The Relevance of Job-Related Concessions for Unemployment Duration Among Recipients of Means-Tested Benefits in Germany

Abstract Unemployment strongly influences a person’s economic resources and life chances. Especially for unemployed individuals who have to rely on means-tested benefits, episodes of unemployment often go along with substantial material restrictions. Therefore, from a policy perspective, measures or regulations that might shorten unemployment episodes and reduce overall unemployment are particularly important. In this paper, we analyse whether concessions regarding the characteristics of the job searched for influence an individual’s unemployment duration. In doing so, we focus on a particular aspect of availability requirements in Germany. This is the fact that for unemployed recipients of means-tested benefits almost all types of jobs count as suitable employment and, therefore, recipients are obliged to make job-related concessions if offered a job requiring such concessions. The results indicate that there is no positive effect of making concessions regarding qualification requirements or status on employment chances. In contrast, there are positive effects of wage concessions. However, searching for a job in a different occupation (that does not necessarily imply a concession) has a comparable, positive effect on finding employment. Thus, it appears that being generally flexible regarding one’s future occupation might be at least as important for employment chances as making concessions.


Introduction
Unemployment strongly influences a person's economic resources and life chances, and thus, being unemployed is an important predictor of a person's position in the inequality structure. Especially for unemployed individuals who have to rely on means-tested benefits, episodes of unemployment often go along with substantial material restrictions. Therefore, strategies by which Jnl. Soc. Pol. (2022), 51, 2, 242-267 © The Author(s), 2021. Published by Cambridge University Press. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited. doi:10.1017/S004727942100009X unemployed individuals might improve their chances of finding a new job are a research topic of high (political) relevance.
On a societal level, (long-term) unemployment has consequences for the sustainability of the welfare state. Moreover, it might be detrimental to aggregate labour supply, productivity, and social cohesion. The recent social policy debate has shown that European welfare states have implemented reforms to address what Bonoli and Natali () call 'new social risks of post-industrial labour markets', such as flexibilisation and the emergence of atypical employment. In addition, retrenchment and cost-containment of the welfare state are important topics. This has led social policies to focus on promoting labour market participation through both human capital investments and activation measures (Bonoli and Natali, ). One risk of these developments is a dualisation of the labour market, preserving protection levels for insiders and introducing activation for outsiders that pushes the latter into poor-quality jobs (Bonoli and Natali, ; Palier and Thelen, ).
Germany's Hartz reforms, which took place in the early s, are an important example of the introduction of activating elements into protection for (long-term) unemployed individuals (Bonoli and Natali, ; Clasen and Clegg, ; Eichhorst et al., ; Dingeldey, ). Availability requirements for recipients of Unemployment Benefit II (UB II) define basically any job as suitable work. This also holds if the quality of the employment relationship offered is significantly lower than that of the previous one; i.e. benefit recipients are expected to make concessions: for example, regarding qualification requirements or employment conditions (c.f. §  SGB II)and they might be sanctioned if they do not meet these expectations (cf. § and § a SGB II).
The assumption behind this definition of suitable work and the resulting requirement to accept lower quality jobs is that lowering one's expectations should improve one's chances of finding employment. The literature shows that accepting employment with less favourable characteristics might indeed influence the chances of reemployment but also points to potential negative effects of making concessions (e.g. Caliendo et al., ; Korpi and Levin, ; Voßemer and Schuck, ).
Studies in this line of research usually apply quasi-experimental methods to identify the effects of transitions into such unfavourable employment relationships. What these studies cannot answer, however, is whether unemployed individuals would be prepared to make a concession. This is because these studies can only observe successful transitions into employment with unfavourable characteristics.
Other studies rely on hypothetical questions about intended behaviour, such as the willingness to accept employment with certain characteristics or reservation wages (e.g. Abraham et al., ; Bähr and Abraham, ; Krueger and Mueller, ), to study the willingness to make concessions. However, the answers to hypothetical questions about intended behaviour are not the same as actual, observed behaviour, even though they are often related (e.g. Petzold and Wolbring, ).
Using the German example, this paper provides an individual-level analysis of whether certain elements of activation have consequences for the labour market participation of unemployed individuals. In doing so, we focus on the requirement to make concessions regarding one's future occupation. To measure such concessions, we apply different operationalisations that do not rely on hypothetical questions. Instead, all these operationalisations utilise information regarding the occupations unemployed persons are searching for during unemployment.
More specifically, we use information on different characteristics of occupations, such as qualification requirements, socioeconomic status, and average wage. We can identify concessions by analysing whether, during their job search, unemployed individuals start searching for a new occupation with characteristics that are less favourable than the ones of the occupation they were previously searching for, i.e. whether they make a job-related concession.
To do so, we use administrative data on benefit receipt, unemployment, employment, and job search in Germany. These data feature complete information on all jobs that unemployed UB II-recipients are searching for, as agreed upon with their Jobcentre caseworkers. Using these data, we apply event history analysis to analyse whether unemployed individuals are prepared to make a concession, as well as the extent to which such a concession influences their employment chances.
The results indicate that there is no positive effect of making concessions regarding qualification requirements or socioeconomic status on employment chances. In contrast, we find positive effects of wage concessions. However, searching for a job in a different occupation (that does not necessarily imply a concession) has a comparable, positive effect on finding employment.
We structured the paper as follows. In the second section, we will discuss previous research on activation and concessions from a social policy, as well as an individual-level perspective. In section three, we give an account of the German context. In section four, we describe our data and methodology. In section five, we present our results before we conclude and discuss the policy implications of these results in section six.

Concessions and their role for leaving unemployment
Scholars of social policy distinguish between various forms of activation measures, which researchers often group into two broad categories: demanding and enabling measures (Eichhorst and Konle-Seidl, ). Both might include several elements (Marchal and van Mechelen, ). In the 'enabling' category,       these are, for example, active and passive labour market measures, investments in human capital, and job search assistance. Examples for the 'demanding' category are job-search requirements, monitoring of job search efforts, and sanctions. Knotz (), for example, shows that various elements of conditionality, such as availability requirements or job search and reporting requirements, as well as sanction rules have become stricter in many OECD countries over recent decades. While Knotz () discusses conditionality and sanctions mainly for unemployment insurance (UI), they also apply to many minimum income protection schemes (Marchal and van Mechelen, ; Frazer and Marlier, ).
In Germany, the Hartz reforms that took place in the early s were an important step in the introduction of demanding elements into the German system of benefits for the unemployed (Bonoli and Natali, ; Eichhorst et al., ; Dingeldey, ). Since these reforms, availability requirements in the newly introduced UB II scheme are among the strictest in international comparison (Immervoll and Knotz, ). Moreover, the regulations consider almost any type of employment to be suitable work. They require recipients of UB II to take up basically any job, regardless of their previous occupation, their qualification, and the quality of the new employment relationshipas long as the new employment helps to end or reduce benefit receipt. Moreover, if UB II-recipients decline acceptable employment, sanctions apply.
A common goal of such demanding requirements is to facilitate a faster reentry of the unemployed into the labour market. The underlying assumption of this type of policy is that lowering one's expectations regarding one's future jobi.e. to make concessionsshould improve one's chances of finding employment.  Moreover, as recipients are potentially unwilling to make concessions, sanctions are an important element of such policies, which shall ensure that the unemployed actually make the concessions required.
There is a substantial literatureparticularly in Economicsfocusing on the effects of benefit sanctions on employment outcomes. While this literature usually only observes sanctions (or the thread of sanctions), it implicitly or explicitly assumes some kind of concession to result from sanctionsmost commonly regarding the unemployed's reservation wagewhich should be responsible for the increase in employment chances of the unemployed.
A comprehensive discussion of this issue is, for example, provided in the literature on optimal unemployment insurance (for an overview, compare e.g. Fredriksson and Holmlund, ; Holmlund, ). In general, this literature aims at evaluating different features of UI-regulations and their consequences for welfare gains created by UI on the one hand and adverse incentive effects (which might e.g. result in longer unemployment duration) on the other (Holmlund, ). In addition to sanctions, particularly prominent design features in this context are, for example, eligibility criteria, the benefit level, the -      potential duration of benefit receipt, or the time profile of benefits over the spell of unemployment.
Indeed, papers focusing on sanctions in UI show that such sanctions increase reemployment rates (Abbring et al., ; Arni et al., ; Lalive et al., ; van den Berg and Vikström, ). Moreover, they do so not only ex posti.e. in cases where caseworkers actually impose sanctionsbut also ex ante, meaning that the mere thread of being sanctioned also seems to have a positive effect on finding a job (Arni et al., ; Lalive et al., ). Moreover, papers focusing on minimum income protection schemes and their recipients have shown that also in these latter schemes sanctions increase recipients' likelihood to find new employment (van den Berg et al., ; van der Klaauw and van Ours, ).
While sanctions thus have a positive effect on finding a job, they also seem to have a negative influence on job qualitye.g. regarding income, working time, or skill requirements (Arni et al., , ; van den Berg and Vikström, ).
Other strands of literature do not focus on the sanctions triggering different concessions, but instead investigate the concessions themselves, i.e. the taking up of employment with unfavourable characteristics, and their potential employment effects. Thus, for example, for fixed-term employment, Korpi and Levin () show that those who take up fixed-term employment spend less time in unemployment later on. Many authors also consider reservation wages (i.e. the lowest wage an unemployed person would accept for a specific job offer) to constituteor be a result ofa (wage) concessions. This is particularly true when the focus is on reservation wages' development over the unemployment spell and the effects on the reemployment chances of the unemployed (e.g. Krueger and Mueller, ; Bender et al., ).
For temporary agency work, Lehmer () demonstrates a positive effect of temporary agency work on workers' later chances of taking up regular employment. For marginal employment, Caliendo et al. () and Lietzmann et al. () show that at least for those with a longer unemployment duration, taking up marginal employment positively influences the likelihood of finding regular employment later on. Moreover, Voßemer and Schuck (), for example, show that taking up employment for which one is overeducated has positive effects on long-term employment chances.
On the other hand, such concessions also entail the risk that the (formerly) unemployed persons will get permanently locked in in the resulting unfavourable employment relationship. Thus, Voßemer and Schuck () show that accepting overeducated employment is actually a trade-off and that those accepting this type of employment run a substantially higher risk of getting stuck in this rather unfavourable type of employment. Another example is provided by Jahn and Rosholm (), who show for temporary agency work that people might get locked in in this type of employmentat least during an economic upturn.
Summing up, the aforementioned literature shows that accepting employment with less favourable features might indeed have a positive influence on the reemployment chances of unemployed individuals but also points to the potential negative effects of making such concessions. Therefore, especially from a policy perspective, it is a relevant question, whether and under which conditions unemployed individuals will be prepared to make a concession regarding their desired job features.
A common characteristic shared by a substantial amount of the literature on this topic is that analyses focusing on the willingness to make concessions often rely on self-reported answers to hypothetical survey questions about respondents' intended behaviour.
An example for this is the empirical research on reservation wages. Research on this topic usually relies on survey items, asking respondents a hypothetical question about the lowest wage or the minimum amount of income they would be willing to work for in a particular job (e.g. Krueger and Mueller, ; Addison et al., ; or, for Germany, Nivorozhkin et al., ; Bender et al., ).
In addition, such surveys might also refer to other dimensions in which the unemployed might make concessions, e.g. commuting distance or unfavourable working conditions. Such items are, for example, applied in studies of employability (e.g. Apel and Fertig,  or Brussig and Knuth, ), which use these items to control for the effect of the unemployed persons' willingness to make concessions on their chances for reemployment.
Another example are studies using vignettes, which can vary several characteristics of a hypothetical job offer to examine job characteristics, such as working hours, contract duration, or commuting distance (Abraham et al., ; Bähr and Abraham, ).
However, answers to hypothetical questions about intended behaviour are not the same as actual, observed behaviour.  This position finds support in experimental evidence comparing behavioural intentions to actual behaviour measured using observation techniques. The results of this line of research often indicate that while predictor variables might show comparable effects for intentions and observed behaviour (e.g. Petzold and Wolbring, ), the actual prevalence of the behaviour of interest often differs substantially from the intentions stated in the survey (Petzold and Wolbring, ; Groß and Börensen, ; Eifler, ).
In a comparable manner, for hypothetical questions on reservation wages, Krueger and Mueller (:  ff.) show that even though the reservation wage is predictive of accepting a job offer, approximately % of those offered a job that pays less than their reservation wage did nevertheless accept it. Likewise, -      % of those receiving a job offer paying a wage that is at least as high as their reservation wage declined the offer.
In this paper, we evaluate how concessions that unemployed UB II recipients in Germany make during job search affect their employment chances. On the individual level, we analyse whetherand to what extentunemployed individuals are prepared to accept new employment with unfavourable characteristics and whether such concessions are associated with a higher probability of finding new employment.
Doing so, a particular feature of our paper is that we do not have to rely on hypothetical questions about intended behaviour to analyse whether respondents are prepared to make a concession. Instead, we use a unique administrative dataset, the 'Sample of Integrated Welfare Benefit Biographies' (SIG). This dataset combines register data on UB II recipients from different sources held by the German Federal Employment Agency (FEA). Moreover, we enrich these data by adding administrative data on respondents' job search histories derived from the job placement process at employment agencies and Jobcentres. As a result, these data feature complete information on all jobs that unemployed recipients search for. Therefore, they allow us to identify whether and at what point in time an unemployed person starts looking for a job in an occupation that is, with respect to some of its characteristics, not as good as the one(s) the person was searching for earlier.
While the specific context of job search when receiving UB II benefits implies, that in principle Jobcentre caseworkers might request benefit recipients to look for jobs implying such concessions (i.e. the unemployed will make concessions not always voluntarily), the information on jobs searched should still be close to actual job search behaviour. This is so for two reasons. First, independent of whether the caseworker or the unemployed person herself initiated the unemployed's search for a particular job, the jobs covered in our data are the ones for which the unemployed persons receive job offers by the Jobcentre. Second, the unemployed benefit recipients have to apply for the positions offered. Those who do not may face sanctions. Thus, we think that overall the data we use should provide more reliable information on the extent to which unemployed benefit recipients are prepared to make a concession and also on the extent to which such a concession influences employment chances than the potential alternatives we discussed above.

Job-Related concessions and welfare: the German case
In Germany, there are two types of benefits for unemployed persons: the Unemployment Benefit (UB I) and the Unemployment Benefit II (UB II). The introduction of the latter in the s is an important example of the general shift towards activation strategies in social and labour market policies       discussed in the previous section. Despite the close similarity in name, both represent quite distinct types of welfare measures. UB I is an individual, contribution-based UI benefit, with payments proportional to one's former net earnings (%, or % for recipients with dependent children). The unemployed receive it only for a limited time and only if they meet the entitlement conditions with respect to previous employment duration. In , , unemployed persons received UB I (BAStat, ).
In contrast, UB II is a household-level, means-tested and tax-financed benefit covering recipients' minimum living standard. In principle, recipients might receive this benefit for an unlimited amount of time. In , ,, unemployed persons received UB II (BAStat, ).
What is most important in the context of our analyses is that when recipients are looking for employment, both types of benefit substantially differ with respect to availability requirements, especially relating to the concessions the unemployed persons have to make regarding their future job. UB I recipients might reject offers for jobs that are, with regard to their characteristics (e.g. wage, or qualification requirements) inferior to their previous position. In contrast, for unemployed UB II recipients more or less all job offers count as suitable work. Therefore, the unemployment administration expects them to accept offers for jobs thatin comparison to the characteristics of their former job might carry several disadvantages. This rule implicitly builds on the assumption that being less selective with regard to their future employment should improve the unemployed persons' chances of finding a new job.
In the remainder of the paper, we want to test this assumption. In doing so, we can exploit the fact that due to the regulations sketched above, job-related concessions should be a more or less common event among UB II recipients. Thus, in our analyses, we use a unique administrative dataset on UB II recipients that allows us to detect changes regarding the job search within spells of unemployment, to identify, whether these constitute a concession, and to analyse whether making job-related concessions affects employment probabilities.

Data and methods
The administrative dataset we use in our analyses is a combination of several data sources provided by the German FEA. The basis for our dataset is the 'Sample of Integrated Welfare Benefit Biographies' (Stichprobe Integrierte Grundsicherungsbiografien -SIG). The SIG is a ten percent random sample of all UB II recipients between  and . This dataset combines register data from different sources: the UB II Recipient History (LHG), which includes information on the receipt of means-tested benefits; the Jobseeker History (ASU/X-ASU), providing information on unemployment and job search; and the Integrated Employment Biographies (IEB; Antoni et al., ), which cover -      information on episodes of employment, as well as on participation in active labour market programmes.
The resulting integrated dataset includes detailed information on times in welfare benefit receipt, past and current dependent employment, unemployment, job search, participation in active labour market programmes (ALMPs), and individual and household characteristics. Moreover, as the standard dataset only includes aggregate information on job searches, we added data on detailed job search histories. These provide information on the occupations searched for, as well as on the periods during which the unemployed individuals search for these occupations.
We use a random five percent sample of the SIG, i.e. a . percent sample of all individuals who received UB II at least once between  and . To avoid left censoring, we analyse entry cohorts instead of point-in-time or ever-begun samples (Bane and Ellwood, ; Dahl and Lorentzen, a). We select individuals who received benefits for the first time between  and . For these individuals, we analyse unemployment episodes that start on or after their first entry into UB II, as well as on or after subsequent UB II episodes. We restrict our sample to the unemployment episodes during which the unemployed individuals are between  and  years of age and are not in (marginal) employment at the start of the episode.
The main process we focus on is unemployment, which we define more broadly than the official definition in Germany. In our definition, unemployment includes the times individuals are registered as unemployed and are searching for a job with the FEA. In contrast to the official definition, we do not consider short periods of sickness (up to a maximum of  weeks) as interruptions of unemployment. In a comparable fashion, we do not consider times in ALMPs that take place during times of unemployment (i.e. one is unemployed before and after the ALMP, or the ALMP takes place at the end of our observation period) as interruptions of or exits from the unemployment episode. Moreover, we do not consider short gaps of up to seven days between two spells of unemployment without a clearly defined status as an interruption of the unemployment spell.  The data are organised in spell format, defining spells by the exact starting and end dates. We performed episode splitting to divide unemployment episodes into subspells when there were changes in the variables of interest or in covariates. Our core variables of interest are the occupations in which a person is looking for a job during a particular period. We use information on these occupations to construct different measures for job-related concessions.
The first measure uses a feature of the German occupational classification, the 'Klassifikation der Berufe' (KldB)  (for details compare Paulus and Matthes, ). The KldB  is a five-digit code. While the first four digits provide increasingly fine-grained information on an occupation's content, the       fifth digit represents the so-called requirement level of the occupation, i.e. the level of qualifications that is usually required to perform this occupation. It distinguishes between occupations requiring no formal qualifications or on the job training only (level ); occupations requiring a vocational qualification (dual system or equivalent) (level ); occupations requiring a master craftsman's degree or a bachelor's degree (level ); and occupations requiring an advanced tertiary degree (e.g. master's degree or diploma) (level ).
As welfare recipients might be searching for a job in more than one occupation at a time (and these occupations might have different requirement levels), defining a concession is not a straightforward issue. We decided to apply a rather conservative definition. Thus, we consider a change in occupation(s) searched for to be a concession only if two conditions apply. First, in the current job search spell, a person is looking for an occupation he or she has not been looking for employment in before. Second, this occupation has a lower requirement level than the least requiring occupation he or she was looking for in the preceding spell.  Our second operationalisation of job-related concessions relies on the International Socio-Economic Index (ISEI), a commonly used measure for the socioeconomic status of occupations (Ganzeboom et al., , Ganzeboom and Treiman,  on the ISEI; for a general overview of occupation-based stratification measures, see Christoph et al., ). We used the current version for ISCO (Ganzeboom and Treiman, ), which can be merged to our data after recoding the KldB  codes to ISCO using the official crosswalks provided by the German FEA (BAStat, ). As there is no obvious way to define an ISEIbased threshold, we applied an ad hoc definition. We consider it a concession if a welfare recipient is looking for an occupation he or she has not been looking for before and if this occupation has an ISEI score at least  points below that of the occupation with the lowest ISEI score the individual was searching for in the preceding spell.  Our third operationalisation focuses on concessions regarding income. Since we do not have information on actual job offers and the potential wages these offers involve, we make use of information on average wages in occupations from an aggregate dataset, the Occupational Panel (OccPan; cf. Dengler et al., ). We use the average gross wages in KldB  occupations in the year , which Dengler and her colleagues calculated using all full-time workers in dependent employment (except for civil servants). This allows us to grasp the potential wage loss that might go along with looking for a job in a different occupation. We consider it a concession if a welfare recipient is looking for employment in an occupation he or she has not been looking for employment in before and if this occupation has a lower average gross monthly wage than the occupation with the lowest average monthly wage in which he or she was looking for a job in the preceding spell. We apply two different thresholds.

-     
The first is ,euro, which corresponds to half a standard deviation of the distribution of average occupational wages in our sample. The second is ,euro, which corresponds to one standard deviation.
Moreover, we look at changes in the occupation(s) searched for that do not necessarily imply a concession. We consider such a change to occur if, first, the number of occupations searched for differs between two subsequent spells. In addition, a change might occur if the number of occupations searched for remains constant, but one or more of the occupations searched for are replaced by an alternative one with a different five-digit KldB  code.
In our analyses, we focus on two questions. First, are welfare recipients prepared to make job-related concessions (in one of the senses defined above), and can we identify differences in the preparedness to do so between recipient groups? Second, will making a concession increase welfare recipients' chances of ending their unemployment spell by finding new employment? To answer these questions, we apply event history analysis for continuous-time data (Cleves et al., ). The event we analyse is whether benefit recipients leave unemployment by finding new employment that persists for at least  days.
In our descriptive analyses, we use nonparametric Kaplan-Meier estimators, which describe the probability of still being in the original state: that is, searching for a job in the same occupation, not having made a concession or still being unemployed, as a function of time. For our multivariate analyses, we use piecewise constant exponential transition rate models to estimate the hazard rate, i.e. the propensity of an event at a given point in time, for individuals who have not yet experienced an event. The main reason for choosing the piecewise constant exponential model was that it allows for a very flexible modelling of the baseline hazard rate. A piecewise exponential model defines certain intervals of the duration in unemployment in which the hazard rate is assumed to be constant but is allowed to vary between the intervals.  As previous research has emphasised the importance of unobserved heterogeneity or frailty in the analysis of welfare or unemployment transitions (Bäckman and Bergmark, ; Dahl and Lorentzen, b), we account for time-constant, unobserved individual characteristics by adding a gamma-distributed frailty term.  As covariates, we include sociodemographic characteristics (gender, age, qualification, citizenship, type of household, and region of residence), year of entry into unemployment, as well as an indicator that controls for concessions that benefit recipients might have made prior to our period of observation. This indicator relies on a comparison between the first occupation searched for in an unemployment episode and the previous occupation of the unemployed person when employed before the current unemployment episode. In addition, there is a second indicator variable for individuals who held no prior employment.

     
However, even though we include important covariates for the processes under study and account for individual unobserved heterogeneity, we cannot rule out that there is endogeneity. Moreover, we do not work with experimental or quasi-experimental data and there isobviouslyno random assignment of concession to unemployed persons. Thus, it is possible that there are unobserved variables that drive the relationship between making concessions and job-finding chances. Therefore, we cannot interpret the effects identified in our models to be causal effects.

Results
As a first step, we present descriptive results of the three core processes of interest. The first process is looking for a job in an occupation other than the one the recipient has been searching for earlier (independent of whether this implies a concession). The second process is to make a concession during this job searchregarding either qualification requirements, socioeconomic status, or average income. The third process is ending the unemployment spell by finding new employment.
We show survivor functions describing the job search and concession processes in Figures a and b and for finding employment in Figure . The survivor functions rely on Kaplan-Meier estimates and describe the probability of still being in the original state as a function of time. As can be seen from the light grey line in Figure a, it is not uncommon to look for a job in another occupation than the one in which one has been originally looking for employment. Almost % of the unemployed welfare recipients look for an alternative occupation at least once during the first five years of unemployment. However, only % make a concession regarding qualification requirements during our fiveyear observation period (black line), and even fewer (%) make a concession regarding socioeconomic status (dark grey line). Figure b shows that approximately % of unemployed welfare recipients make smaller wage concessions of at least ,euro (black line), and only % are prepared to make larger wage concessions of ,euro or more (grey line).
All in all, considering that welfare recipients are required to make concessions, the probability of them actually occurring is lower than we would have expected. What these numbers do not take into account, however, is that the welfare recipients in the sample are often already looking for lower tier jobs, meaning that they do not have much potential for making concessions in the first place. Thus, if one is already looking for a job that requires no formal qualifications, is located at the lower end of the status hierarchy and pays a low wage, it is simply not possible to make a concession.
To take welfare recipients' ability to make a concession into account, for every definition of concession, we restricted the sample to those that are, in -      theory, able to make a concession of this type.  The dotted lines in Figures a and b represent the survivor rates for these groups. The figures shows that for concessions regarding educational requirements and status, being able to make a concession seems to be rather important: among those able to make a concession, % (instead of %) have made a concession regarding educational requirements, and % (instead of %) have made a status concession after  years of unemployment. In contrast, making a concession regarding average wage seems not to depend on being able to make such a concession. For the ,euro       concession, the percentages among all unemployed and those theoretically able to make such a concession are virtually identical.  For the ,euro concession, the difference is only % (% instead of %).
Despite concessions being an unexpectedly uncommon event, more than two-thirds (%) of the unemployed UB II recipients in our sample find a new job and leave unemployment within the five-year period (black line in Figure ). Moreover Figure  shows that those theoretically able to make a concessionwith the exception of the ,euro expected wage concessionare a positively selected group, as their chances to find employment within five years are even higher (% to %).
In the second step of our analyses, we ask which groups of welfare recipients are more likely to make a concession or change the job they are looking for than others. As we discussed in the previous section, we will compare the four different operationalisations of concessions: one based on qualification requirements as captured by the KldB codes, one based on the ISEI and two based on the expected wage. While the results for all persons in our sample are displayed in models  (qualification requirements),  (ISEI),  and  (expected wage) of Table , models , ,  and  are the corresponding models that include only the recipients who were able to make a concession.
The first control variable ('better characteristic/concession') indicates if the occupation searched for at the beginning of the spell already constitutes a concession (when compared with the occupation in the last employment episode and applying the definition of concession used in the respective model). In addition, we added a control variable for those who had no earlier employment episode.      As we would have expected, having already made a concession earlier has a strong negative effect on welfare recipients' preparedness to make another concession during the spell currently under observation. This effect is particularly strong for concessions regarding education requirementsmost likely because there is not much leeway for making repeated concessions of this type, given the four-category requirement scale. With respect to control variables, Table  shows that there are few differences with regard to family type. In contrast, there is a clear pattern regarding education. For concessions regarding socio-economic status and expected wage, the models focusing on all unemployed persons (M, M, M, and M) show that university-educated recipients are more likely to make concessions than the comparison group of those with a vocational degree. Those without a degree, on the other hand, are less likely to make a concession than the comparison group. In fact, we would have expected to find this type of pattern because it reflects the higher potential of the highly educated to actually make a concession. While those with a university degree might still find a qualified job even when making a concession, those without vocational or tertiary education often do not have much potential to make concessions in the first place. The reason is that they might already be searching for a job in an occupation that requires no qualifications, carries a low status, or pays a low wage.
East Germans do not differ from their West German compatriots in any of the models, with the exception of the model using requirement levels. In contrast, women and those without German citizenship are, on average, less likely to make a concession. However, we would argue that interpreting the latter result to indicate that foreign nationals are less prepared to make concessions in general might be misleading. Much rather, this result appears to reflect the restricted potential of this group to make a concession.
The results in models , , and , which only include individuals who can make a concession, support this interpretation. In these models, we find no significant negative effect for women or for foreign nationals.  In the same vein, it appears that at least for required qualification and status (M and M), the less-educated have no lower and in some cases even a higher probability of making a concession than the other groups, once we take their lower ability to do so into account.
The dependent variable in model  also includes changes in the occupation searched for, which are not necessarily a concession (meaning that the occupation searched for might have a comparable or higher requirement level, socioeconomic status, or wage level than the occupation the benefit recipient was looking for earlier). While, as a consequence, we observe a substantially higher number of events in model  than in the earlier models, the effects of the control variables are more or less comparable. The only exceptions are a positive effect for younger persons and East Germans, and statistically insignificant effects for education. In the third step, we analyse whether making a concession has a positive effect on employment chances. Table  displays the results of these analyses. In the models, we operationalised the concession variables as time-varying covariates. They are coded '' before and '' after an unemployed person makes a concession during an unemployment episode. The results show that the effects on employment chances vary by the type of concession. Contrary to our expectation, making a concession with regard to educational requirements or socioeconomic status has no positive effect on employment chances. This finding is the same, independent of whether we restrict the respective model to individuals who are able to make a concession (M/M) or not (M/M).
In contrast, making concessions regarding the average wage of the occupation searched for has a significant and positive effect in three out of four models.  Thus, looking for a job in a lower paying occupation seems indeed to increase employment chances. This is in line with the results from studies reporting positive employment effects of lower reservation wages.
However, model  shows that searching for a job in a different occupation (that does not necessarily imply a concession) also has a substantial and positive effect on employment chances. Thus, it appears that being generally flexible regarding one's future occupation mightwhile being less demanding for the unemployed individualin fact be just as important for employment chances as wage concessions and even more important for finding employment than accepting less qualified or lower status employment.
Further control variables show more or less homogeneous effects across all models and are in line with previous studies on unemployment duration (e.g. Hohmeyer and Lietzmann, ): while employment chances for couples are comparably good, single parents with three or more children have comparably low employment chances. While employment chances for foreign nationals and women are lower than for their German and male counterparts, respectively, the employment chances of the two younger age groups are better than are those of older welfare recipients. Moreover, as we would have expected, high education has a positive effect and low education has a negative effect on employment chances. Moreover, while having made a concession prior to our observation spell has in most cases no effect on employment chances, those without previous employment have considerably lower chances of finding employment. Finally, the chances to find employment are higher for the more recent entry cohorts into unemployment, which should reflect the more positive economic and labour market conditions after .

Conclusion
In this paper, we analysed the extent to which unemployed individuals make job-related concessions and whether making such job-related concessions would -      improve their chances of finding new employment. To answer these questions, we used a unique dataset that combines information from different administrative data sources on recipients of the German UB IIa means-tested and taxfinanced benefit covering recipients' minimum living standard. A specific feature of this dataset is that it includes very detailed information on job search histories, which allowed us to apply different occupation-based measures for job-related concessions.
In general, our results indicate that requiring unemployed individuals to make concessions regarding their future employment has only a very limited potential to actually end their unemployment.
The first reason for this is that, although UB II recipients are required to make job-related concessions, it was unexpectedly rare for them to make such concessions. Even five years after starting their job search, only % to % had made a concession. Part of the explanation for this result is that a substantial proportion of % to % of the unemployed individuals in our data were already searching for low-qualified, low status, or low-paid employment; thus, they simply had no opportunity to make a concession in the first place.
When analysing the determinants of making concessions, women, foreign nationals, and less-educated individuals had a lower probability of making concessions, while those with high education were more often prepared to make a concession. We argued that these results do not necessarily reflect substantial differences but rather indicate these groups' unequal potential to make concessions.
The second reason is that, for most types of concessions, we can identify no positive effect of making such a concession on jobseekers' employment chances. This result holds for concessions regarding qualification requirements, as well as for concessions regarding socioeconomic status (ISEI). In contrast, concessions regarding the average wage in an occupation had a significant and positive effect. However, we could also show that searching for an alternative job in a different occupation (which does not necessarily imply a concession) has a substantial and positive effect on the chances of finding new employment, which is comparable in size. Thus, it appears that being generally flexible regarding one's future occupation might be as importantor even more importantfor employment chances than accepting lower quality employment. If, however, concessions are required from welfare recipients, asking them to search for employment in (mildly) lower paid occupations might be more successful than requiring them to search for less-qualified or lower status employment.
When considering these results in a wider policy context, there are two particularly important implications. First, when applied in an isolated fashion, demanding elements of labour market policies, such as job-related concessions, often fail to reach their intended objectives. Instead, a much more promising strategy would be to combine these elements with enabling elements, such as       investments in the human capital of the unemployed. Such a combination is, however, rather uncommon (Marchal and van Mechelen, ; Oschmiansky and Ebach, ). Second, for employment in particular, our results show that enabling the unemployed to find new occupational perspectivesideally in combination with training and qualification measures for the new occupation is a strategy that is at least as promising as requiring the unemployed to make job-related concessions.

Supplementary material
To view supplementary material for this article, please visit https://doi.org/. /SX Notes  While sociological authors rather emphasize unemployment insurance's potential to reduce unemployment-related losses in job quality, they agree that it also leads to somewhat longer unemployment spells (e.g. Bradley and Stephens, ; Gangl, )  Petzold and Wolbring () list three reasons why intended behaviour reported in a survey might deviate from actual behaviour performed in a real life situation. First, an intention might not result in a corresponding behaviour as obstacles might make the behaviour costly and reduce behavioural control. Second, the information considered by respondents when answering the (hypothetical) survey question is not the same information they have available (and process) in a real life situation.
Third, in a survey situation there might be social desirability bias, i.e. respondents do not report the behaviour they actually intend to perform but the behaviour they think the interviewer and/or society will approve of.  As a stability check, we applied a second definition, 'job search'. We defined 'job search' as being registered with the FEA as searching for a job while not necessarily being unemployed. The difference from our first definition is mainly that we do not consider the take-up of employment to be an exit from the 'job search' if the individuals are still registered as searching for a (possibly better) job. Thus, in a way, we do not define finding any job but finding a satisfactory job to constitute the relevant event we examine. As a consequence, duration in 'job search' is longer than in 'unemployment' and we observe fewer episodes for 'job search' than for 'unemployment'. We report the results in the appendix. While the results are largely comparable, there are some characteristic differences. Most notably, when applying the job search definition, there are no significant effects of concessions on finding employment for all types of concessions.  Quite obviously, this definition implies that respondents searching for jobs in occupations at the lowest requirement level by definition cannot make any concessions. However, this is a rather general problem. Independent of which measure we use to operationalise concessions, those at the lower end of this measure's distribution willby definitionbe unable to make a concession. Nevertheless, the scale of the problem is larger when using a rather crude measure as requirement level. Therefore, we also calculated models that restrict the sample to those that are actually able to make a concession.
-       This is a considerable difference given that the scale values of ISEI for ISCO are between  (Subsistence Farmers, Fishers, Hunters and Gatherers) and just below  (Dentists=.; Judges=.). As robustness checks, we applied thresholds of  and  points difference in the ISEI scores. The results do not differ substantially from those for the -point threshold and are available from the authors on request.  The intervals we used cover time spans of three months during the first two years of unemployment and of six months thereafter. The broader intervals at longer unemployment durations ensure that there are still sufficient events observed during intervals.  As a stability check, we also specified additional models that take the potential interdependence of both processesunemployment duration and the time until a concession takes placeinto account. The models estimated for this purpose are multilevel multi-process (MLMP) models. These models include individual-level random effects to capture unobserved heterogeneity for each of the two processes and allow them to be correlated (Lillard, ; Bartus, ). The results of these models are largely comparable to the ones presented here and are reported in Table A in the appendix.  I.e. those who initially search for a job in an occupation that requires at least a vocational qualification; those who initially search for an occupation that has an ISEI of at least  (which is  points above the lowest ISEI score of ); those who initially search for an occupation that has an average income of  euro; or those who initially search for an occupation that has an average income of  euro, respectively.  As one can see from the case numbers in the multivariate models in Tables  and , there are few unemployed benefit recipients (n=) in our data that look for occupations that would pay as low a wage that making a ,euro wage concession would be technically impossible (i.e. the average wage of most people's occupations is more than ,euro above the lowest wage in the data, which is  euro).  The only significant effect we find is in model , which isfor the reasons we discussed in the previous footnotesimilar to the model that includes all unemployed benefit recipients.  The effects for wage concessions are the only ones, for which we found substantial differences between our main models and the multilevel multi-process models displayed in Table  A in the Appendix. While there is a significant effect of making small wage concessions of ,euro on finding employment in models  and  in Table , we find no significant effect of such small wage concessions on employment in models  and  in Table A.