Testing the Foundations of Signal Detection Theory in Recognition Memory

Abstract

Signal Detection Theory (SDT) plays a central role in the characterization of human judgments in a wide range of domains, most prominently in recognition memory. But despite its success, many of its fundamental assumptions are often misunderstood, especially when its comes to its testability. The present work examines five main assumptions that are characteristic of existing SDT models -- the existence of a random scale representation, independent sampling, monotonic likelihood, Receiver Operating Characteristic (ROC) asymmetry, and a continuous (non-threshold) representation. In each case, we derive a testable consequence and test it against data collected in the appropriately designed recognition memory experiment. We also discuss the connection between yes-no, forced-choice, and ranking judgments. This connection introduces additional behavioral constraints and yields an alternative method of reconstructing yes-no ROC functions. Overall, the reported results provide a strong empirical foundation for SDT modeling in recognition memory.

A Decision-Congruent Heuristic Gives Superior Metacognitive Sensitivity Under Realistic Variance Assumptions

“…Otherwise, theoretically, σ 2 Target/σ 2 Nontarget needs to be estimated in a separate task by using different methods than confidence rating, such as manipulations on target/nontarget presentation probabilities and different incentives for Yes/No responses (for details, see Maniscalco & Lau, 2012. Realistically, however, stable estimation of σ 2 Target/σ 2 Nontarget would be difficult at the individual subjects' level, as empirical ROCs constructed with these methods are usually not as smooth as theoretically supposed to be (e.g., Kellen et al, 2018;Macmillan & Creelman, 2005, pp. 71-72).…”

Bias in confidence: A critical test for discrete-state models of change detection.

“…Note that the assumption that ROC functions are concave (which includes linearity as a boundary case) implies that the likelihood ratio is monotonic. Any point of the ROC with slope larger/smaller than 1 indicates that the value of binary-response criterion τ is more likely under the latent distribution associated with Same/Change trials (for details, seeKellen, Winiger, Dunn, & Singmann, 2019). To obtain biased confidence ratings, one only needs to place the confidence criteria associated with a "change"/"same" response along a range of values in which the latent memory-strength values are more likely under the distribution associated with Same/Change trials.…”

Bias effects in a two-stage recognition paradigm: A challenge for “pure” threshold and signal detection models.

“…Our study adds to previous efforts to test decision models with joint single-item and forced-choice recognition data (Jang et al, 2009;Kellen, Winiger, Dunn, & Singmann, 2019;Smith, & Duncan, 2004). These studies explored a variety of different models and theoretical issues, but one common theme is that all of these studies included the UVSD model, and it consistently provided a good fit to the joint single-item and forced-choice data (or made good predictions about one form of data based on the other).…”