Volume 30, Issue 4. Filling vacancies: Identifying the most efficient recruitment channel

Transcription

1 Volume 30, Issue 4 Fllng vacances: Identfyng the most effcent recrutment channel Mareva Sabater Unversty of Savoe Abstract Ths paper provdes new results on the recrutment behavour of frms by showng that recrutment channels clearly affect the probablty of fllng a vacancy. Usng a French data, we estmate the effectveness of dfferent recrutment strateges, takng nto account that recrutment channels choces are endogenous and that frms strategcally combne several channels. Our results enable us to conclude that nsttutonal ntermedares, such as prvate and publc employment agences, are found to be the most effectve channels for frms. However, these ntermedares are specalzed, n that prvate agences are more effcent at fllng sklled vacances, whereas publc agences are more effcent at fllng non-sklled vacances. We would lke to thank DARES for allowng us access to the OFER survey, Stephen Bazen for helpful comments earler drafts of ths paper. Ctaton: Mareva Sabater, (2010) ''Fllng vacances: Identfyng the most effcent recrutment channel'', Economcs Bulletn, Vol. 30 no.4 pp Submtted: Oct Publshed: December 14, 2010.

2 1-Introducton The search strateges used by jobseekers, and the effectveness of these strateges n terms of reducng unemployment duraton, have been extensvely studed (Mortensen, 1986; Devne and Kefer, 1993). However, lttle s known about the demand sde of recrutment, that s to say, about employers search behavours, manly because mcro-data are rather poor or absent. Recrutment occurs at the end of a matchng process (Mortensen and Pssardes, 1994) n whch both frms and jobseekers develop strateges to meet a potental partner. Hence, the use of napproprate search strateges by employers and/or by prospectve employees can lead to neffcences n the matchng process and to the smultaneous exstence of unemployment and unflled vacances. It s therefore crucal to determne how an employer s search behavour affects the outcome of the recrutment process. Snce Holt and Davd s (1966) poneerng artcle, emprcal analyses of employers search strateges have prmarly focused on the mpact of the search effort, n partcular, on the tme spent searchng and on the number of applcatons receved (Barron and Bshop, 1985; Barron et al., 1985). In ths statc vew, the greater the search effort, the hgher the number of applcants. Search effort s also found to have a postve effect n a dynamc vew, that s to say, on vacancy duraton (Beaumont, 1978, Roper, 1988, Van Ours, 1989). Moreover, several channels (for example, socal networks, advertsements, publc and prvate employment agences) can be used to locate potental employees and to encourage them to apply (Holzer, 1987). As recrutment channels are nformaton vectors, they nfluence the probablty of recevng applcatons (Russo et al., 2000) and the cost of the search (Russo et al., 2005). Roper (1988), usng Brtsh data, concludes that search channels strongly nfluence vacancy duratons. Networks seem to be the most effcent channels for hrng, whereas advertsements, although they are wdely used, tend to result n long vacancy duratons. Conversely, Holzer (1987) and Van Ours and Rdder (1993), usng US and Dutch data respectvely, found that advertsements are the best way to rapdly fll a job vacancy. Ths lack of consensus on the most effcent recrutment channel may be due to the fact that prevous studes have always treated the choce of recrutment channel as exogenous, despte a substantal body of evdence ndcatng ths s not the case. The choce of recrutment channel(s) depends on the characterstcs of the vacancy, on the employer s capacty to use a gven channel and on the prevalng economc stuaton. Gorter et al. (1996, 1993) showed that the choce of recrutment channel also depends on the type of employees needed by the frm, as employers seekng qualfed staff are more lkely to use advertsements to fll vacances, whereas employers seekng personnel n the second segment of the labour market (.e., fxed-term, part-tme and/or low-qualfed jobs, see Doernger and Pore, 1971) are more lkely to use local employment agences. Russo et al. (2000) confrmed these results and also concluded that employers are more lkely to use advertsements when competton between job seekers s low. In addton, employers may combne several channels n order to ensure a hgher return from searches (Holzer, 1996; Gorter and van Ommeren, 1999). As employers generally use a combnaton of recrutment channels, rather than one partcular channel, search strateges can be defned n terms of the type and the number of recrutment channels used. Channel choces should be treated as correlated. Bessy et al. (2007) ponted out that, n France, frms frequently combne several methods to fll vacances. Ths can be explaned by two characterstcs of the French labour market. Frst, frms must notfy the government employment agency (Agence Natonale Pour l Emplo, ANPE) when a vacancy occurs and the ANPE must nform job seekers of that vacancy. Second, unsolcted applcatons are used much more frequently by French jobseekers than they are by jobseekers n other countres:

3 94% of French employers receve at least one unsolcted applcaton per year (Bessy et al., 2007). Consequently, when analysng the success or falure of a recrutment strategy, t s mportant to examne the overall strategy adopted by an employer, that s to say, the type(s) of recrutment channels chosen and how these recrutment channels were combned, rather than to smply examne each ndvdual recrutment channel that was used. The present study dffers from prevous research n three fundamental ways. (1) It focuses on the combnaton of recrutment channels and the lnks between the channels chosen, rather than treatng each recrutment channel separately. (2) It nvestgates the factors determnng the choce of recrutment strategy. (3) It assesses the mpact of these search strateges, treated as endogenous, on the probablty of fllng a vacancy. Our study was based on data from a Job Offer and Recrutment (Offre d emplo et recrutement OFER) survey carred out by the French Employment Mnstry s research and statstcs department (Drecton de l Anmaton de la Recherche des Etudes et des Statstques - DARES). Ths survey provdes data on the search channels used by employers, on the reasons for the vacancy and on the success of the recrutment process. By applyng smulated lkelhood estmaton technques (Cappellar and Jenkns, 2006), we were able to estmate a multvarate probt model that allowed us to determne the effect of the recrutment channel used on the probablty of fllng a vacancy, takng nto account the lnk between the choce of channels and ther endogenety. We begn by presentng our dataset. Ths s followed by a dscusson of our dataset, the presentaton of the emprcal results and our conclusons. 2- Data Our data was taken from a job offer and recrutment ( Offre d'emplo et Recrutement - OFER) survey of employers recrutment behavour that was carred out between January and July 2005 by the French Employment Mnstry s research and statstcs department (Drecton de l Anmaton de la Recherche des Etudes et des Statstques - DARES). The survey looked at French frms that had begun a process to recrut at least one employee durng the 12 months pror to the survey. The OFER survey collected data from 4004 frms, each of whch provded nformaton about the vacancy, the frm s characterstcs and the recrutment procedures used (see Appendx 1 for more detals). Of the observed frms, 88% successfully recruted at least one employee and 12% faled to fll the vacancy. In 17.8% of cases, recrutment was for a sklled poston. The OFER dataset also ndcates whch recrutment channels were used durng the search process. Recrutment channels can be classfed nto fve categores: Market methods, such as advertsements and drect applcatons, ncludng va Internet ; Publc employment agences ; Prvate employment ntermedares, such as prvate agences, schools, professonal assocatons, etc. ; Professonal relatons; and personal relatons (frends and relatves). Descrptve statstcs underlne three nterestng facts. Frst, only 27% of frms reled upon a sngle recrutment channel: the combnaton of several channels s the most used strategy. The fve man search channels avalable to frms can be combned n 31 dfferent ways (see appendx 2). But, among the 31 possble strateges, 11 are used by less than 1% of the frms, that s less than 40 frms. Fnally, observed sub-samples on each strategy are often very small. Second, our data also ndcate that the choce of search channels s nfluenced by the characterstcs of the employer frm, the vacancy and the macroeconomc context. For

4 example, small frms seem to prefer to base ther recrutment strateges on personal relatons, whereas large frms preferred to use prvate employment agences. Ths channel was also the most frequently used when the vacancy concerned an executve poston. Thrd, success rates vary accordng to the macroeconomc context, to frms attrbutes and to the recrutment strategy used. On the opposte of prevous results, whch ndcated that combnng recrutment channels was more effectve than usng a sngle channel (Holzer, 1996; Gorter and van Ommeren, 1999), our descrptve statstcs hghlght more that the recrutment strategy success rates appear to be nfluenced more strongly by the types of channel used than by the number of channels used. Among the 31 dfferent strateges, fve of the most commonly used strateges gave partcularly hgh success rates. Three of these strateges were based on a sngle channel, that s to say, professonal networks (94% success rate), personal networks (95.5% success rate) or market methods (92% success rate). However, even hgher success rates were recorded for two combnatons of two channels: professonal plus prvate networks (96% success rate) and prvate agences plus professonal networks (93.9% success rate). 3- Emprcal results Descrptve statstcs on the OFER data set confrm that employers combned several recrutment channels to fll ther vacances, but not all the combnatons had smlar success rates. Among the 31 possble strateges, some are scarcely used. Ths stops us from drectly estmatng the strateges choces. To overcome ths dffculty, we choose to focus attenton not on strateges choce but on recrutment channels ones, takng nto account ther potental combnaton. We thus jontly estmate the channels choce equatons and the recrutment success equaton, that s the followng system of equatons: Mark_meth = β1' X + ε1 Pub_ag = β3' X + ε 3 Prv_ag = β 2 ' X + ε 2 Prof_netw = β 4 ' X + ε 4 Pers_netw = β5 ' X + ε 5 Recrut_ok = γ ' X + µ where s the frm. The frst fve equatons descrbe the choce of the fve search channels: market methods (mark_meth), publc agences (pub_ag), prvate agences (prv_ag), professonal relatons (prof_netw) and personal relatons (pers_netw). The fnal equaton defnes the probablty of success (recrut_ok). The error terms ε and µ have the followng propertes: E ε j = E µ =, ( ε ) = Var( µ ) = 1 Corr ε ε = ρ ( ) ( ) 0 Var j, ( j, k ) jk and Corr ( ε j µ ) = ρ jµ,. The correlaton coeffcents between the errors of the fve choce equatons are denoted ρ jk. If ρ 0, channel choces are nterdependent. Conversely, f ρ = 0, channel choces are jk ndependent. Smlarly, the µ ρ j coeffcents ndcate the lnk between the choce equatons and the recrutment success equaton. We then have to estmate a system of sx equatons. In each equaton, the dependant varable s a dummy varable. We used a multvarate probt model to estmate the equaton system. However, Tran (2003) showed that probt probabltes do not have a closed-form expresson and must be estmated numercally. In the present study, we used Cappellar and Jenkns (2006) smulated maxmum lkelhood technque to estmate our probt probabltes. jk

5 As correlaton terms between recrutment channels choces are estmated, we can calculate jont probabltes correspondng to the 31 possble combnatons of choosng or not each channel. Estmatng the fve equaton system s then equvalent to studyng all the strateges (Greene, 2008). But, the multvarate probt model has a man advantage: estmated probabltes of usng each strategy can be deduced from the estmated choces of each recrutment channel and from the correlaton terms. Thus, these probabltes can be computed even f some strateges are scarcely used. Moreover, the multvarate probt model allows us to estmate the global probablty of choosng a gven channel, takng nto account that t s combned. Besdes, the X vector ncludes all the observed attrbutes that can nfluence the choce of recrutment channels and also the probablty of fllng a vacancy. Followng Wlde (2000), we suppose that dentfcaton varables are not needed as a large varablty s observed n explanatory varables. These factors can be ndeed dvded nto fve man types: characterstcs of the frm: sze, sectors, fnancal health, attrbutes of the job vacancy: sklled or unsklled, permanent or temporary, part-tme or full-tme; reason for recrutment: replacement, new actvty, ncrease n demand, nternal reorgansaton; search context: prevous recrutment channels used, number of drect applcatons receved; macroeconomc ndcators (taken from Enquête Emplo, INSEE and ANPE, 2005). Table 1 presents the results of ths estmaton. Let us begn to comment the results about the recrutment channel choces. Frst, a number of correlaton terms ( ρ ) between the dfferent recrutment channels used by employers are jk sgnfcantly dfferent from zero. For example, the use of professonal networks postvely correlated wth the use of personal relatons. Ths shows that users of strong relatons have unobserved characterstcs that are postvely related wth those of users of weak relatons. Postve correlatons were also found between the use of publc employment agences and prvate employment agences, and between the use of market methods and publc agences. On the other hand, all the correlatons between the use of prvate agences and market methods or personal networks were negatve. Our results also show that choces of recrutment channels are nterdependent, thus confrmng the mportance of consderng an employer s overall search strategy, rather than, as s often the case n the lterature, the ndvdual channels used. In addton, the results ndcate that the choce of recrutment channels depends not only on the exstence of other search channels, but also on a frm s characterstcs, the type of vacancy and the state of the job market. For example, small frms tend to favour the personal relatons channel, whereas larger frms are more lkely to favour prvate agences and professonal relatons. Ths suggests that t s more dffcult for small frms to use ether expensve recrutment channels, such as prvate agences and advertsements, or to explot professonal networks, whch are probably less well developed than n large frms. A frm s choce of recrutment channels s also lkely to be nfluenced by the busness sector n whch t operates. For example, the recrutment channels most commonly chosen by ndustral frms are prvate and publc agences. Frms n the constructon sector tend to favour prvate employment agences and personal networks, whereas frms n the servce sector tend to favour market methods. Ths pattern may reflect the fact that France s prvate employment agences have a long tradton of ndustral and constructon-sector recrutment but they have only recently begun focusng on other busness sectors. Employment agences not only gve access to large applcant pools, they can also provde help n screenng applcatons (Lesueur, 1997). As a result, many ndustral frms have bult up long-term relatonshps (ether nformal

6 or contractual) wth one or more agences. In addton, econometrc results show that frms that have used prvate or publc agences n the past are more lkely to use them agan n the future. Ths may ndcate the exstence of learnng effects n the use of search methods, whch can reduce search costs. Our results also hghlght that the macroeconomc context affect only the choce of market methods and publc agences as recrutment channels. When the competton between job seekers s low, frms are more lkely to use advertsements, n the lne of the conclusons of Russo et al. (2000). But, when the competton s ntense, French frms prefer to contact the publc ntermedary, manly because the pool of potental applcants s then larger.

7 Table 1: Multvarate probt estmates of recrutment channels choces and the probablty of fllng a vacancy Professonal networks Publc Prvate Market Personal agences agences methods networks Recrutment success Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t constant *** *** * *** *** Sze Small Ref. Ref. Ref. Ref. Ref. Ref. Medum ** *** *** *** Large *** *** *** * Sector Agrculture *** * Constructon *** ** ** *** Industry *** ** *** *** * Servces Ref. Ref. Ref. Ref. Ref. Ref. Economc dffcultes * * Reason for recrutment Replacement * *** *** New actvty * ** *** Demand ncrease *** *** ** Reorgansaton Ref. Ref. Ref. Ref. Ref. Ref. Vacancy characterstcs Executve poston ** *** *** Temporary poston ** Part-tme poston *** *** Urgent recrutment *** Prevous search channels Publc agences *** *** *** ** Prvate agences *** *** Number of drect applcatons receved *** *** Macroeconomc ndcators Number of jobseekers *** *** ** Number of job offers *** ** ***

8 Table 1: Correlaton coeffcents for choce equatons Coef. t rho21: Professonal networks Publc agences rho31: Professonal networks Prvate agences rho41: Professonal networks Market methods rho51: Professonal networks Personal networks *** rho32: Publc agences Prvate agences *** rho42: Publc agences Market methods ** rho52: Publc agences Personal networks rho43: Market methods Prvate agences *** rho53: Personal networks Prvate agences *** rho54: Personal networks Market methods Correlaton coeffcents between choce equatons and recrutment success equatons Coef. t rho61: Recrutment success Professonal networks * rho62: Recrutment success Publc agences *** rho63: Recrutment success Prvate agences * rho64: Recrutment success Market methods *** rho65: Recrutment success Personal networks Number of observatons 4004 Log-Lkelhood Wald test ch2(96) *** NB: Heteroscedastcty was corrected usng the method advocated by Whte (1982). ***: sgnfcant at the 1% level, **: sgnfcant at the 5% level, *: sgnfcant at the 10%level

9 The choce of recrutment channels s also nfluenced by the characterstcs of the vacancy. Frms tend to use prvate agences to fll sklled postons and to use publc agences to fll less-sklled postons. Ths result, whch s n lne wth the results obtaned by Gorter et al. (1996), reflects the fact that French prvate and publc employment agences have tradtonally specalzed n recrutng for dfferent knds of job. Publc agences attract more low-sklled job seekers, whereas prvate agences focus on job offers for sklled people. Thus, frms make strategc choces when decdng whch recrutment channels to use, selectng the channels that are most lkely to gve access to sutable applcants. The results also ndcate that frms recrutng to fll a new poston tend to use employment agences (prvate or publc) because they can provde help wth the screenng and selecton process. On the other hand, frms seekng a new employee to fll an exstng poston can drectly and effcently use market methods or networks, as t s relatvely easy for the frm to defne the characterstcs of the vacancy and therefore t does not need help wth evaluatng and selectng applcants. Fnally, because usng agences s costly (manly n terms of coordnaton), these channels are less lkely to be used when the vacancy must be flled rapdly. In addton, by estmatng a multvarate probt model we were able to analyse the determnants of the probablty of fllng a vacancy. Ths probablty s lower f the frm has economc dffcultes, whch may mean that frms wth fnancal problems fnd t dffcult to attract applcants. However, t may also be due to such frms havng less tme and resources to allocate to the search process. Recrutment success rates are also lower when fllng vacances for newly created posts. In such cases the lower probablty of success may be due to frms fndng t more dffcult to defne and descrbe ther needs and to adopt the best strateges for contactng sutable applcants. Success rates are also lower for frms n the constructon sector and for frms hrng executve-level staff. Ths may reflect structural recrutment dffcultes n partcular segments of the labour market. For example, there s ntense competton for labour n the French constructon ndustry, as ths sector s not attractve due to ts hard workng condtons and relatvely low wages. As a result, there s a lack of sklled applcants. More generally, t s more dffcult to recrut sklled personnel n France than unsklled workers. Ths seems to be confrmed by the fact that macroeconomc condtons have a substantal affect on the probablty of fllng a vacancy. The recrutment process s dffcult when competton among frms s hgh and easer when a lot of jobseekers are avalable, as stressed n matchng models (Mortensen and Pssardes, 1990). However, the most nterestng fndng s that there s a correlaton (correlaton coeffcents sgnfcantly dfferent from zero) between the success equaton for the recrutment procedure and all of the recrutment channel equatons except for the personal network channel equaton. Hence, we can conclude that the probablty of fllng a vacancy s strongly lnked to the choce of recrutment channels. We used the procedure developed by Cappellar and Jenkns (2006) to estmate our multvarate model, whch allowed us to calculate the probablty of fllng a vacancy for each recrutment channel 1. These estmated probabltes take nto account the effect of explanatory factors n the choce equatons, the lnks between recrutment choces and the lnks between the choce equaton and the success equaton, thereby ensurng an unbased estmaton of the real effect of recrutment channels on recrutment success rates. Table 2 gves estmated condtonal-success probabltes for the sample as a whole and for two sub-samples (vacancy for an executve poston, vacancy for a non-executve poston). 1 These condtonal probabltes take nto account the estmated correlaton coeffcents between the fve channels. Alternatvely, the estmated success rate assocated to each of the 31 recrutment strateges could also be computed. Ths gves smlar results (results avalable upon request). -8-

10 Recrutment channel Table 2: Estmated probabltes of fllng a job vacancy for each of the fve possble recrutment channels Percentage of vacances flled Estmated probablty of fllng the vacancy Estmated probablty of fllng the vacancy Executve poston Estmated probablty of fllng the vacancy Non-executve poston Market methods Publc agences Prvate agences Professonal networks Personal networks Table 2 compares the observatons drawn from the descrptve statstcs wth the estmated success rates for the dfferent recrutment channels. Ths comparson ndcates that success rates are affected by the endogenety of the channels and the lnks between them. Frms usng professonal networks successfully flled 87.8% of vacances, suggestng that ths s the most effectve recrutment channel. However, ths apparent effectveness s the result of selecton bas, as the frms that use professonal networks are those that have the best chance of successfully recrutng staff. When ths bas s corrected, the most effectve recrutment channel can be seen to be prvate employment agences. As Houseman et al. (2003) showed for temporary employment agences, prvate agences facltate the hrng process by screenng and selectng jobseekers. Prvate agences also reduce recrutment costs and speed up the matchng process, thereby producng economes of scale and ncreasng the effcency of a frm s search process. Ths more detaled analyss also shows that publc employment agences are not as neffectve as suggested by the descrptve statstcs. Although they obtaned the lowest success rate when consderng all types of vacancy, ths s due to ther low success rate n fllng sklled vacances. In contrast, they provde a very effectve recrutment channel for fllng non-sklled vacances. The results also provde an nsght nto the specalzaton of French prvate and publc employment agences. As prvate and publc agences gve access to dfferent segments of the labour market, and thus to dfferent types of sklls, frms strategcally choose them accordng to ther needs. Frms seekng to fll a vacancy for a sklled poston tend to favour prvate agences, whereas frms seekng unsklled staff or frms that can beneft from publc subsdes tend to favour publc agences. In ths way, frms lmt msmatches n the hrng process. 4- Concluson The am of the present research was to study employers search behavours and the mpact of dfferent recrutment channels on the probablty of fllng a vacancy. We thus assumed that frms use dfferent search methods to form a pool of applcants from whch one or more canddates can be chosen to fll the vacancy. In ths non-sequental search process, recrutment channels are vewed as a means of contactng jobseekers. By combnng recrutment channels, frms can ncrease ther chances of successfully fllng a vacancy and we found that the choce and combnaton of recrutment channels s endogenous. A recursve model was used to analyse employers search behavour. We estmated a recursve model, and then tested ths model aganst French survey data through the use of a multvarate probt analyss. Ths allowed us to take nto account the selectvty bas n recrutment channel choces, as well as possble correlatons between -9-

11 recrutment channel choces and ther mpact on the probablty of fllng a vacancy. Ths analyss produced three nterestng results. Frst, recrutment channel choces depend on the characterstcs of the frm concerned and of the vacances to be flled; hence, they are not exogenous. Second, rather than just choosng one channel, frms choose a combnaton of channels, and choces of whch channels to combne are not random. Thrd, the choce of recrutment strateges strongly nfluences the probablty of fllng a vacancy. Prvate employment agences appear to be the most effectve channels; however frms choces of recrutment channels tend to be based on strategc crtera. For example, prvate agences are more effectve n provdng access to sklled jobseekers, and publc agences are more effectve n provdng access to low-sklled workers. Ths reflects the specalzaton of these two types of agency n the French labour market. Our study provdes evdence for the mportant role of employment agences n the recrutment process. As well as enablng frms to contact jobseekers, employment agences can provde assstance wth the screenng and selecton process. The use of employment agences reduces search costs and ncreases the effcency of fllng a vacancy. These benefts are maxmsed when frms choose the most approprate type of agency, that s to say, prvate agences for recrutng sklled staff and publc agences for fllng low-sklled vacances. By developng long-term relatonshps wth employment agences, frms can facltate the screenng and selecton process and decrease the rsk of msmatches. References Barron J., Bshop J. (1985) "Extensve Search, Intensve Search and Hrng Costs: New Evdence on Employer Hrng Actvty" Economc Inqury 23, Barron J., Bshop J., Dunkelberg W.C. (1985) "Employer Search: The Intervewng and Hrng of New Employees" Revew of Economcs and Statstcs 67, Beaumont P.B. (1978) "The Duraton of Regstered Vacances: An Explanatory Exercse" Scottsh Journal of Poltcal Economy 25, Bessy C., Marchal E., Reucau G. (2007) "L Importance des Canddatures Spontanées en France" Les 4 pages du Centre d Etudes de l Emplo 47. Cappellar L., Jenkns S.P. (2006) "Calculaton of Multvarate Normal Probabltes by Smulaton, wth Applcatons to Maxmum Smulated Lkelhood Estmaton" Dscusson paper number 584, DIW Berln. Devne T.J., Kefer N.M. (1993) "The Emprcal Status of Job Search Theory" Labour Economcs 1, Gorter C., Van Ommeren J. (1999) "Sequencng, Tmng and Fllng Rates of Recrutment Channels" Appled Economcs 31, Gorter C., Njkamp P., Retveld P. (1996) "Employers' Recrutment Behavour and Vacancy Duraton: An Emprcal Analyss for the Dutch Labour Market" Appled Economcs 28, Gorter C., Njkamp P., Retveld P. (1993) "The Impact of Employers Recrutment Behavour on the Allocaton of Vacant Jobs to Unemployed Job Seekers" Emprcal Economcs 18, Greene W. (2008) Econometrc Analyss, 6th Edton, Prentce Hall. Holzer H.J. (1996) What Employers Want, Russel Sage Fundaton, New-York. Holzer H.J. (1987) "Hrng Procedures n the Frm: Ther Economc Determnants and Outcomes" NBER Workng Paper number Houseman Susan N., Kalleberf Arne L., Erckcek Georges A. (2003) "The Role of Temporary Agency Employment n Tght Labor Markets" Industral and Labor Relatons Revew 57, 1,

12 Lesueur J-Y. (1997) "L Intermédaton de l ANPE sur le Marché du Traval: Une Analyse Néo-nsttutonnelle" Revue d Econome Poltque 4, 107. Mortensen D.T. (1986) "Job Search and Labor Analyss", n Handbook of Labor Economcs. O. Ashenfelter and R. Layard (Eds), North-Holland, Amsterdam. Mortensen D.T., Pssardes C.A. (1994) "Job Creaton and Job Destructon n the Theory of Unemployment" Revew of Economc Studes 61, Roper S. (1988) "Recrutment Methods and Vacancy Duratons" Scottsh Journal of Poltcal Economy 35, Russo G., Hassnk W.H.J., Gorter C. ( 2005) "Fllng Vacances: An Emprcal Analyss of the Cost and Beneft of Search n the Labour Market" Appled Economcs 37, Russo G., Retveld P., Njkamp P., Gorter C. (2000) "Recrutment Channel Use and Applcant Arrval: An Emprcal Analyss" Emprcal Economcs 25, Tran K. (2003) Dscrete Choce Methods wth Smulaton, Cambrdge Unversty Press. Van Ours J. (1989) "Duratons of Dutch Job Vacances" De Economst 137, Van Ours J., Rdder G. (1993) "Vacancy Duratons: Search or Selecton?" Oxford Bulletn of Economcs and Statstcs 55, Wlde, J. (2000) "Identfcaton of Multple Equaton Probt Models wth Endogenous Dummy Regressors" Economcs Letters 69,

13 Appendx 1: Lst of varables Varable Descrpton Obs. Mean Recrutment channels Market_meth =1 f market methods are used; =0 otherwse Pub_agency =1 f publc agences are used; =0 otherwse Prv_agency =1 f prvate agences are used; =0 otherwse Prof_netw =1 f professonal relatons are used; =0 otherwse Pers_netw =1 f personal relatons are used; =0 otherwse Recrutment success recrut_ok =1 f the hrng process succeeds; =0 otherwse Frm sze small =1 f the frm has 50 or less employees; =0 otherwse medum =1 f the frm has more than 50 but less than 240 employees; =0 otherwse large =1 f the frm has 250 or more employees; =0 otherwse Sector agrcultural =1 f the frm belongs to the agrcultural sector; =0 otherwse constructon =1 f the frm belongs to the constructon sector; =0 otherwse ndustry =1 f the frm belongs to the manufacturng sector; =0 otherwse servce =1 f the frm belongs to the servce sector; =0 otherwse Reason for recrutment replacement =1 f the vacancy s dude to a replacement; =0 otherwse new actvty =1 f the vacancy s dude to a new actvty; =0 otherwse demand ncrease =1 f the vacancy s dude to a demand ncrease; =0 otherwse reorgansaton =1 f the vacancy s dude to a reorganzaton; =0 otherwse Type of vacancy executve =1 f the vacancy s for an executve poston; =0 otherwse short-term job =1 f the vacancy s for a temporary poston; =0 otherwse part-tme job =1 f the vacancy s for a part-tme poston; =0 otherwse urgent recrutment =1 f the recrutment s urgent; =0 otherwse Prevous recrutment methods pubagen_p =1 f publc agences were used n a prevous search; =0 otherwse prvagen_p =1 f prvate agences were used n a prevous search; =0 otherwse cand_100 =1 f more than 100 drect applcatons were receved; =0 otherwse Macroeconomc context dffcultes =1 f the frm has fnancal problems; =0 otherwse Jobseekers_nb = number of jobseekers n 2005 n the frm s sector Source: Enquête Emplo, INSEE, ,285 offers_nb = number of job offers n the frm s sector n the frst half of 2005 Source: ANPE, ,229 offers_evol = change n the number of job offers n the frm s sector n the frst half of 2005 Source: ANPE

14 n Recrutment strateges Appendx 2: Recrutment strateges and hrng success rates Percentage of frms usng the strategy Success rate (%) 1 Professonal networks (Prof. Netw.) Publc agences (Pub. Ag.) Prvate agences (Prv. Ag.) Market methods (Mark. Meth.) Personal networks (Pers. Netw.) Prof. Netw. + Pub. Ag Prof. Netw. + Prv. Ag Prof. Netw. + Mark. Meth Prof. Netw. + Rel. perso Pub. Ag. + Prv. Ag Pub. Ag. + Mark. Meth Pub. Ag. + Pers. Netw Prv. Ag. + Mark. Meth Prv. Ag. + Pers. Netw Mark. Meth. + Pers. Netw Prof. Netw. + Pub. Ag. + Prv. Ag Prof. Netw. + Pub. Ag. + Mark. Meth Prof. Netw. + Pub. Ag. + Pers. Netw Prof. Netw. + Prv. Ag. + Mark. Meth Prof. Netw. + Prv. Ag. + Pers. Netw Prof. Netw. + Mark. Meth. + Pers. Netw Pub. Ag. + Prv. Ag. + Mark. Meth Pub. Ag. + Prv. Ag. + Pers. Netw Pub. Ag. + Mark. Meth. + Pers. Netw Prv. Ag. + Mark. Meth. + Pers. Netw Prof. Netw. + Pub. Ag. + Inter. prv. + Mark. Meth Prof. Netw. + Prv. Ag. + Mark. Meth. + Pers. Netw Prof. Netw. + Pub. Ag. + Mark. Meth. + Pers. Netw Prof. Netw. + Prv. Ag. + Prv. Ag. + Pers. Netw Pub. Ag. + Prv. Ag. + Mark. Meth. + Pers. Netw Prof. Netw. + Pub. Ag. + Prv. Ag. + Mark. Meth. + Pers. Netw