The Differential in Earnings Premia Between Academically and Vocationally Trained Males in the United Kingdom

Transcription

1 ISSN The Dfferental n Earnngs Prema Between Academcally and Vocatonally Traned Males n the Unted Kngdom Gavan Conlon June 2001

2 Publshed by Centre for the Economcs of Educaton London School of Economcs and Poltcal Scence Houghton Street London WC2A 2AE Gavan Conlon, submtted Aprl 2001 ISBN Indvdual copy prce: 5 The Centre for the Economcs of Educaton s an ndependent research centre funded by the Department of Educaton and Employment. The vew expressed n ths work are those of the authors and do not necessarly reflect the vews of the Department of Educaton and Employment. All errors and omssons reman the authors.

3 Executve Summary Ths paper estmates the rate of return assocated wth alternatve levels and types of qualfcaton n the Unted Kngdom. The analyss s restrcted to males aged between sxteen and ffty-nne and uses alternatve estmaton technques ncludng ordnary least squares, nstrumental varables and a Heckman Selecton approach. In addton, the work presented here utlses dfferent nformaton sources, both cross sectonal (Labour Force Surveys) and longtudnal (Natonal Chld Development Study). Ths s done n an attempt to compare the fndngs when usng alternatve nformaton sources whch contan markedly dfferent nformaton relatng to the personal characterstcs of the ndvduals beng analysed. It s found that there s a statstcally sgnfcant dfferental n the earnngs premum acheved by the academcally and vocatonally qualfed at the every level of qualfcaton wthn the Natonal Vocatonal Qualfcaton classfcaton of qualfcatons. Ths dfferental s nvarant to the method of estmaton and the data source. The dfferental approxmates 5% at Natonal Vocatonal Qualfcaton Level 1, rsng by an addtonal 5% at each successve level of qualfcaton. Ths mples that degree holders acheve a 20% premum over those males n possesson of vocatonal qualfcatons at an equvalent level of qualfcaton attanment (NVQ Level 4). Several reasons are suggested as to why there mght exst an earnngs dfferental between the academcally and vocatonally traned and ths s done to hghlght possbltes for future research. Specfcally, the earnngs dfferental mght be attrbutable to dfferences n the personal characterstcs of the academcally and vocatonally traned (.e. at a gven level of qualfcaton the academcally traned possess a greater level of nnate ablty compared to ther vocatonal counterparts (as measured by readng and mathematcal test scores at the age of seven n the NCDS)); the nature of qualfcaton provson n the Unted Kngdom resultng n the fact that atttudes towards and the treatment of the academcally and vocatonally traned s non-comparable n the Unted Kngdom labour market. Fnally, t suggested that the degree of monopsony power possessed by frms that hre ether academcally or vocatonally traned employees accounts for the dfferental n earnngs between the two types of qualfcaton holder. In other words, the dfferentals that exst mght be as a result of dfferences between frms, however t remans to be seen whether ths s the case or whether the dfferences are attrbutable to dfferences wthn frms.

4 The Dfferental n Earnngs Prema Between Academcally and Vocatonally Traned Males n the Unted Kngdom Gavan Conlon 1. Introducton 1 2. Methodology: Classfcaton of Qualfcatons 3 3. Methodology: NCDS: Ordnary Least Squares NCDS: nstrumental varables approach NCDS: Heckman selecton approach LFS: Ordnary Least Squares LFS: nstrumental varables 8 4. Presentaton of Results: Labour Force Survey Presentaton of results: Natonal Chld Development Study Conclusons 20 Appendx 23 References 26

5 Acknowledgements Gavan Conlon s a research offcer at the Centre for Economc Performance at the London School of Economcs and co-ordnator of the Centre for the Economcs of Educaton. The wrtng of ths paper was aded by the Economc Socal Research Councl and Nuffeld College, Oxford. The Centre for the Economcs of Educaton s an ndependent research centre funded by the Department of Educaton and Employment. The vew expressed n ths work are those of the authors and do not necessarly reflect the vews of the Department of Educaton and Employment. All errors and omssons reman the authors.

6 1. Introducton There have been numerous attempts to ascertan the return assocated wth addtonal years n educaton n the Unted Kngdom and the return to varous qualfcatons or to alternatve levels of hgher educaton. Many of the studes n ths area have as ther man focus the straghtforward rate of return to an addtonal year of schoolng or the earnngs premum assocated wth partcular levels of qualfcaton, such as the dfference n earnng power of an ndvdual holdng an undergraduate degree as opposed to GCE A levels (Blundell et al., 2000). However, there has been lttle analyss of the dfferental n earnngs premum assocated wth dfferent types of qualfcaton 1, ether academc - or vocatonal, holdng the level of qualfcaton constant 2. For nstance, what s the dfference n the earnngs premum to an undergraduate degree and a Cty & Gulds Full Technologcal Certfcate? It could be clamed that t s wholly expected that there would be dfferentals n the earnngs prema between the academcally and vocatonally traned due to the fact that there s self-selecton nto dfferent levels and types of qualfcaton. In other words, more able students undertake addtonal years of schoolng and qualfcatons compared to less able students, and n addton the more able students opt to undertake and complete academc qualfcatons as opposed to vocatonal qualfcatons and t s ths fact that accounts for the dfferental. Thus, there may exst two sources of endogenety, self-selecton nto hgher levels of qualfcaton and selecton nto alternatve types of qualfcaton 3. The mplcaton of these fndngs s that although there are dffcultes assocated wth the estmaton of returns at dfferent levels of qualfcaton due to sample selecton and endogenety bas, the ssue s not such an problem when consderng dfferentals n the rates of return across qualfcaton type. Selecton nto alternatve qualfcaton types does not appear to be determned by ablty characterstcs, but to a sgnfcant extent by famly background characterstcs. 1 Note that ths the decson to classfy qualfcatons accordng to the level at whch the qualfcaton s attaned and whether the qualfcaton s academc or vocatonal s somewhat, though not entrely, arbtrary. Qualfcatons are not smply academc or vocatonal. Often, there are areas of overlap between the two types of qualfcaton. The classfcaton of qualfcatons accordng to type used n ths analyss s presented n Tables 1 and 2. 2 There s dsagreement regardng the classfcaton of qualfcaton to partcular levels of Natonal Vocatonal Qualfcaton framework. In partcular, ths paper broadly classfes qualfcaton accordng to the tme taken to acheve the qualfcaton and the entry requrements needed to commence the qualfcaton. However, as noted by Dearden, McIntosh, Myck and Vgnoles (2000), there are nformatonal and conceptual dffcultes n determnng the tme taken to complete qualfcaton (e block or day release). As a result, the estmates of rates of return are senstve to the assumptons made n the analyss. In ths paper, the earnngs prema of the qualfed over those possessng no formally recognsed qualfcatons s presented rather than rates of return. 3 In prevous work, Conlon (2000), the determnants of both the level and type of qualfcaton attaned were explored usng a sample of males from the Natonal Chld Development Study. Specfcally, usng mathematcal and readng test scores at the age of seven, whch have been used repeatedly n the lterature (Dearden, 1999, Blundell et al., 1997) as a proxy for unobserved ablty, the relatonshp between qualfcaton attanment (level and type) and test scores was questoned. There was a clear and unambguous relatonshp between the hghest level of qualfcaton attaned and the mathematcal and readng test scores posted at the age of seven. However, t was found that mathematcal test scores at the age of seven play a mnor role n determnng whether academc or vocatonal qualfcatons are obtaned. Readng test scores at the age of seven are nfluental n the determnaton of whether a male obtans academc or vocatonal qualfcatons. However, once nteracton terms between the level of parental educaton and the test scores are ntroduced nto the model, the explanatory power of readng test scores on qualfcaton type s greatly reduced. The pror belef that the academcally qualfed are more able than the vocatonally traned and that ths dfferental n apttude can be measured (or llustrated) usng readng and mathematcal test scores remans hghly uncertan.

7 Despte the dffcultes assocated wth Ordnary Least Squares and Instrumental Varables approach to the estmaton of the wage equaton (whch wll be dscussed), ths paper wll use both methods to look at the rates of return to addtonal years of schoolng and the hghest level of academc and vocatonal qualfcatons 4. Ths paper uses nformaton from the Natonal Chld Development Study (5 th Follow Up) and the Labour Force Survey between 1992 and 1998 (both ndvdual cross sectons and pooled) to look at the rate of return to qualfcatons accordng to the Natonal Vocatonal Qualfcaton classfcaton of qualfcatons. Correcton terms wll be added to allow for the fact that only males n employment provde a response to ncome questons. It s found that the estmated return to an addtonal year of schoolng approxmates % accordng to OLS estmates. Ths fgure s n lne wth other studes n the Unted Kngdom (Harmon and Walker, 1995; Berlnsk, 2000; Dearden, McIntosh, Myck and Vgnoles, 2000). Smlarly, when usng Instrumental Varables n an attempt to correct for schoolng endogenety 5 and despte the well documented dffcultes assocated wth ths procedure, the rate of return to addtonal years schoolng s hgher than when nstruments are omtted. There s a statstcally sgnfcant dfferental n the earnngs premum between the academcally and vocatonally qualfed at the every level of qualfcaton. Ths dfferental occurs rrespectve of the method of estmaton. The dfferental approxmates 8-10% at lower levels of qualfcaton (NVQ Levels 1 and 2) and 12-18% at hgher levels of qualfcaton (NVQ Levels 3 and 4). When the data s segmented nto two smaller samples consstng of the hgh qualfed (NVQ level 3 and above) and the low qualfed (NVQ levels 1 and 2), the effect s to margnally reduce the dfferental n the rate of return between the academcally and vocatonally traned for the low qualfed and margnally ncrease the dfferental between qualfcaton types for the more hghly qualfed. In addton, for the sample of low qualfed males, the return to schoolng where no qualfcaton s obtaned ncreases and the return to explct qualfcatons s reduced n absolute terms compared to the entre sample. The converse results are obtaned for the restrcted sample of hgh-qualfed males lendng support to the argument regardng the use of Instrumental Varables by Card (1999). The dfferental n return between qualfcaton types at every level of qualfcaton s nvarant to the method of estmaton and data source. The paper s set out as follows: Secton 2 provdes the methodology behnd the classfcaton of qualfcatons; Secton 3 provdes methodology behnd the estmaton of the returns to academc and vocatonal qualfcatons; Secton 4 dscusses the results whle Secton 5 concludes. 4 Note that many ndvduals undertake and complete both academc and vocatonal qualfcatons. Ths paper analyses the determnants of the hghest qualfcaton attaned. As set out n the appendces, qualfcatons are classfed accordng to the Natonal Vocatonal Qualfcaton classfcaton of qualfcatons. If an ndvdual possesses an academc qualfcaton at NVQ level 4 (say) and a vocatonal qualfcaton at NVQ level 3 (say), then ths ndvdual s coded as possessng an academc qualfcaton as ther hghest qualfcaton for the purposes of ths analyss. If an ndvdual holds both academc and vocatonal qualfcatons at the same level of qualfcaton, then the fnal qualfcaton attaned determnes whether they are coded as beng academcally or vocatonally traned. Ths fnal te break crteron s based on the assumpton that ndvduals obtan qualfcatons of ncreasng dffculty over tme. It s clear that ths does not capture the full story snce many ndvduals acqure academc qualfcatons ntally and undertake addtonal vocatonal qualfcatons n order to do a specfc job. In other words, ndvduals undertake addtonal low levels of vocatonal qualfcaton (say) n order to gan the specfcty of tranng requred for a partcular job, even though ths qualfcaton may (or may not) be below ther exstng qualfcatons. See Conlon (2001) for addtonal analyss of the earnngs prema assocated wth combnatons of academc and vocatonal qualfcatons. 5 See Weale (1993), Heckman (1979), Harmon and Walker (1995), Blundell et al (1997) and Bound, Baker and Jaeger (1995) for dscusson of the ssue of endogenety and specfcaton error. 2

8 2 Methodology: Classfcaton of Qualfcatons In attemptng to compare the earnngs outcomes of the academcally and vocatonally traned, nformaton from the Labour Force Surveys between 1993 and 1998 s utlsed, whch s one of the most detaled nformaton sources combnng formal qualfcatons and economc actvty. The most dffcult ssues assocated wth ths topc n partcular s the codng of the data to reflect comparablty of qualfcatons wthn partcular levels of qualfcaton as well as comparablty across tme. Intally, t was questoned whether to adopt the nomnal classfcatons presented n offcal government statstcal publcatons and to use these classfcatons as a backward lookng benchmark. In other words, we questoned the concept of equvalence between academc and vocatonal qualfcatons. It s clear that t s mpossble to completely classfy qualfcatons accordng to ether the level or the type attaned, whch are by ther very nature dffcult to compare, however, t was decded to consder the specfc entry requrements and the tme taken to compete the qualfcaton n queston as a bass for equvalence. For some qualfcatons, t s clear what the entry requrements are, however, there are stll many stuatons n whch the data does not allow us to be more strngent n our classfcaton. For nstance, for a person who has obtaned a hgher degree, t s clear that the entry requrement wll be an undergraduate degree. However, there s no nformaton avalable regardng the grade of the undergraduate degree requred or the qualty of the nsttuton nvolved. Therefore, we are forced n many respects to adopt the lowest common denomnator n terms of entry requrement. The most problematc ssue assocated wth the classfcaton of vocatonal qualfcatons (apart from the fundamental defnton of what exactly consttutes a vocatonal qualfcaton) s the fact that hstorcally, there has been a complete lack of a centralsed structure concerned wth the award and valdaton of vocatonal qualfcatons (Cruz-Castro, 2000). It s only snce the ntroducton of the Natonal Councl for Vocatonal Qualfcatons that there has been a unfed approach to the provson and award of vocatonal tranng and qualfcatons 6. Therefore, for many of the qualfcatons that we are consderng, there are organsaton specfc entry requrements. The entry requrement for most courses, both academc and vocatonal, s dssmlar between nsttutons. It s mportant to stress ths pont more fully. There s a common belef that the varaton n entry requrements s restrcted to the provson of vocatonal qualfcatons. It s more the case that entry requrements for academc programmes are equally dssmlar across nsttutons, especally as the level of qualfcaton ncreases. In the case of the provson of many academc and vocatonal qualfcatons, the valdatng authorty makes no decson regardng the qualfcatons of the potental students at the pont of entry but smply valdate (or not) the qualfcaton at the pont of ext. The classfcaton of qualfcatons for ths analyss s presented n Tables 1 and 2. 6 Untl recently, there were over 60 organsatons capable of awardng vocatonal qualfcatons. 3

9 Table 1: Descrpton of Academc Qualfcatons by NVQ Equvalent Labour Force Survey 1991 and Natonal Chld Development Study (5 th Follow Up) NCDS Descrpton of Academc Qualfcatons NVQ NVQ LFS 1991 Descrpton of Academc Qualfcatons Unversty or CNAA Hgher Degree (MSc, 5 5 Hgher Degree PhD) Unversty or CNAA Post Graduate Dploma 5 Unversty or CNAA Frst Degree ncludng 4 4 Frst Degree (B.Ed) 4 Other Degree - Unversty or CNAA Dploma or Certfcate ncludng Dp HE and TTCC 4 4 Teachng: Hgher Educaton, Secondary Level or Prmary Level Certfcate (NOT CNAA VALIDATED) 4 Full professonal qualfcaton (Membershp 4 awarded by professonal nsttuton) Part of professonal qualfcatons eg: Part I 4 of two part course More than 1 GCE 'A' Level 3 3 GCE A Level or Equvalent 1 GCE 'A' Level 2 Scottsh CSYS 3 Scottsh Hgher Grade 3 Scottsh Standard Grades grades Scottsh 'O' grade passes or grades A-C 2 GCE O Level passes or grades A-C 2 2 GCE O Level or Equvalent GCSE grades A-C 2 CSE grade 1 2 CSE grade CSE (Not Grade 1) No Qualfcatons 0 0 No Qualfcatons 4

10 Table 2: Descrpton of Vocatonal Qualfcatons by NVQ Equvalent Labour Force Survey 1991 and Natonal Chld Development Study (5 th Follow Up) NCDS Descrpton of Hghest Qualfcatons NVQ NVQ LFS 1991 Descrpton of Qualfcatons Nursng qualfcatons - Nursng NNEB) 4 4 Nursng Qualfcaton BTEC HNC or HND 4 4 BTEC HNC or HND SHNC/SHND 4 ONC/HND (or SNC/SND) 3 3 BTEC ONC or OND Natonal General Certfcate or Dploma 3 JIB/NJC or other Craft/Techncan Cert. 2 Insgna Award n Technology (CGIA) 2 C&G Full Technologcal Cert. (FTC) 4 C&G Advanced/Fnal/Part II or III 3 C&G Craft /Intermed. /Ordnary/ Part I 2 2 Cty and Gulds (ALL) C&G Operatve 1 C&G Can't say whch 2 C&G Other 2 Trade Apprentceshp (No Quals) 2 2 Trade Apprentceshp (No Quals) RSA - Stage 3 3 RSA - Stage 2 2 RSA Stage 1 1 Other Techncal or Busness Qualfcatons ncl.hgv, PSV, etc 1 1 Other Professonal / Vocatonal Qualfcatons Any other qualfcaton 1,2, Youth Tranng Certfcate No qualfcatons 0 0 No Qualfcatons 3. Methodology: NCDS: Ordnary Least Squares The standard Ordnary Least Squares estmatng equaton s as follows: ln 10 ' ' ' ( ω ) β k NVQ k + ζ AB + ρ FAM + ω OTH + ε = k = 0 where ln( ω ) s the natural log of hourly wages for ndvdual, NVQ k I represents the level of qualfcaton obtaned by ndvdual for k=1,2, Ths specfcaton hghlghted s due to the fact that there are 10 categores of qualfcaton rangng from NVQ level 1 to NVQ Level 5, both academc and vocatonal, n the Natonal Vocatonal Qualfcaton framework 7. s coded 1 f ndvdual has obtaned NVQ a vocatonal ki qualfcaton at level k as ther hghest qualfcaton and 0 otherwse. Therefore, the β k coeffcents provde the earnngs premum assocated wth qualfcaton level k. In ths specfcaton of the model, represents the earnngs premum assocated wth an addtonal year of schoolng β 0 where no qualfcaton has been obtaned. 7 Ths paper makes no attempt to estmate the returns to gven subject areas. 5

11 AB s a vector of ablty varables consstng of Mathematcal test scores at the age of seven and eleven Readng test scores at the age of seven and eleven FAM s a vector of personal background varables consstng of Severe Fnancal Dffcultes Aged 11 Severe Fnancal Dffcultes Aged 16 Number of Sblngs Age at whch father left full-tme educaton Age at whch mother left full-tme educaton Accommodaton Aged 16 School Aged 16 Regon Aged 16 Father s Socal Class Aged 16 Father s Employment Status Aged 16 Whether Mother worked Aged 16 OTH s a vector of other job characterstcs relatng to Unon membershp Temporary of Permanent Contract Frm Sze Publc / Prvate Sector Industry 3.1 NCDS: nstrumental varables approach The standard nstrumental varables approach requres the two-stage estmaton of the followng equatons ln where 10 ' ' ( ω ) α SCHOOL + β k NVQ k + ζ AB + ω OTH + ε = SCHOOL k = 1 * = ' γ FAM + ν * SCHOOL s a normally dstrbuted latent varable such that * j µ j 1 SCHOOL µ where the cuts µ j are estmated SCHOOL = f j from the model. As before, ln ( ω ) s the natural logarthm of hourly wages, SCHOOL NVQ represents the years of schoolng obtaned by ndvdual, k for k=1,2,...10 corresponds to the 10 levels of qualfcaton attanable by ndvdual n the NVQ classfcaton of qualfcatons and AB s a vector of ablty varables as prevously dscussed 8. In ths case, the requrements of the explanatory vector FAM s that t contans 8 In partcular, the varables used as nstruments are whether the cohort member experenced severe fnancal dffcultes aged 11, severe fnancal dffcultes aged 16, number of sblngs, age at whch father left full-tme educaton, age at whch mother left full-tme educaton, accommodaton aged 16, school aged 16, regon aged 16, father s socal class aged 16, father s employment status aged 16 and whether the cohort member s mother worked when cohort member was aged 16. As wth the majorty of other studes, whch use nstruments not of a 6

12 varables that can be used to explan the determnants of schoolng but are not contaned n AB or OTH. 3.2 NCDS: Heckman selecton approach In order to mnmze the effect of the bas assocated wth a standard Ordnary Least Squares estmaton the model s extended and the followng equaton s estmated 9 : where NVQ k s as before, λ EMPNOT corresponds to the nverse Mlls rato or employment selecton adjustment term, λ SCHOOL corresponds to schoolng level selecton adjustment term, and λ corresponds to the qualfcaton type selecton adjustment term 10. TYPE ln ' ' ( ω ) β k NVQ k + + ζ AB + ω OTH 3.3 LFS: Ordnary Least Squares The standard Ordnary Least Squares estmatng equaton s as follows: ln 10 = = 0 ' ( ω ) β NVQ + ω Z + 10 = k= 0 k + ξλ + ϑλ + ϕλ + ε SCHOOL TYPE k EMPNOT ε where ln( ω ) s the natural log of hourly wages for ndvdual, NVQ k represents the level I of qualfcaton obtaned by ndvdual for k=1,2,..10. NVQ k s coded 1 f ndvdual I has obtaned a vocatonal qualfcaton at level k as ther hghest qualfcaton and 0 otherwse. Therefore, the β k coeffcents provde the earnngs premum assocated wth qualfcaton level k. In ths specfcaton, β 0 represents the return assocated wth an addtonal year of schoolng where no qualfcaton has been obtaned. Z s a vector of varables consstng of Accommodaton Detals Martal Status polcy nature (such as the change n the mnmum school leavng age n the Unted Kngdom), these nstruments are weak at best. 9 Full detals are provded n the appendces. Note that for the selectvty terms to be vald, each of the equatons for SCHOOL, TYPE and EMPNOT should be ndependent and t s clear that the ncluson of the selecton term nto the type of qualfcaton attaned s restrctve. In most standard analyses (Blundell et al., 1997), no attempt s made to correct for selecton nto the type of qualfcaton attaned, as selecton nto a partcular qualfcaton route s not consdered to be problematc. In alternatve specfcatons where the selecton term nto the type of qualfcaton attaned s suppressed, the results are not substantally dfferent from those presented here, ndcatng that ths mght be ndeed the case. 10 Note that ths specfcaton of the model depends on the ndependence of the selecton terms (level and type of qualfcaton attaned). In the results secton, the estmates presented ndcate that t s n fact the case that selecton nto the type of qualfcaton s not a serous methodologcal ssue warrantng substantal addtonal analyss, whereas, selecton nto the level of qualfcaton and employment remans so. 7

13 Number of Dependent Chldren under 16 Employment/ Unemployment/ Inactvty Status of Other Adults n Household Years Snce Leavng Full-Tme Educaton Regon of Resdence Industry Unon Membershp Temporary or Permanent Contract Frm Sze Publc / Prvate Sector 3.4 LFS: nstrumental varables In the case of the Labour Force Survey, due to the lack of nformaton regardng an ndvdual s famly background and personal characterstcs; as before the model requres the two stage estmaton of ln 10 ' ( ω ) γ ' SCHOOL + β k NVQ k + ω Z + ε = k = 1 SCHOOL = γ + ε * ' Q where Z s a vector of personal characterstcs. Agan, the nstruments used must determne the decson to undertake addtonal years of schoolng of qualfcatons but have no explanatory power n relaton to the earnngs equaton. In the case of the Labour Force Survey, the nstruments that are used are brth month (Angrst and Krueger, 1991) and Mnmum School Leavng Age (Harmon and Walker, 1995). 4. Presentaton of Results: Labour Force Survey The estmates of the rate of return to schoolng and qualfcatons obtaned for year old men n the Unted Kngdom usng pooled data from 1993 to 1998 are presented n Table 3. For each year the earnngs premum assocated wth each addtonal year of schoolng s estmated n solaton. In addton, a second specfcaton of the model s estmated consstng of the resdual premum assocated wth an addtonal year of schoolng (e a year of schoolng where no explct qualfcaton s obtaned) and the premum assocated wth each level of qualfcaton n the Natonal Vocatonal Qualfcaton classfcaton of qualfcatons. In both models, the specfcatons nclude addtonal explanatory terms such as unon membershp or coverage and other job related characterstcs that may determne earnngs. The quarterly cross sectonal data sets for the sx years n queston have been pooled and the returns have been estmated wth the ncluson of yearly and seasonal dummes 11. The OLS estmate of the gross rate of return to a year of schoolng over the entre perod approxmates 6.7%, and s n lne wth many other OLS estmates n the Unted Kngdom (Harmon and Walker, 1995; Dearden et al, 2000 and Berlnsk, 2000). The return s relatvely stable across years and the estmate from any partcular year s never more than 11 The numbers of ndvduals n each year contaned n the pooled sample are unequal. In partcular, due to changes n the Labour Force Survey samplng procedure, there are approxmately twce as many respondents n 1997 and 1998 compared to prevous years. 8

14 1.7 percentage ponts dfferent from any other year s estmate. OLS estmates for each of the years are presented n Table 2. When the level of qualfcaton s explctly ntroduced a meanngful dstncton n the earnngs performance between the academcally and vocatonally qualfed s hghlghted. As would be expected, for both the holders of academc and vocatonal qualfcatons, there s an ncreased return as the level of qualfcaton ncreases. In partcular, accordng to OLS specfcatons, the academcally qualfed at NVQ level 1 acheve a premum of 14.1% over the unqualfed. Table 3: OLS and Instrumental Varables Estmates of the Returns to Educaton 12 : Pooled Labour Force Surveys : Males OLS OLS IV IV Schoolng.067 (.001).020 (.001).088 (.012).045 (.011) Experence.081 (.002).077 (.001).050 (.002).071 (.002) (Experence) 2 / (.025) (.024) (.029) (.029) Academc Level (.011).127 (.012) Academc Level (.011).255 (.012) Academc Level (.015).458 (.015) Academc Level (.013).588 (.012) Academc Level (.018).652 (.018) Vocatonal Level (.011).044 (.012) Vocatonal Level (.010).150 (.010) Vocatonal Level (.012).243 (.013) Vocatonal Level (.013).403 (.015) Vocatonal Level (.081).348 (.083) Sample Sze Adjusted R At NVQ level 2 (5 GCSE A*-Cs), the premum ncreases to 26.9%, whle at degree level, unsurprsngly, the premum over the unqualfed stands at 54%. The outcome for the vocatonally qualfed follows a smlar ncreasng relatonshp as the academcally qualfed. Males possessng academc qualfcatons at NVQ level 1 earn a 5.2% premum over the unqualfed. At NVQ Level 2 (Trade Apprentce/Cty and Gulds Craft), the premum over 12 Standard errors are presented n parenthess. The model specfcatons nclude an ndvdual s ethnc characterstcs, martal status, accommodaton detals, number of dependent chldren under the age of 16 and the economc actvty of other members of the household. Job characterstcs have also been ncluded such as unon membershp, whether the job s full-tme or part-tme, temporary or permanent, n the publc or prvate sector and the sze of the frm the ndvdual s workng n. Yearly dummes have also been ncluded. 9

15 males holdng no qualfcatons stands at 16.4%, whereas at NVQ Level 4 (Hgher Natonal Certfcates and Dplomas) the premum acheved s margnally less that 42%. In addton, note that the resdual return to a year of schoolng when no qualfcatons are acheved approxmates 2%. However, there are mportant dfferentals n earnngs between the academcally and vocatonally qualfed at every level of qualfcaton. At the lower levels of qualfcaton, males wth NVQ level 1 academc qualfcatons acheve an 8.9% premum over ther vocatonal counterparts. At NVQ level 2, the dfferental between the academcally and vocatonally qualfed s 10.5%. As the level of qualfcaton rses, there s an ncrease n the premum pad to the academcally qualfed, but the ncrease s not extreme. At NVQ levels 4 and 5, the dfferental rses to 12.6% and 13% respectvely. The greatest dfferental arses at NVQ level 3, standng at 16.9%, however, care should be exercsed wth ths estmate as the sample sze assocated wth those completng GCE A Levels as ther hghest qualfcaton s substantally smaller than the sample szes assocated wth other qualfcaton levels. It s rare that ndvduals state GCE A levels as beng ther hghest level of qualfcaton, as t s generally the case that those undertakng ths qualfcaton do so n order to progress to degree level educaton. As would be expected, the nstrumental varable approach produces estmates for the return to schoolng substantally hgher than those produced usng an OLS estmator. The pooled estmates are presented n Table 5. In partcular, the return to an addtonal year of schoolng approxmates 8.8% usng pooled estmates compared to the estmate of 6.7% usng an OLS method of estmaton. When explct levels and types of qualfcaton are ntroduced nto the model, the resdual rate of return to a year of schoolng where no qualfcaton s attaned s 4.5%, whch s more than double the OLS estmate. The returns to explct levels and types of qualfcaton are smlar under IV as under OLS. Compared to a male possessng no formal qualfcatons, a male holdng an academc qualfcaton at NVQ level 2 earns a 25.5% wage premum over a male possessng no formal qualfcatons, rsng to 58.8% at NVQ level 4. On the vocatonal sde, a male holdng a trade apprentceshp (or equvalent) acheves a 15% wage premum over the unqualfed, rsng to 40.3% at NVQ level 4. Accordng to the IV estmates of the dfferental n the rate of return between the academcally and vocatonally traned, usng pooled data between 1993 and 1998 stands at 8.3% at NVQ level 1, 10.5% at NVQ level 2, rsng to 18.5% at NVQ level 4. Ths compares to fgures of 8.9%, 10.5% and 12.6% at the correspondng levels of qualfcaton accordng to OLS estmates. The most mportant pont to note s that rrespectve of the method of estmaton, usng nformaton from the Labour Force Survey, the dfferental n returns between the academcally and vocatonally qualfed remans and s robust to the method of estmaton. The only effect of alternatve methods of estmaton s to produce dfferent earnngs prema (n absolute terms) assocated wth partcular levels of qualfcaton but has lttle effect on the dfferental between dfferent types of qualfcaton at gven levels of qualfcaton. At NVQ Levels 1 and 2, the dfferental approxmates 8-10% whle ths dfferental rses to 12-18% at NVQ Level 4. 10

16 Table 4: Ordnary Least Squares Estmates of the Returns to Educaton 13 Labour Force Surveys : Males Schoolng.064 (.003) (.004).064 (.003).017 (.004).061 (.003).019 (.004).078 (.003).024 (.004) 066 (.002).026 (.002).066 (.002).016 (.003) Experence.076 (.005).070 (.005).070 (.005).066 (.005).098 (.006) (.005).077 (.005).083 (.003).079 (.003).080 (.004).074 (.003) (Experence) (.074) (.071) (.073) (.070) (.080) (.078) (.070) (.066) (.048) (.047) (.052) (.050) Academc Level (.039).157 (.037).117 (.041).150 (.030).115 (.021).170 (.023) Academc Level (.029).262 (.029).239 (.033).364 (.034).264 (.024).295 (.026) Academc Level (.042).408 (.041).461 (.048).443 (.040).376 (.030).479 (.032) Academc Level (.035).564 (.035).500 (.040).588 (.035).477 (.025).607 (.026) Academc Level (.053).646 (.051).576 (.057).627 (.051).490 (.035).634 (.036) Vocatonal Level (.034).073 (.032).085 (.036).071 (.031).017 (.022).063 (.024) Vocatonal Level (.026).219 (.027).166 (.032).162 (.026).150 (.020).165 (.022) Vocatonal Level (.035).256 (.034).272 (.039).233 (.032).272 (.023).297 (.024) Vocatonal Level (.038).432 (.036).408 (.040).425 (.035).376 (.026).440 (.027) Vocatonal Level (.154).506 (.107) Sample Sze 3387 Adjusted R Standard errors are presented n parenthess. The model specfcatons nclude an ndvdual s ethnc characterstcs, martal status, accommodaton detals, number of dependent chldren under the age of 16 and the economc actvty of other members of the household. Job characterstcs have also been ncluded such as unon membershp, whether the job s fulltme or part tme, temporary or permanent, n the publc or prvate sector and the sze of the frm the ndvdual s workng n.

17 It has been argued that one of the reasons why a two stage least squares approach provdes hgher estmates of the return to schoolng than OLS estmates, s that returns to educaton vary across groups (Card, 1999). The people that are affected by nstruments are those very people wth hgher returns to educaton. Snce the prevous analyses focus on the average return to schoolng wthout formal qualfcatons and the average return to specfc qualfcatons, n ths secton, the entre sample of males has been dvded nto two subsamples consstng of males wth hgh qualfcatons (NVQ Levels 3,4 and 5) and those wth a low level of qualfcaton (NVQ Levels 1 and 2). The IV estmates of the returns to specfc levels and types of qualfcaton are presented n Table 5. Table 5: Instrumental Varables Estmates of the Returns to Educaton 14 Labour Force Surveys : Males Schoolng.054 (.037) (.035).044 (.034).043 (.031).072 (.037).060 (.035).070 (.020).023 (.026).079 (.022).052 (.021).069 (.033).059 (.031) Experence.051 (.006).069 (.006).046 (.006).057 (.006).077 (.006).085 (.006).062 (.005).071 (.006).058 (.004).072 (.004).053 (.004).075 (.004) (Experence) (.084) (.082) (.083) (.080) (.090) (.087) (.075) (.081) (.055) (.054) (.057) (.056) Academc Level (.039).120 (.038).106 (.042).146 (.031).093 (.022).174 (.024) Academc Level (.032).232 (.031).219 (.035).361 (.037).241 (.026).309 (.028) Academc Level (.043).399 (.042).480 (.048).479 (.041).396 (.030).513 (.033) Academc Level (.034).584 (.033).537 (.038).674 (.031).533 (.025).675 (.028) Academc Level (.048).678 (.047).618 (.056).753 (.044).559 (.035).729 (.039) Vocatonal Level (.034).059 (.047).064 (.038).074 (.032) (.023).072 (.024) Vocatonal Level (.027).189 (.028).151 (.033).148 (.029).123 (.021).172 (.023) Vocatonal Level (.037).198 (.039).215 (.047).231 (.033).229 (.026).305 (.027) Vocatonal Level (.041).417 (.036).382 (.045).427 (.040).348 (.029).465 (.031) Vocatonal Level (.154).505 (.112) Sample Sze 3385 Adjusted R Standard errors are presented n parenthess. The model specfcatons nclude an ndvdual s ethnc characterstcs, martal status, accommodaton detals, number of dependent chldren under the age of 16 and the economc actvty of other members of the household. Job characterstcs have also been ncluded such as unon membershp, whether the job s full-tme or part tme, temporary or permanent, n the publc or prvate sector and the sze of the frm the ndvdual s workng n.

18 Turnng to Table 6, where the returns for those holdng low qualfcatons are presented usng pooled data, t s clear agan that the estmated returns under OLS are less than under the IV approach. In partcular the estmate of the return to schoolng where no qualfcaton has been attaned approxmates 4.5% accordng to OLS estmates. However, the IV approach ndcates that the return to a year of schoolng where no qualfcaton has been attaned s around 5.5% over the correspondng perod. Ths s hgher than the estmates produced accordng to ether model when lookng at the entre sample. The correspondng fgures when analysng the complete sample of males were 2.0% and 4.5% accordng to the OLS and IV estmates respectvely. It s clear that the return to schoolng wth no qualfcatons s greater for the low qualfcaton sample when analysed separately than when the total populaton of males s analysed as a whole. Table 6: OLS and IV Estmates of the Returns to Educaton 15 : Pooled Labour Force Surveys : Males Low and Hgh Qualfcaton Holders OLS OLS IV IV Schoolng.057 (.002).045 (.002).039 (.001).011 (.001).069 (.015).055 (.015).022 (.012).023 (.011) Experence.075 (.002).076 (.002).080 (.003).070 (.002).063 (.002).075 (.002).053 (.003).068 (.003) (Experence) (.029) (.028) (.039) (.037) (.033) (.033) (.043) (.041) Academc Level (.011).115 (.013) Academc Level (.011).251 (.013) Academc Level (.016).485 (.016) Academc Level (.014).625 (.014) Academc Level (.019).707 (.019) Vocatonal Level (.011).049 (.012) Vocatonal Level (.010).152 (.011) Vocatonal Level (.012).272 (.014) Vocatonal Level (.013).438 (.016) Vocatonal Level (.082).435 (.086) Sample Sze Adjusted R Standard errors are presented n parenthess. The model specfcatons nclude an ndvdual s ethnc characterstcs, martal status, accommodaton detals, number of dependent chldren under the age of 16 and the economc actvty of other members of the household. Job characterstcs have also been ncluded such as unon membershp, whether the job s full-tme or part tme, temporary or permanent, n the publc or prvate sector and the sze of the frm the ndvdual s workng n.

19 The returns to specfc qualfcatons are lower for the low qualfed n the analyss of the restrcted sample as opposed to entre populaton. Ths s the case under both methods of estmaton. Vewng the restrcted sample, the return to holdng academc and vocatonal qualfcatons at NVQ Level 2 are 23.2% and 15.0% respectvely under OLS and 25.1% and 15.2% under IV. When lookng at the entre sample the correspondng fgures are 27.0% and 16.4% under OLS and 25.5% and 15.3% under IV. The mplcaton s that when analysng the entre populaton of males that, for the low qualfed, there s an upward bas assocated wth the return to recognsed qualfcatons and a downward bas assocated wth the return assocated wth an addtonal year of schoolng where no qualfcaton s attaned. The dfferental n rate of return to specfc qualfcatons between the academcally and vocatonally qualfed s margnally smaller when lookng at the sample of low qualfed males compared to the entre sample of males. The dfferental n returns at NVQ levels 1 and 2 were 7.7% and 8.2% respectvely accordng to the OLS estmaton and 6.6% and 9.9% accordng to the IV procedure when lookng at the sample of low qualfed males. Ths compares to dfferentals of 9% and 10.6% when lookng at the entre male sample at NVQ levels 1 and 2 respectvely accordng to OLS estmates; and 8.3% and 10.2% respectvely accordng to IV estmates. Lookng agan at Table 6, where the returns correspondng to the hghly qualfed are estmated, unsurprsngly, the rate of return assocated wth a year of schoolng where no qualfcaton s attaned s lower than that of the entre sample. In partcular, accordng to the OLS estmaton, the return approxmates 1.1% per year of addtonal schoolng and 2.3% accordng to the IV estmate. Comparng the returns assocated wth partcular levels of formal qualfcaton, the returns are hgher n absolute terms at every level of qualfcaton under both OLS and IV than the estmates produced from the combned sample. The dfferentals n the earnngs prema by qualfcaton accordng to the sample chosen are presented below n Table 7. When analysng the entre populaton of males, for the hghly qualfed, there s a downward bas assocated wth the estmate of the returns to recognsed qualfcatons and an upward bas assocated wth obtanng an addtonal year of schoolng where no formal qualfcatons are attaned. The dfferental n return between the academcally and vocatonally qualfed s margnally greater when vewng the restrcted sample than when lookng at the entre sample. The dfferental n earnngs prema at NVQ Levels 3, 4 and 5 between the academcally and vocatonally traned stands at 19%, 15.5% and 18.9% respectvely accordng to OLS estmates and 21.3%, 18.7% and 27.2% accordng to IV estmates. Ths compares to dfferentals of 16.9%, 12.6% and 12.6% when lookng at the entre male sample at NVQ levels 3, 4 and 5 respectvely accordng to OLS estmates and 21.5%, 18.6% and 30.6% respectvely accordng to IV estmates. 14

20 Table 7: Dfferental n the Estmates of the Returns to Schoolng Accordng to Sample: Pooled Labour Force Surveys : Males Year of schoolng where no Qualfcaton s obtaned OLS IV OLS IV Low Qualfcaton 16 Hgh Qualfcaton Academc Level Academc Level Academc Level Academc Level Academc Level Vocatonal Level Vocatonal Level Vocatonal Level Vocatonal Level Vocatonal Level Presentaton of results: natonal chld development study In addton to lookng at the dfferentals between the academcally and vocatonally qualfed usng the Labour Force Surveys, a smlar methodology as before has been appled to the NCDS, the reason beng the avalablty of addtonal nformaton relatng to personal characterstcs and famly background of the cohort member. However, n usng the NCDS data, we are restrcted to 33-year-old males, a much smaller sample and a lack of precseness regardng certan employment characterstcs. Note that several dfferent specfcatons of the model used to analyse the Natonal Chld Development Study are presented. The 5 specfcatons nclude a raw analyss where explct qualfcatons, ablty and famly background characterstcs are omtted n order to ascertan the raw return to an addtonal year of schoolng. The fnal specfcaton makes use of the explct nformaton contaned n the survey relatng to qualfcaton levels, ablty and relevant famly background characterstcs. Lookng at the alternatve specfcatons of an ordnary Least Squares approach presented n Table 8, the smple return to an addtonal year of schoolng s estmated at 3.8% (specfcaton 1), whch s just over half the OLS estmate produced usng the LFS. Upon the ntroducton of specfc qualfcatons (specfcaton 2), the resdual return to an addtonal year of schoolng where no qualfcatons are obtaned s estmated at 1.4%, margnally less than the Labour Force Survey pooled OLS estmate of 2.0%. 16 Estmated rate of return for restrcted sample mnus estmated rate of return for entre sample 15

21 Table 8: OLS Estmates of the Returns to Educaton: Males Natonal Chld Development Survey (5 th Follow Up) 17 OLS (1) OLS (2) OLS (3) OLS (4) OLS (5) Schoolng.038 (.002).014 (.003).013 (.003).033 (.002).012 (.003) Academc Level (.035).039 (.035).038 (.035) Academc Level (.032).201 (.032).190 (.032) Academc Level (.056).396 (.056).386 (.056) Academc Level (.035).404 (.036).387 (.037) Academc Level (.047).332 (.048).310 (.048) Vocatonal Level (.038).053 (.038).045 (.038) Vocatonal Level (.032).089 (.032).084 (.032) Vocatonal Level (.033).146 (.033).142 (.034) Vocatonal Level (.038).293 (.039).280 (.039) Maths Test Age 7 2 nd Quartle.037 (.020).035 (.020).039 (.020) 3 rd Quartle.061 (.021).069 (.022).061 (.021) Top Quartle.086 (.023).109 (.024).086 (.023) Readng Test Age 7 2 nd Quartle.021 (.021).037 (.021).021 (.021) 3 rd Quartle.019 (.022).051 (.023).018 (.022) Top Quartle.032 (.023).107 (.023).027 (.023) Age Father Left FTE Years old Years old 17+ Years old.053 (.021).040 (.026).079 (.025) Father s Socal Class Sklled (Man/N.M.) (.015) Sem Sklled (Man/N.M.) (.024) Unsklled (.042) Sample Sze Adjusted R Standard errors are presented n parenthess. The model specfcatons nclude an ndvdual s ethnc characterstcs, martal status, accommodaton detals, number of dependent chldren under the age of 16 and the economc actvty of other members of the household. Job characterstcs have also been ncluded such as unon membershp, whether the job s full-tme or part tme, temporary or permanent, n the publc or prvate sector and the sze of the frm the ndvdual s workng n. 16

22 Accordng to specfcaton 3, where the scores obtaned n mathematcal and readng tests at the age of seven are explctly ntroduced as proxes for unobserved ablty, the return of an addtonal year of schoolng wthout any recognsed qualfcaton s estmated at 1.3%. The dscrepancy between the LFS and NCDS estmates of the raw returns to addtonal schoolng s explcable due to the addtonal nformaton relatng to qualfcatons contaned n the NCDS and the fact that many cohort members who are consdered to possess qualfcatons (PGV, HGV lcenses) accordng to the NCDS classfcaton of qualfcatons are classfed as possessng no formal qualfcatons accordng to the LFS. Other general ponts of note are the fact that the OLS estmates ndcate that the return assocated wth beng postoned n the second quartle n the mathematcal test at the age of seven s 3-4% greater than a male fnshng n the bottom quartle. Ths fgure rses to 8-11% for males fnshng n the top mathematcs test quartle at the age of seven. A smlar pattern s dsplayed when lookng at readng test scores at the age of seven. The combned effect of ncludng mathematcal and readng test scores at the age of seven s to reduce the returns assocated at every level of qualfcaton, addng to the belef that standard rate of return analyses whch do not control for nnate ablty produce estmates of the return to schoolng or qualfcatons that are based. In the fnal specfcaton, explanatory varables nclude readng and mathematcal test scores at the age of seven, explct levels of qualfcaton, famly background characterstcs, ncludng the age at whch the cohort member s parents left full-tme educaton, whether the famly experenced dffcultes, accommodaton status and regonal dummes. In the fnal specfcaton, the return to an addtonal year of schoolng where no qualfcaton was obtaned was estmated at 1.2%. Wth reference to the returns to explct qualfcatons the results from the NCDS do not perfectly concde wth those produced usng LFS data. In partcular, the returns assocated wth low levels of qualfcaton, both academc and vocatonal, are substantally lower usng NCDS data compared to the OLS estmates usng LFS data. At NVQ Level 1, the return assocated wth an academc qualfcaton s a mere 3.8% compared to the LFS estmate of over 14% and the return to vocatonal qualfcatons at the same level stands at 4.5%. At NVQ level 2, the returns assocated wth academc qualfcatons rse to 19%, agan about 7% lower than LFS estmates. On the vocatonal front, there appears to be a large dvergence n the estmated return, wth the vocatonally qualfed at NVQ Level 2 postng a return of 8.4%, compared to 16% accordng to LFS estmates. Upon reachng NVQ level 4, the premum assocated wth academc qualfcaton approxmates 38.7%, whereas the correspondng fgure for the vocatonally traned s 28.0%. These estmates are consderably lower than estmates based on the LFS. However, at most levels of qualfcaton, the dfferental between the academcally and vocatonally qualfed compares favourably wth LFS estmates. At NVQ level 1, the vocatonally qualfed acheve a margnal premum of 0.7% over ther academc counterparts, whereas at NVQ level 2, the academcally qualfed acheve an 11.6% premum over the vocatonally traned. At undergraduate degree level, the dfferental between the academcally and vocatonally traned s exactly 10.7%. Whle some of these fndngs appear to be n contrast to the estmates produced usng LFS nformaton, t must be remembered that the categorsaton of qualfcatons across data sets s not equvalent. In partcular, accordng to the LFS anyone possessng a Cty & Gulds qualfcaton, rrespectve of the level, s consdered to be qualfed at NVQ level 2, whle a clear dstncton can be made usng the NCDS. In fact almost 50% of males wth NVQ level 2 qualfcatons accordng to the LFS are consdered to acheve hgher levels of qualfcaton attanment accordng to the NCDS. Ths would have the effect that models based on the LFS may produce a hgher estmate of the dfferental n return between the academcally and vocatonally qualfed at 17

23 NVQ level 2 than the NCDS and lower estmate of the dfferental at hgher levels of qualfcaton than the NCDS 18. In Table 9, the detals of the estmates of returns to academc and vocatonal qualfcatons usng the Instrumental Varables and Heckman Selecton approaches are presented. Accordng to IV estmates, the smple return to a year of schoolng s estmated at 7.3%, where explct qualfcatons, readng and mathematcal scores at the age of seven are omtted. Ths fgure s almost double the estmate produced under a smlar specfcaton usng OLS. Furthermore, even when recognsed qualfcatons are ntroduced nto the model, the estmate of the rate of return to schoolng, when no qualfcaton s attaned only falls to 3.7%. Upon the ntroducton of mathematcal and readng scores at the age of seven, the return to schoolng falls margnally to 3.6%. However, lookng at the return to the varous academc and vocatonal qualfcatons, the coeffcents produced usng nstruments are substantally larger than the OLS estmates. At NVQ level 2, males holdng academc qualfcatons acheve a 20.6% premum, whereas the holders of vocatonal qualfcatons at the same level acheve a premum of 16.6%. At undergraduate degree level, the premum rses to 47.6% whle ther vocatonal counterparts acheve a 34% premum. Lookng at the estmates of the returns to schoolng (wthout formal qualfcatons) and explct qualfcatons usng Heckman selecton procedure (Table 9 below), the greatest return to an addtonal year of schoolng (where qualfcatons are omtted) s regstered at 10.0%. Upon the ncluson of explct qualfcaton levels, the return to an addtonal year of schoolng where no qualfcaton s obtaned stands at 4.0%, margnally hgher than the estmates produced under IV methods and substantally hgher than the estmates produced under OLS. The returns to explct levels of qualfcaton are lower at every level of qualfcaton than under IV approach. Specfcally, the return assocated wth obtanng an academc qualfcaton at NVQ level 2 s estmated at 19% (when mathematcal and readng test scores at the age of seven are ntroduced to the model) compared to 20.6% under IV and 20.1% under OLS. For those males holdng a vocatonal qualfcaton at NVQ level 2, the return s 14.6% compared to 16.6% under IV and 8.9% under OLS. 18 In Conlon (2000), an analyss of the effect of the reclassfcaton of qualfcatons on the lkelhood of achevng a gven labour market state s undertaken. It s found that the lkelhood of beng employed (or unemployed) s senstve to the method of qualfcaton classfcaton. 18