Comparison of RK and confidence judgement ROCs in recognition memory

Abstract

Several indicators have been used to differentiate familiarity and recollection processes. One dualist theory stipulates that it is possible to decide whether memories come from a feeling of knowing or from a conscious retrieval of the encoding and storage conditions (remembering). Another dualist theory is based on an indirect estimation of familiarity and recollection via the subjective confidence associated with recognition responses, and from an analysis of the derived receiver operating characteristics (ROC). In the present study, participants were presented with target words or faces that they subsequently had to recognise among distractor words or faces. On the recognition phase, the old items were the same size or a different size. In two different conditions, participants had to report (1) their remember/know/guess judgement or (2) their confidence level for each of their recognition responses. The main goal of the experiment was to directly compare different indicators of familiarity and recollection. The results showed that it would be risky to consider the remember/know/guess method and the confidence-judgement method as strictly equivalent.

Keywords:

• Confidence judgement
• Familiarity
• Recollection
• RK judgement
• ROC

Acknowledgements

This research was supported by grant NEUR-012-041 (Neurosciences, neurologie et psychiatrie) from the Agence Nationale de la Recherche (France). C.M. was supported by the Spanish government (Consolider Ingenio 2010 CE-CSD2007-00121). Thanks are extended to Anne Cheylus for her technical assistance and to the volunteer participants.We thank Teresa Bajo, Kenneth Malmberg, and three anonymous reviewers for helpful comments on an earlier version of this paper.

Notes

¹Certain studies using the remember/know procedure suggest that some know responses are not based on feelings of familiarity but rather on guessing: Participants guess that they previously studied an item without experiencing familiarity (knowing) and without recollecting any details from the learning phase (remembering). To distinguish between knowing and guessing, a third category of responses—guess responses—has been introduced.

²In its standard form, this equation (in the Gaussian model) gives the proportion of hits as a function of the proportion of false alarms (FA), a target recollection parameter (Rt), a familiarity parameter (d'), and the decision criterion (c):

³However, Malmberg and Xu (2006) showed that the zROC slope is higher when calculated on the H and FA mean than on the mean slope of the individual zROCs. They showed how averaging can distort the shape of the zROCs such that differences in zROC slopes are observed because they become nonlinear (see also Ratcliff, McKoon, & Tindall, 1994; Ratcliff & Starns, 2009). In the experiment described here, the zROCs and their slopes were determined from aggregate zROCs, so we cannot rule out the possibility of skewing due to the calculation method. If we compute the slopes from the mean of the individual zROC slopes rather than from the slope of the aggregate zROC, the differences between the two calculation methods (confidence and RKG) remain small. More importantly, the conclusions are the same no matter which way the slope is calculated. In fact, Macmillan and Kaplan (1985) showed that the two calculation methods were very similar whenever the interindividual variance was low.

⁴Although faces are usually recognised better than words (Bruce & Young, 1986), there are many exceptions. The face superiority effect—and, more generally, the superiority of pictorial material over verbal material—has mainly been observed with between-subject experimental designs. On the other hand, and although it is not clear why, the opposite effect or no effect has often been observed when within-subject designs are used, as we did here (Roediger III, 2008).

⁵Note, however, that the accuracy level attained for this model was not higher than for UVSD or DPSD, all the while requiring a greater number of parameters.

⁶The overall decision criterion is always more strict in an RKG-judgement situation than in a confidence-judgement situation. However, the difference observed here was not statistically significant for words, but only for faces, t(23)=2.67, p=.014 for unchanged size; t(23)=2.36, p=.027 for changed size. The type of material and the stimulus size did not have a significant effect on the decision criterion.