Statistics in popular science: The argument from correlation to cause

(1)

The Argument from Correlation to

Cause

Martijn H. Demollin

Master’s Thesis

RMA-programme Rhetoric, Argumentation Theory &

Philosophy

Faculty of Humanities

Supervisor:

dr. J. H. M. Wagemans

(2)

Department of Speech Communication, Argumentation Theory

& Rhetoric

Spuistraat 134, 1012 VB Amsterdam, The Netherlands

Preface

Correlations are of vital importance in scientific research and often provide the first piece of the puzzle in the process of identifying relations of cause and effect between certain events. In that capacity, correlations represent a frequent argument in the scientific explanations offered for particular phenomena. In the course of my research on these arguments from correlation to cause, I have become increasingly interested in the relation between the argumentative shape of scientific explanations and the scientific philosophy behind them, which I want to learn more about in the future. This project’s topic has nicely complemented the broad range of compelling subjects I have been able to explore during my Master’s in Rhetoric,

Argumentation Theory and Philosophy.

This thesis has been both fun and challenging to write. As with other projects, I greatly enjoyed the discussions with my supervisor Jean Wagemans on the subject matter and I am very thankful for all his effort and inspiring guidance. I am also grateful to my partner Noor Sollie, and my mother Marianne Ringma for proofreading my work and providing me with helpful feedback. Furthermore, I would like to thank Francisca Snoeck Henkemans and my fellow students for their valuable comments and critique on the project during our academic internship. Finally, I extend my gratitude to all the members of the department of Speech Communication, Argumentation Theory and Rhetoric of the University of Amsterdam. I have studied argumentation with great pleasure over the past two years, and I feel richer for the experience.

(4)

1 Introduction

The popularization of scientific research – also called vulgarisation or accommodated science

writing – is defined as the process of making scientific findings and publications accessible to

a more general public. Popular scientific discourse is often characterized by a less technical vernacular than scientific literature as a professional medium and it aims to both inform and persuade non-scientific outsiders. Popularizing scientific knowledge has been traditionally considered an activity situated outside of the academic domain. That is to say, it was generally deemed unrelated to the process of knowledge production or the validation of scientific findings, both within academic circles and beyond (Shinn & Whitley, 1985, p. 3). However, in more recent developments, science popularization has become progressively widespread and connected to the process of validating scientific knowledge towards society at large. In part, this movement is due to a greater public and political emphasis on the amenability of scientific research, which stresses the societal importance and potential application of scientific findings outside of the academic sphere.

There are a number of dissimilarities between professional and popular scientific literature. Any author making an attempt at accommodated science writing faces Scylla and Charybdis: either one simplifies the source material, thereby concealing its full extent and depth; or one provides a comprehensive and expert account, with the likely result of dissuading an untrained reader (Fahnestock, 1986, p. 276). As a result of this dilemma, popularized publications of scientific results are typically prone to a certain degree of simplification and distortion with respect to their origins. Professional scientific literature is generally characterized by the presence of tentative, careful and qualified propositions of fact or relation. In contrast, popular science tends to present scientific research and its corresponding claims with less reluctance. Furthermore, a primary goal of professional scientific publications consists of providing a validation and justification of a proposed scientific explanation towards an academic audience. A popular scientific publication caters to a much wider and more disparate audience and partially operates outside of the relatively strict academic institutional constraints and conventions.

(5)

From the perspective of the pragma-dialectical theory of argumentation, professional scientific publications can be considered a communicative activity type within the communicative genre of disputation (van Eemeren, 2010, p. 143). Professional scientific publications within the academic domain are aimed at facilitating a number of institutional goals: (1) reporting informatively and persuasively on scientific progress, (2) the accumulation of scientific knowledge and (3) the critical examination of scientific claims by peers. In order to fulfil the institutional goal of critical examination by peers, most academic publications focus on providing a scientific interpretation of observed data or findings, and mounting a substantiated defence for any claims put forward in the process. In other words, a scientific paper will often propose an argumentative standpoint that is to be defended, adduced with a number of arguments that speak in its support.

Popularized scientific texts fulfil a role in the academic domain and the journalistic domain, meaning they also aim to achieve additional institutional goals relevant to the realm of journalism. As such, popular scientific discourse can be considered its own unique communicative activity type, with its own set of characteristics and discursive attributes. This makes the translative process from scientific discourse into a popular scientific publication interesting from an argumentation theoretical point of view. The professional and popular scientific publications serve to fulfil dissimilar institutional goals and cater to different audiences, which affects the argumentative strategies employed in each respective genre. For example, the original standpoint – or interpretation of the observed findings – in a professional scientific publication may be rephrased or emboldened, and its original argumentative structure may be relinquished in favour of another in a popularized text. In doing so, the inferential structure of the original claim and its supporting arguments may be fundamentally altered.

The argument from correlation to cause is a good example of a type of reasoning that is often employed with considerable restraint in professional scientific discourse, but less so in popularized scientific texts. While correlations possess certain predictive capacities, they may only be employed to infer causation under very strict conditions, which is due to the fact that there could remain a number of hidden variables that may play a part in the observed relation between the two measured variables. For example, there may exist a correlation between the number of public schools and movie theatres in cities. Clearly, the construction of a public school does not necessarily or directly cause the construction of a movie theatre or vice versa. Rather, there is a third variable which is responsible for the observed correlation:

(6)

population size. In other words, cities with greater population size will have more demand for movie theatres, public schools and many other social institutions.

In some cases, one could argue that a particular correlation does point to a causal link between the observed variables, when the presence of one variable is believed to directly influence the presence of another. When a correlative argument is adduced in defence of a claim to the existence of a causal relation, this is also known as an argument from correlation

to cause (Walton, 1996, p. 71). In order for such an argument to be acceptable, several other

arguments aside from the correlative connection between the observed variables are needed. However, creating an adequate representation of such tentative and complex argumentation that is suitable for a popular scientific text is complicated by the fact that there exists no real alternative for the concept of correlation that is both accessible and accurate. Also, in light of the partially journalistic nature of popular science, the other necessary elements in the argument from correlation to cause may not always be fully presented for reasons of simplicity or audience demand.

Regrettably, misinterpretation of the difference between correlation and causation is quite common in popular scientific articles about research into causal factors (Govier, 2005, p. 292). At times, popular scientific texts use arguments from correlation to cause with greater inferential authority than they actually warrant, leading to a skewed – or even unreasonable – representation of the scientific findings they are based on. Given the importance of the accuracy with which scientific knowledge is conveyed to a greater public, this discursive phenomenon within popular science is a worthwhile topic of research. However, the popular scientific genre has received relatively little academic attention thus far, which means its communicative characteristics have not yet been fully delineated. Although the argument from correlation to cause has been briefly explicated by a number of scholars, these treatments mostly depart from a view of the argument as a fallacy that may be redeemed with additional arguments. In my view, correlations constitute a preliminary, but essential step in establishing a scientific explanation for a potential causal connection between particular events. That is to say, employing a correlation in an argument supporting a causal relation can be perfectly reasonable – if there is additional evidence. In support of this view, it is important to set out the conditions under which an argument from correlation to cause can be considered reasonable and argumentatively acceptable.

In order to address these theoretical issues and further investigate arguments from correlation in popular science, this thesis aims to fulfil three primary goals: (1) to provide a characterization of popular scientific discourse as a communicative activity type, (2) to

(7)

formulate an argument scheme for the argument from correlation to cause from a pragma-dialectical perspective, and (3) to provide the criteria for the reasonable use of the argument from correlation to cause. In doing so, a theoretical tool is developed that allows for the analysis and evaluation of arguments from correlation to cause in the context of popular scientific discourse.

Chapter 2 will be devoted to a characterization of popular scientific discourse as a communicative activity type. To that end, Section 2.1 provides a brief introduction to some of the concepts of the pragma-dialectical theory of argumentation relevant to this investigation. Section 2.2 discusses the institutional goals of popular scientific publications and contrasts them with the institutional goals of professional scientific publications. In Section 2.3, the focus lies on describing the intended audience of popular scientific texts and how this affects the communicative activity type.

Thereafter, Chapter 3 is dedicated to a précis of the argument from correlation to cause. Section 3.1 is centred on a discussion of the technical definition of a correlative relation, while Section 3.2 provides a pragma-dialectic formulation of the argument from correlation to cause. In Section 3.3, I formulate the criteria for reasonableness that apply to the argumentative use of the argument from correlation to cause.

Lastly, in Chapter 4, I discuss two examples of arguments from correlation to cause in popular scientific discourse based on a correlation put forward in professional scientific publications. These examples serve to illustrate the application of the theoretical instrument developed in Chapter 2 and 3 and highlight some of the complications that may arise in arguments from correlation to cause. For each example, an analytical reconstruction of the argumentation in the professional and popular scientific texts is formulated. Subsequently, the arguments from correlation in the popular scientific literature are evaluated on the basis of their representation of the source material and argumentative acceptability.

(8)

2 Characterizing popular scientific discourse as a communicative activity type

2.1 Introduction to the pragma-dialectical theory of argumentation

The pragma-dialectical theory, as put forward by van Eemeren and Grootendorst (1984, 2004), is a theory of argumentation which is designed to allow for the analysis and evaluation of argumentation as it is actually used in communication. In pragma-dialectics, argumentation is considered as a complex and interlinked array of speech acts, which are directed towards fulfilling the ultimate goal of a critical discussion: the reasonable resolution of a conflict of opinion. Ideally, the resolution of a conflict of opinion consists of four dialectical stages, which are as follows: (1) the confrontation stage, (2) the opening stage, (3) the argumentation stage and finally, (4) the concluding stage. According to the ideal model for critical discussion, the discussants go through each unique stage in order to advance the resolution of their conflict of opinion. In the confrontation stage, arguers establish they have a difference of opinion, which they may decide to attempt to resolve in the opening stage. The argumentation stage is dedicated to providing arguments in support of the standpoints proposed by the arguers taking part in the critical discussion and in the concluding stage the parties determine whether their difference of opinion has been resolved and in who’s favour (van Eemeren & Grootendorst, 2004, pp. 59-62).

The argument from correlation to cause occurs in the argumentation stage of a critical discussion. In order to analyze this type of argumentation, it must be reconstructed in an argumentation structure that makes its standpoint and arguments fully explicit. Such a reconstruction makes clear what the precise relation between the arguments adduced in support of the standpoint is and which unexpressed premises may play a role in the inference the argument proposes. To that end, it is necessary to formulate an argument scheme for the argument from correlation to cause, which reflects the shape one would theoretically expect such an argument to have. This argument scheme primarily performs a heuristic1_{function: it}

1 Due to their nature, argument schemes can be considered inherently heuristic and normative. They can be employed to find arguments in everyday argumentation, but they also stipulate to some degree the conditions under which that type of argumentation can be considered acceptable. For example, an argument scheme with coordinative argumentation normatively implies that only the presence of all the arguments may constitute an

(9)

provides the analyst with a blueprint for the reconstruction of arguments from correlation to cause in actual language use. This constitutes the analytical component of the theoretical tool under construction.

The pragma-dialectical theory stipulates ten rules for a critical discussion (Van Eemeren & Grootendorst, 2004, pp. 190-196), which represent the conditions arguers must uphold in order to ensure the reasonable resolution of their difference of opinion. Any violation of these critical discussion rules constitutes a hindrance towards the reasonable resolution of the conflict of opinion and is considered a fallacious argumentative move. These rules for a critical discussion reflect the normative element of the pragma-dialectical theory of argumentation and allow for an evaluation of the reasonableness of argumentation in actual language use. As such, the pragma-dialectical rules for critical discussion make it possible to establish whether the use of an argument may be deemed within the bounds of a reasonable discussion aimed at resolving a difference of opinion. A combination of the pragma-dialectical rules for critical discussion and the reasonableness criteria for the argument from correlation to cause that will be set forth in Chapter 3 form the evaluative component of the theoretical instrument in development.

A pragma-dialectical analysis of argumentation also takes into account in which communicative context or activity the argumentation occurs in. Interlocutors engaged in a communicative activity are attempting to realise a particular institutional goal, which shapes and governs the argumentation that is commonly employed or even permissible in that specific activity. In increasing order of magnitude, argumentative exchanges that constitute concrete speech events fall within a certain communicative activity type, several of which constitute a genre of communicative activity. In turn, the genres of communicative activities may be further subsumed under a domain of communicative activities (van Eemeren & Grootendorst, 2004, p. 143). To some extent, the communicative practices that are instrumental in achieving the institutional needs of the communicative domain can be considered conventionalized. In other words, a communicative activity type represents a category of communicative activities that (partially) share a particular set of institutional goals, conventions and format regulations. Popularized scientific discourse constitutes such a communicative activity type, with its own set of institutional goals and conventions.

2.2 The institutional goals of popular scientific texts

(10)

Most popular scientific texts are situated between the realms of academics and journalism, which means they aim to fulfil institutional goals relevant to both these communicative domains. Popularized scientific discourse that is intended for a general public is also called

expert-to-layman communication, and constitutes a unique communicative activity type.2_As

van Eemeren notes, when attempting to distinguish communicative activity types from one another, it is important to describe “the specific goals they are to serve in order to fulfil their mission in realizing the institutional point of the communicative activity” (2010, p. 144). In my view, popular scientific discourse can be categorized among the communicative activity types that “prototypically involve the activation of more genres of conventionalized communicative practices” (van Eemeren, 2010, p. 144). This means popular scientific discourse functions as a hybrid communicative activity type. Table 1 below presents an overview of my formulation of the institutional goals popular scientific texts aim to achieve within the journalistic and academic domains.

Communicative

Activity Type Communicative Domain Institutional Goals

Popular scientific discourse

Scholarly communication

1. Providing an adequate and accurate representation of

scientific findings.

2. Persuading the audience of the presented interpretation of

scientific knowledge.

Journalistic communication

3. Simplifying the scientific source material in order to make it accessible to a wider audience.

4. Indicating the societal importance and significance of the

scientific knowledge reported on. Table 1: Institutional goals of the communicative activity type ‘popular scientific discourse’

The institutional goals of popular scientific discourse have some overlap to those of professional scientific publications. As mentioned in Chapter 1, the institutional goals of

2 Depending on the audience that is being addressed, there are a number of different categories of popularization that may be identified. For example, intra-scientific popularized scientific discourse focuses on informing other scientists from different fields, while an article in a popular scientific magazine focuses on an audience with some basic interest and knowledge regarding the topic (Shinn & Whitley, 1985, p. 19). Each of these categories of popularization can be considered its own communicative activity type. However, for the purposes of this paper only expert-to-layman communication is taken into consideration.

(11)

professional scientific discourse consist of (1) reporting informatively and persuasively on scientific progress, (2) the accumulation of scientific knowledge and (3) the critical examination of scientific claims by peers. As such, popular scientific texts share the institutional goal of reporting informatively on scientific progress and persuading an audience of a scientific interpretation. However, the goal of critical examination and evaluation of scientific findings is largely absent for popular scientific discourse, given the fact that it does not target a purely academic audience, but a much more general audience. The primary institutional goals of a popularized scientific article domain as a form of scholarly communication is to persuade a reader by means of a representative interpretation of the scientific source material (Goal 1 & 2, Table 1, p. 9) rather than subject the material to critical examination.

An important factor influencing the institutional goals of popular scientific discourse is that its content is bound to professional scientific material already in existence. Authors of popular scientific texts have an institutional responsibility to provide a representation of existing scientific knowledge that is both adequate and reasonable, although it does not have to be complete or exhaustive. If a popular scientific text strays too far from the source material, its author runs the risk of misrepresenting the scientific material it is based on, possibly leading to the popularized scientific article being rescinded. However, there is no absolute delineation of what constitutes a reasonable or unreasonable representation of scientific source material. Because authors of popular scientific material are not bound by an institutional constraint that requires a full substantiation of claims, they may simplify, distort or omit information in order to make a text more appealing and accessible to a broader audience. In doing so, a popular scientific author is not necessarily telling an untruth, but rather making a selection of information to present that serves his or her purpose – and thus the journalistic institutional goals (Goals 3 & 4, Table 1, p. 9) (Fahnestock, 1986, p. 281). Essentially, when creating a popular scientific text, an author maneuvers between the institutional goals of adequacy and representation, while also attempting to maximally achieve the institutional goals of simplifying the material and indicating its societal importance and relevance in order to ensure that is interesting to a broad public. As such, there exists a certain tension between the academic and journalistic institutional goals of popular scientific discourse, as it may not always be possible to maximally fulfil one without neglecting another.

(12)

In argumentation theoretical terms, professional scientific literature is directed towards a relatively particular audience, while popular scientific texts aim to reach a more universal audience (Perelman & Olbrechts-Tyteca, 1958, p. 35). The adaptation of scientific material to a larger and more heterogeneous audience generally involves relating to the material in a less technical vernacular and a considerably lower word count that can be devoted to a subject. The presumed absence of extensive education and experience with scientific terminology or procedure can lead to scientific knowledge being “presented as certain, important and incontrovertible, typically tied to concerns and issues which are likely to be close to everyday experiences or the widest possible public” (Shinn & Whitley, 1985, p. 17).

The communicative context in which the popularization takes place has an influential bearing on the underlying assumptions a writer of popular scientific discourse makes. A popular scientific article in a newspaper caters to a much wider and disparate audience than an article in a dedicated popular scientific magazine. The relative ‘cognitive distance’ between knowledge producers and the audience it is intended for greatly influences the level of simplification applied to the source material and the desire for further argumentation in support of claims on the basis of scientific knowledge (Shinn & Whitley, 1985, p. 19). This dimension of cognitive distance “incorporates differences in intellectual background, research skills and intellectual goals [and] refers to the extent of common experiences, competences and interests between specialist scientists and their audience” (Shinn & Whitley, 1985, p. 19). In other words, in the popularization context of professional academic publications to a lay-audience the cognitive distance between the knowledge producer and public is relatively high. This means that popularization of scientific material to such a broad and diffuse target audience will have considerable impact, as the source material must be adjusted to bridge the gap in cognitive distance between scientist and audience. As such, the audience that is targeted by popular scientific material is intimately connected to its institutional goal.

The adaptation of scientific knowledge from an academic audience to a lay-audience represents a great shift in discursive context. And as Fahnestock remarks, “with a significant change in rhetorical situation comes a change in genre, and instead of simply reporting facts for a different audience [...] the work of science journalism requires the adjustment of new information to an audience’s already held values and assumptions” (1986, p. 278-279). That is to say, professional scientific material is not simply reported on in popular scientific discourse, but is in fact recontextualized in the communicative activity type of popular science in order to ensure it fulfils both of its institutional goals. In that regard, a popular

(13)

scientific text is a reinterpretation of scientific source material that must be argued for. However, in the popular scientific context of informing and persuading an audience of scientific fact – rather than offering scientific findings for critical evaluation – there is often no room for many qualifications towards what is being communicated. Although a smaller, academic audience may expect an author to provide a full justification for his or her statements and take a tentative approach to inferential claims, the wider audience has much less need for this type of information. Instead, a large and heterogeneous audience is mainly interested in the potential application of research or its appeal to a sense of wonder or progress (Fahnestock, 1986, p. 279). This particular audience demand is often met by the use of vivid imagery and presenting scientific claims with greater certainty than they may actually posses. Additionally, while the significance of what is being communicated in a professional scientific paper is largely understood and made clear contextually, this does not hold true for many popular scientific articles, which generally explicitly indicate the value of the scientific knowledge they report. Such practices tie back to the institutional goal of relating the societal importance of the scientific source material to the target audience, which is more likely to be adequately fulfilled with the use of these discursive strategies.

(14)

3 The argument from correlation to cause

3.1 The statistical meaning of a correlation

Correlations fall within a broad class of statistical relationships that involve dependence and association between two or more variables. In its most common usage, a correlation between two variables refers to the extent to which these variables posses a linear relationship with each other. For example, the relation between the physical appearance of parents and the appearance of their children is a basic example of a correlation, since their relation is clearly dependent, but not completely linear. Correlations may also possess predictive capacities, as they can provide an indication of the likelihood a particular variable may occur. For instance, the weather conditions on a particular day are instrumental in assessing the required level of energy production a power plant should be prepared to meet. In that case, the correlation between extreme weather conditions and a corresponding increased demand for energy in order to heat or cool residences contains a causal relation. In other words, the weather conditions cause the increased need for energy, which means those weather conditions provide a solid indicator of energy demand. However, there are many other factors that may contribute to an increase in energy demand and extreme weather conditions do not always generate the same increase in energy demand, which may be the case during a holiday period. This means the overall relation of dependence between energy demand and weather conditions is still considered a correlation, while it has a causal relation embedded within its relation of dependence. In actuality, what is commonly subsumed under correlation is a broad range of different possible relations between variables.

When two variables or events – A and B, for example – are correlated, there are a number of possible relationships that may exist between those variables, which include the following options (Govier, 2005, p. 288):

- A causes B (direct causation) - B causes A (reverse causation)

- Both A and B are the consequences of a common cause C, but have no causal relation towards each other. (Third-cause variable, ‘standard’ correlation) - There is in fact no actual connection between A and B (coincidence)

(15)

Because of the disparate nature of what a correlation may actually signify3_{, a common}

statistical dictum is correlation does not imply causation. That is to say, on the basis of the fact that event A and B are correlated alone, one cannot draw a conclusion about the existence or the direction of a causal relationship between them. In order to establish whether any causal link between the observed variables exists, additional information is required. As such, the interpretation of a measured correlation as an indication of a certain causal connection is in fact a claim that requires argumentative support. If an arguer employs a correlation to validate the existence of a causal link, he or she can be reasonably expected to provide additional evidence that supports that particular interpretation over another. More specifically, an arguer claiming that a variable shares a causal relation with another may be required to adequately address the impossibility or unlikelihood of the other possible relations between those variables. In the case of an arguer claiming a direct causal relation – A causes B – on the basis of a correlation between them, he or she is using a specific type of argumentation that is also called an argument from correlation to cause (Walton, 1996, p. 71). However, given the fact that the correlation alone is not enough to substantiate a causal relation, the argument from correlation to cause should consist of multiple elements that must be adduced in defence of the causal claim in order to form a cogent inference. Section 3.2 to follow further elaborates on the argument scheme an argument from correlation to cause may take shape in.

3.2 Argument scheme for the argument from correlation to cause

In previous research concerning the use of correlation in argumentation, several formulations of the argument from correlation to cause have been put forward. However, there are considerable differences between these formulations and their theoretical stance towards the reasoning behind the argument from correlation to cause. Firstly, I shall critically discuss a number of the existing formulations by other scholars and briefly highlight the characteristics of their renditions of the argument scheme. Secondly, I present my own formulation of a pragma-dialectical instantiation of the argument scheme.

3 Theoretically, 3 more possible causal relations between the correlated variables could occur than the ones mentioned by Govier (2005, p. 288). A and B could also be the cause of one another, which is a cyclical causation. Also, A could cause a third variable C, which in turn causes variable B, called an indirect causation. Although these relations are more complex, they include a causal relation from A to B and have been accommodated in that category. Lastly, A and B may both cause a third variable C, which is actively conditioned on, which is an example of Berkson’s Paradox (Berkson, 1946) This category has been purposefully excluded, as it exceeds the theoretical scope of this thesis.

(16)

Douglas Walton (1996, p. 71-73) formulates the argument from correlation to cause – which he baptized as such – quite simply:

Premise: There is a positive correlation between A and B. Conclusion: Therefore, A causes B.

Walton adds a number of critical questions that are aimed at identifying whether the relation invoked can be considered reasonable or not. He notes that “quite often, arguments fitting this argumentation scheme are presumptively correct. However, this type of argumentation tends to be very weak in many cases, because other factors are overlooked” (Walton, 1996, p. 71-72). But even if the argumentation employed in an argument from correlation to cause is weak, “the argument may still be a reasonable one, [...] because the critical questions could possibly be answered with further investigation of the case” (Walton, 1996, p. 72). Thus, in Walton’s view, the argument from correlation to cause is not necessarily fallacious, but difficult to make inferentially valid. As stipulated in the previous section, presenting only the correlation between A and B provides insufficient substantiation for the existence or direction of a causal relationship. In my view, for this reason Walton’s formulation of the argument scheme is partially incomplete, as it does not incorporate all of the essential conditions necessary to substantiate a causal claim on the basis of a correlation in the argument scheme itself. Although Walton does integrate some of these requirements in his formulation of the relevant critical questions for the argument scheme, it would be worthwhile to implement the conditions in the argument scheme itself.

In another treatment of argumentation from correlation to cause, Johnson and Blair (1983, p. 121) subsume it under a broader category of arguments that employ a causal claim. As such, their treatment of the argument from correlation to cause departs from the category of post hoc and cum hoc fallacies, which are also called fallacies of questionable cause. Such fallacies have three identifying conditions: (1) there is a causal claim in the standpoint, (2) the protagonist fails to provide adequate evidence for the causal claim and (3) there are grounds to question the acceptability of the claim (Walton, 1996, p. 165). This interpretation of the argument from correlation to cause signals that Johnson and Blair do not consider such arguments inherently fallacious, seeing as a correlation – in combination with other arguments – could satisfy conditions (2) and (3). While Johnson and Blair provide further general criteria for the acceptability of arguments containing a causal claim, the specific requirements for an argument from correlation to cause are not addressed.

(17)

In a more recent educational textbook on logical reasoning, Trudy Govier notes that a “positive correlation may provide reason to suspect a causal relationship, but it does not provide good evidence that a causal relation exists” (2005, p. 288). Again, this interpretation seems to signal that she considers an argument from correlation to cause theoretically acceptable, but not on the basis of a correlation alone. Unfortunately, she does not specify further how a reasonable variant of the argument from correlation to cause may take shape. She does present a formulation of the fallacious argument from objectionable cause, which is closely related to the argument from correlation to cause:

A occurred. B occurred.

We can plausibly connect A to B in a causal relationship. Therefore A caused B.

However, the argument from objectionable cause differs from an argument from correlation to cause in two important respects. First, the co-occurrence of variables A and B is not the same as a correlation between A and B, and secondly, Govier’s use of the term ‘plausibly’ is focussed on a goal of convincing an antagonist, rather than a representation of the inferential steps necessary to provide adequate support for the causal claim in the standpoint.

Groarke & Tindale (2004, p. 303) formulate an extensive argument scheme for the argument from correlation to cause, which is focused on representing the additional theoretical elements necessary to infer a causal relation on the basis of a correlation, like so:

Premise 1: X is correlated with Y.

Premise 2: the correlation between X and Y is not due to chance.

Premise 3: the correlation between X and Y is not due to some mutual cause Z. Premise 4: Y is not the cause of X.

Conclusion: X causes Y.

This scheme appears to contain most of the necessary elements needed to substantiate a claim to causality, but it does not indicate the logical structure in which the premises support the standpoint. Also, the underlying assumptions an arguer uses to justify the inferential step from the premises to the conclusion is not yet represented in this version of the argument scheme. The argument scheme for the argument from correlation to cause proposed by Groarke and Tindale (2004, p. 303) will partially provide the theoretical basis for my own formulation of the scheme.

(18)

I have formulated a pragma-dialectical rendition of the scheme as follows: 1 A causes B.

1.1a A is correlated with B.

1.1b The correlation between A and B is not a coincidence.

1.1c The correlation between A and B is not due to a common cause. 1.1d B does not cause A.

(1.1a’-1.1d’) If premises 1.1a through 1.1d apply, we may assume that A causes B.

(1.1a’-1.1d’).1a If premises 1.1a through 1.1d apply, A is a necessary cause of B. (1.1a’-1.1d’).1b If premises 1.1a through 1.1d apply, A is a sufficient cause of B. In this form, the argument from correlation to cause can be considered a subtype of causal

argumentation (van Eemeren et al., 2002, p. 100-102). The standpoint under 1 that contains

the claims to a causal relation between A and B is defended with a number of coordinative arguments, which only together constitute a complete substantiation of the claim. These arguments are provided in anticipation of critical questions that “relate to specific conditions under which a particular fact can be called the ‘cause’ of another fact” (Wagemans, 2016, p. 103). In support of the standpoint containing the causal claim, four coordinative arguments pertaining to the observed correlation between A and B can be adduced. Argument 1.1a contains the argument that there is a positive correlation between A and B. Argument 1.1b excludes the possibility of the measured correlation being coincidental. In turn, argument 1.1c is adduced in order to exclude the possibility of a third variable causing both A and B. Argument 1.1d eliminates the reverse and cyclical types of causal relations between the variables. These four coordinative arguments are aimed at the elimination of the existence of any other relation between the correlated variables A and B, in order to support the standpoint containing the claim to a direct causation from A to B. The implicit linking premise under (1.1a-1.1d’) reflects the arguer’s assumption that the arguments mounted in defence of the direct causal relation provide an adequate substantiation of the causal claim. In theory, this argument scheme for the argument from correlation to cause represents all the necessary elements for a valid and acceptable inference, in which arguments 1.1a through 1.1d each provide a piece of the puzzle. Finally, the implicit linking premise may be further supported by two coordinative argument relating to the sufficiency and necessity of the causal relation between A and B. Argument (1.1a’-1.1d’).1a states that the causal relation from A to B is necessary, which means that “without the occurrence of the cause, the effect does not occur either” (Wagemans, 2016, p. 103). Argument (1.1a’-1.1d’).1b states that A is a sufficient cause of B, which means that no other factors – such as a third variable – are needed for the

(19)

effect to occur. Section 3.3 to follow will stipulate the criteria for the reasonableness of arguments from correlation to cause.

3.3 Formulating reasonableness criteria for the argument from correlation to cause

The criteria for reasonable arguments form correlation to cause can be formulated as a number of critical questions that examine the argument’s acceptability. Since the argument from correlation to cause is a subtype of causal argumentation, the central critical question that must be answered is does A really lead to B? In the specific case of reasoning from a correlation to a causal relation, reaching an answer to this question can be achieved by posing a number of sub-questions. These critical questions can be subdivided into two main categories: critical questions aimed at critically examining the propositional content of the argument and critical questions aimed validating the inferential structure of the argument. I have formulated five critical question that pertain to the propositional content of the argument as follows:

Critical Question 1: Is there a positive correlation between A and B?

Critical Question 2: Is the correlation between A and B significant?

Critical Question 3: Is the correlation between A and B not due to mere chance?

Critical Question 4: Can we rule out that there is a third variable responsible for the observed correlation between A and B?

Critical Question 5: Can we rule out that B does not cause A?

The first three critical question are aimed at establishing whether there is indeed a positive correlation between the variables, and if that correlation is significant and not due to coincidence. Question 1 can be answered by producing the data the claim to a correlation between A and B is based on, while Question 2 and 3 require a substantiation of the interpretation of that data as a meaningful and significant correlation. For instance, when two traits of human behaviour are observed to be correlated, Question 2 may be answered by indicating that the observed correlation between those variables is strong enough to be significantly different from the norm and thus may be considered representative for the entire

(20)

population. Question 3 is generally addressed by the use of a repeated measurement design, which allows the exclusion of a coincidental correlation on the basis of observing the correlation multiple times.

Question 4 and 5 aim to further complete the necessary grounds for an inference from correlation to cause. The exclusion of a third variable that may be responsible for the measured correlation is one of the most difficult conditions to meet, since many causal relations are open to alternative explanations. This question may be addressed by attempting to eliminate the most plausible other options for the observed correlation between A and B. In turn, Question 5 is not always a necessary critical question, since the exclusion of a causal link from B to A may be excluded on the grounds of the propositional content of the variables. For example, in a correlation between the speed at which windmills spin and the strength of the wind, a reverse causation is logically excluded since windmills can never be the cause of the wind. However, in the case of a correlation between more complex variables – such as unemployment and slow economic growth, for example – a reverse causation is equally plausible, as either could potentially cause the other. In more complex cases, Question 5 must be answered with a further substantiation for the direction of the causal relation. I have also formulated three possible critical questions that are aimed at examining the legitimacy of the inference made in an argument from correlation to cause:

Critical Question 6: Do premises 1.1a through 1.1d provide an adequate substantiation of a causal relation from A to B?

Critical Question 7: Do any exceptions to the causal relation apply, and if so, can they be clearly delineated as exceptions?

Critical Question 8: When a premise in the argument is missing, does this prevent a valid inference to a causal relation?

Question 6 is centred on the evaluation of the implicit linking premise of the argument from correlation to cause, which is the assumption that the coordinative arguments 1.1a through 1.1d in fact constitute an adequate defence of a causal claim. In some cases, it may be necessary to adduce additional arguments in order to fully substantiate the causal claim in the standpoint. For example, in the case of an indirect causation between A and B – this occurs

(21)

when A causes C, which in turn causes B – the argument’s protagonist may be expected to further elaborate on the intervening variable (Walton, 1996, p. 72). Question 7 aims to uproot any exceptions to the causal relation that is invoked in the standpoint and requires an explanation for their exceptional status. In response, the argument’s protagonist could answer by indicating the ranges in which the causal relation between variable A and B does hold. Lastly, Question 8 is an especially important question to consider in the case of partially incomplete arguments from correlation to cause. As mentioned with regard to critical question 5, sometimes not all of the arguments listed in the argument scheme for the argument from correlation to cause are necessary for the argument to be acceptable. Also, as Walton points out, an argument from correlation to cause overlooking or failing to account for one or more of these critical questions does not necessarily entail that it is fallacious (1996, p. 72). As such, determining whether an argument from correlation to cause is reasonable must proceed on the basis of both critical questions relating to the propositional content of the argument, as well as the critical questions that examine its inferential acceptability.

(22)

4 Correlation to cause arguments in popular scientific discourse

4.1 Introduction

This chapter is dedicated to the analysis and evaluation of arguments from correlation to cause as they occur in popularized scientific articles. Two professional scientific publications putting forward a correlation that have been popularized in the popular press or media have been selected for this analysis and evaluation. In the first place, each professional scientific publication will be briefly introduced, after which I will provide a reconstruction of the argumentation structure of the scientific claim made in the publication. In this step, the standpoint and supporting arguments in the professional publication can be made explicit. In the second place, the popularized counterpart of the publication is discussed and reconstructed in an argumentation structure as well. In doing so, the elements present in the arguments from correlation to cause that are employed in these illustrative examples can be made explicit. Finally, in the third place, I will compare and contrast my reconstructions of the argumentation in the professional and popular articles. This allows for an evaluation of their reasonableness in two respects: (1) the acceptability of their representation of the source material and (2) the inferential validity of their use of the argument from correlation to cause.

4.2 ‘Myopia and ambient lighting at night’

The scientific magazine Nature is a well-known and prestigious medium for the publication of professional scientific articles. Generally, scientific articles published in the magazine are shortened and focussed on a slightly broader audience than regular professional scientific discourse. However, due to the stringent reviewing procedure for publication in the magazine, which includes a peer reviewing panel, the medium remains within the realm of professional scientific discourse. In an article in Nature magazine (given in full on page 25) researchers Quinn et al. report a correlation between myopia – better known as nearsightedness – in young children and the presence of night-lighting during sleep. In their words, they “find a strong association4_{between myopia and night-time ambient light exposure during in sleep in}

children before they reach two years of age” (Quinn et al., 1999, p. 113). The authors are

4 Association and correlation are often used synonymously in everyday language, but refer to 2 technically different concepts. An association is a general relationship, in which the variables provide some information about one another, while a correlation can be seen as a more specific subtype of association in which the degree of the linear relation between the variables is also taken into account. However, as is true for correlation, “association should not be confused with causality; [...] associations can arise between variables in the presence and absence of a causal relationship” (Altman & Krzywinski, 2015, p. 899).

(23)

careful to provide several qualifications to this observed association, and indicate that “it does not establish a causal link” and that “the lack of information on whether parental lighting preferences independently associate with other factors known to be correlated with myopia” (Quinn et al., 1999, p. 113). In other words, this indicates that the authors have not yet been able to conclusively rule out a third variable being the cause of the observed association between the variables or a reverse relation of causation between them, and thus cannot yet reach a conclusion on the existence or direction of a direct causal relation. This means the arguments 1.1b, 1.1c and 1.1d in the argument scheme for the argument from correlation to cause cannot be provided yet in support of a causal claim. In spite of this qualification, the authors do suggest that “it seems prudent that infants and young children sleep at night without artificial lighting in the bedroom” (Quinn et al., 1999, p. 113), which forms their ultimate pragmatic standpoint concerning their findings. I have reconstructed the argumentation employed in this professional scientific article as follows:

1 Infants and young children should sleep without artificial night-lighting. 1.1 Artificial night-lighting possibly leads to the development of myopia. (1.1)’ And myopia is undesirable.

1.1.1a Myopia in young children and night-lighting are strongly associated. 1.1.1b Night-lighting being the cause of myopia is plausible.

1.1.1b.1 The eyelids of human adults and infants transmit some visible light. 1.1.1b.2 The scotopic retinal sensitivity of infants is relatively good compared

with that of adults.

1.1.1b.3 The duration of the daily light period has been shown to affect eye growth in chicks.

The pragmatic standpoint is supported by an argument that there is a possible causal connection between night-lighting and myopia, which is further supported by coordinative argumentation regarding the observed association between the variables and the reasons a causal relation between the variables might be plausible.

The article received tremendous attention in the popular press and was popularized in a number of media. Among them, CNN’s internet-based Health section – which produces popularized scientific content – covered the scientific findings in a report claiming that “young children who sleep with a light on may have substantially higher risk of developing nearsightedness as they get older” (Etheridge, May 13, 1999). The article also presents the correlative data between myopia and night-lighting put forward by Quinn et al. and provide the explanation for the plausibility of a causal relation between them. My reconstruction of their argumentation on the basis of the professional scientific is as follows:

(24)

1 Young children who sleep with a light on may have substantially higher risk of developing nearsightedness as they get older.

1.1 Sleeping with a light on and the development of myopia are found to be connected.

1.1.1a The survey was an extension of research in chicks demonstrating that the relative proportions of light and dark greatly affected eye growth and refractive development.

1.1.1b Even low levels of light can penetrate the eyelids during sleep, keeping the eyes working when they should be at rest.

It is clear that CNN’s representation of the original scientific findings is influenced considerably by the institutional goal it aims to fulfil within the journalistic domain of communicative activity. The authors choose to report on the possibility of a causal connection that is “tied to concerns and issues which are likely to be close to everyday experiences or the widest possible public”, as mentioned earlier (Shinn & Whitley, 1985, p. 17). Given the fact that the subject is of broad interest and the findings could indicate a potential risk to health that is relatively easily preventable, this presentation of the professional scientific findings is sure to draw significant attention.

From an evaluative perspective, CNN’s formulation of night-lighting possibly leading to a “substantially higher risk of myopia” does obscure what the exact connection between the two variables is. In other words, if this claim means ‘night-lighting and myopia occur together frequently’, this is a formulation that is allowed on the basis of a correlation alone. However, if the claim is understood as ‘night-lighting causes myopia’, that constitutes the use of a correlation – or association – to defend a causal claim. Such ambiguities may work in favour of the arguer making use of an argument from correlation to cause, as they mask the precise claim that is intended. Ambiguity of language violates the tenth pragma-dialectical rule for critical discussion, which states that arguers may not use insufficiently clear formulations in their argumentation, since this hinders a possible antagonist in distinguishing its intended meaning. In this particular context, it is clear the possibility of a causal effect of the former on the latter is suggested, which is consequently modally modified to indicate its possibility, rather than certainty.

There is no complete and adequate substantiation available for this suggested causal relation between myopia and night-lighting. Still, the claim in CNN’s article to its possibility is not technically incorrect: it is among one of the options the measured association signifies. Moreover, Quinn et al. also explicitly hypothesize the possibility of this causal relation in the

(25)

professional scientific article reporting their findings, and offer further subordinate argumentation for this potential causal relation. In that respect, the representation of the original scientific findings by CNN’s popularized article can be considered adequate and reasonable. The article also mentions the fact that several of the necessary elements for a complete and acceptable causal explanation are missing as of yet, and they have adjusted the strength of their causal claim accordingly. In other words, in response to not being able to answer critical question 4 (Can we rule out a third variable?) and critical question 5 (Can we

rule out that B does not cause A?) the authors have refrained from explicitly claiming a direct

causal relation between night-lighting and myopia on the basis of the scientific findings. The modal modification of the claim is also a solution to the fact that critical question 8 (Does a

missing premise prevent a valid inference to a causal relation?) must be answered with ‘yes’

for the argument form correlation to cause regarding night-lighting and the development of myopia in young children.

CNN’s guarded claim on the basis of the evidence presented in Quinn et al. (1999) can be considered inferentially reasonable, because their inference essentially consists of establishing the causal relation as one of the possible options. In its hypothetical character, CNN’s argument from correlation to cause is mitigated by a hedged claim, which makes up for the lack of a full substantiation for the direct causal relation. The importance of the other arguments necessary to infer direct causation from correlation became clear in an advanced study of the effects of ambient lighting on myopia. No direct cause between the two variables under investigation could be established, but myopia was proven to be strongly hereditary and parents with myopia were much more likely to leave a night-light on in the children’s room. On the basis of these further data it could be established that it was actually a third variable that was responsible for the measured association between night-lighting and myopia in young children: parental myopia.

(26)

(27)

4.3 ‘Facebook usage on smartphones and gray matter volume of the nucleus accumbens’

The professional scientific and peer-reviewed journal for Behavioural Brain Research has recently published a study focussed on the relation between Facebook usage and brain structure by Montag et al. (2017, pp. 221-228). The study used a remote smartphone activity tracker to record the actual Facebook activity of 62 participants and correlated this data with the volume of the nucleus accumbens in the brain. The nucleus accumbens is a region in the brain also known as its ‘reward centre’, and likely plays an important role in a wide range of positive experiences. The resulting data of the study indicate that “in particular higher daily frequency of checking Facebook on the smartphone was robustly linked5_{with smaller gray}

matter volumes of the nucleus accumbens” (Montag et al., 2017, p. 221). In this case, the correlation was negative: “participants who opened the Facebook application more frequently and participants who stayed on Facebook longer had smaller nuclei” (Montag et al., p. 224). However, the authors stipulate that “given the cross-sectional design of the present study, it remains to be elucidated whether lower volumes of the accumbens constitute a factor for increased social media use or whether it results as a consequence of higher usage” (Montag et al., 2017, p. 226). In other words, the study has found the variables to be strongly correlated, but has not been able to establish the existence or direction of a possible causal link between the measured variables. I have formulated a reconstruction of this professional scientific publication’s most relevant argumentation below:

1 The nucleus accumbens could be involved in online social media usage. 1.1a Higher daily frequency of checking Facebook is robustly linked with

smaller gray matter volumes of the nucleus accumbens.

1.1a.1 Left and right nucleus accumbens volume is negatively correlated with Facebook usage.

1.1b The results from other neuroscientific studies suggest that the rewarding aspects of Facebook usage could result in over-usage of this platform. 1.1c Data show that the frequency of daily Facebook checking is also

associated with self-reported tendencies towards online social network addiction.

5 ‘Robustly linked’ means as much as ‘strongly correlated’ in this context. In relation to this discursive phenomenon, Govier points out that “the word linked [...] enables writers to skirt over the distinction between correlational evidence and causal claims conclusions, because it has several meanings that are rarely distinguished from each other. To say that Q is linked to H suggests, but does not assert, a causal connection between Q and H. Because the connection is only suggested, the demands of a rigorous causal argument are avoided. But because the connection is suggested, there is a clear implication that something more than a correlation has been established” (2005, p. 290, italics in original).

(28)

The claim Montag et al. (2017) make on the basis of their observed correlation is careful, and states no more than a possible involvement of the two variables investigated. The claim is supported by the observed negative correlation between the variables and other coordinative argumentation that suggests other evidence for the plausibility of the connection.

In the Health section of Rush Hour Daily news, these professional scientific findings are popularized in an article that claims “Facebook addiction causes decrease in brain’s gray matter” (Ulaky, May 29, 2017). The article goes on to report the measured negative correlation between Facebook usage and nucleus accumbens volume, as well as another study concerning the use of Facebook and markers of general well-being. Furthermore, the popular scientific article contends that researchers Montag et al. used magnetic resonance imaging (MRI) scans, which “ultimately showed the negative reaction too much social activity might have” (Ulaky, May 29, 2017). My reconstruction of the correlation to cause argumentation employed in the popular scientific article is as follows:

1 Increased Facebook usage leads to a decrease in gray matter in the brain.

1.1 A study published in the journal Behavioural Brain Research has discovered the negative side effects and psychological repercussions of Facebook “addiction”.

1.1.1a People who frequently check the application on their smartphones are likely to have less gray matter the ‘reward centres’ of their brains. 1.1.1a.1 Higher daily frequency of checking Facebook is robustly linked with

smaller gray matter volumes of the nucleus accumbens.

1.1.1b MRI-scans of the test subjects showed the negative reaction too much social activity might have.

From an evaluative point of view, the reconstructed argument from correlation to cause in

Rush Hour Daily’s article can be considered fallacious on two accounts:

In the first place, the representation of the scientific source material the article is based on is inadequate and unreasonable towards the original authors. In argument 1.1, the author of the popular scientific article states that researchers Montag et al. have “discovered the negative side effects and psychological repercussions” to the use of Facebook, which signals a relation of cause and effect between the variables. However, Montag et al. explicitly indicate they have not yet discovered the substantive nature of the correlation they have observed (2017, p. 226). In that regard, the popularized article commits a strawman fallacy, by implying the study by Montag et al. has claimed a causal relation between the correlated variables. In doing so, the popularized article violates the third pragma-dialectical rule for

(29)

critical discussion, which states that an arguer may not distort or misrepresent the standpoint of another party.

In the second place, the inference from correlation to cause made in the popular scientific article is unacceptable, since it is improperly substantiated with evidence other than a correlative argument. Argument 1.1.1b is used in the article to support the interpretation of the original study as a relation of cause and effect. In reality, the MRI-scans of the test subjects’ nucleus accumbens indicated the correlation of its volume and Facebook usage, but no causal relation. As such, the correlative argument 1.1.1a and the incorrect interpretation of the data in argument 1.1.1b entail the fallaciousness of this argument from correlation to cause. The author commits the fallacy of cum hoc ergo propter hoc, by making the assumption that the observed variables co-occurring allows for a causal conclusion. The reasoning in the popularized article fails to account for critical questions 4, 5, 6 and 8, since it does not effectively rule out alternative explanation before making a causal claim and does not acknowledge that the remaining premises alone do not adequately substantiate the claim. This use of the argument scheme for the argument from correlation to cause violates the seventh pragma-dialectical rule for critical discussion: the argument scheme rule. This rule states that an arguer may not consider a standpoint conclusively defended if that defence does not take place by means of an appropriate argument scheme that is correctly applied (van Eemeren & Grootendorst, 2004, p. 194).

In my opinion, this example demonstrates the tension between the academic and journalistic institutional goals a popular scientific article aims to fulfil. By presenting the observed connection between Facebook usage and gray matter volumes in the brain as a causal relation, this article has very likely achieved both journalistic institutional goals of simplification and indicating significance. This is further abetted by the nature of the topic, which is of wide application and interest, which now also generates a cause for concern. Although popular science writers “must usually be explicit about the value of scientific discoveries, [as] they cannot rely on the audience to recognize the significance of information” (Fahnestock, 1986, p. 279), in the example this has been achieved at the expense of the institutional goal of adequacy and representation. In sum, the author of this particular popular scientific article appears to have placed too great an emphasis on fulfilling the institutional goals situated within the journalistic domain of communication, thereby overstepping the reasonable bounds for critical discussion.

(30)

5 Conclusion

With this thesis, I have endeavoured to develop an argumentation theoretical instrument for the analysis and evaluation of arguments from correlation to cause in a popular scientific context. The distinction between correlation and cause plays an important role in the process of obtaining valid and substantiated scientific explanations for the relation between observed phenomena. While correlations may provide an important hint to the existence of a causal relation, they do not license a causal claim on their own. An acceptable argument from correlation to cause must contain a number of additional elements that support the inferential step from a correlation to a relation of cause and effect. However, in the context of popularized science, the argument from correlation to cause is employed in an incomplete or inadequate manner with some regularity. In order to be able to investigate these arguments from correlation to cause in popular scientific discourse, I aimed to complete three theoretical steps: (1) a communicative characterization of popular scientific discourse, (2) the formulation of an argument scheme for the argument from correlation to cause and (3) the stipulation of the criteria for its reasonableness.

I have provided a characterization of popular scientific discourse as a unique and hybrid communicative activity type. Popular scientific discourse aims to fulfil institutional goals within both the domains of journalistic and scholarly communication. On the one hand, popularized scientific texts are aimed at providing a convincing and accurate representation of the scientific material they are based on. On the other hand, they intend to simplify the scientific source material in order to make it accessible to the public at large and make explicit its societal importance. Ideally, these institutional goals are all met simultaneously, but this may be complicated by the difficulty in providing an interpretation of scientific knowledge that is relatively simplistic and accurate.

My formulation of the argument scheme for the argument from correlation to cause reflects the arguments that should expectedly be present in an inference from a correlation to a relation of cause. The primary governing factor for the shape of an argument from correlation to cause is the requirement to exclude the other possible causal relations a correlation might entail. As such, the argument scheme for an argument from correlation to cause consists of a number of coordinative arguments that together constitute sufficient and adequate grounds to allow for the inference of a direct causal relation. In order to critically examine the acceptability of an argument from correlation to cause, one may pose several critical questions. These questions can be subdivided in questions concerned with the propositional