This article reports on a study of the impact of unemploymentduration on the probability of becoming a social assistancerecipient and on the time spent on social assistance. The dataare taken from a local Finnish labour market consisting of thecohort of unemployed people at a given date (n = 2,274). Thedata are linked to information about the number of monthsrecipients received social assistance during the subsequentyear. Count data regression models of the hurdle type areestimated. The results of the analysis suggest that the impactof unemployment duration on the probability of becoming asocial assistance recipient is explained to a large extent bychanges in the distribution of types of unemployment benefitsbetween people with different lengths of time spent inunemployment. Unemployment assistance recipients are morelikely to become social assistance recipients and to spendlonger periods on social assistance than are recipients ofunemployment insurance because the unemployment benefitsfor the latter are higher. Among social assistance recipients,time spent on the transfer increases with unemploymentduration only for those who are in frequent need of assistance.
Jan SaarelaÅbo Akademi University, Vasa, Finland
How unemployment duration affects social assistance receipt: evidence from Finland
Key words: unemployment duration, unemployment, socialassistance receipt, Finland
Jan Saarela, Department of Social Sciences, Åbo Akademi Uni-versity, PO Box 311, FIN-65101 Vasa, FinlandE-mail: [email protected]
Accepted for publication October 7, 2003
Introduction
Unemployment rates have a tendency to drive upthe rate of social assistance recipients in manyindustrialised societies (see e.g. Gustafsson, 1984;Stenberg, 1998). This seems to be the case in Finlandas well: the correlation between the unemployment rateand the rate of social assistance recipients at themunicipality level was close to 0.60 in 1996 (author’scalculations based on Stakes, 1997a and StatisticsFinland, 1998). Time series studies display similarpatterns. The sharp rise in unemployment during thefirst half of the 1990s was accompanied by an increasein the rate of social assistance recipients (Figure 1).After 1994 the unemployment rate fell, but the rate ofsocial assistance recipients remained high, which is inconcordance with the delayed macroeconomic effectfound by Stenberg (1998) for Sweden. The averageduration of completed unemployment spells were attheir peak in 1994 and decreased relatively modestlythereafter (Figure 2). Uncompleted unemploymentspells, on the other hand, continued to increase afterthat time. Specifically, for completed spells, the patternis similar to the pattern for the rate of social assistance
recipients. As such, this provides a motive forstudying the impact of unemployment duration onsocial assistance receipt at the individual level.
There are several reasons behind this causality; forexample, a higher degree of social assistance recidivism,longer duration of social assistance spells and increasesin the number of recipients who are outside the labourforce (Gustafsson, 1998). Perhaps the most obviousreason, however, is that unemployed people frequentlybecome social assistance recipients, which is the topicof this article. Data representing a local Finnish labourmarket was used to study both how the length ofindividual unemployment spells affect the probabilityof becoming a social assistance recipient and the timespent on social assistance.
There are only a few earlier studies that have beenconcerned with these issues, largely because of thejudicial and practical difficulties of linking the admin-istrative records of unemployed people with the recordsof social assistance recipients. Previous research is alsofairly restrictive in terms of data quality and statisticalmethods (see Dahl, 1995; Seppänen, 1997). The mostrigorous study was conducted by Gustafsson (1998),who utilised data on first-time unemployed males in a
local Swedish labour market. He found that lengthyperiods of unemployment combined with no access tounemployment compensation increased the risk ofbecoming a social assistance recipient. Among indi-viduals having access to unemployment benefits, on theother hand, the risk did not increase with unemploy-ment duration. The importance of accounting for theinterrelation between the unemployment compensationscheme and the social assistance scheme has also beenput forward by other researchers (see e.g. Barrett,Doiron, Green & Riddell, 1996; Salonen, 1999).
Theoretical and institutional considerations
It is fairly obvious why unemployment duration affectsindividual social assistance receipt. When a personbecomes unemployed, income and private savings arelikely to decrease, becoming less and less the longerthe person stays unemployed (Gustafsson, 1984). Con-sidering that social assistance is a means-tested transferto individuals on a low income, it is thus natural toexpect that the need for assistance may increase with
time spent in unemployment. Similarly, time spentin social assistance could depend on unemploymentduration because it is strongly related to the availabilityof paid work (Gustafsson & Voges, 1998).
In the field of labour economics, most studies ofa related nature have been concerned with how anindividual’s labour supply is affected by time spentin social assistance (see e.g. Danziger, Haveman &Plotnick, 1981; Moffitt, 1992). The motivation behindthis approach is that leisure-activity preferences,household composition and future labour-marketopportunities may change with exposure to the socialassistance programme (Arslanogullari, 2000; Blank,1989; Blank & Ruggles, 1994; Moffitt, 1983).
Causality may be in two directions: fromunemployment (i.e. absence of labour supply) to socialassistance receipt, and from social assistance receipt tounemployment. To separate one effect from the other,it is important to know the timing of each event.
In Finland all categories of residents may becomeeligible for social assistance, which is then paid outafter the application has been processed. In determiningeligibility, the social worker compares the monthlydisposable income of the applicant household with agiven threshold for that particular type of household. Ifthe disposable income is lower than the threshold, theapplicant will receive social assistance amounting to thedifference between the household’s disposable incomeand the threshold. The rules also specify that theapplicant must have no other source of income and thatpersonal assets must be spent before an application canbe approved.
Social assistance consists of three elements: (1) abasic sum, (2) rental expenses for accommodation and(3) other means-tested expenses. The amount of thebasic part depends on the type of household, such asthe presence of a partner or spouse and the age ofchildren. The maximum amount allotted to compensatefor rental expenses is dependent on household size andthe construction or restoration year of the housing.Means-tested expenses are of a temporary nature andare related to health-service costs and additional costsfor housing.
Receipt of an unemployment benefit implies thatsome unemployed people are not eligible for socialassistance. Unemployment benefits in Finland aredivided into two separate categories: unemploymentinsurance (UI) and unemployment assistance (UA). Inorder to apply for one of these, the unemployed personmust be registered as a job seeker at the localEmployment Service. UI is provided to members of thevarious UI funds. Anyone who works in a field coveredby a fund has the right to become a member. To qualifyfor UI, a work requirement is imposed. This means that,while a claimant is a member of a UI fund, he/she musthave had work experience of at least 26 weeks during
Figure 1. Unemployment rate and rate of social assistance recipientsin Finland 1991–1998.Sources: Ministry of Labour, 2000; Stakes, 1997b, 1999, 2000;Statistics Finland, 1994.
Figure 2. Average duration of completed and not completedunemployment spells in Finland 1991–1998.Source: Ministry of Labour, 2000.
Unemployment duration and social assistance in Finland
the 100 weeks prior to registering as unemployed. UIdepends on prior earnings. The replacement rate (UIin relation to prior earnings) is about 0.5. UI can bereceived for a maximum of 100 weeks. Thereafter theperson is only eligible for UA.
UA is provided to applicants who are not membersof any UI fund. It consists of a fixed amount (about $20per day in 1996) that is less than half of the averageamount provided by UI (Saarela, 2000). If the applicantdoes not fulfil the work requirement, or if the maximumpayment period of 100 weeks has been exhausted, thebenefit becomes dependent on the spouse’s earnings,and is thus potentially even lower.
Since income support provided by the UA schemeis lower than that provided by the UI scheme, it isplausible that UI recipients are less dependent on socialassistance than UA recipients, and that the two groupsdiffer also with regard to how unemployment durationaffects social assistance receipt.
Data and methodology
The data utilised in the present study relate to allindividuals who were registered as unemployed at theEmployment Service in the local labour market of theCity of Vasa in Finland on January 1, 1996 (n = 2,274).The date each person became unemployed is known.The City of Vasa is situated in the western part of thecountry and has about 55,000 residents. In terms ofunemployment, social assistance receipt and socio-economic and demographic composition, it is fairly re-presentative of the non-rural part of Finland (Saarela &Finnäs, 2003; Stakes, 1997b, 2000; Statistics Finland, 1998).
These administrative records are linked to informationon how many months the same individuals receivedsocial assistance in 1996, obtained from the local socialservices agencies. It is not known which specific monthsin 1996 individual recipients received social assistance.
In the available data it is possible to follow peoplein the cohort up to 18 September 1996, as well as tostudy the inflow into the unemployment register upto that date. There are, however, causality problemsinvolved in these cases, since the starting point of socialassistance receipt during 1996 is not known. The studytherefore concentrates on reporting how unemploymentduration up to 1 January 1996 of all individuals in theunemployment register at that time affected subsequentsocial assistance receipt during 1996.
Total unemployment duration up to 18 Septemberand certain variables reflecting whether people remainedin the unemployment register during the period between2 January and 18 September have also been used. Theresults did not change to any noteworthy degree.
Since the used part of the data represent the cohortof unemployed people at a specific date, long-termunemployed people are overrepresented. This fact should
be kept in mind when interpreting the results. Estima-tions for people in the inflow to the unemploymentregister were made, but since there are causality problemsinvolved in making inferences about the behaviour ofthese, the results are not reported here.
Originally, an analysis of how unemploymentduration affects the amount of social assistance receivedwas also attempted, since there is information avail-able on the total amounts received during 1996. Theinterpretation of these results is very complex, however,because the unemployment duration variable is relatedto several factors that are not taken into considerationin the data, such as household composition, changes inother income variables and temporary expenses. Theresults of these analyses are therefore not reported here.
Unemployed people are, by definition, people whoreceive an unemployment benefit (UA or UI) and areregistered as job seekers at the employment office. Thedata also contain information about other job seekers,i.e. 33.5 per cent of all job seekers in the cohort. Thesepeople are not unemployed; rather they are either em-ployed or are outside the labour force (e.g. students).1Since the aim of the study is to explore the impact ofunemployment duration, not the job search durationamong all job seekers registered at the employmentoffice, the group of job seekers who are not unemployedhave also been excluded from the analysis.
The data set also contains information on the socio-economic and demographic characteristics of the targetgroup, such as age, gender, education, native language,number of dependant children, labour-market statusbefore registering at the employment office and natureof previous employment, public sector work and super-visory tasks. These characteristics are used as controlvariables in the analysis.
Some descriptive statistics of the data are providedin Table 1. The second column gives the distribution ofsocio-economic and demographic characteristics; thethird column outlines the proportion of social assistancerecipients according to each characteristic. As expected,the proportion of social assistance recipients is greaterin the higher unemployment-duration categories. Theproportion of recipients is also relatively high amongmiddle-aged individuals, males, people with only abasic level of education and people with children. It isalso higher for Finnish-speaking individuals and peoplewith native languages other than Swedish, which seemsto reflect the more favourable labour-market positionof the Swedish-speaking part of the population (Saarela,2002; Saarela & Finnäs, 2003). The proportion ofsocial assistance recipients is also high for individualswho were previously outside the labour force, people
1 At the national level in 1996, about 33.9 per cent of all jobseekers were not classified as unemployed.
without previous managerial tasks and for personsreceiving unemployment assistance.
The distribution of months spent on social assist-ance, conditional to social assistance receipt, is shownin Table 2. Many recipients have been on socialassistance for only one month, others for almost thewhole year (nine to twelve months). It would thusseem that there are two groups: unemployed people intemporary need of social assistance and unemployedpeople in frequent need of assistance. The third columnin Table 2 shows the median duration of unemployment
for each number of months on social assistance. Asexpected from the earlier arguments, individuals whohave spent longer on social assistance have also beenunemployed for longer.
The main question of interest is whether unemploy-ment duration has an independent effect on socialassistance receipt and time spent on social assistance,or if it is just a result of the distribution of socio-economic and demographic characteristics of people indifferent unemployment duration categories. This issuewas analysed with the help of multivariate models.
Count data regression models of the hurdle typewere used (Cameron & Trivedi, 1998: 123–125;Melkersson, 1999; Mullahy, 1986). The principlemotive for using count data regression models hasto do with the nature of the dependent variable, i.e.the number of months on social assistance. It can beargued that the variable is binomially distributedsince it may take the values 0, 1, 2, . . . , 12. The basicidea is consequently that a binomial probabilitygoverns the outcomes of interest. A hurdle approachwas taken because of the large number of zeros; 1,285individuals, 56.5 per cent, have zero months on socialassistance.
The hurdle model is a two-part model. The first partis a binary outcome model – whether or not the personis a social assistance recipient. The second part is atruncated-at-zero count model – the number of monthson social assistance, conditional on social assistancereceipt. In both cases, a logistic specification wasadopted. The partition permitted the interpretation thatpositive observations arise from crossing the zerohurdle, i.e. the zero-months threshold. The first partthus modelled the probability that the threshold wascrossed, whereas the second part modelled time spenton social assistance, given that the threshold was
Table 1. Distribution of background variables in the data, and percentage of social assistance recipients in each category (n = 2,274).
Table 2. Distribution of months on social assistance, conditional on social assistance receipt, and median duration of unemployment in each category (n = 989).
crossed.2 In each of the two stages, the outcome ofinterest was assumed to be dependent on the similarbackground variables.
Results
The unemployment duration variable was divided intothe intervals <3, 3–4, 5–6, 7–10, 11–18 and >18 monthsof unemployment, in order to best fit the data. Theeffect of the variable is not perfectly linear, whichalso suggests a categorisation. Another motive, whichconcerns the choice of setting the first interval at<3 months, is that because it takes time for an unem-ployment benefit application to be processed, thereis often a delay of some weeks in paying out theunemployment benefit, which may generate a highneed for social assistance during the first weeks ofunemployment.
The first part of the hurdle model was estimatedin a stepwise fashion in order to determine the extentto which single sets of variables contributed to theprobability of being a social assistance recipient. InModel 1 in Table 3, only the unemployment durationvariable was used. As already indicated by descriptivestatistics, there was a significant positive effect ofunemployment duration on social assistance receipt. Asan example, the probability of being a social assistancerecipient was 31.4 per cent if an individual had beenunemployed for less than three months, in comparisonwith 64.3 per cent if he or she had been unemployedfor more than 18 months.3
There was no significant change in the estimatedimpact of unemployment duration, or in the impactof each control variable, when additional control vari-ables were included into the estimations. I have there-fore reported only the model in which all variablesrepresenting socio-economic and demographic charac-teristics are used (Model 2). The estimates for theeffects of these variables were as expected and will bementioned only briefly.
The results indicate that social assistance receiptwas lower in the higher age brackets, which might beexplained by a higher degree of accumulated incomeand savings. The probability of being a social assistancerecipient was also lower for females than for males. Theprobability decreased with education as well, which isa reflection of a higher degree of human capital and,thus, income and savings. Similar arguments may
explain why the probability of being a social assistancerecipient is relatively low for individuals who werepreviously employed and for people with previousmanagerial tasks. Unemployed people with childrenare more likely to be social assistance recipients thanthose without children, since such families are in needof greater income support. In comparison with Finnish-speakers, Swedish-speakers are less likely to be socialassistance recipients, and those with other nativelanguages more likely.
Thus, the impact of unemployment duration on socialassistance receipt cannot be attributed to differencesin the distribution of socio-economic and demographiccharacteristics. When type of unemployment benefitwas included in the estimations (Model 3), however,the effect of unemployment duration tapered off sub-stantially. It can also be seen that social assistancereceipt was much more common among UA recipientsthan among UI recipients. The probability of being asocial assistance recipient was 41.8 per cent if thereference individual was a UA recipient, in comparisonwith 10.3 per cent if he or she was a UI recipient.4
As argued previously, it is plausible that the effectof unemployment duration on social assistance receiptdiffers between UA recipients and UI recipients.Model 4, therefore, includes interaction terms betweenunemployment duration and type of unemploymentbenefit. The results indicated that a large part of thepositive effect of unemployment duration on socialassistance receipt disappeared.5 Many of theseestimates did differ from each other at a reasonablelevel of significance.
The findings consequently suggest that the impact ofunemployment duration on social assistance receipt canbe explained to a large extent by differences in theproportion of UA (or UI) recipients in the unemploy-ment duration categories. For example, the proportionof UA recipients, who have a substantially lowerunemployment compensation than UI recipients, was42 per cent among those unemployed for less thansix months, as compared with 85 per cent for thoseunemployed for more than ten months.
For the second part of the hurdle model, a similarmethod was followed. These estimates also changedonly marginally when additional control variables wereincluded, and the results are therefore presented in asimilar manner (Table 4).
It should be noted that a coefficient represents theestimated effect of each variable on the probability(or actually, log odds) of being one month on socialassistance, conditional to social assistance receipt,
2 The approach seems reasonable if it is considered that theremay be a stigma related to social assistance receipt (seeGustafsson, 2002; Moffitt, 1983), which would imply that theunderlying behaviour related to social assistance receipt perse differs from behaviour related to time spent on socialassistance.
under the random-drawings assumption. The lattermeans that social assistance propensity for one monthis independent of that for the following month. Theunderlying probability is thus assumed to be the samefor each potential month in the order, given that theperson is a social assistance recipient.
Since the results apply to social assistance recipientsonly, it is not surprising that the effects of the controlvariables differed to some extent from the effectsreported social assistance receipt. It can be seen thatthe number of months on social assistance washigher for older recipients, for those with a low levelof education, for those with children, for those with
other native languages than Finnish or Swedish and forthose who were previously employed in the publicsector. The impact of gender, previous labour-marketstatus and previous managerial tasks seem to berelated to the type of unemployment benefit the personreceives.
There was, also for this second stage of the hurdlemodel, a significant effect of unemployment duration,which tapered off somewhat when the type of un-employment benefit was included in the estimations.To some extent, the impact of unemployment durationalso diminished when UI recipients and UA recipientswere allowed to be affected differently by the variable.
Figure 3. Predicted probability of being on social assistance different number of months according to unemployment duration, referenceindividual in Model 2.
Figure 4. UI recipients’ predicted probability of being on social assistance different number of months according to unemployment duration,reference individual in Model 4.
Unemployment duration and social assistance in Finland
It is difficult to make a straightforward interpretationof the estimates in terms of time spent on socialassistance. I therefore performed some calculations, theresults of which are reported in Figures 3, 4 and 5.These figures give the predicted probability of being onsocial assistance for a given number of months, in thevarious unemployment duration categories. Figure 3 isbased on the results of Model 2, whereas Figures 4 and5 represent the results of Model 4.
Figure 3 indicates that the probability of beingless than six months on social assistance seems to beinversely related to unemployment duration, whereasthe probability of being six months or more on socialassistance increases with unemployment duration. Thissuggests that time spent in unemployment has anincreasing effect on time spent on social assistance onlyfor individuals who have a frequent need for socialassistance. Figures 4 and 5 show that this patternremained basically the same even when the type ofunemployment benefit received was taken into account.It should be noted, however, that UA recipients aremuch more likely to spend longer periods on socialassistance than are UI recipients, even when the impactof socio-economic and demographic characteristicshave been controlled for.6
Conclusions
The results presented in this article indicate that theimpact of individual unemployment duration on theprobability of becoming a social assistance recipientcan be explained to a large extent by changes in thedistribution of types of unemployment benefit amongpeople with different lengths of time spent inunemployment. The proportion of UA recipients ishigher among those who have been unemployed for alonger period of time. As a result of the poor economiccompensation offered by the unemployment benefitscheme compared with compensation paid to UIrecipients, UA recipients are more likely to becomesocial assistance recipients and to spend longer periodsof time on social assistance. The results also suggestthat, only for social assistance recipients who are infrequent need of social assistance, unemploymentduration has an increasing impact on time spent on thetransfer.
The policy implications of the findings are that, inorder to reduce public spending on social assistance,which is a major issue in many societies, measuresdirected specifically at the unemployed are needed.These measures should be focused on providingemployment opportunities first and foremost for themost economically deprived, such as unemployedindividuals who are UA beneficiaries and socialassistance recipients who spend many months of theyear on social assistance.
The present analysis is limited, however, with respectto a number of factors. In order to avoid causality prob-lems, which could still be present if the possibility of
6 The results were similar when separate models for UIrecipients and UA recipients were estimated. The decline inthe predicted probability for UI recipients who had beenunemployed for more than 18 months was because of a smallnumber of observations; UI is exhausted at 700 days ofunemployment.
Figure 5. UA recipients’ predicted probability of being on social assistance different number of months according to unemployment duration,reference individual in Model 4.
social assistance recidivism is taken into consideration,the data are restricted to the group of unemployed ata given date and in a local labour market. Time spenton social assistance was studied only in terms of thenumber of months on social assistance during the yearfollowing the date at which the cohort sample wasdrawn. Furthermore, the available information concernedonly single spells of unemployment. Future researchefforts should therefore be directed towards obtaininglongitudinal, nationally representative data with linkedinformation on multiple unemployment spells andmultiple social assistance spells. Just constructing suchdata sets needs to be put at the top of the agenda,because no such data sets exist in Finland today.
ReferencesArslanogullari S (2000). Social assistance in Sweden. Licentiate
thesis. Uppsala, Uppsala University, Department ofEconomics (Working Paper No. 90).
Barrett GF, Doiron DJ, Green DA, Riddell WC (1996). Theinteraction of unemployment insurance and social assistance.Quebec, Hull, Human Resources Development Canada (UIEvaluation Technical Reports, Evaluation Brief No. 18).
Blank R (1989). Analyzing the length of welfare spells. Journalof Public Economics 39: 245–273.
Blank R, Ruggles P (1994). Short-term recidivism among public-assistance recipients. American Economic Review 84: 49–53.
Cameron AC, Trivedi PK (1998). Regression analysis of countdata. Cambridge, Cambridge University Press (EconometricSociety Monographs No. 30).
Dahl S-Å (1995). Arbeidslöses bruk av ökonomisk sosialhjelp[Unemployed individuals’ use of social assistance]. Bergen,Stiftelsen for Samfunns- og Naeringslivsforskning (Paperpresented at the 8th Nordic Social Policy Research Meeting).
Danziger S, Haveman R, Plotnick R (1981). How incometransfers affect work, savings, and the income distribution: Acritical review. Journal of Economic Literature 19: 975–1028.
Gustafsson B (1984). Macroeconomic performance, old agesecurity and the rate of social assistance recipients inSweden. European Economic Review 26: 319–338.
Gustafsson B (1998). From the Employment Office to the SocialWelfare Office: social assistance recipiency among first-timeunemployed in Sweden. European Journal of Social Work 1:203–220.
Gustafsson B (2002). Assessing non-use of social assistance.European Journal of Social Work 5: 149–158.
Gustafsson B, Voges W (1998). Contrasting welfare dynamics:Germany and Sweden. In: Leisering L, Walker R, ed. TheDynamics of Modern Society. Bristol, The Policy Press.
Melkersson M (1999). Explaining choice set size forunemployed. Applied Economics 31: 1599–1607.
Ministry of Labour (2000). Finnish Labour Review 43.Moffitt R (1983). An economic model of welfare stigma.
American Economic Review 73: 1023–1035.Moffitt R (1992). Incentive effects of the U.S. welfare system:
A review. Journal of Economic Literature 30: 1–61.Mullahy J (1986). Specification and testing of some modified
count data models. Journal of Econometrics 33: 341–365.Saarela J (2000). Vad innebär en reform av det finländska
arbetslöshetsskyddet? [What are the implications of a reformof the Finnish unemployment benefit system?]. EkonomiskaSamfundets Tidskrift 53: 197–207.
Saarela J (2002). Språkgruppsskillnader i utkomststödstagande[Language-group differences in social assistance receipt].Ekonomiska Samfundets Tidskrift 55: 89–97.
Saarela J, Finnäs F (2003). Unemployment and native language:The Finnish case. Journal of Socio-Economics 32: 59–80.
Salonen T (1999). Den främre parentesen och socialbidraget[The principal parenthesis and social assistance]. Lund,University of Lund, Socialhögskolan (Meddelanden frånSocialhögskolan No. 5).
Seppänen VM (1997). Turvaverkon vankina?Pitkäaikiaistyöttömät ja toimeentulotuki [Caught in the safetynet? The long-term unemployed and social assistance].Helsinki, Ministry of Social Affairs and Health (Publicationsof the Ministry of Social Affairs and Health).
Stakes (1997a). Sosiaali- ja terveydenhuollon taskutieto 1997[Brief information about social and welfare services 1997].Helsinki, National Research and Development Centre forWelfare and Health.
Stakes (1997b). Living allowance 1996. Helsinki, NationalResearch and Development Centre for Welfare and Health(SVT Sosiaaliturva No. 3).
Stakes (1999). Facts about Finnish welfare and health care1999. Helsinki, National Research and Development Centrefor Welfare and Health.
Stakes (2000). Facts about Finnish welfare and health care2000. Helsinki, National Research and Development Centrefor Welfare and Health.
Statistics Finland (1994). Statistical yearbook of Finland 1994.Helsinki, Statistics Finland.
Statistics Finland (1998). Employment statistics. Helsinki,Statistics Finland (Population 1997 No. 15).
Stenberg S-Å (1998). Unemployment and economic hardship –A combined macro- and micro-level analysis of the relation-ship between unemployment and means-tested social assistancein Sweden. European Sociological Review 14: 1–13.
