Gender and ethnicity in social assistance assessments of single applicants with substance abuse problems

ABSTRACT

In Sweden, trained social workers ultimately determine the right to social assistance (SA). This paper examines how a substance-abusing SA applicant’s gender and ethnicity affect assessments on eligibility. It further addresses whether professionals’ gender and migration background have bearing for such patterns. Social work professionals (n = 910) spread over 43 social service offices in 19 Swedish municipalities conducted assessments based on a vignette. Four versions of the vignette were used, in which the name of the applicant was varied. The names were two Swedish-sounding names and two Arabic-sounding names: one of each gender respectively. Data were analysed by means of cross tabulations and multilevel logistic regression analysis. The Swedish-sounding names were linked to a higher chance of granting compared to the Arabic-sounding names, whereas females more often were seen as eligible than males. The major finding was the combined impact of gender and ethnicity: Applicants attributed with an Arabic-sounding male name were less often seen as eligible than those with a Swedish-sounding female name. This bias was specifically manifest among female professionals. Further, native-born professionals generally viewed the applicant as eligible to a higher extent than foreign-born.

SAMMANFATTNING

I Sverige avgörs rätten till ekonomiskt bistånd genom individuella bedömningar som utförs av socialarbetare. I den här studien undersöks hur den sökandes kön och etnicitet påverkar bedömningar av en ensamstående klient med missbruksproblem. Vidare analyseras huruvida de professionellas kön och migrationsbakgrund har betydelse för sådana mönster. Socialarbetare (n = 910) från 43 socialtjänstkontor i 19 svenska kommuner fick göra bedömningar utifrån en vinjett. Fyra versioner av vinjetten användes, i vilka namnet på den sökande varierades mellan två svenskklingande namn (ett manligt och ett kvinnligt) och två arabiskt klingande namn (ett manligt och ett kvinnligt). Data analyserades med hjälp av korstabeller och logistisk flernivåregressionsanalys. De svenskklingande namnen kunde kopplas till en högre chans att beviljas ekonomiskt bistånd jämfört med de arabiskt klingande namnen, medan kvinnor oftare sågs som berättigade till stöd än män. Det viktigaste resultatet var emellertid den kombinerade betydelsen av etnicitet och kön: sökande med arabiskt klingande mansnamn bedömdes mer sällan vara berättigade till ekonomiskt bistånd än de som tilldelats svenskklingande kvinnonamn. Vidare var inrikes födda professionella mer benägna att bevilja ekonomiskt bistånd än sina utrikes födda kollegor.

KEYWORDS:

• Social assistance
• assessments
• gender
• ethnicity
• vignette
• deservingness

NYCKELORD:

• Socialbidrag
• ekonomiskt bistånd
• bedömningar
• genus
• etnicitet

Introduction

Western welfare states have a long history of distinguishing between ‘the deserving’ and ‘the undeserving’ poor in order to legitimise the distribution of targeted benefits. While the moral and material implications of such distinctions vary over time and between societies, ideas of deservingness often carry assumptions related to gender and ethnicity (Fraser, 1987; Shilliam, 2018). In many countries, social work professionals play a key role in implementing policy concerning cash benefits, such as the Swedish social assistance (SA), into practice. Acting with a far-reaching scope of discretionary power, the professionals’ own values as well as societal norms might ultimately affect eligibility assessments (Ellis, 2007). This paper provides empirical input regarding the possible impact of applicants’ gender and ethnicity on assessments of SA eligibility in the Swedish context.

The picture of Sweden as a socially equal society, shaped by ideals of gender equality and colour-blindness, has been increasingly contested in recent years (e.g. Martinsson et al., 2016). From an international perspective, however, the Swedish welfare state is usually described as universalist. Poverty is mainly prevented through earnings-related social insurance programmes that all citizens formally access on the same conditions. SA concerns a minority of the population with constrained access to the more favourable social security programmes, and differs significantly from the major welfare system by being means tested (Nelson, 2004). In addition to the fact that SA targets some of the most disadvantaged in material terms, a significant share of the clients have psychosocial problems, for instance, related to substance abuse (Stranz et al., 2017).

Means testing SA is a municipal responsibility managed at social welfare offices and entitlement is determined through individual assessments. SA assessments in Sweden have shown to vary significantly between municipalities, and differences have been linked to both social worker characteristics and organisational factors (e.g. Gustafsson et al., 1990; Hydén et al., 1995; Stranz, 2007). Further, a number of studies indicate that professionals in the Swedish social services act according to different standards in their meeting with male and female SA clients, and tend to reproduce traditional gender roles in their assessments (e.g. Kullberg, 2005; Nybom, 2012). Such patterns also appear to be coloured by the professional’s gender, or more precisely by the combination of the professional’s gender and the applicant’s gender (Fäldt & Kullberg, 2012).

So far, there is a lack of knowledge of the possible occurrence of ethnic discrimination in SA assessments. This is despite the fact that immigrants are largely overrepresented in SA receipt (Gustafsson, 2013), and that racial/ethnic discrimination have been demonstrated in a number of other areas, including the labour market (e.g. Andriessen et al., 2012; Bursell, 2014; Quillian et al., 2019), the housing market (e.g. Ahmed & Hammarstedt, 2008), and among public officials (Adman & Jansson, 2017). In addition, the possibility that gendered assessment biases might be materialised through categorisations related to ethnicity remains to a large extent unexplored (cf. Hussénius, 2019).

The present paper aims to contribute to this knowledge gap by analysing data from a vignette study. Professionals administrating SA in their day-to-day jobs (n = 910) were provided with a vignette describing a SA applicant with substance abuse problems. The main objective here is to analyse whether the professionals’ willingness to make assessments on granting can be linked to the gender and ethnicity indicated by the applicant’s name. Further, the possible interaction between patterns related to the applicant’s position on the one hand and the professional’s gender and migration background on the other hand, is examined. The following questions are addressed:

1. To what extent, if any, can differences in professionals’ social assistance assessments be related to the gender and ethnicity of the applicant?

2. Is it possible to observe any assessment differences linked to the applicant’s gender and ethnicity that are specific for a) female and male professionals respectively and b) foreign-born and native-born professionals respectively?

Background

SA assessments are ultimately governed by the Social Services Act, which specifies that Swedish municipalities are responsible for ensuring people a ‘reasonable standard of living’ while assisting clients to self-sufficiency (Stranz et al., 2017). Practical social work with SA is located in the municipal personal social services (the PSS), whose other core areas include child welfare and substance abuse treatment for adults. The vast majority of the PSS’ employees are trained social workers holding a bachelor’s degree in social work (Bergmark & Lundström, 2007). As a professional group, they constitute a classic example of street-level bureaucrats whose practices often require balancing acts between the acquaintance with regulatory frameworks, meeting with clients and constraints related to organisational demands (Lipsky, 1980).

The Social Services Act specifies overall objectives while leaving for the municipalities to draw up detailed regulatory guidelines on the implementation of such goals. The possibility for local politicians and administrators of somewhat tailoring the PSS’s work to suit local conditions implies that the organisational setting in which SA assessments take place varies between municipalities, which also applies to the local guidelines regarding means testing (Bergmark & Lundström, 2007).

As regards eligibility for substance-abusing SA clients, the guidance provided by the legislation is limited (cf. Skogens, 2012). It is also not obvious that assessment variation can be prevented with rigorous municipal management. Since social workers often face situations that are too complex to handle with prescribed schemes of responses, higher degrees of detailed regulation might actually increase the professionals’ discretionary power by forcing them to deviate from the rules to be able to perform their tasks (Ellis, 2007; Evans & Harris, 2004; Stranz, 2007).

Whereas there is a lack of overviewing knowledge of the working methods that surround SA administration in Sweden, a few aspects are assumingly reoccurring elements. This includes the national norm, which specifies standardised cost items for certain expenses and allows for calculations considering the client’s level of ‘need’ (cf. Bergmark, 2013).¹ Further, clients must have exhausted all other possible sources of income, including other benefit programmes, and make efforts to achieve self-sufficiency in order to be eligible. In line with the international activation trend (Lodemel & Moreira, 2014), the latter dimension weighs heavily and granted aid is often conditioned by efforts with job-search or participation in activation measures (Nybom, 2012; Thorén, 2008).

The focus on applicants’ willingness to work has been discussed in terms of a work-test logic recognisable from the early poor laws (e.g. Nybom, 2012). Such a morally based criteria has also proven crucial for popular welfare opinions about deservingness: Attitudes towards targeted welfare are closely related to notions of laziness and to what extent recipients are seen as responsible for their own situation (van Oorschot, 2000, 2006). In SA assessments, social workers seem to put more stress on the client’s breadwinning responsibilities if the help-seeking person is male. In addition, the ways in which the problem area of the client is ‘gendered’ seem to affect the assessments (Fäldt & Kullberg, 2012; Kullberg, 2005).

However, research on gender-biased SA assessments mainly concerns judgments related to unemployed single parents without addiction problems. When social workers assess SA eligibility for clients with substance abuse, they might stress somewhat different criteria (cf. Skogens, 2012). Considering that concepts like ‘laziness’ and ‘work ethic’ carry ethnic and gendered stereotypes (e.g. Fraser & Gordon, 1994; Gilens, 1999), it is conceivable that deservingness judgments related to substance abuse do likewise.

Ethnic and gender stereotypes are further intertwined and tend to take different forms depending on the extent to which they reinforce each other (Ridgeway & Kricheli-Katz, 2013). Not least, this has been demonstrated in research on hiring discrimination. When assessing job applications, employers are generally more willing to recruit applicants from ethnic majorities. Meanwhile, gender seems to cut across patterns of ethnic discrimination in ways that are somewhat context bound. Whereas minority men seem to be particularly affected by discrimination in male-dominated occupations, minority women seem, correspondingly, to be especially disadvantaged in female-dominated areas (Bursell, 2014; Derous & Pepermans, 2019; Di Stasio and Larsen, 2020).

Moreover, tendencies of acting according to stereotypes in policy implementation are likely to vary between individuals and groups of professionals (Soss et al., 2008). For instance, same-gender dyads (i.e., when both the professional and the applicant is either male or female) seem to reinforce social workers’ tendency to reproduce traditional gender roles in their assessments (Fäldt & Kullberg, 2012). In addition to this, overlapping inequality systems of gender, ethnicity and class involve material arrangements that are likely to be tangible in the day-to-day practices of social work professionals (Ridgeway & Kricheli-Katz, 2013; Soss et al., 2008). In Sweden, where gender equality has become an integral part of the national self-image, social workers seem inclined to make assumptions about traditional gender roles and cultural backwardness when classifying and individual or family as ‘immigrant’, ‘Arabic’, or ‘Muslim’ (Eliassi, 2015; 2017; Jönsson, 2013; Pringle, 2010).

The recognition of certain demographic groups as particularly at risk of poverty – as well as the notion that help-seeking behaviours are likely to vary between groups – must also be taken into account when interpreting data on actual decisions on SA eligibility (Hussénius, 2019). From studies based on actual applicants, we know that females more often are granted SA than their male counterparts. However, substance abuse problems have been linked to a decreased chance of granting for female clients (Hussénius, 2019). Among those who apply, it has furthermore been shown that foreign-born more commonly are granted SA compared to native-born applicants (Stranz et al., 2017). This is particularly the case for foreign-born women, whereas male natives to lesser extent are granted SA among those who apply (Hussénius, 2019). Corresponding to an intersectional order of unemployment rates, this could indicate that professionals are particularly prone to grant SA to those assumed to face multiple constraints in accessing to other sources of income (cf. Hussénius, 2019; Soss et al., 2008). Speaking against such an interpretation, however, SA transfers have shown to compensate for foreign-born single mothers’ deficit to lesser extent than for native-born single mothers (Stranz & Wiklund, 2011).

Methods

In order to systematically review the occurrence of assessment differences, this study makes use of a vignette: A fictive case description. The study is based on a ‘paper-people-approach’, meaning that the participants made explicit responses to the short scenario described in the vignette (Aguinis & Bradley, 2014). Four versions of the vignette were randomly distributed to the respondents at the sample level (n = 910). Except for the applicant’s name and gendered pronouns, all versions included identical information. The vignette concerned a single adult with addiction problems who had been long-term dependent on SA. S/he was further described as having two children of which the other parent had custody, and no income but housing allowance. The incomes and expenditures were set to imply an obvious deficit in proportion to the national norm, whereas other aspects of the client’s situation were designed to allow for ambiguity. These included that the applicant had started a 12 steps treatment programme for substance abuse, but abstained from showing up in the past weeks. In addition, albeit having access to the children on a regular basis, the client had not seen them in over a month.

The names that were varied in the vignette were two common Swedish-sounding names, Malin (n = 227) and Jonas (n = 223) and two common Arabic-sounding names, Aisha (n = 217) and Mohammed (n = 243). The choice of names was motivated by previous research highlighting Arabic-sounding names as particularly subject to discrimination in Sweden (e.g. Ahmed, 2010; Carlsson, 2010; Khosravi, 2012). In order to increase the degree of external validity (cf. Aguinis & Bradley, 2014), the vignette was constructed in dialogue with trained social workers with specific experience in SA administration. Acknowledging that substance abuse is likely to seem most ‘typical’ for applicants the Swedish-sounding male name, an important aspect was to construct a case that, irrespective of the name attribution, would not be too unlikely for the participants to imagine. The credibility of the vignette was further tested in a pilot including 35 professionals employed in the PSS in a middle-sized Swedish municipality. Since the feedback from the pilot was positive, no subsequent changes were made in the vignette.

The vignette was included in a questionnaire comprising eight vignettes, six of which addressed follow-up analyses of previous studies (Hydén et al., 1995; Stranz, 2007). The experimental design was applied in the two remaining vignettes. Since the latter two covered cases of very divergent nature, the present paper solely focuses on the participants’ responses to one of them, henceforth (and previously) referred to as ‘the vignette’. All vignettes in the questionnaire concerned SA assessments and were essentially designed in the same format, each presented to the respondents together with a fixed set of questions. These included the respondent’s decision (approval/rejection), the size of the grant (if the application was approved) and the possible circumstances motivating the assessment. In addition, the questionnaire contained two sets of items regarding the participants’ background such as age, gender and country of birth (in/outside Sweden), and professional indicators such as education, work experience and work load. Finally, the participants were asked to respond to a set of attitude items covering claims related to for instance conditionality of SA eligibility, and overuse of the benefit.

Forty-three social welfare offices in 19 municipalities participated in the study. These included all city-districts in Sweden’s three largest cities (Stockholm, Gothenburg and Malmö), 10 municipalities in the northern part of the Stockholm region, and 6 municipalities spread across the country.² All in all, the municipalities in the sample range from sparsely populated rural communities to the most densely populated cities. Table 1 gives an overview of the municipalities’ basic demographic characteristics.

Table 1. Municipalities in the sample.

	City-districts	Inhabitants^a	SA households^b	Response rate	Respondents (n)
Major cities
- Stockholm	14	962,154	2.6%	81%	219
- Gothenburg	10	571,868	5.7%	71%	225
- Malmö	5	339,313	9.2%	86%	220
Other municipalities (n = 16)				83%	246
- Min		4055	1.2%	62%	2
- Median		53,288	3.1%	82%	12.5
- Max		127,119	8.9%	100%	59

^a Data from 2018 (Statistics Sweden, 2018).

^b Proportion of the households that received SA at some point during 2018 (data from the National Board of Health and Welfare, 2018).

As a first step of the data collection, complete lists of the staff involved with conducting SA assessments at the concerned PSS units in the municipalities were collected from the managers. These included administrative caseworkers as well as trained social workers and the heads of the units (n = 1263). Thereafter, visits at the welfare offices were booked and took place in the first half of 2018. Each visit was conducted by a member of the research team and followed a given scheme. The respondents were initially gathered to receive oral information about the study, and instructed on how to fill in the questionnaire.³ They were encouraged to make use of regular instrumentalities (such as software for norm calculations), but not allowed to talk to each other concerning the assessments. Each visit lasted for up to two hours. The design of the data collection enabled a high participation. In total, the response rate was approximately 80% (n = 1005). After accounting for item non-responses in the control variables, the sample comprised 910 observations.

Variables and analyses

Table 2 overviews the professionals’ characteristics including the independents used in the multilevel logistic regression models (specified below). As shown, the majority of the respondents were women, which can be regarded as representative for the field of social work in Sweden (Statistics Sweden, 2010). Further, almost a fourth were born outside Sweden, which is worth noting since the share of immigrants in the whole population is approximately 20% (Statistics Sweden, 2019). The small amount of men implied that regression analyses addressing the possible association between respondents’ gender and migration background on the one hand, and name-related patterns on the other, was not applicable on data albeit the large sample size.

Table 2. Respondent variables (n = 910).

Female	87%
Born in Sweden	76%
Age, mean (sd)	37.0 (11.4)
Work position
Manager^a	10.4%
Social worker	82.6%
Administrative caseworker	3.9%
Other	3.8%
Basic education
Social work degree	80%
Other university degree	15%
Other	5%
Years in working with SA, mean (sd)	6.1 (7.3)
Engagement in SA (% of employment), mean (sd)	90.2 (21.0)
Conditionality, mean (sd)	2.78 (0.93)
Number of approvals according to the other vignettes, mean (sd)	3.6 (1.3)

Notes: The category includes unit managers, deputy unit managers and head social workers.

In order to address the research questions, the correlation between the gender and ethnicity indicated by the applicant’s name and the respondent’s willingness to grant SA was initially tested by means of cross tabulations using Pearson’s χ²-test. To account for the fact that the respondents were clustered within 43 social services offices, data were further analysed using multilevel logistic modelling (Wallander, 2012). By defining the social service offices as level 2-units, the multilevel model approach allows the intercept to vary between the clusters. This enables estimates of the name’s ‘average’ impact on the dependent variable, irrespective of local disparities in levels of generosity (Skrondal & Rabe-Hesketh, 2010).

Data were scrutinised using Stata, version 15.1. Three sets of logistic regression multilevel models were conducted, in which the respondent’s answer to whether or not s/he would grant SA served as the dependent variable (1 = yes). Except for the names of the applicant, a number of independents were included to control for other factors likely to co-vary with the assessments. Since social judgments and attitudes to welfare might shift between subgroups (e.g. Byers & Zeller, 1998) this included background characteristics related to the social worker: gender, age, country of birth (within/outside Sweden). Also, a set of professional indicators were included to account for the fact that the respondents’ percentage of working hours attributed to SA varied (engagement in SA), and that assessments might depend on the work position, education and work experience of the respondent (cf. Samuelsson, 2015; Stranz, 2007). Further, an attitude item (conditionality) was included in the model, concerning the respondent’s position on the claim ‘The social services should be given greater opportunity to require treatment of substance abuse and/or psychological problems in relation to applications for social assistance’.⁴ Also, a variable summing up the respondent’s total number of approvals in relation to the other seven vignettes in the questionnaire was constructed (ranging from 0 to 7) to control for the professionals’ individual profile in terms of generosity.

The subsequent sets of logistic multilevel models included mainly the same control variables as presented above. However, to address research question 2, interaction terms in which combinations of the applicant’s name on the one hand, and the professional’s gender (Model 2) and migration background (Model 3) on the other, were used. By altering the reference category between each of the possible combinations, eight versions of Model 2 and Model 3 respectively were conducted. In order to decrease the risk of Type 1 errors, Benjamini and Hochberg’s procedure for false discovery rate correction was applied (Benjamini & Hochberg, 1995).

Results

As shown in Table 3, we can initially state that there is far from any consensus among the professionals considering the applicant’s right to SA. Whereas 54% of the respondents answered that they would grant SA, almost half of them specified that they would not.

Table 3. Eligibility assessments by name of the applicant. Pearson’s χ²-test. n = 910.

Vignette variables	Eligible for SA (%)	Χ^2 (P)
All	54%
Gender		3.4307 (0.064)
Male (Jonas/Mohammed)	51%
Female (Malin/Aisha)	57%
Ethnicity		5.5370 (0.019)
Swedish-sounding name (Malin/Jonas)	58%
Arabic-sounding name (Aisha/Mohammed)	50%
Gender and ethnicity		9.2336 (0.026)
Swedish-sounding f. name (Malin)	59%
Swedish-sounding m. name (Jonas)	56%
Arabic-sounding f. name (Aisha)	54%
Arabic-sounding m. name (Mohammed)	46%

Moving on to the name-related granting rates, Table 3 indicates that the professionals were more inclined to assess those with a Swedish-sounding name as eligible to support, compared to those with an Arabic-sounding name. When comparing assessments based on the applicant’s gender, it is suggested that women are more likely to be assessed as eligible. However, this tendency is less accentuated and should be interpreted with some caution.

Further, Table 3 also suggests that the interaction between gender and ethnicity is crucial to understanding how the name of the applicant is related to assessment differences. The granting rates of the vignettes signed Jonas and Aisha are both close to the average and do not substantially differ from one another. In contrast, the applicant called Malin is assessed as eligible a little more often. Her position is contrasted by Mohammed, whose name is linked to a substantially lower amount of granting than the other names.

In Table 4, the impact of the name on the likelihood of being assessed as eligible is estimated in a multilevel logistic regression model.⁵ Here, the names of the respondent are included as dummy variables, using Mohammed as the reference category.⁶ As shown in Table 4, Mohammed’s chance of approval seems to be lower than for Malin when controlling for social worker factors. However, none of the name differences turn out to be statistically significant at conventional levels. Additional models where the reference category was altered between Aisha, Malin and Jonas were conducted (not presented in the table), but did not reveal any significant differences between these three names relative to one another.

Table 4. Multilevel logistic regression of name variables and social worker characteristics on eligibility assessment. Odds ratios and confident intervals. n = 910.

	Model 1
	OR	CI (95%)
Vignette variables
Mohammed	ref
Jonas	1.59	1.04–2.42
Malin	1.70*	1.12–2.59
Aisha	1.32	0.87–2.01
Respondent variables
Gender (1 = male)	0.92	0.59–1.43
Country of birth (ref = Sweden; 1 = other)	0.50***	0.35–0.73
Age (continuous)	1.00	0.98–1.02
Social work degree (1 = yes)	0.95	0.61–1.47
Work experience (years in work with SA)	1.00	0.97–1.03
Work position (ref = regular social worker)
Manager	1.79	1.05–3.06
Administrative caseworker/other	0.29*	0.01–0.75
Professional engagement in SA (% of employment)	1.00	0.99–1.01
Conditionality (1–4)	0.76***	0.64–0.90
Number of approvals (0–7)	1.36***	1.20–1.54

Note: Benjamini-Hochberg adjusted p-values: *p < 0.10, **p < 0.05, ***p < 0.01.

At the same time, several social worker characteristics can be linked to assessment differences at a general level. Not least, this applies to whether the professional is native or foreign-born. Irrespective of the applicant’s name, social workers born in Sweden were approximately twice as likely to assess the applicant as eligible for support compared to those born outside Sweden. Further, the results indicate that administrative caseworkers made less generous assessments than the regular social workers did. One possible explanation is that the former are less familiar with making decisions on this type of clients, and tend to reject applications in cases of doubt. Finally, and hardly surprising, professionals who preferred a higher degree of conditionality for substance abusing clients made more strict assessments. Also, greater generosity in relation to the other vignettes in the questionnaire corresponded to a higher granting rate.

Whereas a general difference in generosity could be linked to the professional’s migration background, female and male respondents generally seem to make similar assessments. In order to address the possible interaction between the respondent’s gender and country of birth vis-à-vis the applicant’s name, we must however review the following sets of analyses. Starting out with gender, the Model 2 series in Table 5 includes the same set of control variables as Model 1. Here, the separate indicators of the professional’s gender and the applicant’s name have been replaced with dummies representing each possible combination between the two.⁷

Table 5. Multilevel logistic regression of name variables and social worker characteristics on eligibility assessment. Odds ratios. n = 910.

Model	2a	2b	2c	2d	2e	2f	2g	2h
Intercept	2.16	1.63	1.28	1.09	0.70	3.03	2.24	0.94
Female professional
Malin	ref	1.32	1.68	1.99**
Jonas	0.76	ref	1.27	1.50
Aisha	0.59	0.78	ref	1.18
Mohammed	0.50**	0.67	0.85	ref
Male professional
Malin					ref	0.23	0.31	0.74
Jonas					4.37	ref	1.36	3.23
Aisha					3.22	0.74	ref	2.38
Mohammed					1.35	0.31	0.42	ref

Notes: Control variables: respondent’s country of birth, age, social work degree, work experience, work position, engagement in SA, attitude to conditionality and number of approvals. Benjamini-Hochberg adjusted p-values: *p < 0.10, **p < 0.05, ***p < 0.01.

In Models 2a–2d, we can see that the disfavouring of Mohammed in relation to Malin, which was indicated at the overall level, is more pronounced among the female professionals. This is while no assessment difference can be confirmed among the male professionals (Model 2e–h). Note however that a brief look on the coefficients would seem to suggest that, if anything, the men in the sample might assess Jonas more favourably than they assess both Malin and Mohammed. The following set of regression models examines the possible interplay between the respondent’s migration background and the applicant’s name (Table 6).

Table 6. Multilevel logistic regression of name variables and social worker characteristics on eligibility assessment. Odds ratios. n = 910.

Model	3a	3b	3c	3d	3e	3f	3g	3h
Intercept	0.89	0.89	0.74	0.55	0.54	0.44	0.31	0.27
Native professional
Malin	ref	0.99	1.19	1.60
Jonas	1.00	ref	1.20	1.61
Aisha	0.84	0.83	ref	1.34
Mohammed	0.62	0.62	0.75	ref
Foreign-born professional
Malin					ref	1.22	1.71	2.01
Jonas					0.82	ref	1.40	1.65
Aisha					0.58	0.71	ref	1.18
Mohammed					0.50	0.61	0.85	ref

Notes: Control variables: respondent’s gender, age, social work degree, work experience, work position, engagement in SA, conditionality and number of approvals. Benjamini-Hochberg adjusted p-values: *p < 0.10, **p < 0.05, ***p < 0.01.

As was shown in Model 1, there seems to be a general difference between native-born and foreign-born professionals, according to which the latter assess the applicant more strictly. When dividing the professionals by country of birth, no name differences turn out to be statistically significant (Model 3e–3h). In contrast to the gender-specific assessments hierarchies found in Model 2, the relational differences seem to be in the same direction independent of the professional’s country of birth.

Discussion

To conclude, the results can be summarised as follows. Applicants with Swedish-sounding names were to a higher extent granted SA compared to those with Arabic-sounding names. The granting rates were also slightly higher in assessments concerning female applicants than male applicants. The position that most clearly differed from the others was that of the Arabic-sounding male name, Mohammed, which corresponded to a lower granting rate. The disadvantage of Mohammed was most apparent in relation the Swedish-sounding female name, Malin.

The regression models were designed to control for factors at the professional level while accounting for municipal clustering and multiple comparisons. The analyses nuanced the initial picture by indicating that the eligibility gap between Malin and Mohammed primarily was manifest in assessments made by female professionals. Thus, the answer to the first research question – which addressed general assessment differences linked to the applicant’s gender and ethnicity – is that no such linkage could be concluded.

The second research question addressed patterns specifically linked to subgroups of professionals. According to this, as mentioned, an assessment bias was demonstrated among the female professionals, which could not be found among the men in the sample. Recalling that the women constituted 87% of the participants (which is also representative for the profession), a degree of caution may be exercised when interpreting the gender-specificity of this finding. Finally, while native-born professionals were substantially more prone to view the applicant as eligible than foreign-born professionals, the participant’s migration background could not be linked to the degree of discrimination between the names.

How, then, can we understand the findings? The participants were provided with the same information about the applicant’s incomes and expenditures, family situation and substance abuse problems. In relation to the national norm, the applicant’s need for SA was obvious. Although remaining careful in drawing conclusions about the decision-making process, we might assume that the professionals based their assessments on this information, but also, that they did so through assumptions and values that were not specified in the vignette. The applicant’s gender and ethnicity seems to have played a part in this process, at least in the female professionals’ assessments.

If we contextualise the results in relation to demographical data, we know that immigrants are at significantly higher risk of poverty than natives. Further, research on actually conducted SA decisions suggests that the ratio of granting is highest among female immigrant applicants, followed by immigrant men, native-born women and native-born men (Hussénius, 2019). Given the factual socioeconomic disparities – that are certainly tangible for social work professionals – this correspondence could hypothetically follow a compensating logic (in terms of higher deficits corresponding to higher granting rates). It could also be envisaged that general knowledge about the varying degree of need for SA in different subgroups might guide the professionals’ assumptions about individual clients. However, if this were the case, and need were the rationale of the discriminating behaviour, higher granting rates would reasonably have been linked to the Arabic-sounding names in this study. From this perspective, the notion of an ‘immigrant penalty’ highlighted in research on popular welfare opinions might provide more reasonable parallels. According to this, foreign-born across Europe are seen as the least deserving compared to other groups in their countries. This is despite the fact that the degree of need constitutes an otherwise crucial criterion for deservingness opinions, and that immigrants generally face more structural constraints than natives (e.g. Reeskens & van der Meer, 2019; van Oorschot, 2006).

From a wider perspective, the picture sketched by the results fits ill with the narrative of Swedish universalism. The level of disagreement among the respondents was, on the whole, tangible. This strengthens the impression that SA assessments bear an inherent risk of legal uncertainty. Previous research has demonstrated gendered consequences of this scope for variation. The present study adds valuable input to the field by highlighting that also notions of ethnicity – and intersections between gender and ethnicity – might affect SA assessments. This is particularly worrying when considering that SA targets individuals who are, by definition, already in a very exposed situation. In addition, the likelihood of lacking access to more beneficial social security programmes is, by far, higher among ethnic minorities.

Further, the observed bias suggests a gender advantage that primarily benefits women categorised as ‘Swedish’. It also indicates that the corresponding gender disadvantage selectively affects men categorised as ‘Arabic’. This is in line with findings from field experiments on the labour market showing that minority men tend to be more affected by discrimination than women in ethnic minorities (e.g. Arai et al., 2016). While noting that this could point to a somewhat recurring pattern of discrimination, we must also consider that the demonstrated bias might be related to the specific setting of the vignette.

In terms of family situation, the client was described as having two children, who both lived with the other parent. Thereby, the scenario reflected a more gender-traditional setting for the male applicants. In combination with attributions of traditional family roles onto ‘Arabs’, and contrasting ideas of ‘Swedes’ as carriers of gender equality, intersectional stereotypes could be reflected in implicit assumptions about the applicant’s parental ability and the children’s situation.

Likewise, assumptions regarding the implications of substance abuse problems are likely to be affected by processes of categorisation. Thus, the applicant’s name combined with the occurrence of substance abuse problems might have triggered different ideas about, for instance, the individual’s opportunities of receiving support from their relatives. It may also have given rise to different logics of moral reasoning. For instance, stigma interlinking ethnic minorities, poverty and masculinity often involve associations to criminality. In this light, one might consider that the notion of substance abuse problems, when combined with the name of Mohammed, tends to be perceived not only (or primarily) as a psychosocial health impairment affecting the wellbeing of the applicant and his family. It may also allude to the understanding of substance abuse as linked to illegal activities. Following such a reasoning, the occurrence of substance abuse problems may have contributed to strengthen a tendency among the professionals of viewing ‘Arabic’ men with a certain degree of mistrust (cf. Bursell, 2014; Ridgeway & Kricheli-Katz, 2013). On the other hand, it can also not be precluded that the discriminating bias towards Mohammed would have turned out as even more accentuated if the vignette had featured elements that were more consistent with stereotypical beliefs about Arabic men (e.g. religiousness). Similarly, it is possible that the position of Malin more clearly would have anticipated an increased chance of granting, had the applicant not been described to have addiction problems (cf. Hussénius, 2019).

In the case of Aisha, it can be speculated that the professionals categorised the applicant as an immigrant woman from a culture where women’s substance abuse problems are taboo subjects. Thereby, Aisha’s social vulnerability might have been perceived as especially pronounced. Although the granting rates were similar for Aisha and Jonas, corresponding chains of reasoning were probably different in the latter case. Starting from the premise that the name of Jonas is most compatible with the stereotyped image of what a substance abusing SA applicant looks like (namely a man categorised as ‘Swedish’), the professionals may have considered his psychosocial situation as somewhat expected and perhaps less problematic.

Whereas these interpretations are hypothetical, they highlight that the specificity of the vignette has implications for the generalisability of the data. Since the study design did not enable analyses of how the applicant’s intersectional position interacts with other circumstances defined in the vignette (e.g. substance abuse), more research is needed in order to properly contextualise the findings. In future research, scholars may consider to use the factorial survey approach in order to analyse assessment differences related to SA applicants’ gender and ethnicity while varying factors such as the occurrence of (different) psychosocial problems (cf. Wallander, 2012).

As regards the limitations of the study, a few additional aspects should be discussed. The applicant’s ethnicity has been operationalised by reviewing assessment differences related to Swedish- and Arabic-sounding names. This implies that the observed patterns cannot be assumed to apply to differences between applicants categorised as ‘Swedes’ and all other ethnicity-related bases of categorization.

Further, the participants’ assessments of the vignette cannot simply be interpreted as direct reflections of their behaviour in meetings with real clients. Rather than representing the full complexity of real-life situations, vignettes allow us to study key aspects of decision-making processes. Used in the context of a carefully prepared data collection, vignettes serve as methodological tools that provide indications of the participants’ behaviour in real-life assessments (cf. Aguinis & Bradley, 2014). In contexts like SA assessments, any similar procedure would be impossible to carry out in the natural setting, which underlies the value of projects like the present. The fact that similar patterns have been observed in field experiments may be seen as speaking in favour of the study’s external validity. When assessing job applications in the natural setting of a hiring process, employers tend to act according to the same kind of discriminating bias as was displayed by the female professionals that participated in this study (e.g. Bursell, 2014).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Forskningsrådet om Hälsa, Arbetsliv och Välfärd [2016-00438].

Notes on contributors

Klara Hussénius

Klara Hussénius is a PhD student at the Department of Social Work, Stockholm University. Her research interests include social assistance, social policy, feminist theory and intersectionality. Alongside with her PhD project, Klara is also engaged in teaching, primarily on the subjects of poverty and the Swedish welfare state.

Notes

1 The norm was imposed to curb municipal divergences in levels of generosity, and can be seen as a common reference framework of what a ‘reasonable standard of living’ corresponds to in monetary terms. However, the efficacy of the norm to increase the degree of equivalence in SA assessments and prevent levels of compensation from being dumped has been doubtful (Bergmark, 2013).

2 Three additional municipalities were asked but declined participation. Two were small municipalities (<15,000 inhabitants) and one was a middle-sized municipality (20,000–50,000 inhabitants).

3 In order to avoid response bias, the participants were not informed about the name manipulation.

4 Spanning from 1 to 4, higher numbers correspond to greater agreement.

5 To account for the clustering between city-districts in the major cities, additional models in which dummies for Stockholm, Gothenburg and Malmö were included were also conducted. However, this did not affect the results in any substantial way.

6 The choice of reference category was based on the recognition of Mohammed as the name with a granting rate that most clearly differed from the others.

7 For the sake of clarity, odds ratios concerning the control variables are not presented in the table.