When calculate signal detection metrics (e.g. sensitivity) from 2-alternative false choice task behavior data, we assume that the subject is internally comparing a variable to a threshold: if it is higher than threshold, then choose A, otherwise choose B.
However recent neurophysiological data and drift-diffusion model seems to suggest that animal makes a decision whenever an internal variable crosses a threshold.
Does this render the assumptions in signal detection theory invalid?
It turns out that the statistical analysis of any binary decision (choice between two options) can be reduced to a model based on a single decision variable (a scalar variable) crossing a threshold. This is true no matter how the decision making is actually carried out. As I recall (I might be mistaken), a proof of this is given by Jaynes in his text "Probability Theory: The Logic of Science".
The strength of the signal detection theory in psychology is that it can be applied no matter what is taking place in the brain. The theory gives a good way of studying decision performance, as viewed by the external objective observer.
It is wrong to think that the signal detection theory requires or implies that there necessarily or actually exists a single biophysical or neural level of some kind in the brain of a subject, that crosses a threshold of some kind, causing them to make a decision about signal detection. The theory applies even for very complex processes, possibly involving countless unknown levels and thresholds for instance, conscious or unconscious.
The theory organizes and enlightens observed decision behaviour, but it does not explain the underlying decision process in any way.
I am familiar with this matter from automatic target detection, where the outcome of any expert system or AI deciding whether a target is present or absent can be -- ultimately must be -- studied using the signal detection theory. And then also, together with psychologists, from assessing human target detection performance versus AI, which requires the signal detection theory to be uniformly uniformly to people and to AI, regardless of their internal functioning.
The point of signal detection theory is not that an actual single decision-determining internal decision variable exists in the decision maker. The theory does not require this. The point is, as I believe Jaynes proves, that the decisions made can always be treated as if this were the case. Signal detection theory says nothing about how the decision is actually arrived at in the subject's brain.
- Badges
- Science topic
- Similar topics
- Psychology