4.1.1 Incommensurable Paradigms?

It has been argued that qualitative and quantitative methods are incommensurable. Qualitative research focuses on contextualized meanings and interpretation, whereas quantitative research focuses on decontextualized variables and statistics (Coxon, Reference Coxon2005; Flick, Reference Flick2002; Morgan, Reference Morgan2007; Power et al., Reference Power, Velez, Qadafi and Tennant2018). Each makes a different assumption about what should be studied: measurable quanta for quantitative methods and interpretable qualia for qualitative methods (Shweder, Reference Shweder, Jessor, Colby and Shweder1996). One benefit of this incommensurability argument is that it has enabled the development of qualitative methods with tailored quality criteria (Bauer & Gaskell, Reference Bauer and Gaskell2000). As a result, the field of qualitative research has burgeoned to address questions that are beyond the scope of quantitative methods. To ignore these differences, to try and reduce one approach to the other, would be to ignore the added value of each method (Denzin, Reference Denzin2012).

The strong incommensurability argument maintains that qualitative and quantitative methods stem from fundamentally different paradigms. Quantitative methods stem from realist and postpositivist paradigms, while qualitative methods stem from more constructionist paradigms (Guba & Lincoln, Reference Guba and Lincoln1994). Building on this epistemological alignment, some authors have argued that mixed methods stem from a pragmatist paradigm (Feilzer, Reference Feilzer2010; Morgan, Reference Morgan2007).

However, the incommensurability argument should not be overstated. Epistemology does not determine method; one can conduct realist qualitative research and constructionist quantitative research (Onwuegbuzie & Leech, Reference Onwuegbuzie and Leech2005). Furthermore, focusing on what is actually done reveals that qualitative data are often analyzed quantitatively and quantitative results often require qualitative interpretation. Exceptions to any claimed essential differences abound. Qualitative research often makes frequency claims (e.g., “most participants,” “some interviewees,” “rarely observed”). Equally, quantitative research often hinges on qualitative assessments of validity (e.g., face validity, expert raters, human-labeled gold standard data). Accordingly, the qualitative/quantitative distinction has been described as incoherent and misleading (Krippendorff, Reference Krippendorff2019; Sinclair & Rockwell, Reference Sinclair, Rockwell, Schreibman, Siemens and Unsworth2016). At best, the terms “qualitative” and “quantitative” refer to family resemblances, and at worst, they are intergroup affiliations (Coxon, Reference Coxon2005). In either case, the idea of incommensurability is a barrier to mixed methods research (Morgan, Reference Morgan2007) that undermines the credibility of social science research (Onwuegbuzie & Leech, Reference Onwuegbuzie and Leech2005). Mixed methods research argues that qualitative and quantitative methods are not incommensurable despite providing different insights. Indeed, they can be combined in nonduplicative and genuinely synergistic ways precisely because they serve distinct purposes.

A pragmatist approach contributes to differentiating qualitative and quantitative research by recognizing that each method addresses different questions. In one sense, the methods are incommensurable because they do different things and address different questions. A quantitative question about whether there is a between-group difference on a measure cannot, or at least should not, be tackled using qualitative methods. Equally, a qualitative question about how people pragmatically close conversations does not afford a quantitative approach. Nonetheless, despite this incommensurability of purpose, qualitative and quantitative methods are commensurable in other ways. Precisely because they do different things, they are chained together to ask sequences of questions (see Chapter 6). For example, the question “what is it?” and “how frequent is it?” are qualitative and quantitative questions that are nonduplicative and that synergize together.

4.1.2 The Integration Challenge

The core challenge of mixed methods theory is to conceptualize how qualitative and quantitative methods can be combined to produce insights that are not reducible to either method (Fetters & Freshwater, Reference Fetters and Freshwater2015a). This “integration challenge” goes beyond the idea that one has to choose either a qualitative-constructionist or quantitative-realist paradigm (Brown & Dueñas, Reference Brown and Dueñas2020). Instead, it shifts the focus onto how methods can be productively combined. Addressing the integration challenge is essential for legitimating mixed methods research as adding value beyond what qualitative and quantitative research can do alone (Johnson et al., Reference Johnson, Onwuegbuzie and Turner2007) and creating guidance for mixed methods researchers to maximize synergies.

A foundational concept for theorizing integration is the metaphor of triangulation (Denzin, Reference Denzin1970). In navigation and geographic survey work, triangulation refers to calculating an unknown point by drawing a triangle with two known points and then using trigonometry to calculate the unknown location. When applied to qualitative and quantitative methods, the idea is that the findings from each method are compared for either validation or enrichment (Hussein, Reference Hussein2009; Moran-Ellis et al., Reference Moran-Ellis, Alexander and Cronin2006). Triangulation for validation assumes that the results common to both methods have high validity. Triangulation for enrichment assumes that the results found with only one method are not necessarily low in validity but, rather, may reflect a different aspect of (or perspective on) the phenomenon. Validation is duplicative (both methods converge), while enrichment is additive (each method contributes something different).

However, the concept of triangulation has received criticism for being a relatively static geometric metaphor (Fetters & Molina-Azorin, Reference Fetters and Molina-Azorin2017a). The idea of triangulation fails to capture the generative and dynamic aspects of mixing methods. For example, in a review of mixed methods research Boeije and colleagues (Reference Boeije, Slagt and van Wesel2013) found that in addition to validation (e.g., for instrument development) and enrichment (e.g., by providing illustrations and nuance), mixed methods studies often enabled speculating about underlying mechanisms and generating plausible theories. That is to say, integrating (rather than triangulating) methods can reveal contradictions (Greene et al., Reference Greene, Caracelli and Graham1989) and puzzling discrepancies (Bryman, Reference Bryman2006) that spur theory-building.

The core rationale for mixing methods is that it should add value beyond what either method can contribute alone (Fetters & Freshwater, Reference Fetters and Freshwater2015a). The challenge is to specify the relationships between the methods, data, and findings so that the synergy is more than accidental (Moran-Ellis et al., Reference Moran-Ellis, Alexander and Cronin2006). To this end, Fetters and Molina-Azorin (Reference Fetters and Molina-Azorin2017b) identified fifteen dimensions of possible integration, including philosophical, theoretical, researcher, team, literature, sampling, design, research aims, data collection, analysis, and interpretation (see also Schoonenboom & Johnson, Reference Schoonenboom and Johnson2017). These insights direct attention toward the nodal points at which integration occurs, thus potentially isolating how integration can yield more than the sum of the parts (Åkerblad et al., Reference Åkerblad, Seppänen-Järvelä and Haapakoski2021).

4.1.3 A Pragmatist Approach

A pragmatist approach can conceptualize not only how qualitative and quantitative methods are different but also why integrating them can be synergistic. While other paradigms bring into focus social justice (the transformative paradigm; Mertens, Reference Mertens2007) and compatibility (critical realism; Shannon-Baker, Reference Shannon-Baker2016), the pragmatist paradigm emphasizes the research purpose (i.e., what they actually achieve via research questions, hypotheses, aims, goals, and objectives).

At its core, pragmatism is a method for making ideas clear and distinct by focusing on their consequences (Peirce, Reference Peirce1878). Since the birth of philosophy, there have been debates about the meaning of truth, beauty, God, and so on. The tendency has been to rely on axioms and first principles. Pragmatism eschews this approach, instead grounding meaning, and by extension philosophy and science, in human activity. According to James (Reference James1907, p. 22), it entails “looking away from first things, principles, ‘categories’, supposed necessities; and of looking towards last things, fruits, consequences.” For pragmatism, all beliefs, theories, and ideas are guides for action (Cornish & Gillespie, Reference Cornish and Gillespie2009). Simply put, meaning lies in consequences. The meaning of a bike is cycling; the meaning of food is eating; and, by extension, the meaning of a research method is in what it does.

Pragmatism is particularly suited to mixed methods research because it values each method for its contribution (Morgan, Reference Morgan2007). Thus, it offers an alternative to postpositivism and constructionism (Feilzer, Reference Feilzer2010). Pragmatism is inclusive because, in the words of James (Reference James1907, p. 31), it has “no obstructive dogmas, no rigid cannons of what shall count as proof” and “will consider any evidence.” Instead of asking whether methods are epistemologically commensurable, pragmatism asks what each method contributes to the problem at hand. “Pragmatists,” Feilzer (Reference Feilzer2010, p. 14) writes, “do not ‘care’ which methods they use as long as the methods chosen have the potential of answering what it is one wants to know.”

A pragmatist approach can contribute to the integration challenge. Instead of focusing on the rationales for mixed methods research in general (i.e., validating, enriching, developing, explaining), the pragmatist approach directs attention to the underlying qualitative and quantitative purposes, specifically to how these are combined. Thus, in contrast to the many typologies that differentiate qualitative and quantitative methods (Coxon, Reference Coxon2005; Sale et al., Reference Sale, Lohfeld and Brazil2002), we focus on what these methods are used for within the broader logic of the research (see also Denscombe, Reference Denscombe2021). Our aim is not to characterize qualitative and quantitative methods according to essential criteria or even identify family resemblances (Morgan, Reference Morgan2018). We aim to differentiate qualitative and quantitative methods in terms of what they are used for.

4.2.1 Inductive Questions

Induction entails moving from the observation of regularities to a general statement. For example, if one observes a class in session every Monday at noon in room 1A, one infers the rule that there is always a class on Mondays at noon in room 1A. Induction is quintessentially empirical. The prototypical case is that, given previous observations, one infers that the sun always rises in the east. Induction starts with observation and builds theory bottom-up, remaining as close as possible to the data. Inductive research can be either qualitative or quantitative.

Describing Phenomena. Qualitative inductive research contributes thick description, using words to richly document a phenomenon, including its variants and relation to context. It often entails understanding the subjective world of others. Qualitative inductive questions include: What is X? How do they do X? What is the experience of X? Contributions to these questions are evaluated in terms of the subtlety of the description, reflexivity in observation, and participation of the researched. Ideally, the observations form an interlocking web that richly conveys what happened, people’s beliefs and practices, and subjective experiences.

Measuring Phenomena. Quantitative inductive research focuses on measuring observables and abstract constructs using counts, ranks, and scaled scores. Measurement necessarily makes theoretical assumptions about the phenomenon (e.g., identifying valid observable indicators of a phenomenon, determining the most appropriate type of measurement), but it aims to foreground the phenomenon rather than the theory. Measuring questions include: Is measure X reliable and valid? How frequent is X? What statistical model of X emerges? Contributions to these questions are usually evaluated in terms of operationalization, sampling, reliability, and validity. Ideally, a measure captures what it claims to measure and can be generalized to a population.

Inductive research is often devalued as merely descriptive (Gerring, Reference Gerring2012). But detailed observation is the basis of science (Rozin, Reference Rozin2009). When we encounter a strange object, we look at it from different angles, pick it up, and squeeze it to ground our emerging understanding in empirical experiences (Blumer, Reference Blumer1969). Preconception is a liability for induction, potentially suppressing peculiarity (Feyerabend, Reference Feyerabend2001). The aim is to be “open” to the data, to be sensitive to peculiarity, and to be willing to be surprised. Science without induction would produce concepts disconnected from experience. But science based only on induction would cease to have bold and ambitious theories.

4.2.2 Deductive Questions

Deduction entails moving from a general statement, or assumption, to a logically implied conclusion. For example, knowing the finding that classes have been observed in room 1A on Mondays at noon, one expects this to be the case next Monday at noon. The quintessential deduction is a syllogism. All men are mortal (first premise), Socrates is a man (second premise), and therefore, Socrates is mortal (conclusion). Given the theory (first premise) and the observation (second premise), one can deduce the empirically verifiable hypothesis that Socrates will eventually die. Deduction is rationalist; it is what Descartes used to argue that his own doubting was indubitable. Deductive research starts with expectations based on the literature (given X and Y we expect Z) and logic (theory A predicts X, but theory B predicts Y; which is right?). Again, it has both qualitative and quantitative variants.

Theoretical Framing. Qualitative deductive research entails using theory to frame or guide inquiry. Framing uses theory as a conceptual lens, sensitizing the researcher to observable, but conceptual, phenomena. It includes, for example, using theories of conversation, impression management, or nonverbal behavior to analyze a face-to-face interaction. Typical framing questions include: Does idea X provide insight? Does typology X fit the data? Do cases differ as expected? Contributions to these questions are evaluated in terms of how suited the theory is to the phenomenon, whether the framing produces insight, and whether the theory is being overextended, oversimplified, or overly imposed.

Testing Hypotheses. Quantitative deductive research focuses on testing, namely, using theory (i.e., the literature) to specify an expectation that can be compared to observations. The classic case is null hypothesis testing, where a falsifiable statement about the relationship between variables is stated before the research is conducted and then statistics are used to calculate the likelihood of the observed results. Typical testing questions include: Is X associated with Y? Does Z mediate the association? Do the data fit model X? Contributions to these questions are evaluated in terms of the logic that leads to the expectation, priors, operationalization, and potentially confounding variables.

Deductive research is a mainstay of qualitative and quantitative research (Scheel et al., Reference Scheel, Tiokhin, Isager and Lakens2020; Tavory & Timmermans, Reference Tavory and Timmermans2014). Deduction is powerful because it leverages the literature (previous studies and theories) so that the research does not start anew but builds on prior insights. We see farther than our predecessors not because we have better vision but because we stand upon their shoulders (John of Salisbury, Reference McGarry1159). Science without deduction would cease to be cumulative. But science based only on deduction would yield deductions entirely disconnected from practical experience.

4.2.3 Abductive Questions

Abduction entails reasoning from observation and prior expectations to forge a new theory beyond both. For example, observing a class in room 1A on Mondays at noon but finding that there is no class on a particular Monday at noon, one generates the plausible explanation that there might be something called a “timetabling department” that has changed the rules or a thing called “term time” that suspends the rule. In either case, abduction explains the anomaly by introducing an idea outside the data (i.e., the timetabling department, term time). Abduction often begins with a disruptive fact or contradiction (Tavory & Timmermans, Reference Tavory and Timmermans2014); the outcome is an explanation that, although not in the data, explains the data. Einstein’s theory of general relativity explained observables with nonobservables (i.e., spacetime). Darwin’s theory of evolution postulated a mechanism he did not observe (i.e., natural selection). What makes abduction inferential (rather than unconstrained imagination) is that a “hypothesis cannot be admitted, even as a hypothesis, unless it be supposed that it would account for the facts” (Peirce, Reference Peirce1955, p. 151).

Generating Explanations. Qualitative abductive research is widespread and aims to generate explanations and theories (Power et al., Reference Power, Velez, Qadafi and Tennant2018). Being close to raw data, observing particularities, and being relatively free to approach the phenomenon from multiple theoretical standpoints make qualitative research fertile soil for generating plausible explanations (Tavory & Timmermans, Reference Tavory and Timmermans2014). Typical questions include: Why is X? What might cause X? What process might underlie X? Contributions to these questions are not judged by the logic of derivation (deduction) or by the rigor of the process (induction). It does not matter if the abductive leap occurs while lucid dreaming or sitting in the bath; what matters is whether it aids understanding, chimes with prior theories, explains the puzzling observations, and yields productive lines of action.

Exploring Explanations. Quantitative abductive research aims to stimulate ideas by exploring relationships within data. Although often undervalued relative to hypothesis testing, exploring the associations between measures can spur the creation of new hypotheses firmly grounded in data (Tukey, Reference Tukey1977). Typical exploring questions include: What is associated with X? Does anything predict X? Is a confounding variable missing? Contributions to these questions, as with qualitative abduction, are evaluated not in terms of the rigor of the observations or the logic of the deduction but in the fruitfulness of the emergent insights.

Abductive research has received little theorization, possibly because it entails a creative leap outside standardizable procedures (Tavory & Timmermans, Reference Tavory and Timmermans2014). However, induction and deduction are insufficient to explain most scientific breakthroughs in the natural (e.g., heliocentrism, theory of natural selection, dark matter, the structure of the double helix) or social (e.g., equality, feminism, power, impression management, ideology, culture) sciences. Abduction is aided by sensitivity to contradictions and anomalies, and openness to revising one’s expectations. In short, abduction thrives when there is a puzzle to solve. Science without abduction would cease to have revolutions. But science built solely on abduction would perpetually introduce new ideas without any criteria for evaluating them.

4.2.4 Matching Questions to Methods

A pragmatist approach to research questions starts with the insight that methods are tools for action. Instead of trying to distinguish methods from first principles, methods are differentiated in terms of their research purposes.

The typology in Table 4.1 guides when to use qualitative, quantitative, and mixed methods. Scholars have characterized differences in epistemology (Denzin, Reference Denzin2012) and subject matter (Shweder, Reference Shweder, Jessor, Colby and Shweder1996) and distinguished family resemblances (Coxon, Reference Coxon2005; Morgan, Reference Morgan2018). But we propose a contingency table of when to use qualitative, quantitative, and mixed methods. Qualitative methods can describe phenomena, provide theoretical framing, and generate explanations. Quantitative methods can measure phenomena, test hypotheses, and search for explanations. Mixed methods are necessary when research benefits from multiple purposes (e.g., describing a phenomenon to measure it better, generating hypotheses about why an experiment produced surprising results). Differentiating these purposes can guide researchers in selecting suitable methods for the problem they are addressing. This can help researchers avoid the “methodological monologic” described by Bauer and Gaskell (Reference Bauer and Gaskell2000, p. 338):

Differentiating qualitative and quantitative methods in this way gives each method a separate domain of legitimacy. From this standpoint, asking whether qualitative methods are better than quantitative methods, or vice versa, is like asking whether a hammer is better than a saw. It depends on what you want to do. Moreover, to assume that social science can get by with only one of these methods is to dismiss an entire row of indicative questions in Table 4.1. Both qualitative and quantitative methods are legitimate because they are the best tools we currently have for specific purposes.

4.3.1 Creating Questions Inductively

Although high-quality inductive research, such as detailed observation, should be the bedrock of science, it is disappointingly rare in social science (Blumer, Reference Blumer1969; Gerring, Reference Gerring2012) and psychology (Rozin, Reference Rozin2001; Rubin & Donkin, Reference Rubin and Donkin2022). Psychology has been overly focused on deductive development of hypotheses and subsequent confirmatory statistical testing, while devaluing nonconfirmatory or exploratory research (Krpan, Reference Krpan2022; Scheel et al., Reference Scheel, Tiokhin, Isager and Lakens2020). But confirmatory research should not be rushed into; it is the last step in a sequence of questions that begins with patient and detailed description and concept formation (Scheel et al., Reference Scheel, Tiokhin, Isager and Lakens2020). Without these rigorously inductive descriptions, concepts risk becoming vague, disconnected from everyday life, and low in validity despite being reliable. If the creation of new questions comes only from deduction, then it is overly determined by the literature and insufficiently attentive to our continually evolving practical problems.

Beyond psychology, detailed description is widespread (Rozin, Reference Rozin2009). Darwin’s theory of evolution by a process of natural selection did not arise through deductive hypothesis testing; rather, it rests upon numerous patient descriptions. Similarly, Crick and Watson’s breakthrough model of the double helix is, first and foremost, a description. Turning to the social sciences, conversation analysis (Schegloff, Reference Schegloff2007) is also built on elaborate and painstakingly detailed descriptions. In sociology, Goffman’s (Reference Goffman1959) groundbreaking analysis of self-presentation in everyday life is also based on rich descriptions of human interaction. Psychology, in contrast, is rushing headlong into experimentation without extensive description (Rozin, Reference Rozin2009; Rubin & Donkin, Reference Rubin and Donkin2022). Such premature enthusiasm for confirmatory research, with a disregard for descriptive research, is likely counterproductive for a young science, like psychology, that is striving for scientific legitimacy (Krpan, Reference Krpan2022; Scheel et al., Reference Scheel, Tiokhin, Isager and Lakens2020). It can, for example, lead to overextended theories that fail to replicate, and which thus foster skepticism in psychology (Baucal et al., Reference Baucal, Gillespie, Krstić and Zittoun2020; Mede et al., Reference Mede, Schäfer, Ziegler and Weißkopf2021; Open Science Collaboration, 2015).

Blumer (Reference Blumer1969), a neopragmatist building on the work of George Herbert Mead, used the term “inspection” to conceptualize the inductive phase of social research. Inspection aims to describe the peculiarity of a phenomenon. Consider encountering a strange, suggestive, and confusing physical object:

Inspection tries to put aside assumptions and expectations. It entails attending carefully to the empirical particulars. Inspection is part of our natural everyday attitude, and it is also a hallmark of science.

Inspection can be both qualitative and quantitative. For example, inspecting people talking could include counting the number of participants or conversational turns, noting a peculiar tone of voice, or understanding the content of what was said. Qualitative inductive research is good for close-up inspection, rich description, revealing practices, identifying heuristics, and finding puzzles (Becker, Reference Becker1998; Crano et al., Reference Crano, Brewer and Lac2014). Quantitative inductive research identifies differences between groups, changes over time, and outliers (Jaccard & Jacoby, Reference Jaccard and Jacoby2020). And combining these inductive methods can produce added synergies. For example, one might start by using qualitative research to describe a feature of talk (such as conversational repairs; Schegloff, Reference Schegloff1992) and then quantitative research to measure the frequency of the phenomenon (conversational repairs occur every 1.4 minutes; Dingemanse et al., Reference Dingemanse, Roberts and Baranova2015).

Inductive inspection helps generate questions because it ensures that the research is firmly anchored in what is going on. It can reveal the boundary conditions for a previously established phenomenon (Jaccard & Jacoby, Reference Jaccard and Jacoby2020), a challenging or conflicting observation (Crano et al., Reference Crano, Brewer and Lac2014), or simply a novel event (Becker, Reference Becker1998). The critical ingredient is being attuned to peculiarity (Becker, Reference Becker1998; Crano et al., Reference Crano, Brewer and Lac2014; Jaccard & Jacoby, Reference Jaccard and Jacoby2020; McGuire, Reference McGuire1997). In so far as inductive inspection conforms to expectation, it yields data, and in so far as it disrupts expectation, it yields disruptive data that can generate new theory (Chapter 3).

Valuable sources for inductive question generation include participant observation, deviant case analysis, biographies, diaries, news stories, historical events, introspection, role play, and even fictional stories (Jaccard & Jacoby, Reference Jaccard and Jacoby2020). It can also be valuable to talk to practitioners to understand the frontline issues and the existing practical solutions (Jaccard & Jacoby, Reference Jaccard and Jacoby2020). Going further in this direction, participatory action research is a useful methodology for harnessing and building on the insight of communities to generate novel questions (McIntyre, Reference McIntyre2008).

Case studies are a particularly compelling method for generating research questions inductively. Too often, research cuts up phenomena in unnatural ways, with the actual sequence of events suppressed. In qualitative research, quotations from people unfamiliar with one another, perhaps even from different regions, are presented side by side, without regard for the sequencing of what was said. In quantitative research, the focus is on central tendencies and averages (which may not exist in any single case), with outliers discounted or even removed. Case studies, in contrast, focus on a singular event and conceptualize it holistically – as a system of people and parts. There is no statistical average or central tendency; there just is what happened. This bedrock level of “what actually happened” is fertile soil for creating new questions (Eisenhardt, Reference Eisenhardt1989). Why? Because theories can become detached from actual events. And despite the sometimes-low status of descriptive research, observations of particulars (i.e., what actually happened) should always be privileged over abstract theory (i.e., what is expected to happen). Thus, juxtaposing grand theories with specifics can reveal oddities, which grow into disruptive facts, or anomalies, becoming the wellspring of a new idea.

4.3.2 Creating Questions Deductively

Deduction is often proposed to be the ideal method for creating questions. Theories are used to create research questions and hypotheses that are tested empirically (Popper, Reference Popper1969). The key to generating questions deductively is conceptual analysis, especially thinking through the logical implications of theories (Jaccard & Jacoby, Reference Jaccard and Jacoby2020; McGuire, Reference McGuire1997).

A common approach to deductive question generation is to take an established concept and identify subtypes and variants, perhaps manifesting in peculiar contexts (Crano et al., Reference Crano, Brewer and Lac2014). This is aided by defining concepts, refining concepts, challenging definitions, and speculating about subtypes and unaccounted-for variables, and introducing novel concepts (Jaccard & Jacoby, Reference Jaccard and Jacoby2020). For example, the contact hypothesis (Allport, Reference Allport1954) originally proposed that contact between outgroups could reduce prejudice under certain conditions. Initially, these conditions were concise: The contact should be between people of equal status, sharing a common goal, with the parties being interdependent in the achievement of the goal, and the contact should be supported by relevant laws, institutions, and customs. However, since this original formulation, the theory has been refined through numerous questions (Hewstone & Brown, Reference Hewstone, Brown, Hewstone and Brown1986; Paluck et al., Reference Paluck, Green and Green2019). These refining questions include: What if the outcome of the joint activity ends in failure? What if the members of one group are not prototypical? And what about the frequency of contact? Or what about pressure to achieve the goal? What if member behavior is or is not seen to be prototypical of their group membership? What if the contact is face to face, technologically mediated, or even imagined? Such question generation is within-paradigm, leading to refinements of the original theory.

Although identifying subtypes and subconditions is widespread, it is often unsatisfying. In the case of the contact hypothesis it has been criticized for producing “an open-ended laundry-list of conditions” (Pettigrew, Reference Pettigrew1998, p. 69). Each additional condition, mediation, or variable does refine the theory, but it makes the theory conceptually weak, as it becomes a list of if-then propositions. Newell (Reference Newell and Chase1973) famously characterized this type of theorizing as trying to “play 20-questions with nature.” That is to say, deduction is used to pose binary questions to nature that are tested with an experiment. This, Newell argues, necessarily produces fragmented “if-then” type theories. This mode of theorizing, and developing questions, is suited to testing big theories once they are in place, but it is not suited to generating big integrative theories.

A more exciting way to generate research questions deductively is to take an idea that seems obvious and investigate the alternative (Jaccard & Jacoby, Reference Jaccard and Jacoby2020; McGuire, Reference McGuire1997). The world is rarely one-sided, and there are usually contexts in which the opposite holds. For example, Billig (Reference Billig1985) noted that most of the literature on social identity had focused on categorizing (i.e., grouping people into categories), and he made a compelling case for the counterbalancing process of particularizing (i.e., seeing people as individuals and not categories). Relatedly, Gillespie (Reference Gillespie2007a) shows how within moments of differentiation from outgroups there can also be moments of identification. Finally, Moscovici and colleagues (Reference Moscovici, Mucchi-Faina and Maass1994) observed that conformity (or majority influence) must have a counterbalancing mechanism that prevents everyone from becoming the same (i.e., total conformity), and they advanced the influential concept of minority influence. This kind of deductive theory generation is stimulated by taking theories to extremes (e.g., if there was only conformity) and then probing the processes that prevent the extreme.

A final approach to the deductive creation of research questions is to hold multiple theories in mind, such that tensions can be surfaced and competing accounts for observations can be generated (Crano et al., Reference Crano, Brewer and Lac2014; Jaccard & Jacoby, Reference Jaccard and Jacoby2020). We can never have complete certainty in any theory, and thus it is prudent to entertain multiple possible theories. Like the lost explorers described by Midgley (Reference Midgley2003; see Chapter 3), it is prudent to operate with multiple mental maps. The idea that a theory is infallible and timeless is an obstacle to this mode of deductive speculation because it blinds one to anomalies and leads to discarding alternative theories prematurely. When Darwin (Reference Darwin2001) sailed on the HMS Beagle, he entertained both the biblical view of the earth being created a few thousand years ago and the challenging ideas of Lyell that the earth’s geology was shaped over millions of years. Darwin’s theory of evolution was not a direct contribution to the debate about the age of the earth, but the heretical idea that the earth was much older than traditionally assumed was necessary for him to formulate his theory of natural selection. The point is that, given that theories are tools (Chapter 3), more theories enable more diverse ways of thinking about and acting on observations. In short, more paths of action in and through data increase the potential to find a novel pathway, question, or theory.

4.3.3 Creating Questions Abductively

Scientific revolutions have rarely been the result of mere induction or deduction; they have usually entailed a creative leap of abduction. The idea that the planets revolve around the sun, that evolution operates by natural selection, and that space and time are relative to the observer all entailed abductive leaps. Somewhat more controversially, but nonetheless abductively, are the social science theories that conceptualize the mind as a computer, people as utility maximizers, and languages as underpinned by a universal grammar. None of these ideas were deduced from first principles. None of them were created by merely aggregating observations. In each case, there is a creative leap of abduction that goes beyond the evidence and preexisting ideas.

Abduction entails going beyond the data to posit something that explains the data. Although Sherlock Holmes (Doyle, Reference Doyle1892) frequently describes his own method as deduction, it is usually abduction; there is a weaving of a pattern and the positing of a story behind the events that makes sense of the events. “Abductive inference,” Krippendorff (Reference Krippendorff2019, p. 38) writes, “is Sherlock Holmes’s logic of reasoning.” When Sherlock Holmes is confronted with the ostensibly bizarre facts of a case (e.g., the stabbed murder victim in a windowless room locked from the inside), the process of abduction is to imagine possible scenarios. The first step is to attend to the stubborn facts (e.g., the locked door, the footprints, the hesitant responses). The second step is to generate plausible explanations (e.g., was it suicide, was the victim hypnotized, might the victim have been fatally wounded and sought refuge in the room). Abduction is the process of simulating possible explanations until, like a key in a lock, the explanation fits the facts.

Creating questions abductively entails imagination. Abduction is not about what is (either empirically or logically); it is about what might be. It entails leaving behind the here and now to viscerally inhabit alternative scenarios (Zittoun & Gillespie, Reference Zittoun and Gillespie2015). It brackets aside assumptions and aims to generate plausible accounts that might (or might not) explain the known facts. Often, such imagination will not provide added explanatory power. But, sometimes, it provides the seed for a new line of inquiry. Abduction is facilitated by questioning assumptions, challenging taken-for-granted theories, focusing on the most irrefutable details, and being playful in generating alternatives but harsh in evaluating them. Abduction entails tolerating competing and even incompatible theories, seeking peculiarity and anomalies, and pursuing possible lines of inquiry regardless of how surprising or heretical they are.

To generate explanations abductively, the researcher should continually ask “what?” and “why?” (Becker, Reference Becker1998; Jaccard & Jacoby, Reference Jaccard and Jacoby2020). First, it is important to fixate on what actually happened, the sequence of events, and the step-by-step details of the case. These details are the obstinate data to be explained. Second, each datum within the tangled mess of facts should be probed: Why did X happen? Why did Y not happen? What else could have happened? To this end, one needs to be clear about the difference between the facts to be explained (the “what”) and the speculations about those facts (the “why”). As Sherlock Holmes said: “It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts” (Doyle, Reference Doyle1892, p. 163). The “what” (or data) are firm anchor points, the “why” are speculations. The “what” should be shorn of interpretation, the “why” is enhanced by imagination. The “what” can never lead one astray, but the “why” can bring false hope, and become a chimera that inhibits progress with dead ends. The “what” is in the past, and not open to revision, while the “why” should always be open to revision.

To think creatively about the “why” – that is, to generate explanations for the “what” – can be enhanced with analogies and metaphors (Haig, Reference Haig2005; Jaccard & Jacoby, Reference Jaccard and Jacoby2020). As discussed in Chapter 3, human understanding is grounded in the here and now of daily practice. Metaphors pervade social science. They underpin theories in psychology (e.g., the mind as a computer; Leary, Reference Leary1990), communication (e.g., the conduit metaphor of message transmission; Axley, Reference Axley1984), human development (e.g., organic growth; Zittoun & Gillespie, Reference Zittoun and Gillespie2020), sociology (e.g., mechanism and functionalism; Swedberg, Reference Swedberg2020), and economics (e.g., the marketplace; McCloskey, Reference McCloskey1995). Indeed, in this book, we have relied heavily on metaphors, such as likening theories and methods to “tools” and describing abduction as “Sherlock Holmes style” of inference. Hiding metaphors is blinding; being open about them enables discussing the potentials and pitfalls of all metaphors (Swedberg, Reference Swedberg2020), and in so doing, one can liberate the imagination to try alternative metaphors, to see what they elucidate, prompt, and enable.

Finally, thought experiments are a particularly powerful (but often neglected) method for developing research questions abductively. Thought experiments have had a significant impact on science and have been central to several breakthroughs. Einstein (Reference Einstein1982; Norton, Reference Norton, Horowitz and Massey1991) famously used thought experiments, such as trying to chase light and being on a train moving at the speed of light. Searle (Reference Searle1982) powerfully argued against strong artificial intelligence using the thought experiment of a person in a room converting inputs into outputs using an incomprehensible rulebook. Putnam (Reference Putnam1974) argued that meanings are not just in the mind by positing a twin earth in which everything was identical except the chemical composition of water. In the veil of ignorance thought experiment, Rawls (Reference Rawls1971) asks how one would decide upon the rules of society before knowing one’s position in society (i.e., profession, ethnicity, gender, income, wealth, social status). Thought experiments do not need to be elaborate scenarios. For almost any social phenomenon, one can engage in basic thought experiments by asking: Could this have occurred a thousand years ago? Might this occur a thousand years in the future? How would this play out in a world where everyone had perfect information, nobody lied, there was no emotion, or everyone was equal? Literature is an excellent resource for thought experiments. Franz Kafka (Reference Kafka1915) explores what it is like to wake in the morning transformed into a huge insect. Phillip K. Dick (Reference Dick1962) explores what life in America might have been like if it had lost World War II. Cixin Liu (Reference Liu2014) examines what might happen to human society if it was known that aliens were en route and would arrive in 450 years.

While some have argued that thought experiments are merely arguments dressed up as stories (Norton, Reference Norton, Horowitz and Massey1991), this is to underplay their value for us as humans. As argued in Chapter 3, knowledge is “for us” and works best when anchored in everyday life. Thought experiments are invaluable because they give us more than a dry argument, mere numbers, or prediction. Instead, they provide a visceral and meaningful simulation, anchored in an intelligible first-person perspective. They enhance our embodied identification and thus ground abstract thought in everyday human activity (Ludwig, Reference Ludwig2007). In short, thought experiments are “intuition pumps” (Dennett, Reference Dennett1991, p. x) that enable us to “feel” problems from the inside, take ideas to extremes, and run simulations with impossible conditions (McGuire, Reference McGuire1997). This playfulness stretches the space of the possible, opening a semantic space within which abductive questions arise.

4.3.4 Mixing Induction, Deduction, and Abduction

It is a mistake to oppose induction, deduction, and abduction. Morgan (Reference Morgan2007; table 2), in an otherwise excellent article, claims that a pragmatist approach focuses on abduction instead of induction (which he associates with qualitative research) or deduction (which he associates with quantitative research). But Peirce (Reference Peirce1955, Reference Peirce1992) valued all three modes of inference and argued that they work best in tandem. Abandoning any mode of inference would be antipragmatist, because it would be a tribal affiliation to one form of inference; it would fail to leverage the insight that each mode of inference can provide. Peirce’s (Reference Peirce1955) differentiation between induction, deduction, and abduction is conceptual. In any practical context, these modes of inference work together, yielding synergies that cannot be reduced to any one mode of inference operating in isolation.

When trying to solve a problem, we leverage insights from the past (deduction), attend to the concrete particulars of the problem (induction), and make leaps of speculation (abduction). Peirce (Reference Peirce1974, sec. 1.72–1.74) gives the example of Kepler’s discovery of the laws of planetary motion. Throughout a long investigation, Kepler weaves together previous theory, empirical observations, and abductive reasoning to arrive at his theory. Without induction, his theory would have been disconnected from empirical observation. Without deduction, he would not have been able to frame certain observations as surprising. Without abduction, he would not have been able to leap beyond both observations and existing theory to realize that planetary orbits were elliptical rather than circular.

In social science, these three modes of inference are woven into the fabric of most empirical articles. The introduction section is usually deductive, drawing inferences from the existing literature. The empirical findings should be inductive, with inferences being based on the data collected and the analyses performed. Abductive inference is sometimes evident in a surprising research question that posits a novel angle on an established problem and sometimes evident in the interpretation of surprising findings. The key point is that these three modes of inference synergize: Deduction leverages the past to generate expectations; abduction generates ideas that escape the confines of deductive expectation; and induction tames unfounded expectations and excessive speculation.

Moving between modes of inference can help to generate research questions. This moves beyond discovering tensions between observations (induction) or between theories (deduction) and opens the possibility of discovering tensions between what is expected and what is observed, between what should be and what might be, and between what is and what could be (Jaccard & Jacoby, Reference Jaccard and Jacoby2020; McGuire, Reference McGuire1997). The point is that moving between modes of inference opens the research process up to additional and productive tensions that can spur insight and foster new research questions.

Research questions can also be generated by moving between qualitative and quantitative methods. This is an extension of moving between modes of inference. Each mode of inference (i.e., induction, deduction, and abduction) has qualitative and quantitative variants (see Table 4.1). It has long been argued that mixing methods is a powerful means of generating new theories (Greene et al., Reference Greene, Caracelli and Graham1989). The core idea is that integrating qualitative and quantitative methods should lead to a 1 + 1 = 3 synergy, but specifying how this occurs is challenging (Fetters & Freshwater, Reference Fetters and Freshwater2015a). The theoretical literature on mixing methods tries to identify how this creativity is more than accidental and can be traced to particular integrative strategies (Åkerblad et al., Reference Åkerblad, Seppänen-Järvelä and Haapakoski2021) and dimensions of integration (e.g., assumptions, aims, data, interpretation; Fetters & Molina-Azorin, Reference Fetters and Molina-Azorin2017b).

Table 4.1 differentiates both modes of inference and qualitative and quantitative methods. This provides a basis for conceptualizing emergent synergies. It is precisely because each mode of inference and method does something different, and answers different questions, that they are complementary rather than competing. Thus, these methods can be chained together to produce synergistic findings. For example, describing a phenomenon qualitatively (“what is it?”) often leads to measuring the phenomenon quantitatively (“how frequent is it?”); testing a theory (“does the data cluster according to typology X?”) often leads to a qualitative search for an explanation (“why did the clustering not work?”); and generating an explanation (“what might cause X?”) feeds forward into testing the explanation (“does Y cause X?”). These chains of investigation are examined in detail in Chapter 6.

Mixing modes of inference, and associated methods, increases the chances for disruption. Such disruptions are central to scientific progress (Kuhn, Reference Kuhn1962). While much science is routine, entailing fitting data to theories to flesh out a given paradigm, there are key turning points in science, revolutions, that establish new paradigms. Whether engaged in normal (i.e., within-paradigm) or revolutionary (i.e., paradigm-creating) science, the key is to be sensitive to anomalies such as disruptive observations, confounded expectations, and contradictory theories. Normal and especially revolutionary science progress by addressing such anomalies. It follows that any research practices that increase the chance of anomalies arising will advance science and lead to increasingly robust knowledge. Research that is siloed within a subdiscipline, operates with a narrow range of methods, or insulates itself from real-world practices and consequences is protecting itself from disruptive surprises and, thus, the potential for scientific progress.