Should psychology adopt Bhaskar’s critical realist philosophy of science?

Scientific ontologies

According to Bhaskar (2016), the proper role of philosophy is to serve as an underlabourer for science, formulating very general and abstract categories (e.g., causality, lawfulness, necessity) that are tacitly presupposed in scientific practice. In this role, BCR makes explicit use of transcendental arguments, which ask what the world must be like for social practices to be possible. With science in mind, the central question here is: How must the world be structured in order for scientific practice, including experimentation, to be understood? Importantly in this regard, BCR claims that the world comprises a hierarchy of three overlapping domains: the real (containing causal mechanisms), the actual (which admits events), and the empirical (which includes experiences). These levels must be aligned in a closed system for experimental practice to operate successfully. However, the fact that these three domains are regarded as ontological levels, in a strong sense, presents a problem for BCR. Wimsatt (2007) has plausibly argued that our world is complex and messy, comprising multiple compositional levels of organization, and that these levels can be, but need not be, hierarchically arranged. More specifically, Marcus Eronen (2019) maintained that levels in psycho(patho)logical science are ontologically fundamental, but are best regarded as heuristic idealizations, which frequently have blurred boundaries.

I think that psychology is best seen as an ad hoc collection of subdisciplines, assembled largely for pragmatic and administrative convenience. As such, it comprises a varied mix of local ontologies admitting such things as common causes, mechanism traces, data ontologies, and multiple theory/world correspondence relations. Ontologies like these are the products of successful scientific practice in which the contextually specific employment of methods is aligned with their ontological presuppositions. In my view, Bhaskar’s stratified ontology is too rigid and coarse-grained to usefully serve the varied demands of psychology’s subdisciplines.

The hypothetico-deductive method

Part of Archer’s (2024) dismissal of the philosophy of critical rationalism involves the rejection of Popper’s account of the hypothetico-deductive method. Archer’s primary reason for rejecting this method is his belief that it shifts our attention from the process of generating hypotheses to a focus on their testing. I do not find this reason for abandoning Popper’s construal of the method persuasive. Researchers are perfectly entitled to accept Popper’s account of scientific method, or some variant thereof, as an account of theory testing, while maintaining that there are a number of methods that facilitate theory generation. The implausible claim that there is no “logic of discovery” is often appended to characterizations of the hypothetico-deductive method, but that claim is not a part of the method itself. In an environment of methodological pluralism, one is free to use the hypothetico-deductive method for theory-testing purposes and to employ abductive methods to generate explanatory theories. It is relevant to note here that Bhaskar (2016) himself refrained from casting the hypothetico-deductive method in all-embracing terms, remarking that it derives whatever plausibility it has as “a particular phase of scientific inquiry” (p. 79).

One of the attractions of the Popperian account of the hypothetico-deductive method is its insistence on strong tests of hypotheses and theories. However, as is widely known, Popper’s method has been criticized for restricting scientific inference to deductive inference. It has also been criticized for its inability to provide sufficient methodological resources for carrying out strong tests. It is germane to point out here that Mayo’s (2018) error-statistical philosophy offers an important corrective on both counts. Mayo endorses Popper’s strategy of developing scientific knowledge by identifying and correcting errors through strong tests of scientific claims. Making good on Popper’s lack of knowledge of statistics, she also showed how one can properly employ a range of familiar error-statistical methods to implement the requirement that knowledge claims be severely tested. In marked contrast with Popper, Mayo’s conception of falsification stresses the importance of inductive inference in science. This viable account of falsification should alleviate Archer’s worry about being handicapped by the strictures of deductivism—an account, moreover, that provides the operational detail about how to conduct strong tests.

The DREI(C) model of scientific discovery

Archer briefly presents Bhaskar’s preferred alternative DREI(C) model of inquiry as a replacement for Popperian hypothetico-deductive method. This is a wide-ranging five-step procedure. The first step involves the (1) description of empirical patterns or phenomena. The second step employs (2) retroductive reasoning in order to fashion a number of competing candidate theories that possibly explain the empirical patterns. The next step adopts the strategy of (3) eliminating those theories that are judged inapplicable to the case at hand. Step four consists in (4) identifying the salient causal generative mechanism(s) at work. The final step involves the iterative (5) correction of earlier findings in light of the identification of those mechanisms. Although Archer (2024) says little about this method, it is worthy of comment.

Despite Bhaskar’s (2016) claims for its comprehensiveness, the DREI(C) model of inquiry confines itself to a concern with the construction of explanatory theories. Bhaskar (2016) was clear about this restriction: “Critical realism is interested in theoretical . . . rather than empirical generalizations” (p. 79). In fact, the DREI(C) model begins with descriptions of empirical generalization already in hand. In doing so, it ignores the rich insights about experimental practice gained by the philosophy of science over the last 40 years. One aspect of the emergence of the “new experimentalism” is the important scientific process of detecting empirical phenomena (Haig, 2014; Woodward, 1989). Briefly, phenomena are stable recurrent features of the world that we seek to explain. Consistent with my earlier remarks about ontology, it should be emphasized that phenomena comprise a varied ontological bag that includes objects, states, processes, events, and other features that are hard to classify. Archer, BCR, and the literature on replication largely focus on empirical regularities. However, not all phenomena take this form. For example, cognitive science often takes capacities (e.g., the ability to learn a second language) to be the things that need explaining. Moreover, capacities are often known to us, in which case the task is not so much to discover the phenomena (say, through replication), but more to provide an informative specification of them.

Replication

Archer’s perspective on psychology’s handling of the modern replication crisis deserves a more detailed response than space permits. Here, I briefly reply to a number of claims he makes.

Unlike Archer (2024), I do not think that Popper’s philosophy of science has exerted a major influence on psychology’s solutions to the replication crisis (or on psychology, more generally). My work on replication (Haig, 2022), aspects of which Archer criticizes, is framed neither by Popperian philosophy of science nor by BCR. Instead, it selectively draws from the experimentalist literature in modern philosophy of science—something that BCR largely ignores. In doing so, I described practices that are often found in good science, some of which are at variance with a number of claims made by the mainstream methods reform movement.

Without question, replication is important in science, but it is just one element in a wide array of endeavours. By fixating on replication, psychology’s reform movement has overemphasized its importance in science. Further, replication itself is just one of a number of reliabilist strategies for evaluating the consistency of claims about empirical phenomena. Other means of justifying empirical claims include methodological triangulation, controlling for extraneous factors, the calibration of instruments, and meta-analysis. For this reason, one need not subscribe to the DRIE(C) model of inquiry in order to quell the demand for replication.

Much of the substantive and methodological literature on replication places it too close to theory construction endeavours. For the most part, replication speaks directly to phenomena detection, whereas claims about phenomena provide the appropriate evidential grounds for theory construction. For this reason, building theories often takes place without any direct concern with replication. Contrary to what Archer (2024) claims, constructive replication tests the scope of empirical generalizations, not the scope of the application of theories.

Undertaking successful replications is considerably more complex than most researchers and methodologists acknowledge. Even though low levels of replication success are to be expected in parts of psychology, it should be noted that some critical realists argue that (“demi”) regularities do obtain in partially closed systems (Kaidesoja, 2013; Lawson, 1997). I agree with Archer (2024) that critical realism usefully draws attention to the neglect of the ontological dimension of replication but, as already noted, I think its three-tiered, stratified ontology is too inflexible to capture the variations in subject matter that psychological researchers often encounter.

Inference to the best explanation

It is a virtue of Bhaskar’s DREI(C) formulation of the inquiry process that it explicitly acknowledges the importance of abductive, or retroductive, reasoning (here, I ignore Bhaskar’s fine-grained distinction between the two). To his credit, Bhaskar (2016) distinguished between the abductive generation of theories (creative abduction) and their subsequent evaluation by explanatory means (inference to the best explanation; IBE). However, he does not take either of these ideas about explanatory inference very far.

In his attractive formulation of critical realism, Tuukka Kaidesoja (2013) claims that the inferential nature of Bhaskar’s DREI(C) model of inquiry resembles IBE. For Kaidesoja, there are two aspects to this form of inference: the retroductive postulation of candidate explanations, followed by the elimination of those candidates that are judged implausible. Although Kaidesoja does not describe IBE in detail, his characterization bears a similarity to Lipton’s (2004) well-known account of IBE. Kaidesoja (2013) acknowledges the need to develop his account of IBE more fully. Incorporating Lipton’s (2004) systematic perspective on IBE would be one way to do so.

Taking the idea of IBE further, Paul Thagard (1992) formulated a theory of explanatory coherence that provides the researcher with a method for making judgements of the best of competing explanations. This theory states that to infer that a theory is the best of competing explanations is to judge it as more explanatorily coherent than its rivals. Explanatory coherence obtains when the propositions of a theory hold together because of their explanatory relations. The determination of the explanatory coherence of a theory is made in terms of three criteria: explanatory breadth, simplicity, and analogy. These criteria are embedded in a computer programme that facilitates judgements of explanatory coherence in a reliable way.

Thagard’s method of IBE should appeal to critical realists because they are squarely concerned with evaluating theories in terms of explanatory goodness and, importantly, this method enables researchers to carry out such a complex task. As such, it could be made part of what Bhaskar (2016) hoped critical realism might become.

Conclusion

Archer (2024) is to be commended for introducing the philosophy of critical realism to theoretical psychologists. However, those who are attracted to critical realism would do well to shed BCR’s reliance on a priori transcendental arguments and redirect their attention to Kaidesoja’s (2013) naturalist reworking of this philosophy. It has the advantage of employing a good deal of “middle-level,” science-informed philosophy of science—something that is lacking in Bhaskar’s brand of critical realism. For my part, I favour a strategy of employing relevant insights gleaned from local forms of scientific realism (Haig & Evers, 2016), rather than the strictures imposed by scientific realisms of a global nature.