Using naturalistic scenes, we recently demonstrated that confidence-accuracy relations differ depending on whether recognition responses are based on memory for a specific feature or instead on general familiarity: When confidence is controlled for, accuracy is higher for familiarity-based than for feature-based responses. In the present experiment, we show that these results generalize to face recognition. Subjects studied photographs of scenes and faces presented for varying brief durations and received a recognition test on which they (1) indicated whether each picture was old or new, (2) rated their confidence in their response, and (3) indicated whether their response was based on memory for a feature or on general familiarity. For both stimulus types, subjects were more accurate and more confident for their feature-based than for their familiarity-based responses. However, when confidence was held constant, accuracy was higher for familiaritybased than for feature-based responses. These results demonstrate an important similarity between face and scene recognition and show that for both types of stimuli, confidence and accuracy are based on different information.Keywords Human memory . Face recognition .
Recollection . FamiliarityAlthough confidence and accuracy tend to be correlated in recognition memory experiments, the relation is far from perfect. One reason why confidence is not always a good predictor of accuracy is that confidence and accuracy are not always based on the same underlying information. For instance, there is good evidence that metacognitive assumptions sometimes influence confidence in ways that are unrelated to accuracy; for instance, Busey, Tunnicliff, Loftus, and Loftus (2000) found that increasing the luminance of test items in a recognition experiment produced an increase in recognition confidence but no corresponding accuracy increase. The likely reason is that increasing luminance caused the items to become more salient, and subjects assumed that salient items were more likely to be correctly remembered.Reinitz, Peria, Séguin, and Loftus (2011) specifically tested the proposal that accuracy and confidence are based in part on different underlying informational components. Many previous studies have shown that recognition responses are sometimes based on memory for a visual feature, and sometimes on general familiarity. Featurebased responses tend to be more accurate that familiaritybased responses (e.g., Loftus & Bell, 1975;Loftus & Kallman, 1979;Yonelinas, 2002); however, none of these earlier authors measured recognition confidence. Using naturalistic scenes, Reinitz et al. (2011) Psychon Bull Rev (2012) 19:1085-1093 which, in turn, allowed us to produce confidence-accuracy scatterplots, with one point corresponding to each of the 12 response type-duration conditions. Figure 1 includes an example of such a plot, using the natural-scene data from Reinitz et al. (2011). As with all other graphs in this article, up-facing, solid triangles indicate feature-based respons...