Virtues of SIN effects of an immigrant workplace introduction program

Transcription

1 Vrtues of SIN effects of an mmgrant workplace ntroducton program Olof Åslund Per Johansson WORKING PAPER 2006:7

2 The Insttute for Labour Market Polcy Evaluaton IFAU s a research nsttute under the Swedsh Mnstry of Industry, Employment and Communcatons, stuated n Uppsala. IFAU s objectve s to promote, support and carry out: evaluatons of the effects of labour market polces, studes of the functonng of the labour market and evaluatons of the labour market effects of measures wthn the educatonal system. Besdes research, IFAU also works on: spreadng knowledge about the actvtes of the nsttute through publcatons, semnars, courses, workshops and conferences; nfluencng the collecton of data and makng data easly avalable to researchers all over the country. IFAU also provdes fundng for research projects wthn ts areas of nterest. The deadlne for applcatons s October 1 each year. Snce the researchers at IFAU are manly economsts, researchers from other dscplnes are encouraged to apply for fundng. IFAU s run by a Drector-General. The authorty has a tradtonal board, consstng of a charman, the Drector-General and eght other members. The tasks of the board are, among other thngs, to make decsons about external grants and gve ts vews on the actvtes at IFAU. A reference group ncludng representatves for employers and employees as well as the mnstres and authortes concerned s also connected to the nsttute. Postal address: P.O. Box 513, Uppsala Vstng address: Kyrkogårdsgatan 6, Uppsala Phone: Fax: fau@fau.uu.se Papers publshed n the Workng Paper Seres should, accordng to the IFAU polcy, have been dscussed at semnars held at IFAU and at least one other academc forum, and have been read by one external and one nternal referee. They need not, however, have undergone the standard scrutny for publcaton n a scentfc journal. The purpose of the Workng Paper Seres s to provde a factual bass for publc polcy and the publc polcy dscusson. ISSN

3 Vrtues of SIN effects of an mmgrant workplace ntroducton program by Olof Åslund and Per Johansson * June 15, 2006 Abstract We evaluate an mmgrant workplace ntroducton program amed at helpng ndvduals consdered employable but at the same tme expected to experence substantal dffcultes n fndng work. Usng supported employment methods, the SIN program may nfluence outcomes through several channels. We use ndvdual data and a dfference-n-dfferences approach to estmate the effects of the program. The results suggest that the program ncreased transtons from unemployment to work experence schemes, and mproved future employment probabltes for those who entered these schemes. Keywords: unemployment, labor market programs, mmgrants JEL-codes: J61, J64, J68 We thank Helge Bennmarker, Lnus Lndqvst and Alvaro Mranda for preparng the data. We are also grateful to Krstna Sbbmark for ntervewng SIN offcers and gatherng nsttutonal nformaton on the SIN program. Comments and suggestons from Kenneth Carlng, Peter Fredrksson, Erk Mellander, and from semnar partcpants at IFAU have been very valuable n wrtng the paper. IFAU and SNS, olof.aslund@fau.uu.se * IFAU and Uppsala Unversty, per.johansson@fau.uu.se IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 1

4 Table of contents 1 Introducton Program setup and mplementaton Targeted groups The methods of SIN n theory SIN n practce Data and descrptve statstcs The data Descrptve statstcs Who partcpates n SIN? What happens when someone enters SIN? Methodologcal consderatons Conceptual framework Identfcaton and estmaton DD estmaton Emprcal results A frst look at the effects of SIN The mpact through dfferent mechansms Senstvty analyss General robustness checks Analyng an magnary reform n September An attempt to compare the costs and benefts of SIN Concludng remarks References Appendx A IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

5 1 Introducton The ntegraton of mmgrants n the labor market has caused Swedsh poltcans much agony n recent years. In an nternatonal perspectve, Sweden stands out as one of the countres wth the lowest labor force partcpaton and the hghest unemployment levels among the foregn-born OECD In 2002, unemployment stood at 4 percent n the natve populaton. At the same tme t was 15 percent among those born outsde Europe. Also, the chance of movng from unemployment to employment s substantally lower n many mmgrant groups n comparson to natves Åslund & Rooth It s sometmes argued that smple comparsons to other countres do not account for the fact that Sweden has receved a comparatvely large fracton of refugees, who generally have worse outcomes than labor mgrants see e.g. Integratonsverket Stll, the fact that a large proporton of the foregn-born populaton has a dsadvantaged labor market poston remans. The Swedsh government has adopted a varety of polces and programs to combat the problematc stuaton. Measures range from specal local ntroducton programs for the recently arrved, va prorty regulatons at the PES offces, to large-scale neghborhood development programs. Ths paper evaluates a program on tral snce September 2003, whch n some respects goes one step further n ts nterventons. Usng supported employment methods prevously employed for dsabled workers, the SIN Specal INtroducton program targets mmgrants and refugees who are consdered capable of takng a job mmedately, but who are also at rsk of becomng long-term unemployed. Once enrolled n SIN, the job searcher s assgned to a specal employment offcer wth consderably fewer clents at hs/her hands than what s usual. The work s formally dvded nto sx steps. The frst step s to analye the searcher s merts, potentals and preferences. Then job gatherng commences, followed by so-called job analyss where t s nvestgated whether the conceved tasks or workplace need to be adapted n any way. The last three steps are: workplace ntroducton, employment, and follow-up. We wll dscuss the methodology further n secton 2. SIN s run n 20 muncpaltes, many of whch are stuated n the metropoltan areas of Sweden, where the bulk of the targeted populaton lves. Prevous studes have found that partcpaton s qute selectve n terms of beng job ready,.e. partcpants should be capable of managng a job and wllng to make the move nto employment Hernemar 2004, Lndgren Åsbrnk IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 3

6 These facts have mplcatons for the desgn of ths study of the effects of the program. Frst, despte the avalablty of very detaled populaton mcro data on the unemployed, we are unlkely to capture all factors determnng partcpaton n SIN. Second, the mplementaton of the plot opens up for a dfferencen-dfferences analyss comparng the before-after change n SIN muncpaltes to other muncpaltes n the same local labor market. Naturally, ths methodology also allows for the possblty that SIN may have affected unemployed non-partcpants. We wll thus estmate the effects of SIN on the populaton at rsk n the partcpatng muncpaltes,.e. we apply a reduced form estmator at the muncpal level. We consder two alternatve outcomes employment and open unemployment and argue that the man effects of SIN may work through dfferent channels: t can affect the haard from open unemployment to employment orto an ntermedate treatment IMT e.g. employment subsdes or work experence schemes; t may nfluence the flows back to open unemployment from employment or IMT:s; t may affect the transtons from IMT:s to employment. The thrd channel can consst of two parts: a change n the dstrbuton of ntermedate states combned wth the mpact dfferences between ntermedate outcomes n the absence of SIN, and, secondly, a change n the effects from enterng a partcular ntermedate state on the fnal outcome. To get at the mechansms at work, we estmate several sets of Cox regresson models, comparng before-after changes n the SIN locatons to the same changes n the non-sin locatons. Our fndngs suggest that SIN ncreases the rate of transtons nto work experence schemes. Also, entry to work experence schemes s assocated wth hgher chances of becomng employed under SIN than otherwse. SIN does not appear to have affected the flows back to open unemployment from work or from IMT:s. The nterpretaton of a statstcally and economcally postve sgnfcant dfference-n-dfferences estmate on the haard from unemployment to work s complcated by the fact that analyng a fake reform supposed to have occurred one year before the actual reform yelds a smlar estmate. In the next secton we gve some detal on the desgn of the SIN program and ts mplementaton. In secton 3 we present the data, descrbe the SIN partcpants n terms of background characterstcs, and sketch what happens when someone enters SIN. Secton 4 frst dscusses some conceptual methodologcal ssues and then outlnes the emprcal strategy. Secton 5 presents the results on the effects of SIN. Secton 6 concludes. 4 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

7 2 Program setup and mplementaton 1 SIN was ntroduced n 20 Swedsh muncpaltes on September 1, Intally t was scheduled to run untl December 31, 2005, but the tral perod was later extended to December 31, It s admnstered by local PES offces whch have been granted extra fundng for the case workers workng wth the SIN partcpants. In 2005, the resources added amounted to a total of SEK 126 mllon, resultng n some 250 SIN offcers Ams 2006b. Accordng to the December 2005 report from the labor market board, 4,781 ndvduals entered SIN durng 2005 Ams 2006a. The typcal SIN offcer appears to have somewhere between 15 and 30 clents. A cautous comparson to other PES offcers suggests that ths s about one tenth of the normal caseload. Note also that the fundng only covers the SIN offcers resources for e.g. employment subsdes or work experence schemes are taken from the regular budget. 2.1 Targeted groups Accordng to the government bll Förordnng 2003:623, the program may be offered to mmgrants or refugees age 20 or above. The ndvdual can be n SIN for a maxmum of sx months. Under unspecfed extraordnary crcumstances, however, the perod can be prolonged. The partcpants should be reckoned capable of takng a job mmedately, but also at rsk of becomng long-term unemployed. SIN may be granted people who fulfll only the frst crteron, but are n or have completed a local refugee ntroducton program. In other words, the rules for program elgblty are very loose snce t suffces to be at rsk of becomng long-term unemployed, or to have completed a local ntroducton program any perod of tme ago. We wll present characterstcs of the partcpants n secton 3.2. Normally, SIN should not be chosen before testng other alternatves.e. standard job search assstance, other labor market programs. To assgn an n- 1 The secton draws prmarly on Ams 2004a,b 2 Durng 2005, four muncpaltes were effectvely added to the SIN trals. The partcpatng muncpalty of Esklstuna then started servng also Flen, Katrneholm, Nyköpng and Vngåker. These are small towns compared to most of the other partcpatng locatons. We wll attend to the expanson n the robustness checks. IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 5

8 dvdual to the program, the case worker should have estmated the need for support to be larger than the need for qualfed job-matchng assstance. The SIN offcers are not to be consdered just an extra resource for job matchng. 2.2 The methods of SIN n theory SIN s based on so-called supported employment, whch have been developed for, and prevously used n, programs attendng to dsabled workers see Antonsson 2003 or Leach 2002 for dscussons. After havng been judged sutable for SIN, the job searcher s assgned to a SIN offcer, and then the process s dvded nto sx steps: 1. Job searcher analyss 2. Job gatherng 3. Work analyss 4. Workplace ntroducton 5. Follow-up 6. Employment The frst step s the job searcher analyss, n whch the case worker ntervews and maps the ndvdual s merts, potentals and wshes. If the ndvdual does not have a CV or valdated credentals, he or she should be remtted to the regular PES for ths purpose. The second step of the process s job gatherng, when the case worker looks for sutable work. The case worker should then nform the prospectve employers that the am of the program s employment, even though the workplace ntroducton can begn wth a tranee poston. However, there should be a promse of future employment for a partcular workplace to come nto queston for placement. Work analyss s the thrd step, n whch the SIN offcer nvestgates whether tasks and work envronment suts the partcpant. If necessary, the offcer can dscuss possble changes n the tasks wth the employer. In ths step t s also to be made clear whch type of support the offcer can gve durng the workplace ntroducton. Then, the workplace ntroducton begns. The ntroducton s to be performed n close cooperaton between the SIN partcpant, the offcer, the employer, colleagues, and unon representatves. The offcer s presence at the workplace s supposed to facltate the partcpant gettng started wth the job tasks, and to, e.g., help overcomng language barrers or makng sure that the partcpant becomes a part of the workplace communty. When the SIN perod s over usually after sx months, the offcer performs a follow-up of the assgnment. Ths s consdered partcularly mportant snce 6 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

9 employment wthn the program s often on a temporary bass, and there s a wsh to make sure that the ndvdual actually gets hred. 2.3 SIN n practce Some work on SIN has been done prevously. Frst, there are the reports from the labor market board to the central government e.g. Ams These generally carry a postve noton of SIN, where t s argued that even though the measure does not n all cases end n employment, the outcomes are good compared to other programs. One thng worth notng s that SIN offcers had to be recruted externally n about 70 percent of the cases. Accordng to the labor market board, ths meant that SIN was not fully up and runnng untl the fall of 2004 Ams We wll return to ths ssue below. There are also two studes whch have performed ntervews wth partcpants, offcers and n one study employers Hernemar 2005, Lndgren Åsbrnk The overall mpresson from these ntervews s that partcpaton s qute selectve. The offcers stress the mportance of beng job ready. Ths means havng suffcent Swedsh language sklls, not beng n need of any type of rehabltaton and beng wllng to commute or relocate f necessary. Accordng to these two studes and to the Ams 2005 report, the man beneft of the program s that the offcers have consderably more tme wth each clent, ncreasng the potental for a successful match. The ntervews suggest that the workplace ntroducton and follow-up have played a smaller part n the mplementaton. In practce, the offcers do thus not appear to stck to the supported employment methodology. The ntervewed employers express the feelng that they can trust that the ndvduals comng from SIN are really rght for the job. Some employers state that t may be easer handlng mnor conflcts va the offcer than wth the employee drectly. Some offcers, however, express the vew that t would be very stgmatng wth a strong case worker presence durng the ntroducton, snce ths would sgnal low capablty of the applcant. 3 3 In fact, the pont that supported employment methodology may be stgmatng and not transferable to the case of mmgrants was made at a very early plannng stage see Carlng IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 7

10 3 Data and descrptve statstcs 3.1 The data We use data from the IFAU database coverng the entre Swedsh populaton age The data contan detaled ndvdual nformaton on regstered unemployment spells up to November 15, The nformaton ncludes date of unemployment entry, whether the ndvdual s n open unemployment or n some type of program and f so, whch program at any pont n tme durng regstered unemployment, and date of and reason for leavng unemployment e.g. fndng a job, regular educaton, leavng the work force etc.. We nclude ndvduals who entered open unemployment between January 1, 2000 and November 15, 2005, and who were at least 20 and less than 63 years old at the tme of unemployment entry. An ndvdual may have multple spells. In the baselne analyss, we nclude ndvduals who were born outsde the Nordc countres. We wll, however, dscuss robustness checks wth varyng restrctons on regon of brth. There s also a restrcton mplyng that only those who were regstered at a PES located n the local labor market of at least one of the SIN-partcpatng muncpaltes are ncluded. Ths s because we want the comparson group to be subject to the local shocks experenced by the treatment group. Local labor markets are defned by Statstcs Sweden based on observed commutng behavor. A local labor market conssts of one or usually several muncpaltes. No muncpalty belongs to more than one local labor market. We use the 2003 defnton of local labor market regons. The analyss focuses on two fnal outcomes: employment and open unemployment. People are followed for the duraton of open unemployment. Then, we regster four types of transtons: to regular employment permanent or temporary; to subsded employment; to work experence schemes; v to other categores. The fourth type of ext contans those who have deregstered for unknown reasons whch s employment n about 50 percent of the cases see Brng & Carlng 2000, Sahn 2003, Forslund et al. 2004, those who have left for regular educaton or labor market tranng, and those who become enrolled n other types of labor market programs than the ones covered by and. We label ext types, and v ntermedate treatments IMT:s. Smlarly, we montor transtons from employment or IMT:s back to open unemployment, and from IMT:s to employment. Employment s defned as e- 8 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

11 ther beng regstered at the PES n a category ndcatng regular employment temporary, part-tme, lookng for other work, or havng found work as the latest reason for deregstraton. 4 Table 1 dsplays some basc facts about the data. There are just below half a mllon unemployment spells and about 222,000 ndvduals n the data 5. The medan spell of open unemployment lasts 112 days. On average, each spell was preceded by a total of 247 days n labor market programs and 641 days of open unemployment 6. The SIN muncpaltes have worse track records: spells last a bt longer and the unemployed have spent more tme n unemployment and labor market programs. Note that SIN n ths context means beng regstered n a partcpatng muncpalty. The actual SIN partcpants who are a small porton of the total number of ndvduals are descrbed n the next secton. 4 People who fnd temporary, part-tme, or not fully satsfactory work may reman regstered at the PES. 5 Note the dfference to Table 1 whch dsplays the number of ndvduals condtonal on tme perod and locaton. 6 The startng year s IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 9

12 Table 1 Basc data descrpton Tme perod relatve to Sep. 1, 2003 Muncpalty Before After Total Non-SIN # observatons 82,921 58, ,864 comparson # ndvduals 48,542 42,771 91,313 Medan spell length Average tme n LMP Average tme n open unempl SIN # observatons 203, , ,330 treatment # ndvduals 114, , ,764 Medan spell length Average tme n LMP Average tme n open unempl Total # observatons 286, , ,194 # ndvduals 162, , ,077 Medan spell length Average tme n LMP Average tme n open unempl Notes: SIN muncpaltes are the ones that partcpate n the trals startng September 1, Spell length s for completed spells n open unemployment n days, tme n LMP unempl. s days spent n labor market programs open unemployment snce August 1991 pror to the start of the current spell. Note that average spell length s longer n the before perod, whch s partly a result of longer tme untl censorng on November 15, The # ndvduals are condtonal on locaton and tme perod; an ndvdual may be counted n multple cells frst spell n each perod and locaton. The total number of ndvduals observed s 222, Descrptve statstcs Who partcpates n SIN? Let us now descrbe the SIN partcpants. Table 2 below shows that most of the partcpants were born n Asa, Afrca or n European countres outsde EU15. Iraq, Iran and former Yugoslava are the most common countres of brth accordng to Ams Note, however, that the numbers are condtonal on beng born outsde the Nordc countres. As a matter of fact, about 5 percent of the partcpants were born n Sweden. All n all, there are about 7,000 SIN par- 10 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

13 tcpants n the data. The partcpaton rate s between 6 and 7 percent based on the unemployment nflow n n the SIN muncpaltes. Wth such a low probablty of partcpaton, the possblty that people relocate to become elgble for SIN should not be a serous concern for the evaluaton. Nevertheless, we performed an analyss of relocatons nto and out of the non-sin muncpaltes. The results dd not suggest that SIN changed the moblty patterns n any substantal way. Some of the regulatons governng SIN sgnal that the measure s at least partly ntended for recently arrved. It s therefore nterestng to fnd that whle 30 percent of the partcpants have mmgrated snce the year 2000, approxmately half of the sample actually came before We also see that there are somewhat more males among the partcpants, and that the three educaton categores prmary, secondary, tertary each contan roughly one thrd of the partcpants. Somewhat noteworthy s the fact that only about one quarter of the people n SIN has an updated plan for acton accordng to the admnstratve records. 7 Snce the SIN methodology stpulates qute extensve consderatons of dfferent alternatves avalable, ths seems to be an admnstratve flaw n the sense that all nformaton s not entered nto the regsters. The table also shows that 30 percent of the partcpants are not elgble for any type of unemployment benefts. Typcally, these ndvduals are supported by socal assstance. 7 Updated means that the plan was created or changed n the calendar year of the start of the unemployment spell lnked to SIN. IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 11

14 Table 2 Descrpton of the SIN partcpants born outsde the Nordc countres Regon of brth EU Europe except EU Afrca North Amerca 1.59 South Amerca 5.55 Asa Oceana 0.15 Mssng 0.93 Immgraton perod unknown 0.23 Male 56.6 Age at unemployment entry Level of educaton Prmary 31.9 Secondary 36.3 Tertary 31.8 Fracton wth updated acton plan 24.6 Coded as not elgble for benefts Regstered n SIN muncpalty 97.2 # observatons 7,292 Notes: Informaton on SIN partcpaton and unemployment s taken from separate regsters. The dates of SIN start and end do not necessarly comply wth the tmng of unemployment spells. Tme varyng varables are therefore measured for the unemployment spell lnked to SIN partcpaton accordng to the followng herarchcal crtera: SIN start date s wthn the spell; the spell endng at the shortest tme before the SIN start date; the spell begnnng at the shortest tme after the SIN start date. Recall that SIN s supposed to target ndvduals who are beleved to have substantal dffcultes fndng work. Some regulatons hnt that t may be peo- 12 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

15 ple who have lmted labor market experence n Sweden due to short resdence whch s true n about a quarter of the cases. Also, SIN should be consdered the best alternatve and not be chosen before testng other alternatves. Rememberng the low employment rates n some groups of foregn-born, one mght wonder how the SIN partcpants compare to other unemployed wth smlar background. Table 3 shows characterstcs of ndvduals enterng open unemployment durng 2003 and 2004, by eventual observaton n the SIN program. The non-partcpants have been re-weghted to conform to the regon-oforgn/perod-of-mmgraton dstrbuton of the partcpants. In other words, the comparson controls for regonal background and tme spent n Sweden. The dstrbuton of educaton s very smlar across the two groups f anythng SIN partcpants are postvely selected. The partcpants have spent consderably more tme n open unemployment and labor market programs, whch s not surprsng gven that one can qualfy for SIN by beng long-term unemployed. What s perhaps more surprsng s that the partcpants are much less lkely to be coded not elgble for any type of beneft. Even though ths varable taken from an admnstratve regster does not necessarly reflect actual unemployment benefts UB elgblty at a gven pont n tme, 8 the pattern sgnals that SIN partcpants may not have less attachment to the labor market than other unemployed wth smlar backgrounds. Ths concluson s also supported by the fact that the 2002 employment and earnngs statstcs are very smlar across the groups. Even though the outcomes are poor on average, there s among the partcpants clearly a substantal fracton wth experence from the Swedsh labor market. 9 One could ask why ndvduals who may have spent more than 20 years n Sweden and who were employed as late as 2002 are n need of a workplace ntroducton program partly ntended to brdge lngustc and cultural gaps. 8 The varable s based on whch UB fund the ndvdual belongs to. Those who do not belong to any fund or whose benefts have expred should be coded 00 whch s what the table entry s based on. However, one can not be certan that ths nformaton s updated f, e.g., benefts expre durng an unemployment spell. 9 Not surprsngly, employment rates ncrease wth tme spent n Sweden. 47 percent of partcpants arrvng were employed; for those who came earler the fgure was 53 percent. IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 13

16 Table 3 Comparson of SIN partcpants and non-partcpants. Weghted by regon of orgn and mmgraton perod SIN partcpants Non-partcpants Prmary educaton % Secondary educaton % Tertary educaton % Coded as not elgble for benefts % Average tme n open unempl Average tme n LMP Employment Average earnngs ,200 58,350 Notes: Table presents averages of varables for ndvduals who entered open unemployment , by SIN partcpaton before Nov 15, Non-partcpants have been re-weghted to conform to the regon-of-orgn/perod-of-mmgraton dstrbuton of the SIN partcpants. There s a very small number of observatons wthout nformaton on educaton; ths category s excluded from the table and thus the numbers do not add to 100. Measures of employment and average earnngs are lmted to ndvduals observed n What happens when someone enters SIN? Table 4 shows the status of the SIN partcpants 3 days before the start date of SIN. 70 percent are n open unemployment, whereas about 10 percent are classfed as employed. Few are n subsded employment and some are n work experence schemes. A sgnfcant porton s also found n the category other. Further examnaton reveals that about half of these ndvduals are n matchng and gudng actvtes, one quarter s n preparatory educaton and about 15 percent s n labor market tranng. Table 4 Status 3 days before SIN entry Status Freq. Percent Unemployed 5, Employed Subsded employment Work experence Other Total 7, Notes: 3 observatons dropped due to SIN start scheduled after November 15, IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

17 Table 5 then shows the status of the partcpants on the startng day of SIN, and 3, 14 and 28 days later, n total and by status 3 days before SIN start. Obvously, many ndvduals leave open unemployment for work experence schemes or regular or subsded employment. There appears to be transtons just around the SIN start, and then a gradual shft durng the frst weeks of SIN. After four weeks, 36 percent s n regular employment, 20 percent n subsded employment and about 23 percent of the partcpants are n work experence schemes. The latter share s substantally hgher just after entry to SIN, suggestng that some partcpants go va work experence to some form of employment. The share n open unemployment s 11 percent after 28 days n SIN. Of those n the other category, about one thrd s no longer regstered at the PES. Among those who are, 38 percent are n matchng and gudance actvtes, 16 percent are n some form of tranng, 14 percent are n programs prmarly ntended for those wth occupatonal dsabltes, such as wage subsdes or sheltered employment. There are also movements between the other states. Only half of those who were n a work experence scheme three days pror to SIN reman there on the day of SIN entry. The bulk of these transtons are to some form of employment. 28 days after entry, percent of those who started n work experence are n subsded regular employment. IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 15

18 Table 5 Status on and after SIN entry, by status 3 days before entry 3 days before Unempl. Employed Subs empl. Work exp Other Day of entry Unemployed Employed Subs. empl Work exp Other Total days after entry Unemployed Employed Subs. empl Work exp Other Total days after entry Unemployed Employed Subs empl Work exp Other Total days after entry Unemployed Employed Subs empl Work exp Other Total The conclusons from ths descrpton are the followng. Frst, t s clear that SIN usually means entry nto some form of treatment/outcome,.e. regular or subsded employment or work experence schemes. Second, the patterns sgnal that the admnstratve start dates of SIN most lkely do not mark the actual start dates n many cases. Remember that the frst two steps of the SIN methodology are to map out the partcpants capabltes as well as wshes and to fnd a sutable workplace. We would therefore not expect to see so many trans- 16 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

19 tons nto categores ndcatng beng at a workplace around the actual SIN start date. Thus, the SIN start dates must be consdered naccurate n many cases. As a matter of fact, the same s true for the SIN end dates, snce these typcally are set at the same tme as the start dates many tmes resultng n a 180-day perod. Note however, that the emprcal analyss presented below captures a reduced form effect of SIN, whch s not based on observatons of ndvdual SIN partcpaton. 4 Methodologcal consderatons We begn ths secton by outlnng the conceptual framework for the effects of SIN, focusng on the dfferent components entaled n the SIN program. Then we dscuss how to retreve these estmates n practce and whch dentfyng assumptons we need to make. 4.1 Conceptual framework Let us frst dscuss how SIN may affect the outcomes of nterest, here chosen to be employment or alternatvely open unemployment. Fgure 1 llustrates how we thnk about the potental mechansms at work. All ndvduals under study start out n open unemployment. Consder frst the potental effects gong from ths state to other states. SIN may affect the ndvdual s probablty to move drectly from open unemployment to employment. Secondly, t may alter the chances to enter ntermedate treatments IMT:s: work experence, employment subsdes and the summary category other whch ncludes e.g. labor market tranng and gudance actvtes at the PES. The ntermedate treatments may then n turn affect the future labor market opportuntes among the unemployed. A change n the dstrbuton of IMT:s towards treatments that mprove the chances of enterng employment s one potental effect. SIN may also nfluence the lkelhood of movng from a gven IMT to employment. There may also be smlar mpacts on the flows n the other drectons dashed arrows n the fgure: to open unemployment from employment or IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 17

20 from any of the IMT:s respectvely. Of course, one could argue that SIN would nfluence flows between the dfferent IMT:s or the probablty to move from employment to an IMT. We argue, however, that the flows descrbed above are the ones of prmary nterest. 10 Unemployment Employment Work experence Empl. subsdy Other Fgure 1 Conceptual framework of the effects of SIN Followng ths descrpton, the evaluaton of SIN can be splt nto four steps. Frst we estmate how SIN affects the haard or probablty to employment and to three dfferent IM treatments employment subsdy, work experence and all other types of actvtes wthn the PES. Second, we estmate the haard back to open unemployment from employment and the three dfferent IMT:s. Thrd, we consder the mpact of SIN on employment after enterng any IMT regardless of whch. Fourth, for each IMT we estmate the haard to employment,.e. an effect condtonal on enterng a partcular IMT. In the next secton we wll dscuss how to dentfy the parameters of nterest. 4.2 Identfcaton and estmaton If ndvduals were selected randomly to partcpate n SIN from the group of elgble ndvduals, then we could smply estmate the effects usng Cox regresson models comparng partcpants to non-partcpants. The basc estma- 10 A senstvty analyss suggests that SIN dd not affect the excluded types of transtons. 18 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

21 ton problem s that we do not have random selecton nto SIN. The nsttutonal settng suggests that most lkely there s substantal selecton nto SIN at the ndvdual level. We wll therefore not base our estmates on ndvdual SIN partcpaton. Instead we wll use dfference-n-dfferences DD estmators at the muncpal level. Our estmates are therefore the reduced forms effects of enterng unemployment n a SIN muncpalty. The reason why we do not use a smple before-after estmator comparng the flows n the SIN muncpaltes before 1 September 2003 wth the flows after s that there could be changes n macroeconomc condtons that affect the labor market that has nothng to do wth SIN. The DD approach s an attempt to handle ths problem by controllng for macroeconomc changes common to all locatons n the same labor market area. Needless to say, the dentfyng assumpton s that the development over tme n the reference locatons capture the counterfactual development n the SIN locatons, had SIN not been mplemented. We wll pay attenton to the plausblty of the DD strategy n the presentaton of the results DD estmaton In the DD estmatons, the Cox regressons employed are stratfed on munc- practce and all other ntermedate treatments except open unemployment. palty. In other words, the estmatons nclude fxed muncpal effects. The estmatons are performed wth the partal maxmum lkelhood estmator. Let Z = 0, 1, 2, 3 denote employment, employment subsdy, workplace The haard to leave unemployment for Z = 0, 1, 2, 3 s then modeled as e λ t, S = h0 t, mexp δ 1D t + δ 2SD t, = 0,...,3 1 where Dt s an ndcator varable that s equal to 1 when SIN s n operaton,.e. D t = I T > Sep , and h t,, m = 1,..., M s the baselne 0 m haa rd to Z = 0, 1, 2 or 3 for the elgble unemployed n muncpal m. The ndcator varable S s equal to 1 f the observaton s made n a SIN muncpalty, 0 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 19

22 otherwse. Thus, 2 δ measures the effects of SIN on transtons to employment = 0 or to IMT = 1, 2, In the next step we estmate the effects of SIN on the haard back to open unemployment: = + = + = + = + + = = = = = u u S t D t Z S t Z t D t Z t Z t D S t D m t h S t exp,, α α α α α α λ Here s the baselne haard to unemployment for those employed and n IMT, =1, 2, 3 n muncpalty m. Thus,, 0 m t h u 2 α s the effect of SIN on the haard to unemployment from employment, and 3 1,2,, 4 = α estmates the effects wthn a partcular IMT on the transton to open unemployment. Fnally we also estmate the effect of IMT:s on employment = + = + = + = + + = = = = = j j S t D t Z S t Z t D t Z t Z t D S t D m t h S t exp,, β β β β β β λ Here s the baselne haard to employment for those n IMT, =1, 2 3 n muncpal m. Thus,, 0 m t h j 2 β captures the effect on the haard to employment from employment subsdes wthn SIN and 3 2,, 4 = β estmates the effects 11 Note that the coeffcent on S cannot be dentfed n the stratfed regressons. IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 20

23 of SIN operatng through work experence and the other IMT respectvely. Naturally, we can also estmate the model wth the dfferent IMT:s counted as one state. A word of cauton regardng how to nterpret the dfferent sets of estmates s warranted. The general problem s that there may be heterogeneous treatment effects n the populaton at rsk. Suppose that SIN ncreases the haard to employment, and that we see an ncreased rsk of movng back to open unemployment from employment under SIN. Snce more people entered employment, the ncreased flows back to unemployment can be an effect of sample selecton,.e. the ones observed n employment under SIN are dfferent from the ones n employment under the non-sin regme. A smlar case can be made for transtons between the IMT:s and employment. Of course, altered selecton nto employment and dfferent IMT:s may be at work even f we would not see any change n the probabltes of enterng a partcular state. One effect of SIN may smply be to reallocate ndvduals between dfferent states. At the aggregated muncpal level, ths s an effect of major nterest: does SIN mprove the matchng between ndvduals, employers and IMT:s? A drawback s that sample selecton may hde ndvdual wthn IMT effects, whch are of course also of nterest. We wll return to the dscusson on the mpact of sample selecton n the presentaton of the results. 5 Emprcal results We begn the presentaton of the fndngs by a very smple descrpton of the probablty of beng n work 360 days after enterng unemployment n the SIN and non-sin locatons by tme of entry nto unemployment. Ths type of analyss gves a noton of the total reduced form effects of SIN. We then try to shed some lght on dfferent possble mechansms. Frst, we study the flows from unemployment to employment or IMT:s. Then we consder the transtons back to unemployment, and fnally turn to the mpact of SIN on gong from IMT:s to employment. 5.1 A frst look at the effects of SIN Fgure 2 shows the probablty of beng n work 360 days after unemployment entry by locaton and 180-day perod relatve to September 1, 2003 perod 0 begns on ths day. Observatons to the left rght of the left rght vertcal IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 21

24 lne are made before after the mplementaton of the SIN program. Observatons n between the lnes may be partly affected by the SIN program. In the begnnng of the observaton perod, the probablty of beng n work s much hgher n the non-sin locatons. The gap then narrows as the probablty decreases more n the non-sin locatons compared to the partcpatng muncpaltes. In ths fgure t s hard to see any clear effects of SIN by comparng the developments after the mplementaton of the program. There s a dp and then a recovery n the SIN locatons. A very benevolent nterpretaton s that ths recovery resulted from the SIN program. However, many of the comparson locatons exhbt smlar dps and recoveres. The fact that the development over tme dffers across types of locaton s potentally troublng for the DD approach. We have tred restrctng the comparson locatons to ones wth a development before the SIN program smlar to the one n the SIN locaton. However, such an approach s qute arbtrary compared to the one based on offcal classfcatons of local labor markets. Furthermore, the declne n the before perod tends to be larger even wth the restrcted comparsons. Our strategy n the formal analyss below wll be to frst perform DD estmatons of the effects of the actual reform. Then, we analye an magnary reform supposed to have occurred one year before. In case the magnary reform yelds results that are qualtatvely smlar to those followng the actual reform, we are less nclned to beleve n the baselne estmatons. In case we see no effect n the magnary analyss, we fnd the baselne estmates more plausble. 22 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

25 Fracton n work Non-SIN perod SIN perod day perods from September 1, 2003 Non-SIN locaton SIN locaton Fgure 2 Probablty of beng n work 360 days after unemployment entry, by locaton and 180-day perod of entry Notes: Inflow n perod 0 begns on September 1, The nth perods begn +/ n*180 days before/after ths date. Observatons to the left rght of the left rght vertcal lne are made before after the mplementaton of the SIN program. Observatons n between the lnes may be partly affected by the SIN program. 5.2 The mpact through dfferent mechansms We now turn to present the results regardng the dfferent mechansms potentally at work. Throughout the presentaton of the baselne results, we wll dscuss the estmates as showng the effects of SIN on the outcomes.e. assume that the DD setup usng the basc populaton restrctons descrbed above retreves the causal estmates. These assumptons wll then be questoned and dscussed n the senstvty analyss of secton 5.3. Fgure A 1 presents the haards from unemployment to employment, before and after the SIN reform n the SIN and the comparson locatons respectvely. The fgure tells us three thngs: the haard rates are lower n the SIN muncpaltes both before and after the reform; the haards are lower after the reform compared to before; the decrease between the two tme perods ap- IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 23

26 pears to be smaller n the SIN muncpaltes. Ths s also the message of the second row of Table 6, whch dsplays the estmated effect of SIN on the haards from open unemployment to employment. Accordng to the DD setup, there s a postve and sgnfcant effect on the outflow: the haard rate ncreases by about 12 percent. The table also shows that transtons to all of the three IMT:s were postvely affected by SIN. The mpact on employment subsdes and other s qute moderate, but there s a substantal effect on the probablty of enterng work experence schemes. Fgure A 2 shows that what actually occurred was that a dfference present before the SIN reform was elmnated n the post-sin perod. Fgure A 3 shows that very lttle happened to transtons nto employment subsdes, whereas Fgure A 4 shows a smlar declne n the transtons to other IMT:s n the SIN and the comparson locatons. Table 6 Effects on transtons from unemployment to employment or ntermedate treatments Ha. Rato Std. Err. P> Employment D SD Employment subsdy D SD Work exp. D SD Other D SD Notes: Table shows parameter estmates from Cox haard regresson estmaton usng the stratfed partal maxmum lkelhood estmator. D measures the general before-after effect; SD s the DD estmate the dfference n the before-after estmate between the SIN and the non- SIN locatons. The ncreased outflows from unemployment under SIN do not appear to have ncreased the flows back to unemployment, as shown by Table 7. The only statstcally sgnfcant dfferences-n-dfferences estmate s for employment subsdes. Snce there was no sgnfcant mpact on the flows to ths IMT, the somewhat ncreased return transtons may be of less mportance. In terms of levels, Fgure A 5 and Fgure A 6 show that work experence s the IMT 24 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

27 wth the hghest haards back to unemployment, at least durng the frst 200 days. In general, the patterns are relatvely smlar n treatment and control locatons. Let us now turn to transtons to employment among those who enter IMT:s. A potental effect of SIN s to provde a better/worse matchng between ndvduals and IMT:s, or to alter the composton of IMT:s. Another type of effect s to alter the outcomes wthn a partcular IMT. The frst set of estmates n Table 8 captures the total effect after movng to any IMT,.e. the sum of the above-mentoned effects. We see that the pont estmate s small and statstcally nsgnfcant. The estmates presented n the lower part of the table suggest that the only sgnfcant mpact under SIN s that the haard from work experence to employment ncreased by about 15 percent. Fgure A 7 and Fgure A 8 suggest that what happened was that the dfference between SIN and non- SIN locatons dmnshed. Note also n the fgures that the seemngly dramatc changes n the haards from employment subsdes buld on qute a small number of observatons. Table 7 Effects on transtons from employment or ntermedate treatments back to unemployment Ha. Rato Std. Err. P> Employment D SD Employment subsdy D SD Work exp. D SD Other D SD Notes: Table shows parameter estmates from Cox haard regresson estmaton usng the stratfed partal maxmum lkelhood estmator. D measures the general before-after effect; SD s the DD estmate the dfference n the before-after estmate between the SIN and the non-sin locatons. Do the postve wthn-effect for work experence and the ero overall DD estmates of IMT:s make sense? Probably, yes. Work experence has pretty poor outcomes relatve to e.g. employment subsdes. SIN meant an ncrease n the relatve share n work experence, whch counteracts the postve effects wthn the work experence IMT. The latter effect s not so surprsng: those who enter work experence under the SIN program are to have a promse on IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 25

28 employment after fnshng. Even though ths does not hold n all cases, one can suspect that SIN makes case workers more restrctve and cautous n grantng work experence schemes, and at the same tme ncreases pressure on employers to stck to ther promses. Table 8 Effects on movng from IMT:s to employment Ha. Rato Std. Err. P> Any IMT D SD Employment subsdy D SD Work exp. D SD Other D SD Notes: Table shows parameter estmates from Cox haard regresson estmaton usng the stratfed partal maxmum lkelhood estmator. D measures the general before-after effect; SD s the DD estmate the dfference n the before-after estmate between the SIN and the non-sin locatons. 5.3 Senstvty analyss We have performed a number of robustness checks to test the plausblty of the results. We begn by presentng a few general specfcaton tests before proceedng to a more fundamental dscusson on the dfference-n-dfferences strategy General robustness checks A frst type of varaton was to nvestgate the possblty that the populaton n the treatment and the comparson locatons were somehow dfferent, whch may make the DD strategy nvald. The dea s of course that there may be dfferences across groups n the reacton to local shocks. Frst, we smply restrcted the populaton to ndvduals born n the regons most frequent n SIN: Afrca, Asa and Europe outsde EU15. In a more sophstcated robustness check we then estmated the condtonal propensty of enterng SIN usng a logt regresson model wth a large set of covarates for the SIN locatons. The predctons from ths model were then used to remove all comparson observatons wth a propensty to start SIN that was smaller than the smallest propen- 26 IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program

29 sty for the SIN partcpants 846 ndvduals. For dfferent levels of these propenstes 16 groups we then estmate separate treatment effects. We also tred ncludng covarates n the regressons, controllng for potental changes n the dfferences of the populaton composton. None of these varatons made any dfference to the results. Droppng observatons n the local labor market regon where SIN was extended durng 2005 dd not have any mpact on the estmates. As mentoned n the descrpton of the reform, t has been argued that SIN was not workng fully durng the frst year due to the need for recrutng new offcers. We therefore moved the reform date to September 1, 2004 and excluded unemployment entres n the frst year of SIN Sep 2003 Aug Ths gave smlar but somewhat stronger results n the sense that the postve mpact on the transtons to work experence from unemployment was larger, and so was the mpact on the flows from work experence to employment. Ths s vaguely n lne wth the noton that SIN grew more nfluental wth calendar tme. When we perform the analyss separately by gender, we fnd that the qualtatve patterns are smlar, but that the effects appear to be somewhat larger for women than for men. We have also estmated the models separately for some large regons e.g. the Stockholm local labor market. The qualtatve patterns are smlar across regons. Let us also menton that we have approached the SIN reform n many alternatve ways and that we have used the stratfed Cox regresson estmator wth dfferent levels of stratfcaton. Estmatng dfferences-n-dfferences lnear probablty models for the probablty of beng n employment at fxed tmes followng entry to unemployment or IMT:s gves qualtatvely smlar results: an ncreased probablty of beng n employment, 12 workng at least partly through better performance of the work experence schemes. Usng calendar tme stratfcaton, propensty scores and calendar tme and propensty score as stratfcaton unts also gve qualtatvely the same results Analyng an magnary reform n September 2002 The emprcal strategy bulds on the assumpton that the only thng separatng the development over tme n the SIN and the non-sin locatons s the ntro- 12 Although questonable on the bass dscussed n secton IFAU Vrtues of SIN effects of an mmgrant workplace ntroducton program 27