Does ROC asymmetry reverse when detecting new stimuli? Reinvestigating whether the retrievability of mnemonic information is task-dependent

Abstract

Recently, it has been suggested that the mnemonic information that underlies recognition decisions changes when participants are asked to indicate whether a test stimulus is new rather than old (Brainerd et al., 2021, Journal of Experimental Psychology: Learning Memory, and Cognition, advance online publication). However, some observations that have been interpreted as evidence for this assertion need not be due to mnemonic changes, but may instead be the result of conservative response strategies if the possibility of asymmetric receiver operating characteristics (ROCs) is taken into account. Conversely, recent findings in support of asymmetric ROCs rely on the assumption that the mnemonic information accessed by the decision-maker does not depend on whether an old or a new item is considered to be the target Kellen et al. (2021, Psychological Review 128[6], 1022-1050). Here, we aim to clarify whether there is such a difference in accessibility of mnemonic information by applying signal detection theory. To this end, we used two versions of a simultaneous detection and identification task in which we presented participants with two test stimuli at a time. In one version, the old item was the target; in the other, the new item was the target. This allowed us to assess differences in mnemonic information retrieved in the two tasks while taking possible ROC asymmetry into account. Results clearly indicate that there is indeed a difference in the accessibility of mnemonic information as postulated by (Brainerd et al., 2021, Journal of Experimental Psychology: Learning Memory, and Cognition, advance online publication).

Keywords: Old–new recognition; ROC asymmetry; Recognition memory; Signal detection theory; Simultaneous detection and identification.

Fig. 1

The EVGM (left column of panels) with parameters μ_o = 1.25, μ_n = 0.00, and σ_o = σ_n = 1.00 and the UVGM (right column of panels) with parameters μ_o = 1.25, μ_n = 0.00, σ_o = 1.30, and σ_n = 1.00. Top row of panels depicts the probability density functions of old-item (dashed lines) and new-item (solid lines) familiarity distributions, and the bottom row of panels depicts the corresponding ROCs for a single-item yes/no recognition task. Note that the ROC of the EVGM depicted in the bottom left panel is symmetric (i.e., it contains both the points {P(Hit), P(False Alarm)} and {1 − P(Hit), 1 − P(False Alarm)}; see also Kellen et al., ; Killeen & Taylor, 2004), whereas the ROC of the UVGM depicted in the bottom right panel is not

Fig. 2

ROCs of an UVGM with parameters μ_o = 1.33, μ_n = 0.00, σ_o = 1.74, and σ_n = 1.00 for a single-item yes/no recognition task in which either the old item is the target (black) or in which the new item is the target (blue). Left panel: dotted lines and squares indicate the predicted hit rates for either detecting oldness (0.64; black) or newness (0.54; blue) when the predicted false alarm rate (0.24) remains constant. Right panel: dotted lines and squares indicate the predicted false-alarm rates for either detecting oldness (0.19; black) or newness (0.27; blue) when the predicted hit rate (0.60) remains constant

Fig. 3

Illustration of the UVGM with parameters μ_o = 1.25, μ_n = 0.00, σ_o = 1.30, σ_n = 1.00 for both the SDAI and SDAI* task with m = 2 simultaneously presented test stimuli. Dotted vertical lines in the left column of panels indicate the positions of the response criteria Λ = {− 0.5,1,1.5} and black squares in the right column of panels indicate the corresponding predicted response frequencies. Top left panel: PDFs of old-item (dashed line) and new-item (solid line) familiarity distributions. Middle left panel: PDFs of the maximum familiarity value (i.e., the decision variable in an SDAI task) for a trial with one old and one new item (i.e., an SDAI target trial; dashed line) and for a trial with two new items (i.e., an SDAI non-target trial; solid line). Bottom left panel: PDFs of the minimum familiarity value (i.e., the decision variable in an SDAI* task) for a trial with one new and one old item (i.e., an SDAI* target trial; dashed line) and for a trial with two old items (i.e., an SDAI* non-target trial;solid line). Top right panel: zROCs for both the SDAI (dashed lines) and the SDAI* task (solid lines). Middle right panel: ROC (upper line) and IOC (lower line) for the SDAI task. Bottom right panel: ROC (upper line) and IOC (lower line) for the SDAI* task

Fig. 4

Relative response frequencies (crosses) in the data aggregated across participants. The length of the cross lines correspond to 95% bootstrap CIs. Top panel: empirical zROCs and the corresponding ordinary least squares linear regression lines (dashed lines) for both the SDAI (black) and the SDAI* (blue) tasks. Bottom row of panels: empirical ROCs (upper line) and IOCs (lower line) for both the SDAI (left panel, black) and the SDAI* (right panel, blue) tasks

Fig. 5

PDFs of old-item (solid lines) and new-item (dashed line) familiarity distributions according to two separate UVGMs, which were fitted to the data (aggregated across participants) from both the SDAI task (μ_o = 0.61 and σ_o = 1.24; black solid line) and the SDAI* task (μ_o = 0.46 and σ_o = 1.05; blue solid line), respectively

Fig. 6

Mean relative frequencies (black dots) of a correct identification of an old item in the SDAI task (I) and a new item in the SDAI* task (I*). Error bars depict ± 1SE (generalized linear mixed model based), gray dots and lines depict individual responses (i.e., relative response frequencies of each participant)