The scarring effects of unemployment

The longer the spell of unemployment, the harder it becomes to keep and update acquired knowledge and proficiencies. The unemployed individual rapidly loses her/his value in the labour market and is put at risk of becoming unattractive to potential employers. The human capital effects of unemployment are also connected with discouraged worker effects among the long-term unemployed. Repeated rejection in the labour market is connected with lower interest in active job searching, as expectations for future employment are low (Regnér, Reference Regnér1997). We may add to this the serious risk that individuals with long unemployment spells will be at the very bottom of potential employers' lists, owing to the stigmata of long-term unemployment. Research shows that employers have negative attitudes towards hiring long-term unemployed workers (Harkman, Reference Harkman1999) and that employers discriminate against workers with earlier unemployment spells, as such spells are taken as a sign of unreliability and/or low productivity (Maguire, Reference Maguire and McLaughlin1992; Regnér, Reference Regnér1997; Harkman, Reference Harkman1999; Gallie and Paugam, Reference Gallie and Paugam2000; Calmfors et al., Reference Calmfors, Forslund and Hemström2002).

Unemployment scarring has been well documented in previous research. For instance, Groot et al. (Reference Groot, Schippers and Siegers1990) investigated the effects of interruptions in labour market participation and found that workers who become unemployed (or who interrupt their careers for other reasons) suffer from a decreasing human capital stock. Further, the longer the unemployment spell continues, the more underprivileged they become with regard to missing experience, in comparison with those whose careers are not interrupted. Educational investments during or after the unemployment spell reduce depreciation of the human capital stock and have been shown to limit scarring effects (Groot et al., Reference Groot, Schippers and Siegers1990; Gregg and Tominey, Reference Gregg and Tominey2005). This result, showing that career interruptions generate negative human capital effects, seems to correspond well with most other studies within the field that document scarring effects in terms of outcomes. For individuals, unemployment experience is connected with detrimental labour market employment prospects, and previous unemployment experiences have implications for the future labour market experience (Arulampalam et al., Reference Arulampalam, Booth and Taylor2000; Layte et al., Reference Layte, Levin, Hendrickx, Bison, Gallie and Paugam2000; Burgess et al., Reference Burgess, Propper, Rees and Shearer2003).

The experience of unemployment also increases the probability of downward occupational mobility, often to the unskilled manual class (Layte et al., Reference Layte, Levin, Hendrickx, Bison, Gallie and Paugam2000). In a two-country comparison between Germany and the US, Gangl finds that occupational mobility is connected with previous unemployment spells. More than 60 per cent of the US workers were re-employed into occupations other than their pre-unemployment job. The number of workers who experienced the same occupational mobility in Germany was about 30 per cent. Given that the main occupational mobility for the unemployed is downward, he also documents substantial wage losses connected with unemployment. Among US workers, the loss was about 20 per cent and for German workers a somewhat lower 16 per cent (Gangl, Reference Gangl2004).

Unemployment spells thus affect not only the occupation mobility rate but also wages. Relatively short periods in unemployment also have negative consequences for wages. For instance, estimates based on Swedish data showed that a spell of unemployment lasting ten weeks reduces wages by 7 per cent. This effect persists for more than six years (Björklund, Reference Björklund1981). Similar direct and delayed wage effects of unemployment have been widely documented in diverse labour markets. In the UK, the wage penalty associated with unemployment has been found to be about 6 to 10 per cent on re-employment, and 10 to 14 per cent after three years, whereas long-term effects from youth unemployment have been found to be as high as 13 to 21 per cent (Arulampalam, Reference Arulampalam2001; Gregg and Tominey, Reference Gregg and Tominey2005). Studying the situation in Germany, Bender et al. (Reference Bender, Dustmann, Margolis and Meghir1999) found that workers who were re-employed relatively quickly only lost up to 2 per cent of their pre-unemployment wage. Length of unemployment, however, was of importance for scarring, and those who spent over a year in unemployment before they found a new job suffered a ‘permanent additional wage penalty’ of 19 per cent (Bender et al., Reference Bender, Dustmann, Margolis and Meghir1999). In a US setting, wage scarring has been found to be even higher, around 20 per cent (Hipple, Reference Hipple1999; DiPrete, Reference DiPrete2002).

Active labour market programmes – a protection against scarring effects?

The fact that studies have shown that scarring is reduced by investments in education before, during and after unemployment (Groot et al., Reference Groot, Schippers and Siegers1990; Burgess et al., Reference Burgess, Propper, Rees and Shearer2003; Gregg and Tominey, Reference Gregg and Tominey2005) would seem to indicate that ALMPs could be an important tool in trying to reduce unemployment scarring due to long-term unemployment. Such a protective effect would be in accordance with the human capital perspective that has traditionally emphasised the importance of viewing skills and knowledge as a central form of capital in Western economies. From the point of view of human capital theory, productivity – and thus compensation and employability – could basically be reduced to the human capital investments an individual has made. The most important of these would be investments in the form of education and training (Becker, Reference Becker1993). As ALMPs aim to maintain, expand or re-adjust skills and training of the unemployed, we may expect them to have a positive impact on productivity and thus to reduce unemployment scarring. However, it is not possible to reduce human capital loss in unemployment to reductions in the value of existing skills, education and training. Self-confidence and mental wellbeing (see, for example, Kessler et al., Reference Kessler, Turner and House1989; Lahelma, Reference Lahelma1992; Patterson, Reference Patterson1997; Strandh, Reference Strandh2000), social skills and motivation (Regnér, Reference Regnér1997) have been shown to be affected by unemployment. These factors could affect scarring through discouraged worker effects or act as indicators of low productivity for potential employers. Here it is possible that ALMPs reduce scarring through activating the unemployed and increasing their sense of control (Bane and Ellwood, Reference Bane and Ellwood1994).

There are thus good reasons to expect positive effects of ALMPs on unemployment scarring. Yet despite this, findings have been somewhat inconclusive. A number of studies from different European contexts point towards positive effects of ALMPs, at least with regard to returning to the labour market (Lubyova and van Ours, Reference Lubyova and van Ours1999; Eichler and Lechner, Reference Eichler and Lechner2002; Gerfin and Lechner, Reference Gerfin and Lechner2002). In a Swedish and Scandinavian context, there are also a number of studies showing positive effects of ALMPs (Harkman et al., Reference Harkman, Jansson and Tamás1996). For instance, during the 1980s, more than 50 per cent of those attending ALMP training had a job, within the area of the training, in the regular labour market six months after the end of the programme. To this number another 10 per cent can be added: those who had a job but within a different area than that of the ALMP programme. In 1984, as much as 70 per cent got a job after the ALMP training period (Johansson, Reference Johansson and Furåker1989). Moreover, a similar study carried out in 2003 found that ALMP participants have greater probabilities of returning to the regular labour market than do unemployed individuals not participating in ALMP (AMS, 2005). Positive effects have also been found on wages, and programme participation appears to reduce unemployment scarring in the form of wage penalties. Edin (Reference Edin1988) finds that, in the long run, spending ten weeks in an ALMP decreases the risk of negative effects on wage development. In a Norwegian setting, ALMPs have been found to have positive effects on the probability of becoming self-sufficient among unemployed social assistance recipients (Dahl and Lorentzen, Reference Dahl and Lorentzen2005).

While many studies indicate positive effects of ALMPs, others point towards a non-effect or even negative effects. This is also the OECD's conclusion based on analysis of available studies of ALMP effects in OECD countries (OECD, 1996). Macro-economic analyses have pointed towards a non-effect of ALMPs on the outflow from unemployment (Puhani, Reference Puhani2003). This, of course, does not directly translate into a non-effect at the individual level. Such a non-effect, however, has been found in several micro-data analyses of Swedish programmes (AMS, 1997). Even negative effects on the outflow from unemployment to employment in the regular labour market are found. When studying unemployed youth's participation in training programmes, 1992–94, only negative effects were found on the probability of entering a job in the regular labour market as well as on the probability of entering a regular education programme within two years (Larsson, Reference Larsson2000). Other micro-level evaluations of programmes have similarly found no significant effects of individual participation in particular ALMPs, and wage trend analyses largely show that only a few individuals experience wage lifts after ALMP attendance as compared with openly unemployed control groups (Ackum, Reference Ackum1991; Regnér, Reference Regnér1993; AMS, 1997; Larsson, Reference Larsson2000). In a 1995 summary of Swedish studies on wage trend effects of ALMPs, Forslund and Krueger argue that, from a macro-economic point of view, if ALMP attendance raises the yearly income by 3 per cent over a period of 20 years, the value of the outcome balances the level of the costs for the Swedish economy. They find, however, that there is not enough support to draw the conclusion that ALMP attendance has an effect on participants' future wages (Forslund and Krueger, Reference Forslund and Krueger1995).

This inconclusiveness on the effects of ALMPs is also emphasised in a review of the effectiveness of Swedish labour market policy from the early 1980s. The tendency was that positive effects are more commonly found in studies from the 1980s, while studies from the 1990s more often indicated no or negative effects. However, it is not only a matter of when the study was made but also what kind of programme is being evaluated (Calmfors et al., Reference Calmfors, Forslund and Hemström2002). Most evaluations and studies have concentrated on the effects of particular programmes, and these programmes vary substantially in duration, quality and not least in direction. There are basically two types of ALMPs that serve very different purposes. One form is what could be labelled as subsidised employment, where the unemployed are supported as new entrepreneurs, and are temporarily employed in public programmes or through subsidised employment in the regular labour market. Another form of ALMP, which could be labelled ALMP training, aims to maintain, readjust or expand the human capital of the unemployed through training or education. The emphasis of ALMP training is more on the actual skill being achieved: on increasing human capital. In the case of ALMP employment, the aim is more to help individuals stay active, to establish networks and contacts and thus focuses more on creating and maintaining ability and social competence rather than building human capital in a more traditional sense.

Previous research suggests that such differences may very well underlie the differences found in effects, although findings point in different directions concerning which form of ALMP has better outcomes. Some studies indicate great differences between different AMLP programmes. Studying the situation in Switzerland, Gerfin and Lechner (Reference Gerfin and Lechner2002) find positive effects for an ALMP employment-type programme of subsidised temporary jobs in the regular labour market, whereas the effects of ALMP training are not as convincing. A Norwegian study by Dahl and Lorentzen (Reference Dahl and Lorentzen2005), on the other hand, indicates the opposite. Training programmes yield positive results and the employment programmes produce no significant effects. In Ireland, O'Connell (Reference O'Connell2002) finds that programmes with strong links to the labour market help increase employment duration and have positive effects on earnings from employment.

Naturally, the differences found concerning the effects of different forms of ALMPs in different countries may be dependent on the quality of the measures, thus indicating that programme content is of importance for their effects. One main factor for ALMP employment may be the extent to which ALMP employment programmes are tied to the regular labour market. Subsidising employment in the regular labour market does provide the desired network and contacts for the unemployed, the downside being, of course, that the risk of displacement effects is substantial. For ALMP training, the main factor is probably the quality of the training or education received by the unemployed. This is indicated in a Swedish study, where long vocational training (more than 100 days) showed better results than did short vocational training (fewer than 100 days). When comparing those attending vocational training with a control group, a slight positive effect on employment level was found for the group attending longer vocational training, while a negative effect on employment level was found for those with a shorter training period (AMS, 1997). Naturally, one problem associated with providing longer, quality training for the unemployed is the costs involved, especially when unemployment rates are high.

As we have seen, there are contradictory findings on the effect of ALMPs on job chances and unemployment scarring. This is true of both the general effects of ALMPs, although most Swedish studies appear to suggest negative or non-effects, and the differential effects of different types of ALPMs. The OECD points out two main problems with existing studies in this area. The first problem concerns the nature of the programmes. Most studies evaluate individual or only a few programmes. As these programmes have different aims, they serve different purposes and are also probably of varying quality, which makes it difficult to draw general conclusions. The second problem relates to measurement of the effects. In almost all cases, studies are based on immediate effects or on effects arising relatively soon after a completed programme. This is problematic from the perspective of ALMPs as human capital investments, where effects may be expected to be found on long-term labour market outcomes rather than on immediate outcomes (OECD, 1996). This is something which Calmfors et al. (Reference Calmfors, Forslund and Hemström2002) also argue, especially the importance to let more time pass from the point of ALMP activity to evaluation in times when the labour demand is low, as in the 1990s, as the time lag between ALMPs and effects are longer.

We would like to argue that, in order to understand the effects of ALMPs on unemployment scarring, one must first investigate ALMPs as a whole, at the same time as one takes into account the major differences between the two main forms of ALMPs. Second, as the concept of human capital growth concerns long-term investments, we must let time play a role in the study, allowing more time than is usual. Third, because human capital investments are long-term, it is of interest and importance to evaluate the extent to which the two different forms of ALMPs have differential impact over time.

Such differential impacts over time may well be found, given the major differences in how the two forms of programmes affect individual outcomes. The function of ALMPs in the form of training and education is to increase individuals' human capital through an increased level of education and skill, in order to improve long-term employability. Following participation, individuals either continue to further studies or look more broadly for work in the regular labour market, with new skills under their belts. The function of ALMP employment is somewhat different, as it is focused on human capital growth in the form of contacts/networks, ability and motivation. This gain is harder to use broadly in the labour market because it is difficult to formalise and, in the case of subsidised employment in the regular labour market, closely tied to a specific organisation, possibly making it less valuable outside that organisation. The effect of ALMP employment thus should be more immediate and in most cases connected with remaining within the organisation after the subsidisation period.

In this article, we will investigate the impact of the two forms of ALMPs from a long-term perspective. First, we will analyse, over a ten-year period, the extent to which ALMP training and ALMP employment impact on the probability of reaching pre-unemployment incomes, the probability of having an unemployment-free year, and the probability of completely leaving the labour market. Second, we will investigate whether the possible effects of the two forms of ALMPs are differently distributed over time. Our hypothesis here is that the differences in type of human capital investment between the two forms of ALMPs will produce effects that are differently distributed over time. Whereas the effect of ALMP employment will be immediate, with the positive effect declining over time, the effect of ALMP training will be lower, but stable or increasing over time.

Data and methods

Investigating the long-term effects of human capital investments in the unemployed places extreme demands on the data, both in terms of long reliable panels and the size of the panels. Most existing long-term survey-based panels are too small initially to enable detailed study of subgroups such as the unemployed, and suffer over time from problems of attrition. For researchers in the Scandinavian countries, a solution to these problems is to gain access to the comprehensive and electronically updated statistical register databases kept by the national statistics offices. In this study, we have made use of such a database, the LISA database (Longitudinell Integrationsdatabas för Sjukförsäkrings- och Arbetsmarknadsstudier) compiled by Statistics Sweden. LISA is a continuously updated database that contains yearly detailed demographic, income and employment/unemployment information, collected from public registers, on all individuals residing in Sweden.

The LISA data used for this study include information on all residents between the ages of 16 and 54 living in Sweden in 1992, who are followed yearly from 1991 to 2003, with some additional information, for example on class, collected from the 1990 Swedish census. All in all, the dataset from the outset consists of roughly 4.6 million individuals in 1991, with attrition caused by death (which is noted) or emigration leading to roughly 4.4 million individuals still being followed 12 years later. The dataset suffers from virtually no internal falling off and can be considered to represent all cohorts born 1937–1975 living in Sweden during those years, with the exception that it does not take into account individuals within the age span who migrated to Sweden after 1992.

The size and the quality of the data make it possible for us to use a two-pronged strategy for taking into account the selection issues that tend to plague research into ALMPs. There is here the possibility for selection both into unemployment in general and into ALMPs according to both measurable (which is no problem) and immeasurable qualities that can be connected with individual choices and qualities as well as labour market board policies. Our first strategy makes use of the fact that the size of the dataset enables subsampling, which helps us reduce selection effects into unemployment through creating a sample that is initially homogenous and well established in the labour market. The strategy here uses pre-unemployment information in order to minimise selection effects, with the assumption that previous labour market position is a good indicator of both measured and unmeasured qualities. The first years of the dataset are here used as selection years to obtain an unemployment sample that is initially homogenous, well established in the labour market and that has as few features as possible that might indicate a handicap in the labour market. Only those unemployed for more than half of 1993, who during 1992 had no unemployment or ALMP spells, had not been studying, had not had a child and had a labour market income exceeding 58,200Footnote ¹ SEK and a total income more than 90 per cent related to labour market activity (and not transfers) were selected. With these limitations, the dataset used in the study ended up comprising 51,535 individuals followed over a ten-year period after their initial unemployment.

The second strategy used is the application of a Heckman two-step procedure in order to control for selection bias into the two types of ALMPs among the unemployed (see, for instance, Heckman, Reference Heckman1976, Reference Heckman1979; Breen, Reference Breen1996). In the first step the probability in the unemployed sample of participation in the two types of ALMPs was analysed using probit regressions where the available independent variables are included. The effects of the unmeasured characteristics are here assumed to be present in the residuals of the probit analysis. These residuals were then used to construct two selection bias control variables, lambda 1 and lambda 2, which are assumed to reflect the effects of all unmeasured characteristics, which are related to participation in ALMP training and ALMP employment. These are in the second step included in the substantial analyses as additional independent variables, where they reflect the unmeasured characteristics related to participation in ALMP training and ALMP employment, and catch the part of the effect of these characteristics, which are related to the dependent variables. The statistical package used (SPSS) normally provides no procedure for applying the Heckman two-step procedure, but, using procedures developed by Smits (Reference Smits2003), some additional computations can be used to fairly easily construct Heckman selection bias control variables.

What is important when using the Heckman two-step procedure is that the first step contains at least one variable that is not related to the dependent variable in the substantial analysis (the second step). This variable functions as an identifying variable that allows the separate identification of participation in and the impact of the programme. In our case we have succeeded in identifying variables that do have a strong impact on participation in the two types of ALMPs, but no or negligible impact on the dependent outcome variables. These variables were thus used in the first step of the Heckman procedure, but omitted in the substantial analysis.Footnote ²

Dependent variables

The article uses three indicators to investigate whether ALMPs serve to maintain and activate human capital while in unemployment. As the study is longitudinal, time is the central factor in the dependent variables, and dependent variables relate to the probability over time of an event occurring. The main dependent variable used relates to the development over time of yearly incomes from employment or self-employment. In order to get a correct picture of income development, yearly incomes have been calculated in 1991 year prices using the Consumer Price Index (CPI). Income development was then used to create a variable that indicated the time it takes until an unemployed individual reaches her/his pre-unemployment yearly income. By focusing on return to previous income we here try to catch whether the two unemployment groups experience diverse difficulties in returning to their pre-unemployment income level. This should provide a good indicator if participation in the two types of ALMPs helps the unemployed to bounce back quicker and thus limit unemployment scarring. Here, roughly 35,700 individuals had earnings equal to that of the year before the unemployment spell during the time period studied, while about 15,700 still had not reached their previous income level after ten years.

In addition to the income-related variable, two further variables are used that indicate the effect of ALMPs over longer periods of time. Stability in employment and endurance in the labour market are two key indicators of employability and the value of human capital. To investigate this, two variables are used. The first is the time until respondents experience an unemployment-free year while still in the labour market. We would like to argue that one year without interruption by unemployment periods should provide a good indicator of relative stability in the labour market. This corresponds with the Swedish cut-off line for employers where they have to duly notify of non-continuation of employment contracts as well as inform temporary employed of labour market rights. About 44,800 cases experienced a year free from unemployment spells, while about 6,700 cases were right censored in the sense that, during the ten-year period, they did not experience an unemployment-free year.

The second variable, meant to indicate endurance in the labour market, is time until labour market exit (here, being on parental leave is counted as still being in the labour market). The measure itself, to leave the labour market, is a strong indicator of employability and endurance. A weakness with the data is, however, that it is not possible to distinguish between different forms of labour market exits (other than parental leave). The implications of leaving the labour market for early retirement are of course quite different from leaving the labour market for regular education. Despite this limitation, we would like to argue that it serves as a decent indicator of labour market endurance. The typical exit in our sample will definitely not be for educational reasons. Confirming this, the relationship between the exit variable and age is strongly positive. The youngest age groups (for whom education should be a more common possible exit than for others) have very low exit hazards that slowly rise with increasing age up until the oldest age group, those aged 50 and over, who have disproportionally higher hazards of exiting (around one and a half times that of the closest group, the 45–49 year olds). About 8,400 individuals left the labour market, with 43,000 cases right censored, that is remaining in the labour market during the entire period.