Prejudices against the unemployed—empirical evidence from Germany

Prejudices against the unemployed pose an enormous threat to their self‐confidence and can make it difficult for them to re‐enter the labour market, resulting in further long‐term unemployment. Given these high costs for the unemployed and for society as a whole, our knowledge of prejudices against the unemployed is surprisingly scarce. We focus on the question of what determines the strength of prejudice among employees. By applying social identity theory, we assume that people who are disadvantaged in the labour market in general, also hold stronger prejudices. In addition, we assume that social status mediates this association and that self‐efficacy moderates it. We use data from the German panel study “Labour Market and Social Security” (PASS) and show that some groups of people who are themselves disadvantaged in the labour market (women and first‐generation immigrants) have more prejudices against the unemployed; however, people with poor mental health have even fewer prejudices. Low social status (in terms of educational background, income, and job status) is associated with strong prejudices; however, social status does work as mediator to a minor degree only. People with low self‐efficacy in general (main “effect”) and first‐gener‐ ation immigrants in particular (moderating “effect”) have stronger prejudices. These results can be a starting point for developing measures to reduce prejudice and for the onset of a debate about the origins of prejudices against the unemployed.


Introduction
Differences in labour market outcomes according to social groupings-such as women and men, migration background or lack thereof etc.-are of substantial interest in the social sciences. In addition to social groupings, researchers also use social mechanisms to try and explain these differences. These mechanisms can involve (e.g.) meritocratic explanations, such as differences in human capital and soft skills, or the effects of context characteristics, e.g. company size and branch productivity. When differences remain that cannot be explained, researchers often come to the conclusion that these differences are due to discrimination or prejudices by employers against some groups. Additionally, qualitative research among the unemployed hints that these groups do experience being subject to prejudices: "…it's simple: There are people who just see you as an unemployed bum, I´d say. Roughly speaking, well, someone who just wants to spend his life on social welfare. " (German interview translated by Nicolette Steinbach).
investigated directly: Are these prejudices widespread? And who is expressing them? This leads us to the following research questions: Who are the people that actually hold these prejudices? And what factors foster the development of or a predisposition towards prejudice? Who draws and strengthens the boundaries, and who constructs such (spoiled) identities? As identities are built in an interactive process between generalised others (people who are more or less prejudiced) and the target person (a person who is more or less stigmatised), there are two sides of the coin to be considered. This study focuses on the prejudices held by employed people, who may contribute to generating the stigma associated with being unemployed.
Stigma and the perception of stigma have negative effects on people's self-confidence, their achievements in education, their health (Major and O'Brien 2005), their housing situation, their interaction with the legal system, and even their family ties. If unemployed people perceive themselves to be stigmatised (as illustrated in the quotation above), they face these negative consequences, which are responses to the threat of stigma-based exclusion (Miller and Kaiser 2001;Major and Eccleston 2005).
Both sides-the prejudiced and the stigmatised-have negative expectations of the interactions and especially the latter are afraid of being rejected, embarrassed, or ridiculed. These feelings result in the use of avoidance behaviour in interactions, which "mainly means foregone opportunities" (Crocker et al. 1998: 543), and which in turn also means that members of the subordinate group are excluded in different functional relationships. This could contribute to an increasing social divide between status groups, or to social closure. Furthermore, individuals who are affected by prejudices are more likely also to be prejudiced against other groups, resulting in a vicious circle.
The different strategies (e.g. withdrawal, acceptance, denial, and avoidance) for dealing with an inferior status such as unemployment and the negative assumptions associated with it have mostly adverse, momentous effects. One effect could be a less successful reintegration of the unemployed into the labour market. As our own recent (Gurr and Jungbauer-Gans 2017) research shows, the unemployed assume that the employed are not willing to accept them as equals. The unemployed learn how others perceive them through exposure to pejorative media coverage, insulting statements from politicians, and even everyday situations linked with incorporated social norms. This experience results in the unemployed thinking of themselves as not meeting society's expectations. The following interview excerpt from a long-term unemployed individual illustrates this situation: "I mean there are thousands of prejudices, the unemployed are lazy and spend their days in front of the TV? They have too much money anyway? And they all have too many children? And you just have to put in a little effort, then you´ll find the right job. " (German interview translated by Nicolette Steinbach).
In this paper, we analyse whether groups with higher social status-who are more often involved in decisions relating to employee selection-or those who are disadvantaged in the labour market themselves, have greater prejudices against the unemployed. The latter may look for other groups that are also a target of social disapproval in search for a positive social identity, as explained in Sect. 3. Therefore, we enquire into whether prejudices among disadvantaged groups diminish when their social status is controlled for. This finding should reveal whether only those who really have an inferior social position in the social structure have greater prejudices against the unemployed, or, whether those who have (several) ascriptive characteristics that may be the object of discrimination have greater prejudices against the unemployed in general. Second, we analyse whether those who think that everyone is in control of her or his own destiny regard the unemployed as more responsible for their social position. Therefore, we investigate the moderating effects of self-efficacy among members of disadvantaged groups. The results can be used to identify situations (such as job advertising and selection processes) or measures (information campaigns, adjustment of selection procedures or criteria, quotas, etc.) in order to reduce the prejudices themselves or to reduce the opportunities for an influence of prejudices on employment decisions.
The research question regarding which groups have more prejudices against the unemployed is answered by applying social identity theory. To test our hypotheses we analyse data from the German panel study "Labour Market and Social Security" (PASS). In the 7th wave, we have used a newly developed scale to collect data on prejudices and stigmatisation (Gurr and Jungbauer-Gans 2013).

State of research
Modern industrial countries are characterised by their citizens' high level of activity in the labour market to earn a living. Therefore, the active and employed citizen represents an ideal of the modern welfare state (Eichhorst et al. 2008). Throughout Europe, several labour market measures have been implemented to stimulate employability, competitive orientation, and empowerment while also pursuing the strategy of promoting and demanding (e.g. Bröckling et al. 2004;Schönig 2006). These changes in welfare policies (see, for Australia Eardley 1999) can be described as a "global shift" (Sage 2012: 370;Lødemel and Trickey 2001). This global shift turns away from a focus on providing the unemployed with benefit income and towards a focus on instruments and policies aimed at labour market integration and employability. Several studies (Larsen 2008 for a cross-national perspective; Dorey 2010 for the poor in the UK; Sage 2012) have provided evidence that these principles of reciprocal responsibility are associated with a shift in the perception of unemployment. Therefore, these policies reveal a potential impact for everyone-i.e. for the employed as well as the unemployed-and have various implications. Oorschot (2000) and others (Oorschot and Arts 2005) found that "the group who was given the least support was people on social assistance" (Larsen 2008: 149). The public was most in favour of support for old people, followed by support for the sick and disabled, then by needy families with children, and lastly, by the unemployed. It might be reasonable to expect an increase in negative attitudes towards the unemployed in reaction to changes in labour market policies during the past one or two decades.
Recently, Groß (2016) published a noteworthy theoretical and methodological study on amplifiers of prejudices towards unemployed individuals. She asks how the general orientation of the enterprising self translates into individual attitudes in the climate of the recent labour market policy, leading to exclusive and repressive tendencies in society and thus to a strong association of these attitudes, with prejudices against economically inefficient groups. She identifies an increasingly strong mediating effect of neoliberal guiding principles on the devaluation of the unemployed in the eyes of those with a higher status. This kind of "social role model" (Groß 2016: 166) poses a legitimising myth and mediates between social dominance orientation and prejudices against unemployed people. However, the question remains open as to who is holding negative preconceived judgments and what other factors foster the predisposition to and development of prejudice.
The few contributions on the question of who holds negative stereotypes are ambiguous at best. As mentioned by Furnham (1982Furnham ( , 1983; a cross national analysis for New Zealand and the UK: Furnham and Hesketh 1989;for Barbados Furnham 1991;McFadyen 1998), conservatives and less well-educated people tend to support more pejorative individualistic explanations, and both groups view unemployment more often as individual failure. Furthermore, there is evidence from the U.S. that indicates a pervasive victim-blaming view of the poor and the unemployed (Kluegel and Smith 1981: 31;Kluegel 1987). More precisely, they referred to the result appearing to be rather paradoxical at first glance, in that the most negative attitudes towards the unemployed and the poor can be found among the least privileged members of their sample (Furnham 1982;Golding and Middelton 1982). Contrary to these results, Krahn et al. (1987) could not identify any effects of occupation or income on attitudes and only minor effects of education on attitudes. Krahn et al. (1987) and Oorschot (2006) explained that the disapproving attitude towards the allocation of support to the unemployed is somewhat stronger among women, older people, and people with less education but is not affected by work ethic. Mansel and Endrikat (2007) examine the tendency of prejudice towards the long-term unemployed on the basis of a survey on group-focused enmity. They convincingly demonstrated that resentment towards the long-term unemployed increases continuously as an individual's "social position" declines. More precisely, Mansel and Endrikat (2007: 179) reveal in their model that a person's socio-emotional disintegration, individual upwards orientation, experience with powerlessness, and economic orientations lead to a devaluation of the long-term unemployed. However, exceptions include individuals who perceive themselves to be precariously employed; they less often report devaluation of long-term unemployed.
This review of research, into prejudice against the unemployed, shows that few studies to date investigate this question. There are some hints that groups who are disadvantaged in the labour market are more prejudiced against the unemployed, but most of the significant work in this area in the recent past has focused on attitudes towards a specific range of dimensions of the welfare state (Mackonytė et al. 2014; for an overview, see Sundberg and Taylor-Gooby 2013). These attitudes could be a driving force for negative stereotypes against the unemployed or for the differing assumptions on the neediness of specific groups (Oorschot 2006;Groß 2016). However, given the thesis that the social political regime and labour market policy matters, we pose the question of whether the results that have been summarized from different countries in this chapter can be generalized, or, whether additional empirical evidence for Germany is needed.
In the following sections, we do not investigate attitudes about specific welfare policies, social interventions, or redistribution. Rather, we examine specific prejudices against unemployed people and the origins of these prejudices. This is the first study that systematically investigates which groups hold prejudices against the unemployed under the conditions of the German labour market and social regime.

Theory
Theories suitable for our research question are group conflict theory (Sherif and Hovland 1961;Sherif 1966), the theory of social comparison (Festinger 1954), and the approach of prejudice as self-image maintenance (Fein and Spencer 1997). These theories were adopted, integrated, and/or refined by Tajfel and Turner (Turner 1975;Tajfel and Turner 1979) in their social identity theory. We therefore apply social identity theory (Tajfel 1974;Turner 1975;Tajfel and Turner 1979) to explain prejudices against the unemployed. 1 According to Brown (2003), existing empirical evidence on this application of social identity theory is inconclusive. In applying social identity theory, we argue as follows: (a) "Individuals strive to maintain or enhance their self-esteem: they strive for a positive self-concept" (Tajfel and Turner 1979: 40;Festinger 1954;Fein and Spencer 1997). Prejudices can be a method used to achieve this goal (Fein and Spencer 1997) by looking down on others. Therefore, social identity theory is useful to explain the mechanism behind our research question.
(b) "Social groups or categories and the membership of them are associated with positive or negative value connotations. Hence, social identity may be positive or negative according to the evaluations […] of those groups that contribute to an individual's social identity" (Tajfel and Turner 1979: 40). Therefore, people with a negative social identity would be expected to have stronger prejudices. Regarding this assumption, Tajfel and Turner (1979) also state that many definitions of the "social group" are too restrictive in this context. They define a group "as a collection of individuals who perceive themselves to be members of the same social category, share some emotional involvement in this common definition of themselves, and achieve some degree of social consensus about the evaluation of their group and of their membership of it" (Tajfel and Turner 1979: 40). Using this definition, being employed is a sufficient category in the sense of social identity theory. According to Jahoda (1995), people draw a clear distinction between employed and unemployed individuals. Moreover, in Germany, being employed is an important part of most people's self-concept (Jahoda 1995).
Within the group of employed people we can identify several subgroups that are disadvantaged in the labour market. A lot of labour market studies investigate whether gender, ethnic, age or weight discrimination can be observed in the labour market. With respect to wages, re-employment chances, job search duration, or leadership positions; this is sometimes done by using field experiments (e.g. for gender differences see: Behr and Theune 2018, for ethnic discrimination: Brenzel and Reichelt 2015;Nanos and Schluter 2014;Kaas and Manger 2012;Braakmann 2009, for age discrimination: Heywood et al. 2010, for weight discrimination: Katsaiti and Shamsuddin 2016;Bozoyan and Wolbring 2018). However, these studies look at concrete dimensions of labour market integration, and do not measure prejudices directly against the respective target group.
These disadvantages are one form (or at least one result) of a negative evaluation as mentioned above. We assume that (employed) women meet all three criteria defined by Tajfel and Turner (1979). Equally, we assume that people with migration backgrounds are a group of individuals, who perceive themselves to be in the same social category. Although this category can be further differentiated-in general-the same arguments should apply to all the subgroups equally. Moreover, they should share some emotional involvement and have some degree of consensus about the general evaluation of this group. The same applies to the group of people with special needs, mental health problems, poor self-rated health, and overweight or obesity-at least in the context of labour market disadvantages. Furthermore, all of these characteristics, specifying subgroups within the group of employed people, should be part of and relevant for the self-concepts of their members. 2 (c) Tajfel and Turner (1979: 40) state that the "evaluation of one's own group is determined with reference to specific other groups through social comparisons in terms of value-laden attributes and characteristics. Positively discrepant comparisons between in-and out-group produce high prestige". Accordingly, we interpret prejudices as a social comparison in terms of value-laden attributes. This idea is also supported by Fein and Spencer's (1997) selfimage maintenance function of prejudices. Furthermore, these comparisons may occur between the in-group and multiple out-groups. The unemployed are therefore just one of several possible comparison groups, but it is a group for whom people can reach a positively discrepant comparison rather easily.
In applying social identity theory, we assume that groups of employed individuals who are disadvantaged in the labour market perceive their status to be illegitimate compared with that of individuals who are not disadvantaged. Due to the difficult situation of seeking positive identity, these individuals strive for positive discrepant comparisons and thus tend to devalue other groups such as the unemployed, as a group being easy to devalue. Therefore, we assume that disadvantaged groups have stronger prejudices against the unemployed than do more privileged groups (H1).
Additionally, we expect to find that social status is a mediator between disadvantaged groups and the degree of prejudice. We argue that it could, in fact, be a lower (subjective) social status that results in stronger prejudices against the unemployed. As disadvantages in the labour market tend to result in a lower social status and as social status is negatively correlated with prejudices, this social mechanism might partly explain the positive correlation between disadvantaged groups and the holding of strong prejudices. Therefore, the effect coefficients of the disadvantaged groups on prejudices should decrease when the individual social status is controlled for (H2). If this hypothesis holds and we could explain higher prejudices of disadvantaged groups by including the individual social status completely, this would contradict Tajfel and Turner's (1979) social identity theory in parts. We would then conclude that prejudices are presumably not a result of a negative social identity but a negative individual identity.
Finally, we assume that group members' belief systems have a moderating effect. Tajfel and Turner (1979) define these belief systems as a continuum from individual mobility to social change that describes how people believe how society works. Individual mobility means that people are convinced that everyone is able to change her or his position in society on her or his own. Social change means that people are convinced that one's own position in society can only be changed, if the position of one's group in society is also changed. According to Tajfel and Turner (1979) people with a belief system rather on the individual mobility side of the continuum tend not to think in terms of groups. They tend towards individualistic explanations (e.g. for success or failure) and therefore would not be expected to use out-group devaluation in order to gain a positive self-concept. Applying this viewpoint to our research question, we expect to find weaker or no prejudices among disadvantaged groups for those who perceive the individual position in social hierarchy to be individually changeable (H3; for a summary of our hypotheses and the mechanisms assumed, see Fig. 1). These individuals tend to view everybody as being individually responsible for their situation and do not think in terms of groups.

Data and methods
This study uses the factually anonymous data of the panel study "Labour Market and Social Security" (PASS). Data access was provided via a scientific use file supplied by the Research Data Centre (FDZ) of the German Federal Employment Agency (BA) at the Institute for Employment Research (project number 101267). This survey is designed for research on the labour market and poverty in Germany (Trappmann et al. 2010(Trappmann et al. , 2013 and is conducted annually by the Institute for Employment Research (IAB). PASS data provide two subsamples with about 6000 households each per wave. The first subsample includes households which receive Unemployment Benefit II, whereas the households and individuals of the second subsample are a stratified random sample of the German resident population. In the 7th wave of this panel study (conducted in 2013), a newly developed scale is included to measure prejudices against the unemployed. The 7th wave includes the prejudice scale (Gurr and Jungbauer-Gans 2013) answered by employed people only. We use this prejudice scale as our dependent variable.

Measures of prejudices towards the unemployed
Gurr and Jungbauer-Gans (2013) developed the prejudice scale based on qualitative research. This scale follows the percentage approach developed by Brigham (1971), wellestablished in social-psychological research to discover the stereotypical beliefs of the employed towards the unemployed.
The scale includes the following five items 3 : How many of 100 unemployed people… … are really seeking a job? These traits included laziness, complacency, duplicity, and a lack of willingness to work. The mean of the different items is the benchmark of their stereotyping value. Low values of the mean indicate that the respondents on average think that fewer unemployed individuals behave in a disapproved-of manner, whereas high values point to strong prejudices towards the unemployed.
These items differ significantly in the percentage values of the answers, as displayed in Fig. 2. The first item is the only one that is positively worded, i.e. in which a high value corresponds to a low level of prejudice.
We calculate the prejudice index as the average value of the five items (with the first item used in the reverse manner: 100-VAR). The prejudice index has a Cronbach's alpha of 0.71, an empirical range from 2-100, and a mean of 40 (all within the sub-sample of employed people; see Table 1) and is almost normally distributed (see Fig. 3).
We also performed a factor analysis including scree plot, both suggesting a one-factor solution. The percentage approach allows the interpretation: A value of 40 equals the opinion that on average 40% of unemployed people behave and think in a disapproved-of manner.

Cases and variables
We use the 4977 people from the 7th wave of the PASS data that have non-missing values for all five items of the prejudice scale and are employed with a monthly gross income over €400. We multiply imputed all missing values within the independent variables using the mi impute routine implemented in Stata (multiple imputation with chained equations) with k = 30. We use the dependent variable for the imputation model, but exclude cases with missing values within the dependent variable from the analyses as recommended by von Hippel (2007). Table 2 describes the variables without any imputations. Since we describe disadvantaged groups, we coded female gender with 1 and use two dummy variables for migration background (1st and 2nd generation) with no migration background (including 3rd generation) as reference. Special needs is measured by asking, "Do you have a certificate of disability or have you applied for one?" with a simple yes/no answer option. Mental health problems were measured by asking "How much have you suffered

Fig. 3 Distribution of the prejudice index
from mental health problems during the past 4 weeks?" using a 5-point scale from "not at all" to "very much". All four categories from "little" to "very much" were set to "yes", which explains the high share of 60% having mental health problems (see Table 2). Self-rated health is measured as a dummy variable with "very poor" and "poor" health set to 1 and the remaining 3 categories from the original 5-point answer scale to 0 as we expected poor health rather than good health to have an effect. We measure overweight and obesity by using the body mass index (BMI). 4 As weight and height information are unfortunately not available for wave 7, we had to use information provided in wave 6. We think this is reasonable, as we just use this information for creating categories, the time between the interviews is only 1 year and it is the only information we have. For the sake of completeness, we distinguish 4 categories instead of just an indicator for overweight/obesity: BMI < 18.5 is categorized as underweight, 18.5 ≤ BMI < 25 as normal weight (as reference), 25 ≤ BMI < 30 as overweight, and BMI ≥ 30 as obesity (WHO 2020). We measure self-efficacy by using a 0-1-normalised unweighted additive index that includes the following 5 items rated on a 4-point scale (from 1 = fully applicable to 4 = not applicable at all): (a) I have a solution for every problem; (b) Even in unexpected situations, I think I will be ok; (c) I have no difficulty achieving my goals; (d) In unexpected situations, I always know how to behave; (e) I always succeed in solving difficult problems when I try. The index has a Cronbach's alpha value of 0.80, which represents high reliability.
We measure social status by using educational background, income, and occupational status. Educational background is measured by the highest educational attainment with low educational attainment (including the German "Hauptschulabschluss"/"Polytechnische Oberschule" 8th/9th grade and below/no graduation) as reference and two dummy variables indicating medium (including the German "Realschulabschluss"/"Mittlere Reife"/"Polytechnische Oberschule" 10th grade) and high educational attainment ("(Fach-)Abitur"/"Fachhochsch ulreife"/"Erweiterte Oberschule" 12th grade). Income is measured by quantiles of the individual monthly gross income. We use the International Socio-Economic Index of Occupational Status (ISEI) for occupational status, which ranges from 16-90.
Age in years has a minimum of 17 and a maximum of 64. Age is used in the model as control variable to show the job status association with prejudice net of age effects. The agreement on "work is most important in life" is measured on a 4-point-scale with the two agreement categories building the 1 category and controls for the importance of work that is also associated with job status. The "agreeability index" is a 0-1-normalised metric index generated by the unweighted addition of four items that are usually included in the so-called "Big 5" item battery. The share of missing values was especially high for the agreeability index since these data were merged from Wave 5 (from 2011), which included the personality traits. We assume personality traits to be quite constant over the life-course (or at least for shorter time spans such as the 2 years from 2011 to 2013), so a 2-year deviation regarding the time of data collection should prove no more than a minor issue. Marital status was coded with a 1 for being married (including registered partnership) and living together versus all other configurations, which were set to 0. Eight per cent of the respondents live together with (at least) one unemployed household member and 4% of the respondents had been unemployed in the past. The duration of unemployment in the past measured in months ranges from 0 (for the 96% who had never been unemployed) to a long-term unemployed person of 373 months (which amounts to 31 years of unemployment).

Analytical strategy and limitations
We estimate straightforward linear regression models on the prejudice index with robust standard errors accounting for clustering within households. Model 1 includes the variables measuring being part of a disadvantaged group. Model 2 additionally includes measures of social status in order to model the mediation effect. Model 3 finally includes self-efficacy and interaction terms of selfefficacy, and each variable of being part of a disadvantaged group, to model the moderation effect.
In doing so, we face endogeneity issues and respectively, reverse causality, that we cannot solve completely using the PASS data in a cross-sectional way. E.g. healthrelated measures such as special needs or poor self-rated health can also result from poor labour market integration or a low agreeability, both associated with strong prejudices against the unemployed. If this is true, some results should be interpreted with caution. We can partly fix these issues by using control variables within our cross-sectional design. As control variables, we use age, the importance of work in life, and marital status, since these variables are associated with occupational status. Thus, we are able to estimate the association of occupational status and prejudices net of age, marital status and the opinion towards the importance of work. We control for agreeability as a personality trait, to detect the associations with prejudices against the unemployed net of the general personality trait measuring openness towards others. We also control for several experiences with unemployment, since we expect them to be associated with both prejudices against the unemployed and social status in general. Table 3 shows the results. In line with Hypothesis 1 (see Fig. 1) derived from social identity theory, we find stronger prejudices in females (plus 4.8 percentage points), first-generation immigrants (plus 6.1 percentage points), compared to no migration background, people with special needs (plus 2.1 percentage points), and obese people (plus 3.1 percentage points) compared to those with standard weight in Model 1. Surprisingly underweight people also seem to have stronger prejudices (plus 4.3 percentage points, not significantly different from obese people). Contrary to our expectations, having mental health problems is associated with 2.1 percentage points less on the prejudice scale, and self-rated health is not significantly associated with the prejudice scale. With the two exceptions of mental health problems and self-rated health, being part of a disadvantaged group is associated with stronger prejudices.

Disadvantaged groups
Gender (  Hypothesis 2 assumes that social status at least partly explains this association (stronger prejudices held by members of the disadvantaged groups); people with a high status should not feel that their social identity is threatened by unemployed people, and only those with a low status should be able to foster a positive social identity by devaluing the unemployed (see Fig. 1, social status as mediator). In general, a high social status is associated with weaker prejudices against the unemployed and this raises the share of explained variance from 7.0% to 16.4% (comparing the R 2 adjusted of Models 1 and 2). However, the social status variables only slightly reduce the "effect" of the disadvantaged group variables except for obesity (comparing the coefficients of the disadvantaged group variables from Models 1 and 2). The coefficient for obesity is reduced by more than half. 5 Hypothesis 3 postulates that disadvantaged employees have weaker prejudices the higher their self-efficacy, since they do not think in terms of groups and therefore do not strive for a positive social identity but rather a positive individual identity (see Fig. 1, self-efficacy as moderator). We modelled this using interaction terms for each disadvantaged group with self-efficacy. However, the results in Model 3 are mixed: Self-efficacy itself is associated with weaker prejudices (minus 11.5 percentage points with additional interaction terms and minus 10.7 percentage points for Model 3 without interaction terms; not shown in Table 3). This is just the association we expected to find. However, most interaction terms with self-efficacy are not significant except for first-generation immigrants. First-generation immigrants show decreasing prejudice with increasing self-efficacy (see Fig. 4). Therefore we find only very limited support for hypothesis 3.
In summary, people from a disadvantaged group in the labour market-such as female employees, first-generation immigrants, and those with special needs -have stronger prejudices (in line with Hypothesis 1). Although social status is strongly associated with weak prejudices, it does not explain the strong prejudices of the disadvantaged groups (contrary to Hypothesis 2, there is no mediator effect). The stronger prejudices within the disadvantaged groups are not weaker for those with a high self-efficacy except for first-generation immigrants, as stated in Hypothesis 3 (self-efficacy as moderator). But at least Model 3 reveals the expected main effect. People with a high self-efficacy have weaker prejudices, which may be explained by their tendency not to think in terms of groups but of individual mobility. First-generation immigrants show decreasing prejudices against the unemployed, the higher their self-efficacy (Hypothesis 3 can in part be supported).

Conclusion
In modern societies, working is not only important for earning a living but also for social recognition. In this paper, we have investigated which groups hold the greatest prejudices towards the unemployed. We have applied social identity theory to explain differences in negative attitudes towards the unemployed. According to this theory, people who belong to groups that are disadvantaged on the labour market have, in general, difficulties finding a positive identity. Hostility is elicited when they strive for a positive self-identity in comparison with another group. These individuals find such a group in the unemployed, which led to our hypothesis that disadvantaged groups would be expected to have greater prejudices towards unemployed people than do more privileged groups. The second research question asked whether their social position might be the reason for disadvantaged groups in the labour market to hold stronger prejudices towards the unemployed. The third research question was whether only those among the disadvantaged groups who think that everyone is responsible for her or his own destiny are prejudiced against the unemployed.
By analysing data from the panel study "Labour Market and Social Security" (PASS), we found that on average, people think that 40% of the unemployed behave in a disapproved-of manner as measured by the prejudice index that was computed as the mean of the five items. In line with the first hypothesis derived from social identity theory, we found that women, people without German citizenship, people with special needs, those with poor self-rated health, obese, and underweight people hold greater prejudices. However, our results show that people with mental health problems have fewer prejudices than do people without such problems. It is possible that people with mental health problems are more sympathetic and sensitive (which has only partly been considered by means of the agreeability index, controlled for here). Whether they are more empathetic could be investigated subsequently if a suitable scale for measuring empathy becomes available with prejudice data.
The second hypothesis expected prejudices to stem from an inferior social position of disadvantaged groups. In the empirical model, the effects of disadvantaged groups should diminish after social status is controlled for. The coefficients for gender, migration background and people with special needs only slightly decreased in the second model, controlling for social position. The coefficient of obesity is reduced by more than half in this model. All indicators of social position (education, income quartile and occupation) significantly confirm that prejudices are less prevalent among persons with higher social status. We can conclude that social status does not fully explain that women, people with first-generation migration background, people with special needs, underweight and obese people have more prejudices against the unemployed. Hypothesis 2 is therefore confirmed only to a minor degree.
In general, we find a negative effect of self-efficacy on prejudices-being in line with the social identity theory. Nevertheless, the third hypothesis, which expected a moderating effect of self-efficacy, could only be confirmed for people with a first-generation migration background. People with a first-generation migration background and high self-efficacy have even fewer prejudices. The main effect of first-generation migration background increases in the third model. All other interactions of disadvantaged groups and self-efficacy proved not to be significant. Therefore, we conclude that except for people with a migration background there is no moderating effect of the belief that individual agents are responsible for their own social positions.
Future research should address the puzzling results for people with mental health problems. Do these individuals have fewer prejudices because they are more sympathetic and sensitive? The results also indicate that mentally ill people may come from higher social strata, which partly explains their fewer prejudices.
It should be noted that higher self-efficacy, higher levels of education, and a higher social position in the labour market are strongly associated with weaker prejudices towards the unemployed. Among the control variables, stronger prejudices were found together with the attitude that work is the most important thing in life. A high value of the agreeability was associated with fewer prejudices.
Furthermore, one of the most important shortcomings here is that we could not make use of the panel design of the data since the prejudice index has only been measured once thus far. Future research should therefore make use of panel data to investigate the interplay of attitudes, personality, social positions, and labour market conditions. Doing so would also provide further opportunity to address the limitations of this study regarding endogeneity and reverse causality as described in Sect. 4.3.
We can conclude that social identity theory is at least to some extent suitable for explaining most of the results. Most of the investigated groups disadvantaged in the labour market have a higher level of prejudice against unemployed people, which is in line with previous research showing stronger prejudices for less-privileged groups. Negative attitudes towards the unemployed do exist, and they increase if more or larger groups become disadvantaged in the labour market feeling that their social identity is at risk. Moreover only a very small part of these stronger prejudices can be explained by individual social status. And finally, we do not find the expected moderating effects of self-efficacy but we do find the expected main effect just in line with social identity theory. All in all our results rather support social identity theory in explaining prejudices (against the unemployed).
The results from our study do not suggest that the global shift in labour market policy from welfare to workfare (which places more responsibility on individuals) leads to greater prejudices; however, the data are not ideal for addressing this question since self-efficacy may be considered a relatively stable personality trait that is a source of resistance rather than an object of a general policy change. Measuring the degree to which responsibility is truly assumed by individuals would be a more suitable operationalisation of this element.
Actions for reducing prejudices against the unemployed should address the target groups we highlighted in this contribution-especially disadvantaged groups, people with a poor educational background, and those with low job status, all of whom are most likely to have strong prejudices. Otherwise, the tendency of weak groups to develop prejudices could legitimise fine-grained social hierarchies and intensify social division. On the other hand, the result that prejudices are less often found among people with higher social positions suggests that they are not so relevant in the new employee selection process. In order to reduce prejudice against the unemployed, social policy and labour market measures that reduce disadvantages of different social groups or at least the size of disadvantaged social groups seem to be most appropriate.