Suspicious Minds? Media effects on the perception of disability benefit claimants

Abstract

The media are often blamed for widespread perceptions that welfare benefit claimants are undeserving in Anglo-Saxon countries – yet people rarely justify their views through media stories, instead saying that they themselves know undeserving claimants. In this paper, I explain this contradiction by hypothesising that the media shapes how we interpret ambiguous interpersonal contact. I focus on disability benefit claimants, which is an ideal case given that disability is often externally unobservable, and test three hypotheses over three studies (all using a purpose-collected survey in the UK and Norway, n=3,836). In Study 1, I find strong evidence that a randomly-assigned ‘benefits cheat’ story leads respondents to interpret a hypothetical disability claimant as less deserving. Study 2 examines people’s judgements in everyday life, finding that readers of more negative newspapers in the UK are much more likely to judge neighbours as non-genuine – but with effectively no impact on judgements of close family claimants, where ambiguity is lower. However, contra my expectations, in Study 3 I find that Britons are no more likely than Norwegians to perceive known claimants as non-genuine (despite more negative welfare discourses), partly because of different conceptions of what ‘non-genuineness’ means in the two countries.

Introduction

Print/TV media have often been blamed for the widespread perception in Anglo-Saxon countries that benefit claimants are not ‘deserving’. In the UK, longstanding concerns over newspapers (Page, 1984:40) were recently revived amidst a surge in both negative newspaper coverage and ‘reality’ television (e.g. Baumberg et al., 2012; Paterson et al., 2016; Reeves and de Vries, 2016). In the US, claimants have long been represented as disproportionately Black and undeserving (Gilens, 1996; Gilens, 1999; Misra et al., 2003), which has been causally related to harsh attitudes (Gilens, 1996; Hannah and Cafferty, 2006; Iyengar, 1990; Iyengar, 1991; Nelson and Oxley, 1999; Rose and Baumgartner, 2013; Shen and Edwards, 2006; Sotirovic, 2000, 2001).

Yet there is a problem with these accounts: when people are asked to justify perceptions of benefit fraud, they rarely mention the media, and instead cite people they know – “we’ve got a neighbour who does it” (Briant et al., 2011:64). Indeed, in Briant et al.’s (2011) study, every respondent claimed to have ‘first-hand knowledge’ of fraudulent claims, while in Golding and Middleton (1982: 172-3), twice as many justified fraud perceptions via personal observation than media stories. Moreover, when scholars study public conversations about social issues in-depth, they find a “clear hierarchy, with personal experiences and anecdotes at the top, and media stories and statistical information at the bottom” (Rolfe et al., 2018:59; see also Gamson, 1992:123). If most people know undeserving claimants, and these perceptions are trusted more than the media, then the media’s role may have been over-stated.

In this paper, I present and test a new theory that resolves this contradiction, by suggesting that the deservingness we ‘see’ in everyday interactions is influenced by the media-influenced frames we use to make sense of the world. Surprisingly, given well-developed literatures on media effects, this argument is novel – and if it is correct, it would rebut this potential challenge to media effects, explaining how the media can be powerful even in the midst of widespread everyday tales of undeservingness.

I focus on disability benefits, partly because disability is the major category of benefits for working-age people internationally, ¹ and partly because it is an ideal case study: it is often unobservable and fluctuating, with ambiguous signs of deservingness that are susceptible to framing effects. I use novel data to experimentally test if media-style frames can cause people to respond differently to hypothetical contact with a vignette claimant (Study 1), to test if perceptions of claimants in everyday life match my hypotheses (Study 2), and to test if perceptions vary within the different discursive contexts of the UK vs. Norway (Study 3). To begin, however, I situate my hypothesis within the wider media effects literature.

Existing theories of the media and benefits attitudes

To date, researchers have explored two main theories to explain how media coverage influences benefits attitudes: ²

1. 1. Cultivation theory suggests that the more time people spend consuming media, the more their worldview resembles it (Morgan and Shanahan, 2010). ³ Suggestive evidence supports this: attitudes to claimants are more hostile among those who consume more negative media sources (Baumberg et al., 2012; Gilens, 1996; Sotirovic, 2000). However, establishing causality through these cross-sectional studies is difficult, given that people choose their media sources based on pre-existing needs/beliefs (Baumberg et al., 2012; Gilens, 1996).

2. 2. Framing theory suggests that media ‘frames’ – that is, the way that media representations organise the many pieces of information relating to an issue (Scheufele and Tewksbury, 2006) – affect how people think about claimant deservingness. There are many framing studies on welfare attitudes, typically using survey experiments that present people with different narratives (Avery and Peffley, 2003; Hannah and Cafferty, 2006; Iyengar, 1990; Iyengar, 1991; Nelson and Oxley, 1999; Shen and Edwards, 2006; Slothuus, 2007). These permit stronger causal inference, and mostly show that frames impact some – but not all – outcomes (Avery and Peffley, 2003; Hannah and Cafferty, 2006; Nelson and Oxley, 1999; Shen and Edwards, 2006; Slothuus, 2007). Other methodologies also show mixed results, e.g. convincing natural experiments show both significant (Reeves and de Vries, 2016) and null effects (Hedegaard, 2014a). Nevertheless, the overall balance suggests that negative frames are likely to negatively impact attitudes.

Despite this considerable evidence on benefits attitudes-media links, there is almost no evidence that has looked at the connection between media coverage and people’s direct experiences (indeed, few studies have examined this even in the wider media effects literature). Those that do have mostly assumed that the media matters most where people have little direct experience of the phenomenon in the story, often citing Lippmann 1922’s argument that the media have power because “our opinions cover a bigger space, a longer reach of time, a greater number of things, than we can directly observe” (Gilens, 1996:515). This assumption is explicit within some studies on benefits (Sotirovic, 2000; 2001:754); for example, Gilens (1996:530-1) argues that if personal experience does not explain the perceived racial make-up of claimants, media representations must be responsible.

A smaller number of researchers have studied other interactions between media representations and experiences. Some have argued that the media matters most when it resonates with people’s experiences (within cultivation theory and also e.g. Gamson, 1992:125-134). Similarly, Hopkins has convincingly argued that the combination of high local levels of immigration and a ‘politicizing agent’ (such as the media) affects people’s attitudes (Hopkins, 2011). Yet while qualitative studies show how people counter/support media stories about welfare with personal experiences (Briant et al., 2011; Paterson et al., 2016), these ideas have otherwise not been applied to studies of benefits attitudes.

In my theory, in contrast, it is not that the media supplants our direct experiences, or is only powerful if it resonates with them – but rather that the media fundamentally shape our interpretation of these experiences. This is a novel argument; I now develop this into a testable theory.

Developing a new theory

Our theory starts from two observations: we know that stereotypes influence judgements about the people we interact with (Shrum, 2009:54), and that media representations influence such stereotypes (Roskos-Ewoldsen et al., 2009). It is therefore likely that media representations change the way that we judge known others via their influence on stereotypes. Yet surprisingly, this hypothesis has never been tested – neither for welfare attitudes nor more broadly. Gamson (1992:125) comes closest when noting in passing that “even our personal experience is filtered through a culturally created lens… We walk around with hyperreal images from movies and television and use them to code our own experiences” – but this is not developed further. I hypothesise:

H1: Media frames influence how known welfare claimants are judged. If media frames suggest that known claimants are mainly fraudulent, then people will ‘see’ fraud in their everyday contact and judge claimants harshly.

There is almost no previous evidence on H1. While there is considerable evidence of the effect of frames on general welfare attitudes (see above), it is not obvious that general effects will carry over to judgements of specific claimants. I know of only one framing experiment that looked at perceptions of an unrelated vignette claimant, which – despite explicitly describing the vignette to respondents as a ‘perfect example’ of the story – finds only inconsistent effects (Avery and Peffley, 2003).

Secondly, my theory makes specific predictions about the circumstances in which these media effects will be found. Prima facie, stereotypes seem more likely to influence judgements based on ambiguous information; otherwise the direct information will outweigh the stereotype (Aarøe and Petersen, 2014). In the case of known welfare claimants, ambiguity will tend to be greater for less well-known acquaintances vs. close friends (our ‘peripheral’ vs. ‘core’ network; Morgan et al., 1997). I hypothesise:

H2: Media frames primarily influence judgements of peripheral network members (e.g. acquaintances) where ambiguity is greater, with little/no effect on judgements of core network members (e.g. close family).

While this echoes various strands of previous work, the hypothesis itself is novel. For example, intergroup contact positively influences attitudes (the ‘contact hypothesis’; Hewstone and Swart, 2011), and more specifically on welfare, those who know service users – greater policy ‘proximity’ – will be more positive towards services (Hedegaard, 2014b). However, neither theory contrasts core vs. peripheral relationships, ⁴ nor do they examine how media shapes the interpretation of contact. Still, H2 parallels wider accounts that knowing claimants can – contingently – influence welfare attitudes.

Third, if my hypothesis holds at the individual level, we would also expect it to hold at the societal level. That is, in countries where stereotypes of welfare claimants are more negative, I expect that claimants – particularly in peripheral networks – will be judged more harshly. To test this, I compare the UK and Norway: newspaper reporting of benefits fraud is common in the UK but rare in Norway (Larsen and Dejgaard, 2013). I hypothesise that:

H3: Judgements of known welfare claimants will be more positive in Norway (vs. the UK), particularly for peripheral (vs. core) network members.

We must also bear in mind wider cross-national differences in attitudes (general perceptions of undeservingness are lower in Norway ⁵ ) and policy (disability benefit claims in Norway are 2-3 times as prevalent; see Appendix D). But there is no reason to think that these wider differences will lead to cross-national differences in how peripheral vs. core network members are judged – hence H3 provides a useful additional test of my theory.

Applying this to disability benefit claimants

We focus on disability benefits, partly because they are the major category of working-age benefits (they are noticeably more prevalent than unemployment claims in high-income countries¹), and partly because they are an ideal test of my hypotheses. This may seem surprising, as the deservingness literature conventionally argues that disabled claimants in general are perceived to be deserving (Jensen and Petersen, 2017; van Oorschot, 2000, 2006), usually explained by lower average levels of control and higher levels of need (within the ‘CARIN’ model of van Oorschot and Roosma, 2017). However, this evidence is often misinterpreted: it shows that some disabled claimants – paradigmatic “genuine” cases evoked by terms like ‘sick and disabled people’ (van Oorschot, 2006) – are perceived to be deserving, but does not necessarily mean that all disabled people are seen in this way.

In fact, wider evidence shows that only some disabled claimants are judged to be deserving. Most directly, vignette studies show substantial variations in perceptions of different disability benefit claimants (Geiger, 2021). We can also see disabled people being differentially judged in e.g. political debates (Morris, 2016; Pennings, 2011; Soldatic and Pini, 2009) and everyday life (Hay, 2010; Holloway et al., 2007). These distinctions are usually based on whether a claimant is perceived to be ‘genuinely’ disabled, which is often unclear to external observers – there are few demonstrable external signs of impairments related to e.g. pain or mental ill-health (Baumberg et al., 2012:26-7; Holloway et al., 2007). Moreover, disabilities are often fluctuating, which makes even those outward signs of disability on one day an unreliable guide to capacities on another (Boyd, 2012).

Disability claimants are therefore an ideal test of H1 – people make sharp judgements about the disabled people they know, but the outward signs of disability from interpersonal contact are ambiguous, and this ambiguity allows space for media-influenced stereotypes to influence judgements. Disability benefits are also a good test for H2: qualitative studies show that disabled people’s core network tends to support their accounts, with scepticism usually reserved for more peripheral relationships (Hay, 2010:265) – notwithstanding that people are occasionally judgemental about close friends/family (Holloway et al., 2007:1460). In the conclusion I consider how far my findings are likely to extend to other types of claimant.

In the rest of the paper, I test these hypotheses across three studies, all using a purpose-collected survey of nearly 4,000 people in the UK/Norway (described below). Study 1 tests if an experimentally-assigned newspaper-style story influences judgements of a specific vignette claimant (testing H1). Study 2 examines perceptions of ‘non-genuine’ disability claims in everyday life, testing how these vary according to media use and relationship type (testing H2). Study 3 also examines perceptions of ‘non-genuine’ disability claims in everyday life, but now testing differences across countries (testing H3). The studies complement one another’s strengths/weaknesses: Study 1 allows confident causal inferences but weak ecological validity; whereas Studies 2 and 3 provide more tentative causal inferences but with strong ecological validity. Ethical approval was given by the lead author’s institution, and replication code/data are publicly available (see Appendix I).

Study 1: Survey experiment

H1 hypothesises that media frames will influence how a specific disabled benefit claimant is judged. To test this, Study 1 uses a survey framing experiment, investigating if a newspaper-style article on disability benefits fraud affects judgements of a vignette disability benefit claimant. As a test of specificity, I further check that the fraud framing has no impact on judgements of non-disabled claimants (see below). Framing experiments allow strong causal inferences, and have been widely used to study the impact of the media on attitudes (see above).

Data

The experiment was embedded within a survey of 3,836 people in the UK and Norway using YouGov’s opt-in panels. To achieve an approximately representative sample, panel members were invited to participate according to sociodemographic criteria, and the resulting sample was then weighted to known population totals (see Appendix E). The diversity of this sample is a strength compared to the student samples used in most previous welfare framing experiments (Hannah and Cafferty, 2006; Nelson and Oxley, 1999; Shen and Edwards, 2006; Slothuus, 2007), with rare exceptions (Avery and Peffley, 2003; Iyengar, 1991). I conducted the framing experiment in both countries because they vary considerably in welfare discourses (see H3 above); if we find similar effects in both then this supports the generalisability of my findings to other high-income countries.

Procedures and measures

Our experiment uses a sample newspaper story that had previously been developed by Ford and collaborators (embedded within the survey used for e.g. Kootstra and Roosma, 2018) based on stories in UK newspapers. The story is shown in Figure 1 below: it was seen by a random half of respondents (the ‘benefits cheat frame’ group), but not others (the control group).

Figure 1. Mock story used for survey experiment.

We then asked respondents to judge the deservingness of vignette claimants, each representing ‘hypothetical contact’ with a claimant. Rather than choosing a single vignette to represent all claimants, I used a multiple factorial design to investigate the effects of the benefits cheat frame on different claimant types. Most vignettes described disabled claimants, but some described non-disabled unemployed people, who I hypothesised would not be affected by the disability benefits cheat framing (see below). Each disabled claimant vignette contained eight experimentally-varied dimensions, while each unemployment vignette contained six dimensions (see Table 1). I obtained 8,605 disability vignette judgements and 2,468 unemployment vignette judgements from 3,836 people in Feb-May 2017.

Table 1. Dimensions of benefit claimant vignettes (see Appendix F for full text)

¹ Most/least deserving refers to respondents’ average ratings.

Respondents were then asked whether each vignette ‘deserves to receive support from the Government while [he/she] is out of work?’, giving answers on a 0-10 scale. To test H1, I examined whether those in the fraud frame group judged the disability vignettes (but not the unemployment vignettes) more harshly than those in the control group.

More formally, I regress deservingness on the fraud framing vs. the control group using a simple OLS model (using cluster-robust OLS to account for the clustering of vignettes within respondents). I do not weight the data as this can increase bias in survey experiments (Mutz, 2011:114-123), nor do I include control variables as this is unnecessary and may decrease power (Mutz, 2011:124-6). In sensitivity analyses, however, I test if the results are robust to the inclusion of weights and sociodemographic controls (age, gender, children, marital status, qualifications, working status, and own benefit claims; see Appendix A).

Results

The average effect of the ‘benefits cheat’ story on perceived deservingness is shown in Table 2 below. This supports H1: reading the ‘benefits cheat’ story makes respondents judge a vignette claimant more harshly (−0.31, 95% CI −0.46 to −0.15). Table 2 also shows that this effect is found in both the UK and Norway, despite differing welfare/wider contexts, suggesting the effect is generalisable across high-income countries.

Table 2. Effect of mock stories on perceived deservingness of disability benefit claimants (on a 0-10 scale)

Weights/controls not included, although used in sensitivity analyses (see below/Appendix A).

To further test my hypothesised mechanism, I test whether these results are specific to disabled benefit claimants and do not carry over to non-disabled unemployed claimants (given that the story focused on the genuineness of disability, not unemployment). The results are shown in the final column of Table 2, which shows that there is no effect of the benefits cheat story on judgements of non-disabled unemployed vignettes (the difference between the impacts on disabled and unemployed vignettes is strongly significant, p<0.001). This confirms that the impact of the disability benefits cheat story is specific to hypothetical contact with disabled benefit claimants. In further sensitivity analyses (Appendix A), the findings are unchanged if I include controls/weights; or exclude those failing attention checks.

Overall, I find clear support for H1, although the effect is moderately small (0.3 on a 0-10 scale, ≈10% of the standard deviation of 3.06). However, it would be surprising to see a larger effect. The size of effect is not dissimilar – and is more consistent across measures/sensitivity analyses – to previous studies that have been interpreted as showing evidence of framing effects (e.g. Slothuus, 2007). Moreover, the newspaper-style story is muted in tone and makes no attempt to generalise ‘Andy Green’ to claimants in general (unlike real newspaper stories; Baumberg et al., 2012); respondents therefore have to connect a story about one person to a completely different person. Finally, while my vignettes were somewhat ambiguous, they were perhaps less so than everyday contact with peripheral networks. These suggest that my estimates may be lower bounds on real-world effects (though these must be weighed against the limitations of survey experiments), and I therefore turn to examining real-world effects directly.

Study 2: Real-life deservingness judgements

Study 1 provides convincing evidence that media frames can influence judgements of specific hypothetical claimants. However, this comes at the expense of ecological validity: like most social experiments, it invokes one frame without any competing ones, and removes temporal/other barriers between frame and judgement (Barabas and Jerit, 2010). To complement this, Study 2 examines how people’s real-world deservingness judgements relate to media consumption in the UK. It also goes further in specifically testing H2, which hypothesised that media effects would be found for peripheral network members (e.g. acquaintances) but not core network members (e.g. close family), because of the greater amount of ambiguity when judging peripheral network members.

Data and measures of real-world deservingness judgements

We use the same YouGov survey used in Study 1, but here focus only on the UK (as my measure of welfare media coverage is only available in the UK; see below). For each of four types of social relationship (close family, close friends, distant family, and neighbours), I asked respondents (full question wording is given in Appendix G):

• If they personally knew anyone with that relationship (e.g. close family) that they think has probably claimed disability benefits in the past year;

• If they knew any claimants with that relationship (e.g. close family) that they thought ‘are not genuinely sick or disabled’ (hereafter ‘non-genuine’ claimants).

Table 3 shows that nearly half (45.4%) of UK/Norway respondents reported knowing a disability benefit claimant per se – most commonly close family or close friends (17.0-22.6%), and to a lesser extent distant family and neighbours (7.0-7.1%). Yet when looking at non-genuineness, people were more likely to know a non-genuine neighbour than a non-genuine close family member (6.0% vs. 4.1%). This is similar to recent findings in the US (Fang and Huber, 2020).

Table 3. Perceived contact with a non-genuine disability benefit claimant

Data: UK-Norway YouGov survey 2017, n=3,961 (1,973 UK, 1,988 Norway). ¹ Third column is not column 2 divided by column 1; see text and Table 4 for details.

To test H2, we need a fair comparison of the prevalence of non-genuineness for different relationship types, which takes account of the potentially different size of each group. Our outcome variable is therefore reports of knowing non-genuine claimants only among respondents that had already said they knew a claimant of that type, as shown in Table 4 below. ⁶

Table 4. Coding of ‘knows non-genuine claimant’ as outcome variable

Using this outcome variable, we can see a clear pattern in the final column of Table 3, where claimants in core networks are less likely to be seen as non-genuine than those in peripheral relationships (30.2% of neighbours are seen as non-genuine vs. 13.6% among close family).

Measures of media use

To measure media framing of benefit claimants, I use a hand-coded content analysis of a 20% sample of all UK newspaper articles on benefits 1995-2011 (1,291 articles in total), kindly shared by Declan Gaffney (see Baumberg et al., 2012). While television is another pivotal media source and there has been an increasing use of online news sources, I focus on newspapers as in the UK they have long been particularly polarised around welfare (Page, 1984:40), and have therefore been the primary focus of concern (e.g. Baumberg et al., 2012; Reeves and de Vries, 2016). Gaffney’s measure of newspaper negativity encompasses any of the following themes:

• Fraud,

• ‘Shouldn’t be claiming’ (not fraud),

• Never worked/hasn’t worked for very long time,

• Large families,

• Bad parenting/anti-social behaviour,

• Claimants better off on benefits,

• Claimants better off than workers, and

• Compulsion of claimants.

The resulting ‘newspaper negativity’ score is merged into the survey data on the basis of the newspaper that respondents report reading most regularly (see Appendix B). Note that this is a measure of general newspaper negativity to welfare, rather than the disability-specific negativity of the benefits cheat newspaper story in Study 1.

The Gaffney measure reflects a decade-long average several years prior to my survey (with Reeves and de Vries, 2016 showing that short-run changes are possible). I therefore validate Gaffney’s measure in the present-day using my survey: after giving half of respondents the mock article on disability benefit fraud in Study 1, I asked them how often they had read similar newspaper articles, and compared this to the newspapers they regularly read. Across 8 major UK newspapers, newspaper negativity 1995-2011 correlates reassuringly strongly with 2017 reports of disability fraud articles (r=0.86; see Appendix B).

Analytical approach

We regressed ‘self-reported contact with a non-genuine claimant’ on newspaper negativity (treated as a continuous variable) alongside sociodemographic control variables (gender, age, marital status, presence of children in the household, education, working status, own benefits claims, and disability; see Appendix G) using standard logit models, allowing covariates to have different effects for each type of interpersonal contact. (In sensitivity analyses I further control for ideology and social trust). Note that I have 0-4 observations per person, depending on whether they reported knowing claimants within each relationship type, and I use cluster-robust standard errors to account for this clustering within individuals. To avoid possible misinterpretations, all results are presented as average marginal effects (Mood, 2010).

Results

H2 predicted that those who read negative newspapers would perceive greater non-genuineness among neighbours but not among close family/friends. The results are shown in Figure 2, and strongly bear out my hypothesis. There is no systematic relationship between newspaper negativity and perceptions of close friends or family, but a strong relationship for distant family and particularly neighbours (e.g. a 1% rise in newspaper negativity raises perceived non-genuineness among close friends by only 0.1% [95% CI −0.3 to 0.6%], but raises it among neighbours by 2.1% [95% CI 1.0 to 3.1%]; see Appendix H).

Figure 2. How perceived non-genuineness varies by contact type and newspaper negativity (UK).

In further analyses I also control for social trust and political ideology (which may confound the relationship between newspaper negativity and claimant judgements). These still show strong support for H2, only slightly attenuated from my main results (e.g. a 1% rise in newspaper negativity raises perceives perceived non-genuineness among close friends by <0.05% [95% CI −0.5 to 0.5%], but raises it among neighbours by 1.6% [95% CI 0.75 to 2.5%]; see Appendix A). These control variables enable us to compare the judgements of readers of different newspapers who are otherwise identical (in sociodemographics, social trust and ideology), but I should stress that this is cross-sectional observational research, and it is possible that there is both unobserved confounding or reverse causality. This reflects the different trade-offs of Study 1 and Study 2: the latter enhances ecological validity at the expense of confident causal inference.

Study 3: Comparing real-life judgements in the UK vs. Norway

Our final hypothesis was that Britons would be more judgemental of acquaintances vs. close family members than Norwegians, given differences in welfare discourses (H3). This uses the same YouGov data and real-life judgements as Study 2 (but unlike Study 2, now looking across both the UK and Norway). Again, I focus on perceived non-genuineness only among respondents that report knowing a claimant of that type per se (see discussion under Study 2 and below).

Our analytical approach here is simple: I regress perceptions of non-genuine claimants on respondents’ country using logit models, with weights but no controls. (In sensitivity analyses I further include the same controls as in Study 1). As before, I use cluster-robust standard errors to account for the clustering of responses within individuals, and present all results as average marginal effects.

Results

Contrary to expectations, Figure 3 shows no clear difference in perceptions of neighbours between the two countries (about 30% of neighbours who claim disability benefits are perceived to be non-genuine in both). Also against expectations, Norwegians perceived more non-genuineness among close family and friends, again shown in Figure 3. There is some evidence that the difference between perceptions of close family/friends vs. neighbours was slightly stronger in the UK than Norway (the specific prediction of H3) – but this was small and imprecisely estimated. Sensitivity analyses that include sociodemographic controls produce effectively identical results.

Figure 3. Perceived contact with a non-genuine disability benefit claimant: the UK vs. Norway across types of social contact.

This is not because my choice of case studies is incorrect – in my survey, Britons do indeed report seeing more disability fraud articles than Norwegians, albeit by less than expected (42.3% in the UK vs. 28.2% in Norway report seeing articles like the benefits cheat frame in Study 1 ‘very often’ or ‘quite often’, a difference of 14.2% [95% CI 9.7-18.7%]). Yet this does not necessarily mean that the more positive public discourse in Norway has no impact on perceptions of known disability benefit claimants (contra H3), because of three complexities.

Firstly, many respondents initially said they did not know a claimant of a particular relationship type per se, but then reported knowing a non-genuine claimant of that type (see Study 2/ Table 4/note 6). This was more common in the UK, and as these cases of perceived non-genuineness are excluded from my main analysis, this makes cross-national comparisons more difficult. It is possible when testing H3 to use a simpler outcome measure than Study 2, because we can assume that the size of different networks is similar in the two countries (i.e. that people know similar numbers of neighbours in the UK and Norway). If I therefore adopt a simpler outcome measure – reported non-genuineness per se – then I find that Norwegians were only slightly more likely than Britons to know a non-genuine neighbour (17.7% more likely [95% CI 3.7-31.8%]), but much more likely than Britons to know a genuine neighbour (80.7% more likely [95% CI 55.5-105.8%]). This provides some support for H3, but I must stress this is a post-hoc exploratory analysis, rather than my initial approach.

Secondly, at least 2-3 times as many people claim disability benefits in Norway compared to the UK (see Appendix D); in my survey 29.1% of Britons vs. 61.5% of Norwegians report knowing a disability benefit claimant. Assuming that people in Norway do not have greater morbidity, this means that people with less severe disabilities are more likely to be claiming disability benefits in Norway – hence the average severity of disability of claimants in Norway vs. the UK will be lower. Given that severity is linked to perceived deservingness (Geiger, 2021), this will ceteris paribus inflate the rate of perceived non-genuineness in Norway vs. the UK. This may explain why I see greater levels of perceived non-genuineness in Norway for all relationship types.

Third, differences in severity are unlikely to be the full explanation, because the meaning of ‘non-genuine’ appears to differ between the two countries. To explore this, I asked respondents, “Think of the non-genuine claimant that you know best. How could you tell that they were not a genuine claimant?” Norwegian responses were translated into English, and then all responses were coded to an inductively-derived coding frame (in the absence of ‘don’t know/refused’ options, ≈20% of responses in each country were un-code-able or blank). Results for code-able responses are shown in Table 5, showing three summary codes (see Appendix C for further details):

• ‘Able to do more than they claim’: the claimant was believed to be less sick/disabled than they claim. This referred to either general functioning (being healthy, having an active life); specific tasks they have been observed doing (e.g. walking, odd jobs/gardening, sport/partying/socialising, holidays, shopping); lying (e.g. not using a wheelchair/stick when they think they cannot be seen, or vaguer accusations of ‘malingering’); or the claimant admitting fraud/being well.

• ‘Lazy/bad attitude’: the claimant was believed to be lazy, or not trying hard enough to find work (or more rarely, not trying hard enough to get better). In Norway this occasionally included references to people being young, with the implication that efforts should be greater at their age.

• ‘Could work with support/adaptations’: the claimant was believed to be capable of doing some (but not all) work, or comments that they could be working without attaching any blame (sometimes explicitly saying that it is society’s fault that they are given inadequate support).

Table 5. Justifications for saying a known disability benefit claimant is not genuine

Categories are neither complete nor mutually exclusive. Source: authors’ coding of free-text responses from 371 Norwegian and 209 British respondents who explained how they could tell that a disability benefit claimant they knew was not genuine.

This shows a strikingly different pattern of justifications in the two countries. In the UK, accusations of non-genuineness were primarily because people were thought to be ‘faking it’ or exaggerating – over three-quarters (77.9%) of justifications were on the grounds that the person was not as sick/disabled as they claimed to be. While this was still relatively common in Norway, it was much less dominant than in the UK (41.1% of justifications). Instead, Norwegians were much more likely to justify their perceptions of non-genuineness on the grounds that the person was lazy (or at least, not trying hard enough to work), or that despite their genuine sickness/disability, they still had the ability to work (22.6% of justifications in Norway, but virtually absent in the UK).

I revise the theory in the light of these findings in the Concluding section. In the meantime, the evidence does not provide clear support for/against H3.

Conclusions

To the extent that welfare benefit claimants are seen as undeserving in the US/UK, it is often argued that the media are partly to blame. Yet this is seemingly inconsistent with the way US/UK publics primarily justify perceptions of undeservingness through their own contact with claimants, rather than through media reports. In this paper, I have tested the hypothesis that the media may still play a role by shaping how we interpret such contact – particularly contact with peripheral network ties, where outward deservingness cues are few and ambiguous, and where media-driven stereotypes may therefore play a role. I focus on disability benefits, which are not only the most commonly-claimed working-age benefits in high-income countries, but are also an ideal case study given the often ambiguous external signs of disability.

We tested this using three complementary approaches:

• Study 1 used a survey experiment to test if negative newspaper frames influence respondents’ deservingness judgements of a hypothetical person, using a UK-Norway survey of nearly 4,000 people and a wide range of vignettes. Confirming H1, I found that a disability benefits fraud frame did indeed make people judge disability vignettes more negatively.

• Study 2 used perceptions of non-genuine disability claims in real life and how this related to newspaper readership in the UK. This also supported H1: newspaper negativity was associated with a greater probability of viewing known claimants as non-genuine. This also supported H2: these associations were much stronger when judging neighbours, and there was effectively zero association of newspaper negativity with judgements of close family, where perceivers have high information and low ambiguity.

• Study 3 also used perceptions of non-genuine disability claims in real life, but focused on a comparison between the UK vs. Norway (media discourses being more hostile to claimants in the former). Contra H3, I did not find that Britons were more likely to judge neighbours to be non-genuine claimants than Norwegians. In further exploratory analyses, it seemed that this may be partly because Britons were more likely to make reporting errors; partly because the average severity of disability benefit claimants in Norway is noticeably lower; and partly because in Norway there was a different conception of what ‘non-genuine’ meant (see below).

These studies are complementary: survey experiments allow confident causal inferences but are weakly generalisable; whereas real-life associations are strongly generalisable but require more cautious causal inferences. Nevertheless, we must bear in mind the limitations that apply to each study – that framing experiments are some distance from everyday deservingness judgements; and that there is the possibility of unobserved confounding and reserve causation when looking at everyday deservingness judgements. Moreover, while YouGov opt-in panels have been widely used in academic research (e.g. Aarøe and Petersen, 2014; Jensen and Petersen, 2017; Kootstra and Roosma, 2018), and are preferable to the student samples in most previous framing experiments, there is always a risk that the samples are unrepresentative.

Theoretical revisions

Overall, the evidence mostly supports my theory that the media shapes how we interpret our interactions with disability benefit claimants. Nevertheless, H3 was not unequivocally supported, and the post-hoc analyses above suggest that the theory needs to be revised. Welfare state institutions not only shape media framings (Larsen and Dejgaard, 2013), but also the make-up of benefit claimants (extending Larsen, 2006 to consider disability). The lower threshold for disability benefits in Norway means that claimants are fundamentally different groups of people in Norway vs. the UK (indeed, disability itself is defined differently; Kapteyn et al., 2007; O’Brien, 2015).

There are two plausible explanations here. More simplistically, media discourses in Norway are more generous (and lead to more generous judgements under uncertainty), but are counterbalanced by the lower average severity of disability (which leads to harsher judgements). But a more complex interpretation is that the Norwegian system is underpinned by (and underpins) a fundamentally different framing: rather than focusing on whether disability is genuine vs. fraudulent, it focuses on partial work capacity that people have an obligation to make use of, and which the state has an obligation to facilitate (McKowen, 2020). Judgements under ambiguity are therefore not just about genuineness, but also whether worklessness among genuinely disabled people is due to individual effort vs structural constraints.

Going forward, I believe that it is essential to study how benefit claimants are judged under ambiguity, and how the media shapes this. However, these judgements are more complex than my original account allows. In comparative perspective, further research should therefore:

• Extend the research here using multiple media framings, multiple dimensions of judgements and multiple claimant types. I hypothesise that media frames will influence judgements of other claimant groups (e.g. control over unemployment is often unobserved), but multiple dimensions of deservingness will be involved – as they are even for disabled claimants;

• Test and develop the theory by studying mechanisms in more detail. This includes varying ambiguity more directly (e.g. following Petersen and Aarøe, 2013), but also qualitatively studying day-to-day social interactions, examining what external cues are given by claimants to which audiences, and how various cues are interpreted by audiences with different worldviews;

• Explore comparative differences further by studying the role of welfare state institutions in constructing media narratives and claimant groups (and their interactions) – while being sensitive to the many factors that influence media reporting (Baumberg et al., 2012:Ch4; McKendrick et al., 2008).

Implications

Our findings deepen the existing challenge for those seeking public support for more generous benefits policies. Such campaigners in Anglo-Saxon countries are already aware of the challenges of a hostile media context; but one implication of my results is that welfare attitudes will be sticky even in the face of new media frames, which may appear to conflict with ‘common sense’ grounded in everyday experience.

Progressive reformers can choose three different responses to this challenge (Baumberg, 2012). One is to pursue radical reform in moments of crisis, where worldviews rupture. A second is to use ‘sequencing’, where institutional reforms nudge discourses in a more progressive direction – for example, to gradually increase welfare universalism, which will later create wider constituencies of interest for increased generosity. However, more research is necessary to understand how people judge one another within different welfare regimes (see previous section). Finally, we should note that past radical reforms did not require a ‘golden age’ of pro-welfare attitudes (Hudson et al., 2016). Even if hostile welfare attitudes are buttressed by interpretations of ambiguous everyday experience, they are neither immutable nor unchallengeable.

Most of all, though, my theory rebuts a potential challenge to the impact of the media in framing welfare deservingness. When people say that their neighbours cheat benefits, this is not decisive evidence of widespread fraud – instead, these judgements are themselves part of the tangled web of media-fuelled stereotypes, political discourses and institutional design.