The two-stage processing of judgment of confidence: evidence from ERP

Abstract

Background

The judgment of confidence (JOC) refers to the confidence in the accuracy of the target item individuals have just retrieved and is a typical retrospective metacognitive monitoring process. In the classical paradigm of JOC, JOC occurs after the recognition or recall task. While initially viewed as a single-stage monitoring process, recent research on JOC suggests its internal mechanisms may be more complex, potentially encompassing both retrieval and monitoring processes. This study aims to delve into these mechanisms concerning neural temporal processes.

Methods

In this study, event-related potential (ERP) was used to compare N400 and slow-wave ERPs of high and low JOCs at different time windows using a classic JOC paradigm.

Results

Behavioral results showed an inverted-U shaped relationship between response time (RT) and JOCs, peaking at magnitude 3 before declining. There were significantly longer RT for low JOCs compared with high JOCs, along with lower recognition scores. The ERP results showed that low JOCs induced larger N400 in the right frontal lobe and right central area, while high JOCs induced larger slow-wave components (500 ~ 700ms) in the right frontal lobe.

Conclusions

Based on these findings, the present study suggests that JOC involves two processing stages. N400 reflects the process of cue acquisition, while the slow-wave component reflects the process of cue application. Furthermore, a two-stage model was proposed and validated, enriching the study of metacognition monitoring mechanisms, offering insights into the processing mechanisms of retrospective metacognitive monitoring.

Introduction

Metacognitive monitoring refers to the evaluation of an individual’s cognitive activities that are about to take place or have already been completed [1]. Depending on when the evaluation occurs, it can be categorized as prospective monitoring judgments and retrospective monitoring judgments. Prospective monitoring refers to one’s judgments about how well they will perform on an item on a future test, including judgment of learning (JOL), ease of learning (EOL), and feeling of knowing (FOK). Retrospective monitoring refers to one’s assessments of how well they performed on an item just completed, including judgment of confidence (JOC) [2, 3]. Previous research has extensively explored the internal processes of prospective monitoring, while the understanding of retrospective monitoring mechanisms, such as JOC, remains relatively limited. Therefore, it is necessary to further investigate the internal processes of retrospective metacognitive monitoring in order to enhance the understanding of the nature of metacognitive monitoring.

JOC refers the confidence in the correctness of retrieved content after learning [4, 5], which represents a typical retrospective metacognitive monitoring [2, 3]. Hence, in this study, we proposed to investigate the internal mechanisms of JOC to illuminate the processes underlying retrospective metacognitive monitoring. In the classical paradigm of JOC, JOC occurs after recognition/recall tasks and represents an individual’s monitoring of their prior memory performance. JOC differs from prospective metacognitive monitoring, which involves predicting one’s future test performance. Event-related potentials (ERPs) technology studies have found that prospective metacognitive monitoring elicits P200 and N400 components associated with perceptual fluency and retrieval [6–9], as well as slow wave components related to metacognitive monitoring [10]. These results support the notion that prospective metacognitive monitoring involves two stages: cue acquisition and cue application. During the cue acquisition stage, individuals obtain fluency cues based on the ease of encoding or retrieving memory content, while during the cue application stage, they use these cues for evaluation [6, 9]. In contrast, when individuals make JOC, they have already completed the retrieval of memory content. Previous studies indicate that JOC predominantly activate ERP components associated with metacognitive monitoring, such as late positive slow waves [8, 11]. These findings suggest that JOC includes a cue application stage. However, based on current research, it remains unclear whether there are additional stages beyond cue application. Consequently, there is ongoing debate about whether JOC involves two distinct phases: cue acquisition and cue application.

Some studies propose that JOC is a single-stage process involving only monitoring. Chua et al. [12] used functional magnetic resonance imaging (fMRI) to compare the differences in brain areas between JOC and FOK. FOK refers to an individual’s prediction how likely it is that they will be able to retrieve information in the future that they cannot currently access [13]. The results showed that the activation of JOC in the medial parietal lobe was significantly lower than FOK. Moreover, this activation was related to memory extraction in metacognition monitoring [14]. This result supports that JOC is a single-stage process only involving the monitoring.

As research on JOC deepens, emerging studies indicate that the process may not be strictly single-stage. Rather, its internal mechanisms appear to be more intricate, potentially involving multiple processing stages. Roderer and Roebers [15] used eye-tracking device to examine participants’ gaze patterns during the JOC phase. In the memory phase, participants saw Japanese symbols paired with pictures and were asked to memorize which picture corresponded to each Japanese symbol. In this phase, the Japanese symbols acted as cues and the pictures as targets. Later, participants had to choose the correct picture from four options based on the Japanese symbol they had learned. During the JOC phase, the learned Japanese symbols were displayed on the left side of the screen, with a JOC scale on the right. Participants reported their confidence level based on this scale. The study found that participants looked at both the cues (Japanese symbols) and the JOC scale. Moreover, they spent more time fixating on the cues. This suggests that participants engage in memory retrieval to gather useful information for making metacognitive judgments. This involvement of memory retrieval suggests that it plays a role in the JOC process. Previous studies have provided support for this hypothesis. Cruse and Wilding [16] studied the relationship between source memory extraction and JOC by using ERPs. In the experiment, the participants needed to memorize words during learning, and the words were presented in pink or yellow, with an equal number of words in both colors. In the subsequent test task, words which were either from the preceding study phase or were new to the experiment were presented in neutral colors, and the participants needed to make an initial old/new judgment on the presented words. If the presented word was judged as old by the participants, they also had to make a color judgment and report their confidence in the color judgment. The results showed that in the right frontal, the old/new effects for high JOCs (500-1400ms after stimulus presentation) were significantly higher than low JOCs. This indicated the evaluation of the extraction process during JOCs. The research by Addante, Ranganath, and Yonelinas [17] also supported this result. The experiment conducted by Addante et al. found significant differences in the FN400 and the late positive component (LPC) between high and low JOCs. Specifically, the FN400 is associated with familiarity and increases as the JOC increases, while the LPC which is related to recollection and source monitoring, was only observed in high JOCs. This suggests that JOC might involve two distinct processes. However, in this study, Addante et al. did not strictly distinguish recognition tasks and JOC tasks, and their research results might have been affected by the recognition process. Therefore, further verification is needed to determine the reliability of this result. Moreover, recent research has identified activation of the P200 component, related to retrieval and fluency, and the P300 component, associated with memory updating, during JOC tasks [8].

Despite previous studies have offered evidence for possible processing mechanisms in JOC, no study has clearly established whether JOC is a single-stage or two-stage process. It remains unclear whether JOC and prospective metacognitive monitoring involve similar processing stages. Furthermore, existing ERP studies on JOC have primarily used recognition tasks [18], rather than tasks specifically designed for JOC. As a result, these studies might be affected by recognition, which makes it hard to clearly say if JOC has two separate processing steps. This study aims to use a classic JOC paradigm along with ERP technology to explore the processing stages of JOC and to further investigate the similarities and differences between prospective and retrospective metacognitive monitoring. This approach will provide evidence for the generalizability and specificity of metacognition.

Schwartz [18] argued that, similar to prospective metacognitive monitoring, retrospective metacognitive monitoring also involves the use of fluency cues for evaluation, rather than relying solely on the strength of memory traces. Recent developments in the study of metacognitive monitoring suggest that the core of various research paradigms for metacognitive judgments is the evaluation of individual confidence [6]. A meta-analysis study has shown that prospective and retrospective metacognitive monitoring share brain regions, with both activating the anterior and lateral prefrontal cortex, as well as the anterior cingulate cortex [19]. These findings suggest that there are still many commonalities between different types of metacognitive monitoring. Therefore, this study draws on research methods from prospective metacognitive monitoring to explore the internal mechanism of JOC.

Previous research has identified that prospective metacognitive monitoring typically involves two stages: cue acquisition and cue application. The existence of these stages has been demonstrated through differences in amplitude for high and low metacognitive judgments across distinct time windows. Delayed JOL, a type of JOL, involves predicting one’s recognition or recall performance after a delay following the presentation of learning materials or after a number of trials. Liu et al. [9] found that high-level delayed JOL elicited larger early old/new effects in the 400 ~ 600 ms time window and larger late right-frontal old/new effects in the 800 ~ 1200 ms time window compared with low JOLs. These findings indicate that delayed JOLs involve multiple processing stages: the early old/new effect reflects cue recognition, while the late right-frontal effect reflects the evaluation process after retrieval. Irak et al. [7] found that high FOKs produced larger P200 and P300 compared with low FOKs. This suggests that FOK involves both the activation of memory representations and the updating of memory after familiar stimuli [20], as well as the processing of perceptual fluency [21]. EOL occurs early in the learning process and involves assessing the difficulty of learning materials [22]. Cong and Jia[6] found that low EOLs were associated with larger N400 amplitudes in the superior and middle frontal gyri, while high EOLs showed stronger slow waves in the medial temporal lobe, ventromedial prefrontal cortex, and dorsolateral prefrontal cortex. These results suggest that EOL involves both cue acquisition and the evaluation of cues. Therefore, in this study, we plan to investigate the internal processes of JOC by examining ERP amplitude differences between high and low JOC across different time windows If significant differences in ERP amplitudes are observed between high and low JOC across these time windows, it would suggest the existence of two distinct processing stages in JOC.

The N400 component is a central focus of this study. N400 is related to the early old/new effect in memory research [23], reflecting individuals’ memory retrieval processes. The amplitude of the N400 wave increases as the difficulty of retrieval increases [24]. This component is also related to the fluency in metacognitive monitoring processes and serves as an indicator of fluency in metacognitive verification. Previous study has shown that compared with high-level metacognitive judgments, low-level ones involved less fluency and may induce more negative N400 [25]. In other words, as retrieval becomes more difficult, the retrieval fluency decreases and the amplitude of the N400 wave becomes more negative. This reflects the process through which individuals achieve retrieval fluency through their attempts. Therefore, N400 will be one of the key ERP components in this study. Additionally, the time window for the N400 is selected as 300-500ms after stimulus presentation [10, 23, 26]. Another ERP component of interest in the study is the late slow-wave. The slow-wave appears after the N400, and it has been confirmed to be associated with metacognitive monitoring [27]. A related study on metacognitive monitoring have found that within the 350-800ms, stronger positive slow-waves were elicited in the medial frontal regions and stronger negative slow-waves were elicited in the bilateral parietal regions. These results indicate that metacognitive judgments required the utilization of retrieved cues for evaluation, which reflects the monitoring process after retrieval [10]. Similarly, JOCs also depend on retrieved fluency cues. The differences in slow-wave activity in the frontal region between high and low JOCs may indicate individual variations in how fluency cues are utilized during metacognitive monitoring processes. Therefore, this study identifies the late slow-wave as another ERP component of interest. The slow-wave appears after 400ms of stimulus presentation, the time window for the slow-wave is chosen as 500-700ms after stimulus presentation.

To summarize, in previous research on prospective metacognitive monitoring, significant differences were found in the N400 and slow wave between high and low prospective metacognitive judgments. These observed differences within specific time windows allow us to infer that there are two processing stages in prospective metacognitive monitoring. Therefore, in this study, ERPs technology will be used to investigate the differences in ERP components for high and low JOCs at different time windows. Based on this, this study proposes the following hypothesis: JOC is a process involving two stages of cue acquisition and cue application, specifically manifested as differences in ERP components in different time windows between high and low JOCs.

Method

Participants

Forty undergraduate students currently enrolled at the university were randomly recruited through recruitment posters for the study, with an average age of 20.90 years (SD = 2.76). Participants were native Chinese speakers, right-handed, and with normal or corrected-to-normal vision. They had no history of psychiatric illnesses or recent use of psychotropic medications. Prior to the experiment, participants were fully informed about the safety and procedures of the study, demonstrated understanding, and provided informed consent. At the end of the experiment, participants were briefed on the purpose of the study and provided with a certain amount of financial compensation.This study was conducted following the approval of the local ethics committee (No.2022LLSC027).

We conducted power analyses ( repeated measures ANOVA) by using G*Power [28], setting α to 0.05, effect size to 0.25, which yielded power = 0.99. It meets the statistical requirements.

Materials

The experimental materials consisted of 120 pairs of facial images and Chinese surnames (see Fig. 1). The 120 facial images were selected from the CAS-PEAL (Chinese Academy of Sciences-Pose, Expression, Accessory, and Lighting) face database [29]. Among the images, there were 60 male and 60 female photos, all of which with no facial expressions and no recognizable features (such as scars), while retaining hair and clothing information. All photos were processed uniformly, with a size of 225 × 300 pixels and consistent brightness and contrast. The 240 surnames were chosen from the Four Hundred Chinese Surnames [30], all of which were monosyllabic (such as “Li”) and included no homophones. Of these, 120 surnames were randomly paired with facial images, each image being paired with a surname. The 120 pairs of face-surname materials (60 male and 60 female) were divided into six sets of 20 pairs each, which were used as the learning materials in formal experiment. The remaining 120 surnames were used as interference options in recognition. Additionally, there were three pairs of face-surname materials as practice materials.

Fig. 1.

Diagram of face-surname materials

Design and procedure

The experiment employed a single variable (JOC conditions: high or low) within-subject design. The independent variable was the JOCs, while the dependent variables were the response time (RT) and recognition performance of JOCs. The main indicators for ERP analysis were the mean amplitudes of N400 (300 ~ 500ms) and slow-wave (500 ~ 700ms).

The entire experimental procedure was controlled using E-Prime 2.0 software. Stimuli were presented on a 17-inch monitor with a resolution of 1920 × 1080, and the participants were seated approximately 60 cm away from the display.

After wearing the ERP cap, the participants sat naturally relaxed in front of the computer screen. Before the start of the experiment, they were instructed to minimize blinking and to respond as quickly as possible while maintaining accuracy. During the preparatory phase, participants were asked to fixate on the central fixation point “+”, with their left and right hands placed on the keyboard keys “123” and “456” respectively. To prevent participant fatigue, the experiment was divided into six sets with a 1-minute rest interval between each set. Subsequently, participants were tasked with completing a practice exercise emulating the formal procedure, incorporating three pairs of face-surname combinations as practice materials. They were given the opportunity for repeated practice sessions until they felt adept at the experimental task before proceeding to the formal experiment.

The formal experiment was divided into three stages: learning, distraction, recognition and JOC.

1. Learning. A fixation point “+” is presented in the center of the screen for 500ms. Then, the fixation point disappeared, and the face-surname pairs were presented on the screen. Participants were instructed to spend 8 s memorizing the faces displayed on the screen and the surnames of the individuals in the images. In this phase, participants needed to memorize a total of 120 pairs of face-surname materials (see Fig. 2a). All materials were divided into 6 groups, with 20 pairs in each group. To avoid participant fatigue, there was a 1-minute rest period after studying two sets of face-surname materials.

2. Distraction. A mathematical calculation and its answer were presented on the screen, for example, 31 × 4–9 = 125. Participants needed to judge whether the result of the equation is correct within 5s and pressed “1” for correct or “2” for incorrect. A total of 12 equations were presented for evaluation. The distraction phase lasted for 1 minute (see Fig. 2b).

3. Recognition and JOC. A facial image was presented on the screen along with three candidate surnames, where one surname was the correct answer, and the other two were incorrect answers. One of the incorrect answers was a surname that was not learned, and the other one was a learned surname but it was a mismatch. Each memorized surname appeared once as an incorrect distractor. The positions of the three answer options were randomly arranged. Participants needed to respond within 5s and pressed the corresponding number key (1, 2, or 3) to complete the recognition test.

   After each recognition test, participants needed to make a JOC. At this point, the facial image that participants recognized before and the JOC rating options were presented on the screen (see Fig. 2c). Participants needed to give their JOC for their answers in the recognition test. JOC magnitudes increased gradually from 1 to 6 (1 means 0% confidence in the correct answer; 2 means 20% confidence in the correct answer; 3 means 40% confidence in the correct answer; 4 means 60% confidence in the correct answer; 5 means 80% confidence in the correct answer; 6 means 100% confidence in the correct answer). The JOC task needed to be completed within 3 s, and the RT of the participant was recorded.

Fig. 2.

The procedures of the experiment were used in the study

Electroencephalogram recording and preprocessing

The Electroencephalogram (EEG) was recorded by using the Neuroscan 4.5 system, and a 64-channel electrode cap based on the international standard 10–20 system was placed in standard positions to record electrophysiological signals. Electrodes were placed above the left eyebrow and approximately 1 cm below the left eye for vertical electrooculogram (VEO) recording, and two electrodes were placed below both eyebrows, approximately 1 cm from the outer corner of the eyes for horizontal electrooculogram (HEO) recording. The sampling rate was 1000 Hz. M1 was used as the reference electrode to record EEG changes, and the impedance of all electrodes was adjusted to 5kΩ or below.

For the analysis of the EEG data, the ERP data during the JOC phase were processed using the EEGLAB13_0_0b and ERPLAB 7.0 toolboxes running on MATLAB 2013b. Channels HEO, VEO, CB1, and CB2 were removed. The average reference of bilateral mastoids was used for referencing, and the data were bandpass filtered from 0.1 to 30 Hz. Independent Component Analysis (ICA) from the EEGLAB toolbox was used to remove ocular artifacts, and the data were segmented. A baseline was set from 200ms before stimulus onset, and the time window for analysis was from 200ms before to 1000ms after stimulus onset (1200ms) and corrected for baseline. Automated denoising was performed on the segmented data, with trials being excluded if the amplitude exceeded an absolute threshold of ± 100 microvolts at any channel.

Event-related potential analysis

Based on the classification and merging of JOCs by Cong and Jia [6] and Liu et al. [9], levels 1, 2, and 3 were merged into the low JOCs, while levels 4, 5, and 6 were merged into the high JOCs. Averaging the overlapping frequencies of high (4, 5, 6) and low (1, 2, 3) JOCs, the study separately documented the overlapped occurrences of high JOCs (M = 67.18, SD = 14.81) and low JOCs (M = 46.45, SD = 13.35). ERP output the mean amplitude of each electrode within specific time windows of 300-500ms (N400) and 500-700ms (slow-wave).

For the N400 and slow-wave potentials (500-700ms), following the approach of Liu et al. [9] and Müller et al. [10], the time window for the N400 was set as 300-500ms, and the mean amplitude within this window was calculated as the N400. The time window for the slow wave was set as 500-700ms, and the mean amplitude within this window was calculated as the slow wave. Subsequently, statistical analyses were conducted on the mean amplitudes within the 300-500ms and 500-700ms time windows. Referring to Cong and Jia [6], electrodes were categorized into left frontal (FL), frontal-middle (FM), right frontal (FR), left central (CL), central (CM), right central (CR), left parietal (PL), parietal-middle (PM), and right parietal (PR) regions, representing different positions of the frontal, central, and parietal lobes (see Fig. 3).

Fig. 3.

Illustration of brain region distribution

Specific electrode assignments were as follows: FL included F3, F5, F7, FC3, FC5, FT7; FM included F1, Fz, F2, FC1, FCz, FC2; FR included F4, F6, F8, FC4, FC6, FT8; CL included C3, C5, T7, CP3, CP5; CM included C1, C2, Cz, CP1, CPz, CP2; CR included C4, C6, T8, CP4, CP6; PL included P3, P5, P7, PO5, PO7; PM included P1, Pz, P2, PO3, POz, PO4; PR included P4, P6, P8, PO6, PO8. A repeated measures ANOVA of 2 (JOC condition: low, high) x 3 (brain region: F, C, P) x 3 (hemisphere: left, middle, right) was performed on the mean amplitudes within the 300-500ms and 500-700ms. Greenhouse-Geisser correction [31] was used for data that did not meet the sphericity assumption, η_p² was utilized to assess effect size, and Bonferroni correction was applied for multiple comparisons.

Results

The relative accuracy of JOC was calculated by obtaining the Gamma correlation between the JOCs and recognition performance for each participant. Subsequently, a one-sample t-test was conducted using 0 as the test value. The results revealed that the Gamma value was significantly greater than 0, t (39) = 17.91, p < 0.001, Cohen’s d = 2.83, surpassing the random chance level. This indicated that the participants diligently performed the task of JOC in this experiment, and the data obtained from this experiment were valid.

Behavioral results

A repeated measures of variance was conducted on the RT and recognition performance of each JOCs. The recognition performance was the percentage of the correct number of recognition under each JOCs. The results were shown in Table 1.

Table 1.

Performance measures of all participants at different JOCs(M ± SD)

	1	2	3	4	5	6
JOC response time (RT, ms)	703 ± 362	825 ± 281	830 ± 273	751 ± 246	608 ± 193	501 ± 156
Recognition performance	0.33 ± 0.19	0.40 ± 0.12	0.52 ± 0.18	0.60 ± 0.20	0.69 ± 0.19	0.89 ± 0.13

The results showed that there was a significant difference in the RT on JOCs, F (5,195) = 22.19, p < 0.001, η_p2 = 0.36. Further post-hoc multiple comparison analysis revealed that the RT for magnitudes 2, 3, and 4 were longer than those for magnitude 5 (ps < 0.05), and the RT of magnitudes 1, 2, 3, and 4 were longer than those for magnitude 6 (ps < 0.05). Among the magnitudes, RT at magnitudes 2 and 3 were the longest, followed by magnitudes 1, 4, and 5, with magnitude 6 having the shortest RT (see Fig. 4).

Fig. 4.

Response time of each JOC

Repeated measures analysis of variance was conducted on the recognition performance of each JOC, revealing significant differences among the JOCs in recognition performance, F (5, 195) = 79.79, p < 0.001, η_p² = 0.67. Further post- hoc multiple comparisons revealed that the recognition performance of magnitudes 1, 2, and 3 were all worse than those of magnitudes 5 and 6 (ps < 0.05), and the recognition performance increased gradually with the rise in JOCs (see Fig. 5).

Fig. 5.

Recognition performance of each JOC

The repeated measures analysis of variance revealed divergent trends in RT for JOCs between magnitudes 1, 2, 3 and magnitudes 4, 5, 6. This indicates variations in the processing of JOCs across different magnitude ranges. A comprehensive analysis encompassing all magnitudes of JOCs failed to elucidate the underlying reasons for these discrepancies. Therefore, in line with previous studies by Liu et al. [9]and Cong and Jia [6], this study combined magnitudes 1, 2, and 3 as low JOC condition, and magnitudes 4, 5, and 6 as high JOC condition, and conducted paired sample t-tests on RT, recognition performance and relative accuracy under high and low JOC conditions. The recognition performance for the high JOCs were represented as the percentage of correctly recognized items under the high JOC condition, while the recognition performance for the low JOCs were represented as the percentage of correctly recognized items under the low JOC condition (see Table 2).

Table 2.

The JOC response time (RT) and recognition performance between low JOCs and high JOCs (M ± SD)

	Low JOCs	High JOCs
JOC response time (RT, ms)	786 ± 269	620 ± 162
Recognition performance	0.42 ± 0.08	0.75 ± 0.14
Relative Accuracy	0.26 ± 0.26	0.52 ± 0.26

Paired samples t-tests were conducted on RT, recognition performance and relative accuracy for low JOCs (magnitudes 1, 2, and 3) and high JOCs (magnitudes 4, 5, and 6), revealing that RT for low JOCs were longer than those for high JOCs, t(39) = 6.56, p < 0.001, d = 0.75. Recognition performance for low JOCs were worse than those for high JOCs, t(39)= -16.71, p < 0.001, Cohen’s d  = 0.89. The relative accuracy of low JOCs was significantly lower than that of high JOCs, t(39)= -0.25, p < 0.001, Cohen’s d  = 0.99.

Analysis of average event-related potentials

A repeated measures analysis of variance was conducted on the mean amplitudes in 300- 500ms and 500-700ms, with factors of 2 (JOC condition: low JOC, high JOC) × 3 (brain region: F, C, P) × 3 (hemisphere: left, middle, right).

The N400 (300-500ms). A repeated measures analysis of variance with factors of 2 (JOC condition: low JOC, high JOC) × 3 (brain region: F, C, P) × 3 (hemisphere: left, middle, right) was conducted. The results showed that the main effect of JOC condition was not significant, F (1, 39) = 0.51, p = 0.479. The interaction of JOC condition×brain region was not significant, F(2, 78) = 2.28, p = 0.133. However, the interaction of JOCs × hemisphere was significant, F(2, 78) = 4.61, p = 0.030, η_p² = 0.11. Furthermore, the interaction of JOC condition×brain region× hemisphere was significant, F(4, 156) = 7.47, p = 0.006, η_p²= 0.16.

Further simple effects analysis revealed that N400 (300-500ms) was obtained at the right frontal and right central regions. And compared with high JOCs, low JOCs elicited more negative average mean amplitudes (see Figs. 6 and 7).

Fig. 6.

Waveform graphs of JOCscorresponding to each electrode cluster (blue line: low JOCs; red line: high JOCs)

Fig. 7.

Topography of ERP effects in low and high JOCs

Slow-wave (500-700ms). A repeated measures analysis of variance with factors of 2 (JOC condition: low JOC, high JOC) × 3 (brain region: F, C, P) × 3 (hemisphere: left, middle, right) was conducted. The results showed that the main effect of JOC condition was not significant, F(1, 39) = 0.00, p = 0.978. The interaction of JOC condition× hemisphere was not significant, F(2, 78) = 9.59, p = 0.118. However, the interaction of JOC condition× brain region was significant, F(2, 78) = 3.96, p = 0.047, η_p² = 0.09. The interaction among JOC condition× brain region× hemisphere was also significant, F(4, 156) = 5.68, p = 0.013, η_p² = 0.13.

Further simple effects analysis revealed that slow-wave was obtained at the right frontal region, and compared with low JOCs, high JOCs elicited more positive mean amplitudes (see Figs. 6 and 7).

Discussion

The present study employed ERPs to explore the time processing of JOC from a neuro-mechanism perspective. The results revealed differences in both behavioral and ERP data between high and low JOCs.

In terms of behavioral results, JOCs across the six magnitudes followed an inverted U-shaped distribution with RT increasing from magnitude 1 to magnitude 3, reaching a maximum at magnitude 3 before decreasing. Notably, recognition performance increased with JOCs, which is consistent with previous studies on prospective monitoring [6, 7, 9, 32, 33]. These findings suggested that there may be a similar processing mechanism for JOC and prospective monitoring. Additionally, low JOCs had significantly longer RT, poorer recognition performance and lower accuracy compared with high JOCs, indicating that they may represent a different type of judgment. Previous studies have shown that retrieval fluency is an important cue for JOC. Therefore, it was inferred that the difference in RT between JOCs was due to the influence of fluency extraction. According to previous studies [9, 32, 33], in the cue acquisition stage, individuals acquire fluency cues for extraction through two avenues: successful extraction attempts and the duration of extraction time, indicating the fluency of the process. Thus, when extraction is swift, signaling success, or prolonged and unsuccessful, both instances provide sufficient cues for subsequent cue application stage. Consequently, individuals can swiftly employ these acquired cues for rapid judgments. However, if individuals in the cue acquisition stage are not sure whether they can successfully extract, this uncertainty will prompt longer extraction attempts. Subsequently, in the cue application stage, more time is required to weigh the success of the extraction attempt and the length of the extraction time to evaluate the confidence level. Therefore mid JOCs need longer RT. Moreover, since items with low JOCs are less easy to extract for individuals than items with high JOCs, it takes longer for individuals to react when making low JOCs.

Previous studies on prospective metacognitive monitoring have also found a similar inverted U-shaped trend in the response time of metacognitive judgments. Cong and Jia [6] found significantly longer RT at magnitudes 3 and 4 than at magnitudes 1 and 2, and at magnitudes 5 and 6 in EOL. Similar results were also found in the JOL [9]. It can be inferred that different types of metacognitive monitoring may be based on a common psychological mechanism, and its essence may be a decision-making process based on evidence accumulation. The distinction between different types of metacognitive monitoring is mainly due to the difference in obtaining evidence. For example, in EOL, it is mainly based on encoding fluency; while in JOC, it is mainly dependent on extraction fluency. This hypothesis could be further investigated in future studies.

ERP results of JOC revealed differences in high and low JOCs in the N400. Specifically, in the right frontal and right central regions, the mean amplitude of the N400 elicited by low JOCs was greater than that elicited by high JOCs. This finding was consistent with previous studies [17]. N400 in the frontal region was associated with the early old/new effect in memory research [23, 26], reflecting individual’s processes in memory retrieval. This suggests that individuals engage in cue retrieval during JOC. Additionally, N400 was related to fluency cues in metacognitive processes. Undorf et al. [25] studied the processing fluency in high and low JOLs using relevant and irrelevant word pairs as learning materials. The results showed that compared with high metacognitive judgments, low metacognitive judgments had lower retrieval fluency and elicited higher N400. Consequently, the level of retrieval fluency affects individuals’ JOCs. In other words, compared with high JOCs, the information extraction in low JOCs was less fluent, which led to more negative N400. This finding is consistent with the behavioral results. It suggests that when individuals retrieve more difficult items, they require a longer extraction time and experience lower retrieval fluency. This increased difficulty is associated with stronger N400 amplitudes which leads to lower JOCs and accuracy. The difference in N400 between high and low JOC reflects the process by which individuals acquire retrieval fluency cues through attempted retrieval.

We observed that within the 500–700 ms, high JOCs elicited stronger late positive slow wave in the frontal region. It suggests that the positive slow-wave in the frontal region reflected the metacognitive monitoring. This result is consistent with previous studies. Skavhaug et al. [27] found that within the 550–1000 ms time window, both high and low metacognitive monitoring elicited a positive wave, indicating the presence of a monitoring process. Müller et al. [10] also supported this finding, as they found that within the 350–800 ms, metacognitive monitoring elicited higher positive slow-waves in the medial frontal region and stronger negative slow waves in bilateral occipital regions. The ERP findings are consistent with the behavioral results. It indicates that during the cue acquisition phase, shorter retrieval times are associated with higher perceived retrieval fluency. This increased fluency leads to stronger late positive slow wave during the cue application phase. Consequently, individuals are more likely to provide higher JOC and show greater accuracy. In other words, as retrieval fluency increases, individuals acquire more retrieval fluency cues. This leads to a stronger late positive slow wave, which reflects a higher level of JOC. This finding reflects the process by which individuals use retrieval fluency cues for evaluation. Overall, the results of N400 and late slow wave indicate that individuals, when making JOC, first attempt to retrieve information to obtain retrieval fluency cues. Subsequently, individuals use these cues to assess their performance.

However, this study did not observe negative waves in the occipital region, which may be due to the materials used in the experiment. The posterior occipital negative wave is linked to the activation of visual working memory. Müller [10] used images as memory materials, requiring participants to attempt retrieval and recall of these images, thereby activating the occipital region. In contrast, our study focused on surnames, which did not engage the visual working memory areas. Additionally, this study found a right hemispheric dominance for JOC. Overall, these components showed larger amplitudes in the right hemisphere compared with the left, consistent with previous research. Paynter et al. [20] found that the FOK also exhibited greater amplitude in the right hemisphere than in the left. This result suggests that metacognitive monitoring processes may involve specific right hemisphere activity. However, given that the right hemisphere is associated with face recognition [34], the right-hemisphere bias observed in our study might have influenced the JOC due to the materials used. Future research might also use different materials to explore this finding to understand its implications more fully.

Additionally, this study found that JOC has two processing stages. This is similar to prospective metacognitive monitoring. JOC represents a typical retrospective metacognitive monitoring. Therefore, we can infer that the similarity in the processing stages of JOC and prospective metacognitive monitoring indicates some commonality between the two types of metacognitive monitoring. This provides evidence for the generalizability of metacognition. However, this does not imply that prospective and retrospective metacognitive monitoring are identical processes.

Firstly, there are differences in the duration of the two stages between prospective and retrospective metacognitive monitoring. Compared with prospective metacognitive monitoring, retrospective metacognitive monitoring occurs over a shorter time period. This may be because retrospective metacognitive monitoring takes place after recognition or recall tasks. During which memory content has already been retrieved, which leads to a shorter duration for the cue acquisition phase compared with prospective metacognitive monitoring.

Secondly, even though both types of monitoring involve a cue acquisition stage, the actual cues obtained are not entirely the same. For instance, the EOL acquires encoding fluency cues [6], and the FOK is based on familiarity [8]. In contrast, retrospective metacognitive monitoring relies on retrieval fluency cues and the strength of memory traces [18].

Finally, although there are certain similarities in the ERP components elicited by prospective and retrospective metacognitive monitoring, differences in the distribution of these ERP components are observed, particularly during the cue acquisition phase. Research has shown that in prospective monitoring, ERP components related to cue acquisition exhibit differences across the entire brain [8, 9]. In contrast, in retrospective metacognitive monitoring, ERP differences during the cue acquisition phase are localized to the right frontal and central regions. These differences may be attributed to the distinct types of cues employed in prospective versus retrospective metacognitive monitoring. For instance, prospective metacognitive monitoring may rely on cues related to encoding fluency and familiarity, whereas retrospective metacognitive monitoring may utilize cues related to retrieval fluency and the strength of memory traces. In the future, researchers could further investigate these aspects to provide a deeper understanding of the underlying mechanisms.

Limitations

The present study utilized ERP technology to further explore the internal processes of JOC from a neurocognitive perspective. It was found that there were differences in N400 and slow-wave (500-700ms) between high and low JOCs. Specifically, low JOCs elicited larger N400 in the right frontal and right central areas, while high JOCs elicited stronger slow-wave in the right frontal area. This indicated that the process of JOC involved not only metacognitive monitoring but also included stages related to the extraction and the evaluation of fluency-related cues, supporting a two-stage processing model.

However, there were some limitations to the present study. Firstly, we grouped JOCs into high and low categories and conducted a preliminary investigation into the differences in two stages between high and low JOCs. Nevertheless, this categorization is still relatively crude, as research suggests that medium and low metacognitive judgments may involve a more complex processing process [33, 35]. Therefore, future research could take the different processing of medium and low JOCs into consideration, and refine the study of the mechanisms underlying JOC. Various cognitive neuroscience techniques and more appropriate statistical methods could be employed to further investigate the internal processing of JOC, ultimately constructing a dynamic developmental model of JOC.

Secondly, this study only included right-handed participants, which could have led to variability in RT and ERP components due to the use of dominant hands. Future studies should include both right-handed and left-handed participants to balance the impact of handedness on the results. Task difficulty [36] and participant fatigue [37] may also affect the results of ERP. Future research should further control these factors.

Thirdly, although this study observed greater amplitudes in the right hemisphere for ERP components associated with JOC, the potential influence of the experimental materials cannot be completely excluded. Future research should use different types of materials, such as word pairs, to validate these findings and further investigate their implications.

Lastly, while this study identified two processing stages in both prospective and retrospective metacognitive monitoring, the specific characteristics and differences of each stage remain unclear. Further research could investigate these aspects in more detail. In addition, this study only explored the internal processes of JOC. Therefore, its results can only partially represent the processing mechanisms of retrospective metacognitive monitoring. Future research should further validate these findings in other types of retrospective metacognitive monitoring to identify the general mechanisms of retrospective metacognitive monitoring.

Conclusions

The present study utilized event-related potential technology to explore the internal mechanism of JOC. It found that there is an inverted U-shaped relationship between JOC response time and JOC, and there is a significant difference between high and low JOCs in N400 and slow-wave components, which indicates that JOC is a two-stage process involving cue acquisition and cue application, supporting a two-stage processing model.

Author contributions

ZL: study design, data analysis, manuscript draft, and revision work. WD: data collection, and manuscript draft. NJ: data interpretation and revision work. All authors contributed to the article and approved the submitted version. ZL and WD have contributed equally to this work and share first authorship.

Funding

This research was funded by the National Post-funded Projects of Philosophy and Social Sciences (22FJKB019).

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The studies involving human participants were reviewed and approved by Ethical Review Committee of Hebei Normal University. The participants provided their written informed consent to participate in this study.

Consent for publication

All subjects whose faces appear in this article gave informed consent for the publication of the facial images.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Competing interests

The authors declare no competing interests.