Lee, Myeongju; Lee, Gyeonghwan; Lim, Kihoon; Moon, Hyunchul; Doh, Jaehyeok
A icle
Machine lea ning-based causali y analysis o human
esou ce p ac ices on i m pe o mance
Adminis a i e Sciences
P o ided in Coope a ion wi h:
MDPI – Mul idisciplina y Digi al Publishing Ins i u e, Basel
Sugges ed Ci a ion: Lee, Myeongju; Lee, Gyeonghwan; Lim, Kihoon; Moon, Hyunchul; Doh,
Jaehyeok (2024) : Machine lea ning-based causali y analysis o human esou ce p ac ices on i m
pe o mance, Adminis a i e Sciences, ISSN 2076-3387, MDPI, Basel, Vol. 14, Iss. 4, pp. 1-20,
h ps://doi.o g/10.3390/admsci14040075
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/320897
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/
Ci a ion: Lee, Myeongju,
Gyeonghwan Lee, Kihoon Lim,
Hyunchul Moon, and Jaehyeok Doh.
2024. Machine Lea ning-Based
Causali y Analysis o Human
Resou ce P ac ices on Fi m
Pe o mance. Adminis a i e Sciences
14: 75. h ps://doi.o g/10.3390/
admsci14040075
Recei ed: 29 Decembe 2023
Re ised: 5 Ap il 2024
Accep ed: 6 Ap il 2024
Published: 9 Ap il 2024
Copy igh : © 2024 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
adminis a i e
sciences
A icle
Machine Lea ning-Based Causali y Analysis o Human Resou ce
P ac ices on Fi m Pe o mance
Myeongju Lee 1, Gyeonghwan Lee 2, Kihoon Lim 3, Hyunchul Moon 3and Jaehyeok Doh 4,*
1School o Business Adminis a ion, Gyeongsang Na ional Uni e si y, Jinju-si 52725, Republic o Ko ea;
[email p o ec ed]
2College o Business Adminis a ion, Dong-A Uni e si y, Busan 49315, Republic o Ko ea; [email p o ec ed]
3School o Mechanical and Ma e ial Con e gence Enginee ing, Gyeongsang Na ional Uni e si y,
Jinju-si 52725, Republic o Ko ea; [email p o ec ed] (K.L.); [email p o ec ed] (H.M.)
4School o Ae ospace Enginee ing, Gyeongsang Na ional Uni e si y, Jinju-si 52828, Republic o Ko ea
*Co espondence: [email p o ec ed]
Abs ac : An o ganiza ion’s human esou ce managemen p ac ices a e essen ial o i s compe i i e
ad an age. This s udy speci ically examined human esou ce (HR) p ac ices ha p edic co po a e
pe o mance (employee u no e and i m sales) based on a backp opaga ion neu al ne wo k (BPN)-
based causali y analysis. This s udy aims o es how o op imize human esou ce p ac ices o imp o e
o ganiza ional pe o mance. This s udy elucida ed he e ec o HR p ac ices and o ganiza ional-
le el ac o s on p edic ing employee u no e and i m sales. The BPN-based causali y analysis
e ealed he ela i e impo ance o explana o y a iables on i m pe o mance. To es he model, i
employed he Human Capi al Co po a e Panel open da a on Ko ean companies’ HR p ac ices and
o he cha ac e is ics. The analysis iden i ies causal ela ionships be ween speci ic HR p ac ices and
i m pe o mance. The esul s show ha compensa ion- ela ed HR p ac ices a e mos in luen ial in
p edic ing i m sales and employee u no e . Mo eo e , aining- ela ed HR p ac ices we e modes ,
and alen acquisi ion and pe o mance managemen p ac ices had ela i ely weak e ec s on he
wo ou comes. The s udy p o ides insigh s in o how human esou ce p ac ices can be op imized
o imp o e i m pe o mance and enhance o ganiza ional e ec i eness. The indings o his s udy
con ibu e o he g owing body o esea ch on he use o machine lea ning in HR managemen and
sugges p ac ical implica ions o manage s’ insigh s o op imize HR p ac ices.
Keywo ds: BPN-based causali y analysis; i m pe o mance; human co po a e capi al panel; human
esou ce managemen ; machine lea ning
1. In oduc ion
Human esou ces (HR) is an impo an asse in business o ganiza ions because, al-
hough compe i i e ad an age based on HR canno be easily achie ed, he ad an age
p esen s a high ba ie o imi a ion when achie ed, enabling i ms o sus ain hei com-
pe i i e ad an ages (Ba ney 1991;Huselid 1995;Becke 1998). Thus, schola s ha e paid
a en ion o he e ec o a ious HR p ac ices on i m pe o mance. This line o s udy,
s a egic HR managemen (SHRM), has mainly sugges ed he uni e sal e ec s o high-
pe o mance wo k sys ems (HPWSs, a bundle o e ec i e HR p ac ices) on i m-le el
ou comes, such as u no e a e, labo p oduc i i y, inno a ion, and inancial pe o mance.
Me a-analyses ha e suppo ed he hypo hesis ha HPWSs posi i ely co ela e wi h
i m pe o mance (Sub amony 2009). Mo eo e , SHRM esea ch has obse ed media o s
o mode a o s be ween HPWSs and i m pe o mance, such as o ganiza ional cul u e
(
Chow 2012
), o ganiza ional communica ion (Lee e al. 2017), human capi al (
Jiang e al. 2012
),
and employee commi men (Takeuchi e al. 2009). Thus, he gene al sugges ion o SHRM
is ha a ious HR p ac ices posi i ely a ec i ms in many aspec s, he eby imp o ing
o ganiza ional e ec i eness. Following a p e ious esea ch s eam, his s udy explo ed he
Adm. Sci. 2024,14, 75. h ps://doi.o g/10.3390/admsci14040075 h ps://www.mdpi.com/jou nal/admsci
Adm. Sci. 2024,14, 75 2 o 20
ela ionship be ween HR p ac ices and o ganiza ional e ec i eness. In pa icula , i ocused
on employee u no e and i m sales. In addi ion, i explo ed he ela i e impo ance and
di ec ion o he e ec s o HR p ac ices in p edic ing employee u no e and i m sales
using machine-lea ning (ML) echniques. In his espec , his s udy di e s om p e ious
SHRM s udies.
Fi s , his s udy ocused on he e ec o di e en g oups o HR p ac ices on o ganiza-
ional e ec i eness. P e ious SHRM s udies in es iga ed he e ec s o HPWSs measu ed by
he agg ega ed numbe o implemen ed HR p ac ices iden i ied as a componen o HPWSs
(W igh e al. 2005). This app oach assumes ha all HR p ac ices ha e a simila e ec on
a pa icula ou come, such as employee abili y and mo i a ion, and ha i ms can use
di e en HR p ac ices in e changeably as long as hose p ac ices a e componen s o HPWSs.
Howe e , each HR p ac ice has di e en cha ac e is ics; hus, he expec ed e ec s o each
p ac ice on o ganiza ional ou comes can di e (Jiang e al. 2012). The e o e, i examined he
e ec o each g oup o HR p ac ices, no he agg ega ed numbe o HR p ac ices as a whole.
HR p ac ices can gene ally be ca ego ized in o alen acquisi ion, aining, de elopmen ,
pe o mance managemen , and compensa ion (Combs e al. 2006). Using publicly a ailable
da a on HR p ac ices and o he cha ac e is ics o Ko ean i ms ( he Human Capi al Co po-
a e Panel (HCCP)), his s udy classi ied HR p ac ices, which a e widely used in Ko ean
i ms, in o (1) alen acquisi ion p ac ice, (2) aining p ac ice, (3) de elopmen p ac ice,
(4) pe o mance managemen p ac ice, (5) compensa ion p ac ice, and (6) inge bene i s
and explo ed hei ela i e impo ance in p edic ing co po a e ou comes. This app oach
enabled us o co ec ly iden i y he ela i e impo ance o each ype o HR p ac ice.
Second, his s udy explo ed he complex ela ionships be ween each ype o HR
p ac ice and employee u no e and i m sales using ML echniques. He e, ML, which
has been widely in oduced in o he social sciences, was employed o e alua e he ela i e
impo ance and di ec ion o he e ec s o di e en HR p ac ices. ML can disco e pa e ns
om big da a and is an in o ma ion ex ac ion echnique ha can simul aneously handle a
la ge numbe o a iables (Has ie e al. 2009). A p e ious s udy a gued ha ML should
be mo e widely used in managemen s udies o in es iga e he uniden i ied e ec s o
p e iously known independen a iables o heo y building (Choudhu y e al. 2021).
Compa ed wi h he eg ession app oach, ML has se e al ad an ages in e ms o he
aims o he p esen s udy. Dissimila o he eg ession echnique, which is equen ly
used o es he di ec ion o he e ec o independen a iables p ede ined by heo e ical
a gumen s, ML does no cons ain he o m o he e ec s o independen a iables on
he ou comes. Thus, his echnique acili a es he simul aneous disco e y o complex
ela ionship pa e ns be ween a ious explana o y and dependen a iables. This s udy is
ele an o p ac i ione s and schola s because p e ious s udies sugges ha he ela ionship
be ween a pa icula HR p ac ice and o ganiza ional e ec i eness can be nonlinea and
highly complex (Yan e al. 2022;Chadwick 2007). Thus, ou app oach p o ides p ac ical
implica ions o p ac i ione s building HR sys ems by in oducing and implemen ing new
HR p ac ices and enables schola s o disco e new esea ch oppo uni ies.
Thi d, his s udy examined he in luence o o he independen a iables a he em-
ployee and o ganiza ional le els. Al hough his s udy aimed o elucida e he in luence o
HR p ac ices on o ganiza ional e ec i eness, i included o he p edic o s, such as employee
a i udes owa d hei jobs and o ganiza ions, o ganiza ional cul u e, and o ganiza ional
cha ac e is ics, as explana o y a iables. Conside ing ha hese a iables ha e been ec-
ognized as signi ican p edic o s o o ganiza ional e ec i eness, he inclusion helped o
unde s and HR p ac ices’ ela i e impo ance accu a ely. Mo eo e , addi ional indepen-
den a iables ep esen ing changes inside and ou side o ganiza ions can p o ide manage s
wi h addi ional insigh s in o when manage s should e ise exis ing HR p ac ices. Changes
in demand and echnology di ec ly a ec o ganiza ional e ec i eness and equi e manage s
o ealign HR p ac ices wi h he change. Thus, unde s anding he complex ela ionship
be ween changes inside and ou side he o ganiza ion and o ganiza ional e ec i eness
Adm. Sci. 2024,14, 75 3 o 20
enables manage s o de e mine when hey should e ise hei HR sys ems by in oducing
new HR p ac ices o abolishing exis ing p ac ices.
Compa ed wi h p e ious SHRM s udies, his s udy has he ollowing wo con ibu-
ions: While p e ious SHRM s udies ha e es ed he e ec o bundles o HR p ac ices on
o ganiza ional pe o mance, his s udy con ibu es o examining each e ec o HR p ac ices.
In addi ion, unlike eg ession analysis, which was mainly used in p e ious s udies, ML
analysis con ibu es o unde s anding complexi y by analyzing a ious o ganiza ional
a iables a ec ing o ganiza ional pe o mance and each HR p ac ice in one model.
This s udy desc ibes he ollowing h ee esea ch ques ions.
Resea ch Ques ions 1. Wha a e he signi ican de e minan s o i m pe o mance
( u no e a e, sales)?
Resea ch Ques ion 2. Wha is he o de o impo ance o HR p ac ices among he
de e minan s o i m pe o mance ( u no e a e, sales)?
Resea ch Ques ion 3. Wha is he pa e n o he in luence o HR p ac ices on i m
pe o mance ( u no e a e, sales)?
In summa y, his s udy aimed o unde s and he e ec s o HR p ac ices and indi idual
and o ganiza ional le el ac o s in p edic ing employee u no e and i m sales. I in es i-
ga ed he ela i e impo ance and di ec ion o he e ec s o explana o y a iables using
ML. The es o his a icle is o ganized as ollows. The nex sec ion e iews he li e a u e,
and Sec ion 3explains he HR da a sou ce, independen a iables, and cha ac e is ics o
he in eg a ed da ase . Sec ion 4discusses he ML p ocess and o e s he esul s. Sec ion 5
discusses he main indings, implica ions, limi a ions, and u u e esea ch di ec ions.
2. Li e a u e Re iew
Human esou ce managemen (HRM) p ac ices ha e been ound o play a c ucial
ole in shaping a i m’s pe o mance. While he de ini ion o HRM a ies, encompassing
aspec s like HR depa men e ec i eness, indi idual p ac ices, o en i e sys ems, a key
akeaway om Boselie e al.’s (2005) me a-analysis o 104 s udies is ha many iew HRM
as a collec ion o in e connec ed HR sys ems o p ac ices. High-pe o mance wo k sys ems
(HPWS), o example, bundle p ac ices like ec ui men , selec ion, compensa ion, aining,
and job design, posi i ely impac a i m’s pe o mance (Ge ha 2007).
Resea ch on HRM sys ems and i m pe o mance has inc easingly explo ed he mech-
anisms. This includes how hese sys ems in luence a company’s inancial esul s and how
employees pe cei e hei implemen a ion. In e es ingly, e en companies using he same
HRM sys em can expe ience pe o mance a ia ions. This highligh s he ole o o ganiza-
ional cha ac e is ics in shaping how HRM sys ems unc ion wi hin a company. Fu he
esea ch is needed o ully unde s and his ela ionship and he speci ic mechanisms by
which HRM sys ems ul ima ely impac pe o mance (Gues 2011).
In addi ion, esea ch examining he ela ionship be ween s a egic human esou ce man-
agemen and o ganiza ional pe o mance is expanding i s scope. Fe dousi and Abedin (2023)
epo ed a s udy on he speci ici y and pe o mance o human esou ce managemen in
social business o ganiza ions. Compa ed o o he companies, social business o ganiza ions
ha e di icul y aligning HRM and o ganiza ional goals because hey mus pu sue social
and economic goals a he same ime. To ia e al. (2022) discussed he unc ion o HRM o
he sus ainabili y o ca e se ice businesses a ge ing nonp o i social en e p ises in I aly.
This s udy akes a uni e salis ic pe spec i e o examine he connec ion be ween HPWS
and i m pe o mance. The uni e salis ic pe spec i e sugges s ha he e a e bes p ac ices in
HR ha can imp o e pe o mance ac oss di e en o ganiza ions (Dele y and Do y 1996). To
assess his om a s akeholde iewpoin , we ocus on wo key s akeholde s: sha eholde s
and employees. We use inancial pe o mance and employee u no e a e as quan i a i e
measu es o success (Gues 1997;Paauwe and Boselie 2005).
Acco ding o he esou ce-based iew (Ba ney 1991), his s udy a gues ha HR p ac-
ices p omo e excellen human esou ces in companies and posi i ely a ec o ganiza ional
pe o mance as a compe i i e ad an age ac o o companies. The esou ce-based iew
Adm. Sci. 2024,14, 75 4 o 20
p o ides a heo e ical basis o explaining he ela ionship be ween HR p ac ices and i m
pe o mance (Ba ney and W igh 1998). Acco ding o his heo y, a i m’s esou ces and
capabili ies a e c i ical o i s pe o mance and compe i i e ad an age. HR p ac ices, such
as ec ui men and selec ion, aining and de elopmen , and pe o mance managemen ,
a e essen ial esou ces enabling i ms o de elop and main ain a skilled and mo i a ed
wo k o ce. By in es ing in hese p ac ices, human esou ces a e aluable, a e, and no
subs i u able, which is di icul o compe i o s o imi a e. The e o e, hey can be c ucial o
secu ing a sus ainable compe i i e ad an age (Huselid 1995).
P e ious s udies ound s ong e idence ha e ec i e HR p ac ices can imp o e o -
ganiza ional pe o mance. Howe e , hey also ound ha he ela ionship be ween HR
and pe o mance is complex and ha o he ac o s, such as he ex e nal en i onmen
and o ganiza ional cul u e, can also play a ole (Chow 2012). The e o e, his s udy es s
he e ec i eness o HR p ac ices in one model, including en i onmen al ac o s such as
s a egies, o ganiza ional cul u e, o ganiza ional commi men , and o ganiza ional us
by ML. Mo eo e , i explo es HR p ac ices ha simul aneously sa is y wo company-le el
pe o mances o di e en a ibu es by using ML ha can be analyzed, including wo
o mo e dependen a iables in one model. The majo ou comes ha p e ious s udies
ha e in es iga ed can be ca ego ized in o h ee g oups: (1) human esou ce ou comes
(absen eeism a e, u no e a e, indi idual pe o mance, and eam pe o mance), (2) o -
ganiza ional ou comes (p oduc i i y and quali y o se ice), and (3) inancial ou comes
( e u n on in es men and e u n on asse s) (Dye and Ree es 1995).
S udies applying ML in HRM esea ch a e conduc ed in a ious esea ch ields, and
he s udies p esen he ollowing empi ical esul s. Meddeb e al. (2022) examined he
in e sec ion o machine lea ning and causal knowledge disco e y in HRM, highligh ing he
bene i s o inco po a ing domain expe s’ causal knowledge. Loya e-López and Ga cía-
Olaizola (2022) p esen an ML-based me hod o e alua ing he in e nal alue o alen and
ensu ing in e nal equi y in sala y c i e ia, sugges ing ha ML can suppo equi able and
unbiased sala y decisions based on da a. Xiang e al. (2022) p opose an in elligen HRM
sys em ha combines backp opaga ion neu al ne wo k and logis ic eg ession analysis o
imp o e e ec i eness, which is e i ied h ough simula ion es s wi h good p ac ical e ec s.
V on is e al. (2022) analyzed 45 HRM ield jou nals using in elligen au oma ion, including
a i icial in elligence, and explained ha au oma ion echnologies p esen a new app oach
(e.g., echnical and e hical le el) in he HRM esea ch ield. Addi ionally, hese skills
a e iden i ied as a ec ing ec ui men , aining, and wo k pe o mance in o ganiza ions.
Fu he mo e, Ga g e al. (2022) analyzed s udies using ML echnology in HR esea ch. They
ound ha among HR p ac ices, he e ec i eness o ML applica ions was mos signi ican in
ec ui men and pe o mance managemen . ML applica ions imp o e employee expe ience
and p omo e employee pe o mance.
3. Da a Sou ce and In eg a ed Da ase
3.1. O e iew
This s udy was conduc ed using he ollowing s eps. Fi s , he s udy a iables we e
chosen based on p e ious SHRM s udies. The independen a iables we e classi ied
in o HR p ac ices, employee a i udes, and o he i m cha ac e is ics. All he a iables
we e equen ly used in p e ious s udies on he e ec o HR p ac ices on employees and
pe o mance. Second, he independen a iable was gene a ed by ex ac ing in o ma ion
om an HCCP da ase , and hen an in eg a ed da ase was cons uc ed. The s udy a iables
ela ed o HR p ac ice we e composi e a iables gene a ed using a i m-le el da ase . The
in o ma ion o i m cha ac e is ics was also ex ac ed om a i m-le el da ase . O he
a iables ela ed o employee a i udes we e composi e a iables based on in o ma ion om
an employee-le el da ase . Independen a iables a he employee le el we e agg ega ed
a he i m le el. All he s udy a iables we e me ged using he company iden i ica ion
(ID) and yea a iable.
Adm. Sci. 2024,14, 75 5 o 20
3.2. Da a Sou ces
He e, an HCCP da ase was used, which was p o ided by he Ko ea Resea ch Ins i-
u e o Voca ional Educa ion and T aining (KRIVET). This go e nmen - unded ins i u e
p o ides he HCCP da ase o assis s udies on changes in HRM ac i i ies and hei e -
ec s on employees and i m pe o mance in Ko ean i ms. The da a we e collec ed and
dis ibu ed biennially since 2005. Da a om he 3 d o 7 h in es iga ions we e used. The
HCCP comp ises h ee sub-da ase s: i m-le el, employee-le el, and inancial in o ma ion
da ase s. The i m-le el da ase includes gene al i m cha ac e is ics, pe cei ed en i on-
men al cha ac e is ics, HR de elopmen , and managemen ac i i ies. The employee-le el
da ase con ains da a collec ed di ec ly om employees and includes he demog aphics o
esponden s, pe cep ions abou hei i ms and HR p ac ices, a i udes owa d hei jobs
and i ms, and o ganiza ional beha io s. The inancial da ase p o ided by KRIVET mainly
ocuses on accoun ing-based in o ma ion, such as sales, asse s, deb s, and ne incomes.
Dissimila o he o he wo da ase s, which a e collec ed biennially, he inancial da ase
includes annual in o ma ion p o ided by he Na ional In o ma ion & C edi E alua ion
(NICE) in o ma ion se ice, a eliable co po a e in o ma ion p o ide in Ko ea.
3.3. Va iable Selec ion
To cons uc he in eg a ed da ase o he p esen s udy, a iables we e gene a ed
using h ee sub-da ase s. Fi s , ollowing p e ious s udies (Huselid 1995;Choi e al. 2021;
A hu 1994), employee u no e a e and i m sales we e used as he ou come a iables.
The employee u no e a e was calcula ed by di iding he numbe o employees who le
he i m by he o al numbe o employees each yea . The in o ma ion o his a iable was
ex ac ed om he i m-le el da ase .
Second, ega ding HR p ac ices, eigh a iables anging om ec ui men o compen-
sa ion we e gene a ed. The HCCP su eyed whe he o no a pa icula HR p ac ice had
been implemen ed in he p io wo yea s. Based on his in o ma ion, he implemen ed
HR p ac ices we e classi ied in o six dis inc ca ego ies. The ca ego ies we e based on he
gene al HR p ocess, and classi ica ion was guided by a p io s udy ha p o ides a lis o
HR p ac ices ha a e pa o HPWSs (Jiang e al. 2012;Pos huma e al. 2013). Howe e , no
all lis ed HR p ac ices we e in es iga ed in he HCCP, and hus he numbe o HR p ac ices
used in his s udy was limi ed o hose in es iga ed in he HCCP. Fu he mo e, conside ing
compensa ion le el and s uc u e can be an impo an aspec o HR sys ems and a ec
employees’ a i udes and beha io s, hey we e included in he analysis. Speci ically, alen
acquisi ion p ac ice is he numbe o p ac ices used o in e nally de elop o ex e nally
acqui e high pe o me s. T aining p ac ice is he numbe o p ac ices ocused on enhancing
he cu en job pe o mance o employees. De elopmen p ac ice is he numbe o p ac ices
ocused on imp o ing he abili y and knowledge o employees o enhance u u e job pe o -
mance. Pe o mance managemen p ac ice is he numbe o p ac ices used o e alua e job
pe o mance. Pe o mance-based pay p ac ice is he numbe o compensa ion p og ams
based on indi idual, eam, o business-uni pe o mance. Compensa ion le el is he o al
annual compensa ion. Compensa ion s uc u e is he a io o all pe o mance-based pay
o base sala y. F inge bene i is he le el o inge bene i s compa ed wi h hose o o he
i ms in he same indus y each yea . In addi ion o HR p ac ices, h ee a iables ela ed o
HR depa men al ac i i ies, such as s a egic HR planning and he in ol emen o he HR
depa men , we e also included (Ge ha 2007;Huselid and Becke 1997;
Han e al. 2019
).
Annual HR plan is a bina y a iable wi h a alue o ‘1’ i a i m se s an annual HR plan
o ‘0’ i no . HR plan-s a egy alignmen is he deg ee o which a i m’s s a egy is e-
lec ed in i s HR ac i i ies. HR depa men al in ol emen is he deg ee o which an HR
depa men engages in co po a e s a egic planning, a CEO’s decision-making, changes in
HR p ac ices, and co po a e-le el inno a ion. All he a iables we e ex ac ed om he
i m-le el da ase excep o HR depa men al in ol emen , which was ex ac ed om he
employee-le el da ase .
Adm. Sci. 2024,14, 75 6 o 20
Thi d, o accoun o he e ec o employee a i udes owa d hei jobs and o ganiza-
ions, o ganiza ional cul u e, commi men , and us we e included in he model. The ou
aspec s o o ganiza ional cul u e we e measu ed using da a ga he ed om he employee-
le el da ase . Adhoc acy cul u e emphasizes c ea i i y and inno a ion; clan cul u e ocuses
on eamwo k and solida i y; hie a chical cul u e emphasizes ule and p ocess; and ma ke
cul u e ocuses on indi idual capabili y, compe i ion, and pe o mance. Gi en ha an
o ganiza ion possesses some cha ac e is ics o all ou cul u es, he deg ee o each ype
o cul u e was measu ed o all i ms (Quinn 2011). O ganiza ional commi men was
measu ed as he le el o a ec ion o he job, ea o loss, and sense o obliga ion o s ay
(Ma hieu and Zajac 1990;Meye e al. 2002). O ganiza ional us was measu ed as he le el
o us in he managemen eam, o he membe s, and he e alua ion and compensa ion
p ocess (Colqui e al. 2007;Di ks and Fe in 2001). The in o ma ion o hese a iables
was ex ac ed om he employee-le el da ase .
Finally, six a iables ela ed o en i onmen al changes, i m s a egy, i m size, and
inancial condi ion ha could po en ially a ec i m pe o mance and employee u no e
we e measu ed and hen included in he analysis. Changes in managemen en i onmen s
we e measu ed using he ollowing h ee a iables: changes in he demand o p ima y
p oduc s, changes in echnologies used in he p oduc ion p ocess, and changes in he
de elopmen and in oduc ion o new p oduc s. Fi m s a egy was measu ed using a
bina y a iable wi h a alue o ‘1’ i a i m ocuses on ei he quali y imp o emen o new
p oduc de elopmen o ‘0’ i a i m ocuses on cos educ ion. Fi m size was measu ed
by he o al numbe o employees. Le e age was measu ed by he a io o o al deb o
o al equi y. The i s i e a iables we e ex ac ed om he i m-le el da ase , and he las
a iable was ex ac ed om he inancial in o ma ion da ase .
3.4. Da a Cleaning and In eg a ed Da ase
The HCCP pe o med in 2017 is he 7 h in es iga ion, and all he da a om he 1s o
7 h in es iga ions consis ed o 3317 i m–yea obse a ions and 74,774 i m–yea –employee
obse a ions. The i s h ee in es iga ions we e excluded om he s udy because o
changes in su ey i ems be ween he 3 d and 4 h in es iga ions, which esul ed in he
una ailabili y o in o ma ion used in he s udy. In addi ion, he s udy sample was limi ed
o manu ac u ing indus ies o ensu e compa abili y ac oss he i ms in he sample. Thus,
he ini ial sample comp ised 1760 i m–yea obse a ions and 39,906 i m–yea –employee
obse a ions om he 3 d and 7 h in es iga ions. Acco ding o a coding scheme in he
HCCP da ase , non esponse and unknown in o ma ion a e coded
−
9 and
−
8, espec i ely.
These alues eplaced missing alues be o e gene a ing he s udy a iables. This p ocess
esul ed in he exclusion o 539 i m–yea obse a ions (8820 i m–yea –employee obse a-
ions). Thus, he inal i m-le el da ase comp ised 1221 i m–yea obse a ions, and he
employee-le el da ase comp ised 31,086 i m–yea –employee obse a ions. Finally, he
da ase s we e me ged using company ID and yea a iables. Be o e me ging he i m- and
employee-le el da ase s, he a iables gene a ed using he employee-le el da ase we e
a e aged by he yea –company ID a iable. A e wa d, he inal da ase comp ised 25
a iables and 1221 i m–yea obse a ions.
4. Analysis P ocess
4.1. O e iew
He e, a backp opaga ion neu al ne wo k (BPN)-based causali y analysis was con-
duc ed by using he HCCP da ase (Doh e al. 2016). Da a we e p ocessed by ans o ming
he scale o each a iable wi h he s anda diza ion me hod be o e aining a model. The e-
a e , he eg ession model wi h he BPN was gene a ed using he scaled da ase . The
BPN-based causali y analysis was conduc ed based on he in e connec ion weigh ac o s
o neu ons. All p og amming and analysis we e ca ied ou using Tenso Flow based on
Py hon 3 language.
Adm. Sci. 2024,14, 75 7 o 20
4.2. Da a P ocessing o T aining
The da a we e p ep ocessed o ans o m he da a scale o gene a e he a i icial neu al
ne wo k (ANN) model (Nawi e al. 2013). In gene ic da a p ocessing, he e a e wo ypes
o da a no maliza ion. Fi s , no maliza ion ans o ms he ange o he da a as a alue
be ween 0 and 1 based on he maximum and minimum alues o each a iable da a.
Second, s anda diza ion ans o ms da a o p oduce he mean, 0, and s anda d de ia ion, 1,
assuming he da a ollow a s anda d no mal dis ibu ion. This me hod is ad an ageous
as he in luence can educe he ela i e magni ude be ween da a, inc ease he accu acy o
he model, and be insensi i e o ou lie s. In his s udy, he da a we e p ocessed using he
s anda diza ion me hod wi h espec o he inpu and ou pu da a. The s anda diza ion
is ep esen ed in Equa ion (1). He e, xis he eal scale alue o he da a,
µ
is he mean o
he eal scale da a,
σ
is he s anda d de ia ion o he eal scale da a, and zis he z-sco e
(Aga ap 2018), indica ing he s anda dized alue o he eal scale da a.
z=xi−µ
σ(i=1∼N)(1)
Fu he mo e, he a iable ela ionship o gene a e he BPN model comp ised 23 inde-
penden a iables ela i e o HR p ac ices and 2 dependen a iables ega ding “employee
u no e ” and “ i m sales”. These a iables a e summa ized in Table 1.
Table 1. Independen and dependen a iables o he BPN model.
Va iables Type z-Sco e [µ,σ]
Independen
a iables
(Inpu da a)
Pe o mance managemen p ac ice loa [2.62, 1.68]
T aining p ac ice loa [3.35, 1.13]
De elopmen p ac ice loa [1.75, 1.55]
Compensa ion le el loa [44.56, 9.69]
O ganiza ional Commi men loa [3.32, 0.33]
Compensa ion s uc u e double [374.06, 347.01]
Fi m s a egy loa [1.83, 0.69]
HR plan-s a egy alignmen double [2.56, 0.90]
Clan cul u e loa [3.53, 0.38]
Change in echnology double [2.41, 0.78]
HR plan-s a egy alignmen double [0.76, 0.43]
F inge bene i s loa [2.85, 0.83]
Ma ke cul u e loa [3.47, 0.33]
Pe o mance-based pay p ac ice loa [1.45, 1.21]
Change in demand double [3.05, 1.04]
Talen acquisi ion p ac ice loa [1.42, 0.72]
Adhoc acy cul u e loa [3.30, 0.40]
O ganiza ional T us loa [3.43, 0.38]
Change in new p oduc double [2.29, 0.86]
Hie a chical cul u e loa [3.50, 0.29]
HR depa men al in ol emen loa [3.42, 0.41]
Fi m size loa [639.67, 1282.45]
Dependen
a iables
(Ou pu da a)
Le e age loa [1.64, 19.87]
Employee u no e loa [0.14, 0.22]
Fi m sales double [485,352,386,
1890,706,117]
4.3. Backp opaga ion Neu al Ne wo k Model
ML has been widely used in a ious ields. Among many ML echniques, ANN is
an algo i hm inspi ed by human neu al ne wo k a chi ec u es. An ANN model can be
gene a ed using connec ions be ween many neu ons and laye s in a complica ed manne
ega ding s uc u ed and uns uc u ed da a, such as images, ideos, and signal da a
(Abiodun e al. 2018).
Adm. Sci. 2024,14, 75 8 o 20
The de e mina ion o neu al ne wo k a chi ec u es emains a signi ican challenge in
he ield o a i icial in elligence. This challenge a ises due o he absence o speci ic guide-
lines o selec ing an op imal a chi ec u e. Ins ead, he choice o a chi ec u e, including he
numbe o neu ons and hidden laye s, elies on expe ien ial knowledge acqui ed h ough
i e a i e aining p ocesses.
In his s udy, we conduc ed a case s udy aimed a iden i ying an a chi ec u e ha
a oids o e i ing by p og essi ely inc easing he numbe o neu ons wi hin a single hid-
den laye . The decision o choose a single hidden laye was mo i a ed by wo p ima y
ac o s. Fi s ly, i aimed o s eamline he model, simpli ying i s complexi y. Secondly, i
acili a ed he causali y analysis o he ela ionships be ween human esou ce p ac ices (in-
pu ), u no e , and sales (mul i-ou pu ) using he in e connec ed weigh s in he simpli ied
neu al ne wo k model.
The gene ic a chi ec u e o he eed- o wa d neu al ne wo k comp ises e ms o he
in e connec ed weigh o hidden and ou pu laye s ( and w), bias (b), inpu laye (x),
ou pu laye (Op), and ac i a ion unc ion (σac .) in Equa ion (2).
Opi=
Nj
∑
j=1
wjk(σac . y=
Ni
∑
i=1
ijxi+b!) (2)
He e, he a chi ec u e o he BPN model comp ised an inpu laye wi h 23 neu ons, a
single hidden laye wi h 4096 neu ons, and an ou pu laye wi h 2 neu ons (Figu e 1).
Adm. Sci. 2024, 14, x FOR PEER REVIEW 9 o 22
Figu e 1. The a chi ec u e o a single-laye ed neu al ne wo k.
T aining and es da ase s comp ised 80% and 20% o a o al o 1221 da a, espec-
i ely. Based on hese da ase s, he in e connec ed weigh , , and w, be ween he inpu ,
hidden, and ou pu laye s we e ob ained using he s ochas ic g adien descen op imize
(Keska and Soche 2017). This op imize is ad an ageous as con e gence is as du ing
he aining a each s ep and can educe he p obabili y o alling in he local op imum by
s ochas ic shoo ing when upda ing in e connec ed weigh s. Fu he mo e, egula iza ion
is a echnique used in BPN models o p e en o e i ing and imp o e he gene aliza ion
pe o mance o he model. I in ol es adding a penal y e m o he loss unc ion (i.e., MSE,
MAE, and R
2
) ha encou ages he weigh s o he ne wo k o be small. The e a e diffe en
ypes o egula iza ion echniques, such as L1 and L2 egula iza ion, which diffe in he
way he penal y e m is calcula ed. These echniques can educe he complexi y o he
ne wo k, p e en he model om memo izing noise in he aining da a, and imp o e i s
abili y o gene alize o new, unseen da a. In his s udy, he aining pa ame e s we e es-
ablished, as ep esen ed in Table 2. The L1 egula iza ion (Tsu uoka e al. 2009) was em-
ployed wi h Equa ion (4). He e,
λ
is he pa ame e o L1 egula iza ion, and w is he in e -
connec ion weigh o he BPN model.
{
}
1
1
Loss = ( , )
2
i
n
ip
i
L
OO w
n
λ
=
+
(4)
Table 2. T aining pa ame e s o aining he BPN model.
Lea ning Ra e
(h)
Momen um
(g)
L1 Regula iza ion
(l) Epoch Ba ch Size
0.0005 0.9 0.001 200 8
4.4. Valida ion
The accu acy and alidi y o he BPN model we e e alua ed using he k- old c oss-
alida ion (Li e al. 2010) me hod using sequen ial shuffling wi h 20 olds acco ding o he
andomness o he aining and es da ase s a a ixed a e (Figu e 2). Fu he mo e, he
quan i a i e me ics o accu acy and alidi y we e assessed using he mean squa ed e o
(MSE), mean absolu e e o (MAE), and R-squa ed (R
2
) alue, as exp essed in (5)–(7).
He e, O
i
is he ac ual aining o es da a o he i- h inpu da a, O
pi
is he p edic ed alue
Figu e 1. The a chi ec u e o a single-laye ed neu al ne wo k.
The Rec i ied Linea Uni (ReLU) ac i a ion unc ion is a popula choice o deep
neu al ne wo ks because o i s se e al ad an ages in backp opaga ion. I is compu a ionally
e icien , a oids he anishing g adien p oblem, p omo es spa si y, and is easy o op imize.
Compa ed wi h o he ac i a ion unc ions, ReLU in ol es a simple h esholding ope a ion,
and i s de i a i e is always ei he 0 o 1, which makes i mo e e icien and s able du ing
backp opaga ion. ReLU can lead o spa se ep esen a ions in neu al ne wo ks, which helps
o educe o e i ing and imp o e gene aliza ion pe o mance by elimina ing i ele an
ea u es and educing he dimensionali y o he inpu . In his s udy, he
σac .
was employed
by he ReLU unc ion (3), as exp essed in Equa ion (3). This ac i a ion unc ion is a posi i e
alue i σac . exceeds 0; o he wise, σac . is 0.
σac .(y) = ReLU(y) = y(ReLU(y)≥0)
0(ReLU(y)<0)(3)
Adm. Sci. 2024,14, 75 15 o 20
ha e desi able e ec s on he wo ou comes. F inge bene i s dec ease employee u no e
and enhance i m pe o mance. In summa y, HR p ac ices ela ed o compensa ion a e
impo an ac o s in p edic ing i m sales and employee u no e . Howe e , employees in
Ko ean i ms end o eac posi i ely o he ise in o al compensa ion bu eac nega i ely
o an inc ease in he p opo ion o pe o mance-based pay and he numbe o pe o mance-
based pay p ac ices. F inge bene i s can also easily be used o p omo e beha io s ha can
imp o e i m pe o mance and educe u no e .
Second, he e ec s o alen acquisi ion p ac ice and pe o mance managemen p ac ice a e
ela i ely weak, as shown in Figu e 5a,b. Talen acquisi ion p ac ice is he 12 h and 19 h
impo an a iable in p edic ing i m sales and employee u no e , espec i ely. Simila ly,
pe o mance managemen p ac ice is he 18 h mos impo an a iable o i m sales and
employee u no e . The limi ed e ec s migh be because hose p ac ices a e al eady
ins i u ionalized (Meye and Rowan 1977;Boon e al. 2009;Goode ham e al. 1999). In
o he wo ds, se e al Ko ean i ms ely on he ex e nal labo ma ke o acqui e key alen
and use di e se pe o mance managemen p ac ices. Thus, he di e ence in he wo
ou comes explained by hose p ac ices is ela i ely limi ed. Rega ding he di ec ion o
e ec s (Figu e 6e), alen acquisi ion p ac ice exhibi s an in e ed-U shape ela ionship wi h
he wo ou comes o he s udy. Figu e 6 shows hese ela ionships. These esul s sugges ed
ha despi e hei ins i u ionalized na u e, alen acquisi ion and pe o mance managemen
p ac ices can s ill be e ec i e managemen ools o an ex en . Howe e , HR manage s
should de e mine he app op ia e in ensi y o such p ac ices. Fo example, he excessi e
u iliza ion o ex e nal labo ma ke and pe o mance managemen p ac ices may cause
compe i i e in e ac ions inside o ganiza ions, leading o unin ended nega i e e ec s on
o ganiza ional e ec i eness. Gi en ha he de ailed causali y may be mo e complex, he
obse ed ela ionship should be in es iga ed in u u e s udies.
Thi d, compa ed wi h o he HR p ac ices, he e ec s o aining and de elopmen
on he wo ou comes a e modes . T aining p ac ice and de elopmen p ac ice a e he 9 h
and 10 h mos impo an a iables in p edic ing i m sales, espec i ely. They a e also he
4 h and 13 h impo an a iables in p edic ing employee u no e . P e ious s udies ha e
sugges ed ha aining and de elopmen inc ease i m pe o mance and dec ease u no e
in en ion (A hu 1994;Shuck e al. 2014). Howe e , ou esul s showed ha he e ec s
a e mo e complex han p e iously sugges ed. Figu e 6g–h show ha aining a ec s sales
nega i ely, and he e ec o de elopmen p ac ice changes om nega i e o posi i e as
he le el inc eases. These complex ela ionships can be u he explo ed by conside ing
he aining and de elopmen con en and he cha ac e is ics o s a egy and ope a ion.
Conside ing he a o emen ioned con en , a plausible explana ion o he e ec s o aining
and de elopmen p ac ices on employee u no e can be de eloped.
Speci ically, aining enhances employee u no e , bu de elopmen educes employee
u no e . This con as ing e ec migh be caused by he di e ence in he cha ac e is ics o
he wo p ac ices. De elopmen p ac ices, such as ca ee de elopmen p og ams, men o -
ing, lea ning g oups, and job o a ion, a e in e nally implemen ed o enable employees o
accumula e i m-speci ic knowledge. Con a ily, aining p og ams, which a e equen ly
ou sou ced o specialized ex e nal ins i u ions, p o ide employees wi h gene al skills and
he la es knowledge. Thus, al hough enhanced i m-speci ic skills can educe employee
u no e , s eng hened employabili y will lead o employee u no e (Nelissen e al. 2017;
Benson 2006;De Cuype e al. 2011). Fu u e s udies need o explo e his opic by ocusing on
he i be ween he cha ac e is ics o wo ypes o p ac ices and i m s a egy and ope a ion.
Fou h, ega ding HR depa men al ac i i ies, his s udy e alua ed he ela i e im-
po ance and di ec ion o he e ec s o HR plan-s a egy alignmen ,annual HR plan, and HR
depa men al in ol emen . They a e he 7 h, 21s , and 22nd mos impo an a iables o sales
(Figu e 5a). They a e also he 10 h, 15 h, and 20 h mos impo an a iables o employee
u no e (Figu e 5b). Thus, hei e ec s a e ela i ely small and weak. In addi ion, as
shown in Figu e 6i–k, he di ec ions o e ec s a e no clea ly e iden , and he e a e nega i e
e ec s ha equi e u he in es iga ion. Howe e , he esul s should be in e p e ed wi h
Adm. Sci. 2024,14, 75 16 o 20
cau ion because he measu emen ocused on he exis ence o deg ee o he p ac ices, no
hei ac ual con en s. Thus, u u e s udies on hese opics could s a by unde s anding he
goals o p ac ices ha can de e mine he di ec ion o he e ec s.
Fi h, al hough he s udy ocused on HR p ac ices, he e we e in e es ing indings on
he e ec s o employee a i ude and changes in demand and echnology. Many s udies
ha e epo ed he posi i e e ec s o o ganiza ional us and commi men on employee
e en ion and job pe o mance (Ma hieu and Zajac 1990;Meye e al. 2002;Colqui e al.
2007;Di ks and Fe in 2001). Acco ding o ou indings, o ganiza ional us and commi men
ha e a ela i ely mino e ec on he wo ou comes (Figu e 5a,b). In pa icula , hei e ec s
on sales a e minimal, and hose on employee u no e a e conside ably less han hose on
compensa ion p ac ices. Thus, ou indings sugges ed ha compa ed wi h o he ac o s,
o ganiza ional us and commi men may no be in luen ial ac o s in Ko ean i ms. Howe e ,
hey ha e he expec ed e ec s on employee u no e (Figu e 6j–k). No ewo hily, change
in echnology is he second and i h mos impo an a iable in p edic ing i m sales and
employee u no e , espec i ely (Figu e 5a,b). Change in demand is also he eigh h and mos
impo an a iable in p edic ing i m sales and employee u no e , espec i ely. Changes in
he ma ke (demand o he p ima y p oduc ) and inside he o ganiza ion ( echnology used
in he o ganiza ion) also ha e nonlinea ela ionships wi h he wo ou comes (Figu e 6n,o).
Al hough such ela ionships canno be easily explained, he ac ha he e we e in lec ion
poin s, pa icula ly ega ding employee u no e , can p o ide HR manage s wi h guidance
on when HR p ac ices should be e ised.
5.2. Implica ions
The heo e ical implica ions o his s udy a e as ollows. Fi s , his s udy con ibu es o
he exis ing li e a u e on HRM sys ems and i m pe o mance by examining he ela ionship
be ween a la ge numbe o indi idual HR p ac ices and i m pe o mance. P e ious s udies
ha e ypically ocused on a limi ed numbe o indi idual HR p ac ices o on bundles o
HR p ac ices. This s udy’s indings sugges ha a wide ange o HR p ac ices may be
impo an o i m pe o mance. Second, his s udy add esses he issue o endogenei y
p oblem by using a a ie y o me hods o con ol o po en ial con ounding ac o s. This
helps o ensu e ha he esul s o he s udy a e no due o o he ac o s, such as i m size
o indus y. Thi d, his s udy p o ides new insigh s in o he mechanisms by which HR
p ac ices in luence i m pe o mance. The indings sugges ha HR p ac ices can in luence
i m pe o mance h ough a a ie y o channels, including o ganiza ional cul u e and
employee wo k a i udes, employee u no e , and s a egy.
This s udy p esen s he ollowing manage ial implica ions. Fi s , HR manage s can
use he esul s o his s udy o unde s and which HR p ac ices hey should pay a en ion o.
In o mula ing and implemen ing HR p ac ices, HR manage s a e asked o enhance i m
sales and educe o , a leas , main ain employee u no e . Al hough p e ious s udies ha e
p o ided ample e idence ha a bundle o HR p ac ices enables HR manage s o achie e
he wo ou comes o he p esen s udy, hese s udies ha e a ely ocused on he ela i e
impo ance o each HR p ac ice on he wo goals. Focusing on widely used HR p ac ices
and o he indi idual and o ganiza ional cha ac e is ics, his s udy p o ides in o ma ion on
he ela i e impo ance o HR p ac ices on i m sales and employee u no e . The esul s
a e aluable guidelines o HR manage s o e ec i ely alloca e limi ed esou ces o di e en
ypes o HR p ac ices.
Second, ou s udy s essed ha HR manage s should exe cise cau ion when in oduc-
ing new compensa ion p ac ices. Faced wi h a alen sho age, many Ko ean i ms ha e
inc eased pay le els and he p opo ion o pe o mance-based pay o a ac alen om
he labo ma ke . Such p ac ices a e also used o align exis ing employee mo i a ion and
beha io s wi h he o e all goals o he i ms. This end is equen ly accompanied by he
in oduc ion o a ious compensa ion managemen p ac ices. Howe e , as e ealed in he
analysis, al hough compensa ion le el was nega i ely associa ed wi h employee u no e ,
he p opo ion o pe o mance-based pay and he numbe o pe o mance-based pay p ac-
Adm. Sci. 2024,14, 75 17 o 20
ices we e posi i ely associa ed wi h employee u no e . This esul implied ha cu en
compensa ion p ac ices aimed a a ac ing and mo i a ing alen should be econside ed.
Speci ically, gi en ha pe o mance-based pay is a widely used compensa ion p ac ice, ou
analysis sugges s ha i ms seeking o ein o ce pe o mance-based pay should keep hei
compensa ion p ac ices s aigh o wa d by educing he numbe o pe o mance-based
pay p ac ices.
Thi d, his s udy p o ides insigh s in o when exis ing HR p ac ices should be e ised.
F om he pe spec i e o SHRM, HR p ac ices should be e ised when i ms change hei
s a egies o e lec changes in he ex e nal en i onmen , such as changes in demand,
and ha de e mine in e nal condi ions, such as echnology. Acco ding o his s udy,
changes in echnology and demand o p ima y p oduc s a e in luen ial in p edic ing sales
and employee u no e . Gi en ha such changes a e ine i able o a i m o main ain a
compe i i e ad an age, HR manage s in i ms acing such changes should pay a en ion o
hei HR p ac ices. No ably, acco ding o Figu e 6n–o, he in lec ion poin s exceeded ze o,
indica ing ha manage s should ee alua e exis ing HR p ac ices when he changes exceed
he le el aken by compe i o s. Thus, his s udy p o ides mo e de ailed guidelines han
gene al heo e ical sugges ions on when HR manage s should conside e ising cu en
HR p ac ices.
Fou h, his s udy obse ed many complex ela ionships be ween HR p ac ices and
he wo ou comes, which a e challenging o explain heo e ically. No ably, he ela ionships
be ween ce ain p edic o s and he wo ou comes we e nonlinea and canno be easily
explained. These esul s would ha e been de e mined by he ML algo i hm aimed a
imp o ing explana o y powe . Howe e , he complex na u e o he ela ionships sugges s
ha SHRM schola s should u ilize a pa e n disco e y app oach as well as a adi ional
heo y-d i en app oach o p o ide p ac i ione s wi h mo e p ac ice guidance. Thus, he
con en o he p ac ices and he con ex unde which such p ac ices a e implemen ed should
be ca e ully conside ed in u u e s udies.
5.3. Limi a ions
This s udy has ce ain limi a ions ha u u e s udies can add ess. Fi s , i did no
conside he e ec s o each i m’s dis inc i e HR p ac ices. Ou analysis is based on
seconda y da a (HCCP); hus, his s udy did no conside HR p ac ices ha a e no included
in he su ey. Howe e , gi en ha idiosync a ic condi ions o i ms can cause a di e ence
in HR p ac ices be ween i ms, u u e s udies should in es iga e he e ec s o cus omized
HR p ac ices ha e lec each i m’s speci ic goals o pu pose. Second, he s udy’s sample o
i ms was limi ed o he manu ac u ing indus y. Thus, he esul s may no apply o o he
indus ies, such as he se ice indus y. Fu u e s udies may emphasize he di e ence in he
e ec s o he same HR p ac ices ac oss indus ies. Thi d, he analysis did no in es iga e he
nonlinea in e ac ions be ween he e ec s o he wo explana o y a iables on he ou comes.
This s udy aimed o examine he e ec s o HR p ac ices on i m sales and employee u no e ,
emphasizing he di ec e ec s o each HR p ac ice. Howe e , he e ec s o each HR p ac ice
can be u he enhanced o educed depending on o he HR p ac ices and indi idual and
o ganiza ional cha ac e is ics
(Han e al. 2019;Becke and Ge ha 1996)
. Fu u e s udies
can explo e nonlinea in e dependences based on his heo e ical a gumen . This app oach
will be ano he way o eexamine he complex nonlinea e ec s o HR p ac ices ha his
s udy obse ed. Las ly, his s udy u ilizes he empi ical Bayes (EB) es ima ion me hod wi h
he expec a ion–maximiza ion (EM) algo i hm o add ess he cha ac e is ics o he panel
da a, which is an incomple e da ase wi h missing alues. The EM algo i hm-based EB
es ima ion me hod epea edly es ima es he pa ame e s om he gi en incomple e da a and
es ima es he comple e da a su icien s a is ics. I hen u ilizes hese s a is ics o es ima e he
maximum likelihood alue. This me hod does no eplace missing cases o alues. Ins ead,
i es ima es he pa ame e s by es ima ing he comple e da a. The e o e, i can o e come he
limi a ions o missing a iables o cases in he da a (Raudenbush and B yk 2002).
Adm. Sci. 2024,14, 75 18 o 20
In e ms o explana o y capaci y, he limi a ion eme ges wi hin machine lea ning
models g ounded in da a-d i en app oaches, as hey lack he capabili y o explica e he
ela ionship be ween inpu and ou pu a iables, owing o hei exclusi e eliance on
da a. Nume ous esea che s ha e endea o ed o mi iga e his limi a ion by explaining he
ela ionship be ween inpu and ou pu a iables in a a ional manne . Fu he mo e, hey
ha e in oduced he concep o explainable and in e p e able a i icial in elligence models
o add ess his challenge, wi h he objec i e o os e ing a meaning ul comp ehension
o he associa ion be ween inpu and ou pu a iables. Consequen ly, he signi icance
o a machine lea ning model lies in i s capaci y o un eil an in elligible and easoned
in e p e a ion o his associa ion.
6. Conclusions
E ec i e HR p ac ices con ibu e o i ms by suppo ing s a egy implemen a ion.
This suppo ing ole can be achie ed when he HR p ac ices a e aligned wi h s a egic
goals and can acqui e, de elop, mo i a e, and e ain capable employees. Fo example, one
o he main esea ch esul s seems o be a ibu ed o he ollowing wage sys em in he
Ko ean labo ma ke . Fo a long ime, Ko ean companies ha e ope a ed a wage sys em
based on senio i y. The senio i y sys em has he ad an age o making i easy o secu e
and e ain skilled wo ke s and lexibly assign hem o necessa y asks. The e o e, i was
a sys em ha could compensa e o he labo sho age in he Ko ean labo ma ke due
o apid indus ializa ion since he 1970s. A e expe iencing he o eign exchange c isis
and he subp ime mo gage c isis in he 2000s, many Ko ean companies in oduced a
pe o mance-based compensa ion sys em. Howe e , he pe o mance-based compensa ion
sys em is un amilia o Ko ean wo ke s who a e accus omed o he old senio i y-based
o ganiza ional cul u e. The e o e, many companies need o pay a en ion o he ai ness o
he pe o mance-based sys em and communica ion abou he pe o mance sys em o educe
esis ance om wo ke s when in oducing he pe o mance-based compensa ion sys em.
In his s udy, he ela i e impo ance and he di ec ion o he e ec s o widely used
HR p ac ices in Ko ean manu ac u ing i ms we e examined. The s udy iden i ied in lu-
en ial HR p ac ices ha p edic i m sales and employee u no e and p o ided e idence
o complex ela ionships be ween HR p ac ices and he wo ou comes. The esul s em-
phasized ha HR manage s do no need o pay equal a en ion o all HR p ac ices and
should e iew exis ing HR p ac ices depending on ela i e changes o compe i o s’ in-
e nal o ex e nal condi ions. This s udy p o ides p ac i ione s wi h p ac ical guidelines
and encou ages schola ly wo ks ha eexamine he complex e ec s o HR p ac ice on
o ganiza ional e ec i eness.
Au ho Con ibu ions: Concep ualiza ion, M.L. and J.D.; me hodology, J.D. and H.M.; so wa e, H.M.
and K.L.; alida ion, J.D., H.M. and K.L.; o mal analysis, M.L., G.L. and J.D.; in es iga ion, H.M.
and K.L.; da a cu a ion, M.L., G.L. and J.D.; w i ing—o iginal d a p epa a ion, M.L., G.L. and J.D.;
w i ing— e iew and edi ing, M.L., G.L. and J.D.; isualiza ion, H.M. and K.L.; supe ision, M.L.,
G.L. and J.D.; p ojec adminis a ion, M.L., G.L. and J.D.; unding acquisi ion, J.D. All au ho s ha e
ead and ag eed o he published e sion o he manusc ip .
Funding: This wo k was suppo ed by he Minis y o Educa ion o he Republic o Ko ea and he
Na ional Resea ch Founda ion o Ko ea (NRF-2023S1A5A8074321, PI: Jaehyeok Doh) and his esea ch
was suppo ed by he Basic Science Resea ch P og am h ough he Na ional Resea ch Founda ion o
Ko ea (NRF), unded by he Minis y o Educa ion (NRF-2021R1I1A3044394, PI: Jaehyeok Doh).
Ins i u ional Re iew Boa d S a emen : No applicable.
In o med Consen S a emen : No applicable.
Da a A ailabili y S a emen : Publicly a ailable da ase s we e analyzed in his s udy.
Con lic s o In e es : The au ho s decla e no con lic s o in e es .
Adm. Sci. 2024,14, 75 19 o 20
Re e ences
Abiodun, Oluda e Isaac, Aman Jan an, Abiodun Es he Omola a, Kemi Vic o ia Dada, Nachaa AbdEla i Mohamed, and Humai a
A shad. 2018. S a e-o - he-a in a i icial neu al ne wo k applica ions: A su ey. Heliyon 4: e00938. [C ossRe ] [PubMed]
Aga ap, Abien F ed. 2018. Deep lea ning using ec i ied linea uni s ( elu). a Xi a Xi :1803.08375.
A hu , Je ey B. 1994. E ec s o human esou ce sys ems on manu ac u ing pe o mance and u no e . Academy o Managemen Jou nal
37: 670–87. [C ossRe ]
Ba ney, Jay. 1991. Fi m esou ces and sus ained compe i i e ad an age. Jou nal o Managemen 17: 99–120. [C ossRe ]
Ba ney, Jay B., and Pa ick M. W igh . 1998. On becoming a s a egic pa ne : The ole o human esou ces in gaining compe i i e
ad an age. Human Resou ce Managemen : Published in Coope a ion wi h he School o Business Adminis a ion, The Uni e si y o
Michigan and in alliance wi h he Socie y o Human Resou ces Managemen 37: 31–46. [C ossRe ]
Becke , B ian. 1998. High pe o mance wo k sys ems and i m pe o mance: A syn hesis o esea ch and manage ial implica ions.
Resea ch in Pe sonnel and Human Resou ces Managemen 16: 53.
Becke , B ian, and Ba y Ge ha . 1996. The impac o human esou ce managemen on o ganiza ional pe o mance: P og ess and
p ospec s. Academy o Managemen Jou nal 39: 779–801. [C ossRe ]
Benson, Geo ge S. 2006. Employee de elopmen , commi men and in en ion o u no e : A es o ‘employabili y’policies in ac ion.
Human Resou ce Managemen Jou nal 16: 173–92. [C ossRe ]
Boon, Co ine, Jaap Paauwe, Paul Boselie, and Deanne Den Ha og. 2009. Ins i u ional p essu es and HRM: De eloping ins i u ional i .
Pe sonnel Re iew 38: 492–508. [C ossRe ]
Boselie, Paul, G aham Die z, and Co ine Boon. 2005. Commonali ies and con adic ions in HRM and pe o mance esea ch. Human
Resou ce Managemen Jou nal 15: 67–94. [C ossRe ]
Chadwick, Clin . 2007. Examining non-linea ela ionships be ween human esou ce p ac ices and manu ac u ing pe o mance. ILR
Re iew 60: 499–521. [C ossRe ]
Choi, Jung-Gu, Inhwan Ko, Jeongjae Kim, Yeseul Jeon, and Sanghoon Han. 2021. Machine lea ning amewo k o mul i-le el
classi ica ion o company e enue. IEEE Access 9: 96739–50. [C ossRe ]
Choudhu y, P i hwi aj, Ryan T. Allen, and Michael G. End es. 2021. Machine lea ning o pa e n disco e y in managemen esea ch.
S a egic Managemen Jou nal 42: 30–57. [C ossRe ]
Chow, I ene Hau-Siu. 2012. The oles o implemen a ion and o ganiza ional cul u e in he HR–pe o mance link. The In e na ional
Jou nal o Human Resou ce Managemen 23: 3114–32. [C ossRe ]
Colqui , Jason A., B en A. Sco , and Je e y A. LePine. 2007. T us , us wo hiness, and us p opensi y: A me a-analy ic es o hei
unique ela ionships wi h isk aking and job pe o mance. Jou nal o Applied Psychology 92: 909. [C ossRe ] [PubMed]
Combs, James, Yongmei Liu, Angela Hall, and Da id Ke chen. 2006. How much do high-pe o mance wo k p ac ices ma e ? A
me a-analysis o hei e ec s on o ganiza ional pe o mance. Pe sonnel Psychology 59: 501–28. [C ossRe ]
De Cuype , Nele, Saija Mauno, Ulla Kinnunen, and Anne Mäkikangas. 2011. The ole o job esou ces in he ela ion be ween pe cei ed
employabili y and u no e in en ion: A p ospec i e wo-sample s udy. Jou nal o Voca ional Beha io 78: 253–63. [C ossRe ]
Dele y, John E., and D. Ha old Do y. 1996. Modes o heo izing in s a egic human esou ce managemen : Tes s o uni e salis ic,
con ingency, and con igu a ional pe o mance p edic ions. Academy o Managemen Jou nal 39: 802–35. [C ossRe ]
Di ks, Ku T., and Donald L. Fe in. 2001. The ole o us in o ganiza ional se ings. O ganiza ion Science 12: 450–67. [C ossRe ]
Doh, Jaehyeok, Seung Uk Lee, and Jongsoo Lee. 2016. Back-p opaga ion neu al ne wo k-based app oxima e analysis o ue s ess-s ain
beha io s o high-s eng h me allic ma e ial. Jou nal o Mechanical Science and Technology 30: 1233–41. [C ossRe ]
Dye , Lee, and Todd Ree es. 1995. Human esou ce s a egies and i m pe o mance: Wha do we know and whe e do we need o go?
In e na ional Jou nal o Human Resou ce Managemen 6: 656–70. [C ossRe ]
Fe dousi, Fa hana, and Nu en Abedin. 2023. S a egic Human Resou ces Managemen o C ea ing Sha ed Value in Social Business
O ganiza ions. Sus ainabili y 15: 3703. [C ossRe ]
Ga g, Swa i, Shuchi Sinha, A pan Kuma Ka , and Mau icio Mani. 2022. A e iew o machine lea ning applica ions in human esou ce
managemen . In e na ional Jou nal o P oduc i i y and Pe o mance Managemen 71: 1590–610. [C ossRe ]
Ge ha , Ba y. 2007. Ho izon al and e ical i in human esou ce sys ems. Pe spec i es on O ganiza ional Fi 1: 317–48.
Goode ham, Paul N., Odd No dhaug, and K is en Ringdal. 1999. Ins i u ional and a ional de e minan s o o ganiza ional p ac ices:
Human esou ce managemen in Eu opean i ms. Adminis a i e Science Qua e ly 44: 507–31. [C ossRe ]
Gues , Da id E. 1997. Human esou ce managemen and pe o mance: A e iew and esea ch agenda. In e na ional Jou nal o Human
Resou ce Managemen 8: 263–76. [C ossRe ]
Gues , Da id E. 2011. Human esou ce managemen and pe o mance: S ill sea ching o some answe s. Human Resou ce Managemen
Jou nal 21: 3–13. [C ossRe ]
Han, Joo Hun, Saehee Kang, In-Sue Oh, Rebecca R. Kehoe, and Da id P. Lepak. 2019. The goldilocks e ec o s a egic human esou ce
managemen ? Op imizing he bene i s o a high-pe o mance wo k sys em h ough he dual alignmen o e ical and ho izon al
i . Academy o Managemen Jou nal 62: 1. [C ossRe ]
Has ie, T e o , Robe Tibshi ani, Je ome H. F iedman, and J. H. F iedman. 2009. The Elemen s o S a is ical Lea ning: Da a Mining,
In e ence, and P edic ion. New Yo k: Sp inge .
Huselid, Ma k A. 1995. The impac o human esou ce managemen p ac ices on u no e , p oduc i i y, and co po a e inancial
pe o mance. Academy o Managemen Jou nal 38: 635–72. [C ossRe ]
Adm. Sci. 2024,14, 75 20 o 20
Huselid, Ma k A., and B ian E. Becke . 1997. The impac high pe o mance wo k sys ems, implemen a ion e ec i eness, and alignmen
wi h s a egy on sha eholde weal h. In Academy o Managemen P oceedings. B ia cli Mano : Academy o Managemen , pp. 144–48.
Jiang, Kai eng, Da id P. Lepak, Jia Hu, and Judi h C. Bae . 2012. How does human esou ce managemen in luence o ganiza ional
ou comes? A me a-analy ic in es iga ion o media ing mechanisms. Academy o Managemen Jou nal 55: 1264–94. [C ossRe ]
Keska , Ni ish Shi ish, and Richa d Soche . 2017. Imp o ing gene aliza ion pe o mance by swi ching om adam o sgd. a Xi
a Xi :1712.07628.
Lee, Jong-Soo. 2008. Role o A i icial Neu al Ne wo ks in Mul idisciplina y Op imiza ion and Axioma ic Design. In P oceedings o he
KSME Con e ence. Seoul: The Ko ean Socie y o Mechanical Enginee s, pp. 695–700.
Lee, Gyeonghwan, Myeongju Lee, and Yoonhwan Sohn. 2017. High-pe o mance wo k sys ems and i m pe o mance: Mode a ing
e ec s o o ganiza ional communica ion. Jou nal o Applied Business Resea ch (JABR) 33: 951–62. [C ossRe ]
Li, De -Chiang, Yao-Hwei Fang, and Y. M. F ank Fang. 2010. The da a complexi y index o cons uc an e icien c oss- alida ion
me hod. Decision Suppo Sys ems 50: 93–102. [C ossRe ]
Loya e-López, Edu ne, and Igo Ga cía-Olaizola. 2022. Machine lea ning based me hod o deciding in e nal alue o alen . Applied
A i icial In elligence 36: 2151160. [C ossRe ]
Ma hieu, John E., and Dennis M. Zajac. 1990. A e iew and me a-analysis o he an eceden s, co ela es, and consequences o
o ganiza ional commi men . Psychological Bulle in 108: 171. [C ossRe ]
Meddeb, Eya, Ch is ophe Bowe s, and Lynn Nichol. 2022. Compa ing Machine Lea ning Co ela ions o Domain Expe s’ Causal
Knowledge: Employee Tu no e Use Case. In In e na ional C oss-Domain Con e ence o Machine Lea ning and Knowledge Ex ac ion.
New Yo k: Sp inge , pp. 343–61.
Meye , John W., and B ian Rowan. 1977. Ins i u ionalized o ganiza ions: Fo mal s uc u e as my h and ce emony. Ame ican Jou nal o
Sociology 83: 340–63. [C ossRe ]
Meye , John P., Da id J. S anley, Lynne He sco i ch, and La yssa Topolny sky. 2002. A ec i e, con inuance, and no ma i e commi men o
he o ganiza ion: A me a-analysis o an eceden s, co ela es, and consequences. Jou nal o Voca ional Beha io 61: 20–52. [C ossRe ]
Nawi, Naz i Mohd, Walid Hasen A omi, and Mohammad Zubai Rehman. 2013. The e ec o da a p e-p ocessing on op imized
aining o a i icial neu al ne wo ks. P ocedia Technology 11: 32–39. [C ossRe ]
Nelissen, Jill, Anneleen Fo ie , and Ma ijke Ve b uggen. 2017. Employee de elopmen and olun a y u no e : Tes ing he
employabili y pa adox. Human Resou ce Managemen Jou nal 27: 152–68. [C ossRe ]
Paauwe, Jaap, and Paul Boselie. 2005. HRM and pe o mance: Wha nex ? Human Resou ce Managemen Jou nal 15: 68–83. [C ossRe ]
Pos huma, Richa d A., Michael C. Campion, Malika Masimo a, and Michael A. Campion. 2013. A high pe o mance wo k p ac ices
axonomy: In eg a ing he li e a u e and di ec ing u u e esea ch. Jou nal o Managemen 39: 1184–220. [C ossRe ]
Quinn, Robe E. 2011. Diagnosing and Changing O ganiza ional Cul u e: Based on he Compe ing Values F amewo k. San F ancisco:
Jossey-Bass.
Raudenbush, S ephen W., and An hony S. B yk. 2002. Hie a chical Linea Models: Applica ions and Da a Analysis Me hods. Thousand
Oaks: Sage, ol. 1.
Shuck, B ad, De on Twy o d, Thomas G. Reio, J ., and Angie Shuck. 2014. Human esou ce de elopmen p ac ices and employee engagemen :
Examining he connec ion wi h employee u no e in en ions. Human Resou ce De elopmen Qua e ly 25: 239–70. [C ossRe ]
Sub amony, Mahesh. 2009. A me a-analy ic in es iga ion o he ela ionship be ween HRM bundles and i m pe o mance. Human
Resou ce Managemen 48: 745–68. [C ossRe ]
Takeuchi, Riki, Gilad Chen, and Da id P. Lepak. 2009. Th ough he looking glass o a social sys em: C oss-le el e ec s o high-
pe o mance wo k sys ems on employees’a i udes. Pe sonnel Psychology 62: 1–29. [C ossRe ]
To ia, E manno C., Sil ia Sacche i, and F ancisco J. López-A ceiz. 2022. A human g ow h pe spec i e on sus ainable HRM p ac ices,
wo ke well-being and o ganiza ional pe o mance. Sus ainabili y 14: 11064.
Tsu uoka, Yoshimasa, Jun’ichi Tsujii, and Sophia Ananiadou. 2009. S ochas ic g adien descen aining o l1- egula ized log-
linea models wi h cumula i e penal y. P esen ed a he Join Con e ence o he 47 h Annual Mee ing o he ACL and he 4 h
In e na ional Join Con e ence on Na u al Language P ocessing o he AFNLP, Singapo e, Augus 2–7; pp. 477–85.
V on is, Deme is, Michael Ch is o i, Vijay Pe ei a, Shlomo Ta ba, Anna Mak ides, and Eleni T ichina. 2022. A i icial in elligence,
obo ics, ad anced echnologies and human esou ce managemen : A sys ema ic e iew. The In e na ional Jou nal o Human
Resou ce Managemen 33: 1237–66. [C ossRe ]
W igh , Pa ick M., Timo hy M. Ga dne , Lisa M. Moynihan, and Ma hew R. Allen. 2005. The ela ionship be ween HR p ac ices and
i m pe o mance: Examining causal o de . Pe sonnel Psychology 58: 409–46. [C ossRe ]
Xiang, Ting, Ping Zhen Wu, and Shihai Yuan. 2022. Applica ion analysis o combining bp neu al ne wo k and logis ic eg ession in
human esou ce managemen sys em. Compu a ional In elligence and Neu oscience 2022: 7425815. [C ossRe ]
Yan, Liang, Jing Zhao, Qian Zhang, and Ma y D. Sass. 2022. Does high-pe o mance wo k sys em b ing job sa is ac ion? Explo ing
he non-linea e ec o high-pe o mance wo k sys em using he ‘ oo much o a good hing’ heo y. Jou nal o Managemen &
O ganiza ion 2022: 1–25. [C ossRe ]
Disclaime /Publishe ’s No e: The s a emen s, opinions and da a con ained in all publica ions a e solely hose o he indi idual
au ho (s) and con ibu o (s) and no o MDPI and/o he edi o (s). MDPI and/o he edi o (s) disclaim esponsibili y o any inju y o
people o p ope y esul ing om any ideas, me hods, ins uc ions o p oduc s e e ed o in he con en .
