Rules of Conduct for Behavior Analysts in the Presence of Hypothetical Constructs: A Commentary on Eckard and Lattal (2020)

Abstract

Eckard and Lattal (2020) summarized the behavioristic view of hypothetical constructs and theories, and then, in a novel and timely manner, applied this view to a critique of internal clock models of temporal control. In our three-part commentary, we aim to contribute to the authors’ discussion by first expanding upon their view of the positive contributions afforded by constructs and theories. We then refine and question their view of the perils of reifying constructs and assigning them causal properties. Finally, we suggest to behavior analysts four rules of conduct for dealing with mediational theories: tolerate constructs proposed with sufficient reason; consider them seriously, both empirically and conceptually; develop alternative, behavior-analytic models with overlapping empirical domains; and contrast the various models. Through variation and selection, behavioral science will evolve.

Even a cursory survey of any textbook about psychology or its history reveals that most, if not all, psychological theories include hypothetical constructs. These postulated, mediational entities tend to figure more prominently in some approaches than others. They occur in greater numbers and degrees of inference in psychological approaches that conceive of behavior as the output of internal information-processing devices, or as the outcome of computations performed on internal representations, than in behavior-analytic theories. Eckard and Lattal (2020) remind us why. The authors describe the well-known radical behavioristic critique of these constructs, their potential contributions to our understanding of behavior, and the negative effects that are likely to occur when they are used incorrectly. The novelty and perhaps main contribution of Eckard and Lattal’s article is that the authors apply the behavior analytic critique to a specific and reasonably well-defined research domain, that of the temporal control of behavior.1 The authors summarized internal-clock models and theories of temporal control, exposed their philosophical foundations, identified their strengths and weaknesses, and then sketched an alternative behavior-analytic view of temporal control. It is rare to see contemporary scientists engage in internal epistemology—the critique of the assumptions of a specific theory, the analysis of its terms, and the evaluation of its associated research practices. The net effect of Eckard and Lattal’s article is a clearer view of the obstacles to a better understanding of the temporal control of behavior.

Our three-part commentary of Eckard and Lattal’s article aims to contribute to the discussion about the advantages and dangers of the use of hypothetical constructs in psychological theories. We first expand upon some of its statements about the contributions of hypothetical constructs. Next, we refine and occasionally question its view of the perils of these constructs. Finally, we address the quintessential question Eckard and Lattal’s analysis raised: What should behavior analysts do in the presence of mediational models such as the internal clock-based theory of temporal control? We answer the question differently than Eckard and Lattal did.

The Contributions of Hypothetical Constructs and Theories

According to Eckard and Lattal, radical behaviorism acknowledges three potentially useful roles of hypothetical constructs in a science of behavior. First, they "help scientists more effectively wade through pockets of ignorance" (p. 6). The hypothetical status of the construct identifies a gap in the scientific account (e.g., a missing link in a causal chain). In the 18^th century, physicists postulated invisible magnetic fluids to explain the magnetization of metals, and phlogiston to explain combustion; in the 19^th and 20^th centuries, biologists postulated genetic factors to explain the principles of inheritance. In each of these cases, some empirical regularities (the magnetization of a piece of iron when rubbed with loadstone; the calcination of metals when heated; the ratios in the progeny of hybrid crossings) could not be accounted for by current theories. A hypothetical construct attempted to fill the gap. Likewise, when a pigeon exposed to a fixed-interval (FI) schedule refrains from pecking a key during the first two-thirds the interval and then pecks it steadily during the last third of the interval, we ask how it discriminates the two parts of the interval if nothing in its environment—the lights, sounds, smells, and textures that the bird senses—seems to differ consistently between them. Appealing to an internal clock attempts to fill a gap in our knowledge and, in doing so, may point scientists in a direction out of ignorance.

Second, hypothetical constructs may help to integrate disconnected facts, interrelate them, and see them as expressions of the same process. As the authors put it, “it certainly is useful to have condensed terms and expressions by which to refer to a behavioral phenomenon like “timing” (p. 13); and “Quantitative accounts of behavior and hypothetical constructs have proven useful in integrating data across experiments and laboratories by pointing to higher-order behavioral processes” (p. 14). Unfortunately, Eckard and Lattal did not elaborate on this function of hypothetical constructs nor provide an example to explain what a “condensed term” is or to illustrate how a construct succeeds at “integrating data.” Theories with hypothetical constructs, regardless of whether they are cognitive such as the scalar expectancy theory, or behavioral such as the behavioral theory of timing or the learning-to-time model, integrate data from a variety of experimental paradigms by appealing to the same hypothesized process to account for the major temporal properties of behavior in these paradigms (see, e.g., Carvalho, Machado, & Vasconcelos, 2016; Church 2003, 2004).

And third, hypothetical constructs and theories “may provide a useful heuristic framework from which to view behavior that is sensitive to the temporal properties of reinforcement” (p. 12); “It is worth repeating that these constructs initially can serve as useful heuristic tools in developing an account of any behavioral process if they suggest examination of manipulable variables that had yet to be examined” (p. 17). To illustrate, in a suggestively titled article “A Clock Not Wound Runs Down,” Killeen, Hall, and Bizo (1999) examined whether the transition across behavioral states, the bases of temporal control according to the behavioral theory of timing, slows down when reinforcement rate decreased. Likewise, Roberts and Church (1978) examined whether the hypothetical internal clock could function as a stopwatch (i.e., start, stop, and restart without losing the accumulated time) when the stimulus associated with periodic reinforcement is suspended temporarily.

The heuristic value of hypothetical constructs and theories may rest on their metaphorical nature and their use in analogies—rats build cognitive maps, time intervals, expend responses from a reflex reserve, turn on random generators, and strengthen links between internal states and responses. Other sciences show similar uses of metaphor and analogy. William Harvey conceived of the heart and lungs as water bellows; physicists in the 18^th century conceived of a piece of iron as a sponge that could absorb the magnetic fluid; Niels Bohr conceived of the atom as a miniature solar system. Metaphor and analogy may be central to theory construction, at least in the early stages of an investigation, because of their generativity. When scientists use analogies, they superimpose on a new research domain the conceptual structure of a familiar, old domain (e.g., Bunge, 1967). The elements of the new domain and their interrelations are seen in a new light. On occasion, significant advances may occur (see, e.g., Farber, 1950; Harris, 1912; Leary, 1990; Leatherdale, 1974).

The Harmful Effects of Hypothetical Constructs and Theories

According to Eckard and Lattal, hypothetical constructs are used inappropriately when they are reified and assigned causal properties. Then they become an obstacle to scientific progress. When constructs "migrate from being descriptions or metaphors of the behavior to becoming reified and morphing into its cause" (p. 6), they change their theoretical status from mediators to direct causes of behavior, and they change their psychological status from that of a tool that helps researchers interpret results to that of an object for future research. The point is made repeatedly in Eckard and Lattal’s article: “Invoking constructs like the internal clock, however, as a cause of behavior distracts and deflects the focus of the analysis from the environment or organism’s learning history as sources of the controlling variables of behavior" (p. 12); “the transition of the construct from a heuristic to primary causal factor is a serious problem in current approaches to psychological research if they [sic] deter examination of possible influential variables” (p. 17).

There are two problems with the foregoing analysis, though. First, because the consequences of reification are eminently practical,2 they invite empirical checks: Is it the case that when using internal-clock or similar models researchers cease to investigate how environmental variables affect temporal discriminations? When we look at studies inspired directly by such models, even those aimed at measuring parameters or examining the role of putative components of the model, we continue to see the standard manipulation of environmental variables and the standard analysis of their effects on measurable behavior. In fact, we even see the unraveling of new functional relations between environment and behavior, as well as the conception of alternative models.

Take the work by Roberts and Church (1978) mentioned above. Its main purpose, “was to find out the amount of control possible with the rat’s clock” (p. 319), and to examine some of the “qualitative properties of the clock” (p. 334). In particular, to find out whether the internal clock could function as a stopwatch (i.e., an experimental endeavor clearly motivated by the properties assigned to a hypothetical set of constructs), the authors assessed how brief interruptions of the discriminative stimulus that signaled the FI affected temporal performance. They manipulated environmental variables, including the moment of the interruption, and the task (fixed interval, temporal categorization, multiple fixed intervals), and measured temporal performance using standard behavioral methods, plotting absolute and relative response rate as a function of absolute and relative time from interval onset (psychometric functions and points of indifference). Results showed that temporal control persisted even in the presence of stimulus interruptions; the rats continued to track the time during which the stimulus was on and allocate their responses accordingly.3 The authors also showed that when rats start the trial with one signaled FI and then change to a different signaled FI, they are able to take into account the time elapsed in the presence of the first stimulus and subtract it, as it were, from the second one, integrating the two intervals in an absolute rather than proportional fashion. Finally, the authors provided some evidence that responding was better explained as a function of the time since the beginning of the interval as opposed to the time until the end of the interval (i.e., rats seem to time up instead of down). Regardless of the authors’ interpretation that the internal clock can function as a stopwatch, their work provided new information about how environmental variables affect temporal discriminations and inspired many other studies to use the now called gap procedure.

Second, the effects of reification that Eckard and Lattal identified may not be inevitable outcomes (“the transition . . . is a serious problem . . . if they [hypothetical constructs] deter. . .” [p. 17; emphasis added]). Perhaps, we argue, constructs can be reified and assigned causal roles without a researcher straying from the search for the environmental determinants of behavior. To clarify, consider a variant of the argument known as the theoretician’s dilemma (Hempel, 1958): If environmental events, S, relate to internal states, O, according to function f [i.e., O = f(S)], and if these states in turn relate to behavioral events, R, according to function g [i.e., R = g(O)], then R relates to S, [i.e., R = g(f(S)] and O can be eliminated.4 As Eckard and Lattal put it, echoing Skinner,

if the third link [R] is functionally related to the first [S], then we can disregard the second link [O] as a superfluous addition to an otherwise parsimonious account of behavior. In some instances, there may be value in investigating the second link in this chain; however, to the extent that the second link is represented by a hypothetical entity, the intensive work that follows may be for naught. (p. 12)

The first thing to notice about that argument is that even when researchers are interested in O, they cannot avoid manipulating S in ways specified by function f, and measuring R in ways specified by function g. In addition, in empirical sciences, it remains an open question whether the second link can be removed—i.e., whether the role of hypothetical constructs linking S to R can be duplicated by (and the construct reduced to) the effects of current context and history of environment–behavior interactions on behavior.5 The more we know about f and g, the more convinced we become that reduction is possible. For example, the contents of long-term memory in internal-clock models may be reduced, at least in part, to the durations of the reinforced intervals experienced by the animal; in the learning-to-time model, the vectors of associative strengths may be partly reduced to the history of reinforcement rate at successive moments into the trial.

Where we seem to differ from Eckard and Lattal is in the possibility of full reduction. Eckard and Lattal’s position on this issue is unclear. On the one hand, Eckard and Lattal seem to believe that full reduction is possible and perhaps even desirable:

The central view here, however, is that once their heuristic function is exhausted and the facts laid out, the influence of these constructs in formulating causal accounts should be diminished greatly if not removed entirely. In other words, once the circumstances “driving” constituent parts of the “clock” is known, why refer to the clock at all? (p. 6; emphasis added)

In the same vein, when Eckard and Lattal dissect a temporal discrimination “to avoid mediation” (p. 15), they seem to imply that a theory of temporal control need not include any hypothetical constructs. On the other hand, Eckard and Lattal also admit (see p. 10) that some “cogent functional relations” between time intervals and behavior may be due to “certain processes of the clock,” and that the mathematical specification of these processes could help to describe the functional relations in a parsimonious way. In this case, the value of the hypothetical constructs advises against their elimination. But then they add an important caveat: If the mathematical expressions contain parameters that are reducible to neither behavior nor stimulus properties, the theory “is inconsistent with Skinner’s formulation of how a theory of behavior ought to be constructed. . . . In the present case, the clock and its various expressions constitute a dimensional system outside the level of environment–behavior interactions” (p. 14). To summarize, according to Eckard and Lattal, certain processes of the clock may be important to understanding cogent functional relations, but they should be discarded if their mathematical specification includes parameters irreducible to stimuli and responses. The authors seem willing to sacrifice our understanding of cogent function relations, a fundamental goal in a science of behavior, to preserve some form of dimensional purity in the theory. The sacrifice may be for naught.

We are skeptical about the possibility or desirability of full reduction in the use of hypothetical constructs. As Eckard and Lattal acknowledge, some hypothetical constructs may render the theory more parsimonious and easier to use. In this case, although dispensable, the constructs should be retained. And some other hypothetical constructs may not be dispensable because they describe properties of the organism that are irreducible to current context or ontogenetic history, and yet are of fundamental importance to understanding behavior. A case in point may be the processes that yield the scalar property, the equivalent of Weber’s law for time, and arguably the most “cogent functional relation” in the area. The scalar property does not seem derivable from current context or the animal’s learning history.

Two other cases in the domain of temporal control may be the proportionality between the post-reinforcement pause and the FI duration, roughly two-thirds at the steady state (e.g., Schneider, 1969), and the bisection of two temporal intervals at their geometric mean (Catania, 1970; Church & Deluty, 1977; Stubbs, 1968). The derivation of these properties may depend on structural properties of the organism’s nervous system (including sensory and motor apparatuses), properties that, at present, may require for their characterization some form of mediational model that uses hypothetical constructs with surplus meaning. In these cases, the constructs are not dispensable.6

What Should Behavior Analysts do? Four Rules of Conduct

Eckard and Lattal’s radical behavioristic argument concerning hypothetical constructs is likely to conjure up the image of the Roman god Janus, with its two faces looking in opposite directions. One face sees the potential contributions of hypothetical constructs and theories as signposts to our ignorance, as data organizers, and as discovery engines. The other face sees the harmful effects bound to occur when researchers fall into the reification and causal-assignment traps. Advantages and disadvantages pull the researcher in conflicting directions and create a paralyzing tension. What then should behavior analysts do to resolve the tension? How should they act with respect to hypothetical constructs, models, and theories? Eckard and Lattal’s answer consists of two precepts: If you can, avoid hypothetical constructs and mediational models; if you cannot, exploit their potential but “do not reify, do not ‘physiologize’” (p. 14). Because we agree with Eckard and Lattal’s analysis of the virtues of hypothetical constructs, but not with the full thrust of their analysis of the negative effects of reification, we answer the question of what behavior analysts should do differently than Eckard and Lattal did.

Eckard and Lattal offer their critique of the internal clock "as part of the process of selecting, within behavior analysis, those constructs and mechanisms . . . that have the greatest utility in advancing a science of behavior” (p. 17; emphasis added). We believe they may have underestimated the other critical process in the evolution of a science of behavior: variation, mainly theoretical variation. Although they acknowledge that “radical behaviorism and the science that derives from that philosophy broadly benefits from variation in perspectives” (p. 17), they add that not all variation is “equally compatible with or equally useful to the science” (p. 17). That may be true, but it might also be true that there are no adequate criteria to distinguish between potentially useful and useless variations. Incompatibility with behaviorism may be an excessively strict and unwise criterion. For even if we accept that a reified construct has a low probability of capturing a real internal entity or process and therefore that most of our efforts to examine a reified construct may be in vain, we do not recommend rejecting it outright because, as the old adage says, closing the door to all Error is bound to leave Truth outside. Perhaps radical behaviorists should heed to Galileo’s admonition not to act in any way that may proscribe Truth.7

Behavioral science is young. Its large number of metaphors and analogies may attest to a developing science. Given this early stage of growth, with no consensual theories with a history of producing generalizable explanations of behavior and new testable hypotheses, our first rule is greater tolerance to hypothetical constructs. Although metaphors and analogies may push toward autonomous thinking and even seduce reason, they “should be tolerated, faute de mieux, in the period of theory construction. Thus it would have been foolish to reject the hydrodynamical analogies of heat and electricity ‘flow’ in the beginning—as foolish as regarding them as substantial [analogy] rather than formal [analogy]” (Bunge, 1967, pp. 281–282). We accept hypothetical constructs but suggest that their value be probed. The condition we impose upon them at this stage is that of sufficient reason, the same that Kepler imposed on the astronomers of his day: researchers, he said, “should not, in fact, be given absolute license to assume anything whatever without sufficient reason. You ought to be able to give probable reasons for the hypothesis you claim to be true causes of appearances” (cited in Duhem, 1996/1908, p. 144).

Having accepted the constructs and theory, the researcher then ought to consider them seriously. Our second recommendation: probe the power of the constructs to accommodate empirical findings and check their internal consistency. The former is well-known to researchers and does not need further elaboration; the latter is less familiar (Machado & Silva, 2007a, 2007b), but it may be particularly important for at least two reasons. First, when initially proposed, hypothetical constructs always have surplus meaning. The concept of force, for example, evolved significantly over several centuries and part of its evolution consisted in eliminating its initial surplus meanings (Jammer, 1957). Second, nascent theories often have implicit and ambiguous rules to coordinate their constructs (see Machado, Lourenço, & Silva, 2000; Machado & Silva, 2007a). Conceptual analysis exposes surplus meanings and implicit or ambiguous rules, and thereby helps to clarify constructs, refine their meaning, and constrain their uses. It may even reveal contradictions within a theory. When Machado and Silva (2007a) and Machado, Erlhagen, and Malheiro (2009) put scalar expectancy theory under the conceptual microscope, they identified contradictions related to the formation of temporal memories and ambiguities related to the use of threshold-based decision rules. If these shortcomings can be removed, a new and improved theory will emerge.

By taking hypothetical constructs and theories seriously, behavioral analysts reap an immediate and important social benefit: they communicate more with researchers with different philosophical and epistemic persuasions, which can only increase the visibility and influence of their own research. When he was editor of JEAB, Hineline (1984, p. 1) expressed a similar idea: “I would like that variation to include articles that develop and clarify relationships between behavior analysis and research from other scientific traditions.” In the area of temporal control, this communication has occurred on a few occasions (e.g., Fetterman & Killeen, 1995; Gibbon & Church, 1992; Killeen & Fetterman, 1993; Leak & Gibbon, 1995; Machado & Guilhardi, 2000).

Another potential benefit is that behavior analysts will be challenged or inspired to build alternative models and theories with the same empirical domain, which is our third recommendation. Behavioral-analytic models have distinctive and contrasting features. For example, instead of sequestering temporal control “away from other behavioral processes” (p.17), they relate it to other forms of stimulus control (e.g., Carvalho et al., 2016). Moreover, instead of deriving steady state performance from complex internal representations and computations, they try to derive it from the processes of establishing temporal control, that is, by studying the acquisition of temporally discriminated and temporally differentiated behavioral patterns. An illustrative case is our research contrasting behavioral and cognitive models in the bisection and double-bisection temporal tasks (e.g., Carvalho, Machado, & Tonneau, 2015; Carvalho et al., 2016; Machado & Keen, 1999).

Model comparison, our fourth recommendation, follows naturally. Model comparison and ultimately model selection are difficult issues (e.g., Myung, 2000). But they are also unavoidable because internal clock models and theories cannot be ignored. Consider Gibbon and Church’s scalar expectancy theory (e.g., Church, 2003, 2004): it fits a variety of data sets; it has shown significant heuristic value; it is well-developed mathematically; and it has influenced other research areas (e.g., Allan, 1998; Allman, Teki, Griffiths, & Meck, 2014; Gibbon, Malapani, Dale, & Gallistel, 1997; Meck, 1996). Over the last three decades, it has been the standard against which new models and theories of timing are assessed. It has important limitations, but so do all other theories and models. By continuing to compare models, checking the similarities and differences among their components (e.g., Machado et al., 2009; also Hass & Durstewitz, 2016), better theories of temporal control may evolve. The future may even show that the contents of the cognitive representations present in one class of models matches the cumulative effects of the organism’s learning history present in the other class of models.

By means of empirical tests and conceptual analyses, hypothetical constructs, models, and theories are selected: some are short-lived, others survive transformed. Constructs tend to become interconnected in wider networks and these interconnections may gain mathematical expression. This evolutionary process (Hull, 1988) may be conceived of as domesticating the constructs by selecting against their wild features of surplus meaning, implicit, fuzzy, and even contradictory rules. It is somewhat ironic, but a domesticated metaphor is a dead metaphor. Bachelard (1968/1940, p. 119) summarized the view aptly: “Intuitions are very useful: they serve to be destroyed. . . . The diagram of the atom proposed by Bohr a quarter of a century ago [likening it to a miniature planetary system] has, in this sense, acted as a good image: there is nothing left of it.”

Conclusion

The differences between the cognitive and behavioral approaches in explaining behavior seem to run deeper than models, analogies, and metaphors. To borrow an idea from Holton (1988), they seem to lie in

. . . the dimension of fundamental presuppositions, notions, terms, methodological judgements and decisions—in short, of themata or themes—which are themselves neither directly evolved from, nor resolvable into, objective observation on the one hand, or logical, mathematical and other formal analytical raciocination on the other hand. (p. 41)

As intellectual resources, themata are at once powerful and restricting and for that reason we have proposed a balanced set of action guidelines. Consistent with Eckard and Lattal’s arguments, researchers should explain behavior whenever possible not by moving in space, from without to within the organism, but by moving in time, from the present to the past interactions of the organism with its environment (see also Hineline, 1990). It can be argued that this is behavior analysis’s most powerful thematic contribution to psychological science. As we follow this approach, and place learning at the center of timing research, we also tentatively accept a wide range of mediational constructs, theories, metaphors, and analogies. To be blunt, in the context of discovery, where variation and novelty reign supreme, almost anything goes. Thus, a variant proposed with sufficient reason should be entertained. But then we insist on anchoring the constructs to environmental and behavioral variables to give them operational content; on rendering explicit the grammar that coordinates them, preferably via mathematically framed models, to give them analytical content; and on contrasting the models that include them with alternative behavior-analytic models. When we make the transition to the context of justification, where we need to reconcile not only things with things, but also minds with minds (Bachelard, 1984/1934), selection becomes the supreme ruler. Janus, the god of gates and transitions, reappears, but this time we no longer feel the paralyzing tension; we know how to act.

Compliance with Ethical Standards

Conflict of Interest

None of the authors has any conflicts of interest relevant to the research.

Human and Animal Rights and Informed Consent

The research reported here used no human or animal participants.