Job search methods, intensity and success in Britain in the 1990s. René Böheim. and. Mark P Taylor *) Working Paper No.

Transcription

1 DEPARTMENT OF ECONOMICS JOHANNES KEPLER UNIVERSITY OF LINZ Job search methods, ntensty and success n Brtan n the 1990s by René Böhem and Mark P Taylor *) Workng Paper No July 2002 Johannes Kepler Unversty of Lnz Department of Economcs Altenberger Strasse 69 A-4040 Lnz - Auhof, Austra *) correspondng author: taylm@essex.ac.uk

2 Job search methods, ntensty and success n Brtan n the 1990s René Böhem Department of Economcs Johannes Kepler Unversty Lnz, Austra and Mark P. Taylor Insttute for Socal and Economc Research Unversty of Essex Colchester, Essex CO4 3SQ 17 July 2002 Word count: Abstract: We nvestgate the use of varous job search strateges and ther mpact on the probablty of subsequent employment and the re-employment wage among workng age men n Brtan. We fnd that replyng to advertsements and usng Job Centres are the two most common methods of job search and that job search ntensty, and drect applcatons to employers n partcular, result n a hgher probablty of subsequent employment. Condtonal on fndng work, replyng to advertsements results n hgher payng employment. Age, educaton, famly crcumstances and local labour demand emerge as key determnants of job search strategy use. JEL classfcaton: J11, J20, J62, J64, C33 Keywords: Job Moblty, Unemployment, BHPS, Panel data, Job search Acknowledgements: Thanks to Alson Booth, and partcpants at the TSER/CEPR meetng, Labour Demand, Educaton and Socal Excluson, Florence, Italy (June, 2001), the European Socety of Populaton Economcs Annual Congress, Athens, Greece (June, 2001) and the BHPS 2001 conference for helpful comments. Fnancal support from the ESRC, Unversty Correspondng author: taylm@essex.ac.uk 1

3 of Essex and Leverhulme Trust s gratefully acknowledged. Ths research derves from an Insttute for Labour Research programme on Labour Market Dynamcs n a Changng Envronment funded by the Leverhulme Trust. Introducton Osberg (1993) famously compares job search to fshng. Lke a fsherman, the job seeker uses varous forms of lure and tres dfferent locatons n an attempt to catch the bg fsh. However, the actual process of job search has receved relatvely lttle attenton n the job search lterature, whch has generally focussed on the determnants of the reservaton wage n a framework that assumes the job offer arrval rate to be exogenous (e.g. Narendranathan and Nckell, 1985). The probablty of recevng an offer s lkely to depend on an ndvdual s job search strategy - a greater nvestment n search actvty wll yeld more nformaton on vacances whch s lkely to result n a hgher probablty of recevng a job offer. Job search methods dffer n ther tme and money costs and n ther expected returns, whle ndvduals dffer n ther motves for job search, and ther job search competence and constrants. Dfferent job search strateges wll typcally attract dfferent types of employment. Furthermore, f dfferent strateges draw offers from dfferent pools of potental employers wth dfferent dstrbutons of potental wage offers, then t s desrable for ndvduals to vary ther search effort across strateges as the margnal returns n each strategy wll dffer. An ndvdual s choce of search strategy wll reflect ther perceptons of the costs and benefts assocated wth each method. Our am n ths paper s to nvestgate the use of varous job search strateges among unemployed men n Brtan n the 1990s, and examne ther mpact on the probablty of enterng employment and on the re-employment wage. Recent Brtsh studes have shown that unemployment has scarrng effects on ndvduals. Arulampalam et al (2000) conclude that for mature men some 40% of the observed 2

4 persstence n the unemployment probablty s accounted for by state dependence. Gregg (2001) reaches smlar conclusons a man s prevous unemployment experence has mplcatons for hs future labour market behavour. Gregory and Jukes (2001) and Arulampalam (2001) provde evdence suggestng that unemployment results n earnngs some 10% lower than pre-unemployment earnngs. Ths effect s found to persst. Gregory and Jukes (2001) also fnd unemployment duraton to have a permanent mpact on subsequent earnngs, proportonal to the length of the unemployment spell. The latter n partcular hghlghts the mportance of usng effcent job search methods when unemployed f the loss of current ncome durng unemployment s not to be compounded by earnngs reductons and further scarrng on re-entry to work. Socologsts have lnked the probablty of fndng a job to socal coheson Granovetter (1974) fnds that the majorty of whte collar workers report obtanng ther current job through personal contacts, whle more recently Hannan (1999) concludes that nformal contacts and strong socal networks are mportant n fndng work. Ths has also long been known to economsts. Rees (1966) suggests that good jobs are usually found through nformal nformaton networks and personal contacts. More recently however Gregg and Wadsworth (1996) fnd such effects to be neglgble for the long-term unemployed. Although Pssardes (1979) and Gregg and Wadsworth (1996) focus on the use of state employment agences, and Jones (1989) presents evdence on search ntensty n a varety of search methods, there are few Brtsh studes that explctly ncorporate search method and ntensty nto an econometrc framework. Prevous research has shown that Job Centres and replyng to advertsements are the two most commonly used methods of job search for the unemployed n Brtan (Jones, 1989; Gregg and 2

5 Wadsworth, 1996; Labour Market Trends, 1999). The most common ways of obtanng a job are from hearng from someone who already works n the establshment, and from replyng to an advertsement (Labour Market Trends, 1999). In the U.S. and Canada, drect applcatons to frms and usng frends and famly contacts are more frequent methods of job search, and these are also assocated wth an above average probablty of job search success (Holzer, 1988; Osberg, 1993). Holzer (1988) suggests that employers regard referrals from employees as more nformatve and relable than drect applcatons and use them as a relatvely cheap screenng and sgnallng mechansm, whle Rees (1966) ndcates that employees only refer capable workers to ensure that ther own reputaton wth ther employer s not tarnshed. Evdence suggests that job seekers n Brtan use multple search methods rather than rely on a sngle method strategy. Gregg and Wadsworth (1996) report that on average unemployed ndvduals n Brtan use three job search methods, smlar to the number used by unemployed youth n the U.S. (Holzer, 1988) but more than that recently found for the unemployed n Portugal (Addson and Portugal, 1998). A postve relatonshp between job search ntensty and the probablty of recevng and acceptng a job offer s a common fndng n the lterature (Holzer, 1988; Gregg and Wadsworth, 1996). Wanberg et al (1999) suggest that search ntensty s determned by the degree of fnancal hardshp and commtment to the labour market. An ndvdual s choce of job search strategy wll also reflect employers recrutment polces, whch vary accordng to frm and job characterstcs (Gregg and Wadsworth, 1996; Mannng, 2000). Mannng (2000) fnds that approaches to exstng staff and Job Centres are the most frequently used recrutment methods, but hs sample of employers n Brtan s non-random and hs conclusons cannot be generalsed to the behavour of all Brtsh employers. Roper 3

6 (1988) conducts a detaled analyss of employer varaton n recrutment strategy. The author reports that all formal methods of recrutment are sgnfcantly slower n fllng vacances than Job Centres, and that newspaper advertsements are slowest of all. Informal methods are found to be fastest. The choce of recrutment method has the largest effect on the probablty of fllng a job vacancy. We fnd that replyng to advertsements, usng Job Centres and frends and contacts are the most common methods of job search among unemployed men n the 1990s, whle the average unemployed male n Brtan uses three search methods as part of ther job search strategy. Our estmates show that drect contact wth employers s assocated wth a hgher probablty of subsequent employment, all thngs equal, especally f used n combnaton wth respondng to advertsements, Job Centres or frends and contacts. Condtonal on fndng work, replyng to advertsements results n hgher payng employment. Job search ntensty, as measured by the number of search methods used, has a postve and sgnfcant assocaton wth both the probablty of employment at the subsequent date of ntervew and wth the re-employment wage, all thngs equal. Data Panel data are requred to address the mpact of job search methods and ntensty on the probablty of fndng a job. These enable us to observe the search methods and ntensty of the unemployed at tme t and any subsequent change n employment status between tmes t and t+1. They also allow us to use econometrc technques that control for dfferent ndvdual and household crcumstances. Our data source s waves 6, 7, 8 and 9 of the Brtsh Household 4

7 Panel Survey (BHPS) whch provdes detaled nformaton on ndvdual, household and job related characterstcs on an annual bass from 1996 to The frst wave was desgned as a natonally representatve random sample of the populaton of Great Brtan lvng n prvate (non-nsttutonal) households n the Autumn of 1991, consstng of 5,500 households coverng approachng 10,000 ndvduals. These orgnal respondents have been followed and they and any adult co-resdents are ntervewed at annual ntervals. Chldren n orgnal sample households are also ntervewed when they reach the age of sxteen. The sample therefore remans broadly representatve of the Brtsh populaton as t changes through the decade. Our sub-sample conssts of men who are unemployed for at least one of the dates of ntervew at wave 6, 7, 8, or 9 and who are under 65 years of age at that date of ntervew. We do not nvestgate the job search strateges of women because of small sample szes less than one hundred women consdered themselves to be out of work and actvely searchng for employment at each year. To prevent the possblty of countng one unemployment spell that overlaps two or more consecutve dates of ntervew more than once, we restrct analyss to one unemployment spell per ndvdual we only nclude nformaton from the frst date of ntervew at whch a respondent s unemployed. Our defnton of unemployment s currently not workng, havng looked for work n the past four weeks, and beng avalable to start work wthn the next two weeks. Respondents are not requred to be ntervewed at each wave to reman n the sample, and nor are new entrants to the survey prevented from enterng our sample. At each ntervew, respondents are asked detaled questons relatng to ther current employment status and ther household composton, ndvdual demographcs and ncome. 1 Respondents were not asked about ther job search strateges pror to wave 6. 5

8 From ths we observe whether ndvduals are n work, whether they are out of work and lookng for a job, or whether they are economcally nactve. If n work, respondents are asked for nformaton on a range of job characterstcs and for ther usual labour market earnngs and workng hours, allowng calculaton of an hourly wage rate. To these data we have matched the unemployment rate n each ndvdual s travel-to-work area at each date of ntervew to provde nformaton on local labour demand. 2 The job search questons whch are of prmary nterest here are asked of all those n unemployment at the relevant date of ntervew. In partcular, respondents are asked: In the past four weeks what actve steps have you taken to fnd work? Have you.. Appled drectly to an employer? Studed or repled to an advertsement? Used a Job Centre/employment agency? Asked frends or contacts? Taken steps to start your own busness? Respondents are asked to lst all whch apply. 3 Relatng the answers to ths queston to ndvdual characterstcs and demographcs provdes rch nformaton on the determnants of job search strateges whle unemployed, whle relatng them to labour market status at the subsequent date of ntervew provdes detals regardng the effectveness of varous search methods. In addton, job search ntensty can be estmated by addng the number of search methods used by each unemployed ndvdual. Ths allows nvestgaton of the mpact of search ntensty on employment outcomes. 4 2 The local labour market nformaton s taken from the Natonal Onlne Manpower Informaton Servce (NOMIS), and s matched nto the BHPS by month of ntervew and travel-to-work area. 3 Ths queston s not deal, as there s no other category. There are a small proporton of the unemployed who do not use any of the lsted search methods, and we therefore construct a sxth category to allow for ths. 4 Of course ths s only an approxmaton for search ntensty. It s qute possble for an ndvdual who uses one search method to be searchng more ntensvely for work than another ndvdual usng three or four methods. 6

9 These data on the job search strategy of the unemployed are collected for each unemployed ndvdual at each date of ntervew, rather than at regular perods throughout an unemployment spell. Therefore rather than nvestgatng the mpact of search strategy on the hazard rate from unemployment nto employment, we focus on the employment status at the subsequent date of ntervew of currently unemployed ndvduals. Our estmates can be nterpreted as the mpact of job search strategy on the jont probablty of recevng an acceptable job offer between the dates of ntervew (approxmately 12 months apart), and of remanng n employment untl the subsequent date of ntervew. 5 Table 1 provdes nformaton on the proporton of unemployed men who use each job search method. Ths shows that on average, the most commonly used methods of job search over the perod under consderaton are replyng to advertsements, used by 73% of unemployed men, and job centres and frends and contacts used by 69%. Drect applcaton to employers s used by 62% of unemployed men whle only 12% take steps to start ther own busness. These fgures are consstent wth prevous fndngs for both Brtan (Schmtt and Wadsworth, 1993; Gregg and Wadsworth, 1996; Labour Market Trends, 1999) and France (Sabater, 2000), and contrast wth the U.S. where 80% of the unemployed use drect applcatons and frends and famly contacts (Holzer, 1988), and Canada, where drect applcaton s the most common job search method (Osberg, 1993). Heath (1999) fnds that usng newspapers and the meda Unfortunately the data do not provde nformaton on the number of hours spent searchng or on the costs ncurred. Jones (1989) reports that the average unemployed ndvdual spends sx hours per week lookng for work. Jackman et al (1991) report that unemployed men n Brtan make only one or two job applcatons per month on average. St. Lous et al (1986) argue that the most approprate measure of job search effort s the number of actual job contacts made. Such nformaton s not avalable n the BHPS. 7

10 are the most common search methods among unemployed young Australans, followed by the publc employment servce and drect employer contact. Ths evdence suggests that job search n Brtan s more nsttutonalsed than n other countres (see also Wadsworth, 1991). Table 1 also summarses the number of search methods used, whch we use as our measure of job search ntensty. On average 2% of unemployed men, although currently searchng for work, do not use any of the methods lsted, whle 10% use only one method. 22% of unemployed men use two job search methods, whle the mode s three, used by 36%. 26% use four of the lsted search methods whle 5% use all fve. The mean and medan are 3 search methods per unemployed man. Ths s consstent wth the average of 3 for unemployed men and women n Brtan reported n Gregg and Wadsworth (1996), and of 3.3 for Amercan unemployed youth (Holzer, 1988), and s greater than the mean of 2 found for the unemployed n Portugal (Addson and Portugal, 1998). Job search does not appear to be a sngle, unform actvty for the unemployed seekng work. Table 2 nvestgates these ssues n more detal by lstng the most common search strateges wthn the sample. Ths shows that 23% of the unemployed use a combnaton of drect applcaton to employers, replyng to advertsements, vstng a Job Centre and frends and contacts. A further 34% use some combnaton of three out these four search methods. About 4% of the sample reply to advertsements and vst a Job Centre, whle 6% use Job Centres and frends and contacts. Vstng Job Centres s the most common sngle search strategy, used by 4% of the sample. 5 Most prevous studes of job search success use data on how workers found ther current job. Whle ths provdes accurate nformaton on the successful method of job search, such data may suffer from selecton bases 8

11 Table 3 explores the relatonshp between the elapsed duraton of the unemployment spell and the choce of job search method and ntensty. Ths reveals a negatve relatonshp between drectly applyng to potental employers and elapsed duraton, wth 67% of men unemployed for under 6 months usng ths method compared wth 56% of the long-term unemployed. Men who take steps to start a busness are most lkely to do so ether mmedately on entry nto unemployment, or as a response to long-term unemployment. Ths table also reveals a consstent declne n the number of search methods used wth elapsed unemployment duraton. Men who have been unemployed for a short perod (under 3 months) on average use 3.15 search methods, compared to an average of 2.89 among the long-term unemployed. Ths reducton n search ntensty could reflect ether dsncentve effects or men exhaustng search methods wth the passage of tme. Table 4 examnes the success of job search methods and ntensty by focussng on the employment status at the followng date of ntervew of currently unemployed ndvduals. In partcular t provdes the proportons usng each job search method that are n employment at the next date of ntervew. 6 Ths shows that men who apply drectly to frms are the most lkely to be employed (59% are n work at the subsequent wave), followed by men takng steps to start ther own busness (57%), those that reply to advertsements (52%) and who use frends and contacts (51%). The least successful search method n terms of employment at the (see Korp, 2001, for a summary). 6 Ths does not necessarly mply that ndvduals found employment as a drect result of usng any partcular job search method. We only have nformaton on the methods used at the date of ntervew, and ndvduals may vary ther strategy dependng on ther unemployment duraton. Also, t s possble that an ndvdual may have experenced other employment or even unemployment spells n between ther unemployment spell at t and ther job at t+1. Ths measure however provdes an ndcator of the probablty of fndng a job, and remanng n work, assocated wth each search method. 9

12 subsequent date of ntervew s usng a Job Centre or employment agency. Labour Market Trends (1999) reports that hearng from someone already workng at an establshment and replyng to an advertsement are the two most common ways of fndng a job. Gregg and Wadsworth (1996) smlarly fnd personal contacts, meda and Job Centre use as the most effectve job search methods. These dfferent results may be explaned by dfferent defntons of success. Gregg and Wadsworth (1996) and Labour Market Trends (1999) examne the reemployment probablty, whle we focus on the probablty of employment at a pont n the future. Table 4 also reveals a postve relatonshp between the number of search methods used by the unemployed and the probablty of beng employed at the subsequent date of ntervew. Only 40% of men usng one or two job search methods are subsequently employed, compared to 50% of those usng three methods and 59% of men usng four methods. Three quarters of men usng all fve lsted methods are n work at the followng date of ntervew. These fndngs are consstent wth prevous work (Holzer, 1988; Gregg and Wadsworth, 1996) and mply that greater nvestment n search effort yelds more nformaton on exstng job vacances and results n a hgher probablty of recevng an (acceptable) job offer (Sabater, 2000). 10

13 Estmaton Framework and model specfcaton The frst step n our econometrc analyss s to nvestgate the determnants of the choce of job search strategy used by the unemployed. Search strategy s defned as the combnaton of search method and search ntensty used n lookng for work. The (latent) probablty that an unemployed man () uses a partcular job search method ( M ) can be wrtten: j where for =1,..,n; j=1,..,5. * M = Z γ + v [1] j M =1 f Z γ v > 0 j j j j + j M = 0 f Z γ v 0 j j + j Z s a vector of demographc, household and local labour market characterstcs that determne an ndvdual s propensty to use a partcular search method, s the assocated vector of coeffcents to be estmated and v j s random error. Ths set of (fve) equatons are estmated as probt models. 7 These ndependent equatons for each search method used are complemented wth an ordered probt model to nvestgate the determnants of job search ntensty. Job search ntensty takes a value between 0 and 5, dependng on the number of search methods used, and s also modelled as a functon of demographc, household and local labour market characterstcs. The (unobserved) propensty to use S search methods s specfed as: * S = Z β + u [2] γ j where * S = 0 f S µ 1 * S = 1 f µ 1 S µ 2 < 7 Estmatng a jont model s complcated by the fact that the search method categores are not mutually exclusve, and the wde varety of search strateges used (see Table 2). 11

14 * S = 2 f µ 2 < S µ 3 * S = 3 f µ 3 < S µ 4 * S = 4 f µ 4 < S µ 5 * S = 5 f µ < S 5 The results from these procedures provde mportant nformaton on the determnants of job search strateges used by unemployed men n the 1990s. Our second am n ths paper s to examne the effectveness of dfferent job search strateges. We do ths n two ways, by nvestgatng the mpact of varous search strateges on the probablty of re-employment at the subsequent date of ntervew and, condtonal on reemployment, the mpact of search strategy on the re-employment wage. To estmate the reemployment probablty, we specfy the followng (latent) model: R, + = X ϖ + M α M α + ε [3] * t 1 t 1t jt j where R s the unobserved propensty for an unemployed man at t to be n employment at *, t+1 t+1. Ths s observed as a bnary varable where: * R 1 f R 0, t+ 1 =, t+ 1 > * R 0 f R 0, t+ 1 =, t+ 1 The probablty of recevng and acceptng a job offer depends on a set of exogenous demographc, household and local labour market characterstcs, X t, measured at the same pont n tme as the choce of search method, and on the selected method of job search, M 1... M, j =1,..., 5. Estmatng equaton [3] as a bnary choce model yelds consstent t jt estmates of the parameters of nterest f all explanatory varables are exogenous. However, there s the queston of self-selecton n the use of job search methods to be addressed. If job seekers do (or do not) use a partcular search method because of a common but unobservable characterstc, then the estmated coeffcents wll be based. Job centre use, for example, may 12

15 be an ndcator of the relatve unavalablty of other labour market contacts ndvduals wth good contacts have no need to use job centres. To correct for ths, we decompose the random error term n [3] nto two parts: ε = ε + ε 1 2 where ε 1 = M δ M δ * 1 * j j dentfes the unobservable component that s correlated wth the decson to use a partcular job search method and whch depends on the set of latent varables * * M 1 t... M jt, =1,..., 5 j. ε 2 captures the random error component that s uncorrelated wth the * M s. Although the * M s are unobserved, we replace them wth the estmated probabltes resultng from equaton [1]. 8 The mpact of job search strategy on the re-employment wage s estmated usng an OLS wage equaton. However, ncluson nto ths estmatng sample s condtonal on beng n employment at the subsequent date of ntervew. To control for any selecton bas ths may ntroduce, we estmate a two-step model. The frst step s to estmate a reduced form probt equaton where the dependent varable takes the value 1 f an unemployed man at t s n work at t+1. The dependent varable takes the value 0 f the unemployed man at t s ether not ntervewed at t+1 or s not n work at t+1. We calculate the nverse Mll s rato from ths selecton equaton and enter t as a correcton term n the OLS wage equaton (Heckman, 1979). Our estmated hourly wage equaton therefore takes the followng form: 8 A smlar estmaton procedure has been used by Osberg (1993) and Gregg and Wadsworth (1996). 13

16 Y, + ln h + M k M k + λ ρ + σ [4] t 1 = Y, t+ 1 = a + Wt+ 1 1t jt j where Y 1 s the hourly wage receved at the date of ntervew followng the unemployment, t+ spell, W 1 s a set of demographc, household, local labour market and job related, t+ characterstcs that determne wages, M j, t, ndcates the choce of job search method when unemployed, λ s the nverse Mll s rato, σ s random error, and h, k j and ρ are (vectors of) coeffcents to be estmated. 9 Father s employment status and occupaton and pre-sample nformaton on frst labour market experences are used as dentfyng varables they are used to explan selecton nto the estmatng sample, but not wages condtonal on employment. 10 Model specfcaton The vectors of explanatory varables we use n these analyses cover a range of ndvdual, household and local labour market characterstcs. An ndvdual s age s lkely to partly determne ther number of contacts n the labour market, ther atttude towards rsk, ther fnancal and famlal responsbltes, and ther level of savngs. Labour market moblty s also known to be hgher for younger ndvduals, for whom spells of unemployment are less lkely to have a scarrng effect (Arulampalam, Booth, and Taylor, 2000). Martal status, spouses employment status, the number of chldren and level of educaton are all lkely to determne attachment to and opportuntes n the labour market, job search effcency, the utlty of lesure, the margnal value of ncome, job search constrants and the number of 9 Clearly these coeffcents wll be based f hgher wage-earnng workers select dfferent job search strateges than lower wage-earnng workers. To correct for ths, we have also estmated an nstrumental varables specfcaton, replacng the job search method used wth the probablty of usng each method derved from equaton [1]. The results from dong so dffer lttle from those reported here and are therefore excluded for brevty. 10 The results of the selecton probt are shown n Appendx Table A1 and are not dscussed for brevty. 14

17 contacts n the labour market. More hghly educated ndvduals may have access to a geographcally larger labour market and respond to advertsements placed n the natonal or nternatonal meda, whle the less educated may search more locally through frends and local labour market contacts. Montgomery (1991) suggests that there s a socal structure wthn whch hghly sklled, hgh productvty workers are more lkely to assocate wth each other rather than wth lesser sklled, lower productvty workers. Demographc and famly varables are also lkely to affect both search ntensty and margnal productvty, and therefore affect job offer arrval and retenton rates. The number and age of chldren n the famly, for example, may restrct the employment opportuntes of parents (Wanberg et al, 1999). Household ncome captures the level of fnancal hardshp whch n other studes has been found to determne job search ntensty (Wanberg et al, 1999). It may also determne the probablty of acceptng a job offer. Sgnng on mples a requrement for more vsble, ascertanable job search actvty whch may nduce a shft towards more demonstrable methods. 11 It may also be an ndcator of fnancal hardshp. We nclude these varables n the models determnng job search methods, ntensty, and job search success. A key parameter n the job search lterature s the reservaton wage, the wage at whch an ndvdual s ndfferent between acceptng a job offer and rejectng t n favour of contnued search. The BHPS data allow calculaton of the reservaton wage for each ndvdual unemployed at the date of ntervew, defned n the survey as the lowest weekly take home pay you would consder acceptng for a job. Ths s lkely to be an mportant determnant of both choce of search method and ntensty. By drectly nfluencng the probablty of 11 Sgnng on means regstraton wth the unemployment agency for the recept of unemployment related benefts. 15

18 recevng an acceptable job offer, the reservaton wage also determnes the lkelhood of employment at the subsequent date of ntervew. We mght expect an ndvdual s job search strategy to vary accordng to the length of the unemployment spell, ether because the unemployed change ther strateges as dfferent search methods are exhausted, or because of dsncentve effects. Search effort may declne f unemployed workers contact ther most favourable optons at the start of the spell. Schmtt and Wadsworth (1993) fnd unemployment duraton to be one of the most mportant determnants of job search method choce n Brtan. There s also a consstent fndng n the lterature of negatve duraton dependence (e.g. Nckell, 1979; Van den Berg and Van Ours, 1994; Böhem and Taylor, 2000), ndcatng that the probablty of re-employment falls wth the elapsed duraton of the unemployment spell due to ether scarrng effects or unobserved heterogenety. The current state of the labour market also affects the arrval rate of job offers, and there may be cyclcal dependence n job search strateges. The local level of labour demand wll constran the job seeker, and men may change ther job search behavour n response to dfferent labour market condtons (Osberg, 1993). In depressed labour markets for example, more of one s normal contacts may be unemployed or workng n establshments layng off rather than recrutng workers. McGregor (1983) hypothesses that hgher local unemployment rates ncrease search through advertsements and employment agences, whle job seekers n low unemployment areas are more lkely to use frends and contacts. He argues that nformaton about jobs s more lkely to orgnate from employed workers and therefore less nformaton on jobs wll be avalable n hgh unemployment neghbourhoods. Elapsed unemployment duraton and the local unemployment rate are therefore ncluded as explanatory varables n all models. To capture the mpact of prevous unemployment experence, we nclude a varable ( Number of unemployment spells ) measurng the number 16

19 of unemployment spells each respondent has experenced snce 1/9/ Employers may use an ndvdual s prevous unemployment record as a sgnal of low productvty, or prevous unemployment may otherwse scar a worker (see, for example, Heckman and Borjas, 1980; Arulampalam et al, 2000; Böhem and Taylor, 2002a). We also nclude regon of resdence to capture any spatal dmenson n job search strategy choce and success. Other varables wll nfluence only the choce of search strategy and ntensty, and have no drect mpact on the probablty of re-employment. Wanberg et al (1999) show that commtment to the labour market has a drect mpact on job search ntensty. We capture ths through a varable ndcatng whether an ndvdual has had a spell of economc nactvty n the 12 months before beng observed as unemployed. Ths may determne search ntensty, but s unlkely to ndependently nfluence job search success. On the other hand, havng a health condton that lmts the type or amount of work possble s lkely to affect the probablty of recevng an acceptable offer, but not the choce of job search strategy. Smlarly, the probablty of unemployment has been lnked to housng tenure, wth the relatve resdental mmoblty of socal tenants and owner-occupers hypothessed to ncrease ther propensty to experence unemployment and reduce ther ext rate from t (Oswald, 1996, 1998; Böhem and Taylor, 2002b). However, housng tenure s unlkely to drectly nfluence the choce of job search strategy. Our specfcaton of the re-employment wage equaton s emprcally drven. We nclude a range of ndvdual characterstcs and demographcs and employer, workplace and job 12 We also nclude a varable capturng the number of tmes each ndvdual has been ntervewed to control for the fact that ndvduals have been ntervewed a dfferent number of tmes. 17

20 characterstcs whch have a sgnfcant mpact on the re-employment wage. These nclude age, educaton, occupaton, martal status, spouse s employment status, regon of resdence, housng tenure, job tenure and job type (permanent, seasonal or temporary, fxed term contract) and the sample selecton correcton term. We also nclude the wage receved n the most recent prevous job. Estmaton Results Search method used. Table 5 presents the results from the probt models estmatng the determnants of the choce of search method. The frst set of estmates show that unemployed men aged under 35, and partcularly those aged under 25, are more lkely than those aged 45 and over to apply drectly to frms (by percentage ponts). 13 Schmtt and Wadsworth (1993) report smlar results. Men educated to degree or A -Level standard are more lkely than those educated to below O -Level standard to use ths method of job search, all thngs equal (by percentage ponts). Hghly educated and sklled ndvduals may use a more pro-actve approach to job search, and offer ther sklls drectly to potental employers rather than respond to avalable opportuntes (see also Heath, 1999; Sabater, 2000). An nverse relatonshp emerges between the probablty of drect applcaton and the elapsed duraton of the unemployment spell. Ths form of job search s less lkely among men who have been unemployed for a longer perod ndcatng that ether men use drect applcaton early n the job search process, or that ndvduals who use ths search method fnd a job quckly (see also Schmtt and Wadsworth, 1993). Sgnng on and havng an employed spouse are both assocated wth a sgnfcantly hgher probablty of applyng drectly to frms (by 10 and 15 percentage ponts). 18

21 A negatve relatonshp emerges between applyng drectly to frms and an unemployed man s reservaton wage men wth hgher reservaton wages are less lkely to use ths job search approach, all else equal. The second set of estmates show that unemployed men educated to A -Level or degree standard have a hgher probablty of replyng to advertsements than those holdng qualfcatons below O -Level standard (by 18 and 11 percentage ponts). Schmtt and Wadsworth (1993) and Sabater (2000) report smlar fndngs for Brtan and France. Ths form of job search s also more lkely to be used by ndvduals who are sgnng on (by 17 percentage ponts), perhaps reflectng nsttutonal requrements of vsble and demonstrable job search actvty for the recept of unemployment benefts. The probablty of replyng to advertsements declnes wth the local unemployment rate (see also Schmtt and Wadsworth, 1993), fewer jobs are advertsed durng a recesson. Ths hghlghts the mportance of local labour market condtons n explanng job search behavour. Job Centre use s more prevalent among the young, all thngs equal (see also Osberg, 1993, Schmtt and Wadsworth, 1993, Heath, 1999). Men aged under 25 are 14 percentage ponts more lkely than those aged 45 and over to report usng Job Centres or other employment agences as part of ther job search strategy, whle year olds are 11 percentage ponts more lkely. The probablty of usng a Job Centre or employment agency declnes wth the elapsed duraton of the unemployment spell (see also Schmtt and Wadsworth, 1993), and s hgher for men who sgn on. The latter agan reflects a vsble commtment to fndng work and nsttutonal requrements. 13 These margnal effects are evaluated at the varable sample means. 19

22 The hghly educated are less lkely to use nformal networks as part of ther job search process. The coeffcent on the degree level varable s partcularly large and statstcally sgnfcant, reducng the probablty of usng frends and contacts by 27 percentage ponts relatve to an ndvdual wth no qualfcatons. Ths suggests that the less educated are more lkely to use local nformaton networks and search for work n ther mmedate labour market. The negatve coeffcent on household ncome s consstent wth ths argument. The more sklled, educated and wealthy operate wthn a geographcally larger labour market and are less relant on localsed nformal nformaton networks n lookng for work. Men who sgn on have a hgher probablty than those who do not of usng frends and contacts when searchng for work. Unemployed men educated to degree or A -Level standard have a hgher probablty than those wth no hgher or further educaton qualfcatons of takng steps to start ther own busness (by percentage ponts). Ths could be caused by a greater ablty to dentfy potental busness opportuntes. An nverse relatonshp between takng steps to start a busness and the local unemployment rate emerges. Attempts at busness start up are more common when labour demand s hgh. Search ntensty The results of the ordered probt estmates for job search ntensty, measured by the number of search methods used, are shown n Table We mght expect older workers expected 14 We have also estmated count data models to examne the determnants of search ntensty. The results are smlar to those presented here. 20

23 return from search to be lower gven ther shorter actve labour market future and therefore for a negatve relatonshp between search ntensty and age to emerge. The coeffcents on the age varables are consstent wth ths, although only that on the aged varable s statstcally sgnfcant at conventonal levels. Job search ntensty appears to ncrease wth educaton, although only the coeffcent on the A -Level varable s statstcally sgnfcant. Ths relatonshp s consstent wth prevous research (Blau and Robns, 1990; Schmtt and Wadsworth, 1993; Wanberg et al, 1999; Sabater, 2000). Unemployment deprves sklled ndvduals of ther (hgh) wages and may also deprecate ther human captal. Therefore the hghly educated have a bgger ncentve to ext unemployment rapdly and to adopt a greater search effort than the less educated. Workers wth dfferent skll levels may also search n dfferent labour markets, whch could partly determne ther level of search effort. The number of search methods used s negatvely related to elapsed unemployment duraton the longer the unemployment spell the less ntensvely the unemployed worker searches. 15 Ths could be a dsncentve effect, where ndvduals who have been unemployed for a relatvely long tme are dscouraged from further search. Alternatvely, the decrease n search ntensty over the duraton of the unemployment spell may be caused by ndvduals who exhaust search methods as ther unemployment spell lengthens. Our evdence suggests that job search ntensty s also nversely related to the local unemployment rate the hgher the local unemployment rate, the less ntensely ndvduals search for work. Therefore ndvduals ncrease ther search effort as job competton falls and the probablty of recevng a job offer rses. 16 Indvduals who have had a recent spell of economc nactvty have lower levels of search ntensty, all thngs equal. Note that household ncome has no sgnfcant mpact on job 15 Schmtt and Wadsworth (1993) report smlar fndngs for unemployed men n Brtan n the early 1980s. 16 Jones (1989) and Wadsworth (1991) report smlar fndngs. 21

24 search ntensty. Ths suggests that, all thngs equal, the level of fnancal well-beng does not nfluence the job search ntensty of an unemployed worker. However, our results also suggest that job search ntensty s postvely related to sgnng on, whch may ndcate fnancal hardshp and also reflect nsttutonal factors. Probablty of re-employment The results from the models estmatng the probablty of employment at t+1 gven that an ndvdual s unemployed at t are presented n Table Our estmates show that applyng drectly to an employer ncreases the probablty of employment at the subsequent date of ntervew by 27 percentage ponts all else equal. Ths suggests that searchng for a job through makng drect applcatons has a very large postve mpact on an ndvdual s medum term employment prospects. 18 Gregg and Wadsworth (1996) also report a postve (although smaller) effect of drect contact on the re-employment probablty n Brtan, as do Osberg (1993), Addson and Portugal (1998) and Sabater (2000) for Canada, Portugal and France respectvely. It can be argued that applyng to potental employers s the fnal stage of the job search process, and therefore ts postve relatonshp wth the probablty of reemployment s not surprsng. However, our ndcator of re-employment s measured approxmately 12 months later, and s therefore less senstve to ths potental bas than, for example, duraton models. We argue that applyng drectly to frms has a postve mpact on the medum term employment prospects of the currently unemployed. 17 Note that there s no omtted search category. Ths s because the categores are not mutually exclusve. 18 Interactng job search method wth most recent occupaton shows that ths result holds for both manual and non-manual workers. However, the quanttatve mpact s about twce as large for non-manual workers, ncreasng the probablty of subsequent employment by about 40 percentage ponts compared to 20 percentage ponts for manual workers. We have also tred nteractng search method and ntensty wth age to nvestgate whether any dfferental re-employment effects emerge for younger workers. No sgnfcant dfferences were found. 22

25 Usng frends and contacts also has a postve, although more modest and statstcally nsgnfcant mpact. Replyng to advertsements, usng a job centre/employment agency and takng steps to start a busness, however, reduce the probablty of beng employed at t+1, although these effects are not statstcally sgnfcant. 19 The fndng that Job Centre use s not an effectve job search method s consstent wth prevous research. For example, Welgosz and Carpenter (1987), usng U.S. data, conclude that almost all methods of job search are assocated wth sgnfcantly shorter duratons of search when compared to the state employment servce. Osberg (1993) and Sabater (2000) report a negatve relatonshp between publc employment agency use and the probablty of fndng a job for Canada and France respectvely. However, Gregg and Wadsworth (1996), usng Brtsh data, report that the use of Job Centres s assocated wth a hgher than average probablty of re-enterng work. The dfferences between our results and those of Gregg and Wadsworth are not nconsstent and can be explaned by the dfferent defntons of the dependent varables. The dependent varable n the Gregg and Wadsworth study s the probablty of re-employment across a three month perod whle our dependent varable s the probablty of beng employed approxmately one year n the future. Combnng these results suggests that although Job Centres may ncrease the short run probablty of re-employment, ndvduals are no more lkely to fnd themselves n employment n the medum term. Ths mples that ether the jobs people fnd through Job Centres are of low qualty wth relatvely hgh rates of destructon, or 19 These fndngs are robust to ncludng the job search method dummy varables n ndependent equatons. It s possble that these results reflect the tme spent n each job search method. Holzer (1988) for example fnds that young unemployed Amercan men spend more hours searchng through frends and relatves and drect employer contact than through state employment agences or newspapers. We have no nformaton on ths. 23

26 that Job Centres are poor at matchng unemployed workers wth sutable jobs. Ths s a potental avenue for future research. The second specfcaton suggests that job search ntensty, as measured by the number of search methods used, has a postve and statstcally sgnfcant mpact on the probablty of employment at the subsequent date of ntervew, holdng other characterstcs constant. At the sample means, usng one addtonal job search method (.e. usng 4 methods rather than 3) ncreases the probablty of subsequent employment by 7 percentage ponts. Ths s consstent wth prevous studes (Holzer, 1988; Gregg and Wadsworth, 1996; Sabater, 2000). 20 Table 8 and Table 9 nvestgate the mpact of varous job search strateges on the probablty of employment at the subsequent date of ntervew. Table 8 focuses on the mpact of usng each search method, ether alone or n combnaton wth others, relatve to not usng that method. The estmates suggest that combnng drect applcaton wth one or more other search methods sgnfcantly ncreases the probablty of subsequent employment relatve to not usng drect applcaton. The use of other search strateges has lttle statstcally sgnfcant mpact on the probablty of employment 12 months later. Table 9 examnes the mpact of search strateges nvolvng drect applcaton to employers n more detal. Ths shows that applyng drectly to potental employers has the largest quanttatve mpact on the subsequent employment when used together wth replyng to advertsements and frends and contacts, ncreasng the probablty of employment by 46 percentage ponts, and wth just 20 A comparson of the log-lkelhoods of the two specfcatons suggests, however, that the search method used s more nformatve than an aggregaton. The LR test statstc s calculated as 15.6 wth χ 2 (4)=

27 replyng to advertsements (ncreasng the probablty of subsequent employment by 36 percentage ponts). Re-employment wage Fnally we examne whether the job search strategy used by unemployed workers determnes the qualty of subsequent employment measured by the hourly wage. Table 10 presents the selectvty corrected OLS estmates of the natural log of the usual hourly wage at the subsequent date of ntervew, wth search method and ntensty when unemployed as explanatory varables. The results show that usng a Job Centre or other employment agency, drect applcaton to employers, frends or labour market contacts or steps to start a busness as part of a job search strategy have small and statstcally nsgnfcant mpacts on the wage receved at the subsequent date of ntervew, all thngs equal. The coeffcents on these varables are poorly determned. However, replyng to advertsements has a relatvely large, postve and well determned effect on the hourly wage subsequently receved. The coeffcent suggests that replyng to advertsements whle unemployed results n subsequently recevng approxmately 26% hgher earnngs. Therefore usng market methods to seek work are more successful n the sense of ganng relatvely hghly pad employment. The wage receved at the subsequent date of ntervew also ncreases wth the number of search methods used when unemployed. Therefore job search ntensty not only ncreases the probablty of subsequent employment, t also ncreases the subsequent wage, perhaps because search ntensty ncreases the (unobserved) offer arrval rate, provdng ndvduals wth a wder draw from the wage offer dstrbuton. 25

28 Conclusons The job search strateges used by unemployed ndvduals and ther contrbuton to the job matchng process s crucal to understandng ndvduals labour market behavour. We contrbute to ths level of understandng by examnng the determnants of job search strateges, and the mpact these strateges have on subsequent employment and wages. Ths s mportant as dfferent job search strateges wll typcally attract dfferent types of employment and draw offers from dfferent pools of potental employers wth dfferent dstrbutons of potental wages. Furthermore, there s lttle work n the Brtsh lterature that explctly ncorporates job search method and ntensty n the analyss of labour market behavour. We fnd that replyng to advertsements and usng Job Centres or employment agences are the two most common methods of job search, whle the average unemployed man n Brtan uses three search methods as part of ther job search strategy. The most common strategy nvolves a combnaton of drect applcaton, replyng to advertsements, vstng a Job Centre and frends and contacts, used by 23% of unemployed men. Age, educaton and famly crcumstances emerge as key determnants of whch job search strategy ndvduals use. Our estmates show that applyng drectly to potental employers sgnfcantly ncreases the probablty of beng employed at the subsequent date of ntervew, partcularly when used n combnaton wth replyng to advertsements and frends and contacts. Therefore the most common methods of job search used by unemployed men do not correspond to the most successful n terms of the probablty of subsequent employment. Ths suggests that polces amed at returnng the unemployed to work should focus on mprovng specfc job search sklls. Replyng to advertsements results n hgher payng employment, all thngs equal. Job 26

29 search ntensty, as measured by the number of search methods used, s postvely related to both the probablty of employment at the subsequent date of ntervew and, condtonal on workng, a hgher wage. Nevertheless, t appears that the choce of search method s more mportant than search ntensty. Local labour demand s an mportant nfluence on the choce of job search strategy. In partcular, unemployed ndvduals lvng n areas of low labour demand search less ntensvely than those n areas of hgh labour demand. It s therefore mportant to mprove job search effectveness of the unemployed n areas of hgh unemployment f the problem of persstent joblessness s not to deterorate further, and f unemployment s not to become more spatally concentrated. Our analyss reveals sgnfcant dfferences n job search strateges between ndvduals, and furthermore that the choce of job search strategy nfluences the probablty of re-enterng employment. We however focus only on the ndvdual job seeker. Further research s requred to ad understandng of the job search and matchng process, ncorporatng demand sde factors such as how recrutment strateges vary across vacancy type. 27

30 References Addson, J.T. and P. Portugal (1998), Job search methods and outcomes, ZEW Dscusson Paper 98-41, Mannhem: ZEW. Arulampalam, W. (2001), Is unemployment really scarrng? Effects of unemployment experences on wages, Economc Journal, 111 (475), pp.f585-f606. Arulampalam, W., A.L. Booth and M.P. Taylor (2000), Unemployment persstence, Oxford Economc Papers, 52, pp Blau, D.M. and P.K. Robns (1990), Job search outcomes for the employed and unemployed, Journal of Poltcal Economy, 98, pp Böhem, R. and M.P. Taylor (2000), Unemployment duraton and ext states n Brtan, ISER workng paper number , Colchester: Unversty of Essex. Böhem, R. and M.P. Taylor (2002a), The Search for Success: do the unemployed fnd stable employment?, Labour Economcs (forthcomng). Böhem, R. and M.P. Taylor (2002b), Ted down or room to move? Investgatng the relatonshps between housng tenure, employment status and resdental moblty n Brtan, Scottsh Journal of Poltcal Economy (forthcomng). Granovetter, M. (1974), Gettng a job: A study of contacts and careers, London: Havard Unversty Press. Gregg, P. (2001), The mpact of youth unemployment on adult unemployment n the NCDS, Economc Journal, 111 (475), pp.f626-f653. Gregg, P. and J. Wadsworth (1996), How effectve are state employment agences? Jobcentre use and job matchng n Brtan, Oxford Bulletn of Economcs and Statstcs, 58(3), pp Gregory, M. and R. Jukes (2001), Unemployment and subsequent earnngs: Estmatng scarrng among Brtsh men , Economc Journal, 111 (475), pp.f607-f625. Hannan, C. (1999), Beyond networks: socal coheson and unemployment ext rates, Insttute of Labour Research Workng Paper 99/28, Colchester: Unversty of Essex. Heath, A. (1999), Job-search methods, neghbourhood effects and the youth labour market, Research Dscusson Paper , Reserve Bank of Australa. Heckamn, J.J. (1979), Sample selecton bas as a specfcaton error, Econometrca, Vol. 47, pp Heckman, J.J. and G. Borjas (1980) "Does unemployment cause future unemployment? Defntons, questons and answers from a contnuous tme model for heterogenety and state dependence", Economca, 47, pp Holzer, H.J. (1988), Search method use by unemployed youth, Journal of Labor Economcs, Vol. 6 (1), pp Jackman, R., R. Layard and S. Nckell (1991), Job search: The duraton of unemployment, Chapter 5 n Unemployment, macroeconomc performance and the labour market, Oxford: OUP. Jones, S. (1989), Job search methods, ntensty and effects, Oxford Bulletn of Economcs and Statstcs, Vol. 51, pp Korp, T. (2001), Good frends n bad tmes? Socal networks and job search among the unemployed n Sweden, Acta Socologca, 44, pp Labour Market Trends (1999), Methods of job search, January, pp Mannng, A. (2000), Pretty vacant: Recrutment n low-wage labour markets, Oxford Bulletn of Economcs and Statstcs, Vol. 62, Specal Issue, pp