DISCUSSION PAPER SERIES
Fo schungsins i u
zu Zukun de A bei
Ins i u e o he S udy
o Labo
The E ec o B eas eeding on Child en’s
Cogni i e and Noncogni i e De elopmen
IZA DP No. 6697
June 2012
C is ina Bo a
Ma ia Iaco ou
Almudena Se illa
The E ec o B eas eeding on Child en’s
Cogni i e and Noncogni i e De elopmen
C is ina Bo a
Uni e si y o Se ille
Ma ia Iaco ou
Uni e si y o Essex
Almudena Se illa
Uni e si y o Ox o d
and IZA
Discussion Pape No. 6697
June 2012
IZA
P.O. Box 7240
53072 Bonn
Ge many
Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: [email p o ec ed]
Any opinions exp essed he e a e hose o he au ho (s) and no hose o IZA. Resea ch published in
his se ies may include iews on policy, bu he ins i u e i sel akes no ins i u ional policy posi ions.
The Ins i u e o he S udy o Labo (IZA) in Bonn is a local and i ual in e na ional esea ch cen e
and a place o communica ion be ween science, poli ics and business. IZA is an independen nonp o i
o ganiza ion suppo ed by Deu sche Pos Founda ion. The cen e is associa ed wi h he Uni e si y o
Bonn and o e s a s imula ing esea ch en i onmen h ough i s in e na ional ne wo k, wo kshops and
con e ences, da a se ice, p ojec suppo , esea ch isi s and doc o al p og am. IZA engages in (i)
o iginal and in e na ionally compe i i e esea ch in all ields o labo economics, (ii) de elopmen o
policy concep s, and (iii) dissemina ion o esea ch esul s and concep s o he in e es ed public.
IZA Discussion Pape s o en ep esen p elimina y wo k and a e ci cula ed o encou age discussion.
Ci a ion o such a pape should accoun o i s p o isional cha ac e . A e ised e sion may be
a ailable di ec ly om he au ho .
IZA Discussion Pape No. 6697
June 2012
ABSTRACT
The E ec o B eas eeding on Child en’s Cogni i e and
Noncogni i e De elopmen
This pape uses p opensi y sco e ma ching me hods o in es iga e he ela ionship be ween
b eas eeding and child en’s cogni i e and noncogni i e de elopmen . We ind ha
b eas eeding o ou weeks is posi i ely and s a is ically signi ican ly associa ed wi h highe
cogni i e es sco es, by a ound one en h o a s anda d de ia ion. The associa ion be ween
b eas eeding and noncogni i e de elopmen is weake , and is es ic ed o child en o less
educa ed mo he s. We conclude ha in e en ions which inc ease b eas eeding a es would
imp o e no only child en’s heal h, bu also hei cogni i e skills, and possibly also hei
noncogni i e de elopmen .
JEL Classi ica ion: I10, J0
Keywo ds: b eas eeding
Co esponding au ho :
Almudena Se illa
School o Business and Managemen
Queen Ma y, Uni e si y o London
F ancis Banc o Building
Mile End Road
London E1 4NS
Uni ed Kingdom
E-mail: [email p o ec ed]
2
The E ec o B eas eeding on Child en’s Cogni i e and Noncogni i e De elopmen
∗
∗∗
∗
The mos aluable o all capi al is ha in es ed in human beings; and o ha capi al he mos
p ecious pa is he esul o he ca e and in luence o he mo he .
Al ed Ma shall (1890), Pa ag aph VI.IV.11.
1 In oduc ion
This pape examines he ela ionship be ween b eas eeding and child en’s la e cogni i e and
noncogni i e ou comes. This is a opic o conside able impo ance o policy in he UK: he Wo ld Heal h
O ganiza ion ecommends b eas eeding exclusi ely o six mon hs and alongside solid oods o wo yea s, bu
in he UK, ba ely one in h ee in an s is exclusi ely b eas ed du ing he i s ou mon hs o li e. Gi en he
inc easing ecogni ion o he impo ance o e y ea ly in e en ions in child en’s de elopmen and la e
ou comes; and gi en he huge social g adien in b eas eeding a es, wi h he mos p i ileged mo he s cu en ly
being many imes mo e likely o b eas eed han he leas p i ileged mo he s, b eas eeding may well be a
signi ican ou e o he in e gene a ional ansmission o human capi al. Recen esea ch shows a signi ican
impac o beha iou al and psycho-social ou comes on ea nings and educa ion (Duncan and Duni on 1998;
Heckman e al. 2006; Muelle and Plug 2006). Di e ences in child en’s cogni i e de elopmen eme ge a ea ly
ages (Illsey 2002; Feins ein 2003; Cunha e al. 2010), and he impo ance o imely pa en al in es men s (p e-
na al as well as pos -na al) is inc easingly ecognized as a majo ac o in os e ing child de elopmen
(Ca nei o and Heckman 2003; Del Bono e al. 2008). A ulle unde s anding o he ela ionship be ween
b eas eeding and a ious aspec s o child de elopmen is he e o e c ucial o an unde s anding o he
in e gene a ional ansmission o inequali y, and o policy-making aimed a educing inequali y.
The e is a well-es ablished associa ion be ween b eas eeding and a ange o posi i e heal h ou comes
in child en, such as a lowe incidence o as hma and middle ea and u ina y ac in ec ions (Dyson e al.,
2006). A smalle body o esea ch also shows b eas eeding o be ela ed o be e g oss mo o de elopmen
(Sacke e al., 2006), and imp o ed cogni i e abili y (Ande son e al., 1999). O he po en ial e ec s o
b eas eeding, such as cogni i e and noncogni i e ou comes o he ype in es iga ed he e, a e much less well
∗
This pape has bene i ed om commen s p o ided by pa icipan s a he B i ish Socie y o Popula ion S udies and a he
5
h
con e ence o Epidemiological Longi udinal S udies in Eu ope; om ou colleagues a ISER, pa icula ly Emilia del
Bono and Bi gi a Rabe. This wo k was unded by he ESRC unde g an RES-062-23-1693. We a e ex emely g a e ul o
all he amilies who ook pa in his s udy, he midwi es o hei help in ec ui ing hem, and he whole ALSPAC eam,
which includes in e iewe s, compu e and labo a o y echnicians, cle ical wo ke s, esea ch scien is s, olun ee s,
manage s, ecep ionis s and nu ses. The UK Medical Resea ch Council (G an Re : 74882); he Wellcome T us (G an
Re : 076467) and he Uni e si y o B is ol p o ide co e suppo o ALSPAC. This publica ion is he wo k o he au ho s
and hey will se e as gua an o s o he con en s o his pape .
3
esea ched. Despi e he g owing li e a u e in his a ea, scien is s' unde s anding o he mechanisms behind hese
obse ed ela ionships emains incomple e. One heo y is ha , se e al componen s o b eas milk (no ably
long-chain polyunsa u a ed a y acids,which accumula e in he b ain and e ina) may a ec cogni i e
de elopmen ia hei e ec s on neu al de elopmen . (Innis, 2004; Pe yk e al., 2007). In he case o he non-
cogni i e ou comes discussed in his pape , he e ec o senso y s imuli on he de elopmen o he ne ous
sys em may be c ucial. I has been sugges ed ha he skin- o-skin con ac be ween mo he and in an inhe en
in b eas eeding o ms an impo an s imulus o he child (B i on e al., 2006). B eas eeding also eleases
ho mones in he mo he , which a e belie ed o be ela ed o nu u ing beha iou (Pe yk e al., 2007). Recen ly,
esea ch in he eme ging ield o epigene ics shows ha b eas eeding induces gene ic pa hways ela ed o
esponses o s ess (Wea e e al., 2004) o in es inal heal h (Chapkin e al., 2010) which a e qui e di e en
om hose in non-b eas ed indi iduals.
One p oblem which a ises in conside ing he ela ionship be ween b eas eeding and la e ou comes is
he di icul y o iden i ying whe he he obse ed ela ionships a e causal, as opposed o a ising because
b eas eeding is mo e likely o be p ac iced by mo he s whose cha ac e is ics (highe social class, highe IQ,
highe le els o human capi al, highe mo i a ion e c) a ou mo e posi i e ou comes. Fo e hical and p ac ical
easons, he implemen a ion o andomized ials is usually no an op ion in his a ea o esea ch (al hough
expe imen s which andomize he p o ision o acili ies o p omo e b eas eeding a e easible, as discussed by
K ame e al. 2001 and K ame e al. 2008). I is only ela i ely ecen ly ha esea che s ha e made sys ema ic
a emp s o add ess hese issues o causali y. The mos widely p ac iced me hod in he li e a u e is o include a
as a ay o con ol a iables in he ou come eg essions (Ro hs ein 2011, Heikkila e al 2011, Quigley e al.
2011, Bel ield and Kelly 2010, Denny and Doyle 2010, Gibson-Da is and B ooks-Gunn 2006). Some o he
pape s s udy sibling pai s o con ol o unobse ed amily cha ac e is ics (Ro hs ein 2011, Bel ied and Kelly
2010, Rees and Sabia 2009, De e al 2006, E enhouse and Reilly 2005). Ano he app oach is o use an
ins umen al- a iables echnique. Ros hs ein (2011) uses S a e b eas eeding a es and laws abou b eas eeding
in public as ins umen s; Del Bono and Rabe (2011) use whe he he hospi al whe e he child was bo n
pa icipa ed in a b eas eeding p omo ion p og am; Denny and Doyle (2010) use whe he he bi h was ia c-
sec ion; and Bel ield and Kelly (2010) use caesa ean bi h, mo he ’s smoking and alcohol consump ion, and
coun y-le el a iables ega ding heal h ca e and social assis ance.
In his pape we ollow a ou h app oach, which in ol es con olling o selec ion on obse ables wi h
p opensi y sco e ma ching echniques. To ou knowledge only ou e y ecen pape s use his me hod in an
a emp o iden i y causal e ec s o b eas eeding on child en ou comes (Ro hs ein 2011, Jiang e al 2011,
McC o y and Lay e 2011, Bel ield and Kelly 2010). PSM in ol es “ winning” each b eas ed baby wi h one o
mo e babies who we e no b eas ed, bu who in all o he obse able espec s a e simila o he b eas ed baby.
In his way, we e ec i ely simula e an expe imen by c ea ing ma ched “ ea men ” and “con ol” samples,

4
composed espec i ely o women who do and who do no b eas eed, bu who a e iden ical in e e y o he
obse able espec (Rosenbaum and Rubin 1983a). I ma ching is pe ec , di e ences in mean ou comes
be ween bo h samples may be used as es ima es o he causal e ec o b eas eeding. One ad an age o PSM
o e eg ession analysis is ha PSM is non-pa ame ic. Ma ching nei he imposes unc ional o m es ic ions
such as linea i y on he ou come equa ions no assumes a homogeneous ea men e ec ac oss he popula ion.
Bo h assump ions a e usually unjus i ied ei he by economic heo y o by he da a (Zhao 2008).
The da a we use come om he A on Longi udinal S udy o Pa en s and Child en (ALSPAC).
ALSPAC con ains a pa icula ly ich se o a iables on in an eeding, om which we may es ablish he
du a ion o b eas eeding and o exclusi e b eas eeding; he s age a which o mula milk, animal milks and
solid oods we e in oduced; and he ypes o supplemen a y oods which we e in oduced. Howe e , he
ea u e o ALSPAC which makes his da a pa icula ly a ac i e o he implemen a ion o PSM is he ac ha
pa en s we e in e iewed se e al imes p io o he bi h o hei child en, and da a collec ed on hei a i udes o
b eas eeding and whe he hey hemsel es had been b eas ed as babies. The da a a e discussed in Sec ion 2.
The da a a e hus pa icula ly well sui ed o PSM, which es s on he assump ion ha he impo an ac o s
a ec ing an indi idual’s choice o b eas eed a e obse able, and ha he e is no signi ican selec ion on
unobse ables
1
. The e y ich da a we use mean ha we a e able o implemen ma ching o a e y high
s anda d, and hus, ha he isk o bias caused by unobse able he e ogenei y is low.
We ind ha b eas eeding is s ongly ela ed o child en’s cogni i e ou comes. This is s a is ically
signi ican , and s ands a a ound one wen ie h o a s anda d de ia ion a age i e, and on a e age a li le o e
one en h he ea e . I pe sis s o e ime, and is common ac oss all he cogni i e measu es we examine. PSM
es ima es end o be sligh ly lowe han OLS es ima es con olling o he same se o backg ound a iables.
This sugges s ha no aking in o accoun he selec ion e ec o b eas eeding may o e es ima e he ac ual
e u ns o b eas eeding, bu in his case only e y sligh ly. We ind li le e idence o a signi ican ela ionship
be ween b eas eeding and noncogni i e ou comes. I any hing, b eas eeding appea s o be associa ed wi h an
inc ease in beha iou al p oblems o e y small child en; o he ex en ha he e is a posi i e ela ionship, i is
ound la e in child en’s li es, and is ound p edominan ly in he child en o lowe -educa ed mo he s.
When conside ing b eas eeding du a ion wi h gene alized p opensi y sco e (GPS) and OLS me hods,
we ind impo an nonlinea i ies, in ela ion o bo h cogni i e and noncogni i e ou comes. Fo cogni i e
ou comes, we obse e a ma ked dose- esponse ela ionship, wi h a s ong associa ion be ween b eas eeding
and imp o ed cogni i e ou comes o he i s ew mon hs o b eas eeding, becoming la e a longe
1
Thus, PSM di e s impo an ly om Ins umen al Va iable me hods, which ely on he assump ion ha unobse ables a e
impo an and di icul o con ol o ega dless o how p ecise he in o ma ion on he indi idual and he en i onmen is.
IV is o en di icul o implemen in he s udy o b eas eeding, due o a lack o sui able ins umen s. Ne e heless,
selec ion on obse ables is a s ong assump ion and we discuss i s plausibili y a leng h below.
5
du a ions. Fo noncogni i e indica o s, by con as , we ind ha he i s ew mon hs o b eas eeding a e no
s ongly associa ed wi h imp o ed ou comes, bu ha an associa ion eme ges a du a ions longe han wo o
h ee mon hs. O e all, we ind a e y obus posi i e ela ionship be ween b eas eeding and cogni i e
ou comes, which is consis en o di e en de ini ions o b eas eeding; in he noncogni i e case, he
ela ionship is less obus , and obse able only o child en o lowe educa ed mo he s and o olde child en
b eas ed o a leas wo mon hs.
This pape makes a leas wo signi ican con ibu ions o a g owing li e a u e on b eas eeding. Fi s , in
con as o o he pape s s udying he ela ionships be ween b eas eeding and cogni i e and noncogni i e abili ies,
which a e usually based on a single b eas eeding a iable, we use he ich in o ma ion in he ALSPAC da a o
cons uc se e al measu es o b eas eeding: ini ia ion, du a ion o b eas eeding, exclusi e b eas eeding,
o mula- eeding a bi h, and he in oduc ion o supplemen a y oods. We a e hus able o in es iga e whe he
exclusi i y o du a ion o b eas eeding, o bo h, a e decisi e in child en’s de elopmen . To ou knowledge, none
o he p e ious s udies has analyzed he in oduc ion o unsui able oods o used he du a ion o b eas eeding as a
con inuous measu e, allowing o nonlinea i ies. Ou second con ibu ion is ha his is he i s s udy o analyse a
wide se o bo h cogni i e and noncogni i e ou comes o e he long e m (up o 14 yea s o age). We use a
longi udinal da a se whe e bo h cogni i e and noncogni i e ou comes a e obse ed epea edly. Only Del Bono
and Rabe (2011) and Bel ield and Kelly (2010) conside noncogni i e ou comes oge he wi h cogni i e
measu es, bu wi h a much smalle age ange, 3 o 7 yea olds, in he i s case, and 1 o 4, in he second. The only
o he s udy ollowing child en o a compa ably long pe iod o ime is ha o Oddy e al. (2010), who analyze 2 o
14 yea -olds, bu hey conside a mo e limi ed numbe o ou comes ela ed o men al heal h.
This pape is o ganized as ollows. The da a a e discussed in Sec ion 2. Baseline esul s and
he e ogenei y a e conside ed in Sec ion 3. Sec ion 4 examines possible de ia ions om he Condi ional
Independence Assump ion (CIA), Sec ion 5 explo es he sensi i i y o esul s o di e en b eas eeding
measu es, and Sec ion 6 concludes.
2 Da a and desc ip i e s a is ics
The A on Longi udinal Su ey o Pa en s and Child en (ALSPAC) is a longi udinal s udy o a ound
12,000 child en bo n in he A on a ea in he ea ly 1990s (Golding e al. 2001; G egg e al. 2005). Mo he s we e
ec ui ed in o he sample a he poin a which hey i s epo ed hei p egnancy o hei doc o s. Da a we e
collec ed a ou poin s du ing p egnancy and a se e al poin s ollowing bi h – om bo h pa en s, om he
child him/he sel , and om he child’s eache and school. Topic a eas co e ed include physical and men al
heal h, socioeconomic s a us, and child de elopmen ; school-le el da a and child en’s es esul s a e a ailable
ia me ged eco ds. E hical app o al o his s udy was ob ained om he ALSPAC Law and E hics Commi ee
and he Local Resea ch E hics Commi ees.
6
2.1 Sample
Fo ou analysis, we conside a sample o child en in he “co e sample” o ALSPAC. This sample
consis s o 14,541 p egnancies ha esul ed in 14,676 known oe uses o which 14,062 we e li e bi hs and
13,988 we e ali e a one yea .
2
The numbe o child en o which he mo he en olled in he ALSPAC s udy
and had ei he e u ned a leas one ques ionnai e o a ended a “Child en in Focus” clinic by 19/07/99, and
e u ned a leas one pos -bi h ques ionnai e is 12,268. We employ a maximizing s a egy wi h espec o
sample size which implies using as many obse a ions as possible o each ou come measu e po en ially
a ec ed by b eas eeding. Sample sizes hus a y depending on he ou come measu e ac ually used. Table A.1
in he Appendix shows he e ec o he di e en sample selec ions on sample size and he dis ibu ion o
selec ed a iables. O e all he di e en samples e ain he main sample ep esen a i eness. We acknowledge,
none heless, a sligh dec ease in he p opo ion o b eas ed child en, oge he wi h a dec ease in he p opo ion
o high educa ed mo he s, in he sample used in he En y Le el cogni i e assessmen , and a sligh inc ease in
he p opo ion o b eas ed child en, oge he wi h a mino inc ease in he p opo ion o high-educa ed mo he s,
in he sample used in he 42-mon hs noncogni i e e alua ion. In compa a i e e ms, he ep esen a i eness o
hese samples ha co espond o young child en ( i e- and h ee-and-a-hal -yea s-old, espec i ely) is no as
sa is ac o y as he accu acy o he samples ha conside olde child en.
2.2 Main a iables
The cogni i e ou come a iables used in his pape a e S anda d A ainmen Tes (SATs) sco es
3
a
ages 7, 11 and 14
4
, and he esul s o school en y es s a age 5. All sco es a e s anda dized o ha e mean ze o
and s anda d de ia ion 1; hus, all he esul s we p esen may be in e p e ed as p opo ions o a s anda d
de ia ion. In he UK, child en gene ally s a school in he Sep embe ollowing hei ou h bi hday, and mo e
up a school yea e e y Sep embe . Du ing he yea s when he ALSPAC coho we e in school, pupils’ p og ess
was moni o ed by compulso y SATs es s, adminis e ed a he end o each o he “Key S ages” in schools (See
Table 1). When he ALSPAC coho en e ed school he e we e no na ional compulso y en y assessmen s.
Howe e , he ou Local Educa ion Au ho i ies co e ing he o me A on a ea (B is ol, Sou h Glouces e shi e,
Ba h & No h Eas Some se and No h Some se ) all used he same en y assessmen scheme, which was used
by 80% o he local s a e schools.
2
Fo easons o con iden iali y da a on he 13 iple and quad uple child en we e no a ailable o analysis.
3
Sco es we e ma ched o ALSPAC da a om he Na ional Pupil Da abase (NPD), a cen al eposi o y o pupil le el
educa ional da a es ablished in 2002, which assigns e e y pupil in England a Unique Pupil Numbe (UPN).
4
SATs a e also aken a age 16. Howe e , we do no conside hese in his pape , since no all he child en in ou sample
ook he same na ional es a his poin .
7
Table 1: S uc u e o UK schools and es ing p ocedu es
Class Age Key S age Tes s
-- 3 - 4 --
Recep ion 4 – 5 Founda ion Local en y assessmen s
Yea 1 5 - 6 --
Yea 2 6 – 7 1 Na ional es s and asks in English and Ma hs
Yea 3 7 - 8 --
Yea 4 8 – 9 --
Yea 5 9 - 10 --
Yea 6 10 – 11 2 Na ional es s in English, Ma hs and Science
Yea 7 11 - 12 --
Yea 8 12 - 13 --
Yea 9 13 – 14 3 Na ional es s in English, Ma hs and Science
Yea 10 14 - 15 Some child en ake GCSEs
Yea 11 15 – 16 4 Mos child en ake GCSEs, GNVQs e c.
Sou ce: Depa men o Child en, Schools and Families (DCSF) a h p://www.d es.go .uk/
The en y-le el assessmen co e s eigh a eas, o which ou (language, eading, w i ing and
ma hema ics) ela e o cogni i e de elopmen and ou o non-cogni i e de elopmen
5
. The KS1 ba e y
consis s o asks in eading, eading comp ehension, w i ing, spelling and ma hema ics. We epo esul s o
eading, w i ing and ma hema ics a en y le el and KS1; a Key S ages 2 and 3, es s a e adminis e ed in
English, Ma hs and Science, and we epo esul s in all o hese a eas. Sco es a e eco ded on a scale o 2 o 7
(en y le el); 0 o 5 (KS1); and on mo e de ailed anges o ma ks a KS2 and KS3.
The i s se o non-cogni i e ou come a iables is de i ed om pa en al assessmen s a 42 mon hs. When
he child was abou h ee and a hal yea s old, pa en s we e asked o assess hei child on di e en dimensions
ollowing he Re ised Ru e Pa en Scale o P eschool Child en, which is an ex ension o he Ru e beha iou
scale (Elande and Ru e 1996). Pa en s we e gi en desc ip ions o child en, and we e asked o ick he box ha bes
desc ibed hei child. Fo example, o he s a emen “Tends o do hings on his own, a he soli a y” pa en s we e
gi en he choice o answe ing “yes ce ainly”, “yes some imes”, and “no”. The e we e a o al o 43 such ques ions;
esponses we e agg ega ed by he ALSPAC eam o c ea e sco es in i e di e en domains: emo ional di icul ies,
conduc di icul ies, hype ac i i y, beha iou di icul ies, and p osocial beha iou (Golding e al. 2003).
Ano he se o non-cogni i e ou come a iables come om he eache e sion o he S eng hs and
Di icul ies Ques ionnai e (SDQ) (Goodman, 1997), which he child en’s class eache s we e asked o comple e
when child en we e in Yea s 3 and 6 a school. Teache s we e asked o hink abou he child’s beha iou o e
he pas six mon hs, and o ick he esponse which mos closely co esponded o hei imp ession o he child.
Fo example, in answe o he ques ion “Is [ he child] conside a e o o he people’s eelings?” eache s could
answe “no ue”, “somewha ue”, “ce ainly ue”. Sco es on he i e SDQ domains (emo ional symp oms,
5
We do no include esul s om hese non-cogni i e es s in ou analysis, since assessmen in hese a eas was olun a y and only a
small ac ion o he sample (25%) comple ed hese es s.
14
(W i ing, en y le el) o 0.200 (Science, KS2). As we would expec , con olling o addi ional ac o s educes
he es ima es s ill u he – bu in ac , no by e y much. A case in poin is he coe icien on English a KS3,
which alls om 0.450 o 0.174 be ween Speci ica ions 1 and 2, bu only o 0.118 in Speci ica ion 3. This
sugges s ha a e y la ge amoun o he he e ogenei y be ween b eas eeding and non-b eas eeding amilies is
cap u ed by pa en al educa ion, wi h all he o he con ols oge he (including a i udes o b eas eeding and
in en ion o b eas eed) cap u ing only a small p opo ion a e educa ion is con olled o . The es ima es in he
hi d speci ica ion a e o he o de o 10 pe cen o a s anda d de ia ion; mos o he es ima es emain
s a is ically signi ican , al hough some es ima es in he ea ly yea s a e no .
Columns 4 and 5 p esen es ima es o he ATT and ATU ob ained ia PSM echniques when all con ol
a iables a e included. We use he Epanechniko ke nel algo i hm wi h 0.05 bandwid hs, imposing he
common suppo condi ion. Due o he high quali y da a and he easonable sample sizes, ma ching quali y is
e y high. Table A.3 in he Appendix displays he mean alues o he a iables used in he analysis o he
ea men and con ol g oups a e ma ching in each o he ma ching p ocedu es pe o med – one o each
es ing pe iod. O e all, he igu es in Table A.3 con i m ha ou ea men and compa ison g oups, hough
ini ially somewha di e en , look ex emely simila a e ma ching, wi h no signi ican di e ences in any o he
36 backg ound a iables in he PSM o cogni i e ou comes, and e y small di e ences o only wo o he
a iables in he PSM o noncogni i e ou comes.
The PSM es ima es o he ATT (column 4) a e essen ially he same as he OLS coe icien s wi h a la ge
se o con ol a iables. In gene al he signi icance le els a e a li le lowe , bu no o all ou come measu es.
Howe e , he es ima ed ela ionships a e s ill signi ican a all ages a e school en y le el, con i ming ou
p e ious esul ha con olling o a wide ange o ac o s, child en b eas ed o ou weeks o mo e do be e
han child en b eas ed o less han ou weeks by abou one en h o a s anda d de ia ion (sligh ly less a
younge ages, and sligh ly mo e a olde ages). PSM es ima es end o be sligh ly lowe han OLS es ima es
con olling o he same se o backg ound a iables. This sugges s ha no aking in o accoun he selec ion
e ec o b eas eeding may o e es ima e he ac ual e u ns o b eas eeding, bu in his case only e y sligh ly.
Howe e , he PSM esul s con ol o he mo he ’s selec ion in o b eas eeding on he basis o obse ables, i.e.,
o he ac ha mo e able mo he s (who a e mo e likely also o ha e mo e able child en) a e also he ones who
a e mo e likely o b eas eed. The e o e, he di e ence be ween he OLS and PSM esul s gi es an indica ion o
he size and di ec ion o he selec ion e ec .
Column 5 in Table 4 p esen s esul s o he ATU. Di e ences be ween ATT and ATU sugges ha he
ela ionships be ween b eas eeding and child ou comes may di e be ween hose babies who a e cu en ly
likely o be b eas ed, and hose babies who a e no cu en ly likely o be b eas ed. In ac , he ATU es ima es
a e sligh ly highe han ATT es ima es, sugges ing ha o he ex en ha b eas eeding does ha e a causal e ec

15
on cogni i e ou comes, ha e ec would be somewha highe o child en whose mo he s a e less likely o
b eas eed hem.
O e all, ou esul s om Table 4 p o ide e idence ha b eas eeding is esponsible a s a is ically
signi ican , inc ease in SATs sco es – a ound 10 pe cen o a s anda d de ia ion, which ansla es oughly in o
abou wo posi ions in a class o 30 child en. These esul s a e consis en wi h p e ious indings. In pa icula ,
he magni ude o he coe icien s coincides wi h he 9.1%-10.7% o a s anda d de ia ion epo ed by Denny and
Doyle (2010), who also analyze b eas eeding o a leas ou weeks in UK child en (Na ional Child
De elopmen S udy). These di e ences a e isible a all he ages we conside (5, 7, 11 and 14) and ac oss a
ange o subjec a eas. In ac , coe icien s seem o inc ease in magni ude wi h ime, in line wi h Cunha and
Heckman’s (2010) hypo hesis o ea ly in es men s in child en ha ing mul iplie e ec s in la e childhood. This
ac co esponds o he esul s epo ed by Del Bono and Rabe (2011), also o Uni ed Kingdom (Millennium
Coho S udy).
The e emains a ques ion howe e abou he eliabili y o SATS sco es. The p esence o a andom
componen o measu emen e o s eng hens a he han weakens ou conclusions, since his andom elemen
would se e o bias es ima ed coe icien s owa ds ze o. We ha e addi ionally agg ega ed he sco es om all
he es s aken in a pa icula key s age, since his p ocess is known o educe bias due o measu emen e o
and he posi i e esul s o a compa able magni ude pe sis .
8
The e is an addi ional p oblem o wha SATS
ac ually measu e. SATS sco es a e highly in luenced by social class, as child en om ad an ageous
backg ounds a e mo e likely o be p epa ed o he es s, o o go o schools which be e coach hem o hese
es s. We ha e con olled o a wide se o socio-economic household a iables du ing he ma ching p ocess,
which may ake ca e o he ac ha child en om mo e ad an ageous backg ounds may be o mally o
in o mally coached a home. Also, gi en ha in he ALSPAC da a he e a e lo s o child en a ending he same
schools, and he e a e no many schools, he second p oblem is likely o be less o a conce n. As a obus ness
check we included school ixed e ec s in he six ou come a iables measu ed a e school en y. We ob ained
he same esul s, so we a e con iden ha wha we a e cap u ing he e is no he e ec o be e schooling.
9
Addi ionally, one may a gue whe he he obse ed ela ionships could be due o he absence o a di ec
measu e o cogni ion, such as ma e nal IQ. We es ima ed he model on a es ic ed sample o mo he s o which
ma e nal IQ in o ma ion is a ailable (2817 obs.). We did no ind any signi ican esul s, ei he including o
excluding he ma e nal IQ a iable in his sample. We in e p e ha his small sample is highly selec ed among
mo i a ed, willing mo he s, whe e b eas eeding- o -4-weeks a es a e mo e han 67%, ins ead o 55 pe cen o
8
Resul s a ailable upon eques .
9
Resul s a ailable upon eques .
16
ou gene al sample. This ac oge he wi h sample size may p e en us om inding any signi ican
ela ionship be ween b eas eeding and cogni i e ou comes.
10
Table 5 p esen s es ima es o he ela ionship be ween b eas eeding and noncogni i e ou comes
(Ru e sco es and SDQ sco es) om OLS and PSM eg essions. Th ee OLS speci ica ions a e p esen ed. Each
con ols o whe he he child was b eas ed o no a ou weeks o age. The i s speci ica ion includes e y
ew addi ional con ols – only he child’s sex and his o he age a he ime o aking he es . The second
speci ica ion adds in a ange o a iables indica ing pa en al educa ion, while he hi d speci ica ion con ols o
he ull se o a iables desc ibed in Sec ion 2. Es ima es in he i s speci ica ion indica e ha b eas eeding is
associa ed wi h be e noncogni i e skills, al hough he ela ionship is much smalle and less p ecisely
es ima ed han in he case o cogni i e skills p esen ed in Table 4. B eas eeding accoun s o less han one
ou h o a s anda d de ia ion, wi h signi ican coe icien s anging in absolu e alue om 0.196 (hype ac i i y
in Yea 3) o 0.013 (conduc p oblems a 42 mon hs). As soon as pa en al educa ion is con olled o
(Speci ica ion 2) hese es ima es d op sha ply, o app oxima ely wo hi ds o hei o iginal size. Signi ican
coe icien s now ange in absolu e alue om 0.098 (conduc p oblems in Yea 3) o 0.005 (pee p oblems in
Yea 6). The s a is ical signi icance o he coe icien s is u he educed, and many s op being s a is ically
signi ican al oge he . As we would expec , con olling o addi ional ac o s educes he es ima es s ill u he
– al hough as in he case o cogni i e ou pu s in Table 4, he dec ease is no as la ge as when in oducing
con ols o pa en al educa ion. Signi icance le els d op qui e subs an ially, howe e , and only hype ac i i y
and pee p oblems in Yea 3 emain signi ican a he 5% le el, wi h coe icien s o -0.068 and -0.077,
espec i ely. In addi ion, wo measu es no shown o be in luenced by b eas eeding in p e ious esul s –
conduc p oblems and p osocial beha iou a 42 mon hs – now show a signi ican ela ionship, al hough,
su p isingly, wi h a coun e in ui i e sign: i appea s ha b eas eeding may be nega i ely in luencing he
beha iou o ela i ely young child en in hese domains.
10
Resul s a ailable upon eques .
17
Table 5: B eas eeding and Noncogni i e ou comes
(1)
(2)
(3)
(4)
(5)
OLS: Speci ic. 1
OLS: Speci ic. 2
OLS: Speci ic. 3
PSM: ATT
PSM: ATU
Pa en , 42 mon hs
Emo ional di icul ies -0.092 *** -0.062 ** -0.012 0.033 -0.017
(0.021) (0.022) (0.024) (0.033) (0.026)
Conduc p oblems 0.013 0.027 0.075 ** 0.099 *** 0.080 *
(0.021) (0.022) (0.024) (0.028) (0.031)
Hype ac i i y
-
0.095
***
-
0.069
**
-
0.043
-
0.054
*
-
0.006
(0.021) (0.023) (0.024) (0.025) (0.028)
P osocial
-
0.029
-
0.043
-
0.054
*
-
0.092
**
-
0.054
*
(0.021) (0.022) (0.024) (0.030) (0.026)
Teache , Yea 3
Emo ional Symp oms -0.089 ** -0.047 -0.029 -0.041 -0.043
(0.027) (0.029) (0.031) (0.037) (0.039)
Conduc P oblems -0.172 *** -0.098 *** -0.056 -0.021 -0.089 **
(0.025) (0.027) (0.028) (0.031) (0.034)
Hype ac i i y
-
0.196
***
-
0.093
***
-
0.068
*
-
0.017
-
0.115
**
(0.026)
(0.027)
(0.029)
(0.033)
(0.036)
Pee p oblems -0.094 *** -0.084 ** -0.077 * -0.064 -0.080 *
(0.027) (0.029) (0.031) (0.037) (0.033)
P osocial 0.086 *** 0.051 0.015 -0.050 0.038
(0.026) (0.028) (0.030) (0.033) (0.034)
Teache , Yea 6
Emo ional Symp oms -0.091 *** -0.034 -0.006 -0.011 -0.019
(0.025) (0.027) (0.029) (0.029) (0.039)
Conduc P oblems -0.129 *** -0.034 0.019 0.000 0.026
(0.023) (0.025) (0.026) (0.030) (0.040)
Hype ac i i y -0.154 *** -0.028 0.011 0.016 -0.015
(0.023) (0.025) (0.026) (0.034) (0.035)
Pee p oblems -0.034 0.005 0.012 0.025 0.023
(0.025) (0.027) (0.029) (0.036) (0.036)
P osocial 0.020 -0.030 -0.052 -0.061 -0.046
(0.024) (0.026) (0.028) (0.034) (0.036)
No es:
Columns 1, 2, and 3 show OLS eg ession coe icien s o he di e en es sco es on b eas eeding o a leas 4 weeks.
Speci ica ion 1 con ols o he child’s age and sex. Speci ica ion 2 also con ols o ma e nal educa ion. Speci ica ion 3
addi ionally includes he baby’s bi h weigh , c own-heel leng h, and head ci cum e ence; ges a ion and mode o deli e y
( aginal o caesa ean sec ion); whe he he child is win; he mo he ’s age a bi h; he mo he ’s ace and ma i al s a us; he
educa ion le els o bo h pa en s, housing enu e, he size o he home, neighbou hood cha ac e is ics, whe he he mo he
had been in ca e as a child, whe he she had di o ced pa en s, whe he he ca e died p ema u ely; he mo he ’s heal h,
whe he mo he had smoked du ing p egnancy, he mo he ’s men al heal h; he mo he ’s labou ma ke pa icipa ion; and
whe he he mo he and a he had been b eas ed hemsel es as babies and on hei a i udes owa ds b eas eeding,
measu ed p ena ally. S anda d e o s in pa en heses.
Columns 4 and 5 show ATT- and ATU- PSM es ima es o he e ec o b eas eeding o a leas 4 weeks including all he
abo e a iables as con ols. We use he Epanechniko ke nel algo i hm wi h 0.05 bandwid hs, imposing he common
suppo condi ion wi h he psma ch2 S a a command (Leu en and Sianesi 2003).PSM s anda d e o s in pa en heses
compu ed by boo s apping wi h 100 epe i ions.
Signi icance deno ed by as e isks: * = 5%, ** = 1%, *** = 0.1%
Sou ce: ALSPAC co e sample.
18
Columns 4 and 5 in Table 5 p esen PSM es ima es o non-cogni i e ou comes.
11
Mos ela ionships
become s a is ically insigni ican when con olling o selec ion on obse ables, so he i s imp ession is ha
b eas eeding does no ha e a clea ela ionship wi h non-cogni i e ou comes. This ag ees wi h he limi ed
exis ing e idence on he subjec o he Uni ed Kingdom. Heikkila e al. (2011), using logis ic eg ession o
es ima e he p obabili y o abno mal beha iou on 5-yea -old child en, ind no signi ican ela ionship be ween
b eas eeding and any o he compu ed SDQ sco es. They only ind some e idence o a posi i e in luence o
b eas eeding on some noncogni i e measu es o child en exclusi ely b eas ed o mo e han 4 mon hs.
K ame e al. (2008), using da a om a la ge scale, andomized expe imen in he Republic o Bela us, also
epo no consis en and signi ican di e ences in beha iou al s eng hs and di icul ies (SDQ) in 6-yea -old
child en whose mo he s we e encou aged o b eas eed exclusi ely and o a longe du a ion. In he Uni ed
S a es, howe e , Bel ield and Kelly (2010) ind some posi i e ela ionships in he case o mo o sco es and
ma e nal a achmen , i espec i e o he b eas eeding measu e used.
A ew signi ican coe icien s a e e iden . In column 4, which p esen s ATT es ima es, and column 5,
which p esen ATU es ima es, he su p ising nega i e coe icien s on b eas eeding o pa en al assesmen s o
conduc di icul ies and p osocial beha iou a 42 mon hs o age emain. Bo h coe icien s a e somewha below
10 pe cen o a s anda d de ia ion. So appa en ly, when selec ion on obse ables is con olled o , we ind ha
b eas eeding is nega i ely associa ed, in he sho un, wi h child en’s de elopmen in he domains o
an isocial beha iou and conduc di icul ies. Howe e , he ac ha he signi icance o his esul diminishes
when he ATU sample is used calls o addi ional analysis. A po en ial explana ion may be ound in he ac
ha measu emen e o is likely o be la ge in he case o pa en al assessmen s when a child is young. Johns on
e al. (2011) show ha expe psychia ic assessmen places g ea es weigh on eache s’ iews and a he less
on hose o pa en s, indica ing a la ge measu emen e o o pa en s’ assessmen s. Mo eo e , hey also show
ha obse e s, bo h pa en s and eache s, end o o e s a e p oblems o younge child en ela i e o hose o
olde child en and adolescen s, by he s anda ds o he ully-in o med expe psychia ic diagnosis. In he same
ein, K ame e al. (2008) no e ha , compa ed o eache s’ a ings, pa en s’ a ings end o be highe and mo e
ex eme o all SDQ scales.
Two o he signi ican esul s mus be commen ed on. The ATU es ima es in column 5 indica e ha
b eas eeding is associa ed wi h a lowe isk o conduc p oblems and hype ac i i y in Yea 3, bu ha hese
esul s do no pe sis o Yea 6. These esul s, oge he wi h he bene icial esul s ound on hype ac i i y and
pee p oblems o Yea 3 (bu no Yea 6) child en o he OLS me hod (column 3), may indica e ha
b eas eeding has a a ou able e ec o e he medium e m. O e all, none heless, we ind li le e idence o
signi ican e ec s o b eas eeding a leas ou weeks on noncogni i e ou comes.
11
Table A.4 in he Appendix analyses he quali y o ma ching o he p ocedu es used in his able.
19
3.1 He e ogenei y o esul s
By and la ge, esul s om Tables 4 and 5 show ha whe eas he ela ionship be ween b eas eeding and
imp o ed ou comes is pe sis en and s a is ically signi ican o all cogni i e ou comes, he esul s o non-
cogni i e ou comes a e much less conclusi e. I is possible, o cou se, ha a deg ee o bias may s ill be d i ing
ei he o bo h se s o esul s. Recen de elopmen s in ma ching me hods emphasize ha combining exac
ma ching on key co a ia es wi h p opensi y sco e ma ching is a design analogous o blocking in a andomized
expe imen and can lead o la ge educ ions in bias (Dehejia and Wahba 1999; S ua and Rubin 2007).
The e o e in his sec ion, we s a i y he sample wi h espec o especially ele an con ounding ac o s and e-
es ima e he model. In pa icula we in es iga e whe he ou es ima ed esul s a y ac oss ma e nal educa ion
le els.
12
Table 6 epo s he ATT- and ATU- PSM es ima es o he ela ionship be ween b eas eeding and
cogni i e and noncogni i e ou comes o each popula ion g oup. We expec ha s a i ying by ma e nal
educa ion will ha e a ai ly signi ican e ec on esul s, gi en he e ec men ioned ea lie o including his
a iable as a con ol in OLS es ima es. Columns 1 and 2 in each able p esen es ima ed ATT and ATU esul s
o highe educa ed mo he s (Deg ee and A-le el) while columns 3 and 4 o e he co esponding esul s o
lowe educa ed mo he s (O-le el, oca ional and CSE).
13
Wi h espec o cogni i e ou comes, mo e subs an ial
posi i e ela ionships eme ge o lowe -educa ed mo he s. This is in con as o p e ious esul s om Gibson-
Da is and B ooks-Gunn (2006) o Uni ed S a es who only ind posi i e b eas eeding e ec s o mo he s wi h
a leas some pos -seconda y educa ion. Ro hs ein (2011) ob ains la ge e ec s o child en o medium- o low-
educa ed mo he s o some ou comes, bu no o he s.
In espec o noncogni i e ou comes, ou esul s show, in e es ingly, ha al hough b eas eeding ends
o be associa ed wi h wo se sho - un noncogni i e ou comes o child en o bo h high- and low-educa ed
mo he s, he e a e se e al posi i e medium- o long- e m noncogni i e e ec s o child en o low-educa ed
mo he s. Thus, we ind a dec ease in p osocial beha iou a Yea 6 o child en o high-educa ed mo he s, bu
we ind a dec ease in conduc p oblems, hype ac i i y and pee p oblems o he child en o low-educa ed
mo he s a Yea 3. These esul s may be conside ed analogous o he bene i s ound by Bel ield and Kelly
(2010) o child en below he po e y h eshold and by McC o y and Lay e (2011) o socially disad an aged
g oups.
14
12
We also expe imen ed wi h dis inguishing p e- e m bi hs, as Heikkila e al. (2010), and e hnic o igin, as Ro hs ein
(2011), bu he s anda d e o s we e oo la ge, sugges ing ha hose subdi isions o he sample we e oo demanding o he
da a.
13
We ied main aining he i e o iginal educa ion le els bu sample sizes o mos ou come measu es p ecluded ob aining
any eliable esul s.
14
We also ound analogous esul s s a i ying by ma e nal ma i al s a us, wi h child en o non-ma ied mo he s ob aining
g ea e bene i s om b eas eeding. Resul s a ailable upon eques .

20
Table 6: He e ogenei y o esul s by ma e nal educa ion. Cogni i e and noncogni i e ou comes. PSM
es ima es
Deg ee, A-le el O-le .,Voca .,CSE
(1) (2) (3) (4)
ATT
ATU
ATT
ATU
Cogni i e Ou comes
En y le el
Reading 0.068 0.107 * 0.075 * 0.067 *
(0.065)
(0.049)
(0.033)
(0.030)
W i ing
0.011
0.108
*
0.017
-
0.024
(0.055)
(0.050)
(0.033)
(0.037)
Ma hs 0.001 0.060 0.032 0.020
(0.056) (0.040) (0.033) (0.034)
Key S age1
Reading
0.156
*
0.156
*
0.041
0.066
(0.076)
(0.072)
(0.040)
(0.045)
W i ing
0.054
0.107
*
0.062
*
0.069
*
(0.046) (0.043) (0.030) (0.032)
Ma hs 0.091 0.138 ** 0.090 ** 0.089 **
(0.048)
(0.047)
(0.034)
(0.031)
Key S age 2
English
0.000
0.055
0.080
*
0.122
***
(0.040)
(0.041)
(0.032)
(0.035)
Ma hs 0.044 0.101 * 0.096 ** 0.114 ***
(0.043) (0.043) (0.034) (0.032)
Science
0.025
0.083
*
0.142
***
0.154
***
(0.034)
(0.038)
(0.032)
(0.034)
Key S age3
English
0.006
0.081
0.125
***
0.100
***
(0.051) (0.050) (0.031) (0.029)
Ma hs 0.074 0.107 0.093 ** 0.112 **
(0.058)
(0.060)
(0.031)
(0.036)
Science
0.061
0.073
0.083
*
0.110
**
(0.058)
(0.051)
(0.035)
(0.038)
Noncogni i e Ou comes
42 mon hs
Emo ional di icul ies
0.062 -0.004 0.008 -0.020
(0.053) (0.046) (0.030) (0.035)
Conduc p oblems 0.136 *
0.076 0.054 0.075
(0.054)
(0.046)
(0.034)
(0.042)
Hype ac i i y -0.068 -0.082 -0.033 0.042
(0.057) (0.045) (0.035) (0.037)
P osocial -0.096 *
-0.057 -0.072 * -0.052
(0.046) (0.042) (0.031) (0.032)
Yea 3
Emo ional Di icul ies
-
0.077
-
0.030
-
0.010
-
0.044
(0.060) (0.056) (0.039) (0.052)
Conduc P oblems 0.066 0.017 -0.114 ** -0.130 *
(0.042)
(0.055)
(0.042)
(0.051)
Hype ac i i y 0.060 0.013 -0.085 -0.158 ***
(0.058) (0.052) (0.044) (0.044)
Pee p oblems
-
0.038
-
0.098
-
0.094
*
-
0.083
(0.070)
(0.059)
(0.041)
(0.052)
P osocial -0.087 -0.041 -0.015 0.045
(0.059)
(0.052)
(0.044)
(0.047)
21
Table 6: He e ogenei y o esul s by ma e nal educa ion. Cogni i e and noncogni i e ou comes. PSM
es ima es (Con .)
Deg ee, A-le el O-le .,Voca .,CSE
(1)
(2)
(3)
(4)
ATT ATU ATT ATU
Noncogni i e Ou comes
Yea 6
Emo ional
Symp oms
0.012
0.035
-0.017
0.002
(0.054) (0.047) (0.046) (0.046)
Conduc P oblems 0.001 0.032 0.013 0.046
(0.044)
(0.065)
(0.036)
(0.053)
Hype ac i i y
0.013
0.024
0.036
-
0.016
(0.056) (0.062) (0.039) (0.045)
Pee p oblems 0.079 0.006 -0.005 0.041
(0.059) (0.049) (0.037) (0.045)
P osocial
-
0.108
*
-
0.117
*
0.000
-
0.022
(0.054)
(0.055)
(0.041)
(0.045)
No es:
The able shows ATT- and ATU- PSM es ima es o he e ec o b eas eeding o a leas 4 weeks including all as con ols
he child’s age and sex, he baby’s bi h weigh , c own-heel leng h, and head ci cum e ence; ges a ion and mode o
deli e y ( aginal o caesa ean sec ion); whe he he child is win; he mo he ’s age a bi h; he mo he ’s ace and ma i al
s a us; he educa ion le els o bo h pa en s, housing enu e, he size o he home, neighbou hood cha ac e is ics, whe he
he mo he had been in ca e as a child, whe he she had di o ced pa en s, whe he he ca e died p ema u ely; he mo he ’s
heal h, whe he mo he had smoked du ing p egnancy, he mo he ’s men al heal h; he mo he ’s labou ma ke
pa icipa ion; and whe he he mo he and a he had been b eas ed hemsel es as babies and on hei a i udes owa ds
b eas eeding, measu ed p ena ally. We use he Epanechniko ke nel algo i hm wi h 0.05 bandwid hs, imposing he
common suppo condi ion, wi h he psma ch2 S a a command (Leu en and Sianesi 2003). Columns 1 and 2 show ATT
and ATU, espec i ely, compu ed in he sample o highe educa ed mo he s (Deg ee and A-le el). Columns 3 and 4 show
he co esponding esul o lowe educa ed mo he s (O-le el, Voca ional, and CSE).
S anda d e o s in pa en heses compu ed by boo s apping wi h 100 epe i ions.
Signi icance deno ed by as e isks: * = 5%, ** = 1%, *** = 0.1%
Sou ce: ALSPAC co e sample.
4 De ia ions om CIA
The p e ious s a i ied analysis e eals po en ially la ge e ec s o b eas eeding o disad an aged
child en and a clea di e gence be ween cogni i e s. non-cogni i e ou comes. Ou selec ion-on-obse ables
me hod equi es he condi ional independence assump ion (CIA) o gi e es ima es a causal in e p e a ion.
Selec ion on obse ables is mo e plausible i an ample se o con ol a iables is a ailable. We con end ha he
e y ich da a we use mean ha a high p opo ion o he p opensi y o b eas eed is explained by obse able
ac o s and he isk o bias caused by unobse able he e ogenei y is low. The CIA is no a es able assump ion
wi h non-expe imen al da a. Ne e heless, we p o ide some e idence on i s plausibili y using a double s a egy:
pe o ming a alsi ica ion exe cise (Heckman and Ho z 1989) and implemen ing a sensi i i y analysis simila o
ha p oposed by Ichino e al. (2008).
22
4.1 Falsi ica ion Exe cise
In his sec ion we e alua e whe he unobse ables co ela ed wi h b eas eeding incidence and child
ou comes could d i e he p e ious esul s, by pe o ming a alsi ica ion exe cise (Heckman and Ho z 1989).
The exe cise consis s o iden i ying a se o p e- ea men ou comes ha could be ela ed o hese
unobse ables. I ou comes a e measu ed be o e he ea men hey canno possibly be a ec ed by he ea men
unless he e emains some selec ion bias no p ope ly accoun ed o by he PSM me hod. The e o e, CIA is
mo e plausible i he esul s indica e ha he placebo ou come is no in luenced by he ea men a iable.
Imbens and Woold idge (2009) a gue ha i he a iables used in his ex a e closely ela ed o he ou come o
in e es , he es has mo e powe . The placebo ou comes used o his exe cise a e a child’s bi h weigh , head
ci cum e ence, and c own-heel leng h measu es. Gi en ha b eas eeding occu s pos e io o he bi h o he
child, he only way in which b eas eeding could be associa ed wi h any o hese ou comes would be he
p esence o selec ion e ec s, i.e., he p esence common ac o s ha a ec he placebo ou come and he
selec ion in o b eas eeding ha a e no accoun ed o in he PSM. One such ac o may be he mo he ’s IQ, o
example. I mo e able mo he s a e also he ones mo e likely o b eas eed, ailu e o accoun o his
unobse able will lead o o e -es ima ion o he ela ionship be ween b eas eeding and a child’s cogni i e
de elopmen . In he Imbens and Woold idge (2009) amewo k, bi h weigh is closely ela ed o cogni i e and
noncogni i e measu es and he e o e bi h weigh s ands as ou main p e- ea men p oxy ou come.
We also use ma e nal cha ac e is ics as dependen a iables in his placebo exe cise. Va iables like
alcohol and smoking in ake be o e and du ing p egnancy pa ly e lec ma e nal a i udes owa ds child en and
could well be ela ed o unobse ables in luencing child en’s ou comes. I may be illumina ing o conside
whe he he es o obse able cha ac e is ics we use as con ols al eady con ol o hem.
Table 9 p esen s es ima es o he ela ionship be ween b eas eeding and he placebo ou comes, o he
di e en me hods used. Fo mos o he placebo a iables conside ed, he es ima ed e ec s a e insigni ican ,
hus e idencing no selec ion on unobse ables. Howe e , we do ind a nega i e associa ion be ween
b eas eeding and he numbe o ciga e es smoked a he 32nd week o p egnancy, wi h mo he s who go on o
b eas eed o ou weeks smoking on a e age hal a ciga e e less pe day han hose who do no . This does no
in alida e ou me hodology – i may simply demons a e ha ciga e e smoking cap u es di e ences be ween
mo he s which o he a iables do no cap u e, and he e o e ha his a iable is a use ul componen o ou
ma ching models. Howe e , i may indica e a need o addi ional checks o he model.
23
Table 7: Falsi ica ion exe cise.
(1)
(2)
(3)
OLS PSM:ATT PSM: ATU
Bi h weigh -0.324 13.722 6.349
(7.255)
(8.849)
(11.547)
Head ci cum e ence -0.038 -0.037 -0.023
(0.023) (0.029) (0.036)
C own heel leng h
0.043
0.096
*
0.033
(0.035) (0.046) (0.044)
Smoked p e iously -0.002 -0.005 0.014
(0.008)
(0.009)
(0.011)
Ciga e es a 32 w -0.524 *** -0.478 *** -0.587 ***
(0.087) (0.093) (0.136)
P e ious alcohol cons.
0.030
0.020
0.031
(0.017) (0.019) (0.020)
Alcoholic d inks a 8 w -0.098 -0.092 -0.139
(0.088) (0.092) (0.114)
No es:
Column 1 shows OLS eg ession coe icien s o he di e en es sco es on b eas eeding o a leas 4 weeks. Columns 2
and 3 show ATT- and ATU- PSM es ima es o he e ec o b eas eeding o a leas 4 weeks using he Epanechniko
ke nel algo i hm wi h 0.05 bandwid hs, imposing he common suppo condi ion wi h he psma ch2 S a a command
(Leu en and Sianesi 2003). Fo each dependen a iable, he es o co a ia es in Table 3 a e included as con ols. These
a e he child’s age and sex, he baby’s bi h weigh , c own-heel leng h, and head ci cum e ence; ges a ion and mode o
deli e y ( aginal o caesa ean sec ion); whe he he child is win; he mo he ’s age a bi h; he mo he ’s ace and ma i al
s a us; he educa ion le els o bo h pa en s, housing enu e, he size o he home, neighbou hood cha ac e is ics, whe he
he mo he had been in ca e as a child, whe he she had di o ced pa en s, whe he he ca e died p ema u ely; he mo he ’s
heal h, whe he mo he had smoked du ing p egnancy, he mo he ’s men al heal h; he mo he ’s labou ma ke
pa icipa ion; and whe he he mo he and a he had been b eas ed hemsel es as babies and on hei a i udes owa ds
b eas eeding, measu ed p ena ally. S anda d e o s in pa en heses. PSM s anda d e o s compu ed by boo s apping wi h
100 epe i ions.
Signi icance deno ed by as e isks: * = 5%, ** = 1%, *** = 0.1%
Sou ce: ALSPAC co e sample.
4.2 Sensi i i y o Po en ial Con ounde s
Since i is no possible o es ima e he magni ude o selec ion bias wi h non-expe imen al da a, in his
sec ion we add ess he p oblem by conduc ing a sensi i i y analysis. The analysis builds on Rosenbaum and
Rubin (1983b) and assumes ha he CIA is no sa is ied gi en he a ailable obse ables, bu would be sa is ied
i one could obse e an addi ional bina y a iable U. Ichino e al. (2008) sugges simula ing his po en ial
con ounde in he da a and using i as an addi ional co a ia e in combina ion wi h he p e e ed ma ching
es ima o . By compa ing he es ima es ob ained wi h and wi hou ma ching on he simula ed con ounde , he
obus ness o he baseline esul s wi h espec o speci ic sou ces o ailu e o he CIA can be assessed. In
pa icula , he dis ibu ion o he po en ial con ounde U is ully cha ac e ized by he choice o ou pa ame e s,
30
b eas ed child en, bea ing in mind ha hese esul s may no be ully gene alizable o he sample o ne e -
b eas ed child en.
Figu e 1 shows he dose- esponse unc ions, ha is, he e ec o di e en du a ions o b eas eeding on
he agg ega ed cogni i e and noncogni i e ou comes conside ed, oge he wi h he co esponding 95%
con idence bounds.
20
Fo he sake o b e i y, we only show g aphs o he agg ega ed measu es. We ind
consis en esul s o each o he indi idual measu es. Panel A shows he dose- esponse unc ions o cogni i e
ou comes. The e ec is clea ly nonlinea , wi h b eas eeding du a ion inc easing es sco es a a dec easing a e.
The g aphs show a peak a b eas eeding du a ions lowe han hose sugges ed in he OLS eg essions: again,
esul s owa ds his end o he dis ibu ion need o be ea ed wi h cau ion, owing o ela i ely small numbe s o
mo he s b eas eeding a hese du a ions. This aside, we a e easonably con iden in claiming a posi i e and
nonlinea ela ionship be ween b eas eeding and cogni i e ou comes, which is much less p onounced a
du a ions highe han 8 mon hs.
Panel B p esen s he dose- esponse unc ions o he agg ega ed noncogni i e sco es. Following he
ALSPAC S udy Team (2008a) hese agg ega e measu es include only he emo ional di icul ies, conduc
p oblems, hype ac i i y, and pee p oblem domains, associa ed wi h wo se child beha iou , and hus exclude
he p osocial domain.
These esul s a e in e es ing in he ligh o ou p e ious es ima es, which sugges ed e y li le in he
way o ela ionships be ween b eas eeding and noncogni i e ou come measu es. He e, we see a possible
explana ion o his: he cu es a e ex emely la a he lowe end (indeed in he case o he pa en al measu es
a 42 mon hs, he cu e slopes up a low du a ions), and i is no un il wo o h ee mon hs’ b eas eeding
du a ion ha ou comes o b eas ed babies appea be e han ou comes o babies ed o he minimum
du a ion. This may be he eason ha we see so ew di e ences in ou comes when we compa ing babies
b eas ed o less han 4 mon hs wi h babies b eas ed o 4 mon hs o longe : i appea s ha he posi i e
ela ionship be ween b eas eeding and he noncogni i e ou comes we conside do no kick in un il longe
b eas eeding du a ions.
20
The GPS is es ima ed using a no mal dis ibu ion o he loga i hm o he leng h o b eas eeding, gi en he co a ia es.
The balancing p ope y is es ed using 11 s a a and ou ea men in e als. The po en ial ou come a each ea men le el
is es ima ed using a quad a ic app oxima ion o he ea men and he GPS. The dose- esponse unc ion is es ima ed a
T=1,2,…12 mon hs. Con idence bounds a 95% le el a e es ima ed using boo s apping. We use he S a a p og ams
gpsco e and dose esponse (Bia and Ma ei 2008) o pe o m he calcula ions.

31
Figu e 1: Dose- esponse Func ions o B eas eeding Du a ion
Panel A. Agg ega ed Cogni i e Ou comes
En y Le el Key S age 1
0 .1 .2 .3 .4
Agg ega e Cogni i e Measu e. En y Le el
0 4 8 12
Leng h o b eas eeding (mon hs)
0 .1 .2 .3 .4
Agg ega ed Cogni i e Measu e. Key S age 1
0 4 8 12
Leng h o b eas eeding (mon hs)
Key S age 2 Key S age 3
0 .1 .2 .3 .4
Agg ega ed Cogni i e Measu e. Key S age 2
0 4 8 12
Leng h o b eas eeding (mon hs)
0 .1 .2 .3 .4
Agg ega ed Cogni i e Ou come. Key S age 3
0 4 8 12
Lengh o b eas eeding (mon hs)
Panel B. Agg ega ed Noncogni i e Ou comes
42 mon hs Yea 3
-.15 -.1 -.05 0 .05
Agg ega ed Noncogni i e Measu e. 42 mon hs
0 4 8 12
Leng h o b eas eeding (mon hs)
-.8 -.6 -.4 -.2 0
Agg ega ed Noncogni i e Masu e. Yea 3
0 4 8 12
Leng h o b eas eeding (mon hs)
Yea 6
-.8 -.6 -.4 -.2 0
Agg ega ed Noncogni i e Measu e. Yea 6
0 4 8 12
Leng h o b eas eeding (mon hs)
No es:
The igu es show he dose- esponse unc ion o he co esponding ou come sco e and he 95% con idence bounds. All
con ols o Table 3 included in he es ima ion o he GPS. The GPS is es ima ed using a no mal dis ibu ion o he
loga i hm o he leng h o b eas eeding, gi en he co a ia es. The balancing p ope y is es ed using 11 s a a and ou
32
ea men in e als. The po en ial ou come a each ea men le el is es ima ed using a quad a ic app oxima ion o he
ea men and he GPS. The dose- esponse unc ion is es ima ed a T=1,2,…15 mon hs. Con idence bounds a 95% le el
a e es ima ed using boo s apping. We use he S a a p og ams gpsco e and dose esponse (Bia and Ma ei 2008) o pe o m
he calcula ions.
5.2 Feeding indica o s
In Table 10 we p o ide ATT esul s ob ained h ough p opensi y sco e ma ched samples o he
di e en b eas eeding indica o s. Ou i s indica o (column 1) is an ini ia ion measu e, whe he he child was
e e b eas ed. The incidence o e e b eas eeding is 75% in he sample. The p e alence o any b eas eeding
a 4 weeks (ou main indica o , on which all p e ious analyses we e based, column 2) is e y close o he
p e alence o exclusi e b eas eeding (column 3); hese s and a 55% and 42% espec i ely. Following Bel ield
and Kelly (2010) we also use whe he he child was o mula- ed a bi h (column 4). The a e is 30%, which
indica es mode a e o e lap be ween b eas - and o mula- eeding (abou 5% o all child en we e bo h b eas -
and o mula- ed om bi h). An addi ional indica o is a measu e o whe he he child was ed supplemen a y
oods gene ally conside ed o be unsui able when he was 6 mon hs old (column 5): his includes child en who
had e e been gi en co ee, ea, swee s chocola e, coca-cola, and o he izzy d inks a ha age. The a e o use
o unsui able oods in he sample is 27%.
Fo cogni i e ou comes, all he es ima ed e ec s a e consis en , o expec ed sign, and many a e
s a is ically signi ican , as p e iously ound. Resul s o “e e b eas ed” a e b oadly coinciden wi h p e ious
s udies, hough i should be men ioned ha mos p e ious s udies epo jus an agg ega e cogni i e measu e
(Bel ield and Kelly, 2010) o he esul s o p e-school es s (Quigley e al. 2011, Del Bono and Rabe 2011).
Compa ed o hose s udies which epo es ima es o he e ec o b eas eeding ini ia ion on es s compa able o
hose we use, ou esul s, anging om 0.07 o 0.14 o a s anda d de ia ion, a e close o hose o Ro hs ein
(2011) o child en 5 o 6 yea s old (0.06 o 0.10 o a s anda d de ia ion) han o he indings o Jiang e al.
(2011) o 7-yea olds (be ween 0.11 and 0.21 o a s anda d de ia ion), bo h o he Uni ed S a es.
Tu ning o he esul s on exclusi e b eas eeding, we do no ind ha exclusi e b eas eeding is
signi ican ly mo e bene icial han any b eas eeding: when we compa e es ima ed e ec s a ou weeks, he
coe icien on exclusi e b eas eeding is no always la ge han he coe icien on any b eas eeding, and
some imes i is ac ually smalle . This is in con as wi h p e ious indings by Del Bono and Rabe (2011) who
epo signi ican ly posi i e e ec s o exclusi e b eas eeding a 4 weeks, bu no signi ican e ec o any
b eas eeding a 4 weeks. No e, howe e , ha hey signal p oblems o weak iden i ica ion o his la e
measu e. Ou esul s a e mo e simila o hose o Quigley e al. (2011), who epo ha he associa ion be ween
exclusi e b eas eeding and cogni i e sco es is b oadly simila o ha o any b eas eeding.
The nega i e ela ionship be ween cogni i e ou comes and ha ing been o mula- ed a bi h coincide
wi h es ima es epo ed by Bel ield and Kelly (2010) on hei agg ega ed cogni i e measu e. As he measu e
33
e e s o child en exclusi ely bo le- ed om bi h, i s e ec s a e iden ically symme ical o exclusi e
b eas eeding a bi h. Since he e ec s o being o mula- ed a bi h a e no always la ge , in absolu e e ms,
han he posi i e e ec s o e e being b eas ed o pu o he b eas , his is u he e idence ha exclusi e
b eas eeding may no di e signi ican ly om any b eas eeding in his espec . The e o e, we may concu
wi h Quigley e al. (2011) ha bo h exclusi e b eas eeding and any b eas eeding a e equally posi i e o
child en’s cogni i e abili ies. The inding ha he in ake o unsui able oods is a he weakly ela ed o he
cogni i e ou comes conside ed, may also ein o ce he hypo hesis ha i is any b eas eeding a he han
exclusi e b eas eeding which ma e s in his case.
I is also e iden om hese esul s ha he ela ionship be ween b eas eeding and cogni i e ou comes
pe sis s o e he long e m. In ac , he ela i ely inc easing magni ude o he es ima ed e ec s is in line wi h
Cunha and Heckman’s (2007, 2008, 2010) hypo hesis o ea ly in es men in child en ha ing mul iplie e ec s
in la e childhood.
Resul s o noncogni i e ou comes a e also shown in Table 10. Again, no many o he es ima ed
coe icien s a e signi ican ly di e en om ze o. This is consis en wi h he esul s o K ame e al. (2008) om
a la ge-scale, andomized expe imen o he Republic o Bela us. Being o mula- ed a bi h shows signi ican
ha m ul e ec s o Yea 3 child en, bu hese disappea in he longe e m. The in ake o supplemen a y
unsui able oods is also ela ed o ce ain inc eases in beha iou al p oblems, bu only in he e y sho e m (42
mon hs).
O e all hese esul s sugges ha ega dless o he measu e used, b eas eeding shows a posi i e
ela ionship wi h child en’s cogni i e ou comes, and ha he magni ude o his ela ionship inc eases wi h age.
On he con a y, in an eeding seems o ha e no such a clea associa ion wi h child en’s la e beha iou . Ou
esul s o cogni i e ou comes ag ee wi h hose o Ro hs ein (2011) and Jiang e al. (2011), wi h a coe icien on
b eas eeding ini ia ion o less han one en h o a s anda d de ia ion, and wi h Denny and Doyle (2010), wi h a
coe icien on “any b eas eeding o a leas ou weeks” o abou one en h o a s anda d de ia ion. Unlike Del
Bono and Rabe (2011), we do no ind ha exclusi e b eas eeding di e s subs an ially om any b eas eeding.
34
Table 10: O he b eas eeding measu es.Cogni i e and noncogni i e ou comes. PSM es ima es.
Cogni i e Ou comes
(1)
(2)
(3)
(4)
(5)
E e
Any b eas .
Exclusi e
b eas .
Fo mula
-
ed
Unsui able
oods
b eas ed
4
weeks
4 weeks
a bi h
6 mon hs
En y le el
Reading
0.069
0.103
***
0.115
***
-
0.066
*
-
0.037
(0.036)
(0.027)
(0.026)
(0.031)
(0.025)
W i ing
0.031
0.064
*
0.050
-
0.059
-
0.016
(0.037) (0.028) (0.027) (0.031) (0.025)
Ma hs
0.102
**
0.043
0.028
-
0.070
*
-
0.016
(0.035)
(0.027)
(0.026)
(0.030)
(0.024)
Key S age1
Reading
0.082
*
0.076
*
0.043
-
0.092
*
-
0.034
(0.041) (0.033) (0.032) (0.036) (0.029)
W i ing
0.071
*
0.076
**
0.052
*
-
0.074
**
-
0.082
***
(0.031) (0.024) (0.023) (0.027) (0.022)
Ma hs 0.117 *** 0.119 *** 0.071 ** -0.120 *** 0.003
(0.032) (0.025) (0.023) (0.027) (0.022)
Key S age 2
English 0.084 ** 0.090 *** 0.078 *** -0.100 *** -0.082 ***
(0.031) (0.023) (0.022) (0.026) (0.021)
Ma hs 0.078 * 0.126 *** 0.089 *** -0.096 *** -0.028
(0.032) (0.024) (0.022) (0.026) (0.022)
Science 0.128 *** 0.117 *** 0.088 *** -0.111 *** -0.005
(0.030) (0.023) (0.021) (0.025) (0.021)
Key S age3
English 0.112 *** 0.146 *** 0.101 *** -0.105 *** -0.114 ***
(0.033) (0.025) (0.024) (0.028) (0.023)
Ma hs 0.066 0.110 *** 0.055 * -0.080 ** -0.037
(0.035) (0.027) (0.025) (0.030) (0.024)
Science 0.141 *** 0.127 *** 0.084 ** -0.121 *** 0.004
(0.036) (0.028) (0.026) (0.030) (0.025)
Noncogni i e Ou comes
42 mon hs
Emo ional di icul ies -0.004 0.007 -0.012 0.039 -0.002
(0.037)
(0.027)
(0.025)
(0.031)
(0.025)
Conduc p oblems 0.037 0.075 ** 0.026 -0.017 0.067 **
(0.035)
(0.025)
(0.023)
(0.029)
(0.023)
Hype ac i i y 0.091 * -0.043 -0.055 * -0.058 -0.009
(0.037)
(0.027)
(0.025)
(0.031)
(0.025)
P osocial -0.046 -0.089 *** -0.054 * 0.024 0.044
(0.037) (0.026) (0.025) (0.031) (0.025)
Yea 3
Emo ional Symp oms
-
0.044
0.008
0.001
0.061
-
0.006
(0.052) (0.037) (0.035) (0.043) (0.034)
Conduc P oblems
-
0.005
-
0.032
-
0.087
**
0.078
*
0.037
(0.047) (0.032) (0.030) (0.037) (0.030)
Hype ac i i y
-
0.068
-
0.047
-
0.062
0.089
*
-
0.012
(0.051) (0.034) (0.033) (0.041) (0.033)
Pee p oblems 0.053 -0.052 -0.046 0.095 * -0.032
(0.052)
(0.037)
(0.034)
(0.042)
(0.033)
P osocial 0.029 -0.005 0.064 -0.130 ** 0.056
(0.051)
(0.035)
(0.034)
(0.043)
(0.034)
35
Table 10: O he b eas eeding measu es.Cogni i e and noncogni i e ou comes. PSM es ima es. (con )
(1) (2) (3) (4) (5)
E e
Any b eas .
Exclusi e
b eas .
Fo mula
-
ed
Unsui able oods
b eas ed 4 weeks 4 weeks a bi h 6 mon hs
Noncogni i e Ou comes
Yea 6
Emo ional Symp oms -0.021
0.007
-0.015
0.033
-0.01
(0.046) (0.034) (0.032) (0.038) (0.031)
Conduc P oblems -0.017 0.032 0.005 0.034 0.011
(0.041) (0.030) (0.029) (0.035) (0.028)
Hype ac i i y 0.017 0.034 -0.023 0.067 0.03
(0.044)
(0.031)
(0.030)
(0.037)
(0.030)
Pee p oblems 0.024 0.035 -0.020 0.041 0.01
(0.046)
(0.034)
(0.032)
(0.039)
(0.031)
P osocial
-
0.005
-
0.046
0.025
-
0.005
-
0.02
(0.047)
(0.033)
(0.032)
(0.040)
(0.032)
No es:
The able shows ATT es ima es o he e ec o (1) e e b eas ed, (2) b eas eeding o a leas 4 weeks, (3) exclusi e
b eas eeding o a leas 4 weeks, (4) o mula- ed a bi h, and (5) ed unsui able oods a 6 mon hs. The speci ica ions
include as con ols he child’s age and sex, he baby’s bi h weigh , c own-heel leng h, and head ci cum e ence; ges a ion
and mode o deli e y ( aginal o caesa ean sec ion); whe he he child is win; he mo he ’s age a bi h; he mo he ’s ace
and ma i al s a us; he educa ion le els o bo h pa en s, housing enu e, he size o he home, neighbou hood
cha ac e is ics, whe he he mo he had been in ca e as a child, whe he she had di o ced pa en s, whe he he ca e died
p ema u ely; he mo he ’s heal h, whe he mo he had smoked du ing p egnancy, he mo he ’s men al heal h; he mo he ’s
labou ma ke pa icipa ion; and whe he he mo he and a he had been b eas ed hemsel es as babies and on hei
a i udes owa ds b eas eeding, measu ed p ena ally. We use he Epanechniko ke nel algo i hm wi h 0.05 bandwid hs,
imposing he common suppo condi ion, wi h he psma ch2 S a a command (Becke and Ichino 2002). Columns 1 and 2
show ATT and ATU, espec i ely, compu ed in he sample o highe educa ed mo he s (Deg ee and A-le el). Columns 3
and 4 show he co esponding esul o lowe educa ed mo he s (O-le el, Voca ional, and CSE).
S anda d e o s in pa en heses compu ed by boo s apping wi h 100 epe i ions.
Signi icance deno ed by as e isks: * = 5%, ** = 1%, *** = 0.1%
Sou ce: ALSPAC co e sample.
6 Conclusion
This pape implemen s a igo ous analysis o he ela ionship be ween b eas eeding and child en’s
la e cogni i e and non-cogni i e ou comes, a emp ing o disen angle he e ec s o b eas eeding om he
e ec s o mo he ’s cha ac e is ics and o he obse able ac o s. Using ALSPAC, a e y ich longi udinal da a
se , we implemen PSM echniques and a ange o sensi i i y analyses. We ind s a is ically signi ican
ela ionships be ween b eas eeding and cogni i e skills a all ages be ween school en y and Key S age 3 (age
14); hese a e e y simila in magni ude ega dless o he selec ion-on-obse ables me hod implemen ed (OLS,
PSM-ATT and PSM-ATU). Child en b eas ed o ou weeks o mo e do be e han child en b eas ed o less
han ou weeks by abou one en h o a s anda d de ia ion ac oss all he cogni i e measu es we examine. In
ac , he coe icien becomes la ge o e ime, wi h la ge magni udes ound o olde child en. This is in

36
ag eemen wi h Cunha and Heckman’s (2008, 2010) hypo hesis o ea ly in es men s in child en ha ing
mul iplie e ec s in la e childhood.
On he con a y, we ind much less e idence, unde any o he selec ion-on-obse ables me hods used,
o a ela ionship be ween b eas eeding and noncogni i e ou comes - al hough when we s a i y he analysis by
ma e nal educa ion, we do ind ha b eas eeding is posi i ely ela ed wi h some o he beha iou al ou comes
we conside , o he g oup o child en bo n o less-educa ed mo he s.
When conside ing di e en de ini ions o b eas eeding and b eas eeding du a ion, we ind ha he
ela ionship be ween b eas eeding and bo h cogni i e and noncogni i e de elopmen is highly non-linea , wi h
diminishing e u ns o du a ion. Wi h espec o cogni i e skills we ind no signi ican di e ences be ween
exclusi e b eas eeding and any b eas eeding.
Al hough we iew ou esul s as s ong e idence ha mo he ’s socio-economic backg ound canno en i ely
explain he obse ed e ec o b eas eeding on child en’s cogni i e ou pu s, we acknowledge ha unobse ed
he e ogenei y may emain an issue - o example, one de e minan o child en’s cogni i e ou comes which we
a e no able o conside in ou analysis is he mo he ’s IQ. We no e, howe e , ha he ichness o he da a
allows us o con ol o obse ed he e ogenei y in a e y p ecise way. Fu he mo e, we ind ha a e y la ge
amoun o he he e ogenei y be ween b eas eeding and non-b eas eeding amilies is cap u ed by pa en al
educa ion, wi h all he o he con ols oge he cap u ing only a small p opo ion a e educa ion is con olled
o . We con end ha any emaining unobse ed he e ogenei y is likely o be low, as indica ed by he
alsi ica ion exe cise and he sensi i i y analysis pe o med o es he plausibili y o CIA.
The di e gence in he obse ed ela ionships be ween b eas eeding and cogni i e de elopmen on he one
hand, and non-cogni i e de elopmen on he o he , sugges s di e en channels ia which b eas eeding may
a ec di e en aspec s o child de elopmen . I may be ha he con en s o b eas milk a e a de e minan o
cogni i e abili y, whe eas con ac wi h he mo he may be mo e impo an o de eloping he non-cogni i e
skills measu ed he e. To he ex en ha b eas ed and bo le- ed babies ha e a simila deg ee o con ac wi h he
mo he , we may no expec b eas eeding o be closely ela ed o non-cogni i e ou comes. Ne e heless u he
esea ch is needed o disen angle hese issues. This would equi e la ge s udies collec ing de ailed in an
eeding da a, mode o ecei ing b eas milk, and, impo an ly, mo he ’s ime spen wi h he child, eeding and
no eeding he baby.
Ou esul s p o ide insigh s in o he sho e m and long e m ela ionships be ween b eas eeding and a
ange o child ou comes. We ha e no es ablished beyond doub ha hese ela ionships a e causal, bu we ha e
aken sys ema ic s eps o disen angle he e ec s o b eas eeding om he con ounding e ec s o o he ac o s.
To he ex en ha he es ima ed ela ionships a e causal, he ele ance o his esea ch o policy is clea . The
Wo ld Heal h O ganisa ion ecommends exclusi e b eas eeding du ing he i s six mon hs, and he
con inua ion o b eas eeding alongside solid oods o wo yea s (Wo ld Heal h O ganisa ion 2003). The UK
37
Depa men o Heal h has iden i ied b eas eeding p omo ion as a key s a egy in educing inequali ies in
heal h, and has unded se e al ini ia i es and p ojec s which aim o inc ease b eas eeding a es, pa icula ly
among women om disad an aged g oups (Depa men o Heal h 2003). Gi en he huge social g adien in
b eas eeding a es, wi h he mos p i ileged mo he s cu en ly being many imes mo e likely o b eas eed han
he leas p i ileged mo he s, ou esul s show ha b eas eeding may well be a signi ican ou e o he
in e gene a ional ansmission o human capi al, and o policy-making aimed a educing inequali y.
Re e ences
ALSPAC S udy Team, 2008a.Da a Collec ed om he Ques ionnai es ‘Child’s Beha iou and Abili ies’,
‘Abou You Class’, ‘Abou You School’. ALSPAC Documen a ion.
ALSPAC S udy Team, 2008b.Da a Collec ed om he Ques ionnai es ‘The De eloping Child’, ‘Ques ionnai e
o he Class Teache ’, ‘Ques ionnai e o he Head Teache ’. ALSPAC Documen a ion.
Ande son J.W., Johns one B. M., Remley D.T., 1999. B eas - eeding and cogni i e de elopmen : a me a-
analysis, Ame ican Jou nal o Clinical Nu i ion, Vol.70, Issue 4 pp.525-535.
Bake , M. and K. Milligan 2007.“Ma e nal employmen , b eas eeding, and heal h: E idence om ma e ni y
lea e manda es”, NBER Wo king Pape 13188. NBER, Camb idge: Mass.
Bel ield, C. R., Kelly, I.R.,2010.,The Bene i s o B eas eeding ac oss he Ea ly Yea s o Childhood. NBER
Wo king Pape 16496.
Bia, M., Ma ei, A., 2008. "A S a a package o he es ima ion o he dose– esponse unc ion h ough
adjus men o he gene alized p opensi y sco e," S a a Jou nal, 8(3), 354-373.
B i on, J.R., B i on, H.L., G onwald , V., 2006.B eas eeding, sensi i i y, and a achmen .Pedia ics 118,1436
– 43 .
Ca nei o, P. and J. Heckman 2003.Human capi al policy.IZA Discussion Pape n. 821. IZA: Bonn
Chapkin, R. S., Chen Zhao, C., I ano , I., Da idson, L. A. Jenni e S. e al. (2010) Nonin asi e s ool-based
de ec ion o in an gas oin es inal de elopmen using gene exp ession p o iles om ex olia ed
epi helial cells. Am J Physiol Gas oin es Li e Physiol 298: G582–G589.
C own S, C isp AH. 1979. Manual o he C own-C isp Expe imen al Index. London: Hodde & S ough on.
Cunha, F., J. Heckman and S. Schennach, 2010.Es ima ing he Technology o Cogni i e and Noncogni i e
Skill Fo ma ion. Econome ica 78(3), 883-931.
Dehejia, R. ,Wahba, S., 2002.P opensi y Sco e-Ma ching Me hods Fo Nonexpe imen al Causal S udies.
Re iew o Economics and S a is ics 84(1), 151-161.
Del Bono, E., E misch, J. F. And M. F ancesconi 2008.In a- amily esou ce alloca ions: A dynamic model o
child en’s bi h weigh . IZA Discussion Pape s 3704.
Del Bono, E.,Rabe, B., 2011. Es ima ing he Causal E ec s o B eas eeding on Ea ly Child Ou comes.Ins i u e
o Social and Economic Resea ch, mimeo.
Denny K., Doyle O., 2010. The causal e ec o b eas eeding on child en’s cogni i e de elopmen : A quasi-
expe imen al design, Uni e si y College Dublin WP10/05
Depa men o Heal h 2003.In an Feeding Ini ia i e. A epo e alua ing he b eas eeding p ac ice p ojec s
1999-2002. Depa men o Heal h, TSO: London
38
De , G., Ba y, G. D. and I. J. Dea y 2006.“E ec o b eas eeding on in elligence in child en: p ospec i e
s udy, sibling pai s analysis, and me a-analysis”, B i ish Medical Jou nal, ol. 333(7575. 945-xx.
Duncan, G. and R. Duni on, 1998.So skills and long- un labo ma ke success. Resea ch in Labo
Economics17, 123-150
Dyson, L., Ren ew, M., McFadden, A., e al. 2006.P omo ion o b eas eeding ini ia ion and du a ion.E idence
in o p ac ice b ie ing. Na ional Heal h Se ice, Na ional Ins i u e o Heal h and Clinical Excellence:
London
Elande J, Ru e M, 1996.Use and de elopmen o he Ru e pa en s’and eache s’ scales. In e na ional Jou nal
o Me hods in Psychia ic Resea ch 6, 63-78.
E enhouse E., Reilly S., 2005. Imp o ed es ima es o he bene i s o b eas eeding using sibling compa isons o
educe selec ion bias, Heal h Se ices Resea ch 40(6), 1781-1802
Feins ein, L. 2003. Inequali y in he ea ly cogni i e de elopmen o B i ish child en in he 1970 coho .
Economica 70, 73-98
Fe gusson D. M., Woodwa d L.J.,(1999)B eas eeding and la e psychosocial adjus men , Paedia ic and
Pe ina al Epidemiology13(2), 144-157.
Gibson-Da is, C. M., B ooks-Gunn, J., 2006. B eas eedin g and he Ve bal Abili y o Th ee-yea -olds in a
Mul i-ci y Sample.Pedia ics118, 1444-51.
Golding, J., Pemb ey, M., Jones, R. and he ALSPAC S udy Team, 2001.ALSPAC-The A on Longi udinal
S udy o Pa en s and Child en.Paedia ic and Pe ina al Epidemiology 15, 74-87.
Golding, J. Bicke s a e, I. He on, J., S owe, B. and Bishop, J. 2003. Da a Collec ed F om The Ques ionnai e
‘My Son/Daugh e ’s Heal h and Beha iou ’. ALSPAC documen a ion.
Goodman, R. 1997. The s eng hs and di icul ies ques ionnai e: A esea ch no e. Jou nal o Child Psychology
and Psychia y 38, 581-586.
G egg, P. Washb ook, E., P oppe , C., Bu gess, S. 2005. The e ec s o a mo he 's e u n o wo k decision on
child de elopmen in he UK
.
Economic Jou nal 115(501), F48-F80
Heal h P omo ion Agency o No he n I eland 2006.P omo ing b eas eeding o mo he s e u ning o wo k.
Heal h P omo ion Agency o No he n I eland: Bel as
Heckman, J.H., Ho z, V.J., 1989. Choosing among al e na i e nonexpe imen al me hods o es ima ing he
impac o social p og ams: The case o Manpowe aining.Jou nal o he Ame ican S a is ical
Associa ion 84, 862-874.
Heckman, J., Ichimu a, H. Todd, P., 1997. Ma ching as an econome ic e alua ion es ima o : e idence
om e alua ing a job aining p og am.Re iew o Economic S udies64 (4), 605-654.
Heckman, J., S ix ud, J. and U. Se gio, 2006. E ec s o cogni i e and noncogni i e abili ies on labo ma ke
ou comes and social beha iou . Jou nal o Labo Economics 24(3), 411-482.
Heikkilä, K. Sacke , A., Kelly, Y. Ren ew, M. J., Quigley, M. A., 2011.B eas eeding and child beha iou in
he Millennium Coho S udy. A chi es o Disease in Childhood 96, 635-642.
Hi ano, K., and G. W. Imbens. 2004. The p opensi y sco e wi h con inuous ea men s. In Applied B ayesian
Mo deling and Causal I n e ence om Incomple e-Da a Pe sp ec i es , ed. A. Gelman and X.-L. Meng,
73–84. Wes Sussex, England: Wiley In e Science.
Ichino, A., Mealli, F. and Nannicini, T. (2008).‘F om empo a y help jobs o pe manen employmen : wha can
we lea n om ma ching es ima o s and hei sensi i i y?’ Jou nal o Applied Econome ics, No. 23,pp.
305–327.
39
Imbens; G. W.and Woold idge, J. M.(2009)."Recen De elopmen s in he Econome ics o P og am
E alua ion," Jou nal o Economic Li e a u e, Ame ican Economic Associa ion, ol. 47(1), pages 5-86,
Ma ch.
Illsley, R. 2002. A ci y’s schools: F om inequali y o inpu o inequali y o ou come. Ox o d Re iew o
Educa ion 28, 427-445
Innis SM., 2004. Polyunsa u a ed a y acids in human milk: an essen ial ole in in an de elopmen . Ad anced
Expe imen s in Medical Biology 554:27-43.
Ip, S., Chung, M., Raman, G., e al., 2007. B eas eeding and ma e nal and in an heal h ou comes in
de eloped coun ies.Agency o Heal hca e Resea ch and Quali y Publica ion n. 07-E007.
Jiang, M., Fos e , E.M., Gibson-Da is, C.M., 2011. B eas eeding and he Child Cogni i e Ou comes: A
P opensi y Sco e Ma ching App oach. Ma e nal and Child Heal h Jou nal 15(8), 1296-307.
Johns on, D., P oppe , C., Pudney, S., Shields, M., 2011. Child men al heal h and educa ional a ainmen :
mul iple obse e s and he measu emen e o p oblem. Cen e o Mic oda a Me hods and P ac ice
wo king pape CWP27/11.
Jul eze . al., 2006. A en ion beha iou and hype ac i i y a age 4 and du a ion o b eas - eeding,
Ac aPaedia ica Vol.96, Issue 6, pp. 842–847
Kelly, Y. and R. Wa , 2005. B eas eeding ini ia ion and exclusi e du a ion a 6 mon hs by social class:
Resul s om he Millennium Coho S udy. Public Heal h Nu i ion 8, 417-421
K ame , M. S., Chalme s, B., Hodne , E.D. e al., 2001. P omo ion o B eas eeding In e en ion T ial
(PROBIT).A andomized ial in he epublic o Bela us. Jou nal o he Ame ican Medical Associa ion
285(4): 413-420
K ame , M.S., Fombonne, E., Igumno , S., e al., 2008. E ec s o p olonged and exclusi e b eas eeding on
child beha io and ma e nal adjus men : E idence om a la ge andomized ial. Pedia ics121(3),435-
440.
Leu en, E. and B. Sianesi., 2003. "PSMATCH2: S a a module o pe o m ull Mahalanobis and p opensi y
sco e ma ching, common suppo g aphing, and co a ia e imbalance es ing".
h p://ideas. epec.o g/c/boc/bocode/s432001.h ml
Ma shall, A., 1980. P inciples o Economics. Macmillan.
McC o y, C., Lay e, R., 2011. The e ec o b eas eeding on child en’s educa ional es sco es a nine yea s o
age: Resul s o an I ish coho s udy. Social Science and Medicine 72, 1515-1521.
Mo ensen e . al.,, 2002. The Associa ion Be ween Du a ion o B eas eeding and Adul In elligence, The
Jou nal o he Ame ican Medical Associa ion287, 2365-2371
Muelle , G. and E. Plug, 2006. Es ima ing he e ec o pe sonali y on male- emale ea nings. Indus ial and
Labo Rela ions Re iew 60(1), 3-22
Nannicini, T. (2007).‘Simula ion-based sensi i i y analysis o ma ching es ima o s’, S a a Jou nal, Vol. 7,pp.
334–350.
Duke, M.P., &Nowicki, S., J . (1973). Pe sonali y co ela es o he Nowicki-S ickland locus o con ol scale
o adul s. Psychological Repo s , 33 , 267-270.
Oddy, W. H., 2006. Fa y acid nu i ion, immune and men al heal h de elopmen om in ancy h ough
childhood.In J. D. Huang (ed.) F on ie s in Nu i ion Resea ch. No a Science Publishe s