Wage growth and bargaining in the minimum wage era

Transcription

1 Wage growh and barganng n he mnmum wage era Kerry L. Papps * Lle s known abou he long-erm effecs of mnmum wages on workers. As demonsraed n hs paper, a mnmum wage should have a negave effec on a worker s subsequen wage growh when wages are deermned by ndvdual barganng, bu no effec when wages are deermned by collecve barganng. These predcons are esed usng longudnal daa from he Uned Kngdom. Workers who were affeced by he mnmum wage before age 22 are found o have sgnfcanly lower wage growh laer n lfe han ohers, bu only f hey worked a esablshmens ha dd no negoae wh unons over ranng. Ths dfference n wage growh appears o be drven by dfferences n he levels of ranng provded. 1. Inroducon Alhough has been exensvely debaed by economss, a growng consensus has formed ha mnmum wage laws have lle effec on employmen levels, even hough hey rase hourly wage raes (Belman and Wolfson 2014). Ths may mply ha employers are able o pass he coss ono consumers n he form of hgher prces (Allegreo and Rech 2018), absorb he coss and make lower profs, a leas n he shor run (Bell and Machn 2018), or ha workers respond by rasng producvy and reducng urnover (Papps 2012). Alernavely, frms mgh aemp o recoup he coss assocaed wh he mnmum wage by economsng on aspecs of he oal employmen package oher han he basc hourly wage. One way of dong hs s o reduce he amoun of on-he-job ranng offered o workers. In a semnal paper, Becker (1962) esablshed ha when he labour marke s compeve, frms wll only pay for ranng f s specfc o he frm. Ths mples ha frms should cu back on general ranng whenever hey are compelled o rase wages due o he mnmum wage (Rosen 1972). More recen work has shown ha he suaon s more complex when here s no perfec compeon n he labour marke (Acemoglu and Pschke * Deparmen of Economcs, Unversy of Bah. Ths research s based on a repor commssoned by he Low Pay Commsson. The auhor would lke o hank Paul Gregg, Rebecca Rley, Alan Mannng, Tm Bucher, Bob Ello, Seve Machn, Felx Rche for her helpful suggesons, as well as parcpans a he Low Pay Commsson Research Symposum, Sepember 2014, Royal Economc Socey conference, Aprl 2015, and Work, Pensons and Labour Economcs Sudy Group conference, July 2015, and a presenaons a he Unverses of Brsol, Cyprus and Ken and Unversy College London. In addon, he s graeful o he saff of he UK Daa Servce for her asssance n accessng and analysng he ASHE daa. 1

2 1999) and ha frms mgh acually rase general ranng for some workers when he mnmum wage ncreases (Acemoglu and Pschke 2003). Prevous emprcal research has no conclusvely esablshed wheher he mnmum wage affecs eher he prevalence of on-he-job ranng or he rae of wage growh laer n a worker s career (whch should change f ranng levels are adjused). Some auhors have repored a negave relaonshp, whle ohers have found nsgnfcan resuls. Acemoglu and Pschke s work pons o he mporance of he level of compeveness n he labour marke, bu prevous sudes of he relaonshp beween mnmum wages and long-run wage growh have gnored he role of anoher mporan labour marke nsuon: rade unons. The recen resurgence of populary of mnmum wages n Wesern counres has poenally exacerbaed he declne n unons roles n seng wages. However, unons bargan wh frms over more han jus pay levels. Specfcally, f hey negoae over he provson of on-he-job ranng, hey may be able o nsulae young workers from any harmful effecs of he mnmum wage. The naure of he relaonshp beween he mnmum wage and ranng has mporan mplcaons for polcy makers. The fac ha on-he-job ranng has a large effec on subsequen wage growh and mos ranng akes place durng he early sages of workers careers s ofen gven as a jusfcaon for allowng young workers o be pad wages less han he adul mnmum (Eyraud and Sage 2005). However, f here s no lnk beween he mnmum wage and ranng, such youh raes or submnmum wages are unnecessary and wll only serve o preven younger workers from reachng pay equaly wh aduls. Ths paper examnes wheher he long-run effecs of he mnmum wage on wage growh vary accordng o he naure of wage barganng. A smple model of on-hejob ranng s presened, whch mples ha he mnmum wage should have dfferen effecs on a worker s subsequen rae of wage growh dependng on wheher he level of ranng s deermned by ndvdual barganng or collecve barganng. Longudnal daa from he Annual Survey of Hours and Earnngs and Workplace Employmen Relaons Survey n he Uned Kngdom are hen used o es hese predcons. A quas-naural expermen approach s aken, whch nvolves comparng he wage growh of oherwse dencal workers who were affeced by he youh mnmum wage o dfferen degrees because hey enered he labour marke n dfferen years. 2

3 Ths paper provdes he frs evdence of how he effecs of he mnmum wage on wage growh vary by barganng sysem and, consequenly, on he effcacy of youh raes n rasng he wages of low-wage workers over her workng lves. In addon, he resuls sugges a role ha unons mgh play n an era n whch mnmum wage laws have eroded her radonal role of rasng wages. 2. Background Early emprcal sudes of he ranng effecs of he mnmum wage used workers wage growh raes as a proxy for her level of on-he-job ranng. These sudes ended o fnd a negave relaonshp beween he mnmum wage and wage growh (Leghon and Mncer 1981; Lazear and Mller 1981; Hashmoo 1982), whch s conssen wh employers cung back on ranng n response o he mnmum wage. However, as noed by Grossberg and Sclan (1999), here are a number of problems wh usng wage growh as a proxy for ranng: wage growh depends on he fracon of job-specfc ranng ha akes place and he relave barganng power of frms and workers; oher heores of pay srucure (such as Lazear s (1979) delayed compensaon model) predc lower wage growh n response o he mnmum wage for reasons ndependen of producvy; and usng wage changes across all workers fals o accoun for he fac ha mos ranng occurs early n a worker s career. As a consequence of hese problems, more recen sudes have examned he effec of mnmum wages on drec measures of ranng parcpaon. Some of hese fnd negave effecs (Schller 1994; Neumark and Wascher 2001), whereas ohers fnd an nsgnfcan relaonshp (Smpson 1984; Grossberg and Sclan 1999; Acemoglu and Pschke 2003; Arulampalam e al. 2004; Farrs and Pedace 2004). Alhough he use of drec nformaon on ranng avods he ssues noed by Grossberg and Sclan, suffers from a number of oher problems. Frsly, he avalable ranng daa are almos always poor measures of he breadh and exen of acual acves ha rase a worker s fuure producvy. Secondly, he benefs of some ranng mgh be very small or mgh be prmarly receved by frms, meanng ha hey have lle drec effec on a worker s welfare. Fnally, a narrow focus on ranng s unable o provde evdence o polcymakers of how much he nroducon of a youh mnmum wage rae mgh rase young workers ncomes over her enre careers. 3

4 For hese reasons, hs paper revss he lnk beween mnmum wages and subsequen wage growh, usng panel daa and a quas-naural expermenal desgn o address Grossberg and Sclan s concerns. Two recen papers have examned how a person s exposure o he mnmum wage whle young affecs hs/her subsequen wage oucomes. However, neher of hese akes no accoun he exen o whch he mnmum wage affecs a specfc young worker s wage (and poenally level of ranng). Neumark and Nzalova (2007) only observe maure workers and deermne wha mnmum wage hey were lkely o have faced n he pas, based on her age and curren sae of resdence. Cardoso (2009) used longudnal daa over a long perod, bu assgned a mnmum wage based only on he year workers were born and when hey enered he labour marke, no on wheher hey were acually bound by he mnmum wage durng her youh. Neumark and Nzalova found evdence of a negave long-erm effec of he mnmum wage on wages, whereas Cardoso found evdence of a posve effec on wages bu also a negave effec on he wage-enure profle. Furhermore, unlke oher sudes ha have looked a wage growh, he focus here s on he dfference n wage growh beween mnmum wage workers n wo secors of he economy jobs wh ndvdual wage barganng and jobs wh collecve barganng. The only relaed prevous sudy was Acemoglu and Pschke (2003), who analysed how he effecs of he mnmum wage on ranng vary by level of labour marke compeveness, as measured by ndusry wage dfferenals. They found some weak evdence ha ranng s posvely relaed o he mnmum wage among workers n less compeve secors of he Uned Saes economy. However, Acemoglu and Pschke were resrced by her daa source (he Naonal Longudnal Survey of Youh 1979) o focus on ranng underaken beween ages 22 and 29, whch s lkely o be oo lae o capure mos of he general on-he-job ranng underaken by mnmum wage workers. These daa also provded raher mprecse measures of a worker s hourly wage (and hence wheher he worker was bound by he mnmum wage), as hey rely on rerospecve survey responses. 3. The Uned Kngdom Naonal Mnmum Wage A naonal mnmum wage was nroduced n he Uned Kngdom on 1 Aprl 1999, coverng all workers aged 18 and over. Inally, hs conssed of an adul rae for hose aged 22 and over, se a 3.60, and a developmen rae for hose aged 18-21, 4

5 se a On 1 Ocober 2004, a (lower) rae was nroduced for workers aged On 1 Ocober 2010, he age lm for he adul rae was lowered o 21 and an apprence rae was nroduced, for apprences aged or hose aged 19 and over and n her frs year of her apprenceshp. The mnmum wage raes are revewed each year by he Low Pay Commsson, an ndependen advsory body whch makes recommendaons o he Governmen. The Governmen has chosen o ncrease all of he raes each Ocober snce 2000, bu by varyng amouns each year. By Ocober 2010, he adul rae was 5.93, he developmen rae was 4.92, he rae was 3.64 and he apprence rae was The mnmum wage apples o all workers who have a conrac o do work personally, raher han for a cusomer or clen. Hence, apples o casual workers, emporary workers and agency workers, bu no o he self employed. For he purposes of he mnmum wage, a worker s hourly wage s calculaed by dvdng hs/her oal gross pay for he reference perod (normally a monh) by oal hours worked durng ha perod. Incenve pay s ncluded n hs calculaon, bu no premum pay for shf or overme work. All me spen dong on-he-job ranng (and me spen ravellng o and from ) s subjec o he mnmum wage. 4. Theory Consder a model n whch each worker, ndexed by, s assumed o work n he labour marke for T 1 perods and wshes o maxmse dscouned lfeme earnngs: W w T 0 (1 r), (1) where w s he worker s wage n perod and r s he dscoun rae. A worker s margnal revenue produc n each perod defned as he value n a perfecly compeve produc marke of wha he worker produces s assumed o conss of a fxed componen μ, whch capures dfferences n producvy beween workers, and a me-varyng componen γ, whch capures he fac ha producvy nally ncreases over he lfe cycle, before fallng. Workers are able o engage n onhe-job ranng, bu only wh he permsson of her employer. Tranng s assumed o ake place durng he nal perod of employmen only and o produce human capal gans for he followng T perods. The cos of ranng manfess self as a fall of θ n margnal revenue produc n he frs perod, where 0, ], reflecng he [ max 5

6 fac ha workers mus ake me off her regular dues o undergo ranng. The benefs of ranng are a rse of f ( ) n margnal revenue produc n perods 1 o T. To ease exposon, he followng funconal form s used: f ( ), where 1,, (2) 1 and where s he frs dfference operaor. Equaon 2 s conssen wh a seng n whch he benefs from ranng (n erms of ncreases n margnal revenue produc) are realsed over mulple perods, bu ha he larges gans come frs and ha evenually he growh rae n margnal revenue produc falls o ha of unraned workers. All ranng s assumed o be general n naure and, hence, perfecly ransferable o oher jobs. I wll be assumed ha he margnal cos of ranng s ouweghed by he expeced margnal benefs of ranng over a worker s career, so ha: T 1 (1 r) 1. (3) Frms are assumed o maxmse prof n each perod and o operae n an mperfecly compeve produc marke, so ha here s a surplus equal o per worker each perod. 1 Two alernave assumpons abou he naure of barganng n he labour marke are now dscussed n urn. Indvdual barganng If workers bargan ndvdually, s assumed ha hey wll have nsuffcen power o oban any of he produc marke rens. Therefore, worker s wage n any perod s equal o hs/her margnal revenue produc: w 0 f 0. (4) f 1,..., T Suppose nally ha here s no mnmum wage. The level of ranng has no effec on a frm s profs snce s pad for by workers hrough lower wages n perod 0, bu he frm can maxmse he worker s uly by agreeng o provde he opmal level of ranng. Gven he assumpon made abou ranng coss n equaon 3, mus be he case ha all frms choose he maxmum amoun of ranng avalable, 1 The presence of a surplus mplcly assumes ha he workers and frms are barganng over pons on a vercal conrac curve. 6

7 max, whch s he same for all workers. 2 From equaon 4, a worker s wage n he nal perod s w. * 0 0 max Suppose now ha a mnmum wage w s nroduced. The frm s prof maxmsaon problem n perod 0 can be wren as follows: max 0 w0, subjec o w 0 max{ 0, w}. (5) Obvously, f he mnmum wage s se below * w 0, wll have no effec on he amoun of ranng akng place. However, f w 0 max, he mnmum wage wll preven he opmal level of ranng from akng place, because he frm s profs wll fall n hs case. Insead, he level of ranng allowed by he frm wll be: w w w. (6) * 0 max 0 Equaon 6 represens a case of perfec crowdng ou beween he mnmum wage and he amoun of ranng. As Becker (2009) explans, n a perfecly compeve labour marke frms provde general ranng only f hey [do] no have o pay any of he coss. If he mnmum wage rses so hgh ha ranng s cu o zero, he frm wll rean he worker as long as sll makes excess profs from dong so. Furher mnmum wage ncreases wll have no effec on ranng. 3 Therefore, a complee descrpon of he frm s ranng decson s as follows: 0 If w max max w w * 0 f w w f w * 0 f * 0 w 0 0 w 0. (7), he frm wll make negave profs n perod 0, so wll choose 0 o lay off he worker. Snce ranng s only provded n perod 0, f he frm makes non-negave profs n ha perod, mus also do so n all subsequen perods. 2 Ths conrass wh Acemoglu and Pschke s (2003) model, whch assumed ha no ranng akes place n he absence of a mnmum wage. 3 Because boh γ and β ncrease over me, as he mnmum wage rses, wll bnd frs n perod 0, hen n perod 1 and so on. If he supernormal profs are hgh enough, s concevable ha he frm wll choose o rensae ranng once he mnmum wage s rased beyond a ceran level, because knows wll keep all he reurns from he raned worker. Ths s he concep behnd Acemoglu and Pschke (2003). However, no evdence of a posve relaonshp beween he mnmum wage and ranng s found over any wage range (as seen n he frs column of Table 3). 7

8 Collecve barganng Suppose now ha workers bargan collecvely and oban a share,, of he oal surplus. The level of wll be deermned by he barganng power he worker s unon has. The unon can negoae wh he frm o provde s share of he surplus,, as a combnaon of employer-provded ranng n perod 0,, and a wage premum n each perod,. Hence, n he absence of a mnmum wage, he worker s wage schedule wll be: w 0 f 0. (8) f 1,..., T Raher han he level of ranng beng solely deermned by he frm, wll now be deermned by he jon barganng of he unon and frm. If he unon wshes o maxmse a represenave worker s expeced earnngs over hs/her enre career, s objecve wll be as follows: max, T T, subjec o. (9) (1 ) (1 r T 1 (1 r) 0 r 0 ) Snce every pound spen on ranng yelds more over he worker s lfeme han does an addonal pound n wages (and snce hs s he case for every worker), he unon should prorse negoang for he frm o pay for he opmal amoun of * ranng n perod 0, so ha max. Once hs has been ensured, he remnder of he worker s share of he surplus can hen be demanded n he form of a wage premum, so ha: T * (1 r). (10) max 0 In he absence of a mnmum wage, he dsncon beween negoang over ranng and negoang over pay s rrelevan, because ndvdual workers can always arrange o pay for addonal ranng ou of her wages, as n he ndvdual barganng case. Once a mnmum wage s mposed, however, he frm wll aemp o recoup coss by cung back on any ndvdually-arranged ranng. As long as he unon has reached an agreemen ha he frm provde he opmal amoun of ranng, hs wll no be possble and he frm wll have o absorb he coss of he mnmum wage, bu wll connue o hre he worker, as long as makes non-negave expeced profs, whch requres ha w 0 max. Hence, he exsence of 8

9 collecve barganng ensures ha he opmal amoun of ranng connues o be offered, even n he presence of a mnmum wage. Equlbrum wage growh Regardless of he naure of barganng, he change n a worker s wage beween any wo pos-ranng perods s gven by he followng: w. (11) Snce Under ndvdual barganng, he opmal level of ranng s gven by equaon 7. s always posve, a worker s perod 0 wage gap should have a negave effec on hs/her subsequen wage growh (up o a ceran pon), as frms reduce levels of on-he-job ranng. Under collecve barganng, he opmal ranng level s fxed a max, so he wage gap should have no effec on subsequen wage growh. The relaonshp beween wage growh and he wage gap n he wo scenaros s depced n Fgure Daa The analyss draws prmarly on daa from he Annual Survey of Hours and Earnngs (ASHE), whch s an annual survey ha collecs daa on he wages, work hours and oher employmen arrangemens of around 1% of he UK workng populaon (Offce for Naonal Sascs 2013). Basc addonal nformaon, such as age and sex, s also ncluded. The ASHE was nroduced n 2004 and replaced he New Earnngs Survey (NES). However, by applyng ASHE mehodologes o NES daa for he perod, he Offce for Naonal Sascs has produced ASHE daases for 1997 onwards. The analyss n hs sudy uses daa for The ASHE sample s drawn from HM Revenue and Cusoms Pay As You Earn (PAYE) regser, based on he las wo dgs of a worker s Naonal Insurance Number. Survey forms are sen o all employers of he seleced workers o complee. The quesons n he ASHE refer o a reference week, whch s n early Aprl of each year. Snce he responses are provded by employers raher han by employees, he ASHE wage and hours daa are consdered o be hghly accurae, compared o radonal household surveys and censuses. If a person does no work n a gven year, he/she wll no appear n he daase. Therefore, he only way o deermne wheher someone has moved ou of employmen 9

10 s by her absence from he daa n a gven year. Obvously, hs wll also nclude people who have moved abroad or ded. However, as long as he fracon of people makng hese ransons s consan across he wage dsrbuon, hs wll no bas he resuls from he emprcal sraegy oulned n he nex secon. Workers mgh also be absen from he ASHE f her employers fal o respond o he quesonnare or f hey are no ncluded n he PAYE regser because her earnngs fall below he Naonal Insurance Lower Earnngs Lm. For he analyss, he sample s resrced o hose who were aged beween 22 and 31, had full-me jobs and who were employed n he followng year. In cases where a person repored more han one full-me job n a year, only he job on whch he/she worked he mos hours s ncluded. The choce of age range ensures ha all ndvduals n he sample are observed n he ASHE beween ages 16 (when hey are able o leave school) and Ths age range durng whch people are subjec o he year-old rae or he developmen rae for ages (or snce 2010) s nally assumed o be he phase of a worker s career n whch on-he-job ranng akes place (perod 0 n he model of he prevous secon). The sample s furher resrced o hose who had a full-me job a some pon beween 16 and 21 and who were no bound by he mnmum wage when hey frs enered he labour marke. The former resrcon ensures ha mos unversy graduaes are omed from he sample, so ha hose ncluded have relavely low levels of human capal and a greaer need for general on-he-job ranng. The laer resrcon ensures ha here s no downward bas n he esmaed effec of he mnmum wage on subsequen wage growh, snce workers who are nally hred a he mnmum wage may already be recevng less ranng han hey would n he absence of a mnmum wage. As seen n Fgure 1, he magnude of he mnmum wage effec s equal o he dfference beween a person s unconsraned perod 0 wage and he mnmum wage, w * w 0. The emprcal analyss follows hs, as well as he approach used by Lnneman (1982) and Curre and Fallck (1996), n consrucng a reamen varable measurng how much exra an employer mus pay o rean a curren employee afer a mnmum wage ncrease. The reamen varable, WAGEGAP, measures he cumulave effec of all mnmum wage ncreases beween 16 and 21 on person. I s 4 Specfcally, durng he sample perod, young people were allowed o leave school durng he school year n whch hey urned

11 defned as follows, where AGE s he person s age n year, w s hs/her curren hourly wage (n 2012 pounds, adjused usng he RPI) and w( 1 ) s he relevan mnmum wage n he followng year (akng no accoun he worker s age): WAGEGAP max{ w. (12) 16 AGE 21 (1 ) w,0} Abou 11% of he workers n he sample (accounng for 9% of he observaons) were bound by he mnmum wage someme beween ages 16 and 21. Means for he man varables used n he analyss are gven n panel A of Table 1, separaely for hose workers who were bound by he mnmum wage before age 22 and hose who were no. Compared o he unbound workers, he bound workers earn less, are more lkely o be male and are less lkely o have job wh a collecve agreemen. Among bound workers, average annual wage growh was 34 pence, whereas among he unbound workers, was 47 pence. To examne he effec of collecve barganng, he analyss draws on a queson n he ASHE whch asks wheher a worker s pay was se wh reference o an agreemen affecng more han one employee, for example, agreemens wh rade unons or workers commees. A second wage gap varable was calculaed, akng no accoun only hose observaons n whch a worker was covered by a collecve agreemen beween ages 16-21, as follows: COLAGWAGEG AP COLAG max{ w, (13) 16 AGE 21 (1 ) w,0} where COLAG s a dummy for wheher a worker was covered by a collecve agreemen n a gven year. COLAG was defned as he fracon of years durng he ranng perod n whch he ndvdual had a collecve agreemen (and herefore ranges from zero o one). Fgure 2 plos average real wages beween ages 16 and 31, accordng o wheher or no a person was bound by he mnmum wage before age 22 and wheher or no hey had spen mos of her me on collecve agreemen jobs before age 22. Overall, wages ncrease seadly wh age. The unbound group has very smlar wages a each age level, regardless of wheher workers had collecve agreemen jobs or no. The bound group has lower wages han he unbound group and has smlar wages before age 22, regardless of collecve agreemen saus. However, among he bound workers, hose who had collecve agreemens before 22 experenced smlar wage 11

12 growh o he unbound group beween 22 and 31, whereas he non-collecve agreemen group fell furher below he unbound group. The nex secon examnes wheher hs dfference n wage growh persss, once he effecs of oher possble deermnans are conrolled for. The ASHE conans no nformaon on he naure of barganng beween frms and unons or on he prevalence of ranng among workers. Therefore, he analyss draws on supplemenary daa from he Workplace Employmen Relaons Survey (WERS) The WERS 2004 conssed n par of nervews wh 2,295 managers a workplaces wh 5 or more employees, conduced beween February 2004 and Aprl In each workplace, he mos senor manager responsble for employmen relaons and personnel ssues was seleced. Managers were asked wha proporon of experenced workers n he larges occupaon n he workplace had been gven me off work o underake ranng over he precedng 12 monhs and he average number of days of ranng ha was underaken. In addon, hey were asked wheher managemen normally negoaed wh, consuled, nformed or dd no nform unons abou raes of pay, ranng of employees, hours of work, holday enlemens, penson enlemens, recrumen or selecon of employees, grevance procedures, dscplnary procedures, saffng plans, equal opporunes, healh and safey, and performance apprasals. The WERS daa are merged no he ASHE sample a he esablshmen level, usng he workplace seral number. Means for he merged sample are repored n panel B of Table 1. The sample s much smaller han he full ASHE sample. Because he WERS daa are merged n a he esablshmen level, people who worked a large esablshmens when hey were young are more lkely o be n he mached sample. However, alhough average wage growh s slghly lower han n he full ASHE sample, he dsrbuon of oher characerscs s smlar. To examne he effec of specfc ypes of barganng behavour, wo furher reamen varables are defned usng he WERS varables, analogous o COLAGWAGEGAP. These measure he cumulave effec of he mnmum wage on young workers only a frms whch negoaed over ranng or negoaed over pay: NEGTRAINWA GEGAP NEGTRAIN max{ w, (14) 16 AGE AGE 21 (1 ) w,0} NEGPAYWAGE GAP NEGPAY max{ w, (15) (1 ) w,0} 12

13 where NEGTRAIN and NEGPAY are dummy varables ndcang wheher worker worked a a frm ha negoaed over ranng or pay, respecvely, n year. Lkewse, NEGTRAIN and NEGPAY measure he fracon of years beween ages 16 and 21 ha he worker worked a a frm ha negoaed over ranng or pay, respecvely. 6. Analyss The emprcal sraegy s o compare he wage growh raes of oherwse dencal workers who have dfferen values of WAGEGAP because hey enered he labour marke a dfferen pons n me and hence were subjec o dfferen values of he mnmum wage durng her youh. The wage gap wll vary for a number of reasons: he overall nroducon of he naonal mnmum wage n 1999, he nroducon of he and apprence raes, he lowerng of he age of elgbly for he adul rae and he annual ncreases n all raes by dfferen amouns. Resuls usng he ASHE sample To begn wh, he followng basc specfcaon s used: w where ( 1) 0 1WAGEGAP 2w 3 AGE X ψ u. (16) w s a person s real wage durng hs/her frs year n full-me work (beween ages 16 and 21) and conrols for dfferences n wage growh ha are due o dfferences n nheren producvy across people. Hence, he specfcaon compares workers who nally earn he same wage, bu who ener he labour marke n dfferen years and herefore experence dfferen wage rses before 21 due o changes n he mnmum wage. AGE s ncluded as a proxy for he erms n equaon 11. X s a vecor of conrol varables, comprsng a person s wage gap n year (ha s, max{ w( 1 ) w,0} ) and a dummy varable for wheher he person s male. The conemporaneous wage gap conrols for he possbly ha workers wages wll grow because hey connue o be bound by he mnmum wage afer age 21. λ s a year fxed effec and u s an error erm. All sandard errors are clusered by person. The frs column of Table 2 presens he resuls of esmang equaon 16 usng he full ASHE sample. The youh wage gap (WAGEGAP) has an nsgnfcan coeffcen, bu he nal real wage has a sgnfcan posve effec. Among he 13

14 conrol varables, he conemporaneous wage gap has a coeffcen ha s sgnfcanly larger han he value of 1 predced, men are found o have sgnfcanly greaer wage growh han women and wage growh s found o declne wh age. Alhough WAGEGAP has no overall effec on wage growh, he heory n Secon 2 predcs ha should only have a negave effec on wage growh among workers who were covered by ndvdual agreemens. In order o allow he coeffcen on WAGEGAP n equaon 16 o vary accordng o he ype of wage agreemen a person had before age 22, COLAGWAGEGAP s added as a regressor. COLAG s also added n order o conrol for sysemac dfferences n wage growh beween hose who had collecve agreemens when hey were young and hose who dd no, whch mgh be caused by general dfferences n he amoun or ype of ranng provded n he wo secors or by he non-random selecon of workers no each secor: w( 1) 0 1WAGEGAP 2COLAGWAGEG AP 3COLAG 4 5 AGE Xφ v. (17) A sgnfcan esmae of 2 n equaon 17 ndcaes a dfference n he relaonshp beween wage growh and he wage gap beween he collecve and noncollecve agreemen groups. As seen n he hrd column of Table 2, he overall wage gap has a sgnfcan negave coeffcen, ndcang ha he mnmum wage has a negave effec on wage growh among hose workers who were no covered by a collecve agreemen. The value of hs ndcaes ha a wage ncrease of 1 beween he ages of 16 and 21 brough abou by he mnmum wage wll resul n an 11 pence decrease n annual wage growh afer 21, equvalen o an elascy of a he mean. 5 However, conssen wh heory, he effec of he wage gap s sgnfcanly dfferen for hose workers who were covered by a collecve agreemen before 22 and he overall effec of he wage gap s nsgnfcan for hese workers. The assumpon of decreasng reurns o ranng (mpled by he shape of β n he model) means ha he effec of he wage gap should fall as a worker ages. To examne hs, he regresson was run separaely for workers aged and n columns hree and four n Table 2. As predced, he coeffcens on WAGEGAP and COLAGWAGEGAP were much sronger among he younger group. A bndng 1 ncrease n he mnmum wage beween 16 and 21 s found o yeld a 16 pence w 5 Takng no accoun equaons 8 and 12, hs resul mples ha every pound of ranng before age 22 yelds benefs ha grow by 11 pence a year beween ages 22 and 31 on average. 14

15 decrease n annual wage growh beween 22 and 26 (equvalen o an elascy of 0.387), bu no effec on wage growh beween 27 and 31. Overall, he resuls sugges ha he exsence of he 16-17, and apprence raes of he mnmum wage has sgnfcanly rased wage growh among low-wage workers. Compared o a scenaro n whch he adul mnmum wage appled o all workers aged 16 and over, he average worker who was bound by he (acual) mnmum rae experenced 4.0 pence hgher per annum wage growh beween ages 22 and 31 (or 12% a he mean). However, hs was solely drven by workers wh noncollecve agreemen jobs: among hs group only, he effec s 7.4 pence (or 23% a he mean). Robusness ess A lkely explanaon for he larger-han-expeced coeffcens on he conemporaneous wage gap n Table 2 s ha workers near he boom of he wage dsrbuon n a gven year end o experence sronger wage growh regardless of he level of he mnmum wage (Dckens e al. 2015). If a person s poson n he wage dsrbuon s srongly correlaed wh hs/her wage before age 22, hs mgh bas he coeffcens on WAGEGAP and COLAGWAGEGAP. To conrol for hs, a se of dummes for a person s cenle n he enre U.K. wage dsrbuon n a gven year are added as regressors n he frs column of Table 3. The coeffcen on he conemporaneous wage gap falls and s no longer sgnfcanly dfferen from he hypohessed value of 1. Age now has a posve coeffcen. However, he coeffcens on WAGEGAP and COLAGWAGEGAP are lle changed and reman sgnfcan. The model presened n Secon 4 mples ha when he wage gap s hgh enough, wage growh for he ndvdual barganng secor should fall o zero and ha furher mnmum wage ncreases should have no effec on wage growh. To examne wheher equaon 17 obscures a non-lnear relaonshp, he squares of boh WAGEGAP and COLAGWAGEGAP are added as regressors n he second column of Table 3. There s no evdence ha he wage growh-wage gap profle flaens when he mnmum wage s parcularly hgh. The esmaed relaonshp, whch s depced n Fgure 3, s close o lnear and resembles he downward-slopng poron of Fgure 1. Excess profs are lkely o vary by ndusry, due o dfferences n he degree of compeveness of he produc marke. Snce collecve agreemens are lkely o occur 15

16 only n ndusres where here are hgh rens, s possble ha he collecve agreemen effec ha has been dscussed so far smply reflecs dfferences n he level of produc marke compeveness ha are unrelaed o he presence of collecve barganng. In addon, ranng s lkely o vary by ndusry due o dfferen raes of urnover, whch may also be correlaed wh he wage gap measures. To conrol for hs, dummy varables were added o equaon 17 for he fve-dg ndusry sub-class of each person s frs full-me job. 6 Ths s found o have lle effec on he resuls (as repored n he hrd column of Table 3). Even whn narrowly-defned ndusral groups, havng an ndvdual agreemen job resuls n sgnfcanly lower wage growh laer n a worker s career, compared o workers on collecve agreemen jobs wh he same wage gap. In addon o he overall level of rens per worker, he amoun of ranng provded should be affeced by he barganng power of he unon ( n he model presened earler). The degree of barganng power of a unon n he ASHE sample s unknown, bu s lkely o be relaed o he number of workers covered by an agreemen. Usng nformaon repored n he ASHE, workers wh collecve barganng were classfed as beng represened by sngle-company agreemens (organsaonal or workplace agreemens) or mulple-company agreemens (naonal, sub-naonal or ndusry agreemens). Unons whch negoae mulple-company agreemens are expeced o have more barganng power han unons whch negoae a he company level only. As seen n he las column of Table 3, no sgnfcan dfference s found beween he wo ypes of collecve agreemen. However, wage growh among hose who worked a frms wh sngle-company agreemens s slghly lower, conssen wh he predcons of he model. In he resuls presened so far, all work done before age 22 s ncluded n he calculaon of he wage gap measures and all workers who commenced full-me work before hs age are ncluded n he sample. Some of hese workers wll have aaned erary qualfcaons and herefore wll consue a poor comparson group for he mnmum wage workers of neres n hs sudy. To examne how robus he resuls are o hs decson, lower age cu-offs were used. 7 Table A1 reveals ha he coeffcens on WAGEGAP and COLAGWAGEGAP have he same sgns when 19, 20 6 The UK Sandard Indusral Classfcaon of Economc Acves 2003 was used. 7 No nformaon on educaon s avalable n he ASHE. 16

17 or 21 are used as he cu-off age, alhough he coeffcen on WAGEGAP loses s sgnfcance. The dfference n wage growh beween mnmum wage workers wh collecve and ndvdual agreemens remans hghly sgnfcan n all cases. Resuls usng he ASHE-WERS sample The resuls presened so far sugges ha workers who have jobs wh ndvdual barganng experence worse wage growh laer n lfe when he mnmum wage s hgh. However, he model presened n Secon 4 specfcally requres ha unons bargan wh frms o preserve ranng n order o preven frms from cung ranng n response o he mnmum wage. Barganng only over pay s no suffcen, because frms are lef free o cu ranng o offse ncreased wage coss n hs case, as n he ndvdual barganng secor. In order o observe he specfc naure of barganng beween unons and frms and o be ceran ha barganng over ranng s responsble for he observed dvergence n wage growh, he merged ASHE-WERS daa are used. As seen n panel B of Table 1, only 46% of workers n he mached sample worked a frms who negoaed wh unons over ranng when hey were young, even hough 73% worked a frms who negoaed over pay, ndcang ha many unons negoae only over pay and no over oher job condons. Overall, negoaon over ranng s seen o be effecve a rasng ranng levels. In Table A2, measures of ranng are regressed on a seres of negoaon varables a he esablshmen level. Ths confrms ha frms whch negoae over ranng provde ranng o a greaer fracon of her workers (n he frs column) and offer more days of ranng on average (n he second column). Negoaon over pay s assocaed wh lower levels of ranng, perhaps ndcang ha frms aemp o offse he coss of payng hgher wage prema by reducng ranng provson. In he frs column of Table 4, equaon 17 s esmaed on he mached ASHE- WERS sample, hs me replacng COLAGWAGEGAP and COLAG wh NEGTRAINWAGEGAP and NEGTRAIN. Alhough he resuls are weaker han before (reflecng he smaller sample sze), here s a sgnfcan dfference n long-run wage growh beween workers whose ranng levels were negoaed beween frms and unons and workers whose ranng levels were se by he frm unlaerally. When NEGPAYWAGEGAP and NEGPAY are added n he second column, hey are found 17

18 o be nsgnfcan, confrmng ha unons mus negoae wh frms over ranng no jus pay n order for workers o avod poor fuure wage growh. Fnally, he fracon of workers a he esablshmen who were provded ranng was added o he regresson n he fnal column of Table 4. Ths s found o have a sgnfcan posve effec on subsequen wage growh, suggesng ha he ranng repored by frms n he WERS s of value o workers. Moreover, he coeffcen on NEGTRAINWAGEGAP becomes nsgnfcan once ranng s added, conssen wh he noon ha he effecs of he mnmum wage on fuure wage growh operae solely va ranng and no some oher mechansm, such as he whdrawal of delayed compensaon schemes. 7. Concluson Ths paper has examned wheher he mnmum wage leads o lower wage growh laer n a worker s career and how hs dffers accordng o he naure of barganng on a job. A smple model of ranng and prof maxmsaon mples ha beng pad he mnmum wage when young should lead o lower wage growh hereafer for hose who work n secors of he labour marke wh ndvdual barganng, bu ha he mnmum wage should have no effec on wage growh for hose whose wages are deermned by collecve barganng. Ths s because unons are able o negoae for employers o provde and pay for ranng, whch canno be whdrawn o offse he coss of he mnmum wage. Daa from he Annual Survey of Hours and Earnngs n he UK reveal suppor for hese predcons. Workers who receved large wage gans due o he mnmum wage before he age of 22 experence lower wage growh beween he ages of 22 and 31, bu only f hey had ndvdual employmen agreemens before 22; he relaonshp s nsgnfcan for hose who were covered by collecve agreemens. Addonal daa from he Workplace Employmen Relaons Survey confrm ha hs dfference s drven by dfferences n he amoun of ranng receved. Overall, he resuls ndcae he connued mporance of unons n rasng wage raes, despe he populary of mnmum wages, bu sugges ha negoaons over ranng should no be consdered as perpheral o a unon s objecves as hey have radonally been. References 18

19 Acemoglu, Daron, and Jörn-Seffen Pschke The srucure of wages and nvesmen n general ranng. Journal of Polcal Economy, 107(3): Acemoglu, Daron, and Jörn-Seffen Pschke Mnmum wages and on-he-job ranng. Research n Labor Economcs, 22: Allegreo, Sylva, and Mchael Rech Are local mnmum wages absorbed by prce ncreases? Esmaes from nerne-based resauran menus. Indusral and Labor Relaons Revew, 71(1): Arulampalam, Wj, Alson L. Booh and Mark L. Bryan Tranng and he new mnmum wage. Economc Journal, 114(494): C87-C94. Becker, Gary S Invesmen n human capal: A heorecal analyss. Journal of Polcal Economy, 70(5): Becker, Gary S Human capal, 3 rd edon. Chcago, IL: Unversy of Chcago Press. Bell, Bran, and Sephen Machn Mnmum wages and frm value. Journal of Labor Economcs, 36(1): Belman, Dale, and Paul J. Wolfson Wha does he mnmum wage do? Kalamazoo, MI: W.E. Upjohn Insue for Employmen Research. Cardoso, Ana Rue Long-erm mpac of youh mnmum wages: Evdence from wo decades of ndvdual longudnal daa. Dscusson Paper No. 4236, IZA, Bonn. Curre, Jane, and Bruce C. Fallck The mnmum wage and he employmen of youh: Evdence from he NLSY. Journal of Human Resources, 31(2): Dckens, Rchard, Rebecca Rley, and Davd Wlknson A re-examnaon of he mpac of he UK naonal mnmum wage on employmen. Economca, 82(328): Eyraud, Franços, and Caherne Sage The fundamenals of mnmum wage fxng. Geneva: Inernaonal Labour Offce. Farrs, Davd, and Robero Pedace The mpac of mnmum wages on job ranng: An emprcal exploraon wh esablshmen daa. Souhern Economc Journal, 70(3): Grossberg, Adam J., and Paul Sclan Mnmum wages, on-he-job ranng, and wage growh. Souhern Economc Journal, 65(3):

20 Hashmoo, Masanor Mnmum wage effecs on ranng on he job. Amercan Economc Revew, 72(5): Lazear, Edward P Why s here mandaory reremen? Journal of Polcal Economy 87(6): Lazear, Edward P., and Frederck H. Mller, Mnmum wage versus mnmum compensaon. In Repor of he Mnmum Wage Sudy Commsson, vol. 5, Washngon, DC: US Governmen Prnng Offce. Leghon, Lnda, and Jacob Mncer The effecs of mnmum wages on human capal formaon. In (Smon Roenberg, ed.) The Economcs of Legal Mnmum Wages, , Washngon DC: Amercan Enerprse Insue. Lnneman, Peer The economc mpacs of mnmum wage laws: A new look a an old queson. Journal of Polcal Economy, 90(3): Neumark, Davd, and Olena Nzalova Mnmum wage effecs n he longer run. Journal of Human Resources, 42(2): Neumark, Davd, and Wllam Wascher Mnmum wages and ranng revsed. Journal of Labor Economcs, 19(3): Offce for Naonal Sascs Annual Survey of Hours and Earnngs, : Secure Access [compuer fle]. 4h Edon. Colcheser, Essex: UK Daa Archve [dsrbuor], June SN: Papps, Kerry L The effecs of socal secury axes and mnmum wages on employmen: Evdence from Turkey. Indusral and Labor Relaons Revew, 65(3): Rosen, Sherwn Learnng and experence n he labor marke. Journal of Human Resources, 7(3): Schller, Bradley R. (1994). Movng up: The ranng and wage gans of mnmum wage enrans. Socal Scence Quarerly, 75(3): Smpson, Wayne An economerc analyss of ndusral ranng n Canada. Journal of Human Resources, 19(4):

21 Fgure 1 Predced relaonshp beween wage growh and wage gap n perod 0 Δw Δβ θ max Indvdual barganng Collecve barganng θ max * w w 0 21

22 Fgure 2 Mean wages for workers, by collecve agreemen and bound saus when young Noes: Bound workers are hose workers who earned less han he followng year s mnmum wage a any pon beween ages 16 and 21. The collecve agreemen group refers o hose workers wh more perods on jobs wh collecve employmen agreemens beween ages 16 and 21 han on jobs whou collecve agreemens. ASHE survey weghs are used. 22

23 Fgure 3 Predced wage growh by youh wage gap Noes: The predced wage growh s calculaed from he model n he second column of Table 3, seng all oher varables a her means. The horzonal axs ranges from 0 o 4.98, whch s he 99% percenle of WAGEGAP among workers who were bound by he mnmum wage before age

24 Table 1 Descrpve sascs for he esmaon samples A. ASHE sample Varable All workers Bound Unbound workers workers Annual wage growh, Δw (2012 pounds) Youh wage gap, WAGEGAP (2012 pounds) Conemporaneous wage gap (2012 pounds) Real wage (2012 pounds) Collecve agreemen when young, COLAG Male Age Sample sze 64,059 5,551 58,508 B. ASHE-WERS sample Varable All workers Bound Unbound workers workers Annual wage growh, Δw (2012 pounds) Youh wage gap, WAGEGAP (2012 pounds) Conemporaneous wage gap (2012 pounds) Real wage (2012 pounds) Frm negoaed ranng when young, NEGTRAIN Frm negoaed pay when young, NEGPAY Male Age Sample sze 1, ,531 Noes: ASHE survey weghs are used 24

25 Table 2 Resuls for wage growh regressons usng ASHE daa Varable All observaons Aged Aged () () () () Youh wage gap (0.036) 0.112* (0.058) 0.159** (0.074) (0.046) Youh wage gap on collecve 0.128** 0.195** agreemen jobs (0.065) (0.081) (0.091) Collecve agreemen when 0.058** 0.113*** young (0.027) (0.041) (0.098) Inal real wage 0.027** (0.011) 0.028** (0.011) 0.045*** (0.016) (0.034) Conemporaneous wage gap 1.795*** (0.090) 1.796*** (0.090) 1.677*** (0.092) 2.243*** (0.255) Male 0.109*** (0.021) 0.110*** (0.021) 0.145*** (0.031) (0.070) Age 0.029*** (0.006) 0.028*** (0.006) 0.038*** (0.011) (0.027) R-squared Sample sze 64,059 64,059 43,211 20,848 Noes: All regressons nclude a full se of year (13 caegores) dummes. Sandard errors are clusered by person and are presened n parenheses. *, ** and *** denoe sgnfcance a he 10%, 5% and 1% level, respecvely. 25

26 Table 3 Addonal resuls for wage growh regressons usng ASHE daa Varable () () () (v) Youh wage gap 0.104* (0.059) (0.103) 0.111* (0.060) 0.111* (0.060) Youh wage gap squared (0.043) Youh wage gap on collecve agreemen jobs 0.157** (0.067) (0.121) 0.125* (0.066) 0.144* (0.073) Youh wage gap on collecve agreemen jobs squared (0.051) Youh wage gap on snglecompany agreemen jobs (0.057) Collecve agreemen when young 0.077*** (0.026) 0.055** (0.027) (0.028) 0.081* (0.044) Sngle-company agreemen when young (0.042) Inal real wage 0.073*** (0.013) 0.028** (0.011) 0.021* (0.012) 0.021* (0.012) Conemporaneous wage gap 1.236*** (0.204) 1.796*** (0.090) 1.825*** (0.090) 1.826*** (0.091) Male 0.158*** (0.021) 0.109*** (0.021) 0.154*** (0.024) 0.154*** (0.024) Age 0.014** (0.006) 0.028*** (0.006) 0.029*** (0.007) 0.030*** (0.007) Wage cenle dummes Yes No No No Inal ndusry dummes No No Yes Yes R-squared Sample sze 64,059 64,059 64,030 64,030 Noes: All regressons nclude a full se of year (13 caegores) dummes. Sandard errors are clusered by person and are presened n parenheses. *, ** and *** denoe sgnfcance a he 10%, 5% and 1% level, respecvely. 26

27 Table 4 Resuls for wage growh regressons usng ASHE-WERS daa Varable () () () Youh wage gap (0.133) (0.210) (0.125) Youh wage gap a frms ha negoaed over ranng 0.424* (0.235) 0.451* (0.265) (0.365) Youh wage gap a frms ha negoaed over pay (0.229) Frm negoaed over ranng when young (0.109) (0.110) (0.107) Frm negoaed over pay when young (0.156) Proporon of workers recevng ranng 0.312* (0.183) Inal real wage (0.017) (0.017) (0.017) Conemporaneous wage gap 1.844*** (0.345) 1.846*** (0.344) 1.859*** (0.343) Male 0.247** (0.112) 0.247** (0.113) 0.221* (0.115) Age (0.025) (0.025) (0.025) R-squared Sample sze 1,751 1,751 1,694 Noes: All regressons nclude a full se of year (13 caegores) dummes. Sandard errors are clusered by person and are presened n parenheses. *, ** and *** denoe sgnfcance a he 10%, 5% and 1% level, respecvely. 27

28 Appendx Table A1 Resuls for wage regressons usng dfferen ranng age ranges Varable Tranng age range Youh wage gap (0.119) (0.078) (0.049) 0.112* (0.058) Youh wage gap on collecve agreemen jobs 0.197*** (0.075) 0.153*** (0.054) 0.092* (0.054) 0.128* (0.065) R-squared Sample sze 23,291 39,084 51,952 64,059 Noes: All regressons nclude nal real wage, conemporaneous wage gap, male, age, collecve agreemen when young and a full se of year (13 caegores) dummes. In each column, he sample perod runs from he year afer he ranng perod ends o age 31. Sandard errors are clusered by person and are presened n parenheses. *, ** and *** denoe sgnfcance a he 10%, 5% and 1% level, respecvely. 28

29 Table A2 Esablshmen-level resuls for ranng Varable () Proporon recevng ranng () Average days of ranng Frm negoaes ranng 0.044*** (0.012) 0.818*** (0.106) Frm negoaes pay 0.047* (0.024) 0.780*** (0.223) Frm negoaes work hours 0.148*** (0.025) (0.233) Frm negoaes holday enlemens 0.087*** (0.023) 0.716*** (0.203) Frm negoaes pensons 0.032** (0.013) (0.120) Frm negoaes recrumen 0.103*** (0.013) 0.881*** (0.117) Frm negoaes grevance procedures (0.042) (0.368) Frm negoaes dscplnary procedures 0.084** (0.041) (0.364) Frm negoaes saffng plans (0.011) 0.267*** (0.099) Frm negoaes equal opporunes 0.073*** (0.018) 0.413*** (0.154) Frm negoaes healh and safey 0.094*** (0.020) 0.937*** (0.178) Frm negoaes performance apprasals 0.050*** (0.012) 0.935*** (0.108) R-squared Sample sze 4,272 3,916 Noes: All regressons nclude a full se of 3-dg ndusry dummes. *, ** and *** denoe sgnfcance a he 10%, 5% and 1% level, respecvely. 29