Mind Over Matter: the Social Representations of Placebos and Their Effects

(1)

Faculty of Social and Behavioural Sciences

Mind Over Matter: the Social

Representations of Placebos and Their

Effects

Lea Lösch

Amsterdam, 2

nd

of July 2020

Student number: 12332607 Contact: lea.losch@student.uva.nl First supervisor: Dr. Gerben Moerman Second supervisor: Dr. Patrick Brown University of Amsterdam

(2)

Table of contents I

Mind over matter: The social representations of placebos and their effects

Abstract

Despite the repeatedly proven potential of the placebo effect to improve a patient’s health condition, placebos generally carry a rather negative connotation as something fake and not legit. This thesis investigated those public perceptions of placebos. Placebo research, which goes beyond the medical context in general, and specifically investigates public perceptions of placebos, is scarce yet crucial. If the potential of the placebo effect is ever to be used more widely and consciously in everyday medical practice, the perceptions of the public are critical to its success. Public perception was mapped by means of the social representation theory. Qualitative text analysis methods were combined with structural topic modelling to examine the social representations of placebos in German news articles between 2007 and 2019. The study revealed that different, partly contradictory social representations of placebos coexist across a variety of contexts in which the term is used. Although an outdated definition of a “sham drug” prevailed in the articles, there was a fairly sophisticated understanding of the functional mechanisms underlying the placebo effect. If a broader definition was used, medical practice was more often considered a field of application for placebos. Negative connotations slightly predominated, which was partly due to the metaphoric use of the term in other socio-political contexts. Overall, by employing a methodological approach that is novel to placebo research as well as the study of social representations this paper provided insights into the public perceptions of placebos.

1 Introduction

You can now buy a package of 45 placebo pills for $19.95 from Amazon.com (Vermeulen, 2020). In medicine, placebos are generally considered to be drugs and treatments without pharmacological relevant properties (Humphrey, 2002). But why would anyone nevertheless

(4)

1 Introduction 2

be interested in buying placebo pills? Because placebo research carried out over the last two decades has repeatedly proven that they can lead to an improvement in the patient’s health condition. The efficacy has already been demonstrated for e.g. the treatment of (chronic) pain (Finniss et al., 2010), different kinds of inflammation, insomnia (Bélanger et al., 2007), and depression (Kirsch et al., 2008). Various medical professionals reported that they had already administered some kind of placebo in medical practice, e.g. a saline injection to calm the patient (Linde et al. 2018).

Despite this repeatedly proven potential for improved patient care, the term “placebo” seems to have a rather negative connotation. It is associated with being something “inactive”, “non-specific”, “unethical” (Foot & Ridge, 2012); a kind of sham used in a situation where there is – according to the evidence-based medical model – no “real” scientific way of treating the patient’s problem (Justman, 2011). This thesis thus aims to investigate these public understandings of placebos and their effects more closely.

A large part of placebo research from various disciplines has to date mainly focused on explaining the mechanisms of how the placebo effect works. Psychology has brought attention to classical conditioning, expectancy theory and positive emotions. Anthropology has highlighted the performance of rituals and the cultural distinctiveness of symptoms and healing procedures. Sociology has found explanations in the doctor-patient relationship and the “meaning response” (Moerman & Jonas, 2002). This research is extremely valuable, it remains however rather narrowly focused as it attempts to explain the placebo effect always within the medical setting.

In this thesis, I therefore look at placebos from a societal perspective that goes beyond the direct medical context. Specifically, I will investigate the perceptions of placebos held by the general public. Knowledge about these is currently underdeveloped (Hardman et al., 2019). However, it is crucial, as we need to understand what people (doctors and patients) consciously and unconsciously bring to the healing encounter (Foot & Ridge, 2012). These perceptions are

(5)

likely to influence the utilization of placebos and their potential in medical practice. A greater understanding of public perceptions of placebos will contribute to a better comprehension of their current and future application and possible obstacles in medical care.

I will draw on the theory of social representations (SRs), first developed by the social psychologist Serge Moscovici (1961) to theorize the construction and dynamics of common-sense knowledge. I will employ SR theory as a theoretical framework to depict the phenomenon of public perception (Bauer & Gaskell, 2008). The SRs of placebos – as, for example, “fraud” or “great opportunity” – determine which understandings of placebos become socially accepted as “reality”. They are thus likely to play a constitutive role in the future use of the therapeutic potential of placebos in medical practice.

I aim to answer the following question: What are the social representations of placebos and how might they shape the application of placebos in medical practice? The research project focuses on the intersubjective construction of placebos and the assignment of value and meaning to it. The broader objectives are:

▪ To identify the discourses and public understandings of placebos and their effects, and

▪ to understand how these public perceptions might enable or inhibit potential types of clinical application of placebos.

Combining qualitative text analysis with structural topic modelling, an automated text analysis method, I will examine the SRs of placebos in German news articles from 2007 to 2019. Through this inductive approach to naturalistic data, I follow a call raised by researchers in the field (e.g. Hardman et al., 2018) for the need to adopt more naturalistic and contextual research approaches, given the contextual nature of the placebo phenomenon. I will demonstrate the fruitful combination of the chosen methods to study SRs in a comprehensive, yet detailed manner. I will further show why this highly unexplored method of studying SRs is a very suitable and promising one.

(6)

2 Theory 4

The paper begins by providing background information on the conceptual history of the term placebo itself, the nature and findings of placebo research, and by critically discussing the most recent attempt of a discursive exploration of placebos. The following section explains the SR theory. Subsequently, the research design will illuminate how the chosen methods build on each other and the purpose of this interplay. After explaining the conceptualization and implementation of the qualitative analysis methods and the structural topic model, the results will be presented. Finally, the findings and in particular the possible implications of the detected SRs of placebos will be discussed.

2 Theory

2.1 Literature review on placebos and their effects

The extensive placebo research to date provides compelling evidence that the administration of placebos can be enormously effective and beneficial for patients. This has already been demonstrated in the treatment of pain (e.g. Amanzio & Benedetti, 1999), depression (Leuchter et al., 2002), Parkinson’s disease (De la Fuente-Fernández et al., 2001), cardiovascular disorders (Pollo et al., 2003), gastrointestinal dysfunctions (Kaptchuk et al., 2008) as well as autoimmune diseases (Wendt et al., 2014)1. Enck and colleagues (2017) argue that the placebo effect manifests itself not only in the context of placebo treatment but is an immanent component of every medical intervention.

Empirical studies with medical professionals have revealed that a considerable number of doctors and nurses make use of some kind of placebo in medical practice, e.g. a saline injection to calm the patient. Besides inducing placebo effects, medical professionals employed placebos, for example, when other treatment options were exhausted or to meet patient expectations (Fässler et al., 2010). Linde et al. (2018) found in a systematic review and meta-analysis of available surveys that 15% to 89% of general practitioners prescribed placebos at least once a

(7)

month. The great variability is to a great extent caused by the different ways in which the respective studies define placebos.

The placebo concept has a long history of definitions, but there continues to be a lack of conceptual clarity. In the following, this history will be presented in order to understand how the term and possibly its public perception were coined. The term “placebo” is originally derived from Latin and means “I will please”. Interestingly, the word has already been used for centuries e.g. in a religious context (psalm 116, 9th verse) but also in common language. For instance, the metaphor “to sing a placebo” in the XIV century described the behaviour of someone who “seeks to please” and the proverb “no penny, no placebo” in the 16th century meant as much as: “without money you will not receive service or goods” (Wilkinson, 2003).

In the medical context, placebos were first defined in 1785 in Motherby’s medical dictionary as “a commonplace method or medicine”. However, it was not until the 20th century that placebos were used as a methodological tool in clinical trials, which significantly altered the meaning of the term. Placebos became closely associated with the non-active intervention in a randomized controlled trial (RCT), against which effective treatments were tested (Foot & Ridge, 2012). This established a narrow understanding of placebos as “inert” and “inactive”, which gradually replaced alternative perceptions such as “the only single action which all drugs have in common and, in some instances, it is the only useful action which medication can exert” (Modell, 1958). Towards the end of the century, however, a growing number of voices were raised in favour of expanding the definition of placebos again beyond their procedural role as a control in medical research. Shapiro and Shapiro (1997) pointed towards the bigger picture of placebos and called for research into “the nonspecific, as yet mystifying, but very powerful therapeutic potential of the placebo effect“ (p. 237). As Price et al. (2008) and Foot and Ridge (2012) observed, there has been a shift in the conceptualization of placebos from an “inert” substance to a definition that recognizes a more “active” and complicated role in healing. The new evolving understanding stimulated a range of new research agendas exploring the placebo

(8)

2 Theory 6

effect, which in turn inevitably led to a multiplication and fragmentation of definitions of placebos, e.g. across disciplines.

Disciplines ranging from biomedicine to social and behavioural science and the humanities have contributed to our rich, multifaceted understanding of the underlying mechanisms of the placebo effect. They each stressed different aspects, such as classical conditioning (Montgomery & Kirsch, 1997), expectancy (Pollo et al., 2001), positive emotions (Flaten et al., 2011), evolutionary causes (Humphrey, 2002), neurobiological processes (Benedetti et al., 2005), the performance of rituals (Ostenfeld-Rosenthal, 2012), the doctor-patient relationship (Blasini et al., 2018) and the “meaning response”, referring to the patients’ response to the meaning they attribute to their treatment (Moerman & Jonas, 2002). These various explanations do not necessarily contradict each other but rather shed light on different aspects of the placebo phenomenon. Consequently, some theorists have moved away from unitary placebo theories and have begun to regard the placebo effect as a spectrum of factors (Benedetti, 2009; Kaptchuk, 2008). Despite its interdisciplinarity, placebo research is still largely confined within the medical setting and its methods, such as RCTs (Crum & Philips, 2015). By contrast, societal dimensions have been underrepresented in placebo research so far (Enck et al., 2017). Considering the potential and widespread use of placebos, research that provides insights into societal aspects, such as public perceptions of placebos, but also e.g. ethical, legal and economic factors, is reasonable and necessary.

In the struggle over the placebo definition, Howick (2017) offers one of the most comprehensive attempts to develop a universal, cross-disciplinary definition. Building on Grünbaum (1986), he defines a treatment process as placebo when none of the characteristic treatment factors are effective (remedial or harmful) in patients for the target disease. The placebo effect is either “(a) one produced by the incidental features of some treatment (even when the treatment as a whole is a nonplacebo), or (b) any effect of a generic placebo” (Howick, 2017, p.1392). These definitions do not presume, in contrast to former ones, a distinction

(9)

between mind and body. Howick (2017) also avoids the frequently debated, logical paradox, that is inherent in declaring a substance as “inert” and then going on to describe its effects (Bishop et al., 2014). He creates a contextual definition, with reference to the patient, the condition and the therapy. Additionally, he acknowledges that placebo effects are not restricted to placebo treatments, but can modify the outcome of any treatment.

The sophisticated, academic definition of placebos contrasts with how those most directly involved in their use, namely medical professionals and patients define them. Hardman, Geraghty and Lewith (2018) conclude based on a systematic review that the majority of healthcare professionals and patients define placebos as “inert” substances which can have a positive or negative effect on a person. Only very few healthcare professionals referred rather to a process-oriented definition of placebos. According to this perspective, various factors in the process of any form of treatment – not necessarily a pill – can lead to an effect. The latter conceptualization is more in line with Howick (2017) and modern scientific placebo theoretical paradigms, focusing on the psychosocial context of an intervention and its contribution to treatment efficacy (Kisaalita & Robinson, 2012). The majority, however, makes use of an outdated placebo definition as a ‘pharmacologically inert substance (…) having a psychological effect’ (Beecher, 1955, p. 1602), which has long been rejected by placebo researchers as incoherent and untenable. Hardman et al.’s (2018) review suggests that there is a considerable disconnection between modern scientific placebo definitions and how healthcare professionals and patients define them. This can lead to confusion and uncertainty. Moreover, the increasingly nuanced understanding of placebos might be of limited clinical utility. One of the authors’ conclusions is accordingly: “Given the contextual nature of the placebo phenomenon (...) we promote more naturalistic and contextual research approaches, such as ethnography, which are currently underrepresented in placebo studies research.” (Hardman et al., 2018, p. 20). Such an approach might result in a more fruitful relationship between placebo research and clinical practice (Hardman et al., 2019). This demand for more naturalistic and contextual research

(10)

2 Theory 8

approaches is met in this paper by examining naturalistic data with an inductive approach that refrains from predetermined categories.

The – to my knowledge – only other discursive exploration of placebos has been undertaken by Hardman, Geraghty, Howick, Roberts and Bishop (2019). They conducted a discursive analysis of 930 online comments to six news articles reporting on a placebo study in 2013. Hardman et al. (2019) identify two main discursive constructs of clinical placebos. The first of the “placebo pill” regards a placebo as an inert substance. The second, less-prevalent counter-discursive construct perceives placebos as the whole clinical “treatment process” and frames them as “potentially viable within modern evidence-based medicine” (Hardman et al., 2019, p. 1). Commentaries are thus examined as direct indicators of public perception, while I investigate the sources that can contribute to the formation of perceptions.

These are valuable insights, which must however be evaluated against the background of certain limitations of the study. The comments are distributed very unevenly among the articles and there is no information about the number of commentators. Generally, there may be a sampling bias considering that people with certain characteristics might be more likely to write comments than others. Moreover, the authors recognize that the nature of rather short comments “may preclude more nuanced reflection by commentators” (Hardman et al., 2019, p. 7). The authors advocate further research into the public understandings of placebos, ideally including a wider range of media and other interactional environments. Furthermore, they endorse predecessors who advocate a reorientation towards more ethnographic fieldwork in placebo research to explore ‘the endogenous methods employed by members of societies in the co-production of the order and meaning of clinical settings’ (Hutchinson & Moerman, 2018, p. 377).

Thus, based on the history of placebo research as well as most recent findings, I propose that in order to progress in understanding the placebo phenomenon and its potential for application in medical practice it is crucial to understand how people make sense of the phenomena. If we

(11)

are to attain knowledge of how meanings are constructed, we must conduct more research that goes beyond exclusively medical settings, methods and narrow medical definitions. I therefore undertake a sociological examination using the SR theory as conceptual framework and an inductive empirical approach to gain insights into the general public’s perceptions of placebos and their effects.

2.2 Social representation theory

The SR theory was first developed by Serge Moscovici (1961) in his study of understandings of psychoanalysis in different social groups in France. It is also often referred to as a theory of common sense (Bauer & Gaskell, 2008; Sammut & Howarth, 2014) or of socially shared sense-making (Young et al., 2017), which theorizes the constructions and dynamics of common-sense knowledge. I am going to use the SR theory as a “concept with which we, as social scientists, map the phenomenon of common sense and elucidate key underlying processes” (Bauer & Gaskell, 2008). According to Jodelet (1989), a social representation is thus a “socially elaborated and shared form of knowledge” (p. 48). Moscovici (1973) defines SRs as “a system of values, ideas, and practices with a twofold function” (p. xiii). First, they serve to establish a social order which enables individuals to orientate themselves and master their material and social world; and second, to enable communication among members of a community through a shared code for social exchange (Sammut & Howarth, 2014). All representations arise in an effort and with the objective to familiarize the unfamiliar (Moscovici, 1984).

It is one of the central characteristics of the SR theory that the subject of study is marked by an element of novelty, peculiarity or oddity. “Every deviation from the familiar, every rupture of ordinary experience, everything for which the explanation is not obvious creates a supplementary meaning and sets in motion a search for the meaning, and the explanation of what strikes us as strange and troubling.” (Moscovici, 1998, p. 241). The phenomenon of placebos triggers such a process as it seems inexplicable with the current body of knowledge and leaves us puzzled.

(12)

2 Theory 10

For this research, I have favoured the concept of SRs over that of “attitudes” or “opinions”, as it has a stronger social constructivist stance. While attitudes and opinions are more essentialist concepts, residing inside a person, the concept of social representations underlines the social construction of meaning (Burr, 2015). The meaning of something emerges or is “represented” as an understanding between individuals oriented to the same social phenomenon (Sammut et al., 2015). Meaning-making is therefore of utmost importance in the SR approach. Moscovici deliberately allows and even expects cognitive polyphasia, i.e. the coexistence of competing and sometimes contradictory representations within the same public or individual (Howarth et al., 2004). This is one of the main differences to Durkheim’s concept of collective representations, the foundation of Moscovici’s theory, which he sought to adapt to modern, pluralistic and dynamic societies. According to the SR theory, an object can thus have multiple and different representations. This aspect will become relevant again in connection with the topic modelling.

SRs are created and transformed through two processes: anchoring and objectifying. The former refers to a process of classification which locates the novel and unfamiliar within the familiar frame of reference, i.e. associating it with existing SRs (Sammut et al., 2015). Objectification is a process of externalization by which representations are projected into the world through e.g. images, propositions, symbols, metaphors or personifications. SRs are rendered more concrete and tangible, for instance, pictures of “Dolly the sheep” became an objectification of genetic engineering (Bauer & Gaskell, 1999). Anchoring and objectification perform both descriptive and evaluative functions (Jaspal, 2014). Language is a key factor in these processes since the selection of certain words can activate entire systems of meaning (Markus & Plaut, 2001).

However, social representations are not mere cognitive phenomena but translate into social practices. Howarth (2006) points out that SRs do not only influence but constitute social

(13)

practices as we re-present2, transform, resist and promote certain SRs. Moscovici (1998) goes as far as to argue that once a SR is formed, its power will actually supersede the individuals’ own direct experience with a phenomenon. However, SRs can always be renegotiated and modified, or as Moscovici put it – ‘there is a kind of ideological battle, a battle of ideas’ (Moscovici, 1998, p. 403). Howarth (2006) suggests that we need to pay more attention to the struggle, noise and conflict inherent in social re-presentation. She cautions us not to disregard a critical perspective when using the SR theory to address questions such as: how are different meanings of placebos asserted and contested? How do they coexist? What are the consequences of adopting or resisting different SRs? Why do we use some representations of placebos instead of others? SRs reveal themselves in texts but also for example in social practices (Voelklein & Howarth, 2005). In this paper, I focus on social representations manifest in news articles. 2.3 Social representations in the media

Moscovici (1961) developed the SR theory in his book “La psychanalyse, son image et son public” to theorize how scientific knowledge, or abstract ideas, become integrated into everyday knowledge. Although the theory is applied in many areas, one of its main concerns and fields of application continues to be public understandings of scientific topics and innovations (Sammut & Howarth, 2014). Unfortunately, I cannot provide a comprehensive overview of the rich literature on the relations between science and the public in the context of this paper. Nevertheless, I intend to touch upon some relevant theoretical aspects of this relationship and the role of the media in this context.

The SR theory provides a suitable framework to approach the relation and knowledge exchange between the scientific and public sphere. Moscovici (1961) distanced himself from the predominant “deficit concept” of the public at that time. This was the idea of an unbridgeable superiority of science over the erroneous and misunderstanding lay reasoning. He

2_{The hyphen is used to highlight the practice of constant re-negotiation and re-interpretation of the social} representations (Howarth, 2006).

(14)

2 Theory 12

also rejected a “diffusion theory”, in which expert knowledge is disseminated unchanged among a rather passive public (Bauer & Gaskell, 2008).

In the SR theory, in contrast, knowledge is not only diffused but transformed. Individuals bring their previous experiences and predispositions into a communication about a topic, in the process of which this topic undergoes qualitative changes (Bauer & Gaskell, 2008). Furthermore, there is not only a single public discourse, but “a multiplicity of elaborations and re-constructions by social groups for social groups” (Wagner & Hayes, 2005). The SR theory is thus in line with more contemporary conceptualizations of the relationship between science and the public, which recognize that a one-way transfer of knowledge from science to common sense is hardly realistic. Scientific knowledge is equally socially constructed and coexists and competes with different knowledge systems in the struggle over meaning (Howarth, 2006).

Washer and Joffe (2006) demonstrate how the mass media can act as a bridge between scientific or specifically medical and public understandings. The majority of people have no direct access to knowledge about scientific advance and hence depend upon “socially mediated ‘second-hand’ knowledge” (Kronberger, 2015, p. 358). They turn e.g. to the mass media or friends and family as sources of information. The media gain in significance if people have little direct, personal experience with the subject in question and if the phenomenon is not fully understood – as it is often the case with complex scientific or medical issues (Morgan, 2009).

Media not merely inform, but also shape the individual and collective perception of a topic as well as interpersonal communication about it. This is realized through mechanisms such as the use of a specific language, integration into broader schemata, selective focusing, i.e. the news coverage amplifies certain aspects but marginalizes others, or inaccurate reporting (De Vreese & Lecheler, 2012).

The media are thus a place where SRs are formed, (re-)constructed and reflected (Caillaud et al., 2012). If the placebo effect in news articles is, for example, often linked to illusions,

(15)

deception and nonsense, the reader will likely be left with a negative association, which may facilitate an aversion to placebo treatment.

3 Research Design

The overall aim of this study is to contribute to expanding our knowledge of the public’s understanding of placebos and their effects. This is realized through a qualitative and automated text analysis of the social representations of placebos in newspaper articles in Germany from 2007 to 20193. Germany was selected based on empirical considerations and can be regarded as a typical or rather extreme case. The country is a major player in placebo research: the German Research Foundation funds placebo research since 2010 (project: FOR 1328) and in 2013, Tuebingen scholars organized the largest placebo conference up to that date. Furthermore, the term is consistently present in German news media and German is my mother tongue, which is advantageous when analysing SRs constructed through language. I decided to analyse SRs in newspaper articles because of the above-mentioned key role of the media and because it enables a longitudinal analysis of the evolution of SRs over time.

First, an exploratory qualitative pre-analysis of a selection of news articles is undertaken. This serves to check the quality of the retrieved text corpus, to familiarize with the articles and to be more knowledgeable for the development of the automated analysis concept. Subsequently, structural topic modelling (STM), an unsupervised, automated machine learning method, is employed to a) identify various topics in the context of which it is written about placebos and b) to trace the evolution of topics over time. This method is particularly suitable for large amounts of data whose manual qualitative or quantitative analysis would be very labour-intensive and time-consuming. Furthermore, due to similar linguistic and theoretical assumptions, STM lends itself very well to the analysis of SRs – although, surprisingly, this has rarely been done before. Lastly, qualitative analysis methods are consulted again for the purpose

(16)

3 Research Design 14

of a more thorough interpretation of the topics resulting from the topic model and to analyse what it is written about placebos in each topic.

The combination of methods thus fulfils both exploratory and explanatory purposes; both serve to explain and expand on each other’s findings. By alternating and combining qualitative and automated text analysis methods, I seek to draw on the strengths of each method and offset their weaknesses. I can claim, for example, fuller coverage of the representations by using topic modelling and reduce issues related to generalizability and selectivity in qualitative analysis. Through a detailed qualitative analysis of each topic resulting from the topic model, I can detect more subtle nuances and sentiments that otherwise might have been overlooked. Moreover, the systematic thematic clustering of the articles prior to the qualitative analysis enhances reliability. The subsequent in-depth qualitative analysis of articles from each topic verifies measurement validity i.e. the quality of the model and enhances the internal validity or credibility (Guba & Lincoln, 1994) of the findings. All in all, by combining and integrating the two text analysis methods, I aim to capture the social representations of placebos in news articles in a novel, comprehensive and at the same time nuanced way.

Data

The news articles for analysis were retrieved from the database Nexis Uni. Articles were only considered for inclusion if they contained the keyword search term “placebo” at least twice. Furthermore, articles must not contain the words “Los placebos” or “Rock!” or “Brian Molko”. This restriction enabled the exclusion of 66 articles related to two rock bands, which might otherwise have distorted the results. Other than that, the search terms were not further specified, so that texts were included in which references to placebo were both marginal and central. This can be useful, as DiMaggio, Nag and Blei (2013, p. 576) describe, because “casual allusions may reflect prevailing assumptions better than carefully crafted reports” and may be read by readers who would skip an article exclusively dedicated to placebos. Articles which

(17)

were not published in German in Germany and which did not fall under one of the three publication types ‘newspaper articles’, ‘web publications’, and ‘articles from journals and magazines’ were omitted. There is no limitation to specific sources, resulting in a total of 111 national and regional news outlets represented in the sample (see Table A1 in the appendix for a complete overview). The selection was further confined to articles published in the period from 01.01.2007 to 31.12.2019. I have selected 2007 as the starting date because from this year on there is a strong increase in articles on the subject in the database. This is most likely caused by the database, as many major German newspapers, such as ZEIT, Tagesspiegel or the online presence of Süddeutsche Zeitung were not entered into the database until 2007/2008. The number of articles meeting all the criteria set out above increased slightly until 2010 and then remains fairly stable until 2019. Besides the textual data from the articles, the date of publication was additionally retrieved from the database for each article entry.

After removing duplicates using a duplicate detection algorithm, 1970 articles remained. To estimate the effect of the year of publication at a later point in the analysis, 36 articles without information on the publication date were removed. This yields a final text corpus of 1980 articles with a vocabulary of 4125 unique terms.

4 Qualitative pre-analysis

A first qualitative pre-analysis of a sample of 50 articles served to familiarize myself with the data. Systematic sampling ensured that articles from different years and months were included in the sample for analysis. Every 30th article was selected until the sample size was reached. The final sample consisted of 4 to 6 articles from each year between 2010 and 2019. In the first step, I read and re-read the articles and noted down some initial codes, such as dominant or recurring themes. In the second step, I conducted a more systematic analysis and coding with the help of ATLAS.ti. The unit of analysis was the paragraph of an article in which

(18)

4 Qualitative pre-analysis 16

the term placebo occurred. Codes answered the question: “What and how is written about placebos and their effects?”

These primarily thematic codes were generated from a combination of deductive and inductive approaches to the texts. For instance, the code mechanism was derived from the literature as the underlying mechanisms of the placebo effect dominate current placebo research (e.g. Benedetti, 2014). The selection of more inductive codes was guided by Ryan and Bernard’s (2003) techniques to identify themes. For example, the thematic code problems emerged by searching for transitions in content in the texts. Many shifts turned out to address the current difficulties, uncertainties and challenges associated with the placebo effect. I came across twists, such as a change from “it works” to “however, not all people react the same” or from “we’re very positive about the use of placebos” to then limiting it to “it can support but not heal everything”. Altogether the following codes were identified and selected: realness, scientific, effectivity, mechanism, problems, self-healing, group of people, doctor-patient relationship, medical practice, alternative and orthodox medicine, sentiment, tone, general context. Table A2 in the appendix contains detailed information on how I arrived at each code. This qualitative pre-analysis was particularly important for the research process and less for the final results. Altogether it served a threefold purpose: Firstly, it served to check the quality of the sampled text corpus. Secondly, it helped to gain a first impression of the articles, and how they use the placebo concept. For instance, it made me aware of the variety of the various contexts in which the term placebo appears, such as in articles on scientific studies, medical practice, homeopathy, placebo research and politics. It is likely that the meaning of the concept of placebos emerges from the totality of ways how it is used, appropriated and modified in the various contexts. Lastly, the analysis was crucial to make a better-informed decision about the most suitable methods to analyse the social representations of placebos in the whole text corpus. As hand-coding techniques are not well suited to analyse large text corpora, I will rely in the

(19)

next step on an automated text analysis method to organize all articles according to the different contexts.

5 Automated news article analysis

5.1 Method

Due to the extensive corpus, topic modelling, an automated text analysis method, is employed to investigate the occurrence and distribution of contexts in which placebo is mentioned. Topic modelling serves, as DiMaggio et al. (2013) aptly describe, as a “computational lens” for the researcher to look at a large text corpus and reduce its complexity. Furthermore, it is highly suitable for the operationalization of “social representation”, as will be elaborated in section 5.2.

Topic modelling algorithms are a range of machine learning methods for discovering manifest and latent topics in a large collection of text documents (DiMaggio et al., 2013). “Topics” are defined as frequency distributions of words, based on co-occurrences (Blei, 2012). Note that topic modelling is a so-called “bag of words” approach, which disregards other complexity of language, such as grammar, word order or location in the text (Wallach, 2006). It is based solely on the co-occurrence of single words. The underlying linguistic assumption is that words which tend to systematically appear together (cooccur) across multiple texts are also associated thematically (Matthes & Kohring, 2008). Through this relational approach to meaning, i.e. the investigation of the context (other terms) in which a word appears, it is also possible to grasp varying meanings of one word (“polysemy”). Another strength of topic models is that they are mixed membership models, which regard a text document as a mixture of topics. This means that one news article can be assigned to multiple topics. Topic models can thus capture heteroglossia, the co-presence of different perspectives, opinions and topics in one article. This also marks a significant difference to many other clustering methods, where one document is allocated to one cluster.

(20)

5 Automated news article analysis 18

Topic models are often referred to as semi-automated, “unsupervised” methods (Roberts, Stewart, Tingley, 2014). This highlights the fact that it is an inductive, exploratory approach, where – in contrast to supervised machine learning methods – no previously developed theories or categorizations are imposed on the data. Instead, topic modelling enables a data-driven discovery from which clusters of related words result, suggested by language patterns in the texts (Schwartz & Ungar, 2015). This allows for the possibility of obtaining previously unexpected results, which is particularly useful if there is not yet a large body of research on a topic such as the social representations of the placebo effect.

It is important to keep in mind that the unit of analysis is no longer just the paragraph in which “placebo” occurs, but that each word of every article serves as input for the model. The topic model can thus serve to map the news articles according to the different contexts in which it is written about placebos and these semantic contexts might already prime some particular associations or interpretations of placebos in a reader. Subsequent to the modelling, however, one has to carefully analyse the depiction of placebos in example articles for each topic to grasp the SRs of placebos that go beyond the contexts of the articles.

In this research project, I am employing a structural topic model (STM), a specific variant of topic models. STM has the advantage that it allows the inclusion of document-specific covariates, such as ‘date’ or ‘author’, into the topic estimation process (Roberts, Stewart, Tingley, 2014). Latent Dirichlet Allocation (LDA), another example of a topic model, assumes the documents in the corpus to be exchangeable and unstructured. In contrast, STM enables one to take metadata of the news articles, in this case its publication date, into account. In this way, the relationship between the topic prevalence and the time of publication can be examined. Their relation can best be understood by reference to a familiar regression framework, with the topic as the outcome variable and the release date as the explanatory variable (Roberts, Stewart, Tingley, Lucas et al., 2014). Thus, STM allows exploring whether and how the prevalence of the identified topics – the different contexts in which the term “placebo” appears – has changed

(21)

over the period from 2007 to 2019. The topic model was conducted with the stm package in R, developed by Roberts, Stewart and Tingley (2014).

5.2 Connecting SR theory und topic modelling

SR theory has few methodological implications and has been successfully used with a wide range of methods such as surveys, ethnography, discourse analysis and interviews (Sammut et al., 2015). Saadi Lahou’s (1995, 1996) pioneering works were one of the first to develop the application of text mining methods for the study of SRs. Apart from a few scholars who used the same software (ALCESTE) as Lahou, text mining methods have remained marginal in this research program. Chartier and Meunier (2011) pointed out the fallacy of using the software as a method. Subsequently, they outlined the general linguistic assumptions underlying ALCESTE and other automated text analysis methods and how they relate to the SR theory. Lynam (2016) then became the first to use topic modelling to analyse social representations and to highlight the potential of this combination. Apart from this, the use of automatic text analysis methods and in particular topic modelling to study SRs remained very limited and mostly uncommented (e.g. Sterling et al., 2019).

There are several reasons why topic modelling is especially well suited for the operationalization of SRs. As explained above, SRs are significantly formed through language. Segments of a text in a corpus can thus be regarded as instances of SRs (Lahou, 1996). The words constituting these parts of discourse can be grouped into topics based on similar word co-occurrence patterns. Through this structuring based on similarity emerges a kind of “semantic map” of SRs (Lahou, 1996). Elsewhere Lahou (2003, p. 46) also calls these generated classes, here topics, “building blocks” of SRs. Topic modelling thus generates a set of interpretable topics reflective of elements of SRs (Lynam, 2016).

Moreover, the affinity between topic modelling and SR theory arises from a number of similar theoretical assumptions. Topic models can account for cognitive polyphasia, i.e. the co-presence of differing SRs in a community, an individual or an article, since an article can be

(22)

composed of different topics. Structural topic models even allow capturing the dynamic element of SRs, since they can trace their development over time. SR theory assumes that the formation of SRs is particularly initiated by novel and unfamiliar phenomena. The topic modelling fits in well due to its explorative, unsupervised approach – often used when there is a lack of knowledge about the subject. It is important to emphasize though that, depending on the research subject, a detailed analysis of the individual documents per topic might be necessary to reach a profound understanding of the SRs. Overall, I reason and shall demonstrate that topic modelling is very well suited to examine the content but also the structure (e.g. proportions and correlations) of SRs.

5.3 Analysis

Several challenges need to be addressed to ensure a valid application of topic modelling to textual data: a) appropriate pre-processing of the text documents; (b) adequate selection of model parameters; and (c) the evaluation of the model’s reliability and validity (Maier et al., 2018).

a) Pre-processing

Topic models are susceptible to “noise”, which means that they may mistakenly identify patterns of co-occurrence of unimportant but highly repetitive terms. To improve the model’s quality of the result, several standard text pre-processing techniques were performed on the input text. Texts were pre-processed by removing numbers, punctuation, URLs and other symbols such as hashtags. I also converted all terms to lower case and stemmed each term, i.e. reducing inflected words to their word stem (e.g. the words “healer” and “healing” result in the same stem “heal”). Words that primarily serve grammatical functions, so-called “stop words”, often do not convey meaning relevant for the topic modelling. Therefore, a list of common German “stop words” supplied by the quanteda R package was removed as well as some meta-level terms of the news article such as “photo”, “caption” and “end of document”. Lastly, words were removed that appeared in more than 99% of the documents, or less than 5%, based on

(23)

guidelines suggested by Maier et al. (2018). Words that appear in so few, respectively almost all, documents are unlikely to be discriminating and to add value to the topic model.

b) Model selection

One challenging aspect of structural topic modelling is that the researcher needs to determine the number of topics, K, in advance. Assuming too many topics might result in similar topics that cannot be distinguished from each other in a meaningful way; too few topics might lead to several distinct topics being combined in one (Grimmer, 2010). There is no “right” answer to the appropriate number of topics, but it depends on the nature of the corpus, e.g. its heterogeneity, the purpose of the research and some statistical metrics which can give guidance (Roberts, Stewart, Tingley, Lucas et al., 2014). Roberts and colleagues (2019) recommend starting with around 5 to 50 topics for rather small corpora (a few hundred to a few thousand documents). There are various metrics that can be used to inform the search for K within this range. The measure of perplexity, which evaluates the model performance on data to which the model has not yet been applied, has been traditionally used to assess the quality of a model. However, Chang et al. (2009) have discovered that better performance on held-out likelihood does not necessarily equate to more semantically meaningful topics. Therefore, I followed e.g. Fischer-Preßler et al. (2019), Lindstedt (2019) and Roberts, Stewart, Tingley, Lucas et al. (2014) in focusing on metrics that assess the quality of topics, such as semantic coherence and exclusivity. Semantic coherence indicates whether a topic is internally consistent. It is maximized when the most probable words in a given topic frequently co-occur within documents (Mimno et al., 2011). Exclusivity in turn measures the share of top topic words that are distinct to a given topic. A topic is exclusive when the top words for a topic are unlikely to appear within the top words of another topic (Roberts, Stewart, Tingley, Lucas et al., 2014). A topic that scores well on both criteria is more likely to be semantically useful. To find the model which best represents the data, I computed the average semantic coherence and exclusivity for k = 10 to 35 in increments of 5 (Figure 1).

(24)

Figure 1

Model Selection: Semantic Coherence and Exclusivity Plotted for k = 10, 15, 20, 25, 30 and 35.

It can be seen that models with 15, 20 and 25 topics outperform the other models because they achieve better values for semantic coherence and exclusivity. To arrive at the final decision, I fitted models for K = 15, 20 and 25 and examined the most probable words of each topic as well as the semantic coherence and exclusivity for each topic from each model (see Figure A3 in the appendix). Finally, I opted for 20 topics because this number showed the most desirable statistical properties. Moreover, when comparing the most probable words of every topic in each topic model solution, it seemed that the 15-topic solution lacked thematically relevant topics. 20 topics leave more room for differentiation, which seems important for a rather heterogeneous text corpus.

c) Validation and assessment

Producing a topic model solution does not mark the end but the beginning of the analysis. Now it is up to the researcher to assess and demonstrate the validity of the model and to utilize its output information for the analytic question that motivated the research (DiMaggio et al., 2013). There are three overall types of validation (see Quinn et al. (2010) for more differentiated techniques). The first is statistical, including the semantic coherence and exclusivity metrics calculated above. Another form is predictive or external validation (Quinn et al., 2010). It is argued that, for valid topics, external events should be able to explain sudden increases in

(25)

attention for a topic. The extent to which topic prevalence coincides with external events will be discussed in more detail in section 7.2 when the relationship between topic prevalence and the publication date is explored.

Lastly, there is semantic or internal validation (Grimmer & Stewart, 2013), which assesses the extent to which the topics are internally homogenous and semantically interpretable, yet distinctive from other topics. This is usually done through careful reading of representative documents of each topic, also called “exemplar” documents (Roberts, Stewart, Tingley, Lucas et al., 2014). These are newspaper articles in which the highest proportion of words is assigned to a specific topic. After first inspecting the most probable and most exclusive terms of each topic, I analysed the 15 articles most associated with each topic. Based on the examination of significant keywords in conjunction with a close reading of exemplary articles, labels were assigned to the topics. See table 1 for an overview of all topics, their proportions, labels and FREX words, i.e. the most frequent and simultaneously most exclusive words of a topic. Table 1

Overview of All Topics of the Final Topic Model Sorted by Topic Proportions Topic

ID

Assigned label Prevalence FREX words 9 Politics and

economics

10.7% federal government, CDU, government, banks, SPD, state, union, FDP, law, EZB

15 Placebo research 8.9% Enck, expectations, Bingel, effect, expectation, pain, nocebo effect, placebo effect, painkiller, expectation, attitude

6 “Placebo-measure” 8.1% coach, citizen, debate, green, party, politics, request, Mr., politically, mister

18 Homeopathy 7.8% homeopathy, conventional medicine,

complementary medicine, homeopathic, alternative medicine, homeopaths, Hahnemann, globules

16 Women 6.4% heart attack, flibanserin, risk, women, men, sex, placebo group, iron, gel

10 Families 5,8% parent, child, told, hospital, young, mother, asks, suddenly

(26)

19 Belief and spirituality

5.5% Hirschhausen, faith, animal, think, wonder, religion, god, believe, human, our

13 Other experiments 5.1% testosterone, caffeine, behaviour, experiment, participants, test persons, wine, coffee, test persons, scientific

14 Vaccines 4.9% antibody, Alzheimer, Prof. Schwab, virus, infected, aids, vaccinated, immune system, vaccine

8 Studies and research 4.7% Tamiflu, Cochran, data, Roche, analysis, British, authors, published, advice, results 4 Pain treatments 4.4% migraine, chronic, back pain, joint, knee,

chronic pain, intervention, pain, surgery, therapy

20 New medications 4.3% Switzerland, Zurich, big, name, white, Franc, medicine, new

2 Household remedies 4.2% vitamin, fruit, nutrition, detail, eat, paracetamol, food, fish, allergies

5 Body products 3.8% body, customers, beer, Hamburg, athletes, product, sugar, dope, training, athletes

11 Mental health 3.6% antidepressants, depression, depressive, psychiatrists, LSD, psychotherapy, psychiatry, baclofen,

1 RCT 3.1% manufacturer, approval, market, “Bayer”,

medical preparation, approve, company, drug, pharmaceutical company

3 Events 3.0% clock, light therapy, Stuttgart, street, Saturday, light, concert, theatre, tickets

7 Neuropsychology 2.3% oxytocin, stent, face, Prof Hurlemann, hormone, friend, autism, amygdala, love, trust, brain 17 Low semantic coherence 2.1% - 12 Low semantic coherence 1.3% -

Note. The original topic ID of the topic model output was retained for future reference.

As it is often the case in topic models, some topics – in this model topic 12 and 17 – are not interpretable and have no substantive meaning (Maier et al., 2018). However, they can serve to sharpen the clarity of other topics by collecting random or boilerplate content in one place (DiMaggio et al., 2013). My assumption that topics 12 and 17 belong to this type of topics is supported by the fact that they exhibit the lowest semantic coherence (Figure A3 in the

(27)

appendix) and are the least prevalent topics. Except for these two topics, internal validity, i.e. meaningful connections and commonalities between the sampled press releases could be found in all other topics.

The reading of sample articles of each topic did not only serve to check the internal validity but also to analyse the social representations of placebos in each topic. As previously indicated, the STM mainly informs us about the different contexts in which the word “placebo” is used. Only an in-depth qualitative analysis of how precisely placebos are written about in articles of each topic can shed light on the SRs of placebos. This analysis was guided by the procedure as well as the findings of the first, explorative, qualitative analysis.

6 Findings

6.1 Social representations of placebos

For a more concise presentation of the social representations of placebos in the remaining 18 topics, the topics are arranged in four topic clusters. I determined these by calculating the correlations between all topics, (see Table A4 in the appendix) and subsequently visualizing them in a network (Figure 2). Nodes are topics and edges represent significant (>0.05) positive correlations between them. Furthermore, approximate topic proportions can be estimated, indicated by the size of the topic labels. However, one must refrain from making any spatial interpretations of the position of the nodes. The Fruchterman-Reinold algorithm, a force-directed layout algorithm used to plot the network, seeks to make network edges and clustering structures clearly visible by minimizing edge overlap and varying edge lengths. The spatial arrangement of the nodes (topics) is therefore not meaningful and cannot be interpreted.

Based on the correlations between the topics, the topics can be grouped into thematic clusters, i.e. similar contexts in which it is written about placebos. Seven topics focus on clinical research in medicine; three topics cover a variety of other studies and experiments that are not directly related to health and disease; four topics are concerned with applied medicine and the

(28)

6 Findings 26

treatment of diseases; and three topics highlight the use of the term “placebo” in other, societal or political contexts.

Figure 2

Correlation Network of the Structural Topic Model

Note. Edges indicate the correlation between topics. The marked circles indicate the formed thematic clusters.

The topic model has served to identify various contexts in which the word “placebos” is used. The following qualitative analysis of the depiction of placebos in 15 sample articles of each topic determines the specific SRs of placebos in the different contexts. These social representations will be discussed in detail below for each topic cluster. Particular attention will be paid to the definition of the term, the attributed efficacy, the described mechanisms underlying the placebo effect, its scope of application and sentiment.

a) Clinical research

This cluster covers seven topics related to clinical research, whereby the effectiveness of new substances, treatments and products is investigated and assessed. The first topic can be

(29)

summarized as “research and studies”. It constitutes one of the most central nodes in this cluster and could be thought of as a kind of umbrella term for the topics in this cluster. It examines studies and the practice of research from a meta-perspective. Articles are concerned with (un)published, doubtful, weak and available studies and data, e.g. on the influenza drug “Tamiflu”.

The next topic (“RCT”) belonging to this cluster deals with testing the effectiveness of medical interventions in classical randomized controlled trials (RCT). Actors in these articles are, for example, researchers, pharmaceutical companies, manufacturers, patients and test persons. This topic is likewise of central importance for understanding the meaning of this cluster. It has weak positive correlations – or at least no negative ones – to any other topic in the cluster. This is likely due to the terminology used in organizing, conducting and presenting the results of a RCT, which is common in many of the other topics, as they are all concerned with medical research.

One topic is, for instance, more specifically about research on “vaccines”. It is all about vaccinations and similar agents that are intended to stimulate the immune system to produce antibodies or support it in its immune response. Sample articles look at the search for a drug for Alzheimer’s disease, a vaccine against Ebola, HIV, tuberculosis and the Zika virus. The news articles in the fourth topic (“mental health”) inform about the search for drugs against depression and other mental diseases. Articles discuss, for example, antidepressants and Botox to relieve depression and oxytocin to treat borderline disorders and autism. The next topic (“women”) differs from other reports on RCTs in that it focuses on the testing of treatment options for health problems that particularly affect women, amongst others: iron deficiency, menopause and breast cancer. However, sometimes it is just about studies with female participants. This topic is interestingly ranked 5th of the most common topics. Furthermore, it is striking that the word “women” occurs 1124 times in the total text corpus (35th most frequent word) but “men” only 354 times (rank 256th of the most frequent words). It might be

(30)

6 Findings 28

worthwhile for future research to investigate the extent to which there is a gender dimension or differences with regards to placebos and their effectiveness.

The subsequent topic (“household remedies”) is linked to the network through the preceding one. It is about testing the effectiveness of various household remedies and nutrients, such as garlic, elderberry extract, cough suppressants, ginger and sour cherry juice. Finally, the last topic (“new medications”) contains articles on the development, costs, registration, effectiveness and patents of new medicines. Articles report e.g. on planned changes in the pricing of new drugs, and on a critical examination of the process of how new drugs are tested. Interviewed persons are e.g. the scientific advisor at the European Medicines Agency EMA and the head of the Institute for Quality and Efficiency in Health Care.

Social representations in clinical research

In the above-mentioned topics, placebos are usually defined as a “sham drug” opposed to the “correct”, “real” substance – a definition that researchers have repeatedly found, for example, in patients’ and doctors’ conceptualizations of placebos (e.g. Bishop et al., 2012; Sherman & Hickner, 2008). A typical wording is: “One half received the vaccine, the other half an ineffective sham drug (placebo).” This comparison can be seen as part of the anchoring process for the formation of SRs, in which the “unknown” is set in relation to the familiar frame of reference.

The majority of articles in this topic cluster attribute placebos either no or very little effect compared to the real drug. A typical description of the results of a study reads: “Six weeks after the single treatment, 60 percent of them had halved the severity of their symptoms. In the patients in the placebo group, however, the symptoms improved only slightly.” The “mental health”- topic is an exception in this cluster, as the placebo effect is described as effective more often than in the other topics.

First, “more effective” in the sense that it is generally attributed more significance by professionals working in the field, e.g. a psychiatrist acknowledges: “The placebo effect is

(31)

large, that’s quite clear, and it applies to the whole of medicine, not just psychiatry.” Secondly, it is “more effective” in the sense that it is comparatively high in patients with mental health problems, particularly depression, “because about 30 percent of depressives also come out of the crisis with a placebo.” Interestingly, it is reported on two American scientists who make inferences about the disease based on that: “Perhaps the strong placebo effect, especially in mild depression, can be explained by the fact that some of the test patients were not really sick at all.”

“Mental health”- news articles explain the functioning of the placebo effect by referring to the ritual of going to the doctor, the psyche as well as thoughts, feelings, behaviour, expectations and attitudes of a person. All other topics of this cluster are characterized by the fact that they mention only very few underlying mechanisms of the placebo effect. This seems plausible, given that placebos are frequently regarded as ineffective.

Regarding the use of placebos, it can be observed that they are often considered an essential part of modern evidence-based medical research. For example, the topic “household remedies” states: “The gold standard for diagnosing food intolerances and also allergies is the double-blind, placebo-controlled oral food provocation”, or in the “vaccine” topic: “For a reliable and rapid proof of efficacy, one needs a placebo”. Issues with placebos are only mentioned in connection to RCTs. Problems that have been raised several times are whether they are ethically justifiable, that not all people respond equally to placebos, the transferability of the study results to practice and that some problems simply cannot be tested with placebo-controlled studies for methodological reasons.

Overall, the news articles in connection with clinical research are very factual and scientific. The term “placebo” is often used in a relatively undifferentiated manner, usually referring to the RCT procedure. However, the fact that the placebo is consistently compared with the “real” medicine can lead to the impression that the placebo must necessarily be fake, false and illegitimate (Miller & Brody 2011).

(32)

6 Findings 30

In summary, it can be said that placebo in articles in this cluster is mainly defined as a sham drug as opposed to the real substance, with no or little effect – except for the mental health topic –, few explanatory mechanisms are mentioned, the field of application is clearly research and it is reported very factually. Although none of the individual topics itself has an outstanding prevalence, taking all topics belonging to this cluster together, the cluster has the highest prevalence (≈ 0.312), which means that most articles report on placebos in this way.

b) Other research

Similar to cluster a), topics in this cluster involve research and studies. However, the topics also comprise many studies and experiments that are not directly related to health and illness.

The topic “neuropsychology” can best be characterized by the keywords “hormone”, “psyche” and “brain”. Many of the selected articles in this topic deal with the connection between hormones (e.g. oxytocin) and behaviour using the example of eating behaviour and biochemical processes of “falling in love”. The topic “other experiments” is mainly concerned with a variety of experiments and studies e.g. on acting in stressful situations, the influence of the price on the assessment of wine quality and overcoming lovesickness. Some of the articles come very close to a kind of placebo research, such as the last two mentioned examples, which show how the belief in a high price or a drug against lovesickness alone can influence our perception. This resemblance might explain the positive correlation between this and the following “placebo research” topic.

The last topic (“placebo research”) in this cluster focuses on research into the placebo effect and how a non-active substance/treatment can lead to pain relief. As placebos constitute the main subject of the articles, they are discussed in a more differentiated way, for example, a distinction is made between the placebo and nocebo effect and the open and concealed placebo administration. Among the most frequently cited experts in the sample articles are the placebo researchers Ulrike Bingel, a specialist in neurology, and Paul Enck, a professor of medical psychology. It is striking that this topic is the second most widespread of all. This suggests that

(33)

the term “placebo” does not only occur incidentally in articles but that there are also many articles in which placebos and their research constitute one of the central subjects of the article. Social representations in other research

In all three topics, the placebo is regularly referred to as a sham substance. In the topics “neuropsychology” and “other experiments” it is – as in cluster a) – often contrasted with something “proper” or the “real” substance. Despite the similar definition, the topics differ from the cluster on clinical research in the efficacy they attribute to the placebo effect. In the topic “neuropsychology” it is mentioned several times that placebos can significantly influence the neurochemistry of the brain. In one article, for example, it is noted that “doctors and laypersons alike must first get used to the idea that pills without a specific active ingredient stimulate the formation of messenger substances”. This remark also highlights the perception of placebos as something novel and unusual, as it is characteristic for research subjects of the SR theory.

This representation of effectiveness is also found in parts in the “other experiments” topic. Sometimes, however, placebos are simply used relatively uncommented as the following example illustrates: “People who had previously been given testosterone lied significantly less than people who only received a placebo”.

The “placebo research” topic differs from the above definition as it defines a placebo more broadly. It can be a sham drug or any other kind of intervention. For example, it is reported: “What ultimately works in acupuncture is probably a combination of a placebo effect, as it can also occur in conventional medical treatments, and specific neurophysiological mechanisms.” This definition resembles the one identified by Hardman (2019), in which placebo is understood as the entire clinical treatment process with or without a material substance.

As might be anticipated, the “placebo research” topic assumes a strong effect of placebos. For example, Prof. Enck states that placebo effects can often be as strong as a newly developed drug. Another article notes: “the placebo effect does not only apply to pills, but to any drug. It

(34)

6 Findings 32

also works with an ointment or an injection”, and “even if the person concerned knows it is only a placebo”.

While articles in “neuropsychology” and “other experiments” only touch upon some mechanisms of the placebo effect, such as positive expectation and belief in something, the topic “placebo research” examines the different mechanisms of the placebo effect in much greater detail. Apart from the two just mentioned, previous experiences, learning, fear, self-healing powers, the behaviour of the doctors and the ritual of taking the pill are presented.

In addition to the use of placebos in research, there are – to a greater or lesser extent – references to medical practice in the topics of this cluster. In the context of “neuropsychology”, there are occasional recommendations for dealing with placebos in medical practice, such as in the following suggestion by Tom Bschor, chief physician for psychiatry: “There should be a debate about whether it is okay to use placebo effects exclusively”, and “Many doctors already work with illusion and belief without saying so openly.”

Articles on placebo research refer more frequently to medical practice, i.e. how the effect could be given greater consideration in everyday practice. Enck demands, for example: “Doctors must give much more thought to how they affect their patients”, and “‘Talking medicine’ must be strengthened and better paid.” The same articles also identify several problems related to the placebo effect. For example, Enck explains that if doctors do not find physical causes, many do not take their patients seriously. A further difficulty is ethical conflicts. Since doctors are obliged to inform their patients, the placebo effect is difficult to utilize in practice.

Some of the articles are written less scientifically and factually than in the first cluster, which leads to the emergence of certain sentiments associated with the placebo effect. A few occasional, positive and negative, rather subtle connotations of the placebo effect can be identified. The following example from “other experiments” shall demonstrate a more negative one: “‘In the end, the reward and motivation system is playing a trick on us,’ explains Liane