Estimating the probability of a match using microeconomic data for the youth labour market

23
Ž . Labour Economics 8 2001 335–357 www.elsevier.nlrlocatereconbase Estimating the probability of a match using microeconomic data for the youth labour market Martyn J. Andrews a, ) , Steve Bradley b , Richard Upward c a School of Economic Studies, UniÕersity of Manchester, Manchester, M13 9PL, UK b Lancaster UniÕersity, Lancaster, UK c UniÕersity of Nottingham, Nottingham, UK Received 20 July 1999; received in revised form 18 August 2000; accepted 22 February 2001 Abstract In this paper, we estimate the probability of a match for contacts between job seekers and vacancies. We relate the determinants of a match to the characteristics of the job seeker, the vacancy, and labour market conditions. Our main results are: ethnic minorities are discriminated against, but women are not; employers ‘cream’ the market and job seekers are ranked by their labour market state; high wage offers have a lower probability of a match; the probability of filling a job vacancy falls with vacancy duration, the higher stock of unemployed youths in a labour market, and the larger Careers Service; the probability of a match increases with job seeker duration. q 2001 Elsevier Science B.V. All rights reserved. JEL classification: J41; J63; J64 Keywords: Matching probability; Two-sided search 1. Introduction There has been a resurgence of interest in recent years in the matching, or hiring, function because of the potential insights it provides into the operation of the labour market and, in particular, the dynamics of unemployment. However, most of the previous work in this area has been highly aggregated and has focused on the estimation of the matching rate, which is the product of the contact rate and ) Corresponding author. Tel.: q 44-161-275-4874. Ž . E-mail address: [email protected] M.J. Andrews . 0927-5371r01r$ - see front matter q 2001 Elsevier Science B.V. All rights reserved. Ž . PII: S0927-5371 01 00035-5

Transcript of Estimating the probability of a match using microeconomic data for the youth labour market

Ž .Labour Economics 8 2001 335–357www.elsevier.nlrlocatereconbase

Estimating the probability of a match usingmicroeconomic data for the youth labour market

Martyn J. Andrews a,), Steve Bradley b, Richard Upward c

a School of Economic Studies, UniÕersity of Manchester, Manchester, M13 9PL, UKb Lancaster UniÕersity, Lancaster, UK

c UniÕersity of Nottingham, Nottingham, UK

Received 20 July 1999; received in revised form 18 August 2000; accepted 22 February 2001

Abstract

In this paper, we estimate the probability of a match for contacts between job seekersand vacancies. We relate the determinants of a match to the characteristics of the jobseeker, the vacancy, and labour market conditions. Our main results are: ethnic minoritiesare discriminated against, but women are not; employers ‘cream’ the market and jobseekers are ranked by their labour market state; high wage offers have a lower probabilityof a match; the probability of filling a job vacancy falls with vacancy duration, the higherstock of unemployed youths in a labour market, and the larger Careers Service; theprobability of a match increases with job seeker duration. q 2001 Elsevier Science B.V. Allrights reserved.

JEL classification: J41; J63; J64Keywords: Matching probability; Two-sided search

1. Introduction

There has been a resurgence of interest in recent years in the matching, orhiring, function because of the potential insights it provides into the operation ofthe labour market and, in particular, the dynamics of unemployment. However,most of the previous work in this area has been highly aggregated and has focusedon the estimation of the matching rate, which is the product of the contact rate and

) Corresponding author. Tel.: q44-161-275-4874.Ž .E-mail address: [email protected] M.J. Andrews .

0927-5371r01r$ - see front matter q2001 Elsevier Science B.V. All rights reserved.Ž .PII: S0927-5371 01 00035-5

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357336

the probability of a match. This paper represents a substantial departure from theexisting literature, insofar as we estimate the determinants of the probability of amatch using microeconomic data for a particular labour market in the UK. Thedata we use are the computerised records of the Lancashire Careers ServiceŽ . Ž .hereafter LCS and contain information on agents from both sides of the youthlabour market, as well as the interaction between firms and youths in the form ofover 80,000 contacts between job seekers and vacancies. We observe everyvacancy submitted by employers in Lancashire to the Careers Service betweenMarch 1988 and June 1992. For each vacancy, we observe every contact betweenthe employer and every job seeker during the sample period, and for each contact,we observe whether or not a successful match is made. There are only a handful of

Žstudies that use similar microeconomic data, and none for the UK Teyssiere,`.1996; van Ours and Lindeboom, 1996; Russo and van Ommeren, 1998 .

Clearly, there is considerable behavioural content in the matching probability,and our data enable us to shed light on a wide range of issues related to theoperation of the youth labour market. Specifically, the following questions areaddressed.

Ž .1 How does the probability of a match vary with the stock of unemploymentand vacancies in the labour market? Is the probability of a match pro- orcounter-cyclical? How do job and training sub-markets interact: does the stock inone sub-market affect the matching probability in the other?

Ž .2 How does the probability of a match vary with the size of the CareersService?

Ž .3 What is the relationship between the matching probability and the wage?Are high wage vacancies more likely to match? Does it matter whether wages areset by negotiation between firms and job seekers or whether wages are fixed inadvance?

Ž .4 How does the probability of a match vary with the duration of the vacancyand the duration of job search? Does the matching probability vary by currentlabour-market state?

Ž .5 Do firms select the ‘best’ job seekers regardless of their selection criteriaŽ .hereafter referred to as ‘creaming’ ? Is creaming a feature of both sub-markets?

Ž .6 Are ethnic minorities, women or the disabled discriminated against in thematching process?

Ž .7 Is there equal access to all types of youth training programmes? What typeof programme fulfills government policy of providing a ‘guaranteed’ place on the

Ž .Youth Training Scheme hereafter YTS for 16–17 year-old youths?The paper is organised as follows. In Section 2, we present a brief model of the

matching function and the role of the matching probability in such a model. InSection 3, we discuss the institutional background to the youth labour market, theCareers Service, and describe the main features of our data. Model specification isdiscussed in Section 4, and in Section 5 we discuss our findings. Section 6summarises and concludes.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 337

2. A framework for interpreting the probability of a match

ŽIn two-sided search models see, for example, Pissarides, 1990; Burdett and.Wright, 1998 , in a given labour market, the determinants of the flow of new hires

Ž .per period denoted by H are the flow of contacts between job seekers and firmsŽ .per period denoted by M and the probability that each contact is successful

Ž .denoted by m . This decomposition of the hiring function is written as

HsH S,V sm S,V M S,V , 1Ž . Ž . Ž . Ž .

Ž .where S is the stock of job seekers not all of whom are unemployed and V is thestock of vacancies. It is standard to assume that the contact rate M is a function of

Ž .both S and V, via a contact technology M S,V that has the same properties as aproduction function. The matching probability m may also depend on S and V, as

Ž .is discussed below. The resulting hiring function H S,V , synonymously knownas the matching function, has been estimated extensively in the literature, butusually with aggregate time-series data.

In two-sided search models, m is the joint probability that a job seeker finds anemployer acceptable, mw, and an employer finds a job seeker acceptable, me. The

Ž .first of these probabilities the ‘acceptance’ probability will depend on thereservation utility of the job seeker Rw, which in turn depends on labour markettightness VrS, the wage on offer w, the time the job seeker has been in hisrhercurrent labour-market state tw, the costs of search and the average quality ofrelevant jobs.1 The last two variables are both proxied by the characteristics of thejob seeker xw and the characteristics of the employer x e. Similar arguments implythat the employer’s ‘offer’ probability will depend on the same variables, with t e

being vacancy duration. This gives the following general specification of thematching probability function:

msm VrS,w ,tw,t e ,x w,x e sme VrS,w ,t e ,x w,x e mw VrS,w ,tw,x w,x e .Ž . Ž . Ž .2Ž .

We discuss the specification of x e and xw in Sections 4 and 5. Here, we areinterested in the predicted effects of labour-market tightness, VrS, the wage, w,and elapsed duration, t e and tw, on the matching probability. This is complicatedby the fact that we do not observe which party was responsible for whether or nota contact results in a match. That is, we observe m, but not me or mw.

From the employer’s viewpoint, an increase in VrS means that on average,there are fewer job seekers per vacancy. The employer responds by lowering his

1 If the contact function does not exhibit constant returns to scale, the specifications would be S andV, separately.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357338

e e Ž .reservation utility, R , so that m goes up i.e. the employer is less selective .From the job seeker’s viewpoint, Rw increases and mw falls. In Burdett and

Ž . w e ŽWright 1998 , R depends on R negatively a less selective employer generates. e weven fewer offers to job seekers and R depends on R negatively, and so, in

equilibrium, these responses are even stronger than the initial ‘one-sided’ re-sponses. A priori, the effect of labour-market tightness on the matching probability

Ž .is ambiguous. However, Burdett and Wright 1998 show that it is possible to havean equilibrium where job seekers are ‘especially easy’ in the sense that they acceptall job offers. This could happen in a very slack labour market.2 There is evidence

Žthat job seekers rarely refuse job offers from employers Barron et al., 1987;.Holzer, 1988; Barron et al., 1997; van den Berg, 1990; Manning, 2000 . If, indeed,

it is the case that the job seeker’s acceptance probability is close to unity, then thematching probability function can be interpreted as representing employers’ searchbehaviour, with m being positive. In general, as the market gets tighter, it canVrS

Ž .be shown that the function m VrS is inverted-U, turning when the market hasroughly equal numbers of employers and job seekers. In short, it is an empiricalissue as to whether the partial derivative is positive or negative; also thehomogeneity restriction that forms labour-market tightness is easily testable.

ŽThe effect of the wage is absolutely standard in search theory see, for example,. Ž .Mortensen, 1986 : if the mean of the offer distribution in utility terms increases,

the optimal response of the job seeker is to increase Rw, but by not as much as thew Ž .shift in the distribution, and so m increases the job seeker is less selective . By

symmetry, the employer is more selective and so me falls. Again, the employer’sresponse should dominate in slack labour markets.

It is possible that the matching probability will be a function of both theŽ e.vacancy duration t and the time the job seeker has been in hisrher currentŽ w .labour-market state t . For the employer, casual empiricism and some evidence

Ž e.suggests that the initial arrival rate of job seekers l is high and tails off quickly,Ž .for which Coles and Smith 1998 provide a convincing explanation. Once a given

stock of employers and job seekers have contacted and subsequently rejected eachother, then employers will only search the flow of new arrivals of job seekers,which necessarily lowers the rate at which they contact each other. The optimal

Žresponse of the employer is to become less selective andror possibly increase.search intensity . However, if the average quality of the job seeker also falls,

because the most suitable arrive first, then the employer becomes more selective.e eŽ e . wŽ w .The overall effect of t on the matching probability m t , . . . m t , . . . is

ambiguous.3 Similar arguments from the job seeker’s point of view suggest that it

2 Ž .Note that Burdett and Wright 1998 normalise VrSs1.3 e e eŽ e.The effect of t on the employer’s hazard, h , is also ambiguous, as seen by h t s

wŽ w . eŽ e. eŽ e.m t m t l t . However, it should be stressed that in this paper, we are estimating a model of thematching probability, not vacancy duration.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 339

is an empirical issue as to whether job seekers become more or less selective as tw

passes.Finally, it is worth noting that there are only three other studies that estimate

Ž .the probability of a match using microeconomic data. Teyssiere 1996 uses`French data on a sample of contacts between job seekers and firms for a single

Ž .labour market in Marseilles. van Ours and Lindeboom 1996 estimate models ofthe contact rate, the probability of a match and the hiring function, using Dutch

Ž .data collected from a variety of different sources. Russo and van Ommeren 1998also use Dutch vacancy data on the number and gender of job seekers. Compar-isons between the findings in these papers and our own are noted later.

3. Institutional background

The collapse of the youth labour market in the UK in the early 1980s led to theŽ .introduction of the Youth Training Scheme YTS in 1983. It has remained in

place ever since, although in several disguises. The YTS is not a homogeneousprogramme; it can be seen as a route to a wide variety of skilled occupations, orseen as a work-experience programme designed to mop-up the excess supply ofyouth labour. Since its introduction, youths can choose between four labour-marketactivities at the age of 16: different types of YTS, continue their education, get ajob or become unemployed. Employers can also choose whether to recruit youthsvia the YTS or directly into a job.

The Careers Service fulfills a similar role for the youth labour market asEmployment Offices and Job Centres provide for adults. Its main responsibilitiesare to provide vocational guidance for youths and to act as an employment serviceto employers and youths. The latter includes a free pre-selection service foremployers. Use of the Careers Service is voluntary for employers with jobvacancies, whereas notification of YTS vacancies is compulsory, so that thegovernment offer of a guaranteed place for all 16–17-year-old youths can bemonitored. Having notified the Careers Service of the type of vacancy—theoccupation, the wage, a closing date for applications and selection criteria—jobseekers are selected for interview. In other words, a contact is made. Either amatch occurs or the pair each continue their search.

The data we use are the computerised records of the Lancashire Careers ServiceŽ .LCS . The LCS holds records on all youths aged between 15 and 18, includingthose who are seeking employment. We observe every vacancy notified byemployers to the Careers Service between March 1988 and June 1992. All YTSvacancies and about 30% of job vacancies are notified to the Careers Service. Jobvacancies for which the Careers Service is not the method of search are notincluded in the data. Job vacancies require both high- and low-quality job seekers,and are representative of all entry-level jobs in the youth labour market. It follows

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357340

that our data are representative of all job seekers, because we observe all contactsbetween notified job vacancies and job seekers. This is not an issue for YTSvacancies because all of them are notified to the Careers Service.

The structure of the YTS is more complex than is commonly recognised. Atone extreme is the large firm who provides work experience and in-house training.At the other extreme are Training Providers, who act as umbrella organisations,recruit trainees, and provide the off-the-job training, but do not offer workexperience directly. Therefore, one of the main roles of the Training Provider is tocoordinate the off-the-job training and the work experience for trainees and

Ž .participating firms typically small firms . Training Providers receive a ‘fee’ fromgovernment for the service they provide, whereas all firms that participate in theYTS received a subsidy towards wage costs. The ‘fee’ was approximately £100

Ž .per trainee for the sample period 1988–1992 , whereas the subsidy to firmsvaried. The government set the minimum for the YTS allowance paid to thetrainee, which was, during the sample period, approximately £27.50 per week andincreasing to £35.00 per week after 1-year training. A single firm recruiting youngpeople through the YTS received all of the subsidy. Training Providers obtained acontribution towards the allowance from participating firms. This was typicallylow, and very small firms therefore had the greatest incentive to participate in YTSbecause the wage costs of an additional worker constitute a substantial proportionof total variable costs.

There are also differences in the quality of training within the YTS. First, thereare employee-status programmes, where the participants receive the rate of pay forthe job and have greater job security; these are different from trainee-statusprogrammes, which offer a standard allowance and a fixed contract. Employee-status programmes are also more likely to be part of a longer training schemeprovided by the firm, such as a 3- or 4-year apprenticeship. Competition for placeson employee-status programmes is consequently greater, but firms are also morelikely to be selective. Second, at the lower end of the YTS market, there arespecial programmes, usually provided by the voluntary sector, which receive extrafunding to deal with the training needs of the less able. Special programmes areless selective, implying that special programmes bear the brunt of the Govern-ment’s policy of a guaranteed place for all youths.

Throughout our empirical analyses, the data are split between job vacancies andYTS vacancies, since they represent two distinctly different types of sub-market.

4. Data and model specification

To recap, for each vacancy we observe every interview between an employerand a job seeker during the sample period; hereafter, this unit of observation isreferred to as a contact. The sample consists of ms86,978 contacts. In the line

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 341

with the notation developed in Section 2, there are five sorts of information whichdescribe each contact:

Ž . Ž .1. Whether each contact is successful h s1 or not h s0 .i iŽ w .2. Information about the job seeker x , including hisrher current labour-i

Ž .market state unemployed, in a job, on a training scheme, or in school , andŽ w .duration in that state t .i

Ž e. Ž e.3. Information about the employer x , vacancy duration t , and otheri iŽ .characteristics of the vacancy, including the wage w .i

4. Information on the selection criteria of the firm, from which we compute the‘distance’ between job seeker characteristics and vacancy characteristics,

< w e <denoted by x yx .i i

5. Aggregate information about the state of the labour market, including stocksŽ . Ž .of job seekers S and vacancies V .

Our model for the probability of a match is therefore written as

e w e w < w e <Pr H s1 sF S ,V ,w ,t ,t ,x ,x , x yx , is1, . . . ,m , 3Ž . Ž .Ž .i i i i i i i i i i

which, given that F denotes the normal distribution function, is estimated as aProbit regression.

Aggregating over all contacts in the data, the raw matching probability is theŽ .total number of hires in the data h'Ýh divided by the number of contacts:i

mshrm. Table 1 shows the total number of contacts, hires and the aggregateˆmatching probability, stratified by the two sub-markets. Although there areapproximately the same number of applications made to jobs as to YTS vacancies,only 1 in 10 applications to job vacancies results in a match, whereas nearly one in

Table 1Sample size and raw matching probability, 1988–1992

Jobs YTS

Ž .Total number of contacts m 42,698 44,280Ž .Total number of hires h 4364 10,452

Ž .Raw matching probability mshrm 0.102 0.236ˆTotal number of job seekers, of which: 11,093 12,949One contact is made 3175 3480Two or more contacts 7918 9469

Total number of firms, of which: 3159 947One contact made 528 125Two or more contacts 2631 822

Total number of vacancies, of which: 7315 2637One contact made 1987 322Two or more contacts 5328 2315

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357342

Fig. 1. Average matching probability by month.

four applications to YTS vacancies are successful. Table 1 also shows the numberof job seekers, firms and vacancies. Because the unit of observation is a contact,and because firmrjob seeker pairs do not recontact each other, there is no panelelement in these data. However, the majority of job seekers, firms and vacanciesare observed more than once.

Finally, Fig. 1 plots the raw matching probabilities over the sample period.Recruitment is lower between March and May, just before the majority ofschool-leavers enter the labour market, and is much higher in June, July andSeptember. A similar picture arises for YTS vacancies, although more pro-nounced.

5. Results

Ž .Table 2 reports estimates of Eq. 3 , estimated separately for contacts to jobvacancies and contacts to YTS vacancies. Marginal effects, p-values and sample

Ž .means are reported see Table 2, tablenote a . Standard errors are robust and arecorrected for intra-district correlation between contacts. In Section 1, we raised

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 343

Table 2Ž .Probit results dependent variable hi

Job vacancies YTS vacanciesbMarginal P-value Mean Marginal P-value Mean

aeffect effectc( )Labour market characteristics S , Vi i

w x w xLog unemployed -18 y0.0175 0.000 5.112 y0.0098 0.480 5.117w x w xLog job vacancies in the same y0.0052 0.010 2.447 y0.0045 0.333 1.981

doccupationw x w xLog YTS vacancies in the same 0.0015 0.225 4.040 y0.0060 0.068 4.258

doccupationw x w xLog density 0.0008 0.832 1.984 y0.0056 0.686 2.036

Ž w x w xLog Careers Service staffr y0.0258 0.000 y9.169 y0.0217 0.110 y9.224.population

eŽ .Employer and Õacancy characteristics x i

Wage offere eŽ . w x w xw D 1yN y0.0395 0.000 £1.49 y0.0225 0.214 £0.76i i ie ew x w xw D N y0.0681 0.001 £1.42 0.0076 0.795 £0.63i i i

w x w xD N 0.0117 0.382 0.042 y0.0040 0.791 0.040i iŽ . w x1yD N y0.0056 0.362 0.134 0.000i iŽ .1yN D 0.824 0.960i i

eŽ .Duration of vacancy ti

Vacancy open for -1 month 0.500 0.290w x w xVacancy open for 1–2 months y0.0130 0.001 0.266 y0.0009 0.929 0.156w x w xVacancy open for 2–3 months y0.0275 0.000 0.090 y0.0047 0.755 0.178w x w xVacancy open for 3–6 months y0.0349 0.000 0.104 0.0183 0.380 0.378w x w xVacancy open for 6–12 months y0.0321 0.000 0.026 0.0491 0.030 0.153w x w xVacancy open for )12 months y0.0650 0.000 0.014 y0.0257 0.495 0.039

Firm size-11 employees 0.383 0.290

w x w x11–30 employees 0.0070 0.143 0.206 y0.0367 0.014 0.274w x w x30–100 employees 0.0029 0.598 0.192 y0.0302 0.019 0.158w x w x)100 employees 0.0060 0.256 0.219 y0.0227 0.046 0.278

Ž .Firm activity SICw x w xAgriculture 0.0048 0.714 0.007 0.1086 0.025 0.001w x w xEnergy and water supplies y0.0199 0.091 0.007 0.0031 0.853 0.023w x w xExtraction of minerals, metals 0.0007 0.951 0.009 y0.0142 0.720 0.007w x w xMetal goods, engineering 0.0058 0.187 0.118 0.0055 0.841 0.067w x w xOther manufacturing 0.0043 0.401 0.160 0.0858 0.000 0.031w x w xConstruction 0.0026 0.755 0.064 0.0641 0.202 0.029

Distribution, catering and hotels 0.305 0.203w x w xTransport and communication y0.0126 0.001 0.027 0.0030 0.934 0.038w x w xBanking, finance 0.0119 0.002 0.141 0.0037 0.853 0.051w x w xOther services 0.0056 0.323 0.161 y0.0029 0.867 0.235

w xTraining agent 0.000 0.0210 0.441 0.317Location

w x w xFirm in town centre y0.0038 0.319 0.508 0.0076 0.328 0.466Firm outside town centre 0.492 0.534

( )continued on next page

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357344

Ž .Table 2 continued

Job vacancies YTS vacancies

Marginal P-value Mean Marginal P-value Meaneffect effect

eŽ .Employer and Õacancy characteristics x i

Involvement in YTS schemesw xProvides YTS placements y0.0067 0.106 0.378 1.000

No YTS placements 0.622 0.000Occupational type

w x w xSkilled occupation y0.0077 0.136 0.637 y0.0188 0.029 0.635Unskilled occupation 0.363 0.365

w x w xNon-manual occupation y0.0203 0.001 0.581 y0.0226 0.154 0.531Manual occupation 0.419 0.469

Training providedw xAIn-houseB training 0.0059 0.212 0.066 0.000w xDay release training 0.0016 0.732 0.140 0.594w x w xApprenticeship y0.0074 0.309 0.175 y0.0266 0.012 0.406

No or little training provided 0.619 0.000Type of YTS scheme

w xYTS employee y0.0314 0.021 0.302w xSpecial funding 0.0906 0.008 0.018

Standard training programme 0.680Application method

w x w xWritten application required y0.0735 0.000 0.457 y0.0543 0.000 0.610No written application required 0.543 0.390

w x w xLog size of vacancy order 0.0387 0.000 0.313 0.0788 0.000 2.049

wŽ .Job seeker characteristics x i

Ethnicityw x w xNon-white y0.0279 0.002 0.049 y0.0585 0.000 0.046

White 0.951 0.954Gender

w x w xFemale y0.0053 0.691 0.467 y0.0210 0.281 0.426Male 0.533 0.574

f w x w xFemale=proportion of females 0.0483 0.041 0.315 0.0442 0.066 0.297f w x w xProportion of females y0.0242 0.072 0.470 y0.0021 0.937 0.423

Healthw x w xPoor health y0.0029 0.471 0.146 y0.0063 0.200 0.147

Normal health 0.854 0.853Number of occupational choices1 0.172 0.194

w x w x2 y0.0047 0.162 0.279 y0.0283 0.000 0.277w x w x3 y0.0056 0.029 0.281 y0.0452 0.000 0.232w x w x4 y0.0019 0.685 0.268 y0.0644 0.000 0.297

Areas in which job seekers arewilling to workLocal district only 0.120 0.211

w x w xAnywhere in the district y0.0160 0.052 0.779 y0.0515 0.000 0.719w x w xAnywhere in Lancashire y0.0100 0.141 0.102 y0.0387 0.000 0.070

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 345

Ž .Table 2 continued

Job vacancies YTS vacancies

Marginal P-value Mean Marginal P-value Meaneffect effect

wŽ .Job seeker characteristics x i

Special training needsw x w xSpecial training needs y0.0230 0.000 0.041 y0.0101 0.345 0.048

No training needs 0.959 0.952CohortLeft school in 1988 0.270 0.210

w x w xLeft school in 1989 0.0055 0.269 0.289 0.0725 0.000 0.228w x w xLeft school in 1990 0.0018 0.712 0.251 0.1345 0.000 0.254w x w xLeft school in 1991 0.0045 0.607 0.144 0.2277 0.000 0.193w x w xLeft school in 1992 y0.0025 0.855 0.047 0.1419 0.000 0.115

wŽ .Duration in current state tiw x w xF1 month in FE y0.0233 0.012 0.024 y0.0940 0.000 0.012w x w x)1 month in FE 0.0200 0.001 0.049 0.1157 0.000 0.019w x w xF3 months in YTS 0.0076 0.136 0.023 0.0642 0.011 0.012w x w x)3 months in YTS 0.0327 0.000 0.052 0.1667 0.000 0.016w x w xF3 months in employment 0.0854 0.000 0.023 0.1678 0.000 0.009w x w x)3 months in employment 0.1013 0.000 0.018 0.2549 0.000 0.007w x w xF1 month in unemployment y0.0619 0.000 0.344 y0.1388 0.000 0.290w x w x1–2 months in unemployment 0.0047 0.227 0.087 0.0097 0.509 0.046w x w x2–3 months in unemployment 0.0151 0.022 0.065 0.0380 0.030 0.031w x w x)3 months in unemployment 0.0031 0.449 0.106 0.0164 0.505 0.049

gLabour market historyw x w xLog months in FE 0.0014 0.122 y2.949 0.0073 0.046 y3.289w x w xLog months in YTS y0.0011 0.112 y2.357 0.0065 0.000 y3.052w x w xLog months in unemployment y0.0004 0.634 y1.221 0.0118 0.000 y2.464w x w xLog months in employment 0.0007 0.347 y2.460 0.0010 0.677 y3.107

Last school attendedw x w xSingle sex school 0.0090 0.263 0.051 y0.0095 0.409 0.058

Mixed sex school 0.949 0.942h w x w xSchool GCSE performance y0.0035 0.739 0.339 0.0262 0.236 0.344

w eŽ < <.Matching characteristics x yxi i

Agew x w xAge of applicant lower y0.0093 0.116 0.033 y0.0092 0.562 0.036

Age of applicant same 0.939 0.944w x w xAge of applicant higher 0.0079 0.003 0.028 y0.0155 0.219 0.020

Qualificationsw x w xApplicant qualifications lower y0.0152 0.001 0.140 y0.0333 0.000 0.165

Applicant qualifications same 0.465 0.388w x w xApplicant qualifications higher 0.0110 0.001 0.395 0.0231 0.000 0.447

OccupationNo match 0.321 0.350

w x w xPreference matches to two-digits y0.0080 0.114 0.329 0.0136 0.023 0.291w x w xPreference matches to three-digits y0.0088 0.105 0.351 0.0588 0.000 0.358

( )continued on next page

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357346

Ž .Table 2 continued

Job vacancies YTS vacancies

Marginal P-value Mean Marginal P-value Meaneffect effect

w eŽ < <.Matching characteristics x yxi iLocation

w x w xPreference matches y0.0059 0.127 0.183 y0.0178 0.005 0.301No match 0.817 0.699

School subjectsw x w xSubject matches 0.0040 0.145 0.192 0.0046 0.508 0.237

No match 0.808 0.763

Time dummiesJanuary 0.034 0.016

w x w xFebruary 0.0053 0.530 0.060 y0.0130 0.428 0.022w x w xMarch y0.0273 0.001 0.071 y0.0854 0.000 0.073w x w xApril y0.0250 0.000 0.116 y0.1339 0.000 0.223w x w xMay y0.0199 0.010 0.104 y0.1125 0.000 0.135w x w xJune 0.0060 0.531 0.170 y0.0366 0.113 0.259w x w xJuly 0.0241 0.007 0.101 0.1116 0.000 0.087w x w xAugust y0.0047 0.611 0.087 0.0907 0.000 0.060w x w xSeptember 0.0075 0.387 0.064 0.0753 0.001 0.033w x w xOctober 0.0065 0.511 0.090 0.0331 0.224 0.050w x w xNovember y0.0042 0.612 0.063 y0.0269 0.094 0.023w x w xDecember y0.0111 0.168 0.041 y0.0417 0.013 0.020

1988 0.089 0.160w x w x1989 y0.0138 0.017 0.204 y0.0483 0.023 0.220w x w x1990 y0.0233 0.001 0.290 y0.1138 0.000 0.235w x w x1991 y0.0418 0.000 0.282 y0.1677 0.000 0.222w x w x1992 y0.0421 0.000 0.135 y0.1864 0.000 0.163

X ˆxb y1.512 y0.942X ˆŽ .f xb 0.127 0.256X ˆŽ .F xb 0.065 0.173

N 42,698 44,280Log L y11,856.981 y18,220.591

2x 497.788 695.3932McFadden’s pseudo-R 0.158 0.247

2Count R 0.899 0.817

a Marginal effects are computed for continuous variables, whereas for dummy variables the‘marginal effect’ compares the effect of switching the variable from 0 to 1. For each continuous

X Xˆ ˆŽ .variable, the ‘marginal effect’ and the underlying coefficient are in the ratio f xb , where xb is theŽ .regression function evaluated at the means of the data. Regression includes constant not shown .

bP-values refer to test of underlying coefficient being 0. Standard errors are robust and arecorrected for intra-district correlation between contacts.

cS and V vary by district and month, Careers Service staff varies by district and year, andpopulation and area vary by district only.

d Defined as the number of vacancies open in the same month, the same district and the sameoccupation.

eAverage real hourly wage rate for group defined by dummy variables.f Proportion of females applying to that occupation.g Ž .Log 1 dayqduration used to avoid missing values at 0. Excludes duration in current state.h Proportion of year 11 pupils obtaining five or more GCSE grades A) –C.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 347

seven questions, and the results are discussed in relation to those questions. Thisdiscussion does not cover all the covariates reported in Table 2, the rest being afairly standard set of controls.

5.1. The effect of the labour market

See Table 2, Labour market characteristics.The basic theoretical prediction is that, in slack labour markets, labour-market

Ž .tightness VrS has a positive effect on the matching probability. Both Fig. 1 andthe estimates on the year dummies in the regressions suggest that there is, indeed,this pro-cyclical effect.4 Being only suggestive, we seek to detect whether thetightness of a particular market has an effect on the individual job seeker’smatching probability.

Table 2 reports the effect of the stocks of job seekers and vacancies in a givenlabour market, together with the effect of three other ‘aggregate’ variables. Labourmarkets are defined by the 14 districts in Lancashire. Our measure of job seekers,S, is the stock of unemployed in each district-month who are under 18, taken fromthe National Online Manpower Information System database. The stock of jobseekers includes a proportion of those not unemployed, typically those in jobs oron training schemes. As with all the aggregate studies, this is a potential source ofmisspecification. The implications and possible solutions are discussed in BurgessŽ . Ž .1993 and Mumford and Smith 1999 . For example, in Burgess’s job-competi-tion model, an increase in the proportion of on-the-job job seekers has a differen-tial impact on the relative outflow rates out of unemployment and jobs, in favourof the latter. Aggregate studies are unable to detect whether this crowding-out isbecause the matching probability also depends on labour-market state, an effect we

Ž .are able to estimate see Section 5.3 below even though we do not observe theaggregate stocks of each type of job seeker.

Once we correct for the intra-district correlation between contacts, generallythese aggregate variables are insignificant and have weak effects. However, the

Ž .marginal effect of the district-level stock of the unemployed under 18 S convertsŽ .to an elasticity of y0.172 y0.0175r0.102 in the jobs regression—the matching

probability is about 1 percentage-point lower for a district with double unemploy-ment—but is insignificant in the YTS regression.5

Our measures of V are taken from the LCS data. We count the number of joband YTS vacancies in each district-month. We also count the number of vacancies

4 The downward trend in the matching probability is faster for training schemes. This is consistentwith higher staying-on rates into continued education over this period, generating more competitionbetween jobs and YTS. As young people generally prefer jobs to schemes, the matching probability forthe latter declines faster.

5 Throughout, an elasticity is calculated as a marginal effect divided by the raw matchingprobability.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357348

which ‘compete’ with the contact itself, in the sense of being in the samedistrict-month and occupation. We can therefore examine whether the effect of thestock of similar vacancies on the matching probability is greater than an increasein aggregate vacancies.

We find that the total stock of vacancies, whether or not split between jobvacancies and YTS vacancies, has no significant effect on the matching probabil-ity. We therefore report a specification which uses the stock of competingvacancies, split between jobs and YTS. The ‘own effect’—the effect of the stock

Ž . Ž .of job YTS vacancies on the probability of a job YTS contact resulting in amatch—is significant but incorrectly signed, if we maintain the assumption thatthe market is slack and mw is close to 1. In other words, it is inconsistent with thenegative effect of S above. It should, of course, be remembered that both thestocks of S and V are measured with error: we do not observe all vacancies

Ž .notified to the Careers Service about 30% of the total , and we do not observe allnon-unemployed job seekers.

The two other variables which vary at the district level are the number of staffin a given Careers Office, denoted by C, normalised on the population of eachdistrict, denoted by P, and the population density of each district, denoted by PrAŽ .where A is the area of each district . Homogeneity is easily not rejected. Theformer is included because both contact rates and hiring rates may depend on the‘efficiency’ of the Careers Service. If the matching probability is estimated asincreasing in CrP, this suggests that those offices with more staff may be able tospend more time sifting the pool of searchers and select more suitable job seekersfor interview. On the other hand, smaller offices typically deal with a smaller poolof firms and job seekers, and so their higher matching probability may reflectbetter information on both employers and job seekers. Population density isincluded because the rate at which agents contact and match with each other mightbe lower in more dispersed rural labour markets compared with those in citiesŽ .Coles and Smith, 1996 .

The population density variable is insignificant in both regressions. However,the number of Careers Service staff has a significant, negative effect on theprobability of a match in both regressions. For the jobs regression, the elasticity is

Ž .y0.253 y0.0258r0.102 . Although the effects are not large, this finding isconsistent with the view that job seekers do benefit from personal counselling byCareers Service staff. From a policy viewpoint, this finding is interesting, althoughthe implied policy of decentralising the Careers Service would, of course, costmore.

5.2. Wages and training allowances

See Table 2, Employer and Õacancy characteristics.The relationship between the wage offer and the probability of a hire is of

particular interest, especially in two-sided search models. There are three different

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 349

types of wage offer in the data.6 Ninety-six percent of YTS vacancies and 82% ofjob vacancies have a set pre-announced wage, where the wage is non-negotiable.The majority of these vacancies specify age and tenure profiles, which reflects therigid institutional nature of wage setting in the youth labour market.7 Four percentof both YTS and job vacancies have a set pre-announced wage offer, but are stillopen to negotiation. The remaining 13% of job vacancies have a negotiable wageoffer and no pre-announced wage. Clearly, for this third category, there is no wagerecorded in the data.

The important point is that both job seekers and employers take the wage asgiven when they decide whether or not to form a match. Our justification for

Ž .focusing on the Burdett and Wright 1998 model in Section 2 is that they analysewhat happens when the wage is not negotiable after agents meet, and assume thatan agent cannot transfer utility to the other party by varying the wage or by othermeans.8 We argue that this is a very accurate characterisation of the youth labourmarket, given the vast majority of job and YTS vacancies in the data have anon-negotiable wage.

We model these effects as follows. N is defined as a dummy variablei

indicating whether a vacancy has a negotiable wage offer, and D as a dummyi

variable indicating whether the wage is pre-announced. Interacting the dummieswith the log real hourly wage rate w , where it exists, gives:i

Pr H s1 sF b q . . .qb w D 1yN qb D N qb w D NŽ . Ž .Ži 0 1 i i i 2 i i 3 i i i

qb 1yD N . . . , 4Ž . Ž ..4 i i

The four parameters are interpreted as follows:

Ž . Ž .Wage set D s1 Wage not set D s0i i

Ž .Wage not negotiated N s0 b qb w not applicablei 0 1 iŽ .Wage negotiated N s1 b qb qb w b qbi 0 2 3 i 0 4

Ž .In this model, the predictions are clear cut. A high wage or training allowanceŽ w .means that job seekers increase their probability of accepting a job offer m , but

Ž e.employers decrease the offer probability m . In slack labour markets, theŽ Ž ..employer effect should dominate. The key parameter see Eq. 4 is b , the1

impact of the wage on the matching probability for the large majority of job andtraining vacancies, whose wages are set in advance and are non-negotiable. For the

6 The term ‘wage’ also refers to the training allowance for YTS vacancies.7 Our own personal knowledge of this particular market leads us to believe that wage rates do not

Ž .reflect supply and demand conditions see Section 3 .8 Ž .Note that in Pissarides 1990 , utility is transferable, and the wage is determined by splitting the

total surplus of both parties.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357350

ˆ Ž .jobs regression, b converts to an elasticity of y0.387 y0.0395r0.102 , whereas1ˆ Ž .for the YTS regression, b converts to y0.093 y0.0225r0.242 and is insignifi-1

cant. It makes sense that jobs are more responsive, and implies that the employerŽ .effect does outweigh the job-seeker effect. The elasticity see b is twice as3

strong for job vacancies that are negotiable, but is zero for YTS vacancies that arenegotiable; however, these are only a very small proportion of the sample.

5.3. The length of search and labour-market state

See Table 2, Job seeker characteristics and Employer and Õacancy character-istics.

In this subsection, we address the issues raised in the fourth question of Section1. The variables ‘duration of Õacancy’ non-parametrically model the effects of

Ž e.vacancy duration t on the matching probability. For job vacancies, the resultssuggest that the longer the vacancy has been unfilled, the lower the probability of

Ž .a match. The probability of a match falls significantly after 1 month by y0.0130 ,Ž .and again after 2 months y0.0275 . The negative duration effect is then quite

Ž .similar for vacancies open anywhere from 2–12 months fy0.03 , but there isŽ .another drop after 12 months. However, very few 4% of job vacancies remain

unfilled after 6 months. If it is the case that the arrival rate of job seekers declineswith vacancy duration, then this is interpreted as a job-seeker quality effect. Adifferent picture emerges for YTS vacancies. For vacancies unfilled between 6 and12 months, the more likely it is that a match will be found, compared with shorterdurations. This is consistent with the view that the programme absorbs any

Žtemporary excess supply of labour only 4% of YTS vacancies survive beyond 12months because a new batch of YTS vacancies are posted annually onto the

.market and old vacancies cannot compete .ŽVacancies are often posted on to the market in multiple ‘orders’ see Table 2,

.‘log size of vacancy order’ . In fact, the average size of a YTS order is 19vacancies, but only two for job orders. As firms will exploit economies of scale tosearch wherever possible, i.e. the costs of search per vacancy falls, we wouldexpect the matching probability to decrease with the size of an order. On the otherhand, the quality of the job seeker may also fall, and so firms would be lessselective. In the data, it is this the latter effect that dominates—the elasticity inboth regressions is about one-third.

Turning to the other side of the market, we observe job seekers in one of fiveŽ .states: compulsory schooling the base group , post-compulsory education

Ž .hereafter FE , YTS, unemployment and employment. The variables ‘duration incurrent state’ non-parametrically model the effect of the time the job seeker has

Ž w .been in hisrher current labour-market state t on the matching probability.More precisely, the duration in each state is classified into one of two categories:

Žless than x months and greater than x months this does not apply to compulsory.schooling . To get roughly equal proportions in the two categories, xs1 for FE

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 351

and xs3 for YTS and employment; for unemployment, we use four categories.For applications to YTS vacancies, there is a clear ranking in terms of theprobability of a match: employed job seekers are clearly at the front of the queueŽ . Ž . Ž .0.1678 , followed by those on YTS 0.0642 , compulsory schooling base , FEŽ . Ž .y0.0940 , with the unemployed at the back of the queue y0.1388 . A similarpicture emerges for applications to job vacancies, although the probability differ-

Ž . Ž .entials are much smaller. Recall that Blau and Robbins 1990 and Holzer 1988find that employed job seekers have a higher probability of finding a job whencompared with the unemployed.

There is no evidence whatsoever that the matching probability declines after xmonths for any of the four states: in the YTS regression, it goes up by 0.210 for

Ž .FE 0.0940q0.1157 , by 0.103 for YTS, by 0.087 for the employed and by 0.149for the unemployed after 1 month. These are large differentials. The equivalentfour numbers for job vacancies are 0.043, 0.025, 0.016, 0.067, i.e. are muchsmaller but still positive. The implication is clear. Any negative duration depen-dence in vacancy hazards to jobs or YTS vacancies is because the contact ratefalls, not because the matching probability falls, supporting Coles and Smith’sŽ .1998 stock-flow matching hypothesis. See the equation in footnote 3.

There are various reasons why the matching probability rises. The first is astandard search theory response to a fall in the arrival rate of suitable vacancies,irrespective of labour-market state. Second, for those who are employed or onYTS, being in a state for longer than a month signals potential accrual oftransferable skills which increases their employability. The third reason appliesonly to the unemployed, and might be the effect of short-term benefit paymentsreceived by 16- and 17-year-olds, which lasted for only 8 weeks, and may havemade the short-term unemployed more selective in their job search. Once benefitentitlement is exhausted, reservation wages fall, and so the longer-term unem-ployed become less selective as the Careers Service encourage them to accept anyjob or YTS offers. This is why we used four finer categories. For the YTSregression, there is another clear rise of 0.028 after two months, before fallingback after 3 months; similar effects are seen in the jobs regression. This suggeststhat policing the longer-term unemployed by requiring them to attend an interviewwith the Careers Service may have a positive effect on the outflow rate fromunemployment.

In addition to observing each job seekers’ current labour-market state andduration in that state, we are able to construct a set of variables which summarisetheir work history since leaving school. Workers with a poor work history mayface an added disadvantage in the selection process, insofar as employers are less

Žlikely to make job offers to job seekers with repeated spells of unemployment an.effect found by Teyssiere, 1996 . None of the four variables are significant for the`

jobs regression; however, there is a positive effect from all four labour-marketstates on the matching probability for training schemes, which suggests that wemight be picking up some kind of age effect.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357352

5.4. Do firms ‘cream’ the market?

See Table 2, Matching characteristics.The role played by education in the labour market has been the subject of

controversy in labour economics for many years. Human capital theory suggeststhat education has a direct effect on worker productivity, hence the positivecorrelation between educational level and wages. Theories of signalling suggestthere is also an indirect correlation between education and wages, insofar aseducational credentials signal unobservable differences in worker productivityŽ . Ž .Spence, 1974 . Thurow 1975 argues that job seekers are ranked in a queue onthe basis of their ‘trainability’, and workers with better characteristics are selectedbecause they are cheaper to train. Both approaches therefore predict that thebest-qualified job seekers receive job offers, irrespective of the requirements of thejob.9 In view of this debate, we are interested in whether firms ‘cream’ the marketby selecting the most qualified job seekers.

As discussed in Section 4, because each contact contains information on boththe job seeker and the vacancy to which the application is made, it is possible toconstruct measures of the distance between the characteristics of the job seeker

Ž < w e <.and the characteristics of the vacancy denoted x yx . There are numerousi i

specifications which could be adopted for modelling the distance between two setsŽof dummy variables. For those covariates with a natural ordering age and GCSE

.qualifications , we create dummies measuring whether the job seeker’s character-istics are lower or higher than the advertised characteristic. For unordered covari-

Ž .ates subject, occupation and location , we use a single dummy to record whetheror not the job seeker’s characteristic coincides with the advertised vacancycharacteristic.

Job seekers who possess academic qualifications which are superior to thoseŽ .specified on the vacancy ‘applicant qualifications higher’ have a higher probabil-

ity of a match by 0.0231 points for YTS vacancies. Similarly, those with lowerŽ .qualifications ‘applicant qualifications lower’ are less likely to match by 0.0333.

The same effects are found for jobs. However, there is no impact from whether thesubject matches. Nonetheless, these findings can be taken as evidence in favour ofeither the signalling or job-competition models.

ŽThere is some evidence that employers ‘cream’ older job seekers first in the.jobs regression only , although only 6% of the sample do not have the same

Ž .integer age as specified in the vacancy. Job preferences can match at a two-digitor a three-digit level; the latter is more likely to correspond more closely to thevacancy requirements, and should therefore have the highest probability of a

9 Ž .Singell et al. 1997 constructed an educational signal and found that there are positive returns inŽ .wage equations. van Ours and Ridder 1995 find some evidence of job competition when estimating

Ž .Eq. 1 for four types of occupation and four levels of education.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 353

match. In the YTS regression, it is the three-digit match that has the highestmatching probability, by some 0.0588 points compared with the base category ofno match. Curiously, where the travel-to-work intentions of job seekers and thelocation of the firm offering a YTS coincide, the probability of a match falls.

5.5. Discrimination in the matching process

See Table 2, Job seeker characteristics.Discrimination can occur with respect to gender, race and health, although

legislation regarding gender and racial discrimination has been more rigorouslyenforced. Conventionally, wage or employment discrimination has been estimatedas a residual from a wage or employment regression after controlling for all

Žrelevant observables see, for example, Leslie et al., 1998 for a collection of recent.studies .

Our approach is more direct. We are able to investigate whether females orŽ .Asians mainly Pakistani–Bangladeshi in origin , conditional on a contact, have a

lower probability of a match than their male or white counterparts; note that ourcontrols include labour-market state, academic qualifications, occupational choices,mobility, and school background.10

There is clear evidence that job seekers from a minority ethnic backgroundhave a lower probability of a match, particularly in terms of applications to YTS

Ž .vacancies an estimated differential of y0.0585 log-points . This is an alarmingfinding in view of the stringent equal opportunity policy that has been imple-mented in the training market. If the probability of acceptance is close to 1, thenthe implication of our findings is that racial discrimination does indeed exist.

Ž .Teyssiere 1996 finds the same effect.`Turning to gender discrimination, to allow for gender segmentation in occupa-

tional choice, we calculate the proportion of female job seekers to each vacancy,denoted by x, the hypothesis being that jobs that are traditionally associated withwomen are less likely to match. For each of the 74 occupational codes, we alsocalculate the proportion of females among all job seekers.11 In contrast to Russo

Ž .and van Ommeren 1998 , we find that only 27% of vacancies receive applicationsfrom females only, suggesting a greater degree of competition between males andfemales in the youth labour market in the UK. This variable is also interacted withthe female dummy, denoted by F. The results for YTS vacancies give a matching

10 Unlike some studies, our sample of non-white observations is sizable. Also, it is possible that thereis discrimination in the process that generates the contact; if this were so, the degree of discriminationis underestimated.

11 Occupations are classified according to their Occupational Training Family. There are 11 two-digitfamilies each of which contains several three-digit occupational categories.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357354

probability differential of y0.0210q0.0442 x for females compared with men;the higher the proportion of female job seekers, the higher the probability of amatch. Where the proportion of women is greater than 0.0210r0.0442—roughlyone-half—there is a positive differential in favour of women. For the jobsregression, we write the regression result as 0.0483Fxy0.0242 x, giving differen-tials of y0.0242 x if the job seeker is a man and 0.0241 x if a woman. So, for jobsvacancies where x is low, there is no discrimination whatsoever; for vacancieswhere x is high, the matching probability is higher for women and, consistentwith this, lower for men. In conclusion, for jobs and YTS vacancies where there isa high proportion of women job seekers, it is young men that lose out. The onlyevidence of discrimination against young women is for YTS vacancies dominatedby male job seekers.

A dimension of discrimination that has received less attention relates to health.The disabled may be discriminated against because employers focus on the jobseeker’s ‘disability’ rather than their ‘ability’. Similarly, less severe health prob-lems, such as being partially sighted or colour blind may preclude entry to someoccupations. The evidence here suggests that such concerns are unfounded.

Job seekers with a poor social background may also be stigmatised byemployers, especially where they come from council estates with a bad reputation.Involvement in criminal activities may also reduce the likelihood of a job offer.The ‘special training needs’ variable reflects a Careers Service decision that ayouth requires extra help under YTS, and is a composite measure designed tocapture ‘disadvantage’ of these kinds. Our results clearly show that job seekersfrom a disadvantaged background have less chance of obtaining a job, but are notexcluded from participating in YTS.

5.6. Access to YTS programmes and the ‘guarantee’

See Table 2, Employer and Õacancy characteristics.In Section 3, we provided a detailed description of the institutional background

to youth training in the UK. A number predictions were made which are borne outby our results.

First, we can now see that these entry-level jobs are indeed both high and lowŽ .in quality, as reflected by occupation skilled and unskilled , training provision

Ž .norlittle training versus apprenticeship and the hourly wage. Second, it wasclaimed that very small firms are much more likely to partake in the YTS. This isclearly seen in Table 2: the probability of a match is substantially higher for the

Ž .base group that is, the smallest firms with up to 10 employees , with a differentialof about 3 percentage points. Finally, there is evidence of unequal access todifferent types of YTS programmes. Employee programmes are more selectiveŽ .y0.0314 , and special programmes very much more likely to help meet the

Ž .guarantee 0.0906 .

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 355

6. Summary and conclusions

This is the first UK paper to estimate the probability of a match, usingmicroeconomic data relating to both sides of the labour market. Because of therichness of the data, which includes the characteristics of job seekers, vacanciesand firms, we are able to shed light on how the youth labour market operates. Wehave therefore been able to present fresh evidence on a wide variety of issues thathave occupied labour economists for many years.

The main conclusions, which match the issues raised in Section 1, are enumer-ated below. Some of these confirm findings of studies using similar data, asshown.

Ž .1 There is a clear negative, i.e. procyclical, effect of the stock of unemployedŽ .in a labour market on the matching probability for jobs elasticity y0.17 . There

is no equivalent positive effect for the stock of vacancies.Ž . Ž .2 After controlling for the size of the labour market population and area ,

Žlarger Careers Offices have a smaller matching probability elasticity for jobs is.y0.25 .

Ž .3 The majority of wages are set in advance by employers, and are non-nego-Ž .tiable. The effect of the wage is negative for jobs elasticity y0.39 , but

Ž .insignificant for YTS vacancies see also Teyssiere, 1996 .`Ž . Ž .4 a The longer the vacancy has been open, the matching probability is lower

Ž .for job vacancies, but higher for YTS vacancies; b there is a clear ranking of thematching probability by job seekers’ labour-market state: the employed are at the

Ž .front of the job queue and the unemployed are at the back; c for all labour-marketŽ .states, the matching probability increases after 1 or 3 months; c is important

Žbecause any negative duration dependence in vacancy hazards see, for example,.van Ours, 1990 is because the contact rate falls, not because the matching

Ž .probability falls, supporting Coles and Smith 1998 stock-flow matching hypothe-sis.

Ž .5 Employers ‘cream’ the most qualified job seekers in the selection process,which may be regarded as evidence in support of the signalling or job competition

Ž .models see also Teyssiere, 1996; van Ours and Lindeboom, 1996 .`Ž .6 There is clear evidence of racial discrimination. Job seekers from the ethnic

minorities have a lower probability of a match compared to their white counter-Ž . Ž .parts differential of y0.059 for training schemes see also Teyssiere, 1996 .`

However, there is almost no evidence of discrimination in terms of gender andŽ .health see also Russo and van Ommeren, 1998 .

Ž .7 Matching probabilities vary by the type of training offered within the YTS.Also, very small firms offering a training place have a much higher matchingprobability.

One difficulty in interpreting these results arises because the matching probabil-ity is the product of the offer probability by firms and the probability ofacceptance by job seekers. A common assumption is that job seekers are willing to

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357356

accept any job offer for which they apply. While this is plausible in many cases,the probability of a job seeker being offered a training scheme is much higher thanfor a job, simply because one objective of government policy was to mop-up someof the excess supply of youth labour. Nonetheless, when taken together, ourconclusions imply that we are observing the decisions of employers dominatingthose of job seekers.

Acknowledgements

Ž .The authors thank The Leverhulme Trust under grant Fr120rAS for financialassistance. The data were kindly supplied by Lancashire Careers Service. Thecomments of three anonymous referees and Alison Booth are gratefully acknowl-edged, as are those from participants at various presentations. These includeeconomics departments at Manchester and Stirling, and the 1999 Royal Economic

Ž .Society Conference Nottingham . The data used in this analysis are available onrequest.

References

Barron, J., Black, D., Loewenstein, M., 1987. Employer size: the implications for search, training,capital investment, starting wages, and wage growth. Journal of Labor Economics 51, 76–89.

Barron, J., Berger, M., Black, D., 1997. On the Job Training. W.E. Upjohn Institute for EmploymentResearch, Kalamazoo, MI.

Blau, D., Robbins, P., 1990. Job search outcomes for the employed and unemployed. Journal ofPolitical Economy 98, 637–655.

Burdett, K., Wright, R., 1998. Two-sided search with nontransferable utility. Review of EconomicDynamics 1, 220–245.

Burgess, S., 1993. A model of job competition between unemployed and employed job seekers: anapplication to the unemployment outflow rate in Britain. Economic Journal 103, 1190–1204.

Coles, M., Smith, E., 1996. Cross-section estimation of the matching function: evidence from Englandand Wales. Economica 63, 589–597.

Coles, M., Smith, E., 1998. Marketplaces and matching. International Economic Review 39, 239–255.Holzer, H., 1988. Search method use of unemployed youth. Journal of Labor Economics 6, 1–20.Leslie, D., Blackaby, D., Clark, K., Drinkwater, S., Murphy, P., O’Leary, N., 1998. An Investigation of

Racial Disadvantage. Manchester University Press, Manchester.Manning, A., 2000. Pretty vacant: recruitment in low-wage labour markets. Oxford Bulletin of

Ž .Economics and Statistics 62 Supplement 1 , 747–770.Ž .Mortensen, D., 1986. Job search and labor market analysis. In: Ashenfelter, O., Layard, R. Eds. ,

Handbook of Labor Economics, vol. 2, North-Holland, Amsterdam, pp. 849–919.Mumford, K., Smith, P., 1999. The hiring function reconsidered: on closing the circle. Oxford Bulletin

of Economics and Statistics 61, 343–364.Pissarides, C., 1990. Equilibrium Unemployment Theory. Blackwell, Oxford.Russo, G., van Ommeren, J., 1998. Recruitment methods and vacancy duration. Bulletin of Economic

Research 50, 155–166.

( )M.J. Andrews et al.rLabour Economics 8 2001 335–357 357

Singell, L.D., Seaman, P.T., Chatterji, M., 1997. A test of the signalling hypothesis. Department ofEconomic Studies Discussion Paper No. 80, University of Dundee.

Spence, M., 1974. Market Signaling. Harvard University Press, Cambridge, MA.Teyssiere, G., 1996. Matching processes in the labour market: an econometric study. Labour Eco-`

nomics 2, 421–435.Thurow, L., 1975. Generating Inequality. Basic Books, New York.van den Berg, G., 1990. Search behaviour, transitions to non-participation and the duration of

unemployment. Economic Journal 100, 842–865.van Ours, J., 1990. An empirical analysis of employers search. In: Hartog, J., Ridder, G., Theeuwes, J.

Ž .Eds. , Panel Data and Labor Market Studies. North-Holland, Amsterdam, pp. 191–214.van Ours, J., Lindeboom, M., 1996. ASeek and ye shall findB: an empirical analysis of the matching of

job seekers and vacancies. Paper Presented at Labour Market Changes and Income DynamicsConference.

van Ours, J., Ridder, G., 1995. Job matching and job competition: are lower educated workers at theback of job queues? European Economic Review 39, 1717–1731.