The main data source for the analysis, from which the outcome variable and covariates were drawn, was the Eurobarometer, a cross-national survey conducted annually and for 2010, including data on attitudes towards people with mental illness for approximately 26,800 individuals across European countries (22). The survey sample was drawn using multi-stage random probability sampling proportional to population size and density and is representative of the population aged 15 and above in each of the countries [see TNS Opinion and Social (23) for full details of survey methodology]. Interviews were carried out face to face at participant homes, in an appropriate national language (23). The sample for each country includes around 1,000 respondents aside from Luxembourg, Cyprus and Malta where numbers are around 500. This is because sampling is proportionate to population size and so smaller countries have fewer respondents (23). All Eurobarometer data was weighted using the EU27 population weight included in the dataset (22).

The survey used for this analysis represents the second in a special series about mental health across Europe with the first survey taking place in 2006; however, the 2010 Eurobarometer survey is the first and only to explore perceptions of people with mental illness as far as we are aware (22).

The following Eurobarometer survey question was selected as the outcome variable, in which respondents are asked to choose from one of the following two statements: •

You would find it difficult talking to someone with a significant mental health problem.

Answers were recoded into a categorical variable prior to analysis for ease of interpretation. A new variable was derived from the data with 0 = no problem talking to someone with a significant mental health problem and 1 = difficulty talking to someone with a significant mental health problem. “Don't know” responses were excluded from the analysis because it is not possible to interpret this response in relation to desire (or not) for social distance, totalling 9.96% of responses overall (please see Appendix I for a breakdown of responses by country).

Two key macro-economic factors were tested for their association with whether or not respondents described difficulty speaking to a person with a significant mental illness. GDP per capita represents the overall financial resources available to a country, whilst income inequality demonstrates how these resources are distributed in the population. Variables were selected from publicly available European data sources and standardised measurements were used to enable cross-national comparison. Data on the 2010 estimates for GDP per capita (Euros) and income inequality (2010 Gini co-efficient) were derived from Eurostat (25).

Individual demographic variables were sourced from the same Eurobarometer dataset as the outcome variable. Age and gender were included in the analyses since these factors are known to impact on individual attitudes towards people with mental illness (26). Individual perceptions of financial circumstances (measured as difficulty paying bills over the past 12 months) were also included based on previous research that suggests links between individual and contextual economic factors, and exclusionary attitudes towards people from other minority populations (27).

To explore both individual and contextual factors, a random effects logistic regression modelling approach was employed. There is ongoing debate about the acceptable number of contextual levels in multilevel models (28) but in an extensive review of cross-national analyses where multilevel modelling has been used, Bryan and Jenkins (29) recommend that data from upwards of 25 countries should be included to generate reliable estimates, a requirement met by the dataset used here.

All analyses were conducted in Stata version 15.1 (30). Scatter plots were used to explore the relationship between the outcome variable and each of the contextual variables. Individual level variables were added to the model first (model 1), followed by each contextual socioeconomic variable independently (models 2a and b) and a final, full model incorporating all individual and contextual variables (model 3). Likelihood ratio tests confirmed that random effects models were an improved fit compared to ordinary logistic regression once contextual variables were included and intraclass correlations were estimated for each of the random effects models. All effect estimates are reported as odds ratios.

One of the key limitations of this study is the age of the data because it is possible that there have been changes to attitudes since 2010, although more recent research evidence is mixed (4, 34, 41). Without updated data, it is not possible to know whether the findings are a reflection of this specific time and context; however, the study does provide valuable information on the type of economic circumstances that may be associated with differences in attitudes.

It is also worth noting that whilst the study has focused on those factors that could produce greater intolerance towards people with mental illnesses, across the European countries in the sample, only 24.73% of respondents reported that they would experience difficulty talking to a person with a mental illness. It should be highlighted then, that most people in Europe report having no difficulty talking to a person with a significant mental illness, although there are considerable differences by individual country.

Whilst encouraging, this also raises the issue of social desirability bias. The Eurobarometer is conducted via face to face interview for the majority of respondents (22) and so it is possible that responses have been affected by interviewees wanting to appear more tolerant. Henderson et al. (42) found that whilst questions relating to knowledge about mental illness are not associated with social desirability bias, questions relating to intended behaviour towards people with mental illnesses are, particularly in face to face interviews compared to online surveys. The outcome variable in this study represents an intended behaviour question and therefore social desirability bias could be implicated here. In this case, it is possible that the study findings understate the links between economic factors and attitudes towards people with mental illness.

The outcome variable used here does not directly reference prejudice in the question which may help to address some of the issues around social desirability bias but does produce difficulties in determining the exact meaning of responses. A person may be willing to engage with a person experiencing a significant mental illness, but may, for example, lack confidence, therefore this is not necessarily a measure of desire for social distance. More in-depth, qualitative methodological approaches may be needed to assist in understanding the nuances behind these responses, or alternatively the use of more a more detailed scale to understand differences. This applies also to the “don't know” responses in the sample.

Whilst we do not consider it likely that the association between higher levels of income inequality and increased stigmatization could be fully explained by racial/ethnic heterogeneity, our lack of inclusion of this characteristic could be viewed as a limitation of the analysis. Although racial/ethnic heterogeneity has been suggested as a potential confounder in the past by those sceptical of the impact of inequality on health and wellbeing, research has shown that the association between income inequality and health is independent of racial/ethnic heterogeneity in both international and US state comparisons (43). In addition, as income inequality may itself create the social and economic conditions that increase migration/ethnic heterogeneity, which in turn may be a potential cause of higher levels of downward prejudice and stigmatization, we view ethnic heterogeneity as possibly sitting on the causal pathway (i.e., a mediator, not a confounder).

Within a European context, analytical issues around ethnicity are perhaps even more complicated than in the north American context, where most studies adjust for African-American race and Hispanic ethnicity. In the UK, for example, some south Asian groups have high socioeconomic status and do better in terms of wellbeing measures, while others do worse than the White British average. White Eastern European migrants to the UK and White Roma populations can experience considerable disadvantage, so comparing white to non-white groups obscures important complexity.

The analytical treatment of ethnic heterogeneity in relation to income inequality and outcomes, therefore, is a complex matter, worthy of a further paper and beyond the scope of our current analysis. Further exploration of the role of migration status, ethnicity, language and other measures of identity would be a beneficial next step for future research.

Both individual demographic differences and contextual economic factors appear to be relevant in explaining the proportion of people who have difficulty talking to a person with a significant mental health problem across European countries. The wider socioeconomic climate is important because it impacts on all people living in a country. Evidence suggests, for example, that income inequality has detrimental effects on health and social outcomes for all residents in more unequal countries (10). Where the wider socioeconomic climate in a country incorporates greater income inequality or lower GDP per capita the population may be more likely to report difficulty speaking to a person with a significant mental illness, though further research is needed to confirm these findings.

Individual perceptions of financial circumstances were consistently and independently associated with difficulty speaking to a person with a significant mental illness which is in keeping with findings relating to attitudes towards minority groups more broadly. It is feasible that the association between individual financial circumstances and attitudes towards mental illness could be a reflection of the wider socioeconomic climate and this would certainly fit with existing research that suggests income inequality is related to more negative outgroup attitudes at the individual level (9, 11, 36). In more unequal countries, the effects of status anxiety are stronger and this impacts on people at all income levels (44). Status anxiety produces more strained interpersonal relationships because it increases concerns about social evaluation, in turn meaning that people are less likely to associate with others who are viewed as a threat to their social status (12). Goffman (45) conceptualises mental illness as a “discredited” characteristic meaning that it is perceived as conferring an inferior social status. It is possible that in socioeconomic climates where status anxiety means that social position carries more weight, people are more likely to disassociate themselves from those perceived as having a discredited characteristic.

This study contributes to the emerging field in mental health research that focuses on the underlying societal level drivers of stigma. These factors are key to improving understanding of the experiences of individuals with mental illnesses since a country may set out intentions for progressive mental health policies in terms of care and treatment, whilst at the same time making economic decisions that perpetuate disadvantage (46) and place strain on intergroup relations. Further research, using more recent data and exploring the underlying mechanisms behind the associations between both micro and macro-economic factors and public attitudes would be beneficial.

Whilst beyond the scope of the current analysis, future research may wish to focus on additional country-level factors that could be associated with GDP per capita and income inequality, and which may impact on attitudes towards people with mental health problems. In particular, differences in healthcare systems, the political, ethnic and religious composition of each country and the prevalence of mental health problems across different populations could each add important detail to the findings. Similarly at the individual level, additional context in relation to individual demographic variables may further strengthen the analyses, for example, ethnicity, religion and educational level. Although not possible with this particular dataset, it would also be interesting to know whether the perceived socioeconomic status of the person experiencing mental illness had any impact on public attitudes.

Significant social, economic and political change seems likely in the United Kingdom and potentially across Europe in the coming years and this may potentially impact on people with mental health conditions, producing further social exclusion. Reducing existing stigma, whilst ensuring that people with mental health conditions do not become further marginalised in this context, should be key priorities for mental health practitioners and policy makers alike. Targeting income inequality should be the first step to improving intergroup relations, and therefore potentially attitudes towards people experiencing mental illness, ultimately reducing the harmful effects of stigma.