Escude o, Ve ónica; Liepmann, Hannah; Ve ga a, Damián
Wo king Pape
Di ec ed sea ch, wages, and nonwage ameni ies: E idence
om an online job boa d
ILO Wo king Pape , No. 136
P o ided in Coope a ion wi h:
In e na ional Labou O ganiza ion (ILO), Gene a
Sugges ed Ci a ion: Escude o, Ve ónica; Liepmann, Hannah; Ve ga a, Damián (2025) : Di ec ed
sea ch, wages, and nonwage ameni ies: E idence om an online job boa d, ILO Wo king Pape , No.
136, ISBN 978-92-2-041393-7, In e na ional Labou O ganiza ion (ILO), Gene a,
h ps://doi.o g/10.54394/YWML9238
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/315005
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/4.0/
XDi ec ed Sea ch, Wages, and Non-
Wage Ameni ies:
E idence om an Online Job Boa d
Au ho s / Ve ónica Escude o, Hannah Liepmann, Damián Ve ga a
Ma ch / 2025
ILO Wo king Pape 136
© In e na ional Labou O ganiza ion 2025
A ibu ion 4.0 In e na ional (CC BY 4.0)
This wo k is licensed unde he C ea i e Commons A ibu ion 4.0 In e na ional. See: h ps://
c ea i ecommons.o g/licenses/by/4.0/. The use is allowed o euse, sha e (copy and edis ib-
u e), adap ( emix, ans o m and build upon he o iginal wo k) as de ailed in he licence. The
use mus clea ly c edi he ILO as he sou ce o he ma e ial and indica e i changes we e made
o he o iginal con en . Use o he emblem, name and logo o he ILO is no pe mi ed in con-
nec ion wi h ansla ions, adap a ions o o he de i a i e wo ks.
A ibu ion – The use mus indica e i changes we e made and mus ci e he wo k as ollows:
Escude o, V., Liepmann, H., Ve ga a, D. Di ec ed Sea ch, Wages, and Non-Wage Ameni ies: : E idence
om an Online Job Boa d. ILO Wo king Pape 136. Gene a: In e na ional Labou O ice, 2025.© ILO.
T ansla ions – In case o a ansla ion o his wo k, he ollowing disclaime mus be added
along wi h he a ibu ion: This is a ansla ion o a copy igh ed wo k o he In e na ional Labou
O ganiza ion (ILO). This ansla ion has no been p epa ed, e iewed o endo sed by he ILO and should
no be conside ed an o icial ILO ansla ion. The ILO disclaims all esponsibili y o i s con en and ac-
cu acy. Responsibili y es s solely wi h he au ho (s) o he ansla ion.
Adap a ions – In case o an adap a ion o his wo k, he ollowing disclaime mus be added
along wi h he a ibu ion: This is an adap a ion o a copy igh ed wo k o he In e na ional Labou
O ganiza ion (ILO). This adap a ion has no been p epa ed, e iewed o endo sed by he ILO and should
no be conside ed an o icial ILO adap a ion. The ILO disclaims all esponsibili y o i s con en and ac-
cu acy. Responsibili y es s solely wi h he au ho (s) o he adap a ion.
Thi d-pa y ma e ials – This C ea i e Commons licence does no apply o non-ILO copy igh ma-
e ials included in his publica ion. I he ma e ial is a ibu ed o a hi d pa y, he use o such
ma e ial is solely esponsible o clea ing he igh s wi h he igh s holde and o any claims o
in ingemen .
Any dispu e a ising unde his licence ha canno be se led amicably shall be e e ed o a bi a-
ion in acco dance wi h he A bi a ion Rules o he Uni ed Na ions Commission on In e na ional
T ade Law (UNCITRAL). The pa ies shall be bound by any a bi a ion awa d ende ed as a esul
o such a bi a ion as he inal adjudica ion o such a dispu e.
Fo de ails on igh s and licensing, con ac : [email p o ec ed]. Fo de ails on ILO publica ions and
digi al p oduc s, isi : www.ilo.o g/publns.
ISBN 9789220413920 (p in ), ISBN 9789220413937 (web PDF), ISBN 9789220413944 (epub), ISBN
9789220413951 (h ml). ISSN 2708-3438 (p in ), ISSN 2708-3446 (digi al)
h ps://doi.o g/10.54394/YWML9238
The designa ions employed in ILO publica ions, which a e in con o mi y wi h Uni ed Na ions
p ac ice, and he p esen a ion o ma e ial he ein do no imply he exp ession o any opinion
wha soe e on he pa o he ILO conce ning he legal s a us o any coun y, a ea o e i o y
o o i s au ho i ies, o conce ning he delimi a ion o i s on ie s o bounda ies. See: www.ilo.
o g/disclaime .
The opinions and iews exp essed in his publica ion a e hose o he au ho (s) and do no nec-
essa ily e lec he opinions, iews o policies o he ILO.
Re e ence o names o i ms and comme cial p oduc s and p ocesses does no imply hei en-
do semen by he ILO, and any ailu e o men ion a pa icula i m, comme cial p oduc o p o-
cess is no a sign o disapp o al.
In o ma ion on ILO publica ions and digi al p oduc s can be ound a : www.ilo.o g/ esea ch-
and-publica ions
ILO Wo king Pape s summa ize he esul s o ILO esea ch in p og ess, and seek o s imula e
discussion o a ange o issues ela ed o he wo ld o wo k. Commen s on his ILO Wo king Pape
a e welcome and can be sen o escude [email p o ec ed].
Au ho iza ion o publica ion: Richa d Samans, Di ec o , Resea ch Depa men
ILO Wo king Pape s can be ound a : www.ilo.o g/global/publica ions/wo king-pape s
Sugges ed ci a ion:
Escude o, V., Liepmann, H., Ve ga a, D. 2025. Di ec ed Sea ch, Wages, and Non-Wage Ameni ies:
: E idence om an Online Job Boa d, ILO Wo king Pape 136 (Gene a, ILO). h ps://doi.
o g/10.54394/YWML9238
01 ILO Wo king Pape 136
Abs ac
We le e age ich da a om a p ominen online job boa d in U uguay o assess di ec ed sea ch
pa e ns in job applica ions, ocusing on pos ed wages and ad e ised non-wage ameni ies. We
ind obus e idence o di ec ed sea ch based on pos ed wages in he c oss-sec ion, wi h s a k
he e ogenei y by occupa ion: he wage-applica ion co ela ion is d i en by acancies a ached o
lowe -skill occupa ions, wi h applica ions o acancies a ached o highe -skill occupa ions show
-
ing no esponsi eness o pos ed wages. By applying ex analysis o he job ads, we elici ad e -
ised non-wage ameni ies and ind e idence o di ec ed sea ch based on non-wage ameni ies.
Applica ions o acancies a ached o lowe -skill occupa ions a e consis en wi h lexicog aphic
applica ion p e e ences: ameni ies p edic applica ions o hese acancies only when wages a e
no pos ed. Finally, we exploi indus y-by-occupa ion minimum wage a ia ion o demons a e
ha he obse ed occupa ional he e ogenei y in di ec ed sea ch pa e ns is suppo ed by qua-
si-expe imen al di e ence-in-di e ences es ima es o he impac o wages on job applica ions.
JEL codes: E24, J31, J32, J62, J63
Keywo ds: Di ec ed Sea ch, Vacancies, Wages, Non-Wage Ameni ies, Minimum Wages
Abou he au ho s
Ve ónica Escude o joined he Resea ch Depa men o he ILO in 2008 and oday she is Chie
o he Skills, ALMPs and Policy E alua ion Team. Be ween Ma ch 2021 and Feb ua y 2023, she
se ed as a Visi ing Schola wi h CEGA (Cen e o E ec i e Global Ac ion) a he Uni e si y o
Cali o nia Be keley. She is a PhD specialized on labou and de elopmen economics and applied
mic oeconomics. He cu en esea ch ocuses on assessing he e ec i eness o labou ma ke
and social policies on job quali y and social condi ions. Mo e ecen ly, she has been explo ing
opics ela ed o he skills necessa y o os e e ec i e ansi ions o decen wo k wi h a ocus
on low- and middle-income coun ies, h ough he use o online da a on acancies and applica-
ions o labou po als. She holds a PhD in Economics om Pa is School o Economics and he
École des Hau es É udes en Sciences Sociales (EHESS).
Hannah Liepmann joined he Resea ch Depa men o he In e na ional Labou O ganiza ion
in 2018, whe e she wo ks as an Economis in he Skills, Ac i e Labou Ma ke Policies, and Policy
E alua ion Team. As an empi ical labou economis , she is pa icula ly in e es ed in s udying how
labou ma ke and social p o ec ion policies as well as phenomena o s uc u al change a ec
he in eg a ion o ma ginalized g oups in o quali y employmen . Hannah ob ained he PhD in
Economics om Humbold -Uni e si y Be lin and she is an IZA Resea ch Fellow.
Damián Ve ga a is an Assis an P o esso in he Depa men o Economics a he Uni e si y
o Michigan wi h esea ch in e es s in public and labo economics. In he 2023-2024 academic
yea , he was a Pos doc o al Resea ch Associa e a he Indus ial Rela ions Sec ion a P ince on
Uni e si y. Ve ga a ob ained a B.A. and an M.A. in Economics om he Uni e sidad de Chile and
a Ph.D. in Economics om UC Be keley.
02 ILO Wo king Pape 136
Abs ac 01
Abou he au ho s 01
XIn oduc ion 04
Rela ed li e a u e 07
S uc u e o he pape 08
X1 Gene al Con ex , Da a, and Desc ip i e S a is ics 09
1.1 Gene al con ex 09
1.2 Da a 09
1.3 Desc ip i e s a is ics 11
Vacancies 11
Applican s 12
Applica ions 12
Ameni ies 12
X2 C oss-Sec ional Fac s on Job Applica ions 14
2.1 How di e se a e applica ion po olios? 14
Numbe o applica ions 14
Di e si y in applica ions 14
2.2 C oss-sec ional pa e ns o di ec ed sea ch 17
Applica ions and wages 17
Occupa ional he e ogenei y 18
Applican -le el he e ogenei y 20
2.3 The ole o non-wage ameni ies 21
Pos ed wages and ameni ies 21
Applica ions and ameni ies 23
Applican -le el he e ogenei y 24
2.4 Summa y o indings and discussion 26
X3 The Causal E ec o Wages on Applica ions 27
3.1 Se ing and da a 27
Collec i e Ba gaining Ag eemen s 27
CBAs da a 28
Economic signi icance o he minimum wage ac oss occupa ions 29
3.2 Empi ical s a egy 29
Table o con en s
03 ILO Wo king Pape 136
Es ima ing equa ions 30
3.3 Applica ion e ec s o he minimum wage 31
Robus ness checks and wi hin- acancy design 32
He e ogenei y by applican cha ac e is ics 33
3.4 Addi ional esul s 33
Vacancies and openings 33
Ad e ised non-wage ameni ies 33
Vacancy equi emen s 34
XConclusions 35
Re e ences 36
Figu es and Tables 42
A. Me hodology o C ea ing Va iables om F ee Tex En ies 62
A.1 Skills 62
A.2 Occupa ions 64
A.3 Ameni ies 65
B. Addi ional Figu es and Tables 70
Appendix Bibliog aphy 93
Acknowledgemen s 96
04 ILO Wo king Pape 136
XIn oduc ion
How esponsi e a e job seeke s o he cha ac e is ics o acancies? Unpacking he “black box” o
job applica ions in o ms abou he p esence o labo ma ke ic ions and helps o assess key as-
sump ions in ela ed heo e ical wo k, o example, abou andom e sus di ec ed sea ch, wage
pos ing e sus wage ba gaining, o he ole o non-wage ameni ies. Unde s anding he job ap-
plica ion p ocess is pa icula ly ele an gi en ecen documen a ion o impe ec in o ma ion
and belie s in he labo ma ke , om bo h wo ke s and employe s (Cullen, 2024; Jäge e al.,
2024). Mo eo e , as a gued by Holze e al. (1991), job queuing beha io sugges s he exis ence
o ex-an e en s in he labo ma ke . Hence, job seeke s’ esponses o changes in he cha ac e -
is ics o pos ed acancies can shed ligh on he deg ee o which he documen ed indus y- and
i m-le el wage p emia cons i u e e idence o en s in he labo ma ke .
Despi e i s impo ance, he empi ical s udy o job applica ions is challenging because mos da a-
se s eco d equilib ium ou comes which, by de ini ion, a e only obse ed once he job applica ion
p ocess is comple ed. To o e come his challenge, esea che s ha e swi ched gea s o ga he
di ec in o ma ion on he applica ion p ocess. Hall and K uege (2012) and K uege and Muelle
(2016) pionee ed using su ey da a on wo ke s and job seeke s. Mo e ecen ly, economis s ha e
s a ed using acancy-le el da a om p i a e online job boa ds o be e unde s and how i ms
ad e ise jobs and ec ui wo ke s and how job seeke s sea ch and make applica ion decisions
(e.g., Ban i and Villena-Roldan, 2019; Ma inescu and Wol ho , 2020; Skoda, 2022; A nold e al.,
2023; Ba a e al., 2023).
This pape builds on his la e li e a u e and uses da a om a la ge online job boa d in U uguay
o s udy di ec ed sea ch pa e ns in job applica ions ( ha is, he ex en o which job seeke s di-
ec hei sea ch owa d acancies wi h speci ic a ibu es), ocusing on he ole o pos ed wag-
es and ad e ised non-wage ameni ies. The da a comes om BuscoJobs (BJ), a p ominen on-
line job sea ch pla o m ha ope a es in mo e han 30 coun ies. In U uguay, BJ co e s a b oad
se o indus ies and occupa ions and is es ima ed o con ain a ound 60% o o al online p i a e
sec o acancies in he coun y (Escude o e al., Fo hcoming). We ha e access o da a on acan-
cies, applican s, and applica ions o he pe iod 2011-2020, which we link using unique iden i ie s
o applican p o iles and acancies. On op o he comple e applica ion po olio, applican p o-
iles con ain in o ma ion on gende , age, employmen s a us, employmen his o ies, educa ion,
and aining. Vacancies con ain in o ma ion on he numbe o posi ions hey seek o ill, o mal
equi emen s, and i m and indus y iden i ie s. Also, 20% o acancies pos a mon hly wage.
Impo an o ou analysis, we ha e access o he ull job ad ex , which is p ocessed using Na u al
Language P ocessing (NLP) echniques o elici he ollowing addi ional a iables: he skills e-
qui ed by acancies and he occupa ions acancies seek o ec ui (Escude o e al., Fo hcoming);
and he non-wage ameni ies ad e ised in he job pos (Adamczyk e al., Fo hcoming).
The analysis p oceeds in wo pa s. The i s pa de elops a c oss-sec ional analysis ha con i ms
and ex ends he main indings o Ban i and Villena-Roldan (2019) and Ma inescu and Wol ho
(2020). The analysis shows ha acancies a ached o lowe -skill occupa ions ecei e mo e appli-
ca ions when hey pos highe wages o ad e ise non-wage ameni ies, howe e , he e is a lex-
icog aphic applica ion pa e n as he e ec o ameni ies on applica ions anishes in he subse
o acancies ha pos a wage. On he con a y, applica ions o acancies a ached o highe -skill
occupa ions do no eac o pos ed wages bu inc ease when non-wage ameni ies a e ad e ised.
The second pa o he analysis le e ages he ac ha U uguay implemen s Collec i e Ba gaining
Ag eemen s (CBAs) ha dic a e and equen ly adjus minimum wages, which a y a he indus-
y-by-occupa ion le el. We exploi his ea u e o complemen he c oss-sec ional analysis wi h
05 ILO Wo king Pape 136
causal di e ences-in-di e ences es ima es o wage e ec s on job applica ions. We ind ha min-
imum wage hikes inc ease applica ions o acancies a ached o lowe -skill occupa ions, wi h no
e ec on acancies a ached o highe -skill occupa ions. Hence, he occupa ional he e ogenei y
documen ed in he c oss-sec ion is co obo a ed in he quasi-expe imen al exe cise.
To p e iew ou analysis in mo e de ail, he c oss-sec ional analysis is s uc u ed in h ee exe cis-
es. Fi s , we cha ac e ize applica ion po olios a he applican le el and explo e whe he hey
a e di e si ied o concen a ed in a ew indus ies o occupa ions. We ind subs an ial he e o-
genei y in he numbe o applica ions pe job sea ch spell ac oss applican s. We also documen
ha applica ion po olios a e di e si ied. Job seeke s who submi mul iple applica ions in a gi -
en qua e a ely concen a e hei applica ions wi hin a speci ic indus y and/o occupa ion.
Ins ead, wo ke s end o apply o acancies ha span a wide ange o indus ies and occupa-
ions. Fo example, when applican s submi 5 applica ions in a gi en qua e , hei applica ions
span, on a e age, 4.2 2-digi indus ies, 3.5 1-digi indus ies, and 2.8 1-digi occupa ions. This
quali a i e pa e n emains consis en ega dless o he numbe o applica ions made. This se
o indings sugges s ha wo ke s do no exhibi s ong a achmen s o an occupa ion and, es-
pecially, an indus y a he ime o applica ion, implying ha hey possibly conside o he job a -
ibu es when choosing he acancies hey apply o , making di ec ed sea ch pa e ns easible.
The second c oss-sec ional exe cise explici ly explo es di ec ed sea ch based on pos ed wages.
We i s eplica e Ban i and Villena-Roldan (2019) and Ma inescu and Wol ho (2020) main ind-
ing o a posi i e and signi ican co ela ion be ween pos ed wages and acancy-le el applica ions
once app op ia e skill con ols (in ou case, occupa ions) a e included. The main con ibu ion o
his sec ion, howe e , is o documen a s a k he e ogenei y by occupa ion in he wage-applica-
ion elas ici y. We ind ha o acancies a ached o a subse o occupa ions, which we label as
lowe -skill occupa ions (cle ical suppo , se ices and sales, plan and machine ope a o s, and el-
emen a y occupa ions), he elas ici y o applica ions o pos ed wages is la ge, signi ican , and
highly obus o he inclusion o con ols and sample selec ions. On he con a y, o he acan-
cies a ached o he emaining occupa ions, which we label as highe -skill occupa ions (manage s,
p o essionals, echnicians and associa e p o essionals, and c a wo ke s), he ela ionship be-
ween applica ions and pos ed wages is comple ely absen . This inding is consis en wi h wage
pos ing being mo e p e alen in lowe -skill occupa ions and wage ba gaining and indi idual o e
ailo ing being mo e p e alen in highe -skill occupa ions (e.g., Hall and K uege , 2012; Caldwell
and Ha mon, 2019; Lachowska e al., 2022; Caldwell e al., 2024) since pos ed wages may p o-
ide di e en in o ma ion o applican s depending on hei occupa ion, hus media ing he ap-
plica ion esponsi eness. The documen ed he e ogenei y is also consis en wi h ecen e idence
on employe s being mo e likely o use agg ega e in o ma ion o se wages ( ha is, engage in
“sala y benchma king”) when posi ions a e a ached o low-skill occupa ions (Cullen e al., 2024).
In his exe cise, we also ake ad an age o he applican -le el da a and explo e whe he he e-
sponsi eness o applica ions o pos ed wages a ies wi h applican cha ac e is ics. Bo h a a-
cancy-le el analysis and an applica ion-le el analysis show ha applica ions made by male, em-
ployed, olde , college-educa ed, and skilled job seeke s a e signi ican ly mo e esponsi e o wages
han applica ions made by emale, unemployed, young, non-college-educa ed, and unskilled job
seeke s, espec i ely. We also ind ha applican s wi h p esumably wo se labo ma ke p os-
pec s ( emale, unemployed, young, non-educa ed, and unskilled) display nega i e wage elas ic-
i ies when applying o acancies a ached o highe -skill occupa ions. This inding is consis en
wi h models o di ec ed sea ch whe e wo ke s ade-o wages wi h job sea ch spell leng h (e.g.,
Moen, 1997) and models wi h on- he-job sea ch whe e wo se ou side op ions may encou age
wo ke s o apply o low-wage jobs o climb he job ladde in u u e job ansi ions (e.g., Bu de
and Mo ensen, 1998; Pos el-Vinay and Robin, 2002a,b).
12 ILO Wo king Pape 136
Panel (a) o Table 1 also shows he sha es o acancies ha speci y equi emen s o applican s.
14% o acancies equi e a oca ional aining ce i ica e, while 21% equi e a college deg ee, and
19% o acancies equi e knowledge o a language o he han Spanish (in mos cases, English).
These equi emen s a e di ec ly speci ied by i ms in ela ed en ies when pos ing a acancy.
Meanwhile, o he equi emen s a e speci ied in he open ex o job ads, which a e elici ed using
NLP echniques as desc ibed in Sec ion 1.2. We obse e ha 80% o acancies equi e a leas
one cogni i e skill. Likewise, 83% o acancies equi e a leas one socio-emo ional skill, and 38%
equi e a leas one manual skill.
Applican s
Panel (b) o Table 1 p esen s desc ip i e s a is ics o he applican s egis e ed in he BJ pla o m.
We iden i y 698,880 p o iles in he 2010-2020 pe iod, o which 410,955 a e deno ed as “ac i e” p o-
iles, i.e., indi iduals who made a leas one applica ion in he pe iod Oc obe 2011 o Sep embe
2020.8 To ge a sense o he o de o magni ude, he o al popula ion in U uguay was es ima ed
o be 3,530,912 in 2020, o which 2,067,384 (59%) was be ween 20 and 64 yea s old (INE, 2021).
This implies ha he o al numbe o p o iles c ea ed be ween 2010 and 2020 ep esen s ap-
p oxima ely 34% o he wo king-age popula ion in U uguay in 2020.9 Among he ac i e p o iles,
he mean numbe o applica ions made be ween Oc obe 2011 and Sep embe 2020 is 39.7 and
he co esponding median is 11. Among ac i e applican s, 55% a e emale. 11% epo ha ing
a oca ional aining deg ee, and 16% ha e comple ed a college deg ee. The median applican
was bo n in 1990. While he o e all educa ional s uc u e esembles ha o he na ional labo
o ce, BJ applican s a e sligh ly mo e likely o be college g adua es.10 They also include a disp o-
po iona e numbe o younge wo ke s (Escude o e al., Fo hcoming).
Applica ions
We iden i y 16,320,466 applica ions o acancies made be ween Oc obe 2011 and Sep embe
2020. Panel (c) o Table 1 shows ha 42% o applica ions a e made by indi iduals who epo be-
ing employed a he ime o he applica ion. The a e age age a he ime o applica ion is ela i e-
ly young a 27.7 yea s wi h mode a e dispe sion. Finally, based on he open- ex desc ip ions o
cu en and p e ious jobs, we es ima e ha 31% epo ha ing pe o med cogni i e asks, 45%
epo ha ing pe o med socio-emo ional asks, and 13% epo ha ing pe o med manual asks.
Ameni ies
Table 2 shows desc ip i e s a is ics o he ameni ies ad e ised in ou sample o acancies. 45%
o he acancies in ou sample ad e ise a leas one o he 5 ameni ies desc ibed in Sec ion 2.2.
The a e age numbe o ameni ies ad e ised is 0.7, which can be decomposed as 55% o acan-
cies ad e ising ze o ameni ies, 27% ad e ising one, 12% ad e ising wo, 4% ad e ising h ee,
and less han 2% ad e ising ou o i e ameni ies. The sha e o acancies ad e ising ameni-
ies is la ge among acancies ha do no pos wages (46% e sus 39%). Table 2 also shows ha
8244,960 p o iles epo no applica ion in he 2010-2020 pe iod, while 42,965 only made applica ions be ween Janua y 2010 and
Sep embe 2011 and/o Oc obe 2020 and Decembe 2020.
9As an al e na i e benchma k, 114,392 indi iduals made a leas one job applica ion on he pla o m in 2020, which co esponds o
6% o he popula ion aged 20 o 64.
10 O he na ional labo o ce, 14% comple ed a oca ional deg ee and 11% a college deg ee. Thus, he sha e wi hou such quali ica-
ions is la ge o bo h BJ applican s and he o e all labo o ce (see Escude o e al., Fo hcoming, whe e ci ed igu es a e 2010-2020
a e ages based on he U uguayan household su ey).
13 ILO Wo king Pape 136
he h ee mos commonly ad e ised ameni ies a e “human capi al de elopmen ”, “wo king in
eams”, and “wo k en i onmen /impac on socie y”, which a e ea u ed in 23%, 19%, and 16% o
acancies, espec i ely. “Bonuses and commissions” and “schedule lexibili y” a e ad e ised in
7% and 5% o he acancies, espec i ely.
14 ILO Wo king Pape 136
X2 C oss-Sec ional Fac s on Job Applica ions
Ha ing desc ibed he se ing and he da a, we p oceed wi h he c oss-sec ional analysis. We pe -
o m h ee exe cises. Fi s , we explo e how di e se applica ion po olios a e in e ms o indus ies
and occupa ions. Second, we explo e whe he pos ed wages a ec applica ions. Thi d, we explo e
he ole o ad e ised non-wage ameni ies in he applica ion p ocess. We pay pa icula a en-
ion o he e ogenei y analyses a he applican , acancy, and ameni y le els du ing he analysis.
2.1 How di e se a e applica ion po olios?
To s udy applica ion po olios, we analyze he uni e se o applica ions made o he acancies
conside ed in ou analysis. To p oxy g oups o applica ions made in he same job sea ch spell
(applican s may sea ch o jobs a mul iple s ages in hei ca ee s, hus applying o jobs in di -
e en job sea ch spells), we conside an applican ID-by-qua e -by-yea as a uni o obse a ion
and ocus on “ac i e spells”, i.e., applican ID-by-qua e -by-yea combina ions whe e job seeke s
make a leas one applica ion. I ac ual applica ion spells a e longe han a qua e , ou measu e
will unde es ima e he numbe o applica ions by sea ch spell. This s a egy leads o 1,668,348
applican -by-spell obse a ions wi h a leas one applica ion. On a e age, an ac i e p o ile makes
applica ions in 4.1 di e en qua e s be ween 2011 and 2020. The e is, howe e , subs an ial he -
e ogenei y. 34.4% o he applican p o iles a e ac i e only in one qua e , 37.4% be ween 2 and 4
di e en qua e s, 16.0% be ween 5 and 8 qua e s, 6.4% be ween 9 and 12 qua e s, and 5.8%
in 13 o mo e di e en qua e s.11
Numbe o applica ions
Figu e 2 shows he dis ibu ion o he numbe o applica ions a he applican -by-spell le el, again
conside ing ac i e applican -by-spell combina ions wi h a posi i e numbe o applica ions. Panel
(a) shows wide a ia ion in he numbe o applica ions ac oss applican s. While 23% o applican s
wi h posi i e applica ions make a unique applica ion, 51% o applican s submi be ween 2 and
10 applica ions in a gi en spell, and 13% submi be ween 11 and 20 applica ions. Only 3% o ap-
plican s submi mo e han 50 applica ions in a gi en spell (no shown in he his og am). Panels
(b), (c), and (d) show he dis ibu ions sepa a ely by employmen s a us, gende , and educa ion-
al a ainmen . Dis ibu ions look ema kably simila ac oss demog aphic g oups, especially wi h
espec o gende . While employed and college-educa ed applican s end o make ewe appli-
ca ions on a e age han unemployed and non-college-educa ed applican s, hey s ill show wide
dispe sion in he numbe o applica ions.
Di e si y in applica ions
Ha ing es ablished ha applican s a e he e ogeneous in he numbe o applica ions made by
ac i e spell, we hen explo e whe he applica ions made by a gi en applican in a gi en qua e
11 Applican s who a e ac i e only o one qua e may be di e en han he a e age applican . Fo example, hey may en e he BJ web-
si e bu hen use i less ac i ely as hey quickly ind employmen . In Figu e B.2 o Appendix B, we show ha he dis ibu ion o he
numbe o applica ions indeed changes sligh ly when excluding applican s who a e ac i e only du ing one qua e . Compa ed o
Figu e 2, he a e age numbe o applica ions du ing ac i e spells ends o inc ease. Howe e , in Figu e B.4 o Appendix B, we also
show ha he esul s o his subsec ion a e obus o excluding he applican IDs ha apply o jobs only du ing one qua e .
15 ILO Wo king Pape 136
end o a ge acancies in speci ic indus ies o occupa ions o i , ins ead, hei applica ions a e
di e si ied ac oss indus ies and occupa ions. This analysis can in o m abou he ex en o di-
ec ed sea ch in he labo ma ke : i wo ke s a e s ongly a ached o pa icula indus ies and
occupa ions and, he e o e, hei beha io is less esponsi e o wage di e en ials ac oss indus-
ies and occupa ions, we would expec o see hei job applica ions concen a ed wi hin na ow
ca ego ies o acancies.
We explo e di e si ica ion in applica ion po olios using he ollowing s a is ic. Le i index ob-
se a ions (ac i e applican ID-spell combina ions) wi h Ni he o al applica ions made by he
applican in he spell. Each applica ion goes o a acancy a ached o a g oup g ∈ G, wi h #G he
numbe o di e en possible g oups. Fo example, G may be he se o 2-digi indus y codes,
g a pa icula 2-digi indus y, and #G he numbe o di e en 2-digi indus ies. Le #gi ∈ {1, ...,
min{#G, Ni}} be he numbe o g oups spanned by he Ni applica ions o applican i.12 Fo exam-
ple, i Ni = 10, #gi = 5 means ha he 10 applica ions span 5 di e en 2-digi indus ies. When Ni =
1, #gi is mechanically 1. When Ni > 1, he uppe bound o gi is gi en by min{#G, Ni}. We measu e
di e si ica ion wi h he quan i y:
When D(N) = 1, applica ions a e no di e si ied: all a e made o he same g oup o acancies.
When D(N) = N, applica ions a e comple ely di e si ied: all a e made o acancies ha belong o
di e en g oups. This implies ha he dis ance be ween (N, D(N)) and he 45-deg ee line can be
used o isually diagnose he ex en o di e si ica ion in applica ion po olios, aking in o ac-
coun he he e ogenei y in he numbe o applica ions documen ed abo e.
We s udy di e si ica ion ocusing on ou di e en g oups o acancies. We i s conside a na -
ow de ini ion o acancy g oups ha sha e hei 2-digi indus y code and hei 1-digi occu-
pa ion code, hus employing a s ic e de ini ion o a possibly ele an labo ma ke .13 I indus-
y-by-occupa ion cells cons i u e an accu a e de ini ion o he ele an local labo ma ke o he
applican , we should expec job seeke s o make he majo i y o hei applica ions o acancies
in he same indus y-by-occupa ion cell.14 We also conside b oade g oup de ini ions o ele-
an labo ma ke s: 2-digi indus y codes alone, 1-digi indus y codes alone, and 1-digi occu-
pa ion codes alone.
Figu e 3 shows he esul s. We ocus on applican -spell obse a ions making 10 o ewe appli-
ca ions (Ni ∈ {1, ..., 10}). The black do ed cu e is he 45-deg ee line. Figu e B.3 o Appendix B
shows esul s o Ni ∈ {1, ..., 50}, which displays a simila pa e n.15 Two aspec s o he igu e a e
wo h highligh ing. Fi s , when conside ing he na owe g oup de ini ion (blue cu e, 2-digi in-
dus y by 1-digi occupa ion cell), he le els o di e si ica ion a e subs an ial. Fo example, D(2)
= 1.96, which means ha almos e e yone who applies o 2 acancies applies o acancies in 2
di e en indus y-by-occupa ion cells. While D(N)/N dec eases wi h he numbe o applica ions,
i emains la ge ac oss he dis ibu ion o N. Indi iduals making 5 and 10 applica ions span 4.6
and 8.7 indus y-by-occupa ion cells, espec i ely. This esul implies ha job seeke s a ely a ge
12 Fo mally, #gi can be hough o as he ca dinali y o he pa i ion o Ni in he space o G.
13 We ollow he ca ego iza ions o ISCO 08 o occupa ions and ISIC Re . 4 o indus ies.
14 The usual de ini ion o a local labo ma ke also conside s a geog aphical dimension (e.g., Manning and Pe ongolo, 2017). We dis e-
ga d his dimension since mo e han 50% o he coun y li es in he me opoli an a ea o Mon e ideo, he capi al ci y, and he da a
does no allow us o do a mo e g anula analysis wi hin he ci y.
15 When applican s make oo many applica ions, D(N) is mo e likely o be mechanically a ec ed by #G, a ec ing he in e p e abili y o
D(N) in he ail o he dis ibu ion.
16 ILO Wo king Pape 136
indus y-by-occupa ion cells when making applica ions. Second, while mechanically smalle , di-
e si ica ion emains la ge when conside ing b oade g oups (2-digi indus ies, 1-digi indus-
ies, and 1-digi occupa ions alone). Applican s who make 2 applica ions span, on a e age, 1.90
2-digi indus ies, 1.79 1-digi indus ies, and 1.66 1-digi occupa ions. Applican s who make 5
applica ions span, on a e age, 4.2 2-digi indus ies, 3.5 1-digi indus ies, and 2.8 1-digi occu-
pa ions. Applican s who make 10 applica ions span, on a e age, 7.1 2-digi indus ies, 5.1 1-digi
indus ies, and 3.8 1-digi occupa ions.16
The ac ha indus ies seem o be mo e di e si ied han occupa ions is wo h highligh ing. I
implies ha i is mo e accu a e o hink ha , when applying, wo ke s ix occupa ions and a bi-
age indus ies han he o he way a ound, sugges ing ha di ec ed sea ch is plausible. Fo
he e y leas , he analysis ejec s he hypo hesis ha applican s a ge labo ma ke s de ined
by na ow indus y-by-occupa ion cells, which would limi hei sensi i i y o wage di e en ials
ac oss indus ies. In ha spi i , he analysis sugges s ha indus y-wage di e en ials (K uege
and Summe s, 1988; Ca d e al., 2024) canno be a ionalized by wo ke s ha ing s ong a ach-
men o pa icula indus ies. On he con a y, i sugges s ha indus y wage di e en ials may
gi e o m o a job ladde .
A possible ca ea o hese esul s is ha , condi ional on making se e al applica ions, he acancy
o e dis ibu ion a a gi en poin in ime may be limi ed, hus p e en ing job seeke s om im-
plemen ing “non-di e si ied” applica ion po olios. Fo example, a a gi en poin in ime, he e
may be ew acancies (maybe one o none) associa ed wi h a pa icula indus y-by-occupa ion
cell. In ha sense, he spike a 1 in Figu e 2 may pa ially e lec an a e sion o di e si ica ion o
a subse o applican s. Howe e , he ac ha we obse e a non- i ial sha e o job seeke s mak-
ing mul iple applica ions and ha , condi ional on making mul iple applica ions, job seeke s ap-
ply o a wide ange o acancies in e ms o indus ies and occupa ions, shows ha signi ican
numbe s o applican s a e, in ac , di e si ying hei applica ions. Di e si ica ion seems indeed
subs an ial e en condi ional on only making 2 applica ions. Then, i his conce n is d i ing he
esul s, we should obse e all applican s making a unique applica ion, a pa e n ha is s ongly
ejec ed in Figu e 2.17
As a inal es o “willingness o di e si y,” we le e age he ac ha , o he majo i y o employed
job seeke s, we obse e he 1-digi occupa ion o hei cu en job and he occupa ion a ached
o he acancies hey apply o. Then, we can obse e he sha e o on- he-job applica ions ha
a e made o acancies a ached o he same occupa ion as he cu en job. Figu e 4 shows he
esul s spli by occupa ion o he cu en job and numbe o applica ions made in he job sea ch
spell. As a benchma k, i applica ions we e made andomly, he sha e o applica ions a ge ed
o acancies a ached o he same occupa ion would ma ch he dis ibu ion o pos ed acancies
displayed in Panel (b) o Figu e 1. Figu e 4 shows ha job seeke s in all occupa ions display sha es
la ge han he benchma k sugges ed by Figu e 1, which implies ha wo ke s apply mo e o en
o acancies a ached o hei cu en occupa ions. Howe e , he igu e e eals ha job seeke s
who apply o jobs while employed a e also willing o apply o acancies a ached o o he occu-
pa ions, sugges ing ha di e si ica ion is plausible. This pa e n is obse ed e en o applican s
16 This esul is no exclusi ely explained by di e ences in #G, since D(N) emains a below he uppe bounds. The numbe o 2-digi
indus y by 1-digi occupa ion cells obse ed in he acancy da a is 504. The numbe is 70, 14, and 8 o 2-digi indus ies, 1-digi in-
dus ies, and 1-digi occupa ions, espec i ely.
17 Figu e B.5 o Appendix B explo es o he e ogenei ies by applican cha ac e is ics (employmen s a us, gende , educa ion, and job
sea ch spell leng h). The igu es sugges ha all subg oups o applican s a e di e si ying hei applica ions. Pe he conce n desc ibed
abo e, di e ences in di e si ica ion by applican cha ac e is ics may be pa ially e lec ed by he di e en ial dis ibu ions in he num-
be o applica ions obse ed in Figu e 2.
17 ILO Wo king Pape 136
who make only 1 applica ion in a gi en job sea ch spell. This esul is consis en wi h ecen ind-
ings in Al mann e al. (2024) and Fluch mann e al. (Fo hcoming).
2.2 C oss-sec ional pa e ns o di ec ed sea ch
The ac ha job seeke s ha e di e si ied applica ion po olios sugges s hey may di ec hei
sea ch based on cha ac e is ics o he han occupa ion and indus y. This subsec ion explo es di-
ec ed sea ch pa e ns based on pos ed wages. As men ioned p e iously, among ou sample o
77,874 acancies, 15,835 (20.3%) include a sala y ange in he pos ed ad. On a e age, acancies
ha pos a wage ecei e 21% mo e applica ions (wi h a median o 29%) ela i e o hose ha do
no include wage in o ma ion. While he dis ibu ion o indus ies and occupa ions is simila be-
ween acancies ha pos and do no pos wages (see Figu e 1), he di e ence in applica ions
sugges s ha he decision o pos a wage may be endogenous. In Sec ion 3, we he e o e com-
plemen he c oss-sec ional analysis wi h quasi-expe imen al esul s o p o ide u he g ound
o a causal in e p e a ion o he e ec s o wages on applica ions.
Applica ions and wages
We s a by non-pa ame ically explo ing he ela ionship be ween log applica ions and log pos ed
wages, pooling all acancies ha pos a wage in ou da ase . Figu e 5 shows di e en binsca e
plo s ha a y in he con ols conside ed. Panel (a) shows he aw ela ionship be ween log ap-
plica ions and log pos ed wages. The plo shows a noisy and in e se U-shaped ela ionship: a-
cancies ha pos e y low o e y high wages end o ecei e ewe applica ions. Panel (b) shows
ha he same ela ionship is obse ed when excluding he 3% o ou lie acancies ha ecei ed
mo e han 1,000 applica ions and con olling o 2-digi indus y ixed e ec s, yea ixed e ec s,
and he ad e ised non-wage ameni ies in he acancy. As s essed by Ban i and Villena-Roldan
(2019) and Ma inescu and Wol ho (2020), he c oss-sec ional ela ionship may be spu ious when
no p ope ly con olling o he skills associa ed wi h he job asks. Panels (c) and (d) add 1-dig-
i and 2-digi occupa ion ixed e ec s, espec i ely, and sugges ha , wi h he excep ion o he
acancies a he e y op o he pos ed wage dis ibu ion, he ela ionship be ween applica ions
and pos ed wages becomes posi i e, sugges ing he p esence o a wi hin-occupa ion di ec ed
sea ch pa e n in a wide ange o he pos ed wage dis ibu ion.
To summa ize hese pa e ns in e ms o c oss-sec ional wage-applica ion elas ici ies, we un
OLS eg essions o he ollowing ype:
whe e Appj is he numbe o applica ions pe opening o acancy j, wj is he pos ed wage o a-
cancy j, and Xj a e acancy-le el con ols. We clus e s anda d e o s a he 2-digi indus y le el.
Panel (a) o Table 3 shows he es ima e o
α
unde di e en se s o con ols, esembling he anal-
ysis in Figu e 5. Column (1) shows he aw co ela ion, which is posi i e bu small and no s a-
is ically signi ican . Column (2) excludes ou lie s, includes indus y and yea ixed e ec s, and
con ols o he ad e ised ameni ies. Including his se o con ols has a small e ec on he co-
e icien bu sligh ly inc eases p ecision. Columns (3) and (4) add 1-digi and 2-digi occupa ion
ixed e ec s, espec i ely, gene a ing an inc ease in he es ima ed coe icien ha is s a is i-
cally signi ican a con en ional le els. The c oss-sec ional applica ion-wage elas ici y in hese
columns is 0.17 and 0.19, espec i ely. Column (5) excludes he acancies a he op 5% o he
18 ILO Wo king Pape 136
pos ed wage dis ibu ion. Consis en wi h Figu e 5, omi ing he uppe ail inc eases he elas ic-
i y o 0.33. Finally, as a obus ness check, Column (6) le e ages he ac ha se e al i ms in he
pla o m pos mul iple acancies and, he e o e, conside s only acancies pos ed by i ms wi h
10 o mo e pos ed acancies and includes i m ixed e ec s. While i is no clea whe he i m
ixed e ec s a e good con ols (di ec ed sea ch may e lec job ladde s be ween i ms), i is e-
assu ing ha he elas ici y emains posi i e and signi ican , wi h a alue o 0.21.18 We no e ha
ou analysis eplica es he main conclusions in Ban i and Villena-Roldan (2019) and Ma inescu
and Wol ho (2020): di ec ed sea ch a ises a e including app op ia e con ols o he skill a -
ached o he acancy, which we app oxima e wi h occupa ion codes. Ou es ima ed elas ici ies
a e smalle han he ones es ima ed in he a o emen ioned pape s, possibly because hey use
job i les as he skill con ol which a e subs an ially na owe han 1-digi o 2-digi occupa ion-
al codes. The sensibili y o he es ima ed elas ici y o he included con ols, howe e , p o ides a
simila na a i e in quali a i e e ms.19,20
In mos o he exe cises ha ollow, we epo esul s o he same se s o con ols and sample
e inemen s. Gi en he lessons om ela ed li e a u e and he esul s o Panel (a) in Table 3, how-
e e , we designa e he speci ica ion o Column (3) (no ou lie s, 2-digi indus y ixed e ec s, yea
ixed e ec s, ad e ised ameni ies, and 1-digi occupa ion ixed e ec s) as ou p e e ed speci ica-
ion. The choice o 1-digi o e 2-digi occupa ion codes es s solely on he ac ha 2-digi codes
a e no a ailable o all acancies and, he e o e, using 1-digi codes inc eases he sample size.
Occupa ional he e ogenei y
One ca ea o he analysis abo e is ha i pools all acancies when es ima ing he c oss-sec ional
wage-applica ion ela ionship. I could be he case, howe e , ha di e en occupa ions eac di -
e en ly o pos ed wages. Fo example, indings in Hall and K uege (2012), Caldwell and Ha mon
(2019), Lachowska e al. (2022), and Caldwell e al. (2024) sugges ha wage ba gaining is mo e
p e alen in highe -skilled occupa ions, a ea u e ha could media e how job seeke s a ached
o di e en occupa ions in e p e and eac o pos ed wages in online job ads. To explo e o oc-
cupa ional he e ogenei ies, we eplica e he analysis sepa a ely by 1-digi occupa ion ca ego ies.
Figu e 6 p esen s binsca e plo s o he ela ionship be ween applica ions and pos ed wages
by occupa ion. These igu es exclude ou lie s and include indus y ixed e ec s, yea ixed e -
ec s, and con ols o ad e ised ameni ies. The da a e eals he exis ence o wo g oups o oc-
cupa ions ha display opposi e pa e ns. Panel (a) shows esul s o acancies a ached o cle -
ical suppo , se ices and sales, plan and machine ope a o s, and elemen a y occupa ions. We
deno e his g oup o occupa ions as lowe -skill and e e o acancies a ached o hese occupa-
ions as “lowe -skill acancies”. Vacancies in his g oup exhibi a mono one and posi i e ela ion-
ship be ween applica ions and pos ed wages. Panel (b) shows esul s o acancies a ached o
manage s, p o essionals, echnicians and associa e p o essionals, and c a wo ke s. We deno e
18 The 77,874 acancies a e pos ed by 6,214 i ms. 2,341 i ms only pos one acancy, 2,578 i ms pos be ween 2 and 9 acancies, and
1,295 i ms pos 10 o mo e acancies. 2,682 i ms a e esponsible o he 20% o acancies ha pos wage.
19 Table B.3 o Appendix B eplica es Panel (a) o Table 3 using di e en de ini ions o pos ed wage. Panel (a) uses he midpoin o he
sala y ange. Panel (b) uses he midpoin o he sala y ange bu excludes acancies wi h anges la ge han 50% o he midpoin .
Panel (c) uses he maximum o he sala y ange. In hese cases, and consis en wi h Ban i and Villena-Roldan (2019) and Ma inescu
and Wol ho (2020), elas ici ies a e nega i e in he absence o skill con ols bu become posi i e when adding he occupa ion ixed
e ec s. When using he midpoin o he sala y ange, he quali a i e pa e n o Table 3 is con i med and he esul ing elas ici ies a e
signi ican . When using he maximum o he sala y ange, he same quali a i e pa e n is obse ed bu wi h smalle and non-signi -
ican es ima es. These esul s sugges ha he minimum o he sala y ange is p esumably mo e ele an o job seeke s o decide
on applica ions ela i e o he maximum.
20 1.6% (260) o acancies ad e ising a wage had ze o applica ions and a e, he e o e, excluded om he main analysis. Table B.2 o
Appendix B shows ha a Poisson model ha includes he acancies wi h ze o applica ions yields simila esul s.
19 ILO Wo king Pape 136
his g oup o occupa ions as highe -skill and e e o acancies a ached o hese occupa ions as
“highe -skill acancies”. The ela ionship be ween applica ions and pos ed wages is essen ially
la o his g oup o occupa ions. Panel (c) ep oduces he analysis a e g ouping lowe - and
highe -skill occupa ions in o he wo b oad g oups. The lowe -skill g oup exhibi s a clea posi-
i e co ela ion be ween applica ions and pos ed wages, whe eas he highe -skill g oup displays
no such ela ionship.21
Table B.4 in Appendix B p esen s es ima es o equa ion (2) sepa a e by 1-digi occupa ion g oup,
con i ming he pa e ns displayed in Figu e 6. Panel (b) o Table 3 summa izes he esul s by ep-
lica ing Panel (a) o Table 3 using a model wi h in e ac ions:
whe e LS and HS accoun o lowe - and highe -skill occupa ion, espec i ely. When Xj does no
include 1-digi o 2-digi occupa ion ixed e ec s (Columns (1) and (2)), he eg ession con ols
o 1{Occj ∈ LS}. The esul s a e ema kably s able ac oss columns and con i m he pa e n doc-
umen ed in Figu e 6. Vacancies a ached o lowe -skill occupa ions consis en ly display a posi-
i e and signi ican elas ici y o applica ions o pos ed wages, wi h la ge magni udes close o
he alues epo ed in Ban i and Villena-Roldan (2019) and Ma inescu and Wol ho (2020). On
he con a y, acancies a ached o highe -skill occupa ions show no signi ican ela ionship be-
ween pos ed wages and applica ions. In ou p e e ed speci ica ion (Column (3)), he poin es-
ima es a e
α
LS= 0.41 and
α
HS = -0.07, compa ed o he es ima ed
α
= 0.17 om he eg ession
wi h no in e ac ions.
One possible explana ion o his pa e n is ha highe -skill acancies impose mo e equi emen s
on applican s in e ms o o mal quali ica ions o skills, which could p e en job seeke s om
applying o high-wage highe -skill acancies. In ac , Table B.5 o Appendix B shows ha equi e-
men s a e mo e p e alen in highe -skill acancies. Table B.6 o Appendix B, howe e , shows ha
he absence o di ec ed sea ch in highe -skill acancies holds whe he we es ic he sample o
acancies ha pos o do no pos equi emen s. This is ue in e ms o o mal equi emen s
(see Panels (a) and (c), which ocus on oca ional aining, college deg ees, and/o language e-
qui emen s) and skill equi emen s (see Panels (b) and (d), which pe ain o cogni i e, socio-emo-
ional, and/o manual skills). Con e sely, we con inue o ind e idence o di ec ed sea ch among
acancies a ached o lowe -skill occupa ions independen o acancy equi emen s. Ye , o a-
cancies a ached o hese occupa ions, he esponsi eness o pos ed wages is s onge when
no o mal equi emen s a e pos ed (Panels (a) and (c)).22
21 We e e o he wo da a-d i en occupa ional g oups as lowe - and highe -skilled o he sake o exposi ion cla i y and consis ency
wi h exis ing economic li e a u e (e.g., Kuns e al., 2022; Mon obbio e al., 2023) while acknowledging he limi a ions o skill-based
ca ego ies based on b oad occupa ion codes. A a high le el, his ca ego iza ion is aligned wi h he ISCO-08 guidelines (ILO, 2012),
which classi ies he ou 1-digi occupa ions in ou lowe -skilled g oup a he lowes skill le els 1 and 2, and h ee o he ou 1-digi
occupa ions in ou highe -skilled g oup (manage s, p o essionals, and echnicians and associa e p o essionals) a he highes skill
le els 3 and 4. In he ISCO-08 guidelines, skill le els a e de e mined based on he complexi y and ange o asks and du ies ypically
associa ed wi h an occupa ion, as well as he le el o o mal educa ion equi ed o pe o m hose asks. This classi ica ion sys em e-
lies on a b oad gene aliza ion o asks and du ies ypically pe o med wi hin an occupa ion wi hou accoun ing o a ia ion in ask
complexi y ac oss di e en jobs wi hin he same occupa ion o be ween coun ies. Mo eo e , i places g ea e emphasis on o mal
educa ional quali ica ions despi e he impo ance o o he ypes o lea ning, o example, on- he-job (Konings and Vano melingen,
2015; A anasio e al., 2011; Al onsi e al., 2020). I is wo h no ing ha ou highe -skill g oup includes c a and ela ed ades wo k-
e s which ISCO-08 classi ies a skill le el 2. One possible explana ion o why c a and ades wo ke s exhibi simila applica ion pa -
e ns o he es o occupa ions in he highe -skill g oup is he le el o wages. This occupa ion ypically commands a highe a e age
sala y compa ed o o he occupa ions classi ied a a simila ISCO skill le el 2 (see Table B.14 o Appendix B).
22 Fo he sub-sample o acancies no equi ing any cogni i e, socio-emo ional o manual skills, some o he coe icien s a e mo e im-
p ecisely es ima ed, which may be due o he smalle sample size.
20 ILO Wo king Pape 136
Applican -le el he e ogenei y
Finally, we le e age ou applican -le el da a and es whe he di ec ed sea ch pa e ns a e he -
e ogeneous by applican cha ac e is ics. We p oceed in wo ways. Fi s , we es ima e equa ion (3)
using applica ions om pa icula g oups o applican s as dependen a iables. Table 4 p esen s
he esul s. Reg essions exclude ou lie s and include 2-digi indus y ixed e ec s, yea ixed e -
ec s, ameni y con ols, and 1-digi occupa ion ixed e ec s. While he la ge esponsi eness in
acancies a ached o lowe -skill occupa ions ela i e o highe -skill occupa ions is seen ac oss
all g oups o applican s, poin es ima es e eal subs an ial he e ogenei y by g oup o applican s.
Columns (1) and (2) o Panel (a) show esul s o emale and male applican s, espec i ely. Male
applican s a e subs an ially mo e esponsi e o pos ed wages han emale applican s. While he
lowe -skill elas ici y is 0.63 o male applican s, emale applican s e eal a non-signi ican low-
e -skill elas ici y o 0.14 and a nega i e highe -skill elas ici y o -0.27. Columns (3) and (4) o Panel
(a) p o ide a simila compa ison be ween employed and unemployed applican s, wi h employed
applican s showing a much la ge esponsi eness o pos ed wages in lowe -skilled acancies (
αLS
=
0.61) han unemployed applican s (
α
LS
= 0.28). The la e g oup also e eals a nega i e highe -skill
elas ici y o -0.22. Columns (5) and (6) o Panel (a) compa e applica ions om job seeke s aged
25 o less wi h applica ions om job seeke s olde han 25. Younge applican s show a p ecisely
es ima ed ze o lowe -skill elas ici y and a la ge nega i e highe -skill elas ici y o -0.50. On he con-
a y, he es ima ed elas ici ies o olde applican s esemble he indings o male and employed
applican s (
αLS
= 0.74 and
αHS
= 0.20). Panel (b) shows he e ogenei ies by applican s’ educa ion
and skill le el. Columns (1) o (3) o Panel (b) e eals la ge esponsi eness o applican s wi h
e ia y educa ion, especially o job seeke s wi h a college deg ee in lowe -skilled occupa ions
(
αLS
= 0.83). Columns (4) o (6) show ha applican s wi h cogni i e, socio-emo ional, and manu-
al skills, a e also mo e esponsi e o pos ed wages, bo h in lowe - and highe -skill acancies.23
I is no ewo hy ha he g oups o applican s wi h p esumably wo se labo ma ke p ospec s ex-
hibi nega i e highe -skill elas ici ies. The inding o hese g oups is consis en bo h wi h mod-
els o di ec ed sea ch whe e wo ke s ade-o wages wi h job sea ch spell leng h (e.g., Moen,
1997) and wi h models wi h on- he-job sea ch whe e he lack o ou side employmen op ions
may encou age wo ke s o apply o low-wage jobs wi h he aim o climbing he job ladde in u-
u e job ansi ions (e.g., Bu de and Mo ensen, 1998; Pos el-Vinay and Robin, 2002a,b). The
pa icula ly nega i e esponse obse ed o he younge applican s may also e lec ha hese
wo ke s in e nalize in hei applica ion choices he impo ance o labo ma ke expe ience o
access highe -paying jobs.
As a second explo a ion o applican -le el he e ogenei y in di ec ed sea ch beha io , we use ap-
plica ion-le el da a o es whe he applican demog aphics p edic he log pos ed wage o he a-
cancy hey a e applying o condi ional on being a acancy ha pos s wage. Le i index applican s
and j index acancies. Then, o all applica ions made o acancies ha pos a wage, we es ima e:
whe e Ageij is applican i’s age when applying o acancy j; and Femalei , Employedij , Voc.T n.i ,
College
i
, Cogn.Sk.
ij
, Soc.Sk.
ij
, and Man.Sk.
ij
a e indica o a iables aking he alue o 1 i applican
23 No all acancies ecei e applica ions ha span he comple e dis ibu ion o applican s obse ables. As a consequence, he numbe
o acancies conside ed in each eg ession is no cons an . As displayed in Table B.7 o Appendix B, simila esul s a e ob ained om
a Poisson eg ession model ha includes acancies wi h ze o applica ions.
21 ILO Wo king Pape 136
i is, espec i ely, emale, employed when applying o he acancy, has a oca ional aining de-
g ee, has a college deg ee, and epo s ha ing cogni i e, socio-emo ional, and manual skills
when applying.24 As abo e, Xj con ains acancy-le el con ols which may include indus y ixed
e ec s, yea ixed e ec s, ad e ised ameni ies, and occupa ion ixed e ec s. S anda d e o s
a e clus e ed a he applican le el. One ad an age o his app oach ela i e o he acancy-le el
exe cise epo ed in Table 4 is ha , by con olling simul aneously o all applican cha ac e is ics,
i can be e isola e he pa ial co ela ion o a speci ic a ibu e. I applican cha ac e is ics a e
co ela ed wi h each o he , he exe cise abo e may be picking simila a ia ion ac oss columns,
spu iously a ibu ing esul s o pa icula cha ac e is ics.
Table B.8 o Appendix B shows he esul s. The analysis is consis en wi h he esul s o he a-
cancy-le el analysis o Table 4. Es ima es a e all signi ican and ema kably s able ac oss spec-
i ica ions, sugges ing ha applican -le el he e ogenei y is no d i en by di e en ial so ing o
acancies. The p e e ed speci ica ion (Column (3)) sugges s he ollowing poin es ima es, wi h
li le a ia ion ac oss columns. Rela i e o male job seeke s, emale job seeke s apply o acancies
ha pos 5.7% lowe wages. Rela i e o he unemployed, employed applican s apply o acancies
ha pos 5.2% highe wages. Being one yea olde p edic s a 0.6% highe pos ed wage. Rela i e
o applican s wi h no e ia y educa ion, applican s wi h oca ional aining and a college deg ee
apply o acancies ha pos , on a e age, 3.9% and 10.8% highe wages, espec i ely. Likewise,
applican s wi h cogni i e skills, socio-emo ional skills, and manual skills, apply o acancies wi h
5.3%, 0.7%, and 1.5% highe pos ed wages, espec i ely.
2.3 The ole o non-wage ameni ies
The analysis abo e ocused on he ela ionship be ween applica ions and wages, in some cases
con olling o he ameni ies ad e ised in he acancy. I does no , howe e , explo e he con-
c e e ole ameni ies play o job seeke s, which may yield a deepe unde s anding o he p e-
iously shown he e ogenei ies in he esul s on wages. Then, in his hi d exe cise, we explo e
co ela ions ha in o m he ole o non-wage ameni ies in he applica ion p ocess. We s a by
explo ing co ela ions be ween pos ed wages and ameni ies. We hen analyze how ameni ies
co ela e wi h applica ions and show he e ogenei ies by ameni y, acancy cha ac e is ics, and
applican s’ cha ac e is ics.
Pos ed wages and ameni ies
To documen co ela ions be ween ad e ised ameni ies and pos ed wages, we es ima e OLS
eg essions o he o m:
whe e A includes bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on
socie y, wo king in eams, and human capi al de elopmen , and
Am
j
a
= 1{Vacancy j ad e ises
ameni y a}. We also conside eg essions ha only include Amj = 1{Vacancy j ad e ises a leas
1 o he 5 ameni ies}. The po en ial se o con ols is he same as in he p e ious subsec ion, and
24 In ou da a, o mal educa ion indica o s a e ime-in a ian . Howe e , skills a iables a e ime- a ian since hey a e buil om cu en
employmen and, in he case o unemployed indi iduals, employmen his o ies.
28 ILO Wo king Pape 136
CBAs a e nego ia ed a he g oup le el. G oups co espond o b oad economic indus ies. Each
g oup ba gains o e one o mo e CBAs, depending on he numbe o subg oups conside ed. The
objec i e o ha ing di e en con ac s wi hin a g oup is o accommoda e economic di e ences
be ween sub-indus ies, al hough all CBAs wi hin a g oup a e join ly ba gained and, in some cas-
es, exhibi li le wi hin-g oup he e ogenei y. Each g oup has disc e ion o de ine he occupa ions
(i.e., ca ego ies) ha will be subjec o speci ic minimum wages in he ba gained CBA.
Figu e 7 p o ides addi ional in o ma ion abou he CBAs. Panel (a) displays he numbe o g oups
and CBAs (subg oups) by yea . In 2005, CBAs we e nego ia ed in 20 g oups. This inc eased o
24 g oups in 2008 when CBAs became a ailable o economic ac i i ies ha we e his o ically
excluded om he wage councils, such as he domes ic wo ke s g oup and h ee g oups ep-
esen ing ac i i ies o he u al economy. Wi hin each g oup, he e a e he abo e-men ioned
subg oups ha nego ia e di e en CBAs. The e we e 172 subg oups nego ia ing CBAs in 2005,
which eached 221 wi h he inco po a ion o he p e iously excluded g oups in 2008, co e ing
i ually all p i a e sec o employees. Since hen, he inc ease in he numbe o subg oups is ex-
plained by a eo ganiza ion wi hin g oups a he han an inc ease in co e age. Panel (b) shows
he dis ibu ion and e olu ion o he numbe o wage loo s de ined wi hin CBAs. As explained
be o e, g oups ha e au onomy o de ine he ca ego ies ha will be a ec ed by he sec o al min-
imum wages de ined in he CBAs. The numbe o ca ego ies co esponds o he numbe o oc-
cupa ions wi h a ixed minimum wage in he CBA. The e is subs an ial dispe sion in he numbe
o ca ego ies conside ed. In a ypical yea , a CBA in he 25 h pe cen ile de ined 6 di e en mini-
mum wages, while a CBA in he 75 h pe cen ile de ined be ween 25 and 30 di e en minimum
wages. Some g oups de ined mo e han a hund ed ca ego ies, which explains he dis ance be-
ween he median and he a e age. The numbe o ca ego ies wi hin CBAs is e y s able o e
ime. Finally, Panel (c) shows ha , among he 96,598 minimum wage changes ha we iden i y
in he aw CBA da a, mo e han 90% happened ei he in Janua y o July. This ea u e will be im-
po an o ou empi ical s a egy below.
CBAs da a
We ely on in o ma ion o he indus y-by-occupa ion minimum wages de ined in each ipa i e
nego ia ion and eco ded in he CBAs. A e each ba gaining ound, each g oup de ines nominal
wages and biannual adjus men s ha a e alid un il he nex ba gaining ound. CBAs and he
co esponding lis o indus y-by-occupa ion minimum wages a e public in o ma ion. The e o e,
we use digi ized minimum wage da a collec ed om each ound’s con ac .
One ca ea when me ging he CBAs da a wi h he BJ da a is ha g oups and ca ego ies ha de-
ine indus y-by-occupa ion minimum wages do no map one- o-one o he s anda dized codes
a ailable in he BJ da a. The e o e, we manually assign 2-digi ISIC codes o each subg oup, and
1-digi ISCO occupa ion codes o each ca ego y wi hin he con ac . One p oblem wi h his ap-
p oach is ha he impu ed codes can be b oade o na owe han he s anda dized codes. Fo
example, wi hin he g oup “Comme ce”, he e is a CBA o he subg oup “S o es”, o which se -
e al 2-digi ISIC codes apply. Likewise, wi hin he g oup “Food and be e ages manu ac u ing”,
he e a e di e en CBAs o “Whea Mills” and “Rice Mills”, which can be associa ed o he same
2-digi indus y code. A simila issue occu s wi h he ca ego ies wi hin each CBA. One minimum
wage can be associa ed wi h se e al 1-digi occupa ion codes, and se e al minimum wages can
be associa ed wi h he same 1-digi occupa ion code.
Since i is no possible o p ecisely a ach speci ic minimum wages o acancies because o his
mul iplici y p oblem, we build measu es o exposu e o minimum wage changes a he acancy
le el by compu ing summa y s a is ics o all minimum wages ha can be associa ed o a speci ic
29 ILO Wo king Pape 136
indus y-occupa ion combina ion. Then, we code whe he , on a gi en da e, he e is a change in
he compu ed s a is ic.28 Unde his s a egy, non-exposed acancies a e acancies whose indus-
y-occupa ion combina ion ei he canno be associa ed wi h a minimum wage in any CBA, o
acancies o which we can associa e a minimum wage bu i is no changing in he speci ic pe i-
od. The esul ing cells a e de ined a he 2-digi indus y le el and he 1-digi occupa ional le el.
Economic signi icance o he minimum wage ac oss occupa ions
Ou causal analysis aims o assess whe he we can eplica e he occupa ional he e ogenei y
in di ec ed sea ch pa e ns we ound ea lie in he c oss-sec ion. This exe cise equi es occu-
pa ion-speci ic minimum wages o bind in all occupa ions. To alida e his assump ion, we use
su ey da a o explo e whe he minimum wages a e di e en ially binding ac oss occupa ions.
Table B.14 o Appendix B shows ha highe -skill occupa ions exhibi highe a e age wages han
lowe -skill occupa ions bu also highe minimum wages. The a io be ween a e age hou ly min-
imum wages and median hou ly wages compa es well ac oss he di e en occupa ions, anging
om 6% o 18%. This sugges s ha he economic signi icance o he minimum wage is indeed
compa able o he wo occupa ional g oups.
3.2 Empi ical s a egy
Ou empi ical s a egy exploi s he equen a ia ion in minimum wages a he indus y-by-oc-
cupa ion le el p o ided by he CBAs. Each July and Janua y, se e al indus y-by-occupa ion cells
(and, he e o e, he acancies wi hin hose cells) see hei ba gained wage loo adjus ed (see
Panel (c) o Figu e 7). O he indus y-by-occupa ion cells (and hus acancies) see no change in
hei minimum wage, ei he because he con ac is no adjus ing wages in ha adjus men win-
dow, o because some occupa ions do no ha e assigned minimum wages in ce ain sec o al
con ac s. This pa e n o adjus men s gi es ise o na u al ea men and con ol g oups o
which we can es ima e DID models a ound he ime o adjus men . The empi ical s a egy uses
an indus y-by-occupa ion cell as he uni o obse a ion, o which we build a balanced panel
o es ima e s anda d e en -s udy speci ica ions. As a obus ness check, we also es ima e mod-
els using a wi hin- acancy design whe e he uni s o obse a ion a e acancies ha expe ience
a minimum wage inc ease while being ac i e (see Sec ion 3.3).
Ou s a egy may be in e p e ed as conse a i e o wo easons. Fi s , as discussed abo e, we do
no obse e he exac minimum wage ha is a ached o each acancy. Since we use he indus y
and occupa ion a ached o he acancy o measu e exposu e o minimum wage changes, which
do no ma ch one- o-one wi h he de ini ions in he CBAs, ou es ima es should be in e p e ed
as educed- o m in end- o- ea (ITT) es ima ions, possibly inducing a enua ion bias. Second,
we assume ha job applican s a e awa e o he iming o he minimum wage adjus men s and,
he e o e, can upda e hei applica ions a e minimum wages a e inc eased. Ina en ion o min-
imum wages should wo k agains inding applica ion e ec s and, he e o e, should also exe
downwa d bias in ou es ima ions.29
28 Since CBAs change in a coo dina ed ashion, he measu e o exposu e does no depend on he choice o he s a is ic.
29 I he BJ pla o m included many acancies a ge ing sel -employed wo ke s, who a e no co e ed by minimum wages, his would be
ano he sou ce o possible downwa d bias. In p ac ice, he ac ion o such acancies is negligible. Fo example, when sea ching o
he pe inen exp ession “independien e” in job i les, only 110 (0.14%) men ion i .
30 ILO Wo king Pape 136
Es ima ing equa ions
In wha ollows, xi deno es he a iable x o cell
i
in calenda ime (mon h)
, whe e a cell is a
2-digi indus y-by-1-digi occupa ion combina ion. Cells included in he balanced panel a e cells
o which we obse e a leas one pos ed acancy du ing he whole pe iod. In he da a, we ob-
se e acancies spanning 70 2-digi indus ies and 8 1-digi occupa ions. In e ms o ou uni o
obse a ion, we obse e acancies in 506 di e en cells (o a po en ial o 560). Ou pe iod con-
sis s on 108 mon hs be ween Oc obe 2011 o Sep embe 2020, gi ing o m o a o al sample
size o 54,648.
Since ea ed cells po en ially inc ease he minimum wage e e y six mon hs, we implemen a
s acked e en s udy as ollows (Cengiz e al., 2019; Ga dne , 2021; Bake e al., 2022; Dube e al.,
2023). We de ine e en pe iods anging om h ee mon hs be o e a minimum wage inc ease o
wo mon hs a e . This modeling decision means ha e en pe iods un ei he om Oc obe o
Ma ch, o om Ap il o Sep embe , such ha p e- and pos -e en indica o s a e de ined ela i e
o Janua y o July. Each e en is indexed by e. We conside da a om Oc obe 2011 o Sep embe
2020, which ansla es in o 18 di e en e en windows whe e a subse o he cells expe iences a
minimum wage inc ease. In each e en , he subse o con ol cells is composed o cells wi h no
minimum wage inc ease. Then, we es ima e s anda d e en speci ica ions by allowing he cell
ixed e ec s o a y by e en . Since e en pe iods do no o e lap, ime ixed e ec s au oma ically
a y by e en , and e en s a e uniquely de e mined by calenda ime, e( ). To add mo e lexibili y,
we also allow ime ixed e ec s o a y by 1-digi indus ies.
Fo mally, he es ima ing equa ion is gi en by:
Y
i is an ou come o in e es o cell
i
in ime
. Diτe( ) a e e en indica o s, whe e τ deno es he dis-
ance om he e en (in mon hs) meaning ha Diτe( ) is equal o one i cell
i
was ea ed τ mon hs
ago in e en e( ). α
ie( )
a e cell-by-e en ixed e ec s. Y
j(i)
a e mon h-by-1-digi indus y ixed e ec s.
X
i a e con ols o he small sha e o minimum wage changes ha occu in mon hs di e en
om Janua y o July (see Panel (c) o Figu e 7), whose e ec is allowed o a y by e en .30 Unde
he pa allel ends assump ion, βτ iden i ies causal e ec s om he minimum wage inc ease on
Y
i . As i is s anda d in e en s udies, β−1 is no malized o 0. Since minimum wage changes may be
co ela ed wi hin CBA ac oss occupa ions, we clus e s anda d e o s a he 2-digi indus y le el.
To p o ide a quasi-expe imen al es o he c oss-sec ional di ec ed sea ch pa e ns documen ed
in Sec ion 3, we de ine Yi , ou main ou come o in e es , as he median numbe o applica ions
ecei ed by acancies o cell
i
pos ed in mon h . We also conside a a ia ion o equa ion (10)
ha in e ac s he e en indica o s wi h lowe - and highe -skill occupa ion indica o s. The spa si-
y o he balanced panel implies ha Yi = 0 is a equen ou come, so we es ima e he equa ion
in le els and hen compu e back-o - he-en elope es ima es o he implied elas ici y using ex-
e nal da a on a e age minimum wage inc eases. As a complemen o ou di ec ed sea ch es ,
we also es ima e e ec s o o he ou comes a he cell-by- ime le el such as numbe o pos ed
30 Following Cengiz e al. (2019, 2021), Xi is compu ed as ollows. Le R be he mon h in which he a e minimum wage inc ease akes
place. Then, de ine Ea ly = 1{ ϵ { R- 3, R- 2}}, P e = 1{ = R- 1}and Pos = 1{ ϵ { R, R+ 1, R+ 2}}, and le Ra ei be an indica o o cells ha
ace a e minimum wage inc eases. Then Xi includes all he in e ac ions be ween {Ea ly , P e , Pos } × {Ra ei} o each e en sepa a e-
ly.
31 ILO Wo king Pape 136
acancies, openings, sha e o acancies ad e ising non-wage ameni ies, and sha e o acan-
cies pos ing job equi emen s.
To p o ide summa y esul s, we also epo es ima es om s anda d pooled DID eg essions:
whe e Tie( ) is an indica o a iable ha akes alue 1 i cell
i
is ea ed in e en e( ), Pos is an indi-
ca o a iable ha akes alue 1 i mon h
is equal o o la ge han he ea men mon h ( ha
is, i
co esponds o ei he Janua y, Feb ua y, Ma ch, July, Augus , o Sep embe ), and all o h-
e a iables a e de ined as in equa ion (10). The coe icien o in e es in his speci ica ion is β.
Table 11 p esen s desc ip i e s a is ics o he es ima ion sample. 47% o he obse a ions exhib-
i a leas one acancy opening. The mean numbe o openings, including he ze os, is 4.62. The
median and mean numbe o applica ions pe opening is 35.33 and 43.89, espec i ely. 51% o
he obse a ions co espond o ea ed cell-by-e en g oups. Panels (b) and (c) b eak he s a is-
ics by low- and highe -skill occupa ions. The sha e o obse a ions wi h a leas one opening is
ema kably simila ac oss g oups, al hough lowe -skill acancies usually exhibi mo e openings
and mo e applica ions. No su p isingly, lowe -skill occupa ions a e mo e likely o be ea ed han
highe -skill occupa ions (60% e sus 43%).
3.3 Applica ion e ec s o he minimum wage
Ou main esul s use he median numbe o applica ions pe opening a he cell-le el as depend-
en a iable. Figu e 8 shows he es ima ed βτ coe icien s o equa ion (10) wi h hei co espond-
ing 95% con idence in e als. Panel (a) shows he e en s udy ha pools all acancies. The plo
sugges s ha applica ions o acancies in exposed cells inc ease a e he minimum wage ad-
jus men s, al hough he inc ease is small and non-s a is ically signi ican a con en ional le els.
Howe e , as shown in Panel (b), when in e ac ing he e en indica o s wi h he low- and high-
e -skill occupa ion dummies, minimum wage inc eases end o gene a e a signi ican inc ease
in applica ions o exposed lowe -skill acancies wi h no e ec on exposed highe -skill acancies.
Panel (c) plo s he esul s om an analog iple di e ence eg ession, showing ha he di e -
ence be ween lowe - and highe -skill occupa ions is s a is ically signi ican , especially in he i s
mon h a e he minimum wage adjus men . Table 12 shows he esul s o he pooled DID. Panel
(a) shows esul s o he speci ica ion wi h no in e ac ions, which es ima es a noisy inc ease in
2.6 applica ions a he ea ed occupa ion-by-indus y cell le el, wi h an implied elas ici y o 1.15
(Column (1)). Panel (b) shows he esul s o he speci ica ion wi h in e ac ions o he occupa-
ional g oups. The es ima ed e ec o lowe -skill acancies is a signi ican a e age inc ease o
4.5 applica ions pe cell, wi h an implied elas ici y o 1.59. Fo highe -skill acancies, he implied
elas ici y is only 0.02 (Column (1)).
Two aspec s o his esul a e wo h discussing. Fi s , he quasi-expe imen al exe cise mi o s
he c oss-sec ional inding o he e ogeneous pa e ns o di ec ed sea ch ac oss occupa ions.
This is especially ema kable gi en he di e ences be ween he used a ia ion, design, and se
o acancies. Second, he magni ude o he implied wage-applica ions elas ici y in lowe -skill a-
cancies is simila in magni ude, albei on he lowe end o he dis ibu ion, o he labo supply
elas ici y es ima es documen ed in he empi ical monopsony li e a u e (Sokolo a and So ensen,
2021). This benchma k is eassu ing gi en he easons o which we expec a enua ion bias in
ou eg essions.
32 ILO Wo king Pape 136
Robus ness checks and wi hin- acancy design
Table 12 p o ides some obus ness checks o ou main es ima es. Fo he sake o b e i y, we o-
cus he discussion on he speci ica ion including in e ac ions by occupa ional g oup (Panel (b)).
We also discuss below esul s om an al e na i e wi hin- acancy esea ch design ha yields
simila esul s.
Column (2) o Panel (b) shows ha using mean ins ead o median applica ions pe cell-by- ime
gene a es simila , albei sligh ly smalle , es ima es. Ye , he ac ha he panel is spa se and he
dis ibu ion o applica ions pe acancy is skewed sugges s ha he median may be a be e -be-
ha ed measu e a he occupa ion-by-indus y-by-mon h le el. Since he equa ion is es ima ed
in le els, ou lie s may play a signi ican ole in d i ing he esul s. Columns (5) and (6) show ha
excluding he e en window-by-cell uni s o which he median numbe o applica ions pe open-
ing pe cell in a leas one mon h exceeded 750 a enua es he esul s, bu he quali a i e conclu-
sions do no change. Finally, Columns (9) and (10) show ha es ic ing he e en window-by-cell
uni s o which he cell had openings in a leas wo mon hs ba ely changes he poin es ima es,
al hough i dec eases p ecision.31
As an addi ional obus ness check, we es ima e he applica ion e ec s o he minimum wage
using a di e en esea ch design ha exploi s wi hin- acancy a ia ion. We conside he sample
o acancies ha a e open when a minimum wage change po en ially occu s ( ha is, ei he be-
ween June and July o be ween Decembe and Janua y) and es ima e DID eg essions a he a-
cancy le el, compa ing exposed and non-exposed acancies be o e and a e he policy change.
This design allows us o e ine he p e ious analysis as i con ols o acancy ixed e ec s and
he eby cap u es ime-in a ian unobse ed he e ogenei y a he acancy le el. A he same
ime, since acancies a e usually open du ing only one mon h o less (see Table 1), we a e no
able o anspa en ly assess he pa allel ends assump ion. The sample size, mo eo e , dec eas-
es because acancies pos ed in o he mon hs a e no included in he sample and because mos
applica ions o acancies a e made in he i s ew days o he acancy pe iod (see Figu e B.8 o
Appendix B).32 This pa e n implies ha we need o es ic o acancies ha ha e been open o
a ew days when he minimum wage change kicks in. Gi en hese ad an ages and limi a ions,
we see he wi hin- acancy design as a complemen o he p e ious analysis, whe e we a e in e -
es ed in whe he he main conclusions hold ac oss he wo empi ical s a egies.
Ou baseline sample o acancies used in he wi hin- acancy design consis s o he 2,129 acancies
(2.7% o he o al sample) ha we e pos ed be ween June 25 and June 30 o be ween Decembe
26 and Decembe 31 in any o he yea s conside ed. Since we es ima e he eg ession in le els,
we exclude acancies ha ecei e mo e han 1,000 applica ions. Table B.15 o Appendix B shows
desc ip i e s a is ics. Wi h his sample, we es ima e he ollowing eg ession:
whe e Y
j
a e he applica ions pe opening o acancy j in mon h , T
j
is an indica o ha akes alue
1 i he acancy j is ea ed, Pos is an indica o i mon h
is ei he Janua y o July, αj a e acancy
31 We no e ha he implied elas ici ies a e mechanically downwa d biased in he la e exe cise because he exclusion o ze os dis o
he p e-e en mean dependen a iable by cons uc ion.
32 45% o applica ions happen in he i s 2 days, 63% in he i s 5 days, 72% in he i s week. The e a e h ee explana ions o his pa -
e n. Fi s , ecen acancies a e mo e likely o appea i s on he websi e. Second, when acancies a e iled, hey may s op ecei ing
applica ions. Thi d, applican s may op o ecei ing emails wi h weekly upda es o newly pos ed acancies, which again may inc ease
he salience o he ecen ly pos ed ones.
33 ILO Wo king Pape 136
ixed e ec s, and γ a e mon hs (calenda ime) ixed e ec s. As abo e, ea men s a us is de-
e mined based on he indus y-by-occupa ion-by-adjus men window a ached o he acancy.
Table B.16 o Appendix B shows he esul s. We es ima e eg ession (12) o di e en sub-sam-
ples based on he days he acancy was open be o e he po en ial policy change. Columns (1),
(2), (3), and (4) show esul s o acancies ha we e open o 3 days o less, 4 days o less, 5 days
o less, and 6 days o less, espec i ely. We no e wo indings. Fi s , while hey a e noisy, esul s
imply ha ea ed lowe -skill acancies ecei e an inc ease in applica ions a e he minimum
wage inc ease, wi h no co esponding e ec in highe -skill acancies. Tha is, he wi hin- acan-
cy exe cise suppo s he pa e ns documen ed so a . Second, he magni ude o he e ec is de-
c easing in he days open be o e he policy change. The implied elas ici y o lowe -skill acan-
cies is 5.1, 2.6, 1.4, and 0.4 in he co esponding columns. This is consis en wi h acancies being
mo e salien when hey a e ecen ly pos ed.
He e ogenei y by applican cha ac e is ics
Table 13 p esen s esul s o equa ion (11) wi h occupa ion in e ac ions bu using applica ions
om pa icula g oups o applican s as he dependen a iable. All g oups exhibi la ge espons
-
es o lowe -skill acancies ela i e o highe -skill acancies. Consis en wi h he c oss-sec ional
analysis, we ind la ge and mo e signi ican applica ion esponses o wages o male and olde
applican s. The implied wage-applica ion elas ici y o lowe -skill acancies is 2.3 o male appli-
can s, ela i e o a non-signi ican es ima e o 1 o emale applican s. Likewise, he lowe -skill
elas ici y o olde applican s is 1.7, compa ed o a 1.4 es ima e o younge applican s. The im-
plied elas ici ies, howe e , a e no di e en be ween employed and unemployed applican s, and
a e s onge o job seeke s wi h no e ia y educa ion, which con as s om wha was ound
in he c oss-sec ional analysis. This di e ence may be d i en by he ac ha , wi hin occupa ion
and indus y, less educa ed applican s may be mo e a ached o minimum wage jobs han high-
ly educa ed applican s.
3.4 Addi ional esul s
In he emainde o he sec ion, we discuss esul s o complemen a y dependen a iables.
Vacancies and openings
The posi i e e ec o minimum wages on applica ions may come a he expense o a con ac ion
in labo demand in e ms o acancies o openings. We es his hypo hesis by es ima ing simi-
la models as abo e using he o al numbe o acancies and openings pe cell as he depend-
en a iable. Figu e 9 and Table 12 show ha we do no ind any de ec able e ec on acancies
and openings. This esul sugges s ha he inc ease in applica ions may help i ms bu e he
inc ease in labo cos s and/o ha i ms a e adjus ing o he ma gins o pay o he inc eased
minimum wage.
Ad e ised non-wage ameni ies
I p o iding ameni ies is cos ly o i ms, ad e ised non-wage ameni ies could dec ease a e
he minimum wage inc ease (Clemens, 2021). We es his hypo hesis by es ima ing simila mod-
els as abo e using he sha e o acancies ha ad e ise non-wage ameni ies as he dependen
a iable. Figu e B.6 and Table B.17 o Appendix B sugges he absence o nega i e esponses on
34 ILO Wo king Pape 136
ad e ised ameni ies. The only ad e ised ameni y ha exhibi s a non i ial nega i e implied
elas ici y is bonuses and commissions, howe e , he e en s udies sugges ha he nega i e e -
ec is possibly d i en by di e en ial p e- ends.
Vacancy equi emen s
Finally, i ms could eac o inc eased labo cos s by becoming mo e selec i e in e ms o educa-
ion and skills equi emen s. E idence o his na a i e has been p esen ed by Bu schek (2021)
and Clemens e al. (2021). We es o his hypo hesis by es ima ing simila models as abo e us-
ing he sha e o acancies ha impose equi emen s as he dependen a iable. Figu e B.7 and
Table B.18 o Appendix B sugges he absence o inc eases in educa ion and skill equi emen s,
al hough es ima es a e imp ecise enough o make s ong claims abou hese esul s. The only
sligh ly signi ican posi i e es ima e is an es ima ed inc ease in he sha e o highe -skill acan-
cies ha equi e a college deg ee.
35 ILO Wo king Pape 136
XConclusions
In his pape , we assess pa e ns o di ec ed sea ch in job applica ions, ocusing on he ole o
pos ed wages and ad e ised non-wage ameni ies. Using ich da a om a p ominen online job
boa d in U uguay, we a e able o p o ide a se ies o c oss-sec ional ac s on job applica ions,
which we hen co obo a e causally using plausibly exogenous minimum wage a ia ion.
Fi s , we documen subs an ial he e ogenei y ac oss applican s in he numbe o applica ions
hey send wi hin an applica ion spell, and ind a la ge deg ee o di e si ica ion in e ms o he
occupa ions and indus ies o he acancies hey apply o wi hin job seeke s ha send mul iple
applica ions. Second, we ind obus e idence o di ec ed sea ch based on pos ed wages ha is
d i en by acancies a ached o lowe -skill occupa ions, wi h applica ions o acancies a ached
o highe -skill occupa ions showing no esponsi eness o pos ed wages. The di ec ed sea ch
pa e n is ound o be s onge o male, employed, olde , college-educa ed, and skilled job ap-
plican s. Finally, by applying ex analysis o he job ads, we elici ad e ised non-wage ameni ies
and ind ha hey play a key ole in he applica ion p ocess. We ind e idence o di ec ed sea ch
based on ameni ies and show ha applica ions o lowe -skill acancies a e consis en wi h lexi-
cog aphic job p e e ences whe e ameni ies a ec applica ions only when wages a e no pos ed.
We also ind subs an ial he e ogenei y on he ole o non-wage ameni ies by ameni y, occupa ion,
and applican cha ac e is ics. The occupa ional he e ogenei y in di ec ed sea ch is suppo ed by
a quasi-expe imen al exe cise ha uses minimum wage a ia ion a he indus y-by-occupa ion
le el o documen posi i e applica ion e ec s o minimum wage inc eases in lowe -skill occupa-
ions. This exe cise also sugges s he absence o esponses in he numbe o acancies, open-
ings, ad e ised ameni ies, o acancy equi emen s a e minimum wage inc eases.
Ou indings help in o m se e al mechanisms behind he sea ch-and-ma ching p ocess in he
labo ma ke . They a e consis en wi h models o di ec ed sea ch and sugges ha indus y- and
i m-wage di e en ials can be a ionalized by he exis ence o en s a he han s ong wo ke
a achmen o indus ies. They mo eo e un eil impo an occupa ional he e ogenei ies, which
a e consis en wi h he la ge incidence o wage pos ing ( a he han ba gaining) in lowe -skill
occupa ions ha has been documen ed in ela ed li e a u e.
Based on ou indings, se e al a enues o u u e esea ch may be wo h pu suing. Fi s , i seems
p omising o explo e he undamen al di e ences be ween occupa ions mo e deeply. While we
conjec u e ha he di e en ial incidence in wage pos ing and ba gaining can explain hese di -
e ences, u he esea ch is needed o depic a clea e pic u e o ha pa e n. Second, we ha e
been able o exploi plausibly exogenous a ia ion in wages, bu addi ional causal analyses ha
ely on exogenous a ia ion in ameni ies would u he enhance he unde s anding o he job
applica ion p ocess. Beyond hei po en ial o in o ming economic heo y, such analyses ha e
p ac ical alue in ha hey shed ligh on how ce ain go e nmen al in e en ions and i ms’ e-
c ui men s a egies a ec he applican pool.
36 ILO Wo king Pape 136
Re e ences
Adamczyk, Willian, Isau e Delapo e, and Ve ónica Escude o, “Job Ameni ies in De eloping
Coun ies: an analysis o online labou ma ke s,” ILO Wo king Pape , Fo hcoming.
Al onsi, Li ia, O iana Bandie a, Vi o io Bassi, Robin Bu gess, Im an Rasul, Munshi Sulaiman,
and Anna Vi ali, “Tackling You h Unemploymen : E idence F om a Labo Ma ke Expe imen in
Uganda,” Econome ica, 2020, 88 (6), 2369–2414.
Alleg e o, Syl ia and Michael Reich, “A e local minimum wages abso bed by p ice inc eases?
Es ima es om in e ne -based es au an menus,” ILR Re iew, 2018, 71 (1), 35–63.
Al mann, S., R. Mahls ed , M. J. Ra enbo g, and A. Sebald, “Which Occupa ions do Unemployed
Wo ke s Ta ge ? Insigh s om Online Job Sea ch P o iles,” Wo king Pape , 2024.
A nold, Da id, Simon Quach, and Bledi Taska, “The impac o pay anspa ency in job pos ings
on he labo ma ke ,” Wo king Pape , 2023.
Ashen el e , O ley and Š ěpán Ju ajda ˇ , “Minimum wages, wages, and p ice pass- h ough:
The case o McDonald’s Res au an s,” Jou nal o Labo Economics, 2022, 40 (S1), S179—-S201.
A alay, Enghin, Phai Phong hieng ham, Sebas ian So elo, and Daniel Tannenbaum, “The
E olu ion o Wo k in he Uni ed S a es,” Ame ican Economic Jou nal: Applied Economics, 2020, 12
(2), 1–34.
A anasio, O azio, Ad iana Kugle , and Cos as Meghi , “Subsidizing Voca ional T aining o
Disad an aged You h in Colombia: E idence om a Randomized T ial,” Ame ican Economic Jou nal:
Applied Economics, 2011, 3 (3), 188–220.
Aza , José, S e en Be y, and Ioana Elena Ma inescu, “Es ima ing labo ma ke powe ,” Wo king
Pape , 2022.
Bake , And ew, Da id F La cke , and Cha les CY Wang, “How Much Should We T us S agge ed
Di e ence-In-Di e ences Es ima es?,” Jou nal o Financial Economics, 2022.
Ban i, S e ano and Benjamin Villena-Roldan, “Do high-wage jobs a ac mo e applican s?
Di ec ed sea ch e idence om he online labo ma ke ,” Jou nal o Labo Economics, 2019, 37 (3),
715–746.
Ba banchon, Thomas Le, Roland Ra helo , and Alexand a Roule , “Gende di e ences in job
sea ch: T ading o commu e agains wage,” The Qua e ly Jou nal o Economics, 2021, 136 (1),
381–426.
Bassie , Ihsaan, Alan Manning, and Ba ba a Pe ongolo, “Vacancy du a ion and wages,” Wo king
Pape , 2023.
Ba a, Honey, Amanda Michaud, and Simon Mongey, “Online Job Pos s Con ain Ve y Li le
Wage In o ma ion,” Wo king pape , 2023.
37 ILO Wo king Pape 136
Belo , Michèle, Philipp Ki che , and Paul Mulle , “P o iding Ad ice o Jobseeke s a Low Cos :
An Expe imen al S udy on Online Ad ice,” The Re iew o Economic S udies, 2019, 86, 1411–1447.
_ , _ , and _ , “How Wage Announcemen s A ec Job Sea ch - A Field Expe imen ,” Ame ican Economic
Jou nal: Mac oeconomics, 2022, 14 (4), 1–67.
Bó, E nes o Dal, F ede ico Finan, and Ma ín A Rossi, “S eng hening s a e capabili ies: The
ole o inancial incen i es in he call o public se ice,” The Qua e ly Jou nal o Economics, 2013,
128 (3), 1169–1218.
Bonhomme, S ephane and G ego y Joli e , “The pe asi e absence o compensa ing di e -
en ials,” Jou nal o Applied Econome ics, 2009, 24 (5), 763–795.
Bu de , Kenne h and Dale T Mo ensen, “Wage di e en ials, employe size, and unemploy-
men ,” In e na ional Economic Re iew, 1998, pp. 257–273.
Bu schek, Sebas ian, “Raising he ba : Minimum wages and employe s’ hi ing s anda ds,”
Ame ican Economic Jou nal: Economic Policy, 2021.
Caldwell, S., I. Haegele, and J. Heining, “Ba gaining and inequali y in he labo ma ke ,” Wo king
Pape , 2024.
Caldwell, Sydnee and Nikolaj Ha mon, “Ou side op ions, ba gaining, and wages: E idence om
cowo ke ne wo ks,” Wo king Pape , 2019.
Ca d, Da id, “Who se you wage?,” Ame ican Economic Re iew, 2022, 112 (4), 1075–1090.
_ , Jesse Ro hs ein, and Moises Yi, “Indus y wage di e en ials: A i m-based app oach,” Jou nal
o Labo Economics, 2024, 42 (S1), 11–59.
Cengiz, Do uk, A ind aji Dube, A ila Lindne , and Ben Zippe e , “The e ec o minimum
wages on low-wage jobs,” The Qua e ly Jou nal o Economics, 2019, 134 (3), 1405–1454.
_ , _ , _ , and Da id Zen le -Mun o, “Seeing Beyond he T ees: Using Machine Lea ning o Es ima e
he Impac o Minimum Wages on Labo Ma ke Ou comes,” Jou nal o Labo Economics, 2021.
Clemens, Je ey, “How do i ms espond o minimum wage inc eases? Unde s anding he el-
e ance o non-employmen ma gins,” Jou nal o Economic Pe spec i es, 2021, 35 (1), 51–72.
_ , Lisa B Kahn, and Jona han Mee , “D opou s need no apply? The minimum wage and skill
upg ading,” Jou nal o Labo Economics, 2021, 39 (S1), S107–S149.
Co iello, Decio, E ika Dese anno, and Nicola Pe sico, “Minimum wage and indi idual wo ke
p oduc i i y: E idence om a la ge US e aile ,” Jou nal o Poli ical Economy, 2022, 130 (9), 2227– 2500
Cullen, Zoë, “Is pay anspa ency good?,” Jou nal o Economic Pe spec i es, 2024, 38 (1), 153–180.
Cullen, Zoe B, Shengwu Li, and Rica do Pe ez-T uglia, “Wha ’s my employee wo h? The e ec s
o sala y benchma king,” Wo king Pape , 2024.
Deming, Da id and Lisa B. Kahn, “Skill equi emen s ac oss i ms and labo ma ke s: E idence
om job pos ings o p o essionals,” Jou nal o Labo Economics, 2018, 36 (S1), S337–S369.
44 ILO Wo king Pape 136
No es: This igu e plo s he s a is ic desc ibed in equa ion (1), he a e age numbe o “g oups” indi iduals apply o in each qua -
e -by-yea , as a unc ion o he o al numbe o applica ions made in he qua e -by-yea o ou inal sample o pos ed acancies
(see Sec ion 1 o de ails on he sample es ic ions). “G oups” e e o 2-digi indus ies by 1-digi occupa ion cells (blue cu e,
which conside s 504 ca ego ies), 2-digi indus ies ( ed cu e, 70 ca ego ies), 1-digi indus ies (g een cu e, 14 ca ego ies), and
1-digi occupa ions (yellow cu e, 8 ca ego ies). Fo eadabili y, we censo he igu e a 10 applica ions.
No es: This igu e plo s he sha e o applica ions made o acancies a ached o he same 1-digi occupa ion o he cu en em-
ploymen as a unc ion o he o al numbe o applica ions made in he qua e -by-yea o ou inal sample o pos ed acancies
45 ILO Wo king Pape 136
(see Sec ion 1 o de ails on he sample es ic ions). By cons uc ion, his igu e only conside s applican s who a e employed in
he qua e -yea o he applica ion. As an example, a ound 15% o applica ions om job seeke s employed as manage s, who
make 1 applica ion in a gi en qua e -by-yea , a ge manage ial jobs (wi h he emaining applica ions a ge ing jobs in o he
1-digi occupa ions); and a ound 8% o applica ions om job seeke s employed as manage s, who make 10 applica ions in a
gi en qua e -by-yea , a ge manage ial jobs.
No es: This igu e shows binned sca e plo s and co esponding quad a ic i s o he ela ionship be ween he log numbe o
applica ions pe acancy and he log pos ed wage. The analysis conside s all acancies in ou inal sample ha pos a wage
(see Sec ion 1 o de ails on he sample es ic ions). Panel (a) does no include con ols. Panel (b) excludes acancies ecei ing
mo e han 1,000 applica ions and includes 2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies
(indica o s o bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and hu-
man capi al de elopmen ). Panels (c) and (d) augmen Panel (b) speci ica ion by including 1-digi and 2-digi occupa ional ixed
e ec s, espec i ely.
46 ILO Wo king Pape 136
No es: This igu e shows binned sca e plo s and co esponding quad a ic i s o he ela ionship be ween he log numbe o
applica ions pe acancy and he log pos ed wage sepa a ely by occupa ion. The analysis conside s all acancies in ou inal
sample ha pos a wage (see Sec ion 1 o de ails on he sample es ic ions). All plo s exclude acancies ecei ing mo e han
1,000 applica ions and include 2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s
o bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al
de elopmen ). Panel (a) plo s he a o emen ioned ela ionship o lowe -skilled occupa ions (cle ical suppo , se ices and sales,
plan and machine ope a o s, and elemen a y occupa ions). Panel (b) plo s he a o emen ioned ela ionship o highe -skilled
occupa ions (manage s, p o essionals, echnicians and associa e p o essionals, and c a wo ke s). Panel (c) plo s he a o emen-
ioned ela ionship sepa a ely o he wo agg ega e occupa ional g oups.
47 ILO Wo king Pape 136
No es: This igu e p esen s desc ip i e ac s on he scheme o CBAs. Panel (a) shows he numbe o sec o al g oups and ela ed
subg oups ha de e mine he CBAs by yea . Panel (b) shows he dis ibu ion o he numbe o minimum wages ha a e speci-
ied wi hin each CBA by yea . Panel (c) shows he mon hly dis ibu ion o minimum wage adjus men s, pooling all changes ob-
se ed in ou pe iod.
48 ILO Wo king Pape 136
No es: These igu es plo he es ima ed βτ coe icien s o equa ion (10) wi h hei co esponding 95% con idence in e als using
he median numbe o applica ions pe acancy wi hin he sample o acancies a ached o he co esponding indus y-by-oc-
cupa ion cell. Panel (a) pools all acancies. Panel (b) conside s in e ac ions wi h indica o s o lowe - and highe -skill occupa-
ional g oups. Panel (c) plo s he co esponding iple di e ence, whe e he coe icien is in e p e ed as he di e ence be ween
lowe - and highe -skill acancies. Reg essions con ol o cell-by-e en ixed e ec s, calenda mon h-by-1 digi -indus y ixed
e ec s, and minimum wage changes no occu ing in Janua y o July (see Panel (c) o Figu e 7). S anda d e o s a e clus e ed a
he 2-digi indus y le el.
No es: These igu es plo he es ima ed βτ coe icien s o equa ion (10) wi h hei co esponding 95% con idence in e als using
di e en dependen a iables. Coe icien s a e in e ac ed wi h indica o s o lowe - and highe -skill occupa ional g oups. Panel (a)
uses he o al numbe o pos ed acancies as a dependen a iable. Panel (b) uses he o al numbe o openings as he depend-
en a iable. Reg essions con ol o cell-by-e en ixed e ec s, calenda mon h-by-1 digi -indus y ixed e ec s, and minimum
wage changes no occu ing in Janua y o July (see Panel (c) o Figu e 7). S anda d e o s a e clus e ed a he 2-digi indus y le el.
49 ILO Wo king Pape 136
No es: This able shows summa y s a is ics. Panel (a) shows s a is ics o acancies in ou inal sample (see Sec ion 1 o de ails
on he sample es ic ions). Panel (b) shows s a is ics o applican s egis e ed in he BJ pla o m. Panel (c) shows s a is ics o
applican s a he ime o applica ion, only conside ing applica ions o ou inal sample o acancies. In Panel (a), “ oca ional ain-
ing” is de ined as e ia y-le el aining, whe eas he a iable addi ionally cap u es lowe le els o oca ional aining in Panel (b).
50 ILO Wo king Pape 136
No es: This able shows summa y s a is ics o he ameni ies ad e ised in ou inal sample o acancies (see Sec ion 1 o de-
ails on he sample es ic ions). Ad e ised ameni ies we e elici ed ollowing Adamczyk e al. (Fo hcoming). The able de ails
whe eas acancies ad e ise a leas one ameni y, he numbe o ameni ies ad e ised pe acancy, and p o ides in o ma ion
o each o he i e indi idual ameni ies. S a is ics a e also shown sepa a ely be ween acancies ha pos a wage and acancies
ha do no pos a wage.
51 ILO Wo king Pape 136
No es: Panel (a) p esen s he es ima ed α coe icien o equa ion (2). Panel (b) p esen s he es ima ed (αLS, αHS) coe icien s o
equa ion (3). The dependen a iable is he log numbe o applica ions, and he key eg esso is he log pos ed wage, so coe -
icien s a e in e p e ed as c oss-sec ional elas ici ies. Column (1) shows esul s wi h no con ols in Panel (a) and includes a con-
ol o he occupa ional g oup in Panel (b). Column (2) excludes acancies ecei ing mo e han 1,000 applica ions and includes
2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and commissions,
schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Columns (3) and
(4) add 1-digi and 2-digi occupa ion ixed e ec s, espec i ely. Column (5) excludes he acancies a he op 5% o he pos ed
wage dis ibu ion. Column (6) only conside s acancies ha a e pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m
and includes i m ixed e ec s. Compa ed o Column (2) o Table 2, he sample size o he speci icia ion wi hou con ols is ma -
ginally smalle due o a ew acancies ecei ing ze o applica ions; esul s a e obus o using a Poisson model (see Appendix
Table B.2). S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
52 ILO Wo king Pape 136
No es: This able p esen s he es ima ed (αLS, αHS) coe icien s o equa ion (3). The dependen a iable is he log numbe o ap-
plica ions made by applican s wi h he cha ac e is ic depic ed in he column i le, and he key eg esso is he log pos ed wage,
so coe icien s a e in e p e ed as c oss-sec ional elas ici ies. Panel (a) p esen s esul s by gende , employmen s a us, and age.
Panel (b) p esen s esul s by educa ional a ainmen (wi hou e ia y educa ion, oca ional aining, and college deg ee) and
h ee ca ego ies o skills (cogni i e, socio-emo ional, and manual skills). Reg essions exclude acancies ecei ing mo e han
1,000 applica ions and include 2-digi indus y ixed e ec s, yea ixed e ec s, con ols o ad e ised ameni ies (indica o s o
bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de-
elopmen ), and 1-digi occupa ion ixed e ec s. Compa ed wi h Tables 2 and 3, disc epancies in sample sizes s em om acan-
cies ecei ing ze o applica ions om he a ious g oups; esul s a e obus o using a Poisson model (see Appendix Table B.7).
S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
53 ILO Wo king Pape 136
No es: This able p esen s he es ima ed αa coe icien s o equa ion (5). The dependen a iable is he log pos ed wage, and he
key eg esso s a e indica o s o ad e ised ameni ies, so coe icien s a e in e p e ed as c oss-sec ional semi-elas ici ies. Panel
(a) p esen s esul s om eg essions ha include an indica o a iable o ad e ising a leas one ameni y. Panel (b) p esen s
esul s om eg essions ha include i e indica o s associa ed wi h indi idual ameni ies (bonuses and commissions, schedule
lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Column (1) shows esul s
wi h no con ols. Column (2) excludes acancies ecei ing mo e han 1,000 applica ions and includes 2-digi indus y ixed e -
ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and commissions, schedule lexibili y, wo k
en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Columns (3) and (4) add 1-digi and 2-digi
occupa ion ixed e ec s, espec i ely. Column (5) excludes he acancies a he op 5% o he pos ed wage dis ibu ion. Column
(6) only conside s acancies ha a e pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m and includes i m ixed e -
ec s. S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
60 ILO Wo king Pape 136
No es: This able p esen s desc ip i e s a is ics o he es ima ion sample. The uni o obse a ion is a 2-digi indus y by 1-digi
occupa ion cell by calenda mon h. Panel (a) shows summa y s a is ics o all occupa ions combined. Panel (b) shows summa y
s a is ics o he lowe -skilled occupa ional g oup. Panel (c) shows summa y s a is ics o he highe -skilled occupa ional g oup.
No es: Panel (a) p esen s he es ima ed β coe icien o equa ion (11). Panel (b) p esen s he es ima ed β coe icien s in a mod-
el ha conside s in e ac ions wi h indica o s o lowe - and highe -skill occupa ional g oups. Reg essions include cell-by-e en
ixed e ec s, calenda mon h-by-1 digi -indus y ixed e ec s, and minimum wage changes no occu ing in Janua y o July (see
Panel (c) o Figu e 7). The dependen a iables, as depic ed in he column i les, include (in le els) he median numbe o applica-
ions, he mean numbe o applica ions, he o al numbe o acancies, and he o al numbe o openings. Repo ed elas ici ies
61 ILO Wo king Pape 136
a e compu ed by di iding he β-coe icien by he p e-e en a e age ou come wi hin ea ed cells, no malized by he log change
in minimum wage among ea ed cells. In each panel, Columns (1)-(4) p esen he main esul s, Columns (5)-(8) p esen esul s
ha exclude bin-by-e en window obse a ions whe e he median numbe o applica ions exceeded 750, and Columns (9)-(12)
exclude bin-by-e en window obse a ions o which he ou come is 0 mo e han 4 mon hs wi hin he e en window. S anda d
e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
No es: This able p esen s he es ima ed β coe icien o equa ion (11) in a model ha conside s in e ac ions wi h indica o s o
lowe and highe -skill occupa ional g oups. The dependen a iable is he numbe o applica ions made by applican s wi h he
cha ac e is ic depic ed in he column i le. Reg essions include cell-by-e en ixed e ec s, calenda mon h-by-1 digi -indus y
ixed e ec s, and minimum wage changes no occu ing in Janua y o July (see Panel (c) o Figu e 7). Repo ed elas ici ies a e
compu ed by di iding he β-coe icien by he p e-e en a e age ou come wi hin ea ed cells, no malized by he log change in
minimum wage among ea ed cells. Panel (a) p esen s esul s by gende , employmen s a us, and age. Panel (b) p esen s esul s
by educa ional a ainmen (wi hou e ia y educa ion, oca ional aining, and college deg ee) and h ee ca ego ies o skills (cog-
ni i e, socio-emo ional, and manual skills). S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
62 ILO Wo king Pape 136
A. Me hodology o C ea ing Va iables om F ee Tex
En ies
This appendix p o ides an o e iew o he me hodology used o c ea e a iables om ee ex
en ies, which is based on Adamczyk e al. (Fo hcoming) and Escude o e al. (Fo hcoming). We
i s discuss he c ea ion o skills a iables. We hen discuss he c ea ion o occupa ional codes.
Finally, we discuss he elici a ion o ad e ised non-wage ameni ies. Addi ional de ails can be
ound in he a o emen ioned pape s.
A.1 Skills
All skill- ela ed a iables a e based on he me hodology de eloped in Escude o e al. (Fo hcoming).
Thei app oach seeks o p o ide a comp ehensi e ep esen a ion o labo ma ke dynamics ac oss
di e se con ex s ha goes beyond o mal quali ica ion measu es by co e ing he skills demand-
ed by employe s in acancies and highligh ed by wo ke s in hei online p o iles. The au ho s
p opose a axonomy ha g oups skills in o h ee b oad ca ego ies: cogni i e, socio-emo ional,
and manual skills. In u n, each ca ego y is decomposed in o sub-ca ego ies, gi ing o m o a
o al o ou een subca ego ies. See Table A.1 o a desc ip ion o each ca ego y and subca ego-
y and he sou ces each ca ego y was de i ed om.
The axonomy is buil upon exis ing li e a u e om labo economics and psychology and has
been expanded o adap i o indi idual coun y con ex s, wi h a pa icula ocus on eme ging
and de eloping coun ies and online job boa d da a. The s a ing poin o he ca ego iza ion is
es ablished axonomies designed o classi ying skills in online da a wi hin he Uni ed S a es,
pa icula ly Deming and Kahn (2018). O he sou ces used include Heckman and Kau z (2012),
Ku eko á e al. (2016), and Deming and No ay (2020). The i s ex ension is o include manual
skills, which a e o en omi ed in U.S.- cen e ed analyses. Then, he second ex ension expands
he concep ual ounda ions ela ing o cogni i e and socio-emo ional skills o acili a e a mo e
comp ehensi e analysis o online da a beyond indi iduals wi h high o mal quali ica ions. To
achie e hese expansions, he axonomy included addi ional keywo ds and exp essions d awn
om a ious s udies (see Au o e al., 2003; Spi z-Oene , 2006; Almlund e al., 2011; Heckman
and Kau z, 2012; Ku eko á e al., 2016; He shbein and Kahn, 2018; A alay e al., 2020; Deming
and No ay, 2020), as well as he pilo exe cise o O-NET U uguay.
To elici he skill a iables in he BJ da a, he au ho s use a na u al language p ocessing (NLP)
me hodology ha in eg a es p e-p ocessing echniques wi h a ule-based classi ica ion app oach,
guided by he axonomy and he speci ic lis o keywo ds and ph ases associa ed wi h each o
he 14 subca ego ies. While some skills subca ego ies a e closely linked, he keywo ds and ex-
p essions used o cha ac e ize hem a e dis inc , allowing o he unique iden i ica ion o skills
in he da a. In a second s ep, his dic iona y is enla ged by including synonyms o he o iginal
wo ds ob ained h ough sc aping a hesau us websi e (www.wo d e e ence.com) and manually
checking he applicabili y o he e ie ed synonyms.
63 ILO Wo king Pape 136
No es: This able desc ibes he skills subca ego ies p esen ed in Table 1 o Escude o e al. (Fo hcoming), based on he conc e e
keywo ds used in he axonomy. ALM s ands o Au o e al. (2003), APST o A alay e al. (2020), DK o Deming and Kahn (2018),
DN o Deming and No ay (2020), HK o He shbein and Kahn (2018), HK o Heckman and Kau z (2012), KBHT o Ku eko á e
al. (2016), and S-O o Spi z-Oene (2006).
This p ocedu e leads o a o al o 741 dis inc skills, based on he unique keywo ds and exp essions.
The skills-subca ego y a iables a e hen c ea ed using he uns uc u ed ex da a p esen in bo h
he acancies pos ed by i ms and he job spells o applican s a ailable in hei BJ p o iles. F om
acancy da a, he au ho s elici he skills ha a e demanded by he acancy. F om employmen
his o y da a, he au ho s elici he skills applican s epo ha ing. The open- ex desc ip ions o -
e he mos iable app oach o c ea ing skills a iables, as hey con ain de ailed in o ma ion
on skills o all acancies (99.9%) and a majo i y o applican s’ job spells (68.5%). These open-
ex desc ip ions unde go a se ies o p e-p ocessing s eps using NLP echniques, including he
ansla ion o keywo ds and ph ases om English o Spanish, okeniza ion, ex no maliza ion,
lemma iza ion, n-g am c ea ion in he skills axonomy, and n-g am c ea ion in he acancy and
64 ILO Wo king Pape 136
applican s’ da a. These p ocesses a e employed o e o ma he ex da a in o a s uc u ed o -
ma ha acili a es he mapping wi h he skills dic iona y.
Finally, he skill a iables a e es ablished by allying keywo ds and ph ases linked o each skill ca -
ego y and subca ego y ha a e ound wi hin he ex . A skill is conside ed p esen i a leas one
o he keywo ds/ph ases om he dic iona y is iden i ied in he ex . Addi ionally, we calcula ed
he equency o keywo d occu ences o each skill and use his as an indica o o he deg ee o
in ensi y wi h which a pa icula skill used. See Escude o e al. (Fo hcoming) o addi ional de ails.
A.2 Occupa ions
The aw da a p o ided by BJ only classi ies acancies and applican s’ job spells in o ISCO-08 occu-
pa ion codes o a limi ed subse o he da a. This missing da a p oblem p e en s comp ehensi e
analyses a he occupa ion le el. To sol e his p oblem, Escude o e al. (Fo hcoming) employed
a simila me hodology as he one desc ibed abo ed o elici 1- and 2-digi occupa ional codes
o he ull sample o acancies and applican s’ job spells. To elici he occupa ions pos ed acan-
cies seek o ill, he au ho s le e aged ex ual in o ma ion om ou open- ex ields associa ed
wi h each acancy: job i le, job desc ip ion, equi ed le el o educa ion, and hie a chical le el o
he posi ion. To elici he occupa ion associa ed wi h applican s’ job spells, he au ho s used he
same in o ma ion, excep o job i les, which a e no a ailable as a sepa a e en y. This da a un-
de goes NLP p ocedu es simila o he ones used o elici ing skills a iables. The esul ing ex
is hen ca ego ized in o ISCO-08 codes h ough a h ee-s ep p ocess.
The i s s ep is analogous o he ule-based model employed o c ea e he skills a iables. The
au ho s employ a dic iona y o keywo ds, selec ed based on he mos equen ly used wo ds
and ph ases om he subse o bo h acancies and applican s’ job spells al eady classi ied by BJ
in o ISCO-08 occupa ional codes. The dic iona y used o igina es om he o icial ISCO-08 in e -
na ional classi ica ion. This exe cise p o ided he se o ules used o classi y he emaining job
i les in o occupa ional ca ego ies a he 2- digi le el. Addi ionally, he au ho s used in o ma ion
abou he educa ional le el o dis inguish be ween le els 2 and 3, deno ing indi iduals om he
same ield wi h ei he highe educa ion (le el 2) o any o he educa ion (le el 3). Simila ly, in o -
ma ion abou he hie a chical le el is used o iden i y manage s and di ec o s, placing hem in
le el 1 o he ISCO classi ica ion.
To enhance he pe o mance o he p ocedu e, he au ho s in oduced a machine lea ning al-
go i hm (in he o m o a p edic i e model) o assign codes o acancies and job spells ha we e
unclassi ied o o which he o iginal BJ assignmen signi ican ly di e s om he one ha e-
sul s om he algo i hm. This p ocess occu s in wo s eps. Fi s , he model is ained using he
al eady classi ied obse a ions o assign 1-digi ISCO codes. Second, addi ional in o ma ion om
applican s and acancies is inco po a ed in o a second p edic ion model o e ine he code as-
signmen a he 2-digi le el.
Based on a ious es s and sensi i i y analyses, he au ho s chose G adien Boos ing o code
1-digi and 2-digi occupa ions in he acancy da a, and Random Fo es o he applican s’ da a.
As a esul , 100% o acancies ha e an assigned 1-digi occupa ion code, and 94.8% o hem also
ha e a 2-digi occupa ion code. Fo applican s, all job spells wi h a ex desc ip ion we e clas-
si ied a he 1-digi le el, and 97.8% we e also classi ied a he 2-digi le el. See Escude o e al.
(Fo hcoming) o addi ional de ails.
65 ILO Wo king Pape 136
A.3 Ameni ies
The me hodology o iden i ying ad e ised ameni ies in uns uc u ed acancy da a is akin o
he one used o he skills a iables, and i is based on he p ocedu e ou lined in Adamczyk e
al. (Fo hcoming).
To begin, we de eloped a axonomy o ameni ies using he ela ed empi ical li e a u e as a s a -
ing poin and hen ex ending i o be e sui he U uguayan con ex and he na u e o online job
boa ds. As a i s sou ce, we ollow Maes as e al. (2023), who p o ide a lis o nine job a ibu es
based on he esul s om he Ame ican Wo ke Condi ions Su ey (AWCS). The su ey collec s
wo ke s’ assessmen s o nine wo k cha ac e is ics: schedule lexibili y, elecommu ing oppo uni-
ies, physical demands, pace o wo k, au onomy, paid ime o , wo king wi h o he s, job- aining
oppo uni ies, and impac on socie y. To b oaden he scope o he ca ego iza ion, we employ he
comp ehensi e ca ego iza ion p oposed by Sockin (2024), which o ganizes non-wage ameni ies
in 48 ca ego ies de i ed om he li e a u e using a opic-modeling machine lea ning algo i hm
implemen ed in he ex o ameni ies desc ip ions in U.S. employe -employee da a. Table A.2 lis s
addi ional sou ces we use o e ine he p ocedu e o speci ic ameni ies.
Gi en hese ca ego ies, we hen unde ook h ee s eps o b oaden he scope o he ca ego iza-
ion. Fi s , we eo ganized hese ca ego ies o align wi h acancy da a. The li e a u e p ima ily
elies on U.S. wo ke s’ e iews, bu no all ca ego ies a e pe inen o acancy da a because ce -
ain aspec s o a job may no be app op ia e o ad e ise in a pos ed acancy. Second, we sup-
plemen ed he lis o keywo ds and exp essions used in he li e a u e o cha ac e ize di e en
ameni ies, ailo ing hem o be e i he con ex o U uguay. Thi d, we in oduced an addi ional
ameni y ca ego y, “wo k equipmen and allowances,” o e lec he pos -pandemic eali y and o
inco po a e a ibu es o manual wo k ha may hold g ea e impo ance in U uguay and o he
global sou h coun ies ela i e o he U.S. economy.
We g ouped hese addi ional keywo ds in o i e b oad ca ego ies, esul ing in a o al o 16 ameni y
subca ego ies. In some cases, we adjus ed speci ic subca ego ies o ensu e he e was no o e lap
among he keywo ds and exp essions assigned o each subca ego y. The p ocess yielded a inal
se o 659 wo ds and exp essions, comp ising 357 o iginal e ms and 302 di e en e sions o
he same exp essions ( o mul iwo d exp essions). Table A.2 p o ides a lis o hese ca ego ies,
along wi h hei de ini ions and, whe e applicable, hei sou ces in he li e a u e. Mo e de ails
a e a ailable in Adamczyk e al. (Fo hcoming).
66 ILO Wo king Pape 136
To apply his dic iona y o he BJ acancy da a, bo h he e ms in he dic iona y and he ee ex
in o ma ion om he job ad e isemen s need o be o ma ed app op ia ely. The p ocess is
simila o he one used o c ea ing he skills a iables, albei wi h some modi ica ions. These
s eps encompass keywo d de ec ion, okeniza ion (di iding he ex in o single uni s o okens),
no maliza ion ( emo ing capi aliza ion and special cha ac e s), emo ing s op wo ds (including
excep ions o wo ds included in he dic iona y, such as ‘buen,’ ‘mucho,’ ‘g an,’ e c.), and
67 ILO Wo king Pape 136
No es: See Adamczyk e al. (Fo hcoming) o addi ional de ails. S s ands o Sockin (2024), SS o Sockin and Sockin (2019), BE
o Becke s e al. (2008), M o Maes as e al. (2023), SK o Simon and Kaes ne (2004), G o Glassdoo (2015), L o Libe (2016),
Q o Quinn (1974), WZ o Wasme and Zenou (2002); LB o Le Ba banchon e al. (2020), MP o Mas and Pallais (2017), PPB
o Pa k e al. (2021), BKS o B eza e al. (2017), HO o Holmlund (1983), HA o Haywa d e al. (1989), NM o Neuma k and
McLaughlin (2012), FP o File and Pe i (1988), HM o Hame mesh (1990), LLC o Lopes e al. (2014), AAZ o A hey e al. (2000),
AP o Acemoglu and Pischke (1999), PR o Pa en (1999), and BBB o Ba on e al. (1999).
lemma iza ion (associa ing di e en e sions o a wo d, such as conjuga ed e b o ms, wi h a
common oo wo d, like unconjuga ed e bs). Once he ex desc ibing acancies and he key-
wo ds and exp essions om he dic iona y a e in he same o ma , hey can be ma ched using
an NLP ule-based classi ica ion app oach o iden i y ameni ies in he acancy da a. Impo an ly,
his p ocess accommoda es a ia ion in wo d o de wi hin exp essions and allow ma ches wi h
up o one ex e nal wo d in be ween he wo ds om he dic iona y exp ession.
The algo i hm hen allies he occu ences o wo ds and exp essions om he dic iona y in he
acancy ex s and agg ega es hem o each b oade ameni y ca ego y. To simpli y he analysis,
his numbe is ans o med in o an indica o a iable o each ameni y subca ego y. The indica-
o akes he alue o one i any o he keywo ds o exp essions om ha pa icula subca ego-
y a e iden i ied in he job ad e .
Ou o he 86,062 acancies in he BJ da a,33 50.6% we e assigned a leas one o he 16 amen-
i ies. While some acancies lis up o eigh ameni ies, mo e han h ee-qua e s o hose wi h
assigned ameni ies ad e ise only one o wo. The mos equen ly ma ched subca ego ies a e
“human capi al de elopmen ” (22.6% o acancies), “wo king in eams” (18.7%), and “wo k en i-
onmen and impac on socie y” (17.9%). The lowes numbe o ma ches is ound o “ e i emen
con ibu ions” (33 ma ches, o 0.04% o he obse a ions) and “heal h insu ance” (38 ma ches,
0.04% o he obse a ions), possibly because hese a e legally manda ed bene i s ha may no
wa an explici men ion in he U uguayan con ex .
33 The sample size men ioned he e sligh ly de ia es om sample sizes men ioned in Sec ion 1. The il e s o c ea e ou inal analysis
sample a e no ye applied. Ins ead, he sample men ioned he e excludes a ew acancies wi h blank o meaningless job ex de-
sc ip ions (see A alay e al., 2020). In ou main analysis, hese acancies a e coded as ha ing ze o ameni ies.
68 ILO Wo king Pape 136
No es: Analysis done on he base o 86,062 acancies. See Adamczyk e al. (Fo hcoming) o mo e de ails.
A comp ehensi e lis o he sha e o acancies wi h he assigned ameni y can be ound in Table A.3.
Rega ding indi idual keywo ds, “ abaja en equipo ( eamwo k)”, which belongs o he wo king
in eams ca ego y, is he mos equen ly ma ched (wi h a o al o 12,579 ma ches). Typically, in
each subca ego y, a ew keywo ds domina e he majo i y o ma ches, wi h o he e ms making
smalle con ibu ions. Figu e A.1 displays wo d clouds o all ameni y subca ego ies, whe e he
size o a wo d co esponds o i s sha e o ma ches wi hin ha subca ego y. I is impo an o
no e ha he use o keywo ds and exp essions o c ea e ameni y a iables unde wen se e al
ounds o manual e i ica ion o ensu e ha wo ds and exp essions we e con ex ually accu a e.
This e i ica ion was manually conduc ed o a sample o acancies o all wo ds appea ing a
he op o he ma ches o each subca ego y, as well as o a selec ion o o he wo ds deemed
necessa y by he au ho s o his s udy and Adamczyk e al. (Fo hcoming). While he p ocedu e
was pe o med o he comple e lis o 16 ameni ies, in he analysis, we ocus on he 5 ameni-
ies wi h he highes p e alence, namely, bonuses and commissions, schedule lexibili y, wo k
en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen . The es o
he ameni ies a e ound o be ela i ely in equen and also, in some cases, o e addi ional in-
e p e a ion challenges.
69 ILO Wo king Pape 136
No es: Au ho s’ elabo a ion based on Adamczyk e al. (Fo hcoming). The analysis is based on he ull sample o 86,062 acan-
cies. The wo ds displayed in he wo d clouds ep esen he o iginal wo ds used o de ine ameni ies. Fo he ma ching p ocess,
hese o iginal wo ds we e lemma ized o acili a e he ma ching. The inclusion o o iginal wo ds in he igu e is o cla i y and
ease o unde s anding.
76 ILO Wo king Pape 136
No es: This igu e plo s he dis ibu ion o he iming o applica ions o acancies ela i e o he opening da e. “Dis ance” e e s
o he days elapsed since he opening o he acancy.
No es: This able shows summa y s a is ics o he ameni ies ad e ised in ou inal sample o acancies (see Sec ion 2 o de ails
on he sample es ic ions). Ad e ised ameni ies we e elici ed ollowing Adamczyk e al. (Fo hcoming). The able de ails whe eas
77 ILO Wo king Pape 136
acancies ad e ise each indi idual ameni y. This able conside s he ull lis o ameni ies discussed in Appendix A. S a is ics a e
also shown sepa a ely be ween acancies ha pos a wage and acancies ha do no pos a wage.
No es: Panel (a) p esen s he
α
coe icien o equa ion (2) es ima ed using a Poisson model. I β deno es he poin es ima e, he
elas ici y is eco e ed as exp(β) -1. The s anda d e o is es ima ed using he Del a me hod. Panel (b) p esen s he es ima ed
(αLS, αHS) coe icien s o equa ion (3). The dependen a iable is he log numbe o applica ions, and he key eg esso is he log
pos ed wage, so coe icien s a e in e p e ed as c oss-sec ional elas ici ies. Column (1) shows esul s wi h no con ols in Panel (a)
and includes a con ol o he occupa ional g oup in Panel (b). Column (2) excludes acancies ecei ing mo e han 1,000 applica-
ions and includes 2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses
and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ).
Columns (3) and (4) add 1-digi and 2-digi occupa ion ixed e ec s, espec i ely. Column (5) excludes he acancies a he op 5%
o he pos ed wage dis ibu ion. Column (6) only conside s acancies ha a e pos ed by i ms ha pos a leas 10 acancies in he
BJ pla o m and includes i m ixed e ec s. S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
78 ILO Wo king Pape 136
This able p esen s he es ima ed α coe icien o equa ion (2) using di e en de ini ions o pos ed wage. The dependen a iable
is he log numbe o applica ions, and he key eg esso is he log pos ed wage, so coe icien s a e in e p e ed as c oss-sec ional
elas ici ies. Panel (a) conside s he midpoin o he sala y ange. Panel (b) conside s he midpoin o he sala y ange, excluding
acancies whose ange exceeds he 50% o he midpoin . Panel (c) conside s he maximum o he sala y ange. Column (1) shows
esul s wi h no con ols. Column (2) excludes acancies ecei ing mo e han 1,000 applica ions and includes 2-digi indus y ixed
e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and commissions, schedule lexibili y,
wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Columns (3) and (4) add 1-digi and
2-digi occupa ion ixed e ec s, espec i ely. Column (5) excludes he acancies a he op 5% o he pos ed wage dis ibu ion.
Column (6) only conside s acancies ha a e pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m and includes i m
ixed e ec s. S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
79 ILO Wo king Pape 136
Panel (a) p esen s he es ima ed α coe icien o equa ion (2) o lowe -skill occupa ions. Panel (b) p esen s he es ima ed
α
co-
e icien o equa ion (2) o highe -skill occupa ions. The dependen a iable is he log numbe o applica ions, and he key e-
g esso is he log pos ed wage, so coe icien s a e in e p e ed as c oss-sec ional elas ici ies. Wi hin each panel, Columns (1)-(4)
show esul s o indi idual occupa ions in eg essions ha exclude acancies ecei ing mo e han 1,000 applica ions and include
2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and commissions,
schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Wi hin each pan-
el, Columns (5)-(7) show esul s o he b oad occupa ion g oups. Column (5) does no include occupa ion ixed e ec s, while
Columns (6) and (7) include 1-digi and 2-digi occupa ion ixed e ec s, espec i ely. S anda d e o s ( epo ed in pa en heses)
a e clus e ed a he 2-digi indus y le el.
80 ILO Wo king Pape 136
No es: This able shows summa y s a is ics o he p esence o equi emen s (in e ms o o mal quali ica ions, o eign language,
o skills) in ou inal sample o acancies (see Sec ion 1 o de ails on he sample es ic ions). Panel (a) conside s all acancies.
Panel (b) es ic s o acancies ha pos a wage. Wi hin each panel, s a is ics a e shown o all acancies, acancies a ached o
lowe -skill occupa ions, and acancies a ached o highe -skill occupa ions.
81 ILO Wo king Pape 136
82 ILO Wo king Pape 136
No es: This able p esen s he es ima ed (αLS, αHS) coe icien s o equa ion (3). The dependen a iable is he log numbe o ap-
plica ions, and he key eg esso is he log pos ed wage, so coe icien s a e in e p e ed as c oss-sec ional elas ici ies. Panel (a)
conside s acancies ha pos a leas one o mal quali ica ion equi emen (educa ion and/o language). Panel (b) conside s
acancies ha pos a leas one skill equi emen (cogni i e, socio-emo ional, and/o manual). Panels (c) and (d) p esen esul s
wi hou hese equi emen s, espec i ely. Column (1) includes a con ol o he occupa ional g oup. Column (2) excludes acan-
cies ecei ing mo e han 1,000 applica ions and includes 2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad-
e ised ameni ies (indica o s o bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king
in eams, and human capi al de elopmen ). Columns (3) and (4) add 1-digi and 2-digi occupa ion ixed e ec s, espec i ely.
Column (5) excludes he acancies a he op 5% o he pos ed wage dis ibu ion. Column (6) only conside s acancies ha a e
pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m and includes i m ixed e ec s. S anda d e o s ( epo ed in
pa en heses) a e clus e ed a he 2-digi indus y le el.
83 ILO Wo king Pape 136
No es: This able p esen s he (αLS, αHS) coe icien s o equa ion (3) es ima ed using a Poisson model. I β deno es he poin es i-
ma e, he elas ici y is eco e ed as exp(β) -1. The s anda d e o is es ima ed using he Del a me hod. The dependen a iable is
he log numbe o applica ions made by applican s wi h he cha ac e is ic depic ed in he column i le, and he key eg esso is
he log pos ed wage, so coe icien s a e in e p e ed as c oss-sec ional elas ici ies. Panel (a) p esen s esul s by gende , employ-
men s a us, and age. Panel (b) p esen s esul s by educa ional a ainmen (wi hou e ia y educa ion, oca ional aining, and
college deg ee) and h ee ca ego ies o skills (cogni i e, socio-emo ional, and manual skills). Reg essions exclude acancies e-
cei ing mo e han 1,000 applica ions and include 2-digi indus y ixed e ec s, yea ixed e ec s, con ols o ad e ised amen-
i ies (indica o s o bonuses and commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and
human capi al de elopmen ), and 1-digi occupa ion ixed e ec s. S anda d e o s ( epo ed in pa en heses) a e clus e ed a
he 2-digi indus y le el.
84 ILO Wo king Pape 136
No es: This able p esen s he es ima ed (αF, αE, αA, αV, αC, αCS, αSK, αMS) coe icien s o equa ion (4). The dependen a iable is he
log pos ed wage o he applica ion, and he key eg esso s a e indi idual cha ac e is ics o he applican . Column (1) shows e-
sul s wi h no con ols. Column (2) excludes acancies ecei ing mo e han 1,000 applica ions and includes 2-digi indus y ixed
e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and commissions, schedule lexibili y,
wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ). Columns (3) and (4) add 1-digi and
2-digi occupa ion ixed e ec s, espec i ely. Column (5) excludes he acancies a he op 5% o he pos ed wage dis ibu ion.
Column (6) only conside s acancies ha a e pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m and includes i m
ixed e ec s. S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
85 ILO Wo king Pape 136
No es: This able p esen s he es ima ed (
αLS
a
,
αHS
a
) coe icien s o equa ion (6). The dependen a iable is he log pos ed wage,
and he key eg esso s a e indica o s o ad e ised ameni ies, so coe icien s a e in e p e ed as c oss-sec ional semi-elas ici ies.
Column (1) includes a con ol o he occupa ional g oup. Column (2) excludes acancies ecei ing mo e han 1,000 applica ions
and includes 2-digi indus y ixed e ec s, yea ixed e ec s, and con ols o ad e ised ameni ies (indica o s o bonuses and
commissions, schedule lexibili y, wo k en i onmen /impac on socie y, wo king in eams, and human capi al de elopmen ).
Columns (3) and (4) add 1-digi and 2-digi occupa ion ixed e ec s, espec i ely, in exchange o he occupa ional g oup indica-
o . Column (5) excludes he acancies a he op 5% o he pos ed wage dis ibu ion. Column (6) only conside s acancies ha
a e pos ed by i ms ha pos a leas 10 acancies in he BJ pla o m and includes i m ixed e ec s. S anda d e o s ( epo ed
in pa en heses) a e clus e ed a he 2-digi indus y le el.
92 ILO Wo king Pape 136
No es: This able p esen s he es ima ed β coe icien o equa ion (11) in a model ha conside s in e ac ions wi h indica o s o
lowe and highe -skill occupa ional g oups. Reg essions include cell-by-e en ixed e ec s, calenda mon h-by-1 digi -indus y
ixed e ec s, and minimum wage changes no occu ing in Janua y o July (see Panel (c) o Figu e 7). The dependen a iables,
as depic ed in he column i les, include (in le els) he sha e o acancies equi ing oca ional aining, he sha e o acancies e-
qui ing a college deg ee, he sha e o acancies equi ing o eign language knowledge, he sha e o acancies equi ing cogni i e
skills, he sha e o acancies equi ing socio-emo ional skills, and he sha e o acancies equi ing manual skills. Repo ed elas-
ici ies a e compu ed by di iding he β-coe icien by he p e-e en a e age ou come wi hin ea ed cells, no malized by he log
change in minimum wage among ea ed cells. S anda d e o s ( epo ed in pa en heses) a e clus e ed a he 2-digi indus y le el.
93 ILO Wo king Pape 136
Appendix Bibliog aphy
Acemoglu, Da on and J. S. Pischke, “The s uc u e o wages and in es men in gene al ain-
ing,” Jou nal o Poli ical Economy, 1999, 107 (3), 539–572.
Adamczyk, Willian, Isau e Delapo e, and Ve ´onica Escude o, “Job Ameni ies in De eloping
Coun ies: an analysis o online labou ma ke s,” ILO Wo king Pape , Fo hcoming.
Almlund, Ma hilde, Angela Lee Duckwo h, James Heckman, and Tim Kau z, “Chap e 1
- Pe sonali y Psychology and Economics,” in E ic A. Hanushek, S ephen Machin, and Ludge
Woessmann, eds., Handbook o he Economics o Educa ion, Vol. 4, Else ie , 2011, pp. 1–181.
A alay, Enghin, Phai Phong hieng ham, Sebas ian So elo, and Daniel Tannenbaum, “The
E olu ion o Wo k in he Uni ed S a es,” Ame ican Economic Jou nal: Applied Economics, 2020, 12
(2), 1–34.
A hey, Susan, Ch is ophe A e y, and Pe e Zemsky, “Men o ing and Di e si y,” Ame ican
Economic Re iew, 2000, 90 (4), 765–786.
Au o , Da id, F ank Le y, and Richa d Mu nane, “The skill con en o ecen echnological
change: An empi ical explo a ion,” The Qua e ly Jou nal o Economics, 2003, 118 (4), 1279–1333.
Ba banchon, Thomas Le, Roland Ra helo , and Alexand a Roule , “Gende Di e ences in
Job Sea ch: T ading o Commu e agains Wage,” The Qua e ly Jou nal o Economics, 2020, 136
(1), 381–426.
Ba on, John M., Ma k C. Be ge , and Dan A. Black, “Do Wo ke s Pay o On-The-Job T aining?,”
The Jou nal o Human Resou ces, 1999, 34 (2), 235.
Becke s, Debby G.J., Dimi i Van De Linden, Pe e G.W. Smulde s, Michiel A.J. Kompie , Toon
W. Ta is, and Sabine A.E. Geu s, “Volun a y o in olun a y? Con ol o e o e ime and ewa ds
o o e ime in ela ion o a igue and wo k sa is ac ion,” Wo k & S ess, 2008, 22 (1), 33–50.
B eza, Emily, Sup ee Kau , and Yogi a Shamdasani, “The mo ale e ec s o pay inequali y,”
The Qua e ly Jou nal o Economics, 2017, 133 (2), 611–663.
Deming, Da id and Kadeem No ay, “Ea nings Dynamics, Changing Job Skills, and STEM Ca ee s,”
Qua e ly Jou nal o Economics, 2020, 135 (8), 1965–2005.
and Lisa B. Kahn, “Skill equi emen s ac oss i ms and labo ma ke s: E idence om job pos -
ings o p o essionals,” Jou nal o Labo Economics, 2018, 36 (S1), S337–S369.
Escude o, Ve ónica, Hannah Liepmann, and Ana Podjanin, “Using Online Vacancy and Job
Applican s’ Da a o S udy Skills Dynamics,” Resea ch in Labo Economics, Fo hcoming.
File , Randall K. and Pe e A. Pe i, “A Job-Cha ac e is ics Theo y o Re i emen ,” The Re iew o
Economics and S a is ics, 1988, 70 (1), 123.
Glassdoo , “4 in 5 Employees Wan Bene i s o Pe ks Mo e Than a Pay Raise; Glassdoo Employmen
Con idence Su ey (Q3 2015) - Glassdoo US,” 2015.
94 ILO Wo king Pape 136
Hame mesh, Daniel S., “Shi king o P oduc i e Schmoozing: Wages and he Alloca ion o Time
a Wo k,” Indus ial and Labo Rela ions Re iew, 1990, 43 (3), 121S.
Haywa d, Ma k D., William R. G ady, Melissa A. Ha dy, and Da id Somme s, “Occupa ional
in luences on e i emen , disabili y, and dea h,” Demog aphy, 1989, 26 (3), 393–409.
Heckman, James and Tim Kau z, “Ha d e idence on so skills,” Labou Economics, 2012, 19 (4),
451–464.
He shbein, B a and Lisa B. Kahn, “Do ecessions accele a e ou ine-biased echnological
change? E idence om acancy pos ings,” Ame ican Economic Re iew, 2018, 108 (7), 1737–72.
Holmlund, Be il, “Pay oll Taxes and Wage In la ion: The Swedish Expe ience,” The Scandina ian
Jou nal o Economics, 1983, 85 (1), 1.
Ku eko á, Lucia Mý na, Mi osla Bebla ý, Co ina Hai a, and Anna-Elisabe h Thum, “Employe s’
skill p e e ences ac oss Eu ope: be ween cogni i e and non-cogni i e skills,” Jou nal o Educa ion
and Wo k, 2016, 29 (6), 662–687.
Libe , Kelsey, “Employee Bene i s S udy: The Cos and Value o Employee Pe ks,” 2016.
Lopes, Helena, Sé gio Lagoa, and Te esa Calapez, “Wo k au onomy, wo k p essu e, and job
sa is ac ion: An analysis o Eu opean Union coun ies,” The Economic and Labou Rela ions Re iew,
2014, 25 (2), 306–326.
Maes as, Nicole, Ka hleen J Mullen, Da id Powell, Till Von Wach e , and Je ey B Wenge ,
“The alue o wo king condi ions in he Uni ed S a es and he implica ions o he s uc u e o
wages,” Ame ican Economic Re iew, 2023, 113 (7), 2007–2047.
Mas, Alexand e and Amanda Pallais, “Valuing al e na i e wo k a angemen s,” Ame ican Economic
Re iew, 2017, 107 (12), 3722–59.
Neuma k, Da id and Joanne Song McLaughlin, “Ba ie s o La e Re i emen : Inc eases in he
Full Re i emen Age, Age Disc imina ion, and he Physical Challenges o Wo k,” Wo king Pape , 2012.
Pa en , Daniel, “Wages and Mobili y: The Impac o Employe -P o ided T aining,” Jou nal o Labo
Economics, 1999, 17 (2), 298–317.
Pa k, R. Jisung, No a Pank a z, and A. Pa ick Beh e , “Tempe a u e, Wo kplace Sa e y, and
Labo Ma ke Inequali y,” IZA Wo king Pape , 2021, (14560).
Quinn, Robe P., Job Sa is ac ion: Is The e a T end?, Uni e si y o Michigan Lib a y, 1974.
Simon, Kosali Ilaype uma and Robe Kaes ne , “Do Minimum Wages A ec Non-Wage Job
A ibu es? E idence on F inge Bene i s,” ILR Re iew, 2004, 58 (1), 52–70.
Sockin, Jason, “Show Me he Ameni y: A e Highe -Paying Fi ms Be e All A ound?,” Wo king
Pape , 2024.
and Michael Sockin, “A pay scale o hei own: Gende di e ences in a iable pay,” Wo king
Pape , 2019, (3512598).
95 ILO Wo king Pape 136
Spi z-Oene , Alexand a, “Technical Change, Job Tasks, and Rising Educa ional Demands: Looking
ou side he Wage S uc u e,” Jou nal o Labo Economics, 2006, 24 (2), 235–270.
Wasme , E ienne and Y es Zenou, “Does Ci y S uc u e A ec Job Sea ch and Wel a e?,” Jou nal
o U ban Economics, 2002, 51 (3), 515–541.
96 ILO Wo king Pape 136
Acknowledgemen s
We hank Ma celo Be golo, Sydnee Caldwell, Da id Ca d, Rod igo Ceni, Zoe Cullen, Jenni e
Hun , Pa ick Kline, El a López Mou elo, and semina pa icipan s a he Columbia Junio Mic o-
Mac o Labo Con e ence, he 6 h IDSC o IZA Wo kshop (“Ma ching Wo ke s and Jobs Online”),
P ince on Uni e si y, he 8 h RDW Con e ence (ILO), UC Be keley, and Uni e sidad de la República
de Mon e ideo o e y help ul commen s and sugges ions. We especially hank Ma celo Be golo,
Paula Ca asco, Rod igo Ceni, Nicolás G au, and Cecilia Pa ada o gene ously sha ing aw da a
on digi ized collec i e ba gaining ag eemen s, and Willian Bosche i Adamczyk o con ibu ing o
he analysis o non-wage ameni ies. The esponsibili y o opinions exp essed in his a icle es s
solely wi h i s au ho s, and publica ion does no cons i u e an endo semen by he In e na ional
Labou O ice o he opinions exp essed in i . O he usual disclaime s apply.
Resea ch Depa men (RESEARCH)
In e na ional Labou O ganiza ion
Rou e des Mo illons 4
1211 Gene a 22
Swi ze land
T +41 22 799 6530
esea [email p o ec ed]
www.ilo.o g/ esea ch
XAd ancing social jus ice, p omo ing decen wo k
The In e na ional Labou O ganiza ion is he Uni ed Na ions agency o he wo ld o wo k. We b ing oge he go e nmen s, employe s and wo ke s
o imp o e he wo king li es o all people, d i ing a human-cen ed app oach o he u u e o wo k h ough employmen c ea ion, igh s a wo k,
social p o ec ion and social dialogue.
I S B N 9789220413920
9HSTCMA*ebdjca+
9 789220 413920
Con ac de ails
