DISCUSSION PAPER SERIES IZA DP No. 5494 Informaon and Communcaon Technologes and Skll Upgradng: The Role of Inernal vs. Exernal Labour Markes Luc Behaghel Eve Carol Emmanuelle Walkowak February 2011 Forschungsnsu zur Zukunf der Arbe Insue for he Sudy of Labor
Informaon and Communcaon Technologes and Skll Upgradng: The Role of Inernal vs. Exernal Labour Markes Luc Behaghel Pars School of Economcs, INRA and CREST Eve Carol Unversy of Pars Oues Nanerre, EconomX, Pars School of Economcs and IZA Emmanuelle Walkowak Pars XII Unversy, ERUDITE and CEE Dscusson Paper No. 5494 February 2011 IZA P.O. Box 7240 53072 Bonn Germany Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mal: za@za.org Any opnons expressed here are hose of he auhor(s) and no hose of IZA. Research publshed n hs seres may nclude vews on polcy, bu he nsue self akes no nsuonal polcy posons. The Insue for he Sudy of Labor (IZA) n Bonn s a local and vrual nernaonal research cener and a place of communcaon beween scence, polcs and busness. IZA s an ndependen nonprof organzaon suppored by Deusche Pos Foundaon. The cener s assocaed wh he Unversy of Bonn and offers a smulang research envronmen hrough s nernaonal nework, workshops and conferences, daa servce, projec suppor, research vss and docoral program. IZA engages n () orgnal and nernaonally compeve research n all felds of labor economcs, () developmen of polcy conceps, and () dssemnaon of research resuls and conceps o he neresed publc. IZA Dscusson Papers ofen represen prelmnary work and are crculaed o encourage dscusson. Caon of such a paper should accoun for s provsonal characer. A revsed verson may be avalable drecly from he auhor.
IZA Dscusson Paper No. 5494 February 2011 ABSTRACT Informaon and Communcaon Technologes and Skll Upgradng: The Role of Inernal vs. Exernal Labour Markes * Followng he adopon of nformaon and communcaon echnologes (ICT), frms are lkely o face ncreasng skll requremens. They may reac eher by ranng or hrng he new sklls, or by a combnaon of boh. We frs show ha ICT are ndeed skll based and we hen assess he relave mporance of exernal and nernal labour marke sraeges. We show ha skll upgradng followng ICT adopon akes place mosly hrough nernal labour markes adjusmens. The nroducon of ICT s assocaed wh an upward shf n frms occupaonal srucure, of whch one hrd s due o hrng and frng workers from and o he exernal labour marke, whereas wo-hrds are due o promoons. Moreover, we fnd no compellng evdence of exernal labour marke sraeges based on excess urnover. In conras, French frms heavly rely on ranng n order o upgrade he skll level of her workforce, even f hs vares across ndusres. JEL Classfcaon: J23, J24, J41 Keywords: echncal change, labour urnover, skll bas, ranng, nernal labour markes Correspondng auhor: Eve Carol PSE 48 Boulevard Jourdan 75014 Pars France E-mal: Eve.Carol@ens.fr * We are graeful o Andrea Bassann, Marc Gurgand, Kevn Lang, Thomas Lemeux, Thomas Pkey, Murel Roger and wo anonymous referees for very helpful commens and suggesons on an earler verson of he paper. We also hank parcans o he EEA-2007 and AFSE-2007 conferences, o he ERMES-2007 workshop on Organsaon de l enreprse, connassance e nnovaon and o semnars a Harvard Unversy, Pars School of Economcs and Laboraore d Econome d Orleans (LEO) for useful remarks. We are graeful o Julee Camnade for ousandng research asssance. Ths work was suppored by he French Agence Naonale de la Recherche [gran number ANR-08-JCJC-0067-01 o LB and EC].
Inroducon Changes n employmen relaonshs have araced a lo of aenon n recen years, boh n he popular press and n academc research. Followng several downszng epsodes n he USA and n Europe, a wdely shared vew has developed accordng o whch employmen relaonshs have become more nsable han hey used o do be. Accordng o Doernger and Pore (1971) nernal labour markes are characersed by he fac ha wages and career pahs are, o a large exen, deermned by a se of admnsrave rules raher han by pressure arsng from he exernal labour marke. In conras, n recen years, long-erm employeremployee relaonshs have seemed o declne (Cappell, 1999) and he labour marke seems o have been ncreasngly workng lke a spo marke (Aknson, 1999). Correspondngly, he percepon of job nsecury has ncreased n mos OECD counres n he 1990s (OECD, 2003). However, exsng emprcal evdence on rsng job nsably and he declne of nernal labour markes s acually que mxed. Regardng job nsably, he US leraure does no provde compellng evdence of any decrease n reenon raes nor n job enure (see Neumark e al, 1999, Jacoby, 1999, and Sevens, 2008). Regardng France, he revew of he leraure by Germe (2001) suggess ha he evdence of a conracon of nernal labour markes s no clear-cu. However, job nsecury seems o have ncreased n he USA over he pas decades, wheher measured n erms of nvolunary job loss (Farber, 2007) or n erms of probably of dsmssal (Vallea, 2000). Smlar resuls are found on French daa. Gvord and Maurn (2004) and Behaghel (2003) ndeed fnd evdence of an upward rend n annual ranson raes from employmen o unemploymen beween he md-1970s and he early 2000s. So, evdence on he poenal declne of nernal labour markes s far from beng clear-cu. One reason for hs may be ha nernal labour markes are effcen organsaons n he presence of mach-specfc nvesmens, ransacon coss and workers' rsk averson (see Wacher e al, 1990) whch are unlkely o have dsappeared alogeher. Anoher reason may have o do wh he rapd developmen of nformaon and communcaon echnologes (ICT) over he pas 20 years. Alhough hese changes have ofen been seen as poenal causes of a declne n nernal labour markes, hs acually depends on how hey mpac human resource managemen, n parcular hrough new and ncreasng skll requremens. The recen leraure on he skll conen of echnologcal change suggess ha nformaon and communcaon echnologes subsue workers n roune asks, whereas hey complemen hem n neracve and analycal acves (see Auor e al., 2003, and Spz-Oener, 2006). 3/36
The mpac of such changes on he skll level of he workforce may however vary across jobs or secors. Ben-Ner and Urasun (2010) sugges ha ICT could generae some "skll bfurcaon": would evenually rase he skll level of workers n complex jobs whereas would have a deskllng effec on workers n low-sklled jobs. Noneheless, avalable evdence suggess ha, on average, ICT adopon generaes an ncrease n he demand for communcaon and analycal sklls (see Auor e al, 2003). Frms may cope wh n varous ways. They may eher hre he new sklls on he exernal labour marke or ran her own workers, hus relyng on he workng of he nernal labour marke. Hrng he new sklls from ousde he frm may acually ake wo forms. Frms may hre more hghly sklled workers and ge rd of less sklled ones, whch generaes an upward shf n he occupaonal srucure. Alernavely, hey may hre new workers whn a consan occupaonal srucure n order o brng "fresh blood", and presumably new sklls, no her workforce. A number of recen papers have nvesgaed he mpac of new echnologes on changes n he occupaonal srucure whn frms, worker flows or ranng. They fnd a posve mpac of echncal (and organsaonal) change on he employmen or wage-bll share of he more hghly sklled occupaons (see Chennells and Van Reenen, 2002 and Carol and Van Reenen, 2001). On U.S. daa, Neumark and Reed (2004) dsplay a sgnfcanly greaer use of conngen employmen relaonshs n new-economy jobs. Regardng worker flows, Gvord and Maurn (2004) fnd ha he use of new echnologes ncreases he annual ranson rae from employmen o unemploymen, and ha hs s enough o explan he global rend oward greaer job nsecury observed n France. Lookng a worker flows by skll levels n France, Askenazy and Moreno-Galbs (2007) fnd ha frms ha mos nensely use new echnologes and nnovave work pracces experence a hgher urnover among mos of he occupaonal caegores. For Germany, Bauer and Bender (2003) fnd ha new echnologes ncrease churnng raes for sklled and hghly sklled workers and ha mos of he employmen adjusmen paerns assocaed wh echnologcal change are due o hrng and frng raher han o promoons or demoons. These resuls sugges ha, f anyhng, frms mee he new skll requremens followng he nroducon of ICT hrough adjusmens on he exernal raher han nernal labour marke. However, anoher srand of leraure suggess ha he adopon of nformaon echnologes and nnovave workplace pracces also rases frms' nvesmen n ranng. On US daa, Lynch and Black (1998) fnd ha he proporon of workers recevng formal ranng s hgher n frms ha use hgh performance work pracces. Behaghel and Greenan (2010) use 4/36
mached employer-employee daa for France and also fnd ha a more nnovave organsaon ncreases he probably ha workers receve ranng, even once conrollng for selecon and poenal endogeney bases. As underlned by hs revew of he leraure, mos arcles consder only one ype of labour marke response o echnologcal change. They eher sudy nernal or exernal labour marke sraeges, bu rarely boh a he same me. However, as suggesed by Mncer (1989), hese sraeges are lkely o be correlaed wh each oher. Frms may reac o ncreasng skll requremens eher by combnng n some way ranng wh he hrng of new sklls or hey may, on he conrary, rely on one sraegy a he expense of he oher. The characerscs of hs choce and s deermnans are he focus of he presen paper. Invesgang hs ssue on France a he end of he 1990s s of parcular neres because ICT were sll spreadng que quckly across frms, whle n oher counres, lke he USA, a majory of frms had already ransed o ICT-nensve producon processes. To he exen ha we are neresed n how frms cope wh he new skll requremens arsng from he mplemenaon of ICT, France n he lae 90s provdes an neresng case o research. Moreover, ogeher wh Japan, France used o be one of he Wesern counres n whch nernal labour markes were he mos wdespread (Marsden, 1999). I s herefore one of he counres n whch he developmen of ICT s lkely o have had he mos devasang effec on long-erm employmen relaons f new echnologes call for an ongong reshufflng of he workforce. A he same me, nernal labour markes may have been proeced by he srengh of employmen proecon legslaon. Accordng o he OECD ndcaor (Venn, 2009), France s one of he counres wh he hghes level of employmen proecon. Ths s parcularly he case for emporary conracs whch can be used only under very src condons and for a lmed amoun of me. Ths may have made nernal labour markes more susanable n a perod of rapd echncal change. In order o assess he reslence of nernal labour markes n a conex of echnologcal changes, we consder hree ways hrough whch frms may ncrease he skll level of her workforce, whenever necessary: 1) They may shf her occupaonal srucure upward, whch can be done eher by promong ncumben workers from lower o hgher sklled occupaons (nernal labour marke sraegy) or alernavely, by hrng workers n more hghly sklled occupaons and/or frng hem n less sklled ones (exernal labour marke sraegy). 5/36
2) Exernal labour marke adjusmens may also ake he form of "excess urnover" (or churnng).e. urnover n excess o wha s necessary o upgrade he occupaonal srucure f frms ry o acqure new sklls by he adjuncon of "fresh" workers. 3) Evenually, frms may ran her own workers, hus relyng on he nernal labour marke. Accordng o he relave mporance of hese channels, we wll be able o assess how mporan nernal labour marke sraeges reman n a perod of rapd expanson of new ICT or, alernavely, o wha exen exernal labour marke sraeges have aken over. 1 The paper s organsed as follows. Secon 1 oulnes he economerc model. Secon 2 nroduces he daa. The resuls are presened n Secon 3. Some dscusson and concludng remarks are offered n Secon 4. 1. The Economerc model New echnologes and skll upgradng The exsng leraure on skll-based echncal change suggess ha n order o fully explo he poenal of ICT, frms have o upgrade he skll level of her workforce when adopng hem. The correspondng predcon s ha here should be a posve correlaon (all oher hngs kep equal) beween he adopon of more advanced echnologes and an ncrease n he skll level of he workforce. A very smple es s based on he followng regresson: SKILL = z β + ICT δ + ε (1) where ICT s an ndcaor of adopon of new nformaon and communcaon echnologes, SKILL s a measure of he workforce s sklls, and z are conrol varables. ICT adopon s lkely o be endogenous. In he absence of any good nsrumen for echnologcal adopon, we nerpre posve esmaes δˆ as evdence of paral correlaons beween he adopon of new echnologes and upward changes n he skll srucure of he workforce see Secon 4 for furher dscusson of endogeney ssues. 1 Noe ha hese hree channels may overlap; for nsance, frms may promoe workers hey ran. However, as shown n he nex secon, shfs n he occupaonal srucure can be decomposed n shfs hrough nernal movemens and shfs hrough exernal movemens, hus provdng a naural merc o assess he relave mporance of nernal and exernal movemens. 6/36
Sklls, however, can be acqured hrough a varey of channels. In wha follows, we consder hree possble channels: frms may upgrade he occupaonal srucure of her workforce (whch can be acheved hough enres and exs or, alernavely hrough promoons); hey may rely on excess urnover n order o acqure fresh sklls; and/or, hey may ran her own workers. We nerpre upward changes n he occupaonal srucure hrough enres and exs as well as excess urnover as ndcaors of exernal labour marke adjusmens. Conversely, skll upgradng hrough promoons and ranng are seen as ndcaors of nernal labour marke sraeges. Decomposng changes n he occupaonal srucure Changes n frms' occupaonal srucure n relaon wh ICT adopon are usually esmaed usng sandard labour share equaons: S = x β + ICT δ + ε (2) p p where x s a vecor of conrol varables and S s he share of occupaonal group p n he workforce of frm : S = L L L L 1 1 Such changes are he oucome of wo dfferen movemens: () enres and exs of workers a varous levels of he occupaonal srucure and () promoons of workers from lower o hgher occupaons. In order o dsngush boh effecs, we consruc counerfacual changes n labour shares ( S ~ ) descrbng wha would have happened o he occupaonal srucure f here had only been enres and exs a he dfferen occupaonal levels, bu no nernal movemen (promoon or demoon): L + H E L (3) S ~ 1 1 = L 1 + H E L 1 where L 1 s he number of workers n occupaon p n frm a me -1, H s he number of enres n occupaon p n frm beween me -1 and and E s he number of workers formerly employed n occupaon p leavng frm beween me -1 and. Smlarly, L, H, and E respecvely denoe he oal number of workers, enres and exs n frm a me. 7/36
Gven ha we do no have any drec nformaon on promoons, changes n he occupaonal srucure hrough promoons only ( Ŝ ) are defned as he changes n he occupaonal srucure ha would have occurred f here had been none of he enres or exs ha we observe n he daa: L H + E L Ŝ = (4) H + E L L 1 1 where L s he number of workers n occupaon p n frm a me. The number of workers n occupaon p a s compued as a counerfacual ncludng only promoons,.e. he number of workers observed a n occupaon p mnus enres plus exs no ha group beween -1 and. The correspondng labour share s compued by dvdng hs number of workers by wha employmen would have been n he frm a year f no enry nor ex had aken place over he perod. In he case where he oal level of employmen s consan 2, he observed changes are he sum of he wo counerfacual changes. Indeed, we hen have S L = L L 1 1 = L H + E L 1 L 1 H + H = E and L 1 E = Sˆ L L 1 =. Therefore, ~ + S Le us underlne ha our counerfacual measures of changes n he occupaonal srucure ( Ŝ and S ~ ) closely relae o sandard worker flow raes (see Davs and Halwanger, 1999 for a survey). The man dfference beween boh s ha n he former, flows are defned wh respec o he oal sze of he workforce, whereas n he laer, hey are defned wh respec o he sze of occupaon p. Consder equaon (3). When frm sze s consan, ~ H E S =. L Ths dffers from he sandard exernal worker flow rae EWFR: 1. 2 When oal employmen s no consan, he coeffcens we oban for Ŝ and obaned when usng S ~ do no add up o hose S as a dependen varable. Ths s due o he fac ha a complee decomposon of S should nclude a hrd erm, namely he neracon beween changes due o nernal and exernal movemens. We choose o neglec hs erm whch has no clear nerpreaon wh respec o he nernal vs exernal labour marke ssue. I urns ou o be very small n pracce: he coeffcens on Ŝ S ~ almos and add up n our regressons. Robusness checks showng ha alernave addve decomposons yeld almos dencal resuls are avalable upon reques. 8/36
EWFR = H E L 1 by a facor equal o L / 1 L 1. In oher words, he key dfference les n he fac ha we rewegh he usual worker flow raes by he nal occupaonal srucure of he frm, L / 1 L 1. Ths serves our purpose by provdng a drec measure of he mpac of worker flows on he frm s occupaonal srucure. We esmae equaon (2) separaely for each occupaon. 3, S ~ and Ŝ by OLS, equaon by equaon, for Upgradng sklls hrough excess urnover and ranng Anoher way o upgrade he skll srucure of a frm when adopng ICT s hrough he addon of "fresh" workers by means of labour urnover. Turnover s, o some exen, a mechancal consequence of he upgradng of he occupaonal srucure hrough enres/exs: here canno be any upgradng of he occupaonal srucure hrough enres/exs f here are no worker flows. However, as s well known from he leraure on job and worker flows (for French frms, see Abowd, Corbel and Kramarz, 2003), worker flows (urnover) usually largely exceed wha s needed for a gven level of job flows. We herefore use a measure of excess urnover or churnng -,.e. urnover n excess o wha s needed for a gven change n he sze of a group of workers. Specfcally, followng Davs and Halwanger (1999, secon 6) excess urnover e n plan and for group p s defned as: e H + E H E (5) L L We hen esmae: e = x ξ + ICT γ + v (6) p p by OLS separaely for each occupaon, where x s a vecor of conrol varables. ˆγ p s he esmae of neres; posve values ndcae ha ICT adopon s assocaed wh an ncreased 3 Alhough we are esmang a sysem of Seemngly Unrelaed Regressons (SUR), feasble Generalzed Leas Squares (FGLS) are no warraned here, as he same regressors show up n each equaon (see, e.g., Theorem 7.6 n Wooldrdge, 2002). 9/36
urnover of group p, beyond wha s mechancally mpled by he upgradng of occupaonal groups. Smlarly, skll upgradng hrough ranng needs o be analysed whn each occupaonal group. Indeed, ranng raes are hgher n hgh-skll groups, hence upgradng of he occupaonal srucure hrough enres/exs mechancally generaes an ncrease n ranng raes. Our daa allows us o esmae ranng equaons for a gven occupaonal group: T = xψ + ICTη + υ (7) p p whch we esmae by OLS. x s a vecor of conrol varables and ˆη p s he esmae of neres; posve values ndcae ha ICT adopon s assocaed wh an ncrease n ranng, once conrolled for composon effecs. 2. The Daa Measurng echnology adopon and skll upgradng hrough our 3 channels a he frm level requres combnng several daabases. Our man nformaon on ICT comes from he REPONSE survey (RElaons PrOfessonnelles e NégocaonS d Enreprse). In 1998, 2978 esablshmens were surveyed wh senor managers beng asked quesons abou ndusral relaons, mplemenaon of new echnologes and reorgansaons. Regardng ICT, we have nformaon on he proporon of workers usng he Inrane 4 and he Inerne n 1998 (no use, less han 5%, 5 o 19%, 20 o 49%, 50% and more). Gven ha hese echnologes were a he very begnnng of her lfe cycle n France n he md-1990s, we assume ha he proporon of workers usng hem n 1998 provdes a good approxmaon of echnologcal adopon over 1996-1998. More specfcally, we defne a dummy varable equal o 1 f a leas 5% of he workers use he Inerne or a leas 20% of he workers use he Inrane 5. One advanage of hs smple measure of ITC use s ha s relevan for he dfferen secors n he economy: as shown by 4 The exac phrasng s "Inrane and compuer neworks". 5 Usng a bnary varable dscards par of he nformaon avalable n he 5x5 levels of adopon ha are obaned by combnng he orgnal Inerne and Inrane varables. However, our sample s oo small for esmang he effecs of all hese combnaons separaely. We choose o focus on he 5% and 20% hresholds as a way o conras reasonably balanced groups of ITC early adopers (1/3 of he sample) and lae (or non) adopng frms (2/3 of he sample), hus gvng us good sascal power. Compung a more connuous ndex may lead o msspecfcaons as we would have o make lneary assumpons (wh lle sascal power o es hem). We checked ha our resuls are robus o he hresholds we use o generae he ICT adopon ndcaor. 10/36
able A, n our sample, very smlar proporons of frms n he manufacurng and servce secors are classfed as ICT-nensve by hs measure (abou one hrd). In addon o ICT, he REPONSE survey also provdes dealed nformaon on frms and esablshmens whch we use as conrol varables n he regressons: frm characerscs (publc/prvae, frm wh one or several plans, lsed on sock markes/non-lsed), plan characerscs (share of women, share of par-me workers, presence of unon delegaes, employmen growh, local labour marke densy), as well as a se of ndusry and plan sze dummes. In order o capure worker flows, we rely on wo dfferen sources. The DMMO (Données sur les Mouvemens de Man-d Oeuvre) has exhausve daa on enres and exs of workers n and ou of esablshmens wh 50 employees or more. The daa s broken down no four occupaonal caegores: managers and professonals 6, echncans and supervsors, clerks and blue-collars. The EMMO (Enquêe sur les Mouvemens de Man-d Oeuvre) has dencal nformaon on a represenave sample of frms wh less han 50 employees. We use hs daa o compue counerfacual changes n labour shares over 1996-1998,.e. changes ha are due only o enres and exs (resp. promoons) n he varous occupaons over he perod. In order o do so see equaons (3) and (4) we also use nformaon on he level of employmen n each occupaonal cell a he begnnng and a he end of he perod. Ths nformaon s provded by he French survey of employmen srucure: he ESE (Enquêe Srucure des Emplos), as of December 31 s 1995 and 1998. The las channel we consder for skll upgradng s ranng. The so-called 24-83 fscal records provde frm-level daa on he number of workers recevng ranng and he volume of ranng hours 7. Ths nformaon s obvously que lmed snce only refers o connuous and formal ranng. In parcular, apprencesh s excluded whereas may be an mporan skll source (see Bellman and Jank, 2007 and Ryan e al, 2007) whch may be, n some cases, complemenary o nernal labour markes (see Soskce, 1994, on Germany). Smlarly, nformal on-he-job ranng s no ncluded n our daa alhough may represen an mporan par of ranng, n parcular n small frms (see Barron e al, 1997) 8. Ths s a clear lmaon of he 24-83 records. Unforunaely, we are no aware of any daa source ha would provde more exhausve nformaon on ranng n France. Usng he nformaon avalable n 6 Ths caegory also ncludes engneers. 7 The "24-83" records provde frm raher han plan-level daa on ranng. Machng hem wh esablshmenlevel daa generaes some measuremen error ha s lkely o rase he sandard errors n our esmaes. 8 In order o preven hs from basng our resuls, we sysemacally conrol for plan sze n our regressons. 11/36
he 24-83, we compue boh he proporon of workers recevng some ranng and he average number of ranng hours per worker for four occupaonal caegores - dencal o hose n he DMMO-EMMO daabase. These are averaged over 1996-1998 n order o accoun for he fac ha ranng may ake some me o be mplemened afer frms decde o nves n. Machng he four daases and cleanng ou esablshmens wh mplausble values for skll upgradng reduces our sample o 1,114 esablshmens. The low machng rae s prmarly due o he fac ha he EMMO and 24-83 sources are no exhausve (respecvely, no sysemacally coded) see he Daa Appendx for deals. Table A n he Appendx summarzes all he varables used n our models. Our sample consss manly of large plans (53% have more han 200 workers) belongng o mulesablshmen frms of he prvae secor. 77% have a unon delegae and less han half of hem are lsed. The manufacurng secor s over-represened n our sample: accouns for 80% of oal employmen, compared o only 20% n he whole French economy. As a consequence, women accoun for only 35% of he labour force. 37% of he plans employ more han 5% of par-me workers. 3. Resuls 3.1 ICT adopon and skll upgradng sraeges We frs nvesgae he correlaon beween ICT adopon and he sraeges used by frms o upgrade he sklls of her workforce. In order o do so, we esmae equaons (2)-(4), (6) and (7) of our model. Table 1 presens he resuls for he varous forms of skll upgradng: respecvely, upward shf n he occupaonal srucure hrough enres/exs versus promoons, excess urnover and ranng. Panel A of Table 1 provdes evdence of skll-based echncal change. The use of he Inerne or he Inrane s posvely correlaed wh an upward shf n he occupaonal srucure and, more specfcally wh an ncrease n he proporon of managers, engneers and professonals and a decrease n he proporon of clerks n he workforce. Ths occupaonal upgradng s essenally acheved hrough nernal movemens (promoons). These accoun for more han 70% of he ncrease n he proporon of managers as compared o only 30% 12/36
for exernal movemens - and for almos all of he reducon n he share of clerks 9 see Panels B and C. Ths frs se of resuls suggess ha nernal labour markes sll play an mporan role when frms have o cope wh ncreasng skll requremens. Mos of he adjusmen n he occupaonal srucure akes place hrough promoons, whereas he relave mporance of enres and exs o and from he exernal labour marke remans lmed. To gauge he economc sgnfcance of he effecs, s useful o compare hem o he average changes n he occupaonal srucure n our sample durng he 1996-1998 perod (Table A). For nsance, he.80 percenage pon ncrease n he share of managers and professonals occurrng hrough nernal movemens and assocaed wh nensve ICT-use (Table 1) s very close o he overall ncrease n he share of managers and professonals n he economy (+.77 percenage pons). If we were o gve a causal nerpreaon o he esmaes, a back-of-heenvelop calculaon would mply ha abou 35% of he overall ncrease n he demand for managers and professonals s due o he adopon of ICT ha occurred n abou 1/3 of he frms over he perod and whch was sasfed hrough nernal movemens (0.33*.80/0.77=0.34); abou 15% s due o he adopon of ICT bu was sasfed by he exernal labour marke (.33*.3/0.77=.13); and he remanng 50% s due o oher causes. Of course, our esmaes canno necessarly be nerpreed causally; bu hs suggess ha he role of nernal labour markes s far from neglgble n he overall upgradng of he occupaonal srucure. Frms may also ry o upgrade he skll level of her workforce by brngng n "fresh workers" wh new sklls, beyond wha would be necessary o upgrade he occupaonal srucure hrough enres and exs. Panel D of Table 1 nvesgaes he paral correlaons beween ICT adopon and excess urnover. We fnd no compellng evdence of such a correlaon: he use of he Inerne or he Inrane s no assocaed wh excess urnover, whaever he caegory of workers we consder. Overall, he exernal labour marke does no appear as a key provder of new sklls when frms nroduce new nformaon and communcaon echnologes. In conras, frms heavly rely on ranng n order o upgrade he skll level of her workforce. Panels E and F of Table 1 have he paral correlaons beween ICT adopon on he one hand and wo dfferen measures of ranng on he oher hand. As evdenced by Panel E, he nroducon of new echnologes s correlaed wh a greaer access o ranng for all caegores of workers excep for managers and professonals - where he coeffcen s posve 9 Noe ha, gven ha our occupaonal caegores are que broad, we do no ake no accoun all promoons. Those occurrng from one level o he nex one whn a gven occupaonal caegory are no recorded, so ha our resuls on he scope of promoon should be seen as a lower bound. 13/36
bu no sascally sgnfcan. In conras, when consderng he number of hours of ranng per worker, he correlaon s posve and sgnfcan for all occupaonal groups, ncludng managers. Agan, s useful o compare he effecs o he average level of ranng n he sample dsplayed n Table A. Dependng on he occupaon, ICT adopon s assocaed wh an ncrease n he ncdence of ranng by 5 o 15%, and an ncrease n ranng hours by 10 o 15%. Ths frs se of resuls suggess ha he nernal labour marke sll plays a key role n he adjusmen of he skll level of he workforce n frms ha nensely use new nformaon and communcaon echnologes. Ths s confrmed f usng alernave ndcaors of nernal and exernal labour marke sraeges. If esmang he correlaon beween ICT adopon and enry and ex raes n he varous occupaons (see Table 2 Panel 2.1), we do no fnd any sgnfcan assocaon excep for he hrng of echncans and supervsors and, o a lower exen, for managers and professonals. Ths s conssen wh he reduced role of exernal labour marke sraeges whch play a role only hrough upgradng frms' occupaonal srucure a he hgher end see Table 1 panel C. In conras, when esmang he correlaon beween ICT adopon and promoon raes o manageral posons, we do fnd a sgnfcan assocaon, n parcular f consderng promoons from he hree lowes sklled groups (bluecollars, clerks, echncans and supervsors) o managers and professonals see Table 2 Panel 2.2. Promoons from echncans and supervsors o managers are also more frequen n ICT nensve frms. Ths s also he case for promoons from he wo lowes sklled groups (blue-collars and clerks) o he hghes wo (echncans and supervsors; managers and professonals) alhough he correlaon s no sgnfcan a convenonal levels. In order o check he robusness of our fndngs, we ry several alernave specfcaons 10. We frs show ha he mporance of promoons as compared o exernal movemens s no sensve o he decomposon mehod ha we use. We also show ha our resuls are robus o conrollng for employmen growh a he plan level, o usng alernave measures of ICT (dfferen hresholds for ICT adopon or, alernavely, he share of ICT nvesmen n valueadded) and o conrollng for nvesmen n physcal capal. We are also concerned ha our fndngs could be drven by poenally confoundng facors such as he use of nnovave workplace pracces (proxed by qualy crcles, smulaon of workers' parcaon) or o he qualy of ndusral relaons (proxed by he number of srkes). We show ha hs s no he case and ha our resuls reman unchanged f conrollng for such facors. 10 Dealed resuls are avalable from he auhors upon reques. 14/36
The mporan role of he nernal labour marke n upgradng he skll level of he workforce herefore seems o be que reslen n France, even n he conex of developmen of new ICT. One can wonder however wheher hs form of human resource managemen s o be found n all frms or wheher here s some heerogeney n frms' sraeges and along whch dmensons. 3.2 Heerogeney n frms' skll upgradng sraeges A frs poenal dmenson of heerogeney n our daa has o do wh ndusry. Skll upgradng pracces are lkely o be dfferen across secors, f anyhng because he needs and he relave cos of each sraegy are lkely o be dfferen. Gven he fac ha our sample overrepresens he manufacurng secor, our resuls may no be represenave of he average rend n he French economy. A dsaggregaed analyss s herefore needed. When splng our sample across manufacurng and servces, our resuls sugges ha here are ndeed some dfferences, n parcular wh respec o he use of ranng. Table 3.1 suggess ha ICT-nensve frms n he manufacurng secor heavly rely on promoons n order o upgrade he skll level of her workforce. Excess urnover s used o a lmed exen for clerks and blue-collars. Bu he domnan sraegy s based on ranng wh ICT-nensve frms provdng sgnfcanly more ranng han ohers o all caegores of workers. Regardng servces (see Table 3.2), he same paern s observed for changes n he occupaonal srucure wh promoons accounng for mos of he skll upgradng. In conras, frms n he servce secors appear o rely raher lle on ranng when usng ICT, he only sgnfcan correlaon beng for blue-collars. Ths heerogeney n skll upgradng sraeges across secors s, o a large exen, robus o akng no accoun oher poenal sources of heerogeney (see Table 4). The skll upgradng sraeges chosen by frms are lkely o vary across a varey of dmensons. In parcular, one would expec nernal labour markes o be larger and beer organsed n large frms han n smaller ones. In conras, he exernal labour marke s poenally more aracve n hghdensy local labour markes because s lkely o offer a greaer varey of sklls. Evenually, skll upgradng sraeges may be dfferen beween hgh and low-ech secors 11 because he use of he Inerne/Inrane s lkely o be dfferen n boh ypes of ndusres. 11 Secors are classfed as beng low or hgh-ech on he bass of he laes OECD 4-dg classfcaon for manufacurng secors (see Hazchronoglou, 1997) and on he bass of he 4-dg classfcaon proposed by he 15/36
Our resuls sugges ha here s ndeed some heerogeney across several of hese dmensons. The drec effecs of he sze and local labour marke densy varables sugges ha hese affec o a ceran exen frm's skll upgradng sraeges. Large frms essenally rely on promoons o upgrade he occupaonal srucure of her workforce (n favour of echncans and supervsors and, o a lesser exen a he expense of clerks) and hey rely less on excess urnover for boh caegores. More mporanly, ranng s srongly correlaed wh sze wh large frms provdng greaer access and more ranng hours o all caegores of workers. Ineresngly, beng locaed on a hgh-densy local labour marke s correlaed wh a lower urnover for echncans and clerks, whch s no n lne wh he dea ha hgh densy on he local labour marke should make exernal skll upgradng sraeges more aracve o frms. Evenually, belongng o a hgh-ech secor does no seem o be srongly assocaed wh frms' skll sraeges. The only excepon s ha he hgh-ech varable s negavely correlaed wh excess urnover for blue-collars and posvely correlaed wh he number of ranng hours for clerks (a he 10% level). Moreover, he correlaon beween ICT adopon and skll upgradng sraeges s somemes que dfferen across frm sze, ype of local labour marke and echnologcal level of he secor. Ths s he message conveyed by he coeffcens we ge on he neracon erms beween Inerne/Inrane and our conrol varables. The densy of he local labour marke ends o ncrease he relance of ICT-nensve frms upon promoon, n parcular for managers and blue-collars (a he 10% sgnfcance level), conrary o wha could be expeced. 12 However, also ncreases frms' relance on excess urnover for echncans and clerks when adopng ICT, whch s conssen wh he dea ha denser labour markes offer a wder choce of sklls and are hus lkely o beer f frms' needs. In conras, skll upgradng sraeges assocaed wh he use of ICT do no seem o be very heerogeneous across frm sze, excep for promoons whch end o be relavely more mporan for clerks and less so for echncans (a he 10% level). Smlarly, beng n hgh raher han low-ech secors does no massvely affec he skll upgradng sraeges of ICT-nensve frms. One excepon, however, s ha he laer provde more ranng hours (a he 10% level) o echncans and blue-collars n hgh-ech secors. European Foundaon for he Improvemen of Lvng and Workng Condons for servce secors (see European Monorng Cener on Change, 2006). 12 A enave explanaon could be ha, f n dense local labour markes, frms are able o hre workers ha mach her needs more closely; hese successful maches are hen more lkely o lead o promoons. 16/36
The dfferences beween manufacurng and servces are essenally robus o he nroducon of hese addonal conrols. The ncrease n he share of managers and echncans s smaller n ICT-nensve servce frms han n ICT-nensve manufacurng frms, bu here s no dfference n he respecve roles of promoons versus exernal movemens n achevng hese changes. One mporan dfference whch was no srkng when splng he sample across secors s ha labour urnover s much hgher n servces han n manufacurng for all caegores of workers. The specfcy of servces wh respec o ranng s confrmed: ICTnensve frms rely less on ranng for all caegores of workers, wh he coeffcens beng sgnfcan for managers and, o a lesser exen, for blue-collars (when consderng ranng hours). One reason for hs may be he hgh level of urnover. If he ype of ranng requred when frms adop ICT s more cosly han usual, may well be he case ha servce frms are more relucan o ran her workers gven her hgher probably o leave han n he manufacurng secor. The causaon may also go he oher way round n some servces. Wherever jobs are low-qualy and low-pad, ranng s scarce. In such cases, may well be poor job qualy ha leads o hgh labour urnover, raher han he oher way round. Overall, our resuls hghlgh some heerogeney n he skll upgradng sraeges of frms ha have adoped ICT. Noneheless, he message conveyed by our resuls s ha he relance on nernal labour markes remans que wdespread n France even n frms ha have adoped new nformaon and communcaon echnologes. 4. Dscusson and concludng remarks In hs paper, we have aken a fresh look a he fae of nernal labour markes n a perod of rapd echnologcal change. Our resuls sugges ha hey vvdly ress n France even n frms ha have nroduced nformaon and communcaon echnologes. As already evdenced by many papers n he leraure, ICT adopon s assocaed wh an upward shf n frm's occupaonal srucure. Bu we show ha n our sample, hs s largely acheved hrough promoons raher han enres and exs from and o he exernal labour marke. Moreover, when nroducng new echnologes, frms massvely rely on ranng n order o upgrade he skll level of her workforce, whereas he use of excess urnover as a provder of new sklls remans very lmed. Ths ressance of nernal labour markes even where ICT are nensely used s wdespread across frms, whaever her sze or echnologcal 17/36
level. I s slghly sronger n low-densy labour markes and n manufacurng where ranng s more frequenly used han n servces o ncrease he skll level of workers. Le us underlne ha, as all he leraure on skll-based echncal change, our resuls only capure he shor-erm relaonsh beween ICT adopon and skll upgradng sraeges. If here are lags n he way ICT affec frms' skll srucure, our resuls wll capure a mos wha happens n he frs wo years followng he shock (gven ha we are workng on he 1996-1998 perod). We also need o nss upon he fac ha our esmaes do no prove a causal relaonsh: alhough we nroduce a rch se of conrols o rule ou some of he mos lkely sources of spurous correlaons, he ICT varable remans poenally endogenous and we are no able o exhb plausble nsrumens o solve ha problem. Noneheless, our resuls sugges ha he nensve use of ICT does no preclude human resource managemen sraeges based on nernal labour markes: nernal labour markes and ICT can coexs. Moreover, our fndngs make a large negave mpac of ICT on nernal labour markes raher unlkely. Indeed, reconclng a large negave causal mpac wh our fndng of a robus, posve correlaon would requre ha some unobserved varables creae a large upward bas n our esmaes. Le us consder wo ypes of mechansms ha could generae such a bas, and examne wheher hey are conssen wh he evdence we provde. The frs mechansm s one of spurous correlaons due o unobserved shocks a he frm level. A frs possbly would be ha frms ha are h by unfavourable shocks (say, a fallng demand for her produc) have o cu coss. To ha end, hey adop cos savng echnologes; hey also downsze, and he burden of he adjusmen s dsproporonaely born by he unsklled. Ths sequence of evens would generae a spurous correlaon beween ICT and upgradng of he occupaonal srucure hrough enres and exs. Therefore, wrongly gnorng he poenal role of unobserved shocks would make us overesmae he exernal labour marke response o ICT adopon. Ths raher renforces our resul on he predomnance of nernal adjusmens. However, unobserved shocks may also work he oher way round. Frms ha are h by a posve shock may nroduce new echnologes (because hey have more cash avalable for nvesmen). They may also mechancally ncrease her ranng nvesmen: The French regulaon on he fnancng of connuous ranng s such ha over he 1996-1998 perod, all frms had o spend a leas 1.5% of her wage bll on ranng. For he frms for whch hs consran s bndng, any posve shock ha rases he wage bll wll mechancally generae an ncrease n ranng expendure. In hs case, he posve correlaon we fnd 18/36
beween new echnologes and ranng may be spurous. In order o check wheher hs s he case, we re-ran our ranng regressons on he sub-sample of frms spendng more han 2% of her wage bll on ranng. For hese frms, he legal mnmum s no bndng so here s no reason ha an ncrease n her wage bll should lead hem o nves more n ranng. When dong hs, our resuls are vrually unchanged 13. Of course, one can magne oher sources of unobserved heerogeney ha would drve he resuls spurously. Bloom e al. (2008) sugges ha he orgn of capal could be one of hem. They show ha US mulnaonals operang n Europe are more IT-nensve han non-us mulnaonals (and ge a hgher producvy from IT) and ha her IT advanage s prmarly due o more effcen human resource managemen pracces. In order o make sure ha our correlaon beween IT and nernal labour marke sraeges s no drven by US-owned frms, we conrolled for he orgn of capal (US versus non-us) n our baselne regresson. The resuls are lef unchanged. Of course, hs does no rule ou oher possble sources of unobserved heerogeney. Ths s why we reman cauous n nerpreng our resuls. The second mechansm s relaed o a polcy change ha ook place n France n he md 1990s. Sarng n 1993, successve governmens have nroduced fscal measures o reduce he labour cos of low-wage workers. These payroll ax cus have had he larges mpac on frms employng a hgh proporon of low-skll workers. As a resul, hese frms were less nduced o adop new echnologes (as long as hey are subsues o unsklled labour) whle hey had ncenves o hre even more low-skll workers. Agan, hs would lead us o overesmae he mpac of ICT adopon on exernal adjusmens: he share of low-skll workers would ncrease hrough enres and exs n frms less prone o adop ICT. The mpac of he payroll axes on nernal adjusmens s unclear. Low-wage frms mgh refran from promong and ranng her workers, by fear of ncreasng wages above he level elgble for he ax exempon. Ths, however, does no concern hgh-wage workers whose ranng appears o be hghly correlaed wh new echnologes n our daa. Moreover, he addonal conngens of low-wage workers pad a he mnmum wage would probably no ge much wage ncrease followng ranng, gven ha her producvy would, mos probably, reman below he mnmum wage. Overall, s unclear wheher he payroll ax cus should have nduced lower ranng and fewer promoons n frms less lkely o adop ICT. Therefore, s 13 The coeffcens (sandard errors) of he Inerne/Inrane varable n he regresson for he proporon of ranees are: 4.15 (2.18) for managers and professonals, 8.61 (3.83) for echncans and supervsors, 4.87 (2.39) for clerks and 3.72 (2.97) and for blue-collars. The coeffcens (sandard errors) of he Inerne/Inrane varable n he regresson for he nensy of ranng are: 2.38 (1.13) for managers and professonals, 3.09 (1.11) for echncans and supervsors, 1.77 (0.95) for clerks and 2.45 (0.92) for blue-collars. 19/36
unclear why our fndng of a posve correlaon beween ICT and nernal labour marke sraeges should be drven by hs polcy change. Overall, we vew our resuls as provdng evdence ha nernal labour markes have remaned a vable human resource managemen sraegy for French frms even when adopng new nformaon and communcaon echnologes. A queson rased by hese resuls s o wha exen hey may be generalsed. France s ndeed characersed by src employmen proecon legslaon (Venn, 2009) and hs may be he very reason why nernal labour markes have survved: frms would rely on nernal adjusmens because gong on he exernal labour marke would smply be oo cosly. In ha case, he reslence of nernal labour markes n a conex of rapd ICT expanson would be largely dosyncrac or, a leas, specfc o hgh-epl counres. The evdence we provde does no que go n ha drecon. We ndeed fnd srong dfferences n nernal labour marke adjusmens across manufacurng and servces wh more ranng beng used n he former and more labour urnover n he laer - whereas boh secors face he same employmen proecon legslaon. Smlarly, he relance on exernal adjusmens vares accordng o he densy of he local labour marke whereas EPL s equally src all over he French errory. As a consequence, EPL canno enrely accoun for he reslence of nernal labour markes n France. Moeover, recen evdence by Bassann e al. (2007) suggess ha he amoun of ranng provded by frms would acually be negavely correlaed o employmen proecon legslaon. In order o ge a beer undersandng of hese relaons, a drec comparson of France wh oher counres would be poenally very fruful. Ths appears o be a promsng avenue for fuure research. 20/36
References Abowd J., P. Corbel, F. Kramarz (1999), «The Enry and Ex of Workers and he Growh of Employmen,» Revew of Economcs and Sascs 81, 2, 170-187. Askenazy P. and Moreno-Galbs E., 2007, "The Impac of Technologcal and Organzaonal Changes on Labor Flows. Evdence on French Esablshmens", Labour, 21(2), pp. 265-301. Aknson T., 1999, "Is Rsng Inequaly Inevable? A Crque of he Transalanc Consensus", WIDER Annual Lecures 3. Auor D., Levy F. and Murnane R., 2003, "The skll conen of recen echnologcal change: an emprcal nvesgaon", The Quarerly Journal of Economcs, 118(4), pp. 1279-1333. Barron J., M. Berger and D. Black, 1997, On-he-Job Tranng, W.E. Upjohn Insue for Employmen Research, Kalamazoo, Mch. Bassann A., Booh A., Brunello G. De Paola M. and Leuwen E., 2007, "Workplace Tranng n Europe", n Brunello G., Garbald P. and Wasmer E. eds, Educaon and Tranng n Europe, New-York: Oxford Unversy Press, pp. 143-342. Bauer T. and Bender S., 2004, "Technologcal Change, Organzaonal Change and Job Turnover", Labour Economcs, 11(3), pp. 265-91. Behaghel L., 2003 «Insécuré de l emplo : le rôle proeceur de l ancenneé a--l bassé en France?», Econome e sasque, n 366, pp. 1-27. Behaghel L. and Greenan N., 2010, "Tranng and Age-Based Techncal Change: Evdence from French Mcro Daa", Annales d économe e de sasque, forhcomng. Bellmann L. and F. Jank, 2007, "To recru sklled workers or o ran one s own? Vocaonal ranng n he face of uncerany as o he rae of reenon of ranees on compleon of ranng", Zeschrf für ArbesmarkForschung, 2-3, pp. 205-220. Ben Ner A. and A. Urasun, 2010, "Compuerzaon and Skll Bfurcaon: The Role of Task Complexy n Creang Skll Gans and Losses", mmeo Carlson School of Manangemen, Unversy of Mnnesoa. Bloom N., Sadun R. and Van Reenen J., 2008, "Amercans do IT Beer: US Mulnaonals and he Producvy Mracle, Revsed verson of NBER Workng Paper 13085, hp://www.sanford.edu/~nbloom/adib.pdf Cappell P., 1999, "Career Jobs are Dead", Calforna Managemen Revew, 42(1), pp. 146-67. Carol E. and Van Reenen J, 2001, ''Skll based organzaonal change? Evdence from a panel of Brsh and French esablshmens'', The Quarerly Journal of Economcs, 116(4), pp. 1449-92. Chennells L. and Van Reenen J., 2002, "Has Technology hur he less sklled workers? An economerc survey of he effecs of echncal change on he srucure of pay and jobs", n L'Hory Y., Greenan N. and Maresse J. eds, Producvy, Inequaly and he Dgal Economy, MIT Press, pp. 175-225. Davs, S.J., Halwanger, J., 1999. Gross job flows. In: Ashenfeler, O., Card, D. (Eds.), Handbook of Labor Economcs, vol. 3b. Norh-Holland, Amserdam, pp. 2711 2805. Doernger P. and Pore M., 1971, Inernal Labour Markes and Manpower Analyss, Lexngon MA, Heah. 21/36
European Monorng Cenre on Change, 2006, "Trends and drvers of change n he European knowledge-nensve busness servces secor: Mappng repor", European Foundaon for he Improvemen of Lvng and Workng Condons. Farber H., 2007, "Job Loss and he Declne n Job Secury n he Uned Saes", Workng Paper #520, Prnceon Unversy, Indusral Relaons Secon. Germe J.F., 2001, "Au-delà des marchés nernes : quelles moblés, quelles rajecores?", Formaon Emplo, n 76, pp. 129-45. Gvord, P. and Maurn E., 2004, Changes n Job Insecury and her Causes: An Emprcal Analyss for France, 1982-2002, European Economc Revew, 48, pp. 595-615. Hazchronoglou, T. (1997), "Revson of he Hgh-Technology Secor and Produc Classfcaon", OECD Scence, Technology and Indusry Workng Papers, 1997/2, OECD Publshng Jacoby S., 1999, "Are Career Jobs Headed for Exncon?", Managemen Revew, 42(1), pp. 123-45. Lynch L. and Black S., 1998, "Beyond he Incdence of Employer Provded Tranng", Indusral and Labor Relaons Revew, 52(1), pp. 6481. Marsden D., 1999, A Theory of Employmen Sysems: Mcro-foundaons of socal dversy, Oxford Unversy Press. Mncer J., 1989, "Human Capal Responses o Technologcal Change n he Labour Marke", NBER Workng Paper, n 3207. Neumark, D., Polsky, D. and Hansen D., 1999, Has Job Sably declned ye? New Evdence for he 1990s, Journal of Labour Economcs, 17(4), pp. 29-64. Neumark D. and Reed D., 2004, "Employmen Relaonshs n he New Economy", Labour Economcs, 11(1), pp. 1-31. OECD, 2003, "More and Beer Jobs? Aggregae Performance durng he Pas Decade", Chaper 1, Employmen Oulook, OECD, Pars, pp. 17-66. P. Ryan, H. Gospel and P. Lews, 2007, "Large Employers and Apprencesh Tranng n Bran", Brsh Journal of Indusral Relaons, 45(1), pp. 127 153. Soskce D, 1994, "Reconclng Markes and Insuons: The German Apprencesh Sysem", n L. Lynch ed, Tranng and he Prvae Secor. Inernaonal Comparsons, NBER Comparave Labor Marke Seres, Boson, Mass, pp. 25-60. Spz-Oener A., 2006, "Techncal Change, Job Tasks and Rsng Educaonal Demands: Lookng ousde he Wage Srucure, Journal of Labour Economcs, 24(2), pp. 235-70. Sevens A.H., 2008, "No So Fas: Long-Term Employmen n he U.S. 1969-2004." In Lad Off, Lad Low: The Socal and Polcal Impac of Job Insably, Kaherne S. Newman, Edor. New York: Columba Unversy Press. Vallea R., 2000, "Declnng Job Secury", n Neumark D. ed: On he Job: s Long-Term Employmen a Thng of he Pas?, New-York : Russell Sage Foundaon, pp. 227-56. Venn D., 2009, "Legslaon, collecve barganng and enforcemen: updang he OECD employmen proecon ndcaors", OECD-ELSA Workng Paper. Wacher M. and R. Wrgh, 1990, "The Economcs of Inernal Labor Markes", Indusral Relaons, 29(2), pp.240-62. 22/36
Wooldrdge J., 2002, Economerc Analyss of Cross Secon and Panel Daa, MIT Press, Cambrge. 23/36
Table 1 ICT Adopon and Skll Upgradng Channels Managers and professonals Techncans and supervsors Clerks Blue collars A. Overall changes n he occupaonal srucure Inerne/Inrane 1.13*** -0.12-0.65** -0.36 (0.22) (0.30) (0.30) (0.34) Obs 1 114 1 114 1 114 1 114 B. Inernal movemens Inerne/Inrane 0.80*** 0.12-0.58* -0.34 (0.20) (0.31) (0.30) (0.32) Obs 1 114 1 114 1 114 1 114 C. Exernal movemens Inerne/Inrane 0.30* -0.26-0.04 0.00 (0.17) (0.22) (0.20) (0.25) Obs 1 114 1 114 1 114 1 114 D. Excess urnover Inerne/Inrane 1.18 3.15 9.52-7.70 (1.98) (2.56) (6.89) (13.82) Obs 1 090 1 094 1 104 1 046 E. Number of ranees (per 100 workers) Inerne/Inrane 3.40 7.43** 5.50*** 3.81* (2.06) (3.14) (1.93) (2.31) Obs 1 097 1 052 1 087 909 F. Tranng hours per worker Inerne/Inrane 3.04*** 3.31*** 2.06*** 2.24*** (0.98) (0.93) (0.76) (0.72) Obs 1 095 1 042 1 083 907 Noes: (1) Robus sandard errors n parenheses (2) *** p<0.01, ** p<0.05, * p<0.1 (3) A separae OLS regresson s run n each panel and for each occupaonal caegory. Each regresson ncludes conrols for plan sze (dummy varable for plans wh more han 200 employees), beng locaed on a labour marke whose densy s hgher han he medan, mul-esablshmen frms, publc
secor companes, lsed companes, presence of unon delegaes, share of women n he labour force, par-me work (dummy varable for frms wh more han 5% of workers beng par-me) and 15 secors. 25
Table 2 ICT adopon and alernave measures of nernal vs exernal labour marke sraeges Panel 2.1 ICT Adopon, Ex and Hrng Raes Managers and professonals Techncans and supervsors Clerks Blue collars A. Ex rae Inerne/Inrane 0.00 0.02 0.05-0.03 (0.01) (0.01) (0.03) (0.07) Obs 1 090 1 094 1 104 1 046 B. Hrng rae Inerne/Inrane 0.02 0.03* 0.06-0.03 (0.02) (0.02) (0.04) (0.07) Obs 1 090 1 094 1 104 1 046 Panel 2.2 ICT Adopon and promoon raes Promoon rae from blue-collars, clerks or echncans o managers and professonals Promoon rae from blue-collars or clerks o managers and professonals Promoon rae from echncans o managers and professonals Inerne/Inrane 0.02*** 0.05 0.11* (0.01) (0.03) (0.07) Obs 1 114 1 114 1 094 Noes: (1) Robus sandard errors n parenheses (2) *** p<0.01, ** p<0.05, * p<0.1 (3) A separae OLS regresson s run n each panel and for each occupaonal caegory or ype of promoon. Each regresson ncludes conrols for plan sze (dummy varable for plans wh more han 200 employees), beng locaed on a labour marke whose densy s hgher han he medan, mul-esablshmen frms, publc secor companes, lsed companes, presence of unon delegaes, share of women n he labour force, par-me work (dummy varable for frms wh more han 5% of workers beng par-me) and 15 secors.
Table 3.1 ICT Adopon and Skll Upgradng Channels n Manufacurng Secors Managers and professonals Techncans and supervsors Clerks Blue collars A. Overall changes n he occupaonal srucure Inerne/Inrane 1.05*** -0.42-0.20-0.43 (0.26) (0.35) (0.26) (0.39) Obs 641 641 641 641 B. Inernal movemens Inerne/Inrane 0.80*** -0.16-0.11-0.53 (0.23) (0.35) (0.25) (0.36) Obs 641 641 641 641 C. Exernal movemens Inerne/Inrane 0.24-0.15-0.13 0.03 (0.19) (0.23) (0.17) (0.29) Obs 641 641 641 641 D. Excess urnover Inerne/Inrane -1.07 4.04 10.31** 8.99* (1.33) (3.16) (4.27) (5.20) Obs 635 634 634 634 E. Number of ranees (per 100 workers) Inerne/Inrane 8.68*** 13.04*** 8.03*** 4.10** (2.45) (4.48) (2.59) (1.94) Obs 633 625 624 629 F. Tranng hours per worker Inerne/Inrane 4.67*** 4.40*** 3.69*** 1.85** (1.27) (1.12) (1.07) (0.82) Obs 631 620 622 629 Noes: (1) Robus sandard errors n parenheses (2) *** p<0.01, ** p<0.05, * p<0.1 (3) A separae OLS regresson s run n each panel and for each occupaonal caegory. Each regresson ncludes conrols for plan sze (dummy varable for plans wh more han 200 employees), beng locaed on a labour marke whose densy s hgher han he medan, mul-esablshmen frms, publc secor companes, lsed companes, presence of unon delegaes, share of women n he labour force, par-me work (dummy varable for frms wh more han 5% of workers beng par-me) and 7 manufacurng sub secors. 27
Table 3.2 ICT Adopon and Skll Upgradng Channels n Servce Secors Managers and professonals Techncans and supervsors Clerks Blue collars A. Overall changes n he occupaonal srucure Inerne/Inrane 1.07*** 0.40-1.37** -0.10 (0.38) (0.52) (0.65) (0.63) Obs 473 473 473 473 B. Inernal movemens Inerne/Inrane 0.70* 0.53-1.25* 0.03 (0.38) (0.58) (0.67) (0.61) Obs 473 473 473 473 C. Exernal movemens Inerne/Inrane 0.30-0.34 0.04-0.01 (0.30) (0.43) (0.43) (0.44) Obs 473 473 473 473 D. Excess urnover Inerne/Inrane 3.68 2.38 9.05-32.38 (4.91) (4.52) (15.73) (33.71) Obs 455 460 470 412 E. Number of ranees (per 100 workers) Inerne/Inrane -3.65-0.20 2.51 4.34 (3.59) (4.16) (3.03) (6.38) Obs 464 427 463 280 F. Tranng hours per worker Inerne/Inrane 0.88 1.50-0.27 2.93** (1.51) (1.63) (1.04) (1.42) Obs 464 422 461 278 Noes: (1) Robus sandard errors n parenheses (2) *** p<0.01, ** p<0.05, * p<0.1 (3) A separae OLS regresson s run n each panel and for each occupaonal caegory. Each regresson ncludes conrols for plan sze (dummy varable for plans wh more han 200 employees), beng locaed on a labour marke whose densy s hgher han he medan, mulesablshmen frms, publc secor companes, lsed companes, presence of unon delegaes, share of women n he labour force, par-me work (dummy varable for frms wh more han 5% of workers beng par-me) and 8 servce sub secors. 28
Table 4 ICT Adopon and Skll Upgradng Channels: The role of sze, labour marke densy and secor. Managers and professonals Techncans and supervsors Clerks Blue collars A. Overall changes n he occupaonal srucure Inerne/Inrane 0.45 1.01-0.26-1.20* (0.51) (0.69) (0.64) (0.70) Large -0.12 1.03*** -0.29-0.62 (0.20) (0.30) (0.33) (0.40) Inerne/Inrane x Large -0.09-0.89 0.99-0.01 (0.49) (0.64) (0.63) (0.66) Hgh Densy 0.25-0.03 0.11-0.33 (0.26) (0.40) (0.41) (0.47) Inerne/Inrane x Hgh Densy 0.65-0.58-1.15 1.08 (0.51) (0.68) (0.70) (0.71) Servces -0.46* -0.77** 0.32 0.91 (0.27) (0.39) (0.51) (0.57) Inerne/Inrane x Servces 0.15 0.27-0.88 0.46 (0.51) (0.63) (0.69) (0.70) Hghech -0.20 0.18 0.71** -0.69 (0.33) (0.38) (0.35) (0.47) Inerne/Inrane x Hghech 0.77-1.22* -0.72 1.17* (0.54) (0.63) (0.61) (0.67) Obs 1 069 1 069 1 069 1 069 B. Inernal movemens Inerne/Inrane 0.28 1.47** -0.72-1.02 (0.46) (0.72) (0.61) (0.70) Large -0.34 1.10*** -0.61* -0.16 (0.21) (0.32) (0.35) (0.37) Inerne/Inrane x Large 0.13-1.14* 1.14* -0.13 (0.45) (0.67) (0.66) (0.65) Hgh Densy 0.20 0.36 0.09-0.65 (0.20) (0.41) (0.42) (0.47) Inerne/Inrane x Hgh Densy 0.89* -1.02-1.06 1.19* (0.47) (0.74) (0.74) (0.71) Servces -0.27-0.46-0.09 0.82 (0.29) (0.42) (0.50) (0.53) Inerne/Inrane x Servces -0.23 0.27-0.67 0.64 (0.44) (0.67) (0.69) (0.67) Hghech -0.21 0.42 0.22-0.43 (0.31) (0.40) (0.34) (0.42) Inerne/Inrane x Hghech 0.32-1.12* -0.12 0.92 (0.46) (0.65) (0.61) (0.63) Obs 1 069 1 069 1 069 1 069 ( )
Table 4 follow (1) Managers and professonals Techncans and supervsors Clerks Blue collars C. Exernal movemens Inerne/Inrane 0.03-0.32 0.62-0.33 (0.38) (0.53) (0.43) (0.55) Large 0.16-0.09 0.41* -0.48 (0.16) (0.23) (0.23) (0.30) Inerne/Inrane x Large -0.04 0.11-0.34 0.27 (0.35) (0.47) (0.41) (0.50) Hgh Densy 0.06-0.24 0.30-0.12 (0.23) (0.29) (0.29) (0.39) Inerne/Inrane x Hgh Densy -0.16 0.23-0.25 0.18 (0.42) (0.52) (0.45) (0.53) Servces -0.10-0.20 0.12 0.18 (0.22) (0.32) (0.31) (0.37) Inerne/Inrane x Servces 0.31-0.13-0.04-0.14 (0.42) (0.50) (0.49) (0.55) Hghech 0.04-0.25 0.40-0.19 (0.25) (0.27) (0.29) (0.40) Inerne/Inrane x Hghech 0.44-0.01-0.66 0.23 (0.40) (0.45) (0.43) (0.53) Obs 1 069 1 069 1 069 1 069 D. Excess urnover Inerne/Inrane -1.84 1.38-6.50 42.14* (4.11) (9.41) (14.16) (24.06) Large -0.10-12.14* -22.73** 20.44 (4.28) (7.04) (8.97) (35.89) Inerne/Inrane x Large -1.19 2.69 20.22-49.99 (4.51) (7.45) (15.98) (38.67) Hgh Densy -3.47-10.77** -16.44* -29.46 (4.32) (5.05) (9.64) (22.23) Inerne/Inrane x Hgh Densy 3.72 12.13** 43.37** 32.26 (5.53) (5.53) (21.76) (22.83) Servces 14.25** 19.60*** 63.60*** 49.40* (5.70) (5.45) (9.84) (26.43) Inerne/Inrane x Servces 3.55-8.65-18.48-55.77* (4.92) (7.11) (14.73) (30.87) Hghech 1.18 1.10 0.61-17.53** (2.34) (2.57) (5.72) (7.95) Inerne/Inrane x Hghech 1.96-2.74-2.81-2.98 (3.19) (5.17) (13.32) (16.68) Obs 1 045 1 049 1 059 1 005 ( ) 30
Table 4 follow (2) Managers and professonals Techncans and supervsors Clerks Blue collars E. Number of ranees (per 100 workers) Inerne/Inrane 12.13** 8.03 9.26* 3.76 (5.04) (5.53) (4.90) (4.50) Large 12.36*** 15.48*** 8.53*** 9.04* (2.78) (3.55) (2.59) (4.62) Inerne/Inrane x Large -5.27-4.76-2.50-2.54 (4.36) (5.30) (4.32) (5.51) Hgh Densy -1.75-1.45-1.93-0.17 (3.22) (3.91) (2.81) (3.76) Inerne/Inrane x Hgh Densy 2.39 3.98 0.98-0.65 (4.56) (5.50) (4.41) (4.94) Servces 4.52 2.80-2.14 8.18 (3.62) (4.40) (3.51) (5.82) Inerne/Inrane x Servces -16.02*** -7.73-4.81-3.00 (4.94) (5.91) (5.15) (5.81) Hghech -5.26-5.42 1.95 3.59 (3.51) (4.70) (3.49) (2.58) Inerne/Inrane x Hghech 1.26 2.69-1.97 2.39 (4.27) (5.20) (4.96) (3.99) Obs 1 052 1 009 1 043 871 F. Tranng hours per worker Inerne/Inrane 6.09** 2.91 3.79** -1.19 (2.47) (2.03) (1.74) (1.57) Large 6.73*** 5.40*** 3.59*** 2.14*** (1.33) (1.25) (0.91) (0.74) Inerne/Inrane x Large -1.91-0.20-1.52 2.07 (2.13) (2.06) (1.72) (1.47) Hgh Densy 0.32 0.28 0.82 0.18 (1.37) (1.34) (1.17) (0.85) Inerne/Inrane x Hgh Densy -1.40-0.32-1.87 0.33 (2.12) (2.18) (1.67) (1.84) Servces 1.03 2.51* 1.45 0.06 (1.49) (1.44) (1.18) (1.19) Inerne/Inrane x Servces -4.44* -2.01-2.88* 2.35 (2.34) (2.14) (1.73) (1.90) Hghech 1.32 1.94 2.34* -0.56 (1.62) (1.33) (1.24) (1.08) Inerne/Inrane x Hghech 0.63 3.63* 1.54 2.80* (2.14) (1.95) (1.85) (1.62) Obs 1 051 1 001 1 040 869 Noes: (1) Robus sandard errors n parenheses. *** p<0.01, ** p<0.05, * p<0.1 (2) A separae OLS regresson s run n each panel and for each occupaonal caegory. Each regresson ncludes conrols for plan sze (dummy varable for plans wh more han 200 employees), mul-esablshmen frms, publc secor companes, lsed companes, presence of unon delegaes, share of women n he labour force, par-me work (dummy varable for frms wh more han 5% of workers beng par-me) and 15 secors. (3) Hgh Densy ndcaes ha local labour marke densy belongs o he hghes quarle. 31
(3) The coeffcen for "Servces" s compued as he dfference beween he coeffcens for he egh 2-dg servce sub secors and he seven 2-dg manufacurng sub secors, weghed by he shares of each sub secor. (4) The "Hghech" ndcaor uses he OECD defnon, appled o 712 4-dg sub secors. 32
Appendx Table A Descrve Sascs All plans Inerne/Inrane=0 Inerne/Inrane=1 Manufacurng Servces Change of labor share (n %) Managers and professonals 0.77 0.33 1.65 0.93 0,54 Techncans and supervsors 0.56 0.63 0.44 0.72 0,36 Clerks -0.38-0.15-0.84-0.27-0,53 Blue collars -0.95-0.81-1.24-1.38-0,37 Change of labor share hrough nernal movemens (n %) Managers and professonals 0.92 0.60 1.55 0.90 0,93 Techncans and supervsors 0.79 0.73 0.92 0.96 0,56 Clerks -0.64-0.38-1.16-0.41-0,95 Blue collars -1.07-0.95-1.31-1.46-0,55 Change of labor share hrough enres and exs (n %) Managers and professonals -0.07-0.18 0.15 0.07-0,26 Techncans and supervsors -0.13 0.00-0.41-0.17-0,09 Clerks 0.18 0.13 0.29 0.14 0,25 Blue collars 0.02 0.04-0.02-0.04 0,10 Excess urnover (n %) Managers and professonals 24.12 25.04 22.29 18.54 31,89 Techncans and supervsors 25.13 27.38 20.67 13.95 40,55 Clerks 61.10 61.07 61.16 30.13 102,88 Blue collars 49.93 53.38 42.79 27.21 84,89 Number of ranees per 100 workers Managers and professonals 59.60 56.71 65.38 60.58 58,27 Techncans and supervsors 59.59 55.47 67.94 60.94 57,62 Clerks 41.27 37.88 48.06 43.82 37,83 Blue collars 34.18 32.04 39.14 34.17 34,22 Hours of ranng per worker Managers and professonals 21.45 19.54 25.24 22.70 19,75 Techncans and supervsors 19.45 17.55 23.30 19.97 18,69 Clerks 12.07 10.87 14.47 13.08 10,70 Blue collars 9.49 8.27 12.31 10.31 7,63 Inerne/Inrane 0,33 0.00 1.00 0.34 0.32 Indcaor for plan wh more han 200 workers 0,53 0.48 0.62 0.62 0.40 Indcaor for dense local labor marke 0,25 0.21 0.34 0.20 0.32 Indcaor for servce secor 0,42 0.43 0.41 0.00 1.00 Indcaor for mul-esablshmen frm 0,60 0.57 0.66 0.63 0.56 Indcaor for publc secor 0,03 0.03 0.05 0.02 0.05 Indcaor for lsed company 0,43 0.38 0.54 0.55 0.27 Indcaor for presence of unon delegaes 0,77 0.76 0.81 0.84 0.69 Share of women (%) 35,13 35.91 33.54 25.98 47.53 Indcaor for par-me work (>5% of workforce) 0,37 0.37 0.38 0.22 0.58 Number of observaons 1114 747 367 641 473 33
Daa appendx In hs appendx, we deal he key seps aken n preparng he daa. The daa sources are descrbed n he daa secon n he ex. In addon, he daa on local labour marke densy (he number of employed and unemployed workers per square klomeres, compued over 348 local labour markes) comes from he 1990 populaon census. Mergng he dfferen daases We sar from a sample of 2,975 plans from he REPONSE (1998) survey, wh nformaon on ICT adopon as well as key frm and plan characerscs. Machng he RESPONSE sample wh he DMMO-EMMO, ESE and 24-83 sources yelds a sample of 1,537 plans. The relavely low machng rae (53%) s due, n parcular, o he fac ha he EMMO s no an exhausve daa source (plans are sampled a a rae ha depends on her sze) and ha he 24-83 fscal forms are no sysemacally coded. In order o acheve hs machng rae, we exrapolae some of he mssng daa, usng he followng procedures: - when a plan s only presen for 1 (resp. 2) of he 3 years n he 24-83 daabase, we compue he ranng varables as averages over 1 (resp. 2) years nsead of 3; - when a plan s mssng for some quarers n a gven year n he EMMO-DMMO daa, we exrapolae he enres and exs n each occupaon from he enres and exs observed durng he res of he year; f a plan s mssng durng one (or wo) of he hree years, we exrapolae enres and exs from he oher years. Though hese exrapolaons nroduce measuremen error n he movemen and ranng varables, hs does no bas he esmaes as hese are dependen varables n he regressons ( mgh, however, make hem less precse). Moreover, we checked ha resrcng he sample o hose plans ha have complee DMMO-EMMO nformaon does no sgnfcanly aler he resuls. Our resuls are ndeed a lle less precse when usng exrapolaed nformaon. The pon esmaes are generally a b smaller, bu no sgnfcanly. In order o ge nformaon on nvesmen raes and on ICT expendure, we mach our man daase wh he EAE. Gven ha he EAE s no exhausve, we end up wh 874 frms n hs subsample. So, we only use o conduc robusness checks. Oulers and conssency checks I appeared ha for some of he plans, he ranng nformaon had been msrepored (possbly nvered) for sklled and unsklled blue collars n 1997 and 1998, leadng o mplausble ranng raes. As we lacked nformaon o correc he errors, we decded o aggregae he wo caegores n a sngle blue-collar occupaon. In addon o hs, a few observaons had oulyng values for he number of ranees for some years. We correced hese observaons when possble, or se hem o mssng. 34
We hen performed several conssency checks on hs 1537 plan sample. Frs, we checked ha he oal plan sze declared n he ESE does no dffer oo much from he one declared n he DMMO-EMMO sources, a he begnnng and a he end of our perod (Dec 31, 1995 and Dec 31, 1998). We drop all plans for whch he dfference s more han 20% or represens a leas 10 workers. We esed he robusness of our resuls o oher hresholds; he man fndngs are no affeced. Second, we checked for oulers n he changes n he occupaonal srucure. We dropped plans for whch he sum of he absolue changes n he share of he dfferen occupaons s more han 60% and represens more han 10 workers. The resuls are no sensve o usng a srcer hreshold. Wh hese flers, he benchmark sample has 1,114 plans. Droppng any knd of fler and usng he 1537 observaons does no sgnfcanly mpac our resuls on ranng, bu does affec hose concernng nernal and exernal movemens. However, once a mnmum level of conssency across daa sources s requred, he ghness of he flers we use does no maer. Las, we also consdered a sample where we resrced upward nernal mobly for managers and professonals and downward nernal mobly for unsklled blue collars. The dea s ha managers canno be downgraded and ha no one can be promoed o an unsklled blue collar poson. We frs consder a src lm, bu as appears ha he medan of nernal promoon for unsklled workers was 0 (hough obvously no symmercally dsrbued), we evenually sele o a less src lm (+/-2) allowng for measuremen errors (n he frs case he sample conans abou 550 observaons, and abou 950 n he second case). When usng he src rule, resuls do shf a b, bu he sample sze s also dvded by 2. However, when usng he less src rule of +/-2, resuls are n lne wh our benchmark. Sample selecon Our man sample beng consruced by machng 5 dfferen sources, sample selecvy may be an ssue. Frs, a few observaons may be los due o he machng procedure. Frms are mached on a naonal frm denfer (SIREN), managed by he French sascal nsue (INSEE). A few denfer changes may occur over he 3-year perod, n parcular due o mergers; s however reasonable o pu hese frms asde, as he merger may dsrup he lnk beween HR managemen pracces and echnology adopon. More mporanly, he man reason for he relavely low machng rae (53%) s ha 2 of he 5 man daases (EMMO and 24-83) are no exhausve. In hese wo daases, however, we have random samplng of he oal populaon, wh samplng probables varyng by frm sze. Sample selecvy bas s herefore no an ssue, even hough he analyss sample s no represenave of he nal populaon anymore. Insead of usng samplng weghs n he analyss, we prefer o check he robusness of our esmaes by srafyng he sample (or addng neracon erms) n wo key dmensons: frms sze and frms ndusry (servces vs. manufacurng). 35
A second concern s ha, by consrucon, only frms survvng over he 1996-98 perods can be used n he analyss. We canno rule ou sample selecvy ssues here. As we do no observe ICT adopon a he begnnng of he perod, s no possble o analyse he selecve survval of frms, or o mplemen sample selecvy correcons. The dscusson of he possble bas s somewha speculave. However, we beleve ha, f anyhng, our key fndng should be renforced. Wha we are parcularly mssng are he exernal movemens ha would appear f some frms were shrnkng nsead of closng. A lkely hypohess s ha frms ha do no adop ICT are less lkely o survve, so ha we are dsproporonaely mssng exernal movemens among non-adopers. We herefore probably overesmae he correlaon beween exernal movemens and ICT adopon. Consequenly, seems ha correcng for sample selecvy would renforce our fndng ha nernal labour markes play a larger role han he exernal labour marke n a conex of ICT adopon. 36