Retiring Popper: Critical realism,falsificationism,and the crisis of replication

The nature of experimental activity contra Hume

How does Bhaskar establish the transcendental indispensability of his stratified ontology? Well, for Bhaskar, the nature of experimental activity is completely at odds with the dominant Humean understanding of causality, which holds a scientific law to consist solely of observed “constant conjunctions” of atomistic (or unstructured) events. For Hume, the external world consists of nothing more than contingently related events, and so causality is untenably implied to reside in the (observed) invariant regularity of such events: instead of A caused B, we have A occurred followed by B over time. Thus, the basic problem with the Humean approach is its conflation of patterned regularity with natural necessity (i.e., with the causal powers and properties of natural objects). So, for example, every morning when I boil a kettle of water (event A) and make a fresh cup of coffee (event B), Hume cannot account for the fact that such regularity is explained by the irreducible natural properties of water, electricity, and so on, and the necessary connection between my purposeful activity of filling the kettle, switching it on, and so forth. Indeed, my watching the news on the television after I make my coffee every morning (event C) shows just how untenable his position on causality is, since the observed constant conjunction of events A, B, and C precludes the fact that such events involve relations of both necessity and contingency (watching the television is not a necessary coffee-making precondition).

Basically, experimental activity is simply unintelligible in Humean terms. For in experimentally testing the boiling point of water, for instance, we are seeking to explain why water boils at 100 degrees Celsius, which is independent of, and irreducibly contingent to, any observed regularity. And it boils because the water molecules have enough kinetic energy to overcome their intermolecular bonds, break away from each other, and form gaseous molecules of vapour. For Bhaskar, it is such knowledge that provides us with our transcendental warrant for positing real generative causal mechanisms that underlie any observed events. As he puts it,

That generative mechanisms must exist and sometimes act independently . . . and that they must be irreducible to the patterns of events they generate is presupposed by the intelligibility of experimental activity. But is up to it actual experiments to tell us what the mechanisms of nature are. (Bhaskar, 1975/2008, p. 42)

At the same time, it is the latter that underpins his stratification of reality into the strata of the real, the actual, and the empirical. All three strata are real but necessarily differentiated to capture the fact that structures and mechanisms (the real) are distinct from the events to which they give rise (the actual) and that such events are not necessarily experienced (the empirical). All three levels must not, therefore, be conflated or collapsed so that we can provide satisfactory explanatory accounts of scientific activity. Humean empiricism remains wedded to the level (or domain) of the actual, that is, at the level of observable events. Bhaskar has termed this actualism, which entails a denial of the real existence of underlying generative structures that account for things and/or events.

Furthermore, Bhaskar introduces the categories of intransitive and transitive to differentiate ontologically between the independent existence of the objects, mechanisms, and events of the natural world and the processes of conceptual scientific investigation. For Bhaskar, the errors of conflating the intransitive and transitive objects of science can be characterised as committing either the epistemic or ontic fallacies respectively. The epistemic fallacy is the view that statements about being can be reduced to or analysed in terms of statements about knowledge; the ontic fallacy is the view that knowledge is direct and unmediated.

Experimentation, causal tendencies, and the artificial creation of invariant regularity

A further problem with Humean actualism is the assumption of invariant regularity or regularity determinism. This is the idea that for every event y there is an event x or set of events x₁. . .x_n such that the latter are constantly conjoined. This presupposes that nature is a closed system, which is fundamentally not the case. The whole point of experimentation is to create (artificially) invariant regularities to understand what the natural world is like outside the experimental setting because the natural world is an open system. To create artificial (systemic) closure, there are two conditions that must be in place. The first condition is an intrinsic one, which means there must be no change or qualitative variation in the object if causal mechanisms are to operate consistently. The second, extrinsic condition for closure concerns the relationship between the causal mechanisms and those of its external conditions that must remain constant to ensure regularity of outcome. In turn, when both conditions are met, this necessarily precludes the development of any novel sui generis properties.

So, if we return to the example above of the boiling point of water, the successful replication of any experiment(s) to establish its 100 degrees C boiling point derives from the fact that scientists will have ensured both conditions for closure have been met, namely, that there are no impurities in the molecular structure of water (intrinsic condition) and standard atmospheric pressure obtains (extrinsic condition). Bhaskar (1975/2008) himself gives the example of a simple electrical experiment designed to illustrate Ohm’s Law. Here, the wiring of an electric circuit and the generation of an electric current would constitute what he calls experimental production; and the maintenance of appropriate resistance levels, and so on, would then constitute what he calls experimental control. The aim of an experiment is to activate and isolate a single mechanism and record its effects. Outside a closed system, these will normally be affected by the operations of other mechanisms, “either of the same or of different kinds, too, so that no unique relationship between the variables or precise description of the mode of operation of the mechanism will be possible” (Bhaskar, 1975/2008, p. 43).

In the open system that is nature, then, we find that water does not always boil at 100 degrees C, which explains (a) the transcendental intelligibility of experiments and (b) the use of ceteris paribus clauses or reference to law-likeness. Talk of “other things being equal” or law-likeness derives from the fact that we are not dealing with the existence of Humean constant conjunctions. In turn, this signals the importance of causal laws conceptualised as tendencies that dispose things to behave in certain ways, which may be exercised without being realised or manifest in any particular outcome. Indeed, the very reason for Bhaskar’s tripartite distinction between the real, the actual, and the empirical derives from the fact that powers may (a) remain unexercised (power of water to dissolve salt); (b) exercised but unrealised (water does not boil at 100 degrees C because of impurities); or (c) realised but unperceived (water boiling as result of volcanic eruption with no-one to observe). That water does not always boil at 100 degrees C highlights its tendential dispositional nature. Moreover, it underscores the nonexistence of perfect regularities (Anjum & Mumford, 2018). Of course, the fact that water does generally boil at 100 degrees C means that we can talk of tendential (or imperfect) regularity, but not invariant regularity.

Emergence and essentialism

For Bhaskar, there are three senses of stratification. The first sense concerns the causal structures that give rise (tendentially) to events, which furnishes his distinction between the real and the actual. The second sense consists in the multitiered stratification of reality that is revealed in the creative historical development of science. Thus, the observable properties of water are explained by a deeper level of reality that is described by the theory of atomic number and valency, which is further explained by the theory of electrons and atomic structure. The third sense is defined by the special form of multitiered stratification shown by emergence. Emergence concerns the ontologically irreducible powers or properties of objects. Critical realists often cite water as exemplary here, since its power to extinguish fire cannot be derived from its constituents (hydrogen and oxygen are highly inflammable). Equally, our capacity to feel, to think, and so forth, cannot be reduced to our brain chemistry: our emotions and thoughts are sui generis real. For realists, then, the emergent properties of water, emotions, and thinking exist at a different stratum from their constituents.

Furthermore, realism is necessarily committed to the doctrine of essentialism, which is the view that objects possess certain essential properties that make them one kind of thing rather than any other. At some basic level, we need the concept of essence to be able to discriminate between objects, for example, between water and wood or people and animals (Lowe, 2008). The rationale of science is to “dig deeper” as to what makes water precisely what it is, and here we know that its essence is H₂O. In understanding essences and their emergent powers and properties, realism emphasises the importance of distinguishing between internal and external relations. Thus, irreducible necessity characterises water, namely, the internal relationality of hydrogen and oxygen (they each necessarily presuppose each other, and water would not be what it is essentially without them). As we have seen with its tendential power to boil at 100 degrees C, externally or contingently related objects like impurities may interfere with the effects of the exercise of its powers, but they do not affect one another in their respective essential natures.

Transcendental realism and the social sciences

In extending transcendental realism to the social sciences, Bhaskar (2015, 2016) has argued that its applicability turns on the extent to which an independent analysis of the objects of social and psychological knowledge is consistent with his transcendental realist theory of science. Bhaskar argues that there are three ontological limits to such applicability: social structures, unlike natural structures, do not exist independently of (a) the activities they causally condition; (b) agents’ conceptions of what they are doing in their daily activities; and (c) may be only relatively enduring. Furthermore, he underscores the fact that human agency may operate under unacknowledged conditions of action, create unintended consequences, require tacit skills, and involve unconscious motivations. Finally, he argues that social systems are radically open, thereby rendering impossible any attempts at experimental closure. (The actual extent and nature of such “radical” societal openness is a matter to which I will return later.)

As a condition of its applicability, a transcendental realist model of social science requires an intransitive object of study that is characterised by ontological depth, emergence, and contains causal mechanisms. The intransivity of social structures derives from the fact that they both preexist and presuppose human agency; ontological depth concerns their causal irreducibility (social structures possess emergent properties—or essences—that are irreducible to people); and they are causal mechanisms precisely because they necessarily delimit agential activity. As with the essential nature of H₂O, we can talk of internal necessity vis-à-vis the essence of social structure. For example, a teacher presupposes a pupil, a headteacher presupposes teachers, and so on. Here, we are talking about internal relations between roles, ontologically distinct from the individual people who fill them and whom they causally affect. The teacher–pupil relation is an irreducible emergent property because the powers deposited within the role modify the powers of the individuals qua individuals. This modification arises from the combination of internally necessary social relations: a teacher cannot self-award an A-grade certificate, just as a pupil cannot revoke the decision of a national examination board. Such powers do not reside in the properties of individuals but in the social relations that presuppose such individuals for their enduring efficacy and mediation (Archer, 2002a).

Equally, the causal powers of social structure may be conceptualised in terms of causal tendencies (i.e., as conditioning human agency), evidenced by the fact that not all students regularly attend school. In other words, in the case of pupil absenteeism, whilst socioeconomic structures provide objective reasons for attending school (the need to obtain qualifications for future employment, potential status opportunities, and so on), such structures do not operate within a closed system nor determine in hydraulic fashion. Indeed, it is precisely the combination of the conditioning nature of socioeconomic systems and people’s irreducible psycho-biographies that account for imperfect regularities. In strictly causal terms, there is no difference between natural and social reality, which is demonstrated by the reality of contingencies that interfere with the exercise of causal powers (water does not always boil at 100 degrees C and pupils do not always attend school or follow teachers’ instructions). The ontological difference consists in the subjective mediation of objective human-made social structures: hydrogen and oxygen cannot subjectively weigh up the benefits of boiling at 100 degrees C in the way pupils can weigh up the structured benefits (or penalties) of (non)attendance. Clearly, then, it is important to emphasise that whilst Bhaskar rightly accentuates the ontological gap between irreducible social structures and individual persons—such that he contends social science ought to be an account of the former and psychological science an account of the latter—we must not lose sight of the fact that there must be considerable overlap (Groff, 2004), which means that psychologists cannot bracket-off (or give a post hoc nod towards) complex sui generis socioeconomic causal mechanisms that are operative in the open system that is society, just as social scientists cannot bracket-off irreducible psychological causal mechanisms.

Popper’s methodological individualist repudiation of emergent properties

As we have seen in the above example of the teacher–pupil emergent property, social reality, like natural reality, is transcendentally characterised by ontological depth or stratification, since we are dealing with the causal irreducibility of social relations to people. The sui generis nature of emergent social properties (from the micro, to the meso and macro levels) necessarily transcends individual predicates, and so any insistence upon purely individualistic explanations for behaviour would remove the very rationale of sociology, thereby also undermining any adequate theorising of the psychology of agential motivation. It is at this juncture that we can begin to understand just how deeply problematic Popper’s ontology is, and hence why he is both an inappropriate ally for the reformers and for social scientific methodology in general. For, as I will elaborate in more detail later, his repudiation of essentialism in favour of nominalism means that he is unable to sustain an ontology of necessity, that is, of irreducible social and natural causal mechanisms. The task of critical realist social theory is to identify and theorise the essential properties of sociocultural systems and their agential mediation (Archer, 2002a), whereas for Popper,

the task of social theory is to construct and to analyse our sociological models carefully in descriptive or nominalist terms, that is to say, in terms of individuals, of their attitudes, expectations, relations, etc.—a postulate which may be called “methodological individualism.” (Popper, 1961/2004, p. 126)

Such “methodological individualism” nullifies ontological emergence at a stroke, by reducing the essential nature of emergent properties (such as the internal social relation between teacher and pupil) to individual properties. In nominalist fashion, Popper (1961/2004) asserts that when we engage in theoretical abstraction, we are liable to see “a kind of permanent ghost or essence” (p. 126) behind observable events rather than accepting that we are only dealing with concrete entities such as crowds or groups. So, for Popper, we should be talking—reductively—in terms of “people in uniform” in lieu of such abstract concepts as “army.” Whilst permanency is not a defining characteristic of an essence, the salient point is that it is hard to see how Popper can avoid the charge of psychologism (Chalmers, 1985) in view of his repudiation of emergence. Ultimately, this means we can never explain why individuals behave in qualitatively different ways in different social contexts since the (transcendental) grounds for the latter have been ontologically removed. For example, when at school, a teacher is not exercising powers as husband or wife. In explaining the exercise of marital powers, necessarily one differentiates a different spatio-temporal setting (in the marital home, and so on). The fact that children behave in qualitatively different ways at different times (why Christopher undertakes his maths examination at 1 o’clock, watches television at home at 5 o’clock and follows the instructions of his scout leader at 7 o’clock) warrants the autonomy of social structure (Archer, 2002a). Without such autonomy, how do we account for the fact that Christopher is unable to take his maths exam during his scout meeting? Such behaviour cannot be reduced to statements about individuals per se. It is not the characteristics of Christopher as an individual that account for his behaviour in the three irreducibly emergent structural settings. Thus, Chalmers (1985) rightly takes Popper to task for his assertion that the valid aspect of psychologism lies in the fact that “it rightly insists that the ‘behaviour’ and the ‘actions’ of collectives, such as states or social groups, must be reduced to the behaviour and to the actions of human individuals” (Popper, 1966, p. 87). Indeed, Christopher’s scout meeting is not simply a social group composed of individuals who happen to hold specific beliefs about scouting, since such beliefs presuppose an irreducible emergent context as a necessary condition of their intelligibility.

Transcendental realism and psychology

For Martin et al. (2010), unless it can be demonstrated that at least some features of human psychology are irreducible to physical, biological, and sociocultural properties, psychology has “no distinctive subject matter of its own and can readily be absorbed by fields of inquiry judged more fundamental to the constitution of psychological subject matter (e.g., neurophysiology, evolutionary biology, computational science, and cultural studies)” (p. 9). Here, transcendental realism can provide the ontological underpinning that secures the distinctiveness of psychology. As we have seen vis-à-vis social science, a transcendental realist model of psychology requires an intransitive object of study that is characterised by ontological depth, emergence, and contains causal mechanisms. Here, Bhaskar (2015) maintains that intentional human actions are the result of the mental states of actors, and such mental states are the intransitive reasons that (transcendentally) constitute the psychological analogue of generative mechanisms in nature. Ontological depth concerns the irreducibly emergent causal efficacy of reasons, and they are causal mechanisms precisely because they explain human intentionality.

However, Bhaskar commits two mistakes in how he theorises intentional processes: (1) he views reasons as efficient causes and (2) he conceives of action as a subtype of physical event rather than as the execution of purpose (Groff, 2004). For Groff, reasons are not efficient causes but final (i.e., purposeful) causes. She points out that she is not describing the same mistake in two different ways, since (1) and (2) concern Bhaskar’s incorrect answers to the questions: “In what sense are reasons causally efficacious?” and “What are actions?,” respectively. In terms of (1), Groff is making the simple yet crucial point that our reasons for action are irreducibly meaningful and cannot be conceptualised in terms of efficient causation. However, in terms of (2), she is making a more complicated point that Bhaskar does not distinguish adequately between action and behaviour. Briefly, Groff draws upon the work of Charles Taylor (1985), who distinguishes between “causal” and “qualitative” ways of understanding human action. The former considers actions as being events like any others, which happen to have psychological causes and which are physical with nonmaterial causes; the latter holds actions as differing ontologically from other events, where actions and their causes cannot be disaggregated—they “just are the expressions of purposes; they cannot be characterised first in non-purposive terms, and only then. . . connected to a mental or emotional state” (Groff, 2004, p. 126). The problem is that Bhaskar untenably proffers the “causal” view, arguing that actions can be redescribed independently of their reasons.

The false problem of induction, Popper’s “modified essentialism,” and the inevitable failure of deductivism

Popper’s (1969) critical rationalism begins by rejecting induction as a scientific method and replacing it with a continuous process of conjecture and refutation. It is important, therefore, to spell out why the “problem of induction” is a reductio ad absurdum because its ostensible insolubility provides the basis for Popper’s falsificationist deductivism. Basically, the Humean problem of induction concerns our alleged inability to assume that because an event (or series of events) has always been observed to occur in the past, it will do so in the future (e.g., because the sun has always risen in the past, it will do so tomorrow). As we have already seen, Hume proscribes making any causal inference by purely a priori means. Instead, in actualist fashion, he maintains it can only be based on observed constant conjunction. But this unavoidably forces us to reach the absurd conclusion that we lack any firm grounds for trusting all past experiences (the sun may not rise), and thus throws into sharp relief the fact we cannot avoid the transcendental realist argument for natural necessity, that is, the transcendental presumption of irreducible intransitive causal mechanisms. Of course, whilst we know the Sun appears to rise from the horizon (and that it is the Earth’s motion that causes the latter), the crucial point is that we know that the hydrogen fuelling the sun is not likely to run out for a few billion years. In other words, induction is a false problem because science is in the explanatory business of discovering the intransitive nature and structure of the world (and the solar system), with the latter providing us with our rational warrant for trusting that the sun will continue to rise.

Ultimately, Hume’s problem of induction is easily resolved by accepting a realist (stratified) ontology of internally structured and differentiated objects that possess causal powers and liabilities. In turn, we can then distinguish between qualitative change and mere successions of events and thus between necessary or causal changes and relationships and accidental ones (Sayer, 2010). This stands in stark contrast to Hume’s universalisation of contingency, which is the sine qua non of constant conjunction. However, Popper is contradictory in his approach to natural necessity, essentialism, and natural laws. He stated that universal laws “explain regularities or similarities of individual things or singular facts or events,” since they are “conceived . . . as (conjecturally) descriptions of the structural properties of nature—of our world itself” (Popper, 1972, pp. 195–196). Furthermore, “the ‘principle of the uniformity of nature’ can again be regarded as a metaphysical interpretation of a methodological rule – like its near relative, the ‘law of causality’” (Popper, 1959, p. 253). Finally, the principle “expresses the metaphysical faith in the existence of regularities in our world (a faith which I share, and without which practical action is hardly conceivable)” (Popper, 1959, p. 252). I cannot imagine any scientist, let alone critical realist, baulking at such talk of explanation, structural properties, and the principle of the law of causality.

Nevertheless, one can readily “pin down” Popper’s ultimate position, which irredeemably undermines his talk of explanation of regularities in terms of their (necessary) structural provenance, at the same time removing the very possibility of doing experimental science:

I reject all what-is questions: questions asking what a thing is, what is its essence, or its true nature. For we must give up the view, characteristic of essentialism, that in every single thing there is an essence, an inherent principle . . . which necessarily causes is to be what it is, and thus to act as it does . . .. We must give up the view that . . . it is the essential properties inherent in each individual or singular thing which may be appealed to as the explanation of this thing’s behaviour. (Popper, 1972, p. 195)

Furthermore, Popper (1959) abstains “from arguing for or against faith in the existence of regularities in our world” (p. 253), and, moreover, he asserts that we cannot “ever [emphasis added] find out if any given non-logical statement that it is in fact naturally necessary” (p. 454). In fact, for Popper, the concept necessary is a “mere word – a label useful for distinguishing the universality of laws [emphasis added] from ‘accidental’ universality . . . I largely agree with the spirit of Wittgenstein’s paraphrase of Hume: ‘A necessity for one thing to happen because another has happened does not exist. There is only logical necessity [emphasis added]’” (p. 460).

As I said above, Popper’s work is frustrating to read, and his equivocal “largely agree” vis-à-vis the Humean repudiation of natural necessity exemplifies this, since the point, of course, is that one either does or does not endorse natural necessity. In this regard, Popper’s unequivocal rejection of “what-is” questions (essentialism) must surely be taken at face value, together with his assertion that necessary is a mere word, which incontrovertibly means he completely agrees with Wittgenstein and Hume’s denial of natural necessity. In fact, Popper’s response to a reviewer who described his position as a “modified essentialism” is clear in its antiessentialism: “to avoid misunderstanding, I wish to say here that my acceptance of this term [modified essentialism] should not be construed as a concession to the doctrine of essentialist definitions. I fully adhere to the criticism of this doctrine . . . given in my Open Society, Vol. ii” (Popper, 1972, p. 195). And in his Open Society, Vol. 2, he untenably asserts that

methodological nominalism is nowadays fairly generally accepted in the natural sciences. The problems of the social sciences, on the other hand, are still for the most part treated by essentialist methods. This is, in my opinion, one of the main reasons for their backwardness. (Popper, 1966, p. 30)

This is a clear and unequivocal rejection of natural necessity, since science is not in the nominalist business of explaining nature: the essential properties of water should be sufficient to underscore this.

Crucially, then, his denial of natural necessity and essentialism in favour of nominalism necessarily undermines his insistence that we accept the methodological rule “that we are not to abandon the search for universal laws . . . nor ever give up our attempts to explain causally any kind of event we describe” (Popper, 1959, p. 61). For Boulter (2002), Popper thus wants to have “intelligibility on the cheap: he wants to join the empiricists in their anti-necessitarianism, and yet keep all the benefits just the same” (p. 73), which, of course, he cannot do. Indeed, it strikes me that Popper simply wants to have his (ontological) cake and (methodologically) eat it. And in eating his ontological cake, he ends up removing any rational warrant for scientific experimentation and discovery. The nonexistence of such a warrant derives from his repudiation of natural necessity and essentialist “what-is” questions (in favour of logical necessity) because an antiessentialist (i.e., nonstratified) ontology means we have nothing to get our scientific teeth into (given that irreducible natural entities are held no longer to exist). This is precisely what justifies Maxwell’s (1972) argument that it remains completely problematic how we could ever have good grounds for supposing a theory to be falsified, and Sayer’s (2010) point that Popper is forced to commit himself to a universalised ontology of contingency.

The inevitable failure of Popper’s deductivist “solution”

Given that Popper will only countenance logical necessity, it is inevitable that his “solution” to the Humean account of the problem of induction is doomed to fail, since he has removed the ontological rug from underneath his scientific feet. The salient point is that logic does not, and cannot, exhaust reality. For the possibility of argumentation presupposes that logical relations among propositions and conclusions are only part of the story: we cannot make sense of any proposition of an argument if such a proposition is not ontologically grounded, that is, transcendentally it must be about something other than itself. This is precisely the problem: for anything to be falsifiable presupposes that there are correct irreducible ontological grounds for us to be able to falsify in the first place. However, central to Popper’s solution is his “hypothetico–deductive” procedure in which scientists are held to advance “bold” hypotheses or conjectures from which testable propositions can be deduced. Here, Popper claims that by construing science as an attempt to falsify a hypothesis rather than verify it, we shift our focus from the generation of the hypothesis to the process of testing it (Burns, 2009). Essentially, Popper (1969) says we make a deductive prediction from our hypothesis and look for empirical evidence that falsifies the prediction. Yet herein lies the overarching problem: shifting our focus away from hypothesis-generation, that is, from addressing ontological “what-is” questions, results in a truncated or logicist philosophy of science that “has little penetration of what most theoretical disputes are about and is unable to say why we don’t seek to fit just any data into such deductive systems” (Sayer, 2010, p. 114).

Of course, this is to be expected given that Popper denies natural necessity. But the upshot of his deductivist approach is that any conclusion can be deduced from an infinite number of (potentially absurd) premises precisely because there are no ontological grounds both to make sense of, and delimit, such premises. Hence Sayer’s (2010) absurd deductive example of “All woods conduct electricity; copper is a wood, therefore copper conducts electricity” (p. 115). So, whilst Haig (2022a) is quite right to note that any replicative failure would not threaten the inferential structure of the hypothetico–deductive model itself, this is to miss the point, for the model reductively redirects our attention away from natural necessity.

Time to dispense with falsificationism and reclaim the search for truth

I agree with Dykes (1999) that one of the most troubling aspects of Popper’s philosophy is his refusal to countenance anything positive, despite his (inconsistent) acknowledgement that many theories are positively confirmed, that is, not falsified. Again, however, we should not be at all surprised that Popper champions falsification, since his rejection of natural necessity enjoins that there is nothing with which our truth-claims can correspond and explains his disdain for definitions (see Note 5). With consistency, then, he maintains thus:

The way in which knowledge progresses, and especially our scientific knowledge, is by unjustified (and unjustifiable) anticipations, by guesses, by tentative solutions to our problems, by conjectures. These conjectures are controlled by criticism; that is, by attempted refutations . . . They may survive these tests; but they can never be positively justified: they can be established neither as certainly true nor even as “probable.” (Popper, 1969, p. xi)

But knowledge cannot progress if all we are doing is conjecturing and refuting and our conjectures can never be positively justified. Of course, Popper is quite right to assert the latter, with his additional denial of the possibility of probability, for there is nothing to ground such nonprobable conjectures. ⁶ Whilst Popper (1972) commits a performative contradiction when he maintains that “all knowledge is hypothetical” (p. 30), since the latter proposition cannot be conjectural, ultimately, Dykes’ (1999) correct point that Popper does not seem to understand that “truth implies a ‘what is’ question every time critical rationalism tells what is not” (p. 14) stems from his illicit ontological destratification of the intransitive. It is the search for how things really are, which provides the rationale for the natural and social sciences. The Popperian negative injunction to falsify is thus wholly indefensible for transcendental realist reasons: first, the very practice of science is incomprehensible without the presupposition of an ontologically stratified natural world; second, truth is a precondition of all human inquiry, and is best understood as a formal regulative norm (Archer, 2002b).

The DREI(C) and RRREI(C) alternatives to falsification

In contradistinction to Popper’s falsificationist hypothetico–deductive approach, Bhaskar (2016) proffers what he calls the DREI(C) schema, which involves a five-step logic of scientific discovery. Step one, D, consists in the description of some pattern of events or phenomena. Step two, R, involves retroduction, which involves the hypothesising and identification of the causal mechanisms that tendentially give rise to events and phenomena. Step three, E, requires the elimination of those mechanisms that do not apply. What follows next, step four, I, is the identification of the causally efficacious generative mechanism or structure at work. Finally, step five, (C), involves the iterative correction of earlier findings in the light of this identification. For areas of research that are already well theorised, retrodiction, RRREI(C), may be utilised, which stands for resolve using existing theory, redescribe in terms of relevant theories, retrodict back to antecedent states of affairs; identify causal mechanisms, and possibly correct the overall picture considering the fuller explanation (Price & Martin, 2018).

Tendencies and “demi-regularities”

To recap, it is because natural generative mechanisms necessarily operate as tendencies (in concrete open systems, which causally depend on other mechanisms that cannot/do not get included in any experimental set-up) that accounts for the nonexistence of perfect regularities. And, of course, the existence of “imperfect regularities” characterises the social sciences (including psychology) as they, too, deal with open systems. However, there is a generic lack of clarity vis-à-vis the precise nature of regularity in the replication literature, some of which strikes me as perilously close to being parasitic upon a closed-system (i.e., nontendential) ontology. For example, whilst Haig (2022a) is right that, in the context of his discussion of Popper’s insistence upon experimental closure, intersubjective testing practices do not have to take the form of replication, his (unqualified) comment that “the regularity of the occurrence of events underwrites replication and other reliabilist validation strategies” (p. 227) is redolent of presupposing systemic closure, since it is not the regularity of events, pace Haig, that underwrites replication, but rather the artificial creation of such regularity of events through human manipulation. Indeed, he states that “phenomena are stable [emphasis added], recurrent features of the world that we seek to explain. Many of them are noteworthy regularities [emphasis added] and are often called effects” (p. 225). The problem here is that empirical effects are not characterised by invariant “stability” or regularity: such effects are independent of Haig’s “recurrent” stable phenomena. In other words, the psychological stability of agential decision-making processes (forethought, self-reflection, etc.), for example, is subject to open-systemic sociocultural contingencies that may interfere with or override such stable phenomena to produce irregular empirical effects (or none).

Of course, Haig might well retort that (real-world) regularity is not (and cannot be) invariant, and so we do need to be crystal clear about what we mean by regularity. First, then, the use of the stratified metaphor of the real, the actual, and empirical can help dispel any potential confusion, since Haig’s “recurrent features” are not invariant at the levels of the actual and the empirical, and so we must be careful not to confuse or conflate the (relatively) invariant essential nature of mechanisms (level of the real) with their contingent manifestation in an open system (Haig’s “effects”). Second, as Næss (2019) argues, we should differentiate nonclosed systems according to their degree of openness, rather than applying a strict dichotomy between closed and open systems, ⁹ in turn permitting talk of “demi-regularities” or “demi-regs” (which complements Anjum and Mumford’s “imperfect regularities,” 2018). Furthermore, he argues that demi-regs represent probability distributions existing in the real world. Such empirical demi-regs thus derive from relatively enduring causal tendencies (social structure). It is on this basis that Næss maintains that an understanding of demi-regs as repetitive in nature “implies that certain kinds of outcome (e.g. car commuting) are more probable in the given time-space context than alternative outcomes (e.g. commuting by transit) for individuals belonging to a certain category (e.g. being a suburbanite)” (p. 477).

Næss’s (2019) focus on probability relationships fits nicely with my discussion of the teacher–pupil emergent property, since the constellation of causal mechanisms here (education system, central government, economic relations, etc.) constitutes strong probability reasons for the empirical regularity of high pupil attendance. To reiterate, such probability relationships are not invariant—“they vary across time and space in occurrence as well as in strength. Still, in a given context . . . relatively strong relations may persist over a long time (yet with temporal variations in strength)” (p. 478). But we are still left with the task of explaining why some replications have been successful. Basically, such success derives from the fact that the structures of nature are differentially tendential because of their object-specific essential properties and simultaneous existence in an open system. As Anjum and Mumford (2020) argue, tendencies “come in degrees, just as a wineglass and a car windscreen are both fragile but the former to a higher degree than the latter . . . Some of those tendencies, especially in ideal conditions, will almost always manifest” (p. 117). I would suggest this can explain both why replicative success is possible (in some cases), and why such a large proportion of laboratory-based replications are reported to fail. That some generative mechanisms are more amenable to replicative success is attributable to their (object-specific) essential power(s) that underpin their durable stability and the extent of their known causal dependence on other mechanisms. In the case of glass, we are dealing with an invariant stable physical structure, which can be experimentally established by heating a mixture of lead oxide, zinc oxide, and boric acid strongly until it melts, and by changing, for example, the relative contribution of boric acid. We would reasonably expect a high replicative success rate in terms of how much boric acid is required.

Nevertheless, any experimental investigation of how glass actually “plays out” in the real world makes replication a trickier endeavour, since we find here that glass is not invariantly stable but is characterised by real-world degrees of stability (Gaviria et al., 2022). This is why experiments that test the strength of glass continue to remain tentatively exploratory (Santos et al., 2018) precisely because glass “inhabits” an open system where we cannot always know the extent of necessary dependence on other mechanisms and contingent interference. Thus, replication cannot be assumed to succeed on every occasion, as unique and unexpected causal factors may be discovered. However, this does not mean, of course, that we cannot say anything causally meaningful about the properties of glass. What we can say is that under certain conditions glass is more likely to behave in a certain way. It is in virtue of the nature of the object under consideration that we can talk of degrees of likelihood—hence why wine glass is more fragile than car windscreen glass as the former tends to shatter more easily than the latter at the concrete level.

It should be clear from the foregoing that natural science’s endeavour to gain entrée to the open system of natural reality via the door of replication is inherently limited (which is why nonreplication has been found to be a normal part of the scientific process). It follows that replication is equally limited in the social sciences (including psychology) for the same open-systemic reasons. Despite this, it is not just the open-system ontology that underpins my serious reservations about Haig’s (2022b, p. 822) assertion that the social and behavioural sciences can close their systems of study sufficiently to allow for experimental practice, as well as with Stroebe’s (2016) claim that exact replication is likely “the more a social psychological phenomenon is exclusively dependent on basic psychological processes” (p. 140). In addition, as I will argue, ontologically speaking, psychological phenomena do not possess the same kind of relative stability or invariance as physical phenomena because of their innate “plasticity.” So, whilst I agree with Stroebe that there exist basic psychological processes (or “stable recurrent features,” as per above), such processes are inherently susceptible to reductive misidentification if subject to any form of experimentalism (replicative or otherwise).

The intrinsic condition

I mentioned above that psychological phenomena do not possess the same kind of relative stability or invariance as physical phenomena because of their innate “plasticity” or malleability. The notion of plasticity concerns the intrinsic potential for personal systematic change over the course of a lifetime (Lerner, 1984). For such change to be possible requires (transcendentally) that psychological phenomena are not invariant or fixed, unlike the invariant properties of natural science, thereby undermining the intrinsic condition for closure. As Iso-Ahola (2017) neatly puts it,

psychology is a science of subtleties in cognition, affect and behaviour. Its phenomena reside in and arise from the human mind, whether conscious or non-conscious [emphasis added], and as a result, are not static, but instead dynamic and changing, varying with internal and external conditions. (p. 2)

This means that any endeavour to effect experimental closure (replicative or otherwise) ever confronts unconscious mental phenomena, of which the relative causal efficacy cannot be predetermined. It is the latter that necessarily vitiates Stroebe’s “likely” success and Haig’s possibility of closure. It is hardly surprising then, as Haig himself admits, that any rate of replicative success in social psychology can be expected to be relatively low. Equally, the suggested relatively higher rate of success for cognitive psychology might be attributable more to de-concretised narrowness than to uncovering actual causality. But the salient point is that the “subtle nature of psychological phenomena and their sensitivity to social influences means that a previously observed effect may not appear under the seemingly same laboratory conditions at later times” (Iso-Ahola, 2017, p. 3). Such subtlety means they may not appear under the same conditions since there are unconscious influences at play, which cannot be intrinsically manipulated/predicted precisely because they are unconscious and inherently temporal and transient (De Luca Picione, 2021).

However, even though the intrinsic psychological condition for closure cannot generically be met (hence the relatively low level of replicative success in social psychology), this does not mean we cannot talk about causal tendencies and thus demi-regs in psychological terms. For whilst the gap between what we might call (following Anjum and Mumford) the “messy” reality of psychosocial life and the requirements of replication cannot be closed, it is not an unbridgeable one. Such variable “bridge-ability” derives from the degrees of stability that characterise psychosocial phenomena.

The extrinsic condition and the need for real-life (stratified) context

At the same time, the extrinsic condition for closure is equally problematic for replication and experimentalism in general. If we recall, the extrinsic condition for closure concerns the relationship between the causal mechanisms and those of its external conditions that must remain constant to ensure regularity of outcome. At the outset, the simple fact that any psychological experiment involves the complex interplay of conscious and unconscious psychic mechanisms means that the extrinsic condition can never be held a priori to be met, for there is an ever-present susceptibility to contingent psychic generative mechanisms. However, the main problem here concerns the fact that psychological phenomena are “more likely to surface in the real world than in artificial laboratory conditions” (Iso-Ahola, 2017, p. 3). This sums up the case against exact or direct replication and experimentalism in general, since psychological causal mechanisms qua tendencies are intrinsically intractable to abstracted isolation (as measurable variables) by virtue of their interdependent irreducibility, of which any adequate explanation enjoins spatiotemporal and historical contextualisation. This is not dissimilar to my example of glass experimentation, where data are established tentatively precisely because such experimentation necessarily disconnects or cuts-off key causal properties from their natural concrete setting.

In other words, as argued above, natural scientific experimentation unavoidably misses (to varying degrees) causal mechanisms and processes in effecting the extrinsic condition for closure. This applies equally to any psychological experiment since the latter requires an artificially created disconnection from concrete reality. The key point is that, as Van Geert and De Ruiter (2022) put it, “phenomena that researchers claim to control are therefore, in reality, actually entangled processes interacting within persons, in the context of materials, context and time . . . they cannot be manipulated independently of one another” (p. 210). This lack of contextualisation is consistently ignored, paid lip-service to, or even denied within the replication literature. ¹⁰ In essence, replication temporally freezes, ontologically flattens, and reductively decontextualises its subject matter.

To understand what we mean by interdependent irreducibility (i.e., context), critical realism begins with the need to understand the nature of abstraction and the importance of that from which we abstract (Sayer, 2010). An abstraction (or abstract concept) pinpoints a one-sided or partial aspect of an object. That from which we abstract is the multiplicity of features that constitute concrete objects such as people, social structures, and activities. As well as furnishing precision, the things to which abstractions refer have ontological parity with those things referred to by more concrete concepts. The concept of “concrete objects” denotes the fact that objects are constituted by a combination of diverse elements. As Sayer (2010) notes, a particular person qua concrete entity combines influences and properties from a wide range of sources (physique, intelligence, attitudes, and so on) “which might be isolated in thought by means of abstraction, as a first step towards conceptualizing their combined effect. . . . Abstractions should distinguish incidental from essential characteristics” (pp. 59–60). Furthermore, and crucially, sometimes the aspect of an object that is abstracted cannot exist in an abstract form “but only in particular concrete forms [emphasis added], which this ‘contentless abstraction’ ignores. . . [such contentless abstractions] seem harmless enough until we come to put some ‘explanatory weight’ upon them or try to measure [emphasis added] what is abstracted” (p. 67).

Intelligence, for instance, is a “contentless abstraction,” which is why experiments are inappropriate to understand its nature because it can only be understood in concrete form, that is, in the context of the “real world.” And it is this transcendental requirement that makes any endeavours to measure it doomed to failure because of its inherent concrete internal relationality. This is not to deny the sui generis reality of intelligence; it’s simply to recognise that it is embedded within the complex irreducible psycho-biological systems of concrete embodied human beings who themselves are embedded in complex and open sociocultural systems. In experimentally “stripping” intelligence of the latter would be like trying to abstract and measure the nature of sugar in a cake. Or it would be like trying to measure the nature of “pupil effects” and “teacher effects” on a pupil’s examination performance: the irreducible internal social relation between pupil and teacher (and its wider overdetermined psychosocial–systemic anchorage) cannot be measured in any meaningful sense. But this does not mean its effects, which are grounded in underlying social generative causal mechanisms, are not real. The latter point applies with equal force to understanding emotions, since they are internally related to cognition (Green, 1992) and so cannot be measured or isolated in experimentalist abstraction. Furthermore, we have seen that cognitive psychology can be expected to be relatively more successful at replication vis-à-vis social psychology, yet cognitive scientists standardly employ abstract, as opposed to, concrete universals, which necessarily provides only half of the explanatory picture (Shillcock, 2014).

To be clear, then, the basic problem with both replication and experimentalism in general is the de-concretisation of psychosocial reality, thereby resulting in a fundamental loss of explanatory power. There is a lopsided focus on abstraction at the expense of the concrete and disregard of whether objects and processes are internally or contingently related: experimentalism “decontextualises those processes and controls out the complexity associated with real life” (Pilgrim, 2020, p. 26). It is the concrete (real-life) complexity to which Pilgrim refers that defines context. At the same time, such complexity encompasses the stratified and open nature of psychosocial reality, with no a priori limit placed on the number of nested social-cultural and psycho-embodied systems.