School Tracking and Mental Health

This is a sel -a chi ed e sion o an o iginal a icle. This e sion may di e om he o iginal in pagina ion and ypog aphic de ails. Au ho (s): Ti le: Yea : Ve sion: Copy igh : Righ s: Righ s u l: Please ci e he o iginal e sion: CC BY-NC 4.0 h ps://c ea i ecommons.o g/licenses/by-nc/4.0/ School T acking and Men al Heal h © 2020 by The Uni e si y o Chicago Published e sion Böcke man, Pe i; Haapanen, Mika; Jepsen, Ch is ophe ; Roule , Alexand a Böcke man, P., Haapanen, M., Jepsen, C., & Roule , A. (2021). School T acking and Men al Heal h. Jou nal o Human Capi al, 15(2), 291-345. h ps://doi.o g/10.1086/712728 2021 School T acking and Men al Heal h Pe i Böcke man Uni e si y o Jy äskylä, Labou Ins i u e o Economic Resea ch, and IZA Ins i u e o Labo Economics Mika Haapanen Uni e si y o Jy äskylä Ch is ophe Jepsen Uni e si y College Dublin and IZA Ins i u e o Labo Economics Alexand a Roule INSEAD and Cen e o Economic Policy Resea ch To unde s and how he ype o educa ion a ec s long- e m men al heal h, we ex- amine he e ec s o a comp ehensi e school e o m on men al heal h– ela ed hospi aliza ions and dea hs. The e o m pos poned he acking o s uden s in o oca ional and academic schools om age 11 o age 16, hus a ec ing he se o pee s and he cu iculum o which hese s uden s we e exposed. The e o m was implemen ed g adually ac oss Finnish municipali ies be ween 1972 and 1977. We use di e ence-in-di e ences a ia ion and adminis a i e da a. Ou o e all e- sul s show no disce nible e ec s on men al heal h– ela ed hospi aliza ions o dea hs, bu he e ogenei y analysis shows an ad e se e ec on hospi aliza ions due o dep ession o emales om highly educa ed amilies. I. In oduc ion Educa ion leads o mone a y (Ang is and K uege 1991) and nonmone- a y (O eopoulos and Sal anes 2011) gains a he indi idual le el. The po- en ial posi i e e ec on heal h is a c ucial pa o he nonmone a y e u n We a e hank ul o he aluable commen s om Jay Bha acha ya and Isaac Eh lich ( he edi o s), he h ee anonymous e iewe s, A u i Bjö k, Hannu Ka hunen, Sa i Ke , K is ian Koe selman, Jani-Pe i Laamanen, K is a Riukula, Ke s in Schneide , Tuomo Suhonen, he An- nual Con e ence o he Royal Economic Socie y (2019), he Eu opean Associa ion o Labou Economis s (2018), he In e na ional Ins i u e o Public Finance (2018), he I ish Economic Associa ion (2018), semina pa icipan s a he Queen’s Uni e si y Bel as , he Economic and Social Resea ch Ins i u e, he Labou Ins i u e o Economic Resea ch, he Uni e si y o Elec onically Published Ma ch 12, 2021. [Jou nal o Human Capi al, 2021, ol. 15, no. 2] ©2020 by The Uni e si y o Chicago. This wo k is licensed unde a C ea i e Commons A ibu ion-NonComme cial 4.0 In e na ional License (CC BY-NC 4.0), which pe mi s non-comme cial euse o he wo k wi h a ibu ion. Fo comme cial use, con ac [email p o ec ed]. 1932-8575/2021/1502-0003$10.00 p o ided by educa ion. The posi i e co ela ion be ween educa ion and heal h is well es ablished (Cu le and Lle as-Muney 2008). Howe e , quasi- expe imen al e idence using na u al policy expe imen s on he causal link be ween educa ion and heal h ou comes s ill emains inconclusi e (Ga- lama, Lle as-Muney, and an Kippe sluis 2018). We ad ance he unde s anding o he educa ion-heal h ela ionship by s udying he e ec o a change a he age a which s uden s a e spli be- ween academic and oca ional educa ion on men al heal h in adul hood. 1 So a , he li e a u e has ocused only on physical heal h. The e o e, in his s udy, we examine he e ec s o educa ion on men al heal h. The lack o e idence on he e ec s on men al heal h ou comes is a salien gap, since men al heal h is an inc easingly impo an domain o heal h, especially in he de eloped coun ies (F ank and McGui e 2000; Laya d 2013). De- p essi e diso de s a e a leading and o en unde es ima ed cause o he global disease bu den (Vigo, Tho nic o , and A un 2016). Fo example, dep essi e diso de s accoun o 12% o o al yea s li ed wi h disabili y, and dep ession is he la ges con ibu o o he disease bu den a ibu able o non a al heal h ou comes (Üs ün and Cha e ji 2001; Whi e o d e al. 2013). Men al heal h p oblems also lead o subs an ial indi ec cos s, such as absen eeism and p oduc i i y losses a wo k (Bubonya, Cobb-Cla k, and Wooden 2017). Addi ionally, men al heal h p oblems inc ease he isk o poo physical heal h (Sa een e al. 2006). Mo eo e , mos o he li e a u e s udies only he e ec s o one addi ional yea o educa ion on heal h, whe eas o he ele an aspec s o educa ion, such as how long s uden s a e exposed o a common cu iculum be o e be- ing spli be ween academic and oca ional acks, migh also ha e an e ec on heal h, especially on men al heal h. Indeed, school acking undamen- ally a ec s he se o pee s o which s uden s a e exposed as well as he ype o skills hey acqui e and he deg ee o compe i ion hey ace in he class- oom. Pee e ec s and exposu e o compe i ion a e po en ial d i e s o men- al well-being. Unde s anding whe he and how changes in school acking a ec men al heal h p o ides insigh s in o he mechanisms h ough which educa ion ela es o heal h. Many Eu opean coun ies ha e implemen ed comp ehensi e schooling e o ms since he end o he Second Wo ld Wa o delay he age a which s uden s a e selec ed in o di e en acks (B unello, Giannini, and A iga 2007). 2 The p ima y mo i a ion behind such e o m pol- icies was he belie ha ea ly acking sys ems we e un ai o pupils om dis- ad an aged backg ounds ( Jones, Roeme , and Rosa Dias 2014). 1 In ou con ex o school sys ems in Eu ope, acking e e s o he s eaming o s uden s be ween he academic and oca ional educa ional acks, whe eas in he Uni ed S a es acking usually e e s o abili y g ouping wi hin schools (Hall 2012). 2 Ne e heless, age o acking di e s signi ican ly among he OECD (O ganisa ion o Economic Co-ope a ion and De elopmen ) coun ies (OECD 2004, 262). Tampe e, he Uni e si y o Jy äskylä, and he Uni e si y o Queensland. This s udy is pa o a p ojec suppo ed by he S a egic Resea ch Council a he Academy o Finland (p ojec nos. 293120 and 314208), he Al ed Ko delin Founda ion, he OP G oup Resea ch Founda- ion, and he Y jöJahnsson Founda ion (no. 6621). 292 Jou nal o Human Capi al To iden i y he e ec o school acking on men al heal h– ela ed hospi- aliza ions and dea hs, we use di e ence-in-di e ences a ia ion igge ed by he Finnish comp ehensi e school e o m, which was implemen ed g adually ac oss Finnish municipali ies du ing 1972–77. 3 The phase-in o he e o m o e s plausibly exogenous a ia ion in he acking age, and i s occu ence as a back as he 1970s allows us o iden i y long- un heal h e ec s. Key o ou iden i ica ion s a egy is he ac ha he g adual ollou was o hogonal o he incidence o men al heal h p oblems be o e he e- o m. Ou e idence con i ms ha he municipali ies ha we e ea ed i s we e no di e en in e ms o men al heal h om hose ha we e ea ed la e . Mo eo e , he e o m pos poned he acking o s uden s in o oca- ional and academic schools om age 11 o age 16 wi hou a ec ing he leng h o compulso y educa ion. Thus, he e o m p o ides a unique op- po uni y o s udy he e ec o inc easing he age o school acking, hold- ing ixed he numbe o yea s o compulso y educa ion. To iden i y he e ec s o his e o m on men al heal h, we use adminis a- i eda a o heFinnishpopula ionbo nin he1960s.Weha eaccess o comple e egis e s on suicides, men al heal h– ela ed dea hs and hospi al- iza ions, and all-cause mo ali y om he la e 1960s o 2013. The egis e s include all hospi al admissions ela ed o men al heal h diso de s in Fin- land. Using he g adual ollou o he comp ehensi e school e o m ac oss egions and o e ime, we es ima e di e ence-in-di e ences models o iden- i y he e ec s o he e o m on men al heal h ou comes by he age o 45. O e all, we do no ind signi ican e ec s on men al heal h– ela ed hos- pi aliza ions o dea hs. This a e age null esul is p ecisely es ima ed. We con ibu e o he deba e on whe he in es ing in people’s educa ion is an e ec i e way o imp o e hei heal h as well, which ul ima ely hinges on whe he he co ela ion be ween educa ion and heal h is causal ( Jones, Roeme , and Rosa Dias 2014). Ou wo k b ings u he e idence o p e ious esea ch inding negligible causal e ec s (Meghi , Palme, and Simeono a 2018). I ma e s because, as e iewed by Galama, Lle as-Muney, and an Kippe sluis (2018), he ype o educa ion ecei ed seems o a ec heal h be- ha io s mo e han he leng h o educa ion. Hence, ou s udy, in heo y, should be be e able o de ec e ec s on heal h han p e ious s udies ely- ing on compulso y-schooling e o ms ha iden i y he e ec o leng h o educa ion on heal h. Howe e , we s ill do no ind an economically signi - ican causal e ec . II. The Educa ion Re o m and I s Expec ed E ec s on Men al Heal h A. The S uc u e o he Finnish Educa ion Sys em be o e and a e he Re o m Finland had a selec i e wo- ack school sys em un il he 1970s. The e- o m eplaced he old wo- ack sys em wi h a uni o m comp ehensi e 3 P e ious s udies ha e used he e o m o s udy non–men al heal h ou comes (Pekka inen 2008; Pekka inen, Uusi alo, and Pekkala Ke 2009; Pekkala Ke , Pekka inen, and Uusi alo 2013). School T acking and Men al Heal h 293 school sys em (Some ki i 1982). Figu e 1 desc ibes he s uc u es o bo h sys ems (see Sahlbe g 2014). The e o m pos poned he acking age om 11 o 16 yea s. Bo h be- o e and a e he e o m, school s a s a age 7 and is compulso y un il age 16. In he old sys em, pupils we e augh oge he in he same class o only ou yea s, om age 7 o age 11. Then hey we e placed in o aca- demic o oca ional acks o he emaining i e yea s. In con as , in he new sys em, he e is an almos uni o m cu iculum o all nine yea s, un il age 16. 4 A i s co e, he e o m signi ican ly a ec ed he composi ion o he pee s o whom pupils we e exposed be ween ages 11 and 16. 5 The e we e no sys ema ic changes in class oom size, gende composi ion o classes, o eache quali y due o he e o m. The comp ehensi e school e o m was olled ou g adually ac oss Finn- ish municipali ies o e he pe iod 1972–77 ( ig. 2). The iming o he e- o m in he di e en municipali ies was decided by he Na ional Boa d o Educa ion (NBE). Municipali ies made sugges ions ega ding he im- ing o he e o m, bu i was he NBE ha inally app o ed and a i ied hem. Municipali ies we e in cha ge o he p ac ical implemen a ion o he e o m (in collabo a ion wi h he su ounding municipali ies). The e is spa ial co ela ion in he iming o he e o m, because he NBE wan ed Figu e 1.—S uc u e o he Finnish educa ion sys em be o e and a e he comp ehensi e school e o m in he 1970s. ISCED 5In e na ional S anda d Classi ica ion o Educa ion. 4 Abili y g oups in o eign languages and ma hema ics exis ed in he comp ehensi e schools (g ades 7–9) un il 1985 (Sahlbe g 2014, 28). 5 Schooling om age 7 o age 11 (g ades 1–4) emained unchanged a e he e o m as he eache s om he p e e o m sys em we e assigned o he comp ehensi e schools, and cu icula in bo h sys ems we e simila o g ades 1–4 (Some ki i 1982, 28). 294 Jou nal o Human Capi al o make su e ha wi hin la ge a eas s uden s would ge equal oppo uni- ies o mo e o seconda y educa ion (Some ki i 1982). Ne e heless, he e is meaning ul a ia ion in exposu e o he new comp ehensi e school sys- em, bo h ac oss bi h coho s and ac oss municipali ies (see able 1 be- low). 6 This a ia ion p o ides a quasi-expe imen al esea ch se ing. P e ious esea ch on he Finnish comp ehensi e school e o m has shown ha i dec eased he in e gene a ional co ela ion be ween ea n- ings o a he s and sons (Pekka inen, Uusi alo, and Pekkala Ke 2009). The e o m also sligh ly imp o ed he e bal and ma hema ical es sco es o boys belonging o low socioeconomic backg ounds (Pekkala Ke , Pekka inen, and Uusi alo 2013). In addi ion, he e o m inc eased Figu e 2.—Adop ion o he school e o m du ing he pe iod 1972–77. 6 Al hough he e o m mos ly p oceeded om no h o sou h, ig. B6 documen s ha in he adop ion o he e o m, subs an ial wi hin-municipali y a ia ion exis ed among he ou majo socioeconomic egions. School T acking and Men al Heal h 295 he gende di e ence in he p obabili y o choosing an academic ack and ob aining e ia y educa ion (Pekka inen 2008). Finally, Ra es eijn e al. (2017) s udy he e ec o he e o m on all-cause mo ali y. Fo males, using an 11% andom sample o he Finnish popula ion, hey ind occasionally nega i e o posi i e e ec s, depending on he subsample. Fo emales, hey do no ind any e ec on mo ali y. Using o al popula- ion da a, we ocus on he e ec s o he e o m on men al heal h. B. Expec ed E ec s on Men al Heal h A key ea u e o he Finnish e o m is ha i did no a ec he leng h o compulso y schooling. Thus, he e o m did no change he minimum school-lea ing age o 16. This allows us o ocus on examining he e ec s o change in he age o acking on men al heal h ou comes while keep- ing he numbe o yea s o compulso y educa ion ixed. School acking e o m may posi i ely a ec educa ion and income in adul hood. In u n, be e human capi al and a ailabili y o inancial e- sou ces may imp o e heal h (Galama, Lle as-Muney, and an Kippe sluis 2018). Fo his eason, we analyze he po en ial media ing ole o educa- ion and income on men al heal h in sec ion VI. Besides educa ion and income e ec s, o he po en ial mechanisms a e also a play. In e ac ions wi h pee s in school du ing childhood and ado- lescence a e impo an de e minan s o men al heal h ou comes in adul - hood (WHO 2014). In heo y, he p edic ed e ec s o he e o m on heal h ou comes, and especially on men al heal h, a e ambiguous. A e he e- o m, s uden s aged 11–16 now ha e a common se o pee s a he han pee s om hei speci ic ack only. Low-achie ing s uden s, who would ha e been assigned o he oca ional ack, a e being exposed o highe - achie ing pee s in he pos e o m sys em. This change is expec ed o im- p o e educa ion and ela ed economic ou comes o low-achie ing s u- den s, which in u n may imp o e hei heal h ou comes in he long un. I is also possible ha low-achie ing s uden s will s a adop ing and imi a ing heal h beha io s o hei highe -achie ing pee s. Con e sely, highe -achie ing s uden s may be ad e sely a ec ed by being exposed o lowe -achie ing pee s. 7 The comp ehensi e school e o m also a ec ed he abili y anking in he class oom. Voca ional ack s uden s who sco ed high in he abili y anking in he p e e o m sys em a e now, on a e age, anked lowe in comp ehensi e-school class ooms. Cicala, F ye , and Spenkuch (2018) ha e shown ha s uden s’academic achie emen s and dis up i e beha - io s depend on hei o dinal anking among hei pee s. Mo eo e , Elsne and Ispho ding (2018) p o ide e idence ha a s uden ’s o dinal abili y anking in a high school coho is an impo an de e minan o engaging 7 The changes in pee composi ion could also a ec indi iduals’pool o po en ial pa - ne s and hus ma iage ou comes, owing o asso a i e ma ching. We lea e his issue o u- u e s udies. 296 Jou nal o Human Capi al in isky beha io s (such as smoking, d inking, and p oneness o physical igh s). This e idence sugges s ha pupils who would ha e been acked in o oca ional schooling wi hou he e o m could ha e lowe heal h, es- pecially poo men al heal h, a e he e o m. I con adic s he posi i e e - ec o being exposed o highe -achie ing pee s. Mo eo e , he ac ha s uden s in he pos e o m sys em ollow a com- mon cu iculum om age 11 o age 16, a he han a ack-speci ic cu ic- ulum, also means ha , a e he e o m, s uden s in a gi en class oom a e less homogeneous. This makes i mo e di icul o eache s o ailo hei pedagogical app oaches o a mo e he e ogeneous mix o s uden s, he eby esul ing in nega i e educa ion and heal h ou comes (Be s 2011). S u- den s’achie emen may be be e when hey a e su ounded by pee s wi h simila cha ac e is ics. Indeed, he Finnish expe ience sugges s ha mixed- abili y g oups led o lea ning di icul ies; dis up i e beha io s inc eased a - e he e o m, and he numbe o pupils in special educa ion mo e han doubled be ween school yea s 1974–75 and 1979–80 (Some ki i 1982, 40). Finally, one o he main easons o a o he delay in acking is ha he likelihood o a s uden being placed in he “w ong” ack is educed and he anxie y associa ed wi h acking lessens, since he amoun o p e ack- ing in o ma ion abou s uden s’abili ies is highe a he ime o he ack- ing decision (B unello, Giannini, and A iga 2007). This sugges s ha he e o m imp o es men al heal h. Howe e , Pekka inen (2008) a gues ha o boys, his bene i is o se by he ac ha acking now occu s du ing pube y, in con as o gi ls, o whom i occu s a e pube y. In sho , p e ious esea ch does no clea ly p edic he e o m’s e ec on men al heal h, since di e en mechanisms a e pushing in di e en di- ec ions. Howe e , i does sugges ha he e ec likely di e s by gende and academic abili y. III. Da a To e alua e he long- un e ec o he e o m on men al heal h, we link h ee da a se s: (1) census da a co e ing he o al popula ion o Finland, (2) da a on he causes o dea h om he comp ehensi e dea h ce i ica es, and (3) comple e hospi al admissions da a. A. Census Da a We use he popula ion egis e da a o pe manen esiden s o Finland. The da a o igina e om he Longi udinal Popula ion Census Files om S a is ics Finland. Demog aphic and labo ma ke in o ma ion a e a ail- able o he yea s 1975 and 1985 and a e ha annually o e he pe iod 1987–2014. The municipali y o esidence is eco ded annually since 1971. The da a con ain almos comple e household and pa en al links. The da e and municipali y o bi h a e also eco ded. Fu he mo e, he da a also con ain de ailed in o ma ion abou deg ees comple ed om1970onwa d. School T acking and Men al Heal h 297 The co e da a include he indi iduals bo n inFinland be ween 1962 and 1966, ollowing Pekkala Ke , Pekka inen, and Uusi alo (2013, 586). We s a wi h bi h coho 1962 and end wi h bi h coho 1966 o inc ease he homogenei y o he coho s unde s udy. 8 This sample es ic ion also allows us o ollow all bi h coho s up o he maximum age o 45. 9 As shown in able 1, he e a e be ween 72,248 and 74,248 pe sons in each bi h coho . Hence, we ha e app oxima ely 366,000 indi iduals in o al. Annual in o ma ion abou he municipali y o esidence, oge he wi h he bi h da e, de e mines whe he a pupil a ended he acked o com- p ehensi e school sys em. The e o m was e ec i e o s uden s who we e a mos 11 yea s old, ha is, en e ing i h g ade, a he end o he yea in which he e o m was implemen ed in hei egion o esidence. Fo in- s ance, people who u ned 11 in he egion in which he e o m was imple- men ed in 1975 ecei ed pos e o m schooling i hey we e bo n in 1964, 1965, o 1966 and p e e o m schooling i hey we e bo n in 1962 o 1963. We exclude o eign-bo n indi iduals (mos o whom immig a ed o Finland a e he e o m) and hose li ing in he Åland Islands (in o al, 8% o he o iginal sample) om he es ima ion sample, e aining indi id- uals li ing in 465 di e en municipali ies. We also exclude a small num- be o indi iduals who, a ages 11–15, mig a ed be ween municipali ies wi h a di e en yea o adop ion o he e o m (less han 2.6% o he o ig- inal sample), 10 since he e o m indica o canno be assigned unambigu- ously o hese people. Finally, we exclude a e y small numbe o emi- g an s om he o iginal da a. Since measu es o academic abili y be o e acking a e no a ailable in he da a, we in es iga e he e ogenei y by pa en s’educa ion. 11 The a iable is ca ego ized in o h ee alues: (1) low educa ed (53%), i nei he o he pa en s comple ed pos compulso y educa ion, ha is, i hey bo h had a maximum o nine yea s o schooling; (2) mid-educa ed (27%), i ei he o bo h o he pa en s comple ed a oca ional deg ee bu nei he s udied u he ; and (3) highly educa ed (20%), i a leas one o hem comple ed high school o a highe le el o ( e ia y) educa ion (see ig. 1 o a e- minde o he p e e o m sys em). 8 In addi ion, he quali y o heal h da a be o e 1972 is weake (Sund 2012, 507). 9 We ollowed indi iduals un il age o 45 o hei dea h, whiche e came be o e, esul ing in unbalanced panel da a. Since mo ali y is one o he ou comes o in e es o us, we did no condi ion on being in he panel un il age 45. 10 In he o iginal sample, be ween ages 11 and 16, 5.3% o indi iduals mig a ed be ween municipali ies. Less han 2.6% mig a ed be ween he six wa es (yea s) o adop ion (see ig. 2). Only hese la e indi iduals (47.3% o he mig an s) we e excluded om he es ima ion sample. We ha e in es iga ed he po en ial ole o endogenous selec ion o municipali y o esidence by assigning he indi iduals o ea men on he basis o hei municipali y o bi h and da e o bi h, as in Meghi , Palme, and Simeono a (2018); see able B18 and dis- cussion in sec. V.C. 11 The co ela ion be ween pa en s’educa ion and child en’s academic abili y is s ong when he abili y is measu ed by high school comple ion (see able B1). Also no e ha , con- a y o he pa en al subsamples, he ull sample includes a small numbe o indi iduals whose pa en al in o ma ion is missing (less han 5%). Howe e , he esul s emain in ac i we exclude hese indi iduals om he sample. 298 Jou nal o Human Capi al Nex , we conside he e ec s o he e o m on men al heal h– ela ed dea hs, which include suicides, inju ies o unde e mined in en , and ac- ciden al dea hs. The esul s, epo ed in able B2, a e simila o hose ound o suicides. The only no able excep ion is ha o his b oade measu e o men al heal h– ela ed dea hs, imp o ed su i al in males om low-educa ed amilies can be seen. The es ima e indica es educ ion in mo ali y by 0.3 pe cen age poin s (signi ican a he 5% le el), ela i e o he mean ou come o 1.4%. Howe e , a e adjus men o mul iple- hypo hesis es ing, i s signi icance le el is ma ginally ou side he 10% Figu e 3.—E ec o he e o m on suicide a ages 16–45: leads and lags a ound he yea be- o e he e o m. Plo s a e based on baseline eg ession models, whe e he e o m dummy is eplaced wi h yea dummies o he leads and lags a ound he e o m yea . The plo ed poin s ep esen he es ima es on he lead and lag dummies. The omi ed ca ego y is he yea be o e he e o m (21). The es ima ed e ec s a e epo ed, oge he wi h hei 95% con idence in e als (based on obus s anda d e o s clus e ed a he municipal le el). School T acking and Men al Heal h 305 le el ( able B11). The esul s o males om low-educa ed amilies e- main quali a i ely in ac o he al e na i e speci ica ions o he model, bu he signi icance le els a y by model ( able B8). The es ima ed e ec s emain insigni ican in o he subsamples. To e alua e he o e all heal h e ec s o he e o m, able B3 epo s he indings on all-cause mo ali y. Echoing he esul s o suicides, we ind no e idence ha he e o m a ec ed all-cause mo ali y by age 45. The Figu e 4.—E ec o he e o m on men al heal h– ela ed hospi aliza ions a ages 16–45: leads and lags a ound he yea be o e he e o m. Plo s a e based on baseline eg ession models, whe e he e o m dummy is eplaced wi h yea dummies o he leads and lags a ound he e o m yea . The plo ed poin s ep esen he es ima es on he lead and lag dummies. The omi ed ca ego y is he yea be o e he e o m (21). The es ima ed e ec s a e epo ed, oge he wi h hei 95% con idence in e als (based on obus s anda d e - o s clus e ed a he municipal le el). 306 Jou nal o Human Capi al es ima ed e ec s a e close o ze o. 21 Fo males, we can ule ou inc eases o 0.17 pe cen age poin s o dec eases o 0.5 pe cen age poin s a he 5% sig- ni icance le el, ela i e o he mean ou come o 4.6%. Fo emales, we can ule ou inc eases o 0.4 pe cen age poin s o dec eases o 0.15 pe cen age poin s, ela i e o he mean ou come o 1.8%. Despi e he uni e sal heal h ca e sys em, he baseline di e ences in mo ali y by pa en al educa ion a e subs an ial—males om low-educa ed amilies ha e a h ee imes highe mo ali y han hose om highly educa ed amilies—bu he ea men e ec is insigni ican ac oss g oups. Robus ness o hese esul s is docu- men ed in able B9. The conclusions a e also no sensi i e o he use o a Cox p opo ional haza ds model ( able B16). These esul s p o ide oppo uni y o compa e he p ecision o ou es ima es o p io e idence. Meghi , Palme, and Simeono a (2018) epo he e ec s o he e o m on haza d a es o dea h by age 45 in hei appendix able A5 (sample o males and emales). Thei es ima e om a Cox eg ession is 1.0006 (SE o 0.0217). We epo a com- pa able es ima e in he no e o able B16, which indica es a combined e- o m e ec o 0.985 (SE o 0.030). Thus, Meghi , Palme, and Simeono a (2018) can ule ou (wi h 95% con idence) changes in he isk o dea h by age 45 ha a e ou side 24.2% and 14.3%, whe eas we can ule ou changes ha a e ou side 27.3% and 14.6%. In bo h s udies, s anda d e - o s a e clus e ed a he municipal le el. TABLE 3 E ec o he Re o m on Suicide by Age 45 Full Sample Pa en al Educa ion Low Mid High Males T ea men e ec 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) R 2 .0032 .0054 .0077 .0132 Mean ou come .0129 .0099 .0034 .0023 Obse a ions 186,777 94,037 47,224 36,282 Females T ea men e ec 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) R 2 .0025 .0044 .0095 .0148 Mean ou come .0030 .0021 .0007 .0009 Obse a ions 179,537 90,881 46,092 34,828 No e.—The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. None o he coe icien s a e signi ican a he 10% le el ( wo-sided es s). 21 Fo emales, he null e ec s a e consis en wi h hose o he ea lie Finnish s udy o Ra es eijn e al. (2017). Using an 11% andom sample o he popula ion, Ra es eijn e al. (2017) ind longe i y gains (losses) o males om low-income (high-income) amilies. School T acking and Men al Heal h 307 C. Hospi aliza ion Table 4 epo s he esul s ha use he incidence o men al heal h– ela ed hospi aliza ions as an ou come a iable o e he age ange 16–45. The a e - age o he ou come a iable is 10% and 6% o males and emales, espec- i ely. These igu es indica e ha hese inciden s a e no a e in ou da a. The e o m led o an inc ease in men al heal h– ela ed hospi aliza ions, bu o only some subg oups. The i s inding is ha males om low- educa ed amilies who we e exposed o la e acking ha e a 0.74 pe cen - age poin s highe p obabili y o ha ing se e e men al heal h diso de s ha esul in hospi aliza ion han hose who we e educa ed in he p e e o m sys em, ha is, engaged in ea ly acking. The size o he e ec ep esen s a 7% inc ease ela i e o a mean ou come o 10.8%. The e ec is obus o using an ou come measu e ha cap u es also he in ensi e ma gin and no jus he ex ensi e one, ha is, he numbe o days spen in he hospi al o men al heal h easons be ween ages 16 and 45 (see able B17). Impo an ly, he e ec o males om low-educa ed amilies is no longe s a is ically signi ican a e he s anda d e o s a e adjus ed o a Romano-Wol co - ec ion o mul iple-hypo hesis es ing (Romano and Wol 2005). The second inding is ha , because o he pos ponemen o he acking age om 11 o age 16, o emales om highly educa ed amilies he p ob- abili y o se e e men al heal h diso de s inc eased by 0.95 pe cen age poin s. We highligh his inding o se e al easons. Fi s , he quan i a i e magni ude o he e ec is a he la ge, gi en he low baseline p obabili y o men al heal h diso de s o women om highly educa ed amilies (5.2%). TABLE 4 E ec o he Re o m on Men al Heal h–Rela ed Hospi aliza ions a Ages 16–45 Full Sample Pa en al Educa ion Low Mid High Males T ea men e ec .0027 .0074** .0012 2.0058 (.0025) (.0036) (.0045) (.0043) R 2 .0054 .0092 .0136 .0128 Mean ou come .0997 .1080 .0847 .0735 Obse a ions 186,777 94,037 47,224 36,282 Females T ea men e ec .0029 .0022 2.0007 .0095** (.0019) (.0025) (.0034) (.0041) R 2 .0042 .0071 .0106 .0133 Mean ou come .0564 .0593 .0498 .0521 Obse a ions 179,537 90,881 46,092 34,828 No e.—The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y ixed e ec s. S anda d e o s epo ed in pa en hesis a e clus e ed a he municipal le el. Co- e icien s in i alics su i e a Romano-Wol co ec ion o mul iple-hypo hesis es ing (Ro- mano and Wol 2005) a he 10% signi icance le el (see able B11). ** Signi ican a he 5% le el (all wo-sided es s). 308 Jou nal o Human Capi al Second, he e ec emains signi ican (a 10% le el) e en a e he s anda d e o s a e adjus ed o Romano-Wol co ec ion o mul iple-hypo hesis es ing (Romano and Wol 2005; able B11). Thi d, he e ec emains in ac ega dless o whe he we con ol o he incidence o p io men al heal h diso de s and/o pa en s’men al heal h s a us du ing he p e e o m pe- iod ( able B10). Fou h, he inding is also obus o se e al o he sensi i i y checks, o example, accoun ing o he ull se o egion-speci ic linea ime ends ( able B10). Fi h, he es ima ed dynamic esponse models show ha he inding is no d i en by p e e o m ends ( ig. 4). The e o e, i is unlikely ha he esul is a ec ed by unobse ed egional cha ac e is ics ha a e po en ially co ela ed wi h he ollou o he e o m. Six h, he ind- ing emains in ac e en i indi iduals a e classi ied in o ea men on he basis o hei municipali y o bi h ( able B18) o municipali y a age 10 ( a- ble B19) ins ead o municipali y a age 11 (and no elimina ing indi iduals who mig a e be ween ages o 11 and 16). The his o ical iming o he e o m, oge he wi h he a ailabili y o lon- gi udinal da a, enables us o examine he e ec s o he e o m o e he li e cycle. We epo he esul s o ele an age ca ego ies in able 5. We ind ha a ages 36–45, o emales om high-educa ed amilies, men al heal h diso de s a e signi ican ly mo e likely o occu in hose who had been ex- posed o he pos e o m school sys em han in hose who we e educa ed in he p e e o m sys em. Howe e , o his a ec ed g oup, we do no ind sig- ni ican e ec s du ing he school yea s. Addi ionally, o males om low- educa ed amilies, he posi i e e ec o he comp ehensi e school e o m on he p obabili y o hospi aliza ion o men al heal h easons peaks a ages 26–35, bu he e is also a signi ican e ec a la e ages, ha is, be ween he ages o 36 and 45. Table 6 examines whe he any speci ic diso de d i es he es ima ed ea - men e ec s. We obse e a signi ican inc ease in he p obabili y o dep es- si e diso de s o emales om highly educa ed amilies. Con e sely, o males om low-educa ed amilies, he o e all inc ease in hospi aliza ions is d i en by he inc ease in alcohol- ela ed men al diso de s. Alcohol abuse may be caused by sel -medica ion ela ed o pe cei ed s ess (Enoch 2011). The gende pa e n ha we obse e is plausible, since in ou da a, alcohol- ela ed men al diso de s and dep essi e diso de s a e he mos p e alen men al heal h p oblems o males and emales, espec i ely ( able 2). VI. Po en ial Mechanisms The e o m could ha e a ec ed men al heal h h ough a ious channels discussed in sec ion II. On he posi i e side, in addi ion o exposu e o highe -achie ing pee s and access o a la ge se o cogni i e skills o s u- den s who would ha e s a ed oca ional aining a age 11 wi hou he e- o m, he key heo e ical a gumen in a o o pos poning he acking is ha i would ha e allowed he sys em o ha e mo e ele an and accu a e in o ma ion abou abili ies and compa a i e ad an ages when ma ching School T acking and Men al Heal h 309 s uden s o a pa icula educa ion (B unello, Giannini, and A iga 2007). Thus, inc easing he acking age should imp o e he e iciency o he ma ch and subsequen labo ma ke ou comes, which could ansla e in o be e men alheal h la e in adul hood. By con as , we do no ind any im- p o emen in men al heal h as a esul o he comp ehensi e school e o m. In ac , we ind ha he e o m did no lead o an imp o emen in edu- ca ion o labo ma ke ou comes ( ables B20, B21). Fo s uden s om low- educa ed amilies, we e en ind some nega i e e ec s o he e o m on ed- uca ional achie emen s and economic ou comes. These esul s sugges ha he heo e ical mechanisms ela ed o he e iciency o ma ching s u- den s o sui able educa ion a e no a play he e. To u he pin down he po en ial mechanisms, we ollow Acha ya, Blackwell, and Sen (2016) and es ima e he a e age con olled di ec e - ec s o he e o m, as desc ibed in appendix C. The nega i e e ec s on TABLE 5 E ec o he Re o m on Men al Heal h–Rela ed Hospi aliza ions o e he Li e Cycle Full Sample Pa en al Educa ion Low Mid High Males Ages 11–15 2.0002 .0011 2.0016 2.0016 (.0008) (.0009) (.0012) (.0013) Ages 16–25 .0024 .0041 2.0007 .0027 (.0021) (.0031) (.0040) (.0028) Ages 26–35 .0020 .0062*** .0003 2.0057 (.0016) (.0022) (.0031) (.0035) Ages 36–45 .0018 .0046* 2.0018 2.0032 (.0016) (.0023) (.0031) (.0032) Obse a ions 186,777 94,037 47,224 36,282 Females Ages 11–15 2.0002 2.0009 .0024** 2.0017 (.0006) (.0008) (.0010) (.0014) Ages 16–25 .0005 .0007 2.0012 .0009 (.0011) (.0015) (.0021) (.0023) Ages 26–35 .0020 .0011 .0034 .0026 (.0012) (.0016) (.0024) (.0029) Ages 36–45 .0012 .0018 2.0040 .0058** (.0014) (.0021) (.0025) (.0027) Obse a ions 179,537 90,881 46,092 34,828 No e.—The able epo s he ea men e ec o he e o m. Each ow co esponds o a di - e en ou come, and each column co esponds o a di e en sample. See able B1 o he mean alues o he ou come a iables by pa en al educa ion and igu e B5 o a g aphical illus a ion. The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. Indi iduals who died be o e he obse a ion pe iod we e emo ed om he es ima ion sample. All models include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. Coe icien s in i alics su i e a Romano-Wol co ec ion o mul iple-hypo hesis es ing (Ro- mano and Wol 2005) a he 10% signi icance le el (see able B11). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). 310 Jou nal o Human Capi al men al heal h o emales om highly educa ed amilies emain in ac e en when he educa ion and income media o s a e con olled o , as can be seen in able 7. Thus, o emales, he obse ed e ec s on men al heal h canno be explained by he educa ion/income channels. Ins ead, we conjec u e ha pee e ec s cons i u e a po en ial mecha- nism d i ing he obse ed ad e se e ec s o emales om highly educa ed TABLE 6 E ec o he Re o m on Specific Men al Heal h–Rela ed Hospi aliza ions a Ages 16–45 Pa en al Educa ion Full Sample Low Mid High Males Schizoph enia .0005 .0018 2.0013 .0014 (.0008) (.0013) (.0017) (.0018) O he nona ec i e psychosis 2.0005 .0013 2.0007 2.0028 (.0009) (.0015) (.0022) (.0020) Bipola diso de .0007 .0011 .0015 2.0012 (.0005) (.0008) (.0010) (.0013) Dep essi e diso de .0013 .0017 .0007 2.0033 (.0010) (.0015) (.0019) (.0021) Anxie y, s ess, neu o ic diso de .0010 .0026 2.0015 2.0013 (.0013) (.0018) (.0021) (.0018) Subs ance-use diso de .0038** .0048** .0015 .0021 (.0016) (.0023) (.0030) (.0025) Obse a ions 186,777 94,037 47,224 36,282 Females Schizoph enia .0004 .0013 .0001 2.0017 (.0008) (.0010) (.0013) (.0020) O he nona ec i e psychosis .0006 .0015 2.0016 2.0000 (.0010) (.0013) (.0020) (.0021) Bipola diso de .0005 2.0000 2.0007 .0026* (.0005) (.0007) (.0011) (.0014) Dep essi e diso de .0015 .0009 .0004 .0062*** (.0009) (.0015) (.0020) (.0022) Anxie y, s ess, neu o ic diso de .0001 .0003 .0006 2.0010 (.0008) (.0011) (.0012) (.0016) Subs ance-use diso de .0009 .0021 2.0013 .0024 (.0009) (.0014) (.0018) (.0015) Obse a ions 179,537 90,881 46,092 34,828 No e.—The able epo s he ea men e ec o he e o m on a ious ou comes. Each ow co esponds o a di e en ou come, and each column co esponds o a di e en sample. See able B1 o he mean alues o he ou come a iables by pa en al educa ion. The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al ed- uca ion) is missing. All models include coho and municipali y ixed e ec s. S anda d e - o s, epo ed in pa en heses, a e clus e ed a he municipal le el. Coe icien s in i alics su - i e a Romano-Wol co ec ion o mul iple-hypo hesis es ing (Romano and Wol 2005) a he 10% signi icance le el (see able B11). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). School T acking and Men al Heal h 311 TABLE 7 Es ima ed Re o m E ec , Con olling o Educa ion and Income Media o s La e in Li e Ou come Baseline Es ima es Media ion Analysis Educa ion Educa ion and Income A. Males, Low-Educa ed Pa en s Suicide by age 45 2.0016 2.0018 2.0017* (.0011) (.0012) (.0010) Men al heal h– ela ed dea h by age 45 2.0030** 2.0033** 2.0027** (.0014) (.0014) (.0013) Dea h by age 45 2.0034 2.0040* 2.0042** (.0022) (.0022) (.0021) MHD a ages 16–45 .0074** .0061* .0060 (.0036) (.0036) (.0037) MHD a ages 26–35 .0062*** .0056** .0049* (.0022) (.0022) (.0022) MHD a ages 36–45 .0046* .0038 .0032 (.0023) (.0024) (.0024) Subs ance-use diso de a ages 16–45 .0048** .0041* .0033 (.0023) (.0024) (.0023) B. Females, Highly Educa ed Pa en s Suicide by age 45 .0002 .0003 .0005 (.0007) (.0007) (.0007) Men al heal h– ela ed dea h by age 45 .0005 .0005 .0007 (.0008) (.0009) (.0008) Dea h by age 45 2.0008 2.0008 2.0008 (.0012) (.0013) (.0012) MHD a ages 16–45 .0095** .0096** .0097** (.0041) (.0041) (.0042) MHD a ages 26–35 .0026 .0026 .0025 (.0029) (.0029) (.0029) MHD a ages 36–45 .0058** .0057** .0061** (.0027) (.0028) (.0029) Dep ession a ages 16–45 .0062*** .0063*** .0065*** (.0022) (.0022) (.0022) No e.—We epo he esul s only on samples (ou comes) ha show signi ican e ec s in ables 3–6, B2, and B3. See able B23 o co esponding esul s o a ull sample o males o emales. MHD 5Men al heal h diso de equi ing hospi aliza ion. Columns 2 and 3 p e- sen con olled di ec e ec s based on Acha ya, Blackwell, and Sen (2016). Educa ion con- ols include he yea s o schooling and a dummy o ha ing a high school deg ee. Income con ol is he log o axable income a ages 26–35. Each cell epo s he es ima ed e ec o he comp ehensi e school e o m om sepa a e models. All models include coho and municipali y ixed e ec s. S anda d e o s epo ed in pa en hesis a e clus e ed a he mu- nicipal le el. In cols. 2 and 3, s anda d e o s ha e been boo s apped using 1,000 eplica- ions and clus e ed a he municipal le el. Coe icien s in i alics su i e a Romano-Wol co - ec ion o mul iple-hypo hesis es ing (Romano and Wol 2005) a he 10% signi icance le el (see able B22 o mo e de ails). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). amilies. These emales, be o e he e o m, would mos likely ha e been acked in o he selec i e academic cu iculum om age 11 o 16 and ex- posed only o high-abili y pee s. Howe e , a e he e o m, hey a e now exposed o comp ehensi e school and o pee s who would ha e pu sued he oca ional ack wi hou he e o m. P esumably, his new se o pee s, on a e age, is o en less well-beha ed and o lowe academic abili y. This may esul in a highe p obabili y o being exposed o dis up i e beha io s. P io wo k has shown he nega i e associa ion be ween being bullied a school and men al heal h in adul hood (Sigu dson, Wallande , and Sund 2014). In pa icula , bullied gi ls, a he han bullied boys, e en in e- quen ly bullied ones, a e mo e likely o su e om dep ession symp oms in adul hood (B uns ein Klomek e al. 2007, 43). Fu he empi ical e- sea ch is needed o es ablish whe he change in he exposu e o dis up i e beha io s is d i ing ou esul s. An al e na i e in e p e a ion o he obse ed ad e se e ec s elies on he mo e in ensi e compe i ion induced by he e o m. Indeed, he e- o m implied ha child en om disad an aged backg ounds now had a be e chance o compe e o places in he highe educa ion sys em: i equalized oppo uni ies by socioeconomic s a us and dec eased he in e - gene a ional co ela ion o ea nings (Pekka inen, Uusi alo, and Pekkala Ke 2009). A e he e o m, child en om p i ileged backg ounds aced mo e e ec i e compe i ion om he mos alen ed child en om disad- an aged backg ounds. This change may ha e a ec ed emales mo e, as e- males end o espond less a o ably o mo e in ense compe i ion han males (Niede le and Ves e lund 2011). In summa y, we do no ind e idence suppo ing a ole o he educa- ion/income channels (see also able B23). Since he e o m had a sligh ly nega i e e ec on income in adul hood, his could push owa d lowe men- al heal h, and i could be ha , once we con ol o income, he e o m had a posi i e e ec on men al heal h. Howe e , since ou ea lie esul s emain in ac a e we con ol o educa ion/income, we in e p e hese esul s as u he e idence ha he e o m, o e all, had no e ec on men al heal h a he han mul iple e ec s o di e en signs cancelling each o he ou . VII. Conclusion We con ibu e o he limi ed li e a u e on he e ec s o school acking e- gimes in Eu ope. Ou esul s a e based on a comp ehensi e school e o m ha was olled ou g adually ac oss Finnish municipali ies du ing 1972– 77. The e o m esul ed in child en om di e en socioeconomic back- g ounds and po en ially di e en academic abili ies being held in he same classes o i eex ayea s.The e o malsochanged he ypeo educa ion ecei ed, p o iding i e addi ional yea s o gene al educa ion o s uden s be- o e he s a o acking. Consequen ly, ou s udy also con ibu es o a g ow- ing li e a u e illus a ing he impo ance o human capi al measu es o he han simply he quan i y o educa ion. In hei e iew, Galama, Lle as-Muney, School T acking and Men al Heal h 313 and an Kippe sluis (2018) highligh ha o heal h- ela ed beha io s, he ype o educa ion ecei ed—gene al, oca ional, o academic—is mo e im- po an han he leng h o educa ion. As ou come a iables, we ocus on se e e men al diso de s, which cause subs an ial cos s o he heal h ca e sys em and lead o las ing nega i e ou comes a he indi idual le el, such as poo labo ma ke a achmen (E ne , F ank, and Kessle 1997). Al hough he gene aliza ion o ou es ima es o cu en policy se ings is no s aigh o wa d, he long- un heal h e ec s o school e o ms can be iden i ied only o bi h coho s ea ed many decades ago. Fu he mo e, webelie e ha ou esul scanp o idepo en ially aluableinsigh s o o he ypes o school acking sys ems, such as hose ha iden i y “gi ed”o “spe- cial educa ion”s uden s ea ly du ing schooling and p o ide ei he addi- ional oppo uni ies o seg ega ed ins uc ion alongside simila s uden s. O e all, we ind no signi ican e ec s on men al heal h– ela ed hospi- aliza ions o dea hs, e en hough he a e age ze o e ec s a e p ecisely es ima ed. Howe e , since we use egis e da a on hospi aliza ions and dea hs, ou esul s do no ule ou he possible e ec s o he e o m on less se ious men al heal h diso de s. He e ogenei y analysis shows ha pos poning he age o acking had an ad e se long- un e ec on men al heal h ou comes o emales om highly educa ed amilies, as hey we e mo e likely o be hospi alized o dep ession a e he e o m. Thus, in- c easing he age o acking may come a a cos o nega i e men al heal h e ec s o some g oups. Fu he mo e, we ind ha he esul o emales is no accoun ed o by changes in educa ion o income induced by he e o m. Ins ead, we p opose ha pee e ec s can possibly explain ou inding ega ding he a ec ed emales. Explo ing his mechanism is a p omising a enue o u u e esea ch. Appendix A ICD-8,ICD-9, and ICD-10 Diagnosis Codes A1. Men al Heal h–Rela ed Dea hs 1. Suicides a. ICD-8,ICD-9: E950–E959; b. ICD-10: X60–X84, Y87.0. 2. Inju ies o unde e mined in en (i.e., unde e mined whe he acciden ally o pu posely in lic ed) a. ICD-8,ICD-9: E980–E989; b. ICD-10: Y10–Y34, Y87.2, Y89.9. 3. Acciden al dea hs a. Poisoning (i.e., acciden al poisoning and exposu e o noxious subs ances) i. ICD-8: E850–E877; ii. ICD-9: E850–E869; iii. ICD-10: X40–X49. 314 Jou nal o Human Capi al Figu e B4.—E ec o he e o m on all-cause mo ali y a age 16–45: leads and lags a ound he yea be o e he e o m. Plo s a e based on baseline eg ession models, whe e he e o m dummy is eplaced wi h yea dummies o he leads and lags a ound he e o m yea . The plo ed poin s ep esen he es ima es on he lead and lag dummies. The omi ed ca ego y is he yea be o e he e o m (21). The es ima ed e ec s a e epo ed, oge he wi h hei 95% con idence in e als (based on obus s anda d e o s clus e ed a he municipal le el). 321 TABLE B2 E ec o he Re o m on Men al Heal h–Rela ed Dea hs by Age 45 Full Sample Pa en al Educa ion Low Mid High Males T ea men e ec 2.0014 2.0030 a, ** .0010 2.0001 (.0011) (.0014) (.0012) (.0012) R 2 .0033 .0056 .0086 .0147 Mean ou come .0187 .0143 .0051 .0036 Obse a ions 186,777 94,037 47,224 36,282 Females T ea men e ec .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) R 2 .0025 .0050 .0089 .0129 Mean ou come .0043 .0031 .0012 .0012 Obse a ions 179,537 90,881 46,092 34,828 No e.—The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. a Signi icance le el a e Romano-Wol co ec ion o mul iple-hypo hesis es ing (Romano and Wol 2005) is .104 (see able B11). ** Signi ican a he 5% le el ( wo-sided es ). TABLE B3 E ec o he Re o m on All-Cause Mo ali y by Age 45 Full Sample Pa en al Educa ion Low Mid High Males T ea men e ec 2.0016 2.0034 .0022 2.0014 (.0017) (.0022) (.0018) (.0017) R 2 .0032 .0064 .0099 .0118 Mean ou come .0458 .0339 .0132 .0102 Obse a ions 186,777 94,037 47,224 36,282 Females T ea men e ec .0007 2.0005 .0009 2.0008 (.0011) (.0013) (.0011) (.0012) R 2 .0025 .0051 .0098 .0112 Mean ou come .0176 .0122 .0050 .0047 Obse a ions 179,537 90,881 46,092 34,828 No e.—The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. None o he coe icien s a e signi ican a he 10% le el ( wo-sided es s). 322 Figu e B5.—Age-speci ic ea men e ec s on men al heal h– ela ed hospi aliza ions by sex and pa en al backg ound. Each age-speci ic ea men e ec o comp ehensi e schooling is es ima ed om sepa a e models. The dependen a iable is a dummy a iable indica ing whe he o no an indi idual was hospi alized as a esul o men al heal h diso de s du ing he speci ic age ange (e.g., 6–10). The le -hand side o he e ical line deno es he p e- ea men pe iod. T ea men e ec s a e epo ed, oge he wi h hei 95% con idence in- e als (based on obus s anda d e o s clus e ed a he municipal le el). 323 Figu e B6.—Illus a ion o he e o m s a us in municipali ies ac oss NUTS-2 egions. Da a a e agg ega ed o he municipal le el. Municipali ies ha e been classi ied acco ding o he NUTS-2 egional le el. The igu e illus a es he egional alloca ion o indi iduals wi h p e- and pos e o m schooling a age 11. Fo example, in he Helsinki and o he sou he n Fin- land (ou side Helsinki) egions, mos municipali ies ha e oughly he same numbe o in- di iduals wi h p e- and pos e o m schooling. In no he n and eas e n Finland, he sample is no as well balanced be ween indi iduals wi h p e- and pos e o m schooling. Popula ion sizes o he NUTS-2 egions ange om 74,500 o 114,000 inhabi an s. A e age numbe o obse a ions pe municipali y is 788 (median is 410). TABLE B4 Es ima ed Re o m E ec a e Con olling o Al e na i e Regional E ec s Full Sample Pa en al Educa ion Low Mid High A. Con ols o he Six Yea s o Adop ion Males Suicide by age 45 2.0001 2.0015 .0015 .0005 (.0008) (.0011) (.0010) (.0009) Men al heal h– ela ed dea h by age 45 2.0013 2.0027** .0011 .0002 (.0011) (.0013) (.0012) (.0011) Dea h by age 45 2.0012 2.0028 .0030* 2.0011 (.0017) (.0022) (.0018) (.0017) Men al heal h– ela ed hospi aliza ions a age 16–45 .0029 .0079** .0011 2.0052 (.0025) (.0036) (.0045) (.0042) Females Suicide by age 45 2.0002 2.0005 2.0001 .0004 (.0005) (.0005) (.0004) (.0007) Men al heal h– ela ed dea h by age 45 .0003 2.0003 .0003 .0007 (.0005) (.0006) (.0005) (.0008) 324 TABLE B4 (Con inued) Full Sample Pa en al Educa ion Low Mid High Dea h by age 45 .0007 2.0004 .0008 2.0007 (.0011) (.0013) (.0011) (.0012) Men al heal h– ela ed hospi aliza ions a age 16–45 .0031* .0024 2.0004 .0101** (.0019) (.0025) (.0034) (.0040) B. Con ols o he 18 NUTS-3 Regions Males Suicide by age 45 .0001 2.0014 .0007 .0011 (.0008) (.0009) (.0008) (.0009) Men al heal h– ela ed dea h by age 45 2.0009 2.0026** 2.0002 .0009 (.0010) (.0011) (.0010) (.0011) Dea h by age 45 2.0010 2.0033* .0009 2.0007 (.0015) (.0018) (.0016) (.0014) Men al heal h– ela ed hospi aliza ions a age 16–45 .0003 .0036 2.0059 2.0055 (.0028) (.0039) (.0041) (.0036) Females Suicide by age 45 2.0003 2.0006 2.0002 .0002 (.0004) (.0004) (.0003) (.0006) Men al heal h– ela ed dea h by age 45 2.0002 2.0009* 2.0000 .0005 (.0004) (.0005) (.0004) (.0007) Dea h by age 45 .0002 2.0013 .0002 2.0007 (.0009) (.0011) (.0010) (.0011) Men al heal h– ela ed hospi aliza ions a age 16–45 2.0003 2.0014 2.0017 .0043 (.0016) (.0024) (.0030) (.0037) No e.—Each cell epo s he es ima ed e o m e ec om sepa a e models. In panel A, we ha e eplaced he ull municipal ixed e ec s wi h six dummies o he yea s o adop ion (1972–77). In panel B, we ha e eplaced he municipal ixed e ec s wi h 18 NUTS-3 e- gional dummies. All models include coho ixed e ec s. S anda d e o s, epo ed in pa- en heses, a e clus e ed a he municipal le el. * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). TABLE B5 Es ima ed E ec o he Re o m a e Con olling o Mon h o Bi h Full Sample Pa en al Educa ion Low Mid High A. Con olling o Bi h Mon h Males Suicide by age 45 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Men al heal h– ela ed dea h by age 45 2.0014 2.0030** .0010 2.0001 (.0011) (.0013) (.0012) (.0012) Dea h by age 45 2.0016 2.0034 .0022 2.0014 (.0017) (.0022) (.0018) (.0017) Men al heal h– ela ed hospi aliza ions a age 16–45 .0027 .0074** .0012 2.0058 (.0025) (.0036) (.0045) (.0043) 325 TABLE B5 (Con inued) Full Sample Pa en al Educa ion Low Mid High Females Suicide by age 45 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Men al heal h– ela ed dea h by age 45 .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Dea h by age 45 .0007 2.0006 .0009 2.0008 (.0011) (.0013) (.0011) (.0012) Men al heal h– ela ed hospi aliza ions a age 16–45 .0029 .0023 2.0007 .0096** (.0019) (.0025) (.0034) (.0041) B. Con olling o Bi h Mon h and I s In e ac ion wi h Bi h Yea Males Suicide by age 45 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Men al heal h– ela ed dea h by age 45 2.0013 2.0030** .0010 2.0001 (.0011) (.0013) (.0012) (.0012) Dea h by age 45 2.0016 2.0034 .0022 2.0014 (.0017) (.0022) (.0018) (.0017) Men al heal h– ela ed hospi aliza ions a age 16–45 .0028 .0075** .0012 2.0060 (.0025) (.0036) (.0045) (.0043) Females Suicide by age 45 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Men al heal h– ela ed dea h by age 45 .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Dea h by age 45 .0007 2.0006 .0008 2.0009 (.0011) (.0013) (.0011) (.0012) Men al heal h– ela ed hospi aliza ions a age 16–45 .0029 .0024 2.0006 .0096** (.0019) (.0025) (.0034) (.0041) No e.—Each cell epo s he es ima ed e ec o he e o m om sepa a e models. All mod- els include coho as well as municipali y ixed e ec s. S anda d e o s, epo ed in pa en- heses, a e clus e ed a he municipal le el. ** Signi ican a he 5% le el (all wo-sided es s). TABLE B6 Timing o Schooling Rela i e o he Yea o Adop ion o he Re o m in he Municipali y o Residence Bi h coho Adop ion Yea in he Municipali y 1972 1973 1974 1975 1976 1977 1960 212223242526 1961 02122232425 1962 1021222324 1963 210212223 326 TABLE B6 (Con inued) Bi h coho Adop ion Yea in he Municipali y 1972 1973 1974 1975 1976 1977 1964 32102122 1965 4321021 1966 543210 No e.—Fo each bi h coho , people a ec ed by he e o m we e hose who li ed in munic- ipali ies whe e he e o m was adop ed in he yea s in bold ace on . People who we e bo n in 1965 and li ed in a egion whe e he e o m was adop ed in 1974 en e ed he i h g ade 2 yea s a e he i s pos e o m schooling coho in he egion. Also see igu e 2. TABLE B7 Robus ness Checks o he Es ima ed E ec o he Re o m on Suicide Full Sample Pa en al Educa ion Low Mid High Males Baseline esul s 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Con olling o p io men al heal h diso de s 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Con olling o p io men al heal h and pa en s’ men al heal h 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us 2.0002 2.0016 .0013 .0003 (.0008) (.0011) (.0010) (.0010) Con olling o egion-speci ic linea ime ends 2.0010 2.0008 .0008 2.0002 (.0009) (.0013) (.0012) (.0012) Excluding Helsinki me opoli an a ea 2.0008 2.0023* .0019* 2.0011 (.0010) (.0013) (.0011) (.0012) Excluding i <23o >32.0003 2.0011 .0019* .0001 (.0009) (.0012) (.0011) (.0010) Ex ending bi h coho s o 1960–66, excluding i <23o >3 .0002 2.0010 .0020** .0005 (.0009) (.0011) (.0010) (.0010) Females Baseline esul s 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Con olling o p io men al heal h diso de s 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Con olling o p io men al heal h and pa en s’ men al heal h 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us 2.0003 2.0006 2.0001 .0002 (.0005) (.0005) (.0004) (.0007) Con olling o egion-speci ic linea ime ends 2.0006 2.0006 2.0003 .0002 (.0006) (.0006) (.0005) (.0009) Excluding Helsinki me opoli an a ea 2.0006 2.0010 2.0002 .0007 (.0005) (.0006) (.0005) (.0009) Excluding i <23o >32.0010* 2.0012** 2.0002 2.0001 (.0005) (.0006) (.0005) (.0009) 327 TABLE B7 (Con inued) Full Sample Pa en al Educa ion Low Mid High Ex ending bi h coho s o 1960–66, excluding i <23o >32.0007 2.0010** 2.0002 .0004 (.0005) (.0005) (.0005) (.0008) No e.—Baseline esul s a e hose epo ed in able 3. All speci ica ions include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. is he numbe o yea s ela i e o he e o m yea in he municipali y o esidence (see able B6). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). TABLE B8 Robus ness Checks o he Es ima ed E ec o he Re o m on Men al Heal h–Rela ed Dea hs Full Sample Pa en al Educa ion Low Mid High Males Baseline esul s 2.0014 2.0030** .0010 2.0001 (.0011) (.0014) (.0012) (.0012) Con olling o p io men al heal h diso de s 2.0014 2.0030** .0010 2.0001 (.0011) (.0014) (.0012) (.0012) Con olling o p io men al heal h and pa en s’ men al heal h 2.0014 2.0030** .0010 2.0001 (.0011) (.0013) (.0012) (.0012) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us 2.0014 2.0030** .0010 2.0001 (.0011) (.0014) (.0012) (.0012) Con olling o egion-speci ic linea ime ends 2.0019 2.0017 .0004 2.0013 (.0013) (.0016) (.0015) (.0014) Excluding Helsinki me opoli an a ea 2.0026* 2.0043*** .0019 2.0012 (.0013) (.0016) (.0013) (.0015) Excluding i <23o >32.0013 2.0024 .0020 2.0005 (.0012) (.0016) (.0013) (.0013) Ex ending bi h coho s o 1960–66, excluding i <23o >32.0009 2.0023 .0020* 2.0002 (.0011) (.0014) (.0012) (.0012) Females Baseline esul s .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Con olling o p io men al heal h diso de s .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Con olling o p io men al heal h and pa en s’ men al heal h .0002 2.0003 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us .0002 2.0004 .0004 .0005 (.0005) (.0006) (.0005) (.0008) Con olling o egion-speci ic linea ime ends .0001 2.0001 .0004 .0008 (.0006) (.0007) (.0006) (.0010) Excluding Helsinki me opoli an a ea 2.0001 2.0008 .0002 .0011 (.0006) (.0007) (.0006) (.0010) Excluding i <23o >32.0006 2.0009 .0002 .0005 (.0006) (.0008) (.0006) (.0010) 328 TABLE B8 (Con inued) Full Sample Pa en al Educa ion Low Mid High Ex ending bi h coho s o 1960–66, excluding i <23o >32.0003 2.0009 .0003 .0011 (.0006) (.0006) (.0006) (.0009) No e.—Baseline esul s a e hose epo ed in able B2. All speci ica ions include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. is iming o he e o m in he numbe o yea s ela i e o he e o m yea in he municipali y o esidence (see able B6). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). TABLE B9 Robus ness Checks o he Es ima ed E ec o he Re o m on All-Cause Mo ali y Full Sample Pa en al Educa ion Low Mid High Males Baseline esul s 2.0016 2.0034 .0022 2.0014 (.0017) (.0022) (.0018) (.0017) Con olling o p io men al heal h diso de s 2.0016 2.0034 .0023 2.0014 (.0017) (.0022) (.0018) (.0017) Con olling o p io men al heal h and pa en s’ men al heal h 2.0016 2.0034 .0022 2.0014 (.0017) (.0022) (.0018) (.0017) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us 2.0014 2.0033 .0023 2.0014 (.0017) (.0022) (.0018) (.0017) Con olling o egion-speci ic linea ime ends 2.0022 2.0007 .0008 2.0019 (.0020) (.0026) (.0023) (.0020) Excluding Helsinki me opoli an a ea 2.0016 2.0042 .0037* 2.0018 (.0020) (.0026) (.0020) (.0021) Excluding i <23o >32.0010 2.0014 .0035* .0001 (.0019) (.0024) (.0019) (.0017) Ex ending bi h coho s o 1960–66, excluding i <23o >3 .0000 2.0011 .0043** .0015 (.0018) (.0023) (.0019) (.0018) Females Baseline esul s .0007 2.0005 .0009 2.0008 (.0011) (.0013) (.0011) (.0012) Con olling o p io men al heal h diso de s .0007 2.0005 .0009 2.0008 (.0011) (.0013) (.0011) (.0012) Con olling o p io men al heal h and pa en s’ men al heal h .0007 2.0005 .0009 2.0008 (.0011) (.0013) (.0011) (.0012) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us .0007 2.0007 .0010 2.0009 (.0011) (.0013) (.0011) (.0012) Con olling o egion-speci ic linea ime ends .0020* .0015 .0010 2.0002 (.0012) (.0017) (.0013) (.0014) Excluding Helsinki me opoli an a ea .0003 2.0020 .0013 2.0006 (.0013) (.0015) (.0013) (.0016) 329 TABLE B9 (Con inued) Full Sample Pa en al Educa ion Low Mid High Excluding i <23o >3 .0002 2.0005 .0005 2.0013 (.0011) (.0015) (.0012) (.0014) Ex ending bi h coho s o 1960–66, excluding i <23o >3 .0001 2.0001 .0003 2.0010 (.0011) (.0013) (.0012) (.0013) No e.—Baseline esul s a e hose epo ed in able B3. All speci ica ions include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. is iming o he e o m in numbe o yea s ela i e o he e o m yea in he municipali y o esidence (see able B6). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). TABLE B10 Robus ness Checks o he Es ima ed E ec o he Re o m on Men al Heal h–Rela ed Hospi aliza ions a Age 16–45 Full Sample Pa en al Educa ion Low Mid High Males Baseline esul s .0027 .0074** .0012 2.0058 (.0025) (.0036) (.0045) (.0043) Con olling o p io men al heal h diso de s .0028 .0075** .0014 2.0057 (.0025) (.0036) (.0045) (.0043) Con olling o p io men al heal h and pa en s’men al heal h .0029 .0078** .0012 2.0056 (.0025) (.0036) (.0046) (.0043) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us .0029 .0076** .0018 2.0057 (.0025) (.0036) (.0046) (.0042) Con olling o egion-speci ic linea ime ends 2.0012 .0030 2.0011 2.0043 (.0030) (.0045) (.0053) (.0048) Excluding Helsinki me opoli an a ea .0032 .0051 .0049 2.0078 (.0026) (.0038) (.0051) (.0057) Excluding i <23o >3 .0024 .0068* .0022 2.0037 (.0028) (.0041) (.0050) (.0045) Ex ending bi h coho s o 1960–66, excluding i <23o >3 .0018 .0039 .0046 2.0017 (.0024) (.0036) (.0046) (.0043) Females Baseline esul s .0029 .0022 2.0007 .0095** (.0019) (.0025) (.0034) (.0041) Con olling o p io men al heal h diso de s .0030 .0023 2.0007 .0097** (.0019) (.0025) (.0034) (.0041) Con olling o p io men al heal h and pa en s’men al heal h .0029 .0022 2.0007 .0097** (.0019) (.0025) (.0034) (.0041) Con olling o p io men al heal h diso de s and hei in e ac ion wi h ea men s a us .0030 .0022 2.0009 .0100** (.0018) (.0025) (.0034) (.0041) 330 TABLE B18 (Con inued) Ou come Full Sample Pa en al Educa ion Low Mid High Dea h by age 45 .0007 .0000 .0016 2.0002 (.0010) (.0011) (.0010) (.0012) MHD by age 45 .0002 2.0005 2.0022 .0066* (.0017) (.0026) (.0032) (.0037) Obse a ions 186,629 94,138 47,363 36,075 No e.—He e, indi iduals a e classi ied in o ea men on he basis o hei municipali y o bi h ins ead o municipali y a age 11 (see esul s in ables 3, 4, B2, B3, B19). Indi iduals who mig a e be ween ages 11 and 16 a e no d opped om he sample. MHD 5Men al heal h diso de s equi ing hospi aliza ion. The ull sample also includes indi iduals o whom in- o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y o bi h ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus- e ed a he municipal le el. * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). TABLE B19 E ec o he Re o m: T ea men Assignmen Acco ding o he Municipali y a Age 10 Ou come Full Sample Pa en al Educa ion Low Mid High Males Suicide by age 45 2.0003 2.0015 .0013 .0000 (.0006) (.0011) (.0010) (.0011) Men al heal h– ela ed dea h by age 45 2.0010 2.0027** .0012 2.0001 (.0008) (.0013) (.0012) (.0012) Dea h by age 45 2.0004 2.0023 .0030* 2.0012 (.0013) (.0022) (.0018) (.0017) MHD by age 45 .0026 .0065* .0000 2.0055 (.0024) (.0034) (.0047) (.0041) Obse a ions 181,136 95,620 48,189 37,327 Females Suicide by age 45 2.0003 2.0006 2.0003 .0003 (.0003) (.0005) (.0004) (.0007) Men al heal h– ela ed dea h by age 45 .0001 2.0002 .0004 .0005 (.0004) (.0006) (.0005) (.0008) Dea h by age 45 2.0001 2.0004 .0009 2.0008 (.0007) (.0013) (.0011) (.0012) MHD by age 45 .0023 .0021 2.0023 .0080** (.0019) (.0025) (.0035) (.0039) Obse a ions 175,304 92,465 46,998 35,841 No e.—He e, indi iduals a e classi ied in o ea men on he basis o hei municipali y a age 10 ins ead o municipali y a age 11 (see esul s in ables 3, 4, B2, B3, and B18). Indi id- uals who mig a e be ween ages 11 and 16 ha e no been d opped om he sample. MHD 5 Men al heal h diso de s equi ing hospi aliza ion. The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. All models include coho and municipali y-a -age-10 ixed e ec s. S anda d e o s, epo ed in pa en- heses, a e clus e ed a he municipal le el. * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). 337 TABLE B20 Es ima ed E ec o he Re o m on Educa ional Ou comes Ou come Full Sample Pa en al Educa ion Low Mid High Males Yea s o schooling 2.0489** 2.0630** 2.1384*** .0229 (.0222) (.0266) (.0416) (.0562) [12.777] [12.183] [12.790] [14.577] Has a high school deg ee 2.0022 2.0020 2.0163** 2.0020 (.0037) (.0050) (.0068) (.0082) [.291] [.177] [.265] [.636] High school exam sco e (pe cen ile): In na i e language .0007 2.0047 2.0036 .0047 (.0059) (.0080) (.0093) (.0086) [.463] [.431] [.429] [.503] In ad anced ma h 2.0173** 2.0104 2.0370** 2.0147 (.0073) (.0128) (.0164) (.0094) [.538] [.500] [.511] [.571] In basic ma h 2.0103 .0067 2.0329** .0005 (.0070) (.0132) (.0144) (.0117) [.506] [.485] [.493] [.535] Highes deg ee comple ed: Voca ional seconda y educa ion 2.0005 2.0073 .0143** .0040 (.0042) (.0067) (.0066) (.0074) [.459] [.536] [.502] [.232] Voca ional college .0020 .0041 .0005 2.0027 (.0027) (.0038) (.0057) (.0069) [.140] [.116] [.157] [.192] Uni e si y mas e ’s2.0048* 2.0039 2.0151*** .0015 (.0027) (.0028) (.0049) (.0086) [.113] [.055] [.090] [.301] Females Yea s o schooling 2.0139 2.0329 .0157 .0126 (.0207) (.0258) (.0443) (.0498) [13.489] [12.978] [13.532] [14.977] Has a high school deg ee 2.0082** 2.0109** .0024 2.0107 (.0041) (.0053) (.0093) (.0077) [.498] [.383] [.505] [.790] High school exam sco e (pe cen ile): In na i e language 2.0083 2.0084 2.0190** .0019 (.0055) (.0060) (.0086) (.0078) [.497] [.457] [.472] [.568] In ad anced ma h 2.0137* 2.0206* 2.0242* 2.0063 (.0071) (.0124) (.0123) (.0111) [.484] [.447] [.455 [.528] In basic ma h 2.0122** 2.0229*** 2.0241** .0159* (.0061) (.0082) (.0105) (.0084) [.502] [.472] [.487] [.556] Highes deg ee comple ed: Voca ional seconda y educa ion 2.0049 2.0056 2.0174** .0075 (.0045) (.0059) (.0081) (.0074) [.379] [.454] [.394] [.194] Voca ional college .0021 .0011 .0110 2.0052 (.0042) (.0053) (.0074) (.0095) [.254] [.244] [.275] [.264] 338 TABLE B20 (Con inued) Ou come Full Sample Pa en al Educa ion Low Mid High Uni e si y mas e ’s2.0025 2.0070** .0032 .0043 (.0032) (.0035) (.0054) (.0093) [.148] [.086] [.131] [.337] No e.—Each cell epo s he es ima ed e ec o he comp ehensi e school e o m om sep- a a e models. High school exam sco e is a ailable only o hose who ha e g adua ed om a high school. All models include dummy a iables o he yea o e o m and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. Mean alues o dependen a iables a e epo ed in squa e b acke s below he s anda d e - o . The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). TABLE B21 Es ima ed Re o m E ec on Economic Ou comes Ou come Full Sample Pa en al Educa ion Low Mid High Males Log o a e age income: A ages 26–45 2.0092* 2.0187*** 2.0016 .0022 (.0050) (.0065) (.0089) (.0113) [3.307] [3.234] [3.332] [3.528] A ages 26–35 2.0022 2.0144** .0096 .0120 (.0048) (.0057) (.0090) (.0097) [3.111] [3.063] [3.135] [3.247] A ages 36–45 2.0160*** 2.0234*** 2.0121 2.0048 (.0060) (.0080) (.0107) (.0137) [3.444] [3.350] [3.462] [3.707] Employmen a e: A ages 26–45 2.0029 2.0057** 2.0004 .0044 (.0025) (.0031) (.0044) (.0046) [.766] [.765] [.806] [.815] A ages 26–35 2.0017 2.0052* 2.0002 .0067 (.0027) (.0031) (.0054) (.0049) [.740] [.739] [.775] [.776] A ages 36–45 2.0047 2.0077** 2.0008 .0026 (.0027) (.0036) (.0046) (.0050) [.803] [.799] [.841] [.858] Females Log o a e age income: A ages 26–45 2.0083** 2.0145*** 2.0038 .0018 (.0039) (.0049) (.0078) (.0103) [3.047] [2.998] [3.069] [3.204] A ages 26–35 2.0036 2.0079 2.0014 .0025 (.0039) (.0049) (.0071) (.0097) [2.861] [2.825] [2.876] [2.977] A ages 36–45 2.0140*** 2.0211*** 2.0102 2.0006 (.0046) (.0064) (.0093) (.0116) [3.178] [3.113] [3.198] [3.351] 339 TABLE B21 (Con inued) Ou come Full Sample Pa en al Educa ion Low Mid High Employmen a e: A ages 26–45 2.0060** 2.0066** 2.0029 2.0015 (.0025) (.0029) (.0046) (.0055) [.726] [.732] .760] [.764] A ages 26–35 2.0034 2.0056* 2.0000 .0012 (.0025) (.0033) (.0049) (.0059) [.673] [.668] [.701] [.704] A ages 36–45 2.0085*** 2.0078** 2.0060 2.0050 (.0031) (.0035) (.0055) (.0059) [.782] [.790] [.821] [.824] No e.—Each cell epo s he es ima ed e ec o he comp ehensi e school e o m om sep- a a e models. All models include coho and municipali y ixed e ec s. S anda d e o s, e- po ed in pa en heses, a e clus e ed a he municipal le el. Mean alues o dependen a - iables a e epo ed in squa e b acke s below he s anda d e o . The ull sample also includes indi iduals o whom in o ma ion abou pa en s (such as pa en al educa ion) is missing. * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). *** Signi ican a he 1% le el (all wo-sided es s). TABLE B22 Significance Le els o he Media o Resul s a e Adjus men o Mul iple-Hypo hesis Tes ing Ou come Media ion Analysis Educa ion [wi h RW adjus ed p-Value] Educa ion and Income [wi h RW adjus ed p-Value] A. Males, Low-Educa ed Pa en s Suicide by age 45 2.0018 2.0017 [.337] [.285] Men al heal h– ela ed dea h by age 45 2.0033 2.0027 [.0620] [.1139] Dea h by age 45 2.0040 2.0042 [.241] [.144] MHD a ages 16–45 .0061 .0060 [.301] [.367] MHD a ages 26–35 .0056 .0049 [.045] [.118] MHD a ages 36–45 .0038 .0032 [.354] [.524] Subs ance-use diso de a ages 16–45 .0041 .0033 [.184] [.370] B. Females, Highly Educa ed Pa en s Suicide by age 45 .0003 .0005 [.925] [.843] Men al heal h– ela ed dea h by age 45 .0005 .0007 [.897] [.711] Dea h by age 45 2.0008 2.0008 [.861] [.868] 340 TABLE B22 (Con inued) Ou come Media ion Analysis Educa ion [wi h RW adjus ed p-Value] Educa ion and Income [wi h RW adjus ed p-Value] MHD a ages 16–45 .0096 .0097 [.092] [.086] MHD a ages 26–35 .0026 .0025 [.633] [.654] MHD a ages 36–45 .0057 .0061 [.147] [.113] Dep ession a ages 16–45 .0063 .0065 [.022] [.017] No e.—We p esen con olled di ec e ec s based on Acha ya, Blackwell, and Sen (2016), a e con olling o educa ion and/o income. The Romano-Wol (RW) s ep-down ad- jus ed p- alues, epo ed in squa e b acke s, a e obus o mul iple hypo heses es ing ( join ly o he ull sample and he h ee subsamples by pa en al backg ound). They ha e been compu ed using 2,000 boo s ap eplica ions. Signi icance le els a e based on s an- da d e o s ha a e clus e ed a he municipal le el. MHD 5Men al heal h diso de s e- qui ing hospi aliza ion. See also able 7. TABLE B23 Es ima ed Re o m E ec , Con olling o Educa ion and Income Media o s La e in Li e (by Gende ) Ou come Baseline Es ima es (1) Media ion Analysis Educa ion (2) Educa ion and Income (3) A. Males Suicide by age 45 2.0002 2.0004 2.0004 (.0008) (.0008) (.0007) Men al heal h– ela ed dea h by age 45 2.0014 2.0016 2.0015 (.0011) (.0011) (.0010) Dea h by age 45 2.0016 2.0022 2.0022 (.0017) (.0017) (.0016) MHD a ages 16–45 .0027 .0018 .0021 (.0025) (.0026) (.0025) MHD a ages 26–35 .0020 .0016 .0016 (.0016) (.0016) (.0016) MHD a ages 36–45 .0018 .0013 .0014 (.0016) (.0016) (.0016) Subs ance-use diso de a ages 16–45 .0038** .0033** .0031* (.0016) (.0016) (.0016) B. Females Suicide by age 45 2.0003 2.0003 2.0001 (.0005) (.0005) (.0004) Men al heal h– ela ed dea h by age 45 .0002 .0002 .0002 (.0005) (.0005) (.0005) Dea h by age 45 .0007 .0006 .0005 (.0011) (.0011) (.010) MHD a ages 16–45 .0029 .0027 .0029 (.0019) (.0019) (.0020) MHD a ages 26–35 .0020 .0019 .0019 (.0012) (.0012) (.0013) 341 TABLE B23 (Con inued) Ou come Baseline Es ima es (1) Media ion Analysis Educa ion (2) Educa ion and Income (3) MHD a ages 36–45 .0012 .0011 .0013 (.0014) (.0015) (.0016) Dep ession a ages 16–45 .0015 .0014 .0016 (.0009) (.0010) (.0010) No e.—We epo he esul s only on ou comes pa allel o hose in able 7. MHD 5Men al heal h diso de equi ing hospi aliza ion. Columns 2 and 3 p esen con olled di ec e ec s based on Acha ya, Blackwell, and Sen (2016). Educa ion con ols include he yea s o schooling and a dummy o ha ing a high school deg ee. Income con ol is he log o ax- able income a ages 26–35. Each cell epo s he es ima ed e ec o he comp ehensi e school e o m om sepa a e models. All models include coho and municipali y ixed e ec s. S anda d e o s, epo ed in pa en heses, a e clus e ed a he municipal le el. In cols. 2 and 3, s anda d e o s ha e been boo s apped using 1,000 eplica ions and clus e ed a he municipal le el. Coe icien s in i alics su i e a Romano-Wol co ec ion o mul iple- hypo hesis es ing (Romano and Wol 2005) a he 10% signi icance le el (using es ima es o ull sample and he h ee subsamples o low, middle, and high pa en al educa ion wi hin gende ). * Signi ican a he 10% le el (all wo-sided es s). ** Signi ican a he 5% le el (all wo-sided es s). Appendix C Tes ing o Mechanisms In sec ion VI, we examine he ex en o which pos ea men schooling and in- come media e he e ec o he comp ehensi e school e o m on men al heal h– ela ed hospi aliza ions (and dea hs). A simply augmen ed eg ession model wi h pos ea men media o a iables can lead o biased es ima es (see Acha ya, Black- well, and Sen 2016). Acha ya, Blackwell, and Sen (2016) apply a sequen ial p ocedu e ha consis- en ly es ima es he ea men e ec while holding he alues o po en ial medi- a o s ixed. Adop ing hei app oach, we es ima e hese a e age con olled di ec e ec s (ACDEs) as ollows (see able 7): 1. Es ima e an augmen ed model: yij 5a1hj1 1bREFORMj 1d0XPos i 1eij , whe e XPos ia e addi ional pos ea men con ols (i.e., he yea s o schooling and dummy o ha ing a high school deg ee, and/o he log o axable income a ages 26–35). 2. C ea e a demedia ed ou come a iable: ~yij 5yij 2 ^ d0XPos i. 3. Es ima e a model o he demedia ed ou come: ~yij 5a1hj1 1k REFORMj 1eij , whe e kis he ACDE o he e o m. Because he inal es ima ion s ep con ains a gene a ed dependen a iable (~yij ), he s anda d e o s ha e been boo s apped using 1,000 eplica ions o he ull p o- cess (s eps 1–3). Addi ionally, he boo s ap eplica ions ha e been clus e ed a he municipal le el. I he es ima ed ACDE is signi ican ly di e en om ze o, we can conclude ha he comp ehensi e school e o m has in luenced men al heal h– ela ed hospi- aliza ions h ough pa hways o he han he educa ion and income channel. By 342 Jou nal o Human Capi al con as , i he es ima ed ACDE is ze o, hen he e o m has no had an addi ional e ec on men al heal h– ela ed hospi aliza ions once he p oposed mechanisms ha e been accoun ed o . In o he wo ds, he e o m e ec would be exclusi ely d i en by he mechanisms ela ed o he changes in educa ion and income. Re e ences Acha ya, A idi , Ma hew Blackwell, and Maya Sen. 2016. “Explaining Causal Findings wi hou Bias: De ec ing and Assessing Di ec E ec s.”Ame ican Poli . Sci. Re . 110 (3): 512–29. Alexande , Diane, and Molly Schnell. 2019. “Jus Wha he Nu se P ac i ione O - de ed: Independen P esc ip i e Au ho i y and Popula ion Men al Heal h.”J. Heal h Econ. 66:145–62. Ang is , Joshua D., and Alan B. K uege . 1991. “Does Compulso y School A en- dance A ec Schooling and Ea nings?”Q.J.E. 106 (4): 979–1014. A anasio, O azio, A len Gua ín, Ca los Medina, and Cos as Meghi . 2017. “Voca- ional T aining o Disad an aged You h in Colombia: A Long-Te m Follow- Up.”Ame ican Econ. J.: Appl. Econ. 9 (2): 131–43. Be s, Julian R. 2011. “The Economics o T acking in Educa ion.”In Handbook o he Economics o Educa ion, ol. 3, edi ed by E ic A. Hanushek, S ephen Machin, and Ludge Woessmann, 341–81. Ams e dam: No h-Holland. Bhulle , Manudeep, Magne Mogs ad, and Kjell G. Sal anes. 2017. “Li e-Cycle Ea nings, Educa ion P emiums, and In e nal Ra es o Re u n.”J. Labo Econ. 35 (4): 993–1030. Bijwaa d, Go e E., Mikko My skylä, and Pe Tynelius. 2018. “The Impac o Men- al P oblems on Mo ali y and How I Is Mode a ed by Educa ion.”IZA Discus- sion Pape no. 11591, Ins . Labo Econ., Bonn. h p:// p.iza.o g/dp11591.pd . B unello, Gio gio, Massimo Giannini, and Kenn A iga. 2007. “The Op imal Tim- ing o School T acking: A Gene al Model wi h Calib a ion o Ge many.”In Schools and he Equal Oppo uni y P oblem, edi ed by Ludge Woessmann and Paul E. Pe e son, 129–56. Camb idge, MA: MIT P ess. B uns ein Klomek, Ana , F ank Ma occo, Ma jo ie Kleinman, I in Sam Schon eld, and Madelyn S. Gould. 2007. “Bullying, Dep ession, and Suicidali y in Adolescen s.”J. Ame ican Acad. Child and Adolescen Psychia y 46 (1): 40–49. Bubonya, Melisa, Debo ah A. Cobb-Cla k, and Ma k Wooden. 2017. “Men al Heal h and P oduc i i y a Wo k: Does Wha You Do Ma e ?”Labou Econ. 46:150–65. Cicala, S e e, Roland G. F ye , and Jö g L. Spenkuch. 2018. “Sel -Selec ion and Compa a i e Ad an age in Social In e ac ions.”J. Eu opean Econ. Assoc. 16 (4): 983–1020. Cla k, Damon, and Hea he Roye . 2013. “The E ec o Educa ion on Adul Mo - ali y and Heal h: E idence om B i ain.”A.E.R. 103 (6): 2087–120. Cu le , Da id M., and Ad iana Lle as-Muney. 2008. “Educa ion and Heal h: E al- ua ing Theo ies and E idence.”In Making Ame icans Heal hie : Social and Eco- nomic Policy as Heal h Policy, edi ed by Robe F. Schoeni, James S. House, Geo ge A. Kaplan, and Ha old Pollack, 29–60. New Yo k: Russell Sage Found. Elsne , Benjamin, and Ingo E. Ispho ding. 2018. “Rank, Sex, D ugs, and C ime.” J. Human Resou ces 53 (2): 356–81. Enoch, Ma y-Anne. 2011. “The Role o Ea ly Li e S ess as a P edic o o Alcohol and D ug Dependence.”Psychopha macology 214 (1): 17–31. E ne , Susan L., Richa d G. F ank, and Ronald C. Kessle . 1997. “The Impac o Psychia ic Diso de s on Labo Ma ke Ou comes.”Indus. and Labo Rela ions Re . 51 (1): 64–81. School T acking and Men al Heal h 343 F ank, Richa d G., and Thomas G. McGui e. 2000. “Economics and Men al Heal h.”In Handbook o Heal h Economics, ol. 1B, edi ed by An hony J. Culye and Joseph P. Newhouse, 893–954. Ams e dam: No h-Holland. Galama, Ti us J., Ad iana Lle as-Muney, and Hans an Kippe sluis. 2018. “The E ec o Educa ion on Heal h and Mo ali y: A Re iew o Expe imen al and Quasi-expe imen al E idence.”In Ox o d Resea ch Encyclopedia o Economics and Finance. h ps://doi.o g/10.1093/ac e o e/9780190625979.013.7. Ge d ham, Ul -G. 1997. “Equi y in Heal h Ca e U iliza ion: Fu he Tes s Based on Hu dle Models and Swedish Mic o Da a.”Heal h Econ. 6 (3): 303–19. Hakulinen, Ch is ian, Ma ko Elo ainio, Ma i A man, e al. 2019. “Men al Dis- o de s and Long-Te m Labou Ma ke Ou comes: Na ionwide Coho S udy o 2 055 720 Indi iduals.”Ac a Psychia ica Scandina ica 140 (4): 371–81. Hall, Ca oline. 2012. “The E ec s o Reducing T acking in Uppe Seconda y School: E idence om a La ge-Scale Pilo Scheme.”J. Human Resou ces 47 (1): 237–69. Hen iksson, Ma kus M., Hille i M. A o, Mau i J. Ma unen, e al. 1993. “Men al Diso de s and Como bidi y in Suicide.”Ame ican J. Psychia y 150 (6): 935–40. Jones, And ew M., John E. Roeme , and Ped o Rosa Dias. 2014. “Equalising Op- po uni ies in Heal h h ough Educa ional Policy.”Soc. Choice and Wel a e 43 (3): 521–45. Lage , An on, Dominika Seblo a, Daniel Falks ed , and Ma in Lö dén. 2016. “Cogni i e and Emo ional Ou comes a e P olonged Educa ion: A Quasi- expe imen on 320 182 Swedish Boys.”In e na . J. Epidemiology 46:303–11. Lah i, Anniina, Pi kko Räsänen, Kaisa Riala, Si pa Ke änen, and HelinäHakko. 2011. “You h Suicide T ends in Finland, 1969–2008.”J. Child Psychology and Psy- chia y 52 (9): 984–91. Lah i, R., and A. Pen ilä. 2001. “The Validi y o Dea h Ce i ica es: Rou ine Val- ida ion o Dea h Ce i ica ion and I s E ec s on Mo ali y S a is ics.”Fo ensic Sci. In e na . 115 (1): 15–32. Laya d, Richa d. 2013. “Men al Heal h: The New F on ie o Labou Econom- ics.”IZA J. Labo Policy 2:2. Lle as-Muney, Ad iana. 2005. “The Rela ionship be ween Educa ion and Adul Mo ali y in he Uni ed S a es.”Re . Econ. S udies 72 (1): 189–221. Meghi , Cos as, Må en Palme, and Emilia Simeono a. 2018. “Educa ion and Mo ali y: E idence om a Social Expe imen .”Ame ican Econ. J.: Appl. Econ. 10 (2): 234–56. Niede le, Mu iel, and Lise Ves e lund. 2011. “Gende and Compe i ion.”Ann. Re . Econ. 3:601–30. OECD (O ganisa ion o Economic Co-ope a ion and De elopmen ). 2004. Lea ning o Tomo ow’s Wo ld: Fi s Resul s om PISA 2003. Pa is, OECD. h ps://www.oecd-ilib a y.o g/con en /publica ion/9789264006416-en. O eopoulos, Philip, and Kjell G. Sal anes. 2011. “P iceless: The Nonpecunia y Bene i s o Schooling.”J. Econ. Pe spec i es 25 (1): 159–84. Pede sen, Ca s en B., Ole Mo s, Aksel Be elsen, e al. 2014. “A Comp ehensi e Na ionwide S udy o he Incidence Ra e and Li e ime Risk o T ea ed Men al Diso de s.”JAMA Psychia y 71 (5): 573–81. Pekkala Ke , Sa i, Tuomas Pekka inen, and Roope Uusi alo. 2013. “School T ack- ing and De elopmen o Cogni i e Skills.”J. Labo Econ. 31 (3): 577–602. Pekka inen, Tuomas. 2008. “Gende Di e ences in Educa ional A ainmen : E - idence on he Role o T acking om a Finnish Quasi-expe imen .”Scandina- ian J. Econ. 110 (4): 807–25. Pekka inen, Tuomas, Roope Uusi alo, and Sa i Pekkala Ke . 2009. “School T ack- ing and In e gene a ional Income Mobili y: E idence om he Finnish Com- p ehensi e School Re o m.”J. Public Econ. 93 (7–8): 965–73. 344 Jou nal o Human Capi al Pi kola, Sami, Reijo Sund, Eila Sailas, and K is ian Wahlbeck. 2009. “Communi y Men al-Heal h Se ices and Suicide Ra e in Finland: A Na ionwide Small-A ea Analysis.”Lance 373 (9658): 147–53. Ra es eijn, Bas ian, Hans an Kippe sluis, Mau icio A endano, Pekka Ma i- kainen, Hannu Vessa i, and Eddy an Doo slae . 2017. “The Impac o La e T acking on Mo ali y by Pa en al Income in Finland.”Discussion Pape no. 2017–030/V, Tinbe gen Ins . Ri e , Philip L., Ani a L. S ewa , Hulya Kaymaz, Da id S. Sobel, Daniel A. Block, and Ka e R. Lo ig. 2001. “Sel -Repo s o Heal h Ca e U iliza ion Compa ed o P o ide Reco ds.”J. Clinical Epidemiology 54 (2): 136–41. Romano, Joseph P., and Michael Wol . 2005. “S epwise Mul iple Tes ing as Fo - malized Da a Snooping.”Econome ica 73 (4): 1237–82. Sahlbe g, Pasi. 2014. Finnish Lessons 2.0: Wha Can he Wo ld Lea n om Educa ional Change in Finland? 2nd ed. New Yo k: Teache s Coll. P ess, Columbia Uni . Salk, Rachel H., Jane S. Hyde, and Lyn Y. Ab amson. 2017. “Gende Di e ences in Dep ession in Rep esen a i e Na ional Samples: Me a-analyses o Diagnoses and Symp oms.”Psychological Bull. 143 (8): 783–822. San a i a, To s en, Nina San a i a, The esa S. Be ancou , and S ephen E. Gil- man. 2015. “Long Te m Men al Heal h Ou comes o Finnish Child en E a- cua ed o Swedish Families du ing he Second Wo ld Wa and Thei Non- e acua ed Siblings: Coho S udy.”BMJ 350:g7753. Sa een, Ji ende , F ank Jacobi, B ian J. Cox, Shay-Lee Belik, Ian Cla a, and Mu - ay B. S ein. 2006. “Disabili y and Poo Quali y o Li e Associa ed wi h Como - bid Anxie y Diso de s and Physical Condi ions.”A chi es In e nal Medicine 166 (19): 2109–16. Sigu dson, J. F., J. Wallande , and A. M. Sund. 2014. “Is In ol emen in School Bullying Associa ed wi h Gene al Heal h and Psychosocial Adjus men Ou - comes in Adul hood?”Child Abuse and Neglec 38 (10): 1607–17. Some ki i, U ho. 1982. Pe uskoulu: Syn y, Kehi yminen ja Tule aisuus [Comp ehen- si e School: O igins, De elopmen and Fu u e]. Van aa: Kunnallispaino. Sund, Reijo. 2012. “Quali y o he Finnish Hospi al Discha ge Regis e : A Sys em- a ic Re iew.”Scandina ian J. Public Heal h 40 (6): 505–15. Su isaa i, J., T. Aal o-Se älä, A. Tuulio-Hen iksson, e al. 2009. “Men al Diso de s in Young Adul hood.”Psychological Medicine 39 (2): 287–99. Üs ün, T. Bedi han, and Somna h Cha e ji. 2001. “Global Bu den o Dep essi e Diso de s and Fu u e P ojec ions.”In Dep ession: Social and Economic Timebomb, edi ed by Ann Dawson and And e Tylee, 31–43. BMJ Books, 31–43. Vigo, Daniel, G aham Tho nic o , and Ri a A un. 2016. “Es ima ing he T ue Global Bu den o Men al Illness.”Lance Psychia y 3 (2): 171–78. Whi e o d, Ha ey A., Louisa Degenha d , Jü gen Rehm, e al. 2013. “Global Bu - den o Disease A ibu able o Men al and Subs ance Use Diso de s: Findings om he Global Bu den o Disease S udy 2010.”Lance 382 (9904): 1575–86. WHO (Wo ld Heal h O g.). 2014. Social De e minan s o Men al Heal h. Gene a: WHO. Woold idge, Je ey M. 2001. Econome ic Analysis o C oss Sec ion and Panel Da a. MIT P ess, Camb idge, MA. School T acking and Men al Heal h 345