While Pyschology does pride itself to be 'the study of mind and behaviour', the fact of the matter is that the 'mind' is an abstract amorphous entity changes to/of which are realizable but not quantifiable, while 'behaviour' is a consequence of said state changes but not a direct measure of the change itself. The impossibility of direct measurability and observability of internal brain states thus poses a fundamental problem in the "empiricism'' of psychology. To get around this, researchers instead are forced to rely on multiple proxies such as behavioral and physiological changes to infer these internal brain states or change thereof.
For decades, the field has invested substantial resources into developing forward causal models. These models attempt to explain how environmental inputs induce changes in brain states, which subsequently lead to observable and quantifiable behavioral and physiological changes in subjects.
While these physiological changes are not the end goal in itself, they can serve as means to an end. These causal models when formalized in terms of invertible statistical models and fitted to the experimental data can be used to tap into the latent variables. Changes in the physiological reading over time can then be traced back to changes in the latent variable with reasonable reliability.
A thing to be noted here is that only the change in the variable's state is measured, that too through the arbitrary lens of the measurement device (the computational "method"). Accuracy of these measurements still remains an elusive metric by virtue of the latent variable being latent and possessing no 'true' value. Yet, it is possible to gauge the goodness of the method at hand. We could exploit the prior knowledge that a given change in the experimental setup is expected to induce a specific directional change in the latent variable under consideration. A good method thus is one that accurately picks up on the direction of the change, maximizes its apparent value, and performs relatively consistently across different participants undergoing the same experimental setup. This goodness of performance is captured as the method's retrodictive validity, and can be useful in finding the best method for the job.
By continuously refining these models and methodologies, we move closer to bridging the gap between observable phenomena and the elusive brain states they represent.
