HIT on the Psychometric Approach

Click to increase image sizeClick to decrease image size

Notes

¹There are occasions when scientists do explain data—for example, when they suspect that the data are an artifact of experimental procedures (Bogen & Woodward, 1988). However, this is a very different type of explanation than is the primary focus of cognitive neuroscience. We therefore adopt the “accounting for” language when the focus is on measurements to avoid conflation of very different contexts that might both be broadly “explanatory.”

²By interpreting the latent variable causally, the authors are making a metaphysical claim, but it is not the sort of ontological commitment that can help explain cognitive phenomena. In effect, there is not enough metaphysics involved—explanation involves making a commitment to particular phenomena with particular inherent properties, not simply describing a small set of external relations that may or may not track anything going on in the world.

³The authors employ a subtype of formative model, called a “MIMIC” model, which stands for Multiple Indicators, Multiple Causes. Notably, however, the relationship involved in supervenience is one of realization, not of causation. The authors are thus providing a novel interpretation of the MIMIC model, which renders its traditional title misleading in this context.

⁴We have two concerns that are specific to the treatment of latent variables in formative models. The first involves the metaphysical status of the latent variables. There is an ongoing debate regarding the ontological status of latent variables, in which one of the authors of the target article is a participant (Borsboom, 2008; Borsboom et al., 2003). The status of latent variables in formative models is particularly vexed, because (as admitted in the target article) in formative models, “the latent attribute is defined by the choice of predictors” and “a change of predictors implies a change in the nature of the attribute” (p. 72). As a result, scholars disagree over whether to take a broadly realist or antirealist interpretation of latent variables in formative models. This is a topic one would hope to see further addressed by the authors, because it has serious consequences for a metaphysical interpretation of the model. If it is best to adopt an antirealist approach to the latent variable in formative models, then clearly the metaphysical goals of the authors will be unsatisfied. Second, it is unclear whether the formative model actually captures multiple realizability. The authors state that “two people can have different indicator values but the same position at the latent attribute level. Therefore the position on the theoretical attribute is multiply realizable” (p. 74). This statement is doubly perplexing. First, if the latent variable just is some function of the neural values, it would seem to be itself neural in nature, but traditionally, supervenience theory has concerned the multiple neural realizability of psychological features. The authors speak of the latent variable as if it were psychological, but it is unclear why they feel licensed in doing so. Further, if we assume (in accordance with the model) that the “different indicator values” regard N-indicators, then the claim is true only if two people's brains are measured with the very same styles of measurement. Add or subtract to the data and the latent variable changes. Assigning the same numerical value to the latent variables would then not be assigning the “same” value to the “same” latent variable; it would not be multiple realizability.

⁵In a way, the situation is worse for the formative models. Because the latent variable is only posited to determine the measurements, instead of causing them (as on the reflective model—see footnote 2), it is unclear that the formal rendering of the latent variable is invoking a metaphysical claim at all. This is related to the antirealism worry in the previous footnote.

⁶In most accounts, mental simulation is viewed as sufficient to show how the mechanism could generate the phenomenon being explained. But in many biological mechanisms, including ones found in neuroscience, the organization of operations is not sequential and the operations themselves are nonlinear. Accordingly, computational models and application of tools from dynamical systems theory are required to explain how the parts and operations work together to generate the phenomenon. Bechtel and Abrahamsen (2010) characterized such explanations as “dynamic mechanistic explanations.”

⁷Research on the brains of nonhuman animals was possible, though, and yielded important insights into the mechanisms involved in vision (Bechtel, 2008), memory (Craver, 2007), and other phenomena.

⁸Supervenience is a much weaker relation than identity, only requiring that operations characterized in neuroscience terms always map onto the same psychologically characterized process but not vice versa. As noted by Kievit et al., supervenience is compatible with the psychological process being multiply realized. Putnam (1967) argued that mental states are multiply realizable, because different species appear to exhibit the same mental states despite major differences in their physical brains. Multiple realization is often taken as the death knell to identity claims. But, as Bechtel and Mundale (1999) argued, neuroscientific research has long treated activity in brains that exhibit morphological differences as the same. In doing this neuroscientists use a coarse-grained account of neural states that is comparable to the coarse-grained account required to treat the psychological states as the same. If one insists on finer-grained accounts in neuroscience so as to recognize differences in brain processes (as the multiple-realizability argument does), a comparably finer-grained psychological account will also find differences between species (or individuals). The important point to note here is that if the relation of mental states to brain states is to serve as a productive heuristic in guiding research, it is identity, not supervenience, that is required. There may be legitimate cases of multiple realization, resulting for example from convergent evolution, but the heuristic identity theory advocates that researchers settle for supervenience only if serious efforts fail to reveal psychological differences in brains that are differently organized.

⁹Here we gloss over several of the early debates that occurred in the field. Defenders of the modular position, citing the fact that activation of the FFA was slightly greater in response to faces than other stimuli, argued that the FFA is primarily a face recognition area and that the other activations were ancillary or residual (Kanwisher, 2000). This does not alter the situation from the standpoint of HIT, however—the FFA being primarily involved in face recognition is not the same as the claim that it is identical to face recognition. This move by early defenders of the modularity hypothesis, then, was in fact an early recognition of the need for further research, as we illustrate next.

¹⁰For a differing opinion based on a technical objection to the use of high-resolution fMRI, see Baker, Hutchinson, and Kanwisher (2007).