Retirement age reforms and worker substitutability: Implications for employment of older workers

Badalyan, Sona
Wo king Pape
Re i emen age e o ms and wo ke subs i u abili y:
Implica ions o employmen o olde wo ke s
IAB-Discussion Pape , No. 14/2025
P o ided in Coope a ion wi h:
Ins i u e o Employmen Resea ch (IAB)
Sugges ed Ci a ion: Badalyan, Sona (2025) : Re i emen age e o ms and wo ke subs i u abili y:
Implica ions o employmen o olde wo ke s, IAB-Discussion Pape , No. 14/2025, Ins i u ü
A bei sma k - und Be u s o schung (IAB), Nü nbe g,
h ps://doi.o g/10.48720/IAB.DP.2514
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/330590
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-sa/4.0/deed.de
IAB-DISCUSSION PAPER
A icles on labou ma ke issues
14|2025 Re i emen Age Re o ms and Wo ke
Subs i u abili y: Implica ions o Employmen o
Olde Wo ke s
Sona Badalyan
ISSN 2195‑2663
Re i emen Age Re o ms and Wo ke
Subs i u abili y: Implica ions o
Employmen o Olde Wo ke s
Sona Badalyan (Ins i u ü A bei sma k ‑ und Be u s o schung (IAB) & CERGE‑EI)
Mi de Reihe „IAB‑Discussion Pape “ will das Fo schungsins i u de Bundesagen u ü
A bei den Dialog mi de ex e nen Wissenscha in ensi ie en. Du ch die asche Ve b ei ung
on Fo schungse gebnissen übe das In e ne soll noch o D ucklegung K i ik ange eg
und Quali ä gesiche we den.
The “IAB‑Discussion Pape ” is published by he esea ch ins i u e o he Ge man Fede al
Employmen Agency in o de o in ensi y he dialogue wi h he scien i ic communi y. The
p omp publica ion o he la es esea ch esul s ia he in e ne in ends o s imula e
c i icism and o ensu e esea ch quali y a an ea ly s age be o e p in ing.
Con en s
1. In oduc ion............................................................................... 6
2. Ins i u ional se ing and concep ual amewo k..................................... 12
2.1. Ins i u ional se ing..................................................................... 13
2.2. Concep ual amewo k and implica ions ............................................ 15
2.2.1. Fi m’s p oblem o employmen decisions .................................. 15
2.2.2. Wage de e mina ion unde Nash ba gaining............................... 17
3. Da a ..................................................................................... 19
3.1. The Sample o In eg a ed Employe ‑Employee Da a .............................. 19
3.2. Sample cons uc ion o analyses .................................................... 20
4. Iden i ica ion............................................................................ 21
4.1. Reg ession discon inui y design ...................................................... 21
4.2. Desc ip i e e idence on he p esence o discon inui y ........................... 24
5. Resul s .................................................................................. 24
5.1. The e ec o he ise in ERA on employmen s a es .............................. 25
5.2. Robus ness and sensi i i y checks o he baseline RDD esul s ................ 26
5.3. The e ec o he ise in ERA on wages.............................................. 28
6. Labo demand mechanisms: eplacemen cos s..................................... 29
6.1. The ole o job‑speci ic skills ......................................................... 29
6.2. The ole o in e nal and ex e nal subs i u abili y .................................. 33
6.3. The e ec o aised ERA on wages by eplacemen cos s ........................ 39
7. Conclusion .............................................................................. 41
Re e ences...................................................................................... 42
Appendix .................................................................................... 48
A1. Appendix igu es ........................................................................ 48
B1. Appendix ables ......................................................................... 63
IAB‑Discussion Pape 14|2025 3
Abs ac
This pape s udies how labo demand ac o s—speci ically wo ke subs i u abili y and
job‑speci ic skills—shape employmen esponses o a ise in he ea ly e i emen age. Using
a eg ession discon inui y design, I exploi a 1999 Ge man e o m ha elimina ed he op ion
o women o e i e a age 60. Be o e he e o m, olde wo ke s could exi olun a ily,
he eby imposing u no e cos s on i ms. A e wa d, i ms we e be e able o e ain less
subs i u able wo ke s o whom u no e cos s a e highe . A he same ime, he loss o
ea ly pension eligibili y educed wo ke s’ ou side op ions, allowing i ms o o e lowe
wages, o en h ough pa ial e i emen .
Zusammen assung
Dieses Papie un e such , wie a bei snach agesei ige Fak o en – insbesonde e die
E se zba kei on A bei sk ä en und be u sspezi ische Fähigkei en – die
Beschä igungs eak ionen au eine Anhebung des ühes möglichen Ren enal e s
beein lussen. Mi hil e eines Reg ession‑Discon inui y‑Designs analysie e ich eine Re o m in
Deu schland im Jah 1999, die die Möglichkei ü F auen abscha e, be ei s mi 60 Jah en
in Ren e zu gehen. Vo de Re o m konn en äl e e Beschä ig e eiwillig aus dem
E we bsleben ausscheiden, was den Un e nehmen Fluk ua ionskos en e u sach e. Nach
de Re o m wa en Be iebe besse in de Lage, schwe e se zba e A bei sk ä e mi höhe en
Aus i skos en zu hal en. Gleichzei ig e schlech e e sich du ch den Weg all des
o gezogenen Ren enzugangs die Ve handlungsposi ion de Beschä ig en, was es den
Un e nehmen e möglich e, nied ige e Löhne du chzuse zen – häu ig in Fo m on
Al e s eilzei .
JEL
H32, H55, J21, J24, J26
Keywo ds
aging, aise in he e i emen age, in e nal labo ma ke s, human capi al, wo ke
subs i u abili y
IAB‑Discussion Pape 14|2025 4

Acknowledgmen s
I hank Dan Black, Wol gang Dau h, Randall File , Š ěpán Ju ajda, And eas Mense, Nikolas
Mi ag, and Paolo Zacchia o hei eedback; Dan Black o in i ing me o he Uni e si y o
Chicago, whe e pa s o his pape we e w i en; Michael Mo i z, Wol gang Dau h, and he
“Regional Labou Ma ke s” Depa men a he Ins i u e o Employmen Resea ch (IAB) o
hei belie in his p ojec ; Philipp om Be ge and Ka ja Wol o help wi h unde s anding he
da a; Thomas Zwick o sha ing a p og amming ile ha compu es pension eligibili y;
Debo ah No áko á o language edi ing. This pape bene i ed om p esen a ions a HUN
REN 2025; i o Ins i u e 2025; EWMES 2024; EALE 2024 & 2023; IAB B own Bag and Regio Flash
Talks 2024; ESPE 2024; IZA Summe School 2024; Du ch Na ional Bank 2024; Young
Economis s Semina (C oa ian Na ional Bank) 2024; SITES 2023; AIEL 2023 & 2022; S uden
Wo kshop a Ha is School o Public Policy a UChicago 2023; BSE Summe School 2022;
Czech Economic Socie y Biennial Con e ence 2022; A menian Economic Associa ion Annual
Mee ing 2022; CERGE‑EI B own Bag 2023, Applied Mic oeconome ics Reading G oup 2022,
DW 2022 & DPW 2020 Semina s. This s udy was suppo ed by Cha les Uni e si y, GAUK
p ojec No. 333221. This pape is pa o a p ojec ha has ecei ed unding om he
Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme unde he Ma ie
Skłodowska‑Cu ie g an ag eemen No. 870245.
IAB‑Discussion Pape 14|2025 5
1 In oduc ion
The dynamics o labo ma ke s a e p o oundly in luenced by he in e play be ween wo ke
subs i u abili y and i m‑speci ic human capi al. The ease wi h which wo ke s can be
eplaced a ec s a ious labo supply decisions, including absences due o empo a y illness
(Hens ik/Rosenq is , 2019), he du a ion o ac ual pa en al lea e in eac ion o ex ension o
pa en al lea e du a ion (Ginja/Ka imi/Xiao, 2023) and inc ease o paid pa en al lea e
eligibili y co e age (Huebene e al., 2024), and labo supply ollowing a cowo ke ’s dea h
(Jäge /Heining, 2022). Wo ke subs i u abili y has also been associa ed wi h wage losses
a e job displacemen (Jacobson/LaLonde/Sulli an, 1993), as wo ke s wi h mo e speci ic
skills, such as hose ied o a pa icula indus y o occupa ion, ace g ea e di icul y inding
compa able jobs in he ex e nal labo ma ke . Howe e , he ole o wo ke subs i u abili y
in he con ex o e i emen , a signi ican d i e o wo k o ce u no e , emains
unde explo ed.
While subs an ial li e a u e examines how s a u o y e i emen age e o ms impac labo
supply (A alay/Ba e , 2015; B inch/Ves ad/Zweimülle , 2015; Geye /Wel eke, 2021;
Hanel/Riphahn, 2012; He næs e al., 2016; Lali e/S aubli, 2015; Lali e/Magesan/S aubli,
2023; Manoli/Webe , 2016; Mas obuoni, 2009; S aubli/Zweimülle , 2013; Ves ad, 2013),
he e is limi ed unde s anding o how labo demand mechanisms, such as job‑speci ic skills
and wo ke subs i u abili y, shape employmen esponses o such e o ms because hese
pape s o en assume ha labo demand is pe ec ly elas ic a he ele an ma gins. In
con as , my pape a gues ha labo demand is no uni o mly elas ic and highligh s he ole
o wo ke subs i u abili y in shaping i ms’ e en ion decisions. This pape aims o b idge
his gap in he e i emen li e a u e by in eg a ing insigh s om s udies on wo ke
subs i u abili y wi h esea ch on employmen eac ions o e i emen e o ms.
Unde s anding his mechanism is c ucial, as i o e s deepe insigh s in o how wo ke
subs i u abili y in luences labo supply adjus men s o e i emen e o ms and he coping
s a egies adop ed by wo ke s and i ms. This challenges he s anda d assump ion o
uni o mly elas ic labo demand and o e s new insigh s in o he incidence and e iciency o
e i emen e o ms.
The seminal s udy by Becke (1962) posi s ha i m‑speci ic human capi al ende s
incumben wo ke s less subs i u able by ex e nal hi es. In he con ex o e o ms ha aise
he e i emen age, his heo y sugges s ha employmen esponses by olde wo ke s may
exhibi subs an ial he e ogenei y based on hei subs i u abili y and he speci ici y o he
human capi al equi ed o hei oles. A pe inen ques ion a ises: When ea ly e i emen
op ions a e cu ailed, do i ms espond uni o mly ac oss wo ke ypes, o do employmen
gains disp opo iona ely acc ue o hose wi h mo e speci ic skills and lowe subs i u abili y?
IAB‑Discussion Pape 14|2025 6
Such di e ences may e lec how i ms and wo ke s coo dina e—depending on hei
u no e cos s—in esponse o ex ended employmen ho izons. The demand o wo ke s
ises due o i m‑ o job‑speci ic human capi al, o challenges in inding sui able
eplacemen s in e nally o ex e nally. Howe e , in he p esence o ou side op ions in he
o m o pensions, i ms may ha e di icul ies e aining such wo ke s. Re o ms aising he
e i emen age could help i ms o e ain such wo ke s.1
Employmen decisions a olde ages a e a ec ed by many ac o s, including heal h, abili y,
income, and lexibili y o con ac s and i ms; hence, in he absence o exogenous d i e s,
such decisions a e likely endogenous a he indi idual le el. Mo eo e , gi en an op ion o
e i e and ecei e a pension, wo ke s may op o exi he wo kplace and ins ead p io i ize
pe sonal bene i s (such as heal h, leisu e ime wi h amily, e c) o e i m ac o s (such as
hei subs i u abili y and cos s o eplacemen ) in deciding o e i e. A e o m ha aises he
e i emen age shi s he employmen dynamics o hose a ec ed. I o e come his
endogenei y challenge by s udying he e ec s o a e o m in Ge many ha abolished he
women’s pa hway o ea ly e i emen by making he s a u o y e i emen ages gende
neu al. This e o m esul ed in a sha p ise o a leas h ee yea s ( om 60 o 63) in he Ea ly
Re i emen Age (ERA), he ea lies age women could begin o claim a pension. This
discon inuous policy change, which impac ed women bo n om 1952 onwa d, p o ides a
na u al expe imen o causally iden i ying he e ec o aising he e i emen age on
employmen and wages using a Reg ession Discon inui y Design (RDD), and explo ing he
ela ionship o wo ke subs i u abili y wi h a la ge labo supply inc ease.
The Ge man labo ma ke , cha ac e ized by subs an ial a ia ion in wo ke subs i u abili y2
and s ong dismissal p o ec ions, o e s a sui able se ing o in es iga ing whe he wo ke s
delay e i emen based on hei skills and subs i u abili y. The a ailabili y o
comp ehensi e Ge man es ablishmen da a, which encompasses en i e wo k o ces and
employmen his o ies, oge he wi h job cell da a (3‑digi occupa ion g oups wi hin he
es ablishmen s), enables analysis o in e nal ma ke s, measu emen o he a ailabili y o
in e nal subs i u es (wo ke s sha ing he same 3‑digi occupa ion), and a s udy o pe sonnel
p ac ices employed by he es ablishmen s.
To examine how employmen esponses o he ise in e i emen age in e ac wi h wo ke
subs i u abili y, I s a by ske ching a simple model o he in e play be ween he e o m ha
aises he age o he op ion o ecei e pensions, u no e cos s, and employmen decisions
1 S ole/Zwiebel (1996a) and S ole/Zwiebel (1996b) p o ided heo e ical discussions o in a‑ i m ba gaining
and i s ela ion o i m‑speci ic human capi al, while Lazea (2009) and Cahuc/Ma que/Wasme (2008)
ex ended he discussion by a guing ha , simila o i m‑speci ic human capi al, he ease wi h which a i m
can ind a sui able eplacemen could a ec he wages o wo ke s. Howe e , ha ing lowe ba gaining powe
a e emo al o he op ion o ecei e pensions, i ms may be in a s onge posi ion han wo ke s.
2 P e ious li e a u e o Ge many has shown ha ic ions in eplacing wo ke s a e impo an (Jäge /Heining,
2022; Huebene e al., 2024).
IAB‑Discussion Pape 14|2025 7
a 60‑62. I also ou line a Nash ba gaining model wi h implica ions o he e ec s o he
e o m and o subs i u abili y on wages condi ional on employmen a 60‑62.
To es hese implica ions empi ically, I i s cons uc se e al p oxies o wo ke
subs i u abili y (and he e o e u no e cos s). Fi s , I examine whe he wo ke s wi h
speci ic skills a e mo e likely o be e ained a olde ages. Speci ic human capi al and
manage ial oles a e key de e minan s o wo ke subs i u abili y, as ex e nal eplacemen s
o hese skills a e o en sca ce (Bake /Gibbs/Holms om, 1994)3. Consis en wi h heo ies
o i m‑ and job‑speci ic human capi al (Becke , 1962), Be heau (2021) shows ha jobs
equi ing eamwo k and aining wi h senio wo ke s a e mo e o en illed in e nally. In he
con ex o e i emen e o m, his sugges s ha es ablishmen s whe e olde wo ke s’
posi ions ely on job‑ o i m‑speci ic human capi al may bene i mos om he ex ended
e en ion o olde wo ke s. Nex , I explo e in e nal (cowo ke s in he same occupa ion) and
ex e nal (po en ial hi es in a commu ing zone o a gi en occupa ion o indus y) labo
ma ke hickness. Acco ding o Topel/Wa d (1992), bo h in e nal and ex e nal labo ma ke s
a ec wo ke s’ li e‑cycle labo ma ke ou comes. In hin labo ma ke s, inding sui able
eplacemen s is mo e challenging, making wo ke u no e cos ly o i ms (Lazea , 1979).
Au oma ion can subs i u e o some ypes o labo , leading o educed employmen and
wages, pa icula ly in economies wi h aging popula ions like Ge many (Acemoglu/Res epo,
2022). Hence, I es whe he he subs i u abili y ma e s beyond he wo ke le el, by
di iding occupa ions by ou ineness, a p oxy o subs i u ion by au oma ion. Finally, I
conside he adabili y o indus ies as ano he dimension o wo ke subs i u abili y. Fi ms
in adable indus ies can eplace wo ke s no only locally bu also by ou sou cing asks
globally, inc easing subs i u abili y (D enik e al., 2023). While cha ac e is ics such as
manage ial s a us o skill speci ici y may e lec bo h i m‑side cos s and wo ke ‑side
p e e ences, I in e p e he e ogenei y in he e o m’s e ec s p ima ily h ough he lens o
i ms’ e en ion incen i es — ha is, he labo demand channel.
My indings con i m he implica ions o he model and indica e ha he e o m inc eased
employmen among women aged 60–62 by 17.3 pe cen age poin s (a 22% inc ease ela i e
o he con ol mean o wo ke s who we e eligible o e i e a 60). These esul s a e obus o
a ia ions in model speci ica ion. To gauge he po en ial scale o he e o m’s impac , I
conduc a back‑o ‑ he‑en elope calcula ion. This ea men e ec would ansla e in o
oughly 540,000 addi ional women emaining employed a ages 60–62 due o he e o m.4
Condi ional on employmen , he wo ke s whose e i emen age ose by he e o m a e less
likely o ba gain o highe wages a ages 60–62, compa ed o hose p e iously eligible o
pension bene i s. The e o m emo ed access o ea ly e i emen , weakening ou side
3 See also Ba el e al. (2014), F ied ich/Hackmann (2021), Jäge /Heining (2022), Ja a el/Pe ko a/Bell (2018)
4 This is a ough calcula ion based on local ea men e ec s o women bo n in 1951–1952, who we e
employed con inuously a 58‑59 yea s old. The es ima e assumes ha he sample is na ionally
ep esen a i e and ha he e ec gene alizes ac oss coho s a ec ed by he e o m. I does no adjus o
composi ional di e ences o coho ends and should be in e p e ed as illus a i e.
IAB‑Discussion Pape 14|2025 8
2.2 Concep ual amewo k and implica ions
Fi ms ope a ing in impe ec labo ma ke s ace ic ions in eplacing expe ienced wo ke s,
pa icula ly hose wi h occupa ion‑ o i m‑speci ic skills. As hese wo ke s app oach
e i emen age, i ms isk p oduc i i y losses and incu hi ing cos s due o u no e . Ea ly
e i emen eligibili y g an s wo ke s conside able au onomy in deciding when o exi he
labo o ce. This pape s udies how a policy e o m ha aised he ea ly e i emen age
(ERA) om 60 o a leas 63 al e s he in e ac ion be ween wo ke subs i u abili y and
e i emen beha io .
The model builds on he idea ha e i emen is no only a wo ke ’s choice, bu also e lec s
he ela i e ba gaining powe o wo ke s and i ms. When ea ly e i emen is an op ion,
wo ke s wi h aluable skills may le e age his as a ba gaining chip in wage nego ia ions.
When he op ion is emo ed, i ms can e ain e en aluable wo ke s wi hou aising wages.
To unde s and how i ms espond o a ise in ea ly e i emen age, I i s de elop a s a ic
model in which he i m’s decision o e ain a wo ke depends explici ly on he wo ke ’s
subs i u abili y and he policy en i onmen . I hen ex end he amewo k wi h a Nash
ba gaining model, allowing wages o be endogenously de e mined. This yields es able
implica ions condi ional on employmen .10
2.2.1. Fi m's p oblem o employmen decisions
Se up. Conside a i m employing wo ke i, who is app oaching e i emen age. Con inued
employmen a ages 60–62 depends on whe he he ma ch be ween he wo ke and he
i m emains iable. I model his using a la en e en ion condi ion, in which bo h he i m
and he wo ke mus bene i om con inued employmen . While he i m’s willingness o
accommoda e employmen e lec s he economic alue o he ma ch, he wo ke ’s ou side
op ion plays a key ole in he join decision. I do no model ac i e dismissal, consis en wi h
s ong employmen p o ec ions. Ins ead, I in e p e “ e en ion” as he ma ch con inuing
when bo h pa ies ind i p e e able o sepa a ion. The e o m emo es a key olun a y exi
channel (ea ly e i emen ), ex ending employmen among olde wo ke s, especially hose
wi h high speci ici y.
10 The heo e ical amewo k p esen ed in his sec ion builds on Nash ba gaining models o labo ma ke
ic ions (e.g., Pissa ides (2000)), adap ing hem o e i emen con ex s by inco po a ing ou side op ions
shaped by policy. I also d aws on Acemoglu/Pischke (1999) in he implica ions o i m‑speci ic skills o
wage se ing and u no e , and om G ube /Wise (2008) he esponsi eness o e i emen o ins i u ional
incen i es. Las ly, he in e ac ion be ween subs i u abili y and ax incidence in de e mining he incidence o
adjus men cos s is concep ually linked o G ube (1997), and is applied he e o changes in e i emen age.
IAB‑Discussion Pape 14|2025 15

⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎧ ⎫
⎪ ⎪
⎨ ⎬
max πi = di · (y − o(Ri, si)) +(1 − di) · (−c(si)) (1)
di∈{0,1} | {z } | {z }
⎪ ⎪
⎩ su plus i cos i ⎭
ma ch con inues ma ch ends
(∂c(si) > 0)
∂si
whe e:
yi is he wo ke ’s ou pu . I abs ac om he e ogenei y in ou pu ac oss wo ke s, as
subs i u able wo ke s may be ei he mo e o less p oduc i e depending on job i and skill
speci ici y.11
The model uses a single speci ici y pa ame e si o cap u e employe ‑side u no e cos s.
These cos s inc ease when wo ke s a e mo e di icul o eplace (due o specialized
knowledge o ask‑speci ic skills) and when hey ha e limi ed ou side op ions (due o hin
ex e nal ma ke s o hei skills). In his sense, speci ici y si ep esen s a educed‑ o m
measu e encompassing bo h skill speci ici y and subs i u abili y.
o(Ri, si) is he ou side op ion, shaped by he policy e o m Ri and wo ke speci ici y si. I is
dec easing in Ri (∂o(Ri,si) < 0)
∂Ri
because pension eligibili y is delayed pos ‑ e o m, and
dec easing in si because wo ke s wi h speci ic skills ace hinne ex e nal labo ma ke s,
especially a e age 60 (∂o(Ri,si) < 0)
∂si . Such speci ici y could, o example, dec ease he
likelihood o lea ing he social insu ance o o he pa hways han e i emen (mo e o
ano he coun y, s a sel ‑employmen , e c.).12
c(si) deno es he eplacemen cos o an employee, inc easing in speci ici y, and e lec ing
dismissal, se e ance, and o he compensa ion paymen s, as well as hi ing, aining, and
p oduc i i y amp‑up cos s ha ise wi h he deg ee o human capi al speci ici y
.
Solu ion. The ma ch con inues i he join su plus om con inuing exceeds he cos o
sepa a ion:
y − o(Ri, si) + c(si) > 0 (2)
In e p e a ion. While he equa ion is modeled as a i m‑side op imiza ion p oblem, he
11 I hank Wol gang Dau h o his discussion.
12 Such a gumen is also in line wi h he inding o li e a u e on displaced wo ke s: hose unemployed who
swi ch o ano he indus y o occupa ion expe ience much la ge declines in ea nings (e.g. Neal (1995);
Addison/Po ugal (1989)).
IAB‑Discussion Pape 14|2025 16
∂ do(Ri, si) dc(si)
(y − o(Ri, si) + c(si)) = − + > 0 (4)
∂si dsi dsi
| {z } | {z }
<0 >0
∂ do(Ri, si)
(y − o(Ri, si) + c(si)) = − > 0 (3)
∂Ri dRi
| {z }
<0
in e p e a ion e lec s a join ag eemen be ween he i m and he wo ke : con inued
employmen occu s when bo h bene i ela i e o sepa a ion. The e o m shi s his
condi ion by emo ing ea ly e i emen as a allback, hus al e ing he ou side op ion
o(Ri, si) and inc easing he likelihood o ma ch con inua ion, especially o wo ke s wi h
high speci ici y.
Implica ion 1: A highe ERA ises employmen o olde wo ke s.
Delaying pension eligibili y educes o(Ri, si), making con inued employmen mo e
a ac i e o he i m and less a oidable o he wo ke , he eby inc easing employmen .
Implica ion 2: Wo ke s wi h highe speci ici y (si) a e mo e likely o emain employed.
Less subs i u able wo ke s a e mo e likely o emain in employmen due o bo h weake
ou side op ions and highe eplacemen cos s. As a esul , he join su plus o con inued
employmen is la ge , sus aining he ma ch.
2.2.2. Wage de e mina ion unde Nash ba gaining
Condi ional on e en ion (di = 1), he i m and he wo ke ba gain o e he wage wi based
on he o al su plus gene a ed by employmen :
Si = y − o(Ri, si) (5)
Wi h wo ke ba gaining powe β ∈ (0, 1), he Nash wage spli s he su plus be ween he
wo ke and he i m. The wage hus depends on bo h he wo ke ’s p oduc i i y y and hei
ou side op ion o(Ri, si). The gene al o m o he Nash‑ba gained wage is:
IAB‑Discussion Pape 14|2025 17
wi = β · y + (1 − β) · o(Ri, si) (6)
|{z} | {z }
p oduc i i y‑based ewa d ou side‑op ion allback
∂wi do(Ri, si)
= (1 − β) · < 0 (7)
∂Ri dRi
| {z }
<0
In ui ion. The wo ke ’s wage is a weigh ed a e age o wha hey con ibu e o he i m’s
ou pu ( h ough y) and wha hey could ea n elsewhe e ( ia o(Ri, si)). Wo ke s wi h high
p oduc i i y na u ally command highe wages, all else equal. Howe e , hei ou side
op ion—such as e i emen income o al e na i e employmen —also de e mines hei
ba gaining posi ion. I hei allback op ion weakens, he i m can o e a lowe wage e en i
he wo ke is p oduc i e.
To unde s and how wages change due o he e o m and di e ences in subs i u abili y, I
examine how wi esponds o changes in Ri and si.
Implica ion 3: The e o m lowe s wages ia weake ou side op ions.
When he policy aises he ea ly e i emen age (i.e., inc eases Ri), he ou side op ion
o(Ri, si) declines. This educes he wo ke ’s allback posi ion in wage nego ia ions, shi ing
su plus owa d he i m. The wage alls e en hough he wo ke emains employed. This
mechanism is s onge when he wo ke has low ba gaining powe β, and when he
educ ion in ou side op ions is la ge.
Implica ion 4: The e ec o speci ici y on wages is ambiguous.
∂wi dy(si) do(Ri, si)
= β · + (1 − β) · (8)
∂si dsi dsi
Highe speci ici y si a ec s bo h p oduc i i y and ou side op ions. I mo e speci ic wo ke s
a e mo e p oduc i e (dy(si) > 0
dsi ), hen wages may inc ease h ough he i s e m. Howe e ,
speci ici y also educes ou side op ions (do(Ri,si) < 0
dsi ), which lowe s wages h ough he
second e m. I ou side op ions de e io a e as e han p oduc i i y imp o es, o i he
wo ke has low ba gaining powe , he o e all e ec on wages may be nega i e.
IAB‑Discussion Pape 14|2025 18
In ui ion. E en i he wo ke is aluable o he i m (due o di icul ‑ o‑ eplace skills), he
i m may exploi hei lack o ex e nal al e na i es. The e o m ampli ies his asymme y by
emo ing ea ly e i emen as a iable allback, especially o wo ke s in hin ex e nal labo
ma ke s (e.g., manage s, specialis s). This is a o m o monopsony powe , whe e he
employe ’s abili y o se wages below ma ginal p oduc is s eng hened by he wo ke ’s
limi ed exi op ions.
Summa y. Implica ions h ee and ou join ly imply ha he wage esponse o he e o m
depends on he in e ac ion be ween subs i u abili y and ba gaining ic ions. Fo wo ke s
wi h low speci ici y (who a e easy o eplace), wages all mos ly due o he loss o e i emen
op ions. Fo wo ke s wi h high speci ici y, he s o y is mo e nuanced: while hei
p oduc i i y makes hem cos ly o eplace (inc easing employmen ), hei weakened
allback posi ion gi es he i m he le e age o supp ess wages. Thus, employmen may ise
while wages all o s agna e, despi e high skill speci ici y, due o inc eased employe
monopsony powe pos ‑ e o m.
3 Da a
This sec ion consis s o wo pa s. Fi s , I desc ibe he da a I u ilize, i s sampling p ocedu e,
and i s sui abili y o my esea ch ques ion. Second, I desc ibe how I cons uc ed my sample,
he easoning behind each es ic ion, and he esul ing sample size.
3.1 The Sample o In eg a ed Employe -Employee Da a
I use he Sample o In eg a ed Employe ‑Employee Da a (SIEED7518), a andom 1.5%
sample o all es ablishmen s in Ge many. The es ablishmen iden i ie s a e ixed by
indus y, owne ship, and loca ion a he municipali y le el; hence, an es ablishmen is no
equi alen o a i m in all cases. Ne e heless, I use he e ms i ms and es ablishmen s
in e changeably. Employe s a e obliged o epo da a on all o hei employees subjec o
social secu i y con ibu ions. Sel ‑employed and ci il se an s a e no co e ed by he da a.
A he end o each yea , employe s epo he s a and end da e o employmen , wages, and
o he occupa ional, educa ional, and demog aphic indica o s o all o hei wo ke s.
Typically, he da a is a snapsho o he employmen s a e as o June 30 h o each yea .
Employe s a e also obliged o epo changes in employmen con ac s.13
13 One o he da a limi a ions is he lack o wo king hou s; hence, I am limi ed o he analyses o only he
ex ensi e ma gin o employmen .
IAB‑Discussion Pape 14|2025 19
Fo each o hese es ablishmen s, he en i e employmen biog aphies o all employees a e
included o e he obse a ion pe iod 1975‑2018 o Wes Ge many and 1992‑2018 o Eas
Ge many. Hence, he da a also include he es ablishmen s ha did no cons i u e he
andom 1.5% o he es ablishmen s o iginally sampled, in case he wo ke s om he
es ablishmen s o iginally sampled we e e e employed elsewhe e. Obse ing he en i e
wo k o ce o he sampled es ablishmen s is c i ical o my analyses, because I s udy
subs i u abili y mechanisms behind employmen eac ions o he aise in e i emen age,
which equi es obse ing all cowo ke s o a gi en es ablishmen .
Schmid lein/Se h/Vom Be ge (2020) desc ibe he da a sampling in mo e de ail.
3.2 Sample cons uc ion o analyses
To cons uc he inal sample o my analysis, I keep only women bo n in 1951, he con ol
g oup, i.e., women who we e po en ially eligible o w o he women’s pa hway o ea ly
e i emen , i hey accumula ed enough yea s o social secu i y con ibu ions in la e li e;
and 1952, he ea men g oup, i.e., women who expe ienced he ise in he women’s ERA. I
d op women who we e e e employed as mine s and sailo s ( o cla i y) because hei
e i emen ules di e om hose in o he occupa ions.14
To add ess he issue o pa allel spells in he da a, which is possible, o example, due o dual
ea ne s (employed a se e al es ablishmen s simul aneously), I keep he spells in he
andomly selec ed 1.5% es ablishmen s. I bo h spells come om andomly sampled
es ablishmen s, I keep he spells whe e he wo ke accumula ed mo e enu e. In cases
whe e he employee wo ks in wo andomly selec ed es ablishmen s and has accumula ed
an equal amoun o enu e in each o hem, I keep he job wi h he highes wage. D opping
pa allel spells allows me o cons uc Panel da a and s udy he i m mechanisms o only
he es ablishmen s o which he dual wo ke s a e mo e a ached.
The inal da a consis s o pe son‑age en ies (in age‑mon h), whe e I obse e women om
he age o 42 (age‑mon h 504) un il 66 (age‑mon h 792). The choice o his ime ame is
d i en by he ac ha he i s a ec ed coho was 47 yea s old a he ime o he e o m
announcemen in 1999, and in some o my analyses I wan o obse e employmen (1)
be o e he e o m announcemen , (2) be ween he e o m announcemen and i s inac ion a
14 The seminal wo k by Geye /Wel eke (2021) on labo supply esponses o he 1999 e o m makes a es ic ion
o keeping only women who a e eligible o he women’s pa hway o e i emen a he age o 60. I make
es ic ions ha p oxy o eligibili y, ollowing Lo enz e al. (2018). I do no explici ly make sample
es ic ions ha keep he women eligible o he women’s pa hway (e.g., 15 yea s o con ibu ions in o al
and en yea s a e 40 yea s old, e c), because I do no obse e he unemploymen spells ha also
con ibu e o he con ibu ion yea s. Because unemploymen spells s ill coun owa ds he con ibu ions o
social secu i y, no making his es ic ion esul s in smalle ea men e ec s in my sample, compa ed o
ha o Geye /Wel eke (2021)
IAB‑Discussion Pape 14|2025 20

60, (3) and wo ke s who con inue wo king beyond bo h he ERA (60 o a leas 63) and NRA
(65 o 65.5). Fi s , s udying employmen be o e he e o m announcemen shows whe he
he ea men and con ol g oups had di e en labo supply equencies be o e he e o m
announcemen . Second, s udying employmen be ween 47‑60 can show whe he he ise in
ERA leads o di e en employmen choices du ing middle age, in expec a ion o a longe
employmen pe iod. Finally, s udying he e ec s beyond he new ERA shows how he e ec
o aising ERA also spills o e o pos ‑ERA employmen , which could show indi ec
employmen e ec s beyond he age a ge ed by he e o m, u he inc easing i s
e ec i eness in keeping wo ke s in employmen longe .
I keep wo ke s who a e con inuously (in each age‑mon h) employed a 58 and 59. To make
such es ic ion plausible, I ha e o assume ha he employmen a 58‑59 is no is
una ec ed by he e o m. Geye /Wel eke (2021) show ha he e a e no employmen e ec s
be o e he age o 60, he e o e, such es ic ion is no likely o lead o a selec ion bias.
Because mos o he main he e ogenei y a iables a e cons uc ed a he es ablishmen
le el, his es ic ion helps me o ob ain a sample o wo ke s wi h su icien a achmen o
hei es ablishmen s. The inal da a consis s o 32,770 wo ke s, and 9,036,582 wo ke ‑age
mon hs (Table B1 eco ds he numbe o wo ke s a e each es ic ion). Ou o hese
wo ke s, 15,640 a e in he con ol g oup (bo n in 1951), and 17,130 a e in he ea men
g oup (bo n in 1952).
4 Iden i ica ion
Fi s , I desc ibe he iden i ica ion s a egy based on e o m discon inui y in bi h da es, and
hen I p o ide some desc ip i e esul s ha con i m he p esence o discon inui y in he
da a.
4.1 Reg ession discon inui y design
I ollow Geye /Wel eke (2021) o locally iden i y he e ec o he e o m ha aised he ERA
on employmen , τm, in an RDD amewo k15:
15 The e a e se e al di e ences om he iden i ica ion in Geye /Wel eke (2021). Fi s , I do no con ol o he
p esence o child en in my RDD eg ession as I do no obse e such a iables in he da a. Second, because
he mos ecen yea obse ed in my da a is 2018, he da a allow me o pool all he age mon hs
co esponding o 60‑62 yea s o age in he baseline eg ession and beyond 63 in he supplemen a y analyzes,
while Geye /Wel eke (2021) pooled only 60‑62 due o hei igh ‑censo ed da a in 2016. Finally, I use he
mean squa e‑based op imal bandwid h, while hey use a 12‑mon h ad‑hoc bandwid h selec ion p ocedu e.
IAB‑Discussion Pape 14|2025 21
yim = αm + τm
1
{b
i ≥ b
∗
}+ (9)
′
+ β0m
1
{b
i < b
∗
}(bi − b
∗
) + β1m
1
{b
i ≥ b
∗
}(bi − b
∗
) + Xiβm + ϵim
whe e yim
‑ is employmen s a e, eco ded o each woman i a e e y age‑mon hs m; bi is
he bi h coho o he indi idual i;
1
{b
i ≥ b∗} is an indica o showing ha i was bo n a e
he cu o b∗ (Janua y 1952), i.e., expe ienced he ise in he ERA ( ea men g oup); while
1
b
i < b∗}{ includes he indi iduals who a e below he cu o (con ol g oup). I use a local
linea eg ession, and by in e ac ing he unning a iable (bi − b∗) wi h he ea men
indica o , I allow o di e en slopes in ea men and con ol g oups. Figu e 1 shows ha a
linea end in he unning a iable is a plausible assump ion, and he e is a clea
discon inui y ha is unlikely o be a ibu ed o a w ong unc ional o m o polynomials. To
compu e he RDD es ima es, I use a iangula ke nel unc ion and he op imal bandwid h
choice based on mean squa e e o (Imbens/Kalyana aman, 2012). As a esul , I calcula e
he bias‑co ec ed RDD es ima es wi h a obus a iance es ima o .
I also con ol o calenda mon h, a dummy o Wes e n Ge man esidence, wages a he age
o 46, and wo educa ion ca ego ies (ou o 3), because p e ious li e a u e con i ms ha
educa ion is an impo an de e minan o employmen a an olde age (Geye e al., 2022). I
clus e he s anda d e o s a he bi h mon h le el o accoun o he po en ial co ela ion
o s anda d e o s ϵim o he women belonging o he same bi h coho .16 In obus ness
and sensi i i y checks, I e‑ un he eg essions, al e ing all he speci ica ion pa ame e s‑ he
p ocedu es o es ima ing he pa ame e s and co a iance ma ices, polynomial o de ,
ke nel weigh s, bandwid h choice, included co a ia es, and clus e ing le el.
The baseline eg essions pool he 60‑62 age (720‑756 age mon hs) oge he , because his is
he age ame ha was a ec ed by he ERA e o m. This iden i ica ion esul s in a local
a e age ea men e ec o highe ERA on employmen ou comes a ages 60‑62 (coe icien
τm in equa ion Equa ion 9).17
Iden i ica ion assump ions. This iden i ica ion elies on wo main assump ions.
(1) Smoo hness in densi y. This assump ion equi es con inui y o he unning a iable (bi h
coho ) a ound he cu o , which elimina es he possibili y o s a egic bunching
(manipula ion o he ea men s a us) a he cu o . This assump ion holds by cons uc ion
16 Clus e ing a he le el o bi h da es aligns wi h li e a u e sugges ing clus e ing he s anda d e o s a he
ea men le el.
17 Because I canno claim ha all he women included in my sample we e eligible o women’s pa hway o
ea ly e i emen , he coe icien could also cap u e he In en ion‑ o‑T ea (ITT) e ec . Howe e , Lo enz e al.
(2018) show which sample es ic ions a e likely o lead o eligibili y impu a ions, and because mos o my
es ic ions ma ch hei p oposed es ic ions, my sample likely cap u es mos o women eligible o he
women’s pa hway o ea ly e i emen .
IAB‑Discussion Pape 14|2025 22
because i is impossible o change one’s own bi h da e.18 Ne e heless, in he sensi i i y
es s, I e‑es ima e he main eg essions by omi ing he obse a ions close o he cu o and
con i m he obus ness o he esul s.
(2) Smoo hness in co a ia es. This assump ion equi es con inui y o he dis ibu ion o he
obse ed and unobse ed a iables a ound he h eshold, showing ha he assignmen o
he ea men a ound he cu o is as good as andom. Table B2 shows ha he e is no
sizeable signi ican discon inui y in p e‑de e mined a iables. In pa icula , I choose a
a iable showing whe he a woman has Wes e n o igin (p oxied by he place o li ing
acco ding o he i s biog aphical spell) and na ionali y, as hese a iables a e ixed o e
ime and hence a e p e‑de e mined.
Main ou come a iables. In e ms o ou come a iables, a each age mon h, I c ea e h ee
mu ually exclusi e main labo ma ke ca ego ies ‑ employmen , nonemploymen , and
e i emen . I u he disen angle he employmen in o h ee g oups‑ employees liable o
social secu i y, ma ginal pa ‑ ime employmen , and pa ial e i emen . Nonemploymen
s ands o a gap in he employmen age‑mon h spells. I p oxy e i emen wi h he las labo
ma ke ac i i y o a wo ke . Figu e A2 displays he e olu ion o he h ee main employmen
s a es o e age by ea men s a us, i.e., he gap in employmen and e i emen s a uses a
60‑62.
In addi ion o hese employmen s a e ca ego ies, I also de ine wages, because I am also
in e es ed in wages condi ional on employmen .19 Wages a e c ea ed a he de ailed
mon hly le el, and a e non‑ze o only i he wo ke is employed.
E ec he e ogenei y. To s udy he mechanisms behind hese e ec s, I pe o m subsample
analysis using se e al ca ego ies o a iables, which show u no e cos s associa ed wi h
e i emen in he nex sec ion. Because he esea ch ques ion ela es o he labo demand
ac o s in luencing employmen a ages 60‑62, I de ine hese a iables a he age o 58, jus
be o e he p e‑ e o m e i emen age o 60.
18 One could a gue ha he e o m coho s could be chosen by policy‑make s in a way ha iola es he
assump ion, o example, by he coho o baby‑boome s, e c. Howe e , because I compa e coho s bo n
a ound he cu o , and he cu o does no appea in any o he e o ms, policies, o cha ac e is ics (bo h o
hese coho s a e ypically classi ied in he baby‑boome gene a ion) ha would make he 1951 coho
di e en om he 1952 coho , he e is no eason o belie e ha he assump ion is likely o be iola ed.
19 Al hough wages a e op‑coded in he social secu i y da a, his da a ea u e is unlikely o cons i u e an issue
o he analyses as women a e less likely o c oss he h eshold o wage censo ing.
IAB‑Discussion Pape 14|2025 23
4.2 Desc ip i e e idence on he p esence o discon inui y
The abolishmen o women’s pa hway o ea ly e i emen led o a la ge inc ease in
employmen a es a 60‑62, as shown in he igh Panel o Figu e 1. While o e all he e is an
upwa d‑sloping employmen end a 60‑62 o e he bi h coho s, he e is also a clea
discon inui y a ound he 1952 coho . Only a ound 75% o women bo n in 1946‑1951 we e
employed a 60‑6220. Howe e , he employmen a e jumped o app oxima ely 90 pe cen
s a ing wi h he 1952 coho , he e o m cu o .
Figu e A3 ex ends he analyses o display employmen a es by ea men s a us a all age
mon hs (co esponding o he ages be ween 42 and 66), and con i ms he p esence o a
discon inui y in employmen a es a 60‑62 (due o he 1999 e o m ha I s udy) and o a
smalle magni ude o discon inui y a he ages 65‑65.5 (due o he 2007 e o m). Es ima ing
he ea men e ec s o he 2007 e o m is beyond he scope o his pape ; hence, in he
nex sec ion, I causally quan i y he la ges employmen discon inui y ha happens due o
he 1999 e o m, i.e., a 60‑62.
5 Resul s
In his sec ion, I i s ocus on he e ec o he 1999 e o m on employmen , con i ming he
esul s o p io s udies on his e o m (Geye /Wel eke, 2021). I show he e ec s o e i emen
on employmen ajec o ies be o e s udying he labo demand mechanisms o
employmen , because I wan o p o ide a gene al pic u e o he labo supply beha io
o e all be o e zooming in on he o al employmen mechanisms.
Based on he heo e ical amewo k, I expec ha he ise in he ERA should ex end
employmen among a ec ed wo ke s, pa icula ly hose whose exi would impose high
u no e cos s on i ms. These cos s a e likely highe o wo ke s wi h speci ic skills o hose
employed in occupa ions wi h limi ed in e nal o ex e nal subs i u es. The e o e, I expec
he employmen e ec s o he e o m o be s onge o such wo ke s. On wages, he model
p edic s ambiguous e ec s depending on wo ke s’ ou side op ions and eplacemen
di icul y: lowe ba gaining powe due o he loss o pension eligibili y may lead o wage
dec eases, while high eplacemen cos s o speci ic o non‑subs i u able wo ke s could
esul in wage p emiums o incen i ize e en ion.
20 This con ol mean is highe han ha in exis ing li e a u e s udying he labo supply esponse o his e o m
(Geye /Wel eke, 2021), likely because he sampling o SIEED and my sample es ic ion (employmen a he
ages 58‑59) esul s in a sample o wo ke s who a e mo e a ached o he labo o ce.
IAB‑Discussion Pape 14|2025 24
Figu e 3.: The e ec o he ise in ERA on employmen a ages 60‑62 by e u n o expe ience in a
gi en occupa ion and occupa ional hie a chy le el
No es: Coe icien plo s o RDD eg essions a ound he 1952 cu o . Fo compu ing he RDD es ima es, I use
local linea eg essions, a iangula ke nel unc ion, and mean squa e e o ‑based op imal bandwid h choice. I
con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and he highes educa ion.
The subsample analysis in he le Panel is pe o med by he human capi al speci ici y o occupa ion. The igh
Panel s ands o manage ial s a us. The e ical lines indica e 95% con idence in e als based on obus s an‑
da d e o s clus e ed a he bi h mon h le el. The con ol means (on he x‑axis) show he employmen sha e a
he ages o 60‑62 in he co esponding subsample o e he con ol g oup (bo n in 1951). A co esponding able
wi h mo e de ails can be ound in Table B12.
Manage ial s a us. Manage ial occupa ions o en en ail a highe deg ee o i m‑speci ic and
occupa ion‑speci ic human capi al, due o hei eliance on accumula ed ins i u ional
knowledge, leade ship skills, and ela ionship‑speci ic in es men s wi hin he i m.
Manage s a e ypically mo e di icul o eplace han a e non‑manage s, pa icula ly a olde
ages when expe ience and i m‑speci ic knowledge peak. The e o e, dis inguishing
be ween manage s and non‑manage s o e s a meaning ul way o cap u e he e ogenei y in
u no e cos s and he alue o wo ke e en ion ollowing a ise in he ea ly e i emen age.
E en i wo wo ke s ha e he same e u ns o expe ience, manage ial oles may imply ex a
i m‑speci ic alue.
In a ela ed s udy, Jäge /Heining (2022) ind ha he dea h o a manage o a wo ke in a
specialized occupa ion esul s in mo e nega i e e ec s on he cowo ke s in o he
occupa ions. In my se ing, i a wo ke is a manage , she likely has many cowo ke s unde
he hie a chy, communica es wi h hem mo e, and has mo e in o ma ion, making he less
subs i u able, and hus making he ex ension o he wo king li e mo e aluable. I c ea e a
a iable showing manage ial o supe iso y s a us based on he las wo digi s o he 5‑digi
occupa ions. I pool he supe iso s and manage s in o he dummy a iable manage .
28
Depending on occupa ion ype, some occupa ional hie a chies ha e manage s, while o he s ha e
supe iso s as he highes occupa ion le el in a hie a chy. I hank Philipp om Be ge o he help wi h he
da a.
IAB‑Discussion Pape 14|2025
28
31

The igh Panel o Figu e 3 shows ha wo ke s in manage ial posi ions a e signi ican ly
mo e likely o emain employed a olde ages in eac ion o he e o m. The wo ke s in
manage ial posi ions ex end hei employmen by 43.1 p.p. (55.3% ela i e o he con ol
mean‑ he manage s whose e i emen age was no al e ed by he e o m), while he
non‑manage s aise hei e i emen ages by 17.1 p.p. (22.1% inc ease ela i e o he con ol
mean). The di e ence in poin es ima es (26 p.p.) sugges s ha he employmen esponse
o he ea men is subs an ially la ge o wo ke s pe o ming manage ial occupa ions.
Al e na i e measu es o skills and speci ici y. To es whe he he esul s p esen ed abo e
a e sensi i e o app oxima ing wo ke skills, I explo e al e na i e p oxies o wo ke skill
speci ici y. The baseline analysis elies on hie a chical job posi ions as indica o s o
skill‑speci ic oles. As an al e na i e, I use an occupa ional classi ica ion by Bloss eld (1985).
This classi ica ion g oups occupa ions in o en ca ego ies and shows he occupa ional spli
by equi ed skills‑ simple s p o essional.29 Figu e A6 shows ha ac oss all occupa ional
g oups, wo ke s in skilled (i.e., p o essional) ca ego ies exhibi g ea e employmen gains
a e he e o m han hose in co esponding simple oles. Manage s and p o essionals a e
pa icula ly likely o emain employed longe , ein o cing he idea ha skills and job
speci ici y d i e e en ion. Al hough one migh suspec ha his is d i en by longe job
enu e, Figu e A7 shows ha employmen gains do no inc ease mono onically wi h enu e.
This sugges s ha hie a chical posi ion cap u es mo e han enu e alone.30
I u he explo e whe he employmen esponses di e ac oss occupa ional ask ypes,
o e ing ano he dimension o wo ke subs i u abili y. Following Dengle /Ma hes/Paulus
(2014), I ca ego ize jobs along wo dimensions: (i) skill con en —analy ical, in e ac i e,
cogni i e, o manual asks—and (ii) ou ineness— ou ine s non‑ ou ine.31 High‑skilled
wo ke s ypically pe o m analy ical o in e ac i e asks, while low‑skilled wo ke s pe o m
manual non‑ ou ine asks. Figu e A14 shows ha wo ke s in high‑skill ask occupa ions
expe ience he la ges pos ‑ e o m employmen ex ension. In con as , wo ke s in ou ine
occupa ions—o en mo e eplaceable by au oma ion—do no exhibi sys ema ically
di e en employmen esponses. This sugges s ha , in he s udied pe iod, ask ou ineness
and he po en ial o au oma ion play a lesse ole in d i ing employmen e ec s han
o e all skill speci ici y. While au oma ion may become a mo e ele an channel in he
u u e, he e idence he e poin s p ima ily o skill‑ ela ed subs i u abili y as he key
mechanism.
29 I use he codes om ma e ial published by Schmiede / on Wach e /Bende (2016) o implemen his
classi ica ion. Educa ion le el is no a sui able candida e o skill di e en ia ion in his con ex , as i is
di ec ly con olled o in he baseline speci ica ion due o ins i u ional easons (chap e 2 and chap e 4).
The esul s by educa ion le el a e shown in Panel C o Table B22 and exhibi no meaning ul di e ences.
30 Tenu e is an impe ec p oxy o skills in his con ex because eligibili y o e i emen a age 63 depends on
enu e. Thus, i s use con la es eligibili y ules wi h subs i u abili y.
31 This classi ica ion is ma ched o my main da a using he 3‑digi occupa ion iden i ie . Task ypes include
analy ical non‑ ou ine, in e ac i e non‑ ou ine, cogni i e ou ine, manual ou ine, and manual non‑ ou ine.
IAB‑Discussion Pape 14|2025 32
6.2 The ole o in e nal and ex e nal subs i u abili y
The nex g oup o a iables showing he u no e cos s and subs i u abili y o wo ke s is
based on he ma ke s‑ in e nal (by a ailabili y o cowo ke s in he same job cell as an olde
woman) and ex e nal (po en ial hi es in he local labo ma ke ). The main mo i a ion o
s udying in e nal labo ma ke hickness is ha he sca ce he job pe o med is, he mo e
di icul i is o he employe o eplace po en ial e i ees wi h cowo ke s, hus leading o
highe employmen esponses o he e i emen e o m. In e nal subs i u abili y is
pa icula ly impo an , as in e nal wo ke s a e impe ec subs i u es o ex e nal wo ke s
(Jäge /Heining, 2022); hence, o en he in e nal subs i u es weigh mo e han he ex e nal
subs i u es. When ewe wo ke s a e wo king in he speci ic occupa ion o an olde woman
in he commu ing zone, he less subs i u able such a woman is. Simila ly, when ewe
wo ke s a e wo king in he speci ic indus y o an es ablishmen in local labo ma ke s, he
less subs i u able he olde women o such es ablishmen s a e by ex e nal hi es.
A ailabili y o in e nal subs i u es.
To cap u e in e nal subs i u abili y, I use he numbe o a ailable cowo ke s in he same
3‑digi occupa ion as women bo n a ound he e o m cu o . I coun only wo ke s in
employmen posi ions subjec o social secu i y. Following Huebene e al. (2024), I de ine
h ee ca ego ies o such a iables by he a ailabili y o cowo ke s in he same 3‑digi
occupa ion as he a ec ed women: 0, 1‑4, and 5 o mo e in e nal subs i u es. I pe o m he
analyses o es ablishmen s wi h ewe han 100 wo ke s, as he le els o subs i u abili y
will be less dependen on es ablishmen size (such es ic ion also closely ollows Huebene
e al. (2024) de ini ions).
The le Panel o Figu e 4 shows ha when he e a e no cowo ke s who pe o m he same
job as he olde wo ke s, he olde wo ke s a e mo e likely o emain employed a 60‑62
ollowing a e i emen e o m. The g oup wi h mo e han i e subs i u es has signi ican ly
lowe employmen esponses han hose wi h 0 cowo ke s in he gi en job cell. Wo ke s
who ha e no in e nal eplacemen s espond o he e o m by ex ending hei employmen
by 26.8 p.p. (35% inc ease ela i e o he con ol mean o wo ke s who we e allowed o
e i e a 60 and we e employed in non‑subs i u able es ablishmen s). While he e ec s a e
insigni ican o he g oup o wo ke s who ha e be ween one and ou cowo ke s, he
wo ke s who ha e mo e han i e cowo ke s in he same occupa ion ex end hei
employmen by 6.5 p.p. (7.9% inc ease ela i e o he con ol mean). The di e ence in poin
es ima es (20.3 p.p.) sugges s ha he employmen esponse o he e o m ha aised he
ERA is subs an ially la ge o wo ke s who ha e no in e nal subs i u es, ela i e o hose
IAB‑Discussion Pape 14|2025 33
who ha e a leas i e in e nal subs i u es, in line wi h he p edic ion ha i ms e ain
wo ke s who a e mo e di icul o eplace.32
Figu e 4.: Subsample analyses o he e ec o he ise in ERA on employmen a ages 60‑62 by num‑
be o in e nal and ex e nal subs i u es o he gi en occupa ion
No es: Coe icien plo s om RDD eg essions a ound he Janua y 1952 cu o . The es ima es a e ob ained us‑
ing local linea eg essions wi h i s ‑o de polynomials, a iangula ke nel, and mean squa e e o –op imal
bandwid h selec ion. Con ols include calenda mon h o bi h, Wes e n Ge man esidence, wages a age 46,
and educa ion. Subsample analyses a e conduc ed by in e nal subs i u abili y in he le Panel and ex e nal
subs i u abili y in he igh Panel. In e nal subs i u abili y is measu ed by he numbe o cowo ke s in he same
3‑digi occupa ion as he old wo ke , es ic ing he sample o es ablishmen s wi h ewe han 100 wo ke s. The
igh Panel shows ex e nal labo ma ke hickness (ELMT), based on he commu ing zone a mos hal as concen‑
a ed in a gi en occupa ion ela i e o he coun y‑le el (ELMT < 0.5), o a leas hal as concen a ed bu
less concen a ed han he coun y‑le el (0.5 < ELMT < 1), and a leas as concen a ed as he coun y‑le el
concen a ion (ELMT > 1). Ve ical lines ep esen 95% con idence in e als based on obus s anda d e o s
clus e ed a he bi h‑mon h le el. Con ol means (on he x‑axis) e e o he a e age employmen a e a ages
60–62 among he con ol g oup wi hin each g oup’s op imal bandwid h. The co esponding de ailed ables a e
epo ed in Table B13 and Table B17.
Ex e nal labo ma ke hickness (ELMT). I de ine ELMT in wo s eps. Fi s , I c ea e 141 local
labo ma ke s based on high wi hin‑ egion and low be ween‑ egion commu ing o wo k,
ollowing K opp/Schwengle (2011). Nex , I c ea e an index ELMT
kc
, showing he local
labo ma ke sha e o 3‑digi occupa ion (o indus y) employmen (E
kc
/E
c
) o e he
na ional sha e o occupa ion (o indus y) employmen (E
k
/E). I coun only wo ke s
be ween 18 and 64 yea s old who a e ei he in employmen subjec o social secu i y
con ibu ions o ainee wo ke s.
32
Figu e A9 epea s he analyses o all es ablishmen s, ega dless o size. When a leas i e cowo ke s
pe o m he same job as a woman, he e ec s a e s ill la ge, despi e being sligh ly smalle (bu no
signi ican ly smalle ) han hose o women wi h no in e nal wo ke subs i u es. This pa e n could be d i en
by he a ia ion in ea men e ec s by es ablishmen size. Indeed, in la ge es ablishmen s, women a e
mo e likely o wo k longe in eac ion o he e o m han hose in smalle es ablishmen s (Figu e A8); hence,
when analyzing in e nal subs i u abili y, i is impo an o accoun o he es ablishmen size by es ic ing
he sample o hose wi h a mos 100 wo ke s. E en in la ge es ablishmen s, i he e a e no in e nal
subs i u es, he e ec s a e qui e la ge, which highligh s ha , al hough in la ge i ms wo ke s s ay in
employmen longe , hose who ha e no subs i u es s ill wo k longe ega dless o he es ablishmen size.
IAB‑Discussion Pape 14|2025 34
Ekc/Ec
ELMTkc = (10)
Ek/E
whe e k is a 3‑digi occupa ion (o indus y), and c is a commu ing zone, Ekc shows he
numbe o wo ke s employed in he occupa ion (o indus y) k, and in he commu ing zone
c, Ec is he numbe o wo ke s employed in he commu ing zone c and all he occupa ions
(o indus ies) oge he , Ek is he numbe o wo ke s employed in he occupa ion (o
indus y) k in all he commu ing zones oge he , while E is he numbe o wo ke s
employed in all he occupa ions (o indus ies) and all he commu ing zones oge he (i.e.,
coun y).33
Figu e 5 displays an example o his index cons uc ion o he nu sing occupa ion and
hospi al ac i i ies indus y. While Passau has many wo ke s employed in hese indus ies
ela i e o he na ional le el, Leipzig does no . This means ha o an es ablishmen loca ed
in Leipzig, an olde wo ke in a gi en occupa ion and indus y is mo e aluable (i.e., such a
wo ke is associa ed wi h highe u no e cos s) han o an es ablishmen loca ed in
Leipzig. I call an ex e nal labo ma ke hick i his index is o e 1, i.e., i he hickness o an
occupa ion (o indus y) in a gi en commu ing zone is dense han he hickness a he
na ional le el. Addi ionally, I de ine a g oup whe e he index ELMTkc is below 0.5 (i.e., he
commu ing zone a mos hal as concen a ed in a gi en occupa ion o indus y as he
coun y‑le el), be ween 0.5 and one (a leas hal as concen a ed bu less concen a ed
han he coun y‑le el).
33 All o hese a iables a e de ined based on my SIEED da a, bu becuase he sample is ep esen a i e o all
Ge man es ablishmen s in he coun y (and he andom sampling p o ides ep esen a i eness o wo k o ce
subjec o social secu i y a he commu ing zone le el), I expec hese indices o p oxy he coun y‑le el
index well.
IAB‑Discussion Pape 14|2025 35
Figu e 5.: Ex e nal labo ma ke hickness by Ge man indus y and occupa ion in 2010
No es: This map shows he compu ed ex e nal labo ma ke hicknesses (ELMT) o each o he 141 local labo
ma ke s based on he K opp/Schwengle (2011) classi ica ions, which a e cons uc ed based on high wi hin‑
egion and low be ween‑ egion commu ing. I compu e ELMT based on Equa ion 10 o he indus y and occu‑
pa ion la ges sha e o emale employees: “Hospi al ac i i ies indus y” (le Panel) and “Nu sing occupa ion”
( igh Panel). I plo he ELMT indexes (Equa ion 10) on he map based on he en deciles p esen ed in he le
co ne o each g aph.
IAB‑Discussion Pape 14|2025 36

The igh Panel o Figu e 4 displays he RDD esul s spli by ex e nal labo ma ke
hicknesses o occupa ions. I women a e employed in a commu ing zone a mos hal as
concen a ed in a gi en occupa ion as he coun y‑le el, hey ex end hei employmen by
41.3 p.p. (58.4% inc ease ela i e o he con ol g oup). I ind ha i a woman is employed in
a commu ing zone a leas hal as concen a ed bu less concen a ed han he
coun y‑le el, he employmen inc ease is 15.6 p.p. (19.9% inc ease ela i e o he con ol
mean). Finally, in he commu ing zones in which a gi en occupa ion is mo e ep esen ed
han a he na ional le el, he e o m leads o an 11.7 p.p. inc ease in employmen a ages
60‑ 62 (15.3 % inc ease ela i e o he con ol mean). This inc ease in employmen is 29.6
p.p. lowe han in commu ing zones a mos hal as concen a ed in a gi en occupa ion as
he coun y‑le el. This esul indica es ha he esponse o he e o m ha aised he
e i emen age is highe o wo ke s in occupa ions wi h hin ex e nal labo ma ke s, whe e
hey a e less subs i u able han in hicke ma ke s.
I examine he e ogenei y in employmen e ec s along ex e nal labo ma ke hickness a he
indus y le el (Figu e A11). Unlike he baseline occupa ion‑based esul s, which showed
clea di e ences by subs i u abili y, I ind no signi ican he e ogenei y in esponses ac oss
indus ies wi h di e en le els o labo ma ke hickness. One po en ial explana ion is ha
indus y‑le el measu es a e oo b oad o cap u e subs i u abili y o speci ic skills o asks.
Addi ionally, la ge i ms, which a e included in he ull sample, may be less a ec ed by
ex e nal labo ma ke condi ions because hey can ely mo e on in e nal eplacemen
op ions.
To accoun o his conce n, I e‑es ima e he analysis o a subsample o es ablishmen s
wi h ewe han 100 employees, whe e i ms a e less likely o ely on in e nal labo ma ke s.
In his subsample, he e ec s o he e o m do di e signi ican ly by indus y‑le el labo
ma ke hickness: I ind ha wo ke s a e mo e likely o emain employed in indus ies in
which he ex e nal labo ma ke s a e hin. (Figu e A12). This sugges s ha ex e nal
subs i u abili y ma e s mo e when i ms ace igh e ex e nal cons ain s and canno ely
on in e nal hi es.
O e all, he occupa ion‑based measu e o ex e nal subs i u abili y emains mo e
in o ma i e han he indus y‑based measu e, because hick indus y labo ma ke s may
e lec b oade agglome a ion pa e ns a he han job‑le el subs i u abili y. Mo eo e ,
indus y hickness may no map well on o he speci ic skills ha i ms need o eplace.
Gende ‑speci ic subs i u abili y. The main esul s ely on gende ‑neu al measu es o
ex e nal labo ma ke hickness (ELMT), pooling employmen densi ies o bo h men and
women. Howe e , Ge many exhibi s p onounced occupa ional and indus y seg ega ion by
gende , and he e o m exclusi ely a ec ed women. I women ace limi ed compe i ion o
IAB‑Discussion Pape 14|2025 37
hi ing ba ie s in male‑domina ed ields, hei e ec i e subs i u abili y may depend on he
gende composi ion wi hin occupa ions and es ablishmen s.34
To explo e his, I cons uc a gende ‑speci ic e sion o he ELMT index using only emale
employmen densi ies (a modi ica ion o Equa ion 10) and e‑es ima e he main analysis
ac oss he p e iously de ined h ee ELMT ca ego ies. As expec ed, he a ia ion in he
emale‑speci ic ELMT is smalle , and he esul s become s a is ically insigni ican
(Figu e A13). One possible explana ion is ha employe s do no con ine hei eplacemen
pool o women and may conside male hi es ins ead. In such cases, a gende ‑neu al ELMT
measu e may be e e lec he labo supply elas ici y i ms ac ually ace.
Howe e , his in e p e a ion is no de ini i e. Relying solely on emale da a educes
s a is ical powe , and he esul ing ELMT measu e may be noisie and less co ela ed wi h
ue subs i u abili y. The e o e, I canno ully assess gende ‑speci ic subs i u abili y wi h
p ecision in his se ing.
None heless, o es whe he gende seg ega ion in e ac s wi h employmen esponses, I
pe o m addi ional subsample analyses by he gende dominance o occupa ions and
es ablishmen s.35 I ind no signi ican di e ences in employmen esponses be ween male‑
s emale‑domina ed con ex s. One possible in e p e a ion is ha , condi ional on
occupa ion and i m size, women and men a e gene ally subs i u able om he i m’s
pe spec i e, and he subs i u abili y measu es used in he baseline a e obus o gende
composi ion.
Does he ex e nal subs i u abili y ma e beyond he local le el? T adabili y o indus ies.
The esul s abo e show ha wo ke s employed in less subs i u able occupa ions in a gi en
local labo ma ke a e mo e likely o ex end hei employmen in esponse o he e o m. I
analyze he b oad indus y g oups and discuss he esul s in e ms o he con en ional
classi ica ion o indus ies by adabili y o es whe he he wo ke s in adable indus ies
a e mo e likely o espond o he aised e i emen age. Such analyses allow me o es
whe he ex e nal subs i u abili y ma e s beyond he local le el. In adable indus ies,
i ms can eplace wo ke s no only locally bu also by ou sou cing asks globally, inc easing
subs i u abili y (D enik e al., 2023). I classi y he indus ies by adabili y ollowing
G ego y/Salomons/Zie ahn (2022).36 Figu e A15 shows no di e ence be ween adable and
34 Fo example, Illing/Schwank/Tô (2024) ind gende gaps in wages a he hi ing s age o acancies c ea ed by
wo ke dea hs in Ge many.
35 I ollow Topho en e al. (2015) and de ine gende ‑in eg a ed occupa ions o es ablishmen s as hose in
which he p opo ion o men and women anges om 21% o 79%. Gende ‑domina ed occupa ions o
es ablishmen s a e hose in which he sha e o one gende exceeds 80%.
36 T adable indus ies a e: Mining (WZ08: B), Manu ac u ing (WZ08: C), Elec ici y, wa e supply (WZ08: D, E),
T anspo , s o age (WZ08: H), Financial se ices (WZ08: K), Real es a e (WZ08: L), Ag icul u e (WZ08: A),
In o ma ion and communica ion (WZ08: J), Scien i ic and echnical se ices (WZ08: M). Non‑ adable
indus ies a e Cons uc ion (WZ08: F), Wholesale and e ail ade (WZ08: G), Ho els, es au an (WZ08: I),
IAB‑Discussion Pape 14|2025 38
un adable sec o s. The esul implies ha subs i u abili y does no ma e beyond he local
le el when i comes o he e ec s o he e o m on emaining in employmen a e 60.37
To conclude, I ind ha job‑speci ic skills and low in e nal and ex e nal subs i u abili y a e
associa ed wi h a s onge inc ease in employmen a ages 60–62 ollowing he e o m.
While he analysis cap u es equilib ium e ec s — ha is, ma ch‑speci ic a ibu es shaped
by bo h wo ke and i m — he p onounced e en ion o manage s and speci ic wo ke s is
consis en wi h highe eplacemen cos s, poin ing o an impo an ole o labo demand
ic ions.
6.3 The e ec o aised ERA on wages by eplacemen cos s
The heo e ical amewo k in chap e 2 p edic s ha aising he ea ly e i emen age
weakens olde wo ke s’ ou side op ions, mos di ec ly by emo ing he allback o ea ly
pension access. Such elimina ion o ou side op ions in he o m o pensions educes wages
o a ec ed wo ke s on a e age. In his sec ion, I analyze whe he he e ec s on wages
display he e ogenei y by subs i u abili y and job‑speci ic skills.
The e a e wo main opposi e o ces ha display he e ogenei y. On he one hand, he
nega i e e ec migh be mo e p onounced o wo ke s wi h high speci ici y (e.g.,
job‑speci ic skills o high‑le el manage ial oles), because hei ou side op ions may be
especially limi ed. On he o he hand, i such wo ke s a e mo e p oduc i e, i ms may ha e
incen i es o o e wage p emia o e ain hem, po en ially o se ing he nega i e e ec on
hei wages (see he de i a ions in chap e 2). Hence, he e ec o he ise in ERA on wages
by subs i u abili y and he speci ici y o skills equi ed o pe o m he gi en job may be bo h
posi i e and nega i e.
I es his implica ion by es ima ing RDD eg essions wi h mon hly wages as he ou come,
ocusing on subsamples ha di e in job‑speci ici y and subs i u abili y. Figu e 6 p esen s
he esul s. As expec ed, he o e all wage e ec is nega i e, consis en wi h educed ou side
op ions weakening employee ba gaining powe , bu e ec s a y ac oss g oups. Among he
mo e eplaceable wo ke s, wages decline pos ‑ e o m. In con as , manage s and hose in
occupa ions ha a e di icul o eplace ex e nally some imes expe ience wage gains a e
he e o m, likely e lec ing i ms’ eluc ance o lose s a egically impo an employees.
Public adminis a ion (WZ08: O), Educa ion (WZ08: P), Heal h and social se ices (WZ08: Q), Cul u al, social
and pe sonal se ices (WZ08: R, S), Household‑ ela ed se ices (WZ08: T), O he economic se ices (WZ08:
N), Ex a e i o ial o ganiza ions (WZ08: U). I hank Duncan Ro h o he help wi h he da a.
37 In addi ion, he gene alized ca ego ies o indus ies help me o es whe he he ex e nal subs i u abili y
ope a es beyond he na ional le el. I de ine indus ies by mapping based on he IAB es ablishmen Panel,
ollowing he p ocedu e desc ibed in Dau h/Eppelsheime (2020). Figu e A16 does no display signi ican
di e ences by adabili y.
IAB‑Discussion Pape 14|2025 39
Figu e 6.: Subsample analyses o he e ec o he ise in ERA on wages a ages 60‑62 by subs i‑
u abili y measu es
Panel A: Human capi al speci ici y Panel B: Hie a chical posi ions
Panel C: In e nal subs i u abili y in he sample o
small es ablishmen s Panel D: Ex e nal subs i u abili y (occupa ions)
No es: Coe icien plo s om RDD eg essions a ound he Janua y 1952 cu o . The es ima es a e ob ained us‑
ing local linea eg essions wi h i s ‑o de polynomials, a iangula ke nel, and mean squa e e o –op imal
bandwid h selec ion. Con ols include calenda mon h o bi h, Wes e n esidence, wages a age 46, and edu‑
ca ion. The e ical lines ep esen 95% con idence in e als based on obus s anda d e o s clus e ed a he
bi h‑mon h le el. Con ol means (on he x‑axis) e e o he a e age employmen a e a ages 60–62 among
he con ol g oup wi hin each g oup’s op imal bandwid h. The co esponding de ailed ables a e epo ed in
Table B23, Table B24, and Table B25.
This esul may e lec i m e en ion mo i es: when speci ic wo ke s con ibu e mo e o
i m p o i s, i ms may o e wage p emia despi e weak ou side op ions. Howe e , selec ion
in o employmen , whe eby only he mos p oduc i e o c i ical wo ke s emain, may bias
he upwa d wage e ec s obse ed in hese g oups. O e all, hese indings highligh ha he
wage e ec s o he e i emen e o m a e shaped by a complex in e play be ween e en ion
needs and ba gaining powe , condi ional on con inued employmen .
IAB‑Discussion Pape 14|2025 40
Ves ad, Ola Lo he ing on (2013): Labou Supply E ec s o Ea ly Re i emen P o ision. In:
Labou Economics, Vol. 25, p. 98–109.
Zweimülle , Jose ; Win e ‑Ebme , Rudol ; Falkinge , Jose (1996): Re i emen o Spouses
and Social Secu i y Re o m. In: Eu opean Economic Re iew, Vol. 40, No. 2, p. 449–472.
Zwick, Thomas; B uns, Mona; Geye , Johannes; Lo enz, S enja (2022): Ea ly Re i emen o
Employees in Demanding Jobs: E idence om a Ge man Pension Re o m. In: The Jou nal o
he Economics o Ageing, Vol. 22, p. 100 387.
IAB‑Discussion Pape 14|2025 47

Appendix
A1 Appendix igu es
Figu e A1.: The assignmen o no mal e i emen age by bi h coho s
No es: This igu e depic s he assignmen ule o no mal e i emen age by bi h coho s. Be o e he 1952 coho ,
he e was a women’s pa hway o e i emen (dashed line). The e ical dashed line a he Janua y 1952 coho
indica es he bi h cu o om which he women’s pa hway o ea ly e i emen was abolished. S a ing om he
1952 coho , he NRA o people eligible o he egula pa hway o e i emen is equal o he NRA o long‑ e m
insu ed, which used o be 65, bu inc eased by mon hly inc emen s pe bi h yea s a ing om he 1947 coho
(black line).
IAB‑Discussion Pape 14|2025 48
Figu e A2.: F ac ion o women employed, nonemployed, and e i ed a each age‑mon h by ea ‑
men and con ol g oup
No es: This igu e displays he e olu ion o h ee main employmen s a es (employmen in black, nonemploy‑
men in da k g ay, and e i emen in ligh g ay‑ see chap e 4 o mo e de ails) o e age by ea men s a us: (i)
ea ed ‑ women bo n in 1952 (solid lines), and (ii) con ol‑ women bo n in 1951 (dashed lines). The i s sho ‑
dashed e ical line (a age 47) co esponds o he age o he 1s ea ed coho in 1999. The nex wo sho
dashed e ical lines show he age ame be ween he old ERA scheme (a age 60) and he new one (a leas age
63) pe he 1999 e o m, while he las wo sho ‑dashed e ical lines show he old NRA scheme (a age 65) and
he new one (a age 65 yea s and six mon hs) pe he 2007 e o m.
IAB‑Discussion Pape 14|2025 49
Figu e A3.: F ac ion o women employed a each age‑mon h by ea men and con ol g oup
No es: This igu e displays he ac ion o women employed a each age mon h by wo ea men s a uses:
ea ed ( he 1952 bi h coho , in black) and con ol ( he 1951 bi h coho , in g ay). The pe iod be ween he
wo dashed lines a 60 and 63 yea s old indica es he gaps be ween he wo g oups due o he 1999 e o m un‑
de s udy.
IAB‑Discussion Pape 14|2025 50
Figu e A4.: The e ec o he ise in ERA: RDD plo
No es: RDD eg ession o he sha e o employed a ages 60‑62 a ound he 1952 cu o . Fo compu ing he RDD
es ima es, I use i s ‑o de polynomials (uppe g aph) o au oma ic 4 h o de (lowe g aph), iangula ke nel
unc ion, and mean squa e‑based op imal bandwid h selec ion p ocedu e. The e ical line ma ks he bi h
coho h eshold 1952 (e.g., 0 co esponds o Janua y 1952, ‑6 co esponds o people bo n six mon hs be o e,
in June 1951).
IAB‑Discussion Pape 14|2025 51
Figu e A5.: RDD by age in mon hs
No es: Coe icien plo s. Each e ical line co esponds o he RDD eg ession o he sha e o employed a a gi en
age‑mon h. Fo compu ing he RDD es ima es, I use local linea eg essions, a iangula ke nel unc ion, and
mean squa e e o ‑based op imal bandwid h choice. The poin s ep esen he es ima ed obus coe icien s,
and he ba s ep esen he 95% con idence in e als, clus e ed a he bi h mon h le el. The ed solid line ep‑
esen s he con ol mean (wi h co esponding alues displayed on he e e sed y‑axis), while he ed dashed
lines ep esen he con idence in e als o he con ol means.
IAB‑Discussion Pape 14|2025 52

Figu e A6.: Subsample analyses o he e ec o he ise in ERA on employmen a ages 60‑62 by
agg ega e occupa ions
No es: Coe icien plo s o RDD eg essions a ound he 1952 cu o . Fo compu ing he RDD es ima es, I use
local linea eg essions, a iangula ke nel unc ion, and mean squa e e o ‑based op imal bandwid h choice.
I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. I pe o m
subsample analyses by en ca ego ies o occupa ions based on occupa ional classi ica ion. The e ical lines
indica e 95% con idence in e als based on obus s anda d e o s clus e ed a he bi h mon h le el. The con‑
ol means (on he x‑axis) show he employmen sha e a he ages o 60‑62 in he co esponding subsample o e
he con ol g oup (bo n in 1951). A co esponding able wi h mo e de ails can be ound in Table B15.
IAB‑Discussion Pape 14|2025 53
Figu e A7.: Subsample analyses o he e ec o he ise in ERA on employmen a ages 60‑62 by
enu e
Panel A: enu e measu ed a 46 yea s old (Me=4.5 Panel B: enu e measu ed a 58 yea s old (Me=7.7
yea s) yea s)
No es: Coe icien plo s o RDD eg essions a ound he 1952 cu o . Fo compu ing he RDD es ima es, I use
local linea eg essions, a iangula ke nel unc ion, and mean squa e e o ‑based op imal bandwid h choice.
I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. I pe o m
subsample analyses by median spli o enu e eco ded a 46 yea s old (Panel A), and 58 yea s old (Panel B)‑ 4.5
and 7.7 yea s, espec i ely. The e ical lines indica e 95% con idence in e als based on obus s anda d e o s
clus e ed a he bi h mon h le el. The con ol means (on he x‑axis) show he employmen sha e a he ages o
60‑62 in he co esponding subsample o e he con ol g oup (bo n in 1951). A co esponding able wi h mo e
de ails can be ound in Table B14.
IAB‑Discussion Pape 14|2025 54
Figu e A8.: The e ec o he ise in ERA on employmen a ages 60‑62 by es ablishmen size
No es: Coe icien plo s o RDD eg essions a ound he 1952 cu o . Fo compu ing he RDD es ima es, I use
local linea eg essions, a iangula ke nel unc ion, and mean squa e e o ‑based op imal bandwid h choice.
I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. I pe o m
subsample analyses by es ablishmen size ca ego ies. The h ee ca ego ies o es ablishmen size a e (1) up o
19, (2) 20‑249, (3) 250‑999, and (4) mo e han 1,000 wo ke s employed a he es ablishmen . The e ical lines
indica e 95% con idence in e als based on obus s anda d e o s clus e ed a he bi h mon h le el. The con‑
ol means (on he x‑axis) show he employmen sha e a he ages o 60‑62 in he co esponding subsample o e
he con ol g oup (bo n in 1951). A co esponding able wi h mo e de ails can be ound in Table B16.
IAB‑Discussion Pape 14|2025 55
Figu e A9.: Subsample analyses o he e ec o he ise in ERA on employmen a ages 60‑62 by
numbe o in e nal subs i u es
No es: Coe icien plo s o RDD eg essions a ound he 1952 cu o . Fo compu ing he RDD es ima es, I use
local linea eg essions, a iangula ke nel unc ion, and mean squa e e o ‑based op imal bandwid h choice.
I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. I pe o m
subsample analyses by he numbe o cowo ke s in he same 3‑digi occupa ion, es ic ing he sample o es‑
ablishmen s wi h ewe han 100 wo ke s. The e ical lines indica e 95% con idence in e als based on obus
s anda d e o s clus e ed a he bi h mon h le el. The con ol means (on he x‑axis) show he employmen
sha e a he ages o 60‑62 in he co esponding subsample o e he con ol g oup (bo n in 1951). A co espond‑
ing able wi h mo e de ails can be ound in Table B13.
IAB‑Discussion Pape 14|2025 56
B1 Appendix ables
Table B1.: Baseline sample size a e each es ic ion in Ge man social secu i y da a
N women,
(bi h coho 1951)
N women,
(bi h coho 1952) N o al
un es ic ed 34570 36776 71346
dele e mine s 34562 36771 71333
dele e sailo s 34560 36768 71328
dele e pa allel spells ‑ ‑ ‑
dele e age‑mon hs below 42 yea s old 32236 34166 66402
dele e age‑mon hs abo e 66 31988 33936 65924
dele e epea ing age‑mon hs ‑ ‑ ‑
dele e i no employed a 58‑59 15640 17130 32770
No es: This able eco ds he sample size a e each o he es ic ions in Ge man social secu i y da a.
The i s column names he es ic ions. The second and hi d columns lis he sample size o ea ed
and con ol g oups, while he las column eco ds he o al sample size, i.e., he sum o he wo p e‑
ceding columns.
Table B2.: Balance check. The e ec o he ise in ERA on co a ia es
(1)
Wes o igin
(2)
non‑Ge man
The ise in ERA ‑0.007 0.013∗∗∗
(0.009) (0.005)
Bandwid h 2.8 3.4
Obse a ions 1179720 1179720
No es: This able shows he e ec o he ise in
ERA on Wes e n Ge man o igin (column 1) and
non‑Ge man na ionali y (column 2) (RDD eg es‑
sion in Equa ion 9). The cu o is Janua y 1952,
s a ing om which ERA was aised by a leas 3
yea s. I pool all obse a ions om he mon h a ‑
e a wo ke ’s 60 h bi hday o hei 63 d bi hday.
I use a iangula ke nel unc ion and a mean
squa e e o ‑based op imal bandwid h choice. I
con ol o calenda mon h, a dummy o Wes ‑
e n esidence, wages a he age o 46, and educa‑
ion. Robus s anda d e o s in pa en heses a e
clus e ed a he bi h‑mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 63

Table B3.: The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
The ise in ERA 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.027) (0.014) (0.016) (0.005) (0.006) (0.021) (23.368)
Bandwid h 2.9 3.9 3.9 4.5 3.0 3.0 3.4
Con ol mean 0.774 0.455 0.232 0.079 0.050 0.179 1719.644
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose by
a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs
co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen (column
5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion and a
mean squa e e o ‑based op imal bandwid h choice. I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age
o 46, and educa ion. The con ol means a e he a e age alues o he ou comes when I limi he sample o women bo n in 1951 ( he
con ol g oup). Robus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el. The co esponding coe icien plo can
be ound in Figu e 2.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 64
Table B4.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by al e ing he es ima ion p ocedu e
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
Panel A: bias‑co ec ed RD es ima es wi h obus a iance es ima o (baseline)
Robus 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.027) (0.014) (0.016) (0.005) (0.006) (0.021) (23.368)
Panel B: con en ional RD es ima es wi h con en ional a iance es ima o
Con en ional 0.166∗∗∗ 0.078∗∗∗ 0.003 0.051∗∗∗ ‑0.020∗∗∗ ‑0.144∗∗∗ ‑64.181∗∗∗
(0.002) (0.009) (0.014) (0.003) (0.002) (0.002) (21.622)
Panel C: bias‑co ec ed RD es ima es wi h con en ional a iance es ima o
Bias‑co ec ed 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.002) (0.009) (0.014) (0.003) (0.002) (0.002) (21.622)
Bandwid h 2.9 3.9 3.9 4.5 3.0 3.0 3.4
Con ol mean 0.774 0.455 0.232 0.079 0.050 0.179 1719.644
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose by
a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs
co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen (column
5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion and a
mean squa e e o ‑based op imal bandwid h choice. I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age
o 46, and educa ion. The con ol means a e he a e age alues o he ou comes when I limi he sample o women bo n in 1951 ( he
con ol g oup). Panel A shows he bias‑co ec ed RD es ima es wi h obus a iance es ima o , Panel B ‑con en ional RD es ima es
wi h con en ional a iance es ima o , Panel C ‑bias‑co ec ed RD es ima es wi h con en ional bias es ima o . S anda d e o s in
pa en heses a e clus e ed a he bi h‑mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 65
Table B5.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by speci ied polynomial o de
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
Panel A: polynomial unc ion o o de 1 (baseline)
The ise in ERA 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.027) (0.014) (0.016) (0.005) (0.006) (0.021) (23.368)
Bandwid h 2.9 3.9 3.9 4.5 3.0 3.0 3.4
Con ol mean 0.774 0.455 0.232 0.079 0.050 0.179 1719.644
Panel B: polynomial unc ion o o de 2
The ise in ERA 0.254∗∗∗ 0.063∗∗∗ 0.131∗∗∗ 0.056∗∗∗ ‑0.040∗∗∗ ‑0.215∗∗∗ ‑145.377∗∗∗
(0.047) (0.022) (0.023) (0.010) (0.014) (0.032) (33.774)
Bandwid h 3.3 4.6 3.2 4.6 3.4 3.3 4.9
Con ol mean 0.769 0.458 0.232 0.079 0.050 0.181 1724.441
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose by
a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs
co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen (column
5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion and a mean
squa e e o ‑based op imal bandwid h choice. I use a i s ‑o de polynomial unc ion in Panel A, and a second‑o de polynomial in
Panel B. I con ol o calenda mon h, a dummy o Wes e n esidence, wages a 46, and educa ion. The con ol means a e he a e age
alues o he ou comes when I limi he sample o women bo n in 1951 ( he con ol g oup). Robus s anda d e o s in pa en heses
a e clus e ed a he bi h‑mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 66
Table B6.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by he speci ied ke nel unc ion
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
Panel A: iangula weigh s (baseline)
The ise in ERA 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.027) (0.014) (0.016) (0.005) (0.006) (0.021) (23.368)
Bandwid h 2.9 3.9 3.9 4.5 3.0 3.0 3.4
Con ol mean 0.774 0.455 0.232 0.079 0.050 0.179 1719.644
Panel B: Epanechniko ke nel
The ise in ERA 0.171∗∗∗ 0.071∗∗∗ 0.008 0.048∗∗∗ ‑0.020∗∗∗ ‑0.148∗∗∗ ‑99.628∗∗∗
(0.029) (0.016) (0.017) (0.006) (0.006) (0.022) (22.389)
Bandwid h 2.9 3.8 4.0 4.3 3.0 3.0 3.5
Con ol mean 0.774 0.455 0.231 0.079 0.050 0.179 1719.644
Panel C: uni o m ke nel
The ise in ERA 0.168∗∗∗ 0.076∗∗∗ 0.002 0.047∗∗∗ ‑0.023∗∗∗ ‑0.146∗∗∗ ‑133.632∗∗∗
(0.029) (0.021) (0.019) (0.004) (0.006) (0.024) (25.635)
Bandwid h 2.7 3.3 3.1 2.8 2.7 2.9 2.6
Con ol mean 0.774 0.455 0.232 0.080 0.047 0.179 1723.688
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose
by a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age
mon hs co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen
(column 5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion
in Panel A, Epanechniko ke nel in Panel B, and uni o m weigh s in Panel C. I use a mean squa e e o ‑based op imal bandwid h
choice. I con ol o calenda mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. The con ol means a e
he a e age alues o he ou comes when I limi he sample o women bo n in 1951 ( he con ol g oup). Robus s anda d e o s in
pa en heses a e clus e ed a he bi h‑mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 67
Table B7.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by ad‑hoc bandwid h choices
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
Panel A: all he bi h coho s
The ise in ERA 0.132∗∗∗ 0.091∗∗∗ ‑0.010 0.051∗∗∗ ‑0.013∗∗∗ ‑0.119∗∗∗ ‑4.547
(0.014) (0.014) (0.016) (0.005) (0.004) (0.011) (38.270)
Bandwid h 12.0 12.0 12.0 12.0 12.0 12.0 12.0
Con ol mean 0.772 0.458 0.228 0.086 0.050 0.178 1744.540
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
Panel B: excluding Decembe 1951 and Janua y 1952 bi h coho s
The ise in ERA 0.115∗∗∗ 0.123∗∗∗ ‑0.055∗∗∗ 0.047∗∗∗ ‑0.006 ‑0.109∗∗∗ 97.799∗∗∗
(0.023) (0.018) (0.015) (0.007) (0.007) (0.017) (23.570)
Bandwid h 12.0 12.0 12.0 12.0 12.0 12.0 12.0
Con ol mean 0.773 0.458 0.229 0.086 0.050 0.177 1744.764
Obse a ions 1077408 1077408 1077408 1077408 1077408 1077408 895417
N wo ke s 29928 29928 29928 29928 29928 29928 28662
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose by
a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs
co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen (column
5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion and a
12‑mon h ad‑hoc bandwid h choice. Panel A displays he eg essions wi h all coho s bo n 1 yea be o e o a e he Janua y 1952
cu o , while Panel B emo es he obse a ions o women bo n 1 mon h a ound he cu o . I con ol o calenda mon h, a dummy
o Wes e n esidence, wages a he age o 46, and educa ion. The con ol means a e he a e age alues o he ou comes when I limi
he sample o women bo n in 1951 ( he con ol g oup). Robus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el.
The co esponding coe icien plo can be ound in Figu e 2.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 68

Table B8.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by he choice o co a ia es included
(1)
employmen
Panel A: baseline (mon h dummies, educa ion, and wes e n Ge man esidence)
The ise in ERA 0.173∗∗∗
(0.027)
Panel B: addi ionally con olling o egional o igin and o eigne (non‑Ge man) s a us
The ise in ERA 0.173∗∗∗
(0.027)
Panel C: no con ols
The ise in ERA 0.152∗∗∗
(0.024)
Bandwid h 2.8
Con ol mean 0.772
Obse a ions 1179720
N wo ke s 32770
No es: This able shows he e ec o ise in he ERA on employmen (RDD eg ession in Equa‑
ion 9). The cu o is Janua y 1952, s a ing om which ERA ose by a leas 3 yea s. I pool
all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age
mon hs co esponding o ages 60–62). I use a iangula ke nel unc ion and a mean squa e
e o ‑based op imal bandwid h choice. In Panel A, I con ol o calenda mon h, a dummy o
Wes e n esidence, wages a he age o 46, and educa ion. In Panel B, I addi ionally con ol o
wes e n o igin and o eigne s a us. I ha e no con ol a iables in Panel C. The con ol means
a e he a e age alues o he ou comes when I limi he sample o women bo n in 1951. Ro‑
bus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 69
Table B9.: Robus ness and sensi i i y checks. The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s old by he speci ied clus e ing me hod o
s anda d e o s
(1)
employ‑
men
(2)
employees
liable o
social secu i y
(3)
ma ginal
pa ‑ ime
employmen
(4)
pa ial
e i emen
(5)
non‑
employ‑
men
(6)
e i e‑
men
(7)
mon hly
wage
Panel A: clus e ing a he bi h da e le el (baseline)
The ise in ERA 0.173∗∗∗ 0.070∗∗∗ 0.015 0.048∗∗∗ ‑0.021∗∗∗ ‑0.150∗∗∗ ‑116.522∗∗∗
(0.027) (0.014) (0.016) (0.005) (0.006) (0.021) (23.368)
Bandwid h 2.9 3.9 3.9 4.5 3.0 3.0 3.4
Con ol mean 0.774 0.455 0.232 0.079 0.050 0.179 1719.644
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 1179720
N wo ke s 32770 32770 32770 32770 32770 32770 32770
Panel B: clus e ing a he es ablishmen le el
The ise in ERA 0.148∗∗∗ 0.070∗∗ 0.022 0.051∗∗ ‑0.017 ‑0.131∗∗∗ ‑136.181
(0.027) (0.035) (0.022) (0.021) (0.011) (0.026) (100.397)
Bandwid h 3.3 3.2 3.3 3.6 3.3 3.4 2.9
Con ol mean 0.769 0.455 0.232 0.081 0.050 0.181 1723.688
Obse a ions 1179720 1179720 1179720 1179720 1179720 1179720 980014
N wo ke s 32770 32770 32770 32770 32770 32770 31346
No es: These ables show he eg ession discon inui y design es ima es a ound he cu o o 1952, s a ing om which ERA ose by
a leas 3 yea s (Equa ion 9). I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs
co esponding o ages 60–62). The e a e 3 mu ually exclusi e ou come a iables: employmen (column 1), nonemploymen (column
5), and e i emen (column 6). Employmen can be u he decomposed in o columns 2‑4. I use a iangula ke nel unc ion and a
mean squa e e o ‑based op imal bandwid h choice. I con ol o calenda mon h, a dummy o Wes e n Ge man esidence, wages
a he age o 46, and educa ion. The con ol means a e he a e age alues o he ou comes when I limi he sample o women bo n in
1951 ( he con ol g oup). Robus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el in Panel A and es ablishmen
le el in Panel B. The co esponding coe icien plo can be ound in Figu e 2.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 70
Table B10.: Falsi ica ion es : RDD on employmen a 60‑62 yea s old a ound placebo cu o s
(1)
employmen
Panel A: 1948 coho emales
Robus RDD ‑0.025
(0.101)
Bandwid h 4.0
Obse a ions 728892
N wo ke s 20247
Panel B: 1949 coho emales
Robus RDD 0.004
(0.784)
Bandwid h 3.7
Obse a ions 853812
N wo ke s 23717
Panel C: 1950 coho emales
Robus RDD ‑0.004
(0.438)
Bandwid h 3.0
Obse a ions 985104
N wo ke s 27364
Panel D: 1951 coho emales
Robus RDD 0.021 ∗
(0.062)
Bandwid h 3.2
Obse a ions 1083420
N wo ke s 30095
No es: This able shows he e ‑
ec o he ise in ERA on employ‑
men (RDD eg ession in Equa‑
ion 9). Panel A pe o ms RDD
o he women bo n in 1947–
1948, a ound he Janua y 1948
cu o ; Panel B ‑ bo n in 1948–
1949, a ound he Janua y 1949
cu o ; Panel C ‑ bo n in 1949–
1950, a ound he Janua y 1950
cu o ; and Panel D ‑ bo n in
1950–1951, a ound he Janua y
1951 cu o . I pool all obse ‑
a ions om he mon h a e a
wo ke ’s 60 h bi hday o hei
63 d bi hday (age mon hs co e‑
sponding o ages 60–62). I use
a iangula ke nel unc ion and
a mean squa e e o ‑based op i‑
mal bandwid h choice. I con ol
o calenda mon h, a dummy o
Wes e n esidence, wages a he
age o 46, and educa ion. Robus
s anda d e o s in pa en heses
a e clus e ed a he bi h mon h
le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗
(p < 0.01).
IAB‑Discussion Pape 14|2025 71
Table B11.: Falsi ica ion es : RDD on employmen a 60‑62 yea s old a ound he e o m cu o o
males
(1)
employmen
Robus RDD 0.051∗∗∗
(0.016)
Bandwid h 3.2
Obse a ions 1230624
N wo ke s 34184
No es: This able shows he e ‑
ec o he ise in ERA on employ‑
men (RDD eg ession in Equa‑
ion 9) o males. The cu o
is Janua y 1952, s a ing om
which ERA ose by a leas 3
yea s. I pool all obse a ions
om he mon h a e a wo ke ’s
60 h bi hday o hei 63 d bi h‑
day (age mon hs co esponding
o ages 60–62). I use a i‑
angula ke nel unc ion and a
mean squa e e o ‑based op i‑
mal bandwid h choice. I con ol
o calenda mon h, a dummy o
Wes e n esidence, wages a he
age o 46, and educa ion. The
con ol means a e he a e age
alues o he ou comes when I
limi he sample o men bo n in
1951. Robus s anda d e o s in
pa en heses a e clus e ed a he
bi h mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗
(p < 0.01).
IAB‑Discussion Pape 14|2025 72
Table B18.: The e ec o he ise in ERA on employmen by ask ype
employmen
(1)
analy ic
non‑ ou ine
(2)
in e ac i e
non‑ ou ine
(3)
cogni i e
ou ine
(4)
manual
ou ine
(5)
manual
non‑ ou ine
The ise in ERA 0.248∗∗∗ 0.145∗∗∗ 0.246∗∗∗ 0.194∗∗ 0.027∗∗
(0.023) (0.018) (0.058) (0.077) (0.013)
Bandwid h 4.2 3.1 2.9 2.9 3.7
Con ol mean 0.739 0.776 0.768 0.754 0.800
Obse a ions 91152 218952 417384 88416 320724
N wo ke s 2532 6082 11594 2456 8909
No es: This able shows he e ec o he ise in ERA on employmen (RDD eg ession in Equa ion 9). The
cu o is Janua y 1952, s a ing om which ERA ose by a leas 3 yea s. I pool all obse a ions om he
mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs co esponding o ages 60–62).
I use a iangula ke nel unc ion and a mean squa e e o ‑based op imal bandwid h choice. I pe o m
subsample analyses in i e ask‑ ype ca ego ies. I con ol o calenda mon h, a dummy o Wes e n esi‑
dence, wages a he age o 46, and educa ion. The con ol means a e he a e age alues o he ou comes
when I limi he sample o women bo n in 1951. Robus s anda d e o s in pa en heses a e clus e ed a
he bi h‑mon h le el. The co esponding coe icien plo can be ound in Figu e A14.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 79

Table B19.: The e ec o he ise in ERA on employmen a 60‑62 yea s old by indus y by adabili y
employmen
(1)
non‑ adable
(2)
adable
The ise in ERA 0.127∗∗∗ 0.227∗∗∗
(0.015) (0.057)
Bandwid h 3.1 3.0
Con ol mean 0.784 0.737
Obse a ions 838044 334044
N wo ke s 23279 9279
S anda d e o s in pa en heses
∗ ∗∗ ∗∗∗
p < 0.10, p < 0.05, p < 0.01
No es: This able shows he e ec o he ise in
ERA on employmen (RDD eg ession in Equa‑
ion 9). The cu o is Janua y 1952, s a ing
om which ERA ose by a leas 3 yea s. I
pool all obse a ions om he mon h a e a
wo ke ’s 60 h bi hday o hei 63 d bi hday
(age mon hs co esponding o ages 60–62). I
use a iangula ke nel unc ion and a mean
squa e e o ‑based op imal bandwid h choice.
I pe o m subsample analyses by adabili y
o sec o s. I con ol o calenda mon h, a
dummy o Wes e n esidence, wages a he
age o 46, and educa ion. The con ol means
a e he a e age alues o he ou comes when I
limi he sample o women bo n in 1951. Ro‑
bus s anda d e o s in pa en heses a e clus‑
e ed a he bi h‑mon h le el. The co e‑
sponding coe icien plo can be ound in Fig‑
u e A15.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 80
Table B20.: The e ec o he ise in ERA on employmen a 60‑62 yea s old by indus y ca ego ies
employmen
(1)
Ag icul u e,
hun ing
and o es y,
ishing
(2)
Food
and
be e age
(3)
Manu‑
ac u e
o consume
p oduc s
(4)
Manu‑
ac u e o
indus ial
goods
(5)
Manu ac u e
o capi al
and consu
me goods
(6)
Cons‑
uc‑
ion
(7)
Ho el
and es‑
au an
(8)
T ans‑
po ,
s o age
(9)
Edu‑
ca ion
The ise in ERA 0.124 0.158∗∗∗ 0.135∗∗∗ 0.157 0.145∗∗∗ 0.236∗∗∗ 0.164∗∗∗ 0.178∗∗∗ 0.124∗∗∗
(0.092) (0.050) (0.012) (0.128) (0.041) (0.018) (0.010) (0.033) (0.018)
Bandwid h 3.2 4.3 3.2 2.7 4.8 3.5 3.5 3.4 2.9
Con ol mean 0.660 0.740 0.726 0.787 0.732 0.786 0.786 0.771 0.784
Obse a ions 17136 34668 30960 41400 44748 21960 279036 252252 424080
N wo ke s 476 963 860 1150 1243 610 7751 7007 11780
No es: This able shows he e ec o he ise in ERA on employmen (RDD eg ession in Equa ion 9). The cu o is Janua y 1952, s a ing om which ERA ose by
a leas 3 yea s. I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d bi hday (age mon hs co esponding o ages 60–62). I use a
iangula ke nel unc ion and a mean squa e e o ‑based op imal bandwid h choice. I pe o m subsample analyses by indus y ca ego ies. I con ol o calenda
mon h, a dummy o Wes e n esidence, wages a he age o 46, and educa ion. The con ol means a e he a e age alues o he ou comes when I limi he
sample o women bo n in 1951. Robus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el. The co esponding coe icien plo can be ound
in Figu e A16
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 81
Table B21.: The e ec o he ise in ERA on employmen a 60‑62 yea s old by gende domina ion
employmen
Panel A: gende domina ion in occupa ion
gende ‑in eg a ed emale‑domina ed male‑domina ed
The ise in ERA 0.122∗∗∗ 0.288∗∗∗ 0.245∗∗∗
(0.018) (0.027) (0.029)
Bandwid h 2.8 3.7 4.5
Con ol mean 0.736 0.778 0.724
Obse a ions 174600 76752 20376
N wo ke s 4850 2132 566
Panel B: gende domina ion in es ablishmen
gende ‑in eg a ed emale‑domina ed male‑domina ed
The ise in ERA 0.188∗∗∗ 0.207∗∗∗ 0.178∗∗
(0.015) (0.025) (0.072)
Bandwid h 4.4 4.0 4.1
Con ol mean 0.741 0.782 0.681
Obse a ions 144000 95184 19656
N wo ke s 4000 2644 546
No es: This able shows he e ec o he ise in ERA on employmen (RDD eg es‑
sion in Equa ion 9). The cu o is Janua y 1952, s a ing om which ERA ose by a
leas 3 yea s. I pool all obse a ions om he mon h a e a wo ke ’s 60 h bi hday
o hei 63 d bi hday (age mon hs co esponding o ages 60–62). I use a iangu‑
la ke nel unc ion and a mean squa e e o ‑based op imal bandwid h choice. The
subsample analyses a e pe o med by gende dominance o occupa ions (Panel A)
and es ablishmen s (Panel B). Gende ‑in eg a ed occupa ions and es ablishmen s
a e de ined as hose in which he p opo ion o men and women anges om 21%
o 79%. Gende ‑domina ed occupa ions/es ablishmen s a e hose in which he
sha e o one o he gende s exceeds 80%. I con ol o calenda mon h, a dummy
o Wes e n esidence, wages a he age o 46, and educa ion. The con ol means
a e he a e age alues o he ou comes when I limi he sample o women bo n in
1951. Robus s anda d e o s in pa en heses a e clus e ed a he bi h‑mon h le el.
The co esponding coe icien plo can be ound in Figu e A10 in he Appendix.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 82
Table B22.: The e ec o he ise in ERA on employmen a 60‑62 yea s old by demog aphic cha ac‑
e is ics o employees
employmen
Panel A: esidence
Eas Wes
The ise in ERA 0.164∗∗∗ 0.174∗∗∗
(0.039) (0.024)
Bandwid h 2.8 3.0
Obse a ions 228168 949392
N wo ke s 6338 26372
Panel B: esidence o o igin
Eas Wes
The ise in ERA 0.172∗∗∗ 0.171∗∗∗
(0.038) (0.025)
Bandwid h 2.9 2.9
Obse a ions 232776 945756
N wo ke s 6466 26271
Panel C: educa ion
high school oca ional uni e si y
The ise in ERA 0.165∗∗∗ 0.183∗∗∗ 0.094∗∗∗
(0.029) (0.039) (0.024)
Bandwid h 3.4 2.8 3.7
Obse a ions 160740 897840 155340
N wo ke s 4545 24940 4315
No es: This able shows he e ec o he ise in ERA on em‑
ploymen (RDD eg ession in Equa ion 9). The cu o is Jan‑
ua y 1952, s a ing om which he ERA ose by a leas 3 yea s.
I pool all obse a ions om he mon h a e a wo ke ’s 60 h
bi hday o hei 63 d bi hday (age mon hs co esponding o
ages 60–62). I use a iangula ke nel unc ion and a mean
squa e e o ‑based op imal bandwid h choice. Panel A pe ‑
o ms subsample analyses by he esidence o he wo ke s
(dummy a iable); Panel B di ides he wo ke s by Eas e n and
Wes e n Ge man o igin, p oxied by he place o esidence o
he i s wo ke as obse ed in he employmen biog aphy;
and Panel C di ides he sample by educa ional ca ego ies. I
con ol o calenda mon h, a dummy o Wes e n esidence,
wages a he age o 46, and educa ion. Robus s anda d e o s
in pa en heses a e clus e ed a he bi h mon h le el.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 83
Table B23.: The e ec o he ise in ERA on mon hly wages a 60‑62 yea s old by measu es o wo ke
skills
mon hly wages
(1) (2)
Panel A: human capi al speci ici y o occupa ions
low high
The ise in ERA ‑233.398∗∗∗ ‑90.348∗∗∗
(64.777) (15.485)
Bandwid h 2.7 3.9
Con ol mean 1694.602 1744.467
Obse a ions 458872 520927
N wo ke s 14619 16721
Panel B: by hie a chical posi ion
no a manage manage
The ise in ERA ‑144.746∗∗∗ 1360.297∗∗∗
(25.361) (195.102)
Bandwid h 3.4 3.4
Con ol mean 1704.381 3287.176
Obse a ions 968243 11771
N wo ke s 30976 370
No es: This able shows he e ec o he ise in ERA
on mon hly wages (RDD eg ession in Equa ion 9).
The cu o is Janua y 1952, s a ing om which ERA
ose by a leas 3 yea s. I pool all obse a ions om
he mon h a e a wo ke ’s 60 h bi hday o hei 63 d
bi hday (age mon hs co esponding o ages 60–62).
I use a iangula ke nel unc ion and a mean squa e
e o ‑based op imal bandwid h choice. Panel A is pe ‑
o med by ”HK speci ici y”‑ which s ands o human
capi al speci ici y o occupa ion. I is based on he e‑
u n o expe ience in Mince equa ions pe o med sep‑
a a ely o each o he 3‑digi occupa ions. Then, I c e‑
a e a dummy a iable based on a median spli ac oss
all he occupa ions. Panel B s ands o manage ial s a‑
us, which is c ea ed as a dummy om he las 2 digi s
o he 5‑digi occupa ional a iables. I con ol o cal‑
enda mon h, a dummy o Wes e n esidence, wages
a he age o 46, and educa ion. The con ol means a e
he a e age alues o he ou comes when I limi he
sample o women bo n in 1951. Robus s anda d e ‑
o s in pa en heses a e clus e ed a he bi h‑mon h
le el. The co esponding coe icien plo can be ound
in Panel A and Panel B o Figu e 6.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 84

Table B24.: The e ec o he ise in ERA on mon hly wages a 60‑62 yea s old by in e nal subs i‑
u abili y (numbe o cowo ke s in he same occupa ion)
mon hly wages
(1)
0
(2)
1‑4
(3)
a leas 5
Panel A: all he es ablishmen ca ego ies
The ise in ERA ‑217.352 ‑960.250∗∗∗ ‑33.783∗∗∗
(327.504) (156.779) (6.537)
Bandwid h 3.5 3.4 3.6
Con ol mean 1566.372 1349.531 1754.040
Obse a ions 44454 33085 877485
N wo ke s 1427 1054 28054
Panel B: es ablishmen s wi h ewe han 100 wo ke s
The ise in ERA ‑1104.690∗∗∗ ‑997.944∗∗∗ ‑474.748∗∗
(98.434) (160.777) (184.439)
Bandwid h 3.7 3.3 2.8
Con ol mean 1818.764 1761.9 1792.3
Obse a ions 18601 20033 47272
N wo ke s 610 641 1503
No es: This able shows he e ec o he ise in ERA on mon hly
wages (RDD eg ession in Equa ion 9). The cu o is Janua y 1952,
s a ing om which ERA ose by a leas 3 yea s. I pool all obse ‑
a ions om he mon h a e a wo ke ’s 60 h bi hday o hei 63 d
bi hday (age mon hs co esponding o ages 60–62). I use a ian‑
gula ke nel unc ion and a mean squa e e o ‑based op imal band‑
wid h choice. I pe o m subsample analyses by 3 ca ego ies o in‑
e nal subs i u es: 0, 1‑4, and a leas 5 wo ke s. The Panel A dis‑
plays he esul s o all he sizes o es ablishmen s, while Panel B
zooms in on he smalle es ablishmen s wi h ewe han 100 wo k‑
e s. I con ol o calenda mon h, a dummy o Wes e n esidence,
wages a he age o 46, and educa ion. The con ol means a e he
a e age alues o he ou comes when I limi he sample o women
bo n in 1951. Robus s anda d e o s in pa en heses a e clus e ed
a he bi h mon h le el. The co esponding coe icien plo can be
ound in Panel C and Panel D o Figu e 6.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 85
Table B25.: The e ec o he ise in ERA on mon hly wages a 60‑62 yea s old by ex e nal subs i‑
u abili y measu es
mon hly wages
(1)
below 0.5
(2)
0.5‑1
(3)
abo e 1
Panel A: ex e nal labo ma ke hickness (indus y)
The ise in ERA ‑155.193 ‑81.246∗∗∗ ‑90.073∗∗∗
(214.400) (12.265) (21.603)
Bandwid h 4.1 4.0 4.1
Con ol mean 1403.030 1518.215 1903.207
Obse a ions 54534 352647 567063
N wo ke s 1738 11245 18164
Panel B: ex e nal labo ma ke hickness (occupa ion)
The ise in ERA 335.624∗∗∗ ‑78.478∗∗∗ ‑178.565∗∗∗
(59.037) (22.142) (38.253)
Bandwid h 4.1 4.0 3.3
Con ol mean 1320.551 1602.430 1872.761
Obse a ions 38515 430331 505147
N wo ke s 1264 13728 16148
No es: This able shows he e ec o he ise in ERA on mon hly
wages (RDD eg ession in Equa ion 9). The cu o is Janua y 1952,
s a ing om which he ERA ose by a leas 3 yea s. I pool all ob‑
se a ions om he mon h a e a wo ke ’s 60 h bi hday o hei
63 d bi hday (age mon hs co esponding o ages 60–62). I use a
iangula ke nel unc ion and a mean squa e e o ‑based op imal
bandwid h choice. Panel A shows subsample analyses by ex e ‑
nal labo ma ke hickness (ELMT) o a gi en occupa ion, based
on he index aking alues below 0.5, 0.5‑1, and abo e 1. Panel
B shows subsample analyses by ELMT o a gi en indus y. I con‑
ol o calenda mon h, a dummy o Wes e n esidence, wages a
he age o 46, and educa ion. The con ol means a e he a e age
alues o he ou comes when I limi he sample o women bo n in
1951. Robus s anda d e o s in pa en heses a e clus e ed a he
bi h mon h le el. The co esponding coe icien plo can be ound
in Panel E and Panel F o Figu e 6.
∗ (p < 0.10), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
IAB‑Discussion Pape 14|2025 86
Lis o Figu es
Figu e 1: Discon inui y in bi h coho s..................................................... 14
Figu e 2: The e ec o he ise in ERA on he employmen s a e (o e all and om
each ca ego y) ...................................................................... 25
Figu e 3: The e ec o he ise in ERA on employmen a ages 60‑62 by e u n o
expe ience in a gi en occupa ion and occupa ional hie a chy le el ........ 31
Figu e 4: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by numbe o in e nal and ex e nal subs i u es o he gi en
occupa ion
........................................................................... 34
Figu e 5: Ex e nal labo ma ke hickness by Ge man indus y and occupa ion in
2010 .................................................................................. 36
Figu e 6: Subsample analyses o he e ec o he ise in ERA on wages a ages
60‑62 by subs i u abili y measu es............................................... 40
Figu e A1: The assignmen o no mal e i emen age by bi h coho s ................. 48
Figu e A2: F ac ion o women employed, nonemployed, and e i ed a each
age‑mon h by ea men and con ol g oup.................................... 49
Figu e A3: F ac ion o women employed a each age‑mon h by ea men and
con ol g oup ....................................................................... 50
Figu e A4: The e ec o he ise in ERA: RDD plo ......................................... 51
Figu e A5: RDD by age in mon hs............................................................. 52
Figu e A6: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by agg ega e occupa ions ........................................... 53
Figu e A7: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by enu e
Figu e A8: The e ec o he ise in ERA on employmen a ages 60‑62 by
es ablishmen size
Figu e A9: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by numbe o in e nal subs i u es
Figu e A10: The e ec o he ise in ERA on employmen a ages 60‑62 by gende ‑
composi ion o occupa ions and es ablishmen s
Figu e A11: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by ex e nal subs i u abili y o a gi en indus y
............................................................... 54
.................................................................. 55
.................................. 56
.............................. 57
.................. 58
Figu e A12: Subsample analyses o he e ec o he ise in ERA on employmen
a ages 60‑62 by ex e nal subs i u abili y, es ic ing o small
es ablishmen s wi h a mos 100 wo ke s ...................................... 59
Figu e A13: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by ex e nal subs i u abili y, using only he emale wo k o ce
o compu a ions ................................................................... 59
IAB‑Discussion Pape 14|2025 87
Figu e A14: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by ask ype ........................................................... 60
Figu e A15: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by adabili y o indus ies
Figu e A16: Subsample analyses o he e ec o he ise in ERA on employmen a
ages 60‑62 by agg ega e indus y ca ego ies
.......................................... 61
................................... 62
Lis o Tables
Table B1: Baseline sample size a e each es ic ion in Ge man social secu i y
da a ................................................................................... 63
Table B2: Balance check. The e ec o he ise in ERA on co a ia es .................. 63
Table B3: The e ec o he ise in ERA on employmen ou comes a 60‑62 yea s
old ..................................................................................... 64
Table B4: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by al e ing he es ima ion
p ocedu e............................................................................. 65
Table B5: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by speci ied polynomial
o de .................................................................................. 66
Table B6: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by he speci ied ke nel
unc ion ............................................................................... 67
Table B7: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by ad‑hoc bandwid h
choices ................................................................................ 68
Table B8: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by he choice o co a ia es
included .............................................................................. 69
Table B9: Robus ness and sensi i i y checks. The e ec o he ise in ERA on
employmen ou comes a 60‑62 yea s old by he speci ied clus e ing
me hod o s anda d e o s ........................................................ 70
Table B10: Falsi ica ion es : RDD on employmen a 60‑62 yea s old a ound placebo
cu o s ................................................................................ 71
Table B11: Falsi ica ion es : RDD on employmen a 60‑62 yea s old a ound he
e o m cu o o males............................................................. 72
IAB‑Discussion Pape 14|2025 88