Limitations

Before discussing the results of our review some limitations should be addressed. First, our review only included papers written in English or other European languages, which may have resulted in an under-representation of countries where also other languages are in use for disseminating research results. Second, in order to guarantee a minimum of quality we only included peer-reviewed papers and excluded grey literature, research reports available on the Internet and doctoral theses, which also may have biased the sources on which our review is based.

Geographical distribution

Although attitude research has expanded in recent years into other parts of the world studies originating from western countries still predominate. Looking at economic indicators, only 14.6% of all studies included data from at least one low or middle income (LAMI) country (according to World Bank criteria), among them only 0.8% from low income economies, although these countries account for 80% of the world population. Here, psychiatric attitude research mirrors a general problem, the heavy bias of medical and psychiatric research towards high-income countries. It has been pointed out that on the one hand, in LAMI countries with low funding for research in general and few mental health researchers in particular, these restricted resources translate into lower research output (Sharan et al. Reference Sharan, Gallo, Gureje, Lamberte, Mari, Mazzotti, Patel, Swartz, Olifson, Levav, de Francisco and Saxena2009), while on the other hand medical journals could be biased against research from poorer countries (Patel & Kim, Reference Patel and Kim2007). Beyond, attitude research in particular is much more challenging in less affluent than in wealthier countries. Absence of professional market research companies, unequal distribution of landline or mobile telephones and stark differences in urban and rural infrastructure make accomplishing representative population samples highly difficult. These methodological challenges contrast with the high public health relevance of attitude research in countries where barriers to help-seeking are even higher because of lower availability of mental health care (Sharan et al. Reference Sharan, Gallo, Gureje, Lamberte, Mari, Mazzotti, Patel, Swartz, Olifson, Levav, de Francisco and Saxena2009; Becker & Kleinman, Reference Becker and Kleinman2013).

Cross-cultural comparison

Studies comparing public beliefs and attitudes about mental illness across different cultures still are scarce. This applies particularly to the comparison between western and non-western cultures. Typically, in stigma research instruments have been developed and validated in a western culture (mostly in the USA) with people of western European descent and then translated with little or no modification into another cultural context. The strength of this ‘universal’ approach is the generalisability and comparability of results across contexts. However, one inadvertent consequence is the elimination of stigma's culture-specific aspects (Yang et al. Reference Yang, Thornicroft, Alvarado, Vega and Link2014). Only recently, Yang et al. (Reference Yang, Kleinman, Link, Phelan, Lee and Good2007) have proposed a new theory, which allows accounting for how cultural influences impact upon stigma specifically. The concept of ‘moral experience’ or ‘what is most at stake for actors in a local social world’ provides a new interpretative framework for the behaviours of both the stigmatiser and the stigmatised person. The authors maintain that culture affects stigma by threatening a person's capacity to participate in the activities that determine ‘what matters most’ within a cultural context. For example, because the perpetuation of one's lineage is something, which is valued most within Chinese cultures, stigma is seen to most powerfully attack one's ability to extend one's lineage. So far, there exists one single population-based study where this new conceptual approach has been used (Yang et al. Reference Yang, Purdie-Vaughns, Kotabe, Link, Saw, Wong and Phelan2013).

Instruments

Apart from cultural validity, a major problem of population-based stigma research is that most studies use ad hoc developed instruments that have not been psychometrically validated. Moreover, among those instruments with proven psychometric properties, there are some, which have been developed years ago and may therefore not fully reflect the current situation. Just to give an example: In several items of Link's frequently used Devaluation-Discrimination Scale the perception of the stigma experienced by someone who had been in a ‘mental hospital’ is assessed – an institution, which in many western countries nowadays plays a rather marginal role in psychiatric care. The assessment of implicit stigma, a novel approach to measure stigmatising reactions, which is becoming increasingly popular in social-psychological experiments (for example, Rüsch et al. Reference Rüsch, Todd, Bodenhausen and Corrigan2010), has so far not been introduced into population-based research.

Type of mental health problem studied

Despite the spectrum of disorders under study having increased, the main focus of research is still on schizophrenia and depression. Reviews on differences in attitudes towards different disorders show that, for example, alcoholism seems to be stigmatised for very different reasons than schizophrenia (Schomerus et al. Reference Schomerus, Lucht, Holzinger, Matschinger, Carta and Angermeyer2011). This leads to another conceptual problem that should be addressed in future attitude research: Other stigmatised conditions like obesity or poverty often co-occur with mental disorder in individuals who then experience multiple stigmata. The concept of ‘layered stigma’ has been used in studies describing the multiple stigma experiences of persons with HIV (Henkel et al. Reference Henkel, Brown and Kalchman2008). Layered stigma could also be a useful concept when investigating mental illness stigma, but so far, attempts to bring together research on attitudes related to different medical and social conditions is only beginning (for example, Corrigan, Reference Corrigan2014).

Studies of time trends in attitudes and beliefs

The number of studies investigating time trends in attitudes and beliefs has significantly increased. Practically all of them are based on repeated cross-sectional assessment of attitudes with independent samples from the same population. This approach is best suited to detect variations of cultural conceptions of mental illness over time. While also providing insights into changes of people's attitudes on an individual level, panel studies would have to struggle with lack of representativeness of the follow-up assessment due to the huge attrition rate to be expected over time periods of 10 or more years, and, therefore, be less suitable for studying changes at the collective level. Using the same sampling procedure, interview mode and instruments within the same population at different time points has to be considered the gold-standard of time trend analysis. Data from identical surveys at three or more time-points will enable novel statistical approaches such as age-period-cohort (APC) analyses. APC analyses distinguish time-period effects from cohort and age effects and help thus explain whether any change in population attitudes is due to demographic transformation or a general shift of attitudes irrespective of age or cohort effects (Schomerus et al. Reference Schomerus, Van der Auwera, Matschinger, Baumeister and Angermeyer2015).

Influence of societal context on individual attitudes

Research has so far almost exclusively focused on individual beliefs and attitudes. The societal context that may influence individual perceptions and behavioural intentions has rarely attracted attention. Mojtabai (Reference Mojtabai2010) was the first examining the interplay between both individual and collective stigmatizing attitudes and how they are associated with willingness to seek professional help. Using multi-level analysis, Jang et al. (Reference Jang, Lim, Oh, Lee, Kim and Lee2012) were able to show that the desire for social distance from people with mental illness is not only determined by individual factors but also influenced by contextual factors. Adjusting for the individual's perceptions and characteristics, Richardson et al. (Reference Richardson, Morgenstern, Crider and Gonzalez2013), also using multi-level modelling, found collective perception of treatment effectiveness to be positively associated with the use of mental health services while collective perceptions of mental-illness stigma were not. Evans-Lacko et al. (Reference Evans-Lacko, Brohan, Mojtabai and Thornicroft2012) used multi-level analysis to show that self-stigma of individual patients is related to the degree of public stigma in a country. In view of potential implications for anti-stigma and awareness-raising interventions further research efforts in this direction are needed.

Attitudes v. behaviour

The overwhelming focus on attitudes may be looked upon as a limitation of the research reviewed, since attitudes allow predicting behaviour with only less than ideal accuracy. However, rather than using them as proxy for individual behaviours, public attitudes should also be regarded as a relevant phenomenon on their own. At a collective level, they reflect cultural conceptions of mental illness, forming a cultural context that influences the way we think about mental illness and the people who have them. As Link et al. (Reference Link, Angermeyer, Phelan, Thornicroft, Szmukler, Mueser and Drake2011) have pointed out ‘as a context this cultural conception becomes an external reality, something that individuals must take into account when they make decisions and enact behaviour’ (p. 255). Studies on individual stigma experiences yield manifold subtle forms of behaviour that are expressions of stigmatising attitudes but pose sincere problems to measurement (Angermeyer et al. Reference Angermeyer, Beck, Dietrich and Holzinger2004). Altered everyday interactions like not being completely trusted at work or receiving less phone calls by friends, for example, defy scientific evaluation. Moreover, the most frequently reported effect of stigma in a study among persons with schizophrenia was anticipation of discrimination and subsequent avoidance of critical situations (Thornicroft et al. Reference Thornicroft, Brohan, Rose, Sartorius and Leese2009). Social withdrawal because of fear of discrimination is discrimination without discriminatory behaviour, solely based on public attitudes and their perception. So attitudes remain a highly relevant topic of population-based studies. To measure and explain relevant discriminatory behaviour, studies in specific environments (work, school, health care, social media) or studies examining structural discrimination seem more useful than population studies.

Conceptual framework

Papers presenting descriptive results still predominate and only a minority of studies is theory driven. This is deplorable because pure description of reality does little to aid our understanding of attitudes and does not explore possibilities to change them. A good example is how testing specific theories has advanced research into the role of causal attributions for stigma. The dominant biological presentation of mental illness in science and media raised the question whether this shift in illness conceptualisation would help to reduce stigma. Here, conflicting theories predicted different effects: while attribution theory stressed the importance of blame and the potential of biogenetic causal models of mental illness to relieve it (Corrigan, Reference Corrigan2000), genetic essentialism predicted a deepening divide between persons with and without mental illness, should mental illness be explained in biogenetic terms (Phelan, Reference Phelan2005). Comparative testing of both theories revealed that in severe mental illness, essentialist thinking is much more relevant for stigma than attribution of blame (Schomerus et al. Reference Schomerus, Matschinger and Angermeyer2014).

Study design

Most studies are correlational studies not allowing causal inferences. Only a small number of studies used quasi-experimental designs, mainly for examining the influence of the presence of a community mental health facility on the attitudes of those living in the neighbourhood (Veltro et al. Reference Veltro, Raimondo, Porzio, Nugnes and Ciampone2005; Stadler, Reference Stadler2010) or for the evaluation of anti-stigma or awareness campaigns at a population level (Jorm et al. Reference Jorm, Christensen and Griffiths2005a , Reference Jorm, Christensen and Griffiths2006; Wright et al. Reference Wright, McGorry, Harris, Jorm and Pennell2006; Baumann et al. Reference Baumann, Zäske, Decker, Klosterkotter, Maier, Möller and Gaebel2007; Gaebel et al. Reference Gaebel, Zäske, Baumann, Klosterkötter, Maier, Decker and Möller2008; Dietrich et al. Reference Dietrich, Mergl, Freudenberg, Althaus and Hegerl2010). Although a population-based randomised controlled trial (RCT) to examine the effect of an information program on mental health literacy and help-seeking (Jorm et al. Reference Jorm, Griffiths, Christensen, Korten, Parslow and Rodgers2003) had already been conducted prior to the time covered by this review, no such RCT has been conducted since 2005.

As early as in 1963 Phillips (Reference Phillips1963) carried out a first survey experiment, but experimental studies sensu stricto only started again with Link's seminal study on the importance of labelling for mental illness stigma (Link et al. Reference Link, Cullen, Frank and Wozniak1987). Recent experimental studies manipulated for example causal explanations (Phelan, Reference Phelan2005), contextual factors of a disorder (Holzinger et al. Reference Holzinger, Matschinger, Schomerus, Carta and Angermeyer2011), or tested potentially adverse effects of different news media messages (McGinty et al. Reference McGinty, Webster and Barry2013). The thematic diversity among these few experimental studies demonstrates how widely experimental study designs may be used. Conducting online surveys has become increasingly convenient, facilitating complex experimental designs in large samples. The share of population-based experimental online surveys in the literature will thus likely grow.

Interview mode

The diversity of interview modes found in this review reflects the fact that each method carries specific advantages and disadvantages. Face-to-face interviews offer the opportunity that the interviewer can give direct support to the task performance of the respondent. They are therefore particularly suitable for longer interviews with more complex tasks (Holbrook et al. Reference Holbrook, Green and Krosnick2003). On the other hand, the presence of an interviewer may systematically influence the answers of the respondent resulting in interviewer bias, with social desirability bias being the best known example. In addition, as differences between interviewers in the way they do their job are not easy to control interviewer variance may pose a problem (Loosveldt, Reference Loosveldt, de Leeuw, Hox and Dillman2008). Face-to-face interviews are also the most expensive option.

Although face-to-face has remained the most frequently used interview mode, its share has decreased compared to the previous review, while that of telephone interviews has increased. An explanation for this change may be that telephone surveys, particularly those using Random Digit Dial samples, offer a more cost-efficient method for gaining access to the general population. Another advantage is that due to the centralised administration of Computer Assisted Telephone Interview (CATI) systems, complete control over all aspects of the interviewing process is possible (Steeh, Reference Steeh, de Leeuw, Hox and Dillman2008). However, the growing use of mobile phones (and declining availability of landline phones) makes it increasingly difficult to achieve representative samples in telephone surveys (Holborn et al. Reference Holborn, Reavley and Jorm2012). To overcome this problem, in some recent surveys the sample was contacted by random-digit dialling of both landlines and mobile telephones (for example, Reavley & Jorm, Reference Reavley and Jorm2012). However, this mixed design raises new methodological problems, since households are the sample unit for a landline telephone frame, whereas the individual subscriber is the sample unit for a mobile number frame (Steeh, Reference Steeh, de Leeuw, Hox and Dillman2008). Apart from the coverage problem, the principal difficulty has been maintaining adequate response rates as participants tend to object to long telephone interviews and feel more annoyed and disturbed by this approach (Holbrook et al. Reference Holbrook, Green and Krosnick2003). Systematic trend studies show response rates dropping significantly when the interview mode changes from face-to-face to random-digit dialling (Steeh, Reference Steeh1981). Like face-to face interviews, telephone surveys are considered vulnerable to social desirability bias (Henderson et al. Reference Henderson, Evans-Lacko, Flach and Thornicroft2012).

Online surveys are a recent development and their share will likely increase due to the growing availability of online access and low costs of surveys. They have the advantage of eliminating unwanted interviewer effects and providing more privacy when answering sensitive questions (Henderson et al. Reference Henderson, Evans-Lacko, Flach and Thornicroft2012). However, recruiting a sample representative of the general population for online studies is methodologically challenging. Of 14 online studies included in this review, seven used a ‘gold standard’ sampling method, drawing randomised samples from a large nationally representative panel that was recruited offline. Other studies relied on online panels that were not representative of the general population. Methods to approximate a representative sample were quota sampling or weighing data according to socio-demographic variables. Nonresponse is another serious problem for Internet surveys, which is higher than in comparable telephone surveys (de Leeuw & Hox, Reference de Leeuw, Hox, de Leeuw, Hox and Dillman2008).

In sum, the various interview modes carry specific benefits and risks, which, in consequence, may lead to differences in effective study samples and/or responses to survey questions. Direct comparison between surveys using different interview modes appears therefore highly problematic as differences may be the result of mode change rather than any actual difference in attitudes of interest (Dillman & Christian, Reference Dillman and Christian2003).

Use of case-vignettes

Case-vignettes were first used in 1955 by Reference StarStar in a national survey on attitudes of the American public towards mental illness. They have been in use for many years until, in 1990, Angermeyer & Matschinger (Reference Angermeyer and Matschinger1995) introduced for the first time into population studies vignettes utilising explicit criteria for diagnosing mental disorders derived from DSM-III-R (and subsequently DSM-IV). Since then, case-vignettes have enjoyed growing popularity. Over the last 10 years, in about half of all articles on public attitudes the vignette technique has been employed. According to Yang et al. (Reference Yang, Link, Phelan, Arboleda-Florez and Sartorius2008) there are two major reasons why vignettes have gained such a prominent position: first, vignettes present a more elaborate stimulus to respondents than simply asking about ‘mental illness’ or ‘mentally ill people’. Second, they can also be administered via random assignment, which brings the power of the experimental method to hypothesis testing – a possibility which, as indicated above, so far has only rarely been used in population-based studies (for example, Phelan, Reference Phelan2005; Holzinger et al. Reference Holzinger, Matschinger, Schomerus, Carta and Angermeyer2011).

Besides these strengths also some limitations of the vignette methodology have to be mentioned: supposedly equivalent vignettes illustrating the same disorder may yield different responses because of variation in number and type of symptoms described (Sai & Furnham, Reference Sai and Furnham2013; Schlier et al. Reference Schlier, Schmick and Lincoln2014). By their very nature vignettes tend to be highly specific and situated, questioning whether responses are transferable to different persons or situations. Moreover, the focus on paradigmatic cases does not represent the true diagnostic and demographic heterogeneity of a particular mental illness. The utilisation of paradigmatic cases may in fact be assessing the possession of stereotypical information among respondents, as much as their mental health ‘knowledge’ (Aldersey et al. Reference Aldersey, Huynh and Whitley2016). With rare exceptions (for example, Phelan, Reference Phelan2005; von dem Knesebeck et al. Reference von dem Knesebeck, Mnich, Daubmann, Wegscheider, Angermeyer, Lambert, Karow, Härter and Kofahl2013), population-based studies used unlabelled vignettes depicting symptoms of the disorder of interest. This may have clouded potential differences between reactions to unusual behaviour and reactions to psychiatric illness (Angermeyer et al. Reference Angermeyer, Daubmann, Wegscheider, Mnich, Schomerus and Knesebeck2015). In view of these problems there is a pressing need for more systematic research on this methodology.

Future directions of population-based attitude research in psychiatry

Our review shows that there is an avant-garde in attitude research using both well-founded and innovative survey methods to explore the depth of population attitudes towards mental illness, enabling us to exemplify the current gold-standard of population-based attitude research. However, dissemination of these methods is by far incomplete, and many of the shortcomings observed in 2006 persist until today. Additionally, there are challenges ahead: the outlined difficulties to achieve representative population samples particularly with telephone surveys will likely increase and it remains to be seen to what extent carefully selected representative online-panels can be used as a methodologically sound replacement. With regard to its methodology, attitude research needs to show both more rigor and sustain a pioneering spirit.