He, Shouhui; Xu, Si; Zou, A ia Xianya
A icle
Does g een echnological di e si ica ion impac indus ial
ca bon emissions e iciency? The ole o echnological
specialisa ion
Jou nal o Inno a ion & Knowledge (JIK)
P o ided in Coope a ion wi h:
Else ie
Sugges ed Ci a ion: He, Shouhui; Xu, Si; Zou, A ia Xianya (2025) : Does g een echnological
di e si ica ion impac indus ial ca bon emissions e iciency? The ole o echnological
specialisa ion, Jou nal o Inno a ion & Knowledge (JIK), ISSN 2444-569X, Else ie , Ams e dam, Vol.
10, Iss. 3, pp. 1-13,
h ps://doi.o g/10.1016/j.jik.2025.100730
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/327625
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-nc-nd/4.0/
Does g een echnological di e si ica ion impac indus ial ca bon emissions
e iciency? The ole o echnological specialisa ion
Shouhui He
a
, Si Xu
b
, A ia Xianya Zou
c,*
a
School o Logis ics, Linyi Uni e si y, Linyi 276000, China
b
School o In o ma ion Enginee ing, Hangzhou Voca ional & Technical College, Hangzhou 310018, China
c
School o Business, Macau Uni e si y o Science and Technology, Taipa, Macao, China
ARTICLE INFO
JEL codes:
C21
L25
O30
O32
O33
Q59
Keywo ds:
Rela ed g een echnological di e si ica ion
Un ela ed g een echnological di e si ica ion
Indus ial ca bon emissions e iciency
Technological specialisa ion
Pa en ex da a
ABSTRACT
Indus y is he main sou ce o ca bon emissions, and educing emissions is closely linked o ad ancing ca bon
neu ali y. Mo eo e , echnological inno a ion is c ucial o indus ial low-ca bon de elopmen . This s udy uses
p o incial panel da a om China co e ing 2010 o 2023 and employs he S ochas ic Non-smoo h En elopmen o
Da a model o measu e indus ial ca bon emissions e iciency. Combining lis ed companies’ g een pa en ex
da a, we calcula e ela ed and un ela ed g een echnological di e si ica ion o in es iga e he impac o ela ed
and un ela ed g een echnological di e si ica ion on indus ial ca bon emissions e iciency using a panel quan ile
model. The s udy also explo es he mode a ing ole o echnological specialisa ion, e ealing ha ela ed and
un ela ed g een echnological di e si ica ion p omo e indus ial ca bon emissions e iciency a di e en quan ile
le els; howe e , he posi i e e ec diminishes as quan ile le els ise. Fu he mo e, ela ed g een echnological
di e si ica ion has a s onge impac han un ela ed g een echnological di e si ica ion, and ela ed and un e-
la ed g een echnological di e si ica ion ha e a g ea e e ec on accele a ing indus ial ca bon emissions e i-
ciency in non-coas al, low-income and low ma ke sha e egions. Finally, echnological specialisa ion has a
s onge complemen a y impac on ela ed g een echnological di e si ica ion han on un ela ed g een echno-
logical di e si ica ion in enhancing indus ial ca bon emissions e iciency. The conclusions o his s udy p o ide
no el insigh s in o ad ancing indus ial g een ans o ma ion.
In oduc ion
Clima e change poses se ious, long- e m challenges o human socie y
(Liu & Feng, 2022a). E en s a ibu ed o global wa ming esul ing in
economic losses and casual ies ha e become inc easingly equen in
many coun ies. One o he p ima y ac o s con ibu ing o global
wa ming is inc eased ca bon emissions. Acco ding o Ca bon Moni o
da a, indus ial p oduc ion is he p ima y sou ce o hese emissions (Hu
e al., 2024). As indus ialisa ion ad anced, China’s ca bon emissions
su ged, making he na ion he wo ld’s la ges ca bon emi e
(F iedlings ein e al., 2019). In ecen yea s, China has p io i ised he
issue o clima e change, implemen ing a se ies o ca bon mi iga ion
measu es. China has also commi ed o eaching peak ca bon emissions
by 2030 and ca bon neu ali y by 2060; howe e , he na ion is s ill in a
phase o deepening economic de elopmen , and he demand o ene gy
emains s ong. The e o e, i is necessa y and meaning ul o use China as
a case s udy o explo e sui able and e ec i e ca bon mi iga ion
s a egies and sha e hese app oaches and expe iences wi h o he
eme ging coun ies. Ca bon emissions e iciency e lec s he e ec i e-
ness o ca bon mi iga ion om inpu and ou pu pe spec i es, which
aligns wi h he p ac ical need o balance economic de elopmen and
ca bon mi iga ion. Ca bon emissions e iciency has o en been measu ed
using a single- o mul i- ac o index. The single- ac o index has ypi-
cally been cons uc ed using me ics such as he a io o ca bon emis-
sions o o al ene gy consump ion (Liu e al., 2022) and ca bon
emissions pe uni o added alue (Zhang e al., 2024a). Al hough he
calcula ion o he single- ac o index is s aigh o wa d, i does no ac-
coun o o he ac o s ha in luence ca bon emissions e iciency.
Compa ed wi h he single- ac o index, a mul i- ac o s index can p o ide
a mo e accu a e e lec ion o ca bon emissions e iciency as i can ac-
coun o mul iple inpu s and desi able and undesi able ou pu s.
The p ima y me hods o assessing mul i- ac o ca bon emissions
e iciency is s ochas ic on ie analysis (SFA) and da a en elopmen
analysis (DEA). Fo example, Zhang and Chen (2021) used he SFA
* Co esponding au ho a : School o Business, Macau Uni e si y o Science and Technology, Taipa, Macao, China.
E-mail add ess: [email p o ec ed] (A.X. Zou).
Con en s lis s a ailable a ScienceDi ec
Jou nal o Inno a ion & Knowledge
jou nal homepage: www.else ie .com/loca e/jik
h ps://doi.o g/10.1016/j.jik.2025.100730
Recei ed 24 Decembe 2024; Accep ed 6 May 2025
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
A ailable online 22 May 2025
2444-569X/© 2025 The Au ho s. Published by Else ie España, S.L.U. on behal o Jou nal o Inno a ion & Knowledge. This is an open access a icle unde he CC
BY-NC-ND license ( h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/ ).
model o e alua e o e all ca bon emissions e iciency. Howe e , his
model has limi a ions due o i s speci ic unc ional o m and i is only
sui able o a single ou pu (Dellni z & Kleine, 2019). Among a ious
DEA models, he slacks-based measu e (SBM) is widely used o p o-
duc i i y assessmen s, pa icula ly when dealing wi h undesi able ou -
pu s. Wen e al. (2022) used he SBM me hod o e alua e ca bon
emissions e iciency. Ne e heless, slacks in inpu s and ou pu s can
p oduce inaccu acies in he DEA model’s assessmen o indus ial ca -
bon emissions e iciency, esul ing in ecen s udies inc easingly
adop ing modi ied DEA models. Fo ins ance, Xu e al. (2023) applied
he Supe -SBM model o assess ca bon emissions e iciency in he con-
s uc ion sec o , and Zhang e al. (2023) used a Supe -SBM-DEA model
o measu e ca bon emissions e iciency ac oss 32 indus ial sec o s.
Addi ionally, Zhang e al. (2024b) employed a h ee-s age SBM-DEA
model o analyse he ca bon emissions e iciency o land use in he
Yang ze Ri e Del a u ban agglome a ion om 2010 o 2021.
Ca bon mi iga ion is c ucial o ad ancing he sus ainable g ow h o
he global economy, and echnological inno a ion has a i al ole in
educing ca bon emissions (Xie e al., 2021). As echnological inno a-
ion in ol es signi ican isks and expenses, i equi es subs an ial
echnological capabili ies. One aspec o echnological inno a ion is
di e si ica ion, which e e s o companies b oadening inno a ion e o s
ac oss a ious echnological domains (B eschi e al., 2003). Techno-
logical di e si ica ion can be ca ego ised in o ela ed and un ela ed
ypes based on he connec ions be ween new and exis ing echnologies
(Kim e al., 2009). Rela ed g een echnological di e si ica ion ocuses on
inno a ion in mul iple specialised a eas o a speci ic ca ego y, whe eas
un ela ed g een echnological di e si ica ion spans di e en ca ego ies
(He e al., 2017). Ex ensi e esea ch has demons a ed ha g een
echnological inno a ion signi ican ly boos s ene gy e iciency
(Yasmeen e al., 2023). The concep o ela edness is c ucial in he g een
di e si ica ion p ocess (Balland & Boschma, 2021). G een echnologies
o en in ol e inno a i e combina ions o p e iously unused echnical
elemen s, which inc eases hei complexi y and unce ain y compa ed
wi h non-g een echnologies (Ba bie i e al., 2020b). These echnologies
equi e c oss-domain in eg a ion in which knowledge om a ious
ields con e ges o d i e g oundb eaking inno a ions (Ding e al.,
2022a). Al hough g een echnologies ini ially eme ged in de eloped
na ions, he need o echnological di e si ica ion is pa icula ly p ess-
ing o de eloping coun ies (Pe uchas e al., 2020).
Se e al s udies ha e explo ed how echnological di e si ica ion im-
pac s egional economic g ow h. Balland e al. (2020) examined he
ela ionship be ween echnological di e si ica ion and u ban economic
ac i i ies. Ba bie i e al. (2020b) de e mined ha g een echnologies
ha e signi ican ly g ea e knowledge spillo e e ec s han non-g een
echnologies and geog aphic p oximi y is essen ial o c ea ing and
sp eading complex knowledge. The e ec i eness o g een echnology
depends on a ious ac o s, including geog aphy and echnological
de elopmen (Mo eno & Ocampo-Co ales, 2022). Zheng and Ran
(2021) used China’s pa en s o conduc an empi ical analysis, inding
ha un ela ed echnological di e si ica ion has a g ea e impac on
egional economic g ow h han ela ed di e si ica ion. Rocche a e al.
(2022) analysed a sample o 268 Eu opean egions o demons a e he
nonlinea e ec s o echnological di e si ica ion on egional p oduc-
i i y. Howe e , he e ec s o ela ed and un ela ed g een echnological
di e si ica ion on ca bon mi iga ion in de eloping economies ha e no
been examined. Cu en esea ch has la gely assessed echnological di-
e si y using en opy indices (Kim e al., 2016; Choi & Lee, 2021; Lin
e al., 2023) and ocusing on he quan i y o echnologies. In addi ion o
echnological di e si ica ion, echnological specialisa ion is ano he key
ac o o echnological inno a ion. Specialising in g een echnology
p omo es he clus e ing o inno a i e ac i i ies, inc eases g een
knowledge concen a ion and educes he cos s o con e ing aci
knowledge (Liao & Li, 2022). Fu he mo e, specialisa ion suppo s la-
bou ma ke g ow h ocused on speci ic s ages and signi ican ly educes
he ansac ion cos s ela ed o g een inno a ion p ocesses (Pe e ,
2022). Many s udies ha e demons a ed ha echnological specialisa-
ion signi ican ly boos s economic ac i i ies na ionwide. Fo example,
Liao and Li, 2022 analysed he e ec s o g een echnology specialisa ion
on g een de elopmen , and Se gio e al. (2023) explo ed he signi ican
spillo e e ec s o specialisa ion in high- ech sec o s on inno a ion
pe o mance in membe coun ies o he O ganisa ion o Economic
De elopmen and Co-ope a ion. Howe e , esea ch on he impac o
echnological specialisa ion on ca bon educ ion has been limi ed.
Cu en me hods o assessing specialisa ion ha e p ima ily used loca-
ion en opy and he Ho inda index (Liao & Li, 2022). Fu he mo e, i is
c ucial o explo e e ec i e in eg a ion o g een echnological di e si i-
ca ion and echnological specialisa ion.
P e ious esea ch has o e ed aluable insigh s in o he ela ionship
be ween g een echnology di e si ica ion and specialisa ion in e ms o
ca bon emissions; howe e , se e al no able esea ch gaps emain.
Table 1 summa ises he exis ing esea ch and esea ch gaps. Fi s , he
cu en me hodology o e alua ing China’s ca bon e iciency equi es
signi ican imp o emen . DEA is ecognised as a de e minis ic and
en i ely non-pa ame ic app oach and is among he mos equen ly
employed echniques o assessing ca bon emissions e iciency (Wen
e al., 2022; Xu e al., 2023; Zhang e al., 2023, 2024b). Howe e , DEA
canno accoun o s a is ical noise, leading o he conclusion ha any
de ia ion om he e iciency on ie is solely due o ine iciency (Yen
e al., 2023). Second, China lacks speci ic indica o s o g een echnol-
ogy di e si ica ion, and mos ecen esea ch in his a ea ha e o igina ed
om Eu opean s udies (Cas ellani e al., 2022; Mon eso & Qua a o,
2020; Cice one e al., 2023; San oalha e al., 2021a, b; Ba bie i e al.,
2023). Exis ing me ics o echnological di e si ica ion and
Table 1
P e ious esea ch and esea ch gaps.
Resea ch heme Exis ing esea ch Resea ch gaps
Ca bon emissions
e iciency
measu emen
me hod
Many s udies ha e used DEA
models and hei a ian s o
measu e ca bon emissions
e iciency. While hese
me hods can accommoda e
mul i-inpu –ou pu
p oblems (Wen e al., 2022;
Xu e al., 2023; Zhang e al.,
2023, 2024b).
Neglec ing o accoun o
s a is ical noise.
G een echnological
di e si ica ion and
specialisa ion
indica o sys em
Exis ing measu es o
echnological di e si ica ion
and specialisa ion p ima ily
ely on en opy indices ha
ocus on he numbe and
dis ibu ion o echnologies
and canno e lec he
complexi y o g een
echnological di e si ica ion
(Kim e al., 2016; Choi &
Lee, 2021; Liao e al., 2022;
Lin e al., 2023).
Exis ing indica o
sys ems do no
su icien ly cap u e he
quali y o di e si ica ion
and specialisa ion.
Rela ionship be ween
echnological
di e si ica ion and
ca bon e iciency
Cu en s udies ha e
p ima ily ocused on he
impac o echnological
di e si ica ion on economic
g ow h (Kim e al., 2016;
Choi & Lee, 2021; Balland
e al., 2020; Ba bie i e al.,
2020b; Mo eno &
Ocampo-Co ales, 2022;
Zheng & Ran, 2021;
Rocche a e al., 2022; Lin
e al., 2023).
Few s udies ha e
examined he
ela ionship be ween
echnological
di e si ica ion and
ca bon emissions
e iciency, pa icula ly
he egula o y ole o
echnological
specialisa ion.
Regional he e ogenei y
analyses
Exis ing s udies ha e
p edominan ly concen a ed
on analysing egional
di e ences in ca bon
emissions e iciency (Wen
e al., 2022; Zhang e al.,
2023; Wei e al., 2024).
The impac o China’s
egional imbalance on
g een echnological
di e si ica ion and
ca bon e iciency is
ela i ely limi ed.
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
2
specialisa ion has p ima ily used en opy indices (Kim e al., 2016; Choi
& Lee, 2021; Liao & Li, 2022; Lin e al., 2023). Howe e , hese indices
may no ully cap u e he in icacies o g een echnological di e si ica-
ion, and elying on a single index is insu icien o ep esen he na u e
o echnological specialisa ion. Addi ionally, hese me hodologies ocus
on he quan i y and dis ibu ion o echnologies bu neglec he quali y
o echnology. Thi d, p e ious esea ch on he impac o g een ech-
nology di e si ica ion on ca bon emissions ha e been inadequa e,
pa icula ly in analysing egional dispa i ies. The e ec i eness o g een
echnology di e si ica ion is signi ican ly in luenced by local echno-
logical in as uc u e (G asho & Basilico, 2024).
To add ess hese gaps, we employ he S ochas ic Non-smoo h
En elopmen o Da a (S oNED) model o measu e ca bon emissions e -
iciency, de eloping an index o g een echnological di e si ica ion
using ela i e echnological densi y and g een pa en ex da a. Mo e-
o e , he esea ch explo es he impac o g een echnological di e si i-
ca ion on ca bon emissions e iciency while also conside ing egional
dispa i ies. We also examine how echnological specialisa ion mode -
a es he ela ionship be ween g een echnological di e si ica ion and
ca bon emissions e iciency. These indings p o ide speci ic policy ec-
ommenda ions o suppo China’s ca bon peak and neu ali y goals.
The main con ibu ions o his s udy a e as ollows. Fi s , in con as
o p e ious s udies ha ha e only used DEA o SFA, we use he S oNED
model in eg a ing DEA and SFA o measu e indus ial ca bon emissions
e iciency. This model can di ec ly inco po a e undesi able ou pu s and
o e s a simple calcula ion p ocess. Applica ion o he S oNED model o
calcula ing indus ial ca bon emissions e iciency has been limi ed.
Second, unlike p e ious s udies ha used he HI index based on he
numbe o pa en s (Kang & Sohn, 2016), we use g een pa en ex da a
and ollow h ee s eps o iden i y echnological ad an age, calcula e
echnological densi y and classi y ypes o echnological di e si ica ion.
Thi d, we in es iga e he ela ionship be ween g een echnological
di e si ica ion and indus ial ca bon emissions e iciency, conside ing
he e ogeneous de elopmen ac oss a ious egions. We also u he
analyse how echnological specialisa ion ampli ies he e ec o g een
echnological di e si ica ion on indus ial ca bon emissions e iciency.
The emainde o his pape is o ganised as ollows. Sec ion 2 p e-
sen s ou esea ch hypo heses. Sec ion 3 ou lines he a iables and he
model used in he s udy. Sec ion 4 p esen s he empi ical esul s, and
sec ion 5 p o ides he discussion. Finally, sec ion 6 d aws he conclu-
sions and p o ides he policy implica ions.
Resea ch hypo heses
We nex p esen he assump ions ha cla i y he ela ionship be-
ween g een echnological di e si ica ion and ca bon emissions
e iciency and he mode a ing ole o echnological specialisa ion. Fig. 1
illus a es ou heo e ical amewo k.
G een echnological di e si ica ion and ca bon emissions e iciency
G ans and and Sj¨
olande (1990) ini ially in oduced he concep o
echnological di e si ica ion o dis inguish i om p oduc di e si ica-
ion. Technology is conside ed o be a unique o m o knowledge (Xie
e al., 2021; Yen e al., 2023; Kang and Sohn, 2016; Ha is, 2021; Pin-
hei o e al., 2025; Pe e , 2022). Consequen ly, echnological di e si i-
ca ion is de ined as he deg ee o knowledge di e si ica ion among i ms
wi hin a egion (Ceipek e al., 2019). This concep e lec s he ongoing
p ocess o i ms’ inno a ion and he in eg a ion o new echnologies.
Technological di e si ica ion changes en e p ises’ echnological po -
olios and s uc u e, which imp o es p oduc quali y and p omo es new
p oduc de elopmen (Zheng & Ran, 2021). Fi ms engage in echno-
logical di e si ica ion o inc ease he likelihood o success ul echno-
logical inno a ion and imp o e p oduc e iciency (B eschi e al., 2003;
Ca al´
an e al., 2022). This s a egy helps o elimina e echnological
lock-in and educes inhe en esea ch and de elopmen (R&D) isks
(Ga cia-Vega M, 2006; Kim e al., 2016; Choi & Lee, 2021).
G een echnological inno a ions simul aneously enhance na u al
esou ce e iciency and educe ca bon emissions (Ba bie i e al., 2020b;
Meng e al., 2022). G een echnologies ely on in eg a ing knowledge
and expe ise om a ious sou ces (Ba bie i e al., 2020b; Li, Heime iks
& Alkemade, 2021; Mo eno & Ocampo-Co ales, 2022) and depend on
he con e gence o di e se knowledge ac oss di e en domains o
gene a e b eak h ough inno a ions (Ding e al., 2022a). G een ech-
nologies a e ypically mo e complex and no el han non-g een ech-
nologies, esul ing in highe cos s and slowe e u ns (Ba bie i &
Consoli, 2019; Ning & Guo, 2022). The eo ganisa ion o a ious ech-
nological ields p e en s echnology om being locked in o en i on-
men al pollu ion (San oalha & Boschma, 2021a) and acili a es he
eme gence o new g een echnologies (O sa i e al., 2020). G een
echnologies allow i ms o imp o e p oduc s and se ices du ing
echnological di e si ica ion, lowe ing cos s and enhancing ca bon
emissions e iciency. This e idence sugges s ha g een inno a ion de-
pends on i ms’ abili y o le e age he dep h and b ead h o echno-
logically di e se knowledge (Ning & Guo, 2022).
The e ec s o g een echnological di e si ica ion on emissions
educ ion demons a e pa h dependence (Hidalgo e al., 2018; Hidalgo,
2021; Ning & Guo, 2022). Rela ed g een echnology di e si ica ion
enhances ca bon e iciency h ough se e al mechanisms. (i) Co e g een
echnologies lowe he ma ginal cos s o g een inno a ion (B eschi e al.,
2003; Kim e al., 2016; Ba bie i & Consoli, 2019; Ning & Guo, 2022).
Technological ela edness is a c ucial d i ing o ce o echnological
Fig. 1. Theo e ical amewo k o g een echnological di e si ica ion on ca bon emissions e iciency.
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
3
di e si ica ion (F enken & Boschma, 2007; Boschma, 2017; Yeung,
2021; San oalha & Boschma, 2021a). (ii) Simila i ies and complemen-
a i ies ac oss g een echnologies can enhance knowledge dissemina ion
e iciency wi hin a egion (Boschma, 2017). Fi ms a e mo e likely o
de elop g een echnologies wi hin amilia domains, can es ablish new
compa a i e ad an ages (Yeung, 2021). In con as , he emissions
educ ion e ec s o un ela ed g een echnology di e si ica ion ace wo
main cons ain s. Fi s , while i is e ec i e in o e coming echnological
lock-in Le en e al. (2007), c oss-domain knowledge explo a ion ypi-
cally equi es longe de elopmen - o-comme cialisa ion cycles o
eme ging g een echnologies (Hausmann & Hidalgo, 2011; Choi & Lee,
2021). Second, di e si ying in o un amilia echnological a eas en ails
addi ional cos s o knowledge in eg a ion and o ganisa ional coo di-
na ion (Se u, 2014), which may pa ially coun e ac sho - e m emis-
sions educ ion gains (Hausmann & Hidalgo, 2011). Un ela ed g een
echnology di e si ica ion can s imula e he de elopmen o b eak-
h ough g een echnologies bu i s posi i e impac on ca bon emissions
e iciency is condi ional and ime-lagged (Dawley e al., 2019; MacK-
innon e al., 2019; Ha is, 2020; F angenheim e al., 2020; Choi & Lee,
2021). Based on his analysis, we p opose he ollowing hypo heses:
H1a.Rela ed g een echnological di e si ica ion has a posi i e
impac on ca bon emissions e iciency.
H1b.While un ela ed g een echnological di e si ica ion also has
a posi i e impac on ca bon emissions e iciency, i is smalle han
ha o ela ed echnological di e si ica ion.
Impac o egional he e ogeneous cha ac e is ics
While clima e change is a global conce n, implemen ing localised
inno a i e s a egies is essen ial o e ec i ely add ess his challenge
(Pe uchas e al., 2020; Liu e al., 2024). The di e se cha ac e is ics o
di e en egions, s emming om a complex his o ical con ex , can make
s a egies di icul o eplica e in o he a eas (Maskell & Malmbe g,
1999). These cha ac e is ics encompass a ious ac o s such as he
egional economy, geog aphical loca ion and esou ce endowmen .
Fu he mo e, he equency o echnological inno a ion ac i i ies is
closely linked o hese egional ac o s (Du e al., 2019; Milindi &
Inglesi-Lo z, 2022). G een echnology p o essionals’ dis ibu ion ac oss
egions is une en, and signi ican dispa i ies exis in he echnological
capabili ies needed o de elop new g een ini ia i es (Co adini, 2019;
Ba bie i & Consoli, 2019; Pe uchas e al., 2020).
G een echnologies ha depend on complex knowledge a e o en
localised (Dosi e al., 2017). The gene a ion and sp ead o knowledge
can be in luenced by local socio-economic and cul u al ac o s
(Pe uchas e al., 2020). Rela ed g een echnological di e si ica ion is
mo e elian on he local echnological ounda ions han un ela ed
di e si ica ion (Mon eso & Qua a o, 2020; San oalha & Boschma,
2021a) as egions p io i ise echnologies ha align wi h exis ing
capabili ies.
Fi s , high-income egions p omo e g een echnological di e si ica-
ion wi h obus human capi al, a ied knowledge bases and well-
de eloped in as uc u e (Du e al., 2014; Pinhei o e al., 2022). This
di e si ica ion enhances economies o scale by p omo ing knowledge
spillo e s and echnology accumula ion. Howe e , once echnological
di e si ica ion exceeds a ce ain h eshold, i s ma ginal bene i s may
begin o decline (Lin e al., 2023). Excessi e echnological di e si ica-
ion can lead o an o e ly complex knowledge sys em, inc easing in-
o ma ion cos s and impeding knowledge di usion (Pe uchas e al.,
2020). Second, coas al a eas a e cha ac e ised by sophis ica ed ans-
po a ion in as uc u e and a di e se knowledge– echnology base
(G asho e al., 2024). Ne e heless, a high deg ee o echnological
agglome a ion may educe esou ce alloca ion e iciency. In hese a eas,
he bene i s o g een echnological di e si ica ion depend on inno a ion
capabili ies and he abili y o abso b inno a ions, and i s posi i e impac
ends o decline when exceeding he op imal scale (Liu e al., 2021).
Thi d, while ma ke -o ien ed en i onmen s p omo e in o ma ion
sp ead, hey can also in oduce compe i ion ha hinde s g een ech-
nology de elopmen (Hua e al., 2022). In addi ion, echnological isks
such as luc ua ing enewable ene gy p ices and unce ain e u ns on
in es men , along wi h esou ce misma ches due o ma ke segmen a-
ion, could impede he ansi ion o a low-ca bon economy (A ha i and
Ki ikkaleli, 2024; Ding e al., 2022b; Li e al., 2023).
The e o e, whe he egional he e ogenei y is ela ed g een echno-
logical di e si ica ion o un ela ed g een echnological di e si ica ion,
we p opose he ollowing hypo heses:
H2a.G een echnological di e si ica ion has a g ea e impac on
ca bon emissions e iciency in low-income egions han high-
income egions.
H2b.The e ec o g een echnological di e si ica ion on ca bon
emissions e iciency is s onge in non-coas al egions han coas al
egions.
H2c.G een echnological di e si ica ion in low ma ke egions
has a g ea e in luence on ca bon emissions e iciency compa ed
wi h highly ma ke -o ien ed egions.
Mode a ing ole o echnological specialisa ion
The s udy o specialisa ion can be aced back o Adam Smi h’s
p oposi ions in his seminal wo k The Weal h o Na ions (Smi h, 1937).
Subsequen ly, he concep o e ical specialisa ion was in oduced in o
he ield o in e na ional ade in he 1960s and 1970s (Hummels e al.,
2001). Technological specialisa ion i s eme ged in i ms’ echnological
decision-making (G ans and & Oska sson, 1994). Fi ms ha adop his
s a egy can enhance co e compe encies in amilia echnological ields
a lowe cos s (B eschi e al., 2003; Ga cia-Vega, 2006; Liao & Li, 2022).
Regions achie e specialisa ion in pa icula echnological ields, which
educes i ms’ R&D cos s in ela ed a eas (Kim e al., 2016; Choi & Lee,
2021). G een echnology specialisa ion os e s spa ial clus e ing o
speci ic inno a ion ac i i ies, and his clus e ing acili a es in ensi e
knowledge sha ing du ing he inno a ion phase and minimises he
ansac ion cos s associa ed wi h aci knowledge ans e (Kemeny
e al., 2022). Addi ionally, g een echnology specialisa ion p omo es
sha ed labou ma ke s o speci ic inno a ion s ages, which educes
employmen cos s and s abilises inno a ion ac i i ies (Pe e , 2021).
G een echnology specialisa ion is oo ed in common in e es s and
long- e m cul u al accumula ion, which encou ages collabo a ion
among i ms, enhancing inno a ion e iciency (Liao & Li, 2022). G een
echnological di e si ica ion o e s g ea e bene i s o ca bon emissions
e iciency, which a e mo e p onounced o specialised i ms (Ca al´
an
e al., 2022).
Howe e , echnological di e si ica ion and specialisa ion a e wo
dis inc s a egies ha can guide i ms’ echnological inno a ion. Pu -
suing bo h s a egies simul aneously can be pa icula ly challenging o
i ms wi h limi ed esou ces (Chen & Guo, 2023a). G een echnological
di e si ica ion equi es highe coo dina ion and in eg a ion cos s ac-
co ding o Ning and Guo (2022). The e o e, compe i i e i ms may
p e e o en e ma ke s ha a e simila o hei exis ing esou ce p o iles
o educe he cos o echnology ans e (Chiu e al., 2008; Kemeny
e al., 2022). Fi ms can achie e economies o scope by sha ing g een
echnology esou ces, which include echnical expe ise and skilled la-
bou (Ning & Guo, 2022). Howe e , de eloping echnology-di e si ied
indus ies ela ed o a co e indus y has diminishing e u ns o scale
o R&D as in es men in a speci ic echnological ield accumula es (Kim
e al., 2016; Choi & Lee, 2021). G een inno a ion ac i i ies equi e
di e se knowledge capabili ies (A di o e al., 2019) and signi ican
sys emic changes (Ning & Guo, 2022). Inno a o s who engage in
ecombina ion wi h cogni i e dis ance may ace inc eased unce ain y
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
4
(Ba bie i e al., 2020a). Technological specialisa ion can mi iga e he
diminishing e u ns associa ed wi h echnological di e si ica ion
somewha . Howe e , his is pa icula ly challenging in un ela ed ech-
nological domains wi h high unce ain y (Choi & Lee, 2021; G asho
e al., 2024). Based on he abo e, we p opose he ollowing hypo heses:
H3a.G een echnological specialisa ion posi i ely mode a es he
ela ionship be ween ela ed g een echnological di e si ica ion
and ca bon emissions e iciency.
H3b.G een echnological specialisa ion has a g ea e comple-
men a y e ec on ela ed g een echnological di e si ica ion han
un ela ed g een echnological di e si ica ion.
Va iables and models
Va iables desc ip ion
Explained a iable: indus ial ca bon emissions e iciency (CEE)
Re e encing Dai e al. (2024), we employ he S oNED me hod o
assess he indus ial CEE o 30 p o inces and ci ies in China. The
inpu –ou pu indica o s a e de ailed as ollows:
(1) We use ene gy, capi al and labou as inpu indica o s, which a e
ep esen ed by he end consump ion o a ious ene gy sou ces in
he indus ial sec o , he a e age ne alue o indus ial ixed
capi al and he a e age employmen o egional indus y.
(2) The expec ed ou pu indica o is alue added by egional
indus y.
(3) The unexpec ed ou pu indica o is he amoun o indus ial
ca bon emissions calcula ed by he In e go e nmen al Panel on
Clima e Change, which is based on indus ial ene gy consump-
ion. The de ailed o mula is as ollows:
CO2=∑
n
i=1
(CO2)i=∑
n
i=1
Ei×NCVi×CEFi×COFi× (44/12)(1)
whe e CO
2
ep esen s he amoun o indus ial ca bon emissions, i de-
no es he ype o ene gy, Ei signi ies he o al indus ial ene gy con-
sump ion, NCVi e e s o he ne calo i ic alue o he speci ic ype o
ene gy, CEFi indica es he ca bon emissions coe icien pe uni o
calo i ic alue and COFi ep esen s he ca bon monoxide ac o . Addi-
ionally, e e encing Chen (2009), we assign coal a ca bon oxida ion
ac o o 0.99, while c ude oil and na u al gas a e assigned a ca bon
oxida ion ac o o 1. The molecula weigh s o ca bon dioxide and
ca bon a e assigned as 44 and 12, espec i ely. This a io is e e ed o as
he CO
2
gasi ica ion coe icien .
Explana o y a iables: ela ed g een echnology di e si ica ion (RD) and
un ela ed g een echnology di e si ica ion (UD)
We employ a h ee-s ep app oach o measu e ela ed (RD) and un-
ela ed (UD) echnological di e si ica ion.
Fi s , we iden i y he e ealed echnological ad an age (RTA) using a
me hod adap ed om Hidalgo e al. (2007) o e lec echnological
specialisa ion as ollows:
RTAij =⎧
⎪
⎨
⎪
⎩
1,
Pa en ij /∑jPa en ij
∑jPa en ij /∑i∑jPa en ij
≥1
0,else
(2)
whe e i ep esen s he p o ince, j deno es he echnology ield as de ined
by he ou -digi In e na ional Pa en Classi ica ion code. Pa en
ij
in-
dica es he numbe o pa en s in p o ince i a ime in echnology ield
j. RTA
ij
is a bina y a iable used o de e mine whe he p o ince i a
ime in echnology ield j possesses a compa a i e ad an age. When
RTA
ij
=1, i indica es ha p o ince i has achie ed a echnological
ad an age a ime in echnology ield j, o he wise i signi ies ha
p o ince i has no a ained a echnological ad an age a ime in ech-
nology ield j.
Second, we calcula e echnological densi y based on he iden i ied
RTA, ollowing he me hodology ou lined by Balland and Boschma
(2021). This s ep e lec s he capabili ies o exis ing specialised
echnologies.
Densi yij =∑jʹ∈i,jʹ∕∈jϕjjʹ
∑jʹ∕=jϕjjʹ
*100 (3)
whe e j ep esen s a echnological ield ha has no ye been specialised
bu is highly ela ed o he specialised echnological ield j’. We de e -
mine he ela edness o echnological ields using he p inciple o co-
occu ence o pa en applican in o ma ion wi hin he same pa en .
Densi y
ij
deno es he capabili ies o echnology ield j in p o ince i a
ime , which we measu e using he a io o he sum o all compa a i e
ad an age echnological ields j’ ela ed o echnological ield j a ime
o p o ince i, o he sum o all echnological ields j’ ela ed o ech-
nological ield j a ime .
Thi d, we u he assess ela ed and un ela ed g een echnological
di e si ica ion based on echnological densi y, as e e enced by Zheng
and Ran (2021).
ω
ij =Densi yij − 〈Densi yi 〉O
σ
O(Densi yi )(4)
whe e O ep esen s he collec ion o echnology ields ha do no exhibi
a compa a i e ad an age in p o ince i a ime . 〈Densi yi 〉O deno es he
a e age echnological capabili ies wi hin he ensemble.
σ
O(Densi yi )
ep esen s he s anda d de ia ion o he echnological capabili ies in he
ensemble. When
ω
ij >0, i indica es ha he capabili y o echnological
ield j in p o ince i is ela i ely ample and is ela ed o g een echno-
logical di e si ica ion. When
ω
ij <0, i sugges s ha he capabili y o
echnological ield j in p o ince i is ela i ely small and ep esen s un-
ela ed g een echnological di e si ica ion.
In summa y, we quan i y he numbe o echnological ields in
ela ed and un ela ed g een echnological di e si ica ion as he alues o
RD and UD a iables, espec i ely. RD and UD ep esen he ela ed and
un ela ed g een echnological di e si ica ion o p o ince i a ime . A
highe RD alue indica es ha a egion demons a es a s onge p o-
pensi y o achie e di e si ica ion wi hin ele an g een echnology do-
mains. Con e sely, a g ea e UD alue sugges s ha he egion’s g een
echnology di e si ica ion ac i i ies a e inc easingly ocused on un e-
la ed ields.
Mode a ing a iable: echnological specialisa ion (TS)
The mode a ing a iable is TS. Re e encing Balland and Boschma
(2021), when echnological ield j in p o ince i a ime has a
compa a i e ad an age, his indica es ha p o ince i has achie ed TS in
echnological ield j a ime . The e o e, TS =1, o he wise, TS =0.
Con ol a iables
The con ol a iables include indus y s uc u e (IS), ene gy s uc-
u e (ES) and owne ship s uc u e (MS). is measu ed by he p opo ion
o seconda y indus y alue added in egional GDP, ES is calcula ed as
he p opo ion o coal consump ion in he o al indus ial ene gy con-
sump ion in he p o ince and MS is ep esen ed by he a io o he
numbe o employees in s a e-owned en e p ises o he o al numbe o
employees in he egion (Kang & Kim, 2012).
Due o he absence o da a om Tibe , we use panel da a om 30
p o inces and egions in China co e ing 2010–2023. As desc ibed, he
inpu –ou pu da a o CEE include labou , capi al and indus ial ou pu ,
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
5
which we sou ce om he China S a is ical Yea book and he China
Indus ial Economy S a is ics Yea book. We adjus capi al and indus ial
ou pu s acco ding o he ixed asse s in es men p ice index and he
indus ial p oduce p ice index published by he China Na ional Bu eau
o S a is ics. Da a on ene gy consump ion and indus ial ca bon emis-
sions a e ob ained om he China Ene gy S a is ics Yea book and con-
e ed in o s anda d coal equi alen . Addi ionally, we use g een pa en
ex da a o lis ed companies in he China Na ional Knowledge In a-
s uc u e pa en da abase o 2010–2023. Da a o con ol a iables a e
also sou ced om he China S a is ical Yea book. Any missing da a a e
supplemen ed by he annual a e age g ow h a e and in e pola ion
me hods. Table 2 p esen s he desc ip i e s a is ics o all he a iables.
To es he a iables’ s a iona i y, we conduc he Fish-
e –Phillips–Pe on uni oo es o each a iable, p esen ing he esul s
in Table 3. Bo h he ho izon al sequence and he i s -o de di e ence
sequence o all he a iables ejec he null hypo hesis ( ha a uni oo
exis s) a a 1 % signi icance le el. This indica es ha all he a iables a e
s able.
Model cons uc ion
S oNED model
CEE e alua ion o en unco e s undesi able ou pu s. DEA is widely
used o assess complex sys ems wi h mul iple inpu s and ou pu s and is
pa icula ly e ec i e in add essing undesi able ou pu s, such as ca bon
emissions (Wen e al., 2022; Xu e al., 2023; Zhang e al., 2023, 2024b).
Chung e al. (1997) p oposed a no el adial di ec ional dis ance unc ion
(DDF) ha p o ides an ad anced analy ical amewo k o e alua ing
en i onmen al e iciency in he con ex o undesi able ou pu s. G´
ema
e al. (2018) used a non- adial DEA model o assess CEE. Fe ei a e al.
(2023) inco po a ed unce ain y ac o s o e alua e eco-e iciency,
which enhanced he obus ness and eliabili y o he assessmen
amewo k. Fu he mo e, Sala-Ga ido e al. (2023) in eg a ed
c oss-e iciency echniques in o he eco-e iciency e alua ion ame-
wo k. In a ecen s udy, Chen e al. (2023b) in oduced a empo al
dimension in o hei analysis. T adi ional DEA me hods a e ulne able
o s a is ical noise ha can dis o es ima es o CEE. SFA has he sig-
ni ican limi a ion o equi ing a speci ic unc ional o m o p oduc ion
echnology. Kuosmanen and Ko elainen (2012) in oduced he S oNED
echnique o add ess hese challenges, which combines he
non-pa ame ic app oach o DEA wi h he s ochas ic on ie amewo k
o SFA o simul aneously accommoda e lexibili y and noise. This
me hod main ains he non-pa ame ic na u e o DEA and elimina es he
need o speci y a p ede ined p oduc ion unc ion. I also inco po a es he
s ochas ic e o e m de i ed om SFA, which helps decompose de-
ia ions om he e iciency on ie in o andom e o and ine iciency
componen s. The S oNED model is cons uc ed as ollows:
yi = (xi ) +
ε
i ,
ε
i =ui− i (5)
whe e x ep esen s he inpu s, y deno es he ou pu s and signi ies he
bounda y o he p oduc ion possibili y se . We decompose he dis u -
bance e m (
ε
i ) in o a non-e iciency componen (ui) and a andom e o
e m ( i ), whe e ui∼N(0,δ2
u), i ∼N(0,δ2
).
This s udy employs a mul i-s ep app oach o es ima e he S oNED
model. Fi s , we apply he ollowing CNLS-DDF me hod o es ima e
pa ame e s based on mul iple inpu and ou pu da a:
min ∑
n
i=1
ε
2
i
s. .
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
γʹ
iyi=
α
i+βʹ
ixi+δʹ
ibi−
ε
i
α
i+βʹ
ixi+δʹ
ibi−γʹ
iyi≤
α
j+βʹ
jxi+δʹ
jbi−γʹ
jyi
βi≥0,δʹ
i≥0,γi≥0
∀i,j,and,i∕= j
(6)
whe e he i s cons ain de ines he dis ance o he on ie as a linea
unc ion o inpu s and ou pu s. The second se o cons ain s includes
A ia inequali ies, which impose global conca i y. The hi d cons ain
is a no malisa ion cons ain ha ensu es he ansla ion p ope y. The
ou h cons ain imposes a mono onically inc easing limi on he p o-
duc ion unc ion.
Second, we use he ollowing condi ional expec a ion as he poin
es ima o o he i s momen o he non-e iciency e m ui:
Thi d, es ima ion o e iciency is calcula ed using o mula (7) as
ollows:
Ei =yi
(xi )=1−E(ui)
yi +E(ui)− E( i)+E( i )
yi +E(ui)− E( i)=1−E(ui)
yi +E(ui)(8)
To u he in es iga e he egional cha ac e is ics o China’s indus-
ial CEE, we ca ego ise 30 p o inces and municipali ies in o eigh
comp ehensi e economic zones based on he S a egy and Policy o
Regional Coo dina ed De elopmen . These zones a e he No heas Re-
gion (Liaoning, Jilin and Heilongjiang), he No h Coas al Region (Bei-
jing, Tianjin, Hebei and Shandong), he Eas Coas al Region (Shanghai,
Jiangsu and Zhejiang), he Sou h Coas al Region (Fujian, Guangdong
and Hainan), he Middle Yellow Ri e Region (Shaanxi, Shanxi, Henan
and Inne Mongolia), he Middle Yang ze Ri e Region (Hubei, Hunan,
Jiangxi and Anhui), he Sou hwes Region (Yunnan, Guizhou, Sichuan,
Table 2
Desc ip i e s a is ics.
Va iables Obs Mean SD Min Max
CEE 19,295 0.999 0.008 0.369 1.000
RD 19,295 1021.665 389.6049 1.000 1688
UD 19,295 2415.084 652.870 0.000 3445
TS 19,295 0.611 0.487 0 1
IS 19,295 0.374 0.076 0.149 0.601
ES 19,295 0.250 0.051 0.045 0.313
MS 19,295 0.248 0.154 0.034 0.645
Table 3
Uni oo es esul s.
Va iables Le el Fi s di e ence
chi-squa ed p- alue chi-squa ed p- alue
CEE 306.0342*** 0.000 372.4801*** 0.000
RD 305.8847*** 0.000 367.1516*** 0.000
UD 308.8834*** 0.000 367.4188*** 0.000
IS 295.7938*** 0.000 366.1619*** 0.000
ES 293.2550*** 0.000 364.5311*** 0.000
MS 302.2413*** 0.000 368.4304*** 0.000
No es: Null hypo hesis: a uni oo exis s. ***, **and * deno e 1 %, 5 % and 10 %
signi icance le els, espec i ely. p- alues a e in pa en heses.
E(
μ
i/
ε
i) =
μ
*+δ2
*[φ(−
μ
*/δ*)
1−ϕ(1−
μ
*/δ*)],
μ
*= −
ε
iδ2
u/(δ2
u+δ2
),δ2
*=δ2
uδ2
/(δ2
u+δ2
)(7)
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
6
Chongqing and Guangxi) and he No hwes Region (Gansu, Qinghai,
Ningxia and Xinjiang). The CEE o hese eigh comp ehensi e economic
zones and he en i e coun y exhibi s he ollowing cha ac e is ics.
A he na ional le el, CEE exhibi ed a end o ini ial decline ollowed
by a ise om 2010 o 2023. Addi ionally, he CEE in eas coas and
no h coas egions was gene ally high, while i was compa a i ely
lowe in he sou h coas egion. The eas coas and no h coas egions
bene i om signi ican geog aphical ad an ages and ha e high-quali y
indus ies; howe e , en i onmen al p o ec ion e o s in he sou h coas
egion ha e no kep pace wi h economic de elopmen . The CEE in he
Middle Yellow Ri e egion, sou hwes egion and o he a eas was
ela i ely highe , which can be a ibu ed o e ec i e local ca bon
mi iga ion policies. Con e sely, he CEE in he no hwes egion is he
lowes , which is p ima ily a ibu able o he egion’s unde de eloped
economy and a long-s anding ocus on he seconda y indus y. Fig. 2
p esen s a hea map illus a ing CEE ac oss Chinese p o inces, whe e
da ke colou s signi y highe e iciency.
Panel quan ile eg ession model
Compa ed wi h leas squa es eg ession, quan ile eg ession has
mo e lexible applica ion condi ions and can e ec i ely cap u e he ail
cha ac e is ics o a dis ibu ion. When independen a iables ha e
a ying e ec s a di e en posi ions wi hin he dependen a iable’s
dis ibu ion, quan ile eg ession p o ides a mo e comp ehensi e
desc ip ion o hese dis ibu ional cha ac e is ics. Consequen ly, quan-
ile es ima ion is mo e obus han leas squa es es ima ion (Zhu e al.,
2016). The panel quan ile eg ession model is cons uc ed as ollows:
yi =xʹ
i β+ai+ζi (9)
whe e Q(yi |xi ,ai)=xʹ
i β+ai(
τ
),
τ
is a condi ional quan ile, yi is he
dependen a iable, xi is he independen a iable, ai is indi idual ixed
e ec and ζi is he andom e o .
Since a iables’ no mali y is a undamen al assump ion o he
adi ional mean eg ession model, i is essen ial o con i m no mali y
be o e conduc ing panel quan ile eg ession. I he sample da a a e
no mally dis ibu ed, he adi ional mean eg ession model can be
employed o analyse he impac o echnological di e si ica ion on CEE.
Con e sely, i he da a a e no no mally dis ibu ed, he adi ional
mean eg ession model may yield biased es ima es, and he panel
quan ile eg ession model is mo e sui able. This s udy p esen s a
Quan ile-Quan ile (Q-Q) plo o assess he no mali y o he a iables
(Fig. 3). I a a iable ollows a no mal dis ibu ion, he sca e plo
should app oxima e a diagonal line. Based on his analysis, he Q-Q plo
indica es ha none o he a iables ully adhe e o a no mal dis ibu ion,
indica ing ha he panel quan ile eg ession model is he app op ia e
choice.
Empi ical esul s
Reg ession esul s
Benchma k es ima ion
We employ he panel quan ile eg ession model o es ima e he
impac o RD and UD on CEE. The selec ed quan iles a e 25 %, 50 % and
75 %, which co espond o low, medium and high CEE, espec i ely. We
se he boo s ap alue a 500, p esen ing he esul s in Table 4. Column
(1) displays he esul s o he quan ile eg ession o RD on CEE and
column (2) p esen s he esul s o he quan ile eg ession o UD on CEE.
O e all, he coe icien s o RD and UD a e signi ican ly posi i e a he 1
% le el, indica ing ha bo h ac o s p omo e CEE, which suppo s hy-
po heses H
1a
and H
1b
. The coe icien s o 25 %, 50 % and 75 % quan ile
le els g adually dec ease, indica ing ha he e ec o RD and UD on low,
medium and high CEE diminishes as he quan ile ises. Fu he mo e, he
coe icien o RD is gene ally la ge han ha o UD, indica ing ha RD
has a mo e subs an ial impac on CEE. This sugges s ha RD acili a es
echnological inno a ion a a lowe cos , esul ing in highe CEE, which
aligns wi h hypo hesis H
1b
. Rela ed g een echnological di e si ica ion
has a mo e p onounced e ec . This phenomenon occu s because ela ed
g een echnological di e si ica ion can mo e e ec i ely enhance sho -
e m ca bon e iciency, which is achie ed by le e aging he exis ing
echnology base, p omo ing knowledge spillo e s and educing inno-
a ion and o ganisa ional coo dina ion cos s. In con as , while un e-
la ed g een echnological di e si ica ion holds long- e m po en ial, i s
sho - e m impac on ca bon emissions is limi ed by high knowledge
in eg a ion cos s.
Robus ness es s
To con i m he obus ness o pa ame e es ima ion, we employ he
Supe -SBM model o measu e CEE and conduc he panel quan ile
eg ession once mo e. The eg ession esul s a e p esen ed in Table 5.
Bo h RD and UD signi ican ly enhance he CEE, wi h RD ha ing a la ge
impac , which is consis en wi h he benchma k eg ession esul s.
Endogenei y ea men
The ela ionship be ween RD (o UD) and CEE may exhibi e e se
causali y, which can lead o endogenei y conce ns in model se ing. This
s udy employs he wo-s age leas squa es (2SLS) me hod o add ess
endogenei y. Re e encing Li e al. (2021), his s udy conside s he
i s -o de lagged RD and UD as ins umen al a iables (IVs). Columns
(1) and (2) in Table 6 p esen he eg ession esul s o he 2SLS me hod
using he i s -o de lagged RD. The esul s show ha he e ec o he
i s -o de lagged RD on cu en -pe iod RD is posi i e, and ha
cu en -pe iod RD enhances CEE. Column (3) and (4) in Table 6 epo
he eg ession esul s o he 2SLS me hod based on he IV o i s -o de
lagged UD, which is also posi i e and cu en -pe iod UD accele a es
CEE. These esul s demons a e ha he benchma k eg ession is obus .
The Kleibe gen–Paap k Wald LM es con i ms ha he model does no
ha e any uniden i ied issues. Addi ionally, he C agg–Donald Wald F
es and he Kleibe gen–Paap k Wald F es indica e no weak IV
p oblem.
He e ogenei y analysis
In his sec ion, his s udy u he analyses he he e ogenei y o he
impac s o RD (o UD) on CEE. To do so, we ca ego ise he sample in o
h ee g oups based on egional cha ac e is ics o geog aphy, economy
and ma ke le el. These g oups include coas al and non-coas al p o -
inces, high-income and low-income p o inces and p o inces wi h high
and low ma ke le els. I he dummy a iable COST =1, i indica es ha
he p o ince is loca ed in a coas al egion, i he dummy a iable PGDP
=1, i signi ies ha he p o ince’s income is highe han he median
income o sample and i he dummy a iable MARKET =1, i sugges s
ha he p o ince’s ma ke capaci y exceeds he a e age income o he
Fig. 2. Hea map o ca bon emission e iciency.
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
7
sample. Fig. 4 p esen s hea maps ha depic p o incial income le els on
he le side and coas al a eas on he igh side.
Based on his analysis, o dina y leas squa es es ima ion esul s a e
p esen ed in Table 7, spanning six columns. All coe icien s o he
in e ac ion e ms o RD (o UD) and dummy a iables (COST, PGDP,
MARKET) a e signi ican ly nega i e a he 1 % signi icance le el. This
indica es ha RD (o UD) has a mo e p onounced impac on CEE in non-
coas al egions, low-income p o inces and p o inces wi h limi ed ma -
ke access, suppo ing hypo heses H
2a
–H
2c
, espec i ely. Coas al e-
gions, wi h ad anced echnological in as uc u e, ha e achie ed
signi ican g een echnological di e si ica ion; howe e , i s ma ginal
impac ends o diminish, con ibu ing modes ly o CEE imp o emen .
Fig. 3. No mali y g aph o a iables.
Table 4
Benchma k es ima ion esul s.
Va iables (1) (2)
Quan ile le els Quan ile le els
25 h 50 h 75 h 25 h 50 h 75 h
RD 4.01e −09*** 5.87e −09*** 6.09e −09***
(4.110) (4.6e +10) (146.620)
UD 5.05e −09*** 3.90e −09*** 3.47e −09***
(5.410) (3.6e +10) (48.500)
Con ol Va iables YES YES YES YES YES YES
Obs 19,295 19,295 19,295 19,295 19,295 19,295
No e: - alues a e in pa en heses. (2) ***, ** and * deno e 1 %,5 % and 10 % signi icance le els, espec i ely.
Table 5
Robus ness es s.
Va iables (1) (2)
Quan ile le els Quan ile le els
25 h 50 h 75 h 25 h 50 h 75 h
RD 0.00024*** 0.0000013 0.00004
(3638.180) (0.020) (1.320)
UD 0.00016*** 0.00000095 0.00002
(56.590) (0.010) (21.160)
Con ol Va iables YES YES YES YES YES YES
Obs 19,295 19,295 19,295 19,295 19,295 19,295
No e: - alues a e in pa en heses. (2) ***, ** and * deno e 1 %, 5 % and 10 % signi icance le els, espec i ely.
S. He e al.
Jou nal o Inno a ion & Knowledge 10 (2025) 100730
8
