiscompares changes
ininfantmortalityratesand
TSPs
pollutionfrom
1971-1972
forcountiesthat
were
nonattainmentand
attainmentforTSPs
in 1972. Here,we
discuss theeconometric
models used
to estimatethe infantmortality-TSPs
association.For
simplicity, it isassumed
thatthe "true" effectof exposure
toparticulatespollutionishomogeneous
acrossinfantsand
overtime.The
cross-sectionalmodel
predominantly used
inthe literatureon
airpollutionand
healthis:(1)
yc
=
Xc'P +
eTc,+
ec, Ect=
ttc+
Uc,(2) Tct
=
Xct'n+
Tic Tlet=
K
+
Vet,where
yet isthe infantmortalityrateincounty
c inyeart,Xct isavectorof observed
characteristics,T^
isthe
mean
of
TSPs
acrossall monitorsinthecounty,and
8ctand
Tictaretheunobservable
determinantsof
infantmortalityrates
and
TSPs
levels,respectively.The
coefficient9
isthe "true" effectof
TSPs
on
infantmortality.
For
consistent estimation, theleastsquares estimatorof 9
requiresE[ect'nct]=0. Ifthere areomittedpermanent
(ttcand
Xc) ortransitory(Uctand
Vd) factorsthatcovary with both
TSPs
and
infant mortality,then thecross-sectionalestimatorwillbe
biased.With
repeated observations overtime, a"fixed-effects"model
implies that first-differencing thedatawill
absorb
thepermanent county
effects, (Xcand
^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 estimatorof
requiresE[(uct-Uct-i)(Vct-
Vct-i)]=0.That
is, thereareno
unobserved shocks
toTSPs
levelsthatcovary
withunobserved shocks
to infant mortalityrates.While
first-differencingremoves
permanent
sourcesof
bias, itmay
magnify any
attenuation biasdue
tomeasurement
error. Since ourmeasure
ofTSPs
concentrations isan average
across themonitorsina county, itimperfectly capturesindividual
exposure
toTSPs."
Thismismeasurement
will attenuate theestimated
TSPs
regressioncoefficientleadingtoabias thatisproportionaltothe fractionofthetotalvariation in
TSPs
that is attributable tomeasurement
error.To
theextent that theserialcorrelation in "true"TSPs
exposure
is greaterthanthe autocorrelation intheTSPs
measurement
error, first-differencingthedatawill increasethe
magnitude
oftheattenuationbias.'^Now
suppose
thereexistsan
instrumentalvariable, Zc,that causeschanges
inTSPs
withouthaving
adirect effecton
infantmortality ratechanges.Such
a variablewould
purge
theestimatesof
the biasesdue
toboth
omittedvariablesand
measurement
error.One
plausibleinstrumentisthe 1970
C
AAA
regulatoryintervention for
TSPs,
measured
by
the1972
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 istheregulatorystatusofcounty
c in 1972, 1(»)isan
indicatorfunction equaltoone
iftheenclosed statement istrue,
and
T
isthemaximum
concentrationofTSPs
allowed
by
the federal regulations.Regulatory
status in1972
isadiscretefunctionof1970
pollution levels.'^An
attractive featureof
thisresearchdesign is thatthereduced-form
relationsbetween 1972
TSPs
nonattainmentstatus
and
thetwo endogenous
variablesprovidedirect estimatesof
the benefitsof
theregulations. In particular, FItz
from
equation(5)measures
the1971-1972
airqualityimprovement
innonattainment countiesrelativetothe
change
inattairunentcounties. Intheotherreduced-form
equation, (7) yc72-yc7I=
(Xc72-Xc7l)'nyx+
Zc72nyz+
(Uc72"Uc7l),'
' Duringthe 1971-1972period, the
mean
andmedian
numbersofmonitorsinacountyare4.2and2,respectively.'^However,
inour contextitisunclearwhethertheserialcorrelationinthe
TSPs
measurement
errorfrom 1971to1972issmalleror largerthantheserialcorrelation intrue
TSPs
exposure.'^
Forsimphcity,
we
havewritten(6)asifregulatorystatusisa functionofasinglethresholdcrossing. If T^^yg andT^"^^ are theannual geometric
mean
and2" highestdailyTSPs
concentrations, respectively,thenthe actualregulatoryinstrumentusedis 1(T^^q
>
75|ig/m^ orT^^
>
260|ag/m^).Only
sixcountieswerenonattairmientin1972 forexceedingthe2"''
highestdailyconcentrationthreshold, butnottheannual geometric
mean
ceiling. Seefootnote 7andtheData
Appendix
fordetailsonthetiming oftheCAAA
intervention.riyzcapturesthe relative
improvement
inthe infantmortahty
rateinnonattainment countiesbetween
1971
and
1972.Since the instrumentalvariables (IV) estimator(9iv)
of
the effectof
TSPs
on
infant mortalityisexactlyidentified, itis equalto theratioofthe
two
reduced-form
relations-
i.e.,Biv=
Tlyz/TiTz-Two
sufficientconditions for6iv toprovide a consistentestimate
of
theTSPs
effect areHtz
'^and
E[Vc7o(Uc72- Uc7i)]=0-
The
firstconditionrequires that theregulationsinduced
airqualityimprovements.
The
second
conditionrequires that transitory
shocks
toTSPs
levels in1970
are orthogonaltounobserved
shocks to infantmortalityratesfrom
1971-1972.'''Even
ifE[Vc7o(Uc72 -Uc7i)]^
0, causalinferenceson
9
may
be
possibleby
leveragingtheregression discontinuity
(RD)
design implicit in the 1(•) functionthatdeterminesnonattainmentstatus.For example,
ifE[vc7o(Uc72-Uc7i)]=
intheneighborhood of
theregulatoryceiling(i.e.,75
)J,g/m^), thenacomparison of changes
innonattainment
and
attainment counties inthisneighborhood
willcontrolforallomittedvariables.
Below,
we
presentIV
estimatesbased
on
allcountiesand
RD
estimatesinwhich
thesample
is limitedtocounties withTSPs
concentrations "near"the75
|ig/m^thresholdintheregulationselection year.'^ Inthenextsection,
we
show
thattheobservablecharacteristicsof
nonattairmientand
attainment counties nearthe federalTSPs
ceilingarevery
similar.Another
potentialsourceof
biascomes from
uruneasured
state-specificdeterminants ofinfantmortality(e.g., generosity
of
stateMedicaid programs,
stateweather
conditions). Sincethe analysis is atthe
county
level, insome
specificationswe
control for unrestricted state-time effects in infant mortality toabsorb
allunobservables
thatvary
acrossstatesovertime. Here, the effectof
TSPs
is identifiedusing"only
comparisons
between
attainmentand nonattainment
countieswithinthesame
state. Figure3provides agraphical
overview of
the locationof attainmentand
nonattainment counties in 1972.A
county's
shading
indicatesitsregulatorystatus-
lightgray
for attainment,blackfor nonattairmient,and
Inthesimplestcase,the
IV
estimatorwillbeconsistentif1972 nonattainmentstatus isorthogonaltounobservedinfantmortalityshocks
-
i.e.,E[Zc72(Uc72-Uc7i)]=0,which
isa stronger condition thanE[Vc7o(Uc72-Uc7i)]=0. IftherelationshipbetweenVc7oand(Uc72-Uc7i)issufficiently"smooth"attheregulatoryceilings,then causal inferencemay
alsobepossibleby
including smoothflmctionsoftheselection variable,Tc7o,ascontrols. Below,we
alsoprovide
RD
estimatesthatare adjustedfora quadraticin Tc7o.whiteforthe countieswithout
TSPs
pollutionmonitors.The
pervasivenessof
the1972
TSPs
regulatoryprogram
is evident. Inadditiontothe traditional countiesintheRust
Beltand
intheSouth Coast
AirBasin
around
Los
Angeles,scoresof other countieswere
designated nonattainment. Importantly,most
statescontainboth attainmentand
nonattainment counties.IV.
Data
Sources
and
Summary
StatisticsTo
implement
our evaluation strategy,we
usecounty
leveldataon
airpollution, infant health,and
a
number
of
additional control variables for the1969-1974
period. Here,we
describe these dataand
provide
summary
statisticsshowing
thatthenonattainment-attainment research design balancestheobservablecharacteristics
of
counties.More
detailson
thedatasourcesand
variablesarecontainedin theData Appendix.
A.
Data
SourcesThe
healthoutcome
variablesand
anumber
of
the control variablescome
from
theNational MortalityDetail Filesand
National NatalityDetail Files,which
arederivedfrom
censusesof
deathand
birthcertificates.'^
We
merge
the individual birthand
infantdeath recordsatthecounty
levelforeach
yeartocreate
county
by
year cells.For each
cell, the infantmortalityrate is calculated as theratioof
thetotal
number
of
infantdeathsof
a certaintype (ageand
causeof
death) to thetotalnumber
ofbirths.The
mortality rateswithin
24
hours,28 days
(neonatal),and one
yearofbirtharecomputed
separatelyforeach
year. Fatalitiesthatoccur
soon
afterbirtharethoughtto reflectpoor
fetaldevelopment.
Thus,we
use
theestimatedeffects of
TSPs
on 24-hour and
neonatalmortality toprobe
whether
fetalexposure
toTSPs
may
have
adversehealtheffects.We
also calculateseparate mortalityratesfordeathsdue
to internalhealthreasonsand
deathsdue
toexternal"non-health"relatedcauses
such
asaccidentsand
homicides.'' Internalinfant deaths arenot'*
The
NatalityFilesprovidea100%
censusofallbirthsin every yearfrom 1969-1974.The
MortalityFilesprovide a100%
censusofall deathsinevery year exceptfor 1972,inwhicha50%
random
sample ofalldeathsisprovided."
"Internal"and"external"deathsspanallpossiblecausesofdeath. Seefootnote 10forthe definitionofthe internalandexternalcategoriesbased onthe8"'ICD
codes.disaggregatedfurther for
two
broad
reasons. First, coroners assignabout70%
of
thesedeathstotwo
categories
—
"certaincausesof
perinatalmorbidityand mortahty" and
"congenitalanomalies." Second,since thepathophysiological
pathways through
which
TSPs
may
affect fetalor infanthealth are largelyunknown,
thereisno
aprioribasis fordividing these causesof
deathinto ones thatareand
arenot plausiblyrelatedtoTSPs.'^However,
itseems
reasonabletoassume
that thereisno
causalpathway
Unking
airpollution to externalcauses ofdeath.So
we
usethe estimated associationbetween
externalmortalityrates
and
TSPs
pollution as acheck
on
the internal consistencyof
ourfmdings."
The
microdata
on
thecontrolvariables available in theNatalityFiles arealsoaggregated intoannual
county
cells.The
detailedNatalityvariablesincludeinformationon:socioeconomic
and
demographic
characteristicsofthe parents;medical
system
utilization, including prenatalcareusage
(Rosenzweig
and
Schultz 1983;Instituteof
Medicine
1985); maternal healthendowment,
such
asmother's
age
and pregnancy
history(Rosenzweig
and
Schultz 1983;Rosenzweig
and
Wolpin
1991);and
infantbirthweight. Since
TSPs
may
affectfetaldevelopment,
we
useinfantbirthweight
asanotheroutcome
variable.The
Data
Appendix
describes the Natalityvariablesused
intheanalysis indetail.Annual,
monitor-level dataon
TSPs
concentrationswere
obtainedby
filingaFreedom
of
Information
Act
requestwiththeEPA.
This yieldedannual
summary
informationfrom
theEPA's
nationwide
network of
pollutionmonitors, includingthe locationofeach
monitor
and
their readings.The
dataalsocontaininformation
on
thearmual
geometric
mean TSPs
concentrationforeach monitor
inacounty
and
thenumber
ofdailymonitor
readingsexceeding
the federal standardsineach
year.These
measures
areused
todetermine
theannual
nonattaimnent
statusof each
county
(see theData
Appendix
for
more
details).From
the data,we
alsocalculatedthemean
of
TSPs
concentrationsineach
county
and
year,
which
we
use
as ourmeasure of
TSPs
exposure
inthe analysis.'"' SeeUtelland
Samet
(1996)fora
summary
oftheevidenceonthepathophysiologicalrelationshipbetweenTSPs
andhuman
health.Thatis,the associationbetween
TSPs
andexternal infantdeathsmay
resultfrom unobservedsecularfactors that affectall typesofinfant death.Specifically, thearmual
TSPs
concentrationusedistheweightedaverageoftheannualarithmeticmeans
of each monitorin thecounty,withthenumber
of observations permonitor usedasweights.Some
recent researchhas focused onthe effectsofPM-IO
particles.Most
ofthesestudiesmultipliedTSPs
concentrationsbyaconstantfactor toobtainPM-10
measures (e.g..Pope, Dockery,and
Schwartz, 1995).Inadditiontothe
key
infanthealthand
pollution data,we
usea varietyof
othercounty-leveldataas controls. Per-capita
income
datacome
from
theBureau
ofEconomic
Analysis' annual series,which
provides themost
comprehensive
measure of income
availableatthecounty
level. This filealsoprovidesannual, county-level data
on
per-capita net earnings, theratiooftotalemployment
tothetotal population,and
theratioof
totalmanufacturing
employment
tothe totalpopulation.The
RegionalEconomic
InformationSystem
fileprovidesannual, county-level dataon
several different categoriesoftransferpayments.
These
include separateseries fortotal transferpayments;
totalmedical
carepayments;
public expenditureson medical
careforlow-income
individuals(primarilyMedicaid
and
local assistanceprograms);
income
maintenance
benefits;family assistancepayments,
includingAFDC;
Food
Stamps
payments;
and
Unemployment
Insurancebenefits.Table
1 presentssummary
informationfrom 1969-1974
forthe501
countieswithavailableTSPs
and
infantmortality data in 1970, 1971,and
1972. This is ourprimary
sample
inthebelow
analysis. In1969 and
1973-74,thesample
is fiirtherresfrictedtocounties withnonmissing
dataon
TSPs
and
infantmortalityinthoseyears.
The
toprows
of
the tableshow
thatthesecounties accountfor over60%
of
the 3.2-3.7 millionlivebirthsthatoccur annuallyinthe
United
Statesin thisperiod.The
nextrows
present infantfatalitiesper
100,000
live births,by
causeof
death.The
internalinfantmortality ratein 1971 is 1.8deaths within a yearofbirthper100
births,whileinfantdeathsdue
to externalcausesaremuch
rarer. In 1971, about45
and 75
percentofall internalinfantdeathswithinayearof
birthoccur withinthefirst24
hoursand
28
days
of
birth,respectively.The
tablealsoshows
national trends in infant mortality,averageTSPs
concentrations,and
per-capitaincome
across counties. Internalinfant mortalityratesdecreasedsteadilyfrom
1969-74.TSPs, on
theotherhand,fellby
over 18 percentfrom
1971-74, butwere
relativelystablefrom
1969-71. Per-capitaincome
increasedby
8.5 percentfrom
1971-73, beforefallingslightlyfrom
1973-74,
presumably
aresultof
the recession.The
remaining
rows of Table
1 present1969-74
trends forseveralof
the control variablesused
inthe analysis. Importantly, the
magnitude of changes
inthesevariablesissmallwhen
compared
totheTSPs
pollutionchanges
that occurredfrom
1971-72.Our
measures of
parentalbackground
characteristics 14suggestthat theaverage
socioeconomic
status(SES) of
parentswas
decliningfrom
1969-74, as the fractionsof
births to single,high school dropout,and
blackmothers
increased.On
theotherhand,theHkelihood
of
amother
receiving prenatal care orhaving
aprenatalvisitinthe firsttrimesterincreasedslightly.
The
percentageof
births atfributabletoteenagersrose,and
theshareamong
women
aged 35
and
overfell."' Finally,the shareoffirst-timebirths increasedslightly
while
the fractionofmothers
who
had
apriorfetaldeath
remained
stable.B.
Balancing
of Observable
CharacteristicsBefore
proceeding,we
examine
whether
thenonattainment
instrumentalvariable is orthogonalto theobservablepredictorsof
infant mortality.While
itisnotaformaltestof
theexogeneityof
theinstrument,it
seems
reasonabletopresume
thatresearchdesignsthatmeet
thiscriterionmay
sufferfrom
smalleromittedvariablesbias. First, designs thatbalancetheobservable covariatesmay
be
more
likely to alsobalance
theunobservables
(Altonji, Elder,and Taber
2000)."'Second,
iftheinstrumentbalancestheobservables, thenconsistent inference
does
notdepend
on
fiinctionalform
assumptions
on
the relationsbetween
theobservableconfounders
and
infant mortality. Estimatorsthatmisspecify thesefiinctionalforms
(e.g.,hnear
regressionadjustmentwhen
theconditional expectationsfianctionisnonlinear)willbe
biased.
Table
2
shows
the associationof
TSPs
levels,TSPs
changes,and
TSPs
nonattainmentstatuswith otherpotential correlatesof
infantmortalityusingthesame
sample of
counties asinTable
1.Each
column
ofthe tablepresentsthedifferences inthe variablemeans between two
setsof
counties (with standarderrors inparentheses).The
firstcolumn
contains differences inthe 1971 levels ofthe covariatesbetween
counties withhigh (above
themedian)
and
low
(below
themedian)
TSPs
concentrations in 1971.If
TSPs
levelswere randomly
assigned acrosscounties,one
would
expectveryfew
significant differencesbetween
thetwo
groups.However,
thereare significantdifferences for severalkey
variables, including"'
Rees,et. al. (1996)find that
much
ofthecorrelationbetween poorinfanthealthandmother'sageisduetothe higher incidenceoflowbirthweightbirthsamong
teenagersandwomen
35and
older.^^
For example,inexperimental designstestingfordifferencesintheobservablecharacteristicsofthetreatmentand
controlgroups isusedtoassesswhetherthe treatment
was
actuallyrandomly
assigned. 15mother's
race,immigrant
status, useof
prenatalcare,and
likelihoodof being
ateenager.Though
not presentedinthetable, thereare also largeand
significantdifferencesbetween
thecounty
groups in per-capita receipts ofincome
maintenance
benefits, familyassistancepayments,
and
publicexpenditureson
medical
careforlow-income
individuals(e.g., seeChay, Dobkin,
and Greenstone
2003). Overall, thesefindings suggestthat"conventional"cross-sectionalestimates
may
be
biaseddue
toomittedvariables.The
nexttwo columns
ofTable
2perform
a similar analysis for1971-1972
TSPs
changes.They
containdifferences in 1971 variable levelsand 1971-72
variablechanges
between
countiesthathad
large(above
themedian) and
small(below
themedian)
reductionsinTSPs
between
1971and
1972.These
two
county groups
exhibit significant differencesinthe 1971 levels of percapitaincome,
mother'simmigrant
status,prenatal careuse, likelihood
of
being over34
yearsold,and pregnancy
history.There
arealso significantdifferences in the1971-72 changes
inmother'sand
father's educationlevelsand
marginallysignificantdifferences inthe
changes
inthefi-actionsofmothers
who
are teenagersand
who
had
apriorfetaldeath.
Thus,
fixed effectsmodels
may
alsoleadtobiasedinference.The
nexttwo columns
ofthe tableshow
the differencesin 1971 levelsand
197
1-72changes
between nonattainment
and
attaiimientcounties.While
nonattainmentstatusisassociated witha largeand
significantreductioninTSPs
concentrations, it isuncorrected
with nearlyallof
the 1971 levelsand
1971-72 changes
inthe covariates, withtheonly exceptionsbeing
percapitaincome and
mother's racein1971.
The
differences inthechanges
inmother'sand
father'seducationand
inthechanges
in theincidences of
mothers
who
areteenagersand
experienced aprevious fetaldeathare greatlyreduced
by
comparing nonattainment and
attainmentcounties. Also, incontrast totheprevioustwo
setsof county
group
comparisons, nonattainmentand
attainment counties exhibitlittle differencein theirreceiptsof
transfer
payments
firom socialprograms
(resultsavailable fi^omtheauthors). Thus, itappearsthattheattainment-nonattainment researchdesigndoes a betterjob of balancingthe observables."^
^^
Itshould benotedthatthecompositionofcounties
who
arenonattainmentandattainmentin 1972isdifferentfromthecompositionof countieswith "high"and"low"
mean
TSPs
in 1971 (i.e.,column
1).Inthelast
two
columns,
thesample
isfurther restrictedtothe176
countieswith1970
geometricmean TSPs
concentrationsbetween
60 and
90
|ag/m^-
i.e., countieswithTSPs
levels neartheTSPs
regulatoryceiling intheregulationselection year. Itis evidentthat focusing
on
nonattainmentand
attainment counties neartheregulatorythreshold eliminates all
of
the significant differencesinthecovariatelevels
and
changes, including per capita income,mother's
race,and
the infantmortality rateintheyear before
CAAA
enforcement.While
nonattainment
status isorthogonaltoall ofthe other covariates, thesecond
row
shows
thatitis still associatedwith
a sharp reductioninTSPs
levelsbetween
1971
and 1972
forcounties neartheregulatorythreshold."''These
resultsprovide reassuring evidenceon
the quality
of
the instrumental variablesresearchdesignused
in thisstudy.We
probe
thisfurtherbelow.V.
Empirical
Results A. Cross-Sectionaland Fixed
EffectsResultsFirst,
we
replicatethe conventionalcross-sectionalapproach
toestimatingtheassociationbetween
TSPs
pollutionand
infant mortalitythatiscommon
intheliterature.Table
3 presents theregression estimates
of
the effectof
TSPs
on
thenumber
of
internal infantdeaths within a year ofbirthper
100,000
livebirths foreach
cross-sectionfrom
1969-1974.
The
finalrow
containsthe results for thepooled 1969-1974
data.Column
1 presentstheunadjustedTSPs
coefficient;column
2 includesthebasiccontrol variables available inthe Natalitydata;
column
3 includesadditionalNatalityvariablesand
flexiblefunctional
forms
forthevariables;and
column 4
further adjusts forper-capitaincome,
earnings,employment, and
transferpayments by
source."^Columns
5and
6add
unrestrictedstateeffects to thecolumn
2
and
column 4
specifications, respectively.The
sample
sizesand
R''s oftheregressions areshown
in brackets.There
iswide
variabilityintheestimated effectsof
TSPs,
both across specifications foragiven cross-sectionand
acrosscross-sections foragivenspecification.While
theraw
correlationsincolumn
1Giventhenotationinequations(6)and(7),itshouldbe notedthat 1972nonattainmentstatus isorthogonaltothe levelsofcovariatesin 1970aswell(resultsavailablefromthe authors).
^
The
controlvariablesincludedineachspecification arelistedintheData Appendix. 17
are allpositive,only those
from
the 1969 and
1974
cross-sections arestatistically significantatconventional levels."^ Includingthebasic Natalitycontrolsin
column
2reduces thepointestimates substantially inmost
years,even
astheprecision ofthe estimatesincreasesdue
tothegreatlyimproved
fitof
the regressions. In addition, onlytwo
of
the estimatesincolumns
3and 4
are significantatconventionallevels,
and one of
thesehas a perverse negativesign. This isalso trueofthemost
unrestricted specificationin
column
6thatalso adjusts forstatefixedeffects.The
largestpositiveestimatesfrom
the cross-sectionalanalysesimply
thata l-)J,g/m^reductioninmean TSPs
resultsinroughly
3 fewer internalinfantdeaths per100,000
livebirths,which
isan
elasticityof0.14. Overall,
however,
thereis littleevidence ofa systematic cross-sectionalassociationbetween
particulatespollution
and
infantsurvivalrates.While
the1974
cross-sectionproduces
estimates that are positive, significantand
slightlylesssensitive to specification, the 1972
cross-sectionprovides estimatesthatare routinely negative.
The
sensitivityof
the results to theyear analyzedand
thesetof
variablesused
as controlssuggests that omittedvariables
may
playan
importantroleincross-sectional analysis.We
suspectthat thefragility
of
thecross-sectionalresultstochanges
inspecificationand
sample
is likely toapplytotheestimation
of
the relationshipbetween
othermeasures of
human
health(e.g., adult mortality)and environmental
quality(e.g., otherairpollutants).Table
4
presentsthefixed effects estimatesof
the associationbetween
mean TSPs
and
internalinfantmortalityrates
based
on
1969-1974 pooled
data(firstpanel)and 1971-1972
first-differenceddata (secondpanel).The columns
correspond
to similarregressionspecificationsas thoseinTable
3.The
1971-1972
firstdifferences resultsprovide a baselineforcomparison
tothesubsequent
instrumentalvariablesresultsthatusethe
same
samples
and
control variables.The
inclusionofcounty
fixedeffectsor the useof
first-differenceddataeliminatesthe bias in the cross-sectional estimates attributable totime-invariantomittedfactors that
vary
across counties.However,
theseapproaches
willbe
biasedifthereareunobserved shocks
thatare correlatedwith
changes
inTSPs
and
infantmortalityrates.^^
Itisworthnotingthatthecomparison ofcountieswith"high"and "low" 1971
TSPs
levelsinthefirstcolumn
of Table2 impliesanunadjustedTSPs
effectof2.8 (129.8/45.8),whichis75%
greaterthantheestimatedcorrelation for 1971 in Table3 (1.6). ThisisconsistentwithmeasurementerrorinTSPs
leadingtosubstantive attenuationbias,sincetheaggregationinTable2 "averages out"themeasurementerrorsover
many
counties.The
1969-1974
panelshows
thattheinclusionof
county-specificintercepts greatlyimproves
thefit
of
theregressionswhen
compared
tothepooled
cross-sectionalregressions in thelastrow
of Table
3.The
raw
fixed effects correlation incolumn
1 ispositiveand
highlysignificant.However,
theestimatedTSPs
coefficientfalls significantlywhen
the analysis controls for theNatalityvariables,economic
conditions, transfer
payments,
yeareffectsand
state-yeareffects.The
1971-1972
firstdifferences results alsosuggestlittleassociationbetween
changes
inTSPs
and changes
ininfantmortalityrates.The
fixedeffects associationbetween
TSPs
and
infantmortality issmalland
sensitive to specification,which
is consistentwith
potentialbiasesdue
toomittedvariables and/ormeasurement
error.''We
conclude
thatconventionalapproaches
may
notprovidereliableestimatesof
the causal effectof
TSPs
on
infantmortalityand
turnourattention totheinstrumentalvariables design outlined above.B.
The
Impact
of the1970 Clean Air Act
on
Air
Qualityand
InfantMortalityThe
instrumentalvariablesestimateof
theeffectof
TSPs
isa fiinctionof
two
reduced-form
effects: the
impact of 1972
nonttainmentstatuson
improvements
inairquahty,and
its associationwithdeclinesininfant mortality. Here,
we
provide
estimatesof
theairqualityand
infanthealthbenefitsof
the1970 Clean
AirAct
and
presentevidence
thatour designmay
plausibly identifythe causaleffects ofnonattainment
statuson both outcomes.
Figure
4
depicts the1969-1974
time-seriesof
theraw
differencesinTSPs
levels (PanelA)
and
internalinfantmortalityrates(Panel
B) between
countiesthatwere
nonattainment
and
attainmentin 1972. Separate series are plottedforthreesamples: 1) allcountieswithnonmissing
TSPs
data, 2) thesubset of countieswith
a1970
geometricmean TSPs
concentrationbetween
50-1 15 )ig/m^,and
3) thesubsetof
counties
with
1970
TSPs
between 60-90
|ig/m\"^The
lasttwo samples
arerestricted tothose counties withTSPs
concentrationscloser tothe75
|ig/m^regulationceiling intheregulation selection year.^'
The
comparison ofcounties with"big"and"small" 1971-1972TSPs
reductionsinthe thirdcolumn
of Table2 impliesanunadjustedTSPs
effect of-0.73 (18.8/(-25.6)) for 1971-1972firstdifferences. Thissuggeststhatthesmall 1971-1972first-differencesestimatein
column
1 ofTable4 isnotthe resultof attenuationbias duetomeasurement
error. Seefootnotes 12and 26formore
details. ^*The
samplesconsistofthefixedsetofcountieswith continuousmonitorreadingsfrom
1969-1974.The numbers
ofcountiesare401, 262,and 141, respectively.
Panel
A
shows
large trendbreaks inthenonattainment-attainmentdifference inTSPs
thatcorrespond
with
thetimingof
thefirstyearofCAAA
enforcement
in 1972.There
isa clearconvergence
in
TSPs
from
1971 to1972 even
forthe subsetof
countieswith
1970
TSPs
levels closetotheregulatorythreshold.
Among
countieswith1970
TSPs
between 60-90
[ig/m?, the 10.5 unitnonattainment-attainment
gap
inTSPs
in 1971 isalmost
eliminatedby
1972
beforerebounding
somewhat
in 1973.For
countieswith
1970
TSPs
between
50-115 |J.g/m^,thereis also a significant relativereductioninTSPs
from
1971-1972, though
thebreak
intrendappearstohave
starteda yearearlierthanfor the othertwo
county
groups.On
the whole,thetimingand
locationof
theTSPs
reductionssuggestthatenforcement
ofthe 1
970
CAAA
ledtoimproved
airquality.Panel
B
ofthe figureshows
astrikingcorrespondence
between
thebreaksintrendininfantmortalitydifferences
between
nonattainmentand
attainment countiesand
thoseinTSPs
differences.For
thefullsample,thesharp
break
ininfantmortalitytrendsfrom
1971-1972 matches
thebreak
inTSPs
differences.The
differentialtimingof
the relativeshiftdown
ininfant mortalityacrossthe threesamples
also
matches
theTSPs
patterns.While
thefullsample and
thesample
with 1970
TSPs
between 60-90
)Lig/m^
show
relativemortality reductionsfrom
1971-1972
only, thesample with 1970
TSPs
between
50-1 50-15|ag/m^ exhibits arelativemortalityreductiontheprevious yearalso.
Also
consistentwith
the patterns inPanelA,
thesample with 1970
TSPs
between 60-90
jig/m^ exhibitsthe largestrebound
in relativemortalityrates
from 1972
to 1973.Below,
we show
thatsome
ofthe1972-1974 rebound
in infantmortalitydifferences is explained
by
differentialchanges
intheobservablecharacteristicsof
attainmentand
nonattaimnent
countiesafter 1972.Taken
together, theseplots provideevidenceofa direct linkbetween
regulation-inducedimprovements
inairquahty
and
reductionsin infant mortalityrates. First, thetiming of the abrupt declines inTSPs
and
infant mortality correlateremarkably
wellacrossallthreegroups
ofcounties. Further,none of
theobservablecovariates, including per-capitaincome
and
transferpayments,
exhibitssimilartrend
break
patternsinnonattainment-attainmentdifferences. Itseems
unlikelythatany
unobserved
factorshad
nonattainment-attainmentdifferencesthatchanged
as sharplyorabruptlyas thedifferencesin