Theory Figures and Causal Claims in Sociology

Abstract

When sociologists examine the content of sociological knowledge, they typically engage in textual analysis. Conversely, this paper examines the relationship between theory figures and causal claims. Analyzing a random sample of articles from prominent sociology journals, we find several notable trends in how sociologists both describe and visualize causal relationships, as well as how these modes of representation interrelate. First, we find that the modal use of arrows in sociology are as expressions of causal relationship. Second, arrow-based figures are connected to both strong and weak causal claims, but that strong causal claims are disproportionately found in U.S. journals compared to European journals. Third, both causal figures and causal claims are usually central to the overarching goals of articles. Lastly, the strength of causal figures typically fits with the strength of the textual causal claims, suggesting that visualization promotes clearer thinking and writing about causal relationships. Overall, our findings suggest that arrow-based figures are a crucial cognitive and communicative resource in the expression of causal claims.

Introduction

Typically, when sociology itself is the object of study, scholars engage in a textual analysis, focusing on the verbal claims that sociologists have made in order to identify patterns and trends within the discipline (e.g., Abend et al., 2013; Maryl & Wilson, 2020; Vaidyanathan et al., 2016). However, sociologists are not bound to a single medium, using both words and figures (among other mediums) to convey their ideas. As such, a few scholars have found that analyzing the kind of figures sociologists use can also generate important insights; locating affinities between previously incommensurable schools of thought (Bargheer, 2021), providing deeper understanding of theoretical logic (Baldamus, 1992) as well as establishing the grounds for scrutinizing and improving the practices of theorizing and theory visualization within sociology (Brett et al., 2020; Lynch, 1991; Silver, 2020; Swedberg, 2016; Turner, 2010). However, to our knowledge scholars have yet to systematically analyze the relationship between text and figures within sociology.

This relationship is vitally important for sociological practice, especially in that it is now commonplace to have visual representations embedded in textual descriptions. In an early critique of theory visualization, Lynch (1991) argued that most of the theory figures used by sociologists are merely scientistic ornaments which do little beyond the text save create the “impression of rationality”. Conversely, more recent work suggests that, for better or worse, theory figures often do things that their surrounding text simply cannot; they can make our work clearer, more memorable, and more easily communicable than words alone, but they can also easily mislead readers and outstrip the carefully constructed textual claims surrounding them (Brett et al., 2020; Swedberg, 2016; Turner, 2010). To substantiate and contextualize these discussions, what is required is a systematic analysis of how sociological texts and figures interrelate.

This paper focuses specifically on the textual description and visual representation of causality. Causality is a particularly compelling and instructive case, not only because it is a fundamental concern within sociological analysis, but because establishing causality is both a language-bound and visual affair (see Abend et al., 2013; Elwert, 2013; Gangl, 2010; Morgan & Winship, 2015; Pearl, 1998). On the one hand, sociologists are almost unavoidably tasked with describing the nature of the relationship between variables and are typically quite cautious and precise with the language they use when making causal claims (Abend et al., 2013; Vaidyanathan et al., 2016). On the other hand, sociologists regularly use arrow-based figures to illustrate causal relationships (see Brett et al., 2020; Elwert, 2013; Pearl, 1998). While these may be carefully represented in the context of formal causal modelling, they may be less thoughtfully constructed in our more general theoretical efforts, which often produce unclear and under-specified visualizations that may or may not be interpreted as causal (Brett et al., 2020). This contrast between the care and vigilance with which sociologists describe causal relationships and their more relaxed approach to theory visualization suggests a potential disjuncture in the communication of causality.

What is the relationship between arrow-based diagrams and causal language within sociology? We answer this question by analyzing the figures and causal claims in sociology articles, drawing on a random sample of articles from seven prominent sociology journals. Our findings revealed several important trends in how sociologists both describe and visualize causal relationships, as well as how these modes of representation interrelate. First, we find that figures with arrows are usually connected to causal claims within the text, suggesting that the modal use of arrows in sociology are as expressions of causal relationship. Second, the textual causal claims connected to these figures are a relatively even mix of strong and weak causal claims, but strong causal claims are disproportionately found in U.S. journals compared to European journals. Third, in general both causal figures and causal claims are central to the broader aims of the articles as a whole. Lastly, the strength of causal figures typically fits with the strength of the textual causal claims. Overall, our analysis suggests that theory figures are a crucial cognitive and communicative resource in the development of sociological claims.

Literature Review

For many sociologists, establishing causal relationships is the central aim of sociology, but the grounds upon which sociologists are able to make causal claims (if at all) is thorny, both philosophically and methodologically. In recent years this has spurred sociologists to develop and refine how we think about cause and effect: quantitative sociologists have worked to clarify the requirements for valid causal inference (see Gangl, 2010; Morgan & Winship, 2015); historical sociologists have differentiated their broad approaches to causal investigation (Ermakoff, 2019); ethnographers have specified how to construct robust causal claims (Tavory & Timmermans, 2013) and theorists have developed the nature of causal mechanisms and the properties of larger causal chains (Gross, 2009, 2018), establishing them as central to empirical investigation (see Hedström & Ylikoski, 2010). At present, we have several broad perspectives or models for approaching causality within sociological research (see Gangl, 2010; Hirschman & Reed, 2014; Small, 2013).

Rather than debating causality itself, some sociologists have chosen to empirically examine the way sociologists themselves make causal claims in published articles, and how these claims vary by the journals they are published in, the type of data and methods the articles employ, and even national variation in causal claims (Abend et al., 2013; Vaidyanathan et al., 2016). This work has found that sociologists infrequently use direct causal language, and instead rely on metaphors and euphemisms of causality; that quantitative work makes stronger and more direct causal claims than ethnography; and that strong and central causal claims appear more frequently in U.S. generalist journals, amongst other things (Abend et al., 2013; Vaidyanathan et al., 2016). These findings are said to demonstrate the “epistemic machinery” or “thought style” underpinning sociological claims (Abend et al., 2013: 640–641; Vaidyanathan et al., 2016: 5).

While the analysis of text provides some insight about how sociologists think about and communicate causality, the representation of causality in sociology is also highly visual. This is clearest in the visualization of causal models, which has been essential to establishing causal relationships in sociological research. Historically, sociologists have relied heavily on path diagrams, which are a qualitative expression of causal relationships using arrows and variables. Path diagrams were developed by the geneticist Sewall Wright in the 1920’s and 1930’s as a means of identifying, interpreting, and understanding a series of causal relationships (Griesemer, 1991; Wright, 1934). The graphing of causal models garnered interest in sociology starting in the 1960’s with the rise of path diagrams for linear causal modelling (e.g., Blalock, 1964; Duncan, 1966, see Bielby & Hauser, 1977). However, in recent years the focus on causal modelling has shifted toward directed acyclical graphs (DAGs) which emerged from path diagrams (Elwert, 2013; Elwert & Winship, 2014). DAGs are visual representations of qualitative causal assumptions in the data-generating process (Elwert, 2013; Elwert & Winship, 2014; Pearl, 2001). The three graphical elements in DAGs are variables, arrows (indicating possible direct causal effects), and non-arrows (representing a strong assumption of no causal effects) (Elwert, 2013). Importantly, throughout their history, path-based models have all explicitly invited causal interpretation (Blalock, 1964; Duncan, 1966; Elwert, 2013; Elwert & Winship, 2014; Griesemer, 1991). Both path diagrams and DAG’s support the ambitions of much of mainstream sociology: to establish the absence or presence of causal relationships between or among a set of bounded and discrete variables.

It is no coincidence that both path diagrams and DAG’s use arrows to visualize causality, because arrows are well suited to the expression and interpretation of causality. In two experiments on interpreting and producing diagrams of mechanical systems, Heiser and Tversky (2006) found that arrows were connected to functional information (related to sequences of actions or events, i.e., temporal, dynamic, and causal processes) rather than structural information (parts and their relations or arrangement). Depicting a function often entails “depicting movement, changes of state, forces, goals, and outcomes” (Heiser & Tversky, 2006: 582), and arrows supported descriptions based on motion and causes (see also Huey et al., 2021). This suggests that “arrows added to diagrams will be readily interpreted as conveying change, movement, or causality” (Heiser & Tversky, 2006: 589–590).¹

For sociologists, the ease with which arrows are interpreted as expressing causal relationships make them incredibly useful, but also makes them a potential source of unwanted inferences. Outside of the traditions of causal modelling discussed above, theory visualizations often use arrows in an ad-hoc manner, without established conventions or explicit graphical rules to rely on for construction or interpretation. Brett et al. (2020) found that sociologists often use arrows to represent relationships between variables, but rarely do they provide reading rules or legends to guide their visual interpretation. As a result, the meanings of arrows in theory visualization are often vague, making it unclear whether or not the arrow is meant to convey a causal relationship. This undisciplined use of arrows, combined with the typical interpretation of arrows revealed by Heiser and Tversky’s research, suggests that arrows in sociological visualizations are particularly likely to be misinterpreted as expressing causal relations.

Data and Methods

To analyze the relationship between arrow-based figures and causal claims, we draw from a large database of figures from seven leading sociology journals, including three generalist journals (The American Journal of Sociology (AJS), The American Sociological Review (ASR), and The European Journal of Sociology (EJS)) and four theory journals (Sociological Theory (ST), Theory and Society (T&S), The European Journal of Social Theory (EJST), and Theory, Culture, and Society (TCS)). We collected all issues of these journals published until 2015, and randomly selected 60 articles from each journal for analysis (save for EJST, which had fewer than 60 articles with visuals). We coded only those articles that used figures containing arrows.

We coded articles based on six features (see Appendix). (1) Whether the article’s text made a causal claim pertaining to the relationship specified by the arrows used in the figure. (2) The strength of the causal claim in the article’s text. (3) Whether the specific causal claim represented in the diagram was one of the paper’s primary aims. (4) Whether or not the strength of the causal language in the text fit the strength of the causal relationships as represented by the figure. (5) Whether causal arguments or claims were central to the papers’ explicitly stated aims. (6) Whether the figure was a data visualization or a theory visualization.² The first three codes were borrowed from Abend et al. (2013) and adapted to the study of visuals as well as text, while the latter three codes were added and refined based on our independent coding of a small subset of articles and a comparison of the results. This allowed us to discuss our coding decisions, clarify our interpretation and definition of each code, and identify previously unaccounted for variables that necessitated the addition of new codes.

After our preliminary rounds of coding, we coded the rest of the data in full. Coding was conducted independently by three coders, the two authors and a research assistant. While our preliminary coding and discussion was intended to establish consistency across coders, the codes applied in our analysis require subjective and occasionally challenging assessments about the text, figures, and articles as a whole. Because of the complexity and interpretive nature of the coding, we ensured that every article was coded independently by multiple coders. A comparison of the results revealed that while coders tended to be consistent in assessing the more straight-forward features of the articles (e.g., whether figures were theory visualizations or data visualizations), they were less consistent on those that required major interpretive judgement (e.g., whether causality was central to the article as a whole). Because of these interpretive differences, we average out the counts across the two coders of each article, and we use percentages not to establish precise figures, but simply as a means to locate broad patterns in published sociological work.

Findings

The Correlation Between Arrows and Causal Claims

Because of the ease with which arrows are interpreted as expressing causal relations, our first concern was whether or not the text in articles made causal claims pertaining to the relationship specified by the arrows in their figures. Overall, we found that the majority (~60%) of arrow-based figures are connected to claims that explicitly state the directionality of their relationships, rather than using the language of “correlation” or “association,” or simply leaving these relationships undiscussed. As such, arrows are quite often used to express causal relations, making them part of the “epistemic machinery” (Abend et al., 2013) of sociological claims, and important for how sociologists think about and communicate causality. This finding alone should give pause to sociologists using arrows for anything other than expressing causal relations, because it means that arrows are not only usually perceived as causal, but also institutionally expected to express causality.

Importantly, this does not mean that the remaining portion (~40%) of articles neglected discussing causal arrows or exclusively used the language of correlation when discussing them. Oftentimes there were no explicit causal claims made because the arrow itself was clearly used to signify something other than causal relations. Within sociology, arrows are regularly used to visualize the passage of time, hierarchy in an organization chart, the flow of information, or for far more generic purposes like labelling. For example, in Fig. 1a, Wernick (1983) clearly uses arrows to represent signification, and in Fig. 1b, Nielsen (2004) uses arrows to signify both the level of technological advancement as well as common patterns of societal evolution. In such instances, it is clear that the arrow signifies something other than causality, and is therefore logically unaccompanied by textual causal claims.

Fig. 1

Wernick’s a principal lines of signification in the Eves ad, and Nielson’s b visualization of Gerhard Lenski’s ecological-evolutionary taxonomy of societies

Having established the correlation between arrow-based figures and causal claims, our second concern was the nature of these causal claims. This is particularly interesting because, while arrows are readily interpreted as causal, sociologists often have an aversion to causal language, relying on metaphors and euphemisms rather the explicitly using the word “cause” or “causal” (Abend et al., 2013; Vaidyanathan et al., 2016; see King et al., 2021: 76). Following Abend et al. (2013), we first coded for the strength of causal claims. Strong causal claims make definitive statements about the relationship between X and Y, using language like “cause,” “affect,” “produce,” and “determine”. Conversely, weaker causal claims are more ambiguous or euphemistically presented, relying on weaker causal language like “inform,” “foster,” and “facilitate”.

Overall, we found that, among articles making causal claims connected to figures, there was a fairly even split between strong causal claims (~55%) and weak causal claims (~45%). This suggests that arrows are used to represent causal relationships regardless of the strength of the claim. To illustrate, let us compare the figures and text of two articles. First, in Fig. 2, Gates (2009) uses two path-style models to develop a historically contingent theory of business power, one that makes strong causal claims. The text accompanying Fig. 2a specifies that arrow #1 (dotted) involves “assessing whether policies that serve business interests correlate with proximate sources of business power,” while arrow #2 involves “identifying the underlying conditions that produce the proximate sources of business power.” (pg. 60, emphasis added). The following figure, Fig. 2b, details (among other things) the mechanisms involved in the causal relationship specified in Fig. 2a, and continues to consistently use the language of “production,” throughout the article.

Fig. 2

Gates’ a generalized model for a historically contingent theory of business power, and b the sources and conditions of business power in advanced industrial societies

Compare this with Fig. 3 from Wernick (1983), which uses an arrow as part of a larger illustration of how advertisements operate as ideological messages. As we are told in the text, “the arrow at the centre of the figure represents the surface level at which the ad hooks the consumer and attracts her/him towards the (critically undifferentiated) symbolised product” (pg. 24, emphasis added). The arrow therefore effectively signals movement towards the bracket (the “hook”), while the bracket itself indicates that while product and symbols are analytically distinct facets of “symbolized products,” our experience of them is unified (because consumers confuse or fuse the properties of the product and its mythical and psychological value). While the arrow and the text suggest a kind of causality, it is not as strong or direct as in Gates (2009).

Fig. 3

Wernick’s main elements and relations of symbolic ads

While the connection between causal figures and strong causal claims is unsurprising, their coupling with weak causal claims is more notable. In cases where arrows are used to represent weak causal claims, it is likely that the figure will make the causal relationship much more direct and explicit than does the text alone. In general, inferring causal relationships can be an important part of comprehending scholarly texts, and if causal claims are implicit or more ambiguous, causal diagrams make these causal relationships better and more easily understood (Corter et al., 2011; Larkin & Simon, 1987; McCrudden et al., 2007). Phrased another way, these figures may facilitate the inference of causal relationships in ways that weak causal claims do not and that their authors may not want. Alternatively, it may be that authors can and do use these strategically in order to imply causality visually in ways that they are less comfortable or less justified in doing textually. In either case, the use of arrow-based figures likely strengthens the causal interpretation offered by weak causal claims. We return to this issue later in our analysis of “fit”.

We also found that the strength of causal claims varied between U.S. and European sociology journals. While only a quarter (~25%) of the causal claims in European journals (TCS, EJS, and EJST) were strong causal claims, this was more than doubled (~60%) by the causal claims in U.S. journals (ASR, AJD, ST, TS). Abend et al. (2013) found that the strength (and centrality) of causality was much more pronounced in U.S. compared to Mexican journals, and we find a similar pattern in comparison to European journals. However, in contrast to the findings from Abend et al. (2013), this was not exclusively propped up by U.S. generalist journals, with roughly 55% of the causal claims from U.S. theory journals (ST and TS) being strong ones. Overall, this provides further evidence that U.S. journals disproportionately encourage, reward, or are targeted by authors making strong causal claims, and suggests that U.S. journals have a particular commitment to causal explanations as central to sociological knowledge and its advancement.

The Centrality of Causal Claims

After analyzing the strength of causal claims, we next examined their centrality. We looked at both the centrality of the causal relationship in the diagram specifically, coding articles based on whether the specific causal claim represented in the diagram was one of the papers primary aims, as well as the centrality of causality within the article as a whole, assessing whether causal arguments or claims were central to the paper’s general aims. We made these determinations by carefully reading key sections of the text, including the abstract, the introduction, and other sections where the paper’s main aims were explicitly stated (following statements like “this paper seeks to…” or “this paper examines….”). If either the causal relationship in the diagram, or causal arguments more generally were included in these sections, we considered them central to the aims of the paper.

Overall, we found that the relationships specified by arrow-based figures were almost always (~90% of the time) central to the article’s aims. This was the strongest relationship found in our analysis. Take Bakanic et al. (1987) for example, where the first two sentences of the abstract are:

“Peer review and editorial decision-making processes were examined for manuscripts submitted to the American Sociological Review between 1977 and 1981. Authors' professional characteristics, manuscript characteristics, review procedures, and referees’ recommendations accounted for more than 58 percent of the variance in the editors’ final decisions.” (Bakanic et al., 1987: 631)

The path diagram, which is presented several pages later, almost completely captures this information, specifying the variables and analytical steps taken in the paper (see Fig. 4). In fact, the reader could have simply looked at the figures and correctly inferred the overall aim of the paper.

Fig. 4

Bakanic, McPhail, and Simon’s diagram of their multi-stage path analysis of manuscript review and decision making

Similarly, Ozgen (2015) sets up a puzzle in an introductory section which her article then aims to resolve:

“Although constructivist studies have generated important findings, much of this scholarship has focused on the fluidity boundaries, rendering cases of durability a puzzle of sorts. If ethnic boundaries are routinely made and unmade by political, cultural, and economic institutions like nationalist regimes, assimilationist cultural policies, and market competition, what accounts for cases like Turkey where they persist? This article aims to move toward an analytically stronger theory of constructivism by showing how it can help explain both change and stability in an important but understudied region.” (Ozgen, 2015: 34)

Later, Ozgen draws on practice theory to develop a model of stability and change in ethnic boundaries (see Fig. 5). While the model precedes the specific empirical analysis of ethnic boundaries in Turkey, it specifies the mechanisms involved in the maintenance and weakening of ethnic boundaries, and therefore speaks directly to the broader theoretical aims set forth in the introduction.

Fig. 5

Ozgen’s model of ethnic boundary maintenance through marriage choices

In general, this finding suggests that visualization (or at least arrow-based visualization) is rarely a trivial issue. Sociologists by and large do not bother visualizing their minor claims or more peripheral ideas, but reserve them for the most important relationships they wish to establish in their articles. Visualization is usually a meaningful attempt to clarify, develop, illustrate, distil, or make memorable their core ideas, and therefore diagrams are one of the most crucial components of an article. Given their significance, this re-affirms the need for sociologists to improve our visualization efforts by establishing and committing to best practices, and increasing disciplinary standards for visualization at an institutional level (e.g., Brett et al., 2020; Healy & Moody, 2014).

In terms of the centrality of causality within articles as a whole, we found that in the majority (~70%) of articles causality in general was central to their aims. Some papers boldly use the language of cause and effect when stating their articles’ goals, for example Marks et al. (2009: 615) who “propose a causal argument” or Stahura (1979) who explicitly develops and deploys “a causal model of status persistence/evolution” (Stahura, 1979: 937). Others make clear that the ambitions of their article are causal in nature, but without using the word “cause”. For example, Olzak and Shanahan’s (2014) article suggests that “competitive threat from minorities shapes a broad spectrum of social control responses” more generally (Olzak & Shanahan, 2014: 392, emphasis added), and Kyriazis (2006) argues that “the choice of a primarily military-land power or seapower by a state does make a difference in subsequent historical evolution of this state (Kyriazis, 2006: 71, emphasis added). This once again suggests that causality is a major component of sociological theory and research, and that arrow-based figures are a core ingredient in causal theorizing and analysis.

The Fit Between Figures and Causal Claims

As discussed above, the ease with which arrow-based figures are interpreted as causal, combined with the cautious language sociologists use when discussing causal claims, should raise concerns regarding the fit between causal claims and causal diagrams. We treat fit as the (mis)match between the strength of the causal language in the text and the strength of the causal relationships as represented by the figure. Of course, determining the strength of causal relationships in figures is even more subjective than assessing causal language. We suggest that the visual representation of “strong causal relationships” would typically look like a solid arrow, or a series of solid arrows, linking variables in a way that is easily interpreted as causal. Conversely, the visual representation of a “weak causal relationship” might use dotted arrows, lines, or other elements to represent relationships in a way that is less readily interpreted as causal, and may use legends or reading rules to facilitate a softer interpretation. Nevertheless, the challenging nature of determining whether or not the text matches the figure was reflected in the moderate amount of inter-coder variation we had in our assessment of fit. As such, we focus below only on examples where coders agreed there was indeed a mismatch between figure and text.

In general, cases of mismatch involved diagrams which express clear and unambiguous causal relationships accompanied by much weaker and more ambiguous causal language. A clear example of this is Loner’s (2008) research on opinions about genetically modified foods. Loner provides a path style model of the acceptance of genetically modified foods, which guides the empirical analysis (see Fig. 6). The model specifies three factors associated with genetically modified food acceptance, each of which is represented by solid and direct arrows, and is straight-forwardly read as a causal diagram. However, the underlying data do not reveal a clear and direct relationship, specifically for “WORRY”, but instead, as acknowledged in the text (see pgs. 43–44), show a substantial amount of national and regional variation in significance. As such, “WORRY” does not necessarily have the clear and direct relationship that the model implies, and we are not given an “updated” figure that reflects this variation. Therefore, the weakness of the text is inconsistent with the much stronger impression given by the figure.

Fig. 6

Loner’s (2008) multi-sample model of acceptance of GM foods

Archer’s (1983) critique of Basil Bernstein and Pierre Bourdieu’s scholarship on education provides another useful example. Here, it is Archer herself who is critiquing the totalizing causal figure produced by Bernstein ([1975] 2003). As she points out, Bernstein has a dramatically oversimplified conception of society, the educational system, and their relationship, which is encapsulated by his causal diagram. To say that Bernstein’s figure is too strong is not to imply that Bernstein’s figure does not fit his own ideas; rather, it does not fit with the more complex account of social structure developed by Archer, and she annotates the broader implications of her account for Bernstein’s model on the right side of the figure (Fig. 7).

Fig. 7

Archer’s (1983) critical annotations of Bernstein’s diagram

However, our most notable finding regarding fit was that most (~85%) figures matched the strength of the causal language in the text. We found this high degree of fit surprising. In our previous work (Brett et al., 2020) we found that the uses of arrows by sociologists were sometimes vague in terms of the exact kind of relationship they were meant to represent (causal or otherwise), and were occasionally employed to create reductive distillations of more complex theoretical accounts. Conversely, scholarship on causal language has found that sociologists generally avoid clear and explicit causal language in their published work, cautiously employing metaphors and euphemisms to describe relationships (Abend et al., 2013; Vaidyanathan et al., 2016). Therefore, we expected a mismatch between the carefully crafted language in articles and the often under-scrutinized visualization of these relationships.

This surprising finding suggests a new hypothesis: that the use of diagrams encourages authors to think and write in clearer causal terms. As Wright Mills (1959: 213) noted, diagrams are not merely representations; they are also “genuine tools of production” that enable discovery and trigger imagination (see Silver, 2020; Swedberg, 2016). But diagrams are generative not only because they help us solve problems and stimulate new ideas, but also because they place demands on scholars, requiring us to make theoretical commitments (see Silver, 2020). For example, when we construct a path diagram, we commit to a specific causal hypothesis, a particular causal structure for the sake of argument (Griesemer, 1991: 172–173). In this sense, diagrams are “a vehicle for creatively clarifying, confronting, and even discovering what our theories commit us to” (Silver, 2020: 9). This would explain why we see a larger proportion of strong and explicit causal language in our sample relative to prior work on causal claims in sociology (Abend et al., 2013; Vaidyanathan et al., 2016), as the visualization efforts of the author(s) have forced them to take a stand on causality. While it certainly could be the reverse, that people tend to use figures only when they have a clear sense of the kind of causal claims they are willing and able to make, we suggest that the practice of creating figures forces authors to more thoroughly and reflectively come to grips with causality in their scholarship.

Supplementary Analysis

Importantly, this analysis of the relationship between arrow-based figures and causal claims is limited in a few respects. First, it does not take into account the methods used in the articles, nor how methodological standards for adjudicating causality have changed (see Brand et al., 2023; Morgan & Winship, 2015), which may confound the analysis. Second, it does not take into account how sociologists’ preoccupation with causality has waxed (1960s–1980s, 2010s) and waned (1990s–2000s) over time (see Abbott, 1998; Barringer et al., 2013). Lastly, because we sampled on articles with figures, we are unable to compare our results against the causal language of articles without figures, and are therefore limited in our ability to capture their broader significance.

To address these limitations and provide additional support for our claims, we performed a smaller supplementary analysis which compared the causal claims of articles using arrow-based figures to all other articles published within the same issue. To account for the changing standards and scrutiny around causality, we selected two articles and issues from ASR: one published in 1978 (issue 2, totalling 9 articles), at a time in which regression analysis could often be considered a legitimate method for identifying causality, and one published in 2011 (issue 6, totalling 7 article), at a time where the methodological stringency for establishing causality had greatly increased. The first article (Lincoln, 1978) used path diagrams and employed strong causal language both in the description of the figures and throughout the article, which involved establishing “causal relationships” and specifying which variables “affect” strike activity (see Fig. 8). Similarly, the second article (Simons et al., 2011) employed a path diagram that explicitly served as a causal model of aggression (see Fig. 9). We coded every article in these issues for the strength and centrality of its causal claims, whether it used an arrow-based figure, and whether it used regression.

Fig. 8

Lincoln’s (1978) path diagram of causal relationships among urban structure, union organization, and strike variables

Fig. 9

Simons et al.’s (2011) integrated model of aggression

Although our analysis is merely suggestive, the findings are notable. First, related to historical changes in the methodological standards for adjudicating causality, we found that the use of regression analysis was unrelated to both the strength and centrality of causal claims in both issues. Second, addressing the historical variation in sociology’s concern over causality, we found that the nature of causal claims did not differ significantly between issues, with both featuring a relatively even mix of strong and weak causal claims, and articles in which causality was central and non-central. Lastly, we found that the three articles using arrow-based figures (5 including Lincoln, 1978; Simons et al., 2011) all made strong, central causal claims. Though these results are far from definitive, they assuage concerns over confounding historical variables, and support our hypothesis that, independent of other factors, strong and definitive causal thinking and writing is anchored in theory visualization.

Conclusion

This paper examined visual and textual representations of causality in sociology articles. Whether explicit or implicit, literal or metaphorical, sociologists regularly make causal claims about the social relationships they theorize and analyze (Abend et al., 2013; Vaidyanathan et al., 2016). But causal arguments are not limited to text; when sociologists use arrows to visualize social relationships, they are often making a kind of visual causal claim. As we (Brett et al., 2020: 229) previously noted, because much of sociology involves describing, analyzing, and theorizing connections and relationships of various kinds, “to a large degree, the substantive domain of sociology is the visual domain of lines and arrows”. We would be hard pressed to avoid arrows when visualizing sociological thought, and nowhere is this more apparent than in visualizing causal relationships (see Elwert, 2013; Pearl, 1998).

Our findings suggest that arrow-based figures are fundamental to how sociologists think about and communicate causality. When we employ arrows in sociology, they are usually a means for expressing causal relationships specified in the text. Arrows are used to represent both weak and strong causal claims, though stronger claims are much more prominent in U.S. sociology journals compared to European journals. When we do choose to visualize causality, we generally do it to develop, illustrate, distil, or clarify the primary arguments, claims, or relationships in our articles. Moreover, we seem to construct causal figures that generally match the strength of our causal language, which may reflect a level of clarity that is achieved through the process of visualization itself. Overall, our analysis reveals that text and visualization work hand-in-hand in the construction of causal claims within sociology.

Moving forward, analyses of how text and figures work in conjunction may provide new insights into the production of sociological knowledge. At present, such analyses have only been conducted on a narrow set of sociological texts or thinkers (see Baldamus, 1992; Bargheer, 2021). This may reflect the largely taken-for-granted place of theory visualization within sociology. While sociologists have demonstrated the thoroughly visual nature of scientific practice generally (see Coopmans et al., 2014; Lynch & Woolgar, 1990), until recently sociologists had dismissed our own theory figures as “pictures of nothing,” visual ornaments rather than genuine workspaces for solving puzzles (Lynch, 1991). More recently, scholars have come to recognize that, even within sociology, the figures we employ actively shape and direct our thinking (Silver, 2020; Turner, 2014). Certainly, some of the visualizations that guide our work are “visual sketches” that are thrown away well before publication (see Swedberg, 2016). However, those that remain may still be crucial to the production of sociological knowledge, and analyzing them in conjunction with the text may shed light on how and why sociologists find and solve some problems and not others, ask certain kinds questions, make particular assumptions, and, in this case, make certain claims.

The findings from this paper suggest several questions and avenues for future research, especially those extending our supplemental analysis. Specifically, how does causal language vary between articles with figures compared to those without? Between articles with arrow-based figures versus those with non-arrow-based figures? Does the relationship between text and figure change based on the kind of methods employed in the article? It also opens up opportunities for parallel research involving non-causal visualizations. Are there patterns in their visual elements? What kinds of textual claims accompany these sorts of figures? Are they more prominent in certain kinds of journals? We hope this article lays a pathway for future analyses joining the visual and textual dimension of sociological thought.

Appendix (Adapted from Abend et al., 2013)

Variable name	Question	Values	Criteria
Causality	Does the article’s text make a causal claim pertaining to the relationship specified by the arrow?	Yes or No	Causal: If there is a statement in the text about the directionality of the relationship between the two elements (no matter whether the word “cause” occurred or not). Noncausal: If (1) the article’s text makes a claim about association or correlation but implied nothing about the nature of the relationship (or explicitly stated that the correlation should not be interpreted as having causal implications) or (2) if it said nothing at all about the relationship specified by the arrow.
Strength	What is the strength of the causal claim in the article’s text?	Weak or Strong	Strong causal claims: Unambiguous statements in the text about causal relationships between variables, events, or phenomena, and the relationship between X and Y was said or implied to be strong. Weak causal claims: causal claims in the text that are somewhat harder to pinpoint or discern. The relationships between variables are more ambiguous or euphemistically presented.
Centrality (of the diagrams causal relationship)	Was the specific causal claim represented in the diagram one of the paper’s primary aims?	Yes or No	Determined based on whether or not the causal relationship(s) represented in the diagram was stated in the abstract, or in the sections where the paper’s main aims were stated.
Centrality of causality overall	In general, how central are causal arguments or claims to the paper’s explicit overall aims, as expressed in the text?	Yes or No	Determined based on whether or not causal arguments are stated in the abstract, or in the sections where the paper’s main aims were stated.
Fit	Did the strength of the causal language in the text fit the strength of the causal relationships as represented by the figure?	Yes or No	The visual representation of “strong causal relationships” would usually look like a solid arrow, or a series of solid arrows, linking variables in a way that is easily interpreted as causal. The visual representation of a “weak causal relationship” might use dotted arrows, lines, or a variety of other elements to represent relationships in a way that is less readily interpreted as causal. They might also use legends or reading rules to facilitate a softer interpretation.
Type of figure	Is the figure a theory visualization or a data visualization?	Theory or Data	Determined based on whether or not the figure contained empirics.