Are the contributions of processing experience and prior beliefs to confidence ratings domain-general or domain-specific?

Abstract

Previous studies on domain generality of metacognition showed inconsistent results about cross-domain correlation of metacognitive resolution, which might result from the varied relationship between actual performance and the information utilized during confidence rating across tasks. The current study investigated metacognitive domain generality using the Bayesian inference model for metamemory (BIM), which suggests that individuals integrate current processing experience and their prior beliefs to construct confidence ratings. Results from three experiments and a series of meta-analyses showed that the correlation between the contribution of processing experience to confidence ratings (parameter Pexp in BIM) across perceptual and memory domains was significantly positive, while the cross-domain correlation of metacognitive resolution (meta-d’/d’) was relatively weak. Furthermore, meta-d’/d’ was related to specific task requirements, which could lead to very low cross-task correlation of meta-d’/d’ even within the same cognitive domain. These results imply that the cross-domain correlation of metacognitive resolution might underestimate metacognitive domain generality, and the cognitive mechanisms underlying confidence rating process itself may be more domain-general.

Decision criteria in signal detection model are not based on the objective likelihood ratio.

“…100. These results indicated that the dot discrimination task was easier than the color discrimination task, consistent with the findings observed in previous study (Hu et al, 2023).…”

“…The procedure for the recognition memory test was similar to that in previous studies (e.g., Fleming et al, 2014; Hu et al, 2023; McCurdy et al, 2013). The task contained four blocks, and in each block participants were first presented with 50 words simultaneously on the screen (arranged in 10 rows and five columns) and asked to memorize as many words as possible (see Figure 9A).…”

Decision criteria in signal detection model are not based on the objective likelihood ratio

“…The estimated value of d' was reliably greater in the dot discrimination task (M = 2.23, SD = 0.35) than color discrimination task (M = 1.12, SD = 0.33), t( 49 These results indicated that the dot discrimination task was easier than the color discrimination task, consistent with the findings observed in previous study (Hu et al, 2022).…”

“…In Experiment 1, participants were asked to perform two 2AFC perceptual decision tasks, including a color discrimination task and a dot discrimination task, and give their response using a 6-point confidence rating scale. The two tasks were taken from a previous study by Hu et al (2022), in which the performance was reliably higher for dot discrimination task than color discrimination task (i.e., the dot discrimination task is easier). Here we used 2AFC tasks because participants' response in a 2AFC task can be characterized by SDT with two distributions that have equal variance, leading to simplicity in mathematics.…”

“…The procedure for the recognition memory test was similar to that in previous studies (e.g., Fleming et al, 2014;Hu et al, 2022;McCurdy et al, 2013). The task contained four blocks, and in each block participants were first presented with 50 words simultaneously on the screen (arranged in 10 rows and 5 columns) and asked to memorize as many words as possible (see Figure 9A).…”